Reviewer #2 (Public Review):

Summary:

In this paper, the function of trpγ in lipid metabolism was investigated. The authors found that lipid accumulation levels were increased in trpγ mutants and remained high during starvation; the increased TAG levels in trpγ mutants were restored by the expression of active AMPK in DH44 neurons and oral administration of the anti-diabetic drug metformin. Furthermore, oral administration of lipase, TAG, and free fatty acids effectively restored the survival of trpγ mutants under starvation conditions. These results indicate that TRPv plays an important role in the maintenance of systemic lipid levels through the proper expression of lipase. Furthermore, authors have shown that this function is mediated by DH44R2. This study provides an interesting finding in that the neuropeptide DH44 released from the brain regulates lipid metabolism through a brain-gut axis, acting on the receptor DH44R2 presumably expressed in gut cells.

Strengths:

Using Drosophila genetics, careful analysis of which cells express trpγ regulates lipid metabolism is performed in this study. The study supports its conclusions from various angles, including not only TAG levels, but also fat droplet staining and survival rate under starved conditions, and oral administration of substances involved in lipid metabolism.

Weaknesses:

Lipid metabolism in the gut of DH44R2-expressing cells should be investigated for a better understanding of the mechanism. Fat accumulation in the gut is not mechanistically linked with fat accumulation in the fat body. The function of lipase in the gut (esp. R2 region) should be addressed, e.g. by manipulating gut-lipases such as magro or Lip3 in the gut in the contest of trpγ mutant. Also, it is not clarified which cell types in the gut DH44R2 is expressed. The study also mentioned only in the text that bmm expression in the gut cannot restore lipid droplet enlargement in the fat body, but this result might be presented as a figure.

Reviewer #3 (Public Review):

In this manuscript, the authors demonstrated the significance of the TRPγ channel in regulating internal TAG levels. They found high TAG levels in TRPγ mutant, which was ascribed to a deficit in the lipolysis process due to the downregulation of brummer (bmm). It was notable that the expression of TRPγ in DH44+ PI neurons, but not dILP2+ neurons, in the brain restored the internal TAG levels and that the knockdown of TRPγ in DH44+ PI neurons resulted in an increase in TAG levels. These results suggested a non-cell autonomous effect of Dh44+PI neurons. Additionally, the expression of the TRPγ channel in Dh44 R2-expressing cells restored the internal TAG levels. The authors, however, did not provide an explanation of how TRPγ might function in both presynaptic and postsynaptic cells in the non-cell autonomous manner to regulate the TAG storage. The authors further determined the effect of TRPγ mutation on the size of lipid droplets (LD) and the lifespan and found that TRPγ mutation caused an increase in the size of LD and a decrease in the lifespan, which were reverted by feeding lipase and metformin. These were creative endeavors, I thought. The finding that DH44+ PI neurons have non-cell autonomous functions in regulating bodily metabolism (mainly sugar/lipid) in addition to directing sugar nutrient sensing and consumption is likely correct, but the paper has many loose ends. I would like to see a revision that includes more experiments to tighten up the findings and appropriate interpretations of the results.

(1) The authors need to provide interpretations or speculations as to how DH44+ PI neurons have non-cell autonomous functions in regulating the internal TAG stores, and how both presynaptic DH44 neurons and postsynaptic DH44 R2 neurons require TRPγ for lipid homeostasis.

(2) The expression of TRPγ solely in DH44 R2 neurons of TRPγ mutant flies restored the TAG phenotype, suggesting an important function mediated by TRPγ in DH44 R2 neurons. However, the authors did not document the endogenous expression of TRPγ in the DH44R2+ gut cells. This needs to be shown.

(3) While Dh44 mutant flies displayed normal internal TAG levels, Dh44R2 mutant flies exhibited elevated TAG levels (Figure 7A). This suggested that the lipolysis phenotype could be facilitated by a neuropeptide other than Dh44. Alternatively, a Dh44 neuropeptide-independent pathway could mediate the lipolysis. In either case, an additional result is needed to substantiate either one of the hypotheses.

(4) While the authors observed an increased area of fat body lipid droplets (LD) in Dh44 mutant flies (Figure 7F), they did not specify the particular region of the fat body chosen for measuring the LD area.

(5) The LD area only accounts for TAG levels in the fat body, whereas TAG can be found in many other body parts, including the R2 area as demonstrated in Figure 5A-D using Nile red staining. As such, measuring the total internal TAG levels would provide a more accurate representation of TAG levels than the average fat body LD area.

(6) In Figure 5F-I, the authors should perform the similar experiment with Dh44, Dh44R1, and Dh44R2 mutant flies.

(7) The representative image in Figure 6B does not correspond to the GFP quantification results shown in Figure 6C. In trpr1;bmm::GFP flies, the GFP signal appears stronger in starved conditions than in satiated conditions.

(8) In Figure 6H-I, fat body-specific expression of bmm reversed the increased LD area in TRPγ mutants. The authors also showed that Dh44+PI neuron-specific expression of bmm yielded a similar result. The authors need to provide an interpretation as to how bmm acts in the fat body or DH44 neurons to regulate this.

(9) The authors should explain why the DH44 R1 mutant did not represent similar results as the wild type.

(10) It would be good to have a schematic that represents the working model proposed in this manuscript.

eLife assessment

This study investigates the molecular mechanisms underlying chronic pain-related memory impairment by focusing on S1P/S1PR1 signaling in the dentate gyrus (DG) of the hippocampus. Through behavioral tests (Y-maze and Morris water maze) and RNA-seq analysis, the researchers discovered that S1P/S1PR1 signaling is crucial for determining susceptibility to memory impairment, with decreased S1PR1 expression linked to structural plasticity changes and memory deficits. This work has valuable significance and a convincing level of evidence, thus offering new insights into the mechanisms underlying chronic pain-related memory impairment.

Reviewer #1 (Public Review):

This work from Cui, Pan, Fan, et al explores memory impairment in chronic pain mouse models, a topic of great interest in the neurobiology field. In particular, the work starts from a very interesting observation, that WT mice can be divided into susceptible and unsusceptible to memory impairment upon modelling chronic pain with CCI. This observation represents the basis of the work where the authors identify the sphingosine receptor S1PR1 as down-regulated in the dentate gyrus of susceptible animals and demonstrate through an elegant range of experiments involving AAV-mediated knockdown or overexpression of S1PR1 that this receptor is involved in the memory impairment observed with chronic pain. Importantly for translational purposes, they also show that activation of S1PR1 through a pharmacological paradigm is able to rescue the memory impairment phenotype.

The authors also link these defects to reduced dendritic branching and a reduced number of mature excitatory synapses in the DG to the memory phenotype.

They then proceed to explore possible mechanisms downstream of S1PR1 that could explain this reduction in dendritic spines. They identify integrin α2 as an interactor of S1PR1 and show a reduction in several proteins involved in actin dynamic, which is crucial for dendritic spine formation and plasticity.

They thus hypothesize that the interaction between S1PR1 and Integrin α2 is fundamental for the activation of Rac1 and Cdc42 and consequently for the polymerisation of actin; a reduction in this pathway upon chronic pain would thus lead to impaired actin polymerisation, synapse formation, and thus impaired memory.

The work is of great interest and the experiments are of very good quality with results of great importance. I have however some concerns. The main concern I have relates to the last part of the work, namely Figures 8 and 9, which I feel are not at the same level as the results presented in the previous 7 Figures, which are instead outstanding.

In particular:

- In Figure 8, given the reduction in all the proteins tested, the authors need to check some additional proteins as controls. One good candidate could be RhoA, considering the authors say it is activated by S1PR2 and not by S1PR1;

- In addition to the previous point, could the authors also show that the number of neurons is not grossly different between susceptible and unsusceptible mice? This could be done by simply staining for NeuN or performing a western blot for a neuronal-specific protein (e.g. Map2 or beta3-tubulin);

- In Figure 8, the authors should also evaluate the levels of activated RAC1 and activated Cdc42, which are much more important than just basal levels of the proteins to infer an effect on actin dynamics. This is possible through kits that use specific adaptors to pulldown GTP-Rac1 and GTP-Cdc42;

- In Figure 9C, the experiment is performed in an immortalised cell line. I feel this needs to be performed at least in primary hippocampal neurons;

- In Figure 9D, the authors use a Yeast two-hybrid system to demonstrate the interaction between S1PR1 and Integrin α2. However, as the yeast two-hybrid system is based on the proximity of the GAL4 activating domain and the GAL4 binding domain, which are used to activate the transcription of reporter genes, the system is not often used when probing the interaction between transmembrane proteins. Could the authors use other transmembrane proteins as negative controls?;

- In Figure 9E, the immunoblot is very unconvincing. The bands in the inputs are very weak for both ITGA2 and S1PR1, the authors do not show the enrichment of S1PR1 upon its immunoprecipitation and the band for ITGA2 in the IP fraction has a weird appearance. Were these experiments performed on DG lysates only? If so, I suggest the authors repeat the experiment using the whole brain (or at least the whole hippocampus) so as to have more starting material. Alternatively, if this doesn't work, or in addition, they could also perform the immunoprecipitation in heterologous cells overexpressing the two proteins;

- About the point above, even if the results were convincing, the authors can't say that they demonstrate an interaction in vivo. In co-IP experiments, the interaction is much more likely to occur in the lysate during the incubation period rather than being conserved from the in vivo state. These co-IPs demonstrate the ability of proteins to interact, not necessarily that they do it in vivo. If the authors wanted to demonstrate this, they could perform a Proximity ligation assay in primary hippocampal neurons, using antibodies against S1PR1 and ITGA2.

- In Figure 9H, could the authors increase the N to see if shItga2 causes further KD in the CCI?

- To conclusively demonstrate that S1PR1 and ITGA2 participate in the same pathway, they could show that knocking down the two proteins at the same time does not have additive effects on behavioral tests compared to the knockdown of each one of them in isolation.

Other major concerns:

- Supplementary Figure 5: the image showing colocalisation between S1PR1 and CamKII is not very convincing. Is the S1PR1 antibody validated on Knockout or knockdown in immunostaining?;

- It would be interesting to check S1PR2 levels as a control in CCI-chronic animals;

- Figure 1: I am a bit concerned about the Ns in these experiments. In the chronic pain experiments, the N for Sham is around 8 whereas is around 20 for CCI animals. Although I understand higher numbers are necessary to see the susceptible and unsusceptible populations, I feel that then the same number of Sham animals should be used;

- Figures 1E and 1G have much higher Ns than the other panels. Why is that? If they have performed this high number of animals why not show them in all panels?;

- In the experiments where viral injection is performed, the authors should show a zoomed-out image of the brain to show the precision of the injection and how spread the expression of the different viruses was;

- The authors should check if there is brain inflammation in CCI chronic animals. This would be interesting to explain if this could be the trigger for the effects seen in neurons. In particular, the authors should check astrocytes and microglia. This is of interest also because the pathways altered in Figure 8A are related to viral infection;

- If the previous point shows increased brain inflammation, it would be interesting for the authors to check whether a prolonged anti-inflammatory treatment in CCI animals administered before the insurgence of memory impairment could stop it from happening;

- In addition, the authors should speculate on what could be the signal that can induce these molecular changes starting from the site of injury;

- Also, as the animals are all WT, the authors should speculate on what could render some animals prone to have memory impairments and others resistant.

Reviewer #2 (Public Review):

Summary:

The study investigates the molecular mechanisms underlying chronic pain-related memory impairment by focusing on S1P/S1PR1 signaling in the dentate gyrus (DG) of the hippocampus. Through behavioural tests (Y-maze and Morris water maze) and RNA-seq analysis, the researchers segregated chronic pain mice into memory impairment-susceptible and -unsusceptible subpopulations. They discovered that S1P/S1PR1 signaling is crucial for determining susceptibility to memory impairment, with decreased S1PR1 expression linked to structural plasticity changes and memory deficits.

Knockdown of S1PR1 in the DG induced a susceptible phenotype, while overexpression or pharmacological activation of S1PR1 promoted resistance to memory impairment and restored normal synaptic structure. The study identifies actin cytoskeleton-related pathways, including ITGA2 and its downstream Rac1/Cdc42 signaling, as key mediators of S1PR1's effects, offering new insights and potential therapeutic targets for chronic pain-related cognitive dysfunction.

This manuscript consists of a comprehensive investigation and significant findings. The study provides novel insights into the molecular mechanisms of chronic pain-related memory impairment, highlighting the critical role of S1P/S1PR1 signaling in the hippocampal dentate gyrus. The clear identification of S1P/S1PR1 as a potential therapeutic target offers promising avenues for future research and treatment strategies. The manuscript is well-structured, methodologically sound, and presents valuable contributions to the field.

Strengths:

(1) The manuscript is well-structured and written in clear, concise language. The flow of information is logical and easy to follow.

(2) The segregation of mice into memory impairment-susceptible and -unsusceptible subpopulations is innovative and well-justified. The statistical analyses are robust and appropriate for the data.

(3) The detailed examination of S1PR1 expression and its impact on synaptic plasticity and actin cytoskeleton reorganization is impressive. The findings are significant and contribute to the understanding of chronic pain-related memory impairment.

Weaknesses:

(1) Results: While the results are comprehensive, some sections are data-heavy and could be more reader-friendly with summarized key points before diving into detailed data.

(2) Discussion: There is a need for a more balanced discussion regarding the limitations of the study. For example, addressing potential biases in the animal model or limitations in the generalizability of the findings to humans would strengthen the discussion. Also, providing specific suggestions for follow-up studies would be beneficial.

(3) Conclusion: The conclusion, while concise, could better highlight the study's broader impact on the field and potential clinical implications.

eLife assessment

This important work uses an innovative approach to understand similarities between haemodynamic and electrophysiological activity of the human brain. The study provides incomplete evidence to indicate that while similar functional brain networks are used in both modalities, there is a tendency for these multi-modal networks to spatially converge at synchronous rather than asynchronous time points. This work will be of interest to neurophysiological and brain imaging researchers.

Reviewer #1 (Public Review):

The paper proposes an interesting perspective on the spatio-temporal relationship between FC in fMRI and electrophysiology. The study found that while similar network configurations are found in both modalities, there is a tendency for the networks to spatially converge more commonly at synchronous than asynchronous time points. However, my confidence in the findings and their interpretation is undermined by an apparent lack of justification for the expected outcomes for each of the proposed scenarios, and in the analysis pipeline itself.

Main Concerns

(1) Figure 1 makes sense to me conceptually, including the schematics of the trajectories, i.e.<br /> Scenario 1: Temporally convergent, same trajectories through connectome state space<br /> Scenario 2: Temporally divergent, different trajectories through connectome state space

However, based on my understanding I am concerned that these scenarios do not necessarily translate into the schematic CRP plots shown in Figure 2C, or the statements in the main text:

For Scenario 1: "epochs of cross-modal spatial similarity should occur more frequently at on-diagonal (synchronous) than off-diagonal (asynchronous) entries, resulting in an on-/off-diagonal ratio larger than unity"<br /> For Scenario 2: "epochs of spatial similarity could occur equally likely at on-diagonal and off-diagonal entries (ratio≈1)"

Where do the authors get these statements and the schematics in Figure 2C from? Are they based on previous literature, theory, or simulations?<br /> I am not convinced based on the evidence currently in the paper, that the ratio of off- to on-diagonal entries (and under what assumptions) is a definitive way to discriminate between scenarios 1 and 2.

For example, what about the case where the same network configuration reoccurs in both modalities at multiple time points? It seems to me that one would get a CRP with entries occurring equally on the on-diagonal as on the off-diagonal, regardless of whether the dynamics are matched between the two modalities or not (i.e. regardless of scenario 1 or 2 being true).

This thought experiment example might have a flaw in it, and the authors might ultimately be correct, but nonetheless, a systematic justification needs to be provided for using the ratio of off- to on-diagonal entries to discriminate between scenarios 1 and 2 (and under what assumptions it is valid).

In the absence of theory, a couple of ways I can think of to gain insight into this key aspect are:

(1) Use surrogate data for scenarios 1 and 2:<br /> a. For scenario 1: Run the CRP using a single modality. E.g. feed in the EEG into the analysis as both modality 1 AND modality 2. This should provide at least one example of CRP under scenario 1 (although it does not ensure that all CRPs under this scenario will look like this, it is at least a useful sanity check)<br /> b. For scenario 2: Run the CRP using a single modality plus a shuffled version. E.g. feed in the EEG into the analysis as both modality 1 AND a temporally shuffled version of the EEG as modality 2. The temporal shuffling of the EEG could be done by simply splitting the data into blocks of say ~10s and then shuffling them into a new order. This should provide a version of the CRP under scenario 2 (although it does not ensure that all CRPs under this scenario will look like this, it is at least a useful sanity check).

(2) Do simulations, with clearly specified assumptions, for scenarios 1 and 2. One way of doing this is to use a simplified (state-space) setup and randomly simulate N spatially fixed networks that are independently switching on and off over time (i.e. "activation" is 0 or 1). Note that this would result in a N-dimensional connectome state space.

The authors would only need to worry about simulating the network activation time courses, i.e. they would not need to bother with specifying the spatial configuration of each network, instead, they would make the implied assumption that each of these networks has the same spatial configuration in modality 1 and modality 2.

With that assumption, the CRP calculation should simply correspond to calculating, at each time i in modality 1 and time j in modality 2, the number of networks that are activating in both modality 1 and modality 2, by using their activation time courses. Using this, one can simulate and compute the CRPs for the two scenarios:<br /> a. Scenario 1: where the simulated activation timecourses are set to be the same between both modalities<br /> b. Scenario 2: where the simulated activation timecourses are simulated separately for each of the modalities

(2) Choices in the analysis pipeline leading up to the computation of FC in fMRI or EEG will affect the quality of information available in the FC. For example, but not only, the choice of parcellation (in the study, the number of parcels is very high given the number of EEG sensors). I think it is important that we see the impact of the chosen pipeline on the time-averaged connectomes, an output that the field has some idea about what is sensible. This would give confidence that the information being used in the main analyses in the paper is based on a sensible footing and relates to what the field is used to thinking about in terms of FC. This should be trivial to compute, as it is just a case of averaging the time-varying FCs being used for the CRP over all time points. Admittedly, this approach is less useful for the intracranial EEG.

(3) Leakage correction. The paper states: "To mitigate this issue, we provide results from source-localized data both with and without leakage correction (supplementary and main text, respectively)." Given that FC in EEG is dominated by spatial leakage (see Hipp paper), then I cannot see how it can be justified to look at non-spatial leakage correction results at all, let alone put them up front as the main results. All main results/figures for the scalp EEG should be done using spatial leakage-corrected EEG data.

Reviewer #2 (Public Review):

Summary:

The study investigates the brain's functional connectivity (FC) dynamics across different timescales using simultaneous recordings of intracranial EEG/source-localized EEG and fMRI. The primary research goal was to determine which of three convergence/divergence scenarios is the most likely to occur.

The results indicate that despite similar FC patterns found in different data modalities, the time points were not aligned, indicating spatial convergence but temporal divergence.

The researchers also found that FC patterns in different frequencies do not overlap significantly, emphasizing the multi-frequency nature of brain connectivity. Such asynchronous activity across frequency bands supports the idea of multiple connectivity states that operate independently and are organized into a multiplex system.

Strengths:

The data supporting the authors' claims are convincing and come from simultaneous recordings of fMRI and iEEG/EEG, which has been recently developed and adapted.

The analysis methods are solid and involve a novel approach to analyzing the co-occurrence of FC patterns across modalities (cross-modal recurrence plot, CRP) and robust statistics, including replication of the main results using multiple operationalizations of the functional connectome (e.g., amplitude, orthogonalized, and phase-based coupling).

In addition, the authors provided a detailed interpretation of the results, placing them in the context of recent advances and understanding of the relationships between functional connectivity and cognitive states.

Weaknesses:

Despite the impressive work, the paper still lacks some analyses to make it complete.

Firstly, the effect of the window size is unclear, especially in the case of different frequencies where the number of cycles that fall in a window will vary drastically. A typical oscillation lasts just a few cycles (see Myrov et al., 2024), and brain states are usually short-lived because of meta-stability (see Roberts et al., 2019).

Secondly, the authors didn't examine frequencies lower than 1Hz despite similarities between fMRI and infra-slow oscillations found in prior literature (see Palva et al., 2014; Zhang et al., 2023).

On a minor note, the phase-locking value (PLV) is positively biased for EEG data (see Palva et al., 2018) and a different metric for phase coupling could be a more appropriate choice (e.g., iPLV/wPLI, see Vinck et al., 2011). The repository with the code is also unavailable.

eLife assessment

Wang et al. presented visual (dot) motion and/or the sound of a walking person and found that EEG activity tracks the step rhythm, as well as the gait (2-step cycle) rhythm, with tentative demonstration that the gait rhythm is tracked superadditively (power for A+V condition is higher than the sum of the A-only and V-only condition). The findings will be of wide interest to those examining biological motion perception and oscillatory processes more broadly, with the potential to be important. However, at present, due to some analysis concerns - most notably, evidence of double-dipping for one of the core findings - the evidence is incomplete. Furthermore, some of the theoretical interpretations concerning entrainment must remain speculative when the authors cannot dissociate evoked responses from entrained oscillatory effects.

Reviewer #1 (Public Review):

Summary:

Shen et al. conducted three experiments to study the cortical tracking of the natural rhythms involved in biological motion (BM), and whether these involve audiovisual integration (AVI). They presented participants with visual (dot) motion and/or the sound of a walking person. They found that EEG activity tracks the step rhythm, as well as the gait (2-step cycle) rhythm. The gait rhythm specifically is tracked superadditively (power for A+V condition is higher than the sum of the A-only and V-only condition, Experiments 1a/b), which is independent of the specific step frequency (Experiment 1b). Furthermore, audiovisual integration during tracking of gait was specific to BM, as it was absent (that is, the audiovisual congruency effect) when the walking dot motion was vertically inverted (Experiment 2). Finally, the study shows that an individual's autistic traits are negatively correlated with the BM-AVI congruency effect.

Strengths:

The three experiments are well designed and the various conditions are well controlled. The rationale of the study is clear, and the manuscript is pleasant to read. The analysis choices are easy to follow, and mostly appropriate.

Weaknesses:

I only have one potential worry. The analysis for gait tracking (1 Hz) in Experiment 2 (Figures 3a/b) starts by computing a congruency effect (A/V stimulation congruent (same frequency) versus A/V incongruent (V at 1 Hz, A at either 0.6 or 1.4 Hz), separately for the Upright and Inverted conditions. Then, this congruency effect is contrasted between Upright and Inverted, in essence computing an interaction score (Congruent/Incongruent X Upright/Inverted). Then, the channels in which this interaction score is significant (by cluster-based permutation test; Figure 3a) are subselected for further analysis. This further analysis is shown in Figure 3b and described in lines 195-202. Critically, the further analysis exactly mirrors the selection criteria, i.e. it is aimed at testing the effect of Congruent/Incongruent and Upright/Inverted. This is colloquially known as "double dipping", the same contrast is used for selection (of channels, in this case) as for later statistical testing. This should be avoided, since in this case even random noise might result in a significant effect. To strengthen the evidence, either the authors could use a selection contrast that is orthogonal to the subsequent statistical test, or they could skip either the preselection step or the subsequent test. (It could be argued that the test in Figure 3b and related text is not needed to make the point - that same point is already made by the cluster-based permutation test.)

Related to the above: the test for the three-way interaction (lines 211-216) is reported as "marginally significant", with a p-value of 0.087. This is not very strong evidence.

Reviewer #2 (Public Review):

Summary:

The authors evaluate spectral changes in electroencephalography (EEG) data as a function of the congruency of audio and visual information associated with biological motion (BM) or non-biological motion. The results show supra-additive power gains in the neural response to gait dynamics, with trials in which audio and visual information were presented simultaneously producing higher average amplitude than the combined average power for auditory and visual conditions alone. Further analyses suggest that such supra-additivity is specific to BM and emerges from temporoparietal areas. The authors also find that the BM-specific supra-additivity is negatively correlated with autism traits.

Strengths:

The manuscript is well-written, with a concise and clear writing style. The visual presentation is largely clear. The study involves multiple experiments with different participant groups. Each experiment involves specific considered changes to the experimental paradigm that both replicate the previous experiment's finding yet extend it in a relevant manner.

Weaknesses:

The manuscript interprets the neural findings using mechanistic and cognitive claims that are not justified by the presented analyses and results.

First, entrainment and cortical tracking are both invoked in this manuscript, sometimes interchangeably so, but it is becoming the standard of the field to recognize their separate evidential requirements. Namely, step and gate cycles are striking perceptual or cognitive events that are expected to produce event-related potentials (ERPs). The regular presentation of these events in the paradigm will naturally evoke a series of ERPs that leave a trace in the power spectrum at stimulation rates even if no oscillations are at play. Thus, the findings should not be interpreted from an entrainment framework except if it is contextualized as speculation, or if additional analyses or experiments are carried out to support the assumption that oscillations are present. Even if oscillations are shown to be present, it is then a further question whether the oscillations are causally relevant toward the integration of biological motion and for the orchestration of cognitive processes.

Second, if only a cortical tracking account is adopted, it is not clear why the demonstration of supra-additivity in spectral amplitude is cognitively or behaviorally relevant. Namely, the fact that frequency-specific neural responses to the [audio & visual] condition are stronger than those to [audio] and [visual] combined does not mean this has implications for behavioral performance. While the correlation to autism traits could suggest some relation to behavior and is interesting in its own right, this correlation is a highly indirect way of assessing behavioral relevance. It would be helpful to test the relevance of supra-additive cortical tracking on a behavioral task directly related to the processing of biological motion to justify the claim that inputs are being integrated with the service of behavior. Under either framework, cortical tracking or entrainment, the causal relevance of neural findings toward cognition is lacking.

Overall, I believe this study finds neural correlates of biological motion, and it is possible that such neural correlates relate to behaviorally relevant neural mechanisms, but based on the current task and associated analyses this has not been shown.

Reviewer #3 (Public Review):

Summary:

The study demonstrates differential patterns of entrainment to biological motion (BM). At a basic, sensory level, the authors demonstrate entrainment to faster rhythms that make up BM (step-cycle) which seems to be separate from its audio aspects and its visual aspects (though to a much lesser degree). Ultimately this temporal scale seems to reside in a manner that does not indicate much multi-modal integration. At a higher-order, emergent rhythms in motion that are biologically relevant (gait-cycle) seem to be the result of multisensory integration. The work sheds light on the perceptual processes that are engaged in perceiving BM as well as the role of multisensory integration in these processes. Moreover, the work also outlines interesting links between shorter and longer integration windows along the sensory and multisensory processing stages.

In a series of experiments, the authors sought to investigate the role of multisensory integration in the processing of biological motion (BM). Specifically, they study neural entrainment in BM light-point walkers. Visual-only, auditory-only, and audio-visual (AV) displays were compared under different conditions.

Experiments 1a and b mainly characterized entrainment to these stimuli. Here, entrainment to step cycle (at different scales for 1a and 1b) was found to entrain in the presence of the auditory rhythm and to a certain degree also for the visual stimulus (though barely beyond the noise floor in 1b). The AV condition for this temporal scale seemed to follow an additive rule whereby the combined stimulation resulted in entrainment more or less equal to the sum of the unimodal effects. At the slower, gait cycle a slightly different pattern emerges whereby neither unimodal stimulation conditions result in entrainment however the AV condition does.

This finding was further explored in Experiment 2 where two extra manipulations were added. Point-light walkers could generally be either congruently paired with AV or incongruently. In addition, the visual BM stimulus was matched with a control consisting of an inverted BM and thus non-BM movement. This study enabled further discerning among the step- and gait-cycle findings seeing that the pattern that emerged suggested that step-cycle entrainment was consistent with a low-level process that is not selective to BM whilst gait-cycle entrainment was only found for BM. This generally replicated the findings in Experiment 1 and extended them further suggesting that entrainment seen for uni- and multisensory step cycles is reflects a different process than that captured in the gait-cycle multi-modal entrainment. The selective BM finding seemed to demonstrate a link to autistic traits within a sample of 24 participants informing a hypothesis that sensitivity to biological motion might be related to social cognition.

Strengths:

The main strengths of the paper relate to the conceptualization of BM and the way it is operationalized in the experimental design and analyses. The use of entrainment, and the tracking of different, nested aspects of BM result in seemingly clean data that demonstrate the basic pattern. The first experiments essentially provide the basic utility of the methodological innovation and the second experiment further hones in on the relevant interpretation of the findings by the inclusion of better control stimuli sets.

Another strength of the work is that it includes at a conceptual level two replications.

Weaknesses:

The statistical analysis is misleading and inadequate at times. The inclusion of the autism trait is not foreshadowed and adequately motivated and is likely underpowered. Finally, a broader discussion over other nested frequencies that might reside in the point-light walker stimuli would also be important to fully interpret the different peaks in the spectra.

Author response:

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

In addition to our responses to reviewer suggestions below, a minor bug in the calculation of CAIS was brought to our attention by a reader of our preprint. We have corrected this bug and rerun analyses, whose results became slightly stronger as noise was removed. While we were doing that, someone pointed out to us that our equations were almost the same as Kullback-Leibler divergence, which explains why our metric performed so well. We have made the numerically trivial (see before vs. after image below) mathematical change to use Kullback-Leibler divergence instead, and now have a better story, with a solid basis in information theory, as to why CAIS works.

Author response image 1.

Unfortunately, we discovered a second bug that caused our PIC correction code to fail to perform the needed correction for phylogenetic confounding. The previously reported correlation between CAIS (or ENC) with body mass no longer survives PIC-correction. We have therefore removed this analysis from the manuscript. Our story now stands more on the theoretical basis of CAIS and ENC than on the post facto validation than it previously did. We now also present CAIS and ENC on a more equal footing. ENC results are slightly stronger, while CAIS has the complementary advantage of correcting for amino acid frequencies.

The work involved in these changes, as well as some of the responses to reviews below, justifies changing the second author into a co-first author, and adding an additional coauthor (Hanon McShea) who discovered the second bug.

Reviewer #1 (Public Review): 

In this manuscript, the authors propose a new codon adaptation metric, Codon Adaptation Index of Species (CAIS), which they present as an easily obtainable proxy for effective population size. To permit between-species comparisons, they control for both amino acid frequencies and genomic GC content, which distinguishes their approach from existing ones. Having confirmed that CAIS negatively correlates with vertebrate body mass, as would be expected if small-bodied species with larger effective populations experience more efficient selection on codon usage, they then examine the relationship between CAIS and intrinsic structural disorder in proteins. 

The idea of a robust species-level measure of codon adaptation is interesting. If CAIS is indeed a reliable proxy for the effectiveness of selection, it could be useful to analyze species without reliable life history- or mutation rate data (which will apply to many of the genomes becoming available in the near future). 

A key question is whether CAIS, in fact, measures adaptation at the codon level. Unfortunately, CAIS is only validated indirectly by confirming a negative correlation with body mass. As a result, the observations about structural disorder are difficult to evaluate. 

As discussed in the preamble above, we have replaced the body mass validation with a stronger theoretical basis in information theory.

A potential problem is that differences in GC between species are not independent of life history. Effective population size can drive compositional differences due to the effects of GC-biased gene conversion (gBGC). As noted by Galtier et al. (2018), genomic GC correlates negatively with body mass in mammals and birds. It would therefore be important to examine how gBGC might affect CAIS, and to what extent it could explain the relationship between CAIS and body mass. 

Suppose that gBGC drives an increase in GC that is most pronounced at 3rd codon positions in highrecombination regions in small-bodied species. In this case, could observed codon usage depart more strongly from expectations calculated from overall genomic GC in small vertebrates compared to large ones? The authors also report that correcting for local intergenic GC was unsuccessful, based on the lack of a significant negative relationship with body mass (Figure 3D). In principle, this could also be consistent with local GC providing a relatively more appropriate baseline in regions with high recombination rates. Considering these scenarios would clarify what exactly CAIS is capturing. 

Figure 3 (previously Supplementary Figures S5A and S5B) shows that CAIS is negligibly correlated with %GC (not robust to multiple comparisons correction), and ENC not at all. We believe this is evidence against the possibility brought up by the reviewer, i.e. that Ne might affect gBGC (and hence global %GC). This relationship, if present, could act as a confounding effect, but it is not present within our species dataset. 

Note that we expect our genomic-GC-based codon usage expectations to reflect unchecked gBGC in an average genomic region, independently of whether that species has high or low Ne. Our working model is that non-selective forces, include gBGC as well as conventional mutation biases, vary among species, and that they rather than selection determine each species’ genome-wide %GC. By correcting for genome-wide %GC, CAIS and ENC correct for both mutation bias and gBGC, in order to isolate the effects of selection.

This argument, based on an average genomic region, is vulnerable to gene-rich genomic regions having differentially higher recombination rates and hence GC-biased gene conversion. However, we do not see the expected positive correlation between |𝐥𝐨𝐜𝐚𝐥 𝐆𝐂 - global GC| and CAIS (see new Figure 5), again suggesting that gene conversion strength is not a confounding factor acting on CAIS.

Given claims about "exquisitely adapted species", the case for using CAIS as a measure of codon adaptation would also be stronger if a relationship with gene expression could be demonstrated. RSCU is expected to be higher in highly expressed genes. Is there any evidence that the equivalent GCcontrolled measure behaves similarly? 

Correlations with gene expression are outside the scope of the current work, which is focused on producing and exploiting a single value of codon adaptation per species. It is indeed possible that our general approach of using Kullback-Leibler divergence to correct for genomic %GC could be useful in future work investigating differences among genes.  

The manuscript is overall easy to follow, though some additional context may be helpful for the general reader. A more detailed discussion of how this work compares to the approach taken by Galtier et al. (2018), which accounted for GC content and gBGC when examining codon preferences, would be appropriate, for example. In addition, it would have been useful to mention past work that has attempted to explicitly quantify selection on codon usage. 

One key difference between our work and that of Galtier et al. 2018 is that our approach does not rely on identifying specific codon preferences as a function of species. Our approach might therefore be robust to scenarios where different genes have different codon preferences (see Gingold et al. 2014 https://doi.org/10.1016/j.cell.2014.08.011). At a high level, our results are in broad agreement with those of Galtier et al., 2018, who found that gBGC affected all animal species, regardless of Ne, and who like us, found that the degree of selection on codon usage depended on Ne.

Reviewer #2 (Public Review): 

Summary 

The goal of the authors in this study is to develop a more reliable approach for quantifying codon usage such that it is more comparable across species. Specifically, the authors wish to estimate the degree of adaptive codon usage, which is potentially a general proxy for the strength of selection at the molecular level. To this end, the authors created the Codon Adaptation Index for Species (CAIS) that controls for differences in amino acid usage and GC% across species. Using their new metric, the authors find a previously unobserved negative correlation between the overall adaptiveness of codon usage and body size across 118 vertebrates. As body size is negatively correlated with effective population size and thus the general strength of natural selection, the negative correlation between CAIS and body size is expected. The authors argue this was previously unobserved due to failures of other popular metrics such as Codon Adaptation Index (CAI) and the Effective Number of Codons (ENC) to adequately control for differences in amino acid usage and GC content across species. Most surprisingly, the authors also find a positive relationship between CAIS and the overall "disorderedness" of a species protein domains. As some of these results are unexpected, which is acknowledged by the authors, I think it would be particularly beneficial to work with some simulated datasets. I think CAIS has the potential to be a valuable tool for those interested in comparing codon adaptation across species in certain situations. However, I have certain theoretical concerns about CAIS as a direct proxy for the efficiency of selection $sN_e$ when the mutation bias changes across species.  

Strengths 

(1) I appreciate that the authors recognize the potential issues of comparing CAI when amino acid usage varies and correct for this in CAIS. I think this is sometimes an under-appreciated point in the codon usage literature, as CAI is a relative measure of codon usage bias (i.e. only considers synonyms). However, the strength of natural selection on codon usage can potentially vary across amino acids, such that comparing mean CAI between protein regions with different amino acid biases may result in spurious signals of statistical significance (see Cope et al. Biochemica et Biophysica Acta - Biomembranes 2018 for a clear example of this). 

We now cite Cope et al. as an example of how amino acid composition can act as a confounding factor.

(2) The authors present numerous analysis using both ENC and mean CAI as a comparison to CAIS, helping given a sense of how CAIS corrects for some of the issues with these other metrics. I also enjoyed that they examined the previously unobserved relationship between codon usage bias and body size, which has bugged me ever since I saw Kessler and Dean 2014. The result comparing protein disorder to CAIS was particularly interesting and unexpected. 

Unfortunately, our previous PIC correction code was buggy, and in fact the relationship with body size does not survive PIC correction (although it is strong prior to PIC correction). We have therefore removed it from the paper. However, the more novel result on protein disorder remains strong.

(3) The CAIS metric presented here is generally applicable to any species that has an annotated genome with protein-coding sequences. 

Weaknesses 

(1) The main weakness of this work is that it lacks simulated data to confirm that it works as expected. This would be particularly useful for assessing the relationship between CAIS and the overall effect of protein structure disorder, which the authors acknowledge is an unexpected result. I think simulations could also allow the authors to assess how their metric performs in situations where mutation bias and natural selection act in the same direction vs. opposite directions. Additionally, although I appreciate their comparisons to ENC and mean CAI, the lack of comparison to other popular codon metrics for calculating the overall adaptiveness of a genome (e.g. dos Reis et al.'s $S$ statistic, which is a function of tRNA Adaptation Index (tAI) and ENC) may be more appropriate. Even if results are similar to $S$, CAIS has a noted advantage that it doesn't require identifying tRNA gene copy numbers or abundances, which I think are generally less readily available than genomic GC% and protein-coding sequences. 

The main limitation of dos Reis’s test in our view is that, like the better versions of CAI, it requires comparable orthologs across species. See also the discussion below re the benefits of proteome-wide approach. We now also note the advantage of not needing tRNA gene copy numbers and abundances. 

Simulated datasets would be great, but we think it a nice addition rather than must-have, in particular because we are skeptical about whether our understanding of all relevant processes is good enough such that simulations would add much to our more heuristic argument along the lines of Figure 2. E.g. the complications of Gingold et al. 2014 cited above are pertinent, but incorporating them would make simulations quite involved. Instead, we now have a stronger theoretical justification for CAIS grounded in information theory. We have significantly expanded discussion of Figure 2 to give a clearer idea of the conceptual underpinnings of CAIS and ENC.

The authors mention the selection-mutation-drift equilibrium model, which underlies the basic ideas of this work (e.g. higher $N_e$ results in stronger selection on codon usage), but a more in-depth framing of CAIS in terms of this model is not given. I think this could be valuable, particularly in addressing the question "are we really estimating what we think we're estimating?" 

Let's take a closer look at the formulation for RSCUS. From here on out, subscripts will only be used to denote the codon and it will be assumed that we are only considering the case of $r = genome$ for some species $s$. 

\begin{align*} 

RSCUS_i &= \frac{O_i}{E_i} 

\end{align*} 

I think what the authors are attempting to do is "divide out" the effects of mutation bias (as given by $E_i$), such that only the effects of natural selection remain, i.e. deviations from the expected frequency based on mutation bias alone represent adaptive codon usage. Consider Gilchrist et al. MBE 2015, which says that the expected frequency of codon $i$ at selection-mutation-drift equilibrium in gene $g$ for an amino acid with $N_a$ synonymous codons is 

\begin{align} 

E_{i,g} &= \frac{e^{-\Delta M_i-\Delta\eta_i\phi_g}}{\sum_{j=1}^{N_a}e^{-\Delta M_j\Delta\eta_j\phi_g}} 

\end{align} 

where$\Delta M$ is the mutation bias, $\Delta\eta$ is the strength of selection scaled by the strength of drift, and $\phi_g$ is the gene expression level of gene $g$. In this case, $\Delta M$ and $\Delta\eta$ reflect the strength and direction of mutation bias and natural selection relative to a reference codon, for which $\Delta M_ref, \Delta\eta_ref = 0$. Assuming the selection-mutation-drift equilibrium model is generally adequate to model the true codon usage patterns in a genome (as I do and I think the authors do, too), the $E_{i,g}$ could be considered the expected observed frequency codon $i$ in gene $g$ $E[O_{i,g}]$. 

Let's re-write the $E_i = \frac{p_i}{\sum_{j=1}^{N_a}p_j}$ in the form of Gilchrist et al., such that it is a function of mutation bias $\Delta M$. For simplicity, we will consider just the two-codon case and assume the amino acid sequence is fixed. Assuming GC% is at equilibrium, the term $g_r$ and $1 - g_r$ can be written as 

\begin{align*} 

g_r &= \frac{\mu_{AT\rightarrow GC}}{\mu_{AT\rightarrow GC} + \mu_{GC\rightarrow AT}} \1 - g_r & = \frac{\mu_{GC\rightarrow AT}}{\mu_{AT\rightarrow GC} + \mu_{GC\rightarrow AT}}  \end{align*} 

where $\mu_{x\rightarrow y}$ is the mutation rate from nucleotides $x$ to $y$. As described in Gilchrist et al. MBE 2015 and Shah and Gilchrist PNAS 2011, the mutation bias $\Delta M_{NNA,NNG} = log(\frac{\mu_{AT\rightarrow GC}}{\mu_{GC\rightarrow AT}})$. This can be expressed in terms of the equilibrium GC content by recognizing that 

\begin{align*} 

\frac{g_r}{1-g_r} &= \frac{\mu_{AT\rightarrow GC}}{\mu_{GC\rightarrow AT}} \\implies \frac{g_r}{1-g_r} &= e^{\Delta M} 

\end{align*} 

As we are assuming the amino acid sequence is fixed, the probability of observing a synonymous codon $i$ at an amino acid becomes just a Bernoulli process. 

\begin{align*} 

p_i &= g_r^x(1-g_r)^{(1-x)} 

\end{align*} 

If we do this, then 

\begin{align} 

E_{NNA} &= \frac{p_{NNA}}{p_{NNA} + p_{NNG}} \ &= \frac{1-g_r}{g_r + (1-g_r)} \ &=

\frac{1}{\frac{g_r}{1-g_r} + 1} \ &= \frac{1}{e^{\Delta M} + 1} \ & = \frac{e^{-\Delta M}}{1 + e^{-\Delta M}} 

\end{align} 

Recall that in the Gilchrist et al. framework, the reference codon has $\Delta M_{NNG,NNG} = 0 \implies e^{-\Delta M_{NNG,NNG}} = 1$. Thus, we have recovered the Gilchrist et al. model from the formulation of $E_i$ under the assumption that natural selection has no impact on codon usage and codon NNG is the pre-defined reference codon. To see this, plug in 0 for $\Delta\eta$ in equation (1). 

We can then calculate the expected RSCUS using equation (1) (using notation $E[O_i]$) and equation (6) for the two codon case. For simplicity assume, we are only considering a gene of average expression (defined as $\phi_g = 1$). 

Assume in this case that NNG is the reference codon ($\Delta M_{NNG},\Delta\eta_{NNG} = 0$). 

\begin{align} 

E[RSCUS_{NNA}] &= \frac{E[O_{NNA}]}{E_{NNA}} \ &= \frac{e^{-\Delta\eta_{NNA}}(e^{-\Delta

M_{NNA}}+e^{-\Delta M_{NNG}})}{e^{-\Delta M_{NNA}-\Delta\eta_{NNA}} + e^{-\Delta M_{NNG}\Delta\eta_{NNG}}} \ & = \frac{e^{-\Delta M_{NNA} - \Delta\eta_{NNA}} + e^{-\Delta M_{NNG} -

\Delta\eta_{NNA}}}{e^{-\Delta M_{NNA} - \Delta\eta_{NNA}} + e^{-\Delta M_{NNG} -

\Delta\eta_{NNG}}} \ &= \frac{e^{-\Delta M_{NNA} - \Delta\eta_{NNA}} + e^{- \Delta\eta_{NNA}}}{e^{\Delta M_{NNA} - \Delta\eta_{NNA}} + 1} 

\end{align} 

This shows that the expected value of RSCUS for a two-codon amino acid is expected to increase as the strength of selection $\Delta\eta$ increases, which is desired. Note that $\Delta\eta$ in Gilchrist et al. is formulated in terms of selection *against* a codon relative to the reference, such that a negative value represents that a codon is favored relative to the reference. If $\Delta\eta = 0$ (i.e. selection does not favor either codon), then $E[RSCUS] = 1$. Also note that the expected RSCUS does not remain independent of the mutation bias. This means that even if $sN_e$ (i.e. the strength of natural selection) does not change between species, changes to the strength and direction of mutation bias across species could impact RSCUS. Assuming my math is right, I think one needs to be cautious when interpreting CAIS as representative of the differences in the efficiency of selection across species except under very particular circumstances. One such case could be when it is known that mutation bias varies little across the species of interest. Looking at the species used in this manuscript, most of them have a GC content ranging around 0.41, so I suspect their results are okay. 

Although I have not done so, I am sure this could be extended to the 4 and 6 codon amino acids. 

We thank Reviewer 2 for explicitly laying out the math that was implicit in our Figures 1 and 2. While we keep our more heuristic presentation, our revised manuscript now more clearly acknowledges that the per-site codon adaptation bias depicted in Figure 1 has limited sensitivity to s*Ne. The reason that we believe our approach worked despite this, is that we think the phenomenon is driven by what is shown in Figure 2. I.e., where Ne makes a difference is by determining the proteome-wide fraction of codons subject to significant codon adaptation, rather than by determining the strength of codon adaptation at any particular site or gene. We have made multiple changes to the texts to make this point clearer.

Another minor weakness of this work is that although the method is generally applicable to any species with an annotated genome and the code is publicly available, the code itself contains hard-coded values for GC% and amino acid frequencies across the 118 vertebrates. The lack of a more flexible tool may make it difficult for less computationally-experienced researchers to take advantage of this method. 

Genome-wide %GC values are hard-coded because they were taken from the previous study of James et al. (2023) https://doi.org/10.1093/molbev/msad073. As summarized in the manuscript, genome-wide %GC was a byproduct of a scan of all six reading frames across genic and intergenic sequences available from NCBI with access dates between May and July 2019. The more complicated code used to calculate the intergenic %GC, and the code used to calculate amino acid frequencies is located at https://github.com/MaselLab/CodonAdaptation-Index-of-Species. Luckily, someone else just wrote a simpler end to end pipeline for us, on the basis of our preprint. We now note this in the Acknowledgements, and link to it: https://github.com/gavinmdouglas/handy_pop_gen/blob/main/CAIS.py.

eLife assessment

This study develops a useful metric for quantifying codon usage adaptation - the Codon Adaptation Index of Species (CAIS). This metric permits direct comparisons of the strength of selection at the molecular level across species. The study is based on solid evidence, and the authors identify relationships between CAIS and the presence of disordered protein domains. Other correlations, such as the one between CAIS and body size, are weak and non-significant. In summary, the study introduces an interesting new approach to quantifying codon usage across species, which may be helpful in attempts to measure selection at the molecular level.

Reviewer #2 (Public Review):

MotorNet aims to provide a unified interface where the trained RNN controller exists within the same TensorFlow environment as the end effectors being controlled. This architecture provides a much simpler interface for the researcher to develop and iterate through computational hypotheses. In addition, the authors have built a set of biomechanically realistic end effectors (e.g., a 2 joint arm model with realistic muscles) within TensorFlow that are fully differentiable.

MotorNet will prove a highly useful starting point for researchers interested in exploring the challenges of controlling movement with realistic muscle and joint dynamics. The architecture features a conveniently modular design and the inclusion of simpler arm models provides an approachable learning curve. Other state-of-the-art simulation engines offer realistic models of muscles and multi-joint arms and afford more complex object manipulation and contact dynamics than MotorNet. However, MotorNet's approach allows for direct optimization of the controller network via gradient descent rather than reinforcement learning, which is a compromise currently required when other simulation engines (as these engines' code cannot be differentiated through).

The paper has been reorganized to provide clearer signposts to guide the reader. Importantly, the software has been rewritten atop PyTorch which is increasingly popular in ML and computational neuroscience research.

eLife assessment

This work will be of interest to the motor control community as well as neuroAI researchers interested in how bodies constrain neural circuit function. The authors present "MotorNet", a useful software package to train artificial neural networks to control a biomechanical model of an effector. The manuscript provides solid evidence that MotorNet is easy to use and can reproduce past results in the field, both at the neural and behavioural levels. Validation is limited to planar arm-like plants or point-masses, so future work exploring three-dimensional movements and other types of plants would strengthen the impact of the tool.

Reviewer #1 (Public Review):

Summary:

Codol et al. present a toolbox that allows simulating biomechanically realistic effectors and training Artificial Neural Networks (ANNs) to control them. The paper provides a detailed explanation of how the toolbox is structured and several examples demonstrating its utility.

Main comments:

(1) The paper is well-written and easy to follow. The schematics facilitate understanding of the toolbox's functionality, and the examples give insight into the potential results users can achieve.

(2) The toolbox's latest version, developed in PyTorch, is expected to offer greater benefits to the community.

(3) The new API, being compatible with Gymnasium, broadens the toolbox's application scope, enabling the use of Reinforcement Learning for training the ANNs.

Impact:

MotorNet is designed to simplify the process of simulating complex experimental setups, enabling the rapid testing of hypotheses on how the brain generates specific movements. Implemented in PyTorch and compatible with widely-used machine learning toolboxes, including Gymnasium, it offers an end-to-end pipeline for training ANNs on simulated setups. This can greatly assist experimenters in determining the focus of their subsequent efforts.

Additional context:

The main outcome of the work, a toolbox, is supplemented by a GitHub repository and a documentation webpage. Both the repository and the webpage are well-organized and user-friendly. The webpage guides users through the toolbox installation process, as well as the construction of effectors and Artificial Neural Networks (ANNs).

Author response:

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

Reviewer #1 (Public Review):

Summary:

Codol et al. present a toolbox that allows simulating biomechanically realistic effectors and training Artificial Neural Networks (ANNs) to control them. The paper provides a detailed explanation of how the toolbox is structured and several examples that demonstrate its usefulness.

Main comments:

(1) The paper is well written and easy to follow. The schematics help in understanding how the toolbox works and the examples provide an idea of the results that the user can obtain.

We thank the reviewer for this comment.

(2) As I understand it, the main purpose of the paper should be to facilitate the usage of the toolbox. For this reason, I have missed a more explicit link to the actual code. As I see it, researchers will read this paper to figure out whether they can use MotorNet to simulate their experiments, and how they should proceed if they decide to use it. I'd say the paper provides an answer to the first question and assures that the toolbox is very easy to install and use. Maybe the authors could support this claim by adding "snippets" of code that show the key steps in building an actual example.

This is an important point, which we also considered when writing this paper. We instead decided to focus on the first approach, because it is easier to illustrate the scientific use of the toolbox using code or interactive (Jupyter) notebooks than a publication format. We find the “how to proceed” aspect of the toolbox can more easily and comprehensively be covered using online, interactive tutorials. Additionally, this allows us to update these tutorials as the toolbox evolves over different versions, while it is more difficult to update a scientific article. Consequently, we explicitly avoided code snippets on the article itself. However, we appreciate that the paper would gain in clarity if this was more explicitly stated early. We have modified the paper to include a pointer to where to find tutorials online. We added this at the last paragraph of the introduction section:

The interested reader may consult the full API documentation, including interactive tutorials on the toolbox website at https://motornet.org.

(3) The results provided in Figures 1, 4, 5 and 6 are useful, because they provide examples of the type of things one can do with the toolbox. I have a few comments that might help improving them:

a. The examples in Figures 1 and 5 seem a bit redundant (same effector, similar task). Maybe the authors could show an example with a different effector or task? (see point 4).

The effectors from figures 1 and 5 are indeed very similar. However, the tasks in figure 1 and 5 present some important differences. The training procedure in figure 1 never includes any perturbations, while the one from figure 5 includes a wide range of perturbations of different magnitudes, timing and directions. The evaluation procedure of figure 1 includes center-out reaches with permanent viscous (proportional to velocity) external dynamics, while that of figure 5 are fixed, transient, square-shaped perturbation orthogonal to the reach direction. Finally, the networks in figure 1 undergo a second training procedure after evaluation while the network of figure 5 do not.

While we agree that some variation of effectors would be beneficial, we do show examples of a point-mass effector in figure 6. Overall, figure 5 shows a task that is quite different from that of figure 1 with a similar effector, while the opposite is true for figure 6. We have modified the text to clarify this for the reader, by adding the following.

End of 1st paragraph, section 2.4.

Therefore, the training protocol used for this task largely differed from section 2.1 in that the networks are exposed to a wide range of mechanical perturbations with varying characteristics.

1st paragraph of section 2.5

[…] this asymmetrical representation of PMDs during reaching movements did not occur when RNNs were trained to control an effector that lacked the geometrical properties of an arm such as illustrated in Figure 4c-e and section 2.1.

b. I missed a discussion on the relevance of the results shown in Figure 4. The moment arms are barely mentioned outside section 2.3. Are these results new? How can they help with motor control research?

We thank the reviewer for this comment. This relates to a point from reviewer 2 indicating that the purpose of each section was sometimes difficult to grasp as one reads. Section 2.3 explains the biomechanical properties that the toolbox implements to improve realism of the effector. They are not new results in the sense that other toolboxes implement these features (though not in differentiable formats) and these properties of biological muscles are empirically well-established. However, they are important to understand what the toolbox provides, and consequently what constraints networks must accommodate to learn efficient control policies. An example of this is the results in figure 6, where a simple effector versus a more biomechanically complex effector will yield different neural representations.

Regarding the manuscript itself, we agree that more clarity on the goal of every paragraph may improve the reader’s experience. Consequently, we ensured to specify such goals at the start of each section. Particularly, we clarify the purpose of section 2.3 by adding several sentences on this at the end of the first paragraph in that section. We also now clearly state the purpose of section 2.3 with the results of figure 6 and reference figure 4 in that section.

c. The results in Figure 6 are important, since one key asset of ANNs is that they provide access to the activity of the whole population of units that produces a given behavior. For this reason, I think it would be interesting to show the actual "empirical observations" that the results shown in Fig. 6 are replicating, hence allowing a direct comparison between the results obtained for biological and simulated neurons.

These empirical observations are available from previous electrophysiological and modelling work. Particularly, polar histograms across reaching directions like panel C are displayed in figures 2 and 3 of Scott, Gribble, Graham, Cabel (2001, Nature). Colormaps of modelled unit activity across time and reaching directions like panel F are also displayed in figure 2 of Lillicrap, Scott (2013, Neuron). Electrophysiological recordings of M1 neurons during a similar task in non-human primates can also be seen on “Preserved neural population dynamics across animals performing similar behaviour” figure 2 B (https://doi.org/10.1101/2022.09.26.509498) and “Nonlinear manifolds underlie neural population activity during behaviour” figure 2 B as well (https://doi.org/10.1101/2023.07.18.549575). Note that these two pre-prints use the same dataset.

We have added these citations to the text and made it explicit that they contain visualizations of similar modelling and empirical data for comparison:

This heterogeneous set of responses matches empirical observations in non-human primate primary motor cortex recordings (Churchland & Shenoy, 2007; Michaels et al., 2016) and replicate similar visualizations from previously published work (Fortunato et al., 2023; Lillicrap & Scott, 2013; Safaie et al., 2023).

(4) All examples in the paper use the arm26 plant as effector. Although the authors say that "users can easily declare their own custom-made effector and task objects if desired by subclassing the base Plant and Task class, respectively", this does not sound straightforward. Table 1 does not really clarify how to do it. Maybe an example that shows the actual code (see point 2) that creates a new plant (e.g. the 3-joint arm in Figure 7) would be useful.

Subclassing is a Python process more than a MotorNet process, as python is an object-oriented language. Therefore, there are many Python tutorials on subclassing in the general sense that would be beneficial for that purpose. We have amended the main text to ensure that this is clearer to the reader.

Subclassing a MotorNet object, in a more specific sense, requires overwriting some methods from the base MotorNet classes (e.g., Effector or Environment classes, which correspond to the original Plant and Task object, respectively). Since we made the decision (mentioned above) to not include code in the main text, we added tutorials to the online documentation, which include dedicated tutorials for MotorNet class subclassing. For instance, this tutorial showcases how to subclass Environment classes:

https://colab.research.google.com/github/OlivierCodol/MotorNet/blob/master/examples/3-environments.ipynb

(5) One potential limitation of the toolbox is that it is based on Tensorflow, when the field of Computational Neuroscience seems to be, or at least that's my impression, transitioning to pyTorch. How easy would it be to translate MotorNet to pyTorch? Maybe the authors could comment on this in the discussion.

We have received a significant amount of feedback asking for a PyTorch implementation of the toolbox. Consequently, we decided to enact this, and the next version of the toolbox will be exclusively in PyTorch. We will maintain the Application Programming Interface (API) and tutorial documentation for the TensorFlow version of the toolbox on the online website. However, going forward we will focus exclusively on bug-fixing and expanding from the latest version of MotorNet, which will be in PyTorch. We now believe that the greater popularity of PyTorch in the academic community makes that choice more sustainable while helping a greater proportion of research projects.

These changes led to a significant alteration of the MotorNet structure, which are reflected by changes made throughout the manuscript, notably in Figure 3 and Table 1.

(6) Supervised learning (SL) is widely used in Systems Neuroscience, especially because it is faster than reinforcement learning (RL). Thus providing the possibility of training the ANNs with SL is an important asset of the toolbox. However, SL is not always ideal, especially when the optimal strategy is not known or when there are different alternative strategies and we want to know which is the one preferred by the subject. For instance, would it be possible to implement a setup in which the ANN has to choose between 2 different paths to reach a target? (e.g. Kaufman et al. 2015 eLife). In such a scenario, RL seems to be a more natural option Would it be easy to extend MotorNet so it allows training with RL? Maybe the authors could comment on this in the discussion.

The new implementation of MotorNet that relies on PyTorch is already standardized to use an API that is compatible with Gymnasium. Gymnasium is a standard and popular interfacing toolbox used to link RL agents to environments. It is very well-documented and widely used, which will ensure that users who wish to employ RL to control MotorNet environments will be able to do so relatively effortlessly. We have added this point to accurately reflect the updated implementation, so users are aware that it is now a feature of the toolbox (new section 3.2.4.).

Impact:

MotorNet aims at simplifying the process of simulating complex experimental setups to rapidly test hypotheses about how the brain produces a specific movement. By providing an end-to-end pipeline to train ANNs on the simulated setup, it can greatly help guide experimenters to decide where to focus their experimental efforts.

Additional context:

Being the main result a toolbox, the paper is complemented by a GitHub repository and a documentation webpage. Both the repository and the webpage are well organized and easy to navigate. The webpage walks the user through the installation of the toolbox and the building of the effectors and the ANNs.

Reviewer #2 (Public Review):

MotorNet aims to provide a unified interface where the trained RNN controller exists within the same TensorFlow environment as the end effectors being controlled. This architecture provides a much simpler interface for the researcher to develop and iterate through computational hypotheses. In addition, the authors have built a set of biomechanically realistic end effectors (e.g., an 2 joint arm model with realistic muscles) within TensorFlow that are fully differentiable.

MotorNet will prove a highly useful starting point for researchers interested in exploring the challenges of controlling movement with realistic muscle and joint dynamics. The architecture features a conveniently modular design and the inclusion of simpler arm models provides an approachable learning curve. Other state-of-the-art simulation engines offer realistic models of muscles and multi-joint arms and afford more complex object manipulation and contact dynamics than MotorNet. However, MotorNet's approach allows for direct optimization of the controller network via gradient descent rather than reinforcement learning, which is a compromise currently required when other simulation engines (as these engines' code cannot be differentiated through).

The paper could be reorganized to provide clearer signposts as to what role each section plays (e.g., that the explanation of the moment arms of different joint models serves to illustrate the complexity of realistic biomechanics, rather than a novel discovery/exposition of this manuscript). Also, if possible, it would be valuable if the authors could provide more insight into whether gradient descent finds qualitatively different solutions to RL or other non gradient-based methods. This would strengthen the argument that a fully differentiable plant is useful beyond improving training time / computational power required (although this is a sufficiently important rationale per se).

We thank the reviewer for these comments. We agree that more clarity on the section goals may improve the reader’s experience and ensured this is the case throughout the manuscript. Particularly, we added the following on the first paragraph of section 2.3, for which an explicit goal was most missing:

In this section we illustrate some of these biomechanical properties displayed by MotorNet effectors using specific examples. These properties are well-characterised in the biology and are often implemented in realistic biomechanical simulation software.

Regarding the potential difference in solutions obtained from reinforcement or supervised learning, this would represent a non-trivial amount of work to do so conclusively and so may not be within the scope of the current article. We do appreciate however that in some situations RL may be a more fitting approach to a given task design. In relation to this point we now specify in the discussion that the new API can accommodate interfacing with reinforcement learning toolboxes for those who may want to pursue this type of policy training approach when appropriate (new section 3.2.4.).

Reviewer #3 (Public Review):

Artificial neural networks have developed into a new research tool across various disciplines of neuroscience. However, specifically for studying neural control of movement it was extremely difficult to train those models, as they require not only simulating the neural network, but also the body parts one is interested in studying. The authors provide a solution to this problem which is built upon one of the main software packages used for deep learning (Tensorflow). This allows them to make use of state-of-the-art tools for training neural networks.

They show that their toolbox is able to (re-)produce several commonly studied experiments e.g., planar reaching with and without loads. The toolbox is described in sufficient detail to get an overview of the functionality and the current state of what can be done with it. Although the authors state that only a few lines of code can reproduce such an experiment, they unfortunately don't provide any source code to reproduce their results (nor is it given in the respective repository).

The possibility of adding code snippets to the article is something we originally considered, and which aligns with comment two from reviewer one (see above). Hopefully this provides a good overview of the motivation behind our choice not to add code to the article.

The modularity of the presented toolbox makes it easy to exchange or modify single parts of an experiment e.g., the task or the neural network used as a controller. Together with the open-source nature of the toolbox, this will facilitate sharing and reproducibility across research labs.

I can see how this paper can enable a whole set of new studies on neural control of movement and accelerate the turnover time for new ideas or hypotheses, as stated in the first paragraph of the Discussion section. Having such a low effort to run computational experiments will be definitely beneficial for the field of neural control of movement.

We thank the reviewer for these comments.

Author response:

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

Reviewer #1 (Public Review):

Summary: The global decline of amphibians is primarily attributed to deadly disease outbreaks caused by the chytrid fungus, Batrachochytrium dendrobatidis (Bd). It is unclear whether and how skin-resident immune cells defend against Bd. Although it is well known that mammalian mast cells are crucial immune sentinels in the skin and play a pivotal role in immune recognition of pathogens and orchestrating subsequent immune responses, the roles of amphibian mast cells during Bd infections is largely unknown. The current study developed a novel way to enrich X. laevis skin mast cells by injecting the skin with recombinant stem cell factor (SCF), a KIT ligand required for mast cell differentiation and survival. The investigators found an enrichment of skin mast cells provides X. laevis substantial protection against Bd and mitigates the inflammation-related skin damage resulting from Bd infection. Additionally, the augmentation of mast cells leads to increased mucin content within cutaneous mucus glands and shields frogs from the alterations to their skin microbiomes caused by Bd. 

Strengths: This study underscores the significance of amphibian skin-resident immune cells in defenses against Bd and introduces a novel approach to examining interactions between amphibian hosts and fungal pathogens. 

We thank the reviewer for recognizing the significance and the novelty of our work.

Weaknesses: The main weakness of the study is lack of functional analysis of X. laevis mast cells. Upon activation, mast cells have the characteristic feature of degranulation to release histamine, serotonin, proteases, cytokines, and chemokines, etc. The study should determine whether X. laevis mast cells can be degranulated by two commonly used mast cell activators IgE and compound 48/80 for IgE-dependent and independent pathway. This can be easily done in vitro. It is also important to assess whether in vivo these mast cells are degranulated upon Bd infection using avidin staining to visualize vesicle releases from mast cells. Figure 3 only showed rSCF injection caused an increase in mast cells in naïve skin. They need to present whether Bd infection can induce mast cell increase and rSCF injection under Bd infection causes a mast cell increase in the skin. In addition, it is unclear how the enrichment of mast cells provides the protection against Bd infection and alternations to skin microbiomes after infection. It is important to determine whether skin mast cell release any contents mentioned above. 

We would like to thank the reviewer for taking the time to review our work and providing us with valuable feedback.

Please note, that as indicated in our previous rebuttal to reviewers, amphibians do not possess the IgE antibody isotype1.

To our knowledge, there are no published works describing the approaches used in studying mammalian mast cell degranulation towards examining amphibian mast cells. While there are commercially available kits and reagents for examining mammalian mast cell granule content, most of these do not cross-react with amphibian counterparts. This is especially true of cytokines and chemokines, which diverged quickly with evolution and thus do not share substantial protein sequence identity across species as diverged as frogs and mammals. We would also like to highlight the fact that several studies suggest that amphibian mast cells lack histamine2, 3, 4, 5 and serotonin2, 6. While following up on these findings would be possible, we would like to respectfully emphasize that adopting approaches used in mammalian research to comparative immunology work is not always straightforward.

As we highlight in our manuscript, frog mast cells upregulate their expression of interleukin-4 (IL4), a hallmark cytokine associated with mammalian mast cells7. The additional findings presented in our revised manuscript indicate that mast cells respond to Bd by upregulating IL4 expression in vitro and in vivo. Together, this suggests that IL4 may be a central means by which frog mast cells confer protection against Bd, by counteracting Bd-elicited inflammation, including minimizing neutrophil infiltration, maintaining skin integrity, and promoting cutaneous mucus production. Please find that these additional results are presented in Figure 8 and are described in the results and discussion sections of our revised manuscript.

Our attempts to elicit degranulation of frog mast cells using compound 48/80 have so far not been successful. This may reflect technical issues with assays optimized for mammalian mast cells or biological difference between frog and mammalian mast cells, such as species differences in mas-related G-protein coupled receptors, through which compound 48/80 acts8. We will continue to explore means to study frog mast cell degranulation both in vitro and in vivo but also respectfully point out that while degranulation is a feature commonly associated with mammalian mast cells, this is not the only means by which the mammalian mast cells confer their immunological effects. Indeed, our studies suggest that frog mast cell IL4 production may be a key means by which these cells offer anti-Bd protection.

Please note that we successfully adopted an avidin staining approach to visualize mast cell heparin content in vitro and to evaluate cutaneous mast cell numbers in vivo in control and mast cell-enriched, mock- and Bd-infected animals. This additional work is depicted in Figure 4 and addressed in the results and discussion sections of our revised manuscript.

Reviewer #2 (Public Review):

Summary: In this study, Hauser et al investigate the role of amphibian (Xenopus laevis) mast cells in cutaneous immune responses to the ecologically important pathogen Batrachochytrium dendrobatidis (Bd) using novel methods of in vitro differentiation of bone marrow-derived mast cells and in vivo expansion of skin mast cell populations. They find that bone marrow-derived myeloid precursors cultured in the presence of recombinant X. laevis Stem Cell Factor (rSCF) differentiate into cells that display hallmark characteristics of mast cells. They inject their novel (r)SCF reagent in the skin of X. laevis and find that this stimulates expansion of cutaneous mast cell populations in vivo. They then apply this model of cutaneous mast cell expansion in the setting of Bd infection and find that mast cell expansion attenuates skin burden of Bd zoospores and pathologic features including epithelial thickness and improves protective mucus production and transcriptional markers of barrier function. Utilizing their prior expertise with expanding neutrophil populations in X. laevis, the authors compare mast cell expansion using (r)SCF to neutrophil expansion using recombinant colony stimulating factor 3 (rCSF3) and find that neutrophil expansion in Bd infection leads to greater burden of zoospores and worse skin pathology. Combining these two observations, they demonstrate that mast cell expansion using rSCF attenuates cutaneous neutrophilic infiltration. They further show that mast cell expansion correlates to cutaneous IL-4 expression, and that treatment with exogenous rIL-4 reduces neutrophilic infiltration and restores markers of epithelial health, offering a mechanism by which mast cell expansion protects from Bd infection. 

Strengths: The authors report a novel method of expanding amphibian mast cells utilizing their custom-made rSCF reagent. They rigorously characterize expanded mast cells in vitro and in vivo using histologic, morphologic, transcriptional, and functional assays. This establishes solid footing with which to then study the role of rSCF-stimulated mast cell expansion in the Bd infection model. This appears to be the first demonstration of exogenous use of rSCF in amphibians to expand mast cell populations and may set a foundation for future mechanistic studies of mast cells in the X. laevis model organism. Building on prior work, they are able to contrast mast cell expansion with their neutrophil expansion model, allowing them to infer a mechanistic link between mast cell expansion and IL-4 production and subsequent suppression of neutrophil infiltration and cutaneous dysbiosis. 

We thank the reviewer for recognizing the rigorousness and utility of the studies presented in our manuscript.

Weaknesses: The main weaknesses derive from technical limitations inherent to the Xenopus model at this time. For example, in mice a mechanistic study would be expected to use IL-4 knockouts, preferably mast cell-specific, to prove the link between mast cell expansion and IL-4 production being necessary and sufficient to suppress neutrophils. However, the novel reagents in this manuscript present a compelling technical advance and a step forward in the tools available to study amphibian biology. 

We agree with the reviewer that an IL4 knock-out animal model would be a great way to support our findings. Unfortunately, working with a non-mammalian model such as X. laevis poses limitations that include lack of knock-out lines for immunology research. Moreover, as mentioned in our manuscript, we do not believe that IL4 is the sole mast cell-produced component responsible for the conferred antifungal protection. We thank the reviewer for acknowledging the limitations of our model system and recognizing the novelty, technical advances, and merits of the work presented in our manuscript.

In addition to their discussion, one open question from the revised manuscript is how a single treatment with rSCF leads to a peak in mast cell numbers and then decline to baseline in mock-infected frogs, while Bd infection either sustains rSCF-boosted mast cells or leads to steady mast cell increase over time in control-treated frogs. Whether this is mediated by endogenous SCF or some other factor remains unexplored.

This is an interesting question that we hope to explore in future studies. We did not see significant differences in skin SCF gene expression at 21 days post Bd infection. This does not rule out the possibility that the observed Bd-mediated effects to frog skin mast cell composition are not due to changes in skin SCF gene expression at earlier infection times, alone or in combination with other host or pathogen derived factors. We know that other factors are responsible for homing/retention of antimicrobial and immunosuppressive granulocyte subsets within frog skin9 and we postulate that some of these may be distinct mast cell types. Additionally, Bd is known to produce a myriad of immunomodulatory factors10, which may well also directly affect frog skin mast cell composition. Mammalian mast cells are heterogenous and are homed or recruited into tissues by an extensive array of host as well as microbiome-derived components11, 12. Undoubtedly, the frog skin mast cell composition is likewise complex, dynamic, and contingent on a plethora of host, cutaneous microbial flora- and in this case also Bd-produced factors.

References

(1) Flajnik, M.F. A cold-blooded view of adaptive immunity. Nat Rev Immunol 18, 438-453 (2018).

(2) Mulero, I., Sepulcre, M.P., Meseguer, J., Garcia-Ayala, A. & Mulero, V. Histamine is stored in mast cells of most evolutionarily advanced fish and regulates the fish inflammatory response. Proc Natl Acad Sci U S A 104, 19434-19439 (2007).

(3) Reite, O.B. A phylogenetical approach to the functional significance of tissue mast cell histamine. Nature 206, 1334-1336 (1965).

(4) Reite, O.B. Comparative physiology of histamine. Physiol Rev 52, 778-819 (1972).

(5) Takaya, K., Fujita, T. & Endo, K. Mast cells free of histamine in Rana catasbiana. Nature 215, 776-777 (1967).

(6) Galli, S.J. New insights into "the riddle of the mast cells": microenvironmental regulation of mast cell development and phenotypic heterogeneity. Lab Invest 62, 5-33 (1990).

(7) Babina, M., Guhl, S., Artuc, M. & Zuberbier, T. IL-4 and human skin mast cells revisited: reinforcement of a pro-allergic phenotype upon prolonged exposure. Archives of dermatological research 308, 665-670 (2016).

(8) Hermans, M.A.W. et al. Human Mast Cell Line HMC1 Expresses Functional Mas-Related G-Protein Coupled Receptor 2. Front Immunol 12, 625284 (2021).

(9) Hauser, K. et al. Discovery of granulocyte-lineage cells in the skin of the amphibian Xenopus laevis. FACETS 5, 571 (2020).

(10) Rollins-Smith, L.A. & Le Sage, E.H. Batrachochytrium fungi: stealth invaders in amphibian skin. Curr Opin Microbiol 61, 124-132 (2021).

(11) Halova, I., Draberova, L. & Draber, P. Mast cell chemotaxis - chemoattractants and signaling pathways. Front Immunol 3, 119 (2012).

(12) West, P.W. & Bulfone-Paus, S. Mast cell tissue heterogeneity and specificity of immune cell recruitment. Front Immunol 13, 932090 (2022).

---

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

Reviewer #1 (Public Review):

Summary:

The global decline of amphibians is primarily attributed to deadly disease outbreaks caused by the chytrid fungus, Batrachochytrium dendrobatidis (Bd). It is unclear whether and how skin-resident immune cells defend against Bd. Although it is well known that mammalian mast cells are crucial immune sentinels in the skin and play a pivotal role in the immune recognition of pathogens and orchestrating subsequent immune responses, the roles of amphibian mast cells during Bd infections are largely unknown. The current study developed a novel way to enrich X. laevis skin mast cells by injecting the skin with recombinant stem cell factor (SCF), a KIT ligand required for mast cell differentiation and survival. The investigators found an enrichment of skin mast cells provides X. laevis substantial protection against Bd and mitigates the inflammation-related skin damage resulting from Bd infection. Additionally, the augmentation of mast cells leads to increased mucin content within cutaneous mucus glands and shields frogs from the alterations to their skin microbiomes caused by Bd.

Strengths:

This study underscores the significance of amphibian skin-resident immune cells in defenses against Bd and introduces a novel approach to examining interactions between amphibian hosts and fungal pathogens. 

We thank the reviewer for acknowledging the novelty and importance of the work presented in our manuscript.

Weaknesses:

The main weakness of the study is the lack of functional analysis of X. laevis mast cells. Upon activation, mast cells have the characteristic feature of degranulation to release histamine, serotonin, proteases, cytokines, and chemokines, etc. The study should determine whether X. laevis mast cells can be degranulated by two commonly used mast cell activators IgE and compound 48/80 for IgE-dependent and independent pathways. This can be easily done in vitro. It is also important to assess whether in vivo these mast cells are degranulated upon Bd infection using avidin staining to visualize vesicle releases from mast cells. Figure 3 only showed rSCF injection caused an increase in mast cells in naïve skin. They need to present whether Bd infection can induce mast cell increase and rSCF injection under Bd infection causes a mast cell increase in the skin. In addition, it is unclear how the enrichment of mast cells provides protection against Bd infection and alternations to skin microbiomes after infection. It is important to determine whether skin mast cells release any contents mentioned above. 

We would like to thank the reviewer for taking the time to review our work and providing us with valuable feedback. We feel that we have successfully incorporated the reviewer’s suggestions into our revised manuscript, thereby improving this work.

Please note that amphibians do not possess the IgE antibody isotype1.

To our knowledge there have been no published work assimilating approaches used when studying mammalian mast cell degranulation towards examining amphibian mast cells. While there are commercially available kits and reagents for examining mammalian mast cell granule content, most of these reagents do not cross-react with amphibian counterparts. This is especially true of cytokines and chemokines, which diverged quickly with evolution and thus do not share substantial protein sequence identity across species as diverged as frogs and mammals. Additionally, several studies suggest that amphibian mast cells lack histamine2, 3, 4, 5 and serotonin2, 6. Respectfully, while following up on these findings is possible, we would not consider adopting approaches used in mammalian research to comparative immunology work as easy.

As noted in our manuscript, frog mast cells upregulate their expression of interleukin-4 (IL4), which is a hallmark cytokine associated with mammalian mast cells7. The additional findings, presented in our revised manuscript indicate that mast cells respond to Bd by upregulating IL4 expression in vitro and in vivo. In turn, our work indicates that IL4 may be a central means by which frog mast cells confer protection against Bd, by counteracting Bd-elicited inflammation, including minimizing neutrophil infiltration, maintaining skin integrity, and promoting mucus production by skin mucus glands. Please find that these additional findings are presented in Figure 8 of our revised manuscript and are described in the results and discussion sections of the paper.

Our attempts to elicit degranulation of frog mast cells using compound 48/80 have so far not been successful. This may reflect technical issues with assays optimized for mammalian mast cells or biological difference between frog and mammalian mast cells, such as species differences in mas-related G-protein coupled receptors, through which compound 48/80 acts8. We will continue explore means to study frog mast cell degranulation both in vitro and in vivo but would also like to respectfully point out that while mast cell degranulation is a feature most associated with mammalian mast cells, this is not the only means by which the mammalian mast cells confer their immunological effects. Indeed, our additional studies suggest that mast cell IL4 production may be a key means by which these cells offer anti-Bd protection.

Please find that we have adopted an avidin-staining approach to visualize mast cell heparin content in vitro and to evaluate mast cell numbers in vivo in the skins of control and mast cell-enriched, mock- and Bd-infected animals. This additional work is depicted in Figure 4 of our revised manuscript and addressed in the results and discussion sections of our revised paper.

Reviewer #2 (Public Review):

Summary:

In this study, Hauser et al investigate the role of amphibian (Xenopus laevis) mast cells in cutaneous immune responses to the ecologically important pathogen Batrachochytrium dendrobatidis (Bd) using novel methods of in vitro differentiation of bone marrow-derived mast cells and in vivo expansion of skin mast cell populations. They find that bone marrow-derived myeloid precursors cultured in the presence of recombinant X. laevis Stem Cell Factor (rSCF) differentiate into cells that display hallmark characteristics of mast cells. They inject their novel (r)SCF reagent into the skin of X. laevis and find that this stimulates the expansion of cutaneous mast cell populations in vivo. They then apply this model of cutaneous mast cell expansion in the setting of Bd infection and find that mast cell expansion attenuates the skin burden of Bd zoospores and pathologic features including epithelial thickness and improves protective mucus production and transcriptional markers of barrier function. Utilizing their prior expertise with expanding neutrophil populations in X. laevis, the authors compare mast cell expansion using (r)SCF to neutrophil expansion using recombinant colony-stimulating factor 3 (rCSF3) and find that neutrophil expansion in Bd infection leads to greater burden of zoospores and worse skin pathology. 

Strengths:

The authors report a novel method of expanding amphibian mast cells utilizing their custom-made rSCF reagent. They rigorously characterize expanded mast cells in vitro and in vivo using histologic, morphologic, transcriptional, and functional assays. This establishes solid footing with which to then study the role of rSCF-stimulated mast cell expansion in the Bd infection model. This appears to be the first demonstration of the exogenous use of rSCF in amphibians to expand mast cell populations and may set a foundation for future mechanistic studies of mast cells in the X. laevis model organism. 

We thank the reviewer for recognizing the breadth and extent of the undertaking that culminated in this manuscript. Indeed, this manuscript would not have been possible without considerable reagent development and adaptation of techniques that had previously not been used for amphibian immunity research. In line with the reviewer’s sentiment, to our knowledge this is the first report of using molecular approaches to augment amphibian mast cells, which we hope will pave the way for new areas of research within the fields of comparative immunology and amphibian disease biology.

Weaknesses:

The conclusions regarding the role of mast cell expansion in controlling Bd infection would be stronger with a more rigorous evaluation of the model, as there are some key gaps and remaining questions regarding the data. For example: 

(1) Granulocyte expansion is carefully quantified in the initial time courses of rSCF and rCSF3 injections, but similar quantification is not provided in the disease models (Figures 3E, 4G, 5D-G). A key implication of the opposing effects of mast cell vs neutrophil expansion is that mast cells may suppress neutrophil recruitment or function. Alternatively, mast cells also express notable levels of csfr3 (Figure 2) and previous work from this group (Hauser et al, Facets 2020) showed rG-CSF-stimulated peritoneal granulocytes express mast cell markers including kit and tpsab1, raising the question of what effect rCSF3 might have on mast cell populations in the skin. Considering these points, it would be helpful if both mast cells and neutrophils were quantified histologically (based on Figure 1, they can be readily distinguished by SE or Giemsa stain) in the Bd infection models. 

We thank the reviewer for this insightful suggestion. Please find that we successfully adopted an in situ hybridization approach to evaluate neutrophil numbers in the skins of control and mast cell-enriched, mock- and Bd-infected animals based on expression of the neutrophil marker, myeloperoxidase (MPO9).  Please find these results are presented in Figures 6 and 8 of our revised manuscript and addressed in the appropriate sections of our revised paper.

Our findings suggest that rSCF administration results in the accumulation of mast cells that are polarized such, that they ablate the inflammatory response elicited by Bd infection, such as through mechanisms like IL4 production. Mammalian mast cells, including peritonea-resident mast cells, express csf3r10, 11. For this reason, we used MPO expression to visualize neutrophil skin infiltration in Figures 6 and 8 of our revised work. While the X. laevis animal model does not permit nearly the degree of immune cell resolution afforded by mammalian animal models, we do know that the adult X. laevis peritonea contain a myriad of immune cell subsets. We anticipate that the high kit expression reported by Hauser et al., 2020 in the rCSF3-recruited peritoneal leukocytes reflects the presence of mast cells therein.

Please find that we have used avidin-staining and MPO in situ hybridization to respectively visualize and enumerate mast cells and neutrophils in the skin of control and mast cell-enriched, mock- and Bd-infected animals. Indeed, our results show interesting, experimental condition-dependent changes in both the skin neutrophil and mast cell numbers. The results of these additional studies are presented in Figures 4, 6 and 8 of the revised manuscript and addressed in the results and discussions sections of our revised paper.

(2) Epithelial thickness and inflammation in Bd infection are reported to be reduced by rSCF treatment (Figure 3E, 5A-B) or increased by rCSF3 treatment (Figure 4G) but quantification of these critical readouts is not shown.

We thank the reviewer for this suggestion. We scored epithelial thickness under the distinct conditions described in our manuscript and presented the quantified data in Figures 5 and 8 of the revised paper.

(3) Critical time points in the Bd model are incompletely characterized. Mast cell expansion decreases zoospore burden at 21 dpi, while there is no difference at 7 dpi (Figure 3E). Conversely, neutrophil expansion increases zoospore burden at 7 dpi, but no corresponding 21 dpi data is shown for comparison (Figure 4G). Microbiota analysis is performed at a third time point,10 dpi (Figure 5D-G), making it difficult to compare with the data from the 7 dpi and 21 dpi time points. Reporting consistent readouts at these three time points is important to draw solid conclusions about the relationship of mast cell expansion to Bd infection and shifts in microbiota.

We thank reviewer for noting this discrepancy. Please find that we have repeated our mast cell-enrichment, Bd-challenge studies, examining days 10 and 21 post infection. Our new findings indicate that compared to control animals, mast cell-enrichment does result in significant reduction in Bd loads at both 10 and 21 dpi. The difference in Bd loads between r-ctrl and rSCF-treated animals at 10 dpi corroborates the other parameters that are altered between the two treatment groups at this experimental time point.

Our question regarding the roles of inflammatory granulocytes/neutrophils during Bd infections was that of ‘how’ rather ‘when’ these cells affect Bd infections.  Thus, and because the central focus of this work was mast cells and not other granulocyte subsets; when we saw that rCSF3-recruited granulocytes adversely affect Bd infections at 7 days, we did not pursue the kinetics of these observations further. We plan to explore the roles of inflammatory mediators and immune cell subsets during the course of Bd infections but feel that these future studies are more peripheral to the central thesis of the present manuscript regarding the roles of frog mast cells during Bd infections.

(4) Although the effect of rSCF treatment on Bd zoospores is significant at 21 dpi (Figure 3E), bacterial microbiota changes at 21 dpi are not (Figure S3B-C). This discrepancy, how it relates to the bacterial microbiota changes at 10 dpi, and why 7, 10, and 21 dpi time points were chosen for these different readouts (Figure 5F-G), is not discussed.

Please find that our additional studies indicate that compared to control animals, frog skin mast cell-enrichment results in significant reduction in Bd loads at 10 dpi. This corroborate our other findings including the observation that at 10 dpi, control animals exhibit reduced microbial richness whereas mast cell-enriched frogs were protected from this disruption of their microbiome. The amphibian microbiome serves as a major barrier to these fungal infections12 and we anticipate that Bd-mediated disruption of microbial richness facilitates host skin colonization by this pathogen. In turn, we anticipate that frog mast cells are conferring the observed anti-Bd protection in part by preventing microbial disassembly and thus interfering with optimal Bd colonization and growth on frog skins. Please find that we acknowledge and discuss these notions in our revised manuscript.

(5) The time course of rSCF or rCSF3 treatments relative to Bd infection in the experiments is not clear. Were the treatments given 12 hours prior to the final analysis point to maximize the effect? For example, in Figure 3E, were rSCF injections given at 6.5 dpi and 20.5 dpi? Or were treatments administered on day 0 of the infection model? If the latter, how do the authors explain the effects at 7 dpi or 21 dpi given mast cell and neutrophil numbers return to baseline within 24 hours after rSCF or rCSF3 treatment, respectively?

Please find that in our revised manuscript, we underlined the kinetics of our animal treatments and Bd-infections. In brief, for mast cell-enrichment, animals were injected with r-ctrl or rSCF, challenged 12 hours later with Bd and examined after 10 (per reviewers’ suggestions) and 21 days of infection. For neutrophil enrichment, animals were injected with r-ctrl or rCSF3, challenged 12 hours later with Bd and examined after 7 days of infection.

The title of the manuscript may be mildly overstated. Although Bd infection can indeed be deadly, mortality was not a readout in this study, and it is not clear from the data reported that expanding skin mast cells would ultimately prevent progression to death in Bd infections.

We acknowledge this point. The revised manuscript will be titled: “Amphibian mast cells: barriers to chytrid fungus infections”.

Reviewer #3 (Public Review):

Summary:

Hauser et al. provide an exceptional study describing the role of resident mast cells in amphibian epidermis that produce anti-inflammatory cytokines that prevent Batrachochytrium dendrobatidis (Bd) infection from causing harmful inflammation, and also protect frogs from changes in skin microbiomes and loss of mucin in glands and loss of mucus integrity that otherwise cause changes to their skin microbiomes. Neutrophils, in contrast, were not protective against Bd infection. Beyond the beautiful cytology and transcriptional profiling, the authors utilized elegant cell enrichment experiments to enrich mast cells by recombinant stem cell factor, or to enrich neutrophils by recombinant colony-stimulating factor-3, and examined respective infection outcomes in Xenopus.

Strengths:

Through the use of recombinant IL4, the authors were able to test and eliminate the hypothesis that mast cell production of IL4 was the mechanism of host protection from Bd infection. Instead, impacts on the mucus glands and interaction with the skin microbiome are implicated as the protective mechanism. These results will press disease ecologists to examine the relative importance of this immune defense among species, the influence of mast cells on the skin microbiome and mucosal function, and open the potential for modulating mucosal defense.

We thank the reviewer for recognizing the utility of the work presented in our manuscript.

Weaknesses:

A reduction of bacterial diversity upon infection, as described at the end of the results section, may not always be an "adverse effect," particularly given that anti-Bd function of the microbiome increased. Some authors (see Letourneau et al. 2022 ISME, or Woodhams et al. 2023 DCI) consider these short-term alterations as encoding ecological memory, such that continued exposure to a pathogen would encounter an enriched microbial defense. Regardless, mast cell-initiated protection of the mucus layer may negate the need for this microbial memory defense.

We thank the reviewer their insightful comment. We have revised our discussion to include this notion.

While the description of the mast cell location in the epidermal skin layer in amphibians is novel, it is not known how representative these results are across species ranging in chytridiomycosis susceptibility. No management applications are provided such as methods to increase this defense without the use of recombinant stem cell factor, and more discussion is needed on how the mast cell component (abundance, distribution in the skin) of the epidermis develops or is regulated.

We thank the reviewer for this suggestion. Please find that we have added a paragraph to our revised manuscripts to address possible source(s) of skin mast cells and a statement acknowledging that greater understanding of mast cell biology across distinct amphibian species may be used to develop future strategies for management of amphibian diseases.

We are very thankful to the reviewer for this excellent suggestion but would like to point out that the work presented in our manuscript was driven by comparative immunology questions more than by conservation biology. As such and considering just how little is known about mast cells outside of mammals; we chose not to speculate too much into possible utilities of altering amphibian skin mast cell composition and instead to focus our discussion on the immediate takeaways of the work presented by our paper.

References

(1) Flajnik, M.F. A cold-blooded view of adaptive immunity. Nat Rev Immunol 18, 438-453 (2018).

(2) Mulero, I., Sepulcre, M.P., Meseguer, J., Garcia-Ayala, A. & Mulero, V. Histamine is stored in mast cells of most evolutionarily advanced fish and regulates the fish inflammatory response. Proc Natl Acad Sci U S A 104, 19434-19439 (2007).

(3) Reite, O.B. A phylogenetical approach to the functional significance of tissue mast cell histamine. Nature 206, 1334-1336 (1965).

(4) Reite, O.B. Comparative physiology of histamine. Physiol Rev 52, 778-819 (1972).

(5) Takaya, K., Fujita, T. & Endo, K. Mast cells free of histamine in Rana catasbiana. Nature 215, 776-777 (1967).

(6) Galli, S.J. New insights into "the riddle of the mast cells": microenvironmental regulation of mast cell development and phenotypic heterogeneity. Lab Invest 62, 5-33 (1990).

(7) Babina, M., Guhl, S., Artuc, M. & Zuberbier, T. IL-4 and human skin mast cells revisited: reinforcement of a pro-allergic phenotype upon prolonged exposure. Archives of dermatological research 308, 665-670 (2016).

(8) Hermans, M.A.W. et al. Human Mast Cell Line HMC1 Expresses Functional Mas-Related G-Protein Coupled Receptor 2. Front Immunol 12, 625284 (2021).

(9) Buchan, K.D. et al. A transgenic zebrafish line for in vivo visualisation of neutrophil myeloperoxidase. PLoS One 14, e0215592 (2019).

(10) Aponte-Lopez, A., Enciso, J., Munoz-Cruz, S. & Fuentes-Panana, E.M. An In Vitro Model of Mast Cell Recruitment and Activation by Breast Cancer Cells Supports Anti-Tumoral Responses. Int J Mol Sci 21 (2020).

(11) Jamur, M.C. et al. Mast cell repopulation of the peritoneal cavity: contribution of mast cell progenitors versus bone marrow derived committed mast cell precursors. BMC Immunol 11, 32 (2010).

(12) Walke, J.B. & Belden, L.K. Harnessing the Microbiome to Prevent Fungal Infections: Lessons from Amphibians. PLoS Pathog 12, e1005796 (2016).

Reviewer #2: (Recommendations For The Authors): 

We thank the reviewer for their excellent suggestions, their time reviewing this work and their help with this manuscript.

While we were not able to incorporate some of these changes, please find that we have significantly altered our manuscript in accordance with the reviewer’s suggestions from their public review. We feel that we have substantially altered our paper, including providing considerable additional data, supporting the key findings therein.

(1) The heatmap in Figure 1I appears to be scaled data, similar to Figure 4A, in which case the indicated scale numbers are not correct (e.g. they should be -2 to 2, or -3 to 3) 

Thank you for the suggestion. Please find that we have changed this figure accordingly.

(2) For Figure 1, additional curated gene lists might better illustrate the difference in cell types, e.g. include the data for a panel of mast cell genes in a heatmap (mcpt1, tpsab1, etc.) and another panel of curated neutrophil genes (e.g. lyz) in a heatmap. If the authors still have leftover RNA, qPCR verification of some of the critical genes (e.g. kit) would add to the rigor of the analysis, as this study is the foundation of a new method for culturing amphibian mast cells. 

We thank the reviewer for this suggestion. Unfortunately, we do not have leftover RNA/cDNA and we have not been able to locate mcpt1 or tpsab1 in our DEGs. We anticipate that this issue may stem from the suboptimal annotation of the Xenopus laevis genome. We agree that curating more mast cell/neutrophil genes would be ideal but feel that we have adequately highlighted those genes that are differentially expressed between the two populations in our analysis.

(3) The presentation of counts in Figure 2 is a bit hard to interpret. Although it is mentioned that everything is statistically significant, explicitly showing statistics for each gene would be better. One possibility would be to use a volcano plot (p-value vs log2 fold change) and highlight the genes shown in Figure 2, potentially with an accompanying heat map to show replicate variability. 

We thank the reviewer for this suggestion. We entertained presenting the data as volcano plots or heat maps, but in the end felt that the bar graphs better conveyed the information that we are hoping to get across. Please note that the error bars in the bar graph depict the replicate variability. Please also note that to highlight that all the depicted genes were differentially expressed, we italicized the statement in the corresponding figure legend: “All depicted genes were significantly differentially expressed between the two populations”.

(4) Narratively, it might make more sense to put Figure 4A-C with Figure 3. 

We thank the reviewer for this suggestion. Please find that we significantly revised most of our figures to better convey the content therein. We combined the content of Figure 4A-C with Figure 5A-C and added data on epidermal thickness under different conditions into this figure; Figure 5 of our revised manuscript.

(5) If possible, complementing the skin RNA-seq from rSCF treatment in Bd infection with skin RNA-seq from rCSF3 treatment to compare effects on transcriptional programs of barrier function, etc would elevate this study and add additional insights into cutaneous inflammation in the setting of Bd infection. 

We thank the reviewer for this suggestion. We anticipate that the skin inflammation caused by Bd infection is not due solely to neutrophil infiltration and artificially altering the frog skin neutrophil content would thus not recapitulate chytridiomycosis progression. We completely agree that it would be valuable to examine barrier functions in control and mast cell-enriched, Bd-infected frogs. This is something that we hope to pursue further in future studies but feel that together with our additional findings, we are presenting a significant amount of data to constitute a stand-alone story.

(6) In Figure S1A, analyzing only 3 AMP genes by qPCR is perhaps too focused. As a control, it would be useful to also test some genes known to be functionally important in neutrophil anti-microbial responses, e.g. lyz. Expanding on this experiment by performing RNA-seq on Bd-treated, bone-marrow-derived mast cells and neutrophils would be a great addition to the manuscript and an important resource for future studies in the field. The fact that the use of rSCF (or rCSF3) enables the differentiation of these cells in large numbers of pure populations presents this unique opportunity. Although IL-4 did not end up affecting mucus production, clues to the mediator(s) of this mast cell-dependent effect may be found with unbiased RNA-seq after exposure to Bd. 

We thank the reviewer for this suggestion but would like to point out that our manuscript is focused on mast cells rather than neutrophils. We also believe that in vitro exposure of leukocytes to Bd is not the most physiologically relevant model of what would happen to skin-resident and incoming immune cell subsets, since Bd primarily infects top-most keratinocytes. We anticipate that rather than coming into direct contact with the fungus, cells like mast cells and neutrophils are responding to Bd-produced and infected cell-produced products. For this reason, we did not perform RNA-seq analysis of in vitro derived mast cells or neutrophils stimulated with Bd. As we develop more X. laevis-specific reagents, we hope to revisit the question of infected skin mast cell and neutrophil gene expression profiles but are not in a position to ask these questions at this time.

This work is also guided by a finite budget, and we feel that together with our significant additional findings described in our revised manuscript, we are presenting a substantial amount of work to constitute a stand-alone story and manuscript.

Reviewer #3 (Recommendations For The Authors): 

The following are minor edits needed in the text and figure legends: 

Standardize terms such as IL4 instead of il4 or ril4 vs rIL4 throughout. Also, r-SCF vs rSCF. 

Thank you. Please find that we have standardized such terms throughout our revised manuscript. Please note that we are adhering to the convention that gene names are in lower case, protein names are in upper case and recombinant protein names are preceded by an ‘r’.

Pg 9 Change "In contract" to "In contrast". 

Thank you and changed accordingly.

Fig 4 - Perhaps indicate if results in addition to 7dpi are also available. 

Please find that we analyzed Bd loads in control and mast cell-enriched, infected frogs after 10 dpi. This data is presented in Figures 3 and 4 of our revised manuscript.

Similarly in Fig. 5, are results other than 10dpi available in the supplement? 

Please find that the results from the microbiome studies are presented in supplemental figure 3 (Fig. S3). Please note that the results presented in original manuscript Fig. 5A-C - revised manuscript Fig. 5B-E depict data for 21 dpi, which is the longest examined infection timepoint. We present data from 1 and 10 dpi in Fig. 4 of our revised manuscript.

Indicate why these days were chosen in the methods. 

Please find that we indicated why the experimental timepoints were chosen, in the methods section of our revised manuscript.

Fig S1 legend has errors in describing which panels are for which asterisks. 

Fig. S3 legend indicates panels F and G. 

Thank you. Please find that we revised our supplemental figures and amended the corresponding figure legends.

eLife assessment

This important study reveals the role of skin-resident mast cells in amphibians in mediating antimicrobial responses. The data are compelling and highlight species-specific biology that can cross-inform human mast cell biology in a species that does not rely on IgE as a primary mechanism for antimicrobial skin responses.

Reviewer #1 (Public Review):

Summary:

The global decline of amphibians is primarily attributed to deadly disease outbreaks caused by the chytrid fungus, Batrachochytrium dendrobatidis (Bd). It is unclear whether and how skin-resident immune cells defend against Bd. Although it is well known that mammalian mast cells are crucial immune sentinels in the skin and play a pivotal role in immune recognition of pathogens and orchestrating subsequent immune responses, the roles of amphibian mast cells during Bd infections is largely unknown. The current study developed a novel way to enrich X. laevis skin mast cells by injecting the skin with recombinant stem cell factor (SCF), a KIT ligand required for mast cell differentiation and survival. The investigators found an enrichment of skin mast cells provides X. laevis substantial protection against Bd and mitigates the inflammation-related skin damage resulting from Bd infection. Additionally, the augmentation of mast cells leads to increased mucin content within cutaneous mucus glands and shields frogs from the alterations to their skin microbiomes caused by Bd.

Strengths:

This study underscores the significance of amphibian skin-resident immune cells in defenses against Bd and introduces a novel approach to examining interactions between amphibian hosts and fungal pathogens.

Weaknesses:

The main weakness of the study is lack of functional analysis of X. laevis mast cells. Upon activation, mast cells have the characteristic feature of degranulation to release histamine, serotonin, proteases, cytokines, and chemokines, etc. The study should determine whether X. laevis mast cells can be degranulated by two commonly used mast cell activators IgE and compound 48/80 for IgE-dependent and independent pathway. This can be easily done in vitro. It is also important to assess whether in vivo these mast cells are degranulated upon Bd infection using avidin staining to visualize vesicle releases from mast cells. Figure 3 only showed rSCF injection caused an increase in mast cells in naïve skin. They need to present whether Bd infection can induce mast cell increase and rSCF injection under Bd infection causes a mast cell increase in the skin. In addition, it is unclear how the enrichment of mast cells provides the protection against Bd infection and alternations to skin microbiomes after infection. It is important to determine whether skin mast cell release any contents mentioned above.

Reviewer #2 (Public Review):

Summary:

In this study, Hauser et al investigate the role of amphibian (Xenopus laevis) mast cells in cutaneous immune responses to the ecologically important pathogen Batrachochytrium dendrobatidis (Bd) using novel methods of in vitro differentiation of bone marrow-derived mast cells and in vivo expansion of skin mast cell populations. They find that bone marrow-derived myeloid precursors cultured in the presence of recombinant X. laevis Stem Cell Factor (rSCF) differentiate into cells that display hallmark characteristics of mast cells. They inject their novel (r)SCF reagent in the skin of X. laevis and find that this stimulates expansion of cutaneous mast cell populations in vivo. They then apply this model of cutaneous mast cell expansion in the setting of Bd infection and find that mast cell expansion attenuates skin burden of Bd zoospores and pathologic features including epithelial thickness and improves protective mucus production and transcriptional markers of barrier function. Utilizing their prior expertise with expanding neutrophil populations in X. laevis, the authors compare mast cell expansion using (r)SCF to neutrophil expansion using recombinant colony stimulating factor 3 (rCSF3) and find that neutrophil expansion in Bd infection leads to greater burden of zoospores and worse skin pathology. Combining these two observations, they demonstrate that mast cell expansion using rSCF attenuates cutaneous neutrophilic infiltration. They further show that mast cell expansion correlates to cutaneous IL-4 expression, and that treatment with exogenous rIL-4 reduces neutrophilic infiltration and restores markers of epithelial health, offering a mechanism by which mast cell expansion protects from Bd infection.

Strengths:

The authors report a novel method of expanding amphibian mast cells utilizing their custom-made rSCF reagent. They rigorously characterize expanded mast cells in vitro and in vivo using histologic, morphologic, transcriptional, and functional assays. This establishes solid footing with which to then study the role of rSCF-stimulated mast cell expansion in the Bd infection model. This appears to be the first demonstration of exogenous use of rSCF in amphibians to expand mast cell populations and may set a foundation for future mechanistic studies of mast cells in the X. laevis model organism. Building on prior work, they are able to contrast mast cell expansion with their neutrophil expansion model, allowing them to infer a mechanistic link between mast cell expansion and IL-4 production and subsequent suppression of neutrophil infiltration and cutaneous dysbiosis.

Weaknesses:

The main weaknesses derive from technical limitations inherent to the Xenopus model at this time. For example, in mice a mechanistic study would be expected to use IL-4 knockouts, preferably mast cell-specific, to prove the link between mast cell expansion and IL-4 production being necessary and sufficient to suppress neutrophils. However, the novel reagents in this manuscript present a compelling technical advance and a step forward in the tools available to study amphibian biology.

In addition to their discussion, one open question from the revised manuscript is how a single treatment with rSCF leads to a peak in mast cell numbers and then decline to baseline in mock-infected frogs, while Bd infection either sustains rSCF-boosted mast cells or leads to steady mast cell increase over time in control-treated frogs. Whether this is mediated by endogenous SCF or some other factor remains unexplored.

Author response:

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

Reviewer #1 (Public Review):

Summary:

The study entitled "Rifampicin tolerance and growth fitness among isoniazid-resistant clinical Mycobacterium tuberculosis isolates: an in-vitro longitudinal study" by Vijay et al. provides valuable insights into the association of rifampicin tolerance and growth fitness with isoniazid resistance among clinical isolates of M. tuberculosis. Antibiotic tolerance in M. tuberculosis is an important topic since it contributes to the lengthy and complicated treatment required to cure tuberculosis disease and may portend the emergence of antibiotic resistance. The authors found that rifampicin tolerance was correlated with bacterial growth, rifampicin minimum inhibitory concentrations, and isoniazid-resistance mutations.

Strengths:

The large number of clinical isolates evaluated and their longitudinal nature during treatment for TB (including exposure to rifampin) are strengths of the study.

Weaknesses:

Some of the methodologies are not well explained or justified and the association of antibiotic tolerance with growth rate is not a novel finding. In addition, the molecular mechanisms underlying rifampicin tolerance only in rapidly growing isoniazid-resistant isolates have not been elucidated and the potential implications of these findings for clinical management are not immediately apparent.

We thank the reviewer for the comments, we have modified the method section and figure 1 to clarify the method as suggested by the reviewer.

Although we agree that previous studies have shown the association of slow growth rate with antibiotic tolerance, ours is the most comprehensive assessment of rifampicin tolerance among clinical isolates, to our knowledge. In particular, we show that the degree of tolerance in clinical isolates can vary over several orders of magnitude: which had not been previously documented or appreciated. Furthermore, the association of high tolerance among IR isolates is a new finding, and given the potential for tolerance to increase risk of de novo drug resistance, our study suggests that IR isolates with high rifampicin tolerance may present a risk for development of MDR-TB.

In addition, we have also analysed the longitudinal isolates and the genetic variants emerging in them associated with increase in rifampicin tolerance. This analysis reveals possible multiple pathways to increase in rifampicin tolerance among clinical M. tuberculosis isolates. Possible clinical implication includes associating high rifampicin tolerance and isoniazid resistance as a risk factor for tuberculosis treatment failure. This study helps to develop further clinical studies to evaluate the role of rifampicin tolerance in IR isolates and treatment outcome. We have focused on these aspects in the discussion of the revised manuscript.

Reviewer #2 (Public Review):

Summary:

This study by Vijay and colleagues addresses a clinically important, and often overlooked aspect of Tb treatment. Detecting for variations in the level of antibiotic tolerance amongst otherwise antibiotic-susceptible isolates is difficult to routinely screen for, and consequently not performed. The authors, present a convincing argument that indeed, there is significant variation in the susceptibility of isoniazid-resistant strains to killing by rifampicin, in some cases at the same tolerance levels as bona fide resistant strains. On the whole, the study is easy to follow and the results are justified. This work should be of interest to the wider TB community at both a clinical and basic level.

Weaknesses:

The manuscript is long, repetitive in places, and the figures could use some amending to improve clarity (this could be a me-specific issue as they look ok on my screen, yet the colour is poor when printed).

We thank the reviewer for the comments, we have modified the revised manuscript as per the reviewer suggestions.

It would have been great to have seen some correlation between increased rifampicin tolerance and treatment outcome, although I'm not sure if this data is available to the researchers. I agree with the researchers the use of a single media condition is a limitation. However, this is true of a lot of studies. Rifampicin tolerance and treatment outcome analysis.

We agree with the reviewer that correlation between rifampicin tolerance and treatment outcome is important. This needs to be performed in future studies with better design to correlate rifampicin tolerance with treatment progression or outcome data.  

Reviewer #3 (Public Review):

Summary:

The authors have initiated studies to understand the molecular mechanisms underlying the devolvement of multi-drug resistance in clinical Mtb strains. They demonstrate the association of isoniazid-resistant isolates by rifampicin treatment supporting the idea that selection of MDR is a microenvironment phenomenon and involves a group of isolates.

Strengths:

The methods used in this study are robust and the results support the authors' claims to a major extent.

Weaknesses:

The manuscript needs a thorough vetting of the language. At present, the language makes it very difficult to comprehend the methodology and results.

We thank the reviewer for the comments, we have revised the manuscript as per the reviewer’s suggestions.

Reviewer #1 (Recommendations For The Authors):

Major comments:

(1) Methods: The authors attempt to differentiate between "fast"- and "slow"-growing bacteria in order to determine if the growth rate is associated with rifampicin tolerance. This is accomplished by assessing growth on solid agar at 15 and 60 days post-incubation, respectively. However, mycobacterial growth rate is not a binary phenomenon but rather a continuous variable. Moreover, it is not clear why 15 and 60 days were selected. Also, instead of a "slow growth" phenotype, the 60-day time point might simply reflect a longer lag phase. Were the plates examined at any interval time points? It would be interesting to know whether colony growth was delayed overall in the populations observed only at 60 days, or simply if the appearance of microcolonies visible to the naked eye was delayed (with normal growth afterwards).

We thank the reviewer for the comments, we want to clarify that we have not used agar plates but most-probable number method to determine the survival fraction post antibiotic treatment. We have clarified this in the revised manuscript and revised figure 1. The MPN method is a binary measure (growth/ no growth) and therefore cannot differentiate between long lag time and other mechanisms. In our original analysis, we included an intermediate time point of 30 days, but these data (included as supp fig. 1) cannot address the issue of lag phase directly. Since the 30-day time point did not add to the overall analysis and interpretation, we had not included them in the original submission.

(2) Methods/Results/Discussion: Some important clinical information is missing-how were the patients treated who had IR isolates? Did they receive the standard regimen for DS TB or was another drug substituted for isoniazid? Exposure to different drugs could affect the rifampicin-tolerant populations during the intensive phase (Figure 5).

Thank you for this comment, we have included the information regarding the treatment regimen in the revised manuscript.

Were there differences in microbiological (sputum culture conversion rate at 8 weeks or time to culture negativity) or clinical outcomes based on isoniazid susceptibility? Perhaps more importantly, were there differences in microbiological/clinical outcomes based on the proportion of bacterial subpopulations with rifampicin tolerance for a particular isolate? There should be more discussion on the potential clinical implications of the study's findings.

We agree with the reviewer that correlation between rifampicin tolerance and treatment progression or outcome is important. This needs to be performed in future studies with better design to correlate rifampicin tolerance with treatment progression or outcome data.  

(3) Results (Figure 3A): Although an interesting finding, the increased rifampicin tolerance observed only in the "rapidly" growing populations of isoniazid-resistant isolates (IR) vs. isoniazid-susceptible (IS) isolates is not explained. In contrast, equally, increased rifampicin tolerance is seen in the "slowly" growing populations of both IR and IS isolates. It would be interesting to know if these slowly growing populations show specific tolerance to rifampicin or if, as expected, slow growth confers tolerance to a range of different bactericidal antibiotics.

We thank the reviewer for the suggestions. we agree these will be interesting to investigate in a future study but are outside the scope of the current study.

(4) Results (Figure 3B): The basis for the classification into tertiles is not clear and appears somewhat arbitrary-does this represent the survival of a particular isolate following rifampicin exposure relative to the other isolates based on isoniazid susceptibility (IS or IR) or the % growth relative to other populations for the same isolate? Figure 3B is missing a y-axis label. Is it a log10 MPN ratio?

We thank the reviewer for pointing this, we want to clarify that for the classification into tertiles, first we pooled both group of isolates isoniazid susceptible (IS) and isoniazid resistant (IR) into a single population. Subsequently, we categorized this unified population into three distinct groups: low, medium, and high, based on their survival fraction following rifampicin treatment. Consequently, the 'low,' 'medium,' and 'high' tertiles represent the survival of each isolate following rifampicin exposure relative to the total number of isolates  combing both IS and IR isolates.

For clarity, we provide a breakdown of the criteria for each tertile:

+Low tertile: Consists of isolates with the lowest survival fraction (bottom 25%).

+Medium tertile: Encompasses isolates with survival fractions that fall between the bottom 25% and the top 25%.

+High tertile: Comprises isolates with the highest survival fractions (top 25%). This we have modified in the revised manuscript to clarify.

We have also modified the Figure 3B to correct the y-axis label.

(5) Results (lines 185-186): For correlating relative growth in the absence of antibiotics, 19 clinical isolates "outliers" were removed without explanation.

We have added explanation for the “outliers” which were removed earlier due to deviation from normal distribution, we have also provided the supplementary figure 3 which includes these outliers.

(6) Results (lines 203-211): The authors attempted to investigate a potential association between the mechanism of M. tuberculosis isoniazid resistance and the degree of rifampicin tolerance. However, the vast majority of IR clinical isolates (n=71) had a katG_S315X mutation and only 8 isolates had alternative mutations (inhA_I21T and fabG1_C-15X). Given the wide range of rifampicin tolerance observed within these isoniazid-resistant isolates, they concluded that other genetic or epigenetic determinants must be playing a role. WGS of longitudinally collected isolates from the same patients during TB treatment yielded non-synonymous SNPs in a list of genes previously reported to be associated with persistence, tolerance, and mycobacterial survival. However, precise mechanisms (including, e.g., expression of efflux pumps) are not investigated.

We thank the reviewer for summarising the findings. Yes, we agree that investigating the precise mechanism of rifampicin tolerance is beyond the scope of the current work.

Minor comments:

(1) Abstract (line 41): The nonstandard abbreviations "IR" and "IS" have not been introduced prior to this usage.

We have modified this in the abstract.

(2) Introduction (line 60): Insert "phenomena" or "mechanisms" after "two".

We have modified this in the introduction.

(3) Introduction (lines 66-69): This sentence is confusing, especially the second part ("supporting this studies...").

We have modified the lines to clarify.

(4) Introduction (line 84): In the current text, it appears as if "IR" is the abbreviation for "isoniazid". Therefore, I recommend changing "resistance to isoniazid" to "isoniazid resistance".

We have modified this in the revised manuscript.

(5) Results (line 141): Insert "the" before "rest".

We have modified this in the revised manuscript.

(6) Results (line 187): Replace "did not had" with "did not have".

We have modified this in the revised manuscript.

Reviewer #2 (Recommendations For The Authors):

Abstract:

The abstract is long and repetitive. It needs reworking and shortening to improve clarity and highlight the main takeaway message.

We thanks the reviewer for the suggestions and have modified this in the revised manuscript.

The introduction is interesting and contains relevant information. However, it is long and takes a while to get to the point of the study. It needs re-writing to emphasise key prior results and the purpose of this study.

We thanks the reviewer for the suggestions and we have modified this in the revised manuscript.

Results:

As the study relies predominately on the use of MPN, I think a simple schematic of how the experiment is performed would be informative. Could this be added to Figure 1?

We have revised the figure 1 in the manuscript to include the schematic representation.

Some of the differences in MKD90, whilst they may be significant, are small so it would at least provide context as to the relevance of these differences. This may also alleviate my confusion as to how the authors can measure the time required to achieve MDK90 as 1.23-1.31 days when the first time point that is taken is day 2 (the data in Figure 2). They have FigS6 but this is small and hard to follow.

We thank the reviewer for this suggestion, we have modified this in the revised manuscript and figureS6.

Figure 2:

Would be helpful to have -1 on the Y axis.

The grey dots don't print very well (Might be my printer)

We have modified this in the revised manuscript, figure 2.

Line 142: The authors note a difference in RIF tolerance at day 15 that disappeared by day 60. I assume they are referring to the day 5 timepoint although this isn't clear as written.

Yes, it is referring to the day 5 time point and we have clarified this in the revised manuscript.

The section starting at line 148 (fig 3) is interesting, but it is difficult to read and follow what the difference is between this data and the prior data in Figure 2. It also wasn't until about line 165 that the purpose became clear. Overall the conclusions are sound and interesting.

We have modified this in the revised manuscript.

Line 154: What are the early and late time recovery time points?

Is Figure 3A the same data as Figure 2?

We have clarified this in the revised manuscript, the figure 3A is the same data as Figure 2.

I found Figure 6 hard to follow. I'm not sure how better to present this data, but it should be improved. Some further clarification in the text would be helpful.

We thank the reviewer for the suggestions. We have added more explanation in the text to clarify figure 6.

Conclusions:

The conclusions are sound, based on the data presented. The clinical relevance is highlighted, yet appropriately phrased to not be too far-reaching.

Again, I think the conclusions could be condensed considerably. It is repetitive in places, which distills the main outcomes of this otherwise interesting and important study. The authors appropriately highlight some of the limitations of their study.

We thank the reviewer for these comments and have modified this in the revised manuscript.

Reviewer #3 (Recommendations For The Authors):

The manuscript "Rifampicin tolerance and growth fitness among isoniazid-resistant clinical Mycobacterium tuberculosis isolates: an in-vitro longitudinal study" by Srinivasan et.al., details the identification/ development of isoniazid-resistant strains in clinical isolates following testament with rifampicin. This is an important aspect of understanding MDR development in TB strains. the results are promising and gel well with the hypothesis. However, the manuscript requires a thorough language modification. While the overall idea is clear the methodology does not come out clearly.

Specific comments:

(1) It is not clear whether rifampicin treatments were given for 2 and 5 days before kill curves or for 15 and 60 days? The methodology needs to be phased clearly. Why was this time interval of 15 days and 60 days taken? is there a rationale for this?

We thank the reviewer for the suggestions, we have modified the method and figure 1 to clarify this in the revised manuscript.

(2) A concentration of 2ug/ml was used for in vitro culture in this study. While the authors themselves indicate that this is well above the MIC, this might represent a non- natural dose and hence may force the evolution of strains. What will be the scenario in the natural course of antibiotic treatment (dose at MIC or less than MIC)?

We have observed that till 5 days there is no significant resistant emergence but after 5 days only resistance emerges, therefore we avoided determining the survival fraction after resistance emergence, the kill curve represents mostly tolerant sub population. ADD: Pharmacokinetic studies of rifampicin dosing suggest that peak concentrations of >2-32 µg/mL are typical for standard doses of the drug, therefore we believe the chosen concentration of 2 µg/mL to be physiologically relevant.

(3) As described in line 155, the survival spanned a broad distribution, across a million times in difference. This is rather surprising that 5 days of rifampicin treatment would lead to such a spread in resistance patterns. Did the authors study the different populations to understand this phenomenon? This is important given the scale of resistance developed in this short time.

We want to clarify that the broad range of survival fraction reflect the difference in tolerant sub-populations but not resistant sub-population to rifampicin as they are determined post rifampicin treatment in rifampicin free media, this has been clarified in the revised figure 1.

Overall, the manuscript is a detailed study with new insights into the development of multi-drug resistance by Mtb. A thorough vetting for language is essential for a greater impact of the study.

We thank the reviewer and have attempted to improve the clarity of the language to increase the potential impact of our findings.

eLife assessment

This valuable study demonstrates that there is significant variation in the susceptibility of isoniazid-resistant Mycobacterium tuberculosis clinical isolates to killing by rifampicin, in some cases at the same tolerance levels as bona fide resistant strains. The evidence provided is solid, with no clear genetic marker for increased tolerance, suggesting that there may be multiple routes to achieving this phenotype. The work will be of interest to infectious disease researchers.

Reviewer #3 (Public Review):

Summary:

The authors have initiated studies to understand the molecular mechanisms underlying the devolvement of multi drug resistance in clinical Mtb strains. They demonstrate the association of isoniazid resistant isolates by rifampicin treatment supporting the idea that selection of MDR is a microenvironment phenomenon and involves a group of isolates.

Strengths:

The methods used in this study are robust and the results support the authors claims to a major extent.<br /> The language has now been corrected and is better to comprehend.

Reviewer #3 (Public Review):

Summary:

In this paper Hajra et al have attempted to identify the role of Sirt1 and Sirt3 in regulating metabolic reprogramming and macrophage host defense. They have performed gene knock down experiments in RAW macrophage cell line to show that depletion of Sirt1 or Sirt3 enhances the ability of macrophages to eliminate Salmonella Typhimurium. However, in mice inhibition of Sirt1 resulted in dissemination of the bacteria but the bacterial burden was still reduced in macrophages. They suggest that the effect they have observed is due to increased inflammation and ROS production by macrophages. They also try to establish a weak link with metabolism. They present data to show that the switch in metabolism from glycolysis to fatty acid oxidation is regulated by acetylation of Hif1a, and PDHA1.

Strengths:

The strength of the manuscript is that the role of Sirtuins in host-pathogen interactions has not been previously explored in-depth making the study interesting. It is also interesting to see that depletion of either Sirt1 or Sirt3 results in a similar outcome.

Weaknesses:

The major weakness of the paper is the low quality of data, making it harder to substantiate the claims. Also, there are too many pathways and mechanisms being investigated. It would have been better if the authors had focussed on either Sirt1 or Sirt3 and elucidated how it reprograms metabolism to eventually modulate host response against Salmonella Typhimurium. Experimental evidence is also lacking to prove the proposed mechanisms. For instance they show correlative data that knock down of Sirt1 mediated shift in metabolism is due to HIF1a acetylation but this needs to be proven with further experiments.

eLife assessment

This study presents valuable findings on the role of the sirtuins SIRT1 and SIRT3 during Salmonella Typhimurium infection. Although the work increases our understanding of the mechanisms used by this pathogen to interact with its host and may have implications for other intracellular pathogens, the reviewers found that the evidence to support the claims is incomplete. In particular, the discrepancy between results obtained using cultured cell lines and the animal model of infection, as well as potential indirect effects through the microbiome stand out.

Reviewer #2 (Public Review):

Dipasree Hajra et al demonstrated that Salmonella was able to modulate the expression of Sirtuins (Sirt1 and Sirt3) and regulate the metabolic switch in both host and Salmonella, promoting its pathogenesis. The authors found Salmonella infection induced high levels of Sirt1 and Sirt3 in macrophages, which were skewed toward the M2 phenotype allowing Salmonella to hyper-proliferate. Mechanistically, Sirt1 and Sirt3 regulated the acetylation of HIF-1alpha and PDHA1, therefore mediating Salmonella-induced host metabolic shift in the infected macrophages. Interestingly, Sirt1 and Sirt3-driven host metabolic switch also had an effect on the metabolic profile of Salmonella. Counterintuitively, inhibition of Sirt1/3 led to increased pathogen burdens in an in vivo mouse model. Overall, this is a well-designed study.

Comments on revised version:

The authors have performed additional experiments to address the discrepancy between in vitro and in vivo data. While this offers some potential insights into the in vivo role of Sirt1/3 in different cell types and how this affects bacterial growth/dissemination, I still believe that Sirt1/3 inhibitors could have some effect on the gut microbiota contributing to increased pathogen counts. This possibility can be discussed briefly to give a better scenario of how Sirt1/3 inhibitors work in vivo. Additionally, the manuscript would improve significantly if some of the flow cytometry analysis and WB data could be better analyzed.

Author response:

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

Reviewer #1 (Public Review):

Summary:

The current manuscript by Hajra et al deals with the role of the prominent Sirtuins SIRT1 and -3 during infection of macrophages with Salmonella Typhimurium (ST). Apparently, ST infection induces upregulation of host cell SRTs to aid its own metabolism during the intracellular lifestyle and to help reprogramming macrophage polarization. The manuscript has two parts, namely one part that deals with Salmonella infection in cells, where RAW 264.7 murine macrophage-like cells, sharing some features with primary macrophages, were employed. Infected RAW cells displayed a tendency to polarize towards wound-healing M2 and not inflammatory M1 macrophages, which was dependent on SRT. Consequently, the inflammatory response in RAW was more robust in the absence of SRT. Moreover, loss of SRTs leads to impaired bacterial proliferation in these cells, which was attributed to defects in metabolic adaption of the bacteria in the absence of SRT-activity and to the increased M1 inflammatory response.

Unfortunately, the line of argumentation remains incomplete because corresponding assays in mice showed the opposite result as compared to the experiments using RAW 264.7 cells. i.e. loss of SRTs leads to increased bacterial load in animals (versus impaired proliferation in RAW 264.7 cells). The authors cannot explain this discrepancy.

Strengths:

Extensive analysis of Salmonella infection in RAW macrophage-like cells and mice in the context of SRT1/3 function.

Weaknesses:

Lack of connection between the cell-based and organismic data, which are not supportive of each other.

We are highly grateful for your valuable and insightful comments. Thank you for appreciating the merit of our manuscript. We agree with the opposing phenotypes among the RAW264.7 cell line (Fig. 2A), primary peritoneal macrophages (ex vivo) (Fig.2B), and in vivo mouse model (Fig.8) findings. Both RAW264.7 macrophage and peritoneal macrophage infection show attenuated intracellular bacterial proliferation owing to the heightened proinflammatory burst. This is in sharp contrast to our in vivo mouse model of infection which shows increased organ burden and bacterial dissemination. The higher bacterial load in the organs including the spleen (Fig.8B) is attributed to increased pro-inflammatory cytokine burst and ROS production (Fig.8F-H, Fig.S9) triggering bacterial dissemination. The pro-inflammatory arsenals like IL-6, IL-1β and ROS that limit bacterial proliferation within the macrophages (F4/80+ macrophages within the spleen or in RAW264.7 macrophages or primary peritoneal macrophages) are facilitating bacterial dissemination in blood and to the other organs (Fig. 8I-L, Fig.S3F-G). This is in line with the following previous findings-

Klebsiella pneumoniae infection triggers an inflammatory response via secretion of IL-6 upon HIF-1α activation that induces bacterial dissemination (Holden VI, Breen P, Houle S, Dozois CM, Bachman MA. Klebsiella pneumoniae Siderophores Induce Inflammation, Bacterial Dissemination, and HIF-1α Stabilization during Pneumonia. mBio. 2016 Sep 13;7(5):e01397-16. doi: 10.1128/mBio.01397-16. PMID: 27624128; PMCID: PMC5021805.).

Correlation analysis of immune responses to Salmonella infection revealed that increased innate immune “cassette” opposes the adaptive immune arm leading to increased bacterial load in mice (Hotson AN, Gopinath S, Nicolau M, Khasanova A, Finck R, Monack D, et al. Coordinate actions of innate immune responses oppose those of the adaptive immune system during Salmonella infection of mice. Science signaling. 2016;9(410):ra4). 

In our revised manuscript, we have assessed additional splenic populations including CD45+, Ly6C+, and CD11c+ populations. Our results show that the CD45+ splenic population depicts increased bacterial loads like that of the total splenic population within the SIRT1/3 inhibited cohorts. However, CD45+ monocytes and Ly6C positive splenic population exhibit compromised burden within the SIRT1/3 inhibited cohorts. Moreover, within the CD11c+ population, CD45+ granulocytes or lymphocytes show comparable organ loads to that of the vehicle control or SIRT1 activator-treated mice group (Fig. M-S, Fig.S8). Overall, our data suggest heterogeneous bacterial burden in diverse splenic populations.

Reviewer #2 (Public Review):

Dipasree Hajra et al demonstrated that Salmonella was able to modulate the expression of Sirtuins (Sirt1 and Sirt3) and regulate the metabolic switch in both host and Salmonella, promoting its pathogenesis. The authors found Salmonella infection induced high levels of Sirt1 and Sirt3 in macrophages, which were skewed toward the M2 phenotype allowing Salmonella to hyper-proliferate. Mechanistically, Sirt1 and Sirt3 regulated the acetylation of HIF-1alpha and PDHA1, therefore mediating Salmonella-induced host metabolic shift in the infected macrophages. Interestingly, Sirt1 and Sirt3-driven host metabolic switch also had an effect on the metabolic profile of Salmonella. Counterintuitively, inhibition of Sirt1/3 led to increased pathogen burdens in an in vivo mouse model. Overall, this is a well-designed study. There are a few comments below that would further strengthen the current study.

Major comments:

In the in vivo study (lines 436-446) - the authors noticed increased pathogen burden in the EX-527 or the 3TYP-treated mice cohorts but decreased pathogen burden within the F4/80+ macrophage population. What are the other cell types that have increased pathogen burden in splenocytes from EX-527 or the 3TYP treated? Can this be further explored and explained?

While the authors indicated that IL-6 cytokine storm and elevated ROS production could result in bacterial dissemination in vivo, one could also argue that Sirt1/3 inhibitors might have an impact on gut function and/or gut microbiota (PMID: 22115311). Did Sirt1/3 inhibitors also lead to increased pathogen burdens in the gut? If so, the potential effect of these in vivo treatments on gut microbiota/colonization resistance should be discussed.

Minor comment:

Sirt1 has been shown to be degraded during Salmonella infection (PMID: 28192515), which is different from the current study. An explanation should be provided for this.

We thank you for your encouraging and gracious comments. We deeply appreciate your time and efforts in providing constructive feedback for the betterment of our work. As per your precious suggestions, we have assessed additional splenic populations including CD45+, Ly6C+, and CD11c+ populations apart from F4/80+ macrophage populations. Our analysis suggests that the CD45+ splenic population show increased bacterial loads similar to the total splenic population within the SIRT1/3 inhibited cohorts. However, CD45+ monocytes and Ly6C positive splenic population exhibit compromised burden within the SIRT1/3 inhibited cohorts. Moreover, CD11c+ population, CD45+ granulocytes or lymphocytes show comparable organ loads to that of the vehicle control or SIRT1 activator treated mice group (Fig. 8M-S). Overall, our data suggest heterogeneous bacterial burden in diverse splenic populations.

We immensely appreciate the reviewer for this insightful question about the effect of SIRT1/3 on the gut per se. To answer your question, we observed increased pathogen loads within the mesenteric lymph nodes of the gut in the SIRT1/3 inhibitor-treated mice groups (Fig.8B). In our revised manuscript, we evaluated gut inflammation via IL1-β estimation in the mice's ileal tissues and have observed heightened IL-1β production in the inhibitor-treated mice cohorts in comparison to the vehicle control (Fig. S3G). We have also examined gut epithelial pathology via Haematoxylin-Eosin (H&E) staining of the ileal sections to address the effect of in vivo treatment on gut microbiota and colonization resistance which is appended here. However, the gut microbiota crosstalk and their effect on colonization resistance is a part of another current study and it is being examined in detail there. Therefore, this appended H&E has not been incorporated in the revised manuscript.

Author response image 1.

In line with the reference PMID: 28192515, where Sirt1 has been shown to be degraded during Salmonella infection at later time points of infection, our study also has shown that both SIRT1 mRNA (Fig. 1A) and protein levels (Fig. S1A) show an elevated expression at 2h and 6h post-infection and show a downregulation at 16h in comparison to the 6h time point.  However, SIRT3 expression levels remain elevated even at later time points of infection. Therefore, we speculate that there is a shared role between SIRT1 and SIRT3 that facilitates the phenotypes reported in our study.

Reviewer #3 (Public Review):

Summary:

In this paper, Hajra et al have attempted to identify the role of Sirt1 and Sirt3 in regulating metabolic reprogramming and macrophage host defense. They have performed gene knockdown experiments in RAW macrophage cell lines to show that depletion of Sirt1 or Sirt3 enhances the ability of macrophages to eliminate Salmonella Typhimurium. However, in mice, inhibition of Sirt1 resulted in dissemination of the bacteria but the bacterial burden was still reduced in macrophages. They suggest that the effect they have observed is due to increased inflammation and ROS production by macrophages. They also try to establish a weak link with metabolism. They present data to show that the switch in metabolism from glycolysis to fatty acid oxidation is regulated by acetylation of Hif1a, and PDHA1.

Strengths:

The strength of the manuscript is that the role of Sirtuins in host-pathogen interactions has not been previously explored in-depth making the study interesting. It is also interesting to see that depletion of either Sirt1 or Sirt3 results in a similar outcome.

Weaknesses:

The major weakness of the paper is the low quality of data, making it harder to substantiate the claims. Also, there are too many pathways and mechanisms being investigated. It would have been better if the authors had focussed on either Sirt1 or Sirt3 and elucidated how it reprograms metabolism to eventually modulate host response against Salmonella Typhimurium. Experimental evidence is also lacking to prove the proposed mechanisms. For instance, they show correlative data that the knockdown of Sirt1-mediated shift in metabolism is due to HIF1a acetylation but this needs to be proven with further experiments.

We appreciate the reviewer’s critical analysis of our work. In the revised manuscript, we aimed to eliminate the low-quality data sets and have tried to substantiate them with better and conclusive ones, as directed in the recommendations for the author section. We agree with the reviewer that the inclusion of both Sirtuins 1 and 3 has resulted in too many pathways and mechanisms and focusing on one SIRT and its mechanism of metabolic reprogramming and immune modulation would have been a less complicated alternative approach. However, as rightly pointed out, our work demonstrated the shared and few overlapping roles of the two sirtuins, SIRT1 and SIRT3, together mediating the immune-metabolic switch upon Salmonella infection. As per the reviewer’s suggestion, we have performed additional experiments with HIF-1α inhibitor treatment in our revised manuscript to substantiate our correlative findings on SIRT1-mediated regulation of host glycolysis (Fig.7G).

Reviewer #1 (Recommendations For The Authors):

The authors state "SIRT1 and SIRT3 inhibition resulted in increased pathogen loads in organs and triggered enhanced bacterial dissemination, together leading to increased susceptibility of the mice to S. Typhimurium infection owing to increased ROS and IL-6 production." How can this be reconciled? To the reviewer, this is not a convincing explanation. The reviewer is not a mouse pathologist, so maybe did not understand the argument in full.

However, in order to clarify whether these phenomena can be brought into context and explained by for instance cell-autonomous (in (RAW) macrophages) versus non-autonomous (in mice) mechanisms, it would be required to bring in context the organismic phenotype with a cellular phenotype, using more physiologic primary macrophages.

(1) The authors show in Figure 8 that in general SRT inhibition leads to increased infection whereas SRT activation results in decreased infection. This is even true for e the spleen (e.g. Figure 8B), which should be full of macrophages upon infection.

(2) Only Figure 8L implies that endogenous primary, splenic macrophages show a higher infection rate upon pharmacologic SRT activation, which would potentially mirror the RAW results. This is however not supportive of their own explanation: Who would now produce more ROS and IL6 if these macrophages are more supportive of intracellular ST? Is there a difference in the roles or SRTs between different types of macrophages and/or neutrophils? And between macrophages and somatic cells concerning ST infection? The reviewer tends to believe that RAW cells display a defective killing response (such as ROS production) as they are highly transformed cells. Therefore, the authors should use cultured peritoneal macrophages or BMDMs in addition to RAW264.7 cells.

The literature cited by the authors also implies that the inflammatory response in mice is higher in the absence of SRTs. This is in line with a role for SRTs in (negatively) regulating M1 inflammatory polarization but probably not with increased bacterial burden in mice. If it was, then increased dissemination could be explained by increased tissue damage. However, the flow cytometry experiments from infected organs then do not confirm that, as the infection of individual cells is higher upon SRT inhibition. Thus there seems a broad gap between the role of SRTs in ST infection in RAW264.7 cells versus non-transformed cells.

I would not discard the RAW results, as I am convinced that they contain valuable data. However, it needs to be clarified what aspect of the host response RAW 264.7 cells represent. Primary macrophages might likely be more aggressive towards the bacteria. Finally, the question arises: what is the role of the metabolic switch in the in vivo setting?

The reviewer recommends repeating some key experiments by in-vitro-infecting BMDMs or isolated peritoneal macrophages (after some days of culturing) to bridge between the present RAW-derived data and the mouse data. How is the bacterial load with and without SRT inhibitor/activator in primary macrophages, when infected outside of the body? Can ex-vivo infection also affect polarization of e.g. peritoneal macrophages or the metabolic switch? If it is possible to find a conclusive explanation for their data, then this story might really add to our understanding of another aspect of how ST manipulates the host to survive.

In case the reviewer understands the mouse experiments correctly, all assays on peritoneal cells were performed after in-vivo-infection and/or treatment.

Together, RAW 264.7 murine macrophage-like cells might not be the right model to understand the phenotypes in full. As far as the reviewer knows, these cells are not capable of killing bacteria as effectively as activated primary macrophages or neutrophils.

A few of the key findings of RAW264.7 macrophages have been replicated in primary peritoneal macrophages (Fig. 2B, S3E-F, S6B, S7B-D). We wanted to clarify that the peritoneal macrophage experiments were performed ex vivo, wherein peritoneal macrophages were isolated from mice were then subjected to SIRT1/3 inhibitor treatments and Salmonella infection and not after in vivo treatment or infection. In ex vivo setting, we have examined the effect of SIRTs on the metabolic switch during Salmonella infection (Fig. S7B-D) which resembled our RAW264.7 macrophage data. Additionally, in in vivo setting, we have analyzed the transcript level expression of host metabolic genes and corresponding bacterial metabolic genes in infected mice liver and spleen tissue under SIRT1/3 inhibitor treatment (Fig.S7E-F, Fig.6C-D). Our primary peritoneal macrophage data exactly mirrors the RAW264.7 macrophage findings showing attenuated intracellular bacterial proliferation owing to the heightened proinflammatory burst upon SIRT1/3 knockdown or inhibition (Fig.2A-B). This is opposite to our in vivo mouse model of infection which shows increased organ burden and bacterial dissemination (Fig.8A-H). The pro-inflammatory arsenals that limit bacterial proliferation within the macrophages (F4/80+ macrophages within the spleen or in RAW264.7 macrophages or primary peritoneal macrophages) are facilitating bacterial dissemination in blood and to the other organs owing to tissue damage (Fig.8E-L). This is in line with the following previous findings-

Klebsiella pneumoniae infection triggers an inflammatory response via secretion of IL-6 upon HIF-1α activation that induces bacterial dissemination (Holden VI, Breen P, Houle S, Dozois CM, Bachman MA. Klebsiella pneumoniae Siderophores Induce Inflammation, Bacterial Dissemination, and HIF-1α Stabilization during Pneumonia. mBio. 2016 Sep 13;7(5):e01397-16. doi: 10.1128/mBio.01397-16. PMID: 27624128; PMCID: PMC5021805.).

Correlation analysis of immune responses to Salmonella infection revealed that increased innate immune “cassette” opposes the adaptive immune arm leading to increased bacterial load in mice (Hotson AN, Gopinath S, Nicolau M, Khasanova A, Finck R, Monack D, et al. Coordinate actions of innate immune responses oppose those of the adaptive immune system during Salmonella infection of mice. Science Signaling. 2016;9(410):ra4). 

As per the reviewer’s suggestions, we have analyzed other populations apart from F4/80+ macrophages and have observed that the CD45+ splenic population depicts increased bacterial loads like that of the total splenic population within the SIRT1/3 inhibited cohorts. However, CD45+ monocytes and Ly6C positive splenic population exhibit compromised burden within the SIRT1/3 inhibited cohorts. Moreover, the CD1c+ population, CD45+ granulocytes, or lymphocytes show comparable organ loads to that of the vehicle control or SIRT1 activator-treated mice group (Fig.8M-S, Fig.S8). Overall, our data suggest heterogeneous bacterial burden in diverse splenic populations.

Reviewer #3 (Recommendations For The Authors):

Abstract

The authors state that perturbing Sirt1 and Sirt3 results in a shift in Salmonella's metabolism. On the contrary, the data reflects the metabolism in the host cell and not the bacteria. This statement is wrong. They only show increased expression of some of the glycolytic genes in Salmonella, which is not sufficient to make the claim that the switch to fatty acid oxidation in macrophages is due to utilisation of glucose by the bacteria.

We value the reviewer’s response and have accordingly reframed our sentence in the abstract (Line 24-25).

Fig 1: Expression of Sirt1 - The data needs to be supported with a western blot for Sirt1 and Sirt3 but the Western blots shown in the supplementary figure are of very poor quality and do not support the authors' claim.

We have repeated the western blot and have supplemented the previous blot with an alternate blot in Fig. S1A as per your precious input.

Why haven't the authors shown any representative blots for Sirt1 and Sirt3 upon infection with Salmonella mutants? They need to italicize the genes when they describe mRNA expression.

Previously we had only performed transcript-level expression of Sirt1 and Sirt3 upon infection with Salmonella mutants and therefore representative blot image was absent. The gene names have been duly italicized while describing mRNA expression (Line 126-154). We regret the inconvenience caused. We have performed the western blotting to assess the protein expression profile upon infection with Salmonella mutants as per the reviewer’s suggestion and the representative blot image has been duly appended in the revised manuscript (Fig. S1B).

What is the rationale for examining Sirt1 and Sirt3 mRNA in M1 and M2 macrophages? Salmonella infection on its own will polarise the macrophages towards M1. How long were these macrophages infected? The time points are missing.

The rationale behind the examination of Sirt1 and Sirt3 mRNA in M1 and M2 polarized was to ascertain whether indeed M1 polarized macrophages exhibit decreased expression of Sirt1 or Sirt3 and polarization of macrophages toward M2 state show upregulation of Sirt1 and Sirt3 upon Salmonella infection. After confirming these above-mentioned findings through this preliminary experiment, we then hypothesized whether Salmonella infection on its own will polarise the macrophages toward an immunosuppressive M2 state at a later time course of infection as infection drives the induction of SIRT expression and whether this is mediated by Sirt1 and Sirt3 (Fig. 3). We are extremely apologetic for not mentioning the 16h time-point in the figure and the missing time point has been duly documented in the revised manuscript (Line 155).

Fig S2 knockdown of Sirt1 and Sirt3 are not convincing.

We are extremely sorry for the inconclusive knockdown blot. An alternative blot has been substantiated in the revised manuscript (Fig. S2,C-D).

Fig 2A and 2B the time point post infection has not been mentioned. Although it is stated that 2h and 16h post-infection samples were analysed. Only one time point has been shown.

We are sorry for the confusion. We wanted to clarify that Fig.2A and Fig. 2B show the fold proliferation where fold proliferation was calculated as CFU at 16hr divided by CFU at 2hr as mentioned in the materials and methods section under the heading of Intracellular proliferation or gentamicin protection assay.

Fold Proliferation= [CFU at 16h]/[CFU at 2h]

The cytokines data are intriguing in that the increase in IL-6 relative to control is seen only at 2h and 20h but not at 6h. Il-6 at 20h in untransfected cells is comparable to uninfected cells. Did the authors investigate cell death? Salmonella induces various forms of cell death which could account for the decreased cytokine production at later time points.

We have investigated the cell death upon Salmonella infection via MTT assay. At later time points of infection, we indeed observed around 16 percent decrease in cell survival compared to the initial time point of 2h. The results have been appended here and it supports our eminent reviewer’s reasoning for the decreased cytokine production at later time points.

Author response image 2.

Additional cytokines such as IL-1b would be helpful. Also, not sure how uninfected macrophages produce nearly 200pg of IL-10.

As per the author’s critical suggestion, we have assessed the IL-1b cytokine production at 16h post-infection in RAW264.7 macrophages and peritoneal macrophages and mice serum samples at 5th day post-infection (Fig.S3C, S3E-F). Our results indicate increased production of IL-b in the infected SIRT1/3 knockdown RAW264.7 macrophages, SIRT1/3 inhibitor-treated peritoneal macrophages and in mice serum samples under SIRT1/3 inhibitor treatment in comparison to the vehicle control. Additionally, we have quantified IL-1b in mice ileal tissues under SIRT1/3 inhibitor treatment (Fig.S3G) and have obtained heightened intestinal IL-1b production in the inhibitor-treated cohorts. We thank the reviewer for raising the concern for 200pg of IL-10 in the uninfected macrophages. We have repeated the experiment and have provided an alternative representative graph for the experiment wherein the IL-10 levels in the uninfected cohorts range between 20-40pg/ml (Fig. S3B).

It is surprising that the authors have found increased Sirt1 binding to NFkB, however there is no change in acetylated NFkB upon infection (Fig 4B). Acetylated p65 is equally high in uninfected Scrambled siRNA, UI shSirt1, STM Scr, and STM shSirt1. Furthermore, increased binding of Sirt1 with NFkb would mean decreased acetylation hence decreased inflammation. However, Salmonella induces profound inflammation.

We thank the reviewers for their insightful and critical questioning. We truly acknowledge that due to oversaturation there was no apparent change in the acetylated p65 among the different sample sets. Therefore, in the revised manuscript we have provided an image at lower exposure where the changes in the acetylation of the p65 subunit are apparent. Salmonella induces inflammation upon challenge similar to any other pathogens and induces acute inflammatory responses. This heightened acute inflammation at the initial phases of infection subsides at a later phase of infection. Here, we have performed the Sirt1 interaction with NFκB at 16hr post-infection where increased binding of Sirt1 with NFκB facilitates the resolution of the Salmonella-_induced acute inflammation. This is in line with previous reports that suggest SIRT1 suppresses acute inflammation through the promotion of p65 acetylation and inhibition of NFκB activity. (Yang H, Zhang W, Pan H, et al. SIRT1 activators suppress inflammatory responses through promotion of p65 deacetylation and inhibition of NF-κB activity. _PLoS One. 2012;7(9):e46364. doi:10.1371/journal.pone.0046364, Liu TF, Yoza BK, El Gazzar M, Vachharajani VT, McCall CE. NAD+-dependent SIRT1 deacetylase participates in epigenetic reprogramming during endotoxin tolerance. J Biol Chem. 2011;286(11):9856–64., Liu TF, Vachharajani V, Millet P, Bharadwaj MS, Molina AJ, McCall CE. Sequential actions of SIRT1-RELB-SIRT3 coordinate nuclear-mitochondrial communication during immunometabolic adaptation to acute inflammation and sepsis. J Biol Chem. 2015;290(1):396–408.)

Please explain how the acetylated p65 was analysed.

Total endogenous p65 subunit was immunoprecipitated using Anti-NFκB p65 antibody and the immunoprecipitated fraction was probed with Anti-Acetylated Lysine antibody to assess acetylated p65.

An increase in ROS production is seen in a relatively small percentage of cells- not more than 4% of cells. How does this contribute to such a significant difference in intracellular bacterial burden? Also, it is not clear how the authors calculated the fold change in proliferation. It is better to show the actual bacterial burden logarithmically.

We strongly agree with the reviewer’s concerns, and we have reanalyzed the flow cytometric data set. The revised data have been presented in Fig. S5 which shows a considerable increase in DCFDA positive population. For instance, the infected scrambled control shows around 2.44% of ROS-producing cells, however knockdown of SIRT1 and SIRT3 increases the ROS-producing cells to 27.34% and 28.64% respectively.

Fold proliferation was calculated as CFU at 16hr divided by CFU at 2hr as mentioned in the materials and methods section under the heading of Intracellular proliferation or gentamicin protection assay. Fold proliferation has been calculated as opposed to absolute CFU values to nullify the differential phagocytosis of bacteria to the macrophages among the samples.

Fold Proliferation= [CFU at 16h]/[CFU at 2h]

An increase in metabolic genes is not sufficient to show that the macrophages are metabolically reprogrammed.

We thank the reviewer for the valuable comment. We agree that an increase in metabolic gene profile is not sufficient to claim metabolic reprogramming. Therefore, in addition to the metabolic gene profile, we have estimated lactate production (end-product of glycolysis) as an indicator of glycolysis (Fig. 5 C-E) and have performed the fatty acid β oxidation activity (Fig. 5G-H) to support our claims.

Figure 5F the band intensities do not visually match the bands shown for PFK. For instance, shSIRT1 STM (1.00) and shSIRT3 STM (0.81).

We are extremely sorry for the erroneous band intensity for shSIRT3. Upon reanalysis of the band intensities, we have corrected the band intensity for shSIRT3 to 2.28 (Fig.5F).

It is surprising that HADHA is not expressed in uninfected samples.

We are extremely apologetic for the inappropriate representative blot. We feel that the discrepancy might have arisen due to the usage of old antibodies. We have provided an alternate blot for the HADHA gene where fresh antibody staining solution was used for probing which shows expression even in the uninfected samples (Fig.5F).

Figure 6A - What is the significance of PFA fixed samples (PI) compared to SI samples? This has not been discussed.

PFA-fixed samples are paraformaldehyde-treated bacterial samples that harbor the immune signals or Pattern Associated Molecular Patterns (PAMPs). The rationale for using PI in addition to SI samples was to show whether the phenomena is driven by live metabolically active pathogens or is mediated by PAMPs.

I understand that the hypothesis is that during the later phase of infection, there is an increase in fatty acid oxidation which correlates with a decrease in inflammation. However, at 6h there is no increase in genes regulating fatty acid oxidation. Why did the authors choose 6h when the previous experiments have been done at 16h?

We indeed agree with the reviewer’s understanding of our hypothesis that there is an increase in fatty acid oxidation along the progression of infection which correlates with a decrease in inflammation. The Salmonella intracellular replication has been reported to commence at 6h post-internalization when SPI-2 effector expression is fully established (Helaine S, Thompson JA, Watson KG, Liu M, Boyle C, Holden DW. Dynamics of intracellular bacterial replication at the single cell level. Proc Natl Acad Sci U S A. 2010;107(8):3746-3751. doi:10.1073/pnas.1000041107). Therefore, we have assessed the 6h timepoint post-infection in addition to the initial and later timepoints of 2h and 16h respectively. Additionally, the nanostring gene profiling data of both host and bacterial genes indicate the onset of both metabolic (Fig. 5A, 6A) and immune genes (Fig. 3A) modulation at 6h post-infection. We have validated these results via qPCR studies and have observed an upregulation in the transcript level of fatty acid oxidation genes as depicted in Fig. S7A in RAW264.7 macrophages.

Line 355 it is mentioned that Sirt1 and Sirt3 abrogate metabolic shift by reducing glycolytic flux. This is incorrect as experiments such as carbon chase assays have not been performed to investigate glycolytic flux.

As per the reviewer’s valuable suggestion, we have removed the word ‘flux’ from the above-mentioned statement(Line 351, Line 353).

Lines 392-393: "We immunoprecipitated PDHA1 and checked for its interaction with SIRT3 or SIRT1 under knockdown condition of SIRT3 or upon SIRT3 inhibitor treatment (Fig.7 G-H)"

What is the rationale for checking PDHA1 interaction with Sirt under Sirt knockdown conditions?

We are thankful to the reviewer for the critical comments. The rationale for checking PDHA1 interaction with Sirt was to ascertain that indeed Sirt interacted with PDHA1 under S. Typhimurium infection and abrogation of either protein expression (knockdown) or their enzymatic activity (inhibitor treatment) diminished the interaction.

Moreover, the blots are very confusing and do not represent the authors' claims.

(1) In the input blot I do not see Sirt3 depletion in shSirt3 knockdown sample.

The knockdown has been quantified in the input blot as per your suggestion. A knockdown of 40% has been obtained in the uninfected dataset whereas a knockdown of 47.1% has been obtained in the infected data set at 16h post-infection (Fig.7H).

(2) Why does Sirt1 interact with PDHA1 similar to Sirt3. Do both the proteins bind to PDHA1 at the same time/ competitively? If so do they both deacetylate?

In literature, Sirt3 has been shown to interact with PDHA1 and deacetylate PDHA1. However, the interaction of Sirt1 with PDHA1 has not been reported previously and therefore we are unable to comment on the exact dynamics of the interaction. Future studies need to be performed to explore these phenomena in depth. However, SIRT1 agonist SRT1720 has been shown to impact PDH phosphorylation and its activity (Han Y, Sun W, Ren D, Zhang J, He Z, Fedorova J, Sun X, Han F, Li J. SIRT1 agonism modulates cardiac NLRP3 inflammasome through pyruvate dehydrogenase during ischemia and reperfusion. Redox Biol. 2020 Jul;34:101538).

(3) Figure 7I in the IP: IgG samples Sirt3 seem to bind to IgG non-specifically, which questions the specificity of Sirt3 binding to PDHA1.

We appreciate the reviewer for pointing out this concern. The immunoprecipitation experiment has been repeated and the same has been appended in the revised manuscript and we observe no non-specific binding of Sirt3 antibody to IgG.

(4) In Figure 7I all the bands Ac PDHA1, PDHA1, and Sirt3 look similar with double bands, which has not been seen in other blots. How is this possible?

This cannot explain the increase in beta-oxidation observed.

We thank the reviewer for raising this concern. We have repeated the experiment and provided the alternative blot as per the reviewer’s suggestion.

The rationale for performing this experiment was to show that SIRT plays an important role in the activation of downstream TCA cycle pathways via PDHA1 deacetylation during Salmonella infection. The deacetylation of PDHA1 has been previously reported to cause transcriptional activation of the downstream TCA cycle and oxidative phosphorylation (Zhang Y, Wen P, Luo J, et al., Cell Death Dis.,2021). Additionally, PDHA1 hyperacetylation has been reported to cause lactate overproduction (An, S., Yao, Y., Hu, H. et al. PDHA1 hyperacetylation-mediated lactate overproduction promotes sepsis-induced acute kidney injury via Fis1 lactylation. Cell Death Dis 14, 457 (2023)). In our study, increased lactate production and PDHA1 hyperacetylation have been observed during SIRT3 inhibition conditions upon Salmonella infection.

eLife assessment

The work by Han and collaborators describes valuable findings on the role of Akkermansia muciniphila during ETEC infection. If confirmed, these findings will add to a growing list of beneficial properties of this organism. However, as it stands, the strength of the evidence used to justify the conclusions in the manuscript is incomplete.

Reviewer #2 (Public Review):

The authors indicated that the adherence of ETEC is to intestinal epithelial cells. However, it is also possible that the majority of ETEC may reside in the intestinal mucus, particularly under in vivo infection condition. The colonization of ETEC in the jejunum and colon of piglets (Fig 2C) and in the intestines of mice (Fig S2A) does not necessarily reflect the adherence of ETEC to epithelial cells. Please verify these observations with other methods, such as immunostaining. Also, while Salmonella enterica serovar Typhimurium or Listeria monocytogenes can invade organoids within 1 hour, it is unknown if ETEC invade into organoids in this study. Clarifying this will help resolve if A. muciniphila block the adherence and/or invasion of ETEC. Please also address if A. muciniphila metabolites could prevent ETEC infection in the organoid models.

Reviewer #3 (Public Review):

Summary:

The manuscript by Ma et al. describes a multi-model (pig, mouse, organoid) investigation into how fecal transplants protect against E. coli infection. The authors identify A. muciniphila and B. fragilis as two important strains and characterize how these organisms impact the epithelium by modulating host signaling pathways, namely the Wnt pathway in lgr5 intestinal stem cells.

Strengths:

The strengths of this manuscript include the use of multiple model systems and follow up mechanistic investigations to understand how A. muciniphila and B. fragilis interacted with the host to impact epithelial physiology.

Weaknesses:

After revision, the bioinformatics section of the methods is still jumbled and may indicate issues in the pipeline. Important parameters are not included to replicate analyses. Merging the forward and reverse reads may represent a problem for denoising. Chimera detection was performed prior to denoising.

Potential denoising issues for NovaSeq data was not addressed in the response. The authors did not clarify if multiple testing correction was applied; however, it may be assumed not as written. The raw sequencing data made available through the SRA accession (if for the correct project) indicates it was a MiSeq platform; however, the sample names do not appear to link up to this experimental design and metadata not sufficient to replicate analyses.

Author response:

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

Reviewer #1 (Public Review):

Summary:

In this paper, the authors investigate the impact of fecal microbiota transfer (FMT) on intestinal recovery from enterotoxigenic E. coli infection following antibiotic treatment. Using a piglet model of intestinal infection, the authors demonstrate that FMT reduces weight loss and diarrhea and enhances the expression of tight junction proteins. Sequencing analysis of the intestinal microbiota following FMT showed significant increases in Akkermansia muciniphila and Bacteroides fragilis. Using additional mouse and organoid models, the authors examine the impact of these microbes on intestinal recovery and modulation of the Wnt signaling pathway. Overall, the data support the notion that FMT following ETEC infection is beneficial, however, additional investigation is required to fully elucidate the mechanisms involved.

Strengths:

Initial experiments used a piglet model of infection to test the value of FMT on recovery from E. coli. The FMT treatment was beneficial and the authors provide solid evidence that the treatment increased the diversity of the microbiota and enhanced the recovery of the intestinal epithelium. Sequencing data highlighted an increase in Akkermansia muciniphila and Bacteroides fragilis after FMT.

The mouse data are consistent with the observations in pigs, and reveal that daily gavage with A. muciniphila or B. fragilis enhances intestinal recovery based on histological analysis, expression of tight junction proteins, and analysis of intestinal barrier function.

The authors demonstrate the benefit of probiotic treatment following infection using a range of model systems.

Weaknesses:

Without sequencing the pre-infection pig microbiota or the FMT input material itself, it's challenging to firmly say that the observed bloom in Akkermansia muciniphila and Bacteroides fragilis stemmed from the FMT.

Response: We have determined the relative abundance of each bacterium in fecal bacterial suspension, referring to Hu et al. (2018). The absolute abundances of Akkermansia muciniphila and Bacteroides fragilis in the FMT were 1.3 × 103 ± 2.6 × 103 and 4.5 × 103 ± 6.1 × 103 respectively.

Reference:

Hu LS, Geng SJ, Li Y, et al. Exogenous Fecal Microbiota Transplantation from Local Adult Pigs to Crossbred Newborn Piglets. Front. Microbiol. 2018, 8.

The lack of details for the murine infection model, such as weight loss and quantification of bacterial loads over time, make it challenging for a reader to fully appreciate how treatment with Akkermansia muciniphila and Bacteroides fragilis is altering the course of infection. Bacterial loads of E. coli were only quantified at one time point, and the mice that received A. muciniphila and B. fragilis had very low levels of E. coli. Therefore, it is not clear if all mice were subjected to the same level of infection in the first place. The reduced translocation of E. coli to the organs and enhanced barrier function may just reflect the low level of infection in these mice. Further, the authors' conclusion that the effect is specific to A. muciniphila or B. fragilis would be more convincing if the experiments included an inert control bacterium, to demonstrate that gavage with any commensal microbe would not elicit a similar effect.

The weight loss was added in Figure S2A. All mice were subjected to the same level of infection in the first place.

Many of the conclusions in the study are drawn from the microscopy results. However, the methods describing both light microscopy and electron microscopy lack sufficient detail. For example, it is not clear how many sections and fields of view were imaged or how the SEM samples were prepared and dehydrated. The mucus layer does not appear to be well preserved, which would make it challenging to accurately measure the thickness of the mucus layer.

For light microscopy, 3-4 fields were selected from each mouse to count about 30 crypts. The method of electron microscopy was complemented on line 263-270. We have removed data of the mucus layer.

Gene expression data appears to vary across the different models, for example, Wnt3 expression in mice versus organoids. Additional experiments may be required to clarify the mechanisms involved. Considering that both of the bacteria tested elicited similar changes in Wnt signaling, this pathway might be broadly modulated by the microbiota.

The reason why the Wnt3 expression pattern is different in mice and in porcine intestinal organoids may be caused by the different infection periods of ETEC in vivo and in vitro. Furthermore, in vivo, the stem cell niche of intestinal stem cells is not only regulated by intestinal epithelial cells, but also affected by mesenchymal cells in connective tissues (Luo et al., 2022). However, in vitro models, stem cell niche is only regulated by epithelial secretory factors, which may also account for the differences in in vitro and in vivo results.

It has been reported that B. fragilis pretreatment significantly increased the relative abundance of A. muciniphila in the intestine of CDI mice, and the growth and maintenance of A. muciniphila were involved in the restoration of intestinal barrier integrity after CDI infection, indicating that there might exist a bacterial metabolic symbiosis between A. muciniphila and B. fragilis (Deng et al., 2018).

References:

Luo HM, Li MX, Wang F, et al. The role of intestinal stem cell within gut homeostasis: Focusing on its interplay with gut microbiota and the regulating pathways. Int. J. Biol. Sci. 2022, 18(13): 5185-5206.

Deng HM, Yang SQ, Zhang YC, et al. Bacteroides fragilis Prevents Clostridium difficile Infection in a Mouse Model by Restoring Gut Barrier and Microbiome Regulation. Front. Microbiol. 2018, 9.

The unconventional choice to not include references in the results section makes it challenging for the reader to put the results in context with what is known in the field. Similarly, there is a lack of discussion acknowledging that B. fragilis is a potential pathogen, associated with intestinal inflammation and cancer (Haghi et al. BMC Cancer 19, 879 (2019) ), and how this would impact its utility as a potential probiotic.

Bacteroides fragilis is one of the symbiotic anaerobes within the mammalian gut and is also an opportunistic pathogen which often isolated from clinical specimens. Bacteroides fragilis was first isolated from the pathogenic site and considered to be pathogenic bacteria. However, with the deepening of research, it is gradually realized that in the long-term evolution process, Bacteroides fragilis colonized in the gut has established a friendly relationship with the host, which is an essential component for maintaining the health of the host, especially for obesity, diabetes and immune deficiency diseases. We have supplemented the discussion on line 598-603.

Reviewer #2 (Public Review):

Ma X. et al proposed that A. muciniphila was a key strain that promotes the proliferation and differentiation of intestinal stem cells by acting on the Wnt/β-catenin signaling pathway. They used various models, such as the piglet model, mouse model, and intestinal organoids to address how A. muciniphila and B. fragilis offer protection against ETEC infection. They showed that FMT with fecal samples, A. muciniphila or B. fragilis protected piglets and/or mice from ETEC infection, and this protection is manifested as reduced intestinal inflammation/bacterial colonization, increased tight junction/Muc2 proteins, as well as proper Treg/Th17 cells. Additionally, they demonstrated that A. muciniphila protected basal-out and/or apical-out intestinal organoids against ETEC infection via Wnt signaling. While a large body of work has been performed in this study, there are quite a few questions to be addressed.

Major comments:

- The similar protective effect of FMT with fecal samples, A. muciniphila or B. fragilis is perhaps not that surprising, considering that FMT likely restores microbiota-mediated colonization resistance against ETEC infection. While FMT with fecal samples increases SCFAs, it is unclear whether/how FMT with A. muciniphila or B. fragilis alter the microbiota composition/abundance as well as metabolites in the current models in a way that offers protection.

We examined changes in the gut microbiota of mice treated with A. muciniphila and B. fragilis through 16s rRNA, and results showed that both A. muciniphila and B. fragilis improved the alpha and beta diversities of the microbiota, while these results were not included in this manuscript.

- Does ETEC infection in piglets/mice cause histological damage in the intestines? These data should be shown.

The results of scanning electron microscopy (Figure 3A) showed the intestinal damage of piglets after ETEC infection. H&E staining and transmission electron microscopy (Figure 5A and 5B) showed the intestinal damage of mice after ETEC infection.

- Line 447, "ETEC adheres to intestinal epithelial cells". However, there is no data showing the adherence (or invasion) of ETEC to intestinal epithelial cells, irrespective of piglets/mouse/organoids.

The scanning electron microscope (Figure 3A bottom) showed that ETEC K88 infected piglets existed obvious rod-shaped bacterial adhesion on the surface of microvilli. Figure 2C showed the colonization of ETEC K88 in the jejunum and colon of piglets. Figure S2A showed the E. coli colonization in intestines and other tissues of mice.

- In both basal-out and apical-out intestinal organoid models, A. muciniphila protects organoids against ETEC infection. Did ETEC enter into intestinal epithelial cells at all after only one hour of infection? Is the protection through certain A. muciniphila metabolites?

It has been reported that the duration of the co-culture for studying the host-microbiota cross-talk by apical-out organoids model is 1 hour (Poletti et al., 2021). In addition, Co et al. (2019) used apical-out organoids model to study host-pathogen interactions, with Salmonella enterica serovar Typhimurium or Listeria monocytogenes invading organoids for an hour.

References:

Poletti M, Arnauts K, Ferrante M, et al. Organoid-based Models to Study the Role of Host-microbiota Interactions in IBD. J. Crohns Colitis. 2021, 15(7): 1222-1235.

Co JY, Margalef-Catala M, Li XN, et al. Controlling Epithelial Polarity: A Human Enteroid Model for Host-Pathogen Interactions. Cell Reports. 2019, 26(9): 2509-2520.

Reviewer #3 (Public Review):

Summary:

The manuscript by Ma et al. describes a multi-model (pig, mouse, organoid) investigation into how fecal transplants protect against E. coli infection. The authors identify A. muciniphila and B. fragilis as two important strains and characterize how these organisms impact the epithelium by modulating host signaling pathways, namely the Wnt pathway in lgr5 intestinal stem cells.

Strengths:

The strengths of this manuscript include the use of multiple model systems and follow-up mechanistic investigations to understand how A. muciniphila and B. fragilis interacted with the host to impact epithelial physiology.

Weaknesses:

The major weakness is that, as presented, the manuscript is quite difficult to follow, even for someone familiar with the field. The lack of detail in figure legends, organization of the text, and frequent use of non-intuitive abbreviated group names without a clear key (ex. EP/EF, or C E A B) make comprehension challenging. The results section is perhaps too succinct and does not provide sufficient information to understand experimental design and interpretation without reading the methods section first or skipping to the discussion (as an example: WNT-c59 treatment). Extensive revisions could be encouraged to aid in communicating the potentially exciting findings.

The abbreviations of experimental groups are firstly defined in the Methods and Materials, and we have supplemented the experimental design in the results section on line 397-399, 439-442 and 516-520.

The bioinformatics section of the methods requires revision and may indicate issues in the pipeline. Merging the forward and reverse reads may represent a problem for denoising. Also since these were sequenced on a NovaSeq, the error learning would have to be modified or the diversity estimates would be inappropriately multiplied. "Alpha diversity and beta diversity were calculated by normalized to the same sequence randomly." Not sure what this means, does this mean subsampled? "Blast was used for sequence alignment", does this mean the taxonomic alignment? This would need to be elaborated on and database versions should be included. The methods, including if any form of multiple testing was included, for LEFSE was also not included.

Denoising was conducted using UNOISE3 to correct for sequencing errors. Subsequent analysis of alpha diversity and beta diversity were all performed based on the output normalized data. Multiple sequence alignment was performed using MUSCLE (v3.8.31) software to obtain the phylogenetic relationships of all OTUs sequences. We have supplemented the method of multiple testing on line 323-328.

Reviewer #1 (Recommendations For The Authors):

At some points, the rationale for using both porcine and murine models was unclear, and it would be helpful for the reader to elaborate on the benefits of these models and why they were used in the introduction. Similarly, it would be helpful to describe the benefits of basal-in organoids versus injecting standard organoids with bacteria.

The main subject of this study was piglets, supplemented by a mouse model for validation. Interpretation of measurements from organoid microinjection experiments must account for multiple confounding variables such as heterogeneous exposure concentrations and durations, as well as impacts of disrupting the organoid wall. We have added the description in the introduction on line 88-90.

Line 165 -- The number of piglets used seems high, is it correct approximately 100 pigs were used?

Nine litters were selected for processing, while only 18 piglets were finally slaughtered.

There is very little discussion of the preliminary experiment that the authors used to determine how much bacteria to use. I recommend either discussing the data and how the doses were chosen or omitting it. It was not clear if the authors used pasteurized or live bacteria in the experiments. It would also be interesting to include a discussion of the observation that relatively low levels of Akkermansia (10^6 CFU) appeared more beneficial than the higher doses, typically used in these types of experiments.

We removed these results. The experiments used live bacteria.

Microscopy methods for both light microscopy and EM would be stronger with added details including how many sections and fields of view were imaged and how the numbers of goblet cells normalized across samples. Without having a clear cross-section of a crypt, it is not clear to me how the images can be used to accurately quantify the number of cells per crypt. Additional details in the methods on how many total crypts were counted should also be included.

For light microscopy, 3-4 fields were selected from each mouse to count about 30 crypts. We have removed the data of the mucus layer and goblet cells.

Line 236 -- missing which gene was used.

The Genbank Accession was added on line 232-233.

Line 310 -- OTU nomenclature.

We have supplemented the OTU nomenclature on line 314.

Line 413 -- This line seems inconsistent with the data analysis described in the methods section. The authors may need to expand their description of the 16S data analysis to be clear and reproducible.

We have redescribed the 16S data analysis on line 312-328.

Line 413 -- it is not surprising that 16s analysis did not capture species, it will have limited resolution beyond the genus level.

We deleted this sentence.

Methods are missing some details on the data analysis, eg. methods/programs and statistical analysis of PCoA and NMDS, LefSe.

The methods and statistical analysis of PCoA, NMDS and LEfSe were supplemented on line 323-328.

Fig 4C -- The images do not clearly capture the mucus layer or how it was analyzed. The sections appear to be cut at a slight angle, with multiple partial sections of crypts. I think this might make it challenging to count goblet cells, especially if the counts are normalized over the number of crypts or villi. The mucus layer does not appear well preserved. For example, I would expect to see an intact mucus layer lining the colon in the PBS control group. Re-cutting sections with a clean cross-section through the tissue will make data analysis easier.

We have removed data of the mucus layer.

Fig 4D -- The images appear to be of the mouse proximal colon, whereas the mucus layer and most muc2 will be in the distal colon. If the authors have tissue sections of the distal colon, this may give a clearer image of the mucus layer and might be more consistent with the TEM images in Fig. 4B.

We apologize for the absence of the distal colon sections.

To fully preserve the mucus layer, in addition to fixing in Carnoy's solution, the embedding process must be run without the standard washes in 70% ethanol (see: Johansson and Hansson. Methods Mol Biol. (2012) 229; doi: 10.1007/978-1-61779-513-8_13). The mucus will wash away during standard paraffin embedding if the tissue is washed with 70% ethanol, and I wonder if that has occurred in these samples.

The tissue wasn’t washed with 70% ethanol.

Fig 6A and 6B -- Although the legend indicates that the data is representative of two independent experiments, it is not clear how many fields of view or cells were imaged. In the bar graphs, it is not clear how many crypts were analyzed and from how many fields of view.

3-4 fields were selected from each mouse to count about 30 crypts.

**For all of the bar graphs, this could be addressed by displaying all of the data points, rather than just the mean, to give the reader a sense of how many cells were counted. (as was done in Fig 7B).

We have changed the bar graphs with data points.

498-501 -- The text says that the gene expression patterns in the organoids are consistent with the in vivo data, but the data patterns of gene expression appear to be different. For example, patterns for Wnt3 and B-catenin expression in mice, appear to be the opposite of what was observed in the organoid?

Lines 509-512 mean that the expression patterns of mice in organoids and in vivo is consistent. Figure 7C was incorrectly written as Figure 8C, we have changed it.

Since Akkermansia does not grow under aerobic conditions, it should be made clear that the organoid co-culture treatment does not involve actively growing bacterial cultures.

Reunanen et al. found that Akkermansia can tolerate oxygen, more than 90% Akkermansia can keep for 1 h under oxic, 5% CO2 conditions.

Reference:

Reunanen J, Kainulainen V, Huuskonen L, et al. Akkermansia muciniphila Adheres to Enterocytes and Strengthens the Integrity of the Epithelial Cell Layer. Appl. Environ. Microbiol. 2015, 81(11): 3655-3662.

Minor points

Line 50 -"evidence".

We have changed to “evidence” on line 49.

Line 64, 422 - italicize, check italics throughout.

We have checked italics throughout the manuscript.

Line 64 - may need to be reworded.

We have changed to “Clostridioides difficile” on line 66.

Line 77 - pathogen.

We have changed to “pathogen” on line 77.

Line 161 - the.

We have removed “the” on line 161.

Line 178 - mouse.

We have changed to “mouse” on line 179.

Line 313 -- wording is confusing.

We have changed the description on line 319-320.

Line 318 -- Silva version #.

The version is Silva 132. We have added it on line 316.

Line 334 - Manufacturer for Live/Dead cell stain?

The Live/Dead cell stain was used BD Biosciences FVS510. We have added it on line 345.

Line 433 -- FD4 not defined until here.

We have refined the FD4 on line 218-219.

Line 512 -- but did not promote.

We have changed to “but did not promote” on line 526.

Line 517 -- Looks like this should be "basal-in organoids" instead of basal-out?

We have changed the "basal-out" to "apical-to" on line 531.

Line 546 -- induced neonatal should be protected?

They are in separate pens.

Jumps from Fig 7B to Fig 8C in the text.

We apologize for the wrong writing, and we have change it.

Reviewer #2 (Recommendations for The Authors):

The title itself is a bit misleading. Please consider changing it. The authors meant that A. muciniphila prevents pathogen invasion, but does not function in pathogen invasion.

We have changed the title.

Major comments:

- Figures 4A, 4D, and 6B should include presentation of cross-section pictures.

We provided cross-section pictures to the journal.

- Figures 7, 8, and 9 should indicate clearly whether mouse or piglet organoids are used. For instance, in the main text, line 490, it indicates piglet organoids, but in Figure 7A legend, it indicates mouse tissue.

We apologize for the misspelling, and have changed to “mice” on line 501-502.

- In Figure 7A, the 3rd row, 2nd panel, crypts formed into spherical organoids; whereas in Figure 8, ETEC infection of basal-out organoids formed budding organoids. This needs to be better explained.

Mouse intestinal organoids were cultured ex vivo from crypts isolated from mice infected with ETEC, while porcine intestinal organoids were co-cultured with ETEC in vitro.

Minor comments:

- In the result section, the numbering of Figures or supplementary Figures is problematic, i.e it should start with Figure 1..., Figure S1, but not directly go to Figure S2A etc.

The Figure 1 was in Materials and Methods.

- Line 458, please add the gating strategy used in the flow cytometry study.

The gating strategy was added on line 351-356.

- The effect of A. muciniphila on the proliferation of intestinal epithelium through the Wnt/β-catenin signaling pathway is well known (such as PMID: 32138776). The authors should discuss this in detail.

We have supplemented the discussion on line 637-639.

Reviewer #3 (Recommendations For The Authors):

It is somewhat unusual that the results from the piglets are in the supplement as this is a major strength of the manuscript (Fig S2).

We have put these results into Figure 2 of the manuscript.

"Collectively, our results may provide theoretical basis that FMT is a promising mitigation method for pathogenic bacteria infection and a new strategy for precise application of FMT in clinical and livestock production"- This is somewhat of an odd statement as the introduction of the manuscript completely skips over most of what is known about FMTs in the context of C. difficile. Also if anything, does the authors' own data not point mostly at using A. muciniphila on its own? Clinical trials are well underway in humans.

We have changed the sentences to “Collectively, our results may provide theoretical basis that A. muciniphila is a promising method to repair intestinal barrier damage and a new strategy for the precise application of A. muciniphila in livestock production.” on line 98-100.

Line 26: I am not sure probiotic is the right word here given its strict scientific definition. Perhaps beneficial or protective would be more appropriate.

We have changed “probiotic” to “beneficial” on line 25.

Line 27: I believe AIMD is antibiotic-induced microbiome-depletion in most usages which may be more accurate and informative than dysregulated.

The type, dosing, and time of antibiotic we used were applied to induce microbiota disorder.

It would appear that there are issues in the reference formatting where a number of journal names are missing.

We have re-edited the reference formatting.

Line 64- I believe eLife requires the standard practice of italicizing genus and species names. Also Clostridium difficile should now be referred to as Clostridioides difficile.

We have changed to “Clostridioides difficile” and italicized it on line 66 and 569. The italicizing genus and species names were checked throughout the manuscript.

Figure S2C: is it not clear why the melt curve was included here, but the legend should make it more clear what is being shown. I assume this is to provide evidence of specificity?

The melting curve was used to demonstrate that only the ETEC K88 could be amplified by the primers we used. We have added an illustration in the figure legend.

Figure 2D: there should be a quantitative analysis done on the staining of Muc2.

We have quantified the staining of MUC2 in Figure 3D.

Figure 3: The legends are not sufficient. For example: it is not clear what Figure 3A actually shows as the y-axis is not labelled and it is not clear what the relationship is between this and the anosim which is a function for permanova.

Anosim analysis was performed using the R software with anosim package function based on the rank order of Bray-Curtis distance values to test the significance of differences between groups. The y-axis is the rank of the distance between samples.

Line 416- OTU not OUT.

We have changed to “OTU” on line 428.

Figure 4- the naming key needs to be included in the figure legend. C, E, A, and B are immediately obvious.

The naming key was included in the figure legend.

Methods: additional information on the flow cytometry gating strategy/controls should be included.

The gating strategy was added on line 351-356.

Author response:

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

Reviewer #1 (Public Review):

I'll begin by summarizing what I understand from the results presented, and where relevant how my understanding seems to differ from the authors' claims. I'll then make specific comments with respect to points raised in my previous review (below), using the same numbering. Because this is a revision I'll try to restrict comments here to the changes made, which provide some clarification, but leave many issues incompletely addressed.

As I understand it the main new result here is that certain recurrent network architectures promote emergence of coordinated grid firing patterns in a model previously introduced by Kropff and Treves (Hippocampus, 2008). The previous work very nicely showed that single neurons that receive stable spatial input could 'learn' to generate grid representations by combining a plasticity rule with firing rate adaptation. The previous study also showed that when multiple neurons were synaptically connected their grid representations could develop a shared orientation, although with the recurrent connectivity previously used this substantially reduced the grid scores of many of the neurons. The advance here is to show that if the initial recurrent connectivity is consistent with that of a line attractor then the network does a much better job of establishing grid firing patterns with shared orientation.

Beyond this point, things become potentially confusing. As I understand it now, the important influence of the recurrent dynamics is in establishing the shared orientation and not in its online generation. This is clear from Figure S3, but not from an initial read of the abstract or main text. This result is consistent with Kropff and Treves' initial suggestion that 'a strong collateral connection... from neuron A to neuron B... favors the two neurons to have close-by fields... Summing all possible contributions would result in a field for neuron B that is a ring around the field of neuron A.' This should be the case for the recurrent connections now considered, but the evidence provided doesn't convincingly show that attractor dynamics of the circuit are a necessary condition for this to arise. My general suggestion for the authors is to remove these kind of claims and to keep their interpretations more closely aligned with what the results show.

We would like to clarify that the simple (flexible) attractor is a weaker condition than the ones previously used to align grid cells. However, by no means we claim that it is a necessary condition for grid maps to align. Other architectures, certainly more complex ones but perhaps even simpler ones, can align grid maps in our model.

Major (numbered according to previous review)

(1) Does the network maintain attractor dynamics after training? Results now show that 'in a trained network without feedforward Hebbian learning the removal of recurrent collaterals results in a slight increase in gridness and spacing'. This clearly implies that the recurrent collaterals are not required for online generation of the grid patterns. This point needs to be abundantly clear in the abstract and main text so the reader can appreciate that the recurrent dynamics are important specifically during learning.

We respectfully disagree with the interpretation of this result. In this model cells self-organize to produce aligned grid maps. In such systems it makes sense to characterize the equilibrium states of the system. We turned learning off in Figure S3 to show that the recurrent connections have a contractive effect on grid spacing. But artificially turning off learning means that one can no longer make claims about the equilibrium states of the system, since it can no longer evolve freely. In a functional network, if the recurrent attractor is removed, the system will evolve towards poor gridness and no alignment no matter what the starting point is, as also shown in Figure S3. Several experimental results invite us to think of grid cells as the equilibrium solution of a series of constraints that is ready to change at any time: Barry et al, 2012; Yoon et al, 2013; Carpenter et al, 2015; Krupic et al, 2015; Krupic et al, 2018; Jayakumar et al, 2019.

One point in which we perhaps agree with the reviewer is that information about the hexagonal maps is kept in the feedforward weights, while behavior and the recurrent collaterals act as constraints of which these feedforward weights are the equilibrium solution.

(2) Additional controls for Figure 2 to test that it is connectivity rather than attractor dynamics (e.g. drawing weights from Gaussian or exponential distributions). The authors provide one additional control based on shuffling weights. However, this is far from exhaustive and it seems difficult on this basis to conclude that it is specifically the attractor dynamics that drive the emergence of coordinated grid firing.

Again, we do not claim that this is the only way in which grid maps can be aligned, but it is the simplest one proposed so far. We were asked if it was the specific combination of input weights to a cell rather than the organization provided by the attractor which resulted in aligned maps. By shuffling the inputs to a cell we keep the combination of inputs invariant but lose the attractor architecture. Since grid maps in this new situation are not aligned, we can safely conclude that it is not the combination of inputs per se, but the specific organization of these inputs that allows grid alignment. It is not fully clear to us what ‘exhaustive’ means in this context.

(3) What happens if recurrent connections are turned off? The new data clearly show that the recurrent connections are not required for online grid firing, but this is not clear from the abstract and is hard to appreciate from the main text.

This point is related to (1). Absent this constraint, Figure S3 shows that the system evolves toward larger spacing, with poorer gridness and no alignment.

(4) This is addressed, although the legend to Fig. S2D could provide an explanation / definition for the y-axis values.

We have now added: Mean input fields are the sum of all inputs of a given kind entering a neuron at a given moment in time, averaged across cells and time.

(5) Given the 2D structure of the network input it perhaps isn't surprising that the network generates 2D representations and this may have little to do with its 1D connectivity. The finding that the networks maintain coordinated grids when recurrent connections are switched off supports my initial concern and the authors explanation, to me at least, remain confusing. I think it would be helpful to consider that the connectivity is specifically important for establishing the coordinated grid firing, but that the online network does not require attractor dynamics to generate coordinated grid firing.

This point is related to (1) and (3). We agree with the reviewer that the input lies within a 2D manifold, but this is not something that the network has to find out because it receives one datapoint of information at a time. This alone is not enough to form aligned grid cells, since each grid cell can find a roughly equivalent equilibrium in a different direction. It is only the constraint imposed by the recurrent collaterals that aligns grid maps, and, as we show, this constraint does not need to be constructed ad hoc to work on 2D, as previously thought. When recurrent connections are switched off, the system evolves toward unaligned grid maps, with larger spacing and lower gridness. Regarding the results obtained after modifying the network and turning off learning, we think they have a very limited scope (in this case showing the contractive effect of recurrent collaterals on grid spacing), given that the system is artificially being kept out of its natural equilibrium.

(6) Clarity of the introduction. This is somewhat clearer, but I wonder if it would be hard for someone not familiar with the literature to accurately appreciate the key points.

We have made our best effort to improve the clarity of the introduction.

(7) Remapping. I'm not sure why this is ill posed. It seems the proposed model can not account for remapping results (e.g. Fyhn et al. 2007). Perhaps the authors could just clearly state this as a limitation of the model (or show that it can do this).

We view our model as perfectly consistent with Fyhn et al, 2007. Remapping is not triggered by the network itself, though, but rather by a re-arrangement of the inputs requiring the network to learn new associations. Different simulations of the same model with identical parameters can be interpreted as remapping experiments.

Reviewer #3 (Public Review):

Summary:

The paper proposes an alternative to the attractor hypothesis, as an explanation for the fact that grid cell population activity patterns (within a module) span a toroidal manifold. The proposal is based on a class of models that were extensively studied in the past, in which grid cells are driven by synaptic inputs from place cells in the hippocampus. The synapses are updated according to a Hebbian plasticity rule. Combined with an adaptation mechanism, this leads to patterning of the inputs from place cells to grid cells such that the spatial activity patterns are organized as an array of localized firing fields with hexagonal order. I refer to these models below as feedforward models.

It has already been shown by Si, Kropff, and Treves in 2012 that recurrent connections between grid cells can lead to alignment of their spatial response patterns. This idea was revisited by Urdapilleta, Si, and Treves in 2017. Thus, it should already be clear that in such models, the population activity pattern spans a manifold with toroidal topology. The main new contributions in the present paper are (i) in considering a form of recurrent connectivity that was not directly addressed before. (ii) in applying topological analysis to simulations of the model. (iii) in interpreting the results as a potential explanation for the observations of Gardner et al.

We wanted to note that we do not see this paper as proposing an alternative to the attractor hypothesis, given that we use attractor networks, but rather as an exploration of possibilities not yet visited by this hypothesis.

Strengths:

The exploration of learning in a feedforward model, when recurrent connectivity in the grid cell layer is structured in a ring topology, is interesting. The insight that this not only align the grid cells in a common direction but also creates a correspondence between their intrinsic coordinate (in terms of the ring-like recurrent connectivity) and their tuning on the torus is interesting as well, and the paper as a whole may influence future theoretical thinking on the mechanisms giving rise to the properties of grid cells.

Weaknesses:

(1) In Si, Kropff and Treves (2012) recurrent connectivity was dependent on the head direction tuning, in addition to the location on a 2d plane, and therefore involved a ring structure. Urdapilleta, Si, and Treves considered connectivity that depends on the distance on a 2d plane. The novelty here is that the initial connectivity is structured uniquely according to latent coordinates residing on a ring.

The recurrent architectures in the cited works are complex and require arranging cells in a 2D manifold to calculate connectivity based on their relative 2D position. In other words, the 2D structure is imprinted in the architecture, as in our 2D condition. In this work the network is much simpler and only requires neighboring relations in 1D. Such relationships have been shown to spontaneously emerge in the hippocampal formation (Pastalkova et al, 2008; Gonzalo Cogno et al, 2024).

(2) The paper refers to the initial connectivity within the grid cell layer as one that produces an attractor. However, it is not shown that this connectivity, on its own, indeed sustains persistent attractor states. Furthermore, it is not clear whether this is even necessary to obtain the results of the model. It seems possible that (possibly weaker) connections with ring topology, that do not produce attractor dynamics but induce correlations between neurons with similar locations on the ring would be sufficient to align the spatial response patterns during the learning of feedforward weights.

Regarding the first part of the comment, the recurrent collaterals create one or at times multiple bumps of activity in the network so that neighboring (interconnected) cells activate together. An initial random state of activity rapidly falls into this dynamic, constrained by the attractor. To us this is not surprising given that this connectivity is the classical means of creating a continuous attractor. Perhaps there is some deeper meaning in this comment that we are not fully grasping.

Regarding the second part of the comment, we fully agree with the reviewer. We are presenting what so far is the simplest connectivity that can align grid maps, but by no means we claim that it is the simplest possible one. Regarding weaker connections with ring topology, we show in Figure S2 that a ring attractor with too weak or too strong connections is incapable of aligning grids, since a balance between feedforward and feedback inputs is required.

(3) Given that all the grid cells are driven by an input from place cells that span a 2d manifold, and that the activity in the grid cell network settles on a steady state which is uniquely determined by the inputs, it is expected that the manifold of activity states in the grid cell layer, corresponding to inputs that locally span a 2d surface, would also locally span a 2d plane. The result is not surprising. My understanding is that this result is derived as a prerequisite for the topological analysis, and it is therefore quite technical.

We understand that the reviewer is referring to the motivation behind studying local dimensionality. We agree that the topological analysis approach is quite technical, but it provides unique insights. The theorem of closed surfaces, which allows us to deduce a toroidal topology from Betti numbers (1,2,1), only applies to closed surfaces. One thus needs to show that the point cloud is a surface (local dimensionality of 2) and is closed (no borders or singularities). If borders or singularities were present, a toroidal topology could not be claimed from these Betti numbers. Thus, it is a crucial step of the analysis.

(4) The modeling is all done in planar 2d environments, where the feedforward learning mechanism promotes the emergence of a hexagonal pattern in the single neuron tuning curve. Under the scenario in which grid cell responses are aligned (i.e. all neurons develop spatial patterns with the same spacing and orientation) it is already quite clear, even without any topological analysis that the emerging topology of the population activity is a torus.

However, the toroidal topology of grid cells in reality has been observed by Gardner et al also in the wagon wheel environment, in sleep, and close to boundaries (whereas here the analysis is restricted to the a sub-region of the environment, far away from the walls). There is substantial evidence based on pairwise correlations that it persists also in various other situations, in which the spatial response pattern is not a hexagonal firing pattern. It is not clear that the mechanism proposed in the present paper would generate toroidal topology of the population activity in more complex environments. In fact, it seems likely that it will not do so, and this is not explored in the manuscript.

We agree that our work was constrained to exploration in 2D and that the situations posed by the reviewer are challenging, but we do not see them as unsurmountable. The wagon wheel shows a preservation of toroidal topology locally, where the behavior of the animal is rather 2-dimensional. Globally, hexagonal maps are lost, which is compatible with some flexibility in the way grid maps are formed. If sleep meant that all inputs are turned off, our model would predict a dynamic dictated by the architecture (1D for the ring attractor, for example), but we do not really know that this is the case. In the future, we intend to explore predictive activity along the linear attractor, which could both result in path integration and in some level of preservation of the activity when inputs are completely turned off.

Regarding boundaries, as we have argued before, the cited work chooses to filter away what looks like more than half of the overall explained variance through PCA, and this is only before applying a non-linear dimensionality reduction algorithm. It is specifically shown that the analyzed components are the ones with global periodicity throughout the environment. Thus, it is conceivable that through this approach, local irregularities found only at the borders are disregarded in favor of a clearer global picture. While using a different methodology, our approach follows a similar spirit, albeit with far less noisy data.

(5) Moreover, the recent work of Gardner et al. demonstrated much more than the preservation of the topology in the different environments and in sleep: the toroidal tuning curves of individual neurons remained the same in different environments. Previous works, that analyzed pairwise correlations under hippocampal inactivation and various other manipulations, also pointed towards the same conclusion. Thus, the same population activity patterns are expressed in many different conditions. In the present model, this preservation across environments is not expected. Moreover, the results of Figure 6 suggest that even across distinct rectangular environments, toroidal tuning curves will not be preserved, because there are multiple possible arrangements of the phases on the torus which emerge in different simulations.

We agree with this observation. A symmetry in our implementation results in the fact that only ~50% of times the system falls in the preferred solution, and the rest of the times it falls into other local minima. Whether this result is at odds with current observations can be debated on the basis of probabilities. However, we believe that the symmetry we found is purely circumstantial, and that it can be broken by elements such as head direction modulation or other ingredients used to achieve path integration. In other words, we acknowledge that symmetry is an issue of the implementation we show here (which has been kept as simple as possible to serve as a proof-of-principle) but we do not think that it is a defining feature of flexible attractors in general. We expect that future implementations that incorporate path integration capabilities will not present this kind of symmetry in the space of solutions.

Regarding the rigid phase translation across modalities, while this effect is very clear in Gardner et al, it is less so in other datasets. The analyses shown in Hermansen et al (2024) can rather be interpreted as somewhere in the way between perfect rigid translation and fully randomized phases across navigation modalities.

(6) In real grid cells, there is a dense and fairly uniform representation of all phases (see the toroidal tuning of grid cells measured by Gardner et al). Thus, the highly clustered phases obtained in the model (Fig. S1) seem incompatible with the experimental reality. I suspect that this may be related to the difficulty in identifying the topology of a torus in persistent homology analysis based on the transpose of the matrix M.

We partly agree with this observation and note that a pattern of ordered phases is an issue not only for the 1D attractor but also for the 2D one, which appears much more uniform than in experimental data. The low number of neurons we used for computational economy and the full connectivity could be key ingredients to generate these phase patterns. To show that this is not a defining feature of flexible attractors, apart from the fact that these patterns appear also with non-flexible 2D architectures, we included in Figure S1 simulations with ‘fragmented 1D’ architectures. In this case the architecture is a superposition of 20 random 1D stripe-like attractors. While the alignment of maps achieved with this architecture is almost at the same level as the one obtained with 1D and 2D attractors, the phases are much more similar to what has been observed experimentally, and less uniform than what is obtained with 2D attractors.

(7) The motivations stated in the introduction came across to me as weak. As now acknolwledged in the manuscript, attractor models can be fully compatible with distortions of the hexagonal spatial response patterns - they become incompatible with this spatial distortions only if one adopts a highly naive and implausible hypothesis that the attractor state is updated only by path integration. While attractor models are compatible with distortions of the spatial response pattern, it is very difficult to explain why the population activity patterns are tightly preserved across multiple conditions without a rigid two-dimentional attractor structure. This strong prediction of attractor models withstood many experimental tests - in fact, I am not aware of any data set where substantial distortions of the toroidal activity manifold were observed, despite many attempts to challenge the model. This is the main motivation for attractor models. The present model does not explain these features, yet it also does not directly offer an explanation for distortions in the spatial response pattern.

Some interesting examples are experiments in 3D, where grid cells presumably communicate with each other through the same recurrent collaterals, but global periodicity is lost and only some local order is preserved even away from boundaries (Ginosar et al, 2021; Grieves et al, 2021). While these datasets have not been explored using topological analysis, they serve as strong motivators to understanding 2D grid cells as one equilibrium solution that arises under some set of constraints, but belongs to a wider space of possible solutions that may arise as well under more flexible constraints. Even (and especially) if one adheres to the hypothesis that grid cells are pre-wired into a 2D torus, a concept like flexible attractors might become useful to understand how their activity is rendered in 3D. Another strong motivation is our lack of understanding of how a perfectly balanced 2D structure is formed and maintained. Simpler architectures could be thought of as alternatives, but also as an intermediate step towards it.

Regarding the rigid phase translation across modalities, while this effect is very clear in Gardner et al, it is less so in other datasets. The analyses shown in Hermansen et al (2024) can rather be interpreted as somewhere in the way between perfect rigid translation and fully randomized phases.

In a separate point, although it might not be strictly related to the comment, we do not fully share the idea that persistent activity patterns during sleep are necessary or sufficient conditions for attractor dynamics, although we do agree that attractors could be the mechanism behind them and any alternative is at least as complex as attractors. On the necessity side, attractors in the hippocampus are not constantly engaged (Wills et al, 2005). For sufficiency, one should prove that no other network is capable of reproducing the phenomenon, and to our best knowledge we are still far from that point.

(8) There is also some weakness in the mathematical description of the dynamics. Mathematical equations are formulated in discrete time steps, without a clear interpretation in terms of biophysically relevant time scales. It appears that there are no terms in the dynamics associated with an intrinsic time scale of the neurons or the synapses (a leak time constant and/or synaptic time constants). I generally favor simple models without lots of complexity, yet within this style of modelling, the formulation adopted in this manuscript is unconventional, introducing a difficulty in interpreting synaptic weights as being weak or strong, and a difficulty in interpreting the model in the context of other studies.

We chose to keep the model as simple as possible and in the line of previous publications developing it. However, we see the usefulness of putting it in what in the meantime has become a canonical framework. Fortunately this has been done by D’Albis and Kempter (2017). In our simplified version of the model there is no leak term and adaptation on its own brings down activity in the absence of input, but we agree that such a term could be added, albeit not without modifying all other network parameters.

In my view, the weaknesses discussed above limit the ability of the model, as it stands, to offer a compelling explanation for the toroidal topology of grid cell population activity patterns, and especially the rigidity of the manifold across environments and behavioral states. Still, the work offers an interesting way of thinking on how the toroidal topology might emerge.

Reviewer 1:

Reviewer #1 (Recommendations For The Authors):

See comments above. In addition:

(1) Abstract: '...interconnected by a two-dimensional attractor guided by path integration'. This is unclear. I think the intended meaning might be along the lines of '...their being computed by a 2D continous attractor that performs path integration'?

'path integration allowing for no deviations from the hexagonal pattern' This is incorrect. Local modulation of the gain of the speed input to a standard CAN would distort the grid pattern.

'Using topological data analysis, we show that the resulting population activity is a sample of a torus' Activity in the model?

'More generally, our results represent a proof of principle against the intuition that the architecture and the representation manifold of an attractor are topological objects of the same dimensionality, with implications to the study of attractor networks across the brain' I guess one might hold this intuition, but it strikes me as obvious that if you impose an sufficiently strong n-dimensional input on a network then it it's activity could have the same dimensionality. I don't really see this as being a point worth highlighting. Perhaps the more interesting point, it that during learning the recurrent connectivity aligns the grid fields of neurons in the network, and this may be a specific function of the 1D attractor dynamcis, although I don't think the authors have made this point convincing.

'The flexibility of this low dimensional attractor allows it to negotiate the geometry of the representation manifold with the feedforward inputs'. See above for comments on the use of 'negotiate'.

'while the ensemble of maps preserves features of the network architecture'. I don't understand this. What is the 'ensemble of maps' and what are the features referred to.

We have reviewed the abstract considering these points. Regarding the ‘strong n-dimensional input’, we want to point out that it is not the input itself that generates a torus (the no attractor condition does not lead to a torus) but rather the interplay between the input and the attractor.

‘Perhaps the more interesting point …’, we do not fully understand how this sentence deviates from our own conclusions. We here show that a strong n-dimensional input is not enough to align grid cells (produce a n-torus), it is the interplay between inputs and attractor dynamics that does so, even if the attractor is not n-dimensional in terms of architecture.

The ensemble of maps refers to the transpose of the population activity matrix, where each point in the cloud is a map, and the features refer to the persistent homology.

(2) The manuscript still fails to clarify the difference between a model that path integrates in two dimensions and a model that simply represents information with a given dimensionality. The argument that it's surprising that a network with 1D architecture represents a higher dimensional input strikes me as incorrect and an unnecessary attempt to argue for conceptual importance. At least to me this isn't surprising. It would be surprising if the 1D network could path integrate but this doesn't seem to be the case.

In response to the reviewer’s concerns, we have made clear in the introduction and discussion that this model has no path integration capabilities, although we aim to develop a model capable of path integration using the kind of simple architecture presented here. We want to highlight here that equating attractor dynamics with path integration would be a conceptual mistake.

(3) Other wording also seems to make unnecessary conceptual claims. E.g. The repeated use of 'negotiate' implies some degree of intelligence, or at least an exchange of information, that isn't shown to exist. I wonder if more precise language could be used? As I understand it the dimensionality is bounded by the inputs on the one hand, and the network connectivity on the other, with the actual dimensionality being a function of the recurrent and feedforward synaptic weights. There's clearly some role for the relative weights and the properties of plasticity rules, but I don't see any evidence for a negotiation.

An interesting observation in Figure S2 is that grid maps are aligned only if the relative strength of feedforward and recurrent inputs is similar. If one of them can impose over the other, grid maps do not align. This equilibrium can metaphorically be thought of as a negotiation instance, where the negotiation is an emergent property of the system rather than something happening at an individual synapse.

---

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

Reviewer #1:

Reviewer #1 (Recommendations For The Authors):

Major

(1) What is the evidence that, after training, the 1D network maintains its attractor dynamics when feedforward inputs are active? If the claim is that it does then it's important to provide evidence, e.g. responses to perturbations, or other tests. The alternative is that after training the recurrent inputs are drowned out by the feed forward spatial inputs.

We agree with the reviewer on the importance of this point. In our model, networks are always learning, and the population activity represented by aligned grid maps in a trained network is a dynamic equilibrium that emerges from the interplay between feedforward and collateral constraints. If Hebbian learning is turned off, one gets a snapshot of the network at that moment. We now show in Fig. S3 that in a trained network without feedforward Hebbian learning the removal of recurrent collaterals results in a slight increase in gridness and spacing. The expansion is due to the fact that, as we argue in the Results section, the attractor has a contractive effect on grid maps, which could relate to observations in novel environments (Barry et al, 2007). If Hebbian learning is turned on in the same situation, the maps, no longer constrained by the attractor, drift toward the equilibrium solution of the ‘No attractor’ condition, with significantly larger spacing, no alignment and lower individual gridness. Thus, the attractor is the force preventing them to do so when feedforward Hebbian learning is on.

These observations point to the key role played by the attractor not only in forming but also in sustaining grid activity. The dynamic equilibrium framework fits well known properties of the system, such as its capacity to recalibrate very fast (Jayakumar et al, 2019), although this particular feature cannot be modeled with the current version of our model, that lacks path integration capabilities.

(2) It would be useful to include additional control conditions for Figure 2 to test the hypothesis that it is simply connectivity, rather than attractor dynamics, that drives alignment.

This could be achieved by randomly assigning strengths to the recurrent connections, e.g. drawing from exponential or Gaussian distributions.

We agree and have included Fig. S2b-d, showing that the same distribution of collateral input weights entering each neuron, but lacking the 1D structure provided by the attractor, does not align grid maps. This is achieved by shuffling rows in the connectivity matrix, while avoiding self connections to make the comparison fair (self connections substantially alter the dynamic of the network, making it much more rigid). We observed that individual grid maps have very low gridness levels, even lower than in the no-attractor condition. In contrast, they have levels of population gridness slightly higher than in the no-attractor condition, but closer to 0 than to levels achieved with attractors. Our interpretation of these results is that irregular connectivity achieves some alignment in a few arbitrary directions and/or locations, which improves the coordination between maps at the expense of impairing rather than improving hexagonal responses of individual cells. Such observations stand in clear context to what is observed with continuous attractors with an orderly architecture.

These results suggest that it is the structure of the attractor that allows grid cells to be aligned rather than the mere presence of recurrent collateral connections.

(3) It seems conceivable that once trained the recurrent connections would no longer be required for alignment. Can this be evaluated by considering what happens if the recurrent connections are turned off after training (or slowly turned off during training)? Does the network continue to generate aligned grid fields?

This point has elements in common with point 1. As we argued in that response, the attractor has two main effects on grid maps: it aligns them and it contracts them. If the attractor is turned off, feedforward Hebbian learning progressively drives maps toward the solution obtained for the ‘no attractor’ condition, characterized by maps with larger spacing, poorer gridness and lack of alignment.

(4) After training what is the relative strength of the recurrent and feedforward inputs to each neuron?

Both recurrent and feedforward synaptic-strength matrices are normalized throughout training, so that the overall incoming synaptic strength to each neuron is invariant. Because of this, although individual feed-forward and recurrent input fields vary dynamically, their average is constant, with the exception of the very first instances of the simulation, before a stable regime is reached in grid-cell activity levels. We have included Fig. S2d, showing the dynamics of feedforward and recurrent mean fields throughout learning as well as their ratio. In addition, Fig. S2a shows that the strength of recurrent relative to feedforward inputs is an important parameter, since alignment is only obtained in an intermediate range of ratios.

(5) It would be helpful to also evaluate the low dimensional structure of the input to the network. Assuming it has a 2D structure, as it represents 2D space, can an explanation be provided for why it is surprising that the trained network also encodes activity with a 2D manifold? It strikes me that the more interesting finding might relate to alignment of the grids rather than claims about a 1D attractor encoding a 2D representation. Either way, stronger evidence and clearer discussion would be helpful.

The reviewer is correct in assuming that the input has a 2D structure, that can be represented by a sheet embedded in a high dimensional space and thus has the Betti numbers [1,0,0]. The surprising element in our results is that we are showing for the first time that the population activity of an attractor network is constrained to a manifold that results from the negotiation between the architecture of the attractor and the inputs, and does not merely reflect the former as previously assumed. In this sense, the alignment of grid cells by a 1D attractor is an instance of the more general case that 1D attractors can encode 2D representations.

It is certainly the case that the 2D input is a strong constraint pushing population activity toward a 2D manifold. However, the final form of the 2D manifold is strongly constrained by the attractor, as shown by the contrast with the no-attractor condition (a 2D sheet, as in the input, vs a torus when the attractor is present). The 1D attractor is able to flexibly adapt to the constraint posed by the inputs while doing its job (as demonstrated in previous points), which results in 2D grid maps aligned by a 1D attractor. Generally speaking, this work provides a proof of principle demonstrating that the topology of the attractor architecture and the manifold of the population activity space need not be identical, as previously widely assumed by the attractor community, and need not even have the same dimensionality. Instead, a single architecture can potentially be applied to many purposes. Hence, our work provides a valuable new perspective that applies to the study of attractors throughout the brain.

(6) The introduction should be clearer about the different types of grid model and the computations they implement. E.g. The authors' previous model generates grid fields from spatial inputs, but if my understanding is correct it isn't able to path integrate. By contrast, while the many 2D models with continuous attractor dynamics also generate grid representations, they do so by path integration mechanisms that are computationally distinct from the spatial transformation implemented by feedforward models (see also general comments above).

We agree with the reviewer and have made this point explicit in the introduction.

(7) A prediction from continuous attractor models is that when place cells remap the low dimensional manifold of the grid activity is unaffected, except that the location of the activity bump is moved. It strikes me as important to test whether this is the case for the model presented here (my intuition is that it won't be, but it would be important to establish either way).

We want to emphasize that our model is a continuous attractor model, so the question regarding the difference between what our model and continuous attractor network models predict is an ill-posed one. One of our main conclusions is precisely that attractors can work in a wider spectrum of ways than previously thought.

In lack of a better definition, our multiple simulations could be thought of as training in different arenas. It is true that in our model maps take time to form, but this is also the case in novel environments (Barry et al, 2007 ), and continuous attractor models exclusively or strongly guided by self motion cues struggle to replicate this phenomenon. We show that the current version of our model accepts multiple solutions (in practice four but conceptually infinite countable), all of them resulting in a torus for the population activity (i.e. the same topology or low dimensional manifold). It is not clear to us how easy it would be to differentiate between most of these solutions in experimental data, with only incomplete information. This said, incorporating a symmetry-breaking ingredient to the model, for example related to head direction modulation, could perhaps lead to the prevalence of a single type of solution. We intend to explore this possibility in the future in order to add path-integration capabilities to the system, as described in the discussion.

(8) The Discussion implies that 1D networks could perform path integration in a manner similar to 2D networks. This is a strong claim but isn't supported by evidence in the study. I suggest either providing evidence that this is the case for models of this kind or replacing it with a more careful discussion of the issue.

The current version of our model has no path integration capabilities, as is now made explicit in the Introduction and Discussion. In addition, we have now made clear that the idea that path integration could perhaps be implemented using 1D networks is, although reasonable, purely speculative.

Minor

(1) Introduction. 'direct excitatory communication between them'. Suggest rewording to 'local synaptic interactions', as communication can also be purely inhibitory (e.g. Burak and Fiete, 2009) or indirect by excitation of local interneurons (e.g. Pastoll et al., Neuron, 2013).

We agree and have adopted this phrasing.

(2) The decision to focus the topology analysis on the 60 cm wide central square appears somewhat arbitrary. Are the irregularities referred to a property of the trained networks or would they also emerge with analysis of simulated ideal data? Can more justification be expanded and supplementary analyses be shown when the whole arena is used?

In practical terms, a subsampling of the data to around half was needed because the persistent homology packages struggle to handle large amounts of data, especially in the calculation of H2. We decided to cut a portion of contiguous pixels in the open field at least larger than the hexagonal tile representing the whole grid population period (as represented in Figure 6). Leaving the borders aside was a logical choice since it is known that the solution at the borders is particularly influenced by the speed anisotropy of the virtual rat (see Si, Kropff & Treves, 2012), in a way that mimics how borders locally influence grid maps in actual rats (Krupic et al, 2015). The specific way in which our virtual rat handles borders is arbitrary and might not generalize. A second issue around borders is that maps are differently affected by incomplete smoothing, although this issue does not apply to our data because we did not smooth across neighboring pixels. In sum, considering the central 60 cm wide square was sufficient to contain the whole torus and a reasonable compromise that would allow us to perform all analyses in the part of the environment less influenced by boundaries.

(3) It could help the general reader to briefly explain what a persistence diagram is.

This is developed in the Appendix, but we have now added a reference to it and a brief description in the main text.

(4) For the analyses in Figure 3-4, and separately for Figure 5, it might help the reader to provide visualizations of the low dimensional point cloud.

All these calculations take place in the original high-dimensional point cloud. Doing them in a reduced space would be incorrect because there is no dimensionality reduction technique that guarantees the preservation of topology. In Figure 7 we reduce the dimensionality of data but emphasize that it is only done for visualization purposes, not to characterize topology. We also point out in this Figure that the same non-linear dimensionality reduction technique applied to objects with identical topology yields a wide variety of visualizations, some of them clear and some less clear. This observation further exemplifies why one cannot assume that a dimensionality-reduction technique preserves topology, even for a low-dimensional object embedded in a high-dimensional space.

(5) The detailed comparison of the dynamics of each model is limited by the number of data points. Why not address this by new simulations with more neurons?

We are not sure we understand this comment. In Figure 2, the dynamics for each model are markedly different. These are averages over 100 simulations. We are not sure what benefit would be obtained from adding more neurons. Before starting this work we searched for the minimal number of neurons that would result in convergence to an aligned solution in 2D networks, which we found to be around 100. Optimizing this parameter in advance was important to reduce computational costs throughout our work.

(6) Could the variability in Figure 7 also be addressed by increasing the number of data points?

As we argued in a previous point, there is no reason to expect preservation of topology after applying Isomap. We believe this lack of topology preservation to be the main driver of variability.

(7) Page/line numbers would be useful.

We agree. However, the text is curated by biorxiv which, to our best knowledge, does not include them.

Reviewer 2:

Reviewer #2 (Recommendations For The Authors):

(1) I highly suggest that the author rewrite some parts of the Results. There are lots of details which should be put into the Methods part, for example, the implementation details of the network, the analysis details of the toroidal topology, etc. It will be better to focus on the results part first in each section, and then introduce some of the key details of achieving these results, to improve the readability of the work.

This suggestion contrasts with that of Reviewer #1. As a compromise, we decided to include in the Results section only methodological details that are key to understanding the conclusions, and describe everything else in the Methods section.

(2) 'Progressive increase in gridness and decrease in spacing across days have been observed in animals familiarizing with a novel environment...' From Fig.2c I didn't see much decrease. The authors may need to carry out some statistical test to prove this. Moreover, even the changes are significant, this might be not the consequence of the excitatory collateral constraint. To prove this, the authors may need to offer some direct evidence.

We agree that the decrease is not evident in this figure due to the scale, so we are adding the correlation in the figure caption as proof. In addition, several arguments, some related to new analyses, demonstrate that the attractor contracts grid maps. First, the ‘no attractor’ condition has a markedly larger spacing compared to all other conditions (Fig. 2a). We also now show that spacing monotonically decreases with the strength of recurrent relative to feedforward weights, in a way that is rather independent of gridness (Fig. S2a). Second, as we now show in Fig. S2b-d, simulations with a shuffled 1D attractor, such that the sum of input synapses to each neuron are the same as in the 1D condition but no structure is present, lead to a spacing that is mid-way between the ‘no attractor’ condition and the conditions with attractors. Third, as we now show in Fig. S3a, turning off both recurrent connections and feedforward learning in a trained network results in a small increase in spacing. Fourth, as we now show in Fig. S3b, turning off recurrent connections while feedforward learning is kept on increases grid spacing to levels comparable to those of the ‘no attractor’ condition. All these elements support a role of the attractor in contracting grid spacing.

(3) Some of the items need to be introduced first before going into details in the paper, for instance, the stipe-like attractor network, the Betti number, etc.

We have added in the Results section a brief description and references to full developments in the Appendix.

Reviewer 3 (Public Review):

(1) It is not clear to me that the proposal here is fundamentally new. In Si, Kropff and Treves (2012) recurrent connectivity was dependent on the head direction tuning and thus had a ring structure. Urdapilleta, Si, and Treves considered connectivity that depends on the distance on a 2d plane.

In the work of Si et al connectivity is constructed ad-hoc for conjunctive cells to represent a torus, it depends on head-directionality but also on the distance in a 2D plane. The topology of this architecture has not been assessed, but it is close to the typical 2D ‘rigid’ constraint. In the work of Urdapilleta et al, the network is a simple 2D one. The difference with our work is that we focus on the topology of the recurrent network and do not use head-direction modulation. In this context, we prove that a 1D network is enough to align grid cells and, more generally, we provide a proof of principle that the topology of the architecture and the representation space of an attractor network do not need to be identical, as previously assumed by the attractor community. These two important points were neither argued, speculated nor self-evident from the cited works.

(2) The paper refers to the connectivity within the grid cell layer as an attractor. However, would this connectivity, on its own, indeed sustain persistent attractor states? This is not examined in the paper. Furthermore, is this even necessary to obtain the results in the model? Perhaps weak connections that do not produce an attractor would be sufficient to align the spatial response patterns during the learning of feedforward weights, and reproduce the results? In general, there is no exploration of how the strength of collateral interactions affects the outcome.

The reviewer makes several important points. Local excitation combined with global inhibition is the archetypical architecture for continuous attractors (see for example Knierim and Zhang, Annual review of neuroscience, 2012). Thus, in the absence of feedforward input, we observe a bump of activity. As in all continuous attractors, this bump is not necessarily ‘persistent’ and instead is free to move along the attractor.

We cannot prove that there is not a simpler architecture that has the same effect as our 1D or 1DL conditions, and we think that there are some interesting candidates to investigate in the future. What we now prove in new Fig. S2b-d is that it is not the strength of recurrent connections themselves, but instead the continuous attractor structure that aligns grid cells in our model. To demonstrate this, we shuffle incoming recurrent connections to each neuron in the 1D condition (while avoiding self-connections for fairness), and show that training does not lead to grid alignment. We also show in Fig. S1 that an architecture represented by 20 overlapping 1DL attractors, each formed by concatenating 10 random cells, aligns grid cells to levels slightly lower but similar to the 1D or 1DL attractors. This architecture can perhaps be considered as simpler to build in biological terms than all the others, but it is still constituted by continuous attractors.

The strength of recurrent collaterals, or more precisely the recurrent to feedforward ratio, is crucial in our model to achieve a negotiated outcome from constraints imposed by the attractor and the inputs. We now show explicit measures of this ratio in Fig. S2, as well as examples showing that an imbalance in this ratio impairs grid alignment. When the ratio is too high or too low, both individual and population gridness are low. Interestingly, grid spacing behaves differently, decreasing monotonically with the relative strength of recurrent connections.

(3) I did not understand what is learned from the local topology analysis. Given that all the grid cells are driven by an input from place cells that spans a 2d manifold, and that the activity in the grid cell network settles on a steady state that depends only on the inputs, isn't it quite obvious that the manifold of activity in the grid cell layer would have, locally, a 2d structure?

The dimensionality of the input is important, although not the only determinant of the topology of the activity. The recurrent collaterals are the other determinant, and their architecture is a crucial feature. For example, as we now show in Figure S2b-d, shuffled recurrent synaptic weights fail to align grid cells. In the 1D condition, if feedforward inputs were absent, the dynamics of the activity would be confined to a ring. The opposite condition is our ‘no attractor’ condition, in which activity in the grid cell layer mimics the topology of inputs, a 2D sheet (and not a torus). It is in the intermediate range, when both feedforward and recurrent inputs are important, that a negotiated solution (a torus) is achieved.

The analyses of local dimensionality and local homology of Figure 3 are crucial steps to demonstrate toroidal topology. According to the theorem of classification of closed surfaces, global homology is not enough to univocally define the topology of a point cloud, and thus this step cannot be skipped. The step is aimed to prove that the point cloud is indeed a closed surface.

(4) The modeling is all done in planar 2d environments, where the feedforward learning mechanism promotes the emergence of a hexagonal pattern in the single neuron tuning curve. This, combined with the fact that all neurons develop spatial patterns with the same spacing and orientation, implies even without any topological analysis that the emerging topology of the population activity is a torus.

We cannot agree with this intuition. In the ‘no attractor’ condition, individual maps have hexagonal symmetry with standardized spacing, but given the lack of alignment the population activity is not a closed surface and thus not a torus. It can rather be described as a 2D sheet embedded in a high dimensional space, a description that also applies to the input space.

While it is rather evident that an ad hoc toroidal architecture folds this 2D population activity into a torus, it is less evident and rather surprising that 1D architectures have the same capability. This is the main novelty in our work.

(5) Moreover, the recent work of Gardner et al. demonstrated much more than the preservation of the topology in the different environments and in sleep: the toroidal tuning curves of individual neurons remained the same in different environments. Previous works, that analyzed pairwise correlations under hippocampal inactivation and various other manipulations, also pointed towards the same conclusion. Thus, the same population activity patterns are expressed in many different conditions. In the present model, the results of Figure 6 suggest that even across distinct rectangular environments, toroidal tuning curves will not be preserved, because there are multiple possible arrangements of the phases on the torus which emerge in different simulations.

We agree with the reviewer in the main point, although the recently found ring activity in the absence of sensory feedback (Gonzalo Cogno et al, 2023) suggests that what is happening in the EC is more nuanced than a pre-wired torus. Solutions in Figure 6 are different ways of folding a 1D strip into a torus, with or without the condition of periodicity in the 1D strip. Whether or not these different solutions would be discernible from one another in a practical setup is not clear to us. For example, global homology, as addressed in the Gardner paper, is the same for all these solutions. Furthermore, while our solutions of up to order 3 are highly discernable, higher order solutions, potentially achievable with other network parameters, would be impossible to discern by eye in representations similar to the ones in Figure 6. In addition, while we chose to keep our model in the simplest possible form as a clear proof of principle, new elements introduced to the model such as head directionality could break the symmetry and lead to the prevalence of one preferred solution for all simulation replicates. We plan to investigate this possibility in the future when attempting to incorporate path-integration capabilities to the model.

(6) In real grid cells, there is a dense and fairly uniform representation of all phases (see the toroidal tuning of grid cells measured by Gardner et al). Here the distribution of phases is not shown, but Figure 7 suggests that phases are non uniformly represented, with significant clustering around a few discrete phases. This, I believe, is also the origin for the difficulty in identifying the toroidal topology based on the transpose of the matrix M: vectors representing the spatial response patterns of individual neurons are localized near the clusters, and there are only a few of them that represent other phases. Therefore, there is no dense coverage of the toroidal manifold that would exist if all phases were represented equally. This is not just a technical issue, however: there appears to be a mismatch between the results of the model and the experimental reality, in terms of the phase coverage.

As mentioned in the results section, Figure 7 is meant for visualization purposes only, and serves more as cautionary tale regarding the imprevisible risks of non-linear dimensionality reduction than as a proof of the organization of activity in the network. Isomap is a non-linear transformation that deforms each of our solutions in a unique way so that, while all have the topology of a torus embedded in a high dimensional space, only a few of them exhibited one of two possible toroidal visualizations in a 3D Isomap reduction. Isomap, as well as all other popular dimensionality reduction techniques, provide no guarantee of topology invariance. A better argument to judge the homogenous distribution of phases is persistent homology, which identifies relatively large holes (compared to the sampling spacing) in the original manifold embedded in a high dimensional space. In our case, persistent homology identified only two holes significantly larger than noise (the two cycles of a torus) and one cavity in all conditions that included attractors. Regarding the specific distribution of phases in different conditions, however, see our reply below.

(7) The manuscript makes several strong claims that incorrectly represent the relation between experimental data and attractor models, on one hand, and the present model on the other hand. For the latter, see the comments above. For the former, I provide a detailed list in the recommendations to the authors, but in short: the paper claims that attractor models induce rigidness in the neural activity which is incompatible with distortions seen in the spatial response patterns of grid cells. However, this claim seems to confuse distortions in the spatial response pattern, which are fully compatible with the attractor model, with distortions in the population activity patterns, which would be incompatible with the attractor model. The attractor model has withstood numerous tests showing that the population activity manifold is rigidly preserved across conditions - a strong prediction (which is not made, as far as I can see, by feedforward models). I am not aware of any data set where distortions of the population activity manifold have been identified, and the preservation has been demonstrated in many examples where the spatial response pattern is disrupted. This is the main point of two papers cited in the present manuscript: by Yoon et al, and Gardner et al.

First of all, we would like to note that our model is a continuous attractor model. Different attractor models have different outcomes, and one of the main conclusions of our manuscript is that attractors can do a wider range of operations than previously thought.

We agree with the reviewer that distortions in spatial activity (which speak against a purely path-integration guided attractor) should not be confused with distortions in the topology of the population activity (which would instead speak against the attractor dynamics itself). We have rephrased these observations in the manuscript. In fact, we believe that the capacity of grid cells to present distorted maps without a distortion of the population activity topology, as shown for example by Gardner and colleagues, could result from a tension between feedforward and recurrent inputs, the potential equilibriums of which our manuscript aims to characterize.

(8) There is also some weakness in the mathematical description of the dynamics. Mathematical equations are formulated in discrete time steps, without a clear interpretation in terms of biophysically relevant time scales. It appears that there are no terms in the dynamics associated with an intrinsic time scale of the neurons or the synapses, and this introduces a difficulty in interpreting synaptic weights as being weak or strong. As mentioned above, the nature of the recurrent dynamics within the grid cell network (whether it exhibits continuous attractor behavior) is not sufficiently clear.

We agree with the reviewer that our model is rather simple, and we value the extent to which this simplicity allows for a deep characterization. All models are simplifications and the best model in any given setup is the one with the minimum amount of complexity necessary to describe the phenomenon under study. We believe that to understand whether or not a 1D continuous attractor architecture can result in a toroidal population activity, a biophysically detailed model, with prohibitive computational costs, would have been unnecessarily complex. This argument does not intend to demerit biophysically detailed models, which are capable of addressing a wider range of questions regarding, for example, the spiking dynamics of grid cells, which cannot be addressed by our simple model.

Reviewer #3 (Recommendations For The Authors):

The work points to an interesting scenario for the emergence of toroidal topology, but the interpretation of this idea should be more nuanced. I recommend reconsidering the claims about limitations of the attractor theory, and acknowledging the limitations of the present theory.

I don't see the limitations mentioned above as a reason to reject the ideas proposed in this manuscript, for two main reasons: first, additional research might reveal a regime of parameters where some issues can be resolved (e.g. the clustering of phases). In addition, the mechanism described here might act at an early stage in development to set up initial dynamics along a toroidal manifold, while other mechanisms might be responsible for the rigidity of the toroidal manifold in an adult animal. But all this implies that the novelty in the present manuscript is weaker than implied, the ability to explain experimental observations is more limited than implied, and these limitations should be acknowledged and discussed.

I recommend reporting on the distribution of grid cell phases and, if indeed clustered, this should be discussed. It will be helpful to explore whether this is the reason for the difficulty in identifying the toroidal topology based on the collection of spatial response patterns (using the transpose of the matrix M).

Ideally, a more complete work would also explore in a more systematic and parametric way the influence of the recurrent connectivity's strength on the learning, and whether a toroidal manifold emerges also in non-planar, such as the wagon-wheel environment studied in Gardner et al.

Part of these recommendations have been addressed in the previous points (public review). Regarding the reason why the transpose of M does not fully recapitulate architecture with our conservative classification criteria, we believe that there is no reason why it should in the first place. We view the fact that the transpose of M recapitulates some features of the architecture as a purely phenomenological observation, and we think it is important as a proof that M is not exactly the same for the different conditions. We imagined that if M matrices were exactly the same this could be due to poor spatial sampling by our bins. Knowing that they are intrinsically different is important even if the reason why they have these specific features is not fully clear to us.

Although we do not think that the distribution of phases is related to the absence of a cavity in the transpose of M or to the four clusters found in Isomap projections, it remains an interesting question that we did not explore initially. We are now showing examples of the distribution of phases in Figure S1. We observed that in both 2D and 1D conditions phases are distributed following rather regular patterns. Whether or not these patterns are compatible with experimental observations of phase distribution is to our view debatable, given that so far state-of-the-art techniques have only allowed to simultaneously record a small fraction of the neurons belonging to a given module. This said, we think that it is important to note that ordered phase patterns are an anecdotal outcome of our simulations rather than a necessary outcome of flexible attractors or attractors in general. To prove this point, we simulated a condition with a new architecture represented by the overlap of 20 short 1DL attractors, each recruiting 10 random neurons from the pool of 100 available ones.

The rest of the parameters of the simulations were identical to those in the other conditions.

By definition, the topology of this architecture has Betti numbers [20,0,0]. We show in Figure S1 that this architecture aligns grid cells, with individual and population gridness reaching slightly lower levels compared to the 1D condition. However, the distribution of phases of these grid cells has no discernible pattern. This result is an arbitrary example that serves as a proof-of-principle to show that flexible attractors can align grid cells without exhibiting ordered phases, not a full characterization of the outcome of this type of architecture, which we leave for future work. For the rest of our work, we stick to the simplest versions of 1D architectures, which allow for a more in-depth characterization.

The wagon-wheel is an interesting case in which maps loose hexagonal symmetry although the population activity lies in a torus, perhaps evidencing the tension between feedforward and recurrent inputs and suggesting that grid cell response does not obey the single master of path integration. If we modeled it with a 1D attractor, we believe the outcome would strongly depend on virtual rat trajectory. If the trajectory was strictly linear, the population activity would be locally one-dimensional and potentially represented by a ring. Instead, if the trajectory allowed for turns, i.e. a 2D trajectory within a corridor-like maze, the population activity would be toroidal as in our open field simulations, while maps would not have perfect hexagonal symmetry, mimicking experimental results.

More minor comments:

Recurrent dynamics are modeled as if there is no intrinsic synaptic or membrane time constant. This may be acceptable for addressing the goals of this paper, but it is a bit unusual and it will be helpful to explain and justify this choice.

As mentioned above, we believe that the best model in a given setup is the one with the lowest number of complexities that can still address the phenomenon under study. One does not use general relativity to build a bridge, although it provides a ‘more accurate’ description of the physics involved. All models are simplifications, and the more complex a model, the more it has to be taken as a black box.

The Introduction mentions that in most models interaction between co-modular neurons occurs through direct excitatory communication, but in quite a few models the interaction is inhibitory. The crucial feature is that the interaction is strongly inhibitory between neurons that differ in their tuning, and either less inhibitory or excitatory between neurons with similar phases.

We agree that directed inhibition has been shown to be as efficient as directed excitation, and we have modified the introduction to reflect this.

The Discussion claims that the present work is the first one in which the topology of the recurrent architecture differs from the topology of the emergent state space. However, early works on attractor models of grid cells showed how neural connectivity which is arranged on a 2d plane, without any periodic boundary conditions, leads to a state space that exhibits the toroidal topology. Therefore, this claim should be revised.

We agree, although the 2D sheet in this case acts as a piece of the torus, and locally the input space and architecture are identical objects. It could be argued that architectures that represent a 2D local slice of the torus, the whole torus, or several cycles around the torus form a continuous family parametrized by the extension of recurrent connections, and as a consequence it is not surprising that these works have not made claims about the incongruence between architecture and representation topologies. The 2D sheet connectivity is still constructed ad hoc to organize activity in a 2D bump, and there is no negotiation between disparate constraints because locally the constraints imposed by input and architecture are the same. We believe this situation is conceptually different from our flexible 1D attractors. We have adapted our claim to include this technical nuance.

Why are neural responses in the perimeter of the environment excluded from the topological analysis? The whole point of the toroidal manifold analysis on real experimental data is that the toroidal manifold is preserved regardless of the animal's location and behavioral condition.

We agree, although experimental data needs to go through extensive pre-processing such as dimensionality reduction before showing a toroidal topology. Such manipulations might smooth away the specific effects of boundaries on maps, together with other sources of noise. In our case, the original reason to downsample the dataset is related to the explosion in computational time that we experience with the ripser package when using more than ~1000 data points. For a proof-of-principle characterization we were much more interested in what happened in the center of the arena, where a 1D attractor could fold itself to confine population activity into a torus. The area we chose was sufficiently large to contain the whole torus. Borders do affect the way the attractor folds (they also affect grid maps in real rats). We feel that these imperfections could be interesting to study in relation to the parameters controlling how our virtual rat behaves at the borders, but not at this proof-of-principle stage.

The periodic activity observed in Ref. 29 could in principle provide the basis for the ring arrangement of neurons. However, it is not yet clear whether grid cells participate in this periodic activity.

We agree. So far it seems that entorhinal cells in general participate in the ring, which would imply that all kinds of cells are involved. However, it could well be that only some functional types participate in the ring and grid cells specifically do not, as future experiments will tell.

eLife assessment

This valuable work explores death coding data to understand the impact of COVID-19 on cancer mortality. The work provides solid evidence that deaths with cancer as a contributing cause were not above what would be expected during pandemic waves, suggesting that cancer did not strongly increase the risk of dying of COVID-19. These results are an interesting exploration into the coding of causes of death that can be used to make sense of how deaths are coded during a pandemic in the presence of other underlying diseases, such as cancer.

Reviewer #1 (Public Review):

Summary:

In the paper, the authors study whether the number of deaths in cancer patients in the USA went up or down during the first year (2020) of the COVID-19 pandemic. They found that the number of deaths with cancer mentioned on the death certificate went up, but only moderately. In fact, the excess with-cancer mortality was smaller than expected if cancer had no influence on the COVID mortality rate and all cancer patients got COVID with the same frequency as in the general population. The authors conclude that the data are consistent with cancer not being a risk factor for COVID and that cancer patients were likely actively shielding themselves from COVID infections.

Strengths:

The paper studies an important topic and uses sound statistical and modeling methodology. It analyzes both, deaths with cancer listed as the primary cause of death, as well as deaths with cancer listed as one of the contributing causes. The authors argue, correctly, that the latter is a more important and reliable indicator to study relationships between cancer and COVID. The authors supplement their US-wide analysis with analysing three states separately.

For comparison, the authors study excess mortality from diabetes and from Alzheimer's disease. They show that Covid-related excess mortality in these two groups of patients was expected to be much higher (than in cancer patients), and indeed that is what the data showed.

Author response:

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

eLife assessment

This valuable work explores death coding data to understand the impact of COVID-19 on cancer mortality. The work provides solid evidence that deaths with cancer as a contributing cause were not above what would be expected during pandemic waves, suggesting that cancer did not strongly increase the risk of dying of COVID-19. These results are an interesting exploration into the coding of causes of death that can be used to make sense of how deaths are coded during a pandemic in the presence of other underlying diseases, such as cancer.

We thank the editor and reviewers for the time they took to review our manuscript and for the thoughtful suggestions they provided. We have completed several revisions based on their feedback and we feel our paper is stronger as a result. However, none of these revisions change the overall conclusions of our study.

Reviewer #1 (Public Review):

Summary:

In the paper "Disentangling the relationship between cancer mortality and COVID-19", the authors study whether the number of deaths in cancer patients in the USA went up or down during the first year (2020) of the COVID-19 pandemic. They found that the number of deaths with cancer mentioned on the death certificate went up, but only moderately. In fact, the excess with-cancer mortality was smaller than expected if cancer had no influence on the COVID mortality rate and all cancer patients got COVID with the same frequency as in the general population. The authors conclude that the data show no evidence of cancer being a risk factor for COVID and that the cancer patients were likely actively shielding themselves from COVID infections.

Strengths:

The paper studies an important topic and uses sound statistical and modeling methodology. It analyzes both, deaths with cancer listed as the primary cause of death, as well as deaths with cancer listed as one of the contributing causes. The authors argue, correctly, that the latter is a more important and reliable indicator to study relationships between cancer and COVID. The authors supplement their US-wide analysis by analysing three states separately.

Weaknesses:

The main findings of the paper can be summarized as six numbers. Nationally, in 2022, multiple-cause cancer deaths went up by 2%, Alzheimer's deaths by 31%, and diabetes deaths by 39%. At the same time, assuming no relationship between these diseases and either Covid infection risk or Covid mortality risk, the deaths should have gone up by 7%, 46%, and 28%. The authors focus on cancer deaths and as 2% < 7%, conclude that cancer is not a risk factor for COVID and that cancer patients must have "shielded" themselves against Covid infections.

However, I did not find any discussion of the other two diseases. For diabetes, the observed excess was 39% instead of "predicted by the null model" 28%. I assume this should be interpreted as diabetes being a risk factor for Covid deaths. I think this should be spelled out, and also compared to existing estimates of increased Covid IFR associated with diabetes.

And what about Alzheimer's? Why was the observed excess 31% vs the predicted 46%? Is this also a shielding effect? Does the spring wave in NY provide some evidence here? Why/how would Alzheimer's patients be shielded? In any case, this needs to be discussed and currently, it is not.

We thank the reviewer for their positive feedback on the paper and for these suggestions. It is true that we have emphasized the impact on cancer deaths, as this was the primary aim of the paper. In the revised version, we have expanded the results and discussion sections to more fully describe the other chronic conditions we used as comparators (lines 267-284;346 – 386).

Note that we are somewhat reluctant to designate any of these conditions as risk factors based solely on comparing the time series model with the demographic model of our expectations. As we mention in the discussion, there is considerable uncertainty around estimates from the demographic model in terms of the size of the population-at-risk, the mean age of the population-at-risk, and the COVID-19 infection rates and infection fatality ratios. Our demographic model is primarily used to demonstrate the effects of competing risks across types of cancers and chronic conditions, since these findings are robust to model assumptions. In contrast, the demographic model should be used with caution if the goal is to titrate the level of these risk factors (as the level of imputed risk is dependent on model assumptions). In the updated version of the manuscript, we have included uncertainty intervals in Table 3, using the upper and lower bounds of the estimated infection rates and IFRs, to better represent this uncertainty. We have also discussed this uncertainty more explicitly in the text and ran sensitivity analyses with different infection rate assumptions in the discussion (lines 354-362; 367 -370).

We would like to note that rather than interpreting the absolute results, we used this demographic model as a tool to understand the relative differences between these conditions. From the demographic model we determined that we would expect to see much higher mortality in diabetes and Alzheimer’s deaths compared to cancer deaths due to three factors (1. Size of population-at-risk, 2. Mean age of the population-at-risk, 3. Baseline risk of mortality from the condition), that are separate from the COVID-19 associated IFR. And in general, this is what we observed.

In comparing the results from the demographic model to the observed excess, diabetes does standout as an outlier from cancer and Alzheimer’s disease in that the observed excess is consistently above the null hypothesis which does lend support to the conclusion that diabetes is in fact a risk factor for COVID-19. A conclusion which is also supported by many other studies. Our findings for hematological cancers are also similar, in that we find consistent support for this condition being a risk factor. We have commented on this in the discussion and added a few references (lines 346-354; 395-403).

Our hypothesis regarding non-hematological cancer deaths (lower than anticipated mortality due to shielding) could also apply to Alzheimer’s deaths. Furthermore, we used the COVID-19 attack rate for individuals >65 years (based on the data that is available), but we estimate that the mean age of Alzheimer’s patients is actually 80-81 years, so this attack rate may in fact be a bit too high, which would increase our expected excess. We have commented on this in the discussion (lines 363-377).

Reviewer #2 (Public Review):

The article is very well written, and the approach is quite novel. I have two major methodological comments, that if addressed will add to the robustness of the results.

(1) Model for estimating expected mortality. There is a large literature using a different model to predict expected mortality during the pandemic. Different models come with different caveats, see the example of the WHO estimates in Germany and the performance of splines (Msemburi et al Nature 2023 and Ferenci BMC Medical Research Methodology 2023). In addition, it is a common practice to include covariates to help the predictions (e.g., temperature and national holidays, see Kontis et al Nature Medicine 2020). Last, fitting the model-independent for each region, neglects potential correlation patterns in the neighbouring regions, see Blangiardo et al 2020 PlosONE.

Thank you for these comments and suggestions. We agree there are a range of methods that can be used for this type of analysis, and they all come with their strengths, weaknesses, and caveats. Broadly, the approach we chose was to fit the data before the pandemic (2014-2019), and project forward into 2020. To our knowledge it is not a best practice to use an interpolating spline function to extrapolate to future years. This is demonstrated by the WHO estimates in Germany in the paper you mention. This was our motivation for using polynomial and harmonic terms.

Based on the above:

a. I believe that the authors need to run a cross-validation to justify model performance. I would suggest training the data leaving out the last year for which they have mortality and assessing how the model predicts forward. Important metrics for the prediction performance include mean square error and coverage probability, see Konstantinoudis et al Nature Communications 2023. The authors need to provide metrics for all regions and health outcomes.

Thank you for this suggestion. We agree that our paper could be strengthened by including cross validation metrics to justify model performance. Based on this suggestion, and your observations regarding Alzheimer’s disease, we have done two things. First, for the full pre-pandemic period (2014-2019) for each chronic condition and location we tested three different models with different degree polynomials (1. linear only, 2. linear + second degree polynomial, 3. linear + second degree polynomial + third degree polynomial) and used AIC to select the best model for each condition and location. Next, also in response to your suggestion, we estimated coverage statistics. Using the best fit model from the previous step, we then fit the model to data from 2014-2018 only and used the model to predict the 2019 data. We calculated the coverage probability as the proportion of weekly observed data points that fell within the 95% prediction interval. For all causes of death and locations the coverage probability was 100% (with the exception of multiple cause kidney disease in California, which is only shown in the appendix). The methods and results have been updated to reflect this change and we have added a figure to the appendix showing the selected model and coverage probability for each cause of death and location (lines 504 – 519; 847-859; Appendix 1- Figure 11).

b. In the context of validating the estimates, I think the authors need to carefully address the Alzheimer case, see Figure 2. It seems that the long-term trends pick an inverse U-shape relationship which could be an overfit. In general, polynomials tend to overfit (in this case the authors use a polynomial of second degree).It would be interesting to see how the results change if they also include a cubic term in a sensitivity analysis.

Thank you for this observation. Based on the changes described above, the model for Alzheimer’s disease now includes a cubic term in the national data and in Texas and California. The model with the second-degree polynomial remained the best fit for New York (Appendix 1 – Figure 11).

c. The authors can help with the predictions using temperature and national holidays, but if they show in the cross-validation that the model performs adequately, this would be fine.

At the scale of the US, adding temperature or environmental covariates is difficult and few US-wide models do so (see Goldstein 2012 and Quandelacy 2014 for examples from influenza). Furthermore, because we are looking at chronic disease outcomes, it is unclear that viral covariates or national holidays would drive these outcomes in the same way as they would if we were looking at mortality outcomes more directly related to transmissible diseases (such as respiratory mortality). Our cross validation also indicates that our models fit well without these additional covariates.

d. It would be nice to see a model across the US, accounting for geography and spatial correlation. If the authors don't want to fit conditional autoregressive models in the Bayesian framework, they could just use a random intercept per region.

We think the reviewer is mistaken here about the scale of our national analysis. Our national analysis did not fit independent models for each state or region. Rather, we fit a single model to the weekly-level national mortality data where counts for the whole of the US have been aggregated. We have clarified in the text (lines 156, 464). As such, we do not feel a model accounting for spatial correlation would be appropriate nor would we be able to include a random intercept for each region. We did fit three states independently (NY, TX, CA), but these states are very geographically distant from each other and unlikely to be correlated. These states were chosen in part because of their large population sizes, yet even in these states, confidence intervals were very wide for certain causes of death. Fitting models to each of the 50 US states, most of which are smaller than those chosen here, would exacerbate this issue.

(2) I think the demographic model needs further elaboration. It would be nice to show more details, the mathematical formula of this model in the supplement, and explain the assumptions

Thank you for this comment. We have added additional details on the demographic model to the methods. We have also extended this analysis to each state to further strengthen our conclusions (lines 548-590).

Reviewing Editor Recommendations:

I think that perhaps something that is missing is that the authors never make their underlying assumption explicit: they are assuming that if cancer increases the risk of dying of COVID-19, this would be reflected in the data on multiple causes of death where cancer would be listed as one of the multiple causes rather than as the underlying cause, and that their conclusions are predicated on this assumption. I would suggest explicitly stating this assumption, as opposed to other reasons why cancer mortality would increase (ex. if cancer care worsened during pandemic waves leading to poorer cancer survival).

Response: Thank you for this suggestion. We have added a few sentences to the introduction to make this assumption clear (lines 106-112).

Reviewer #1 (Recommendations For The Authors):

- It could make sense to add "in the United States" into the title, as the paper only analyses US data.

- It may make sense to reformulate the title from "disentangling the relationship..." into something that conveys the actual findings, e.g. "Lack of excess cancer mortality during Covid-19 pandemic" or something similar. Currently, the title tells nothing about the findings.

Thank you for these suggestions. We have added “in the US” to the title. However, we feel that our findings are a bit more subtle than the suggested reformulation would imply, and we prefer to leave it in its current form.

- Abstract, lines 42--45: This is the main finding of the paper, but I feel it is simplified too strongly in the abstract. Your simulations do *not* "largely explain" excess mortality with cancer; they give higher numbers! Which you interpret as "shielding" etc., but this is completely absent from the abstract. This sentence makes the impression that you got a good fit between simulated excess and real excess, which I would say is not the case.

Thank you for this comment. We have rephrased the sentence in the abstract to better reflect our intentions for using the demographic model (lines 46-49). As stated above, the purpose of the demographic model was not to give a good fit with the observed excess mortality. Rather, we used the demographic model as a tool to understand the relative differences between these conditions in terms of expected excess mortality given the size, age-distribution, and underlying risk of death from the condition itself, assuming similar IFR and attack rates. And based on this, we conclude that it is not necessarily surprising that we see higher excess mortality for diabetes and Alzheimer’s compared to cancer.

- Results line 237: you write that it's "more consistent with the null hypothesis", however clearly it is *not* consistent with the null hypothesis either (because 2% < 7%). You discuss in the Discussion that it may be due to shielding, but it would be good to have at least one sentence about it already here in the Results, and refer to the Discussion.

We have mentioned this in the results and refer to the discussion (lines 277-278).

- Results line 239: why was it closer to the assumption of relative risk 2? If I understand correctly, your model prediction for risk=1 was 7% and for risk=2 it was 13%. In NY you observed 8% (line 187). How is this closer to risk=2?

Thank you for this observation. We have updated the demographic model with new data, extended the model to state-level data, and included confidence intervals on these estimates. We have also added additional discussion around the differences between our observations and expectations (lines 249-284).

- Discussion line 275: "we did not expect to see large increases" -- why exactly? Please spell it out here. Was it due to the age distribution of the cancer patients? Was it due to the high cancer death risk?

We demonstrate that it is the higher baseline risk of death for cancer that seems to be driving our low expectations for cancer excess mortality (lines 304-320). We have added this to the sentence to clarify our conclusions on this point and have added a figure to better illustrate this concept of competing risks (Figure 6).

- Methods, line 405: perhaps it makes sense to cite some other notable papers on Covid excess mortality such as Msemburi et al Nature 2023, Karlinsky & Kobak eLife 2021, Islam et al BMJ 2021, etc.

Thank you for mentioning this oversight. We certainly should have cited these papers and have included them in the updated version.

- Methods line 410: why did you use a 5-week moving average? Why not fit raw weekly death counts? NB regression should be able to deal with it.

Smoothing time series data with a moving average prior to running regression models is a very common practice. We did a sensitivity analysis using the raw data. This produced excess estimates with slightly larger confidence intervals, but does not change the overall conclusions of the paper.

- Methods line 416: please indicate the software/library/package you used for fitting NB regression.

We fit the NB regression using the MASS package in R version 4.3. We have added this to the methods (line 519).

- Line 489: ORCHID -> ORCID

eLife assessment

This work provides a valuable characterization of neural activity in the anterior insular cortex during fear. Using behavior, single unit recording, and optogenetic control of neural activity, the paper provides convincing data on the role of anterior insular circuits in bidirectionally controlling fear. The study is a great starting point on the path to testing hypotheses about bidirectional control of behavior via neural activity in anatomically defined output populations.

Reviewer #1 (Public Review):

The authors tested whether anterior insular cortex neurons that increase or decrease firing during fear behavior, freezing, bidirectionally control fear via separate, anatomically defined outputs. Using a fairly simple behavior where mice were exposed to tone-shock pairings, they found roughly equal populations that increased or decreased firing during freezing. They next tested whether these distinct populations also had distinct outputs. Using retrograde tracers they found that the anterior insular cortex contains non-overlapping neurons which project to the mediodorsal thalamus or amygdala. Mediodorsal thalamus-projecting neurons tended to cluster in deep cortical layers, while amygdala-projecting neurons were primarily in more superficial layers. Stimulation of insula-thalamus projection decreased freezing behavior, and stimulation of insula-amygdala projections increased fear behavior. Given that the neurons which increased firing were located in deep layers, that thalamus projections occurred in deep layers, and that stimulation of insula-thalamus neurons decreased freezing, the authors concluded that the increased-firing neurons were likely thalamic projections. Similarly, given that decreased-firing neurons tended to occur in more superficial layers, that insula-amygdala projections were primarily superficial, and that insula-amygdala stimulation increased freezing behavior, authors concluded that the decreased firing cells were likely amygdala projections. The study has several strengths though also some caveats. Overall, the authors provide a valuable contribution to the field by demonstrating bidirectional control of behavior, linking the underlying anatomy and physiology.

Strengths:

The potential link between physiological activity, anatomy, and behavior is well laid out and is an interesting question. The activity contrast between the units that increase/decrease firing during freezing is clear.

It is nice to see the recording of extracellular spiking activity, which provides a clear measure of neural output, whereas similar studies often use bulk calcium imaging, a signal which rarely matches real neural activity even when anatomy suggests it might.

Weaknesses:

The link between spiking, anatomy, and behavior requires assumptions/inferences: the anatomically/genetically defined neurons which had distinct outputs and opposite behavioral effects can only be assumed the increased/decreased spiking neurons, based on the rough area of cortical layer they were recorded. This is, of course, discussed as a future experiment.

Reviewer #2 (Public Review):

In this study, the authors aim to understand how neurons in the anterior insular cortex (insula) modulate fear behaviors. They report that the activity of a subpopulation of insula neurons is positively correlated with freezing behaviors, while the activity of another subpopulation of neurons is negatively correlated to the same freezing episodes. They then used optogenetics and showed that activation of anterior insula excitatory neurons during tones predicting a footshock increases the amount of freezing outside the tone presentation, while optogenetic inhibition had no effect. Finally, they found that two neuronal projections of the anterior insula, one to the amygdala and another to the medial thalamus, are increasing and decreasing freezing behaviors respectively.

Author response:

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

Reviewer #1 (Public Review):

The authors sought to test whether anterior insular cortex neurons increase or decrease firing during fear behavior and freezing, bi-directionally control fear via separate, anatomically defined outputs. Using a fairly simple behavior where mice were exposed to tone-shock pairings, they found roughly equal populations that do indeed either increase or decrease firing during freezing. Next, they sought to test whether these distinct populations may also have distinct outputs. Using retrograde tracers they found that the anterior insular cortex contains non-overlapping neurons which project to the mediodorsal thalamus or amygdala. Mediodorsal thalamus-projecting neurons tended to cluster in deep cortical layers while amygdala-projecting neurons were primarily in more superficial layers. Stimulation of insula-thalamus projection decreased freezing behavior, and stimulation of insula-amygdala projections increased fear behavior. Given that the neurons that increased firing were located in deep layers, that thalamus projections occurred in deep layers, and that stimulation of insula-thalamus neurons decreased freezing, the authors concluded that the increased firing neurons may be thalamus projections. Similarly, given that decreased-firing neurons tended to occur in more superficial layers, that insula-amygdala projections were primarily superficial, and that insula-amygdala stimulation increased freezing behavior, authors concluded that the decreased firing cells may be amygdala projections. The study has several strengths though also some caveats.

Strengths:

The potential link between physiological activity, anatomy, and behavior is well laid out and is an interesting question. The activity contrast between the units that increase/decrease firing during freezing is clear.

It is nice to see the recording of extracellular spiking activity, which provides a clear measure of neural output, whereas similar studies often use bulk calcium imaging, a signal that rarely matches real neural activity even when anatomy suggests it might (see London et al 2018 J Neuro - there are increased/decreased spiking striatal populations, but both D1 and D2 striatal neurons increase bulk calcium).

Weaknesses:

The link between spiking, anatomy, and behavior requires assumptions/inferences: the anatomically/genetically defined neurons which had distinct outputs and opposite behavioral effects can only be assumed the increased/decreased spiking neurons, based on the rough area of the cortical layer they were recorded.

Yes, we are aware that we could not provide a direct link between spiking, anatomy and behavior. We have specifically noted this in the discussion section and added a possible experiment that could be carried out to provide a more direct link in a future study.

[Lines 371-375] We would like to provide a more direct evidence between the neuronal response types and projection patterns in future studies by electrophysiologically identifying freezing-excited and freezing-inhibited aIC neurons and testing whether those neurons activates to optogenetic activation of amygdala or medial thalamus projecting aIC neurons.

The behavior would require more control to fully support claims about the associative nature of the fear response (see Trott et al 2022 eLife) - freezing, in this case, could just as well be nonassociative. In a similar vein, fixed intertrial intervals, though common practice in the fear literature, pose a problem for neurophysiological studies. The first is that animals learn the timing of events, and the second is that neural activity is dynamic and changes over time. Thus it is very difficult to determine whether changes in neural activity are due to learning about the tone-shock contingency, timing of the task, simply occur because of time and independently of external events, or some combination of the above.

Trott et al. (2022) stated that "...freezing was the purest reflection of associative learning." The nonassociative processes mentioned in the study were related to running and darting behaviors, which the authors argue are suppressed by associative learning. Moreover, considerable evidence from immediate postshock freezing and immediate postshock context shift studies all indicate that the freezing response is an associative (and not nonassociative) response (Fanselow, 1980 and 1986; and Landeira-Fernandez et al., 2006). Thus, our animals' freezing response to the tone CS presentation in a novel context, following three tone CS-footshock US pairings, most likely reflects associative learning. 

Concerning the issue of fixed inter-trial intervals (ITIs), which are standard in fear conditioning studies, particularly those with few CS-US paired trials, we acknowledge the challenge in interpreting the neural correlates of behavior. However, the ITIs in our extinction study was variable and we still found neural activities that had significant correlation with freezing. The results of our extinction study, carried out with variable it is, suggest that the aIC neural activity changes measured in this study is likely due to freezing behavior associated with fear learning, not due to learning the contingencies of fixed ITIs.

Reviewer #2 (Public Review):

In this study, the authors aim to understand how neurons in the anterior insular cortex (insula) modulate fear behaviors. They report that the activity of a subpopulation of insula neurons is positively correlated with freezing behaviors, while the activity of another subpopulation of neurons is negatively correlated to the same freezing episodes. They then used optogenetics and showed that activation of anterior insula excitatory neurons during tones predicting a footshock increases the amount of freezing outside the tone presentation, while optogenetic inhibition had no effect. Finally, they found that two neuronal projections of the anterior insula, one to the amygdala and another to the medial thalamus, are increasing and decreasing freezing behaviors respectively. While the study contains interesting and timely findings for our understanding of the mechanisms underlying fear, some points remain to be addressed.

We are thankful for the detailed and constructive comments by the reviewer and addressed the points. Specifically, we included possible limitations of using only male mice in the study, included two more studies about the insula as references, specified the L-ratio and isolated distance used in our study, added the ratio of putative-excitatory and putative-inhibitory neurons obtained from our study, changed the terms used to describe neuronal activity changes (freezing-excited and freezing-inhibited cells), added new analysis (Figure 2H), rearranged Figure 2 for clarity, added new histology images, and added atlas maps with viral expressions (three figure supplements).

Reviewer #1 (Recommendations For The Authors):

- I would suggest keeping the same y-axis for all figures that display the same data type - Figure 5D, for example.

Thank you for the detailed suggestion. We corrected the y-axis that display the same data type to be the same for all figures.

- In the methods, it says 30s bins were used for neural analysis (line 435). I cannot imagine doing this, and looking at the other figures, it does not look like this is the case so could you please clarify what bins, averages, etc were used for neural and behavioral analysis?

Bin size for neural analysis varied; 30s, 5s, 1s bins were used depending on the analysis. We corrected this and specified what time bin was used for which figure in the methods.

Bin size for neural and freezing behavior was 30s and we also added this to the methods.

- I would not make any claims about the fear response here being associative/conditional. This would require a control group that received an equal number of tone and shock exposures, whether explicitly unpaired or random.

The unpaired fear conditioning paradigm, unpaired tone and shock, suggested by the reviewer is well characterized not to induce fear behavior by CS (Moita et al., 2003 and Kochli et al., 2015). In addition, considerable evidence from immediate post-shock freezing and immediate post-shock context shift studies all indicate that the freezing response is an associative (and not nonassociative) response (Fanselow, 1980 and 1986; and Landeira-Fernandez et al., 2006). Thus, our animals' freezing response to the tone CS presentation in a novel context, following three tone CS-footshock US pairings, most likely reflects associative learning.

- I appreciate the discussion about requiring some inference to conclude that anatomically defined neurons are the physiologically defined ones. This is a caveat that is fully disclosed, however, I might suggest adding to the discussion that future experiments could address this by tagging insula-thalamus or insula-amygdala neurons with antidromic (opto or even plain old electric!) stimulation. These experiments are tricky to perform, of course, but this would be required to fully close all the links between behavior, physiology, and anatomy.

As suggested, we have included that, in a future study, we would like to elucidate a more direct link between physiology, anatomy and behaviors by optogenetically tagging the insula-thalamus/insula-amygdala neurons and identifying whether it may be a positive or a negative cell (now named the freezing-excited and freezing-inhibited cells, respectively) in the discussion.

[Lines 371-375] We would like to provide a more direct evidence between the neuronal response types and projection patterns in future studies by electrophysiologically identifying freezing-excited and freezing-inhibited aIC neurons and testing whether those neurons activates to optogenetic activation of amygdala or medial thalamus projecting aIC neurons.

Reviewer #2 (Recommendations For The Authors):

Major comments:

(1) As all experiments have been performed only in male mice, the authors need to clearly state this limit in the introduction, abstract, and title of the manuscript.

With increasing number of readers becoming interested in the biological sex used in preclinical studies, we also feel that it should be mentioned in the beginning of the manuscript. As suggested, we explicitly wrote that we only used male mice in the title, abstract, and introduction. In addition, we discussed possible limitations of only using male mice in the discussion section as follows:

[Lines 381-386] Another factor to consider is that we have only used male mice in this study. Although many studies report that there is no biological sex difference in cued fear conditioning (42), the main experimental paradigm used in this study, it does not mean that the underlying brain circuit mechanism would also be similar. The bidirectional fear modulation by aIC→medial thalamus or the aIC→amygdala projections may be different in female mice, as some studies report reduced cued fear extinction in females (42).

(2) The authors are missing important publications reporting findings on the insular cortex in fear and anxiety. For example, the authors should cite studies showing that anterior insula VIP+ interneurons inhibition reduces fear memory retrieval (Ramos-Prats et al., 2022) and that posterior insula neurons are a state-dependent regulator of fear (Klein et al., 2021). Also, regarding the anterior insula to basolateral amygdala projection (aIC-BLA), the author should include recent work showing that this population encodes both negative valence and anxiogenic spaces (Nicolas et al., 2023). 

We appreciate the detailed suggestions and we added appropriate publications in the discussion section. The anterior insula VIP+ interneuron study (Ramos-Prats et al., 2022) is interesting, but based on the evidence provided in the paper, we felt that the role of aIC VIP+ interneuron in fear conditioning is low. VIP+ interneurons in the aIC seem to be important in coding sensory stimuli, however, it’s relevance to conditioned stimuli seems to be low; overall VIP intracellular calcium activity to CS was low and did not differ between acquisition and retrieval. Also, inhibition of VIP did not influence fear acquisition. VIP inhibition during fear acquisition did reduce fear retrieval (CS only, no light stimulation), but this does not necessarily mean that VIP activity will be involved in fear memory storage or retrieval, especially because intracellular calcium activity of VIP+ neurons was low during fear conditioning and retrieval.

Studies by Klein et al. (2021) and Nicolas et al. (2023) are integrated in the discussion section as follows.

[Lines 297-301] Group activity of neurons in the pIC measured with fiberphotometry, interestingly, exhibited fear state dependent activity changes—decreased activity with high fear behavior and increased activity with lower fear behavior (29)—suggesting that group activity of the pIC may be involves in maintain appropriate level of fear behavior.

[Lines 316-319] Another distinction between the aIC and pIC may be related with anxiety, as a recent study showed that group activity of aIC neurons, but not that of the pIC, increased when mice explored anxiogenic space (open arms in an elevated plus maze, center of an open field box) (32).

(3) The authors should specify how many neurons they excluded after controlling the L-ratio and isolation distance. It is also important to specify the percentage of putative excitatory and inhibitory interneurons recorded among the 11 mice based on their classification (the number of putative inhibitory interneurons in Figure 1D seems too low to be accurate).

We use manual cluster cutting and only cut clusters that are visually well isolated. So we hardly have any neurons that are excluded after controlling for L-ratio and isolation distance. The criterion we used was L-ratio<0.3 and isolation distance>15, and we specified this in the methods as follows.

[Lines 454-458] We only used well-isolated units (L-ratio<0.3, isolation distance>15) that were confirmed to be recorded in the aIC (conditioned group: n = 116 neurons, 11 mice; control group: n = 14 neurons, 3 mice) for the analysis (46). The mean of units used in our analysis are as follows: L-ratio = 0.09 ± 0.012, isolation distance = 44.97 ± 5.26 (expressed as mean ± standard deviation).

As suggested, we also specified the percentage of putative excitatory and inhibitory interneurons recorded from our study in the results and methods section. The relative percentage of putative excitatory and inhibitory interneurons were similar for both the conditioned and the control groups (conditioned putative-excitatory: 93.1%, putative-inhibitory: 6.9%; control putative-excitatory: 92.9%, putative-inhibitory: 7.1%). Although the number of putative-interneurons isolated from our recordings is low that is what we obtained. Putative inhibitory neurons, probably because of their relatively smaller size, has a tendency to be underrepresented than the putative excitatory cells.

[Lines 83-87] Of the recorded neurons, we analyzed the activity of 108 putative pyramidal neurons (93% of total isolated neurons) from 11 mice, which were distinguished from putative interneurons (n = 8 cells, 7% of total isolated neurons) based on the characteristics of their recorded action potentials (Figure 1D; see methods for details).

[Lines 464-467] The percentage of putative excitatory neurons and putative inhibitory interneurons obtained from both groups were similar (conditioned putative-excitatory: 93.1%, putative-inhibitory: 6.9%; control putative-excitatory: 92.9%, putative-inhibitory: 7.1%).

(4) While the use of correlation of single-unit firing frequency with freezing is interesting, classically, studies analyze the firing in comparison to the auditory cues. If the authors want to keep the correlation analysis with freezing, rather than correlations to the cues, they should rename the cells as "freezing excited" and "freezing inhibited" cells instead of positive and negative cells.

As suggested, we used the terms “freezing-excited” and “freezing-inhibited” cells instead of positive and negative cells.

(5) To improve clarity, Figure 2 should be reorganized to start with the representative examples before including the average of population data. Thus Panel D should be the first one. The authors should also consider including the trace of the firing rate of these representative units over time, on top of the freezing trace, as well as Pearson's r and p values for both of them. Then, the next panels should be ordered as follows: F, G, H, C, A, B, I, and finally E.

We have rearranged Figure 2 based on the suggestions.

(6) It is unclear why the freezing response in Figure 2 is different in current panels F, G, and H. Please clarify this point.

It was because the freezing behaviors of slightly different population of animals were averaged. Some animals did not have positive/negative (or both) cells and only the behavior of animals with the specified cell-type were used for calculating the mean freezing response. With rearrangement of Figure 2, now we do not have plots with juxtaposed mean neuronal response-types and behavior.

(7) Even though the peak of tone-induced firing rate change between negative and positive cells is 10s later for positive cells, the conclusion that this 'difference suggests differential circuits may regulate the activities of different neuron types in response to fear' is overstating the observation. This statement should be rephrased. Indeed, it could be the same circuits that are regulated by different inputs (glutamatergic, GABA, or neuromodulatory inputs).

We agree and delete the statement from the manuscript.

(8) The authors mention they did not find tone onset nor tone offset-induced responses of anterior insula neurons. It would be helpful to represent this finding in a Figure, especially, which were the criteria for a cell to be tone onset or tone offset responding.

We added how tone-onset and tone-offset were analyzed in the methods section and added a plot of the analysis in Figure 2H.

(9) Based on the spread of the viral expression shown in Figure 3B, it appears that the authors are activating/inhibiting insula neurons in the GI layer, whereas single-unit recordings report the electrodes were located in DI, AID, and AIV layers. The authors should provide histology maps of the viral spread for ChR2, NpHR3, and eYFP expression.

Thank you for the excellent suggestion. Now the histological sample in Figure 3B is a sample with expression in the GI/DI/AID layers and it also has an image taken at higher resolution (x40) to show that viral vectors are expressed inside neurons. We also added histological maps with overlay of viral expression patterns of the ChR2, eYFP, and NpHR3 groups in Figure 3—figure supplement 1.

(10) In Figure 5B, the distribution of terminals expressing ChR2 appears much denser in CM than in MD. This should be quantified across mice and if consistent with the representative image, the authors should refer to aIC-CM rather than aIC-MD terminals.

Overall, we referred to the connection as aIC-medial thalamus, which collectively includes both the CM and the MD. Microscopes we have cannot determine whether terminals end at the CM or MD, but the aIC projections seems to pass through the CM to reach the MD. The Allen Brain Institute’s Mouse brain connectivity map (https://connectivity.brain-map.org/projection/experiment/272737914) of a B6 mouse, the mouse strain we used in our study, with tracers injected in similar location as our study also supports our speculation and shows that aIC neuronal projections terminate more in the MD than in the CM. In addition, the power of light delivered for optogenetic manipulation is greatly reduced over distance, and therefore, the MD projecting terminals which is closer to the optic fiber will be more likely to be activated than the CM projecting terminals. However, since we could not determine whether the aIC terminate at the CM or the MD, we collectively referred to the connection as the aIC-medial thalamus throughout the manuscript.

Author response image 1.

(11) Histological verifications for each in vivo electrophysiology, optogenetic, and tracing experiments need to include a representative image of the implantation/injection site, as well as a 40x zoom-in image focusing on the cell bodies or terminals right below the optic fiber (for optogenetic experiments). Moreover, an atlas map including all injection locations with the spread of the virus and fiber placement should be added in the Supplement Figures for each experiment (see Figure S1 Klein et al., 2021). Similarly, the authors need to add a representation of the spread of the retrograde tracers for each mouse used for this tracing experiment.

As suggested, we added a histology sample showing electrode recording location for in-vivo electrophysiology in Figure 1 and added atlas maps for the optogenetic and tracing experiments in supplementary figures. We also provide a 40x zoom-in image of the expression pattern for the optogenetic experiments (Figure 3B).

(12) To target anterior insula neurons, authors mention coordinates that do not reach the insula on the Paxinos atlas (AP: +1.2 mm, ML: -3.4 mm, DV: -1.8 mm). If the DV was taken from the brain surface, this has to be specified, and if the other coordinates are from Bregma, this also needs to be specified. Finally, the authors cite a review from Maren & Fanselow (1996), for the anterior insula coordinates, but it remains unclear why.

AP and ML coordinates are measurement made in reference to the bregma. DV was calculated from the brain surface. We specified these in the Methods. We did not cite a review from Maren & Fenselow for the aIC coordinates.

Minor comments:

(1) A schematic of the microdrive and tetrodes, including the distance of each tetrode would also be helpful.

We used a handcrafted Microdrives with four tetrodes. Since they were handcrafted, the relative orientation of the tetrodes varies and tetrode recording locations has to be verified histologically. We, however, made sure that the distance between tetrodes to be more than 200 μm apart so that distinct single-units will be obtained from different tetrodes. We added this to the methods as follows.

[Lines 430-431] The distance between the tetrodes were greater than 200 μm to ensure that distinct single-units will be obtained from different tetrodes.

(2) Figure 2E: representation of the baseline firing (3-min period before the tone presentation) is missing.

Figure 2E is the 3 min period before tone presentation

(3) Figure 2: Averages Pearson's correlation r and p values should be stated on panels F, G, and H (positive cell r = 0.81, P < 0.05; negative cell r = -0.68, P < 0.05).

They were all originally stated in the figures. But with reorganization of Figure 2, we now have a plot of the Pearson’s Correlation with r and p values in Figure 2F.

(4) Figure 2I: Representation of the absolute value of the normalized firing is highly confusing. Indeed, as the 'negative cells' are inhibited to freezing, firing should be represented as normalized, and negative for the inhibited cells.

To avoid confusion, we did not take an absolute value of the “negative cells”, which are now called the “freezing-inhibited cells”.

(5) Figure 4E (retrograde tracing): representation of individual values is missing.

Figure 4E now has individual values.

References:

London, T. D., Licholai, J. A., Szczot, I., Ali, M. A., LeBlanc, K. H., Fobbs, W. C., & Kravitz, A. V. (2018). Coordinated ramping of dorsal striatal pathways preceding food approach and consumption. Journal of Neuroscience, 38(14), 3547-3558.

Trott, J. M., Hoffman, A. N., Zhuravka, I., & Fanselow, M. S. (2022). Conditional and unconditional components of aversively motivated freezing, flight and darting in mice. Elife, 11, e75663.

Fanselow, M. S. (1980). Conditional and unconditional components of post-shock freezing. The Pavlovian journal of biological science: Official Journal of the Pavlovian, 15(4), 177-182.

Fanselow, M. S. (1986). Associative vs topographical accounts of the immediate shock-freezing deficit in rats: implications for the response selection rules governing species-specific defensive reactions. Learning and Motivation, 17(1), 16-39.

Landeira-Fernandez, J., DeCola, J. P., Kim, J. J., & Fanselow, M. S. (2006). Immediate shock deficit in fear conditioning: effects of shock manipulations. Behavioral neuroscience, 120(4), 873.

Moita, M. A., Rosis, S., Zhou, Y., LeDoux, J. E., & Blair, H. T. (2003). Hippocampal place cells acquire location-specific responses to the conditioned stimulus during auditory fear conditioning. Neuron, 37(3), 485-497.

Kochli, D. E., Thompson, E. C., Fricke, E. A., Postle, A. F., & Quinn, J. J. (2015). The amygdala is critical for trace, delay, and contextual fear conditioning. Learning & memory, 22(2), 92-100.

Ramos-Prats, A., Paradiso, E., Castaldi, F., Sadeghi, M., Mir, M. Y., Hörtnagl, H., ... & Ferraguti, F. (2022). VIP-expressing interneurons in the anterior insular cortex contribute to sensory processing to regulate adaptive behavior. Cell Reports, 39(9).

Klein, A. S., Dolensek, N., Weiand, C., & Gogolla, N. (2021). Fear balance is maintained by bodily feedback to the insular cortex in mice. Science, 374(6570), 1010-1015.

Nicolas, C., Ju, A., Wu, Y., Eldirdiri, H., Delcasso, S., Couderc, Y., ... & Beyeler, A. (2023). Linking emotional valence and anxiety in a mouse insula-amygdala circuit. Nature Communications, 14(1), 5073.

Maren, S., & Fanselow, M. S. (1996). The amygdala and fear conditioning : Has the nut been cracked? Neuron, 16(2), 237‑240. https://doi.org/10.1016/s0896-6273(00)80041-0

eLife assessment

This important study reports that FBXO24 is essential for the normal formation and function of the sperm flagellum, motility, and male fertility in mice. The evidence supporting the direct role of this protein in preventing RNP granule formation in the sperm flagellum is compelling. This work will be of interest to biomedical researchers who work on testicular biology and male fertility.

Reviewer #1 (Public Review):

Summary:

The main goal of the authors was to study the testis-specific role of the protein FBXO24 in the formation and function of the ribonucleoprotein granules (membrane-less electron-dense structures rich in RNAs and proteins).

Strengths:

The wide variety of methods used to support their conclusions (including transgenic models)

Weaknesses:

The complex phenotype observed, in some situations, cannot be fully explained by the experiments presented by the authors (i.e., AR or the tail structure).

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

The main goal of the authors was to study the testis-specific role of the protein FBXO24 in the formation and function of the ribonucleoprotein granules (membraneless electron-dense structures rich in RNAs and proteins).

We appreciate the summary comment of reviewer #1.

Strengths:

The wide variety of methods used to support their conclusions (including transgenic models)

We appreciate the positive comment of reviewer #1.

Weaknesses:

The lack of specific antibodies against FBXO24. Some of the experiments showing a specific phenotype are descriptive and lack of logical explanation about the possible mechanism (i.e. AR or the tail structure).

Because we could not obtain specific antibodies against FBXO24, we generated Fbxo24-FLAG transgenic mice, which can be used to show the interaction between FBXO24 and IPO5. For the mechanism of impaired acrosome reaction, we added some results and discussion as written in the response to the question (1) of reviewer #1 (public review). For the mechanism of abnormal flagellar structure, we added new results and fixed the manuscript as written in the response to the major comments of reviewer #3 (recommendations for the authors).

Questions:

The paper is excellent and employs a wide variety of methods to substantiate the conclusions. I have very few questions to ask:

(1) KO mice cannot undergo acrosome reaction (AR) even spontaneously. How do you account for this, given that no visible defects were observed in the acrosome?

One possibility is that Fbxo24 KO spermatozoa cannot undergo capacitation; however, it is difficult to analyze the capacitation status such as tyrosine phosphorylation because most Fbxo24 KO spermatozoa are not alive (Figure S3A). Other possibility is that AR-related proteins are affected in Fbxo24 KO spermatozoa. Therefore, we analyzed the amounts of AR-related proteins with mass spectrometry (Figure S3C). Although previous studies indicate that the assembly of the SNARE complex is a key event prior to AR [Hutt et al., 2005 (PMID: 15774481); Katafuchi et al., 2000 (PMID: 11066067); Schulz et al., 1997 (PMID: 9356173); Tomes et al., 2002 (PMID: 11884041)], no clear differences were detected for SNARE proteins (Figure S3C and D). PLCD4 that is important for AR [Fukami et al., 2001 (PMID: 11340203)) was also detected in Fbxo24 KO spermatozoa (Figure S3C). Although we could not find differences in the amounts of AR-related proteins, it is still possible that FER1L5, another AR-related protein [Morohoshi et al., 2023 (PMID: 36696506)] not detected in the mass spectrometry analyses, or AR-related proteins not yet identified are affected in Fbxo24 KO spermatozoa. We added these results and discussion (line 160-166 and 305-312).

(2) KO sperm are unable to migrate in the female tract, and, more intriguingly, they do not pass through the utero-tubal junction (UTJ). The levels of ADAM3 are normal, suggesting that the phenotype is influenced by other factors. The authors should investigate the levels of Ly6K since mice also exhibit the same phenotype but with normal levels of ADAM3.

We detected LY6K in Fbxo24 KO spermatozoa with immunoblotting, but no difference was found.

We added the results (Figure S3E and line 172–175).

(3) In Figure 4A, the authors assert that "RBGS Tg mice revealed that mitochondria were abnormally segmented in Fbxo24 KO spermatozoa." I am unable to discern this from the picture shown in that panel. Could you please provide a more detailed explanation or display the information more explicitly?

We are sorry for the ambiguous explanation on the morphology of sperm mitochondria sheath. Fbxo24 KO cauda epidydimal spermatozoa shows disorganized mitochondria sheath rather than “segmented”. We fixed the sentence (line 190-192) and added white arrowheads that indicate the disorganized regions (Figure 4A).

Reviewer #2 (Public Review):

Summary:

The manuscript by Kaneda et al "FBXO24 ensures male fertility by preventing abnormal accumulation of membraneless granules in sperm flagella" is a significant paper on the role of FBXO24 in murine male germ cell development and sperm ultrastructure and function. The body of experimental evidence that the authors present is extraordinarily strong in both breadth and depth. The authors investigate the protein's functions in male germ cells and sperm using a wide variety of approaches but focusing predominantly on their novel mouse model featuring deletion of the Fbxo24 gene and its product. Using this mouse, and a cross of it with another model that expresses reporters in the head and midpiece, they logically build from one experiment to the next. Together, their data show that this protein is involved in the regulation of membraneless electron-dense structures; loss of FBXO24 led to an accumulation of these materials and defects in the sperm flagellum and fertilizing ability. Interestingly, the authors found that several of the best-known components of electron-dense ribonucleoprotein granules that are found in the intermitochondrial cement and chromatoid body were not disrupted in the Fbxo24 knockout, suggesting that the electron-dense material and these structures are not all the same, and the biology is more complicated than some might have thought. They found evidence for the most changes in IPO5 and KPNB1, and biochemical evidence that FBXO24 and IPO5 could interact.

We appreciate the summary comment of reviewer #2.

Strengths:

The authors are to be commended for the thoroughness of their experimental approaches and the extent to which they investigated impacts on sperm function and potential biochemical mechanisms. Very briefly, they start by showing that the Fbxo24 message is present in spermatids and that the protein can interact with SKP1, in a way that is dependent on its F-box domain. This points toward a potential function in protein degradation. To test this, they next made the knockout mouse, validated it, and found the males to be sterile, although capable of plugging a female. Looking at the sperm, they identified a number of ultrastructural and morphological abnormalities, which they looked at in high resolution using TEM. They also cross their model with RBGS mice so that they have reporters in both the acrosome and mitochondria. The authors test a variety of sperm functions, including motility parameters, ability to fertilize by IVF, cumulus-free IVF, zona-free-IVF, and ICSI. They found that ICSI could rescue the knockout but not other assisted reproductive technologies. Defects in male fertility likely resulted from motility disruption and failure to get through the utero-tubal junction but defects in acrosome exocytosis also were noted. The authors performed thorough investigations including both targeted and unbiased approaches such as mass spectrometry. These enabled them to show that although the loss of the FBXO24 protein led to more RNA and elevated levels of some proteins, it did not change others that were previously identified in the electron-dense RNP material.

The manuscript will be highly significant in the field because the exact functions of the electron-dense RNP materials have remained somewhat elusive for decades. Much progress has been made in the past 15 years but this work shows that the situation is more complex than previously recognized. The results show critical impacts of protein degradation in the differentiation process that enables sperm to change from non-descript round cells into highly polarized and compartmentalized mature sperm, with an equally highly compartmentalized flagellum. This manuscript also sets a high bar for the field in terms of how thorough it is, which reveals wide-ranging impacts on processes such as mitochondrial compaction and arrangement in the midpiece, the correct building of the major cytoskeletal elements in the flagellum, etc.

We appreciate the positive comment of reviewer #2.

Weaknesses:

There are no real weaknesses in the manuscript that result from anything in the control of the authors. They attempted to rescue the knockout by expressing a FLAG-tagged Fbxo24 transgene, but that did not rescue the phenotype, either because of inappropriate levels/timing/location of expression, or because of interference by the tag. They also could not make anti-FBXO24 that worked for coimmunoprecipitation experiments, so relied on the FLAG epitope, an approach that successfully showed co-IP with IPO5 and SKP1.

We could not rescue the phenotype with Fbxo24-FLAG transgene, but different Fbxo24 mutant mice show the same phenotypes (Figure S6G). Further, another group showed that Fbxo24 KO mice exhibited abnormal mitochondrial coiling [Li et al., 2024 (PMID: 38470475)], confirming that

FBXO24 is involved in the mitochondrial sheath formation.

Reviewer #3 (Public Review):

Summary:

In this manuscript, the authors found that FBXO24, a testis-enriched F-box protein, is indispensable for male fertility. Fbxo24 KO mice exhibited malformed sperm flagellar and compromised sperm motility.

We appreciate the summary comment of reviewer #3.

Strengths:

The phenotype of Fbxo24 KO spermatozoa was well analyzed.

We appreciate the positive comment of reviewer #3.

Weaknesses:

The authors observed numerous membraneless electron-dense granules in the Fbxo24 KO spermatozoa. They also showed abnormal accumulation of two importins, IPO5 and KPNB1, in the Fbxo24 KO spermatozoa. However, the data presented in the manuscript do not support the conclusion that FBXO24 ensures male fertility by preventing the abnormal accumulation of membraneless granules in sperm flagella, as indicated in the manuscript title.

Fbxo24 KO mice showed abnormal accumulation of membraneless granules in sperm flagella and male infertility, suggesting that FBXO24 is involved in these processes, but there are no results that show the direct relationship as reviewer #3 mentioned. Therefore, we fixed the title.

Recommendations For The Authors:

Reviewer #2 (Recommendations For The Authors):

On page 4, lines 152-154, the authors introduce the RBGS mouse model and use it in their experiments.

However, they left out an obvious but helpful sentence that tells the reader that they crossed the Fbxo24-null mouse with the RBGS. As one continues reading it is clear, but best to avoid even slight confusion.

We revised the explanation in the result section (line 150-153).

Reviewer #3 (Recommendations For The Authors):

In this manuscript, the authors found that FBXO24, a testis-enriched F-box protein, is indispensable for male fertility. Fbxo24 KO mice exhibited malformed sperm flagellar and compromised sperm motility. The phenotype of Fbxo24 KO spermatozoa was well analyzed.

The authors observed numerous membraneless electron-dense granules in the Fbxo24 KO spermatozoa. They also showed abnormal accumulation of two importins, IPO5 and KPNB1, in the Fbxo24 KO spermatozoa. However, the data presented in the manuscript do not support the conclusion that FBXO24 ensures male fertility by preventing the abnormal accumulation of membraneless granules in sperm flagella, as indicated in the manuscript title.

Fbxo24 KO mice showed abnormal accumulation of membraneless granules in sperm flagella and male infertility, suggesting that FBXO24 is involved in these processes, but there are no results that show the direct relationship as reviewer #3 mentioned. Therefore, we fixed the title.

Major comments:

In the title, abstract, introduction, and some sections such as lines 275-276, the authors conclude that FBXO24 prevents the accumulation of importins and RNP granules during spermiogenesis. However, the provided data do not substantiate this claim. To provide conclusive evidence to support the current title, the authors need to present evidence supporting: 1) direct degradation of IPO5 and KPNB1 by FBXO24; 2) the direct requirement of IPO5 for the formation of the membraneless granules, and 3) infertility resulting from the presence of membraneless granules, rather than other issues such as abnormal ODF and AX.

(1) direct degradation of IPO5 and KPNB1 by FBXO24.

To examine if IPO5 can be degraded by FBXO24, we performed a ubiquitination assay using HEK293T cells. Ubiquitination of IPO5 was upregulated in the presence of WT FBXO24 but not with the mutant ΔF-box FBXO24, suggesting that IPO5 can be ubiquitinated by FBXO24. We did not examine the ubiquitination of KPNB1 because we failed to construct a plasmid vector expressing mouse KPNB1. We think that KPNB1 is not the substrate because we did not detect the interaction between FBXO24 and KPNB1 (Figure 5E). We added the results of the ubiquitination assay (Figure

5F and line 261-265) and mentioned it in the abstract (line 35).

(2) the direct requirement of IPO5 for the formation of the membraneless granules.

(3) infertility resulting from the presence of membraneless granules, rather than other issues such as abnormal ODF and AX.

We revealed that IPO5 aggregate under stress condition in COS7 cells (Figure 6C and D); however, we did not examine whether IPO5 is required for the formation of the membraneless granules. We consider that protein degradation systems such as PROTAC or Trim-Away to knockdown IPO5 at the protein level in Fbxo24 KO mice could be a good way to see if the membraneless granules are diminished and male fertility is rescued. However, it takes time to apply the degradation systems in vivo. Therefore, we would like to leave this rescue experiment for future studies. We fixed the title and  abstract (line 37-38), and removed the last sentence of the introduction.

Also, the other group reported the analyses of Fbxo24 KO mice [Li et al., 2024 (PMID: 38470475)] right after we submitted our manuscript to the eLife. They reported not only disorganized flagellar structures but also abnormal head morphology, which may lead to male infertility. The differences from our study may be due to different mouse genetic backgrounds. We mentioned it in the discussion section (line 348-353).

Minor comments:

(1) The authors claimed a significant increase in the total amount of RNAs in Fbxo24 KO spermatozoa (lines 259-261), suggesting that the ...contain RNAs. More direct evidence supporting this claim should be provided.

We show that the amounts of IPO5 and KBNB1 increased in Fbxo24 KO spermatozoa (Figure 5A and B), both of which could be incorporated into RNP granules in COS7 cells (Figure 6C and D), supporting the idea that membraneless electron-dense structures may be RNP granules. However, because we did not show direct evidence that electron-dense structures contain RNAs, we removed the sentences (line 259-261 of the 1st submission manuscript). 

(2) The author should provide an explanation for the absence of a FLAG band in the input Tg in Figure 5D and the larger size of the IPO5 band in the FLAG-IP group compared to the input. Similar observations are also noted in Figure 5E.

The FLAG band is weak because the protein amount is low. When we increase the contrast, we can see the FLAG band. We added an image with high contrast (Figure 5D). Sometimes, proteins run differently with SDS-PAGE after immunoprecipitation, likely due to varying protein composition in the sample. We explained it in the figure legend (line 868-869).

(3) In Line 526, clarify the procedure for sperm purification, and determine the potential for contamination from somatic cells.

We did not perform sperm purification, but when we observed spermatozoa obtained from cauda epididymis, we rarely observed either somatic cells or immature spermatogenic cells. We added  pictures in Figure S7. Further, we added detailed explanation about how to collect spermatozoa from the epididymis (line 549-550).

(4) Define the Y-axis in Figure 2E, F, and G.

We have revised the figures.

Author response:

Reviewer #1 (Public Review):

Using the UK Biobank, this study assessed the value of nuclear magnetic resonance measured metabolites as predictors of progression to diabetes. The authors identified a panel of 9 circulating metabolites that improved the ability in risk prediction of progression from prediabetes to diabetes. In general, this is a well-performed study, and the findings may provide a new approach to identifying those at high risk of developing diabetes. I have some comments that may improve the importance of this study.

We deeply appreciate the reviewer's invaluable time dedicated to the review of this manuscript and the insightful comments to enhance its overall quality.

(1) It is unclear why the authors only considered the top 20 variables in the metabolite selection and why they did not set a wider threshold.

Thank you for the comment. We set the top 20 variables in the metabolite selection balancing the performance of the final diabetes risk prediction model and the clinical applicability due to measurement costs. We have added this explanation in the “Methods” section.

“We chose the intersection set of the top 20 most important variables selected by the three machine learning models, after balancing the performance of the final diabetes risk prediction model and the clinical applicability associated with measurement costs of metabolites.”

(2) The methods section would benefit from a more detailed exposition of how parameter tuning was conducted and the range of parameters explored during the training of the RSF model.

According to the reviewer’s suggestion, we have added a more detailed description of parameters tunning and the range of parameters explored during the training of the RSF model in the “Method S2” section in the Supplementary material.

“The RSF model was fitted using the “randomForestSRC” package and the grid search method was used for hyperparameter tuning. Specifically, the grid search method was used to tune hyperparameters among the RSF model, through minimizing out-of-sample or out-of-bag error1. Each tree in the RSF is constructed from a random sample of the data, typically a bootstrap sample or 63.2% of the sample size (as in the present study). Consequently, not all observations are used to construct each tree. The observations that are not used in the construction of a tree are referred to as out-of-bag observations. In an RSF model, each tree is built from a different sample of the original data, so each observation is “out-of-bag” for some of the trees. The prediction for an observation can then be obtained using only those trees for which the observation was not used for the construction. A classification for each observation is obtained in this way and the error rate can be estimated from these predictions. The resulting error rate is referred to as the out-of-bag error. Through calculating the out-of-bag error in each iteration, the best hyperparameters were finally determined.

The hyperparameters to be tuned and range of grid search in the present study were below: number of trees (50-1000, by 50), number of variables to possibly split at each node (3-6, by 1), and minimum size of terminal node (1-20, by 1)2.”

(3) It is hard to understand the meaning of the decision curve analysis and the clinical implications behind the net benefit, which are required to clarify the application values of models.

Thank you for the comment. We have added more description and discussion about the decision curve analysis in the “Methods” and “Discussion” sections.

“Furthermore, we used decision curve analysis (DCA) to assess the clinical usefulness of prediction model-based guidance for prediabetes management, which calculates a clinical “net benefit” for one or more prediction models in comparison to default strategies of treating all or no patients3.”

“Most importantly, a model with good discrimination does not necessarily have high clinical value. Hence, DCA was used to compare the clinical utility of the model before and after adding the metabolites, and this showed a higher net benefit for the latter than the basic model, suggesting the addition of the metabolites increased the clinical value of prediction, i.e., the potential benefit of guiding management in individuals with prediabetes3,4. These results provided novel evidence supporting the value of metabolic biomarkers in risk prediction and stratification for the progression from prediabetes to diabetes.”

(4) Notably, the NMR platform utilized within the UK Biobank primarily focused on lipid species. This limitation should be discussed in the manuscript to provide context for interpreting the results and acknowledge the potential bias from the measuring platform.

Thank you for the comment. We acknowledged this limitation that NMR platform within the UK Biobank primarily focused on lipid species and the potential bias from the measuring platform and have added this in “Discussion” section.

“Third, the Nightingale metabolomics platform primarily focused on lipids and lipoprotein sub-fractions, and thus the predictive value of other metabolites in the progression from prediabetes to diabetes warranted further research using an untargeted metabolomics approach.”

(5) The manuscript should explain the potential influence of non-fasting status on the findings, particularly concerning lipoprotein particles and composition. There should be a detailed discussion of how non-fasting status may impact the measurement and the findings.

According to the reviewer’s suggestion, we have added more details to explain the potential influence of non-fasting status on our findings in the “Discussion” section.

“Additionally, the use of non-fasting blood samples might increase inter-individual variation in metabolic biomarker concentrations, however, fasting duration has been reported to account for only a small proportion of variation in plasma metabolic biomarker concentrations5. Therefore, we believe the impact of non-fasting samples on our findings would be minor.”

(6) Cross-platform standardization is an issue in metabolism, and further descriptions of quality control are recommended.

Thank you for the comment. We have added more description of quality control in the “Method S1” section in the Supplementary material.

“Metabolic biomarker profiling by Nightingale Health’s NMR platform provides consistent results over time and across spectrometers. Furthermore, the sample preparation is minimal in the Nightingale Health’s metabolic biomarker platform, circumventing all extraction steps. These aspects result in highly repeatable biomarker measurements. Pre-specified quality metrics were agreed between UK Biobank and Nightingale Health to ensure consistent results across the samples, and pilot measurements were conducted. Nightingale Health performed real-time monitoring of the measurement consistency within and between spectrometers throughout the UK Biobank samples. Two control samples provided by Nightingale Health were included in each 96-well plate for tracking the consistency across multiple spectrometers. Furthermore, two blind duplicate samples provided by the UK Biobank were included in each well plate, with the position information unlocked only after results delivery. Coefficient of variation (CV) targets across the metabolic biomarker profile were pre-specified for both Nightingale Health’s internal control samples and UK Biobank’s blind duplicates. The targets were met for each consecutively measured batch of ~25,000 samples. For the majority of the metabolic biomarkers, the CVs were below 5% (https://biobank.ndph.ox.ac.uk/showcase/refer.cgi?id=3000). Further, the distributions of measured biomarkers from 5 sample batches indicated absence of batch effects (https://biobank.ctsu.ox.ac.uk/ukb/ukb/docs/nmrm_app1).”

Reviewer #2 (Public Review):

Deciphering the metabolic alterations characterizing the prediabetes-diabetes spectrum could provide early time windows for targeted preventive measures to extend precision medicine while avoiding disproportionate healthcare costs. The authors identified a panel of 9 circulating metabolites combined with basic clinical variables that significantly improved the prediction from prediabetes to diabetes. These findings provided insights into the integration of these metabolites into clinical and public health practice. However, the interpretation of these findings should take account of the following limitations.

We appreciate the reviewer’s positive comments and encouragement.

(1) First, the causal relationship between identified metabolites and diabetes or prediabetes deserves to be further examined particularly when the prediabetic status was partially defined. Some metabolites might be the results of prediabetes rather than the casual factors for progression to diabetes.

Thank you for your insightful comments. We agree with you that the panel of metabolites in this study might not be the causal factor for progression from prediabetes to diabetes, which needs further validation in experimental studies. We have added this limitation in the “Discussion” section.

“Fifth, we could not draw any conclusion about the causality between the identified metabolites and the risk for progression to diabetes due to the observational nature, which remained to be validated in further experimental studies.”

(2) The blood samples were taken at random (not all in a non-fasting state) and so the findings were subjected to greater variability. This should be discussed in the limitations.

According to the reviewer’s suggestion, we have added more details to explain the potential influence of non-fasting status on our findings in the “Discussion” section.

“Additionally, the use of non-fasting blood samples might increase inter-individual variation in metabolic biomarker concentrations, however, fasting duration has been reported to account for only a small proportion of variation in plasma metabolic biomarker concentrations5. Therefore, we believe the impact of non-fasting samples on our findings would be minor.”

(3) The strength of NMR in metabolic profiling compared to other techniques (i.e., mass spectrometry [MS], another commonly used metabolic profiling method) could be added in the Discussion section.

According to the reviewer’s suggestion, we have added the strength of NMR in metabolic profiling compared to other techniques in the “Discussion” section.

“Circulating metabolites were quantified via NMR-based metabolome profiling within the UK Biobank, which offers metabolite qualification with relatively lower costs and better reproducibility6.”

(4) Fourth, the applied platform focuses mostly on lipid species which may be a limitation as well.

Thank you for the comment. We acknowledged this limitation that NMR platform within the UK Biobank primarily focused on lipid species and the potential bias from the measuring platform and have added this in the “Discussion” section.

“Third, the Nightingale metabolomics platform primarily focused on lipids and lipoprotein sub-fractions, and thus the predictive value of other metabolites in the progression from prediabetes to diabetes warranted further research using an untargeted metabolomics approach.”

(5) it is a very large group with pre-diabetes, but the results only apply to prediabetes and not to the general population. This should be clear, although the authors have also validated the predictive value of these metabolites in the general population.

Thank you for the comment. We agree with you that the results only apply to prediabetes and not to the general population, though they also showed potential predictive value among participants with normoglycemia. We have accordingly modified the relevant expressions in the “Conclusion” section to restrict these findings to participants with prediabetes.

“In this large prospective study among individuals with prediabetes, we detected a panel of circulating metabolites that were associated with an increased risk of progressing to diabetes.”

References

(1) Janitza S, Hornung R. On the overestimation of random forest's out-of-bag error. PLoS One. 2018;13(8):e0201904.

(2) Tian D, Yan HJ, Huang H, et al. Machine Learning-Based Prognostic Model for Patients After Lung Transplantation. JAMA Netw Open. 2023;6(5):e2312022.

(3) Vickers AJ, van Calster B, Steyerberg EW. A simple, step-by-step guide to interpreting decision curve analysis. Diagn Progn Res. 2019;3:18.

(4) Li J, Xi F, Yu W, Sun C, Wang X. Real-Time Prediction of Sepsis in Critical Trauma Patients: Machine Learning-Based Modeling Study. JMIR Form Res. 2023;7:e42452.

(5) Li-Gao R, Hughes DA, le Cessie S, et al. Assessment of reproducibility and biological variability of fasting and postprandial plasma metabolite concentrations using 1H NMR spectroscopy. PLoS One. 2019;14(6):e0218549.

(6) Geng T-T, Chen J-X, Lu Q, et al. Nuclear Magnetic Resonance–Based Metabolomics and Risk of CKD. American Journal of Kidney Diseases. 2023.

eLife assessment

This valuable study combines prospective cohort, metabolomics, and machine learning to identify a panel of 9 circulating metabolites that improved the ability in risk prediction of progression from prediabetes to diabetes. The findings are solid and the methods, data, and analyses support the claims. However, the interpretation would benefit from a more rigorous description. With revision of these weaknesses, this paper would provide insights into the integration of these metabolites into clinical and public health practice.

Reviewer #1 (Public Review):

Using the UK Biobank, this study assessed the value of nuclear magnetic resonance measured metabolites as predictors of progression to diabetes. The authors identified a panel of 9 circulating metabolites that improved the ability in risk prediction of progression from prediabetes to diabetes. In general, this is a well-performed study, and the findings may provide a new approach to identifying those at high risk of developing diabetes.

I have some comments that may improve the importance of this study.

(1) It is unclear why the authors only considered the top 20 variables in the metabolite selection and why they did not set a wider threshold.

(2) The methods section would benefit from a more detailed exposition of how parameter tuning was conducted and the range of parameters explored during the training of the RSF model.

(3) It is hard to understand the meaning of the decision curve analysis and the clinical implications behind the net benefit, which are required to clarify the application values of models.

(4) Notably, the NMR platform utilized within the UK Biobank primarily focused on lipid species. This limitation should be discussed in the manuscript to provide context for interpreting the results and acknowledge the potential bias from the measuring platform.

(5) The manuscript should explain the potential influence of non-fasting status on the findings, particularly concerning lipoprotein particles and composition. There should be a detailed discussion of how non-fasting status may impact the measurement and the findings.

(6) Cross-platform standardization is an issue in metabolism, and further descriptions of quality control are recommended.

Reviewer #2 (Public Review):

Deciphering the metabolic alterations characterizing the prediabetes-diabetes spectrum could provide early time windows for targeted preventive measures to extend precision medicine while avoiding disproportionate healthcare costs. The authors identified a panel of 9 circulating metabolites combined with basic clinical variables that significantly improved the prediction from prediabetes to diabetes. These findings provided insights into the integration of these metabolites into clinical and public health practice. However, the interpretation of these findings should take account of the following limitations.

First, the causal relationship between identified metabolites and diabetes or prediabetes deserves to be further examined particularly when the prediabetic status was partially defined. Some metabolites might be the results of prediabetes rather than the casual factors for progression to diabetes.

Second, the blood samples were taken at random (not all in a non-fasting state) and so the findings were subjected to greater variability. This should be discussed in the limitations.

Third, the strength of NMR in metabolic profiling compared to other techniques (i.e., mass spectrometry [MS], another commonly used metabolic profiling method) could be added in the Discussion section.

Fourth, the applied platform focuses mostly on lipid species which may be a limitation as well.

Fifth, it is a very large group with pre-diabetes, but the results only apply to prediabetes and not to the general population. This should be clear, although the authors have also validated the predictive value of these metabolites in the general population.

eLife assessment

This study provides important findings regarding the stability over time of the response properties of neurons in the auditory cortex, including their nonlinear sensitivity to sound context. The data obtained from chronic recordings combined with nonlinear stimulus-response estimation provide convincing evidence that auditory cortical representations are stable over a period of days to weeks. While this study should be of widespread interest to sensory neuroscientists, the paper would be strengthened by a more thorough assessment and discussion of the effects of context and of the stability of the responses, as well as by the inclusion of more information about the location and types of neurons that were sampled.

Reviewer #1 (Public Review):

Summary:

Recent studies have used optical or electrophysiological techniques to chronically measure receptive field properties of sensory cortical neurons over long time periods, i.e. days to weeks, to ask whether sensory receptive fields are stable properties. Akritas et al expand on prior studies by investigating whether nonlinear contextual sensitivity, a property not previously investigated in the context of so-called 'representational drift,' remains stable over days or weeks of recording. They performed chronic tetrode recordings of auditory cortical neurons over at least five recording days while also performing daily measurements of both the linear spectro-temporal receptive field (principal receptive field, PRF) and non-linear 'contextual gain field' (CGF), which captures the neuron's sensitivity to acoustic context. They found that spike waveforms could be reliably matched even when recorded weeks apart. In well-matched units, by comparing the correlation between tuning within one day's session to sessions across days, both PRFs and CGFs showed remarkable stability over time. This was the case even when recordings were performed over weeks. Meanwhile, behavioral and brain state, measured with locomotion and pupil diameter, respectively, resulted in small but significant shifts in the ability of the PRF/CGF model to predict fluctuations in the neuronal response over time.

Strengths:

The study addresses a fundamental question, which is whether the neural underpinnings of sensory perception, which encompasses both sensory events and their context, are stable across relevant timescales over which our experiences must be stable, despite biological turnover. Although two-photon calcium imaging is ideal for identifying neurons stably regardless of their activity levels and tuning, it lacks temporal precision and is therefore limited in its ability to capture the complexity of sensory responses. Akritas et al performed painstaking chronic extracellular recordings in the auditory cortex with the temporal resolution to investigate complex receptive field properties, such as neural sensitivities to acoustic context. Prior studies, particularly in the auditory cortex, focused on basic tuning properties or sensory responsivity, but Akritas et al expand on this work by showing that even the nonlinear, contextual elements of sensory neurons' responses can remain stable, providing a mechanism for the stability of our complex perception. This work is both novel and broadly applicable to those investigating cortical stability across sensory modalities.

Weaknesses:

Apart from some aspects such as single-unit versus multi-unit, the study largely treats their dataset as a monolith rather than showing how factors such as firing rate, depth, and cell type could define more or less stable subpopulations. It is likely that their methodology did not enable an even sampling over these qualities, and the authors should discuss these biases to put their findings more in context with related studies.

Reviewer #2 (Public Review):

Summary:

This study explores the fundamental neuroscience question of the stability of neuronal representation. The concept of 'representational-drift' has been put forward after observations made using 2-photon imaging of neuronal activity over many days revealed that neurons contribute in a time-limited manner to population representation of stimuli or experiences. The authors contribute to the still contested concept of 'drifts' by measuring representation across days using electrophysiology and thus with sufficient temporal resolution to characterize the receptive fields of neurons in timescales relevant to the stimuli used. The data obtained from chronic recordings over days combined with nonlinear stimulus-response estimation allows the authors to conclude that both the spectrotemporal receptive fields as well as contextual gain fields dependent on combination sensitivity to complex stimuli were stable over time. This suggests that when a neuron is responsive to experimental parameters across long periods of time (days), its sensory receptive field is stable.

Strengths:

The strength of this study lies in the capacity to draw novel conclusions on auditory cortex representation based on the experimentally difficult combination of stable recordings of neuronal activity, behavior, and pupil over days and state-of-the-art analysis of receptive fields.

Weaknesses:

It would have been desirable, but too ambitious in the current setting, to be able to assess what proportion if any of the neurons drop out or in to draw a closer parallel with the 2-photon studies.

Reviewer #3 (Public Review):

Summary:

In their study on "Nonlinear sensitivity to acoustic context is a stable feature of neuronal responses to complex sounds in auditory cortex of awake mice", Akritas et al. investigate the stability of the response properties of neurons in the auditory cortex of mice. They estimate a model with restricted non-linearities for individual neurons and compare the model properties between recordings on the same day and subsequent days. They find that both the linear and nonlinear components of the model stay rather constant over this period and conclude that on the level of the tuning properties, there is no evidence for representational drift on this time scale.

Strengths:

- The study has a clear analytical approach that goes beyond linear models and investigates this in a rigorous way, in particular comparing across-day variability to within-day variability.<br /> - The use of tetrodes is a rather reliable way in electrophysiological recordings to assess neuron identity over multiple days.<br /> - The comparison with pupil and motion activity was useful and insightful.<br /> - The presentation of the study is very logical and pretty much flawless on the writing level.

Weaknesses:

- The stability results across cells show a good amount of variability, which is only partially addressed.<br /> - In particular, no attempt is made to localize the cells in space, in order to check whether these differences could be layer or area-dependent.<br /> - The full context model also includes the possibility to estimate the input non-linearity, which was not done here, but could have been insightful.

Author response:

Reviewer #1 (Public Review):

Summary:

Recent studies have used optical or electrophysiological techniques to chronically measure receptive field properties of sensory cortical neurons over long time periods, i.e. days to weeks, to ask whether sensory receptive fields are stable properties. Akritas et al expand on prior studies by investigating whether nonlinear contextual sensitivity, a property not previously investigated in the context of so-called 'representational drift,' remains stable over days or weeks of recording. They performed chronic tetrode recordings of auditory cortical neurons over at least five recording days while also performing daily measurements of both the linear spectro-temporal receptive field (principal receptive field, PRF) and non-linear 'contextual gain field' (CGF), which captures the neuron's sensitivity to acoustic context. They found that spike waveforms could be reliably matched even when recorded weeks apart. In well-matched units, by comparing the correlation between tuning within one day's session to sessions across days, both PRFs and CGFs showed remarkable stability over time. This was the case even when recordings were performed over weeks. Meanwhile, behavioral and brain state, measured with locomotion and pupil diameter, respectively, resulted in small but significant shifts in the ability of the PRF/CGF model to predict fluctuations in the neuronal response over time.

Strengths:

The study addresses a fundamental question, which is whether the neural underpinnings of sensory perception, which encompasses both sensory events and their context, are stable across relevant timescales over which our experiences must be stable, despite biological turnover. Although two-photon calcium imaging is ideal for identifying neurons stably regardless of their activity levels and tuning, it lacks temporal precision and is therefore limited in its ability to capture the complexity of sensory responses. Akritas et al performed painstaking chronic extracellular recordings in the auditory cortex with the temporal resolution to investigate complex receptive field properties, such as neural sensitivities to acoustic context. Prior studies, particularly in the auditory cortex, focused on basic tuning properties or sensory responsivity, but Akritas et al expand on this work by showing that even the nonlinear, contextual elements of sensory neurons' responses can remain stable, providing a mechanism for the stability of our complex perception. This work is both novel and broadly applicable to those investigating cortical stability across sensory modalities.

Weaknesses:

Apart from some aspects such as single-unit versus multi-unit, the study largely treats their dataset as a monolith rather than showing how factors such as firing rate, depth, and cell type could define more or less stable subpopulations. It is likely that their methodology did not enable an even sampling over these qualities, and the authors should discuss these biases to put their findings more in context with related studies.

We did, in fact, investigate whether firing rate and other physiological response properties of units might differentiate subpopulations with different stability. This analysis is shown in Figure 7B-D. There was no apparent relationship between stability of nonlinear contextual gain fields and physiological properties such as mean evoked firing rate, signal-to-noise ratio for evoked firing, or predictive power of the context model (a measure of model goodness-of-fit).

The reviewer is correct, however, that we did not address possible differences between units recorded at different cortical depths or of different cell types, due to limitations of our methodology and sampling.

Reviewer #2 (Public Review):

Summary:

This study explores the fundamental neuroscience question of the stability of neuronal representation. The concept of 'representational-drift' has been put forward after observations made using 2-photon imaging of neuronal activity over many days revealed that neurons contribute in a time-limited manner to population representation of stimuli or experiences. The authors contribute to the still contested concept of 'drifts' by measuring representation across days using electrophysiology and thus with sufficient temporal resolution to characterize the receptive fields of neurons in timescales relevant to the stimuli used. The data obtained from chronic recordings over days combined with nonlinear stimulus-response estimation allows the authors to conclude that both the spectrotemporal receptive fields as well as contextual gain fields dependent on combination sensitivity to complex stimuli were stable over time. This suggests that when a neuron is responsive to experimental parameters across long periods of time (days), its sensory receptive field is stable.

Strengths:

The strength of this study lies in the capacity to draw novel conclusions on auditory cortex representation based on the experimentally difficult combination of stable recordings of neuronal activity, behavior, and pupil over days and state-of-the-art analysis of receptive fields.

Weaknesses:

It would have been desirable, but too ambitious in the current setting, to be able to assess what proportion if any of the neurons drop out or in to draw a closer parallel with the 2-photon studies.

We certainly agree that this comparison would have been desirable in principle. In practice, however, it was technically infeasible and would have been likely to produce misleading results. Our criteria for spike waveform matching across days were extremely conservative, to minimise the potential for a false positive match (which could artifactually decrease apparent stability of unit responses). Therefore, we were likely to have missed some neurons that did in fact remain active over days, due to small changes in extracellular waveform or just noise (which could artifactually decrease apparent stability of population representations). Two-photon imaging is more appropriate for analysing population stability, because cell identity is determined by spatial location. However, as we mention in the paper, electrophysiology is more appropriate for analysing receptive-field stability, because the temporal resolution is sufficient to resolve structure at the millisecond timescales relevant to auditory perception.

Reviewer #3 (Public Review):

Summary:

In their study on "Nonlinear sensitivity to acoustic context is a stable feature of neuronal responses to complex sounds in auditory cortex of awake mice", Akritas et al. investigate the stability of the response properties of neurons in the auditory cortex of mice. They estimate a model with restricted non-linearities for individual neurons and compare the model properties between recordings on the same day and subsequent days. They find that both the linear and nonlinear components of the model stay rather constant over this period and conclude that on the level of the tuning properties, there is no evidence for representational drift on this time scale.

Strengths:

- The study has a clear analytical approach that goes beyond linear models and investigates this in a rigorous way, in particular comparing across-day variability to within-day variability.

- The use of tetrodes is a rather reliable way in electrophysiological recordings to assess neuron identity over multiple days.

- The comparison with pupil and motion activity was useful and insightful.

- The presentation of the study is very logical and pretty much flawless on the writing level.

Weaknesses:

- The stability results across cells show a good amount of variability, which is only partially addressed.

- In particular, no attempt is made to localize the cells in space, in order to check whether these differences could be layer or area-dependent.

- The full context model also includes the possibility to estimate the input non-linearity, which was not done here, but could have been insightful.

We agree with these comments and acknowledge these limitations, which arise from technological constraints. In particular, the tangential trajectory of our chronic tetrode implant, used to maximise stability of chronic recordings, limited our ability to sample cells from different cortical layers/areas and to explore how these factors might relate to variability in stability across units. Estimating input nonlinearities would have been valuable but also would have increased the number of parameters in the model and the data required to obtain reliable, predictive model fits.

eLife assessment

This study shows that a peptide called galanin can decrease or increase seizure activity in experimental models of seizures depending on the way seizures are induced (genetic vs. pharmacological). The authors use zebrafish and several methods to address the effects of galanin. The study will be useful to researchers who use zebrafish as experimental animals and who are interested in how the peptides in the brain (neuropeptides) regulate seizures. However, the strength of evidence was considered incomplete at the present time due to several limitations of the results.

Reviewer #1 (Public Review):

Summary:

In this study, the 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. The 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:

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.

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).

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.

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 for 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.

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.

Reviewer #2 (Public Review):

Summary:

This study is 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 or seizure event conditions. The authors' Eaat2a-/- model (introduced in their Glia 2022 paper, PMID 34716961) that shows recurrent seizure activity alongside suppression of neuronal activity and locomotion in the time periods lacking seizures is used in this paper in comparison to the well-known pentylenetetrazole (PTZ) pharmacological model of epilepsy in zebrafish. Given the literature cited in their Introduction, the authors reasonably hypothesize that galanin will exert a net inhibitory effect on brain activity in models of epilepsy and at homeostatic baseline, but were surprised to find that this hypothesis was only moderately supported in their Eaat2a-/- model. In contrast, under PTZ challenge, fish with galanin overexpression showed increased seizure number and reduced duration while fish with galanin KO showed reduced seizure number and increased duration. These results would have been greatly enriched by the inclusion of behavioral analyses of seizure activity and locomotion (similar to the authors' 2022 Glia paper and/or PMIDs 15730879, 24002024). In addition, the authors have not accounted for sex as a biological variable, though they did note that sex sorting zebrafish larvae precludes sex selection at the younger ages used. It would be helpful to include smaller experiments taken from pilot experiments in older, sex-balanced groups of the relevant zebrafish to increase confidence in the findings' robustness across sexes. A possible major caveat is that all of the various genetic manipulations are non-conditional as performed, meaning that developmental impacts of galanin overexpression or galanin or galr1a knockout on the observed results have not been controlled for and may have had a confounding influence on the authors' findings. Overall, this study is important and solid (yet limited), and carries clear value for understanding the multifaceted functions that neuronal galanin can have under homeostatic and disease conditions.

Strengths:

- The authors convincingly show that galanin is upregulated across multiple contexts that feature seizure activity or hyperexcitability in zebrafish, and appears to reduce neuronal activity overall, with key identified exceptions (PTZ model).

- The authors use both genetic and pharmacological models to answer their question, and through this diverse approach, find serendipitous results that suggest novel underexplored functions of galanin and its receptors in basal and disease conditions. Their question is well-informed by the cited literature, though the authors should cite and consider their findings in the context of Mazarati et al., 1998 (PMID:982276). The authors' Discussion places their findings in context, allowing for multiple interpretations and suggesting some convincing explanations.

- Sample sizes are robust and the methods used are well-characterized, with a few exceptions (as the paper is currently written).

- Use of a glutamatergic signaling-based genetic model of epilepsy (Eaat2a-/-) is likely the most appropriate selection to test how galanin signaling can alter seizure activity, as galanin is known to reduce glutamatergic release as an inhibitory mechanism in rodent hippocampal neurons via GalR1a (alongside GIRK activation effects). Given that PTZ instead acts through GABAergic signaling pathways, it is reasonable and useful to note that their glutamate-based genetic model showed different effects than did their GABAergic-based model of seizure activity.

Weaknesses:

- The authors do not include behavioral assessments of seizure or locomotor activity that would be expected in this paper given their characterizations of their Eaat2a-/- model in the Glia 2022 paper that showed these behavioral data for this zebrafish model. These data would inform the reader of the behavioral phenotypes to expect under the various conditions and would likely further support the authors' findings if obtained and reported.

- No assessment of sex as a biological variable is included, though it is understood that these specific studied ages of the larvae may preclude sex sorting for experimental balancing as stated by the authors.

- The reported results may have been influenced by the loss or overexpression of galanin or loss of galr1a during developmental stages. The authors did attempt to use the hsp70l system to overexpress galanin, but noted that the heat shock induction step led to reduced brain activity on its own (Supplementary Figure 1). Their hsp70l:gal model shows galanin overexpression anyways (8x fold) regardless of heat induction, so this model is still useful as a way to overexpress galanin, but it should be noted that this galanin overexpression is not restricted to post-developmental timepoints and is present during development.

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 has provided convincing evidence for the 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 a 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. The authors also used a heat shock protein line (hsp70I:gal) where galanin transcript levels are induced by activation of heat shock protein, but this line also shows higher basal transcript levels of galanin. Again, the higher level of galanin in hsp70I:gal larval zebrafish resulted in a reduction of calcium events and a reduction in the amplitude of events. In contrast, galanin knockout (gal-/-) increased calcium activity, indicated by an increased number of calcium events, but a reduction in amplitude and duration. Knockout of the galanin receptor subtype galr1a via crispants also increased the frequency of calcium events.

In subsequent experiments in eaat2a-/- mutants were crossed with hsp70I:gal or gal-/- to increase or decrease galanin expression, respectively. 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 an increased number of seizures and reduced seizure duration during PTZ. In contrast, gal-/- mutants showed an increased normalized area under the curve and a stark reduction in the number of detected seizures, a reduction in seizure amplitude, but an increase in seizure duration. The authors then ruled out the role of Galr1a in modulating this effect during PTZ, since the number of seizures was unaffected, whereas the amplitude and duration of seizures were increased.

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. In particular, the relationship between galanin transcript levels and brain activity in Figures 1 & 2 was convincing.

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

(3) Focus on the galanin receptor subtype galr1a provided good evidence for the important role of this receptor in controlling brain activity during interictal and/or seizure-free periods.

Weaknesses:

(1) Although the relationship between galanin and brain activity during interictal or seizure-free periods was clear, the manuscript currently lacks mechanistic insight in the role of galanin during seizure-like activity induced by PTZ.

(2) Calcium imaging is the primary data for the paper, but there are no representative time-series images or movies of GCaMP signal in the various mutants used.

(3) For Figure 3, the authors suggest that hsp70I:gal x eaat2a-/-mutants would further increase galanin transcript levels, which were hypothesized to further reduce brain activity. However, the authors failed to measure galanin transcript levels in this cross to show that galanin is actually increased more than the eaat2a-/- mutant or the hsp70I:gal mutant alone.

(4) Similarly, transcript levels of galanin are not provided in Figure 2 for Gal-/- mutants and galr1a KOs. Transcript levels would help validate the knockout and any potential compensatory effects of subtype-specific knockout.

(5) The authors very heavily rely on calcium imaging of different mutant lines. 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).

Author response:

Reviewer #1 (Public Review):

Summary:

In this study, the 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. The 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:

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.

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).

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.

Indeed, our observation of the unexpected hypoactivity in EAAT2a mutants, described in our description of this mutant (Hotz et al., 2022), prompted us to initiate this study formulating the hypothesis that the observed upregulation of galanin is a neuroprotective response to epilepsy.

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 have performed a transcriptome analysis that we are still evaluation. We can already state that AMPA receptor genes are not significantly altered in the mutant.

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 for 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.

We agree that upregulation of galanin transcripts is at best one of a suite of regulatory mechanisms that lead to hypoactivity in EAAT2 zebrafish mutants.

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.

Reviewer #2 (Public Review):

Summary:

This study is 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 or seizure event conditions. The authors' Eaat2a-/- model (introduced in their Glia 2022 paper, PMID 34716961) that shows recurrent seizure activity alongside suppression of neuronal activity and locomotion in the time periods lacking seizures is used in this paper in comparison to the well-known pentylenetetrazole (PTZ) pharmacological model of epilepsy in zebrafish. Given the literature cited in their Introduction, the authors reasonably hypothesize that galanin will exert a net inhibitory effect on brain activity in models of epilepsy and at homeostatic baseline, but were surprised to find that this hypothesis was only moderately supported in their Eaat2a-/- model. In contrast, under PTZ challenge, fish with galanin overexpression showed increased seizure number and reduced duration while fish with galanin KO showed reduced seizure number and increased duration. These results would have been greatly enriched by the inclusion of behavioral analyses of seizure activity and locomotion (similar to the authors' 2022 Glia paper and/or PMIDs 15730879, 24002024). In addition, the authors have not accounted for sex as a biological variable, though they did note that sex sorting zebrafish larvae precludes sex selection at the younger ages used. It would be helpful to include smaller experiments taken from pilot experiments in older, sex-balanced groups of the relevant zebrafish to increase confidence in the findings' robustness across sexes. A possible major caveat is that all of the various genetic manipulations are non-conditional as performed, meaning that developmental impacts of galanin overexpression or galanin or galr1a knockout on the observed results have not been controlled for and may have had a confounding influence on the authors' findings. Overall, this study is important and solid (yet limited), and carries clear value for understanding the multifaceted functions that neuronal galanin can have under homeostatic and disease conditions.

Strengths:

- The authors convincingly show that galanin is upregulated across multiple contexts that feature seizure activity or hyperexcitability in zebrafish, and appears to reduce neuronal activity overall, with key identified exceptions (PTZ model).

- The authors use both genetic and pharmacological models to answer their question, and through this diverse approach, find serendipitous results that suggest novel underexplored functions of galanin and its receptors in basal and disease conditions. Their question is well-informed by the cited literature, though the authors should cite and consider their findings in the context of Mazarati et al., 1998 (PMID:982276). The authors' Discussion places their findings in context, allowing for multiple interpretations and suggesting some convincing explanations.

- Sample sizes are robust and the methods used are well-characterized, with a few exceptions (as the paper is currently written).

- Use of a glutamatergic signaling-based genetic model of epilepsy (Eaat2a-/-) is likely the most appropriate selection to test how galanin signaling can alter seizure activity, as galanin is known to reduce glutamatergic release as an inhibitory mechanism in rodent hippocampal neurons via GalR1a (alongside GIRK activation effects). Given that PTZ instead acts through GABAergic signaling pathways, it is reasonable and useful to note that their glutamate-based genetic model showed different effects than did their GABAergic-based model of seizure activity.

Weaknesses:

- The authors do not include behavioral assessments of seizure or locomotor activity that would be expected in this paper given their characterizations of their Eaat2a-/- model in the Glia 2022 paper that showed these behavioral data for this zebrafish model. These data would inform the reader of the behavioral phenotypes to expect under the various conditions and would likely further support the authors' findings if obtained and reported.

We agree that a thorough behavioral assessment would have strengthened the study, but we deemed it outside of the scope of this study.

- No assessment of sex as a biological variable is included, though it is understood that these specific studied ages of the larvae may preclude sex sorting for experimental balancing as stated by the authors.

The study was done on larval zebrafish (5 days post fertilization). The first signs of sexual differentiation become apparent at about 17 days post fertilization (reviewed in Ye and Chen, 2020). Hence sex is no biological variable at the stage studied. 

- The reported results may have been influenced by the loss or overexpression of galanin or loss of galr1a during developmental stages. The authors did attempt to use the hsp70l system to overexpress galanin, but noted that the heat shock induction step led to reduced brain activity on its own (Supplementary Figure 1). Their hsp70l:gal model shows galanin overexpression anyways (8x fold) regardless of heat induction, so this model is still useful as a way to overexpress galanin, but it should be noted that this galanin overexpression is not restricted to post-developmental timepoints and is present during development.

The developmental perspective is an important point to consider. Due to the rapid development of the zebrafish it is not trivial to untangle this. In the zebrafish we first observe epileptic seizures as early as 3 days post fertilization (dpf), where the brain is clearly not well developed yet (e.g. behavioral response to light are still minimal). Even the 5 dpf stage, where most of our experiments have been conducted, cannot by far not be considered post-development.  

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 has provided convincing evidence for the 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 a 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. The authors also used a heat shock protein line (hsp70I:gal) where galanin transcript levels are induced by activation of heat shock protein, but this line also shows higher basal transcript levels of galanin. Again, the higher level of galanin in hsp70I:gal larval zebrafish resulted in a reduction of calcium events and a reduction in the amplitude of events. In contrast, galanin knockout (gal-/-) increased calcium activity, indicated by an increased number of calcium events, but a reduction in amplitude and duration. Knockout of the galanin receptor subtype galr1a via crispants also increased the frequency of calcium events.

In subsequent experiments in eaat2a-/- mutants were crossed with hsp70I:gal or gal-/- to increase or decrease galanin expression, respectively. 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 an increased number of seizures and reduced seizure duration during PTZ. In contrast, gal-/- mutants showed an increased normalized area under the curve and a stark reduction in the number of detected seizures, a reduction in seizure amplitude, but an increase in seizure duration. The authors then ruled out the role of Galr1a in modulating this effect during PTZ, since the number of seizures was unaffected, whereas the amplitude and duration of seizures were increased.

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. In particular, the relationship between galanin transcript levels and brain activity in Figures 1 & 2 was convincing.

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

(3) Focus on the galanin receptor subtype galr1a provided good evidence for the important role of this receptor in controlling brain activity during interictal and/or seizure-free periods.

Weaknesses:

(1) Although the relationship between galanin and brain activity during interictal or seizure-free periods was clear, the manuscript currently lacks mechanistic insight in the role of galanin during seizure-like activity induced by PTZ.

We completely agree and concede that this study constitutes only a first attempt to understand the (at least for us) perplexing complexity of galanin function on the brain.

(2) Calcium imaging is the primary data for the paper, but there are no representative time-series images or movies of GCaMP signal in the various mutants used.

We are in the process of preparing some time series images and will include them in the next revision.

(3) For Figure 3, the authors suggest that hsp70I:gal x eaat2a-/-mutants would further increase galanin transcript levels, which were hypothesized to further reduce brain activity. However, the authors failed to measure galanin transcript levels in this cross to show that galanin is actually increased more than the eaat2a-/- mutant or the hsp70I:gal mutant alone.

This is an excellent suggestion. We will perform the necessary qPCR experiments and will include the data in the next revision.

(4) Similarly, transcript levels of galanin are not provided in Figure 2 for Gal-/- mutants and galr1a KOs. Transcript levels would help validate the knockout and any potential compensatory effects of subtype-specific knockout.

(5) The authors very heavily rely on calcium imaging of different mutant lines. 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).

Again, we agree and concede that a number of additional approaches are needed to get more insight into the complex role of galanin in regulation overall brain activity. These include, among others, also behavioral, multiple single cell recordings and pharmacological interventions.

eLife assessment

The work introduces a valuable new method for depleting the ribosomal RNA from bacterial single-cell RNA sequencing libraries and shows that this method is applicable to studying the heterogeneity in microbial biofilms. The evidence for a small subpopulation of cells at the bottom of the biofilm which upregulates PdeI expression is solid. However, more investigation into the unresolved functional relationship between PdeI and c-di-GMP levels with the help of other genes co-expressed in the same cluster would have made the conclusions more significant.

Reviewer #1 (Public Review):

Summary:

In this manuscript, Yan and colleagues introduce a modification to the previously published PETRI-seq bacterial single-cell protocol to include a ribosomal depletion step based on a DNA probe set that selectively hybridizes with ribosome-derived (rRNA) cDNA fragments. They show that their modification of the PETRI-seq protocol increases the fraction of informative non-rRNA reads from ~4-10% to 54-92%. The authors apply their protocol to investigating heterogeneity in a biofilm model of E. coli, and convincingly show how their technology can detect minority subpopulations within a complex community.

Strengths:

The method the authors propose is a straightforward and inexpensive modification of an established split-pool single-cell RNA-seq protocol that greatly increases its utility, and should be of interest to a wide community working in the field of bacterial single-cell RNA-seq.

Weaknesses:

The manuscript is written in a very compressed style and many technical details of the evaluations conducted are unclear and processed data has not been made available for evaluation, limiting the ability of the reader to independently judge the merits of the method.

Reviewer #2 (Public Review):

Summary:

This work introduces a new method of depleting the ribosomal reads from the single-cell RNA sequencing library prepared with one of the prokaryotic scRNA-seq techniques, PETRI-seq. The advance is very useful since it allows broader access to the technology by lowering the cost of sequencing. It also allows more transcript recovery with fewer sequencing reads. The authors demonstrate the utility and performance of the method for three different model species and find a subpopulation of cells in the E.coli biofilm that express a protein, PdeI, which causes elevated c-di-GMP levels. These cells were shown to be in a state that promotes persister formation in response to ampicillin treatment.

Strengths:

The introduced rRNA depletion method is highly efficient, with the depletion for E.coli resulting in over 90% of reads containing mRNA. The method is ready to use with existing PETRI-seq libraries which is a large advantage, given that no other rRNA depletion methods were published for split-pool bacterial scRNA-seq methods. Therefore, the value of the method for the field is high. There is also evidence that a small number of cells at the bottom of a static biofilm express PdeI which is causing the elevated c-di-GMP levels that are associated with persister formation. Given that PdeI is a phosphodiesterase, which is supposed to promote hydrolysis of c-di-GMP, this finding is unexpected.

Weaknesses:

With the descriptions and writing of the manuscript, it is hard to place the findings about the PdeI into existing context (i.e. it is well known that c-di-GMP is involved in biofilm development and is heterogeneously distributed in several species' biofilms; it is also known that E.coli diesterases regulate this second messenger, i.e. https://journals.asm.org/doi/full/10.1128/jb.00604-15).<br /> There is also no explanation for the apparently contradictory upregulation of c-di-GMP in cells expressing higher PdeI levels. Perhaps the examination of the rest of the genes in cluster 2 of the biofilm sample could be useful to explain the observed association.

eLife assessment

This descriptive study reports the genetic requirements for growth and fitness of multiple clinical strains of a relatively understudied species of mycobacteria, Mycobacterium intracellulare. The findings are valuable however, the study is incomplete as the primary claims related to hypoxia adaptation need additional experimental support and data presentation requires more clarity. The work will be of interest to microbiologists.

Reviewer #1 (Public Review):

Summary:

Tateishi et al. report a Tn-seq-based analysis of genetic requirements for growth and fitness in 8 clinical strains of Mycobacterium intracellulare Mi), and compare the findings with a type strain ATCC13950. The study finds a core set of 131 genes that are essential in all nine strains, and therefore are reasonably argued as potential drug targets. Multiple other genes required for fitness in clinical isolates have been found to be important for hypoxic growth in the type strain.

Strengths:

The study has generated a large volume of Tn-seq datasets of multiple clinical strains of Mi from multiple growth conditions, including from mouse lungs. The dataset can serve as an important resource for future studies on Mi, which despite being clinically significant remains a relatively understudied species of mycobacteria.

Weaknesses:

The paper lacks clarity in data presentation and organization. For example, some of the key data on cfu counts of clinical Mi strains in a mouse model can be presented along with the Tn-seq dataset in Figure 6, the visualization of which can be improved with volcano plots. etc. Improvement in data visualization is perhaps necessary throughout the paper.

The primary claim of the study that the clinical strains are better adapted for hypoxic growth is not well-supported by the data presented in Figure 7.

The title of the paper is misleading as the study doesn't provide any mechanistic aspect of hypoxic adaptation in Mi.

Reviewer #2 (Public Review):

Summary:

In the study titled "Functional genomics reveals the mechanism of hypoxic adaptation in nontuberculous mycobacteria" by Tateishi et al., the authors have used TnSeq to identify the common essential and growth-defect-associated genes that represent the genomic diversity of clinical M. intracellulare strains in comparison to the reference type strain. By estimating the frequency of Tn insertion, the authors speculate that genes involved in gluconeogenesis, the type VII secretion system, and cysteine desulfurase are relatively critical in the clinical MAC-PD strains than in the type strain, both for the extracellular survival and in a mouse lung infection model.

Based on their analysis, the authors proposed to identify the mechanism of hypoxic adaptation in nontuberculous mycobacteria (NTM) which offer promising drug targets in the strains causing clinical Mycobacterium avium-intracellulare complex pulmonary disease (MAC-PD).

Strengths:

A major strength of the manuscript is the performance of the exhaustive set of TnSeq experiments with multiple strains of M. intracellulare during in vitro growth and animal infection.

Weaknesses:

(1) The study suffers from the authors' preconceived bias toward a small subset of genes involved in hypoxic pellicle formation in ATCC13950.

(2) An important set of data with the ATCC13950 reference strain is missing in the mouse infection study. In the absence of this, it is difficult to establish whether the identified genes are critical for infection/intracellular proliferation, specifically in the clinical isolates that are relatively more adapted for hypoxia.

(3) Statistical enrichment analysis of gene sets by GSEA wrongly involves genes required for hypoxic pellicle formation in ATCC13950 together with the gene sets found essential in the clinical MAC-PD strains, to claim that a significant % of genes belong to hypoxia-adaptation pathways. It could be factually incorrect because a majority of these might overlap with those found critical for the in vitro survival of MAC-PD strains (and may not be related to hypoxia).

(4) Validation of mouse infection experiments with individual mutants is missing.

(5) Phenotypes with TnSeq and CRISPRi-based KD exhibit poor correlation with misleading justifications by the authors.

In summary, this study is unable to provide mechanistic insights into why and how different MAC-PD mutant strains exhibit differential survival (in vitro and in animals) and adaptation to hypoxia. It remains to understand why the clinical strains show better adaptation to hypoxia and what is the impact of other stresses on their growth rates.

Reviewer #3 (Public Review):

Summary:

The study by Tateishi et al. utilized TnSeq in nine genetically diverse M. intracellulare strains, identifying 131 common essential and growth-defect-associated genes across those strains, which could serve as potential drug targets. The authors also provided an overview of the differences in gene essentiality required for hypoxic growth between the reference strain and the clinical strains. Furthermore, they validated the universal and accessory/strain-dependent essential genes by knocking down their expression using CRISPRi technique. Overall, this study offers a comprehensive assessment of gene requirements in different clinical strains of M. intracellular.

(1) The rationale for using ATCC13950 versus clinical strains needs to be clarified. The reference strain ATCC13950 was obtained from the abdominal lymph node of a patient around 10 years ago and is therefore considered a clinical strain that has undergone passages in vitro. How many mutations have accumulated during these in vitro passages? Are these mutations significant enough to cause the behavior of ATCC13950 to differ from other recently sampled clinical strains? From the phylogenetic tree, ATCC13950 is located between M018 and M.i.27. Did the authors observe a similarity in gene essentiality between ATCC13950 and its neighbor strains? What is the key feature that separates ATCC13950 from these clinical strains? The authors should provide a strong rationale for how to interpret the results of this comparison in a clinical or biological context.

(2) Regarding the 'nine representative strains of M. intracellulare with diverse genotypes in this study,' how were these nine strains selected? To what extent do they represent the genetic diversity of the M. intracellulare population? A phylogenetic tree illustrating the global genetic diversity of the M. intracellulare population, with these strains marked on it, would be important to demonstrate their genetic representativeness.

(3) The authors observed a considerable amount of differential gene requirements in clinical strains. However, the genetic underpinning underlying the differential requirement of genes in clinical strains was not investigated or discussed. Because M. intracellulare has a huge number of accessory genes, the authors should at least check whether the differential requirement could be explained by the existence of a second copy of functional analogous genes or duplications.

(4) Growth in aerobic and hypoxic conditions: The authors concluded that clinical strains are better adapted to hypoxia, as reflected by their earlier entry into the log phase. They presented the 'Time at midpoint' and 'Growth rate at midpoint.' However, after reviewing the growth curves, I noticed that ATCC13950 had a longer lag phase compared to other strains under hypoxic conditions, and its phylogenetic neighbor M018 also had a longer lag phase. Hence, I do not believe a conclusion can be drawn that clinical strains are better adapted to hypoxia, as this behavior could be specific to a particular clade. It's also possible that the ATCC13950 strain has adapted to aerobic growth. I would suggest that the authors include growth curves in the main figures. The difference in 'Time at midpoint' could be attributed to several factors, and visualizing the growth curves would provide additional context and clarity.

(5) Lack of statistical statement: The authors emphasized the role of pellicle-formation-associated genes in strain-dependent essential and accessory essential genes. Additionally, the authors observed that 10% of the genes required for mouse infection are also required for hypoxic pellicle formation. However, these are merely descriptive statements. There is no enrichment analysis to justify whether pellicle-formation-associated genes are significantly enriched in these groups.

eLife assessment

This important study shows the effect of gut dysbiosis on the colonization of mycobacteria in the lung. The data with comprehensive analysis of gene expression profiles in the lung with dysbiotic mice is compelling and goes beyond the current state of the art. However, the mechanistic insight and the experiments with Mtb infection are incomplete. With those parts strengthened, this paper would be of interest to researchers working on Mtb infection.

Reviewer #1 (Public Review):

Summary:

This work sought to demonstrate that gut microbiota dysbiosis may promote the colonization of mycobacteria, and they tried to prove that Nos2 down-regulation was a key mediator of such gut-lung pathogenesis transition.

Strengths:

They did large-scale analysis of RNAs in lungs to analyze the gene expression of mice upon gut dysbiosis in MS-infected mice. This might help provide an overview of gene pathways and critical genes for lung pathology in gut dysbiosis. This data is somewhat useful and important for the TB field.

Weaknesses:

(1) They did not use wide-type Mtb strain (e.g. H37Rv) to develop mouse TB infection models, and this may lead to the failure of the establishment of TB granuloma and other TB pathology icons.

(2) The usage of in vitro assays based on A542 to examine the regulation function of Nos2 expression on NO and ROS may not be enough. A542 is not the primary Mtb infection target in the lungs.

(3) They did not examine the lung pathology upon gut dysbiosis to examine the true significance of increased colonization of Mtb.

(4) Most of the studies are based on MS-infected mouse models with a lack of clinical significance.

Reviewer #2 (Public Review):

The manuscript entitled "Intestinal microbiome dysbiosis increases Mycobacteria pulmonary colonization in mice by regulating the Nos2-associated pathways" by Han et al reported that using clindamycin, an antibiotic to selectively disorder anaerobic Bacteriodetes, intestinal microbiome dysbiosis resulted in Mycobacterium smegmatis (MS) colonization in the mice lungs. The authors found that clindamycin induced damage of the enterocytes and gut permeability and also enhanced the fermentation of cecum contents, which finally increased MS colonization in the mice's lungs. The study showed that gut microbiota dysbiosis up-regulated the Nos2 gene-associated pathways, leading to increased nitric oxide (NO) levels and decreased reactive oxygen species (ROS) and β-defensin 1 (Defb1) levels. These changes in the host's immune response created an antimicrobial and anti-inflammatory environment that favored MS colonization in the lungs. The findings suggest that gut microbiota dysbiosis can modulate the host's immune response and increase susceptibility to pulmonary infections by altering the expression of key genes and pathways involved in innate immunity. The authors reasonably provided experimental data and subsequent gene profiles to support their conclusion. Although the overall outcomes are convincing, there are several issues that need to be addressed:

(1) In Figure S1, the reviewer suggests checking the image sizes of the pathological sections of intestinal tissue from the control group and the CL-treatment group. When compared to the same intestinal tissue images in Figure S4, they do not appear to be consistently magnified at 40x. The numerical scale bars should be presented instead of just magnification such as "40x".

(2) In Figure 4d, the ratio of Firmicutes in the CL-FMT group decreased compared to the CON-FMT group, whereas the CL-treatment group showed an increase in Firmicutes compared to the Control group in Figure 3b. The author should explain this discrepancy and discuss its potential implications on the study's findings.

(3) In Figure 6, did the authors have a specific reason for selecting Nos2 but not Tnf for further investigation? The expression level of the Tnf gene appears to be the most significant in both RT-qPCR and RNA-sequencing results in Figure 5f. Tnf is an important cytokine involved in immune responses to bacterial infections, so it is also a factor that can influence NO, ROS, and Defb1 levels.

eLife assessment

The manuscript by Carbo et al. reports a novel role for the MltG homolog AgmT in gliding motility in M. xanthus. The authors provide convincing data to demonstrate that AgmT is a cell wall lytic enzyme (likely a lytic transglycosylase), its lytic activity is required for gliding motility, and that its activity is required for proper binding of a component of the motility apparatus to the cell wall. The findings are valuable as they contribute to our understanding of the molecular mechanisms underlying the interaction between gliding motility and the bacterial cell wall.

Reviewer #1 (Public Review):

Summary:

This manuscript nicely outlines a conceptual problem with the bFAC model in A-motility, namely, how is the energy produced by the inner membrane AglRQS motor transduced through the cell wall into mechanical force on the cell surface to drive motility? To address this, the authors make a significant contribution by identifying and characterizing a lytic transglycosylase (LTG) called AgmT. This work thus provides clues and a future framework work for addressing mechanical force transmission between the cytoplasm and the cell surface.

Strengths:

(1) Convincing evidence shows AgmT functions as an LTG and, surprisingly, that mltG from E. coli complements the swarming defect of an agmT mutant.

(2) Authors show agmT mutants develop morphological changes in response to treatment with a -lactam antibiotic, mecillinam.

(3) The use of single-molecule tracking to monitor the assembly and dynamics of bFACs in WT and mutant backgrounds.

(4) The authors understand the limitations of their work and do not overinterpret their data.

Weaknesses:

(1) A clear model of AgmT's role in gliding motility or interactions with other A-motility proteins is not provided. Instead, speculative roles for how AgmT enzymatic activity could facilitate bFAC function in A-motility are discussed.

(2) Although agmT mutants do not swarm, in-depth phenotypic analysis is lacking. In particular, do individual agmT mutant cells move, as found with other swarming defective mutants, or are agmT mutants completely nonmotile, as are motor mutants?

(3) The bioinformatic and comparative genomics analysis of agmT is incomplete. For example, the sequence relationships between AgmT, MltG, and the 13 other LTG proteins in M. xanthus are not clear. Is E. coli MltG the closest homology to AgmT? Their relationships could be addressed with a phylogenetic tree and/or sequence alignments. Furthermore, are there other A-motility genes in proximity to agmT? Similarly, does agmT show specific co-occurrences with the other A-motility genes across genera/species?

(4) Related to iii, what about the functional relationship of the endogenous 13 LTG genes? Although knockout mutants were shown to be motile, presumably because AgmT is present, can overexpression of them, similar to E. coli MltG, complement an agmT mutant? In other words, why does MltG complement and the endogenous LTG proteins appear not to be relevant?

(5) Based on Figure 2B, overexpression of MltG enhances A-motility compared to the parent strain and the agmT-PAmCh complemented strain, is this actually true? Showing expanded swarming colony phenotypes would help address this question.

(6) Cell flexibility is correlated with gliding motility function in M. xanthus. Since AgmT has LTG activity, are agmT mutants less flexible than WT cells and is this the cause of their motility defect?

Reviewer #2 (Public Review):

The manuscript by Carbo et al. reports a novel role for the MltG homolog AgmT in gliding motility in M. xanthus. The authors conclusively show that AgmT is a cell wall lytic enzyme (likely a lytic transglycosylase), its lytic activity is required for gliding motility, and that its activity is required for proper binding of a component of the motility apparatus to the cell wall. The data are generally well-controlled. The marked strength of the manuscript includes the detailed characterization of AgmT as a cell wall lytic enzyme, and the careful dissection of its role in motility. Using multiple lines of evidence, the authors conclusively show that AgmT does not directly associate with the motility complexes, but that instead its absence (or the overexpression of its active site mutant) results in the failure of focal adhesion complexes to properly interact with the cell wall.

An interpretive weakness is the rather direct role attributed to AgmT in focal adhesion assembly. While their data clearly show that AgmT is important, it is unclear whether this is the direct consequence of AgmT somehow promoting bFAC binding to PG or just an indirect consequence of changed cell wall architecture without AgmT. In E. coli, an MltG mutant has increased PG strain length, suggesting that M. xanthus's PG architecture may likewise be compromised in a way that precludes AglR binding to the cell wall. However, this distinction would be very difficult to establish experimentally. MltG has been shown to associate with active cell wall synthesis in E.c oli in the absence of protein-protein interactions, and one could envision a similar model in M. xanthus, where active cell wall synthesis is required for focal adhesion assembly, and MltG makes an important contribution to this process.

eLife assessment

This important study demonstrates a potential mechanism by which adjuvants influence T-cell responses. The observation that adjuvant impacts the exogenous peptide repertoire presented by MHC II molecules is fascinating and the strength of the evidence is solid, with studies comparing different adjuvants and an H pylori vaccine in murine models and in vitro systems, analysis of MHCII: peptide complexes in antigen-presenting cells and assessment of differential peptide binding affinities. This work will be of broad interest to vaccinologists as well as immunologists.

Reviewer #1 (Public Review):

Summary:

Li et al investigated how adjuvants such as MPLA and CpG influence antigen presentation at the level of the Antigen-presenting cell and MHCII : peptide interaction. They found that the use of MPLA or CpG influences the exogenous peptide repertoire presented by MHC II molecules. Additionally, their observations included the finding that peptides with low-stability peptide:MHC interactions yielded more robust CD4+ T cell responses in mice. These phenomena were illustrated specifically for 2 pattern recognition receptor activating adjuvants. This work represents a step forward for how adjuvants program CD4+ Th responses and provides further evidence regarding the expected mechanisms of PRR adjuvants in enhancing CD4+ T cell responses in the setting of vaccination.

Strengths:

The authors use a variety of systems to analyze this question. Initial observations were collected in an H pylori model of vaccination with a demonstration of immunodominance differences simply by adjuvant type, followed by analysis of MHC:peptide as well as proteomic analysis with comparison by adjuvant group. Their analysis returns to peptide immunization and analysis of strength of relative CD4+ T cell responses, through calculation of IC:50 values and strength of binding. This is a comprehensive work. The logical sequence of experiments makes sense and follows an unexpected observation through to trying to understand that process further with peptide immunization and its impact on Th responses. This work will premise further studies into the mechanisms of adjuvants on T cells

Weaknesses:

While MDP has a different manner of interaction as an adjuvant compared to CpG and MPLA, it is unclear why MDP has a different impact on peptide presentation and it should be further investigated, or at minimum highlighted in the discussion as an area that requires further investigation.

It is alluded by the authors that TLR activating adjuvants mediate selective, low affinity, exogenous peptide binding onto MHC class II molecules. However, this was not demonstrated to be related specifically to TLR binding. I wonder if some work with TLR deficient mice (TLR 4KO for example) could evaluate this phenomenon more specifically.

It is unclear to me if this observation is H pylori model/antigen-specific. It may have been nice to characterize the phenomenon with a different set of antigens as supplemental. Lastly, it is unclear if the peptide immunization experiment reveals a clear pattern related to high and low-stability peptides among the peptides analyzed.

Reviewer #2 (Public Review):

Adjuvants boost antigen-specific immune responses to vaccines. However, whether adjuvants modulate the epitope immunodominance and the mechanisms involved in adjuvant's effect on antigen processing and presentation are not fully characterized. In this manuscript, Li et al report that immunodominant epitopes recognized by antigen-specific T cells are altered by adjuvants.

Using MPLA, CpG, and MDP adjuvants and H. pylori antigens, the authors screened the dominant epitopes of Th1 responses in mice post-vaccination with different adjuvants and found that adjuvants altered antigen-specific CD4+ T cell immunodominant epitope hierarchy. They show that adjuvants, MPLA and CpG especially, modulate the peptide repertoires presented on the surface of APCs. Surprisingly, adjuvant favored the presentation of low-stability peptides rather than high-stability peptides by APCs. As a result, the low stability peptide presented in adjuvant groups elicits T cell response effectively.

eLife assessment

This study provides valuable insights into how IL-1 cytokines may protect cells against SARS-COV-2 infection. By inducing a non-canonical RhoA/ROCK signaling pathway, IL-1beta appears to inhibit the ability of SARS-COV-2 infected cells to fuse with uninfected cells and produce syncytia. The evidence underlying the identification of the key signaling components required for this inhibitory phenotype in vitro is solid and could be further improved by addressing key weaknesses. However, data supporting this specific mechanism of inhibition in IL-1-mediated control of SARS-COV-2 infection in vivo remains incomplete.

Reviewer #1 (Public Review):

Summary:

SARS-CoV-2 infection induces syncytia formation, which promotes viral transmission. In this paper, the authors aimed to understand how host-derived inflammatory cytokines IL-1α/β combat SARS-CoV-2 infection.

Strengths:

First, they used a cell-cell fusion assay developed previously to identify IL-1α/β as the cytokines that inhibit syncytia formation. They co-cultured cells expressing the spike protein and cells expressing ACE2 and found that IL-1β treatment decreased syncytia formation and S2' cleavage.

Second, they investigated the IL-1 signaling pathway in detail, using knockouts or pharmacological perturbation to understand the signaling proteins responsible for blocking cell fusion. They found that IL-1 prevents cell-cell fusion through MyD88/IRAK/TRAF6 but not TAK1/IKK/NF-κB, as only knocking out MyD88/IRAK/TRAF6 eliminates the inhibitory effect on cell-cell fusion in response to IL-1β. This revealed that the inhibition of cell fusion did not require a transcriptional response and was mediated by IL-1R proximal signaling effectors.

Third, the authors identified RhoA/ROCK activation by IL-1 as the basis for this inhibition of cell fusion. By visualizing a RhoA biosensor and actin, they found a redistribution of RhoA to the cell periphery and cell-cell junctions after IL-1 stimulation. This triggered the formation of actin bundles at cell-cell junctions, preventing fusion and syncytia formation. The authors confirmed this molecular mechanism by using constitutively active RhoA and an inhibitor of ROCK.

Diverse Cell types and in vivo models were used, and consistent results were shown across diverse models. These results were convincing and well-presented.

Weaknesses:

As the authors point out in the discussion, whether IL-1-mediated RhoA activation is specific to viral infection or regulates other RhoA-regulated processes is unclear. We would also require high-magnification images of the subcellular organization of the cytoskeleton to appreciate the effect of IL-1 stimulation.

Reviewer #2 (Public Review):

Summary:

In this study, Zheng et al investigated the role of inflammatory cytokines in protecting cells against SARS-CoV-2 infection. They demonstrate that soluble factors in the supernatants of TLR-stimulated THP1 cells reduce fusion events between HEK293 cells expressing SARS-CoV-2 S protein and the ACE2 receptor. Using qRT-PCR and ELISA, they demonstrate that IL-1 cytokines are (not surprisingly) upregulated by TLR treatment in THP1 cells. Further, they convincingly demonstrate that recombinant IL-1 cytokines are sufficient to reduce cell-to-cell fusion mediated by the S protein. Using chemical inhibitors and CRISPR knock-out of key IL-1 receptor signaling components in HEK293 cells, they demonstrate that components of the myddosome (MYD88, IRAK1/4, and TRAF6) are required for fusion inhibition, but that downstream canonical signaling (i.e., TAK1 and NFKB activation) is not required. Instead, they provide evidence that IL-1-dependent non-canonical activation of RhoA/Rock is important for this phenotype. Importantly, the authors demonstrate that expression of a constitutively active RhoA alone is sufficient to inhibit fusion and that chemical inhibition of Rock could reverse this inhibition. The authors followed up these in vitro experiments by examining the effects of IL-1 on SARS-COV-2 infection in vivo and they demonstrate that recombinant IL-1 can reduce viral burden and lung pathogenesis in a mouse model of infection. However, the contribution of the RhoA/Rock pathway and inhibition of fusion to IL-1-mediated control of SARS-CoV-2 infection in vivo remains unclear.

Strengths:

(1) The bioluminescence cell-cell fusion assay provides a robust quantitative method to examine cytokine effects on viral glycoprotein-mediated fusion.

(2) The study identifies a new mechanism by which IL-1 cytokines can limit virus infection.

(3) The authors tested IL-1 mediated inhibition of fusion induced by many different coronavirus S proteins and several SARS-CoV-2 strains.

Weaknesses:

(1) The qualitative assay demonstrating S2 cleavage and IL-1 mediated inhibition of this phenotype is extremely variable across the data figures. Sometimes it appears like S2 cleavage (S2') is reduced, while in other figures immunoblots show that total S2 protein is decreased. Based on the proposed model the expectation would be that S2 abundance would be rescued when cleavage is inhibited.

(2) The text referencing Figure 1H suggests that TLR-stimulated THP-1 cell supernatants "significantly" reduce syncytia, but image quantification and statistics are not provided to support this statement.

(3) The authors conclude that because IL-1 accumulates in TLR2-stimulated THP1 monocyte supernatants, this cytokine accounts for the ability of these supernatants to inhibit cell-cell fusion. However, they do not directly test whether IL-1 is required for the phenotype. Inhibition of the IL-1 receptor in supernatant-treated cells would help support their conclusion.

(4) Immunoblot analysis of IL-1 treated HEK293 cells suggests that this cytokine does not reduce the abundance of ACE2 or total S protein in cells. However, it is possible that IL-1 signaling reduces the abundance of these proteins on the cell surface, which would result in a similar inhibition of cell-cell fusion. The authors should confirm that IL-1 treatment of their cells does not change Ace2 or S protein on the cell surface.

(5) In Figure 5A, expression of constitutively active RhoA appears to have profound effects on how ACE2 runs by SDS-PAGE, suggesting that RhoA may have additional effects on ACE2 biology that might account for the decreased cell-cell fusion. This phenotype should be addressed in the text and explored in more detail.

(6) The experiments linking IL-1 mediated restriction of SARS-COV-2 fusion to the control of virus infection in vivo are incomplete. The reported data demonstrate that recombinant IL-1 can restrict virus replication in vivo, but they fall short of confirming that the in vitro mechanism described (reduced fusion) contributes to the control of SARS-CoV2 replication in vivo. A critical piece of data that is missing is the demonstration that the ROCK inhibitor phenocopies IL-1RA treatment of SARS-COV-2 infected mice (viral infection and pathology).

eLife assessment

In this valuable study, the authors propose a model wherein the bacterial redox state plays a crucial role in the differentiation of Chlamydia trachomatis into elementary and reticulate bodies. They provide evidence to argue that a highly oxidising environment favours the formation of elementary bodies while a reducing condition slows down development. Whilst aspects related to the role of AhpC in regulating redox, and implications on differentiation, are solid, more precise measurements of the redox potential are required to convincingly demonstrate the role of redox in developmental progression.

Reviewer #1 (Public Review):

Summary:

Chlamydia spp. has a biphasic developmental cycle consisting of an extracellular, infectious form called an elementary body (EB) and an intracellular, replicative form known as a reticular body (RB). The structural stability of EBs is maintained by extensive cross-linking of outer membrane proteins while the outer membrane proteins of RBs are in a reduced state. The overall redox state of EBs is more oxidized than RBs. The authors propose that the redox state may be a controlling factor in the developmental cycle. To test this, alkyl hydroperoxide reductase subunit C (ahpC) was overexpressed or knocked down to examine effects on developmental gene expression. KD of ahpC induced increased expression of EB-specific genes and accelerated EB production. Conversely, overexpression of phpC delayed differentiation to EBs. The results suggest that chlamydial redox state may play a role in differentiation.

Strengths:

Uses modern genetic tools to explore the difficult area of temporal gene expression throughout the chlamydial developmental cycle.

Weaknesses:

The environmental signals triggering ahpC expression/activity are not determined.

Reviewer #2 (Public Review):

The factors that influence the differentiation of EBs and RBs during Chlamydial development are not clearly understood. A previous study had shown a redox oscillation during the Chlamydial developmental cycle. Based on this observation, the authors hypothesize that the bacterial redox state may play a role in regulating the differentiation in Chlamydia. To test their hypothesis, they make knock-down and overexpression strains of the major ROS regulator, ahpC. They show that the knock-down of ahpC leads to a significant increase in ROS levels leading to an increase in the production of elementary bodies and overexpression leads to a decrease in EB production likely caused by a decrease in oxidation. From their observations, they present an interesting model wherein an increase in oxidation favors the production of EBs.

Major concern:

In the absence of proper redox potential measurements, it is not clear if what they observe is a general oxidative stress response, especially when the knock-down of ahpC leads to a significant increase in ROS levels. Direct redox potential measurement in the ahpC overexpression and knock-down cells is required to support the model. This can be done using the roGFP-based measurements mentioned in the Wang et al. 2014 study cited by the authors.

Reviewer #3 (Public Review):

Summary:

The study reports clearly on the role of the AhpC protein as an antioxidant factor in Chlamydia trachomatis and speculates on the role of AhpC as an indirect regulator of developmental transcription induced by redox stress in this differentiating obligate intracellular bacterium.

Strengths:

The question posed and the concluding model about redox-dependent differentiation in chlamydia is interesting and highly relevant. This work fits with other propositions in which redox changes have been reported during bacterial developmental cycles, potentially as triggers, but have not been cited (examples PMID: 2865432, PMID: 32090198, PMID: 26063575). Here, AhpC over-expression is shown to protect Chlamydia towards redox stress imposed by H2O2, CHP, TBHP, and PN, while CRISPRi-mediated depletion of AhpC curbed intracellular replication and resulted in increased ROS levels and sensitivity to oxidizing agents. Importantly, the addition of ROS scavengers mitigated the growth defect caused by AhpC depletion. These results clearly establish the role of AhpC affects the redox state and growth in Ct (with the complicated KO genetics and complementation that are very nicely done).

Weaknesses:

However, with respect to the most important implication and claims of this work, the role of redox in controlling the chlamydial developmental cycle rather than simply being a correlation/passenger effect, I am less convinced about the impact of this work. First, the study is largely observational and does not resolve how this redox control of the cell cycle could be achieved, whereas in the case of Caulobacter, a clear molecular link between DNA replication and redox has been proposed. How would progressive oxidation in RBs eventually trigger the secondary developmental genes to induce EB differentiation? Is there an OxyR homolog that could elicit this change and why would the oxidation stress in RBs gradually accumulate during growth despite the presence of AhpC? In other words, the role of AhpC is simply to delay or dampen the redox stress response until the trigger kicks in, again, what is the trigger? Is this caused by increasing oxidative respiration of RBs in the inclusion? But what determines the redox threshold?

I also find the experiment with Pen treatment to have little predictive power. The fact that transcription just proceeds when division is blocked is not unprecedented. This also happens during the Caulobacter cell cycle when FtsZ is depleted for most developmental genes, except for those that are activated upon completion of the asymmetric cell division and that is dependent on the completion of compartmentalization. This is a smaller subset of developmental genes in caulobacter, but if there is a similar subset that depends on division on chlamydia and if these are affected by redox as well, then the argument about the interplay between developmental transcription and redox becomes much stronger and the link more intriguing. Another possibility to strengthen the study is to show that redox-regulated genes are under the direct control of chlamydial developmental regulators such as Euo, HctA, or others and at least show dual regulation by these inputs -perhaps the feed occurs through the same path.

This redox-transcription shortcoming is also reflected in the discussion where most are about the effects and molecular mitigation of redox stress in various systems, but there is little discussion on its link with developmental transcription in bacteria in general and chlamydia.

eLife assessment

This important study examines the role of TNF in modulating energy metabolism during parasite infection. The authors perform an elegant set of studies, however the evidence supporting the major claims of the manuscript is incomplete. This work integrates an interesting set of observations that will be of interest to the Plasmodium and pathogenesis communities with an expanded set of experiments.

Reviewer #1 (Public Review):

Summary:

The manuscript by Kely C. Matteucc et al. titled "Reprogramming of host energy metabolism mediated by the TNF-iNOS-HIF-1α axis plays a key role in host resistance to Plasmodium infection" describes that TNF induces HIF-1α stabilization that increases GLUT1 expression as well as glycolytic metabolism in monocytic and splenic CD11b+ cells in P. chabaudi infected mice. Also, TNF signaling plays a crucial role in host energy metabolism, controlling parasitemia, and regulating the clinical symptoms in experimental malaria.

Weaknesses:

Even though iNOS deficiency reduced the expression of the glycolytic enzymes as well as reduced GLUT1 expression and lower ECAR in splenic monocytes, there is no data to support that RNI induces the expression and stabilization of HIF-1α.

This paper involves an incredible amount of work, and the authors have done an exciting study addressing the TNF-iNOS-HIF-1α axis as a critical role in host immune defense during Plasmodium infection.

Reviewer #2 (Public Review):

Summary:

The premise of the manuscript by Matteucci et al. is interesting and elaborates on a mechanism via which TNFa regulates monocyte activation and metabolism to promote murine survival during Plasmodium infection. The authors show that TNF signaling (via an unknown mechanism) induces nitrite synthesis, which (via yet an unknown mechanism), and stabilizes the transcription factor HIF1a. Furthermore, HIF1a (via an unknown mechanism) increases GLUT1 expression and increases glycolysis in monocytes. The authors demonstrate that this metabolic rewiring towards increased glycolysis in a subset of monocytes is necessary for monocyte activation including cytokine secretion, and parasite control.

Strengths:

The authors provide elegant in vivo experiments to characterize metabolic consequences of Plasmodium infection, and isolate cell populations whose metabolic state is regulated downstream of TNFa. Furthermore, the authors tie together several interesting observations to propose an interesting model.

Weaknesses:

The main conclusion of this work - that "Reprogramming of host energy metabolism mediated by the TNF-iNOS-HIF1a axis plays a key role in host resistance to Plasmodium infection" is unsubstantiated. The authors show that TNFa induces GLUT1 in monocytes, but never show a direct role for GLUT1 or glucose uptake in monocytes in host resistance to infection (nor the hypoglycemia phenotype they describe).

eLife assessment

The work provides a valuable assessment of how antibiotics impact the human gut microbiota in diverse observational cohorts. Although the data presented are solid, some of the assumptions underlying their models may have affected the interpretation of their findings. The study is relevant for researchers and clinicians interested in antimicrobial resistance.

Reviewer #1 (Public Review):

Summary:

In this manuscript, the authors provide a study among healthy individuals, general medical patients and patients receiving haematopoietic cell transplants (HCT) to study the gut microbiome through shotgun metagenomic sequencing of stool samples. The first two groups were sampled once, while the patients receiving HCT were sampled longitudinally. A range of metadata (including current and previous (up to 1 year before sampling) antibiotic use) was recorded for all sampled individuals. The authors then performed shotgun metagenomic sequencing (using the Illumina platform) and performed bioinformatic analyses on these data to determine the composition and diversity of the gut microbiota and the antibiotic resistance genes therein. The authors conclude, on the basis of these analyses, that some antibiotics had a large impact on gut microbiota diversity, and could select opportunistic pathogens and/or antibiotic resistance genes in the gut microbiota.

Strengths:

The major strength of this study is the considerable achievement of performing this observational study in a large cohort of individuals. Studies into the impact of antibiotic therapy on the gut microbiota are difficult to organise, perform and interpret, and this work follows state-of-the-art methodologies to achieve its goals. The authors have achieved their objectives and the conclusion they draw on the impact of different antibiotics and their impact on the gut microbiota and its antibiotic resistance genes (the 'resistome', in short), are supported by the data presented in this work.

Weaknesses:

The weaknesses are the lack of information on the different resistance genes that have been identified and which could have been supplied as Supplementary Data. In addition, no attempt is made to assess whether the identified resistance genes are associated with mobile genetic elements and/or (opportunistic) pathogens in the gut. While this is challenging with short-read data, alternative approaches like long-read metagenomics, Hi-C and/or culture-based profiling of bacterial communities could have been employed to further strengthen this work. Unfortunately, the authors have not attempted to perform corrections for multiple testing because many antibiotic exposures were correlated.

Impact:

The work may impact policies on the use of antibiotics, as those drugs that have major impacts on the diversity of the gut microbiota and select for antibiotic resistance genes in the gut are better avoided. However, the primary rationale for antibiotic therapy will remain the clinical effectiveness of antimicrobial drugs, and the impact on the gut microbiota and resistome will be secondary to these considerations.

Reviewer #2 (Public Review):

Summary:

In this manuscript by Peto et al., the authors describe the impact of different antimicrobials on gut microbiota in a prospective observational study of 225 participants (healthy volunteers, inpatients and outpatients). Both cross-sectional data (all participants) and longitudinal data (a subset of 79 haematopoietic cell transplant patients) were used. Using metagenomic sequencing, they estimated the impact of antibiotic exposure on gut microbiota composition and resistance genes. In their models, the authors aim to correct for potential confounders (e.g. demographics, non-antimicrobial exposures and physiological abnormalities), and for differences in the recency and total duration of antibiotic exposure. I consider these comprehensive models an important strength of this observational study. Yet, the underlying assumptions of such models may have impacted the study findings (detailed below). Other strengths include the presence of both cross-sectional and longitudinal exposure data and the presence of both healthy volunteers and patients. Together, these observational findings expand on previous studies (both observational and RCTs) describing the impact of antimicrobials on gut microbiota.

Weaknesses:

(1) The main weaknesses result from the observational design. This hampers causal interpretation and corrects for potential confounding necessary. The authors have used comprehensive models to correct for potential confounders and for differences between participants in duration of antibiotic exposure and time between exposure and sample collection. I wonder if some of the choices made by the authors did affect these findings. For example, the authors did not include travel in the final model, but travel (most importantly, south Asia) may result in the acquisition of AMR genes [Worby et al., Lancet Microbe 2023; PMID 37716364). Moreover, non-antimicrobial drugs (such as proton pump inhibitors) were not included but these have a well-known impact on gut microbiota and might be linked with exposure to antimicrobial drugs. Residual confounding may underlie some of the unexplained discrepancies between the cross-sectional and longitudinal data (e.g. for vancomycin).

In addition, the authors found a disruption half-life of 6 days to be the best fit based on Shannon diversity. If I'm understanding correctly, this results in a near-zero modelled exposure of a 14-day-course after 70 days (purple line; Supplementary Figure 2). However, it has been described that microbiota composition and resistome (not Shannon diversity!) remain altered for longer periods of time after (certain) antibiotic exposures (e.g. Anthony et al., Cell Reports 2022; PMID 35417701). The authors did not assess whether extending the disruption half-life would alter their conclusions.

(2) Another consequence of the observational design of this study is the relatively small number of participants available for some comparisons (e.g. oral clindamycin was only used by 6 participants). Care should be taken when drawing any conclusions from such small numbers.

(3) The authors assessed log-transformed relative abundances of specific bacteria after subsampling to 3.5 million reads. While I agree that some kind of data transformation is probably preferable, these methods do not address the compositional data of microbiome data and using a pseudocount (10-6) is necessary for absent (i.e. undetected) taxa [Gloor et al., Front Microbiol 2017; PMID 29187837]. Given the centrality of these relative abundances to their conclusions, a sensitivity analysis using compositionally-aware methods (such as a centred log-ratio (clr) transformation) would have added robustness to their findings.

(4) An overall description of gut microbiota composition and resistome of the included participants is missing. This makes it difficult to compare the current study population to other studies. In addition, for correct interpretation of the findings, it would have been helpful if the reasons for hospital visits of the general medical patients were provided.

eLife assessment

This valuable study reports data showing the link between a disruption in testicular mineral (phosphate) homeostasis, FGF23 expression, and Sertoli cell dysfunction. The data supporting the conclusion remains incomplete. This work will be of interest to biomedical researchers working on testis biology and male infertility.

Reviewer #1 (Public Review):

Summary:

Despite the study being a collation of important results likely to have an overall positive effect on the field, methodological weaknesses and suboptimal use of statistics make it difficult to give confidence to the study's message.

Strengths:

Relevant human and mouse models approached with in vivo and in vitro techniques.

Weaknesses:

The methodology, statistics, reagents, analyses, and manuscripts' language all lack rigour.

(1) The authors used statistics to generate P-values and Rsquare values to evaluate the strength of their findings.

However, it is unclear how stats were used and/or whether stats were used correctly. For instance, the authors write: "Gaussian distribution of all numerical variables was evaluated by QQ plots". But why? For statistical tests that fall under the umbrella of General Linear Models (line ANOVA, t-tests, and correlations (Pearson's)), there are several assumptions that ought to be checked, including typically:

(a) Gaussian distribution of residuals.

(b) Homoskedasticity of the residuals.

(c) Independence of Y, but that's assumed to be valid due to experimental design.

So what is the point of evaluating the Gaussian distribution of the data themselves? It is not necessary. In this reviewer's opinion, it is irrelevant, not a good use of statistics, and we ought to be leading by example here.

Additionally, it is not clear whether the homoscedasticity of the residuals was checked. Many of the data appear to have particularly heteroskedastic residuals. In many respects, homoscedasticity matters more than the normal distribution of the residuals. In Graphpad analyses if ANOVA is used but equal variances are assumed (when variances among groups are unequal then standard deviations assigned in each group will be wrong and thus incorrect p values are being calculated.

Based on the incomplete and/or wrong statistical analyses it is difficult to evaluate the study in greater depth.

While on the subject of stats, it is worth mentioning this misuse of statistics in Figure 3D, where the authors added the Slc34a1 transcript levels from controls in the correlation analyses, thereby driving the intercept down. Without the Control data there does not appear to be a correlation between the Slc34a1 levels and tumor size.

There is more. The authors make statements (e.g. in the figure levels as: "Correlations indicated by R2.". What does that mean? In a simple correlation, the P value is used to evaluate the strength of the slope being different from zero. The authors also give R2 values for the correlations but they do not provide R2 values for the other stats (like ANOVAs). Why not?

(2) The authors used antibodies for immunos and WBs. I checked those antibodies online and it was concerning:

(a) Many are discontinued.

(b) Many are not validated.

(c) Many performed poorly in the Immunos, e.g. FGF23, FGFR1, and Kotho are not really convincing. PO5F1 (gene: OCT4) is the one that looks convincing as it is expressed at the correct cell types.

(d) Others like NPT2A (product of gene SLC34A1) are equally unconvincing. Shouldn't the immuno show them to be in the plasma membrane?

If there is some brown staining, this does not mean the antibodies are working. If your antibodies are not validated then you ought to omit the immunos from the manuscript.

Reviewer #2 (Public Review):

Summary:

This study set out to examine microlithiasis associated with an increased risk of testicular germ cell tumors (TGCT). This reviewer considers this to be an excellent study. It raises questions regarding exactly how aberrant Sertoli cell function could induce osteogenic-like differentiation of germ cells but then all research should raise more questions than it answers.

Strengths:

Data showing the link between a disruption in testicular mineral (phosphate) homeostasis, FGF23 expression, and Sertoli cell dysfunction, are compelling.

Weaknesses:

Not sure I see any weaknesses here, as this study advances this area of inquiry and ends with a hypothesis for future testing.

eLife assessment

This valuable study shows how an intersecting network of regulators acting on genes with differences in their RNA metabolism explains why the loss of some regulators of RNAi in C. elegans can selectively impair the silencing of some target genes. The evidence presented is convincing, as the authors use a combination of computational modeling and RNAi assays to support their conclusions.

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

The manuscript addresses a fundamental question about how different types of communication signals differentially affect brain state and neurochemistry. In addition, their manuscript  highlights the various processes that modulate brain responses to communication signals, including prior experience, sex, and hormonal status. Overall, the manuscript is well-written and the research is appropriately contextualized.

That being said, it remains important for the authors to think more about their analytical approaches. In particular, the effect of normalization and the explicit outlining and interpretations of statistical models. As mentioned in the original review, the normalization of neurochemical data seems unnecessary given the repeated-measures design of their analysis and by normalizing all data to the baseline data and including this baseline data in the repeated measures analysis,   one artificially creates a baseline period with minimal variation that dramatically differs in variance from other periods (akin to heteroscedasticity). If the authors want to analyze how a stimulus changes neurochemical concentrations, they could analyze the raw data but depict normalized data in their figures (similar to other papers). Or they could analyze group differences in the normalized data of the two stimulus periods (i.e., excluding the baseline period used for normalization).

We appreciate the reviewer’s point on the difference in variance caused by including the 100% baseline values in the analysis. After consulting with our statistician, we chose the latter of the two approaches suggested by the reviewer. Specifically, we reran the analysis to exclude the baseline and focus only on the playback windows and the group differences. The text in the results, the significance signs in the figures, and the discussion are corrected accordingly. Despite these changes, our major conclusions remains as before.

We also followed this reviewer’s suggestions to clarify the statistical model in studying the experience effect. After further consultation with our statistician, we reran the analysis on experience effect, including all the groups of EXP and INEXP animals together. We have corrected text in the figure captions, results, discussion, and data analysis sections of the manuscript related to the effect of experience and its interactions. This has not changed the conclusion made related to the experience effect in the dataset.

It would also be useful for the authors to provide further discussion of the potential contributions of different types of experiences (mating vs. restraint) to the change in behavior and neurochemical responses to the vocalization playbacks and to try to disentangle sensory and  motor contributions to neurochemical changes.

We have acknowledged in the Discussion that previous studies suggest that the effect of experience involving stress could be generalized. We believe that this is an important area of future research. Our Discussion acknowledges that the relationship between sensory and motor contributions to neurochemical changes remains an area of interest. We further point out that the time resolution of microdialysis data renders the suggested discussion highly speculative. We plan to use other methods to assess this in future experiments.

Reviewer #3 (Public Review):

The work by Ghasemahmad et al. has the potential to significantly advance our understanding of how neuromodulators provide internal-state signals to the basolateral amygdala (BLA) while an animal listens to social vocalizations.

Ghasemahmad et al. made changes to the manuscript that have significantly improved the work. In particular, the transparency in showing the underlying levels of Ach, DA, and 5HIAA is excellent. My previous concerns have been adequately addressed.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

I appreciate the authors responses to my previous queries (and to the comments by other reviewers). The introduction does a better job contextualizing the data, and the additional details in the results and Methods sections help readers digest the material. I continue to think the topic  is interesting and the manuscript is potentially impactful. However, I continue to be concerned about their analytical approaches and other aspects of the revised manuscript.

(a) Normalization

In my original review I wrote: "The normalization of neurochemical data seems unnecessary   given the repeated-measures design of their analysis and could be problematic; by normalizing     all data to the baseline data (p. 24), one artificially creates a baseline period with minimal   variation (all are "0"; Figures 2, 3 & 5) that could inflate statistical power." I continue to feel that an analysis of normalized data that includes the baseline data is inappropriate because of the minimal variation in the normalized data for the baseline period. When the normalized data for   the baseline period is included in the analysis, there is clearly variation in the extent of variability within each of the time periods (no variability at baseline, variability during periods 1 & 2; analogous to heteroscedasticity). For example, when analyzing the RAW DATA about the change in ACh release in experienced males listening to restraint vocalizations (thank you for releasing the raw data), there was a non-significant effect of time (baseline, period 1, and period 2; linear mixed effects model; F(2,12)=3.2, p=0.0793). However, when the normalized data for  this dataset was analyzed (with baseline values being set at 100% for each mouse), there was a statistically significant effect (F(2,12)=4.5, p=0.0352). This example is just to illustrate how normalization can affect (e.g., inflate) statistical power.

That being said, I do think that it is reasonable to analyzed normalized data if the period used for normalization is NOT included in the analysis (see Figure 3 of one of the paper the authors listed in their response to reviewers: Galvez-Marquez et al., 2022). However, from the reading of this manuscript, it does seem like normalized baseline data are analyzed to assess how stimuli affect neurochemical concentrations.

We appreciate the reviewer’s point on the difference in variance caused by including the 100% baseline values in the analysis. After consulting with our statistician, we chose one of the two approaches suggested by the reviewer. Specifically, we reran the analysis to exclude the baseline and focus only on the playback windows and the group differences. The text in the results, the significance signs in the figures, and the discussion are corrected accordingly. Despite these changes, our major conclusions remains as before. We have included some descriptive statistics in the text because we think these are informative.

We decided to take this approach because the inter-individual variability in the raw data levels, caused by non-experimental factors, is too great to be useful. As we have stated before, these values are affected by probe placement, collection process, or differences in the HPLC or LC/MS runs. These effects are widely recognized in the field.

It is worth pointing out a few things about the papers listed by the authors. Li et al. (2023) does depict normalized microanalysis data but it isn't clear that any analysis of the normalized data is conducted. The same can be said about Holly et al. (2016). Further, in Bagley et al (2011), the authors depict normalized data in the figures but conduct analyses on the raw data ("After  chronic morphine treatment, systemic naloxone injection increased GABA outflow in PAG by 41% (from 24.6 {plus minus} 2.9 nM to a peak of 34.8 {plus minus} 3.8 nM, n = 6, P = 0.016), but did not alter GABA levels after vehicle treatment (39.8 {plus minus} 8.3 to 38.6 {plus  minus} 7.4 nM with naloxone at matched peak time, n = 4; Fig. 3a)". This latter approach (analyzing raw data in a repeated-measures manner and depicted normalized data) seems reasonable for the authors of the current study.

(b) Clarification and modification of statistical models

When analyzing the effect of experience on neuromodulator release, the authors analyze the experienced and inexperienced mice independently (e.g., figure 3 vs. 6). The ideal way to assess the effects of experience is to create a factorial model. For example, one could analyze a full factorial model with experience (exp vs. inexp), stimulus time (mating vs. restraint) and time  (baseline, period 1 vs period 2, assuming raw data are used). If one wanted to exclude the  baseline period because group differences in baseline are not informative, conducting a factorial analysis of normalized data with just the data from period 1 and 2 seems fine. I believe an analysis like this will help increase the legitimacy of the analysis. For example, when analyzing the normalized data (periods 1 and 2) of experienced and inexperienced males in response to mating or restraint vocalizations, you find a significant interaction between experience and stimulus type. Finding an effect of experience in an analysis that includes both experienced and inexperienced mice is ideal from an analytical framework.

In Figure 6, it is not clear what the statistical model is and what the interactions mean. For example, in the figure legend for figure 6, the authors report time*context and time*sex interactions. However, in this analysis there are two groups of inexperienced males (males that   are listening to restraint vocalizations, males that are listening to mating vocalizations) and one group of females (females that are listening to mating vocalizations); in other words, this is an unbalanced analysis. So, when the authors indicate a time*context interaction, does that mean  they are comparing the male-restraint group to the combination of males and females listening to mating vocalizations? And when they talk about a time*sex interaction, are they analyzing how males listening to either mating or restraint vocalizations differ from females listening to a   mating vocalization? This all seems peculiar to me.

- A similar set of questions could be raised about interaction effects depicted in Figure 4.

Overall, I would like this manuscript to be reviewed by a statistician to provide additional input on how best to analyze the data.

We followed the reviewer’s suggestions to clarify the statistical model in studying the experience effect. After further consultation with the statistician, we reran the analysis on experience effect, including all the groups of EXP and INEXP animals together.

Design: Intercept + Sex +Context + Experience+ Sex* Experience + Context* Experience.

The model is not full factorial as recommended by the statistician, because we don’t have females in the restraint group and that would make an unbalanced design. Therefore, running GLM based on the above model and included factors, as advised by the statistician, is the best way of approaching the analysis for the current dataset.

We have corrected text in the figure captions, results, discussion, and data analysis sections of the manuscript related to the effect of experience and its interactions. The GLM models are clarified for all the figures in the “data analysis” section of the manuscript. We have clarified that the major effect of experience on neuromodulators was seen in the ACh data.

(c) Analysis of post-stimulus period

I agree with Reviewer 3 that analyzing the post-stimulus period would be useful. As mentioned     in the original review, these data could serve as an opportunity to show that the neurochemical levels returned to baseline and add further support for the model described in Figure 6. In   addition, these data could help reveal the link  between  neurochemical  release,  auditory responses, and behavior. If neurochemical changes reflect auditory responses, then these should back to baseline during the post-stimulus period. In addition, if behavioral variation (e.g.,    between mice hearing mating vs. restraint stimuli) persists following the termination of playback, then one could similarly assess whether neurochemical variation persists following playback. If   the latter is the case, then the neurochemical release could be more related to the behavior than to the playback stimulus itself.

We did not change this analysis. Our response to Reviewer 3’s comment is shown below.

“We decided not to include analyses of the post-stimulus period because this period is subject to wider individual and neuromodulator-specific effects and because it weakens statistical power in addressing the core question—the change in neuromodulator release DURING vocal playback. We agree that the general question is of interest to the field, but we don’t think our study is best designed to answer that question.”

This was accepted by Reviewer 3. We also note that release patterns have multiple time courses (e.g., Aitta-aho et al., 2018 for ACh), and thus may not support an assumption that levels should return to baseline shortly after playback offset.

Minor comments:

Page 7, line 15: I suggest changing "vocalization-dependent" to "stimulus-dependent" because the former could connote patterns of release related to the animal itself vocalizing.

Changed to: “There were also distinct patterns of ACh and DA release into the BLA depending on the type of vocalization playback (Fig 3C,D).”

Discussion section: The authors should point out a few caveats with their experiments in the Discussion section. First, experienced animals received both mating (social) and restraint experiences, and it is not clear to what degree each type of experience affected neural and behavioral responses (i.e., specificity of experience effects). For example, mating experience can lead to a wide range of physiological changes, including a resilience to stress (e.g., Leuner et al., PLoS One, 2010; Arnold et al., Hormones and Behavior, 2019), so it is possible that mating experiences by themselves could have induced these changes. Or it could be that experiencing restraint stress affects responses to mating stimuli. This could be added to the first paragraph in page 16. (The authors could also discuss which aspects of the sexual encounters might be most important for the behavioral and neural plasticity.)

We have added text to raise this issue, stating that it is unknown wither the experience effects are specific and citing the above references concerning the generalized effects of certain experiences.

Discussion section: It would also be useful for the authors to discuss the extent to which behavior might be driving the neurochemical changes. Some of the analyses suggest that the release is independent of the behavior (e.g., reflects a sensory responses) but this could be emphasized    more in the Discussion.

We believe that we have addressed this issue sufficiently in our previous response to related issues raised by this reviewer. As we note, there are limitations in the time resolution of microdialysis data that render the suggested discussion highly speculative. We plan to use other methods to assess this in future experiments.

Figure 2, legend: Please note that the text above the images describes the stimulus played back to these animals and their hormonal state, and not the type of experienced they underwent (i.e.,  clarify the titles)

Changed as requested.

I also agree with Reviewer 3 that "mating experience" is a misnomer for this manuscript. "Social experience with a female" is a more accurate descriptor. If they wanted to specifically provide mating experience, males should have only been tested with estrus (receptive females). I don't think this wording change detracts from their findings.

We have not changed this term. As noted in our previous response to Reviewer #3, we stated: “In the mating experience, mounting or attempted mounting was required for the animal to be included in subsequent testing.” Due to this requirement, the term “mating behavior” is informative and appropriate. In our view, “Social experience with a female” does not adequately describe our inclusion criterion or the experience.

Reviewer #3 (Recommendations For The Authors):

The work by Ghasemahmad et al. has the potential to significantly advance our understanding of how neuromodulators provide internal-state signals to the basolateral amygdala (BLA) while an animal listens to social vocalizations.

Ghasemahmad et al. made changes to the manuscript that have significantly improved the work. In particular, the transparency in showing the underlying levels of Ach, DA, and 5HIAA is excellent. My previous concerns have been adequately addressed. I only have a few minor suggestions for the text and one figure.

Minor suggestions:

Page 2, Ln 9: add adult before male and female mice

Changed as requested

Page 4, Ln 10: add a period after Tsukano et al., 2019)

Changed as requested

Page 6, Ln 9: what did you mean by "their interaction"? Being more specific, but concise, would help the readers.

We revised the wording to clarify that the neuromodulatory systems interact in the emission of positive and negative vocalizations.

Page 6, Ln 17: You mention Stim 1 and Stim 2, but the stimuli are not defined at this point. The clear explanation is provided in the following paragraph. Maybe consider switching the order  and define the stimuli before you describe the liquid chromatography/mass spectrometry technique.

We have revised and merged these paragraphs so that Stim 1 and Stim 2 are defined on first use. We also revised our description of the depiction and analysis of neurochemical data.

Page 11, Ln 12: replace well-proven with well-documented

Changed as requested

Figure 2: There are two arrows pointing towards a single track. I assume one of the arrows is a duplicate. If so, delete one of the arrows. If not, please explain what the second arrow represents.

Arrow removed

eLife assessment

This provocative manuscript from presents valuable comparisons of the morphologies of Archaean bacterial microfossils to those of microbes transformed under environmental conditions that mimic those present on Earth during the same Eon, although the evidence in support of the conclusions is currently incomplete. The reasons include that taphonomy is not presently considered, and a greater diversity of experimental environmental conditions is not evaluated – which is significant because we ultimately do not know much about Earth's early environments. The authors may want to reframe their conclusions to reflect this work as a first step towards an interpretation of some microfossils as 'proto-cells,' and less so as providing strong support for this hypothesis.

eLife assessment

This important study aims to understand how the regulation of oligodendrocyte progenitor cell (OPC) remyelination and function contributes to the treatment of multiple sclerosis. The authors provide convincing evidence for the platelets mediating OPC differentiation and remyelination. This work will be of interest to several disciplines.

Reviewer #1 (Public Review):

Summary:

The authors have studied the effects of platelets in OPC biology and remyelination. For this, they used mutant mice with lower levels of platelets as a demyelinating/remyelinating scenario, as well as in a model with large numbers of circulating platelets.

Strengths:

-The work is very focused, with defined objectives.<br /> -The work is properly done.

Revision comments:

Having consulted the new version of the work by Amber et al., the modifications and the point-by-point cover letter explaining them give direct answers to my previous comments.

Reviewer #2 (Public Review):

Summary:

This paper examined whether circulating platelets regulate oligodendrocyte progenitor cell (OPC) differentiation for the link with multiple sclerosis (MS). They identified that the interaction with platelets enhances OPC differentiation although persistent contact inhibits the process in the long-term. The mouse model with increased platelet levels in the blood reduced mature oligodendrocytes, while how platelets might regulate OPC differentiation is not clear yet.

Strengths:

The use of both partial platelet depletion and thrombocytosis mouse models gives in vivo evidence. The presentation of platelet accumulation in a time-course manner is rigorous. The in vitro co-culture model tested the role of platelets in OPC differentiation, which was supportive of in vivo observations.

Revision comments:

Although the mechanisms are limited, the authors addressed the major experiments I suggested.

Author response:

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

Reviewer #1 (Public Review):

Summary:

The authors have studied the effects of platelets in OPC biology and remyelination. For this, they used mutant mice with lower levels of platelets as a demyelinating/remyelinating scenario, as well as in a model with large numbers of circulating platelets.

Strengths:

-The work is very focused, with defined objectives.

-The work is properly done.

Weaknesses:

-There is no clear effect on a single cell type and/or mechanism involved.

We appreciate the reviewer’s feedback. We understand that from our in vivo studies we are unable to distinguish whether the effects of platelets are directly exerted on OPCs or indirectly through a different cell type. However, data obtained from the platelet depleted model as well as the new data provided in this revised version in CALRHet mice indicate that, at least, macrophages / microglia do not contribute to the observed effects in OPCs. In addition to this, in vitro data support the direct effects of platelets on OPC function.

Reviewer #2 (Public Review):

Summary:

This paper examined whether circulating platelets regulate oligodendrocyte progenitor cell (OPC) differentiation for the link with multiple sclerosis (MS). They identified that the interaction with platelets enhances OPC differentiation although persistent contact inhibits the process in the longterm. The mouse model with increased platelet levels in the blood reduced mature oligodendrocytes, while how platelets might regulate OPC differentiation is not clear yet.

Strengths:

The use of both partial platelet depletion and thrombocytosis mouse models gives in vivo evidence. The presentation of platelet accumulation in a time-course manner is rigorous. The in vitro co-culture model tested the role of platelets in OPC differentiation, which was supportive of in vivo observations.

Weaknesses:

How platelets regulate OPC differentiation is not clear. What the significance of platelets is in MS progression is not clear.

We thank reviewer’s view and assessment of our manuscript. We understand both of the reviewer’s concerns. Firstly, we performed additional in vitro studies and we have confirmed that platelet-contained factors are, at least in part, responsible for modulating OPC differentiation and, thus, direct cell-cell contact is not essential. Secondly, in this revised version, we added references arguing that the plasma levels of platelet microparticles and platelet-specific factors correlate with MS progression and severity.  

Reviewer #1 (Recommendations For the Authors):

To ameliorate the quality of their work and make it suitable for its publication in eLife, I strongly suggest the authors to: 

(1) In vitro co-culture platelets and OPCs to check the effects on this latter cell type biology. 

Response: We have performed in vitro studies, in which OPCs were co-cultured with washed platelets (WP). We observed that OPC differentiation was boosted after a short exposure to WP, however, prolonged exposure to WP suppressed this effect (revised Figure 3A and B). Also, our new data using platelet lysate (PL) indicate that platelet-contained molecules are responsible for this effect (revised Figure 3C and D). Finally, we showed that by removing PL after sustained exposure (6 DIV) the ability of platelets to promote OPC differentiation is rescued (revised Figure 3E and F).

(2) In the CALR model, can the authors check effect of chronic exposure to large numbers of platelets? Is this affecting macrophages (including their polarization)? 

Response: Yes, compared to wild type mouse, in the CALRHET model we confirmed the presence of larger number of platelets within demyelinated lesions (Figure 4A and C). Also, in this revised version we added data showing in the CALRHET model that thrombocytosis does not affect macrophage / microglia numbers and polarization (revised Supplementary Figure 2). 

(3) Some aspects of the Introduction section seems too old-fashioned (ex.: the use of bFGF instead of FGF2 to refer to Fibroblast Growth Factor 2), as well as it would be convenient to include more recent references on the role of FGF2 and PDGFa in OPC biology. 

Response: We agree with the reviewer. In this revised version we have changed bFGF for FGF2 and we added more recent references addressing the role of FGF2 and PDGFa in OPC biology.

(4) There are some constructions and typos that could be corrected. 

Response: We have checked language constructions as well as typos, and these have been corrected.

(5) Please revise spelling of some names in the bibliography list (ex.: the correct surname is ffrenchConstant, not Ffrench-Constant).

We have revised the spelling of names within the bibliography, and we have corrected them accordingly.

Reviewer #2 (Recommendations For the Authors):

Mechanisms of platelet-OPC interactions 

-  transwell co-culture assay will examine if the OPC phenotype is through direct or indirect interactions with platelets; 

We have performed additional in vitro studies, in which OPCs were exposed to platelet lysate (PL). New results indicate that a short exposure to PL can promote OPC differentiation (revised Figure 3C and D), while a sustained exposure supresses this effect (revised Figure 3E and F). These findings indicate that platelet-contained factors are, at least in part, responsible for modulating OPC differentiation and, thus, direct cell-cell contact is not essential for such an effect.

-  can you revert the phenotype of OPCs co-cultured long with platelets (maturation blocked) by removing platelet (then OPC differentiate again?) to see if the phenotype is reversible or not? 

We would like to thank the reviewer for bringing up this interesting question. We have performed additional in vitro studies to address this issue. We found that by removing PL upon 6-days of sustained exposure rescues the ability of platelets to promote OPC differentiation (revised Figure 3E and F). These findings indicate that the supressing effect of prolonged exposure to platelets in OPC differentiation is reversible.  

Clinical correlation 

-  How many MS patients has an abnormal number of or exposure to platelets? 

We have added new information in the introduction section. Indeed, previous studies have shown that MS patients display higher levels of circulating platelet microparticles (PMPs) (MarcosRamiro et al., 2014) as well as increased plasma levels of platelet-specific factors such as, P-selectin and PF4 (Cananzi et al., 1987; Kuenz et al., 2005).

do platelets amount correlate with diseases severeness? 

We have added new information in the introduction section. Indeed, PMPs are indicative of the clinical status of the disease (Saenz-Cuesta et al., 2014). Also, plasma levels of P-selectin and PF4 correlate with disease course and severity, respectively (Cananzi et al., 1987; Kuenz et al., 2005).

Image quantification 

-  please state how many sections were counted. How many animals were used per condition. Is the practice of blinded observers done for each dataset?

We added this information in the figure legends and in methods section. We counted between 3-5 sections per lesion. We used 3 to 6 animals per experimental group and data was analysed by blinded observers.

Reviewer #2 (Public Review):

Summary:

Ma et al. employed a myeloid progenitor/microglia differentiation protocol to produce human-induced pluripotent stem cell (hiPSC)-derived microglia in order to examine the potential of microglial cell replacement as a treatment for retinal disorders. They characterized the iPSC-derived microglia by gene expression and in vitro assay analysis. By evaluating xenografted microglia in the partly microglia-depleted retina, the function of the microglia was further assessed.

Overall, the study and the data are convincing, and xenografted microglia were also tested in a RPE injury paradigm.

eLife assessment

The authors have improved a method to differentiate human iPSC-derived microglial cells with immune responses and phagocytic abilities; and through transplantation into the adult mouse retina, the authors further demonstrated their integration and occupation of native microglial cell space, and functional response to retinal injuries. The study is important and the data are convincing for potential microglial replacement therapy to treat retinal and CNS diseases.

Reviewer #1 (Public Review):

Summary:

This paper reported a protocol of using human-induced pluripotent stem cells to generate cells expressing microglia-enriched genes and responding to LPS by drastically upregulation of proinflammatory cytokines. Upon subretinal transplantation in mice, hiPSC-derived cells integrated into the host retina and maintained retinal homeostasis while they responded to RPE injury by migration, proliferation, and phagocytosis. The findings revealed the potential of using hiPSC-derived cell transplantation for microglia replacement as a therapeutic strategy for retinal diseases.

Strengths:

The paper demonstrates a method of consistently generating a significant quantity of hiPSC-derived microglia-like cells for in vitro study or for in vivo transplantation. RNAseq analysis offers an opportunity for comprehensive transcriptome profiling of the derived cells. It is impressive that following transplantation, these cells were well integrated into the retina, migrated to the corresponding layers, adopted microglia-like morphologies, and survived for a long term without generating apparent harm. The work has laid a foundation for future utilization of hiPSC-derived microglia in lab and clinical applications.

Weaknesses:

(1) The primary weakness of the paper concerns its conclusion of having generated "homogenous mature microglia", partly based on the RNAseq analysis. However, the comparison of gene profiles was carried out only between "hiPSC-derived mature microglia" and the proliferating myeloid progenitors. While the transcriptome profiles revealed a trend of enrichment of microglia-like gene expression in "hiPSC-derived mature microglia" compared to proliferating myeloid progenitors, this is not sufficient to claim they are "mature microglia". It is important that one carries out a comparative analysis of the RNAseq data with those of primary human microglia, which may be done by leveraging the public database. To convincingly claim these cells are mature microglia, questions to be addressed include how similar the molecular signatures of these cells are compared with the fully differentiated primary microglia cell or if they remain progenitor-like or take on mosaic properties, and how they distinguish from macrophages.

(2) While the authors attempted to demonstrate the functional property of "hiPSC-derived mature microglia" in culture, they used LPS challenge, which is an inappropriate assay. This is because human microglia respond poorly to LPS alone but need to be activated by a combination of LPS with other factors, such as IFNγ. Their data that "hiPSC-derived mature microglia" showed robust responses to LPS indeed implicates that these cells do not behave like mature human microglia.

(3) The resolution of Figs. 4 - 6 is so low that even some of the text and labels are hardly readable. Based on the morphology shown in Fig. 4 and the statement in line 147, these hiPSC-derived "cells altered their morphology to a rounded shape within an hour of incubation and rapidly internalized the fluorescent-labeled particles". This is a peculiar response. Usually, microglia do not respond to fluorescent-labeled zymosan by turning into a rounded-shaped morphology within an hour when they internalize them. Such a behavior usually implicates weak phagocytotic capacity.

(4) Data presented in Fig. 5 are not very convincing to support that transplanted cells were immunopositive for "human CD11b (Fig.5C), as well as microglia signature markers P2ry12 and TMEM119 (Fig.5D)" (line 167). The resolution and magnification of Fig. 5D are too low to tell the colocalization of tdT and human microglial marker immunolabeling. In the flat-mount images (C, I), hCD11b immunolabeling is not visible in the GCL or barely visible in the IPL. This should be discussed.

(5) Microglia respond to injury by becoming active and losing their expression of the resting state microglial marker, such as P2ry12, which is used in Fig. 6 for the detection of migrated microglia. To confirm that these cells indeed respond to injury like native microglia, one should check for activated microglial markers and induction of pro-inflammatory cytokines in the sodium iodate-injury model.

Author response:

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

Reviewer #1 (Public review):

(1) The primary weakness of the paper concerns its conclusion of having generated "homogenous mature microglia", partly based on the RNAseq analysis. However, the comparison of gene profiles was carried out only between "hiPSC-derived mature microglia" and the proliferating myeloid progenitors. While the transcriptome profiles revealed a trend of enrichment of microglia-like gene expression in "hiPSC-derived mature microglia" compared to proliferating myeloid progenitors, this is not sufficient to claim they are "mature microglia". It is important that one carries out a comparative analysis of the RNAseq data with those of primary human microglia, which may be done by leveraging the public database. To convincingly claim these cells are mature microglia, questions need to be addressed including how similar the molecular signatures of these cells are compared with the fully differentiated primary microglia cell or if they remain progenitor-like or take on mosaic properties, and how they distinguish from macrophages.

We greatly appreciate the insightful comments and suggestions from the reviewers, which were instrumental in enhancing our data analysis and organization. In response to the feedback, we have updated the terminology from “mature microglia” to simply “microglia” while clarifying in our text that these are fully differentiated microglia under single-type cell culture conditions.

Guided by the reviewer's advice, we incorporated RNA-seq data from human brain microglia studies conducted by Dr. Poon and Dr. Blurton-Jones' Lab (Abud et al., Neuron, 2017) and Dr. Huitinga's Lab (van der Poel et al., Nat Commun, 2019). We then conducted a comparative analysis of the gene expression profiles between our fully differentiated hiPSC-derived microglia and those from fetal/adult brain microglia (see Fig.2. Suppl. B, C and D; Suppl. table 1 and table 2). The correlation analysis revealed that our hiPSC-derived microglia closely resemble fetal and adult brain microglia, distinguishing them significantly from monocytes and inflammatory monocytes.

(2) While the authors attempted to demonstrate the functional property of "hiPSC-derived mature microglia" in culture, they used LPS challenge, which is an inappropriate assay. This is because human microglia respond poorly to LPS alone but need to be activated by a combination of LPS with other factors, such as IFNγ. Their data that "hiPSC-derived mature microglia" showed robust responses to LPS indeed implicates that these cells do not behave like mature human microglia.

We appreciate the feedback received. In response, we cultured hiPSC-derived microglia cells and subjected them to treatments with IFNγ, LPS, and a combination of both IFNγ+LPS, as illustrated in Figure 3 suppl. Our findings revealed that the IFNγ+LPS combination notably enhanced the expression of IL1a, IL1b, TNFa, CCL8, and CXCL10, whereas IL6 and CCL2 levels remained unchanged. Treatment with IFNγ alone significantly elevated the expression of TNFa, CCL8, CXCL10, and CCL2. These outcomes align with the findings reported by Rustenhoven et al. (Sci Rep, 2016), suggesting that the functionality of our hiPSC-derived microglia cells closely mirrors that of primary human adult microglia cells.

(3) The resolution of Figs. 4 - 6 is so low that even some of the text and labels are hardly readable. Based on the morphology shown in Fig. 4 and the statement in line 147, these hiPSC-derived "cells altered their morphology to a rounded shape within an hour of incubation and rapidly internalized the fluorescent-labeled particles". This is a peculiar response. Usually, microglia do not respond to fluorescent-labeled zymosan by turning into a rounded shaped within an hour when they internalize them. Such a behavior usually implicates weak phagocytotic capacity.

Thank you for your insightful comments. During submission, the main text's PDF version was converted online, resulting in low-quality output. We have since updated this with a high-resolution version. The observed alterations in cell morphology following zymosan phagocytosis may be attributed to the high zymosan concentration used (2mg/ml). We conducted an assessment to understand the impact of zymosan concentration on the morphology of hiPSC-derived microglial cells, as shown in Figure 4 suppl B. Our findings indicate that microglia cells adopt an amoeboid, rounded shape at zymosan concentrations exceeding 20ug/ml. To clarify this point, we have amended the text to read: "The cells altered their morphology and rapidly internalized the fluorescent-labeled particles."

(4) Data presented in Fig. 5 are not very convincing to support that transplanted cells were immunopositive for "human CD11b (Fig.5C), as well as microglia signature markers P2ry12 and TMEM119 (Fig.5D)" (line 167). The resolution and magnification of Fig. 5D is too low to tell the colocalization of tdT and human microglial marker immunolabeling. In the flat-mount images (C, I), hCD11b immunolabeling is not visible in the GCL or barely visible in the IPL. This should be discussed.

We are grateful for the reviewer's comments. As previously mentioned, the low quality of the images was due to the online conversion of the PDF version. We have now submitted both high-quality PDF and Word versions for the reviewer's assessment. In these high-quality versions, the colocalization of tdT with human P2ry12 and TMEM119 is distinctly visible. Additionally, we have updated the hTMEM119 staining images in Figure 5D. The results from hCD11b staining align with those observed in mouse CD11b staining, notably showing more effective staining in the outer plexiform layer (OPL) microglia cells. The reason for this—whether it pertains to a staining issue, a variance in CD11b expression among microglia cells in the OPL and ganglion layer (GL), or differences in the samples due to varying conditions—is not yet clear and warrants further investigation.

(5) Microglia respond to injury by becoming active and lose their expression of the resting state microglial marker, such as P2ry12, which is used in Fig. 6 for detection of migrated microglia. To confirm that these cells indeed respond to injury like native microglia, one should check for activated microglial markers and induction of pro-inflammatory cytokines in the sodium iodate-injury model.

The reviewer's insights are spot-on. We utilized preserved retinas to extract mRNA, which was then reverse-transcribed to cDNA for conducting qRT-PCR using human-specific primers, as detailed in the updated Table 5. The findings revealed that following retinal pigment epithelium (RPE) injury for 3 days, the transplanted hiPSC-derived microglial cells exhibited an increase in the production of inflammatory cytokines and upregulated genes related to phagocytosis, migration, and adhesion. Conversely, there was a decrease in the expression of microglia-specific signature genes and neurotrophic factors, as demonstrated in Figure 7 suppl.

Reviewer #1 (Recommendations For The Authors):

Line 52: "Microglia cell repopulation research suggests that: 1) if no injury or infection occurs, retinal microglia cells can sustain their homeostasis indefinitely" - this statement is too strong or delivers a confusing message; it needs clarification or to be backed up by evidence. Recent single cell RNA sequencing analyses suggest that even under a normal condition, residential microglia do not present as a single homeostatic cell cluster, rather a subpopulation of activated inflammatory microglia are constantly detectable in the normal retina. This is likely because normal retinal neurons can be stressed due to various reasons, such as the temporal accumulation of misfolded proteins, exposed to strong light, or ageing, etc.

We appreciate the comments. We changed the sentence to read, "Microglia cell repopulation research suggests that: 1) retinal resident microglia cells can sustain their population with the local dividing and migration if any perturbations do not exceed the threshold of the recovery speed by local neighbor microglia cells."

Line 83: "we applied an appropriate protocol for culturing human iPSC-derived microglia cells" - it would be more appropriate if the word "appropriate" can be replaced by either "unique" or a phrase like "we adopted a (previously published) protocol...".

Thanks! We changed it to “We modified a previously published protocol to culture human iPSC-derived microglia cells.".

Fig. 1F,G: A method of flow cytometry will provide more comprehensive cell quantification for percentages of positively labeled cells than cell counts under high magnification confocal images.

Thanks for the comments! We agreed with the reviewer. Given the experimental resources available, the quantifications of confocal images did provide a reasonable assessment. We will perform flow cytometry analysis in future experiments.

Reviewer #2 (Public review):

Weaknesses:

Gene expression analysis of mature microglia cells should be better interpreted and it would be beneficial to compare the iPSC-derived microglia gene set to a human microglial cell line (for example, HMC3) instead of myeloid progenitor cells.<br /> The way that the manuscript has been written, unfortunately, is not optimal. I recommend that the entire manuscript be edited and proofread in English. The text contains spelling and grammar mistakes, and the manuscript is inconsistent in several parts. The manuscript should also be revised for a scientific paper format.

We appreciate the reviewer's comments and have taken them into consideration along with similar inquiries from Reviewer 1. Following the suggestions, we conducted a comparison of gene expression profiles between our hiPSC-derived microglia and those from fetal/adult brain microglia, as depicted in the updated Fig.2. Suppl. B, C and D; as well as in the Suppl. table 1 and table 2. The correlation analysis demonstrated that the hiPSC-derived microglia cells closely resemble fetal and adult brain microglia, significantly differing from monocytes and inflammatory monocytes. Additionally, we have revised the manuscript to adhere more closely to the conventional scientific format.

Reviewer #2 (Recommendations For The Authors):

Specific suggestions for improvement:

- Regarding the characterization of human iPSC-derived microglia, P2RY12 is a general hematopoietic cell marker. One cannot judge the maturity of microglia only by P2RY12 expression (for example, line 261). The expression of more specific markers such as TMEM119 and PROS1 should be studied and discussed.

We are thankful for the reviewer's valuable feedback. In response:

We have removed the term "mature" and clarified that the hiPSC-derived microglia we studied are fully differentiated within single-type cell culture conditions.

We performed a comparative analysis of the gene expression profiles between our hiPSC-derived microglia and microglia from human brains, as illustrated in the updated Fig.2. Suppl. B, C and D. The results affirm that hiPSC-derived microglia closely resemble human fetal and adult microglia.

We noted that the expression of TMEM119 in hiPSC-derived microglia under in vitro single-type cell culture conditions is notably low, as shown in the below A. This suggests that the stimulatory factors in our single-type cell culture might not sufficiently induce TMEM119 expression in microglia. The necessity for a retinal environment or interaction with neuronal and/or other glial cells for TMEM119 expression mirrors the behavior of infiltrating peripheral monocytes in pathological conditions, which initially lack TMEM119 but later differentiate into microglial-like macrophages that express TMEM119, as reported by Ma et al. in Sci Rep (2017).

Additionally, our findings suggest that PROS1 is not uniquely characteristic of microglia but is expressed across a variety of cell types. Within our specific culture conditions, we noted a higher expression of PROS1 in microglial progenitor cells, as shown in Author response image 1B and C.

Author response image 1.

- In Figure 2, Part E, the names of the genes or pathways in the figure are not clear, and are these genes the set that are the most differentially expressed between iPSCs-derived microglia and MPC? The analysis needs more explanation.

We regret any confusion caused by our previous explanation. To clarify, we compiled a list of microglia-enriched genes from the research conducted by Barres BA Lab (Bennett et al., Proc Natl Acad Sci U S A, 2016) and from our own RNA sequencing data of mouse retinal microglia, identifying a total of 130 genes predominantly expressed in microglia (Suppl. Table 3). We then applied this gene list to analyze our hiPSC-derived microglia RNA sequencing data, resulting in the identification of 71 microglia-specific genes. These 71 genes were subjected to Ingenuity Pathway Analysis (IPA) to visualize the signaling pathways involved. The details of these microglia genes can be found in the updated suppl. table 3.

- Lines 124 to 128 mention that high expression of Stat3, IL1b, and IL6 and their central role in pathway analysis emphasize the efficiency of the maturation protocol. Regarding the fact that Stat3, IL1b, and IL6 are contributors to proinflammatory pathways, it is not convincing that the high expression of these genes in iPSC-derived microglia demonstrates the efficiency of the maturation protocol, given that microglia are not stimulated.

Thanks for the comments! We added the sentences about the comparison results between hiPSC-derived microglia and human brain microglia. We have also replaced the “mature” with “functional.” The sentence reads, “Thus, our method of obtaining differentiated microglia is a reliable method to generate a large number of homogenous functional microglia cells.”

- Statistical analysis is missing for some graphs, for example, figures 1-3 and 5.

We appreciate the comments. We have added the statistical results in the revised version.

- The legend for Figure 3 needs to be rewritten. The graphs or applied assays should be explained in the legend, not the interpretation of the data.

The legend was rewritten.

- There is no Figure 3 in the supplement figures file.

We added Figure 3. Suppl.

- hTMEM119 staining in Figure 5, Part D, is mostly background. Please provide another image.

The images were unclear after on-line converting due to the low number of pixels. We replaced them with new hTMEM119 staining images in Figure 5D.

- In line 176, figure 5I has been forgotten to be mentioned.

Thank you very much! We added 5I.

- Lines 241 to 244 state that more than 50% of the AMD-associated genes are highly expressed in retinal microglia according to Fig. discussion suppl A & B. It is not clear that the gene set that was used for analysis is from a healthy retinal microglia or AMD-related ones. Please explain precisely.

Thank you for your feedback. The gene list we referenced originates from a Genome-Wide Association Study (GWAS) that compared patients with Age-related Macular Degeneration (AMD) to healthy cohorts. We did not directly utilize this list in our experiments but referred to it to underscore the importance of microglia cells in the context of AMD.

Some of the English proofreading and manuscript format comments:

Line 805: Iba1 is written in lowercase. Is it human IBA1? It is not consistent with the way it is written in the text (in line 117, for example).

Thank you for pointing out the error. We reformed all Iba1 as “Iba1”. The Iba1 we used here are all from Wako (#019–19741), which labels both mouse and human microglial cells.

Line 814: microglia-enriched gene expression instead of microglia-enrich gene expression

Thank you! We changed it.

Line 345: Starting a sentence with lower case letter.

Thank you! We changed it.

Line 342: Myeloid lineage instead of myeloid cell linage.

Thank you! We changed it.

Line 815: What does FPKM stand for? The abbreviations should be explained.

The FPKM is the abbreviation of Fragments Per Kilobase of transcript per Million mapped reads. We added it in the text.

Line 309: The manuscript has occasionally referred to PLX-5622 without a minus. Please follow a uniform format.

We changed all “PLX5622” to “PLX-5622”.

Lines 327-331: should be rewritten.

The mentioned paragraph was rewritten.

Lines 335-340: should be rewritten.

The mentioned sentence was rewritten.

Line 135: qRT-PCR instead of QPCR," as it is also mentioned in the methods and material. The correction also applies to all the QPCRs in the text.

We changed “QPCR” with “qRT-PCR”

Figure 3: Graph B should be right side of graph A

Images description: It is better to have the images description in the left side of the image, for example, figure 5 part B, GL, IPL and OPL

Thanks for the suggestion. We changed the image organization as per the reviewer’s advice.

Lines 258 to 260 in the discussion have also been repeated with the same words in the introduction.

The mentioned paragraph was rewritten.

Lines 327-331 should be rewritten.

The mentioned paragraph was rewritten.

Lines 335-340 should be rewritten.

The mentioned paragraph was rewritten.

eLife assessment

This useful study investigates the evolution of the Mycobacterium tuberculosis Complex (MTBC) pangenome using state-of-the-art bioinformatic tools to analyse 324 complete and 11 new genomes representing all known lineages and sublineages, yielding data indicating that MTBC has a closed pangenome with relatively few accessory genes. The strength of the evidence is solid for gene presence-absence analysis and inadequate for the deletion bias claim. Their conclusions regarding pangenome evolution being driven by deletions in sublineage-specific regions of difference are difficult to interpret due to the description of methods not being complete and data interpretation not being adequate.

Reviewer #1 (Public Review):

Summary:

In this paper, Behruznia and colleagues use long-read sequencing data for 335 strains of the Mycobacterium tuberculosis complex to study genome evolution in this clonal bacterial pathogen. They use both a "classical" pangenome approach that looks at the presence and absence of genes, and a more general pangenome graph approach to investigate structural variants also in non-coding regions. The two main results of the study are that (1) the MTBC has a small pangenome with few accessory genes, and that (2) pangenome evolution is driven by deletions in sublineage-specific regions of difference. Combining the gene-based approach with a pangenome graph is innovative, and the former analysis is largely sound apart from a lack of information about the data set used. The graph part, however, requires more work and currently fails to support the second main result. Problems include the omission of important information and the confusing analysis of structural variants in terms of "regions of difference", which unnecessarily introduces reference bias. Overall, I very much like the direction taken in this article, but think that it needs more work: on the one hand by simply telling the reader what exactly was done, on the other by taking advantage of the information contained in the pangenome graph.

Strengths:

The authors put together a large data set of long-read assemblies representing most lineages of the Mycobacterium tuberculosis context, covering a large geographic area. State-of-the-art methods are used to analyze gene presence-absence polymorphisms (Panaroo) and to construct a pangenome graph (PanGraph). Additional analysis steps are performed to address known problems with misannotated or misassembled genes in pangenome analysis.

Weaknesses:

The study does not quite live up to the expectations raised in the introduction. Firstly, while the importance of using a curated data set is emphasized, little information is given about the data set apart from the geographic origin of the samples (Figure 1). A BUSCO analysis is conducted to filter for assembly quality, but no results are reported. It is also not clear whether the authors assembled genomes themselves in the cases where, according to Supplementary Table 1, only the reads were published but not the assemblies. In the end, we simply have to trust that single-contig assemblies based on long-reads are reliable.

One issue with long read assemblies could be that high rates of sequencing errors result in artificial indels when coverage is low, which in turn could affect gene annotation and pangenome inference (e.g. Watson & Warr 2019, https://doi.org/10.1038/s41587-018-0004-z). Some of the older long-read data used by the authors could well be problematic (PacBio RSII), but also their own Nanopore assemblies, six of which have a mean coverage below 50 (Wick et al. 2023 recommend 200x for ONT, https://doi.org/ 10.1371/journal.pcbi.1010905). Could the results be affected by such assembly errors? Are there lineages, for example, for which there is an increased proportion of RSII data? Given the large heterogeneity in data quality on the NCBI, I think more information about the reads and the assemblies should be provided.

The part of the paper I struggled most with is the pangenome graph analysis and the interpretation of structural variants in terms of "regions of difference". To start with, the method section states that "multiple whole genomes were aligned into a graph using PanGraph" (l.159/160), without stating which genomes were for what reason. From Figure 5 I understand that you included all genomes, and that Figure 6 summarizes the information at the sublineage level. This should be stated clearly, at present the reader has to figure out what was done. It was also not clear to me why the authors focus on the sublineage level: a minority of accessory genes (107 of 506) are "specific to certain lineages or sublineages" (l. 240), so why conclude that the pangenome is "driven by sublineage-specific regions of difference", as the title states? What does "driven by" mean? Instead of cutting the phylogeny arbitrarily at the sublineage level, polymorphisms could be described more generally by their frequencies.

I fully agree that pangenome graphs are the way to go and that the non-coding part of the genome deserves as much attention as the coding part, as stated in the introduction. Here, however, the analysis of the pangenome graph consists of extracting variants from the graph and blasting them against the reference genome H37Rv in order to identify genes and "regions of difference" (RDs) that are variable. It is not clear what the authors do with structural variants that yield no blast hit against H37Rv. Are they ignored? Are they included as new "regions of difference"? How many of them are there? etc. The key advantage of pangenome graphs is that they allow a reference-free, full representation of genetic variation in a sample. Here reference bias is reintroduced in the first analysis step.

Along similar lines, I find the interpretation of structural variants in terms of "regions of difference" confusing, and probably many people outside the TB field will do so. For one thing, it is not clear where these RDs and their names come from. Did the authors use an annotation of RDs in the reference genome H37Rv from previously published work (e.g. Bespiatykh et al. 2021)? This is important basic information, its lack makes it difficult to judge the validity of the results. The Bespiatykh et al. study uses a large short-read data (721 strains) set to characterize diversity in RDs and specifically focuses on the sublineage-specific variants. While the authors cite the paper, it would be relevant to compare the results of the two studies in more detail.

As far as I understand, "regions of difference" have been used in the tuberculosis field to describe structural variants relative to the reference genome H37Rv. Colloquially, regions present in H37Rv but absent in another strain have been called "deletions". Whether these polymorphisms have indeed originated through deletion or through insertion in H37Rv or its ancestors requires a comparison with additional strains. While the pangenome graph does contain this information, the authors do not attempt to categorize structural variants into insertions and deletions but simply seem to assume that "regions of difference" are deletions. This, as well as the neglect of paralogs in the "classical" pangenome analysis, puts a question mark behind their conclusion that deletion drives pangenome evolution in the MTBC.

Reviewer #2 (Public Review):

Summary:

The authors attempted to investigate the pangenome of MTBC by using a selection of state-of-the-art bioinformatic tools to analyse 324 complete and 11 new genomes representing all known lineages and sublineages. The aim of their work was to describe the total diversity of the MTBC and to investigate the driving evolutionary force. By using long read and hybrid approaches for genome assembly, an important attempt was made to understand why the MTBC pangenome size was reported to vary in size by previous reports.

Strengths:

A stand-out feature of this work is the inclusion of non-coding regions as opposed to only coding regions which was a focus of previous papers and analyses which investigated the MTBC pangenome. A unique feature of this work is that it highlights sublineage-specific regions of difference (RDs) that were previously unknown. Another major strength is the utilisation of long-read whole genomes sequences, in combination with short-read sequences when available. It is known that using only short reads for genome assembly has several pitfalls. The parallel approach of utilizing both Panaroo and Pangraph for pangenomic reconstruction illuminated the limitations of both tools while highlighting genomic features identified by both. This is important for any future work and perhaps alludes to the need for more MTBC-specific tools to be developed.

Weaknesses:

The only major weakness was the limited number of isolates from certain lineages and the over-representation others, which was also acknowledged by the authors. However, since the case is made that the MTBC has a closed pangenome, the inclusion of additional genomes would not result in the identification of any new genes. This is a strong statement without an illustration/statistical analysis to support this.

Author response:

Reviewer #1 (Public Review): 

Summary: 

In this paper, Behruznia and colleagues use long-read sequencing data for 335 strains of the Mycobacterium tuberculosis complex to study genome evolution in this clonal bacterial pathogen. They use both a "classical" pangenome approach that looks at the presence and absence of genes, and a more general pangenome graph approach to investigate structural variants also in non-coding regions. The two main results of the study are that (1) the MTBC has a small pangenome with few accessory genes, and that (2) pangenome evolution is driven by deletions in sublineage-specific regions of difference. Combining the gene-based approach with a pangenome graph is innovative, and the former analysis is largely sound apart from a lack of information about the data set used. The graph part, however, requires more work and currently fails to support the second main result. Problems include the omission of important information and the confusing analysis of structural variants in terms of "regions of difference", which unnecessarily introduces reference bias. Overall, I very much like the direction taken in this article, but think that it needs more work: on the one hand by simply telling the reader what exactly was done, on the other by taking advantage of the information contained in the pangenome graph. 

Thank you for your constructive feedback. We have hopefully positively addressed all your concerns. Please see our detailed responses below.

Strengths: 

The authors put together a large data set of long-read assemblies representing most lineages of the Mycobacterium tuberculosis context, covering a large geographic area. State-of-the-art methods are used to analyze gene presence-absence polymorphisms (Panaroo) and to construct a pangenome graph (PanGraph). Additional analysis steps are performed to address known problems with misannotated or misassembled genes in pangenome analysis. 

Thank you for your positive feedback. We are pleased that you found these aspects of our work noteworthy and valuable.

Weaknesses: 

The study does not quite live up to the expectations raised in the introduction. Firstly, while the importance of using a curated data set is emphasized, little information is given about the data set apart from the geographic origin of the samples (Figure 1). A BUSCO analysis is conducted to filter for assembly quality, but no results are reported. It is also not clear whether the authors assembled genomes themselves in the cases where, according to Supplementary Table 1, only the reads were published but not the assemblies. In the end, we simply have to trust that single-contig assemblies based on long-reads are reliable. 

The BUSCO results are present for all the genomes in Supplementary Table S1. Genome assemblies were obtained from public databases and other studies that performed the assemblies. We did not perform assemblies for any of the public datasets except the 11 genomes sequenced by ourselves, for which we included the assembly statistics. The public genomes from NCBI were marked as closed based on the NCBI pipelines so there are additional checks on quality undertaken there before we included in our analysis. Marin et al (2024; BioRxiv) also performed additional checks on the vast majority of the genomes before they were included here.  We are confident that these genomes represent the highest quality M. tuberculosis dataset possible, but we will check that all genomes are present in the GTDB list, which performs additional tests including CheckM, to add another layer of confidence. Some of the accessions to the final genomes were not included as these papers were not released yet but will be in the next version. Supplementary Table S1 will be updated to include the assembly information for each genome.

One issue with long read assemblies could be that high rates of sequencing errors result in artificial indels when coverage is low, which in turn could affect gene annotation and pangenome inference (e.g. Watson & Warr 2019, https://doi.org/10.1038/s41587-018-0004-z). Some of the older long-read data used by the authors could well be problematic (PacBio RSII), but also their own Nanopore assemblies, six of which have a mean coverage below 50 (Wick et al. 2023 recommend 200x for ONT, https://doi.org/ 10.1371/journal.pcbi.1010905). Could the results be affected by such assembly errors? Are there lineages, for example, for which there is an increased proportion of RSII data? Given the large heterogeneity in data quality on the NCBI, I think more information about the reads and the assemblies should be provided. 

We have shown elsewhere (Marin et al (2024; BioRxiv)) that short read sequencing is significantly worse for these types of problems. For this reason, we have included only closed genomes which we believe will reduce the potential for such errors. However, we agree that older sequencing technologies, such as PacBio RSII, can introduce errors in the assemblies and subsequent downstream analyses. We will look for correlation between platform and accessory genome presence/absence to see if the type of sequencing influences the results.

Wick et al. (2023) recommend a coverage of 200x for ONT sequencing; however, newer analyses from Wick have shown that with modern basecalling and sequencing very low error rates can be achieved with much lower coverage (see https://rrwick.github.io/2023/10/24/ont-only-accuracy-update.html). We are quite confident that gene presence/absence patterns should be robust to this in our analysis but will confirm with some additional analysis on our sequenced genomes.

The part of the paper I struggled most with is the pangenome graph analysis and the interpretation of structural variants in terms of "regions of difference". To start with, the method section states that "multiple whole genomes were aligned into a graph using PanGraph" (l.159/160), without stating which genomes were for what reason. From Figure 5 I understand that you included all genomes, and that Figure 6 summarizes the information at the sublineage level. This should be stated clearly, at present the reader has to figure out what was done.

All genomes were included in the pangenome graph construction and to look for regions of differences. We then grouped genomes into sub-lineages to undertake the additional analyses as there is not enough genomes per sub-sub-lineages and lower for robust analyses. We will make this clearer in the next version, likely with a flowchart of analyses.

It was also not clear to me why the authors focus on the sublineage level: a minority of accessory genes (107 of 506) are "specific to certain lineages or sublineages" (l. 240), so why conclude that the pangenome is "driven by sublineage-specific regions of difference", as the title states? What does "driven by" mean? Instead of cutting the phylogeny arbitrarily at the sublineage level, polymorphisms could be described more generally by their frequencies. 

We acknowledge the importance of polymorphisms, but our study primarily aimed to investigate the presence and absence of genes/genomic regions, as highlighted in our focus on structural differences rather than SNPs (L67-69). We attempted to clarify our goal of exploring gene content variation both between and within lineages (L69) to avoid confusion.

Our focus on the sub-lineage level addresses the gap in understanding gene content distribution beyond the broad lineage level, where previous pangenome studies have concentrated. The decision to focus on sub-lineages allows for a more detailed exploration of genetic diversity. Due to the limited number of genomes available to represent all sub-sub-lineages and lower levels of classification, we aimed to investigate gene content differences at the sub-lineage level. This decision allows for a more detailed and comprehensive exploration of gene content differences within the MTBC.

I fully agree that pangenome graphs are the way to go and that the non-coding part of the genome deserves as much attention as the coding part, as stated in the introduction. Here, however, the analysis of the pangenome graph consists of extracting variants from the graph and blasting them against the reference genome H37Rv in order to identify genes and "regions of difference" (RDs) that are variable. It is not clear what the authors do with structural variants that yield no blast hit against H37Rv. Are they ignored? Are they included as new "regions of difference"? How many of them are there? etc. The key advantage of pangenome graphs is that they allow a reference-free, full representation of genetic variation in a sample. Here reference bias is reintroduced in the first analysis step. 

Genomic analysis of Mycobacterium tuberculosis is H37Rv reference-centric, meaning that RDs are typically defined based on their presence or absence relative to the reference strain. Our approach comparing variants to the H37Rv reference was primarily to identify and name the known regions of differences (RDs). For structural variants that did not yield a BLAST hit against H37Rv, we assigned them as new RDs in Supplementary Table S4 to provide a reference-free approach for investigating gene content differences. Further clarifications on the definition and identification of RDs will be added.

Along similar lines, I find the interpretation of structural variants in terms of "regions of difference" confusing, and probably many people outside the TB field will do so. For one thing, it is not clear where these RDs and their names come from. Did the authors use an annotation of RDs in the reference genome H37Rv from previously published work (e.g. Bespiatykh et al. 2021)? This is important basic information, its lack makes it difficult to judge the validity of the results. The Bespiatykh et al. study uses a large short-read data (721 strains) set to characterize diversity in RDs and specifically focuses on the sublineage-specific variants. While the authors cite the paper, it would be relevant to compare the results of the two studies in more detail. 

Indeed the term regions of difference (RDs) is somewhat M. tuberculosis specific. These are large polymorphisms which are differentially present in clades (primarily lineages) of M. tuberculosis. Annotations and naming of these is based on Bespiatykh et al. (2021) and RDscan tool which identify RD regions based on the H37Rv genomic coordinates. We obtained the corresponding Rv locus for RD regions by matching their genomic coordinates on the H37Rv genome and confirmed the RDs using the bed file from RDscan. We have used their names where our findings overlap and any new RDs we report are not found in their data. We will ensure this is clearer in the next version.

As far as I understand, "regions of difference" have been used in the tuberculosis field to describe structural variants relative to the reference genome H37Rv. Colloquially, regions present in H37Rv but absent in another strain have been called "deletions". Whether these polymorphisms have indeed originated through deletion or through insertion in H37Rv or its ancestors requires a comparison with additional strains. While the pangenome graph does contain this information, the authors do not attempt to categorize structural variants into insertions and deletions but simply seem to assume that "regions of difference" are deletions. This, as well as the neglect of paralogs in the "classical" pangenome analysis, puts a question mark behind their conclusion that deletion drives pangenome evolution in the MTBC. 

The term regions of difference or RDs has traditionally been used to describe structural variants relative to the H37Rv genome, often interpreted as deletions. Consistent with our study, Bespiatykh et al. (2021) observed two types of deletions: those associated with repeat sequences or mobile genetic elements, and conserved RDs that are phylogenetically informative deletions inherited by all descendants of a strain.

In our study, we employed a phylogenetic approach to identify deletions. If RDs are present in genomes both upstream and downstream of a phylogenetic branch but are absent in one specific branch, we interpret this as evidence of gene deletion (Figure 5B). This method was systematically applied to all RDs identified as deletions in our study; we will clarify this better in the next version.

We acknowledge the importance of considering paralogs in pangenome analysis. While the evolution of genomes is driven by duplication, loss and transfer, we know that transfer is not a mechanism in modern MTBC evolution and we have focussed here on loss. Duplication (paralog) analysis from annotations continues to be difficult to quantify due to the difficult of reliably confirming paralogy. We have addressed the effect of different Panaroo options, including merge paralogs, on the genomic diversity and pangenome estimation of MTBC in our associated paper (Marin et al 2024). This study showed that most structural variation in Mycobacterium tuberculosis is attributed to rearrangements of existing sequences rather than novel sequence content. For example, the transposable element IS6110 accounts for a significant portion of sequence variation. This hints that paralogs are not very important in terms of gene content differences in MTBC.

However, we will attempt to build on this by looking at Panaroo outputs without merged paralogs and looking for potentially duplicated genomic stretches in the Pangraph analyses. This will hopefully show more robustly that the MTBC diversity is primarily deletion driven.

Reviewer #2 (Public Review): 

Summary: 

The authors attempted to investigate the pangenome of MTBC by using a selection of state-of-the-art bioinformatic tools to analyse 324 complete and 11 new genomes representing all known lineages and sublineages. The aim of their work was to describe the total diversity of the MTBC and to investigate the driving evolutionary force. By using long read and hybrid approaches for genome assembly, an important attempt was made to understand why the MTBC pangenome size was reported to vary in size by previous reports. 

Strengths: 

A stand-out feature of this work is the inclusion of non-coding regions as opposed to only coding regions which was a focus of previous papers and analyses which investigated the MTBC pangenome. A unique feature of this work is that it highlights sublineage-specific regions of difference (RDs) that were previously unknown. Another major strength is the utilisation of long-read whole genomes sequences, in combination with short-read sequences when available. It is known that using only short reads for genome assembly has several pitfalls. The parallel approach of utilizing both Panaroo and Pangraph for pangenomic reconstruction illuminated the limitations of both tools while highlighting genomic features identified by both. This is important for any future work and perhaps alludes to the need for more MTBC-specific tools to be developed. 

Thank you for recognising the strengths of our work.

Weaknesses: 

The only major weakness was the limited number of isolates from certain lineages and the over-representation others, which was also acknowledged by the authors. However, since the case is made that the MTBC has a closed pangenome, the inclusion of additional genomes would not result in the identification of any new genes. This is a strong statement without an illustration/statistical analysis to support this. 

The language around open and closed pangenomes is difficult to convey and indeed we will improve this for the next version. We aimed to show that with a set of highly curated genomes that span the breadth of known diversity within the MTBC, we see no evidence for a large, open pangenome as has been previously suggested. We instead suggest that adding new genomes is unlikely to bring large additions to the accessory genome, therefore showing that the MTBC pangenome tends towards being closed. We will add additional visualisations such as gene accumulation plots to better support this argument.

Reviewer #2 (Public Review):

Summary:

The study explores a new strategy of lysin-derived antimicrobial peptide-primed screening to find peptidoglycan hydrolases from bacterial proteomes. Using this strategy authors identified five peptidoglycan hydrolases from A. baumannii. They further tested their antimicrobial activities on various Gram-positive and Gram-negative pathogens.

Strengths:

Overall, the study is good and adds new members to the peptidoglycan hydrolases family. The authors also show that these lysins have bactericidal activities against both Gram-positive and Gram-negative bacteria. The crystal structure data is good, and reveals different thermostablility to the peptidoglycan hydrolases. Structural data also reveals that PhAb10 and PHAb11 form thermostable dimers and data is corroborated by generating variant protein defective in supporting intermolecular bond pairs. The mice bacterial infection shows promise for the use of these hydrolases as antimicrobial agents.

Weaknesses:

While the authors have employed various mechanisms to justify their findings, some aspects are still unclear. Only CFU has been used to test bactericidal activity. This should also be corroborated by live/dead assay. Moreover, SEM or TEM analysis would reveal the effect of these peptidoglycan hydrolases on Gram-negative /Gram-positive cell envelopes. The authors claim that these hydrolases are similar to T4 lysozyme, but they have not correlated their findings with already published findings on T4 lysozyme. T4 lysozyme has a C-terminal amphipathic helix with antimicrobial properties. Moreover, heat, denatured lysozyme also shows enhanced bactericidal activity due to the formation of hydrophobic dimeric forms, which are inserted in the membrane. Authors also observe that heat-denatured PHAb10 and PHAb11 have bactericidal activity but no enzymatic activity. These findings should be corroborated by studying the effect of these holoenzymes/ truncated peptides on bacterial cell membranes. Also, a quantitative peptidoglycan cleavage assay should be performed in addition to the halo assay. Including these details would make the work more comprehensive.

eLife assessment

This valuable study explores a new strategy of lysin-derived antimicrobial peptide-primed screening to find peptidoglycan hydrolases from bacterial proteomes. Using this strategy, the authors identified five peptidoglycan hydrolases from Acinetobacter baumannii, which they tested on various Gram-positive and Gram-negative pathogens for antimicrobial activity. The data presented are solid and will be of interest to microbiologists.

Reviewer #1 (Public Review):

Summary:

Li Zhang et al. characterized two new Gram-negative endolysins identified through an AMP-targeted search in bacterial proteomes. These endolysins exhibit broad lytic activity against both Gram-negative and Gram-positive bacteria and retain significant antimicrobial activity even after prolonged exposure to high temperatures (100{degree sign}C for 1 hour). This stability is attributed to a temperature-reversible transition from a dimer to a monomer. The authors suggest several potential applications, such as complementing heat sterilization processes or being used in animal feed premixes that undergo high-temperature pelleting, which I agree with.

Strengths:

The claims are well-supported by relevant and complementary assays, as well as extensive bioinformatic analyses.

Weaknesses:

There are numerous statements in the introduction and discussion sections that I currently do not agree with and consider need to be addressed. Therefore, I recommend major revisions.

Major comments:

Introduction and Discussion:

The introduction and the discussion are currently too general and not focused. Furthermore, there are some key concepts that are missing and are important for the reader to have an overview of the current state-of-the-art regarding endolysins that target gram-negatives. Specifically, the concepts of 'Artilysins', 'Innolysins', and 'Lysocins' are not introduced. Besides this, the authors do not mention other high-throughput mining or engineering strategies for endolysins, such as e.g. the VersaTile platform, which was initially developed by Hans Gerstmans et al. for one of the targeted pathogens in this manuscript (i.e., Acinetobacter baumannii). Recent works by Niels Vander Elst et al. have demonstrated that this VersaTile platform can be used to high-throughput screen and hit-to-lead select endolysins in the magnitude tens of thousands. Lastly, Roberto Vázquez et al. have recently demonstrated with bio-informatic analyses that approximately 30% of Gram-negative endolysin entries have AMP-like regions (hydrophobic short sequences), and that these entries are interesting candidates for further wet lab testing due to their outer membrane penetrating capacities. Therefore, I fully disagree with the statement being made in the introduction that endolysin strategies to target Gram-negatives are 'in its infancy' and I urge the authors to provide a new introduction that properly gives an overview of the Gram-negative endolysin field.

Results:

It should be mentioned that the halo assay is a qualitative assay for activity testing. I personally do not like that the size of the halos is used to discriminate in endolysin activity. In this reviewer's opinion, the size of the halo is highly dependent on (i) the molecular size of the endolysin as smaller proteins can diffuse further in the agar, and (ii) the affinity of the CBD subdomain of the endolysin for the bacterial peptidoglycan. It should also be said that in the halo assay, there is a long contact time between the endolysin and the bacteria that are statically embedded in the agar, which can result in false positive results. How did the authors mitigate this?

Testing should have been done at equimolar concentrations. If the authors decided to e.g. test 50 µg/mL for each protein, how was this then compensated for differences in molecular weight? For example, if PHAb10 and PHAb11 have smaller molecular sizes than PHAb7, 8, and 9, there is more protein present in 50 µg/mL for the first two compared to the others, and this would explain the higher decrease in bacterial killing (and possibly the larger halos).

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

The authors show that upon treatment with Doxorubicin (Doxo), there is an increase in senescence and inflammatory markers in the muscles. They also show these genes get upregulated in C2C12 myoblasts when treated with conditioned media or 15d-PGJ2. 15dPGJ2 induces cell death in the myoblasts, decreases proliferation (measured by cell numbers), and decreases differentiation and fusion. 15d-PGJ2 modified Cys184 of HRas, which is required for its activation as indicated by the FRET analysis with RAF RBD. They also showed that 15d-PGJ2 activates ERK signaling, but not Akt signaling, through the electrophilic center. 15d-PGJ2 inhibits Golgi localization of HRAS (only WT, not C181 or C184 mutant). They also showed that expressing the WT HRas followed by 15d-PGJ2 treatment led to a decrease in the levels of MHC mRNA and protein, and this defect is dependent on C184. This is a well-written manuscript with interesting insights into the mechanism of action of 15d-PGJ2. However, some clarification and experiments will help the paper advance the field significantly.

Strengths:

The data clearly shows that 15d-PGJ2 has a negative role in the myoblast cells and that it leads to modification of HRas protein. Moreover, the induction of biosynthetic enzymes in the PGD2 pathway also supports the induction of 15d-PGJ2 in Doxorubicin-treated cells. Both conditioned media experiments and the 15d-PGJ2 experiments show that 15d-PGJ2 could be the active component secreted by the senescent myoblasts.

Weaknesses:

The genes that are upregulated in the muscles upon injection with Doxo are also markers for inflammation. Since Doxo is also known to induce systemic inflammation, it is important to delineate these two effects (Inflammatory cells vs senescent cells). The expression of beta Gal and other markers of senescence in the tissue sections will help to delineate these.

As pointed out Doxo induces systemic inflammation along with inducing DNA damage-mediated senescence. Therefore, along with the inflammatory markers of the SASP (CXCL1/2, TNF1α, IL6, PTGS1/2, PTGDS) we also observed an increase in the mRNA levels of canonical markers of DNA damage-mediated senescence. We observed an increase in the mRNA levels of cell cycle and senescence associated proteins p16 and p21 (Fig. 1C). We also observed an increased nuclear accumulation of p21 (Fig. 1A) and increased levels of phosphorylated H2A.X in the nucleus (Fig. 1B).

In Figure 2, where the defect in the differentiation of myoblasts upon treatment with 15d-PGJ2 is shown, most of the cells die within 48 hours at higher concentrations, making it difficult to perform the experiments. This also shows that 15d-PGJ2 was toxic to these cells. Lower concentrations show a decrease in the differentiation based on the lower number of nuclei in fibers and low expression of MyoD, MyoG, and MHC. However, it is unclear if this is due to increased cell death or defective differentiation. It would be a lot more informative if the cell count, cell division, and cell death could be plotted for these concentrations of the drug during the experiment.

We measured the viability of C2C12 cells after 24 hours of treatment with 15d-PGJ2 using the MTT assay and observed that the viability of cells was decreased after treatment with 15d-PGJ2 (10 µM) but not with 15d-PGJ2 (1 µM, 2 µM, 4 µM, or 5 µM) (see Fig. S2A of the updated manuscript). The results and figures of the manuscript have been updated accordingly.

Also, in the myoblast experiments, are the effects of treatment with Dox reversible?

The treatment with Doxorubicin is irreversible as the senescent phenotype was not reversed after withdrawal of Doxorubicin, even after 20 days.

In Figure 3, most of the experiments are done at a high concentration, which induces almost complete cell death within 48 hours.

Figure 3 is an acute experiment for only 1 hour, at which time no cell death was observed. Specifically, we measured the phosphorylation of Erk and Akt proteins after 1 hour of treatment with 15d-PGJ2 (10 µM) during which we did not observe any cell death.

Even at such a high concentration of 15dPGJ2, the increase in ERK phosphorylation is minimal.

We observe a ~30% increase in the phosphorylation of Erk proteins after treatment with 15d-PGJ­2 in 0.2% serum medium compared to treatment with vehicle (DMSO). This is reproducible and significant.

The experiment Figure 4C shows that C181 and C84 mutants of the HRas show higher levels in Golgi compared with WT. However, this could very well be due to the defect in palmitoylation rather than the modification with 15d-PGJ2.

Our data does not suggest higher levels of C184S mutant in the Golgi compared with WT (Fig. S4A). We observed that the ratio of HRas levels in the Golgi to the HRas levels in the plasma membrane were similar in C2C12 cells expressing HRas C184S and HRas WT (Fig. S4A graph columns 1 and 5).

Though the authors allude to the possibility that intracellular redistribution of HRas by 15d-PGJ2 requires C181 palmitoylation, the direct influence of C184 modification on C181 palmitoylation is not shown. To have a meaningful conclusion, the authors need to compare the palmitoylation and modification with 15d-PGJ2.

Palmitoylation of HRas C181S is required for the localization of HRas at the plasma membrane. The inhibition of palmitoylation of C181, either by mutation (C181S) or treatment with protein palmitoyl transferase inhibitor (2-Bromopalmitate), results in the accumulation of HRas at Golgi(Rocks et al., 2005) (Fig. S4A). Modification of HRas at C184 by 15d-PGJ2 (Fig. 3A) could inhibit the palmitoylation of HRas at C181. However, our data does not support this hypothesis as modification of HRas WT by 15d-PGJ2 does not increase the level of HRas at the Golgi, like in the case of inhibition of cysteine palmitoylation due to C181S mutation.

To test if the inhibition of myoblast differentiation depends on HRas, they overexpressed the HRas and mutants in the C2C12 lines. However, this experiment does not take the endogenous HRAs into consideration, especially when interpreting the C184 mutant. An appropriate experiment to test this would be to knock down or knock out HRas (or make knock-in mutations of C184) and show that the effect of 15d-PGJ2 disappears.

Endogenous HRas (wild type) is present in the C2C12 cells overexpressing the EGFP-tagged HRas constructs. Therefore, we only observe a partial rescue in the differentiation after 15d-PGJ2 treatment in C2C12 cells expressing the C184S mutant (Fig. 4D and E). However, since HRas is expressed under high expression CMV promoter and in the absence of other regulatory elements, the overexpressed constructs do show a dominant effect over the endogenous HRas, showing cysteine mutant dependent inhibition of differentiation of myoblasts after treatment with 15d-PGJ2 (Fig. 4D and E).

Moreover, in this specific experiment, it is difficult to interpret without a control with no HRas construct and another without the 15d-PGJ2 treatment.

The mRNA levels of MyoD, MyoG, and MHC in C2C12 cells expressing HRas constructs after treatment with 15d-PGJ2 were normalized to the mRNA levels in C2C12 cells expressing corresponding constructs and were treated with vehicle (DMSO). mRNA levels in C2C12 cells treated with vehicle were not shown as they were normalized to 1. MHC protein levels in C2C12 cells expressing HRas constructs after 15d-PGJ2 treatment were normalized to that in C2C12 cells treated with vehicle (DMSO). Since the hypothesis to study the effect of HRas cysteine mutations on the differentiation of myoblasts after treatment with 15d-PGJ2, C2C12 cells expressing HRas WT serve as adequate control. Fig. 2 shows the effect of 15d-PGJ2 on muscle differentiation when HRas was not overexpressed.

Moreover, the overall study does not delineate the toxic effects of 15d-PGJ2 from its effect on the differentiation.

The inhibition of differentiation in C212 cells after treatment with 15d-PGJ2 cannot be attributed to the general toxicity of 15d-PGJ2 in cells. We show that the inhibition of differentiation of myoblasts after 15d-PGJ2 depends on modification of HRas at C184 i.e. failure to modify HRas at C184 (Fig. 3A) and resultant activation (Fig. 3B) by 15d-PGJ2 rescues this inhibition of differentiation of C2C12 cells (Fig. 4D and E), dissecting the inhibition of differentiation of myoblasts by 15d-PGJ2 from general toxic effects of 15d-PGJ2 on cell physiology.

Please note that the effect of 15d-PGJ2 on cell physiology is context-specific. On one hand, 15d-PGJ2 has been shown to exert tumor-suppressor effects by inhibiting the proliferation of ovarian cancer cells and lung adenocarcinoma cells (de Jong et al., 2011; Slanovc et al., 2024), 15d-PGJ2 also exerts pro-carcinogenic effects by induction of epithelial to mesenchymal transition in breast cancer cells MCF7 and inhibition of tumor-suppressor protein p53 in MCF7 and PC-3 cells (Choi et al., 2020; Kim et al., 2010).

Reviewer #2 (Public Review):

Summary:

In this study, Swarang and colleagues identified the lipid metabolite 15d-PGJ2 as a potential component of senescent myoblasts. They proposed that 15d-PGJ2 inhibits myoblast proliferation and differentiation by binding and regulating HRas, suggesting its potential as a target for restoring muscle homeostasis post-chemotherapy.

Strengths:

The regulation of HRas by 15d-PGJ2 is well controlled.

Weaknesses:

(1) I still think the novelty is limited by previous published findings. The authors themselves noted that the accumulation of 15d-PGJ2 in senescent cells has been reported in various cell types, including human fibroblasts, HEPG2 hepatocellular carcinoma cells, and HUVEC endothelial cells (PMCID: PMC8501892). Although the current study observed similar activation of 15d-PGJ2 in myoblasts, it appears to be additive rather than fundamentally novel. The covalent adduct of 15d-PGJ2 with Cys-184 of H-Ras was reported over 20 years ago (PMID: 12684535), and the biochemical principles of this interaction are likely universal across different cell types. The regulation of myogenesis by both HRas and 15d-PGJ2 has also been previously extensively reported (PMID: 2654809, 1714463, 17412879, 20109525, 11477074). The main conceptual novelty may lie in the connection between these points in myoblasts. But as discussed in another comment, the use of C2C12 cells as a model for senescence study is questionable due to the lack of the key regulator p16. The findings in C2C12 cells may not accurately represent physiological-relevant myoblasts. It is recommended that these findings be validated in primary myoblasts to strengthen the study's conclusions.

This is the first study to show a molecular mechanism where activation of HRas signaling in skeletal myoblasts due to covalent modification by 15d-PGJ2 at C184 of HRas inhibits the differentiation of skeletal myoblasts.

(2) The C2C12 cell line is not an ideal model for senescence study.

C2C12 cells are a well-established model for studying myogenesis. However, their suitability as a model for senescence studies is questionable. C2C12 cells are immortalized and do not undergo normal senescence like primary cells as C2C12 cells are known to have a deleted p16/p19 locus, a crucial regulator of senescence (PMID: 20682446). The use of C2C12 cells in published studies does not inherently validate them as a suitable senescence model. These studies may have limitations, and the appropriateness of the C2C12 model depends on the specific research goals.

Several reports have shown that cells undergo senescence independent of p16 expression. MCF7 human breast adenocarcinoma cells have been shown to undergo DNA damage mediated and Oncogene induced senescence as seen after treatment with Doxorubicin (PMID: PMC7025418) and expression of constitutively active HRas (PMID: 17135242), despite the homozygous deletion of p16 locus (ISBN 9780124375512 Chapter 17 Table 2) by upregulation of cell cycle inhibitor protein p21. In this study, we observe an increase in the senescence markers in C2C12 cells after treatment with Doxo (Fig. 1). We also observed an increase in the markers of DNA damage-mediated senescence in MCF7 after treatment with Doxo (Data will be included in the revised manuscript). Based on these observations, we have concluded that C2C12 cells undergo senescence despite lacking the p16/p19 locus.

In the study by Moustogiannis et al. (PMID: 33918414), they claimed to have aged C2C12 cells through multiple population doublings. However, the SA-β-gal staining in their data, which is often used to confirm senescence, showed almost fully confluent "aged" C2C12 cells. This confluent state could artificially increase SA-β-gal positivity, suggesting that these cells may not truly represent senescence. Moreover, the "aged" C2C12 cells exhibited normal proliferation, which contradicts the definition of senescence. Similar findings were reported in another study of C2C12 cells subjected to 58 population doublings (PMID: 21826704), where even at this late stage, the cells were still dividing every 2 or 3 days, similar to younger cells at early passages. More importantly, I do know how the p16 was detected in that paper since the locus was already mutated. In terms of p21, there was no difference in the proliferative C2C12 cells at day 0.

In the study by Moiseeva et al. in 2023 (PMID: 36544018), C2C12 cells were used for senescence modeling for siRNA transfection. However, the most significant findings were obtained using primary satellite cells or confirmed with complementary data.

In conclusion, while molecular changes observed in studies using C2C12 cells may be valid, the use of primary myoblasts is highly recommended for senescence studies due to the limitations and questionable senescence characteristics of the C2C12 cell line.

(3) Regarding source of increased PGD in the conditioned medium, I want to emphasize that it's unclear whether the PGD or its metabolites increase in response to DNA damage or the senescence state. Thus, using a different senescent model to exclude the possibility of DNA damage-induced increase will be crucial.

Though Senescence can be induced by several stress stimuli like DNA damage, Oncogene expression, ROS, Mitochondrial Dysfunction, etc., DNA damage remains critical for the induction of the SASP (reviewed in PMID: 20078217). Also, other models of senescence, like Oncogene Induced Senescence (reviewed in PMID: 17671427), ROS Induced Senescence (PMID: 24934860), Mitochondrial Dysfunction Associated Senescence (MiDAS) (PMID: 26686024) have shown upregulation of DNA damage-associated signaling pathways. In this study, we have explored the SASP of cells undergoing senescence upon chemotherapy drug Doxorubicin-mediated DNA damage.

(4) Similarly for the in vivo Doxorubicin (Doxo) injection, both reviewers have raised concerns about the potential side effects of Doxo, including inflammation, DNA damage, and ROS generation. These effects could potentially confound the results of the study. The physiological significance of this study will heavily rely on the in vivo data. However, the in vivo senescence component is confounded by the side effects of Doxo.

We concur that this is a limitation of this study and the subsequent work will demonstrate the origin of prostaglandin biosynthesis after treatment with Doxo in vivo.

(5) Figure 2A lacks an important control from non-senescent cells during the measurement of C2C12 differentiation in the presence of conditioned medium. The author took it for granted that the conditioned medium from senescent cells would inhibit myogenesis, relying on previous publications (PMID: 37468473). However, that study was conducted in the context of myotonic dystrophy type 1. To support the inhibitory effect in the current experimental settings, direct evidence is required. It would be necessary to include another control with conditioned medium from normal, proliferative C2C12 cells.

Conditioned medium of senescent cells of several types, like senescent myoblasts in case of DM1 (PMID: 37468473), adipocytes undergoing senescence due to H2O2 treatment, Insulin Resistance, and Replicative senescence (PMID: 37321332), has been shown to inhibit the differentiation of myoblasts. Therefore, in this study, we measured the effect of prostaglandin PGD2 and its metabolites on the differentiation of myoblasts by inhibiting the biosynthesis of PGD2 in senescent myoblasts by treatment with AT-56. We inhibited the synthesis of PGD2 in senescent cells by treatment with AT-56, and then collected the conditioned medium. Conditioned medium collected from senescent C2C12 cells treated with vehicle (DMSO) served as a control for the experiment.

(6) Statistical analyses problems.

Only t-test was used throughout the study even when there are more than two groups. Please have a statistician to evaluate the replicates and statistical analyses used.

In experiments with more than two groups, the t-test was used for column-wise comparison of the experiment samples to the control sample. Multiple sample comparisons using one-way or two-way ANOVA were avoided as experimental samples were individually compared to the control sample.

For the 15d-PGJ2/cell concentration measurements in Figure 1F, there were only two replicates, which was provided in the supplementary table after required. Was that experiment repeated with more biological replicates?

Additional replicates of the experiment will be included in the revised manuscript.

For figure 1C, Fig 1F, 1G, 1J, 2C, 2E, 3A, 3E, 3F, 4D, 4E, please include each data points in bar graphs as used in Fig 1D, or at least provide how many biological replicates were used for each experiment?

Appropriate revisions will be made in the figure legends of the revised manuscript.

There is no error bar in a lot of control groups (Fig 2C, 2E, 3EF, 4E, S4B).

There are no error bars for the control groups in the figures 2C, 2E, 3E, 3F, 4E, and S4B as the experimental samples of each replicate were normalized to the corresponding control sample, rendering the values for the control sample of each replicate to 1.

For qPCR data in Figure 1C, the author responded in that the data in was plotted using 2-ΔCT instead of 2-ΔΔCT to show the variability in the expression of mRNAs isolated from animals treated with Saline. This statement does not align with the method section. Please revise.

Appropriate revisions will be made to the method sections of the revised manuscript.

(7) For Figure 1, the title may not be appropriate as there is insufficient data to support the inhibition of myoblast differentiation.

Appropriate revisions will be made to the revised manuscript.

Recommendations for the authors:

After careful review, the editors advise you to carefully address the following concerns.

(1) There were concerns that in the revised manuscript, the DMSO and Doxo experiments depicted in Figure 1H appeared quite homogenous despite the author's description to the contrary. This leads to concerns about the type of statistics employed and the possible low number of replicates of experiments shown in Fig. 1.

(2) Experiments in Figure 1F, 1I, and 1J had as few as n=2 experiments. Figures 1C, 1D, 1F, 1G, and 1J, the statistics used a two-tailed student's t-test; for all other experiments, they marked N/A for statistics. Using a t-test for multi-group comparisons (as indicated in the figure legend) and relying on only 2 replicates for many experiments are not appropriate.

Additional replicates for the experiments shown in figures 1F, 1I, and 1J have been done and the data will be revised along with updated statistical tests during the revision of the manuscript.

(3) In several experiments, the difference between technical replicates is too high.

Reviewer #1 (Recommendations For The Authors):

Most of my concerns were addressed in the revised manuscript.

We thank the reviewer for their time in reviewing the manuscript and consideration of the author’s response to their comments in during the previous round of review.

Reviewer #2 (Recommendations For The Authors):

Validating the findings in a primary myoblast is highly recommended for senescence studies due to the limitations and questionable senescence characteristics of the C2C12 cell line.

We have explained the statistical tests used in the manuscript in the general comment section of the reviewer’s comments.

Validate the finding in a different senescent model to exclude the possibility of DNA damage-response.

We have explained the statistical tests used in the manuscript in the general comment section of the reviewer’s comments.

For Fig 2A, add another control with a conditioned medium from normal, proliferative C2C12 cells.

We have explained the statistical tests used in the manuscript in the general comment section of the reviewer’s comments.

Please have a statistician to evaluate the replicates and statistical analyses used.

We have explained the statistical tests used in the manuscript in the general comment section of the reviewer’s comments.

For the barplots (figure 1C, Fig 1F, 1G, 1J, 2C, 2E, 3A, 3E, 3F, 4D, 4E), please include each data points, or at least provide how many biological replicates were used for each experiment.

Appropriate revisions will be made in the figure legends of the revised manuscript.

For Figure 1, the title may not be appropriate as there is insufficient data to support the inhibition of myoblast differentiation.

Appropriate revisions will be made to the revised manuscript.

---

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

eLife assessment

This manuscript provides useful information about the lipid metabolite 15d-PGJ2 as a potential regulator of myoblast senescence. The authors provide experimental evidence that 15d-PGJ2 inhibits myoblast proliferation and differentiation by binding and regulating HRas. However, the manuscript is incomplete in its current form, as it lacks robust support from the data regarding the main conclusions related to senescence and technical concerns related to the senescence models used in this study.

We are grateful to the editors and the reviewers for their time and comments in sharpening the science and the writing of the manuscript. We have attached a detailed response to emphasize that the manuscript does include robust evidence regarding the claims, which could have been missed during the review process. We have provided a better context for these points now.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

The authors show that upon treatment with Doxorubicin (Doxo), there is an increase in senescence and inflammatory markers in the muscles. They also show these genes get upregulated in C2C12 myoblasts when treated with conditioned media or 15d-PGJ2. 15dPGJ2 induces cell death in the myoblasts, decreases proliferation (measured by cell numbers), and decreases differentiation and fusion. 15d-PGJ2 modified Cys184 of HRas, which is required for its activation as indicated by the FRET analysis with RAF RBD. They also showed that 15d-PGJ2 activates ERK signaling, but not Akt signaling, through the electrophilic center. 15d-PGJ2 inhibits Golgi localization of HRAS (only WT, not C181 or C184 mutant). They also showed that expressing the WT HRas followed by 15d-PGJ2 treatment led to a decrease in the levels of MHC mRNA and protein, and this defect is dependent on C184. This is a well-written manuscript with interesting insights into the mechanism of action of 15d-PGJ2. However, some clarification and experiments will help the paper advance the field significantly.

Strengths:

The data clearly shows that 15d-PGJ2 has a negative role in the myoblast cells and that it leads to modification of HRas protein. Moreover, the induction of biosynthetic enzymes in the PGD2 pathway also supports the induction of 15d-PGJ2 in Doxorubicin-treated cells. Both conditioned media experiments and the 15d-PGJ2 experiments show that 15d-PGJ2 could be the active component secreted by the senescent myoblasts.

Weaknesses:

The genes that are upregulated in the muscles upon injection with Doxo are also markers for inflammation. Since Doxo is also known to induce systemic inflammation, it is important to delineate these two effects (inflammatory cells vs senescent cells). The expression of beta Gal and other markers of senescence in the tissue sections will help to delineate these.

As pointed out Doxo induces systemic inflammation along with inducing DNA damage-mediated senescence. Therefore, along with the inflammatory markers of the SASP (CXCL1/2, TNF1α, IL6, PTGS1/2, PTGDS) we also observed an increase in the mRNA levels of canonical markers of DNA damage-mediated senescence. We observed an increase in the mRNA levels of cell cycle and senescence associated proteins p16 and p21 (Fig. 1C). We also observed an increased nuclear accumulation of p21 (Fig. 1A) and increased levels of phosphorylated H2A.X in the nucleus (Fig. 1B).

In Figure 2, where the defect in the differentiation of myoblasts upon treatment with 15d-PGJ2 is shown, most of the cells die within 48 hours at higher concentrations, making it difficult to perform the experiments. This also shows that 15d-PGJ2 was toxic to these cells. Lower concentrations show a decrease in the differentiation based on the lower number of nuclei in fibers and low expression of MyoD, MyoG, and MHC. However, it is unclear if this is due to increased cell death or defective differentiation. It would be a lot more informative if the cell count, cell division, and cell death could be plotted for these concentrations of the drug during the experiment.

We measured the viability of C2C12 cells after 24 hours of treatment with 15d-PGJ2 using the MTT assay and observed that the viability of cells was decreased after treatment with 15d-PGJ2 (10 µM) but not with 15d-PGJ2 (1 µM, 2 µM, 4 µM, or 5 µM) (see Fig. S2A of the updated manuscript). The results and figures of the manuscript have been updated accordingly.

Also, in the myoblast experiments, are the effects of treatment with Dox reversible?

The treatment with Doxorubicin is irreversible as the senescent phenotype was not reversed after withdrawal of Doxorubicin, even after 20 days.

In Figure 3, most of the experiments are done at a high concentration, which induces almost complete cell death within 48 hours.

Figure 3 is an acute experiment for only 1 hour, at which time no cell death was observed. Specifically, we measured the phosphorylation of Erk and Akt proteins after 1 hour of treatment with 15d-PGJ2 (10 µM) during which we did not observe any cell death. 

Even at such a high concentration of 15dPGJ2, the increase in ERK phosphorylation is minimal.

We observe a ~30% increase in the phosphorylation of Erk proteins after treatment with 15d-PGJ2 in 0.2% serum medium compared to treatment with vehicle (DMSO). This is reproducible and significant.

The experiment Figure 4C shows that C181 and C84 mutants of the HRas show higher levels in Golgi compared with WT. However, this could very well be due to the defect in palmitoylation rather than the modification with 15d-PGJ2.

Our data does not suggest higher levels of C184S mutant in the Golgi compared with WT (Fig. S4A). We observed that the ratio of HRas levels in the Golgi to the HRas levels in the plasma membrane were similar in C2C12 cells expressing HRas C184S and HRas WT (Fig. S4A graph columns 1 and 5).

Though the authors allude to the possibility that intracellular redistribution of HRas by 15d-PGJ2 requires C181 palmitoylation, the direct influence of C184 modification on C181 palmitoylation is not shown. To have a meaningful conclusion, the authors need to compare the palmitoylation and modification with 15d-PGJ2.

Palmitoylation of HRas C181S is required for the localization of HRas at the plasma membrane. The inhibition of palmitoylation of C181, either by mutation (C181S) or treatment with protein palmitoyl transferase inhibitor (2-Bromopalmitate), results in the accumulation of HRas at Golgi(Rocks et al., 2005) (Fig. S4A). Modification of HRas at C184 by 15d-PGJ2 (Fig. 3A) could inhibit the palmitoylation of HRas at C181. However, our data does not support this hypothesis as modification of HRas WT by 15d-PGJ2 does not increase the level of HRas at the Golgi, like in the case of inhibition of cysteine palmitoylation due to C181S mutation.

To test if the inhibition of myoblast differentiation depends on HRas, they overexpressed the HRas and mutants in the C2C12 lines. However, this experiment does not take the endogenous HRAs into consideration, especially when interpreting the C184 mutant. An appropriate experiment to test this would be to knock down or knock out HRas (or make knock-in mutations of C184) and show that the effect of 15d-PGJ2 disappears. 

Endogenous HRas (wild type) is present in the C2C12 cells overexpressing the EGFP-tagged HRas constructs. Therefore, we only observe a partial rescue in the differentiation after 15d-PGJ2 treatment in C2C12 cells expressing the C184S mutant (Fig. 4D and E). However, since HRas is expressed under high expression CMV promoter and in the absence of other regulatory elements, the overexpressed constructs do show a dominant effect over the endogenous HRas, showing cysteine mutant dependent inhibition of differentiation of myoblasts after treatment with 15dPGJ2 (Fig. 4D and E).

Moreover, in this specific experiment, it is difficult to interpret without a control with no HRas construct and another without the 15d-PGJ2 treatment.

The mRNA levels of MyoD, MyoG, and MHC in C2C12 cells expressing HRas constructs after treatment with 15d-PGJ2 were normalized to the mRNA levels in C2C12 cells expressing corresponding constructs and were treated with vehicle (DMSO). mRNA levels in C2C12 cells treated with vehicle were not shown as they were normalized to 1. MHC protein levels in C2C12 cells expressing HRas constructs after 15d-PGJ2 treatment were normalized to that in C2C12 cells treated with vehicle (DMSO). Since the hypothesis to study the effect of HRas cysteine mutations on the differentiation of myoblasts after treatment with 15d-PGJ2, C2C12 cells expressing HRas WT serve as adequate control. Fig. 2 shows the effect of 15dPGJ2 on muscle differentiation when HRas was not overexpressed.

Moreover, the overall study does not delineate the toxic effects of 15d-PGJ2 from its effect on the differentiation.

The inhibition of differentiation in C212 cells after treatment with 15d-PGJ2 cannot be attributed to the general toxicity of 15d-PGJ2 in cells. We show that the inhibition of differentiation of myoblasts after 15d-PGJ2 depends on modification of HRas at C184 i.e. failure to modify HRas at C184 (Fig. 3A) and resultant activation (Fig. 3B) by 15d-PGJ2 rescues this inhibition of differentiation of C2C12 cells (Fig. 4D and E), dissecting the inhibition of differentiation of myoblasts by 15d-PGJ2 from general toxic effects of 15d-PGJ2 on cell physiology.

Please note that the effect of 15d-PGJ2 on cell physiology is context-specific. On one hand, 15d-PGJ2 has been shown to exert tumor-suppressor effects by inhibiting the proliferation of ovarian cancer cells and lung adenocarcinoma cells (de Jong et al., 2011; Slanovc et al., 2024), 15d-PGJ2 also exerts pro-carcinogenic effects by induction of epithelial to mesenchymal transition in breast cancer cells MCF7 and inhibition of tumor-suppressor protein p53 in MCF7 and PC-3 cells (Choi et al., 2020; Kim et al., 2010).

Reviewer #2 (Public Review):

Summary:

In this study, Swarang and colleagues identified the lipid metabolite 15d-PGJ2 as a potential component of senescent myoblasts. They proposed that 15d-PGJ2 inhibits myoblast proliferation and differentiation by binding and regulating HRas, suggesting its potential as a target for restoring muscle homeostasis post-chemotherapy.

Strengths:

The regulation of HRas by 15d-PGJ2 is well controlled.

Weaknesses:

The novelty of the study is compromised as the activation of PGD and 15d-PGJ2, as well as the regulation of HRas and cell proliferation, have been previously reported. 

Literature does not support this statement, and it is important to clarify this misimpression for the field as a whole. 

Let us clarify- 

Covalent modification of HRas by 15d-PGJ2 has been reported only twice in the literature(Luis Oliva et al., 2003; Yamamoto et al., 2011) in fibroblasts and neurons respectively. 

Interaction between Hras and 15d-PGJ2 in skeletal muscles has not been shown before, even though both Hras and 15d-PGJ2 are shown to be key regulators of muscle homeostasis. 

Activation of Hras by 15d-PGJ2 was reported first by Luis Oliva et al (Luis Oliva et al., 2003). However, this study does not comment on the functional implications of activation of Hras signaling. 

Recently, our lab contributed to a study where the functional implication of activation of Hras signaling due to covalent modification by 15d-PGJ2 was shown in the maintenance of senescence phenotype (Wiley et al., 2021). 

15d-PGJ2 was shown to inhibit the differentiation of myoblasts by Hunter et al (Hunter et al., 2001). This study hypothesized that the inhibition of myoblast differentiation is via 15d-PGJ2 mediated activation of the PPARγ signaling, the study also showed inhibition of myoblast differentiation independent of PPARγ activity, suggesting the presence of other mechanisms.

This is the first study to show a molecular mechanism where activation of Hras signaling in skeletal myoblasts due to covalent modification by 15d-PGJ2 at C184 of Hras inhibits the differentiation of skeletal myoblasts.

Additionally, there are major technical concerns related to the senescence models, limiting data interpretation regarding the relevance to senescent cells.

Major concerns:

(1) The C2C12 cell line is not an ideal model for senescence study due to its immortalized nature and lack of normal p16 expression. A more suitable myoblasts model is recommended, with a more comprehensive characterization of senescence features.

C2C12 is a good model for DNA damage-based senescence that is used in this manuscript. Several reports in the literature have shown the induction of senescence in C2C12 cells. Moiseeva et al 2023 show induction of senescence in C2C12 cells after etoposide-mediated DNA damage. Moustogiannis et al 2021 show the induction of replicative senescence in C2C12 cells. In this study, we show that C2C12 cells undergo DNA damage-mediated senescence after treatment with Doxo. We measured the induction of senescence in C2C12 cells upon DNA damage using several physiological (Nuclear Size, Cell Size, and SA β-gal) and molecular markers (mRNA levels of p21 and SASP factors (IL6 and TGFβ), protein levels of p21) of senescence (see Fig. 1 of the updated manuscript). The results and the figures in the manuscript have been updated accordingly.

(2) The source of increased PGD or its metabolites in the conditioned medium is unclear. Including other senescence models, such as replicative or oncogeneinduced senescence, would strengthen the study.

Fig. 1E shows time-dependent increase in the expression of PGD2 biosynthetic enzymes in senescent C2C12 cells. Fig. 1F shows an increase in the levels of 15dPGJ2 secreted by senescent C2C12 cells in the conditioned medium. This data shows that senescent C2C12 cells are the source of PGD and its metabolites in the conditioned medium.

Again, C2C12 is not suitable for replicative senescence due to its immortalized status.

We and others have shown that C2C12 cells undergo senescence, and this manuscript only used DNA damage induced senescence.

(3) In the in vivo part, it is unclear whether the increased expression of PTGS1, PTGS2, and PTGDS is due to senescence or other side effects of DOXO.

We concur that this is a limitation of this study and the subsequent work will demonstrate the origin of prostaglandin biosynthesis after treatment with Doxo in vivo.

(4) Figure 2A lacks an important control from non-senescent cells during the measurement of C2C12 differentiation in the presence of a conditioned medium.

Figure 2A tests the effect of prostaglandin PGD2 and its metabolites secreted by the senescent cells on the differentiation of myoblasts. Therefore, we inhibited the synthesis of PGD2 in senescent cells by treatment with AT-56, and then collected the conditioned medium. Conditioned medium collected from senescent C2C12 cells treated with vehicle (DMSO) served as a control for the experiment, whereas differentiation of C2C12 cells without any treatment serves as a positive control.

There is no explanation of how differentiation was quantified or how the fusion index was calculated.

The fusion index was calculated using a published myotube analyzer software (Noë et al., 2022). Appropriate information has been added to the materials and methods section of the manuscript.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Line 3: Expand SA in "SA β-gal".

The manuscript has been updated accordingly (See line 3).

Line 68: HRas is highly regulated by lipid modifications.

The manuscript has been updated accordingly (See line 67).

Figures

Figure S1A seemed incomplete (maybe some processing issue).

The Figure has been updated in the revised manuscript (See Fig. S1A).

Figure S1B-H are mislabeled.

The figure has been updated in the revised manuscript (See Fig. S1C, D, E, and F).

Figures S1E-H are not mentioned in the manuscript.

The manuscript has been updated accordingly (See line 120).

Many supplementary figures are not cited in the article.

The manuscript has been updated accordingly. (See lines 85, 120, 123, 166, 225, 356, 364, 412, and 413)

Reviewer #2 (Recommendations For The Authors):

(1) Clarify the injection method for Doxorubicin in B6J mice on line 83 (IP or IM).

Mice were injected intraperitoneally with Doxorubicin (as mentioned in the materials and methods, see lines 83 and 794)

(2) Address missing information in figures or figure legends.

There is missing piece in Sup Fig 1A.

The figure has been updated in the revised manuscript (See Fig. S1A).

Correct labels in Sup Fig 1C and 1D.

The figure has been updated in the revised manuscript (See Fig. S1C, D, E, and F).

How would the authors explain the dramatic differences in the morphology of C2C12 cells treated with DOXO between bright field and SA-beta-gal staining images in Sup Fig 1B and 1C.

The SA β-gal image after treatment with Doxo does show a flattened cell morphology. Another field of view from the same experiment has been added in the figure to show the difference in the cell morphology more prominently in the revised manuscript (See Fig. 1H).

Provide explanations for Sup Fig 1E-1G, including the meaning of the y-axis and the blue dots and red lines.

We have provided an explanation for the multiple reaction monitoring mass spectrometry used to measure the concentration of 15d-PGJ2 in the conditioned medium in the revised manuscript (see lines 119-130 and the legends of Fig. S1C, D, and E)

(3) Please review the calculation of qPCR data in Figure 1C for correctness, ensuring reference samples with an average expression level of 1.

The data in Fig. 1C was plotted using 2-ΔCT instead of 2-ΔΔCT to show the variability in the expression of mRNAs isolated from animals treated with Saline.

(4) Please explain the calculation of 15d-PGJ2/cell concentration in Figure 1F and provide raw data for review, considering the substantial changes and small error bars. The method or result section lacks an explanation of how this calculation was performed. Additionally, there is no mention of the cell number count.

All the raw values (concentration of 15d-PGJ2 measured using mass spec and cell numbers counted at the time of collection of conditioned medium) are provided in the supplementary table 1. The standard curve to calculate the concentration of 15dPGJ2 in the conditioned medium is shown in Fig. S1F. The cell number was counted after trypsinization using a hemocytometer on the day of collection of the conditioned medium.

(5) Please clarify how cell number normalization and doubling time calculation were done in Fig 2B. Consider replacing the figure with a growth curve showing confluence on the y-axis for easier interpretation.

Cells were counted every 24 hours and the normalization was done to the number of cells counted on day 0 of the treatment (to consider attaching efficiency and other cell culture parameters). Doubling time was calculated as the reciprocal of the slope of the graph of log2(normalized cell number) vs time.

eLife assessment

This manuscript outlines an interaction between senescence-related 15d-PGJ2 and the proliferation and differentiation of myoblasts, with potential implications for muscle health. This manuscript is useful in understanding the role of lipid metabolite 15d-PGJ2 in myoblast proliferation and differentiation. However, in its current form, the manuscript is incomplete as there are several concerns in the statistical analysis, lack of clarity on the mechanistic details, and concerns about the use of an immortalized C2C12 myoblasts cell line to draw major conclusions related to senescence-associated secreted phenotype.

Reviewer #1 (Public Review):

Summary:

The authors show that upon treatment with Doxorubicin (Doxo), there is an increase in senescence and inflammatory markers in the muscles. They also show these genes get upregulated in C2C12 myoblasts when treated with conditioned media or 15d-PGJ2. 15dPGJ2 induces cell death in the myoblasts, decreases proliferation (measured by cell numbers), and decreases differentiation and fusion. 15d-PGJ2 modified Cys184 of HRas, which is required for its activation as indicated by the FRET analysis with RAF RBD. They also showed that 15d-PGJ2 activates ERK signaling, but not Akt signaling, through the electrophilic center. 15d-PGJ2 inhibits Golgi localization of HRAS (only WT, not C181 or C184 mutant). They also showed that expressing the WT HRas followed by 15d-PGJ2 treatment led to a decrease in the levels of MHC mRNA and protein, and this defect is dependent on C184. This is a well-written manuscript with interesting insights into the mechanism of action of 15d-PGJ2. However, some clarification and experiments will help the paper advance the field significantly.

Strengths:

The data clearly shows that 15d-PGJ2 has a negative role in the myoblast cells and that it leads to modification of HRas protein. Moreover, the induction of biosynthetic enzymes in the PGD2 pathway also supports the induction of 15d-PGJ2 in Doxorubicin-treated cells. Both conditioned media experiments and the 15d-PGJ2 experiments show that 15d-PGJ2 could be the active component secreted by the senescent myoblasts.

Weaknesses:

The genes that are upregulated in the muscles upon injection with Doxo are also markers for inflammation. Since Doxo is also known to induce systemic inflammation, it is important to delineate these two effects (Inflammatory cells vs senescent cells). The expression of beta Gal and other markers of senescence in the tissue sections will help to delineate these.

In Figure 2, where the defect in the differentiation of myoblasts upon treatment with 15d-PGJ2 is shown, most of the cells die within 48 hours at higher concentrations, making it difficult to perform the experiments. This also shows that 15d-PGJ2 was toxic to these cells. Lower concentrations show a decrease in the differentiation based on the lower number of nuclei in fibers and low expression of MyoD, MyoG, and MHC. However, it is unclear if this is due to increased cell death or defective differentiation. It would be a lot more informative if the cell count, cell division, and cell death could be plotted for these concentrations of the drug during the experiment. Also, in the myoblast experiments, are the effects of treatment with Dox reversible?

In Figure 3, most of the experiments are done at a high concentration, which induces almost complete cell death within 48 hours. Even at such a high concentration of 15dPGJ2, the increase in ERK phosphorylation is minimal.

The experiment Figure 4C shows that C181 and C84 mutants of the HRas show higher levels in Golgi compared with WT. However, this could very well be due to the defect in palmitoylation rather than the modification with 15d-PGJ2. Though the authors allude to the possibility that intracellular redistribution of HRas by 15d-PGJ2 requires C181 palmitoylation, the direct influence of C184 modification on C181 palmitoylation is not shown. To have a meaningful conclusion, the authors need to compare the palmitoylation and modification with 15d-PGJ2.

To test if the inhibition of myoblast differentiation depends on HRas, they overexpressed the HRas and mutants in the C2C12 lines. However, this experiment does not take the endogenous HRAs into consideration, especially when interpreting the C184 mutant. An appropriate experiment to test this would be to knock down or knock out HRas (or make knock-in mutations of C184) and show that the effect of 15d-PGJ2 disappears. Moreover, in this specific experiment, it is difficult to interpret without a control with no HRas construct and another without the 15d-PGJ2 treatment.

Moreover, the overall study does not delineate the toxic effects of 15d-PGJ2 from its effect on the differentiation.

Reviewer #2 (Public Review):

Summary:

In this study, Swarang and colleagues identified the lipid metabolite 15d-PGJ2 as a potential component of senescent myoblasts. They proposed that 15d-PGJ2 inhibits myoblast proliferation and differentiation by binding and regulating HRas, suggesting its potential as a target for restoring muscle homeostasis post-chemotherapy.

Strengths:

The regulation of HRas by 15d-PGJ2 is well controlled.

Weaknesses:

(1) I still think the novelty is limited by previous published findings. The authors themselves noted that the accumulation of 15d-PGJ2 in senescent cells has been reported in various cell types, including human fibroblasts, HEPG2 hepatocellular carcinoma cells, and HUVEC endothelial cells (PMCID: PMC8501892). Although the current study observed similar activation of 15d-PGJ2 in myoblasts, it appears to be additive rather than fundamentally novel. The covalent adduct of 15d-PGJ2 with Cys-184 of H-Ras was reported over 20 years ago (PMID: 12684535), and the biochemical principles of this interaction are likely universal across different cell types. The regulation of myogenesis by both HRas and 15d-PGJ2 has also been previously extensively reported (PMID: 2654809, 1714463, 17412879, 20109525, 11477074). The main conceptual novelty may lie in the connection between these points in myoblasts. But as discussed in another comment, the use of C2C12 cells as a model for senescence study is questionable due to the lack of the key regulator p16. The findings in C2C12 cells may not accurately represent physiological-relevant myoblasts. It is recommended that these findings be validated in primary myoblasts to strengthen the study's conclusions.

(2) The C2C12 cell line is not an ideal model for senescence study.<br /> C2C12 cells are a well-established model for studying myogenesis. However, their suitability as a model for senescence studies is questionable. C2C12 cells are immortalized and do not undergo normal senescence like primary cells as C2C12 cells are known to have a deleted p16/p19 locus, a crucial regulator of senescence (PMID: 20682446). The use of C2C12 cells in published studies does not inherently validate them as a suitable senescence model. These studies may have limitations, and the appropriateness of the C2C12 model depends on the specific research goals.<br /> In the study by Moustogiannis et al. (PMID: 33918414), they claimed to have aged C2C12 cells through multiple population doublings. However, the SA-β-gal staining in their data, which is often used to confirm senescence, showed almost fully confluent "aged" C2C12 cells. This confluent state could artificially increase SA-β-gal positivity, suggesting that these cells may not truly represent senescence. Moreover, the "aged" C2C12 cells exhibited normal proliferation, which contradicts the definition of senescence. Similar findings were reported in another study of C2C12 cells subjected to 58 population doublings (PMID: 21826704), where even at this late stage, the cells were still dividing every 2 or 3 days, similar to younger cells at early passages. More importantly, I do know how the p16 was detected in that paper since the locus was already mutated. In terms of p21, there was no difference in the proliferative C2C12 cells at day 0.<br /> In the study by Moiseeva et al. in 2023 (PMID: 36544018), C2C12 cells were used for senescence modeling for siRNA transfection. However, the most significant findings were obtained using primary satellite cells or confirmed with complementary data.<br /> In conclusion, while molecular changes observed in studies using C2C12 cells may be valid, the use of primary myoblasts is highly recommended for senescence studies due to the limitations and questionable senescence characteristics of the C2C12 cell line.

(3) Regarding source of increased PGD in the conditioned medium, I want to emphasize that it's unclear whether the PGD or its metabolites increase in response to DNA damage or the senescence state. Thus, using a different senescent model to exclude the possibility of DNA damage-induced increase will be crucial.

(4) Similarly for the in vivo Doxorubicin (Doxo) injection, both reviewers have raised concerns about the potential side effects of Doxo, including inflammation, DNA damage, and ROS generation. These effects could potentially confound the results of the study. The physiological significance of this study will heavily rely on the in vivo data. However, the in vivo senescence component is confounded by the side effects of Doxo.

(5) Figure 2A lacks an important control from non-senescent cells during the measurement of C2C12 differentiation in the presence of conditioned medium. The author took it for granted that the conditioned medium from senescent cells would inhibit myogenesis, relying on previous publications (PMID: 37468473). However, that study was conducted in the context of myotonic dystrophy type 1. To support the inhibitory effect in the current experimental settings, direct evidence is required. It would be necessary to include another control with conditioned medium from normal, proliferative C2C12 cells.

(6) Statistical analyses problems.<br /> Only t-test was used throughout the study even when there are more than two groups. Please have a statistician to evaluate the replicates and statistical analyses used.<br /> For the 15d-PGJ2/cell concentration measurements in Figure 1F, there were only two replicates, which was provided in the supplementary table after required. Was that experiment repeated with more biological replicates?<br /> For figure 1C, Fig 1F, 1G, 1J, 2C, 2E, 3A, 3E, 3F, 4D, 4E, please include each data points in bar graphs as used in Fig 1D, or at least provide how many biological replicates were used for each experiment?<br /> There is no error bar in a lot of control groups (Fig 2C, 2E, 3EF, 4E, S4B).<br /> For qPCR data in Figure 1C, the author responded in that the data in was plotted using 2-ΔCT instead of 2-ΔΔCT to show the variability in the expression of mRNAs isolated from animals treated with Saline. This statement does not align with the method section. Please revise.

(7) For Figure 1, the title may not be appropriate as there is insufficient data to support the inhibition of myoblast differentiation.

Author response:

Please find below our provisional author response, outlining the revisions we plan to undertake to address the Recommendations received:

Reviewer #1 (Recommendations For The Authors):

(1) A set of recent advances have shown that embeddings of unsupervised/self-supervised speech models aligned to auditory responses to speech in the temporal cortex (e.g. Wav2Vec2: Millet et al NeurIPS 2022; HuBERT: Li et al. Nat Neurosci 2023; Whisper: Goldstein et al. bioRxiv 2023). These models are known to preserve a variety of speech information (phonetics, linguistic information, emotions, speaker identity, etc) and perform well in a variety of downstream tasks. These other models should be evaluated or at least discussed in the study.

We plan to evaluate two of these other models, Wav2Vec2 and HuBERT, in the brain encoding and RSA parts.

(2) The test statistics of the results in Fig 1c-e need to be revised. Given that logistic regression is a convex optimization problem typically converging to a global optimum, these multiple initializations of the classifier were likely not entirely independent. Consequently, the reported degrees of freedom and the effect size estimates might not accurately reflect the true variability and independence of the classifier outcomes. A more careful evaluation of these aspects is necessary to ensure the statistical robustness of the results.

We plan to address this point to ensure the statistical robustness of our results.

(3) In Line 198, the authors discuss the number of dimensions used in their models. To provide a comprehensive comparison, it would be informative to include direct decoding results from the original spectrograms alongside those from the VLS and LIN models. Given the vast diversity in vocal speech characteristics, it is plausible that the speaker identities might correlate with specific speech-related features also represented in both the auditory cortex and the VLS. Therefore, a clearer understanding of the original distribution of voice identities in the untransformed auditory space would be beneficial. This addition would help ascertain the extent to which transformations applied by the VLS or LIN models might be capturing or obscuring relevant auditory information.

We plan to include direct decoding results from the original spectrograms in addition from the VLS and LIN models.

Reviewer #2 (Recommendations For The Authors):

We plan to address the following points raised by Reviewer #2:

(1) English mistakes, rewordings:

a. L31: 'in voice' > consider rewording (from a voice?).

b. L33: consider splitting sentence (after interactions).

c. L39: 'brain' after parentheses.

d. L45-: certainly DNNs 'as a powerful tool' extend to audio (not just image and video) beyond their use in brain models.

e. L52: listened to / heard.

f. L63: use second/s consistently.

g. L64: the reference to Figure 5D is maybe a bit confusing here in the introduction.

h. L79-88: this section is formulated in a way that is too detailed for the introduction text (confusing to read). Consider a more general introduction to the VLS concept here and the details of this study later.

i. L99-: again, I think the experimental details are best saved for later. It's good to provide a feel for the analysis pipeline here, but some of the details provided (number of averages, denoising, preprocessing), are anyway too unspecific to allow the reader to fully follow the analysis.

We will correct the mistakes, apply the suggested rewordings, and clarify the points raised.

(2) Clarification.

• L159: what was the motivation for classifying age as a 2-class classification problem? Rather than more classes or continuous prediction? How did you choose the age split?

• L263: Is the test of RDM correlation>0 corrected for multiple comparisons across ROIs, subjects, and models?

• L379: 'these stimuli' - weren't the experimental stimuli different from those used to train the V/AE?

• L443: what are 'technical issues' that prevented subject 3 from participating in 48 runs??

• L444: participants were instructed to 'stay in the scanner'!? Do you mean 'stay still', or something?

• L463: Hearing thresholds of 15 dB: do you mean that all had thresholds lower than 15 dB at all frequencies and at all repeated audiogram measurements?

• L472: were the 4 category levels balanced across the dataset (in number of occurrences of each category combination)?

• L482: the test stimuli were selected as having high energy by the amplitude envelope. It is unclear what this means (how is the envelope extracted, what feature of it is used to measure 'high energy'?)

• L500 was the audio filtered to account for the transfer function of the Sensimetrics headphones?

• L500: what does 'comfortable level' correspond to and was it set per session (i.e. did it vary across sessions)?

• L526- does the normalization imply that the reconstructed spectrograms are normalized? Were the reconstructions then scaled to undo the normalization before inversion?

• L606: does the identity GLM model the denoised betas from the first GLM or simply the BOLD data? The text indicates the latter, but I suspect the former.

• L704: could you unpack this a bit more? It is not easy to see why you specify the summing in the objective. Shouldn't this just be the ridge objective for a given voxel/ROI? Then you could just state it in matrix notation.

• L716: you used robust scaling for the classifications in latent space but haven't mentioned scaling here. Are we to assume that the same applies?

• L720: Pearson correlation as a performance metric and its variance will depend on the choice of test/train split sizes. Can you show that the results generalize beyond your specific choices? Maybe the report explained variance as well to get a better idea of performance.

• Could you specify (somewhere) the stimulus timing in a run? ISI and stimulus duration are mentioned in different places, but it would be nice to have a summary of the temporal structure of runs.

We will clarify the points raised.

Reviewer #3 (Recommendations For The Authors):

We plan to address the following points raised by Reviewer #3:

Comments:

• Code and data are not currently available.

• In the supplementary material, it would be beneficial to present the different analyses as boxplots, as in the main text, but with the ROIs in the left and right hemispheres separated, to better show potential hemispheric effect. Although this information is available in the Supplementary Tables, it is currently quite tedious to access it.

• In Figure 3a, it might be beneficial to order the identities by age for each gender in order to more clearly illustrate the structure of the RDMs,

• In Figure 3b, the variance for the correlations for the aTVA is higher than in other regions, why?

• Please make sure that all acronyms are defined, and that they are redefined in the figure legends.

• Gender and age are primarily encoded by different brain regions (Figure 5, pTVA vs aTVA). How does this finding compare with existing literature?

We will upload the code and the preprocessed data; improve the supplementary material figures; Fix Figure 3 according to the Reviewer’s suggestion, and clarify the points raised.

eLife assessment

This study used deep neural networks (DNN) to reconstruct voice information (viz., speaker identity), from fMRI responses in the auditory cortex and temporal voice areas, and assessed the representational content in these areas with decoding. A DNN-derived feature space approximated the neural representation of speaker identity-related information. While some of the neural decoding results are valuable, the overall evidence for general representational and computational principles is incomplete as the results rely on a very specific model architecture.

Reviewer #1 (Public Review):

Summary:

In this study, the authors trained a variational autoencoder (VAE) to create a high-dimensional "voice latent space" (VLS) using extensive voice samples, and analyzed how this space corresponds to brain activity through fMRI studies focusing on the temporal voice areas (TVAs). Their analyses included encoding and decoding techniques, as well as representational similarity analysis (RSA), which showed that the VLS could effectively map onto and predict brain activity patterns, allowing for the reconstruction of voice stimuli that preserve key aspects of speaker identity.

Strengths:

This paper is well-written and easy to follow. Most of the methods and results were clearly described. The authors combined a variety of analytical methods in neuroimaging studies, including encoding, decoding, and RSA. In addition to commonly used DNN encoding analysis, the authors performed DNN decoding and resynthesized the stimuli using VAE decoders. Furthermore, in addition to machine learning classifiers, the authors also included human behavioral tests to evaluate the reconstruction performance.

Weaknesses:

This manuscript presents a variational autoencoder (VAE) to evaluate voice identity representations from brain recordings. However, the study's scope is limited by testing only one model, leaving unclear how generalizable or impactful the findings are. The preservation of identity-related information in the voice latent space (VLS) is expected, given the VAE model's design to reconstruct original vocal stimuli. Nonetheless, the study lacks a deeper investigation into what specific aspects of auditory coding these latent dimensions represent. The results in Figure 1c-e merely tested a very limited set of speech features. Moreover, there is no analysis of how these features and the whole VAE model perform in standard speech tasks like speech recognition or phoneme recognition. It is not clear what kind of computations the VAE model presented in this work is capable of. Inclusion of comparisons with state-of-the-art unsupervised or self-supervised speech models known for their alignment with auditory cortical responses, such as Wav2Vec2, HuBERT, and Whisper, would strengthen the validation of the VAE model and provide insights into its relative capabilities and limitations.

The claim that the VLS outperforms a linear model (LIN) in decoding tasks does not significantly advance our understanding of the underlying brain representations. Given the complexity of auditory processing, it is unsurprising that a nonlinear model would outperform a simpler linear counterpart. The study could be improved by incorporating a comparative analysis with alternative models that differ in architecture, computational strategies, or training methods. Such comparisons could elucidate specific features or capabilities of the VLS, offering a more nuanced understanding of its effectiveness and the computational principles it embodies. This approach would allow the authors to test specific hypotheses about how different aspects of the model contribute to its performance, providing a clearer picture of the shared coding in VLS and the brain.

The manuscript overlooks some crucial alternative explanations for the discriminant representation of vocal identity. For instance, the discriminant representation of vocal identity can be either a higher-level abstract representation or a lower-level coding of pitch height. Prior studies using fMRI and ECoG have identified both types of representation within the superior temporal gyrus (STG) (e.g., Tang et al., Science 2017; Feng et al., NeuroImage 2021). Additionally, the methodology does not clarify whether the stimuli from different speakers contained identical speech content. If the speech content varied across speakers, the approach of averaging trials to obtain a mean vector for each speaker-the "identity-based analysis"-may not adequately control for confounding acoustic-phonetic features. Notably, the principal component 2 (PC2) in Figure 1b appears to correlate with absolute pitch height, suggesting that some aspects of the model's effectiveness might be attributed to simpler acoustic properties rather than complex identity-specific information.

Methodologically, there are issues that warrant attention. In characterizing the autoencoder latent space, the authors initialized logistic regression classifiers 100 times and calculated the t-statistics using degrees of freedom (df) of 99. Given that logistic regression is a convex optimization problem typically converging to a global optimum, these multiple initializations of the classifier were likely not entirely independent. Consequently, the reported degrees of freedom and the effect size estimates might not accurately reflect the true variability and independence of the classifier outcomes. A more careful evaluation of these aspects is necessary to ensure the statistical robustness of the results.

Reviewer #2 (Public Review):

Summary:

Lamothe et al. collected fMRI responses to many voice stimuli in 3 subjects. The authors trained two different autoencoders on voice audio samples and predicted latent space embeddings from the fMRI responses, allowing the voice spectrograms to be reconstructed. The degree to which reconstructions from different auditory ROIs correctly represented speaker identity, gender, or age was assessed by machine classification and human listener evaluations. Complementing this, the representational content was also assessed using representational similarity analysis. The results broadly concur with the notion that temporal voice areas are sensitive to different types of categorical voice information.

Strengths:

The single-subject approach that allows thousands of responses to unique stimuli to be recorded and analyzed is powerful. The idea of using this approach to probe cortical voice representations is strong and the experiment is technically solid.

Weaknesses:

The paper could benefit from more discussion of the assumptions behind the reconstruction analyses and the conclusions it allows. The authors write that reconstruction of a stimulus from brain responses represents 'a robust test of the adequacy of models of brain activity' (L138). I concur that stimulus reconstruction is useful for evaluating the nature of representations, but the notion that they can test the adequacy of the specific autoencoder presented here as a model of brain activity should be discussed at more length. Natural sounds are correlated in many feature dimensions and can therefore be summarized in several ways, and similar information can be read out from different model representations. Models trained to reconstruct natural stimuli can exploit many correlated features and it is quite possible that very different models based on different features can be used for similar reconstructions. Reconstructability does not by itself imply that the model is an accurate brain model. Non-linear networks trained on natural stimuli are arguably not tested in the same rigorous manner as models built to explicitly account for computations (they can generate predictions and experiments can be designed to test those predictions). While it is true that there is increasing evidence that neural network embeddings can predict brain data well, it is still a matter of debate whether good predictability by itself qualifies DNNs as 'plausible computational models for investigating brain processes' (L72). This concern is amplified in the context of decoding and naturalistic stimuli where many correlated features can be represented in many ways. It is unclear how much the results hinge on the specificities of the specific autoencoder architectures used. For instance, it would be useful to know the motivations for why the specific VAE used here should constitute a good model for probing neural voice representations.

Relatedly, it is not clear how VAEs as generative models are motivated as computational models of voice representations in the brain. The task of voice areas in the brain is not to generate voice stimuli but to discriminate and extract information. The task of reconstructing an input spectrogram is perhaps useful for probing information content, but discriminative models, e.g., trained on the task of discriminating voices, would seem more obvious candidates. Why not include discriminatively trained models for comparison?

The autoencoder learns a mapping from latent space to well-formed voice spectrograms. Regularized regression then learns a mapping between this latent space and activity space. All reconstructions might sound 'natural', which simply means that the autoencoder works. It would be good to have a stronger test of how close the reconstructions are to the original stimulus. For instance, is the reconstruction the closest stimulus to the original in latent space coordinates out of using the experimental stimuli, or where does it rank? How do small changes in beta amplitudes impact the reconstruction? The effective dimensionality of the activity space could be estimated, e.g. by PCA of the voice samples' contrast maps, and it could then be estimated how the main directions in the activity space map to differences in latent space. It would be good to get a better grasp of the granularity of information that can be decoded/ reconstructed.

What can we make of the apparent trend that LIN is higher than VLS for identity classification (at least VLS does not outperform LIN)? A general argument of the paper seems to be that VLS is a better model of voice representations compared to LIN as a 'control' model. Then we would expect VLS to perform better on identity classification. The age and gender of a voice can likely be classified from many acoustic features that may not require dedicated voice processing.

The RDM results reported are significant only for some subjects and in some ROIs. This presumably means that results are not significant in the other subjects. Yet, the authors assert general conclusions (e.g. the VLS better explains RDM in TVA than LIN). An assumption typically made in single-subject studies (with large amounts of data in individual subjects) is that the effects observed and reported in papers are robust in individual subjects. More than one subject is usually included to hint that this is the case. This is an intriguing approach. However, reports of effects that are statistically significant in some subjects and some ROIs are difficult to interpret. This, in my view, runs contrary to the logic and leverage of the single-subject approach. Reporting results that are only significant in 1 out of 3 subjects and inferring general conclusions from this seems less convincing.

The first main finding is stated as being that '128 dimensions are sufficient to explain a sizeable portion of the brain activity' (L379). What qualifies this? From my understanding, only models of that dimensionality were tested. They explain a sizeable portion of brain activity, but it is difficult to follow what 'sizable' is without baseline models that estimate a prediction floor and ceiling. For instance, would autoencoders that reconstruct any spectrogram (not just voice) also predict a sizable portion of the measured activity? What happens to reconstruction results as the dimensionality is varied?

A second main finding is stated as being that the 'VLS outperforms the LIN space' (L381). It seems correct that the VAE yields more natural-sounding reconstructions, but this is a technical feature of the chosen autoencoding approach. That the VLS yields a 'more brain-like representational space' I assume refers to the RDM results where the RDM correlations were mainly significant in one subject. For classification, the performance of features from the reconstructions (age/ gender/ identity) gives results that seem more mixed, and it seems difficult to draw a general conclusion about the VLS being better. It is not clear that this general claim is well supported.

It is not clear why the RDM was not formed based on the 'stimulus GLM' betas. The 'identity GLM' is already biased towards identity and it would be stronger to show associations at the stimulus level.

Multiple comparisons were performed across ROIs, models, subjects, and features in the classification analyses, but it is not clear how correction for these multiple comparisons was implemented in the statistical tests on classification accuracies.

Risks of overfitting and bias are a recurrent challenge in stimulus reconstruction with fMRI. It would be good with more control analyses to ensure that this was not the case. For instance, how were the repeated test stimuli presented? Were they intermingled with the other stimuli used for training or presented in separate runs? If intermingled, then the training and test data would have been preprocessed together, which could compromise the test set. The reconstructions could be performed on responses from independent runs, preprocessed separately, as a control. This should include all preprocessing, for instance, estimating stimulus/identity GLMs on separately processed run pairs rather than across all runs. Also, it would be good to avoid detrending before GLM denoising (or at least testing its effects) as these can interact.

Reviewer #3 (Public Review):

Summary:

In this manuscript, Lamothe et al. sought to identify the neural substrates of voice identity in the human brain by correlating fMRI recordings with the latent space of a variational autoencoder (VAE) trained on voice spectrograms. They used encoding and decoding models, and showed that the "voice" latent space (VLS) of the VAE performs, in general, (slightly) better than a linear autoencoder's latent space. Additionally, they showed dissociations in the encoding of voice identity across the temporal voice areas.

Strengths:

- The geometry of the neural representations of voice identity has not been studied so far. Previous studies on the content of speech and faces in vision suggest that such geometry could exist. This study demonstrates this point systematically, leveraging a specifically trained variational autoencoder.

- The size of the voice dataset and the length of the fMRI recordings ensure that the findings are robust.

Weaknesses:

- Overall, the VLS is often only marginally better than the linear model across analysis, raising the question of whether the observed performance improvements are due to the higher number of parameters trained in the VAE, rather than the non-linearity itself. A fair comparison would necessitate that the number of parameters be maintained consistently across both models, at least as an additional verification step.

- The encoding and RSM results are quite different. This is unexpected, as similar embedding geometries between the VLS and the brain activations should be reflected by higher correlation values of the encoding model.

- The consistency across participants is not particularly high, for instance, S1 seemed to have demonstrated excellent performances, while S2 showed poor performance.

- An important control analysis would be to compare the decoding results with those obtained by a decoder operating directly on the latent spaces, in order to further highlight the interest of the non-linear transformations of the decoder model. Currently, it is unclear whether the non-linearity of the decoder improves the decoding performance, considering the poor resemblance between the VLS and brain-reconstructed spectrograms.

Author response:

We thank the reviewers for their comments and will revise the manuscript to provide more comprehensive clarifications to aide readers’ understanding of behaviorMate. Additionally, we intend to take several steps which could provide further insights and improve the ease of use for new behaviorMate users: (1) to release an expanded and annotated library of existing settings and VR scene files, (2) improve the online documentation of context lists and decorators which allow behaviorMate to run custom experimental paradigms without writing code, and (3) release online API details of the JSON messaging protocol that is used between behaviorMate, the Arduinos, and the VRMate program which could be especially helpful to developers interested in expanding or modifying the system. Here we provide a few brief points of clarification to some of the concerns raised by the reviewers.

Firstly, we clarify the system’s focus on modularity and flexibility. behaviorMate leverages the “Intranet of Things” framework to provide a low-cost platform that relies on asynchronous message passing between independent networked devices. While our current VR implementation typically involves a PC, 2 Arduinos, and an Android device per VR display, the behaviorMate GUI can be configured without editing any source code to listen on additional ports for UDP messages which will be automatically timestamped and logged. Since the current implementation of the behaviorMate GUI can be configured through the settings file to send and receive JSON-formatted messages on arbitrary ports, third-party devices could be configured to listen and respond to these messages also without editing the UI source code. More specialized responsibilities or tasks that require higher temporal precision (such as position tracking) are handled by dedicated circuits so as to not overload the general purpose one. This provides a level of encapsulation/separation of concerns since components can be optimized for performance of a single tasks—a feature that is especially desirable given resource limitations on the most common commercially available microcontrollers.

A number of methods exist for synchronizing recording devices like microscopes or electrophysiology recordings with behaviorMate’s time-stamped logs of actuators and sensors. For example, the GPIO circuit can be configured to send sync triggers, or receive timing signals as input, alternatively a dedicated circuit could record frame start signals and relay them to the PC to be logged indecently of the GPIO (enabling a high-resolution post-hoc alignment of the time stamps). The optimal method to use varies based on the needs of the experiment. For example, if very high temporal precision is needed, such as during electrophysiology experiments, a high-speed data acquisition (DAQ) circuit to capture a fixed interval readout might be beneficial. behaviorMate could still be set up as normal to provide closed and open-loop task control at behaviorally relevant timescales alongside a DAQ circuit recording events at a consistent temporal resolution. While this would increase the relative cost of the recording setup, identical rigs for training animals could still be configured without the DAQ circuit avoiding the additional cost and complexity.

VRMate provides the interface between Unity and behaviorMate—therefore using the two systems together mean that no Unity or C# programming is necessary. VRMate provides a prespecified set of visual cues that can be scaled in 3 dimensions and have textures applied to them, permitting a wide variety of different scenes to be displayed. All VRMate scene details are additionally logged by behaviorMate to allow for consistency checks across experiments. The VRMate project also includes “editor scripts” that provide a drag-and-drop utility in Unity Editor for developing new scenes. Since the details pertaining to specific scenes and view angle are loaded at runtime via JSON-formatted UDP messages, it is not necessary to recompile VRMate in order to use this feature. Since we send individual position updates to VRMate from the PC, any issues with clock drift would be limited to the refresh rate of the Unity program that fast enough to be perceived as instantaneous and we have thoroughly tested the timing differences between displays using high-speed cameras and found them to be negligible. While we find using 5 separate Android computers to render scenes as described an optimal solution to maximize flexibility, it would also be possible to render all scenes on a single PC to further mitigate this concern depending on experimental demands. Finally, our treadmill implementations of behaviorMate use no monitor displays, however due to the modular design of behaviorMate virtual cues could be seamlessly added by added to any such setup by a VR context to the settings files.

One last point to mention is that while our project is not affected by the recent changes in pricing structure of the Unity project, since the compiled software does not need to be regenerated to update VR scenes, or implement new task logic since this is handled by the behaviorMate GUI. This means the current state of the VRMate program is robust to any future pricing changes or other restructuring of the Unity program and does not rely on continued support of Unity. Additionally, the solution presented in VRMate has many benefits, however, a developer could easily adapt any open-source VR Maze project to receive the UDP-based position updates from behaviorMate or develop their own novel VR solutions. We intend to update the VR section of the manuscript to make all of this information clearer in the document as well as to provide the additional online documentation in the materials linked in the supplemental information.

Reviewer #1 (Public Review):

Summary:

Bowler et al. present a thoroughly tested system for modularized behavioral control of navigation-based experiments, particularly suited for pairing with 2-photon imaging but applicable to a variety of techniques. This system, which they name behaviorMate, represents a valuable contribution to the field. As the authors note, behavioral control paradigms vary widely across laboratories in terms of hardware and software utilized and often require specialized technical knowledge to make changes to these systems. Having a standardized, easy-to-implement, and flexible system that can be used by many groups is therefore highly desirable. This work will be of interest to systems neuroscientists looking to integrate flexible head-fixed behavioral control with neural data acquisition.

Strengths:

The present manuscript provides compelling evidence of the functionality and applicability of behaviorMate. The authors report benchmark tests for real-time update speed between the animal's movement and the behavioral control, on both the treadmill-based and virtual reality (VR) setups. Further, they nicely demonstrate and quantify reliable hippocampal place cell coding in both setups, using synchronized 2-photon imaging. This place cell characterization also provides a concrete comparison between the place cell properties observed in treadmill-based navigation vs. visual VR in a single study, which itself is a helpful contribution to the field.

Documentation for installing and operating behaviorMate is available via the authors' lab website and linked in the manuscript.

Weaknesses:

The following comments are mostly minor suggestions intended to add clarity to the paper and provide context for its significance.

(1) As VRMate (a component of behaviorMate) is written using Unity, what is the main advantage of using behaviorMate/VRMate compared to using Unity alone paired with Arduinos (e.g. Campbell et al. 2018), or compared to using an existing toolbox to interface with Unity (e.g. Alsbury-Nealy et al. 2022, DOI: 10.3758/s13428-021-01664-9)? For instance, one disadvantage of using Unity alone is that it requires programming in C# to code the task logic. It was not entirely clear whether VRMate circumvents this disadvantage somehow -- does it allow customization of task logic and scenery in the GUI? Does VRMate add other features and/or usability compared to Unity alone? It would be helpful if the authors could expand on this topic briefly.

(2) The section on "context lists", lines 163-186, seemed to describe an important component of the system, but this section was challenging to follow and readers may find the terminology confusing. Perhaps this section could benefit from an accompanying figure or flow chart, if these terms are important to understand.

(2a) Relatedly, "context" is used to refer to both when the animal enters a particular state in the task like a reward zone ("reward context", line 447) and also to describe a set of characteristics of an environment (Figure 3G), akin to how "context" is often used in the navigation literature. To avoid confusion, one possibility would be to use "environment" instead of "context" in Figure 3G, and/or consider using a word like "state" instead of "context" when referring to the activation of different stimuli.

(3) Given the authors' goal of providing a system that is easily synchronizable with neural data acquisition, especially with 2-photon imaging, I wonder if they could expand on the following features:

(3a) The authors mention that behaviorMate can send a TTL to trigger scanning on the 2P scope (line 202), which is a very useful feature. Can it also easily generate a TTL for each frame of the VR display and/or each sample of the animal's movement? Such TTLs can be critical for synchronizing the imaging with behavior and accounting for variability in the VR frame rate or sampling rate.

(3b) Is there a limit to the number of I/O ports on the system? This might be worth explicitly mentioning.

(3c) In the VR version, if each display is run by a separate Android computer, is there any risk of clock drift between displays? Or is this circumvented by centralized control of the rendering onset via the "real-time computer"?

Author response:

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

Reviewer #1 (Public Review):

Summary:

The present paper introduces Oscillation Component Analysis (OCA), in analogy to ICA, where source separation is underpinned by a biophysically inspired generative model. It puts the emphasis on oscillations, which is a prominent characteristic of neurophysiological data.

Strengths:

Overall, I find the idea of disambiguating data-driven decompositions by adding biophysical constrains useful, interesting and worth-pursuing. The model incorporates both a component modelling of oscillatory responses that is agnostic about the frequency content (e.g., doesn’t need bandpass filtering or predefinition of bands) and a component to map between sensor and latent space. I feel these elements can be useful in practice.

Thank you for the positive evaluation!

Weaknesses:

Lack of empirical support: I am missing empirical justification of the advantages that are theoretically claimed in the paper. I feel the method needs to be compared to existing alternatives.

Thank you for bringing up this important issue.  We agree that a direct performance comparison would be important to demonstrate.  We performed additional analyses to compare OCA with ICA and one easy frequency domain exploratory technique in both simulated and real human data (see Section How does OCA compare to conventional approaches? and Supporting Text: Comparison of OCA to traditional approaches in experimental EEG data).  The results of the simulated data are shown in the revised Figure 3.  Although the slow and alpha oscillations in this simulation are statistically independent under the generative model, ICA identifies components that mix these independent signals, as one would expect based on the above discussion (i.e., all components are Gaussian).  Meanwhile, OCA is able to recover distinct slow and alpha components.  We repeated this analysis in real human EEG during propofol-induced unconsciousness and found a similar result where ICA produced components that mixed slow and alpha band signals whereas OCA identified distinct oscillatory components (see Figure S4.1).

Reviewer #1 (Recommendations For The Authors):

Major

Theoretical justification. About the limitation of ICA In M/EEG, lines 24-28 seem to suggest that, almost by necessity (if Gaussianity approximately holds as argued), ICA doesn’t work on these modalities. But a body of work indicates that it does work to a reasonable extent, and that it is useful in practice; see https://www.pnas.org/doi/pdf/10.1073/pnas.1112685108?download=true. How then this theoretical claim be reconciled with the empirical evidence suggesting otherwise? I am putting this as a major comment because the limitations of ICA are one of the main motivations for this work, so it needs to be well-justified.

Thanks for bringing this forward this important point and for suggesting the reference Brookes, et al. Their work actually supports our claim. In the fifth paragraph of the discussion section, Brookes, et al. states “ICA has been used previously and extensively for artifact rejection in MEG; however, its use in identification of oscillatory signals has remained limited. This limitation is likely due to its susceptibility to interference and the fact that amplitude-modulated oscillatory signals exhibit a largely Gaussian statistical distribution (and ICA relies on non-Gaussianity in recovered sources).” For this reason, they use the Hilbert envelope as the input to the ICA procedure rather than the original time-series. These Hilbert envelopes represent the instantaneous amplitude of neural oscillatory activity, i.e., they follow the amplitude modulation of the oscillatory activity. The method does not extract any oscillatory activity or disambiguate different oscillatory sources, but only assess the connectivity pattern within pre-defined bands, i.e., how different areas of the brain are harmonized through modulation of the oscillations or vice-versa inside those pre-defined bands. The paper did not show extracted independent time signals (tICs), focusing instead on the spatial pattern that these tICs activated. In that way, their use of ICA was totally justified.  Overall, our assessment of the limitations of ICA are very well aligned with Brookes, et al. We have added the against our claim in the introduction (see page 3 line 23) and revised the discussion section to refer to this paper (see page 21 lines 426-432).

Empirical justification. The synthetic example is good, but I’m not quite sure what to make out of the real data examples. One can see reasonable spectra in the different bands and not-soeasy to interpret spatial topologies. But the main question is how OCA compares to more standard, easier approaches. Could the authors show explicitly how the benefits that were spelled out in the introduction/discussion manifest in practice, when compared to other methods?

Thank you for bringing up this important issue.  We agree that a direct performance comparison would be important to demonstrate. We performed additional analyses to compare OCA with ICA and one easy frequency domain exploratory technique in both simulated and real human data (see Section How does OCA compare to conventional approaches? and Supporting Text: Comparison of OCA to traditional approaches in experimental EEG data).  The results of the simulated data are shown in the revised Figure 3 in page 12. Although the slow and alpha oscillations in this simulation are statistically independent under the generative model, ICA identifies components that mix these independent signals, as one would expect based on the above discussion (i.e., all components are Gaussian).  Meanwhile, OCA is able to recover distinct slow and alpha components. We repeated this analysis in real human EEG during propofol-induced unconsciousness and found a similar result where ICA produced components that mixed slow and alpha band signals whereas OCA identified distinct oscillatory components (see Figure S4.1 in Supporting Text: Comparison of OCA to traditional approaches in experimental EEG data).

Minor

"a recently-described class of state-space models" -> of the three references, one is from the sixties, another from the eighties, and the last one is 21 years old. Is this really a recent idea?

Maybe rephrase "recently-described", or else think of more recent references that bring something new?

We have amended the wording as suggested. (See page 4, line 53)

Lines 72-74. It might be useful to unwrap in *intuitive* terms why the elements of this vector are closely related to the real and imaginary parts of the analytic signal.

Thanks for the helpful comment. The sentence now reads:

“These elements of this state vector traces out two time-series that maintains an approximate π/ 2 radian phase difference and therefore are closely related to the real and imaginary parts of an analytic signal…”. (See page 5, lines 72-75)

Also, relatedly, I don’t seem to have access to the SI which is supposed to explain this. It doesn’t show up in the BiorXiv preprint either.

We are sorry to hear that. BiorXiv merges all the supporting information and posts them under the Supplementary Material.

In Eq(1) should it be R(f) instead of R(2 \pi f / f_s) ?

Thank you for catching this typo.

As I understand from lines 182-195, the input for the method is not channels but PCA components. Since R is learned, presumably the variance of the lower-order PCs (i.e. the latest elements of the diagonal of R) will estimated to be small. This, in turn, would make the likelihood to be heavily weighed on these components (because one basically divides their contribution by their variance). Would this potentially bias the estimation towards these lower-order PCs, at the expense of higher-order PCs. In a different context, this is shown here: https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1008580 Maybe it would be worth commenting on this?

We agree with reviewer’s initial observations but disagree with the assessment. Our loglikelihood calculation reweights the components appropriately to counter the weighting coming due to spatial whitening, thus negating the above-mentioned bias. The main contribution of the spatial whitening and PCA are to make the learning numerically stable, i.e., it does not encounter underflow or overflow in the iterative steps. We also note that this spatial whitening, and the PCA are also reverted at the end to obtain the spatial components and estimated noise covariance. So, as long as we use all the components with strictly positive variances, we will not bias the log-likelihood one way or other.

eLife assessment

This method paper proposes a valuable Oscillation Component Analysis (OCA) approach, in analogy to Independent Component Analysis (ICA), in which source separation is achieved through biophysically inspired generative modeling of neural oscillations. The empirical evidence justifying the approach's advantage is solid. This work will be of interest to researchers in the fields of cognitive neuroscience, neural oscillation, and MEG/EEG.

Reviewer #1 (Public Review):

Summary:

The present paper introduces Oscillation Component Analysis (OCA), in analogy to ICA, where source separation is underpinned by a biophysically inspired generative model. It puts the emphasis on oscillations, which is a prominent characteristic of neurophysiological data.

Strengths:

Overall, I find the idea of disambiguating data-driven decompositions by adding biophysical constrains useful, interesting and worth pursuing. The model incorporates both a component modelling of oscillatory responses that is agnostic about the frequency content (e.g. doesn't need bandpass filtering or predefinition of bands) and a component to map between sensor and latent-space. I feel these elements can be useful in practice.

Weaknesses:

Lack of empirical support: I am missing empirical justification of the advantages that are theoretically claimed in the paper. I feel the method needs to be compared to existing alternatives.

Comments on the revised version: This concern has been addressed in the revised version.

eLife assessment

The study answers the important question of whether the conformational dynamics of proteins are slaved by the motion of solvent water or are intrinsic to the polypeptide. The results from neutron scattering experiments, involving isotopic labelling, carried out on a set of four structurally different proteins are convincing, showing that protein motions are not coupled to the solvent. A strength of this work is the study of a set of proteins using spectroscopy covering a range of resolutions. A minor weakness is the limited description of computational methods and analysis of data. The work is of broad interest to researchers in the fields of protein biophysics and biochemistry.

Reviewer #1 (Public Review):

Summary:

Zheng et al. study the 'glass' transitions that occurs in proteins at ca. 200K using neutron diffraction and differential isotopic labeling (hydrogen/deuterium) of the protein and solvent. To overcome limitations in previous studies, this work is conducted in parallel with 4 proteins (myoglobin, cytochrome P450, lysozyme and green fluorescent protein) and experiments were performed at a range of instrument time resolutions (1ns - 10ps). The author's data looks compelling, and suggests that transitions in the protein and solvent behavior are not coupled and contrary to some previous reports, the apparent water transition temperature is a 'resolution effect'; i.e. instrument response limited. This is likely to be important in the field, as a reassessment of solvent 'slaving' and the role of the hydration shell on protein dynamics should be reassessed in light of these findings.

Strengths:

The use of multiple proteins and instruments with a rate of energy resolution/ timescales.

Weaknesses:

The paper could be organised to better allow the comparison of the complete dataset collected.<br /> The extent of hydration clearly influences the protein transition temperature. The authors suggest that "water can be considered here as lubricant or plasticizer which facilitates the motion of the biomolecule." This may be the case, but the extent of hydration may also alter the protein structure.

Reviewer #2 (Public Review):

Summary:

The manuscript entitled "Decoupling of the Onset of Anharmonicity between a Protein and Its Surface Water around 200 K" by Zheng et al. presents a neutron scattering study trying to elucidate if at the dynamical transition temperature water and protein motions are coupled. The origin of the dynamical transition temperature is highly debated since decades and specifically its relation to hydration.

Strengths:

The study is rather well conducted, with a lot of efforts to acquire the perdeuterated proteins, and some results are interesting.

Weaknesses:<br /> The MD data presented appears to be missing description of the methods used.<br /> If these data support the authors claim that different levels of hydration do not affect the protein structure, careful analysis of the MD simulation data should be presented that show the systems are properly equilibrated under each condition. Additionally, methods are needed to describe the MD parameters and methods used, and for how long the simulations were run.

Author response:

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

eLife assessment:

The study answers the important question of whether the conformational dynamics of proteins are slaved by the motion of solvent water or are intrinsic to the polypeptide. The results from neutron scattering experiments, involving isotopic labelling, carried out on a set of four structurally different proteins are convincing, showing that protein motions are not coupled to the solvent. A strength of this work is the study of a set of proteins using spectroscopy covering a range of resolutions, however, it suffers from some scholarly shortcomings and limited discussion of results. The work is of broad interest to researchers in the fields of protein biophysics and biochemistry.

Reply 1: We thank the editors and reviewers for the positive and encouraging comments.

Reviewer #1 (Public Review):

Summary:

Zheng et al. study the 'glass' transitions that occur in proteins at ca. 200K using neutron diffraction and differential isotopic labeling (hydrogen/deuterium) of the protein and solvent. To overcome limitations in previous studies, this work is conducted in parallel with 4 proteins (myoglobin, cytochrome P450, lysozyme, and green fluorescent protein) and experiments were performed at a range of instrument time resolutions (1ns - 10ps). The author's data looks compelling, and suggests that transitions in the protein and solvent behavior are not coupled and contrary to some previous reports, the apparent water transition temperature is a 'resolution effect'; i.e. instrument response is limited. This is likely to be important in the field, as a reassessment of solvent 'slaving' and the role of the hydration shell on protein dynamics should be reassessed in light of these findings.

Strengths:

The use of multiple proteins and instruments with a rate of energy resolution/ timescales.

Reply 2: We thank the reviewer for highlighting our key findings.

Weaknesses:

The paper could be organised to better allow the comparison of the complete dataset collected. The extent of hydration clearly influences the protein transition temperature. The authors suggest that "water can be considered here as lubricant or plasticizer which facilitates the motion of the biomolecule." This may be the case, but the extent of hydration may also alter the protein structure.

Reply 3: Following the reviewer’s suggestion, we studied the secondary structure content and tertiary structure of CYP protein at different hydration levels (h = 0.2 and 0.4) through molecular dynamics simulation. As shown in Table S2 and Figure S6, the extent of hydration does not alter the protein secondary structure content and overall packing. Thus, this result also suggests that water molecules have more influence on protein dynamics than on protein structure.

Reviewer #2 (Public Review):

Summary:

The manuscript entitled "Decoupling of the Onset of Anharmonicity between a Protein and Its Surface Water around 200 K" by Zheng et al. presents a neutron scattering study trying to elucidate if at the dynamical transition temperature water and protein motions are coupled. The origin of the dynamical transition temperature has been highly debated for decades, specifically its relation to hydration.

Strengths:

The study is rather well conducted, with a lot of effort to acquire the perdeuterated proteins, and some results are interesting.

Reply 4: We thank the reviewer for highlighting our key findings.

Weaknesses:

The present work could certainly contribute some arguments, but I have the feeling that not all known facts are properly discussed.

The points the authors should carefully discuss are the following:

(1) Daniel et al. (10.1016/S0006-3495(98)77694-5) have shown that enzymes can be functional below the dynamical transition temperature which is at odds with some of the claims of the authors.

Reply 5: Following the reviewer’s suggestion, we added the following paragraph into the Introduction into the revised main text.

“Although exceptions have been reported (Biophys. J. 1998, 75, 2504.), the dynamical transition has been linked to the thermal onset of function in a number of proteins, e.g, myoglobin (Biochemistry, 1975, 14, 5355-5373), ribonuclease (Nature, 1992, 357, 423-424.), elastase ( Biochemistry, 1994, 33, 9285-9293.) and bacteriorhodopsin (PNAS, 1993, 90, 9668-9672.), all of which become inactive below the dynamical transition temperature.”

(2) It is not as easy to say that protonated proteins in D2O reflect protein dynamics while perdeuterated proteins in H2O reflect water dynamics. A recent study by Nidriche et al. (PRX LIFE 2, 013005 (2024)) reveals that H <-> D exchange is much faster than usually assumed and has important consequences for such studies.

Reply 6: For the sample preparation, all the H-proteins were dissolved in D2O to allow full deuterium exchange of all exchangeable hydrogen atoms and then lyophilized for 12 hours to obtain the dry sample. The lyophilized H-protein is then put into a desiccator with D2O, placed in the glove box purged with nitrogen gas, to absorb D2O till the desired hydration level, h (gram water/gram protein). In contrast, the preparation of the deuterated proteins was conducted in the opposite way. The D-proteins were dissolved in H2O to allow full hydrogen exchange of all exchangeable deuterium atoms and then lyophilized for 12 hours to obtain the dry sample. The lyophilized D-protein is then put into a desiccator with H2O to absorb H2O till the desired h. This procedure can avoid H-D exchange during experiments. We added the above methods into the revised SI.

(3) A publication by Jasnin et al. (10.1039/b923878f) on heparin sulfate shows a resolution effect.

Reply 7: Based on the data from Jasnin et al. (10.1039/b923878f), we found that the dynamical transition of heparin sulfate did not exhibit a strong resolution effect. Estimating the dynamical transition of mean square displacement (MSD) for nanosecond motions in all heparan sulfate samples is challenging due to the absence of data on nanosecond motion of HS-dry.

(4) The authors should discuss the impact of the chosen q-range on their findings (see Phys. Chem. Chem. Phys., 2012, 14, 4927-4934, where the authors see a huge effect!).

Reply 8: Following the reviewer's suggestion, we calculated Ton of H-protein in D2O in the q-range from 0.45-0.9 Å⁻¹ and 1.1-1.75 Å⁻¹. The results are summarized in Table S2 and Table S3. As shown in Tables S2-3., the q-range does not alter the Ton of proteins. We added the above results into the revised SI.

(5) The authors underline that the dynamical transition is intrinsic to the protein. However, Cupane et al. (ref 12) have shown that it can also be found in a mixture of amino acids without any protein backbone.

Reply 9: Following the reviewer’s suggestion, we added the following discussion into the revised main text.

“Unfreezing of the protein structural relaxation might facilitate these conformational jumps, turning on its functionality. However, as revealed by Ref (Journal of biological physics, 2010, 36, 291-297.), the denatured form of lysozyme also exhibits a dynamical transition, similar to that seen in its folded native form. Additionally, the dynamical transition also can be found in the mixture of amino acids (Physical Review Letters, 2012, 109, 128102.). Hence, one can argue that the activation of the structural relaxation of the biomolecule above the dynamical transition temperature is a necessary but insufficient condition for the protein to function, as the latter also requires the biomolecule assuming the correctly folded 3-dimensional structure.”

(6) The authors say that they find similar dependences from MSD. They should explain that the MSD is inversely proportional to the summed intensities squared.

Reply 10: Following the reviewer’s suggestion, we added the estimation of mean-squared atomic displacement (MSD) in the revised SI.

“The mean-squared atomic displacement was estimated by performing Gaussian approximation, where . The values of q used for Gaussian fitting ranges from 0.45 to 0.9 Å (Biophys. J. 2006, 91, 2573.).”

(7) A decoupling between water dynamics and membrane dynamics has already been discussed by K. Wood, G. Zaccai et al.

Reply 11: Following the reviewer’s suggestion, we added the discussion in revised main text. “The results from the neutron scattering experiments suggest that the dynamical transition in proteins is an intrinsic property of the biomolecule and strongly depends on the amount of water surrounding it. Such an intrinsic transition can result either from a critical phase transition, e.g., water to ice (PNAS 2007, 104, 18049-18054.; JPCB, 1999, 103, 8036-8050), or from freezing of the structural relaxation of the system beyond the equilibrium time (~100-1000 s) of the experiment, in analogy to the glass transition in polymers from rubbery state to the glass form (Philosophical Magazine, 2004, 84, 1341-1353.; Science, 1995, 267, 1939-1945.; Colloid and Polymer Science, 1995, 273, 413-420.).”

(8) The fact that transition temperature in lipid membranes is higher when the membrane is dry is also well known (A.V. Popova, D.K. Hincha, BMC Biophys. 4, 11 (2011)).

Reply 12: We agree with the reviewer that transition temperature in lipid membranes is higher when the membrane is dry is well known. We cited this work as reference.

(9) The authors should mention the slope (K/min) they used for DSC and discuss the impact of it on the results.

Reply 13: Following the reviewer’s suggestion, we added DSC measurements in revised SI. “DSC measurements were performed by using the METTLER instruments DSC3+. The sample was sealed in a pan of aluminum. An empty pan was used as a reference. All the experiments were carried out in the temperature range from 150 to 300 K with a heating rate of 1 K/min. The heating rate of DSC is the same as neutron experiments.”

(10) In the introduction, the authors should present the different explanations forwarded for the dynamical transition.

Reply 14: Following the reviewer’s suggestion, we added different explanations forwarded for the dynamical transition in the Introduction in revised main text.

“The dynamical transition of protein represents a significant change in the internal mobility of proteins, which has garnered various explanations. One theory suggests it's due to the behavior of water in the hydration shell, transitioning from rigid to fluid at certain temperatures, thus influencing protein flexibility. Another theory considers the transition as an inherent property of the protein, where thermal energy allows the protein to access a wider range of conformations. ”

Reviewer #1 (Recommendations For The Authors):

A major strength of the work is the parallel experiments performed on each of the 4 proteins. To allow better comparison of these it would be helpful to present these combined data in relevant figures to make a side-by-side comparison easier. A summary table of Ton (and potentially TDSC) values would also be helpful.

Reply 15: Following the reviewer’s suggestion, we summarized the Ton of proteins in Table S5 and Table S6.

The effect of hydration on protein structure should be considered. Alterations in protein secondary and tertiary structure would be expected to alter dynamics and thus could be seen as a change in Ton.

Reply 16: The detailed analysis and discussion are presented in Reply 3.

No uncertainty (error) in Ton values is presented. Could these be estimated from e.g. a comparison of protein Ton values measured under identical sample conditions with different spectrometers?

Reply 17: It would be hard to compare Ton of proteins measured with different spectrometers because different spectrometers have different energy resolutions. For example, the energy resolutions of HFBS, DNA and OSIRIS are 1 μeV, 13 μeV, 25.4 μeV and 100 μeV, respectively.

More detail is needed to correctly describe/define the proteins used for the study - e.g. P450 is a family of enzymes, so which one was used?

Reply 18: We used P450 from Pseudomonas putida for the study. The PDB ID is 2ZAX. We added this information in the revised SI.

P450 and myoglobin also have heme cofactors. Were these deuterated as part of the protein preparation?

Reply 19: The heme cofactors were deuterated as part of the protein preparation.  For D-protein, all the cell culture for E.coli is deuterated.

eLife assessment

This study provides important new insight into how non-synaptic interactions affect the activity of adjacent gustatory neurons housed within the same sensillum. The conclusions are supported by convincing electrophysiological, behavioral, and genetic data. This work will be of interest to neuroscientists studying chemosensory processing or regulation of neuronal excitability.

Reviewer #1 (Public Review):

Summary:

This study identifies new types of interactions between Drosophila gustatory receptor neurons (GRNs) and shows that these interactions influence sensory responses and behavior. The authors find that HCN, a hyperpolarization-activated cation channel, suppresses the activity of GRNs in which it is expressed, preventing those GRNs from depleting the sensillum potential, and thereby promotes the activity of neighboring GRNs in the same sensilla. HCN is expressed in sugar GRNs, so HCN dampens excitation of sugar GRNs and promotes excitation of bitter GRNs. Impairing HCN expression in sugar GRNs depletes the sensillum potential and decreases bitter responses, especially when flies are fed on a sugar-rich diet, and this leads to decreased bitter aversion in a feeding assay. The authors' conclusions are supported by genetic manipulations, electrophysiological recordings, and behavioral assays.

Strengths:

(1) Non-synaptic interactions between neurons that share an extracellular environment (sometimes called "ephaptic" interactions) have not been well-studied, and certainly not in the insect taste system. A major strength of this study is the new insight it provides into how these interactions can impact sensory coding and behavior.

(2) The authors use many different types of genetic manipulations to dissect the role of HCN in GRN function, including mutants, RNAi, overexpression, ectopic expression, and neuronal silencing. Their results convincingly show that HCN impacts the sensillum potential and has both cell-autonomous and nonautonomous effects that go in opposite directions. Temporally controlled RNAi experiments suggest that the effect is not due to developmental changes. There are a couple of conflicting or counterintuitive results, but the authors discuss potential explanations.

(3) Experiments comparing flies raised on different food sources suggest an explanation for why the system may have evolved the way that it did: when flies live in a sugar-rich environment, their bitter sensitivity decreases, and HCN expression in sugar GRNs helps to counteract this decrease. New experiments in the revised paper show the timecourse of how sugar diet affects GRN responses and sensillum potential.

Weaknesses/Limitations:

(1) The RNAi Gal80ts experiment only compares responses of experimental flies housed at different temperatures without showing control flies (e.g. Gal4/+ and UAS/+ controls) to confirm that observed differences are not due to nonspecific effects of temperature. Certainly temperature cannot account for sugar and bitter GRN firing rates changing in opposite directions, but it may have some kind of effect.

(2) The experiments where flies are put on sugar vs. sorbitol food show that the diet clearly affects GRN responses and sensillum potential, even for food exposures as short as 1-4 hours, but it is not clear to what extent the GRNs in the labellum are being stimulated during those incubation periods. The flies are most likely not feeding over a 1 hour period if they were not starved beforehand, in which case it is not clear how many times the labellar GRNs would contact the food substrate.

(3) The authors mention that HCN may impact the resting potential in addition to changing the excitability of the cell through various mechanisms. It would be informative to record the resting potential and other neuronal properties, but this is very difficult for GRNs, so the current study is not able to determine exactly how HCN affects GRN activity.

Reviewer #2 (Public Review):

Summary:

In this manuscript, the authors show that HCN loss-of-function mutation causes a decrease in spiking in bitter GRNs (bGRN) while leaving sweet GRN (sGRN) response in the same sensillum intact. They show that a perturbation of HCN channels in sweet-sensing neurons causes a similar decrease while increasing the response of sugar neurons. They were also able to rescue the response by exogenous expression. Ectopic expression of HCN in bitter neurons had no effect. Next, they measure the sensillum potential and find that sensillum potential is also affected by HCN channel perturbation. These findings lead them to speculate that HCN in sGRN increases sGRN spiking, which in turn affects bGRNs. To test this idea, they carried out multiple perturbations aimed at decreasing sGRN activity. They found that reducing sGRN activity by either using receptor mutant or by expressing Kir (a K+ channel) in sGRN increased bGRN responses. These responses also increase the sensillum potential. Finally, they show that these changes are behaviorally relevant as conditions that increase sGRN activity decrease avoidance of bitter substances.

Strengths:

There is solid evidence that perturbation of sweet GRNs affects bitter GRN in the same sensillum. The measurement of transsynaptic potential and how it changes is also interesting and supports the author's conclusion

Weaknesses:

The ionic basis of how perturbation in GRN affects the transepithelial potential, which in turn affects the second neuron, is unclear.

Reviewer #3 (Public Review):

Ephaptic inhibition between neurons housed in the same sensilla has been long discovered in flies, but the molecular basis underlying this inhibition is underexplored. Specifically, it remains poorly understood which receptors or channels are important for maintaining the transepithelial potential between the sensillum lymph and the hemolymph (known as the sensillum potential), and how this affects the excitability of neurons housed in the same sensilla.

Lee et al. used single-sensillum recordings (SSR) of the labellar taste sensilla to demonstrate that the HCN channel, Ih, is critical for maintaining sensillum potential in flies. Ih is expressed in sugar-sensing GRNs (sGRNs) but affects the excitability of both the sGRNs and the bitter-sensing GRNs (bGRNs) in the same sensilla. Ih mutant flies have decreased sensillum potential, and bGRNs of Ih mutant flies have a decreased response to the bitter compound caffeine. Interestingly, ectopic expression of Ih in bGRNs also increases sGRN response to sucrose, suggesting that Ih-dependent increase in sensillum potential is not specific to Ih expressed in sGRNs. The authors further demonstrated, using both SSR and behavior assays, that exposure to sugars in the food substrate is important for the Ih-dependent sensitization of bGRNs. The experiments conducted in this paper are of interest to the chemosensory field. The observation that Ih is important for the activity in bGRNs albeit expressed in sGRNs is especially fascinating and highlights the importance of non-synaptic interactions in the taste system.

Comments on the revised version:

The authors performed additional analyses/experiments to address my previous major points. I'm satisfied with most of their answers:

(1) Sensilla types are labeled in all figures. Proper GAL4 and UAS controls were added to the figures.<br /> (2) Fig. 2A was added to illustrate the important concepts of SP. Fig. 5E was added to show a working model, which could be better but is alright.<br /> (3) Although not in my list of major points, I appreciate the newly added Fig. 5A and 5B, which demonstrate the long-lasting effect of exposure to sugars.<br /> (4) Post-stimulus histogram was added for Fig. 4.<br /> (5) Regarding the expression of Ih in bGRNs and sGRNs, the authors referred to their preprint (Lee et al., 2023, Fig 5C, D, suppl movie 1 and 2). The authors stated that "On the other hand, bGRNs labeled by Gr66a-LexA appeared to colocalize only partially with GFP when the confocal stacks were examined image by image." This interpretation unfortunately does not align with my viewing of the images and the movies. Just looking at the images and the movies alone, one would conclude that Ih is indeed expressed in both bGRNs and sGRNs. Notably, the Ih-TG4.0 is expressed in other non-neuronal cells in the labellum. That being said, I agree with the authors that even if Ih is indeed expressed in bGRNs, it would not affect SP (Fig. 1C, D of this paper, Fig. 5B of Lee et al., 2023 preprint), so I think the authors have addressed my major concern.

Author response:

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

Reviewer #1 (Public Review):

Summary:

This study identifies new types of interactions between Drosophila gustatory receptor neurons (GRNs) and shows that these interactions influence sensory responses and behavior. The authors find that HCN, a hyperpolarization-activated cation channel, suppresses the activity of GRNs in which it is expressed, preventing those GRNs from depleting the sensillum potential, and thereby promoting the activity of neighboring GRNs in the same sensilla. HCN is expressed in sugar GRNs, so HCN dampens the excitation of sugar GRNs and promotes the excitation of bitter GRNs. Impairing HCN expression in sugar GRNs depletes the sensillum potential and decreases bitter responses, especially when flies are fed on a sugar-rich diet, and this leads to decreased bitter aversion in a feeding assay. The authors' conclusions are supported by genetic manipulations, electrophysiological recordings, and behavioral assays.

Strengths:

(1) Non-synaptic interactions between neurons that share an extracellular environment (sometimes called "ephaptic" interactions) have not been well-studied, and certainly not in the insect taste system. A major strength of this study is the new insight it provides into how these interactions can impact sensory coding and behavior.

We appreciate the reviewer’ view that our findings may allow researchers to better understand sensory coding and behavior. However, we respectfully disagree that the SP homeostasis in Drosophila gustation we describe here pertains to ephaptic interaction. Although SP reduction was proposed as the basis of post-ephaptic hyperpolarization in Drosophila olfaction, we find that SP changes are found to be too slow to mediate the fast action of ephaptic inhibition in gustation, reported in the ref#17. We observed a slow, sweet-dependent SP depletion (Fig. 5B, revised), which takes more than one hour. The real-time change of SP was also slow even upon contact with 200-mM sucrose; this result was set aside for another manuscript in preparation. Therefore, we believe the main findings in this paper concern the homeostatic preservation of SP for the maintenance of gustatory function, not ephaptic interaction.

(2) The authors use many different types of genetic manipulations to dissect the role of HCN in GRN function, including mutants, RNAi, overexpression, ectopic expression, and neuronal silencing. Their results convincingly show that HCN impacts the sensillum potential and has both cell-autonomous and nonautonomous effects that go in opposite directions. There are a couple of conflicting or counterintuitive results, but the authors discuss potential explanations.

(3) Experiments comparing flies raised on different food sources suggest an explanation for why the system may have evolved the way that it did: when flies live in a sugar-rich environment, their bitter sensitivity decreases, and HCN expression in sugar GRNs helps to counteract this decrease.

Weaknesses/Limitations:

(1) The genetic manipulations were constitutive (e.g. Ih mutations, RNAi, or misexpression), and depleting Ih from birth could lead to compensatory effects that change the function of the neurons or sensillum. Using tools to temporally control Ih expression could help to confirm the results of this study.

We attempted to address this point by using the tub-Gal80ts system. The result is now included as Fig. 1-figure supplement 2. At 29C, a non-permissive temperature for GAL80ts which allows GAL4-dependent expression Ih-RNAi, we observed that bGRN responses were decreased and sGRN responses were increased compared to the control maintained at 18°C, and this is in parallel with the result in Fig. 1C,D. For this experiment, we inserted “To exclude the possibility that Ih is required for normal gustatory development, we temporally controlled Ih RNAi knockdown to occur only in adulthood, which produced similar results (Fig. 1-figure supplement 2).” (~line 113).

(2) The behavioral experiment shows a striking loss of bitter sensitivity, but it was only conducted for one bitter compound at one concentration. It is not clear how general this effect is. The same is true for some of the bitter GRN electrophysiological experiments that only tested one compound and concentration.

We conducted additional behavioral experiments with other bitters such as lobeline and theophylline (Fig. 5-figure supplement 1), which showed sensitivity losses in Ih mutants similar to caffeine. For these results, the following is inserted at ~line 274: “These results were recapitulated with other bitters, lobeline and theophylline (Fig. 5-figure supplement 1).”

We also added single sensillum recording data with bitters, berberine, lobeline, theophylline and umbelliferone, which yielded results similar to those obtained with caffeine (Fig. 1-figure supplement 1). This is described with the sentence at ~line 105 “Other bitter chemical compounds, berberine, lobeline, theophylline, and umbelliferone, also required Ih for normal bGRN responses (Fig. 1-figure supplement 1).”

(3) Several experiments using the Gal4/UAS system only show the Gal4/+ control and not the UAS/+ control (or occasionally neither control). Since some of the measurements in control flies seem to vary (e.g., spiking rate), it is important to compare the experimental flies to both controls to ensure that any observed effects are in fact due to the transgene expression.

We appreciate the reviewers for raising this point. Indeed, there was a small logical flaw with the controls. We have now included all the necessary controls for Fig. 1C-F, Fig. 2I,J, Fig. 4E, and Fig. 5D, as reviewers suggested. These experiments remained statistically significant after including the new control groups.

(4) I was surprised that manipulations of sugar GRNs (e.g. Ih knockdown, Gr64a-f deletion, or Kir silencing) can impact the sensillum potential and bitter GRN responses even in experiments where no sugar was presented.

We are afraid there is a misunderstanding on the early part of the paper. We suspected that the manipulations impacted bGRNs and SP due to the sweetness in the regular cornmeal food, as stated in lines 214-220 “Typically, we performed extracellular recordings on flies 4-5 days after eclosion, during which they were kept in a vial with fresh regular cornmeal food containing ~400 mM D-glucose. The presence of sweetness in the food would impose long-term stimulation of sGRNs, potentially requiring the delimitation of sGRN excitability for the homeostatic maintenance of gustatory functions. To investigate this possibility, we fed WT and Ihf03355 flies overnight with either non-sweet sorbitol alone (200 mM) or a sweet mixture of sorbitol (200 mM) + sucrose (100 mM).”

I believe the authors are suggesting that the effects of sugar GRN activity (e.g., from consuming sugar in the fly food prior to the experiment) can have long-lasting effects, but it wasn't entirely clear if this is their primary explanation or on what timescale those long-lasting effects would occur. How much / how long of a sugar exposure do the flies need for these effects to be triggered, and how long do those effects last once sugar is removed?

We attempted to address this point with additional experiments (Fig. 5A,B). The reduction of SP could be observed in WT and HCN-deficient mutants with similar degrees 1 hr after the flies were transferred from nonsweet sorbitol-containing vials to sweet sucrose-containing ones. Moreover, the mutants, but not WT, showed further depression of SP when the sweetness persisted in the media for 4 hrs and overnight. This long-term exposure to sweetness longer than 1 hr may simulates the feeding on the regular sweet cornmeal food. The recovery of SP was also tested by removing flies from the sweet media after overnight-long sweet exposure and placing them in sorbitol food. SPs of WT and the mutants were recovered to the similar levels 1 hr after separating the animals from sweetness, although the HCN-lacking mutants showed much lower SP right after overnight sweetness exposure. The unimpaired recovery of the mutants suggests that HCN is independent of generating transepithelial potential itself. Therefore, regardless of HCN, SP changes are not fast even in the presence of strong sweetness, and SP is much better guarded when sGRNs express HCN in a sweet environment.

We inserted the following at ~line 260 to describe the newly added recovery experiment: “Following overnight sweet exposure, SPs of WT and Ihf03355 were recovered to similar levels after 1-hr incubation with sorbitol only food. However, it was after 4 hrs on the sorbitol food that the two lines exhibited SP levels similar to those achieved by overnight incubation with sorbitol only food (Fig. 5B). These results indicate that SP depletion by sweetness is a slow process, and that the dysregulated reduction and recovery of SPs in Ihf03355 manifest only after long-term conditioning with and without sweetness, respectively.”.

(5) The authors mention that HCN may impact the resting potential in addition to changing the excitability of the cell through various mechanisms. It would be informative to record the resting potential and other neuronal properties, but this is very difficult for GRNs, so the current study is not able to determine exactly how HCN affects GRN activity.

On this point, we cannot but rely on previous studies of biophysical and electrophysiological characterization on mammalian HCN channels and a heterologous expression study that revealed a robust hyperpolarization-activated cation current from Drosophila HCN channels (PMID: 15804582).

Reviewer #2 (Public Review):

Summary:

In this manuscript, the authors start by showing that HCN loss-of-function mutation causes a decrease in spiking in bitter GRNs (bGRN) while leaving sweet GRN (sGRN) response in the same sensillum intact. They show that a perturbation of HCN channels in sweet-sensing neurons causes a similar decrease while increasing the response of sugar neurons. They were also able to rescue the response by exogenous expression. Ectopic expression of HCN in bitter neurons had no effect. Next, they measure the sensillum potential and find that sensillum potential is also affected by HCN channel perturbation. These findings lead them to speculate that HCN in sGRN increases sGRN spiking which in turn affects bGRNs. To test this idea that carried out multiple perturbations aimed at decreasing sGRN activity. They found that decreasing sGRN activity by either using receptor mutant or by expressing Kir (a K+ channel) in sGRN increased bGRN responses. These responses also increase the sensillum potential. Finally, they show that these changes are behaviorally relevant as conditions that increase sGRN activity decrease avoidance of bitter substances.

Strengths:

There is solid evidence that perturbation of sweet GRNs affects bitter GRN in the same sensillum. The measurement of transsynaptic potential and how it changes is also interesting and supports the authors' conclusion.

Weaknesses:

The ionic basis of how perturbation in GRN affects the transepithelial potential which in turn affects the second neuron is not clear.

We speculate that HCN-dependent membrane potential regulation, rather than ionic composition change, is responsible for the observed SP preservation, as further discussed as an author response in the section of “Recommendations for the authors”. The transepithelial potential can be dissipated by increased conductance through receptor-linked ion channels following gustatory receptor activation in GRNs. The volume of the sensillum lymph is very small according to electron micrographs of horizontally sliced bristles (PMID: 11456419). Therefore, robust excitation of a gustatory neuron may easily deplete the extracellular potential built as a form of polarized ion concentrations across the tight junction. When the consumption is too strong and extended, the neighboring neuron, which share TEP with the activated GRN, can be negatively affected. We propose that HCN suppresses overexcitation of sGRNs by means of membrane potential stabilization. This stabilization prevents sGRNs from excessively reducing the TEP, thereby protecting the activity of neighboring bGRNs.

Reviewer #3 (Public Review):

Ephaptic inhibition between neurons housed in the same sensilla has been long discovered in flies, but the molecular basis underlying this inhibition is underexplored. Specifically, it remains poorly understood which receptors or channels are important for maintaining the transepithelial potential between the sensillum lymph and the hemolymph (known as the sensillum potential), and how this affects the excitability of neurons housed in the same sensilla.

Although a reduction of sensillum potential was proposed to underlie membrane hyperpolarization of post-ephaptic olfactory neurons in Drosophila, our preliminary data (not shown due to a manuscript in preparation) and the results included in the paper (Fig. 5B) strongly suggest that SP reduction is not a requisite for ephaptic inhibition at least in GRNs. Ephaptic inhibition is expected to be instantaneous, whereas we find that SP reduction in gustation is very slow. Therefore, we would like to indicate that the findings we report in this manuscript are not directly related to ephaptic inhibition.

Lee et al. used single-sensillum recordings (SSR) of the labellar taste sensilla to demonstrate that the HCN channel, Ih, is critical for maintaining sensillum potential in flies. Ih is expressed in sugar-sensing GRNs (sGRNs) but affects the excitability of both the sGRNs and the bitter-sensing GRNs (bGRNs) in the same sensilla. Ih mutant flies have decreased sensillum potential, and bGRNs of Ih mutant flies have a decreased response to the bitter compound caffeine. Interestingly, ectopic expression of Ih in bGRNs also increases sGRN response to sucrose, suggesting that Ih-dependent increase in sensillum potential is not specific to Ih expressed in sGRNs. The authors further demonstrated, using both SSR and behavior assays, that exposure to sugars in the food substrate is important for the Ih-dependent sensitization of bGRNs. The experiments conducted in this paper are of interest to the chemosensory field. The observation that Ih is important for the activity in bGRNs albeit expressed in sGRNs is especially fascinating and highlights the importance of non-synaptic interactions in the taste system.

Despite the interesting results, this paper is not written in a clear and easily understandable manner. It uses poorly defined terms without much elaboration, contains sentences that are borderline unreadable even for those in the narrower chemosensory field, and many figures can clearly benefit from more labeling and explanation. It certainly needs a bit of work.

We would like to revise the language aspect of the manuscript after finalizing the scientific revision.

Below are the major points:

(1) Throughout the paper, it is assumed that Ih channels are expressed in sugar-sensing GRNs but not bitter-sensing GRNs. However, both this paper and citation #17, another paper from the same lab, contain only circumstantial evidence for the expression of Ih channels in sGRNs. A simple co-expression analysis, using the Ih-T2A-GAL4 line and Gr5a-LexA/Gr66a-LexA line, all of which are available, could easily demonstrate the co-expression. Including such a figure would significantly strengthen the conclusion of this paper.

We did conduct confocal imaging with Ih-T2A-Gal4 in combination with GRN Gal4s (ref#17 version2). The expression is very broad, including both neurons and non-neuronal cells. We observed much stronger sGRN expression than bGRN expression. But the promiscuous expression of the reporter in many cells hindered us from clearly demonstrating the void of the reporter in bGRNs. However, the functional and physiological examination of Ih-T2A-Gal4 with the neuronal modifiers such as TRPA1 and Kir2.1 in ref#17 indicates the strong and little expression of Ih in sGRNs and bGRNs, respectively. Furthermore, the RNAi kd results present another line of evidence that HCN expressed in sGRNs regulates SP and bGRN activity (Fig. 1C,D, Fig. 1-figure supplement 2). Ih-RNAi expression in bGRNs did not result in any statistically significant changes in the activities of sGRNs and bGRNs compared to controls (Fig. 1C,D, revised), advocating that Ih acts in sGRNs for the functional homeostasis of SP and GRNs, as we claim.

(2) Throughout this paper, it is often unclear which class of labellar taste sensilla is being recorded. S-a, S-b, I-a, and I-b sensilla all have different sensitivities to bitters and sugars. Each figure should clearly indicate which sensilla is being recorded. Justification should be provided if recordings from different classes of sensilla are being pooled together for statistics.

We mainly performed SSR (single sensillum recording) on i-type bristles as they have the simplest composition of GRNs compared to s- and L-type bristles. As single s-types also contain each of s- and bGRN, we measured SP also for s-types (Figs. 2, 3F and 4D). In case of Fig.3-figure supplement 1, L-types were tested for the relationship between water cell activity and SP. Now all the panels are labelled with the tested bristle types.

(3) In many figures, there is a lack of critical control experiments. Examples include Figures 1C-F (lacking UAS control), Figure 2I-J (lacking UAS control), Figure 4E (lacking the UAS and GAL4 control, and it is also strange to compare Gr64f > RNAi with Gr66a > RNAi, instead of with parental GAL4 and UAS controls.), and Figure 5D (lacking UAS control). Without these critical control experiments, it is difficult to evaluate the quality of the work.

Thank you for pointing this out. We appreciate the feedback and have addressed these concerns by including all the requested controls in the figures. Specifically, we have added the UAS controls for Figs 1C-F and 2I-J, as well as the UAS and GAL4 controls for Fig. 4E. We have also included the UAS control for Fig. 5D.

(4) Figure 2A could benefit from more clarification about what exactly is being recorded here. The text is confusing: a considerable amount of text is spent on explaining the technical details of how SP is recorded, but very little text about what SP represents, which is critical for the readers. The authors should clarify in the text that SP is measuring the potential between the sensillar lymph, where the dendrites of GRNs are immersed, and the hemolymph. Adding a schematic figure to show that SP represents the potential between the sensillar lymph and hemolymph would be beneficial.

SP was defined at lines 55-56 in the first paragraph of introduction, which also contains the background information for SP as a transepithelial potential. As reviewer suggested, we now also included a sentence describing SP (“SP is known as a transepithelial potential between the sensillum lymph and the hemolymph, generated by active ion transport through support cells”, line 126) and a drawing to illustrate the concept of SP (Fig. 2A), and revised the legend.

(5) The sGRN spiking rate in Figure 4B deviates significantly from previous literature (Wang, Carlson, eLife 2022; Jiao, Montell PNAS 2007, as examples), and the response to sucrose in the control flies is not dosage-dependent, which raises questions about the quality of the data. Why are the responses to sucrose not dosage-dependent? The responses are clearly not saturated at these (10 mM to 100 mM) concentrations.

Our recordings show different spiking frequencies from others’ work, because the frequencies are from 5-sec bins not only first 0.5 sec. This lowers the frequencies, as spikes are relatively more frequent in the beginning of the recording (Fig. 4-figure supplement 1).

Why are the responses to sucrose not dosage-dependent? The responses are clearly not saturated at these (10 mM to 100 mM) concentrations.

We were also puzzled with the flat dose dependence to sucrose. This result may suggest the existence of another mechanism moderating sucrose responses of sGRNs. This flat curve reappeared with other genotypes with the same concentration range (5-50 mM) in Fig. 4E. However, 1-mM sucrose produced much lower spiking frequencies (Fig. 4E), suggesting that sGRN responses are saturated at 5 mM sucrose with our recording/analysis condition.

(6) In Figure 4C, instead of showing the average spike rate of the first five seconds and the next 5 seconds, why not show a peristimulus time histogram? It would help the readers tremendously, and it would also show how quickly the spike rate adapts to overexpression and control flies. Also, since taste responses adapt rather quickly, a 500 ms or 1 s bin would be more appropriate than a 5-second bin.

Taste single sensillum recording starts by contacting stimulants, which bars us from recording pre-stimulus responses of GRNs. Therefore, we showed post-stimulus graphs with 1-sec bins (Fig. 4-figure supplement 1) as we reviewer suggested.

(7) Lines 215 - 220. The authors state that the presence of sugars in the culture media would expose the GRNs to sugar constantly, without providing much evidence. What is the evidence that the GRNs are being activated constantly in flies raised with culture media containing sugars? The sensilla are not always in contact with the food.

We agree with reviewer. We replaced “long-term stimulation of sGRNs” with “strong and frequent stimulation of sGRNs for extended period”. The word long-term may be interpreted to be constant.

(8) Line 223. To show that bGRN spike rates in Ih mutant flies "decreased even more than WT", you need to compare the difference in spike rates between the sorbitol group and the sorbitol + sucrose group, which is not what is currently shown.

The data were examined by ANOVA and a multiple comparison test (Dunn’s) between all the groups regardless of genotypes and conditions in the panel (all the groups sharing the y axis). Therefore, the differences were statistically examined. However, the cited expression we used read like it was about the slope or extent of the decrease. We intended to indicate the difference in the absolute values of spiking frequencies after overnight sweet exposure between the genotypes, while bGRN activities were statistically indifferent between WT and Ih mutants when they were kept only on sorbitol food. We revised it to “decreased to the level significantly lower than WT”. We also changed the graph style to effectively present the trend of changes in bGRN sensitivity with comparison between genotypes. Again, the groups were statistically examined together regardless of the genotypes and conditions.

(9) To help readers better understand the proposed mechanisms here, including a schematic figure would be helpful. This should show where Ih is expressed, how Ih in sGRNs impacts the sensillum potential, how elevated sensillum potential increases the electrical driving force for the receptor current, and affects the excitability of the bGRNs in the same sensilla, and how exposure to sugar is proposed to affect ion homeostasis in the sensillum lymph.

As reviewer suggested, we included two panels to show working model for gustatory homeostasis via SP maintenance by HCN (Fig. 5E,F).

Reviewer #1 (Recommendations For The Authors):

(1) The relationship between this paper and the authors' bioRxiv preprint posted last year is not clear. In the introduction they made it seem like this paper is a follow-up that builds on the preprint, but most or all of the experiments in this paper were already performed in the preprint. I guess the authors are planning to divide the original paper into two papers. I would suggest updating the preprint to avoid confusion.

Thank you for the comment. We updated the preprint to be without a part of Fig.6 and entire Fig.7 along with associated texts. As reviewer pointed out, our eLife paper was spun off from the part of the preprint paper, because we feel that the two stories could confuse readers when presented together.

(2) Have the authors considered testing responses of water GRNs? They reside in the same sensilla as sugar neurons, so are they also increased affected by Ih mutation or RNAi in sugar neurons? This would strengthen the evidence that the indirect (non-cell autonomous) effects of Ih are due to the sensillum potential and not some specific interaction between sweet and bitter cells.

As reviewer proposed, we appraised water GRN activity in the L-type bristles of WT, Ihf03355 and a genomic rescue line for Ihf03355. Spiking responses in water GRNs were evoked by hypo-osmolarity of electrolyte (0.1 mM tricholine citrate-TCC). Interestingly, the Ih mutant showed reduced 0.1 mM TCC-provoked spiking frequencies compared to WT. This impairment was rescued by the genomic fragment containing an intact Ih locus (Figure 3-figure supplement 1A).

Additionally, SPs in L-type bristles were reduced by Ih deficiencies but increased in Gr64af, suggesting that HCN regulates sGRNs in L-type bristles as well (Figure 3-figure supplement 1B). Again, the bristles of animals with both mutations together exhibited SPs similar to those of WT.

Furthermore, when we conducted cDNA rescue experiments in L bristles, introduction of Ih-RF cDNA in sGRNs restored SPs, while expressing it in bGRNs did not unlike the results from the i- and s-bristles (Fig. 2K,L), likely because L-bristles lack bGRNs. These cDNA rescue and genetic interaction experiments were conducted using flies fed on fresh cornmeal food with strong sweetness, suggesting that the sweetness in the media is the likely key factor producing the genetic interaction and necessitating HCN, consistent with other results in the manuscript. Therefore, SP regulation by HCN is observed in the L-type bristles.

Minor comments:

Line 52: typo, "Many of"

Thank you. Corrected

Line 95: typo, "sensilla do an sGRN"

Corrected

Line 98: typo, "we observed reduced the spiking responses"

Corrected

Line 206: typo, "a relatively low sucrose concentrations"

Corrected

Line 260: "inverse relationship between the two GRNs in excitability" - I am not exactly sure what data you are referring to.

Although alleles did not show increased sGRN activities, knockdown of Ih decreased bGRN activity but increased sGRN activity (Fig. 1C,D, Fig.1-figure supplement 2B), while suppression of sGRNs increased bGRN activity (Fig. 3). To clarify this point, we revised the phrase to “the inverse relationship between the two GRNs in excitability observed in Fig. 1C,D, Fig. 1-figure supplement 2B, and Fig. 3”.

Methods: typo, "twenty of 3-5 days with 10 males and 10 females"

Corrected to “Twenty flies, aged 3-5 days and consisting of 10 males and 10 females,”

Methods: typo, "Kim's wipes" should be "Kimwipes"

Corrected

Reviewer #2 (Recommendations For The Authors):

(1) More clarification is necessary on Transepithelial potential (TEP). TEP is typically created by having pumps and tight junctions between the sensillar lymph and the hemolymph.

We have an introduction to TEP or SP in the context of sensory functions (lines 40-57) with relevant references. The involvement of pumps and tight junction was mentioned in the same paragraph; “Glia-like support cells exhibit close physical association with sensory receptor neurons, and conduct active transcellular ion transport, which is important for the operation of sensory systems” (line 40) and “Tight junctions between support cells separate the externally facing sensillar lymph from the internal body fluid known as hemolymph” (line 53).

It is not clear how HCN channels in one of the neurons might change the composition of the sensillum lymph. An explanation of their model of how TEP depends on HCN is necessary.

Although the ionic composition of the sensillum lymph is a contributing factor to the sensillum potential, it is more conceptually relevant to describe our findings with the perspective of membrane potential regulation given the role of HCN in membrane potential stabilization as discussed in our manuscript.

We speculate that HCN controls the membrane potential at rest and/or in motion to modulate sGRN activity towards saving SP despite the sweetness in the niche. We positioned our results in relation to SP in discussion; “Our results provide multiple lines of evidence that HCN suppresses HCN-expressing GRNs, thereby sustaining the activity of neighboring GRNs within the same sensilla. We propose that this modulation occurs by restricting SP consumption through HCN-dependent neuronal suppression rather than via chemical and electrical synaptic transmission.” (lines 252-255). Moreover, it is unclear whether HCN is localized to the dendrite bathed in the sensillum lymph to influence the ionic composition of the lymph. It would be very interesting to study in future whether the ionic flow through HCN channels itself is critical for the function of HCN in this context, and whether HCN is exclusively present in the dendrite to support the postulation. However, we would like to remind reviewer that Kir2.1 and HCN channels in sGRNs showed similar effects on SP and bGRNs, while they differ in Na+ conductance.

In the initially submitted manuscript (lines 325-343), we discussed the potential mechanism by which Kir2.1 and HCN channels commonly increase SP in terms of how the membrane potential regulation in the soma can control the SP consumption in the dendrite of sGRNs.

Another point about the TEP that needs some explanation is that these sensilla are open to the environment as tastants must flow in and are different from mechanical sensilla in that sense.

This is a very important question regarding the general physiology of the taste sensilla, as the sensillum lymph is in contact with the external environment through the pore of the sensillum. It is indeed interesting to consider how the composition and potential of the lymph are maintained despite the relatively vast volume of food the sensilla encounter during gustation and the continuous evaporation to air between episodes of gustation. However, we believe that this question, while important, is distinct from the primary focus of our manuscript.

Are the TEP measurements in Figure 2 under control conditions where there are no tastants?

There is no tastant in the SP-measuring glass electrode other than the electrolyte. We apologize that we did not specify the recording electrode condition. We inserted a clause in the method; “For SP recordings, the recording electrode contained 2 mM TCC as the electrolyte, and…”

Does the TEP change dynamically as sGRN is activated?

SP does shift in response to sweets. Please see Fig. 5B. Also, we showed SP changes by mechanical stimuli, which depended on the mechanoreceptor, NompC (Fig. 2D-F). Mechanoreceptor neurons share the sensillum lymph with GRNs.

(2) More clarification on the potential transduction mechanism and how TEP affects one neuron differentially. Essentially, sGRN perturbation affects sGRN activity and it affects the TEP. More explanation is needed for the potential ionic mechanism of each.

Our results strongly suggest that HCN lowers the activity of HCN-expressing GRNs, mitigating SP consumption. This modulation is crucial because the SP serves as a driving force for neuronal activation within the sensillum. HCN is particularly necessary in sGRNs because of the flies’ sweet feeding niche, which is expected to result in frequent and strong activation of sGRNs. The SP saved by HCN-dependent delimitation of sGRNs can be used to raise the responsibility of bGRNs.

(3) The authors refer to their own unreviewed paper (Reference 17). This paper is on a similar topic and there seems to be some overlap. Clarification on this point would be important.

We revised the biorxiv preprint, so that the preprint version 2 does not contain the parts overlapping with this eLife paper. This eLife paper was originally part of the preprint paper, but it was separated to clarify the messages of the two stories. As we explained in Discussion (lines 276-297), HCN provides resistance to both hyperpolarization and depolarization of the membrane potential. Simply put, one paper focuses on the role of HCN in resisting hyperpolarization, while the other (this paper in eLife) focuses on resisting depolarization.

(4) Methods are sparse. Many details on the method are necessary. For example, Sensilla recordings are being done by the tip-dip method (I assume). What does "number of experiments" mean in Figure 1? Is it the number of animals or the number of sensilla? How many trials/sensilla?

We indicated the extracellular recording was performed by the tip-dip method; “In vivo extracellular recordings were performed by the tip-dip method as detailed previously”. We also added a statement on the number of experiments; “The number of experiments indicated in figures are the number of naïve bristles tested. The naïve bristles were from at least three different animals.”

(5) Figure 1: I understand the author's interpretation. But if one compares WT in Figure 1A to Gr64a-IhRNAi in 1C, we can come to the conclusion that there is no change. In other words, the control in Figure 1C (grey) has a much higher response than WT. Similar conclusions can be made for other experiments. Is the WT response stable enough to make the conclusions made here?

The genetic background of each genotype may influence GRN activity to some extent. RNAi knockdown experiments are well-known for their hypomorphic nature, and their effects should be evaluated by comparison with their parental controls such as Gal4 and UAS lines. As all reviewers pointed out, we added the results from UAS control. This effort confirms that Gr89a>Ih RNAi is statistically indifferent to UAS control as well as Gr64f-Gal4 control in bGRN spiking evoked by 2-mM caffeine, while Gr64f>Ih RNAi showed reduced bGRN responses to 2 mM caffeine compared to all the controls.

(6) Figure 3: Why is bGRN spiking not plotted against sensillum potential to observe the dependence more directly?

This is a very interesting suggestion. We are not, however, equipped to measure spiking and sensillum potential simultaneously. Therefore, they are independent experiments, and we treated them accordingly.

(7) Figure 4: Why bGRN response is only affected at high caffeine concentrations is not clear.

We were also surprised by the differences in the dose dependence results of b- and sGRNs, genetically manipulated to mis-express and over-express HCN in Fig. 4A and 4E, respectively. Each gustatory neuron likely has distinct sets of players and parameters that set its own membrane potential and excitability.

We can think of a possibility that there might be a range of membrane potentials within which HCN does not engage. In bGRNs, the resting membrane potential may lie low within this range, so that some degrees of membrane depolarization by low concentrations of caffeine do not significantly close HCN channels, thus preventing their hyperpolarizing effects. On the other hand, the membrane potential of sGRNs may be high within this range, showing suppressive effects at all tested sucrose concentrations. However, we find this explanation is too speculative to include in the main text, while we stated in the original manuscript, “implying a complex cell-specific regulation of GRN excitability.” (line 210).

(8) Minor:

L98 - there is a small typo

Corrected

L274: "funny" !?

“Funny” currents, denoted If, were initially observed by electrophysiologists and later attributed to HCN channels, now indicated by Ih (thus the gene name Ih in Drosophila). These currents were termed "funny" due to their unusual properties compared to other currents. For more detailed information, please refer to the cited references.

L257: Neuropeptide seemed to be abrupt

We attempted to discuss possible mechanisms that mediate excitability changes across GRNs beyond the mechanism by SP shifts. Neuropeptides, which are chemical neurotransmitters along with small neurotransmitters, were mentioned following the discussion on synaptic transmission to suggest alternative pathways for excitability regulation. This inclusion is meant to provide a comprehensive overview of potential mechanisms influencing GRN activity.

Reviewer #3 (Recommendations For The Authors):

Congratulations on your fascinating research! The results are certainly of interest to the chemosensory field. However, I suggest using academic editing services to enhance the clarity of your text and ensure that the terminology and jargon align with standard usage in the field. The current choice of words may not be consistent with commonly used terms. As it is now, the writing might not fully showcase the compelling story and the effort behind your study, and is underselling your interesting results. Proper refinement could make sure your valuable findings are appropriately recognized.

We appreciate your comments and apologize for any difficulties reviewers faced during the review process. We are currently prioritizing the review of scientific content and plan to address language issues in a subsequent revision. It would be very helpful for future revisions if the problematic sentences or expressions could be indicated in detail after this revision. This will allow us to ensure that our terminology and expression align with standard usage in the field, and that our findings are clearly and effectively communicated.

Minor points:

(1) Line 110: what is Ih-RF?

We apologize that we relied on a reference in describing the cDNA. The following clause was inserted with additional reference and the Flybase id: “(Flybase id: FBtr0290109), which previously rescued Ih deficiency in other contexts17,26 ,”  

(2) Line 158: Gr64af mutant flies still have Gr5a and a residual response to fructose and sucrose (Slone, Amrein 2007).

We revised the line to “is severely impaired in sucrose and glucose sensing”, since there is a substantial loss of sucrose and glucose sensing in both Gr64af from Kim et al 2018 and DGr64 from Slone et al 2007, when they were examined by the proboscis extension reflex assay. This was also confirmed in the study by Jiao et al 2009. We also deleted “sugar-ageusic” and instead describe the mutant “impaired in sucrose and glucose sensing” in Fig. 3 legend.

(3) Lines 264-273 seem unnecessary. This paper is not about the function of HCN in mammals, and these discussions seem largely irrelevant.

We feel that it is important to position our results within a broader context by discussing the potential implications of our findings for sensory systems of other animals. As we stated, HCN channels have been localized in mammalian sensory systems, but their roles are often not well understood. By including this discussion, we aim to highlight the relevance of our findings beyond the model organism used in our study and suggest possible areas for future research in mammalian systems.

eLife assessment

This important study enhances our understanding of how habitat fragmentation and climate change jointly influence bird community thermophilization in a fragmented island system. The evidence supporting some conclusions is incomplete, as while the overall trends are convincing, some methodological aspects, particularly the isolation metrics and interpretation of colonization/extinction rates, require further clarification. This work will be of broad interest to ecologists and conservation biologists, providing crucial insights into how ecosystems and communities react to climate change.

Reviewer #1 (Public Review):

Summary:

This study reports on the thermophilization of bird communities in a network of islands with varying areas and isolation in China. Using data from 10 years of transect surveys, the authors show that warm-adapted species tend to gradually replace cold-adapted species, both in terms of abundance and occurrence. The observed trends in colonisations and extinctions are related to the respective area and isolation of islands, showing an effect of fragmentation on the process of thermophilization.

Strengths:

Although thermophilization of bird communities has been already reported in different contexts, it is rare that this process can be related to habitat fragmentation, despite the fact that it has been hypothesized for a long time that it could play an important role. This is made possible thanks to a really nice study system in which the construction of a dam has created this incredible Thousand Islands lake. Here, authors do not simply take observed presence-absence as granted and instead develop an ambitious hierarchical dynamic multi-species occupancy model. Moreover, they carefully interpret their results in light of their knowledge of the ecology of the species involved.

Weaknesses:

Despite the clarity of this paper on many aspects, I see a strong weakness in the authors' hypotheses, which obscures the interpretation of their results. Looking at Figure 1, and in many sentences of the text, a strong baseline hypothesis is that thermophilization occurs because of an increasing colonisation rate of warm-adapted species and extinction rate of cold-adapted species. However, there does not need to be a temporal trend! Any warm-adapted species that colonizes a site has a positive net effect on CTI; similarly, any cold-adapted species that goes extinct contributes to thermophilization.

Another potential weakness is that fragmentation is not clearly defined. Generally, fragmentation sensu lato involves both loss of habitat area and changes in the spatial structure of habitats (i.e. fragmentation per se). Here, both area and isolation are considered, which may be slightly confusing for the readers if not properly defined.

Reviewer #2 (Public Review):

Summary:

This study addresses whether bird community reassembly in time is related to climate change by modelling a widely used metric, the community temperature index (CTI). The authors first computed the temperature index of 60 breeding bird species thanks to distribution atlases and climatic maps, thus obtaining a measure of the species realized thermal niche.

These indices were aggregated at the community level, using 53 survey transects of 36 islands (repeated for 10 years) of the Thousand Islands Lake, eastern China. Any increment of this CTI (i.e. thermophilization) can thus be interpreted as a community reassembly caused by a change in climate conditions (given no confounding correlations).

The authors show thanks to a mix of Bayesian and frequentist mixed effect models to study an increment of CTI at the island level, driven by both extinction (or emigration) of cold-adapted species and colonization of newly adapted warm-adapted species. Less isolated islands displayed higher colonization and extinction rates, confirming that dispersal constraints (created by habitat fragmentation per se) on colonization and emigration are the main determinants of thermophilization. The authors also had the opportunity to test for habitat amount (here island size). They show that the lack of microclimatic buffering resulting from less forest amount (a claim backed by understory temperature data) exacerbated the rates of cold-adapted species extinction while fostering the establishment of warm-adapted species.

Overall these findings are important to range studies as they reveal the local change in affinity to the climate of species comprising communities while showing that the habitat fragmentation VS amount distinction is relevant when studying thermophilization. As is, the manuscript lacks a wider perspective about how these results can be fed into conservation biology, but would greatly benefit from it. Indeed, this study shows that in a fragmented reserve context, habitat amount is very important in explaining trends of loss of cold-adapted species, hinting that it may be strategic to prioritize large habitats to conserve such species. Areas of diverse size may act as stepping stones for species shifting range due to climate change, with small islands fostering the establishment of newly adapted warm-adapted species while large islands act as refugia for cold-adapted species. This study also shows that the removal of dispersal constraints with low isolation may help species relocate to the best suitable microclimate in a heterogenous reserve context.

Strength:

The strength of the study lies in its impressive dataset of bird resurveys, that cover 10 years of continued warming (as evidenced by weather data), 60 species in 36 islands of varying size and isolation, perfect for disentangling habitat fragmentation and habitat amount effects on communities. This distinction allows us to test very different processes mediating thermophilization; island area, linked to microclimatic buffering, explained rates for a variety of species. Dispersal constraints due to fragmentation were harder to detect but confirms that fragmentation does slow down thermophilization processes.

This study is a very good example of how the expected range shift at the biome scale of the species materializes in small fragmented regions. Specifically, the regional dynamics the authors show are analogous to what processes are expected at the trailing and colonizing edge of a shifting range: warmer and more connected places display the fastest turnover rates of community reassembly. The authors also successfully estimated extinction and colonization rates, allowing a more mechanistic understanding of CTI increment, being the product of two processes.

The authors showed that regional diversity and CTI computed only by occurrences do not respond in 10 years of warming, but that finer metrics (abundance-based, or individual islands considered) do respond. This highlights the need to consider a variety of case-specific metrics to address local or regional trends. Figure Appendix 2 is a much-appreciated visualization of the effect of different data sources on Species thermal Index (STI) calculation.

The methods are long and diverse, but they are documented enough so that an experienced user with the use of the provided R script can follow and reproduce them.

Weaknesses:

While the overall message of the paper is supported by data, the claims are not uniformly backed by the analysis. The trends of island-specific thermophilization are very credible (Figure 3), however, the variable nature of bird observations (partly compensated by an impressive number of resurveys) propagate a lot of errors in the estimation of species-specific trends in occupancy, abundance change, and the extinction and colonization rates. This materializes into a weak relationship between STI and their respective occupancy and abundance change trends (Figure 4a, Figure 5, respectively), showing that species do not uniformly contribute to the trend observed in Figure 3. This is further shown by the results presented in Figure 6, which present in my opinion the topical finding of the study. While a lot of species rates response to island areas are significant, the isolation effect on colonization and extinction rates can only be interpreted as a trend as only a few species have a significant effect. The actual effect on the occupancy change rates of species is hard to grasp, and this trend has a potentially low magnitude (see below).

While being well documented, the myriad of statistical methods used by the authors ampere the interpretation of the figure as the posterior mean presented in Figure 4b and Figure 6 needs to be transformed again by a logit-1 and fed into the equation of the respective model to make sense of. I suggest a rewording of the caption to limit its dependence on the method section for interpretation.

By using a broad estimate of the realized thermal niche, a common weakness of thermophilization studies is the inability to capture local adaptation in species' physiological or behavioral response to a rise in temperature. The authors however acknowledge this limitation and provide specific examples of how species ought to evade high temperatures in this study region.

Reviewer #3 (Public Review):

Summary:

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

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

Strengths:

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

Weaknesses:

The metric of island isolation based on the distance to the mainland seems a bit too oversimplified as in real life the study system rather represents an island network where the islands of different sizes are in varying distances to each other, such that smaller islands can potentially draw from the species pools from near-by larger islands too - rather than just from the mainland. Thus a more holistic network metric of isolation could have been applied or at least discussed for future research. The fact, that the authors did find a signal of island isolation does support their method, but the variation in responses to this metric could hint at a more complex pattern going on in real-life than was assumed for this study.<br /> Further, the link between larger areas and higher habitat diversity or heterogeneity could be presented by providing evidence for this relationship. The authors do make a reference to a paper done in the same study system, but a more thorough presentation of it would strengthen this assumption further.

Despite the general clear patterns found in the paper, there were some idiosyncratic responses. Those could be due to a multitude of factors which could be discussed a bit better to inform future research using a similar study design.

Author response:

We would like to 1) response one comment from the public review, which is also related to the eLife assessment, and 2) give provisional author responses.

(1) Regarding the definition of the colonization-extinction rate, the first reviewer may misunderstand it: “However, there does not need to be a temporal trend! Any warm-adapted species that colonizes a site has a positive net effect on CTI; similarly, any cold-adapted species that goes extinct contributes to thermophilization.” We here clarify the definition:

In a single iteration of our MSOM (Multi-species occupancy model), the occupancy rate of species[n] in transect[i] from year[t-1] to year[t] is related to the colonization rate and extinction rate, and is defined as:<br /> muz[n,i,t] = z[n,i,t-1]*(1-eps[n,i,t-1]) + (1-z[n,i,t-1])*gam[n,i,t-1], (also shown in Line411 in our MS).

If the colonization rate (gam) and extinction rate (eps) remain constant, the occupancy rate(muz) will be a constant number which is related to the state of real occupancy (0 or 1). The occupancy rate will only increase if colonization rate increases (or the extinction rate decreases). That is why we are considering the temporal trend in colonization/extinction rate.

(2) Provisional author responses:

We will revise and improve the manuscript according to the public reviews and mainly focus on:

(1) clarify the general definition of habitat fragmentation in the Introduction.

(2) provide a wider perspective about how our results can be applied to conservation biology in the Discussion.

(3) discuss the diversity of isolation metrics for future research and provide more evidence about the link between larger areas and higher habitat diversity or heterogeneity.

eLife assessment

This potentially important study employs advanced imaging techniques to directly visualize molecular dynamics and of the immune receptor kinase FLS2 in specific microenvironments. The evidence supporting the ligand-induced association with remorin and the requirement of a previously reported phosphosite as presented is solid, although support by independent methods would be welcome. The work will be of interest to plant biologists working on cell surface receptors.

eLife assessment

This important study explores how cells maintain subcellular structures in the face of constant protein turnover, focusing on neurons, whose synapses must be kept stable over long periods of time for memory storage. Using proteins from knock-in mice expressing tagged variants of the synaptic scaffold protein PSD95, nanobodies, and multiple imaging methods, there is compelling evidence that PSD95 proteins form complexes at synapses in which single protein copies are sequentially replaced over time. This happens at different rates in different synapse types and is slowest in areas where PSD95 lifetime is the longest and long-term memories are stored. While of general relevance to cell biology, these findings are of particular interest to neuroscientists because they support the notion put forward by Francis Crick that stable synapses, and hence stable long-term memories, can be maintained in the face of short protein lifetimes by sequential replacement of individual subunits in synaptic protein complexes.

Joint Public Review:

The present study explored the principles that allow cells to maintain complex subcellular proteinaceous structures despite the limited lifetimes of the individual protein components. This is particularly critical in the case of neurons, where the size and protein composition of synapses define synaptic strength and encode memory.

PSD95 is an abundant synapse protein that acts as a scaffold in the recruitment of transmitter receptors and other signaling proteins and is required for proper memory formation. The authors used super-resolution microscopy to study PSD95 super-complexes isolated from the brains of mice expressing tagged PSD variants (Halo-Tag, mEos, GFP). Their results show compellingly that a large fraction (~25%) of super-complexes contains two PSD95 copies about 13 nm apart, that there is substantial turnover of PSD95 proteins in super-complexes over a period of seven days, and that ~5-20% of the super-complexes contain new and old PSD95 molecules. This percentage is higher in synaptic fractions as compared to total brain lysates, and highest in isocortex samples (~20%). These important findings support the notion put forward by Crick that sequential subunit replacement gives synaptic super-complexes long lifetimes and thus aids in memory maintenance. Overall, this is very interesting, providing key insights into how synaptic protein complexes are formed and maintained. On the other hand, the actual role of these PSD95 super-complexes in long-term memory storage remains unknown.

Strengths

(1) The study employed an appropriate and validated methodology.

(2) Large numbers of PSD95 super-complexes from three different mouse models were imaged and analyzed, providing adequately powered sample sizes.

(3) State-of-the-art super-resolution imaging techniques (PALM and MINFLUX) were used, providing a robust, high-quality, cross-validated analysis of PSD95 protein complexes that is useful for the community.

(4) The result that PSD95 proteins in dimeric complexes are on average 12.7 nm apart is useful and has implications for studies on the nanoscale organization of PSD95 at synapses.

(5) The finding that postsynaptic protein complexes can continue to exist while individual components are being renewed is important for our understanding of synapse maintenance and stability.

(6) The data on the turnover rate of PSD95 in super-complexes from different brain regions provide a first indication of potentially meaningful differences in the lifetime of super-complexes between brain regions.

Weaknesses

(1) The manuscript emphasizes the hypothesis that stable super-complexes, maintained through sequential replacement of subunits, might underlie the long-term storage of memory. While an interesting idea, this notion requires considerably more research. The presented experimental data are indeed consistent with this notion, but there is no evidence that these complexes are causally related to memory storage.

(2) Much of the presented work is performed on biochemically isolated protein complexes. The biochemical isolation procedures rely on physical disruption and detergents that are known to alter the composition and structure of complexes in certain cases. Thus, it remains unclear how the protein complexes described in this study relate to PSD95 complexes in intact synapses.

(3) Because not all GFP molecules mature and fold correctly in vitro and the PSD95-mEos mice used were heterozygous, the interpretation of the corresponding quantifications is not straightforward.

(4) It was not tested whether different numbers of PSD95 molecules per super-complex might contribute to different retention times of PSD95, e.g. in synaptic vs. total-forebrain super-complexes.

(5) The conclusion that the population of 'mixed' synapses is higher in the isocortex than in other brain regions is not supported by statistical analysis.

(6) The validity of conclusions regarding PSD95 degradation based on relative changes in the occurrence of SiR-Halo-positive puncta is limited.

Author response:

(1) The manuscript emphasizes the hypothesis that stable super-complexes, maintained through sequential replacement of subunits, might underlie the long-term storage of memory. While an interesting idea, this notion requires considerably more research. The presented experimental data are indeed consistent with this notion, but there is no evidence that these complexes are causally related to memory storage. 

We agree with the reviewer that, while our data support the idea that subunit exchange in supercomplexes could underlie long-term memory storage, more research is necessary to conclusively validate this hypothesis. The experimental data presented are consistent with the idea that stable supercomplexes, maintained through sequential replacement of subunits, play a role in memory retention. However, establishing a causal relationship between these supercomplexes and memory storage will require additional experiments and in-depth analyses.

(2) Much of the presented work is performed on biochemically isolated protein complexes. The biochemical isolation procedures rely on physical disruption and detergents that are known to alter the composition and structure of complexes in certain cases. Thus, it remains unclear how the protein complexes described in this study relate to PSD95 complexes in intact synapses. 

Whilst it could be the case that biochemical isolation procedures have the potential to alter the composition and structure of protein complexes, we have previously published the protocol used to isolate PSD95-containing supercomplexes (Nat Commun. 2016; 7: 11264). In that study, we demonstrated that the isolated supercomplexes are approximately 1.5 MDa in size and contain multiple proteins, including other scaffolding proteins (e.g., PSD93) and receptors (e.g., NMDARs). Importantly, these supercomplexes remain stable when exposed to detergents and dilution, strongly indicating that they represent the native complexes present in intact synapses.

(3) Because not all GFP molecules mature and fold correctly in vitro and the PSD95-mEos mice used were heterozygous, the interpretation of the corresponding quantifications is not straightforward. 

Although genetic tagging ensures a 1:1 labeling stoichiometry, we acknowledge that the presence of unfolded GFP and the use of heterozygous PSD95-mEos mice can complicate the analysis. We have highlighted this limitation in the manuscript. Nonetheless, our results show a high level of consistency across the different genetic fusions used in this study.

(4) It was not tested whether different numbers of PSD95 molecules per super-complex might contribute to different retention times of PSD95, e.g. in synaptic vs. total-forebrain super-complexes. 

The potential impact of varying numbers of PSD95 molecules per super-complex on retention times was considered. However, our analysis showed minimal differences in the distribution of molecule numbers per super-complex between the synaptic and forebrain samples.

(5) The conclusion that the population of 'mixed' synapses is higher in the isocortex than in other brain regions is not supported by statistical analysis. 

The conclusion that the population of 'mixed' synapses is higher in the isocortex than in other brain regions is indeed supported by statistical analysis. All relevant statistical data are detailed in Table S2, and the finding is statistically significant. We will emphasize this point in the revised manuscript.

(6) The validity of conclusions regarding PSD95 degradation based on relative changes in the occurrence of SiR-Halo-positive puncta is limited.

We recognize that conclusions based solely on the relative changes in SiR-Halo-positive puncta concerning PSD95 degradation have limitations. To address this, we also quantified the “new” PSD95 by analyzing AF488-Halo-positive molecules.