Author response:

Reviewer #2 (Public Review):

M. El Amri et al., investigated the functions of Marcks and Marcks like 1 during spinal cord (SC) development and regeneration in Xenopus laevis. The authors rigorously performed loss of function with morpholino knock-down and CRISPR knock-out combining rescue experiments in developing spinal cord in embryo and regeneration in tadpole stage.

For the assays in the developing spinal cord, a unilateral approach (knock-down/out only one side of the embryo) allowed the authors to assess the gene functions by direct comparing one-side (e.g. mutated SC) to the other (e.g. wild type SC on the other side). For the assays in regenerating SC, the authors microinject CRISPR reagents into 1-cell stage embryo. When the embryo (F0 crispants) grew up to tadpole (stage 50), the SC was transected. They then assessed neurite outgrowth and progenitor cell proliferation. The validation of the phenotypes was mostly based on the quantification of immunostaining images (neurite outgrowth: acetylated tubulin, neural progenitor: sox2, sox3, proliferation: EdU, PH3), that are simple but robust enough to support their conclusions. In both SC development and regeneration, the authors found that Marcks and Marcksl1 were necessary for neurite outgrowth and neural progenitor cell proliferation.

The authors performed rescue experiments on morpholino knock-down and CRISPR knock-out conditions by Marcks and Marcksl1 mRNA injection for SC development and pharmacological treatments for SC development and regeneration. The unilateral mRNA injection rescued the loss-of-function phenotype in the developing SC. To explore the signalling role of these molecules, they rescued the loss-of-function animals by pharmacological reagents They used S1P: PLD activator, FIPI: PLD inhibitor, NMI: PIP2 synthesis activator and ISA-2011B: PIP2 synthesis inhibitor. The authors found the activator treatment rescued neurite outgrowth and progenitor cell proliferation in loss of function conditions. From these results, the authors proposed PIP2 and PLD are the mediators of Marcks and Marcksl1 for neurite outgrowth and progenitor cell proliferation during SC development and regeneration. The results of the rescue experiments are particularly important to assess gene functions in loss of function assays, therefore, the conclusions are solid. In addition, they performed gain-of-function assays by unilateral Marcks or Marcksl1 mRNA injection showing that the injected side of the SC had more neurite outgrowth and proliferative progenitors. The conclusions are consistent with the loss-of-function phenotypes and the rescue results. Importantly, the authors showed the linkage of the phenotype and functional recovery by behavioral testing, that clearly showed the crispants with SC injury swam less distance than wild types with SC injury at 10-day post surgery.

Prior to the functional assays, the authors analyzed the expression pattern of the genes by in situ hybridization and immunostaining in developing embryo and regenerating SC. They confirmed that the amount of protein expression was significantly reduced in the loss of function samples by immunostaining with the specific antibodies that they made for Marcks and Marcksl1. Although the expression patterns are mostly known in previous works during embryo genesis, the data provided appropriate information to readers about the expression and showed efficiency of the knock-out as well.

MARCKS family genes have been known to be expressed in the nervous system. However, few studies focus on the function in nerves. This research introduced these genes as new players during SC development and regeneration. These findings could attract broader interests from the people in nervous disease model and medical field. Although it is a typical requirement for loss of function assays in Xenopus laevis, I believe that the efficient knock-out for four genes by CRISPR/Cas9 was derived from their dedication of designing, testing and validation of the gRNAs and is exemplary.

Weaknesses,

(1) Why did the authors choose Marcks and Marcksl1? The authors mentioned that these genes were identified with a recent proteomic analysis of comparing SC regenerative tadpole and non-regenerative froglet (Line (L) 54-57). However, although it seems the proteomic analysis was their own dataset, the authors did not mention any details to select promising genes for the functional assays (this article). In the proteomic analysis, there must be other candidate genes that might be more likely factors related to SC development and regeneration based on previous studies, but it was unclear what the criteria to select Marcks and Marcksl1 was.

To highlight the rationale for selecting these proteins, we reworded the sentence as follows: “A recent proteomic screen … after SCI identified a number of proteins that are highly upregulated at the tadpole stage but downregulated in froglets (Kshirsagar, 2020). These proteins included Marcks and Marcksl1, which had previously been implicated in the regeneration of other tissues (El Amri et al., 2018) suggesting a potential role for these proteins also in spinal cord regeneration.”

(2) Gene knock-out experiments with F0 crispants,

The authors described that they designed and tested 18 sgRNAs to find the most efficient and consistent gRNA (L191-195). However, it cannot guarantee the same phenotypes practically, due to, for example, different injection timing, different strains of Xenopus laevis, etc. Although the authors mentioned the concerns of mosaicism by themselves (L180-181, L289-292) and immunostaining results nicely showed uniformly reduced Marcks and Marcksl1 expression in the crispants, they did not refer to this issue explicitly.

To address this issue, we state explicitly in line 208-212: “We also confirmed by immunohistochemistry that co-injection of marcks.L/S and marcksl1.L/S sgRNA, which is predicted to edit all four homeologs (henceforth denoted as 4M CRISPR) drastically reduced immunostaining for Marcks and Marcksl1 protein on the injected side (Fig. S6 B-G), indicating that protein levels are reduced in gene-edited embryos.”

(3) Limitations of pharmacological compound rescue

In the methods part, the authors describe that they performed titration experiments for the drugs (L702-704), that is a minimal requirement for this type of assay. However, it is known that a well characterized drug is applied, if it is used in different concentrations, the drug could target different molecules (Gujral TS et al., 2014 PNAS). Therefore, it is difficult to eliminate possibilities of side effects and off targets by testing only a few compounds.

As explained in the responses to reviewer 1, we have completely rewritten and toned down our presentation of the pharmacological result and explicitly mention in our discussion now the possibility of side effects.

Author response:

We agree with reviewer #1 to remove the mGluR6b data. It is indeed a weakness and is too preliminary. We will gladly remove it from the revised version.

We will address the issue of the bulk responses (depicted in Figures 5 and 6) by showing the significance data, arguing that although we cannot prove that prey-detection is increased for lower intensities, the bulk effect is significant, so prey detection is effectively stronger.

Author response:

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

Reviewer #1 (Public Review):

Summary:

Previous work demonstrated a strong bias in the percept of an ambiguous Shepard tone as either ascending or descending in pitch, depending on the preceding contextual stimulus. The authors recorded human MEG and ferret A1 single-unit activity during presentation of stimuli identical to those used in the behavioral studies. They used multiple neural decoding methods to test if context-dependent neural responses to ambiguous stimulus replicated the behavioral results. Strikingly, a decoder trained to report stimulus pitch produced biases opposite to the perceptual reports. These biases could be explained robustly by a feed-forward adaptation model. Instead, a decoder that took into account direction selectivity of neurons in the population was able to replicate the change in perceptual bias.

Strengths:

This study explores an interesting and important link between neural activity and sensory percepts, and it demonstrates convincingly that traditional neural decoding models cannot explain percepts. Experimental design and data collection appear to have been executed carefully. Subsequent analysis and modeling appear rigorous. The conclusion that traditional decoding models cannot explain the contextual effects on percepts is quite strong.

Weaknesses:

Beyond the very convincing negative results, it is less clear exactly what the conclusion is or what readers should take away from this study. The presentation of the alternative, "direction aware" models is unclear, making it difficult to determine if they are presented as realistic possibilities or simply novel concepts. Does this study make predictions about how information from auditory cortex must be read out by downstream areas? There are several places where the thinking of the authors should be clarified, in particular, around how this idea of specialized readout of direction-selective neurons should be integrated with a broader understanding of auditory cortex.

While we have not used the term "direction aware", we think the reviewer refers generally to the capability of our model to use a cell's direction selectivity in the decoding. In accordance with the reviewer's interpretation, we did indeed mean that the decoder assumes that a neuron does not only have a preferred frequency, but also a preferred direction of change in frequency (ascending/descending), which is what we use to demonstrate that the decoding in this way aligns with the human percept. We have adapted the text in several places to clarify this, in particular expanding the description in the Methods substantially.

Reviewer #2 (Public Review):

The authors aim to better understand the neural responses to Shepard tones in auditory cortex. This is an interesting question as Shepard tones can evoke an ambiguous pitch that is manipulated by a proceeding adapting stimulus, therefore it nicely disentangles pitch perception from simple stimulus acoustics.

The authors use a combination of computational modelling, ferret A1 recordings of single neurons, and human EEG measurements.

Their results provide new insights into neural correlates of these stimuli. However, the manuscript submitted is poorly organized, to the point where it is near impossible to review. We have provided Major Concerns below. We will only be able to understand and critique the manuscript fully after these issues have been addressed to improve the readability of the manuscript. Therefore, we have not yet reviewed the Discussion section.

Major concerns

Organization/presentation

The manuscript is disorganized and therefore difficult to follow. The biggest issue is that in many figures, the figure subpanels often do not correspond to the legend, the main body, or both. Subpanels described in the text are missing in several cases.

We have gone linearly through the text and checked that all figure subpanels are referred to in the text and the legend. As far as we can tell, this was already the case for all panels, with the exception of two subpanels of Fig. 5.

Many figure axes are unlabelled.

We have carefully checked the axes of all panels and all but two (Fig. 5D) were labeled. As is customary, certain panels inherit the axis label from a neighboring panel, if the label is the same, e.g. subpanels in Fig. 6F or Fig. 5E, which helps to declutter the figure. We hope that with this clarification, the reviewer can understand the labels of each panel.

There is an inconsistent style of in-text citation between figures and the main text. The manuscript contains typos and grammatical errors. My suggestions for edits below therefore should not be taken as an exhaustive list. I ask the authors to consider the following only a "first pass" review, and I will hopefully be able to think more deeply about the science in the second round of revisions after the manuscript is better organized.

While we are puzzled by the severity of issues that R2 indicates (see above, and R3 qualifies it as "well written", and R1 does not comment on the writing negatively), we have carefully gone through all specific issues mentioned by R2 and the other reviewers. We hope that the revised version of the paper with all corrections and clarifications made will resolve any remaining issues.

Frequency and pitch

The terms "frequency" and "pitch" seem to be used interchangeably at times, which can lead to major misconceptions in a manuscript on Shepard tones. It is possible that the authors confuse these concepts themselves at times (e.g. Fig 5), although this would be surprising given their expertise in this field. Please check through every use of "frequency" and "pitch" in this manuscript and make sure you are using the right term in the right place. In many places, "frequency" should actually be "fundamental frequency" to avoid misunderstanding.

Thanks for pointing this out. We have checked every occurrence and modified where necessary.

Insufficient detail or lack of clarity in descriptions

There seems to be insufficient information provided to evaluate parts of these analysis, most critically the final pitch-direction decoder (Fig 6), which is a major finding. Please clarify.

Thanks for pointing this out. We have extended the description of the pitch-direction decoder and highlighted its role for interpreting the results.

Reviewer #3 (Public Review):

Summary:

This is an elegant study investigating possible mechanisms underlying the hysteresis effect in the perception of perceptually ambiguous Shepard tones. The authors make a fairly convincing case that the adaptation of pitch direction sensitive cells in auditory cortex is likely responsible for this phenomenon.

Strengths:

The manuscript is overall well written. My only slight criticism is that, in places, particularly for non-expert readers, it might be helpful to work a little bit more methods detail into the results section, so readers don't have to work quite so hard jumping from results to methods and back.

Following this excellent suggestion, we have added more brief method sketches to the Results section, hopefully addressing this concern.

The methods seem sound and the conclusions warranted and carefully stated. Overall I would rate the quality of this study as very high, and I do not have any major issues to raise.

Thanks for your encouraging evaluation of the work.

Weaknesses:

I think this study is about as good as it can be with the current state of the art. Generally speaking, one has to bear in mind that this is an observational, rather than an interventional study, and therefore only able to identify plausible candidate mechanisms rather than making definitive identifications. However, the study nevertheless represents a significant advance over the current state of knowledge, and about as good as it can be with the techniques that are currently widely available.

Thanks for your encouraging evaluation of our work. The suggestion of an interventional study has also been on our minds, however, this appears rather difficult, as it would require a specific subset of cells to be inhibited. The most suitable approach would likely be 2p imaging with holographic inhibition of a subset of cells (using ArchT for example), that has a preference for one direction of pitch change, which should then bias the percept/behavior in the opposite direction.

Reviewer #1 (Recommendations For The Authors):

MAJOR CONCERNS

(1) What is the timescale used to compute direction selectivity in neural tuning? How does it compare to the timing of the Shepard tones? The basic idea of up versus down pitch is clear, the intuition for the role of direction tuning and its relation to stimulus dynamics could be laid out more clearly. Are the authors proposing that there are two "special" populations of A1 neurons that are treated differently to produce the biased percept? Or is there something specific about the dynamics of the Shepard stimuli and how direction selective neurons respond to them specifically? It would help if the authors could clarify if this result links to broader concepts of dynamic pitch coding in general or if the example reported here is specific (or idiosyncratic) to Shepard tones.

We propose that the findings here are not specific to Shepard tones. To the contrary, only basic properties of auditory cortex neurons, i.e. frequency preference, frequency-direction (i.e. ascending or descending) preference, and local adaptation in the tuning curve, suffice. Each of these properties have been demonstrated many times before and we only verified this in the lead-up to the results in Fig. 6. While the same effects should be observable with pure tones, the lack of ambiguity in the perception of direction of a frequency step for pure tone pairs, would make them less noticeable here. Regarding the time-scale of the directional selectivity, we relied on the sequencing of tones in our paradigm, i.e. 150 ms spacing. The SSTRFs were discretized at 50 ms, and include only the bins during the stimulus, not during the pause. The directional tuning, i.e. differences in the SSTRF above and below the preferred pitchclass for stimuli before the last stimulus, typically extended only one stimulus back in time. We have clarified this in more detail now, in particular in the added Methods section on the directional decoder.

(2) (p. 9) "weighted by each cell's directionality index ... (see Methods for details)" The direction-selective decoder is interesting and appears critical to the study. However, the details of its implementation are difficult to locate. Maybe Fig. 6A contains the key concepts? It would help greatly if the authors could describe it in parallel with the other decoders in the Methods.

We have expanded the description of the decoder in the Methods as the reviewer suggests.

LESSER CONCERNS

p. 1. (L 24) "distances between the pitch representations...." It's not obvious what "distances" means without reading the main paper. Can some other term or extra context be provided?

We have added a brief description here.

p. 2. (L 26) "Shepard tones" Can the authors provide a citation when they first introduce this class of stimuli?

Citation has been added.

p. 3 (L 4) "direction selective cells" Please define or provide context for what has a direction. Selective to pitch changes in time?

Yes, selective to pitch changes in time is what is meant. We have further clarified this in the text.

p. 4 (L 9-19). This paragraph seems like it belongs in the Introduction?

Given the concerns raised by R2 about the organization of the manuscript we prefer to keep this 'road-map' in the manuscript, as a guidance for the reader.

p. 4 (L 32) "majority of cells" One might imagine that the overlap of the bias band and the frequency tuning curve of individual neurons might vary substantially. Was there some criterion about the degree of overlap for including single units in the analysis? Does overlap matter?

We are not certain which analysis the reviewer is referring to. Generally, cells were not excluded based on their overlap between a particular Bias band and their (Shepard) tuning curve. There are several reasons for this: The bias was located in 4 different, overlapping Shepard tone regions, and all sounds were Shepard tones. Therefore, all cells overlapped with their (Shepard) tuning curve with one or multiple of the Biases. For decoding analysis, all cells were included as both a response and lack of a response is contributing to the decoding. If the reviewer is referring only to the analysis of whether a cell adapts, then the same argument applies as above, i.e. this was an average over all Bias sequences, and therefore every responding cell was driven to respond by the Bias, and therefore it was possible to also assess whether it adapted its response for different positions inside the Bias. We acknowledge that the limited randomness of the Bias sequences in combination with the specific tuning of the cells could in a few cases create response patterns over time that are not indicative of the actual behavior for repeated stimulation, however, since the results are rather clear with 91% of cells adapting, we do not think this would significantly change the conclusions.

p. 5 (L 17) "desynchronization ... behaving conditions" The logic here is not clear. Is less desynchronization expected during behavior? Typically, increased attention is associated with greater desynchronization.

Yes, we reformulated the sentence to: While this difference could be partly explained by desynchronization which is typically associated with active behavior or attention [30], general response adaptation to repeated stimuli is also typical in behaving humans [31].

p. 7 (L 5) "separation" is this a separation in time?

Yes, added.

p. 7 (L 33) "local adaptation" The idea of feedforward adaptation biasing encoding has been proposed before, and it might be worth citing previous work. This includes work from Nelken specifically related to SSA. Also, this model seems similar to the one described in Lopez Espejo et al (PLoS CB 2019).

Thanks for pointing this out. We think, however, that neither of these publications suggested this very narrow way of biasing, which we consider biologically implausible. We have therefore not added either of these citations.

p. 11 (L. 17) The cartoon in Fig. 6G may provide some intuition, but it is quite difficult to interpret. Is there a way to indicate which neuron "votes" for which percept?

This is an excellent idea, and we have added now the purported perceptual relation of each cell in the diagram.

p. 12 (L. 8). "classically assumed" This statement could benefit from a citation. Or maybe "classically" is not the right word?

We have changed 'classically' to 'typically', and now cite classical works from Deutsch and Repp. We think this description makes sense, as the whole concept of bistable percepts has been interpreted as being equidistant (in added or subtracted semitone steps) from the first tone, see e.g. Repp 1997, Fig.2.

p. 12 (L. 12) "...previous studies" of Shepard tone percepts? Of physiology?

We have modified it to 'Relation to previous studies of Shepard tone percepts and their underlying physiology", since this section deals with both.

p. 12 (L. 25) "compatible with cellular mechanisms..." This paragraph seems key to the study and to Major Concern 1, above. What are the dynamics of the task stimuli? How do they compare with the dynamics of neural FM tuning and previously reported studies of bias? And can the authors be more explicit in their interpretation - should direction selective neurons respond preferentially to the Shepard tone stimuli themselves? And/or is there a conceptual framework where the same neurons inform downstream percepts of both FM sweeps and both normal (unbiased) and biased Shepard tones?

The reviewer raises a number of different questions, which we address below:

- Dynamics of the task stimuli in relation to previously reported cellular biasing: The timescales tested in the studies mentioned are similar to what we used in our bias, e.g. Ye et al 2010 used FM sweeps that lasted for up to 200ms, which is quite comparable to our SOA of 150ms.

- Preferred responses to Shepard tones: no, we do not think that there should be preferred responses to Shepard tones, but rather that responses to Shepard tones can be thought of as the combined responses to the constituent tones.

- Conceptual framework where the same neurons inform about FM sweeps and both normal (unbiased) and biased Shepard tones: Our perspective on this question is as follows: To our knowledge, the classical approach to population decoding in the auditory system, i.e. weighted based on preferred frequency, has not been directly demonstrated to be read out inside the brain, and certainly not demonstrated to be read out in only this way in all areas of the brain that receive input from the auditory cortex. Rather it has achieved its credibility by being linked directly with animal performance or match with the presented stimuli. However, these approaches were usually geared towards a representation that can be estimated based on constituent frequencies. Additional response properties of neurons, such as directional selectivity have been documented and analyzed before, however, not been used for explaining the percept. We agree that our use of this cellular response preference in the decoding implicitly assumes that the brain could utilize this as well, however, this seems just as likely or unlikely as the use of the preferred frequency of a neuron. Therefore we do not think that this decoding is any more speculative than the classical decoding. In both cases, subsequent neurons would have to implicitly 'know' the preference of the input neuron, and weigh its input correspondingly.

We have added all the above considerations to the discussion in an abbreviated form.

p. 15 (L. 15). Is there a citation for the drive system?

There is no publication, but an old repository, where the files are available, which we cite now: https://code.google.com/archive/p/edds-array-drive/

p. 16 (L. 24) "position in an octave" It is implied but not explicitly stated that the Shepard tones don't contain the fundamental frequency. Can the authors clarify the relationship between the neural tuning band and the bands of the stimulus. Did a single stimulus band typically fall in a neuron's frequency tuning curve? If not 1, how many?

Yes, it is correct that the concept of fundamental frequency does not cleanly apply to Shepard tones, because it is composed of octave spaced pure tones, but the lowest tone is placed outside the hearing range of the animal and amplitude envelope (across frequencies). Therefore one or more constituent tones of the Shepard tone can fall into the tuning curve of a neuron and contribute to driving the neuron (or inhibiting it, if they fall within an inhibitory region of the tuning curve). The number of constituent tones that fall within the tuning curve depends on the tuning width of the neurons. The distribution of tuning widths to Shepard tones is shown in Fig. S1E, which indicated that a lot of neurons had rather narrow tuning (close to the center), but many were also tuned widely, indicated that they would be stimulated by multiple constituent tones of the Shepard tone. As the tuning bandwidth (Q30: 30dB above threshold) of most cortical neurons in the ferret auditory cortex (see e.g. Bizley et al. Cerebral Cortex, 2005, Fig.12) is below 1, this means that typically not more than 1 tone fell into the tuning curve of a neuron. However, we also observed multimodal tuning-curves w.r.t. to Shepard tones, which suggests that some neurons were stimulated by more than 2 or more constituent tones (again consistent with the existence of more broadly tuned neurons (see same citation). We have added this information partly to the manuscript in the caption of Fig. S1E.

p. 17 (L. 32). "Fig 4" Correct figure ref? This figure appears to be a schematic rather than one displaying data.

Thanks for pointing this out, changed to Fig. 5.

p. 18 (L. 25). "assign a pitchclass" Can the authors refer to a figure illustrating this process?

Added.

p. 19 (L. 17). Is mu the correct symbol?

Thanks. We changed it to phi_i, as in the formula above.

p. 19 (L 19). "convolution" in time? Frequency?

Thanks for pointing this out, the term convolution was incorrect in this context. We have replaced it by "weighted average" and also adapted and simplified the formula.

p. 19 (L 25) "SSTRF" this term is introduced before it is defined. Also it appears that "SSTRF" and "STRF" are sometimes interchanged.

Apologies, we have added the definition, and also checked its usage in each location.

p. 23 (Fig 2) There is a mismatch between panel labels in the figure and in the legend. Bottom right panel (B3), what does time refer to here?

Thanks for pointing these out, both fixed.

p. 24 (L 23) "shifts them away" away from what?

We have expanded the sentence to: "After the bias, the decoded pitchclass is shifted from their actual pitchclass away from the biased pitchclass range ... "

p. 25 (L 7) "individual properties" properties of individual subjects?

Thanks for pointing this out, the corresponding sentence has been clarified and citations added.

p. 26 (L 20) What is plotted in panel D? The average for all cells? What is n?

Yes, this is an average over cells, the number of cells has now been added to each panel.

p. 28 (L 3) How to apply the terms "right" "right" "middle" to the panel is not clear. Generally, this figure is quite dense and difficult to interpret.

We have changed the caption of Panel A and replaced the location terms with the symbols, which helps to directly relate them to the figure. We have considered different approaches of adding or removing content from the figure to help make it less dense, but that all did not seem to help. For lack of better options we have left it in its current form.

MINOR/TYPOS

p. 3 (L 1) "Stimulus Specific Adaptation" Capitalization seems unnecessary

Changed.

p. 4 (L 14) "Siple"

Corrected.

p. 9 (L 10) "an quantitatively"

Corrected

p. 9 (L 20) "directional ... direction ... directly ... directional" This is a bit confusing as directseems to mean several different things in its different usages.

We have gone through these sentences, and we think the terms are now more clearly used, especially since the term 'direction' occurs in several different forms, as it relates to different aspects (cells/percept/hypothesis). Unfortunately, some repetition is necessary to maintain clarity.

Reviewer #2 (Recommendations For The Authors):

Detailed critique

Stimuli

It would be very useful if the authors could provide demos of their stimuli on a website. Many readers will not be familiar with Shepard tones and the perceptual result of the acoustical descriptions are not intuitive. I ended up coding the stimuli myself to get some intuition for them.

We have created some sample tones and sequences and uploaded them with the revision as supplementary documents.

Abstract

P1 L27 'pitch and...selective cells' - The authors haven't provided sufficient controls to demonstrate that these are "pitch cells" or "selective" to pitch direction. They have only shown that they are sensitive to these properties in their stimuli. Controls would need to be included to ensure that the cells aren't simply responding to one frequency component in the complex sound, for example. This is not really critical to the overall findings, but the claim about pitch "selectivity" is not accurate.

Fair point. We have removed the word 'selective' in both occurrences.

Introduction

P2 L14-17: I do not follow the phonetic example provided. The authors state that the second syllable of /alga/ and /arda/ are physically identical, but how is this possible that ga = da? The acoustics are clearly different. More explanation is needed, or a correction.

Apologies for the slightly misleading description, it has now been corrected to be in line with the original reference.

P2,L26-27: Should the two uses of "frequency" be "F0" and "pitch" here? The tones are not separated in frequency by half and octave, but "separated in [F0]" by half an octave, correct? Their frequency ranges are largely overlapping. And the second 'frequency', which refers to the percept, should presumably be "pitch".

Indeed. This is now corrected.

P3 L2-6: Unclear at this point in the manuscript what is the difference between the 3 percepts mentioned: perceived pitch-change direction, Shepard tone pitches, and "their respective differences". (It becomes clear later, but clarification is needed here).

We have tried a few reformulations, however, it tends to overload the introduction with details. We believe it is preferable to present the gist of the results here, and present the complete details later in the MS.

P3 L6-7 What does it mean that the MEG and single unit results "align in direction and dynamics"? These are very different signals, so clarification is needed.

We have phrased the corresponding sentence more clearly.

Results

Throughout: Choose one of 'pitch class', 'pitchclass', or 'pitch-class' and use it consistently.

Done.

P4L12 - would be helpful at this point to define 'repulsive effect'

We have added another sentence to clarify this term.

P4, L14 "simple"

Done

P4, L12 - not clear here what "repulsive influence" means

See above.

P4, L17 - alternative to which explanation? Please clarify. In general, this paragraph is difficult to interpret because we do not yet have the details needed to understand the terms used and the results described. In my opinion, it would be better to omit this summary of the results at the very beginning, and instead reveal the findings as they come, when they can be fully explained to the Reader.

We agree, but we also believe that a rather general description here is useful for providing a roadmap to the results. However, we have added a half-sentence to clarify what is meant by alternative.

P4 L30 - text says that cells adapt in their onset, sustained and offset responses, but only data for onset responses are shown (I think - clarification needed for fig 2A2). Supp figure shows only 1 example cell of sustained and offset, and in fact there is no effect of adaptation in the sustained response shown there.

Regarding the effect of adaptation and whether it can be discerned from the supplementary figure: the shown responses are for 10 repetitions of one particular Bias sequence. Since the response of the cell will depend on its tuning and the specific sequence of the Shepard tones in this Bias, it is not possible to assess adaptation for a given cell. We assess the level of adaptation, by averaging all biases (similar to what is shown in Fig. 2A2) per cell, and then fit an exponential to it, separately by response type. The step direction of the exponential, relative to the spontaneous rate is then used to assess the kind of adaptation. The vast majority of cells show adaptation. We have added this information to the Methods of the manuscript.

P4, L32 - please state the statistical test and criterion (alpha) used to determine that 91% of cells decreased their responses throughout the Bias sequence. Was this specifically for onset responses?

Thanks for pointing this out, test and p-value added. Adaptation was observed for onset, sustained and offset responses, in all cases with the vast majority showing an adapting behavior, although the onset responses were adapting the most.

P4 L36 - "response strength is reduced locally". What does "locally" mean here? Nearby frequencies?

We have added a sentence here to clarify this question.

Figure 1 - this appears to be the wrong version of the figure, as it doesn't match the caption or results text. It's not possible to assess this figure until these things are fixed. Figure 1A schematic of definition of f(diff) does not correspond to legend definition.

As far as we can tell, it is all correct, only the resolution of the figure appears to be rather low. This has been improved now.

Fig 2 A2 - is this also onset responses only?

Yes, added to the caption.

Fig 2 A3 - add y-axis label. The authors are comparing a very wide octave band (5.5 octaves) to a much narrower band (0.5 octaves). Could this matter? Is there something special about the cut-off of 2.5 octaves in the 2 bands, or was this an arbitrary choice?

Interesting question.... essentially our stimulus design left us only with this choice, i.e. comparing the internal region of the bias with the boundary region of the bias, i.e. the test tones. The internal region just corresponds to the bias, which is 5 st wide, and therefore the range is here given as 2.5 st relative to its center, while the test tones are at the boundary, as they are 3 st from the center. The axis for the bias was mislabelled, and has now been corrected. The y-axis label is matched with the panel to the left, but has now been added to avoid any confusion.

Fig 2A4 - does not refer to ferret single unit data, as stated in the text (p5L8). Nor does supp Fig2, as stated. Also, the figure caption does not match the figure.

Apologies, this was an error in the code that led to this mislabelling. We have corrected the labels, which also added back the recovery from the Bias sequence in the new Panel A4.

P5 l9 - Figure 3 is not understandable at this point in the text, and should not be referred to here. There is a lot going on in Fig 3, and it isn't clear what you are referring to.

Removed.

P5 L12 - by Fig 2 B1, I assume you mean A4? Also, F2B1 shows only 1 subject, not 2.

Yes, mislabeled by mistake, and corrected now.

Fig2B2 -What is the y-axis?

Same as in the panel to its left, added for clarity.

Stimuli: why are tones presented at a faster rate to ferrets than to humans?

The main reason is that the response analysis in MEG requires more spacing in time than the neuronal analysis in the ferret brain.

P5 L6 - there is no Fig 5 D2? I don't think it is a good idea to get the reader to skip so far ahead in the figures at this stage anyway, even if such a figure existed. It is confusing to jump around the manuscript

Changed to 'see below'

P5 L8 - There is no Figure 2A4, so I don't know whether this time constant is accurate.

This was in reference to a panel that had been removed before, but we have added it back now.

P5 L16: "in humans appears to be more substantial (40%) than for the average single units under awake conditions". One cannot directly compare magnitude of effects in MEG and single unit signals in this way and assume it is due to behavioural state. You are comparing different measures of neural activity, averaged over vastly different numbers of numbers, and recorded from different species listening to different stimuli (presentation rates).

Yes, that's why the next sentence is: "However, comparisons between the level of adaptation in MEG and single neuron firing rates may be misleading, due to the differences in the signal measured and subsequent processing.", and all statements in the preceding sentences are phrased as 'appears' and 'may'. We think we have formulated this comparison with an appropriate level of uncertainty. Further, the main message here is that adaptation is taking place in both active and passive conditions.

P5 L25 -I do not see any evidence regarding tuning widths in Fig s2, as stated in the text.

Corrected to Fig. S1.

P5 l26 - Do not skip ahead to Fig 5 here. We aren't ready to process that yet.

OK, reference removed.

P5 l27 - Do you mean because it could be tuning to pitch chroma, not height?

Yes, that is a possible interpretation, although it could also arise from a combination of excitatory and inhibitory contributions across multiple octaves.

P5 l33 - remove speculation about active vs passive for reasons given above.

Removed.

P6L2-6 'In the present...5 semitone step' - This is an incorrect interpretation of the minimal distance hypothesis in the context of the Shepard tone ambiguity. The percept is ambiguous because the 'true' F0 of the Shepard tones are imperceptibly low. Each constituent frequency of a single tone can therefore be perceived either as a harmonic of some lower fundamental frequency or as an independent tone. The dominant pitch of the second tone in the tritone pair may therefore be biased to be perceived at a lower constituent frequency (when the bias sequence is low) or at a higher constituent frequency (when the bias sequence is high). The text states that the minimal distance hypothesis would predict that an up-bias would make a tritone into a perfect fourth (5 semitones). This is incorrect. The MDH would predict that an up-bias would reduce the distance between the 1st tone in the ambiguous pair and the upper constituent frequency of the 2nd tone in the pair, hence making the upper constituent frequency the dominant pitch percept of the 2nd tone, causing an ascending percept.

The reviewer here refers to a “minimal distance hypothesis”, which without a literature reference,is hard for us to fully interpret. However, some responses are given below:

- "The percept is ambiguous because the 'true' F0 of the Shepard tones are imperceptibly low." This statement appears to be based on some misconception: due to the octave spacing (rather than multiple/harmonics of a lowest frequency), the Shepard tones cannot be interpreted as usual harmonic tones would be. It is correct that the lowest tone in a Shepard tone is not audible, due to the envelope and the fact that it could in principle be arbitrarily small... hence, speaking about an F0 is really not well-defined in the case of a Shepard tone. The closest one could get to it would be to refer to the Shepard tone that is both in the audible range and in the non-zero amplitude envelope. But again, since the envelope is fading out the highest and lowest constituent tones, it is not as easy to refer to the lowest one as F0 (as it might be much quieter than the next higher constituent.

- "The dominant pitch of the second tone in the tritone pair may therefore be biased to be perceived at a lower constituent frequency (when the bias sequence is low) or at a higher constituent frequency (when the bias sequence is high)." This may relate to some known psychophysics, but we are unable to interpret it with certainty.

- "The text states that the minimal distance hypothesis would predict that an up-bias would make a tritone into a perfect fourth (5 semitones). This is incorrect." We are unsure how the reviewer reaches this conclusion.

- "The MDH would predict that an up-bias would reduce the distance between the 1st tone in the ambiguous pair and the upper constituent frequency of the 2nd tone in the pair, hence making the upper constituent frequency the dominant pitch percept of the 2nd tone, causing an ascending percept." Again, in the absence of a reference to the MDH, we are unsure of the implied rationale. We agree that this is a possible interpretation of distance, however, we believe that our interpretation of distance (i.e. distances between constituent tones) is also a possible interpretation.

Fig 4: Given that it comes before Figure 3 in the results text, these should be switched in order in the paper.

Switched.

PCA decoder: The methods (p18) state that the PCA uses the first 3 dimensions, and that pitch classes are calculated from the closest 4 stimuli. The results (P6), however, state that the first 2 principal components are used, and classes are computed from the average of 10 adjacent points. Which is correct, or am I missing something?

Thanks for pointing this out, we have made this more concrete in the Methods to: "The data were projected to the first three dimensions, which represented the pitch class as well as the position in the sequence of stimuli (see Fig. 43A for a schematic). As the position in the Bias sequence was not relevant for the subsequent pitch class decoding, we only focussed on the two dimensions that spanned the pitch circle." Regarding the number of stimuli that were averaged: this might be a slight misunderstanding: Each Shepard tone was decoded/projected without averaging. However, to then assign an estimated pitch class, we first had to establish an axis (here going around the circle), where each position along the axis was associated with a pitch class. This was done by stepping in 0.5 semitone steps, and finding the location in decoded space that corresponded to the median of the Shepard tones within +/- 0.25st. To increase the resolution, this circular 'axis' of 24 points was then linearly interpolated to a resolution of 0.05st. We have updated the text in the Methods accordingly. The mentioning of 10 points for averaging in the Results was correct, as there were 240 tones in all bias stimuli, and 24 bins in the pitch circle. The mentioning of an average over 4 tones in the Methods was a typo.

Fig 3A: axes of pink plane should be PC not PCA

Done.

Fig 3B: the circularity in the distribution of these points is indeed interesting! But what do the authors make of the gap in the circle between semitones 6-7? Is this showing an inherent bias in the way the ambiguous tone is represented?

While we cannot be certain, we think that this represents an inhomogeneous sampling from the overall set of neural tuning preferences, and that if we had recorded more/all neurons, the circle would be complete and uniformly sampled (which it already nearly is, see Fig.4C, which used to be Fig. 3C).

Fig 3B (lesser note): It'd be preferable to replace the tint (bright vs. dark) differentiation of the triangles to be filled vs. unfilled because such a subtle change in tint is not easily differentiable from a change in hue (indicating a different variable in this plot) with this particular colour palette

We have experimented with this suggestion, and it didn't seem to improve the clarity. However, we have changed the outline of the test-pair triangles to white, which now visually separates them better.

P6 l32 - Please indicate if cross-validation was used in this decoder, and if so, what sort. Ideally, the authors would test on a held-out data set, or at least take a leave-one-out approach. Otherwise, the classifier may be overfit to the data, and overfitting would explain the exceptional performance (r=.995) of the classifier.

Cross-validation was not used, as the purpose of the decoder is here to create a standard against which to compare the biased responses in the ambiguous pair, which were not used for training of the decoder. We agree that if we instead used a cross-validated decoder (which would only apply to the local average to establish the pitch class circle) the correlation would be somewhat lower, however, this is less relevant for the main question, i.e. the influence of the Bias sequence on the neural representation of the ambiguous pair. We have added this information to the corresponding section.

Fig 3D: I understood that these pitch classifications shown by the triangles were carried out on the final ambiguous pair of stimuli. I thought these were always presented at the edges of the range of other stimuli, so I do not follow how they have so many different pitchclass values on the x-axis here.

There were 4 Biases, centered at 0,3,6 or 9 semitones, and covering [-2.5,2.5]st relative to this center. Therefore the edges of the bias ranges (3st away from their centers) happen to be the same as the centers, e.g. for the Bias centered at 3, the ambiguous pair would be a 0-6 or 6-0 step. Therefore there are 4 locations for the ambiguous tones on the x-axis of Fig. 4D (previously 3D).

Figure 4: This demonstration of the ambiguity of Shepard pairs may be misleading. The actual musical interval is never ambiguous, as this figure suggests. Only the ascending vs descending percept is ambiguous. Therefore the predictions of the ferret A1 decoding (Fig 3D) and the model in Fig 5 are inconsistent with perception in two ways. One (which the authors mention) is the direction of the bias shift (up vs down). Another (not mentioned here) is that one never experiences a shift in the shepard tone at a fraction of a semitone - the musical note stays the same, and changes only in pitch height, not pitch chroma.

We are unsure of the reviewer’s direction with this question. In particular the second point is not clear to us: "...one (who?) never (in this experiment? in real life?) experiences a bias shift in the Shepard tone at a fraction of a semitone" (why is this relevant in the current experiment?). Pitch chrome would actually be a possible replacement for pitch class, but somehow, the previous Shepard tone literature has referred to it as pitch class.

P7 l12 - omit one 'consequently'

Changed to 'Therefore'.

P7 l24 - I encourage the authors to not use "local" and "global" without making it clear what space they refer to. One tends to automatically think of frequency space in the auditory system, but I think here they mean f0 space? What is a "cell close to the location of the bias"? Cells reside in the brain. The bias is in f0 space. The use of "local" and "global" throughout the manuscript is too vague.

Agreed, the reference here was actually to the cell's preferred pitch class, not its physical location (which one might arguably be able to disambiguate, given the context). We have changed the wording, and also checked the use of global/local throughout the manuscript. The main use of 'global/local' is now in reference to the range of adaptation, and is properly introduced on first mention.

P7 L26 -there is no Fig 5D1. Do you mean the left panel of 5D?

Thanks. Changed.

FigS3 is referred to a lot on p7-8. Should this be moved to the main text?

The main reason why we kept it in the supplement is that it is based on a more static model, which is intended to illustrate the consequences of different encoding schemes. In order to not confuse the reader about these two models, we prefer to keep it in the supplement, which - for an online journal - makes little difference since the reader can just jump ahead to this figure in the same way as any other figure.

Fig 5C, D - label x-axis.

Added.

Fig 5E - axis labels needed. I don't know what is plotted on x and y, and cannot see red and green lines in left plot

Thanks for noticing this, colors corrected, axes labeled.

Page 8 L3-15 - If I follow this correctly, I think the authors are confusing pitch and frequency here in a way that is fundamental to their model. They seem to equate tonotopic frequency tuning to pitch tuning, leading to confused implications of frequency adaptation on the F0 representation of complex sounds like Shepard tones. To my knowledge, the authors do not examine pure tone frequency tuning in their neurons in this study. Please clarify how you propose that frequency tuning like that shown in Fig 5A relates to representation of the F0 of Shepard tones. Or...are the authors suggesting these neural effects have little to do with pitch processing and instead are just the result of frequency tuning for a single harmonic of the Shepard tones?

We agree that it is not trivial to describe this well, while keeping the text uncluttered, in particular, because often tuning properties to stimulus frequency contribute to tuning properties of the same neuron for pitch class, although this can be more or less straightforward: specifically, for some narrowly tuned cells, the Shepard tuning is simply a reflection of their tuning to a single octave range of the constituent tones (see Fig. S1). For more broadly tuned cells, multiple constituent tones will contribute to the overall Shepard tuning, which can be additive, subtractive, or more complex. The assumption in our approach is that we can directly estimate the Shepard tuning to evaluate the consequence for the percept. While this may seem artificial, as Shepard tones do not typically occur in nature, the same argument could be made against pure tones, on which classical tuning curves and associated decodings are often based. Relating the Shepard tuning to the classical tuning would be an interesting study in itself, although arguably relating the tuning of one artificial stimulus to another. Regarding the terminology of pitch, pitch class and frequency: The term pitch class is commonly used in the field of Shepard tones, and - as we indicated in the beginning of the results: "the term pitch is used interchangeably with pitch class as only Shepard tones are considered in this study". We agree that the term pitch, which describes the perceptual convergence/construction of a tone-height from a range of possible physical stimuli, needs to be separated from frequency as one contributor/basis for the perception of a pitch. However, we think that the term pitch can - despite its perceptual origin - also be associated with neuron/neural responses, in order to investigate the neural origin of the pitch percept. At the same time, the present study is not targeted to study pitch encoding per se, as this would require the use of a variety of stimuli leading to consistent pitch percepts. Therefore, pitch (class) is here mainly used as a term to describe the neural responses to Shepard tones, based on the previous literature, and the fact that Shepard tones are composite stimuli that lead to a pitch percept. The last sentence has been added to the manuscript for clarity.

P7-9: I wasn't left with a clear idea of how the model works from this text. I assume you have layers of neurons tuned to frequency or f0 (based on the real data?), which are connected in some way to produce some sort of output when you input a sound? More detail is needed here. How is the dynamic adaptation implemented?

The detailed description of the model can be found in the Methods section. We have gone through the corresponding paragraph and have tried to clarify the description of the model by introducing a high-level description and the reference to the corresponding Figure (Fig. 5A) in the Results.

Fig6A: Figure caption can't be correct. In any case, these equations cannot be understood unless you define the terms in them.

We have clarified the description in the caption.

Fig 6/directionality analysis: Assuming that the "F" in the STRFs here is Shepard tone f0, and not simple frequency?

We have changed the formula in the caption and the axis labels now.

Fig 6C - y-axis values

In the submission, these values were left out on purpose, as the result has an arbitrary scale, but only whether it is larger or smaller than 0 counts for the evaluation of the decoded directionality (at the current level of granularity). An interesting refinement would be to relate the decoded values to animal performance. We have now scaled the values arbitrarily to fit within [-1,1], but we would like to emphasize that only their relative scale matters here, not their absolute scale.

Fig 6E - can't both be abscissa (caption). I might be missing something here, but I don't see the "two stripes" in the data that are described in the caption.

Thank you. The typo is fixed. The stripes are most clearly visible in the right panel of Fig. 6E, red and blue, diagonally from top left to bottom right.

Fig 6G -I have no idea what this figure is illustrating.

This panel is described in the text as follows: "The resulting distribution of activities in their relation to the Bias is, hence, symmetric around the Bias (Fig. 6G). Without prior stimulation, the population of cells is unadapted and thus exhibits balanced activity in response to a stimulus. After a sequence of stimuli, the population is partially adapted (Fig. 6G right), such that a subsequent stimulus now elicits an imbalanced activity. Translated concretely to the present paradigm, the Bias will locally adapt cells. The degree of adaptation will be stronger, if their tuning curve overlaps more with the biased region. Adaptation in this region should therefore most strongly influence a cell’s response. For example, if one considers two directional cells, an up- and a down-selective cell, cocentered in the same frequency location below the Bias, then the Bias will more strongly adapt the up-cell, which has its dominant, recent part of the SSTRF more inside the region of the Bias (Fig. 6G right). Consistent with the percept, this imbalance predicts the tone to be perceived as a descending step relative to the Bias. Conversely, for the second stimulus in the pair, located above the Bias, the down-selective cells will be more adapted, thus predicting an ascending step relative to the previous tone."

I might be just confused or losing steam at this point, but I do not follow what has been done or the results in Fig 6 and the accompanying text very well at all. Can this be explained more clearly? Perhaps the authors could show spike rate responses of an example up-direction and down-direction neuron? Explain how the decoder works, not just the results of it.

We agree that we are presenting something new here. However, it is conceptually not very different from decoding based on preferred frequencies. We have attempted to provide two illustrations of how the decoder works (Fig. 6A) and how it then leads to the percept using prototypical examples of cellular SSTRFs (Fig. 6G). We have added a complete, but accessible description to the Methods section. Showing firing rates of neurons would unfortunately not be very telling, given the usual variability in neural response and the fact that our paradigm did not have a lot of repetitions (but instead a lot of conditions), which would be able to average out the variability on a single neuron level.

Discussion - I do not feel I can adequately critique the author's interpretation of the results until I understand their results and methods better. I will therefore save my critique of the discussion section for the next round of revisions after they have addressed the above issues of disorganization and clarity in the manuscript.

We hope that the updated version of the manuscript provides the reviewer now with this possibility.

Methods

P15L7 - gender of human subjects? Age distribution? Age of ferrets?

We have added this information.

P16L21 - What is the justification for randomizing the phase of the constituent frequencies?

The purpose of the randomization was to prevent idiosyncratic phase relationships for particular Shepard tones, which would depend in an orderly fashion on the included base-frequencies if non-randomized, and could have contributed to shaping the percept for each Shepard tone in a way that was only partly determined by the pitch class of the Shepard tone. Added to the section.

P17L6 - what are the 2 randomizations? What is being randomized?

Pitch classes and position in the Bias sequence. Added to the section.

P16 Shepard Tuning section - What were the durations of the tones and the time between tones within a trial?

Thanks, added!

Equations - several undefined terms in the equations throughout the manuscript.

Thanks. We have gone through the manuscript and all equations and have introduced additional definitions where they had been missing.

Reviewer #3 (Recommendations For The Authors):

P3L10: "passive" and "active" conditions come totally out of the blue. Need introducing first. (Or cut. If adaptation is always seen, why mention the two conditions if the difference is not relevant here?)

We have added an additional sentence in the preceding paragraph, that should clarify this. The reason for mentioning it is that otherwise a possible counter-argument could be made that adaptation does not occur in the active condition, which was not tested in ferrets (but presents an interesting avenue for future research).

P3L14 "siple" typo

Corrected.

P4L1 "behaving humans" you should elaborate just a little here on what sort of behavior the participants engaged in.

Thanks for pointing this out. We have clarified this by adding an additional sentence directly thereafter.

P4 adaptation: I wonder whether it would be useful to describe the Bias condition a bit more here before going into the observations. The reader cannot know what to expect unless they jump ahead to get a sense of what the Bias looks like in the sense of how many stimuli are in it, and how similar they are to each other. Observations such as "the average response strength decreases as a function of the position in the Bias sequence" are entirely expected if the Bias is made up of highly repetitive material, but less expected if it is not. I appreciate that it can be awkward to have Methods after Results, but with a format like that, the broad brushstroke Methods should really be incorporated into the Results and only the tedious details should be reserved for the Methods to avoid readers having to jump back and forth.

Agreed, we have inserted a corresponding description before going into the details of the results.

Related to this (perhaps): Bottom of P4, top of P5: "significantly less reduced (33%, p=0.0011, 2 group t-test) compared to within the bias (Fig. 2 A3, blue vs. red), relative to the first responses of the bias" ... I am at a loss as to what the red and blue symbols in Fig 2 A3 really show, and I wonder whether the "at the edges" to "within the Bias" comparison were to make sense if at this stage I had been told more about the composition of the Bias sequence. Do the ambiguous ('target') tones also occur within the Bias? As I am unclear about what is compared against what I am also not sure how sound that comparison is.

We have added an extended description of the Bias to the beginning of this section of the manuscript. For your reference: the Shepard tones that made up the ambiguous tones were not part of the Bias sequence, as they are located at 3st distance from the center of the Bias (above and below), while the Bias has a range of only +/- 2.5st.

Fig 2: A4 B1 B2 labels should be B1 B2 B3

Corrected.

Fig 2 A2, A3: consider adjusting y-axis range to have less empty space above the data. In A3 in particular, the "interesting bit" is quite compressed.

Done, however, while still matching the axes of A2 and A3 for better comparability.

I am under the strong impression that the human data only made it into Fig 2 and that the data from Fig 3 onwards are animal data only. That is of course fine (MEG may not give responses that are differentiated enough to perform the sort of analyses shown in the later figures. But I do think that somewhere this should be explicitly stated.

Yes, the reviewer's observation is correct. The decoding analyses could not be conducted on the human MEG data and was therefore not further pursued. Its inclusion in the paper has the purpose of demonstrating that even in humans and active conditions, the local adaptation is present, which is a key contributor to the two decoding models. We now state this explicitly when starting the decoding analysis.

P5L2 "bias" not capitalized. Be consistent.

All changed to capitalized.

P5L8 reference to Fig 2 A4: something is amiss here. From legend of Fig 2 it seems clear that panel A4 label is mislabeled B1. Maybe some panels are missing to show recovery rates?

Apologies for this residual text from a previous version of the manuscript. We have gone through all references and corrected them.

P6L7 comma after "decoding".

Changed.

Fig 3, I like this analysis. What would be useful / needed here though is a little bit more information about how the data were preprocessed and pooled over animals. Did you do the PCA separately for each animal, then combine, or pool all units into a big matrix that went into the PCA? What about repeat, presentations? Was every trial a row in the matrix, or was there some averaging over repeats? (In fact, were there repeats??)

Thanks for bringing up these relevant aspects, which were partly insufficiently detailed in the manuscript. Briefly, cells were pooled across animals and we only used cells that could meaningfully contribute to the decoding analysis, i.e. had auditory responses and different responses to different Shepard tones. Regarding the responses, as stated in the Methods, "Each stimulus was repeated 10 times", and we computed average responses across these repetitions. Single trials were not analyzed separately. We have added this information in the Methods, and refer to it in the Results.

Also, there doesn't appear to be a preselection of units. We would not necessarily expect all cortical neurons to have a meaningful "best pitch" as they may be coding for things other than pitch. Intuitively I suspect that, perhaps, the PCA may take care of that by simply not assigning much weight to units that don't contribute much to explained variance? In any event I think it should be possible, and would be of some interest, to pull out of this dataset some descriptive statistics on what proportion of units actually "care about pitch" in that they have a lot (or at least significantly more than zero) of response variance explained by pitch. Would it make sense to show a distribution of %VE by pitch? Would it make sense to only perform the analysis in Fig 3 on units that meet some criterion? Doing so is unlikely to change the conclusion, but I think it may be useful for other scientists who may want to build on this work to get a sense of how much VE_pitch to expect.

We fully agree with the reviewer, which is why this information is already presented in Supplementary Fig.1, which details the tuning properties of the recorded neurons. Overall, we recorded from 1467 neurons across all ferrets, out of which 662 were selected for the decoding analysis based on their driven firing rate (i.e. whether they responded significantly to auditory stimulation) and whether they showed a differential response to different Shepard tones The thresholds for auditory response and tuning to Shepard tones were not very critical: setting the threshold low, led to quantitatively the same result, however, with more noise. Setting the thresholds very high, reduced the set of cells included in the analysis, and eventually that made the results less stable, as the cells did not cover the entire range of preferences to Shepard tones. We agree that the PCA based preprocessing would also automatically exclude many of the cells that were already excluded with the more concrete criteria beforehand. We have added further information on this issue in the Methods section under the heading 'Unit selection'.

P9 "tones This" missing period.

Changed.

P10L17 comma after "analysis"

Changed.

Author response:

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

Reviewer #3 (Public Review):

Some critical comments are provided below：

(1) The data quality still needs to be improved. There are many outliers in the experimental data shown in some figures, e.g. Figure 2D-G. The presence of these outliers makes the results unreliable. The author should thoroughly review the data analysis in the manuscript. In addition, a couple of western blot bands, such as IL-1β in Figure 3C, are not clear enough, please provide clearer western blot results again to support the conclusion.

Following our comparative analysis, we have determined that these data do not affect our conclusions. Moreover, our experimental design included a total of six mice per group, with all mouse samples being subjected to testing.

(2) As shown in Figure 1G-I, foot thickness and IL-1β content in foot tissues of the Aged+Abx group were significantly reduced, but there was no difference in serum uric acid level. In addition, the Abx-untreated group should be included at all ages.

Thank you for your comment. We have included this data in Supplemental Material 4.

(3) Since FMT (Figure 4) and butyrate supplementation (Figure 8) have different effects on uric acid synthesis enzyme and excretion, different mechanisms may lie behind these two interventions. Transplantation with significantly enriched single strains from young mice, such as Bifidobacterium and Akkermansia, is the more reliable approach to reveal the underlying mechanism between gut microbiota and gout.

Thank you for your comment. Due to the involvement of multiple bacterial genera in gout and hyperuricemia, and the practical challenge of testing all strains, our focus shifted to the functional implications and metabolism of the microbiota. Experimental validation confirmed that butyrate exerts a dual-therapeutic effect in mitigating gout and hyperuricemia.

(4) In Figure 2F, the results showed the IL-1β, IL-6, and TNF-α content in serum, which was inconsistent with the authors' manuscript description (Line 171).

Thank you for your comment. The modifications to the results have been implemented.

(5) Figures 2F-H duplicate Supplementary Figures S1B-D. The authors should prepare the article more carefully to avoid such mistakes.

Thank you for your comment. We have corrected it in the manuscript.

(6) In lines 202-206, the authors stated that the elevated serum uric acid levels in the Young+Old or Young+Aged groups, but there is no difference in the results shown in Figure 4A.

Thank you for your comment. We have corrected it in the manuscript.

(7) Please visualize the results in Table 2 in a more intuitive manner.

The results have been presented in Table 2 with a more intuitive visual format. The detailed information is presented in Supplement 4.

(8) The heatmap in Figure 7A cannot strongly support the conclusion "the butyric acid content in the faeces of Young+PBS group was significantly higher than that in the Aged+PBS group". The author should re-represent the visual results and provide a reasonable explanation. In addition, please provide the ordinate unit of Supplementary Figure 7A-H.

Thank you for your comment. Figure 7A and Supplementary Figure 7A-H together illustrate "the butyric acid content in the faeces of Young+PBS group was significantly higher than that in the Aged+PBS group", and the specific units of short-chain fatty acids have been annotated in the manuscript.

(9) Uncropped original full-length western blot should be provided.

Thank you for your comment. We have made relevant notes in the paper.

Reviewer #1 (Recommendations For The Authors):

Gout, a prevalent form of arthritis among the elderly, exhibits an intricate relationship with age and gut microbiota. The authors found that gut microbiota plays a crucial role in determining susceptibility to age-related gout. They observed that age-related gut microbiota regulated the activation of the NLRP3 inflammasome pathway and modulated uric acid metabolism. "Younger" microbiota has a positive impact on the gut microbiota structure of old or aged mice, enhancing butanoate metabolism and butyric acid content. Finally, they found butyric acid exerts a dual effect, inhibiting inflammation in acute gout and reducing serum uric acid levels. This work's insights emphasize the potential of "young" gut microbiome in mitigating senile gout. The whole study was interesting, but there were some minor errors in the overall writing of the paper. The author should carefully check the spelling of the words in the text and the case consistency of the group names.

Questions:

(1) Line 118, line 142, and elsewhere 24 months in the same format as before.

Thank you for your comment. We have corrected it in the manuscript.

(2) Lines 123, Old and Aged group should be a complex number.

Thank you for your suggestion. We have corrected it in the manuscript.

(3) Why does line 133 mention the use of ABX? Please add a brief explanation.

Thank for your suggestion. The aim of utilizing ABX is to construct the linkage between gut microbiota, age, and gout.

(4) Lines 172-175, the description of TNF does not match the description of the result figure, may be the picture placement error, please correct this.

Thank you for your careful review. The error has been corrected and the accurate result has been inserted into the original manuscript.

(5) Lines183-185 and lines193-lines195, Pro-Caspase-1 and Pro-IL activate excess write.

Thank you for your careful review. We have corrected the error at the original location.

(6) Line 400, the text should not be written as increased.

Thank you for your careful review. We have corrected the error at the original location.

(7) "ns" needs to be added in the legend to indicate that there is no significant difference.

Thank you for your careful review. We have corrected the error at the original location.

(8) Lines 1080-1084 "Old or Aged control group and the old or aged group", group names should be case-sensitive.

Thank you for your suggestion. We have made the correct modification to the group names.

(9) Lines 1072-1073, "Representative western blot images of foot tissue NLRP3 pathways proteins" add band density.

Thank you for your suggestion. We have corrected the error on lines 1072-1073 of the article.

Reviewer #2 (Recommendations For The Authors):

Specific comments:

(1) In Figures 1G-H, the Aged+PBS group with antibiotic treatment shows a significant reduction in foot swelling and IL-1β compared to the Young+PBS and Old+PBS groups. The authors state that age-related changes in the gut microbiota exacerbate gout. However, why does only the Aged+PBS group improve with antibiotic treatment? It seems that butyrate alone cannot explain this phenomenon.

We utilize antibiotics for treatment in order to establish the relationship between gut microbiota, age, and gout. Different age groups are directly given antibiotics for treatment. We found that after clearing the gut microbiota and then stimulating with MSU, the trend of inflammation factors changing with age disappears.

(2) In Figure 2, the fecal transplantation from young mice improved the infiltration of inflammatory cells and inflammatory cytokines in the Old and Aged groups. However, in Supplementary Figure 1A, there is no improvement observed in the percentage of foot swelling. Is it appropriate to conclude that inflammation was improved even though foot swelling was not suppressed?

Although we did not observe changes in the swelling of the mice's feet, there were changes in the inflammatory cell infiltration and inflammation factors in the slices. We rely on a comprehensive assessment of various indicators to determine whether the inflammatory condition has improved or worsened.

(3) In line #249, the authors state that "the fecal microbiota from mice in the young group promotes uric acid elimination, inhibits reabsorption, and may contribute to the integrity of the intestinal barrier structure." However, Supplementary Figure 3F-H shows no significant alterations in Occludin and ZO-1 mRNA expression levels among all groups. Therefore, it is difficult to conclude that the fecal microbiota from the young group promotes the integrity of the intestinal barrier structure. A functional barrier assay, such as oral administration of FITC-dextran, would be necessary to verify the authors' conclusion.

In Supplementary Figure 3F-H, we observed that the mRNA expression of Occludin and ZO-1 increased but showed no significant difference. However, after the elderly mice were transplanted with the intestinal microbiota of young mice, the mRNA expression of JAMA showed a significant upward trend. Additionally, due to the scarcity of old mice, we were unable to perform the oral administration of FITC-dextran. However, we supplemented with immunohistochemical slices of Zo-1 and Occludin to support our viewpoint.

(4) In Figure 4, when comparing the young+PBS group with the old+PBS or aged+PBS groups, there are hardly any differences in the proteins involved in uric acid synthesis (ADA, GDA, XOD) or the genes involved in uric acid transport (URAT1, GLUT9, OAT1, OTA3, ABCG2). Since no changes in uric acid synthesis or transport pathways are observed with aging, it is questionable to conclude that fecal transplantation from young mice improves these pathways and lowers blood uric acid levels.

In the calculation process, we used different age groups of the control group as references, instead of directly using young mice. We then compared the data of mice of different ages, and the results are in Supplementary Material 4.

(5) In line 276, the authors describe "the Young +Old and Young+Aged groups tended to be closer to the Old+PBS and Aged+PBS groups, and the Old+Young and Aged+young groups tended to be closer to the Young+PBS group (Figure 5D)". Please conduct a statistical analysis.

(6) In line 298, the authors hypothesize that butyrate might be the key molecule responsible for controlling gout, as Bifidobacterium and Akkermansia were abundant in the Young group, and the butyrate pathway was prominent. However, neither Bifidobacterium nor Akkermansia are butyrate-producing bacteria. Thus, the conclusion appears to be biased toward butyrate, raising questions about this interpretation.

Upon comparison, we discovered other bacteria genera that produce butyrate, such as Lachnoclostridium. Additionally, literature (PMID:38126785, 26420851) reports have indicated that Bifidobacteria combined with other genera can enhance the production of butyrate. Meanwhile, Akkermansia, particularly the species Akkermansia muciniphila, has been found to confer several beneficial traits, as evidenced by preclinical studies. These traits include promoting the growth of butyrate-producing bacteria through the production of acetate, which leads to a decrease in the loss of the colonic bilayer and subsequent reduction in inflammation (PMID:35468952). Based on the predicted results of microbiome functions, we observed that the Butanoate_metabolism of the microbiota in young mice and the elderly mice recipients of young mouse microbiota was enhanced. Considering that Lachnoclostridium can produce butyrate, and that Bifidobacteria and Akkermansia can promote the production of butyrate by the intestinal microbiota, we speculated that butyrate might play a role in gout and hyperuricemia.

(7) In Supplementary Figure 7, acetic acid and propionic acid also show the same behavior as butyric acid. It is possible that these metabolites may also affect the development of gout.

Thank you for your suggestion. Indeed, Figure 7 does show a similar trend for acetic and propionic acids as for butyric acid. However, considering the predictive data of microbial function and the non-targeted metabolomic data, there is an enhancement of Butanoate_metabolism in both young mice and elderly mice receiving young mouse intestinal microbiota transplants. Therefore, we prioritized butyrate as the subject of our study. Due to the scarcity of elderly mice, we are unable to conduct subsequent experiments with acetic and propionic acids, which is one of the limitations of this study. This work will be addressed in our follow-up research.

(8) In Figure 6, the secondary bile acid biosynthesis pathway was also changed. However, there is little mention of secondary bile acid in the discussion section. Please carefully discuss other possibilities besides butyrate.

Thank you for your suggestion. We have incorporated a discussion about secondary bile acids into the relevant section of our manuscript.

(9) In line #330, the authors state, 'the metabolites identified as showing differential abundance between the groups were enriched in the butanoate metabolism pathway (Figure 6A-D).' However, there does not appear to be much difference in the butanoate metabolism pathway. Specifically, in Figure 6C, the butanoate metabolism pathway in the Old group does not differ from that in the Young group. Please explain in more detail whether the butanoate metabolism pathway is relevant in the Old group.

The metabolites identified as showing differential abundance between the groups were enriched in the butanoate metabolism pathway. The differential metabolites are enriched in the butyrate metabolism pathway; however, the non-targeted metabolomics did not reveal the extent of their enrichment.

(10) In Figure 7, the authors measured the levels of short-chain fatty acids in the Young and Aged groups. They found butyrate in the feces of mice in the Young group was higher than that in the Aged group. However, I wonder whether the Old group also had low levels of butyrate or not.

In the experiment, we selected three representative groups to verify the hypothesis that butyrate may play a significant role in gout and hyperuricemia. Subsequently, we found that supplementing 18-month-old and 24-month-old mice with butyrate indeed reduced blood uric acid levels and alleviated gout symptoms. Since 18-month-old mice are difficult to obtain, we only conducted microbiome sequencing and non-targeted metabolomic analysis.

Minor issues:

(11) In line 74, what does MSU stand for? Please describe the abbreviation.

In line 74, MSU refers to Monosodium urate crystals.

(12) In line 136, please insert a space between "IL-1β" and "and".

Thank you for your suggestion. We have corrected the error of the article.

(13) In line 570, please describe the method of butyrate administration and also correct the grammatical errors.

Thank you for your suggestion. We have corrected the error of the article.

(14) Change the title of x axis in Figure 2F-H, "Serum ~" to "Peritoneal fluid ~", according to the legend.

Thank you for your suggestion. We have corrected this error in the manuscript.

(15) In line 302, "succinates" should be "butyric acid or butyrate".

Thank you for your suggestion. We have corrected this error in the manuscript.

Reviewer #3 (Recommendations For The Authors):

(1) The authors showed the results of IL-1β levels in foot tissues in Figure 1C and Figure 1H, and serum IL-1β, IL-6, and TNF-α levels in Figure 2F-H. Could the authors also provide the results of IL-6 and TNF-α in foot tissue in Figure 1?

Thank you for your suggestion. We have added the results of of IL-6 and TNF-α in foot tissue in supplementary material 4.

(2) There are some errors in the reference citation format, such as missing page numbers.

Thank you for your careful review. We have revised the references in our manuscript.

(3) There are too many writing errors in the manuscript, which greatly affect the understanding of the text. The manuscript must be carefully revised to improve its readability. It's recommended that a professional English writer or native speaker proofread the paper before submission. Some errors, but not limited to these errors, are listed below.

a. Line 107: The abbreviation for "short-chain fatty acid" should be SCFA, not SFCA.

Thank you for your careful review. We have corrected this error in the manuscript.

b. Line 136: There is a missing space between IL-1β and and. B.

Thank you for your careful review. We have corrected this error in the manuscript.

c. Line 145, the phrase "on gout on gout", and line 471, "that transplantation" are repeated.

Thank you for your careful review. We have corrected this error in the manuscript.

d. Line 152: "Age+PBS" should be "Aged+PBS".

Thank you for your careful review. We have corrected this error in the manuscript.

e. In Figure 1e, "Aded+PBS" should be "Aged+PBS".

Thank you for your careful review.  We have corrected the error in Figure 1e.

f. Line 152: The phrase "by via" is repeated.

Thank you for your suggestion. We have deleted the phrase "by via" in line 152.

g. "16S rDNA" in line 92 is inconsistent with the "16S rRNA" in line 652.

Thank you for your suggestion. We have revised the error in the manuscript to maintain consistency in professional terminology.

Author response:

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

Reviewer #1 (Public Review):

Tleiss et al. demonstrate that while commensal Lactiplantibacillus plantarum freely circulate within the intestinal lumen, pathogenic strains such as Erwinia carotovora or Bacillus thuringiensis are blocked in the anterior midgut where they are rapidly eliminated by antimicrobial peptides. This sequestration of pathogenic bacteria in the anterior midgut requires the Duox enzyme in enterocytes, and both TrpA1 and Dh31 in enteroendocrine cells. This effect induces muscular muscle contraction, which is marked by the formation of TARM structures (thoracic ary-related muscles). This muscle contraction-related blocking happens early after infection (15mins). On the other side, the clearance of bacteria is done by the IMD pathway possibly through antimicrobial peptide production while it is dispensable for the blockage. Genetic manipulations impairing bacterial compartmentalization result in abnormal colonization of posterior midgut regions by pathogenic bacteria. Despite a functional IMD pathway, this ectopic colonization leads to bacterial proliferation and larval death, demonstrating the critical role of bacteria anterior sequestration in larval defense.

This important work substantially advances our understanding of the process of pathogen clearance by identifying a new mode of pathogen eradication from the insect gut. The evidence supporting the authors' claims is solid and would benefit from more rigorous experiments.

(1) The authors performed the experiments on Drosophila larvae. I wonder whether this model could extend to adult flies since they have shown that the ROS/TRPA1/Dh31 axis is important for gut muscle contraction in adult flies. If not, how would the authors explain the discrepancy between larvae and adults?

We have linked the adult phenotype to the larval model to explore the ROS/TrpA1/Dh31 axis in both contexts.  As highlighted in the discussion, however, there are key behavioral differences between larvae and adult flies. Unlike larvae, which remain in the food environment, adult flies have the ability to move away. This difference could impact the relevance of gut muscle contraction and bacterial clearance mechanisms between the two stages. Specifically, in larvae, the rapid ejection of gut contents due to muscle contraction poses a unique risk: larvae may inadvertently re-ingest the expelled material within minutes, which could influence their immune defenses. We have clarified this distinction and our hypothesis in the final section of the discussion, as it emphasizes the adaptive nature of this mechanism in larvae.

(2) The authors performed their experiments and proposed the models based on two pathogenic bacteria and one commensal bacterial at a relatively high bacterial dose. They showed that feeding Bt at 2X1010 or Ecc15 at 4X108 did not induce a blockage phenotype. 

I wonder whether larvae die under conditions of enteric infection with low concentrations of pathogenic bacteria. 

To address this, we have provided new data (Movie 5), in which larvae were fed a lower dose of Bt-GFP at 1.3 × 10^10 CFU/mL. In this video, we observe that when larvae ingest fewer bacteria, no blockage occurs, and the bacteria are able to reach the posterior midgut. As the bacterial load is lower, the fluorescence signal is weaker, but the movie clearly shows the excretion of bacteria. Importantly, under these conditions, no larval death was observed. These findings suggest that below a certain bacterial threshold, the pathogenicity is insufficient to: (1) trigger the blockage response, and (2) kill the larvae. In such cases, bacteria are likely eliminated through normal peristaltic movements rather than through the blockage mechanism described in our study.

If larvae do not show mortality, what is the mechanism for resisting low concentrations of pathogenic bacteria? 

As mentioned in our previous response, we hypothesize that the larvae’s ability to resist low concentrations of pathogenic bacteria is likely due to being below the threshold of virulence. At lower bacterial doses, the pathogenic load is insufficient to trigger the blockage mechanism or cause larval death. In these cases, it is probable that classical peristaltic movements of the gut efficiently eliminate the bacteria, preventing them from colonizing the posterior midgut or causing significant harm. Thus, the larvae rely on standard gut motility and immune mechanisms, rather than the blockage response, to clear lower doses of bacteria.

Why is this model only applied to high-dose infections? 

The reason this model primarily applies to high-dose infections is that lower concentrations of pathogenic bacteria do not trigger the blockage mechanism. As we mentioned in the manuscript, for low bacterial concentrations, where the GFP signal remains detectable, wild-type larvae are still able to resist live bacteria in the posterior part of the intestine.

Regarding the bacterial doses used in our experiments, it's important to clarify that we calculate the bacterial load based on colony-forming units (CFU). In our setup, there are approximately 5 × 10^4 CFU per midgut. For each experiment, we prepare 500 µl of contaminated medium containing 4 × 10^10 CFU. Fifty larvae are placed into this 500 µl of medium, meaning each larva ingests around 5 × 10^4 CFU within one hour of feeding.

This leads us to two key points:

(1) Continuous feeding might trigger the blockage response even at lower doses, as extended exposure to bacteria could lead to higher accumulation within the gut.

(2) Other defense mechanisms, such as the production of reactive oxygen species (ROS) or classical peristaltic movements, could be sufficient to eliminate lower bacterial doses (around 10^3 CFU or below).

We also refer to the newly provided Movie 5, where larvae fed with Bt-GFP at 1.3 × 10^10 CFU/mL show no blockage at low ingestion levels and successfully eliminate the bacteria.

(3) The authors claim that the lock of bacteria happens at 15 minutes while killing by AMPs happens 6-8 hours later. 

Our CFU data indicate that it’s after 4 to 6 hours that the quantity of bacteria decreases. We fixed this in the text.

What happened during this period? 

During the 4 to 6-hour period, several defense mechanisms are activated. ROS play a bacteriostatic and bacteriolytic role, helping to control bacterial growth. Concurrently, the IMD pathway is activated, leading to the transcription, translation, and secretion of antimicrobial peptides. These AMPs exert both bacteriostatic and bacteriolytic effects, contributing to the eventual clearance of the pathogenic bacteria.

More importantly, is IMD activity induced in the anterior region of the larval gut in both Ecc15 and Bt infection at 6 hours after infection? 

We have provided new data (Supplementary Figure 6) that includes RT-qPCR analysis of the whole larval gut in wt, TrpA1- and Dh31- genetic background after feeding with Lp, Ecc15, Bt, or yeast only. We monitored the expression of three different AMP-encoding genes and found that while AMP expression varied depending on the food content, there were no significant differences between the genotypes tested.

Additionally, we included new imaging data (Supplementary Figure 11) from AMP reporter larvae (Dpt-Cherry) fed with fluorescent Lp or Bt. In larvae infected with Bt, which is blocked in the anterior part of the gut, the dpt gene is predominantly induced in this region, indicating strong IMD pathway activity in response to Bt infection. Conversely, in larvae fed with Lp-GFP, the Dpt-Cherry reporter shows weak expression in the anterior midgut, and is barely detectable in the posterior midgut where Lp-GFP establishes itself. This aligns with previous findings by Bosco-Drayon et al. (2012), which demonstrated low AMP expression in the posterior midgut due to the presence of negative regulators of the IMD pathway, such as amidases and Pirk.

Are they mostly expressed in the anterior midgut in both bacterial infections? Several papers have shown quite different IMD activity patterns in the Drosophila gut. Zhai et al. have shown that in adult Drosophila, IMD activity was mostly absent in the R2 region as indicated by dpt-lacZ. Vodovar et al. have shown that the expression of dpt-lacZ is observable in proventriculus while Pe is not in the same region. Tzou et al. showed that Ecc15 infection induced IMD activity in the anterior midgut 24 hours after infection. 

Based on our new data (Supplementary Figure 11), we observe that Dpt-RFP expression is primarily localized in the anterior midgut and likely in the beginning of acidic region in larvae infected with Bt, Ecc and Lp. 

Using TrpA1 and Dh31 mutants, the authors found both Ecc15 and Bt in the posterior midgut. Why are they not evenly distributed along the gut? 

We observe that bacteria are not evenly distributed along the gut in wild-type larvae as well, with LP. This suggests that the transit time in the anterior part of the gut may be relatively short due to active peristaltism, which would make this region function as a "checkpoint" for bacteria that are not supposed to be blocked. Indeed, we confirmed that peristaltism is active during our intoxication experiments, which could explain the rapid movement of bacteria through the anterior midgut.

In contrast, bacteria tend to remain longer in the posterior midgut, which corresponds to the absorptive functions of intestinal cells in this region. This would explain why we observe more bacteria in the posterior midgut for Lp in control larvae and for Ecc15 and Bt in the TrpA1- and Dh31- mutants. Although a few bacteria are still found in the anterior midgut, they are consistently in much lower numbers compared to the posterior, as shown in Figures 1A and 3A of our manuscript.

Last but not least, does the ROS/TrpA1/Dh31 axis affect AMP expression?

We investigated whether the ROS/TrpA1/Dh31 axis influences AMP expression by performing RT-qPCR on the whole gut of larvae in wild-type, TrpA1-, and Dh31- genetic backgrounds. Larvae were fed with Lp, Ecc, Bt, or yeast (new data: Supplementary Figure 6). We monitored the expression of three different AMP-encoding genes and found that while AMP expression varied depending on the food content, there were no significant differences in AMP expression between the different genotypes.

Additionally, we provide imaging data from AMP reporter larvae (pDpt-Cherry) fed with fluorescent Lp or Bt (new data: Supplementary Figure 11). These results further confirm that the ROS/TrpA1/Dh31 axis does not significantly affect AMP expression in our experimental conditions.

(4) The TARM structure part is quite interesting. However, the authors did not show its relevance in their model. Is this structure the key-driven force for the blocking phenotype and killing phenotype? 

We agree that the TARM structures are a fascinating aspect of this study and acknowledge the interest in their potential role in the blocking and killing phenotypes. While we are keen to explore the specific contributions of these structures during bacterial intoxication, the current genetic tools available for manipulating TARMs target both TARM T1 and T2 simultaneously, as demonstrated by Bataillé et al., 2020 (Fig. 2). Of note, these muscles are essential for proper gut positioning in larvae, and their absence leads to significant defects in food intake and transit, which would confound the results of our intoxication experiments (see Fig. 6 from Bataillé et al., 2020).

Therefore, while TARMs are likely involved in these processes, the current limitations in selectively targeting them prevent us from definitively testing their role in bacterial blocking and killing at this stage. We hope to address this in future studies as more refined genetic tools become available.

Is the ROS/TrpA1/Dh31 axis required to form this structure?

To determine whether the ROS/TrpA1/Dh31 axis is required for the formation of TARM structures, we examined larval guts from control, TrpA1-, and Dh31- mutant backgrounds. Our new data (Supplementary Figure 8) show that the TARM T2 structures are still present in the mutants, indicating that the formation of these structures does not depend on the ROS/TrpA1/Dh31 axis.

Reviewer #2 (Public Review):

This article describes a novel mechanism of host defense in the gut of Drosophila larvae. Pathogenic bacteria trigger the activation of a valve that blocks them in the anterior midgut where they are subjected to the action of antimicrobial peptides. In contrast, beneficial symbiotic bacteria do not activate the contraction of this sphincter, and can access the posterior midgut, a compartment more favorable to bacterial growth.

Strengths:

The authors decipher the underlying mechanism of sphincter contraction, revealing that ROS production by Duox activates the release of DH31 by enteroendocrine cells that stimulate visceral muscle contractions. The use of mutations affecting the Imd pathway or lacking antimicrobial peptides reveals their contribution to pathogen elimination in the anterior midgut.

Weaknesses:

The mechanism allowing the discrimination between commensal and pathogenic bacteria remains unclear.

Based on our findings, we hypothesize that ROS play a crucial role in this discrimination process, with uracil release by pathogenic or opportunistic bacteria potentially serving as a key signal.

To test whether uracil could trigger this discrimination, we conducted experiments where Lp was supplemented with uracil. However, our results show that uracil supplementation alone was not sufficient to induce the blockage response (new data: Supplementary Figure 5). This suggests that while uracil may be a factor in bacterial discrimination, it is likely not the sole trigger, and additional bacterial factors or signals may be required to activate the blockage mechanism. 

The use of only two pathogens and one symbiotic species may not be sufficient to draw a conclusion on the difference in treatment between pathogenic and symbiotic species.

To address this concern, we performed additional intoxication experiments using Escherichia coli OP50, a bacterium considered innocuous and commonly used as a standard food source for C. elegans in laboratory settings. The results, presented in our updated data (new data: Fig 1B), show that E. coli OP50, despite being from the same genus as Ecc, does not trigger the blockage response. This further supports our conclusion that the gut’s discriminatory mechanism is specific to pathogenic bacteria, and not merely based on bacterial genus.

We can also wonder how the process of sphincter contraction is affected by the procedure used in this study, where larvae are starved. Does the sphincter contraction occur in continuous feeding conditions? Since larvae are continuously feeding, is this process physiologically relevant?

In our intoxication protocol, the larvae are exposed to contaminated food for 1 hour, during which the blockage ratio is quantified. Since this period involves continuous feeding with the contaminated food, we do not consider the larvae starved during the quantification process. Our observations show differences in the blockage response depending on the bacterial contaminant and the genetic background of the host. Additionally, we were able to trigger the blocking phenomenon using exogenous hCGRP.

Regarding the experimental setup for movie observations, it is true that larvae are immobilized on tape in a humid chamber, which is not a fully physiological context. However, in the new movie we provide (Movie 3), co-treatment with fluorescent Dextran (Red) and fluorescent Bt (Green) shows that both are initially blocked, followed by the posterior release of Dextran once the bacterial clearance begins.

Furthermore, to address the question of continuous exposure, we extended the exposure period to 20 hours instead of 1 hour. Even after prolonged exposure, we observed that pathogens are still blocked in the anterior part of the gut (new data: Supplementary Figure 2B). This supports the physiological relevance of the sphincter contraction and its ability to function under continuous feeding conditions.

Reviewer #1 (Recommendations For The Authors):

(1) The authors performed the experiments on Drosophila larvae. I wonder whether this model could extend to adult flies since they have shown that the ROS/TRPA1/Dh31 axis is important for gut muscle contraction in adult flies. If not, how would the authors explain the discrepancy between larvae and adults?

We link the adult phenotype to the one we describe in larvae in order to have the candidate approach toward the ROS/TrpA1/Dh31 axis. As we already mention in the discussion, while larvae stay in the food, adult flies can go away. If larvae eject their gut content, they may ingest it within minutes. We clarify our idea in the last part of the discussion.

(2) The authors performed their experiments and proposed the models based on two pathogenic bacteria and one commensal bacterial at a relatively high bacterial dose. They showed that feeding Bt at 2X1010 or Ecc15 at 4X108 did not induce a blockage phenotype. 

I wonder whether larvae die under conditions of enteric infection with low concentrations of pathogenic bacteria. 

Video provided with Bt-GFP 1.3 10^10 CFU/mL (new data: Movie 5). When larvae eat less, there is no blockage and bacteria can reach the posterior midgut. Note that the fluorescence is weak due to the low amount of bacteria ingested. The movie shows an excretion of the bacteria. There is also no death of the larvae. Together these results suggest that below a given threshold, the virulence of the bacteria is too weak to i) trigger a blockage and 2/ kill the larva. The bacteria are likely eliminated through classical peristaltism.

If larvae do not show mortality, what is the mechanism for resisting low concentrations of pathogenic bacteria? 

Maybe we are below the threshold of virulence. See our response just above.

Why is this model only applied to high-dose infections? 

As mentioned in the manuscript, lower concentrations do not trigger the blockage and for lower concentrations with a GFP signal still detectable, wild-type animals resist the presence of live-bacteria within the posterior part of the intestine.

About the doses, the CFU should be considered. Indeed, there are around 5.10^4 CFU per midgut. In our experimental procedure we calculate the amount of bacteria for 500 µl of contaminated medium (i.e. 4.10^10 CFU/500µl of medium). Then around 50 larvae were deposited in the 500µl of contaminated media. In this condition, one larva ingests 5.10^4 CFU. Moreover, larvae are only fed for 1h. 

So 1/ continuous feeding may also trigger locking even at lower doses and 2/ the other mechanisms of defenses (such as ROS) or peristalsis may be sufficient to eliminate lower doses (i.e. 10^3 CFU or below). See the new movie 5 we provide with Bt-GFP 1.3 10^10 CFU/mL

(3) The authors claim that the lock of bacteria happens at 15 minutes while killing by AMPs happens 6-8 hours later. 

Our CFU data indicate that it’s after 4 to 6 hours that the quantity of bacteria decreases. We fixed this in the text.

What happened during this period? 

ROS activity (bacteriostatic and bacteriolytic), IMD activation, AMP transcription, translation, secretion and bacteriostatic as well as bacteriolytic activity.

More importantly, is IMD activity induced in the anterior region of the larval gut in both Ecc15 and Bt infection at 6 hours after infection? 

We provide new data for larval whole gut RT-qPCR data in wt, TrpA1- and Dh31- genetic background fed with Lp or Ecc or Bt or yeast only (new data: SUPP6). We monitored 3 different AMP-encoding genes and found differences related to the food content, but no differences between genotypes. In addition, we provide images from AMP reporter animals (Dpt-Cherry) fed with fluorescent Lp or Bt (new data: SUPP11) showing that with Bt blocked in the anterior part of the intestine, the dpt gene is mainly induced in this area. Note that in the larva infected with Lp-GFP, the Dpt-Cherry reporter is weakly expressed in the anterior midgut. In the posterior midgut, the place where Lp-GFP is established, Dpt-Cherry is barely detectable. This observation is in line with the previous observation made by Bosco-Drayon et al., (2012) demonstrating the low level of AMP expression in the posterior midgut due to the expression of the IMD negative regulators such as amidases and pirk. In the larva infected with Bt-GFP, note the obvious expression of DptCherry in the anterior midgut colocalizing with the bacteria (new data: SUPP11).

Are they mostly expressed in the anterior midgut in both bacterial infections? Several papers have shown quite different IMD activity patterns in the Drosophila gut. Zhai et al. have shown that in adult Drosophila, IMD activity was mostly absent in the R2 region as indicated by dpt-lacZ. Vodovar et al. have shown that the expression of dpt-lacZ is observable in proventriculus while Pe is not in the same region. Tzou et al. showed that Ecc15 infection induced IMD activity in the anterior midgut 24 hours after infection. 

In ctrl animals fed Bt, Ecc and Lp we see Dpt-RFP in anterior midgut and likely in the beginning of acidic region. See the new data: SUPP11 images provided for the previous remark.

Using TrpA1 and Dh31 mutants, the authors found both Ecc15 and Bt in the posterior midgut. Why are they not evenly distributed along the gut? 

Same is true with Lp in wt; not evenly distributed. As if the transit time in the anterior part is very short due to peristaltism which would fit for a check point area if you’re not supposed to be blocked. Indeed, peristaltism is active during our intoxications. Then, it stays longer in the posterior part, fitting with the absorptive skills of the intestinal cells in this area. With Lp in ctrl or Ecc and Bt in TrpA1- and Dh31- mutants, there are always a few in the anterior midgut but always much less compared to the posterior. See our figure 1A and 3A.

Last but not least, does the ROS/TrpA1/Dh31 axis affect AMP expression?

We provide larval whole gut RT-qPCR data in wt, TrpA1- and Dh31- genetic background fed with Lp or Ecc or Bt or yeast only (new data: SUPP6). We monitored 3 different AMPencoding genes and found differences related to the food content, but no differences between genotypes. In addition, we provide images from AMP reporter animals (pDptCherry) fed with fluorescent Lp or Bt, (new data: SUPP11).

(4) The TARM structure part is quite interesting. However, the authors did not show its relevance in their model. Is this structure the key-driven force for the blocking phenotype and killing phenotype? 

Indeed, we would like to explore the roles of these structures and the putative requirement upon bacterial intoxication using some driver lines developed by the team that studied these muscles in vivo. However, the genetic tools currently available will target TARMsT1 and T2 at the same time. See Fig 2 form Bataillé et al, . 2020. Moreover, these TARMs are, at first, crucial for the correct positioning of the gut within the larvae and their absence lead to a global food intake and transit defect that will bias the outcomes of our intoxication protocol (see fig 6 from Bataillé et al,. 2020).

Is the ROS/TrpA1/Dh31 axis required to form this structure?

We provide images of larval guts from ctrl, TrpA1 and Dh31 mutants demonstrating the presence of the TARMs T2 structures despite the mutations (new data: SUPP8). In addition, we provide representative movies of peristalsis in intestines of Dh31 mutants fed or not with Ecc to illustrate that muscular activity is not abolished (new data: Movie 9 and Movie 10).

Minor points:

(1) Why not use the Pros-Gal4/UAS-Dh31 strain in Figure 3B in addition to hCGRP?

We opted for exogenous hCGRP addition because it allowed us precise timing control over Dh31 activation. Overexpression of Dh31 from embryogenesis or early larval stages could have significant and unintended effects on intestinal physiology, potentially confounding the results. While temporal control using TubG80ts could be an alternative, our focus was on identifying the specific cells responsible for the phenomenon.

To achieve this, we perturbed Dh31 production via RNAi, specifically targeting a limited number of enteroendocrine cells (EECs) using the DJ752-Gal4 driver, as described by Lajeunesse et al., 2010. Our new data (Supplementary Figure 4) demonstrate that Dh31 expression in this subset of cells is indeed necessary for the blockage phenomenon.

(2) Section title (line 287) refers to mortality, but no mortality data is in the figure.

We agree that the title referenced mortality, whereas no mortality data was presented in this section. We have updated the title to better reflect the data discussed in this part of the manuscript.

(3) It may be better to combine ROS-related contents in the same figure.

While it is technically feasible to consolidate the ROS-related content into one figure, doing so would require splitting essential data, such as the Gal4 controls for the RNAi assays and parts of the survival phenotype data. We believe that the current structure of the study, which first explores the molecular aspects of the phenomenon and then demonstrates its relevance to the animal’s survival, provides a clearer and more logical flow. For these reasons, we prefer to maintain the current figure layout.

Reviewer #2 (Recommendations For The Authors):

Major recommendation

(1) Other wild-type backgrounds should be added (including the w Drosdel background of the AMP14 deficient flies) to check the robustness of the phenotype.

To address the concern regarding the robustness of the phenotype across different wildtype backgrounds, we have tested additional genetic backgrounds, including w1, the isogenized w1118 and Oregon animals. 

The results (new data: Figure 1C) demonstrate that Lp is able to transit freely to the posterior part of the intestine in all backgrounds, while Ecc and Bt are blocked in the anterior part. These findings confirm the robustness of the phenotype across different wildtype strains.

(2) Although we recognize that this may be limited by the number of GFP-expressing species, other commensal and pathogenic bacteria should be tested in this assay (e.g. E. faecalis and Acetobacter).

We performed new intoxication experiments using Escherichia coli OP50, a wellestablished innocuous bacterial strain. The data, presented in Figure 1B (new data), show that E. coli OP50, despite being from the same genus as Ecc, does not trigger the blockage response. This further supports our hypothesis that the blockage phenomenon is specific to pathogenic bacteria and not simply related to the bacterial genus.

(3) It is important to test whether sphincter closure also occurs in continuous feeding conditions. This does not mean repeating all the experiments but just shows that this mechanism can take place in conditions where larvae are kept in a vial with food.

While the movies we provide involve larvae immobilized on tape in a humid chamber, which is not a fully physiological context, we now provide new data (Movie 3) showing that, after co-treatment with fluorescent Dextran (Red) and fluorescent Bt (Green), both substances are initially blocked in the anterior midgut. Later, the dextran is released posteriorly once bacterial clearance has begun.

Additionally, we extended the feeding period in our experiments from 1 hour to 20 hours to simulate more continuous exposure to contaminated food. Even under these prolonged conditions, we observed that pathogens are still blocked in the anterior part of the gut (new data: Supplementary Figure 2B). This confirms that the sphincter mechanism can function in continuous feeding conditions as well.

(4) What are the molecular determinants discriminating innocuous from pathogenic bacteria? Addressing this point will increase the impact of the article. The fact that Relish mutants have normal valve constriction suggests that peptidoglycan recognition is not involved. Is there a sensing of pathogen virulence factors? 

Our data suggest that uracil could be a key molecular determinant in discriminating between innocuous and pathogenic bacteria, as previously described by the W-J Lee team in several studies on adult Drosophila. However, in our experiments, exogenous uracil addition using the blue dye protocol (Keita et al., 2017) did not induce any significant changes in the larvae. Similarly, uracil supplementation in adult flies failed to trigger the Ecc expulsion and gut contraction phenotype, as reported by Benguettat et al., 2018. 

To further investigate this, we tested the addition of uracil during Lp-GFP intoxication. In these experiments, we did not observe any blockage of Lp (new data: Supplementary Figure 5). These results suggest that uracil might not be the sole trigger for the blockage response, or we may not be providing uracil exogenously in the most effective way. Alternatively, there could be other pathogen-specific virulence factors that contribute to this discrimination mechanism.

To address this question, the authors should infect larvae with Ecc15 evf- mutants or Ecc15 lacking uracil production. 

Thank you for your suggestion to use Ecc15 evf- mutants or Ecc15 lacking uracil production to explore the role of uracil in bacterial discrimination. While we have provided some data using uracil supplementation (new data: Supplementary Figure 5), we agree that testing mutants like PyrE would be an important next step. Unfortunately, we currently lack access to fluorescent PyrE or Ecc15 evf- mutants.

We are planning to address this by developing a new protocol involving fluorescent beads alongside bacteria. This approach will allow us to test several bacterial strains in parallel and better define the size threshold of the valve. However, we do not have the relevant data yet, but this will be a key focus of our future work.

Similarly, does feeding heat-killed Ecc15 or Bt induce sequestration in the anterior midgut (larvae may be fed dextran-FITC at the same time to track bacteria)?

Unfortunately, in our attempts to test heat-killed or ethanol-killed fluorescent Ecc15 for these experiments, we encountered an issue: while we were able to efficiently kill the bacteria, we lost the GFP signal required to track their position in the gut. This made it challenging to assess whether sequestration in the anterior midgut occurs with non-viable bacteria.

Is uracil or Bt toxin feeding sufficient to induce valve closure? 

As previously mentioned, uracil is a strong candidate for bacterial discrimination, and we have tested its role by adding exogenous uracil during Lp-GFP intoxication. However, in these experiments, Lp was not blocked (new data: Supplementary Figure 5). This suggests that uracil alone may not be sufficient to induce valve closure, or it may not be the only factor involved. It is also possible that our method of exogenous uracil supplementation may not be effectively mimicking the endogenous conditions.

Regarding Bt, we used vegetative cells without Cry toxins in our experiments. Cry toxins are only produced during sporulation and are enclosed in crystals within the spore. The Bt strain we used, 4D22, has been deleted for the plasmids encoding Cry toxins. As a result, there were no Cry toxins present in the Bt-GFP vegetative cells used in our assays. This has been clarified in the Materials and Methods section of the manuscript.

Would Bleomycin induce the same phenotype? 

Indeed, Bleomycin, as well as paraquat, has been shown to damage the gut and trigger intestinal cell proliferation in adult Drosophila through mechanisms involving TrpA1. Testing whether Bleomycin induces a similar phenotype in larvae would indeed be interesting.

However, one challenge we face in our intoxication protocol is that larvae tend to stop feeding when chemicals are added to their food mixture. We encountered similar difficulties in our DTT experiments, which were challenging to set up for this reason. Consequently, we aim to avoid approaches that might impair the general feeding activity of the larvae, as it can significantly affect the outcomes of our experiments.

Could this process of sphincter closure be more related to food poisoning?

If gut damage were the primary trigger for sphincter closure, we would indeed expect the blockage phenomenon to occur later following bacterial exposure. However, in our experiments, we observe the blockage occurring early after bacterial contact, suggesting that damage may not be the main trigger for this response.

That said, we have not yet tested bacterial mutants lacking toxins, nor have we tested a direct damaging agent such as Bleomycin, as proposed. These would be valuable future experiments to explore the potential role of gut damage more thoroughly in this process.

(5) Is Imd activation normal in trpA1 and DH31 mutants? The authors could use a diptericin reporter gene to check if Diptericin is affected by a lack of valve closure in trpA1.

To address this, we performed RT-qPCR on whole larval guts from wt, TrpA11 and Dh31KG09001 genetic background. Larvae were fed with Lp, Ecc, Bt or yeast only (new data: SUPP6). We monitored the expression of three different AMP-encoding genes and found that while AMP expression varied depending on the food content, there were no significant differences in AMP expression between the genotypes.

Additionally, we provide imaging data from AMP reporter animals (pDpt-Cherry) in a wildtype background, fed with fluorescent Lp or Bt (new data: Supplementary Figure 11). These images also support the conclusion that Diptericin expression is not significantly affected by a lack of valve closure in trpA1 and Dh31 mutants.

(6) Are the 2-6 DH31 positive cells the same cells described by Zaidman et al., Developmental and Comparative Immunology 36 (2012) 638-647.

The cells identified as hemocytes in the midgut junctions by Zaidman et al. are likely the same cells we describe in our study, as they are located in the same region and are Dh31 positive. We have added a reference to this paper and included lines in the manuscript acknowledging this connection.

Although confirming whether these cells are Hml+, Dh31+, and TrpA1+ would clarify their exact identity, this falls outside the scope of our current study. However, the possibility that these cells play a role in physical barrier immunity and also possess a hemocyte identity is indeed intriguing, and we hope future research will explore this further.

Minor points

(1) The mutations should be appropriately labelled with the allele name.

This has been fixed in the main text, in Fig Legends, and in figures. 

(2) Line 230-231: the sentence is unclear to me.

We simplified the sentence and do not refer to the expulsion in larvae.

(3) Discussion: although the discussion is already a bit long, it would be interesting to see if this process is likely to happen/has been described in other insects (mosquito, Bactrocera, ...).

We reviewed the available literature but were unable to find specific examples describing the blockage phenomenon in other insects. Most studies we found focused on symbiotic bacteria rather than pathogenic or opportunistic bacteria. However, as mentioned in our manuscript, the anterior localization of opportunistic or pathogenic bacteria has been observed in Drosophila by independent research groups.

(4) Line 546: add the Caudal Won-Jae Lee paper to state the posterior midgut is less microbicidal.

We added the reference at the right place, mentioning as well that it concerns adults. 

(5)  Figure 6 indicates what the cells are, shown by the arrow.

The sentence ‘the arrows point to TARMs’ is present in the legend of Fig6.

(6) Does the sphincter closure depend on hemocytes?

As mentioned above, the cells we identify as TrpA1+ in the midgut junction may be the same cells described by Zaidman et al., 2012, and earlier by Lajeunesse et al., 2010. Inactivating hemocytes using the Hml-Gal4 driver may also affect these Dh31+ cells, as they share similarities with hemocytes, as pointed out by Zaidman et al. However, distinguishing between hemocytes and Dh31+/TrpA1+ cells would require a genetic intersectional approach, which is beyond the scope of our current study.

Nevertheless, the possibility that these cells play a dual role in immunity (through blockage) and share characteristics with hemocytes while functioning as enteroendocrine cells (EECs) is quite intriguing and deserves further exploration in future studies.

Author response:

Reviewer #1 (Public review):

Lu et al. use their workflow to visualize RNA expression of five enzymes that are each involved in the biosynthetic pathway of different neurotransmitters/modulators, namely chat (cholinergeric), gad (GABAergic), tbh (octopaminergic), th (dopaminergic), and tph (serotonergic). In this way, they generate an anatomical atlas of neurons that produce these molecules. Collectively these markers are referred to as the "neuronpool." They overstate when they write, "The combination of these five types of neurons constitutes a neuron pool that enables the labeling of all neurons throughout the entire body." This statement does not accurately represent the state of our knowledge about the diversity of neurons in S. mediterranea. There are several lines of evidence that support the presence of glutamatergic and glycinergic neurons, including the following. The glutamate receptor agonists NMDA and AMPA both produce seizure-like behaviors in S. mediterranea that are blocked by the application of glutamate receptor antagonists MK-801 and DNQX (which antagonize NMDA and AMPA glutamate receptors, respectively; Rawls et al., 2009). scRNA-Seq data indicates that neurons in S. mediterranea express a vesicular glutamate transporter, a kainite-type glutamate receptor, a glycine receptor, and a glycine transporter (Brunet Avalos and Sprecher, 2021; Wyss et al., 2022). Two AMPA glutamate receptors, GluR1 and GluR2, are known to be expressed in the CNS of another planarian species, D. japonica (Cebria et al., 2002). Likewise, there is abundant evidence for the presence of peptidergic neurons in S. mediterranea (Collins et al., 2010; Fraguas et al., 2012; Ong et al., 2016; Wyss et al., 2022; among others) and in D. japonica (Shimoyama et al., 2016). For these reasons, the authors should not assume that all neurons can be assayed using the five markers that they selected. The situation is made more complex by the fact that many neurons in S. mediterranea appear to produce more than one neurotransmitter/modulator/peptide (Brunet Avalos and Sprecher, 2021; Wyss et al., 2022), which is common among animals (Vaaga et al., 2014; Brunet Avalos and Sprecher, 2021). However the published literature indicates that there are substantial populations of glutamatergic, glycinergic, and peptidergic neurons in S. mediterranea that do not produce other classes of neurotransmission molecule (Brunet Avalos and Sprecher, 2021; Wyss et al., 2022). Thus it seems likely that the neuronpool will miss many neurons that only produce glutamate, glycine or a neuropeptide.

In response to your comments, we agree that our initial statement regarding the "neuron pool" overstated the extent of neuronal coverage provided by the five selected markers. We have revised the sentence as “The combination of these five types of neurons constitutes a neuron pool that enables the labeling of most of the neurons throughout the entire body, including the eyes, brain, and pharynx”. 

Furthermore, we chose the five neurotransmitter systems (cholinergic, GABAergic, octopaminergic, dopaminergic, and serotonergic) based on their well-characterized roles in planarian neurobiology and the availability of reliable markers. However, we acknowledge the limitations of this approach and recognize that it does not encompass all neuron types, particularly those involved in glutamatergic, glycinergic, and peptidergic signaling, which have been documented in S. mediterranea. We will also add the content about other neuron types in our revised manuscript “Additionally, there is considerable diversity among glutamatergic, glycinergic, and peptidergic neurons in planarians. Many neurons in S. mediterranea express more than one neurotransmitter or neuropeptide, which adds further complexity to the system.”

The authors use their technique to image the neural network of the CNS using antibodies raised vs. Arrestin, Synaptotagmin, and phospho-Ser/Thr. They document examples of both contralateral and ipsilateral projections from the eyes to the brain in the optic chiasma (Figure 1C-F). These data all seem to be drawn from a single animal in which there appears to be a greater than normal number of nerve fiber defasciculatations. It isn't clear how well their technique works for fibers that remain within a nerve tract or the brain. The markers used to image neural networks are broadly expressed, and it's possible that most nerve fibers are too densely packed (even after expansion) to allow for image segmentation. The authors also show a close association between estrella-positive glial cells and nerve fibers in the optic chiasma. 

Thank you for your detailed feedback. While we did not perform segmentation of all neuron fibers, we were able to segment more isolated fibers that were not densely packed within the neural tracts. We use 120 nm resolution to segment neurons along the three axes. Our data show the presence of both contralateral and ipsilateral projections of visual neurons. Although Figure 1C-F shows data from one planarian, we imaged three independent specimens to confirm the consistency of these observations. In the revised manuscript, we will include a discussion on the limitations of TLSM in reconstructing neural networks, particularly when it comes to resolving fibers within densely packed regions of the nerve tracts.

The authors count all cell types, neuron pool neurons, and neurons of each class assayed. They find that the cell number to body volume ratio remains stable during homeostasis (Figure S3C), and that the brain volume steadily increases with increasing body volume (Figure S3E). They also observe that the proportion of neurons to total body cells is higher in worms 2-6 mm in length than in worms 7-9 mm in length (Figure 2D, S3F). They find that the rate at which four classes of neurons (GABAergic, octopaminergic, dopaminergic, serotonergic) increase relative to the total body cell number is constant (Figure S3G-J). They write: "Since the pattern of cholinergic neurons is the major cell population in the brain, these results suggest that the above observation of the non-linear dynamics between neurons and cell numbers is likely from the cholinergic neurons." This conclusion should not be reached without first directly counting the number of cholinergic neurons and total body cells. Given that glutamatergic, glycinergic, and peptidergic neurons were not counted, it also remains possible that the non-linear dynamics are due (in part or in whole) to one or more of these populations. 

We have removed the statement "Since the pattern of cholinergic neurons is the major cell population in the brain, these results suggest that the above observation of the non-linear dynamics between neurons and cell numbers is likely from the cholinergic neurons." We changed this statement into “These results suggest that the above observation of the non-linear dynamics between neurons and cell numbers is not likely from the octopaminergic, GABAergic, dopaminergic and serotonergic neurons. Since our neuron pool may not include glutamatergic, glycinergic, and peptidergic neurons, we would like to add the possibility that the non-linear dynamics may be from cholinergic neurons or other neurons not included in our staining.”

Reviewer #2 (Public review): 

Weaknesses: 

(1) The proprietary nature of the microscope, protected by a patent, limits the technical details provided, making the method hard to reproduce in other labs. 

Thank you for your comment. We understand the importance of reproducibility and transparency in scientific research. We would like to point out that the detailed design and technical specifications of the TLSM are publicly available in our published work: Chen et al., Cell Reports, 2020. Additionally, the protocol for C-MAP, including the specific experimental steps, is comprehensively described in the methods section of this paper. We believe that these resources should provide sufficient information for other labs to replicate the method.

(2) The resolution of the analyses is mostly limited to the cellular level, which does not fully leverage the advantages of expansion microscopy. Previous applications of expansion microscopy have revealed finer nanostructures in the planarian nervous system (see Fan et al. Methods in Cell Biology 2021; Wang et al. eLife 2021). It is unclear whether the current protocol can achieve a comparable resolution. 

Thank you for raising this important point. The strength of our C-MAP protocol lies in its fluorescence-protective nature and user convenience. Notably, the sample can be expanded up to 4.5-fold linearly without the need for heating or proteinase digestion, which helps preserve fluorescence signals. In addition, the entire expansion process can be completed within 48 hours. While our current analysis focused on cellular-level structures, our method can achieve comparable or better resolution and we will add this information in the revised manuscript.

(3) The data largely corroborate past observations, while the novel claims are insufficiently substantiated. 

A few major issues with the claims: 

(4) Line 303-304: While 6G10 is a widely used antibody to label muscle fibers in the planarian, it doesn't uniformly mark all muscle types (Scimone at al. Nature 2017). For a more complete view of muscle fibers, it is important to use a combination of antibodies targeting different fiber types or a generic marker such as phalloidin. This raises fundamental concerns about all the conclusions drawn from Figures 4 and 6 about differences between various muscle types. Additionally, the authors should cite the original paper that developed the 6G10 antibody (Ross et al. BMC Developmental Biology 2015). 

We appreciate the reviewer’s insightful comments and acknowledge that 6G10 does not uniformly label all muscle fiber types. We agree that this limitation should be recognized in the interpretation of our results. we will revise the manuscript to explicitly state the limitations of using 6G10 alone for muscle fiber labeling and highlight the need for additional markers. We would also clarify that the primary objective of our study was not to distinguish all muscle fiber types but rather to demonstrate the application of our 3D tissue reconstruction method in addressing traditional research questions. Nonetheless, we agree that expanding the labeling strategy in future studies would allow for a more thorough investigation of muscle fiber diversity. We will ensure all citations are properly revised and updated in our next version.

(5) Lines 371-379: The claim that DV muscles regenerate into longitudinal fibers lacks evidence. Furthermore, previous studies have shown that TFs specifying different muscle types (DV, circular, longitudinal, and intestinal) both during regeneration and homeostasis are completely different (Scimone et al., Nature 2017 and Scimone et al., Current Biology 2018). Single-cell RNAseq data further establishes the existence of divergent muscle progenitors giving rise to different muscle fibers. These observations directly contradict the authors' claim, which is only based on images of fixed samples at a coarse time resolution. 

Thank you for your valuable feedback. Our intent was not to suggest that DV muscles regenerate into longitudinal fibers. Our observations focused on the wound site, where DV muscle fibers appear to reconnect, and longitudinal fibers, along with other muscle types, gradually regenerate to restore the structure of the injured area. We will revise the relevant sections of the manuscript to clarify this dynamic process more accurately.

(6) Line 423: The manuscript lacks evidence to claim glia guide muscle fiber branching. 

We will remove this statement from the revised version. Instead, we will focus on describing our observations of the connections between glial cells and muscle fibers.

(7) Lines 432/478: The conclusion about neuronal and muscle guidance on glial projections is similarly speculative, lacking functional evidence. It is possible that the morphological defects of estrella+ cells after bcat1 RNAi are caused by Wnt signaling directly acting on estrella+ cells independent of muscles or neurons. 

We understand that this approach is insufficient and we will revise the manuscript to more clearly state the limitations of our data. We will describe our observations as preliminary and suggest that further experiments are required.

(8) Finally, several technical issues make the results difficult to interpret. For example, in line 125, cell boundaries appear to be determined using nucleus images; in line 136, the current resolution seems insufficient to reliably trace neural connections, at least based on the images presented. 

We use two setups for imaging cells and neuron projections. For cellular resolution imaging, we utilized a 1× air objective with a numerical aperture (NA) of 0.25 and a working distance of 60 mm (OLYMPUS MV PLAPO). The voxel size used was 0.8×0.8×2.5 µm3. This configuration resulted in a resolution of 2×2×5 µm3 and a spatial resolution of 0.5×0.5×1.25 µm3 with 4× isotropic expansion. Alternatively, for sub-cellular imaging, we employed a 10×0.6 SV MP water immersion objective with 0.8 NA and a working distance of 8 mm (OLYMPUS). The voxel size used in this configuration was 0.26×0.26×0.8 µm3. As a result of this configuration, we achieved a resolution of 0.5×0.5×1.6 µm3 and a spatial resolution of 0.12×0.12×0.4 µm3 with a 4.5× isotropic expansion. The higher resolution achieved with sub-cellular imaging allows us to observe finer structures and trace neural connections.

Regarding your question about cell boundaries, we will revise the manuscript to specify that the boundaries we identified are those of each nucleus, rather than entire cells. This distinction will be made clear in the revised version.

Reviewer #3 (Public review): 

Weaknesses: 

(1) The work would have been strengthened by a more careful consideration of previous literature. Many papers directly relevant to this work were not cited. Such omissions do the authors a disservice because in some cases, they fail to consider relevant information that impacts the choice of reagents they have used or the conclusions they are drawing. 

For example, when describing the antibody they use to label muscles (monoclonal 6G10), they do not cite the paper that generated this reagent (Ross et al PMCID: PMC4307677), and instead, one of the papers they do cite (Cebria 2016) that does not mention this antibody. Ross et al reported that 6G10 does not label all body wall muscles equivalently, but rather "predominantly labels circular and diagonal fibers" (which is apparent in Figure S5A-D of the manuscript being reviewed here). For this reason, the authors of the paper showing different body wall muscle populations play different roles in body patterning (Scimone et al 2017, PMCID: PMC6263039, also not cited in this paper) used this monoclonal in combination with a polyclonal antibody to label all body wall muscle types. Because their "pan-muscle" reagent does not label all muscle types equivalently, it calls into question their quantification of the different body wall muscle populations throughout the manuscript. It does not help matters that their initial description of the body wall muscle types fails to mention the layer of thin (inner) longitudinal muscles between the circular and diagonal muscles (Cebria 2016 and citations therein). 

Ipsilateral and contralateral projections of the visual axons were beautifully shown by dye-tracing experiments (Okamoto et al 2005, PMID: 15930826). This paper should be cited when the authors report that they are corroborating the existence of ipsilateral and contralateral projections. 

Thank you for your feedback. We will incorporate these citations and clarifications into the revised manuscript. We acknowledge the limitations of this approach and recognize that it does not encompass all neuron types, particularly those involved in glutamatergic, glycinergic, and peptidergic signaling. We will also add the content about other neuron types in our revised version.

(2) The proportional decrease of neurons with growth in S. mediterranea was shown by counting different cell types in macerated planarians (Baguna and Romero, 1981; https://link.springer.com/article/10.1007/BF00026179) and earlier histological observations cited there. These results have also been validated by single-cell sequencing (Emili et al, bioRxiv 2023, https://www.biorxiv.org/content/10.1101/2023.11.01.565140v). Allometric growth of the planaria tail (the tail is proportionately longer in large vs small planaria) can explain this decrease in animal size. The authors never really discuss allometric growth in a way that would help readers unfamiliar with the system understand this. 

Thank you for your feedback. We will incorporate these citations and clarifications into the revised manuscript.

(3) In some cases, the authors draw stronger conclusions than their results warrant. The authors claim that they are showing glial-muscle interactions, however, they do not provide any images of triple-stained samples labeling muscle, neurons, and glia, so it is impossible for the reader to judge whether the glial cells are interacting directly with body wall muscles or instead with the well-described submuscular nerve plexus. Their conclusion that neurons are unaffected by beta-cat or inr-1 RNAi based on anti-phospho-Ser/Thr staining (Fig. 6E) is unconvincing. They claim that during regeneration "DV muscles initially regenerate into longitudinal fibers at the anterior tip" (line 373). They provide no evidence for such switching of muscle cell types, so it is unclear why they say this. 

We acknowledge that some of our conclusions were overclaimed given the current data, and we appreciate the opportunity to clarify and refine these claims in the revised manuscript. Regarding the statement that "DV muscles initially regenerate into longitudinal fibers at the anterior tip" (line 373), as addressed in our previous response, this phrasing was unclear. Our intent was not to imply that DV muscles switch into longitudinal fibers. Instead, we observed that muscle fibers reconnect at the wound site, with longitudinal fibers and other muscle types gradually restoring the structure. We will revise this section to better describe the dynamic changes observed during regeneration.

(4) The authors show how their automated workflow compares to manual counts using PI-stained specimens (Figure S1T). I may have missed it, but I do not recall seeing a similar ground truth comparison for their muscle fiber counting workflow. I mention this because the segmented image of the posterior muscles in Figure 4I seems to be missing the vast majority of circular fibers visible to the naked eye in the original image. 

Thank you for raising this important point. We will include a ground truth comparison of our automated muscle fiber counting with manual counts in the supplementary figures. Regarding the observation of missing circular fibers in Figure 4I, we agree that the segmentation appears to have missed a significant number of circular fibers in this particular image. This may have been due to limitations in the current parameters of the segmentation algorithm, especially in distinguishing fibers in regions of varying intensity or overlap. We are revisiting the segmentation parameters to improve the accuracy of detecting circular fibers, and we will provide an updated version of Figure 4I in the revised manuscript.

(5) It is unclear why the abstract says, "We found the rate of neuron cell proliferation tends to lag..." (line 25). The authors did not measure proliferation in this work and neurons do not proliferate in planaria. 

Thank you for bringing this to our attention. What we intended to convey was the increase in neuron number during homeostasis. We will revise the abstract to avoid this mistake in this context and instead describe it as the increase in neuron numbers due to progenitor cell differentiation during homeostasis.

(6) It is unclear what readers are to make of the measurements of brain lobe angles. Why is this a useful measurement and what does it tell us? 

The measurement of brain lobe angles is intended to provide a quantitative assessment of the growth and morphological changes of the planarian brain during regeneration. Additionally, the relevance of brain lobe angles has been explored in previous studies, such as Arnold et al., Nature, 2016, further supporting its use as a meaningful parameter.

(7) The authors repeatedly say that this work lets them investigate planarians at the single-cell level, but they don't really make the case that they are seeing things that haven't already been described at the single-cell level using standard confocal microscopy. 

Thank you for your comment. We agree that single-cell level imaging has been previously achieved in planarians using conventional confocal microscopy. However, our goal was to extend the application of expansion microscopy by combining C-MAP with tiling light sheet microscopy (TLSM), which allows for faster and high-resolution 3D imaging of whole-mount planarians. This combination offers several key advantages over traditional confocal microscopy. For example, it enables high-throughput imaging across entire organisms with a level of detail and speed that is not easily achieved using confocal methods. This approach allows us to investigate the planarian nervous system at multiple developmental and regenerative stages in a more comprehensive manner, capturing large-scale structures while preserving fine cellular details. The ability to rapidly image whole planarians in 3D with this resolution provides a more efficient workflow for studying complex biological processes. We believe this distinction is significant and represents an advance over previous methods. We will clarify this point in the manuscript to better distinguish our approach from standard techniques.

Author response:

Reviewer #1 (Public review):

Summary:

In this article the authors described mouse models presenting with backer muscular dystrophy, they created three transgenic models carrying three representative exon deletions: ex45-48 del., ex45-47 19 del., and ex45-49 del. This article is well written but needs improvement in some points.

Strengths:

This article is well written. The evidence supporting the authors' claims is robust, though further implementation is necessary. The experiments conducted align with the current state-of-the-art methodologies.

Weaknesses:

This article does not analyze atrophy in the various mouse models. Implementing this point would improve the impact of the work

We thank the reviewer for their constructive suggestions and comments on this work. Muscle hypertrophy is shown with growth in dystrophin-deficient skeletal muscle in mdx mice; thus, we did not pay attention to the factors associated with muscle atrophy in BMD mice. As the reviewer suggested, the examination of the association between type IIa fiber reduction and muscle atrophy is important, and the result is considered to be helpful in resolving the cause of type IIa fiber reduction in BMD mice.

Thus, we are planning to:

(1) Evaluate the cross-sectional areas (CSA) of muscles and compare them with the changes in the proportion of type IIa fibers.

(2) Evaluate the expression levels of Murf1 and Atrogin1 as markers of muscle atrophy using RT-PCR.

Reviewer #2 (Public review):

Summary

Miyazaki et al. established three distinct BMD mouse models by deleting different exon regions of the dystrophin gene, observed in human BMD. The authors demonstrated that these models exhibit pathophysiological changes, including variations in body weight, muscle force, muscle degeneration, and levels of fibrosis, alongside underlying molecular alterations such as changes in dystrophin and nNOS levels. Notably, these molecular and pathological changes progress at different rates depending on the specific exon deletions in the dystrophin gene. Additionally, the authors conducted extensive fiber typing, revealing a site-specific decline in type IIa fibers in BMD mice, which they suggest may be due to muscle degeneration and reduced capillary formation around these fibers.

Strengths:

The manuscript introduces three novel BMD mouse models with different dystrophin exon deletions, each demonstrating varying rates of disease progression similar to the human BMD phenotype. The authors also conducted extensive fiber typing across different muscles and regions within the muscles, effectively highlighting a site-specific decline in type IIa muscle fibers in BMD mice.

Weaknesses:

The authors have inadequate experiments to support their hypothesis that the decay of type IIa muscle fibers is likely due to muscle degeneration and reduced capillary formation. Further investigation into capillary density and histopathological changes across different muscle fibers is needed, which could clarify the mechanisms behind these observations.

We thank the reviewer for these positive comments and the very important suggestion about type IIa fiber reduction and capillary change around muscle fibers in BMD mice. From the results of the cardiotoxin-induced muscle degeneration and regeneration model, type IIa and IIx fibers showed delayed recovery compared with that of type-IIb fibers. However, this delayed recovery of type IIa and IIx could not explain the cause of the selective muscle fiber reduction limited to type IIa fibers in BMD mice. Therefore, we considered vascular dysfunction as the reason for the selective type IIa fiber reduction, and we found morphological capillary changes from a “ring pattern” to a “dot pattern” around type IIa fibers in BMD mice. However, the association between selective type IIa fiber reduction and the capillary change around muscle fibers in BMD mice remains unclear due to the lack of information about capillaries around type IIx and IIb fibers. The reviewer pointed out this insufficient evaluation of capillaries around other muscle fibers (except for type IIa fibers), and this suggestion is very helpful for explaining the association between selective type IIa fiber reduction and vascular dysfunction in BMD mice.

Thus, we are planning to:

(1) Evaluate the changes in capillary formation around other muscle fibers, except for type IIa fibers (e.g., type IIx and IIb fibers).

(2) Evaluate the endothelial area around other muscle fibers, except for type IIa fibers.

Author response:

Reviewer #1 (Public review):

Summary:

The authors have assembled a cohort of 10 SiNET, 1 SiAdeno, and 1 lung MiNEN samples to explore the biology of neuroendocrine neoplasms. They employ single-cell RNA sequencing to profile 5 samples (siAdeno, SiNETs 1-3, MiNEN) and single-nuclei RNA sequencing to profile seven frozen samples (SiNET 4-10).

They identify two subtypes of siNETs, characterized by either epithelial or neuronal NE cells, through a series of DE analyses. They also report findings of higher proliferation in non-malignant cell types across both subtypes. Additionally, they identify a potential progenitor cell population in a single-lung MiNEN sample.

Strengths:

Overall, this study adds interesting insights into this set of rare cancers that could be very informative for the cancer research community. The team probes an understudied cancer type and provides thoughtful investigations and observations that may have translational relevance.

Weaknesses:

The study could be improved by clarifying some of the technical approaches and aspects as currently presented, toward enhancing the support of the conclusions:

(1) Methods: As currently presented, it is possible that the separation of samples by program may be impacted by tissue source (fresh vs. frozen) and/or the associated sequencing modality (single cell vs. single nuclei). For instance, two (SiNET1 and SiNET2) of the three fresh tissues are categorized into the same subtype, while the third (SiNET9) has very few neuroendocrine cells. Additionally, samples from patient 1 (SiNET1 and SiNET6) are separated into different subtypes based on fresh and frozen tissue. The current text alludes to investigations (i.e.: "Technical effects (e.g., fresh vs. frozen samples) could also impact the capture of distinct cell types, although we did not observe a clear pattern of such bias."), but the study would be strengthened with more detail.

We thank the reviewer for the thoughtful and constructive review. Due to the difficulty in obtaining enough SiNET samples, we used two platforms to generate data - single cell analysis of fresh samples, and single nuclei analysis of frozen samples. We opted to combine both sample types in our analysis while being fully aware of the potential for batch effects. We therefore agree that this is a limitation of our work, and that differences between samples should be interpreted with caution.

Nevertheless, we argue that the two SiNET subtypes that we have identified are very unlikely to be due to such batch effect. First, the epithelial SiNET subtype was not only detected in two fresh samples but also in one frozen sample (albeit with relatively few cells, as the reviewer correctly noted). Second, and more importantly, the epithelial SiNET subtype was also identified in analysis of an external and much larger cohort of bulk RNA-seq SiNET samples that does not share the issue of two platforms (as seen in Fig. 2f). Moreover, the proportion of samples assigned to the two subtypes is similar between our data and the external data. We therefore argue that the identification of two SiNET subtypes cannot be explained by the use of two data platforms. However, we agree that the results should be further investigated and validated by future studies, as is often done in research on rare tumors.

The reviewer also commented that two samples from the same patient which were profiled by different platforms (SiNET1 and SiNET6) were separated into different subtypes. We would like to clarify that this is not the case, since SiNET6 was not included in the subtype analysis due to too few detected Neuroendocrine cells, and was not assigned to any subtype, as noted in the text and as can be seen by its exclusion from Figure 2 where subtypes are defined. We apologize that our manuscript may have gave the wrong impression about SiNET6 classification (it is labeled in Fig. 4a in a misleading manner). In the revised manuscript, we will correct the labeling in Fig. 4a and clarify that SiNET is not assigned to any subtype. We will further acknowledge the limitation of the two platforms and the arguments in favor of the existence of two SiNET subtypes.

(2) Results:<br /> Heterogeneity in the SiNET tumor microenvironment: It is unclear if the current analysis of intratumor heterogeneity distinguishes the subtypes. It may be informative if patterns of tumor microenvironment (TME) heterogeneity were identified between samples of the same subtype. The team could also evaluate this in an extension cohort of published SiNET tumors (i.e. revisiting additional analyses using the SiNET bulk RNAseq from Alvarez et al 2018, a subset of single-cell data from Hoffman et al 2023, or additional bulk RNAseq validation cohorts for this cancer type if they exist [if they do not, then this could be mentioned as a need in Discussion])

We agree that analysis of an independent cohort will assist in defining the association between TME and the SiNET subtype. However, the sample size required for that is significantly larger than the data available. In the revised manuscript we will note that as a direction for future studies.

(3) Proliferation of NE and immune cells in SiNETs: The observed proliferation of NE and immune cells in SiNETs may also be influenced by technical factors (including those noted above). For instance, prior studies have shown that scRNA-seq tends to capture a higher proportion of immune cells compared to snRNA-seq, which should be considered in the interpretation of these results. Could the team clarify this element?

We agree that different platforms could affect the observed proportions of immune cells, and more generally the proportions of specific cell types. However, the low proliferation of Neuroendocrine cells and the higher proliferation of immune cells (especially B cells, but also T cells and macrophages) is consistently observed in both platforms, as shown in Fig. 4a, and therefore appears to be reliable despite the limitations of our work. We will clarify this consistency in the revised manuscript. 

(4) Putative progenitors in mixed tumors: As written, the identification of putative progenitors in a single lung MiNEN sample feels somewhat disconnected from the rest of the study. These findings are interesting - are similar progenitor cell populations identified in SiNET samples? Recognizing that ideally additional validation is needed to confidently label and characterize these cells beyond gene expression data in this rare tumor, this limitation could be addressed in a revised Discussion.

We agree with this comment and will add the need for additional validation for this finding in the revised Discussion.

Reviewer #2 (Public review):

Summary:

The research identifies two main SiNET subtypes (epithelial-like and neuronal-like) and reveals heterogeneity in non-neuroendocrine cells within the tumor microenvironment. The study validates findings using external datasets and explores unexpected proliferation patterns. While it contributes to understanding SiNET oncogenic processes, the limited sample size and depth of analysis present challenges to the robustness of the conclusions.

Strengths:

The studies effectively identified two subtypes of SiNET based on epithelial and neuronal markers. Key findings include the low proliferation rates of neuroendocrine (NE) cells and the role of the tumor microenvironment (TME), such as the impact of Macrophage Migration Inhibitory Factor (MIF).

Weaknesses:

However, the analysis faces challenges such as a small sample size, lack of clear biological interpretation in some analyses, and concerns about batch effects and statistical significance.

Reviewer #3 (Public review):

Summary:

In this study, the authors set out to profile small intestine neuroendocrine tumors (siNETs) using single-cell/nucleus RNA sequencing, an established method to characterize the diversity of cell types and states in a tumor. Leveraging this dataset, they identified distinct malignant subtypes (epithelial-like versus neuronal-like) and characterized the proliferative index of malignant neuroendocrine cells versus non-malignant microenvironment cells. They found that malignant neuroendocrine cells were far less proliferative than some of their non-malignant counterparts (e.g., B cells, plasma cells, epithelial cells) and there was a strong subtype association such that epithelial-like siNETs were linked to high B/plasma cell proliferation, potentially mediated by MIF signaling, whereas neuronal-like siNETs were correlated with low B/plasma cell proliferation. The authors also examined a single case of a mixed lung tumor (neuroendocrine and squamous) and found evidence of intermediate/mixed and stem-like progenitor states that suggest the two differentiated tumor types may arise from the same progenitor.

Strengths:

The strengths of the paper include the unique dataset, which is the largest to date for siNETs, and the potentially clinically relevant hypotheses generated by their analysis of the data.

Weaknesses:

The weaknesses of the paper include the relatively small number of independent patients (n = 8 for siNETs), lack of direct comparison to other published single-cell NET datasets, mixing of two distinct methods (single-cell and single-nucleus RNA-seq), lack of direct cell-cell interaction analyses and spatially-resolved data, and lack of in vitro or in vivo functional validation of their findings.

The analytical methods applied in this study appear to be appropriate, but the methods used are fairly standard to the field of single-cell omics without significant methodological innovation. As the authors bring forth in the Discussion, the results of the study do raise several compelling questions related to the possibility of distinct biology underlying the epithelial-like and neuronal-like subtypes, the origin of mixed tumors, drivers of proliferation, and microenvironmental heterogeneity. However, this study was not able to further explore these questions through spatially-resolved data or functional experiments.

Author response:

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

Reviewer #1 (Public Review): 

Summary: 

The authors examined the salt-dependent phase separation of the low-complexity domain of hnRN-PA1 (A1-LCD). Using all-atom molecular dynamics simulations, they identified four distinct classes of salt dependence in the phase separation of intrinsically disordered proteins (IDPs), which can be predicted based on their amino acid composition. However, the simulations and analysis, in their current form, are inadequate and incomplete. 

Strengths: 

The authors attempt to unravel the mechanistic insights into the interplay between salt and protein phase separation, which is important given the complex behavior of salt effects on this process. Their effort to correlate the influence of salt on the low-complexity domain of hnRNPA1 (A1-LCD) with a range of other proteins known to undergo salt-dependent phase separation is an interesting and valuable topic. 

Weaknesses: 

(1) The simulations performed are not sufficiently long (Figure 2A) to accurately comment on phase separation behavior. The simulations do not appear to have converged well, indicating that the system has not reached a steady state, rendering the analysis of the trajectories unreliable.

We have extended the simulations for an additional 500 ns, to 1500 ns. The last 500 ns show reasonably good convergence (see Figure 2A).

(2) The majority of the data presented shows no significant alteration with changes in salt concentration. However, the authors have based conclusions and made significant comments regarding salt activities. The absence of error bars in the data representation raises questions about its reliability. Additionally, the manuscript lacks sufficient scientific details of the calculations.  

We have now included error bars. With the error bars, the salt dependences of all the calculated properties (exception for Rg) show a clear trend. Additionally, we have expanded the descriptions of our calculations (p. 15-16).

(3) In Figures 2B and 2C, the changes in the radius of gyration and the number of contacts do not display significant variations with changes in salt concentration. The change in the radius of gyration with salt concentration is less than 1 Å, and the number of contacts does not change by at least 1. The authors' conclusions based on these minor changes seem unfounded. 

The variation of ~ 1 Å for the calculated Rg is similar to the counterpart for the experimental Rg. As for the number of contacts, note that this property is presented on a per-residue basis, so a value of 1 means that each residue picks up one additional contact, or each protein chain gains a total of 131 contacts, when the salt concentration is increased from 50 to 1000 mM.

Reviewer #2 (Public Review): 

This is an interesting computational study addressing how salt affects the assembly of biomolecular condensates. The simulation data are valuable as they provide a degree of atomistic details regarding how small salt ions modulate interactions among intrinsically disordered proteins with charged residues, namely via Debye-like screening that weakens the effective electrostatic interactions among the polymers, or through bridging interactions that allow interactions between like charges from different polymer chains to become effectively attractive (as illustrated, e.g., by the radial distribution functions in Supplementary Information). However, this manuscript has several shortcomings: 

(i) Connotations of the manuscript notwithstanding, many of the authors' concepts about salt effects on biomolecular condensates have been put forth by theoretical models, at least back in 2020 and even earlier. Those earlier works afford extensive information such as considerations of salt concentrations inside and outside the condensate (tie-lines). But the authors do not appear to be aware of this body of prior works and therefore missed the opportunity to build on these previous advances and put the present work with its complementary advantages in structural details in the proper context.

(ii) There are significant experimental findings regarding salt effects on condensate formation [which have been modeled more recently] that predate the A1-LCD system (ref.19) addressed by the present manuscript. This information should be included, e.g., in Table 1, for sound scholarship and completeness. 

(iii) The strengths and limitations of the authors' approach vis-à-vis other theoretical approaches should be discussed with some degree of thoroughness (e.g., how the smallness of the authors' simulation system may affect the nature of the "phase transition" and the information that can be gathered regarding salt concentration inside vs. outside the "condensate" etc.). Accordingly, this manuscript should be revised to address the following. In particular, the discussion in the manuscript should be significantly expanded by including references mentioned below as well as other references pertinent to the issues raised. 

(1) The ability to use atomistic models to address the questions at hand is a strength of the present work. However, presumably because of the computational cost of such models, the "phase-separated" "condensates" in this manuscript are extremely small (only 8 chains). An inspection of Fig.1 indicates that while the high-salt configuration (snapshot, bottom right) is more compact and droplet-like than the low-salt configuration (top right), it is not clear that the 50 mM NaCl configuration can reasonably correspond to a dilute or homogeneous phase (without phase separation) or just a condensate with a lower protein concentration because the chains are still highly associated. One may argue that they become two droplets touching each other (the chains are not fully dispersed throughout the simulation box, unlike in typical coarse-grained simulations of biomolecular phase separation). While it may not be unfair to argue from this observation that the condensed phase is less stable at low salt, this raises critical questions about the adequacy of the approach as a stand-alone source of theoretical information. Accordingly, an informative discussion of the limitation of the authors' approach and comparisons with results from complementary approaches such as analytical theories and coarsegrained molecular dynamics will be instructive-even imperative, especially since such results exist in the literature (please see below). 

We now discuss the limitations of our all-atom simulations and also other approaches (p. 13; see below).

(2) The aforementioned limitation is reflected by the authors' choice of using Dmax as a sort of phase separation order parameter. However, no evidence was shown to indicate that Dmax exhibits a twostate-like distribution expected of phase separation. It is also not clear whether a Dmax value corresponding to the linear dimension of the simulation box was ever encountered in the authors' simulated trajectories such that the chains can be reliably considered to be essentially fully dispersed as would be expected for the dilute phase. Moreover, as the authors have noted in the second paragraph of the Results, the variation of Dmax with simulation time does not show a monotonic rank order with salt concentration. The authors' explanation is equivalent to stipulating that the simulation system has not fully equilibrated, inevitably casting doubt on at least some of the conclusions drawn from the simulation data. 

First off, with the extended simulations, the Dmax values converge to a tiered order rank, with successively decreasing values from low salt (50 mM) to intermediate salt (150 and 300 mM) to high salt (500 and 1000 mM). Secondly, as we now state (p. 13), our low-salt simulations mimic a homogenous solution whereas our high-salt simulations mimic the dense phase of a phase-separated system. The intermediate-salt simulations also mimic the dense phase but at a somewhat lower concentration (hence the intermediate Dmax value).

(3) With these limitations, is it realistic to estimate possible differences in salt concentration between the dilute and condensed phases in the present work? These features, including tie-lines, were shown to be amenable to analytical theory and coarse-grained molecular dynamics simulation (please see below).  

The differences in salt effects that we report do not represent those between two phases. Rather, as explained in the preceding reply, they represent differences between a homogenous solution at low salt and the dense phase at higher salt. We also acknowledge salt effects calculated by analytical theory and coarse-grained simulations (p. 13).

(4) In the comparison in Fig.2B between experimental and simulated radius of gyration as a function of [NaCl], there is an outlier among the simulated radii of gyration at [NaCl] ~ 250 mM. An explanation should be offered.  

After extending the simulations and analyzing the last 500 ns, the Rg data no longer show an outlier though still have some fluctuations from one salt concentration to another.

(5) The phenomenon of no phase separation at zero and low salt and phase separation at higher salt has been observed for the IDP Caprin1 and several of its mutants [Wong et al., J Am Chem Soc 142, 24712489 (2020) [https://pubs.acs.org/doi/full/10.1021/jacs.9b12208], see especially Fig.9 of this reference]. This work should be included in the discussion and added to Table 1. 

We now have added Caprin1 to Table 1 (new ref 26) and discuss this paper (p. 13).

(6) The authors stated in the Introduction that "A unifying understanding of how salt affects the phase separation of IDPs is still lacking". While it is definitely true that much remains to be learned about salt effects on IDP phase separation, the advances that have already been made regarding salt effects on IDP phase separation is more abundant than that conveyed by this narrative. For instance, an analytical theory termed rG-RPA was put forth in 2020 to provide a uniform (unified) treatment of salt, pH, and sequence-charge-pattern effects on polyampholytes and polyelectrolytes (corresponding to the authors' low net charge and high net charge cases). This theory offers a means to predict salt-IDP tie-lines and a comprehensive account of salt effect on polyelectrolytes resulting in a lack of phase separation at extremely low salt and subsequent salt-enhanced phase separation (similar to the case the authors studied here) and in some cases re-entrant phase separation or dissolution [Lin et al., J Chem Phys 152. 045102 (2020) [https://doi.org/10.1063/1.5139661]]. This work is highly relevant and it already provided a conceptual framework for the authors' atomistic results and subsequent discussion. As such, it should definitely be a part of the authors' discussion. 

We now cite this paper (new ref 34) in Introduction (p. 4). We also discuss its results for Caprin1 (new ref 18; p. 13).

(7) Bridging interactions by small ions resulting in effective attractive interactions among polyelectrolytes leading to their phase separation have been demonstrated computationally by Orkoulas et al., Phys Rev Lett 90, 048303 (2003) [https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.90.048303]. This result should also be included in the discussion. 

We now cite this paper (new ref 41; p. 11).

(8) More recently, the salt-dependent phase separations of Caprin1, its RtoK variants and phosphorylated variant (see item #5 above) were modeled (and rationalized) quite comprehensively using rG-RPA, field-theoretic simulation, and coarse-grained molecular dynamics [Lin et al., arXiv:2401.04873 [https://arxiv.org/abs/2401.04873]], providing additional data supporting a conceptual perspective put forth in Lin et al. J Chem Phys 2020 (e.g., salt-IDP tie-lines, bridging interactions, reentrance behaviors etc.) as well as in the authors' current manuscript. It will be very helpful to the readers of eLife to include this preprint in the authors' discussion, perhaps as per the authors' discretion along the manner in which other preprints are referenced and discussed in the current version of the manuscript. 

We now cite this paper (new ref 18) and discuss it along with new ref 26 in Discussion (p. 13).

Reviewer #3 (Public Review): 

Summary: 

This study investigates the salt-dependent phase separation of A1-LCD, an intrinsically disordered region of hnRNPA1 implicated in neurodegenerative diseases. The authors employ all-atom molecular dynamics (MD) simulations to elucidate the molecular mechanisms by which salt influences A1-LCD phase separation. Contrary to typical intrinsically disordered protein (IDP) behavior, A1-LCD phase separation is enhanced by NaCl concentrations above 100 mM. The authors identify two direct effects of salt: neutralization of the protein's net charge and bridging between protein chains, both promoting condensation. They also uncover an indirect effect, where high salt concentrations strengthen pi-type interactions by reducing water availability. These findings provide a detailed molecular picture of the complex interplay between electrostatic interactions, ion binding, and hydration in IDP phase separation. 

Strengths: 

Novel Insight: The study challenges the prevailing view that salt generally suppresses IDP phase separation, highlighting A1-LCD's unique behavior. 

Rigorous Methodology: The authors utilize all-atom MD simulations, a powerful computational tool, to investigate the molecular details of salt-protein interactions. 

Comprehensive Analysis: The study systematically explores a wide range of salt concentrations, revealing a nuanced picture of salt effects on phase separation. 

Clear Presentation: The manuscript is well-written and logically structured, making the findings accessible to a broad audience. 

Weaknesses: 

Limited Scope: The study focuses solely on the truncated A1-LCD, omitting simulations of the full-length protein. This limitation reduces the study's comparative value, as the authors note that the full-length protein exhibits typical salt-dependent behavior. A comparative analysis would strengthen the manuscript's conclusions and broaden its impact.

Perhaps we did not impress on the reviewer how expensive the all-atom MD simulations on A1-LCD were: the systems each contained half a million atoms and the simulations took many months to complete. That said, we agree with the reviewer that, ideally, a comparative study on a protein showing the typical screening class of salt dependence would have made our work more complete. However, we are confident of the conclusions for several reasons. First, the three salt effects – charge neutralization, bridging, and strengthening of pi-types of interactions – revealed by the all-atom simulations are physically sound and well-supported by other studies. Second, these effects led us to develop a unified picture for the salt dependence of homotypic phase separation, in the form of a predictor for the classes of salt dependence based on amino-acid composition. This predictor works well for nearly 30 proteins. Third, recent studies using analytical theory and coarse-grained simulations (new ref 18) also strongly support our conclusions.

Reviewer #1 (Recommendations For The Authors): 

(1) In Figure 1, the color scheme should be updated and the figure remade, as the current set of color choices makes it very difficult to distinguish the magenta spheres.  

We have increased the sizes of ions in Figure 1 to make them distinguishable.

(2) Within the framework of atomistic simulations, the influence of salt concentration alteration on protein conformational plasticity is worth investigating. This could be correlated (with proper details) with the effect of salt-concentration-modulated protein aggregation behavior. 

We now use RMSF to measure conformational plasticity, which shows a clear salt-dependent trend with a 27% reduction in fluctuations from 50 mM to 1000 mM NaCl (new Fig. S1).

(3) The authors should mention the protein concentrations employed in the simulations and whether these are consistent with experimentally used concentrations.  

We have mentioned the initial concentration (3.5 mM). We now further state that this concentration is maintained in the low-salt simulations, indicating absence of phase separation, but is increased to 23 mM in the high-salt simulations, indicating phase separation. The latter value is consistent with the measured concentrations in the dense phase (last two paragraphs of p. 5).

(4) It would be useful to test the salt effect for at least two extreme salt concentrations at various protein concentrations, consistent with experimental protein concentration ranges.  

In simulation studies of short peptides (ref 37), we have shown that the initial concentration does not affect the final concentration in the dense phase, as expected for phase-separation systems. We expect that the same will be true for the A1-LCD system at intermediate and high salt where phase separation occurs. Though this expectation could be tested by simulations at a different initial protein concentration, such simulations would be expensive but unlikely to yield new physical insight.

(5) Importantly, the simulations do not appear to have converged well enough (Figure 2A). The authors should extend the simulation trajectories to ensure the system has reached a steady state.  

We extended the simulations for an additional 500 ns, which now appear to show convergence. In Figure 2A we now see Dmax values converge to a tiered order rank, with successively decreasing values from low salt (50 mM) to intermediate salt (150 and 300 mM) to high salt (500 and 1000 mM). 

(6) The authors mention "phase separation" in the title, but with only a 1 μs simulation trajectory, it is not possible to simulate a phenomenon like phase separation accurately. Since atomistic simulations cannot realistically capture phase separation on this timescale, a coarse-grained approach is more suitable. To properly explore salt effects in the context of phase separation, long timescale simulation trajectories should be considered. Otherwise, the data remain unreliable. 

Our all-atom simulations revealed rich salt effects that might have been missed in coarse-grained simulations. It is true that coarse-grained models allow the simulations of the phase separation process, but as we have recently demonstrated (refs 36 and 37), all-atom simulations on the μs timescale are also able to capture the spontaneous phase separation of peptides and small IDPs. A1-LCD is much larger than those systems, so we had to use a relatively small chain number (8 chains here vs 64 used in ref 37 and 16 used in ref 37). S2ll, we observe the condensation into a dense phase at high salt. We discuss the pros and cons of all-atom vs. coarse-grained simulations in p. 13.

(7) In Figure 5E, the plot does not show that g(r) has reached 1. If it does, the authors should show the full curve. The same issue remains with supplementary figures 1, 2, 3, etc.  

We now show the approach to 1 in the insets of Figs. S2, S3, S4, and 5E.

(8) None of the data is represented with error bars. The authors should include error bars in their data representations. 

We have now included error bars in all graphs that report average values.

(9) The authors state that "the net charge of the system reduces to only +8 at 1000 mM NaCl (Figure 3C)" but do not explain how this was calculated. 

We now add this explanation in methods (p. 16).

(10). The authors mention "similar to the role played by ATP molecules in driving phase separation of positively charged IDPs." However, ATP can inhibit aggregation, and its induction of phase separation is concentration-dependent. Given ATP's large aromatic moiety, its comparison to ions is not straightforward and is more complex. This comparison can be at best avoided. 

In this context we are comparing the bridging capability of ATP molecules in driving phase separation of positively charged IDPs in ref 36 to the bridging capability of the ions here. In ref 36 the authors show ATP bridging interactions between protein chains similar to what we show here with ions.

(11) Many calculations are vaguely represented. The process for calculating the number of bridging ions, for example, is not well documented. The authors should provide sufficient details to allow for the reproducibility of the data. 

We have now expanded the methods section to include more detailed information on calculations done.

Reviewer #3 (Recommendations For The Authors): 

Include error bars or standard deviations for all results averaged over four replicates, particularly for the number of ions and contacts per residue. This would provide a clearer picture of the data's reliability and variability. 

We have now included error bars in all graphs that report averaged values.

Strengthen the support for the conclusion that "each Arg sidechain often coordinates two Cl- ions, multiple backbone carbonyls often coordinate a single Na+ ion." While Fig. 3A clearly demonstrates ArgCl- coordination, the Na+ coordination claim for a 131-residue protein requires further clarification. Consider including the integration profile of radial distribution functions for Na+ ions to bolster this assertion. 

We now report the number of Na+ ions that coordinate with multiple backbone carbonyls (p. 7) as well as the number of Na+ ions that bridge between A1-LCD chains via coordination with multiple backbone carbonyls (p. 9). Please note that Figure 4A right panel displays an example of Na+ coordinating with multiple backbone carbonyls.

Address the following typographical errors in the main text: o Page 11, line 25: "distinct classes of sat dependence" should be "distinct classes of salt dependence" o Page 14, line 9: "for Cl- and 3.0 and 5.4 A" should be "for Cl- and 3.0 and 5.4 √Ö" o Page 14, line 18: "As a control, PRDFs for water were also calculated" should be "As a control, RDFs for water were also calculated" (assuming PRDF was meant to be RDF) 

We have now corrected these typos.

Consider expanding the study to include simulations of the full-length protein to provide a more comprehensive comparison between the truncated A1-LCD and the complete protein's behavior in various salt concentrations. 

As we explained above, even with eight chains of A1-LCD, which has 131 residues, the systems already contain half a million atoms each and the all-atom simulations took many months to complete. Full-length A1 has 314 residues so a multi-chain system would be too large to be feasible for all-atom simulations.

Author response:

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

Reviewer #1 (Public review):

Summary:

Crosslinking mass spectrometry has become an important tool in structural biology, providing information about protein complex architecture, binding sites and interfaces, and conformational changes. One key challenge of this approach represents the quantitation of crosslinking data to interrogate differential binding states and distributions of conformational states.

Here, Luo and Ranish present a novel class of isobaric crosslinkers ("Qlinkers"), conduct proof-of-concept benchmarking experiments on known protein complexes, and show example applications on selected target proteins. The data are solid and this could well be an exciting, convincing new approach in the field if the quantitation strategy is made more comprehensive and the quantitative power of isobaric labeling is fully leveraged as outlined below. It's a promising proof-of-concept, and potentially of broad interest for structural biologists.

Strengths:

The authors demonstrate the synthesis, application, and quantitation of their "Q2linkers", enabling relative quantitation of two conditions against each other. In benchmarking experiments, the Q2linkers provide accurate quantitation in mixing experiments. Then the authors show applications of Q2linkers on MBP, Calmodulin, selected transcription factors, and polymerase II, investigating protein binding, complex assembly, and conformational dynamics of the respective target proteins. For known interactions, their findings are in line with previous studies, and they show some interesting data for TFIIA/TBP/TFIIB complex formation and conformational changes in pol II upon Rbp4/7 binding.

Weaknesses:

This is an elegant approach but the power of isobaric mass tags is not fully leveraged in the current manuscript.

First, "only" Q2linkers are used. This means only two conditions can be compared. Theoretically, higher-plexed Qlinkers should be accessible and would also be needed to make this a competitive method against other crosslinking quantitation strategies. As it is, two conditions can still be compared relatively easily using LFQ - or stable-isotope-labeling based approaches. A "Q5linker" would be a really useful crosslinker, which would open up comprehensive quantitative XLMS studies.

We agree that a multiplexed Qlinker approach would be very useful. The multiplexed Qlinkers are more difficult and more expensive to synthesize. We are currently working on different schemes for synthesizing multiplexed Qlinkers.

Second, the true power of isobaric labeling, accurate quantitation across multiple samples in a single run, is not fully exploited here. The authors only show differential trends for their interaction partners or different conformational states and do not make full quantitative use of their data or conduct statistical analyses. This should be investigated in more detail, e.g. examine Qlinker quantitation of MBP incubated with different concentrations of maltose or Calmodulin incubated with different concentrations of CBPs. Does Qlinker quantitation match ratios predicted using known binding constants or conformational state populations? Is it possible to extract ratios of protein populations in different conformations, assembly, or ligand-bound states?

With these two points addressed this approach could be an important and convincing tool for structural biologists.

We agree that multiplexed Qlinkers would open the door to exciting avenues of investigation such as studying conformational state populations.  We plan to conduct the suggested experiments when multiplexed Qlinkers are available.

Reviewer #2 (Public review):

The regulation of protein function heavily relies on the dynamic changes in the shape and structure of proteins and their complexes. These changes are widespread and crucial. However, examining such alterations presents significant challenges, particularly when dealing with large protein complexes in conditions that mimic the natural cellular environment. Therefore, much emphasis has been put on developing novel methods to study protein structure, interactions, and dynamics. Crosslinking mass spectrometry (CSMS) has established itself as such a prominent tool in recent years. However, doing this in a quantitative manner to compare structural changes between conditions has proven to be challenging due to several technical difficulties during sample preparation. Luo and Ranish introduce a novel set of isobaric labeling reagents, called Qlinkers, to allow for a more straightforward and reliable way to detect structural changes between conditions by quantitative CSMS (qCSMS).

The authors do an excellent job describing the design choices of the isobaric crosslinkers and how they have been optimized to allow for efficient intra- and inter-protein crosslinking to provide relevant structural information. Next, they do a series of experiments to provide compelling evidence that the Qlinker strategy is well suited to detect structural changes between conditions by qCSMS. First, they confirm the quantitative power of the novel-developed isobaric crosslinkers by a controlled mixing experiment. Then they show that they can indeed recover known structural changes in a set of purified proteins (complexes) - starting with single subunit proteins up to a very large 0.5 MDa multi-subunit protein complex - the polII complex.

The authors give a very measured and fair assessment of this novel isobaric crosslinker and its potential power to contribute to the study of protein structure changes. They show that indeed their novel strategy picks up expected structural changes, changes in surface exposure of certain protein domains, changes within a single protein subunit but also changes in protein-protein interactions. However, they also point out that not all expected dynamic changes are captured and that there is still considerable room for improvement (many not limited to this crosslinker specifically but many crosslinkers used for CSMS).

Taken together the study presents a novel set of isobaric crosslinkers that indeed open up the opportunity to provide better qCSMS data, which will enable researchers to study dynamic changes in the shape and structure of proteins and their complexes. However, in its current form, the study some aspects of the study should be expanded upon in order for the research community to assess the true power of these isobaric crosslinkers. Specifically:

Although the authors do mention some of the current weaknesses of their isobaric crosslinkers and qCSMS in general, more detail would be extremely helpful. Throughout the article a few key numbers (or even discussions) that would allow one to better evaluate the sensitivity (and the applicability) of the method are missing. This includes:

(1) Throughout all the performed experiments it would be helpful to provide information on how many peptides are identified per experiment and how many have actually a crosslinker attached to it.

As the goal of the experiments is to maximize identification of crosslinked peptides which tend to have higher charge states, we targeted ions with charge states of 3+ or higher in our MS acquisition settings for CLMS, and ignored ions with 2+ charge states, which correspond to many of the normal (i.e., not crosslinked) peptides that are identified by MS. As a result, normal peptides are less likely to be identified by the MS procedure used in our CLMS experiments compared to MS settings typically used to identify normal peptides. Our settings may also fail to identify some mono-modified peptides. Like most other CLMS methods, the total number of identified crosslinked peptide spectra is usually less than 1% of the total acquired spectra and we normally expect the crosslinked species to be approximately 1% of the total peptides. 

We added information about the number of crosslinked and monolinked peptides identified in the pol I benchmarking experiments (line 173).  The number of crosslinks and monolinks identified in the pol II +/- a-amanitin experiment, the TBP/TFIIA/TFIIB experiment and the pol II experiment +/- Rpb4/7 are also provided.

(2) Of all the potential lysines that can be modified - how many are actually modified? Do the authors have an estimate for that? It would be interesting to evaluate in a denatured sample the modification efficiency of the isobaric crosslinker (as an upper limit as here all lysines should be accessible) and then also in a native sample. For example, in the MBP experiment, the authors report the change of one mono-linked peptide in samples containing maltose relative to the one not containing maltose. The authors then give a great description of why this fits to known structural changes. What is missing here is a bit of what changes were expected overall and which ones the authors would have expected to pick up with their method and why have they not been picked up. For example, were they picked up as modified by the crosslinker but not differential? I think this is important to discuss appropriately throughout the manuscript to help the reader evaluate/estimate the potential sensitivity of the method. There are passages where the authors do an excellent job doing that - for example when they mention the missed site that they expected to see in the initial the pol II experiments (lines 191 to 207). This kind of "power analysis" should be heavily discussed throughout the manuscript so that the reader is better informed of what sensitivity can be expected from applying this method.

Regarding the Pol II complex experiment described in Figures 4 and 5, out of the 277 lysine residues in the complex, 207 were identified as monolinked residues (74.7%), and 817 crosslinked pairs out of 38,226 potential pairs (2.1%) were observed. The ability of CLMS to detect proximity/reactivity changes may be impacted by several factors including 1) the (low) abundance of crosslinked peptides in complex mixtures, 2) the presence of crosslinkable residues in close proximity with appropriate orientation, and 3) the ability to generate crosslinked peptides by enzymatic digestion that are amenable to MS analysis (i.e., the peptides have appropriate m/z’s and charge states, the peptides ionize well, the peptides produce sufficient fragment ions during MS2 analysis to allow confident identification). Future efforts to enrich crosslinked peptides prior to MS analysis may improve sensitivity.

It is very difficult to estimate the modification efficiency of Qlinker (or many other crosslinkers) based on peptide identification results. One major reason for this is that trypsin is not able to cleave after a crosslinker-modified lysine residue.  As a result, the peptides generated after the modification reaction have different lengths, compositions, charge states, and ionization efficiencies compared to unmodified peptides. These differences make it very difficult to estimate the modification efficiencies based on the presence/absence of certain peptide ions, and/or the intensities of the modified and unmodified versions of a peptide. Also, 2+ ions which correspond to many normal (i.e., unmodified) peptides were excluded by our MS acquisition settings.

It is also very difficult to predict which structural changes are expected and which crosslinked peptides and/or modified peptides can be observed by MS.  This is especially true when the experiment involves proteins containing unstructured regions such as the experiments involving Pol II, and TBP, TFIIA and TFIIB. Since we are at the early stages of using qCLMS to study structural changes, we are not sure which changes we can expect to observe by qCLMS. Additional applications of Qlinker-CLMS are needed to better understand the types of structural changes that can be studied using the approach.

We hope that our discussions of some the limitations of CLMS for detecting conformational/reactivity changes provide the reader with an understanding of the sensitivity that can be expected with the approach.  At the end of the paragraph about the pol II a-amanitin experiment we say, “Unfortunately, no Q2linker-modified peptides were identified near the site where α-amanitin binds. This experiment also highlights one of the limitations of residue-specific, quantitative CLMS methods in general. Reactive residues must be available near the region of interest, and the modified peptides must be identifiable by mass spectrometry.” In the section about Rbp4/7-induced structural changes in pol II we describe the under-sampling issue. And in the last paragraph we reiterate these limitations and say, “This implies that this strategy, like all MS-based strategies, can only be used for interpretation of positively identified crosslinks or monolinks. Sensitivity and under sampling are common problems for MS analysis of complex samples.”

(3) It would be very helpful to provide information on how much better (or not) the Qlinker approach works relative to label-free qCLMS. One is missing the reference to a potential qCLMS gold standard (data set) or if such a dataset is not readily available, maybe one of the experiments could be performed by label-free qCLMS. For example, one of the differential biosensor experiments would have been well suited.

We agree with the reviewer that it will be very helpful to establish gold standard datasets for CLMS. As we further develop and promote this technology, we will try to establish a standardized qCLMS.

Reviewer #1 (Recommendations for the authors):

Only a very minor point:

I may have missed it but it's not really clear how many independent experiments were used for the benchmarking quantitation and mixing experiments for Figure 1. What is the reproducibility across experiments on average and on a per-peptide basis?

Otherwise, I think the approach would really benefit from at least "Q5linkers" or even "Q10linkers", if possible. And then conduct detailed quantitative studies, either using dilution series or maybe investigating the kinetics of complex formation.

We used a sample of BSA crosslinked peptides to optimize the MS settings, establish the MS acquisition strategies and test the quantification schemes.  The data in Figure 1 is based on one experiment, in which used ~150 ug of purified pol I complexes from a 6 L culture. We added this information to the Figure 1 legend. We also provide information about the reproducibility of peptide quantification by plotting the observed and expected ratios for each monolinked and crosslinked peptide identified in all of the runs in Figure S3.

We agree with the reviewer that the Qlinker approach would be even more attractive if multiplex Qlinker reagents were designed. The multiplexed Qlinkers are more difficult and more expensive to synthesize. We are currently working on different schemes for synthesizing multiplexed Qlinkers.

Reviewer #2 (Recommendations for the authors):

In addition to the public review I have the following recommendations/questions:

(1) The first part of the results section where the synthesis of the crosslinker is explained is excellent for mass spec specialists, but problematic for general readers - either more info should be provided (e.g. b1+ ions - most readers will have no idea why that is) - or potentially it could be simplified here and the details shifted to Materials and Methods for the expert reader. The same is true below for the length of spacer arms.

However - in general this level of detail is great - but can impact the ease of understanding for the more mass spec affine but not expert reader.

We have added the following sentence to assist the general reader: A b1+ ion is an ion with a charge state of +1 corresponding to the first N-terminal amino acid residue after breakage of the first peptide bond (lines 126-128).

(2) The Calmodulin experiment (lines 239 to 257) - it is a very nice result that they see the change in the crosslinked peptide between residues K78-K95, but the monolinks are not just detected as described in the text but actually go 2 fold up. This would have been actually a bit expected if the residues are now too far away to be still crosslinked that the monolinks increase. In this case, this counteraction of monolinks to crosslinked sites can also be potentially used as a "selection criteria" for interesting sites that change. Is that a possible interpretation or do the authors think that upregulation of the monolinks is a coincidence and should not be interpreted?

We agree with the reviewer that both monolinks and crosslinks can be used as potential indicators for some changes. However, it is much more difficult to interpret the abundance information from monolinks because, unlike crosslinks, there is little associated structural/proximity information with monolinks. Because it is difficult to understand the reason(s) for changes in monolink abundance, we concentrate on changes in crosslink abundances, which provide proximity/structural information about the crosslinked residues.

(3) Lines 267 to 274: a small thing but the structural information provided is quite dense I have to say. Maybe simplify or accompany with some supplemental figures?

We agree that the structural information is a bit dense especially for readers who are not familiar with the pol II system.  We added a reference to Figure 3c (line 177) to help the reader follow the structural information. 

As qCLMS is still a relatively new approach for studying conformational changes, the utility of the approach for studying different types of conformational changes is still unclear. Thus, one of the goals of the experiments is to demonstrate the types of conformational changes that can be detected by Q2linkers.  We hope that the detailed descriptions will help structural biologists understand the types of conformational changes that can be detected using Qlinkers.

(4) Line 280: explain maybe why the sample was fractionated by SCX (I guess to separate the different complexes?).

SCX was used to reduce the complexity of the peptide mixtures. As the samples are complex and crosslinked peptides are of low abundance compared to normal peptides, SCX can separate the peptides based on their positive charges.  Larger peptides and peptides with higher charge states, such as crosslinked peptides, tend to elute at higher salt concentration during SCX chromatography.  The use of SCX to fractionate complex peptide mixtures is described in the “General crosslinking protocol and workflow optimization” section of the Methods, and we added a sentence to explain why the sample was fractionated by SCX (lines 278-279).

(5) Lines 354 to 357: "This suggests that the inability to identity most of these crosslinked peptides in both experiments is mainly due to under-sampling during mass spectrometry analysis of the complex samples, rather than the absence of the crosslinked peptides in one of the experiments."

This is an extremely important point for the interpretation of missing values - have the authors tried to also collect the mass spec data with DIA which is better in recovery of the same peptide signals between different samples? I realize that these are isobaric samples so DIA measurements per se are not useful as the quantification is done on the reporter channels in the MS2, but it would at least give a better idea if the missing signals were simply not picked up for MS2 as claimed by the authors or the modified peptides are just not present. Another possibility is for the authors to at least try to use a "match between the run" function as can be done in Maxquant. One of the strengths of the method is that it is quantitative and two states are analyzed together, but as can be seen in this experiment, more than two states might want to be compared. In such cases, the under-sampling issue (if that is indeed the cause) makes interpretation of many sites hard (due to missing values) and it would be interesting if for example, an analysis approach with a "match between the runs" function could recover some of the missing values.

We agree that undersampling/missing values is an important issue that needs to be addressed more thoroughly. This also highlights the importance of qCLMS, as conclusions about structural changes based on the presence/absence of certain crosslinked species in database search results may be misleading if the absence of a species is due to under-sampling. We have not tried to collect the data with DIA since we would lose the quantitative information. It would be interesting to see if match between runs can recover some of the missing values. While this could provide evidence to support the under-sampling hypothesis, it would not recover the quantitative information.

We recommend performing label swap experiments and focusing downstream analysis on the crosslinks/monolinks that are identified on both experiments. Future development of multiplexed Qlinker reagents should help to alleviate under-sampling issues. See response to Reviewer #1.

(6) Lines 375 to 393 (the whole paragraph): extremely detailed and not easy to follow. Is that level of detail necessary to drive home that point or could it be visualized in enough detail to help follow the text?

We agree that the paragraph is quite detailed, but we feel that the level of detailed is necessary to describe the types of conformational changes that can be detected by the quantitative crosslinking data, and also illustrate the challenges of interpreting the structural basis for some crosslink abundance changes even when high resolution structural data exists.

To make it easier to follow, we added a sentence to the legend of Figure 5b. “In the holo-pol II structure (right), Switch 5 bending pulls Rpb1:D1442 away from K15, breaking the salt bridge that is formed in the core pol II structure (left). The increase in the abundances of the Rpb1:15-Rpb6:76 and Rpb1:15-Rpb6:72 crosslinks in holo-pol II is likely attributed to the salt bridge between K15 and D1442 in core pol II which impedes the NHS ester-based reaction between the epsilon amino group of K15 and the crosslinker.”

(7) Final paragraph in the results section - lines 397 and 398: "All of the intralinks involving Rpb4 are more abundant in holo-pol II as expected." If I understand that experiment correctly the intralinks with Rpb4 should not be present at all as Rpb4 has been deleted. Is that due to interference between the 126 and 127 channels in MS2? If so, then this also sets a bit of the upper limit of quantitative differences that can be seen. The authors should at least comment on that "limitation".

Yes, we shouldn’t detect any Rpb4 peptides in the sample derived from the Rpb4 knockout strain. The signal from Rpb4 peptides in the DRpb4 sample is likely due to co-eluting ions. To clarify, we changed the text to:

All of the intralinks involving Rpb4 are more abundant in the holo-pol II sample (even though we don’t expect any reporter ion signal from Rpb4 peptides derived from the ∆Rpb4 pol II sample, we still observed reporter ion signals from the channel corresponding to the DRpb4 sample, potentially due to the presence of low abundance, co-eluting ions)(lines 395-399).

(8) Materials and Methods - line 690: I am probably missing something but why were two different mass additions to lysine added to the search (I would have expected only one for the crosslinker)?

The 297 Da modification is for monolinked peptides with one end of the crosslinker hydrolyzed and 18 Da water molecule is added. The 279 Da modification is for crosslinks and sometimes for looplinks (crosslinks involving two lysine residues on the same tryptic peptide).

Author response:

Review #1:

Also, they observed no difference in the binding free energy of phosphatidylserine with wild TREM2-Ig and mutant TREM2-Ig, which is a bit inconsistent with the previous report with experiment studies by Journal of Biological Chemistry 293, (2018), Alzheimer's and Dementia 17, 475-488 (2021), Cell 160, 1061-1071 (2015).

We directly note this contrast with experimental findings in the body of our work, particularly given the known limitations of free energy calculations in MD simulations, as outlined in the Limitations section. Our claim is that the loss of function in the R47H variant extends beyond decreased binding affinities and also impacts binding patterns. As stated in our manuscript: ‘Our observations for both sTREM2 and TREM2 indicate that R47H-induced dysfunction may result not only from diminished ligand binding but also an impaired ability to discriminate between different ligands in the brain, proposing a novel mechanism for loss-of-function.’

Perhaps the authors made significant efforts to run a number of simulations for multiple models, which is nearly 17 microseconds in total; none of the simulations has been repeated independently at least a couple of times, which makes me uncomfortable to consider this finding technically true. Most of the important conclusions that authors claimed, including the opposite results from previous research, have been made on the single run, which raises the question of whether this observation can be reproduced if the simulation has been repeated independently. Although the authors stated the sampling number and length of MD simulations in the current manuscript as a limitation of this study, it must be carefully considered before concluding rather than based on a single run.

The reviewer raises an interesting point regarding the repetition of individual simulations, a consideration we carefully evaluated during the design of this study. However, we believe our approach—running multiple independent models of the same system—offers a more rigorous methodology than simply repeating simulations of the same docked model. This strategy allows us to sample several distinct starting configurations, thereby minimizing biases introduced by docking algorithms and single-model reliance.

In our study, we demonstrate that within the 150 ns timescale of our protein/ligand (PL) simulations, the relatively small ligands are able to move from their initial docking positions to a specific binding site. While ideally, replicates of these independent models would further strengthen the findings, this was not computationally feasible given the unprecedented total duration of our simulations. Importantly, our conclusions are seldom based on the results of a single protein/PL simulation.

Moreover, the ergodic hypothesis suggests that over sufficiently long timescales, simulations will explore all accessible states. Additionally, we have performed several replicate simulations of our WT and R47H Ig-like domain models in solution, specifically to investigate CDR2 loop dynamics.

In this case, since the system involves only the protein and lacks the independent replicates seen in the protein/PL simulations, these runs were chosen to effectively capture the stochastic nature of CDR2 loop movement.

sTREM2 shows a neuroprotective effect in AD, even with the mutations with R47H, as evidenced by authors based on their simulation. sTREM2 is known to bind Aβ within the AD and reduce Aβ aggregation, whereas R47H mutant increases Aβ aggregation. I wonder why the authors did not consider Aβ as a ligand for their simulation studies. As a reader in this field, I would prefer to know the protective mechanism of sTREM2 in Aβ aggregation influenced by the stalk domain.

Our initial approach for this study used Aβ as a ligand rather than phospholipids. However, we noted the difficulties in simulating Aβ, particularly in choosing relevant Aβ structures and oligomeric states (n-mers). We believe that phospholipids represent an equally pertinent ligand for TREM2, given its critical role in lipid sensing and metabolism. Furthermore, there is growing recognition in the AD research community of the need to move beyond Aβ and focus on other understudied pathological mechanisms.

In a similar manner, why only one mutation is considered "R47H" for the study? There are more server mutations reported to disrupt tethering between these CDRs, such as T66M. Although this "T66M" is not associated with AD, I guess the stalk domain protective mechanism would not be biased among different diseases. Therefore, it would be interesting to see whether the findings are true for this T66M.

In most previous studies, the mechanism for CDR destabilization by mutant was explored, like the change of secondary structures and residue-wise interloop interaction pattern. While this is not considered in this manuscript, neither detailed residue-wise interaction that changed by mutant or important for 'ligand binding" or "stalk domain".

These are both excellent points that deserve extensive investigation. While R47H is the most common and prolific mutation in literature, an extensive catalog of other mutations is important to explore. We are currently preparing two separate publications that will delve into these gaps in more detail, as addressing them was beyond the scope of the present study.

The comparison between the wild and mutant and other different complex structures must be determined by particular statistical calculations to state the observed difference between different structures is significant. Since autocorrelation is one of the major concerns for MD simulation data for predicting statistical differences, authors can consider bootstrap calculations for predicting statistical significance.

We are currently working to address this comment to strengthen the validity of our results and statistical conclusions in the revised manuscript.  

Review #2:

The authors state that reported differences in ligand binding between the TREM2 and sTREM2 remain unexplained, and the authors cite two lines of evidence. The first line of evidence, which is true, is that there are differences between lipid binding assays and lipid signaling assays. However, signaling assays do not directly measure binding. Secondly, the authors cite Kober et al 2021 as evidence that sTREM2 and TREM2 showed different affinities for Abeta1-42 in a direct binding assay. Unfortunately, when Kober et al measured the binding of sTREM2 and Ig-TREM2 to Abeta they reported statistically identical affinities (Kd = 3.8 {plus minus} 2.9 µM vs 5.1 {plus minus} 3.7 µM) and concluded that the stalk did not contribute measurably to Abeta binding.

We appreciate the reviewer’s insight and acknowledge the need to clarify our interpretation of Kober et al. (2021). We will adjust and refocus how we reference this evidence from Kober et al. in our revised manuscript. 

In line with these findings, our energy calculations reveal that sTREM2 exhibits weaker—but still not statistically significant—binding affinities for phospholipids compared to TREM2. These results suggest that while overall binding affinity might be similar, differences in binding patterns or specific lipid interactions could still contribute to functional differences observed between TREM2 and sTREM2.

The authors appear to take simulations of the Ig domain (without any stalk) as a surrogate for the full-length, membrane-bound TREM2. They compare the Ig domain to a sTREM2 model that includes the stalk. While it is fully plausible that the stalk could interact with and stabilize the Ig domain, the authors need to demonstrate why the full-length TREM2 could not interact with its own stalk and why the isolated Ig domain is a suitable surrogate for this state.

We believe that this is a major limitation of all computational work of TREM2 to-date, and of experimental work which only presents the Ig-like domain. This is extensively discussed in the limitations section of our paper. Hence, we are currently working toward a manuscript that will be the first biologically relevant model of TREM2 in a membrane and will challenge the current paradigm of using the Ig-like domain as an experimental surrogate for TREM2.

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

PPARgamma is a nuclear receptor that binds to orthosteric ligands to coordinate transcriptional programs that are critical for adipocyte biogenesis and insulin sensitivity. Consequently, it is a critical therapeutic target for many diseases, but especially diabetes. The malleable nature and promiscuity of the PPARgamma orthosteric ligand binding pocket have confounded the development of improved therapeutic modulators. Covalent inhibitors have been developed but they show unanticipated mechanisms of action depending on which orthosteric ligands are present. In this work, Shang and Kojetin present a compelling and comprehensive structural, biochemical, and biophysical analysis that shows how covalent and noncovalent ligands can co-occupy the PPARgamma ligand binding pocket to elicit distinctive preferences of coactivator and corepressor proteins. Importantly, this work shows how the covalent inhibitors GW9662 and T0070907 may be unreliable tools as pan-PPARgamma inhibitors despite their widespread use.

Strengths:

- Highly detailed structure and functional analyses provide a comprehensive structure-based hypothesis for the relationship between PPARgamma ligand binding domain co-occupancy and allosteric mechanisms of action. - Multiple orthogonal approaches are used to provide high-resolution information on ligand binding poses and protein dynamics.

- The large number of x-ray crystal structures solved for this manuscript should be applauded along with their rigorous validation and interpretation.

Weaknesses

- Inclusion of statistical analysis is missing in several places in the text. - Functional analysis beyond coregulator binding is needed.

We added additional statistical analyses as recommended (Source Data 1, a Microsoft Excel spreadsheet).

Related to functional analysis, we cite and studies from our previous publication (Hughes et al. Nature Communications 2014 5:3571) where we demonstrated that the covalent inhibitor ligands (GW9662 and T0070907) do not block the activity of other ligands using a PPARγ transcriptional reporter assay and gene expression analysis in 3T3-L1 preadipocytes. Our study here expands on this finding and other published studies showing the structural mechanism for the lack of blocking activity by the covalent inhibitors.

Reviewer #2 (Public Review):

Summary:

The flexibility of the ligand binding domain (LBD) of NRs allows various modes of ligand binding leading to various cellular outcomes. In the case of PPARγ, it's known that two ligands can co-bind to the receptor. However, whether a covalent inhibitor functions by blocking the binding of a non-covalent ligand, or co-bind in a manner that weakens the binding of a non-covalent ligand remains unclear. In this study, the authors first used TR-FRET and NMR to demonstrate that covalent inhibitors (such as GW9662 and T0070907) weaken but do not prevent non-covalent synthetic ligands from binding, likely via an allosteric mechanism. The AF-2 helix can exchange between active and repressive conformations, and covalent inhibitors shift the conformation toward a transcriptionally repressive one to reduce the orthosteric binding of the non-covalent ligands. By co-crystal studies, the authors further reveal the structural details of various non-covalent ligand binding mechanisms in a ligand-specific manner (e.g., an alternate binding site, or a new orthosteric binding mode by alerting covalent ligand binding pose).

Strengths:

The biochemical and biophysical evidence presented is strong and convincing.

Weaknesses:

However, the co-crystal studies were performed by soaking non-covalent ligands to LBD pre-crystalized with a covalent inhibitor. Since the covalent inhibitors would shift the LBD toward transcriptionally repressive conformation which reduces orthosteric binding of non-covalent ligands, if the sequence was reversed (i.e., soaking a covalent inhibitor to LBD pre-crystalized with a non-covalent ligand), would a similar conclusion be drawn? Additional discussion will broaden the implications of the conclusion.

This is an interesting point, which we now expand upon in a new (third) paragraph of the discussion in our revised manuscript:

“In our previous study, we observed synthetic and natural/endogenous ligand co-binding via co-crystallography where preformed crystals of PPARγ LBD bound to unsaturated fatty acids (UFAs) were soaked with a synthetic ligand, which pushed the bound UFA to an alternate site within the orthosteric ligand-binding pocket 8. In the scenario of synthetic ligand cobinding with a covalent inhibitor, it is possible that soaking a covalent inhibitor into preformed crystals where the PPARγ LBD is already bound to a non-covalent ligand may prove to be difficult. The covalent inhibitor would need to flow through solvent channels within the crystal lattice, which may not be a problem. However, upon reaching the entrance surface to the orthosteric ligand-binding pocket, it may be difficult for the covalent inhibitor to gain access to the region of the orthosteric pocket required for covalent modification as the larger non-covalent ligand could block access. This potential order of addition problem may not be a problem for studies in solution or in cells, where the non-covalent ligand can more freely exchange in and out of the orthosteric pocket and over time the covalent reaction would reach full occupancy.”

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

- IC50 or EC50 values are not reported for the coregulator interaction assays, R2 for fit should also be reported where Ki and IC50s are disclosed.

We now report fitting statistics and IC50/EC50 values when possible in Figure 2B and Source Data 1 along with R2 values for the fit. We note that some data do not show complete or robust enough binding curves to faithfully fit to a dose response equation.

-  Reporter gene or qPCR should be performed for the combinations of covalent and noncovalent ligands to show how these molecules impact transcriptional activities rather than just coregulator binding profiles.

We previously performed PPARγ transcriptional reporter assay and gene expression analysis in 3T3-L1 preadipocytes to demonstrate that cotreatment of a covalent inhibitor (GW9662 or T0070907) with a non-covalent ligand does not block activity of the non-covalent ligand and showed cobinding-induced activation relative to DMSO control (Hughes et al., 2024 Nature Communications). We did not specifically mention this in our original manuscript, but we now call this out in the first paragraph of the results section.

- Inclusion of a structure figure to show the different helix 12 orientations should be included in the introduction. Likewise, how the overall structure of the LBD changes as a result of the cobinding in the discussion or a summary model would be helpful.

Our revised manuscript includes a structure figure called out in the introduction describing the active and repressive helix 12 PPARγ LBD conformations (new Figure 1). There are no major changes to the overall structure of the LBD compared to the active conformation that crystallized, so we did not include a summary model figure but we do refer readers to our previous paper (Shang and Kojetin, Structure 2021 29(9):940-950) in the penultimate paragraph of the discussion. We also added the following sentence to the crystallography results section related to the overall LBD changes:

“The structures show high structural similarity to the transcriptionally active LBD conformation with rmsd values ranging from 0.77–1.03Å (Supplementary Table S2)”

A typo in paragraph 3 of the discussion says "long-live" when it should probably say "long-lived."

We corrected this typo.

Reviewer #2 (Recommendations For The Authors):

It's interesting that ligand-specific binding mode of non-covalent ligands was observed. Would modifications of the chemical structure of a covalent inhibitor alter the allosteric binding behavior of non-covalent ligands in a predictive manner? If so, how can such SAR be used to guide the design of covalent inhibitors to more broadly and effectively inhibit agonists of various chemical structures? Discussion on this topic could be valuable.

This is an interesting point, which we now discuss in the penultimate and last paragraphs of the discussion:

“Another way to test this structural model could be through the use of covalent PPARγ inverse agonist analogs with graded activity 23, where one might posit that covalent inverse agonist analogs that shift the LBD conformational ensemble towards a fully repressive LBD conformation may better inhibit synthetic ligand cobinding.”

“It may be possible to use the crystal structures we obtained to guide structure-informed design of covalent inhibitors that would physically block cobinding of a synthetic ligand. This could be the potential mechanism of a newer generation covalent antagonist inhibitor we developed, SR16832, that more completely inhibit alternate site ligand binding of an analog of MRL20, rosiglitazone and the UFA docosahexaenoic acid (DHA)

21 and thus may be a better choice for the field to use as a covalent ligand inhibitor of PPARγ.”

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

How plants perceive their environment and signal during growth and development is of fundamental importance for plant biology. Over the last few decades, nano domain organisation of proteins localised within the plasma-membrane has emerged as a way of organising proteins involved in signal pathways. Here, the authors addressed how a non-surface localised signal (viral infection) was resisted by PM localised signalling proteins and the effect of nano domain organisation during this process. This is valuable work as it describes how an intracellular process affects signalling at the PM where most previous work has focused on the other way round, PM signalling effecting downstream responses in the plant. They identify CPK3 as a specific calcium dependent protein kinase which is important for inhibiting viral spread. The authors then go on to show that CPK3 diffusion in the membrane is reduced after viral infection and study the interaction between CPK3 and the remorins, which are a group of scaffold proteins important in nano domain organisation. The authors conclude that there is an interdependence between CPK3 and remorins to control their dynamics during viral infection in plants.

Strengths:

The dissection of which CPK was involved in the viral propagation was masterful and very conclusive. Identifying CPK3 through knockout time course monitoring of viral movement was very convincing. The inclusion of overexpression, constitutively active and point mutation non functioning lines further added to that.

Weaknesses:

My main concerns with the work are twofold.

(1) Firstly, the imaging described and shown is not sufficient to support the claims made. The PM localisation and its non-PM localised form look similar and with no PM stain or marker construct used to support this. The sptPALM data conclusions are nice and fit the narrative. However, no raw data or movie is shown, only representative tracks. Therefore, the data quality cannot be verified and in addition, the reporting of number of single particle events visualised per experiment is absent, only number of cells imaged is reported. Therefore, it is impossible for the reader to appreciate the number of single molecule behaviours obtained and hence the quality of the data.

(2) Secondly, remorins are involved in a lot of nanodomain controlled processes at the PM. The authors have not conclusively demonstrated that during viral infection the remorin effects seen are solely due to its interaction with CPK3. The sptPALM imaging of REM1.2 in a cpk3 knockout line goes part way to solve this but more evidence would strengthen it in my opinion. How do we not know that during viral infection the entire PM protein dynamics and organisation are altered? Or that CPK3 and REM are at very distant ends of a signalling cascade. Negative control experiments are required here utilising other PM localised proteins which have no role during viral infection. In addition, if the interaction is specific, the transiently expressed CPK3-CA construct (shown to from nano domains) should be expressed with REM1.2-mEOS to show the alterations in single particle behaviour occur due to specific activations of CPK3 and REM1.2 in the absence of PIAMV viral infection and it is not an artefact of whole PM changes in dynamics during viral infection.

In addition, displaying more information throughout the manuscript (such as raw particle tracking movies and numbers of tracks measured) on the already generated data would strengthen the manuscript further.

Overall, I think this work has the potential to be a very strong manuscript but additional reporting of methods and data are required and additional lines of evidence supporting interaction claims would significantly strengthen the work and make it exceptional.

Reviewer #2 (Public Review):

Summary:

The paper provides evidence that CPK3 plays a role in plant virus infection, and reports that viral infection is accompanied by changes in the dynamics of CPK3 and REM1.2, the phosphorylation substrate of CPK3, in the plasma membrane. In addition, the dynamics of the two proteins in the PM are shown to be interdependent.

Strengths:

The paper contains novel, important information.

Weaknesses:

The interpretation of some experimental data is not justified, and the proposed model is not fully based on the available data.

Reviewer #3 (Public Review):

Summary:

This study examined the role that the activation and plasma membrane localisation of a calcium dependent protein kinase (CPK3) plays in plant defence against viruses.<br /> The authors clearly demonstrate that the ability to hamper the cell-to-cell spread of the virus P1AMV is not common to other CPKs which have roles in defence against different types of pathogens, but appears to be specific to CPK3 in Arabidopsis. Further they show that lateral diffusion of CPK3 in the plasma membrane is reduced upon P1AMV infection, with CPK3 likely present in nano-domains. This stabilisation however, depends on one of its phosphorylation substrates a Remorin scaffold protein REM1-2. However, when REM1-2 lateral diffusion was tracked, it showed an increase in movement in response to P1AMV infection. These contrary responses to P1AMV infection were further demonstrated to be interdependent, which led the authors to propose a model in which activated CPK3 is stabilised in nano-domains in part by its interaction with REM1.2, which it binds and phosphorylates, allowing REM1-2 to diffuse more dynamically within the membrane.

The likely impact of this work is that it will lead to closer examination of the formation of nano-domains in the plasma membrane and dissection of their role in immunity to viruses, as well as further investigation into the specific mechanisms by which CPK3 and REM1-2 inhibit the cell-to-cell spread of viruses.

Strengths:

The paper provided compelling evidence about the roles of CPK3 and REM1-2 through a combination of logical reverse genetics experiments and advanced microscopy techniques, particularly in single particle tracking.

Weaknesses:

There is a lack of evidence for the downstream pathways, specifically whether the role that CPK3 has in cytoskeletal organisation may play a role in the plant's defence against viral propagation. Also, there is limited discussion about the localisation of the nano-domains and whether there is any overlap with plasmodesmata, which as plant viruses utilise PD to move from cell to cell seems an obvious avenue to investigate.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Viral spread work in CPK mutants with time courses is beautiful!

Regarding my public points on my issues with the imaging:

- Figure 2A shows 'PM' localisation of CPK3 and 'non-PM' imaging of CPK3-G2A. The images are near identical both showing cell outlines and cytoplasmic strands. Here a PM marker (such as Lti6B) tagged with a different fluorophore or PM stain should be used in conjunction with surface views (such as in Figure 2C) to show it really is at the PM and the G2A line is not.

Impaired membrane localization of CPK3-G2A is documented in Mehlmer et al., 2010 using microsomal fractionation. Although Figure 2A main purpose is to show correct expression of the constructs in the lines used for PlAMV propagation (Figure 2B), we replaced the images with wider view pictures to be more representative of the subcellular localization of CPK3 and CPK3-G2A.

- Regarding Figure 2C, this is extremely noisy and PM heterogeneity is barely observable over the noise from the system (looking at the edges of surface imaged). You mention low resolution was an issue. I notice from the methods you have taken confocal images on an Zeiss 880 with Airyscan. These images must be confocal but If imaged with Airyscan the PM heterogeneity would be much clearer (see work from John Runions lab).

Indeed, these are tangential views images obtained by Zeiss 880 with Airyscan. Based on tessellation analysis (Figure 2H-J), CPK3 is rather homogeneously distributed and forms ND of around 70nm of diameter. Objects of such size cannot be resolved using pixel reassignment methods such as Airyscan. Note also that AtREM in our study are less heterogeneously distributed than what was described in the literature for StREM1.3.

- Regarding all sptPALM data. At least an example real data image and video is required otherwise the data can’t be assessed. The work of Alex Martiniere (sptPALM) or Alex Jonson (TIRF) all show raw data so the reader can appreciate the quality of the data. In addition, number of events (particles tracked) has to be shown in the figure legend, not just number of cells otherwise was one track performed per cell? Or 10,000? Obviously where the N sits in this range gives the reader more or less confidence of the data.

We agree and we added example videos of sptPALM experiments in the supplementary data, we also indicated the number of tracked particles in the figure legends.

- On a slight technical aside, how do you know the cells being imaged for sptPALM with PIAMV are actually infected with the virus? In Fig 2C you use a GFP tagged version but in sptPALM you use none tagged. I think a sentence in methods on this would help clarify.

PlAMV-GFP was used for spt-PALM experiment and cell infection was assessed during PALM experiment. This is now precised in the corresponding figures and methods.

- I also have a concern over some of the representative images showing the same things between different figures. Your clustering data in 3F looks very convincing. However, in Figure 2H the mock and PIAMV-GFP look very similar. How is Figure 3F so different for the same experiment? Especially considering the scale bars are the same for both figures. Same for CPK3-mRFP1.2 in Fig 2C and 3A, the same thing is being imaged, at the same scale (scale bars same size) but the images are extremely different.

Figure 2 data were generated using CPK3 stably expressed in A. thaliana while Figure 3 data were obtained upon transient over-expression of CPK3 in N. benthamiana. We do not have a clear explanation for such a difference in CPK3 PM behavior, it could lie on a different PM composition or actin organization between those two species, this point is now addressed in the discussion.

- Line 193&194 - you state that the CA CPK3 is reminiscent of the CPK3 upon PIAMV expression. I don't agree, while CPK3CA is less mobile (2D), the MSD shows it is in-between CPK3 and CPK3 + PIAMV. Therefore, can’t the opposite also be true? That overall the behaviour of CPK3-CA is reminiscent of WT CPK. I think this needs rewording.

We agree and we reworded that part

- Line 651 - what numerical aperture are you using for the lens during confocal microscopy. This is fundamentally important information directly related to the reproducibility of the work. You report it for the sptPALM.

The numerical aperture is now indicated in the methods.

Regarding my bigger point about specific interactions between CPK3 and remorin during viral infection to strengthen your claim the following need doing. I am not suggesting you do all of these but at least two would significantly enhance the paper.

(1) Image a none related PM protein during viral infection using sptPALM and demonstrate that its behaviour is not altered (such as lti6b). This would show the affects on remorin behaviour are specific to CPK3 and not a whole scale PM alteration in dynamics due to viral infection.

(2) Two colour SPT imaging of CPK3 and REM1.2. You show in absence of proteins (knockouts effect on each other) but your only interaction data is from a kinase assay (where CPK1 and 2 also interact, even though they are not localised at the same place) and colocalisation data (see below). A two colour SPT imaging experiment showing interaction and clustering of CPK3 and REM1.2 with each other and the change in their behaviours when viral infected and simultaneously imaged would address all of my concerns.

- On another note, the co-localisation data (fig 5 sup 4) needs additional analysis. I would expect most PM proteins to show the results you show as the data is very noisy. In order to improve I would zoom in to fill the field of view and then determine correlation and also when one image is rotated 90 degrees (as described in Jarsch et al., plant cell) to enhance this work.

(3) In the absence of viral infection, but presence of CPK3-CA, is sptPALM REM1.2 behaviour in the PM altered, if so then the interaction is specific and changes in remorin dynamics are not due to whole scale PM changes during viral infection and the manuscript substantially strengthened.

(4) Building on from 3), if you have a CPK3 mutated with both CPK3-CA and G2A this would be constitutively active and non-PM localised and as such should not affect Remorin behaviour if your model is true, this would strengthen the case significantly but I appreciate is highly artificial and would need to be done transiently.

Regarding the first point, since the role of PM proteins involved in potexvirus infection is barely assessed, picking a non-related PM protein might be tricky. The data obtained with mEOS3.2-REM1.2 expressed in cpk3 null-mutant point towards a specific role of CPK3 in PlAMV-induced REM1.2 diffusion and not a general alteration of PM protein behavior.

Regarding the second point, we already reported the in vivo interaction between AtCPK3CA and AtREM1.2/AtREM1.3 by BiFC in N.benthamiana (Perraki et al 2018) and AtCPK3 was shown to co-IP with AtREM1.2 (Abel et al, 2021). While we agree on the relevance of doing dual color sptPALM with CPK3 and REM1.2, it is so far technically challenging and we would not be able to implement this in a timely manner. For the colocalization, although the whole cell is displayed in the figure, the analysis was performed on ROI to fill the field of analysis.

We agree with the relevance of adding the colocalization analysis of randomized images (mTagBFP2 channel rotated 90 degrees), this is now added to Figure 5 – supplement figure 5.

Finally, for the third and fourth points, spt-PALM analysis of REM1.2 in presence of CPK3-CA and CPK3-CA-G2A was performed (Figure 5 - figure supplement 4). The results suggest a specific role of CPK3-CA in REM1.2 diffusion.

Minor points:

Line 59 - from, I think you mean from.

Line 63 - Reference needed after latter.

Line 68 - Reference required after viral infection.

Line 85 - Propose not proposed.

Line 156 - Allowed us to not allows to.

Line 204 - add we previously 'demonstrated'

Line 622 and 623 - You say lines obtained from Thomas Ott. This is very odd phrasing considering he is an author. I appreciate citing the work producing the lines but maybe reword this

These points were corrected, thank you.

Reviewer #2 (Recommendations For The Authors):

The paper provides evidence that CPK3 plays a role in plant virus infection, and reports that viral infection is accompanied by changes in the dynamics of CPK3 and REM1.2, the phosphorylation substrate of CPK3, in the plasma membrane. In addition, the dynamics of the two proteins in the PM are shown to be interdependent. The paper contains novel, important information that can undoubtedly be published in eLife. However, I have some concerns that should be addressed before it can be accepted for publication.

Major concerns

When the authors say that CPK3 plays a role in viral propagation, it should be clarified what is meant by 'propagation', - replication of the viral genome, its cell-to-cell transport, or long-distance transport via the phloem. By default the readers will tend to assume the former meaning. In my opinion, the term 'propagation' is misleading and should be avoided.

We purposely chose the term “propagation” because it sums replication and cell-to-cell movement. Nevertheless, we previously showed that group 1 StREM1.3 doesn’t alter PVX replication (Raffaele et al., 2009 The Plant Cell). In this paper, as we do not investigate the role of AtREM1.2 or AtCPK3 in the replication of the viral PlAMV genome, we cannot state that these proteins are strictly involved in cell-to-cell movement of the virus.

The authors show that viral infection is associated with decreased diffusion of CPK3 and increased diffusion of REM1.2 in the PM. However, it remains unclear whether these changes are related to partial resistance to viral infection involving CPK3 and REM1.2, or whether they are simply a consequence of viral infection that may lead to altered PM properties and altered dynamics of PM-associated proteins. Therefore, the model presented in Fig. 6 appears to be entirely speculative, as it postulates that changes in CPK3 and REM1.2 dynamics are the cause of suppressed virus 'propagation'. In addition, the model implies that a decrease in CPK3 mobility leads to activation of its kinase activity. This view is not supported by experimental data (see my next comment). The model should be deleted (both as the figure and its description in the Discussion) or substantially reworked so that it finally relies on existing data.

For the first point, the results obtained from the additional experiments proposed by reviewer #1 supports the hypothesis of a direct impact of CPK3 on REM1.2 diffusion (Figure 5 - figure supplement 4).

We agree with the second point and reworked the model to remove the link between CPK3 activation and its increased diffusion.

The statement that 'changes in CPK3 dynamics upon PlAMV infection are linked to its activation' (line 194) is based on a flawed logic, and the conclusion in this section of Results ('changes in CPK3 dynamics upon PlAMV infection are linked to its activation') is incorrect, as it is not supported by experimental data. In fact, the authors show that CPK3 dynamics and clustering upon viral infection is somewhat reminiscent of the behavior of a CPK3 deletion mutant, which is a constitutively active protein kinase. However, this partial similarity cannot be taken as evidence that CPK3 dynamics upon PlAMV infection are related to its activation. Furthermore, the authors emphasize the similarity of the mutant and CPK3 in infected cells without taking into account a drastic difference in their localization (Fig. 3A, middle and right panels) showing that the reduced dynamics or the compared proteins may have different causes. I suggest the removal of the section 'CPK3 activation leads to its confinement in PM ND' from the paper, as the results included in this section are not directly related to other data presented.

The PM lateral organization of PM-bound CPKs in their native or constitutively active form as well as the role of lipid in such phenomenon was never shown before. We believe that this section contains relevant information for the community. We kept the section but reworded it to tamper the correlation made between CPK3 PM organization upon viral infection and its activation.

Line 270 - 'group 1 REMs might play a role in CPK3 domain stabilization upon viral infection'. This is an overstatement. The size of the CPK3-containing NDs may have no correlation with their stability.

We reworded the sentence.

Minor points

Line 204 - we previously that Line 234 and hereafter - "the D" sounds strange. Suggest using "the diffusion coefficient".

This was reworded.

Reviewer #3 (Recommendations For The Authors):

The authors have previously demonstrated that there was an increase in REM1.2 localisation to plasmodesmata under viral challenge. It would be useful to see if there was any co-localisation of REM1.2 and CPK3 with plasmodesmata in response to PlAMV and how this is affected in the mutants. This could be carried out relatively simply using aniline blue.

These experiments were added to the supplementary data of Figure 2 – figure supplement 2.  and Figure 4 – figure supplement 4. , no enrichment of CPK3 or REM1.2 at plasmodesmata could be observed upon PlAMV infection.

Fig 3 supplementary figure 2 would be better incorporated into the main body of Figure 3 as this underpins discussion on the involvement of lipids such as sterols in the formation of nanodomains.

We moved Figure 3 – Supplementary figure 2 to the main body of Figure 3.

Minor corrections:

Whilst the paper is generally well written there are a number of grammatical errors:

Line 1 & 2: Title doesn't quite read correctly, suggest a rewording for clarity.

L31: Insert "a"after only

L33: Replace "are playing" with "play"

L34: Begin sentence "Viruses are intracellular pathogens and as such the role..."59: replace "form" with "from"

L63: Insert "was demonstrated" after REM1.2)

L85: Replace "proposed" with "propose"

L86: replace "encouraging to explore" with "which will encourage further exploration of "

L129: replace "we'll focus on" with "we concentrated on"

L131: insert "an" before ATP

L138: change "among" to "amongst"

L156: change "allows to analyse" to "allows the analysis of"

L204: Insert "showed" after previously.

L232: "root seedlings" should this be the roots of seedlings?

L235: insert "to" after "as"

L280: insert "a" after "only"

L281: change " to play" with "as playing": change CA to superscript

L307: Insert "was" after "transcription"

L320: change "display" to "displaying"

L321: change "form" to forms"

L340: "hampering" should come before viral

L365: insert"us' after "allow"

Thank you, these were corrected

Author response:

Public Reviews: 

Reviewer #1 (Public review): 

Summary: 

This paper provides a computational model of a synthetic task in which an agent needs to find a trajectory to a rewarding goal in a 2D-grid world, in which certain grid blocks incur a punishment. In a completely unrelated setup without explicit rewards, they then provide a model that explains data from an approach-avoidance experiment in which an agent needs to decide whether to approach or withdraw from, a jellyfish, in order to avoid a pain stimulus, with no explicit rewards. Both models include components that are labelled as Pavlovian; hence the authors argue that their data show that the brain uses a Pavlovian fear system in complex navigational and approach-avoid decisions. 

We thank the reviewer for their thoughtful comments. To clarify, the grid-world setup was used as a didactic tool/testbed to understand the interaction between Pavlovian and instrumental systems (lines 80-81) [Dayan et al., 2006], specifically in the context of safe exploration and learning. It helps us delineate the Pavlovian contributions during learning, which is key to understanding the safety-efficiency dilemma we highlight. This approach generates a hypothesis about outcome uncertainty-based arbitration between these systems, which we then test in the approach-withdrawal VR experiment based on foundational studies studying Pavlovian biases [Guitart-Masip et al., 2012, Cavanagh et al., 2013].

Although the VR task does not explicitly involve rewards, it provides a specific test of our hypothesis regarding flexible Pavlovian fear bias, similar to how others have tested flexible Pavlovian reward bias without involving punishments (e.g., Dorfman & Gershman, 2019). Both the simulation and VR experiment models are derived from the same theoretical framework and maintain algebraic mapping, differing only in task-specific adaptations (e.g., differing in action sets and temporal difference learning for multi-step decisions in the grid world vs. Rescorla-Wagner rule for single-step decisions in the VR task). This is also true for Dayan et al. [2006] who bridge Pavlovian bias in a Go-No Go task (negative auto-maintenance pecking task) and a grid world task. Therefore, we respectfully disagree that the two setups are completely unrelated and that both models include components merely labelled as Pavlovian.

We will rephrase parts of the manuscript to prevent the main message of our manuscript from being misconveyed. Particularly in the Methods and Discussion, to clarify that our main focus is on Pavlovian fear bias in safe exploration and learning (as also summarised by reviewers #2 and #3), rather than on its role in complex navigational decisions. We also acknowledge the need for future work to capture more sophisticated safe behaviours, such as escapes and sophisticated planning which span different aspects of the threat-imminence continuum [Mobbs et al., 2020], and we will highlight these as avenues for future research.

In the first setup, they simulate a model in which a component they label as Pavlovian learns about punishment in each grid block, whereas a Q-learner learns about the optimal path to the goal, using a scalar loss function for rewards and punishments. Pavlovian and Q-learning components are then weighed at each step to produce an action. Unsurprisingly, the authors find that including the Pavlovian component in the model reduces the cumulative punishment incurred, and this increases as the weight of the Pavlovian system increases. The paper does not explore to what extent increasing the punishment loss (while keeping reward loss constant) would lead to the same outcomes with a simpler model architecture, so any claim that the Pavlovian component is required for such a result is not justified by the modelling. 

Thank you for this comment. We acknowledge that our paper does not compare the Pavlovian fear system to a purely instrumental system with varying punishment sensitivity. Instead, our model assumes the coexistence of these two systems and demonstrates the emergent safety-efficiency trade-off from their interaction. It is possible that similar behaviours could be modelled using an instrumental system alone. In light of the reviewer’s comment, we will soften our claims regarding the necessity of the Pavlovian system, despite its known existence.

We also encourage the reviewer to consider the Pavlovian system as a biologically plausible implementation of punishment sensitivity. Unlike punishment sensitivity (scaling of the punishments), which has not been robustly mapped to neural substrates in fMRI studies, the neural substrates for the Pavlovian fear system (e.g., the limbic loop) are well known (see Supplementary Fig. 16).

Additionally, we point out that varying reward sensitivities while keeping punishment sensitivity constant allows our PAL agent to differentiate from an instrumental agent that combines reward and punishment into a single feedback signal. As highlighted in lines 136-140 and the T-maze experiment (Fig. 3 A, B, C), the Pavlovian system maintains fear responses even under high reward conditions, guiding withdrawal behaviour when necessary (e.g., ω = 0.9 or 1), which is not possible with a purely instrumental model if the punishment sensitivities are fixed. This is a fundamental point.

We will revise our discussion and results sections to reflect these clarifications.

In the second setup, an agent learns about punishments alone. "Pavlovian biases" have previously been demonstrated in this task (i.e. an overavoidance when the correct decision is to approach). The authors explore several models (all of which are dissimilar to the ones used in the first setup) to account for the Pavlovian biases. 

Thank you, we respectfully disagree with the statement that our models used in the experimental setup are dissimilar to the ones used in the first setup. Due to differences in the nature of the task setup, the action set differs, but the model equations and the theory are the same and align closely, as described in our response above. The only additional difference is the use of a baseline bias in human experiments and the RLDDM model, where we also model reaction times with drift rates which is not a behaviour often simulated in grid world simulations. We will improve our Methods section to ensure that model similarity is highlighted.

Strengths: 

Overall, the modelling exercises are interesting and relevant and incrementally expand the space of existing models. 

We thank reviewer #1 for acknowledging the relevance of our models in advancing the field. We would like to further highlight that, to the best of our knowledge, this is the first time reaction times in Pavlovian-Instrumental arbitration tasks have been modelled using RLDDM, which adds a novel dimension to our approach.

Weaknesses: 

I find the conclusions misleading, as they are not supported by the data. 

First, the similarity between the models used in the two setups appears to be more semantic than computational or biological. So it is unclear to me how the results can be integrated. 

We acknowledge the dissimilarity between the task setups (grid-world vs. approach-withdrawal). However, we believe these setups are computationally similar and may be biologically related, as suggested by prior work like Dayan et al. [2006], which integrates Go-No Go and grid-world tasks. Just as that work bridged findings in the appetitive domain, we aim to integrate our findings in the aversive domain. We will provide a more integrated interpretation in the discussion section of the revised manuscript.

Dayan, P., Niv, Y., Seymour, B., and Daw, N. D. (2006). The misbehavior of value and the discipline of the will. Neural networks, 19(8):1153–1160.

Secondly, the authors do not show "a computational advantage to maintaining a specific fear memory during exploratory decision-making" (as they claim in the abstract). Making such a claim would require showing an advantage in the first place. For the first setup, the simulation results will likely be replicated by a simple Q-learning model when scaling up the loss incurred for punishments, in which case the more complex model architecture would not confer an advantage. The second setup, in contrast, is so excessively artificial that even if a particular model conferred an advantage here, this is highly unlikely to translate into any real-world advantage for a biological agent. The experimental setup was developed to demonstrate the existence of Pavlovian biases, but it is not designed to conclusively investigate how they come about. In a nutshell, who in their right mind would touch a stinging jellyfish 88 times in a short period of time, as the subjects do on average in this task? Furthermore, in which real-life environment does withdrawal from a jellyfish lead to a sting, as in this task? 

Thank you for your feedback. As mentioned above, we invite the reviewer to potentially think of Pavlovian fear systems as a way how the brain might implement punishment sensitivity. Secondly, it provides a separate punishment memory that cannot be overwritten with higher rewards (see also Elfwing and Seymour 2017, and Wang et al, 2021)

Elfwing, S., & Seymour, B. (2017, September). Parallel reward and punishment control in humans and robots: Safe reinforcement learning using the MaxPain algorithm. In 2017 Joint IEEE International Conference on Development and Learning and Epigenetic Robotics (ICDL-EpiRob) (pp. 140-147). IEEE. 

Wang, J., Elfwing, S., & Uchibe, E. (2021). Modular deep reinforcement learning from reward and punishment for robot navigation. Neural Networks, 135, 115-126.

The simulation setups such as the following grid-worlds are common test-beds for algorithms in reinforcement learning [Sutton and Barto, 2018].

Any experimental setup faces the problem of having a constrained experiment designed to test and model a specific effect versus designing a lesser constrained exploratory experiment which is more difficult to model. Here we chose the former, building upon previous foundational experiments on Pavlovian bias in humans [Guitart-Masip et al., 2012, Cavanagh et al., 2013].  The condition where withdrawal from a jellyfish leads to a sting, though less realistic, was included for balancing the four cue-outcome conditions. Overall the task was designed to isolate the effect we wanted to test - Pavlovian fear bias in choices and reaction times, to the best of our ability. In a free operant task, it is very well likely that other components not included in our model could compete for control.

Crucially, simplistic models such as the present ones can easily solve specifically designed lab tasks with low dimensionality but they will fail in higher-dimensional settings. Biological behaviour in the face of threat is utterly complex and goes far beyond simplistic fight-flight-freeze distinctions (Evans et al., 2019). It would take a leap of faith to assume that human decision-making can be broken down into oversimplified sub-tasks of this sort (and if that were the case, this would require a meta-controller arbitrating the systems for all the sub-tasks, and this meta-controller would then struggle with the dimensionality j). 

We agree that safe behaviours, such as escapes, involve more sophisticated computations. We do not propose Pavlovian fear bias as the sole computation for safe behavior, but rather as one of many possible contributors. Knowing about the existence about the Pavlovian withdrawal bias, we simply study its possible contribution. We will include in our discussion that such behaviours likely occupy different parts of the threat-imminence continuum [Mobbs et al., 2020].

Dean Mobbs, Drew B Headley, Weilun Ding, and Peter Dayan. Space, time, and fear: survival computations along defensive circuits. Trends in cognitive sciences, 24(3):228–241, 2020.

On the face of it, the VR task provides higher "ecological validity" than previous screen-based tasks. However, in fact, it is only the visual stimulation that differs from a standard screen-based task, whereas the action space is exactly the same. As such, the benefit of VR does not become apparent, and its full potential is foregone. 

We thank the reviewer for their comment. We selected the action space to build on existing models [Guitart-Masip et al., 2012, Cavanagh et al., 2013] that capture Pavlovian biases and we also wanted to minimize participant movement for EEG data collection. Unfortunately, despite restricting movement to just the arm, the EEG data was still too noisy to lead to any substantial results. We will explore more free-operant paradigms in future works.

On the issue of the difference between VR and lab-based tasks, we note the reviewer's point. Note however that desktop monitor-based tasks lack the sensorimotor congruency between the action and the outcome. Second, it is also arguable, that the background context is important in fear conditioning, as it may help set the tone of the fear system to make aversive components easier to distinguish.

If the authors are convinced that their model can - then data from naturalistic approach-avoidance VR tasks is publicly available, e.g. (Sporrer et al., 2023), so this should be rather easy to prove or disprove. In summary, I am doubtful that the models have any relevance for real-life human decision-making. 

We thank the reviewers for their thoughtful inputs. We do not claim our model is the best fit for all naturalistic VR tasks, as they require multiple systems across the threat-imminence continuum [Mobbs et al., 2020] and are currently beyond the scope of the current work. However, we believe our findings on outcome-uncertainty-based arbitration of Pavlovian bias could inform future studies and may be relevant for testing differences in patients with mental disorders, as noted by reviewer #2. At a general level, it can be said that most well-controlled laboratory-based tasks need to bridge a sizeable gap to applicabilty in real-life naturalistic behaviour; although the principle of using carefully designed tasks to isolate individual factors is well established

Finally, the authors seem to make much broader claims that their models can solve safety-efficiency dilemmas. However, a combination of a Pavlovian bias and an instrumental learner (study 1) via a fixed linear weighting does not seem to be "safe" in any strict sense. This will lead to the agent making decisions leading to death when the promised reward is large enough (outside perhaps a very specific region of the parameter space). Would it not be more helpful to prune the decision tree according to a fixed threshold (Huys et al., 2012)? So, in a way, the model is useful for avoiding cumulatively excessive pain but not instantaneous destruction. As such, it is not clear what real-life situation is modelled here. 

We thank the reviewer for their comments and ideas. In our discussion lines 257-264, we discuss other works which identify similar safety-efficiency dilemmas, in different models. Here, we simply focus on the safety-efficiency trade-off arising from the interactions between Pavlovian and instrumental systems. It is important to note that the computational argument for the modular system with separate rewards and punishments explicitly protects (up to a point, of course) against large rewards leading to death because the Pavlovian fear response is not over-written by successful avoidance in recent experience. Note also that in animals, reward utility curves are typically convex. We will clarify this in the discussion section.

We completely agree that in certain scenarios, pruning decision trees could be more effective, especially with a model-based instrumental agent. Here we utilise a model-free instrumental agent, which leads to a simpler model - which is appreciated by some readers such as reviewer #2. Future work can incorporate model-based methods.

A final caveat regarding Study 1 is the use of a PH associability term as a surrogate for uncertainty. The authors argue that this term provides a good fit to fear-conditioned SCR but that is only true in comparison to simpler RW-type models. Literature using a broader model space suggests that a formal account of uncertainty could fit this conditioned response even better (Tzovara et al., 2018). 

We thank the reviewer for bringing this to our notice. We will discuss Tzovara et al., 2018 in our discussion in our revised manuscript.

Reviewer #2 (Public review): 

Summary: 

The authors tested the efficiency of a model combining Pavlovian fear valuation and instrumental valuation. This model is amenable to many behavioral decision and learning setups - some of which have been or will be designed to test differences in patients with mental disorders (e.g., anxiety disorder, OCD, etc.). 

Strengths: 

(1) Simplicity of the model which can at the same time model rather complex environments. 

(2) Introduction of a flexible omega parameter. 

(3) Direct application to a rather advanced VR task. 

(4) The paper is extremely well written. It was a joy to read. 

Weaknesses: 

Almost none! In very few cases, the explanations could be a bit better. 

We thank reviewer #2 for their positive feedback and thoughtful recommendations. We will ensure that, in our revision, we clarify the explanations in the few instances where they may not be sufficiently detailed, as noted.

Reviewer #3 (Public review): 

Summary: 

This paper aims to address the problem of exploring potentially rewarding environments that contain the danger, based on the assumption that an independent Pavlovian fear learning system can help guide an agent during exploratory behaviour such that it avoids severe danger. This is important given that otherwise later gains seem to outweigh early threats, and agents may end up putting themselves in danger when it is advisable not to do so. 

The authors develop a computational model of exploratory behaviour that accounts for both instrumental and Pavlovian influences, combining the two according to uncertainty in the rewards. The result is that Pavlovian avoidance has a greater influence when the agent is uncertain about rewards. 

Strengths: 

The study does a thorough job of testing this model using both simulations and data from human participants performing an avoidance task. Simulations demonstrate that the model can produce "safe" behaviour, where the agent may not necessarily achieve the highest possible reward but ensures that losses are limited. Interestingly, the model appears to describe human avoidance behaviour in a task that tests for Pavlovian avoidance influences better than a model that doesn't adapt the balance between Pavlovian and instrumental based on uncertainty. The methods are robust, and generally, there is little to criticise about the study. 

Weaknesses: 

The extent of the testing in human participants is fairly limited but goes far enough to demonstrate that the model can account for human behaviour in an exemplar task. There are, however, some elements of the model that are unrealistic (for example, the fact that pre-training is required to select actions with a Pavlovian bias would require the agent to explore the environment initially and encounter a vast amount of danger in order to learn how to avoid the danger later). The description of the models is also a little difficult to parse. 

We thank reviewer #3 for their thoughtful feedback and useful recommendations, which we will take into account while revising the manuscript.

We acknowledge the complexity of specifying Pavlovian bias in the grid world and appreciate the opportunity to elaborate on how this bias is modelled. In the human experiment, the withdrawal action is straightforwardly biased, as noted, while in the grid world, we assume a hardwired encoding of withdrawal actions for each state/grid. This innate encoding of withdrawal actions could be represented in the dPAG [Kim et. al., 2013]. We implement this bias using pre-training, which we assume would be a product of evolution. Alternatively, this could be interpreted as deriving from an appropriate value initialization where the gradient over initialized values determines the action bias. Such aversive value initialization, driving avoidance of novel and threatening stimuli, has been observed in the tail of the striatum in mice, which is hypothesized to function as a Pavlovian fear/threat learning system [Menegas et. al., 2018].

Additionally, we explored the possibility of learning the action bias on the fly by tracking additional punishment Q-values instead of pre-training, which produced similar cumulative pain and step plots. While this approach is redundant, and likely not how the brain operates, it demonstrates an alternative algorithm.

We thank the reviewer for pointing out these potentially unrealistic elements, and we will revise the manuscript to clarify and incorporate these explanations and improve the model descriptions.

Eun Joo Kim, Omer Horovitz, Blake A Pellman, Lancy Mimi Tan, Qiuling Li, Gal Richter-Levin, and Jeansok J Kim. Dorsal periaqueductal gray-amygdala pathway conveys both innate and learned fear responses in rats. Proceedings of the National Academy of Sciences, 110(36):14795–14800, 2013

William Menegas, Korleki Akiti, Ryunosuke Amo, Naoshige Uchida, and Mitsuko Watabe-Uchida. Dopamine neurons projecting to the posterior striatum reinforce avoidance of threatening stimuli. Nature neuroscience, 21(10): 1421–1430, 2018

Author response:

Public Reviews: 

Reviewer #1 (Public review): 

Summary: 

The study significantly advances our understanding of how exosomes regulate filopodia formation. Filopodia play crucial roles in cell movement, polarization, directional sensing, and neuronal synapse formation. McAtee et al. demonstrated that exosomes, particularly those enriched with the protein THSD7A, play a pivotal role in promoting filopodia formation through Cdc42 in cancer cells and neurons. This discovery unveils a new extracellular mechanism through which cells can control their cytoskeletal dynamics and interaction with their surroundings. The study employs a combination of rescue experiments, live-cell imaging, cell culture, and proteomic analyses to thoroughly investigate the role of exosomes and THSD7A in filopodia formation in cancer cells and neurons. These findings offer valuable insights into fundamental biological processes of cell movement and communication and have potential implications for understanding cancer metastasis and neuronal development. 

Weaknesses: 

The conclusions of this study are in most cases supported by data, but some aspects of data analysis need to be better clarified and elaborated. Some conclusions need to be better stated and according to the data observed. 

We appreciate the reviewer's recognition of the impact of our study.  We will address the concerns about data analysis and statement of our conclusions in our full response to reviewers.

Reviewer #2 (Public review): 

Summary: 

The authors show that small EVs trigger the formation of filopodia in both cancer cells and neurons. They go on to show that two cargo proteins, endoglin, and THSD7A, are important for this process. This possibly occurs by activating the Rho-family GTPase CDC42. 

Strengths: 

The EV work is quite strong and convincing. The proteomics work is well executed and carefully analyzed. I was particularly impressed with the chick metastasis assay that added strong evidence of in vivo relevance. 

Weaknesses: 

The weakest part of the paper is the Cdc42 work at the end of the paper. It is incomplete and not terribly convincing. This part of the paper needs to be improved significantly.

We appreciate the reviewer's recognition of the impact of our study.  Indeed, more work needs to be done to clarify the role of Cdc42 in the induction of filopodia by exosome-associated THSD7A.  We anticipate that this will be a separate manuscript, delving in-depth into how exosome-associated THSD7A interacts with recipient cells to activate Cdc42 and carrying out a variety of assays for Cdc42 activation.

Reviewer #3 (Public review): 

Summary: 

The authors identify a novel relationship between exosome secretion and filopodia formation in cancer cells and neurons. They observe that multivesicular endosomes (MVE)-plasma membrane (PM) fusion is associated with filopodia formation in HT1080 cells and that MVEs are present in filopodia in primary neurons. Using overexpression and knockdown (KD) of Rab27/HRS in HT1080 cells, melanoma cells, and/or primary rat neurons, they found that decreasing exosome secretion reduces filopodia formation, while Rab27 overexpression leads to the opposite result. Furthermore, the decreased filopodia formation is rescued in the Rab27a/HRS KD melanoma cells by the addition of small extracellular vesicles (EVs) but not large EVs purified from control cells. The authors identify endoglin as a protein unique to small EVs secreted by cancer cells when compared to large EVs. KD of endoglin reduces filopodia formation and this is rescued by the addition of small EVs from control cells and not by small EVs from endoglin KD cells. Based on the role of filopodia in cancer metastasis, the authors then investigate the role of endoglin in cancer cell metastasis using a chick embryo model. They find that injection of endoglin KD HT1080 cells into chick embryos gives rise to less metastasis compared to control cells - a phenotype that is rescued by the co-injection of small EVs from control cells. Using quantitative mass spectrometry analysis, they find that thrombospondin type 1 domain containing 7a protein (THSD7A) is downregulated in small EVs from endoglin KD melanoma cells compared to those from control cells. They also report that THSD7A is more abundant in endoglin KD cell lysate compared to control HT1080 cells and less abundant in small EVs from endoglin KD cells compared to control cells, indicating a trafficking defect. Indeed, using immunofluorescence microscopy, the authors observe THSD7A-mScarlet accumulation in CD63-positive structures in endoglin KD HT1080 cells, compared to control cells. Finally, the authors determine that exosome-secreted THSD7A induces filopodia formation in a Cdc42-dependent mechanism. 

Strengths: 

(1) While exosomes are known to play a role in cell migration and autocrine signaling, the relationship between exosome secretion and the formation of filopodia is novel. 

(2) The authors identify an exosomal cargo protein, THSD7A, which is essential for regulating this function. 

(3) The data presented provide strong evidence of a role for endoglin in the trafficking of THSD7A in exosomes. 

(4) The authors associate this process with functional significance in cancer cell metastasis and neurological synapse formation, both of which involve the formation of filopodia. 

(5) The data are presented clearly, and their interpretation appropriately explains the context and significance of the findings. 

Weaknesses: 

(1) A better characterization of the nature of the small EV population is missing: 

It is unclear why the authors chose to proceed to quantitative mass spectrometry with the bands in the Coomassie from size-separated EV samples, as there are other bands present in the small EV lane but not the large EV lane. This is important to clarify because it underlies how they were able to identify THSD7A as a unique regulator of exosome-mediated filopodia formation. Is there a reason why the total sample fractions were not compared? This would provide valuable information on the nature of the small and large EV populations. 

We would like to clarify that there are two sets of proteomics data in the manuscript. The first was comparing bands from a Coomassie gel from two samples: small EVs and large EVs from B16F1 cells. In this proteomics experiment, we identified endoglin as present in small EVs, but not large EVs. For this experiment, we only sent 4 bands from the small EV lane, chosen based on their obvious banding pattern difference on the Coomassie gel.

In the second proteomics experiment, we used quantitative iTRAQ proteomics to compare small EVs purified from B16F1 control (shScr) and endoglin KD (shEng1 and shEng2) cell lines. In this experiment, we sent total protein extracted from small EV samples for analysis. So, these samples included the entire EV content, not just selected bands from a gel. In this experiment, we identified THSD7A as reduced in the shEng small EVs.

(2) Data analysis and quantification should be performed with increased rigor: 

a) Figure 1C - The optical and temporal resolution are insufficient to conclusively characterize the association between exosome secretion and filopodia. Specifically, the 10-second interval used in the image acquisitions is too close to the reported 20-second median time between exosome secretion and filopodia formation. Two-5 sec intervals should be used to validate this. It would also be important to correlate the percentage of filopodia events that co-occur with exosome secretion. Is this a phenomenon that occurs with most or only a small number of filopodia? Additionally, resolution with typical confocal microscopy is subpar for these analyses. TIRF microscopy would offer increased resolution to parse out secretion events. As the TIRF objective is listed in the Methods section, figure legends should mention which images were acquired using TIRF microscopy. 

We acknowledge that the frame rate naturally limits our estimates of the timing of filopodia formation after exosome secretion. We set out to show a relationship between exosome secretion and filopodia formation, based on their proximity in timing. While our data set shows a median time interval of 20 seconds, the true median could be between 10-30 seconds, based on our frame rate.  Regardless of the exact timing, our data show that exosome secretion is rapidly followed by filopodia formation events.

To address the question of the percentage of filopodia events that are preceded by exosome secretion, the reviewer is correct in stating that we might need TIRF microscopy to get an accurate calculation of this number.  Nonetheless, we will review our live imaging data for this experiment to determine if this calculation is possible. Again, we will be limited by the frame rate we used to capture the images, so we could possibly be missing secretion events taking place between the 10 second time intervals.  Regardless, for the secretion events that we visualized, we always observed subsequent filopodia formation.

No TIRF imaging was used in this manuscript.  A TIRF objective was used for selected neuron imaging (see methods); however, it was used for spinning disk confocal microscopy, not for TIRF imaging.  We will clarify this in the methods.

b) Figure 2 - It would be important to perform further analysis to concretely determine the relationship between exosome secretion and filopodia stability. Are secretion events correlated with the stability of filopodia? Is there a positive feedback loop that causes further filopodia stability and length with increased secretion? Furthermore, is there an association between the proximity of secretion with stability? Quantification of filopodia more objectively (# of filopodia/cell) would be helpful. 

Our data shows that manipulation of general exosome secretion, via Hrs knockdown, affects both de novo filopodia formation and filopodia stability (Fig 2g,h). Interestingly, knockdown of endoglin only affects de novo filopodia formation, while filopodia stability is unaffected (Fig 4g,h). These results suggest that filopodia stability is dependent upon exosome cargoes besides endoglin/THSD7A.  Such cargoes might include other extracellular matrix molecules, such as fibronectin. We previously showed that exosomes promote nascent cell adhesion and rapid cell migration, through exosome-bound fibronectin (Sung et al., Nature Communications, 6:7164, 2015). We also previously found that inhibition of exosome secretion affects the persistence of invadopodia, which are filopodia-dependent structures (Hoshino et al., Cell Reports, 5:1159-1168, 2013).  We agree that this is an interesting research direction, and perhaps future work could focus on exosomal factors that are responsible for filopodia persistence.

With regard to the way we plotted the filopodia data, we plotted the cancer cell data as filopodia per cell area so that it matched the neuron data, which was plotted as filopodia per 100 mm of dendrite distance. Since the neurons cannot be imaged as a whole cell, the quantification is based on the length of the dendrite in the image. We found that graphing the cancer cell data as filopodia per cell gave similar results as filopodia per cell area, as there were no significant differences in cell area between conditions and experiments. We plan to include a new supplementary figure showing the data in Figure 2 plotted as filopodia per cell to show that this quantification gives the same results.

c) Figure 6 - Why use different gel conditions to detect THSD7A in small EVs from B16F1 cells vs HT1080 and neurons? Why are there two bands for THSD7A in panels C and E? It is difficult to appreciate the KD efficiency in E. The absence of a signal for THSD7A in the HT1080 shEng small EVs that show a signal for endoglin is surprising. The authors should provide rigorous quantification of the westerns from several independent experimental repeats. 

Detection of THSD7A via Western blot was, unfortunately, not straightforward and simple. Due to the large size (~260 kDa) of THSD7A, its low level of expression in cancer cells, as well as the inconsistency of commercially available THSD7A antibodies, we had to troubleshoot multiple conditions.  We found that it was much easier to detect THSD7A in the human fibrosarcoma cell line HT1080 than in the mouse B16F1 cells, both in the cell lysates and in the small EVs. We were usually unable to detect THSD7A using these same conditions for the mouse melanoma B16F1 samples, but were successful using native gel conditions. We also detected THSD7A in rat primary neuron samples. All these samples were from different source organisms (human, mouse, rat) and from either cell lysates or extracellular vesicles, further complicating the analyses. Expression and maturation of THSD7A in these different cell types and compartments could involve different post-translational modifications, such as glycosylation, thus requiring different methods needed to detect THSD7A on Western blots and leading to different banding patterns. Based on our THSD7A trafficking data, we believe that in control cells, most of the THSD7A is getting trafficked and secreted via small EVs. As you can see in Figure 7A, the band for THSD7A in the shScr cell lysate is relatively light and also shows a double band similar to Figure 6E (both HT1080 samples).

With regard to the level of knockdown of THSD7A in the Western blot shown in Figure 6E, the normalized level is quantitated below the bands.  If you compare that quantitation to the filopodia phenotypes in the same panel, they are quite concordant.  Figures 7B and 7C show quantification of triplicate Western blots, highlighting the significant accumulation of THSD7A in shEng cell lysates, as well as significant small EV secretion of THSD7A in control and WT rescued conditions.

(3) The study lacks data on the cellular distribution of endoglin and THSD7A: 

a) Figure 6 - Is THSD7A expected to be present in the nucleus as shown in panel D (label D is missing in the Figure). It is not clear if this is observed in neurons. a Western of endogenous THSD7A on cell fractions would clarify this. The authors should further characterize the cellular distribution of THSD7A in both cell types. Similarly, the cellular distribution of endoglin in the cancer cells should be provided. This would help validate the proposed model in Figure 8. 

The image in figure 6D shows an HT1080 cell stained with phalloidin-Alexa Fluor 488 to visualize F-actin with or without expression of THSD7A-mScarlet.  In order to fully visualize the thin filopodia protrusions, the cellular plane of focus of the images for this panel was purposely taken at the bottom of the cell, where the cell is attached to the coverslip glass. Thus, we interpret the red signal across the cell body as THSD7A-mScarlet expression on the plasma membrane underneath the cell, not in the nucleus. The neuron images only include the dendrite portion of the neurons; therefore, there is no nucleus present in the neuronal images.

b) Figure 7 - Although the western blot provides convincing evidence for the role of endoglin in THSD7A trafficking, the microscopy data lack resolution as well as key analyses. While differences between shSCR and shEng cells are clear visually, the insets appear to be zoomed digitally which decreases resolution and interferes with interpretation. It would be crucial to show the colocalization of endoglin and THSD7A within CD63-postive MVE structures. What are the structures in Figure 7E shSCR zoom1? It would be important to rule out that these are migrasomes using TSPAN4 staining. More information on how the analysis was conducted is needed (i.e. how extracellular areas were chosen and whether the images are representative of the larger population). A widefield image of shSCR and shEng cells and DAPI or HOECHST staining in the higher magnification images should be provided. Additionally, the authors should quantify the colocalization of external CD63 and mScarlet signals from many independently acquired images (as they did for the internal signals in panel F). Is there no external THSD7A signal in the shEng cells? 

The images for Figure 7E were taken with high resolution on a confocal microscope.  Insets for Figure 7E were zoomed in so that readers could see the tiny structures.  Zoom 1 in Figure 7E shows areas of extracellular deposition. In these areas, we can see small punctate depositions that are positive for CD63 and/or THSD7A-mScarlet. Our interpretation of this staining is that the cells are secreting heterogeneous small EVs that are then attached to the glass coverslip. The images and zooms in Fig 7E were chosen to be representative and indeed reveal that there is more extracellular deposition of THSD7A-mScarlet outside the control shScr cells compared to the shEng cells, consistent with more export of THSD7A into small EVs from shScr cells when compared to those of shEng cells (Fig 7A,B). However, we did not quantify this difference, as these experiments were conducted with transient transfection of THSD7A-mScarlet and it is challenging to determine which cell the extracellular THSD7A-mScarlet came from, complicating any quantitative analysis on a per-cell basis.  Quantification of internal THSD7A localization is much more straightforward in this experimental regime.  Indeed, in Figure 7F we assessed internal colocalization of THSD7A-mScarlet and CD63, which we obtained by choosing only cells that were visually positive for THSD7A-mScarlet in each transient transfection and omitting all extracellular signals. Quantifying the extracellular colocalization of THSD7A and CD63 could certainly be a future direction for this project and would require establishing cells that stably express THSD7A-mScarlet.

Author response:

We thank the reviewers for their thoughtful feedback and valuable comments. We plan to fully address their concerns by including the following experiments and analyses:

Reviewer 1 suggested exploring data scaling trends for encoding models, as successful scaling would justify larger datasets for language ECoG studies. To estimate scaling effects, we will develop encoding models on subsets of our data.

Reviewer 2 expressed uncertainty about the baseline for model-brain correlation and recommended adding control LLMs with randomly initialized weights. In response, we will generate embeddings using untrained LLMs to establish a more robust baseline for encoding results.

Reviewer 2 also proposed incorporating control regressors such as word frequency and phonetic features of speech. We will re-run our modeling analysis using control regressors for word frequency, 8 syntactic features (e.g., part of speech, dependency, prefix/suffix), and 3 phonetic features (e.g., phonemes, place/manner of articulation) to assess how much these features contribute to encoding performance.

Reviewer 3 raised concerns that the “plateau in maximal encoding performance” was actually a decline for the largest models. We will add significance tests in Figure 2B to clarify this issue.

Reviewer 3 also noted that in Supplementary Figure 1A, the decline in encoding performance was more pronounced when using PCA to reduce embedding dimensionality, in contrast to the trend observed when using ridge regression. To address this, we will attempt to replicate the observed scaling trends in Figure 2B using PCA combined with OLS.

Additionally, we will provide a point-by-point response and revise the manuscript with updated analyses and figures in the near future.

Author response:

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

Reviewer #1 (Public Review): 

Summary: 

Colomb et al have further explored the mechanisms of action of a family of three immunodulatory proteins produced by the murine gastrointestinal nematode parasite Heligmosomoides polygyrus bakeri. The family of HpARI proteins binds to the alarmin interleukin 33 and depending on family members, exhibits differential activities, either suppressive or enhancing. The present work extends previous studies by this group showing the binding of DNA by members of this family through a complement control protein (CCP1) domain. Moreover, they identify two members of the family that bind via this domain in a non-specific manner to the extracellular matrix molecule heparan sulphate through a basic charged patch in CCP1. The authors thus propose that binding to DNA or heparan sulphate extends the suppressive action of these two parasite molecules, whereas the third family member does not bind and consequently has a shorter half-life and may function via diffusion. 

Strengths: 

A strength of the work is the multifaceted approach to examining and testing their hypotheses, using a well-established and well-defined family of immunomodulatory molecules using multiple approaches including an in vivo setting. 

Weaknesses: 

There are a few weaknesses of the approach. Perhaps some discussion and speculation as to how these three family members might operate in concert during Heligmosomoides polygyrus bakeri infection would help place the biology of these molecules in context for the reader, e.g. when and where they are produced. 

We agree that the roles of these proteins during infection requires further study and is not fully elucidated in infection here. We have added further discussion to the manuscript on their potential roles during infection (track changes manuscript, lines 277 – 283).

Reviewer #2 (Public Review): 

Summary: 

Colomb et al. investigated here the heparin-binding activity of the HpARI family proteins from H. polygyrus. HpARIs bind to IL-33, a pleiotropic cytokine, and modulate its activities. HpARI1/2 has suppressive functions, while HpARI3 can enhance the interaction between IL-33 and its receptor. This study builds upon their previous observation that HpARI2 binds DNA via its CCP1 domain. Here, the authors tested the CCP1 domain of HpARIs in binding heparan sulfate, an important component of the extracellular matrix, and found that 1/2 bind heparan, but 3 cannot, which is related to their half-lives in vivo. 

Strengths: 

The authors use a comprehensive multidisciplinary approach to assess the binding and their effects in vivo, coupled with molecular modeling. 

Weaknesses: 

(1) Figure 1C should include Western. 

We apologise for this oversight, and now include an uncropped western blot image as a Figure 1, Figure Supplement 1.

(2) Figure 1E: Why does HpARI1 stop binding DNA at 50%? 

It is currently unclear why HpARI1 does not bind to all DNA in the EMSA assay, however this was our repeated finding. With our revised findings we can now state definitively that HpARI1 has a lower affinity for HS compared to HpARI2, and in each of our assays (EMSA (Fig 1D-E), size exclusion chromatography (Fig 4A), HS-bead pull-down (Fig 4B), lung cell surface binding (Fig 4C) and ITC (Fig 4D)) HpARI1 always shows a weaker response compared to HpARI2. We hypothesise that HpARI1 binds more weakly to DNA/HS to allow it to diffuse further from the site of deposition, but we have yet to demonstrate this during infection. We add further discussion of this point (track changes manuscript, lines 262 – 266).

(3) ITC binding experiment with HpARI1? Also, the ITC results from HpARI2 do not seem to saturate, thus it is difficult to really determine the affinity. 

We have now included HpARI1-HS ITC, and re-ran the HpARI2 experiment to saturation (Fig 4D-E).

(4) It would be helpful to add docking results from HpARI1. 

We have now included HpARI1-HS docking, in Figure 5B.

(5) Some conclusions are speculative and need to remain in the Discussion. e.g.: a) That HpARI3 may be able to diffuse farther 

We have rewritten these points to remove the speculation on localisation from the abstract (lines 18-19) and introduction (line 78).

b) That DNA/HS may trap HpARI1/2 at the infection site. 

Likewise, these points have been rewritten in the abstract and introduction as above, and we have made it clearer that this is a model that we are proposing in the discussion (line 277-283).

Reviewer #1 (Recommendations For The Authors): 

The paper is well-written and the data well-presented. I have one small comment that the authors may like to consider. In the discussion, second paragraph, line 17, perhaps, "evolved" rather than "developed". 

Thank you for this suggestion, we have made this change (line 248).

Author response:

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

Reviewer #1 (Recommendations For The Authors):

To hopefully contribute to more strongly support the conclusions drawn by the authors, I am including a series of concerns regarding the manuscript, as well as some suggestions that could be useful to address these issues:

(1) The main results of this study derive from the use of auxin-inducible degron (AID)-tagged proteins. Despite the great advantages of the AID strategy to conditionally deplete proteins, the AID tag can affect the normal function of a protein. In fact, some of the AID-labeled DDC components generated in this work are shown to be hypomorphic. Hence, the manuscript would have benefited from the additional confirmation of some of the observations using a different way to eliminate the proteins (e.g., temperature-sensitive mutants).

Most ts mutants are also hypomorphic; hence we don’t see there is much advantage to their use. The addition of the AID to these proteins alone does not interfere with the ability to sustain checkpoint arrest as demonstrated in Figure S1. Instead we found that by overexpressing Rad9-AID we could demonstrate that inactivating Rad9 after 15 h behaved the same way as the inactivation of Ddc2, significantly strengthening our finding that the DDC checkpoint becomes dispensable while the SAC takes over. 

(2) In cells depleted of Rad53-AID, the deletion of CHK1 stimulates an earlier release from a mitotic arrest induced by two DSBs (Figures 2D and 3C). Likewise, the authors claim that a faster escape from the cell cycle block can also be observed when upstream factors such as Ddc2, Rad9, or Rad24 are depleted in the absence of CHK1 (Figures 2A-C and Figures 3D-F). However, this earlier release from the cell cycle arrest, if at all, is only slightly noticeable in a Rad9-AID background (Figures 2B and 3E). In this sense, it is also worth pointing out that Rad9-AID chk1Δ (Figure 3E) and Rad24-AID chk1Δ (Figure 3F) cells were only evaluated up to 7 h, while in all other instances, cells were followed for 9 h, which hinders a fair assessment of the differences in the release from the cell cycle arrest.

As noted above, we have now been able to examine Rad9 over the long-time frame.

(3) Although only 25% of the cells depleted for Dun1 remained in G2/M arrest 7 h following the induction of two DSBs, it is shocking that Rad53 was nonetheless still phosphorylated after the cells had escaped the cell cycle blockage (Figure 4A).

This persistence of Rad53 phosphorylation is also seen with the inactivation of Mad2, allowing escape in spite of continued Rad53 phosphorylation.

(4) Generation of Rad9-AID2 and Rad24-AID2 strains did not fully restore the function of these proteins, since most cells had adapted 24 h after induction of two DSBs (Figure S1C). Nonetheless, Rad9-AID2 and Rad24-AID2 are still likely more stable than their AID counterparts, and hence the authors could have instead used the AID2 proteins for the experiments in Figure 2 to better evaluate the role of Rad9 and Rad24 in the maintenance of the DDC-dependent arrest.

We note again that we have found a way to study Rad9 up to 24 h. 

(5) Deletion of BFA1 has been shown to promote the escape from a cell cycle arrest triggered by telomere uncapping (Wang et al. 2000, Hu et al. 2001, Valerio-Santiago et al. 2013). Likewise, while cells carrying the cdc5-T238A allele cannot adapt to a checkpoint arrest induced by one irreparable DSB, BFA1 deletion rescues the adaptation defect of this mutant CDC5 allele (Rawal et al., 2016). The authors show how, using AID-degrons of Bfa1 and Bub2, that only Bub2, but not Bfa1, is required to maintain a prolonged cell cycle arrest after the induction of two DSBs. To reinforce this point, and as shown for mad2Δ cells (Figure S6A), the authors could perform a complete time course using both the Bfa1-AID and a bfa1Δ mutant to demonstrate that they do indeed show the same behavior in terms of the adaptation to a two DSB-induced cell cycle arrest.

We thank the reviewer for noting these other instances where bfa1D promoted an escape from arrest. We tested a 2-DSB bfa1 deletion, data has been added to Figure S9E-F. We did not observe a difference in the percentage of cells escaping arrest between the 2-DSB bfa1 deletion and the 2-DSB BFA1-AID strains.

(6) Bypass or adaptation of a checkpoint-induced cell cycle arrest in S. cerevisiae often leads to cells entering a new cell cycle without doing cytokinesis and, hence, to the accumulation of rebudded cells. However, the experiments shown in the manuscript only account for G1 or budded cells with either one or two nuclei. Do any of the mutants show cytokinesis problems and subsequent rebudding of the cells? If so, this should have been also noted and quantified in the corresponding assays.

In the cases we have studied we have not seen instances where the cells re-bud without completing mitosis (at least as assessed by the formation of budded cells with two distinct DAPI staining masses). In the morphological assays we have done, we score the continuation of the cell cycle by the appearance of multiple buds, G1, and small budded cells. In our adaptation assays when cells escaped G2/M arrest they formed microcolonies indicating no short-term deficiency in cell division.

(7) The location of the DSB relative to the centromere of a chromosome seems to be a factor that determines the capacity of the SAC to sustain a prolonged cell cycle arrest. The authors discuss the possibility that the DSB could somehow affect the structure of the kinetochore. Did they evaluate whether Mad1 or Mad2 were more actively recruited to kinetochores in those strains that more strongly trigger the SAC after induction of the DSBs?

We have not attempted to follow Mad1/2 recruitment. ChIP-seq could be used to monitor Mad1/2 localization at the 16 centromeres in response to DSBs and the spread of g-H2AX across the centromere. Our previous data showed that g-H2AX could spread across the centromere region and could create a change that would be detected by Mad1/2.  This change does not, however, affect the mitotic behavior of a strain in which the H2A genes have been modified to the possibly phosphomimetic H2A-S129E allele.

(8) The authors could speculate in the discussion about the reasons that could explain why the DDC is required for the maintenance of checkpoint arrest at early stages but then becomes dispensable for the preservation of a prolonged cell DNA DSB-induced cycle arrest, which is instead sustained at later stages by the SAC.

Our suggestion is that cells would have adapted, but modification of the centromere region engages SAC.

Finally, some minor issues are:

(1) The lines in the graphs that display the results from adaptation assays (e.g., Figures 1B and 1E) or cell and nuclear morphology (e.g., Figures 1D and 1G) are too thick. This makes it sometimes difficult to distinguish the actual percentages of cells in each category, particularly in the experiments monitoring nuclear division.

Fixed

(2) While both the adaptation assay and the analysis of nuclear division in Figures 1E and 1G, respectively, show a complete DDC-dependent arrest at 4h, the Western blot in Figure 1F suggests that Rad53 is not phosphorylated at that time point. Do these figures represent independent experiments? Ideally, the analysis of cell budding and nuclear division, which is performed in liquid cultures, and the Western blot displaying Rad53 phosphorylation should correspond to the same experiment.

Cell budding in liquid cultures and adaptation assays were performed in triplicate with 3 biological replicates and the collective results are shown in each graph showing the percentage of large-budded cells. Western blot samples were collected in each liquid culture experiment. The western blot in 1G is a representative western blot.

(3) It is somewhat confusing that the blots for the proteins are not displayed in the same order in Figures 2A (Rad53 at the top) and 2B or 2C (Rad53 in the middle).

Fixed.  We place Rad53 – the relevant protein - at the top.

Reviewer #2 (Recommendations For The Authors):

(1) Yeast with the two breaks responds to DNA damage checkpoint (DDC) until sometimes 4-15 h post DNA damage. Since the auxin-induced degradation does not completely deplete all the tagged proteins in cells, the results should be more carefully considered and not to interpret if the checkpoint entry or maintenance depends on each target protein's ability to induce Rad53 phosphorylation. It should be theoretically possible if checkpoint maintenance requires only a modest amount of checkpoint factors especially because the experiments involve the induction of one or two DSBs. The low levels of DDC factors may be insufficient for Rad53 activation but could still be effective for cell cycle arrest. Indeed, the Haber group showed that the mating type switch did not induce Rad53 phosphorylation but still invoked detectable DNA damage response. To test such possibilities, the authors might consider employing yet another marker for DDC such as H2A or Chk1 phosphorylation besides Rad53 autophosphorylation. Alternatively, the authors might check if auxin-induced depletion also disrupts break-induced foci formation for checkpoint maintenance or their enrichment at DNA breaks using ChIP assays at various points post-damage.

DAPI staining of Ddc2-AID cells show that when IAA is added 4 h after DSB induction (Figure S3A), cells escape G2/M arrest as evidenced by the increase in large-budded cells with 2 DAPI signals, small budded cells, and G1 cells. Overexpression of Ddc2 can sustain the checkpoint past 24 h, but without SAC proteins like Mad2 they will eventually adapt (Figure S6B).

That Rad9-AID or Rad24-AID in the absence of added auxin (but in the presence of TIR1) is unable to sustain arrest suggests to us that low levels of Rad9 or Rad24 are not sufficient to maintain arrest.  As the reviewer notes, normal MAT switching doesn’t cause Rad53 phosphorylation or arrest, though early damage-induced events such as H2A phosphorylation do occur.  But our point is that Rad9 or Ddc2 is needed to maintain arrest only up to a certain point, after which they become superfluous and a different checkpoint arrest is imposed. At that point apparently a low level of these proteins plays no obvious role.

(2) It is interesting that DDC no longer responds to the damage signaling after 15 h of DSB-induced prolonged checkpoint arrest after two DNA double-strand breaks. Is this also applicable to other adaptation mutants? The results might improve the broad impact of the current conclusions. It is also possible that the transition from DDC to SPC depends on simply the changes in signaling or in part due to the molecular changes in the status of DNA breaks or its flanking regions. Indeed, the proposed model suggests that the spreading of H2A phosphorylation to centromeric regions induces SAC and thus mitotic arrest. The authors could measure H2A phosphorylation near the centromere using ChIP assays at various intervals post-DNA damage. It is particularly interesting if depletion of Ddc2 at 15 h post DNA damage does not alter the level of H2A phosphorylation at or near centromere.

Our previous data have suggested that the involvement of the SAC in prolonging DSB-induced arrest involved post-translational modification of centromeric chromatin such as the Mec1- and Tel1-dependent phosphorylation of the histone H2A (Dotiwala). In budding yeast there is also a similar DSB-induced modification of histone H2B (Lee et al.). To ask if there is an intrinsic activation of the SAC if the regions around centromeres were modified by checkpoint kinase phosphorylation, we examined cell cycle progression in strains in which histone H2A or histone H2B was mutated to their putative phosphomimetic forms (H2A-S129E and H2B-T129E).  As shown in Figure S11, there was no effect on the growth rate of these strains, or of the double mutant, suggesting that cells did not experience a delay in entering mitosis because of these modifications. We note that although histone H2A-S129E is recognized by an antibody specific for the phosphorylation of histone H2A-S129, the mutation to S129E may not be fully phosphomimetic. 

(3) It is puzzling why Rad9-AID or Rad24-AID are proficient for DDC establishment but cannot sustain permanent arrest in the two break cells. It appears Rad53 phosphorylation for DDC is weaker in cells expressing Rad9-AID or Rad24-AID according to Fig.2B and C even though their protein level before IAA treatment is still robust. This might also explain why the results of depleting Rad53 and Rad9 are very different. It also raises concern if the effect of Rad24 depletion on checkpoint maintenance is in part due to the weaker checkpoint establishment. It might be necessary to use the AID2 system to redo Rad24 depletion to exclude such a possibility.

We believe that the AID mutants are very sensitive to the low level of IAA present in yeast.  The instability of the protein is entirely dependent on the TIR1 SCF factor, so the proteins themselves are not intrinsically defective; they are just subject to degradation.  Overexpressing Rad9 allowed us to evaluate its role at late time points. 

(4) It is intriguing that the switch from DDC to SAC might take place at around 12 h when yeasts with a single unrepairable break ignore DDC and resume cell cycling (so-called "adaptation"). Since 4h and 15h are far apart and the transition point from DDC to SAC likely takes place between these two points, it will be very helpful to analyze and compare cell cycle exit after 24 h by treating IAA at multiple points between 4-15h.

When we add IAA to Mad2-AID and Mad1-AID 4 h after DSB induction, cells remain arrested for up to 12 h after DSB induction. At 15 h cells begin to exit checkpoint arrest indicating that the handoff of checkpoint arrest must occur between 12 to 15 h after DSB induction. If we degraded DNA damage checkpoint proteins at any point before Mad2, Mad1, and Bub2 begin to contribute to checkpoint arrest, then arrested cells will likely adapt in a similar manner to when IAA was added 4 h after DSB induction.

(5) Some of the Western blot quality is poor. For instance, in Figure 6C, Mad1-AID level after IAA addition is not compelling especially because the TIR level (the loading control) is also very low.

In Figure 6C, while the relative levels of TIR1 are similar in the IAA treated and untreated samples, there is no detectable amount of Mad1-AID in the IAA treated samples indicating that Mad1-AID was successful degraded with the AID system.

(6) Fig. 8 is complex. It might be helpful to define the different types of arrows in the figure. The legend also has a spelling error, Rad23 should be Rad24.

We’ve defined what each arrow means in the legend and corrected the spelling error in the figure legend.

Reviewer #3 (Recommendations For The Authors):

Major concerns:

Much of the manuscript states that two unrepairable DSBs lead to a long and severe G2/M arrest. Two main cytological approaches are used to make this statement: bud size and number on plates after micromanipulation (microcolony assay), and cell and nuclear morphology in liquid cultures. While the latter gives a clear pattern that can be assigned to a G2/M block as expected by DDC, i.e. metaphase-like mononucleated cells with large buds, the former can only tell whether cells eventually reach a second S phase (large budded cells on the plate can be in a proper G2/M arrest, but can also be in an anaphase block or even in the ensuing G1). The authors always performed the microcolony assay, but there are several cases where the much more informative budding/DAPI assay is missing. These include Dun1-aid and others, but more importantly chk1D and its combinations with DDC proteins. Incidentally, for the microcolony assay, it is more accurate to label the y-axis of the corresponding graphs (and in the figure legends and main text) with something like "large budded cells"; "G2/M arrested cells" is misleading.

Figures have been updated to more accurately reflect what we are measuring.

The results obtained with the Bfa1/Bub2 partner are intriguing. These two proteins form a complex whose canonical function is to prevent exit from mitosis until the spindle is properly aligned, acting in a distinct subpathway within the SAC that blocks MEN rather than anaphase onset. The data presented by the authors suggest that, on the one hand, both SAC subpathways work together to block the cell cycle. However, why does canonical SAC (Mad1/Mad2) inactivation not lead to a transition from G2/M (metaphase-like) arrested cells to anaphase-like arrest maintained by Bfa1-Bub2? Since Bfa1-Bub2 is a target of DDC, is it possible that DDC knockdown also inactivates this checkpoint, allowing adaptation? On the other hand, can the authors provide more data to confirm and strengthen their claim of a Bfa1-independent Bub2 role in prolonged arrest? Perhaps long-term protein localization and PTM changes. Bub2-independent roles for Bfa1 have been reported, but not vice versa, to the best of my knowledge.

In the mitotic exit network Bfa1/Bub2 prime activation of the pathway by bringing Tem1 to spindle pole bodies. Phosphorylation of Bfa1 causes Tem1 to be released and phosphorylate Cdc5 to trigger exit by MEN. It has been shown that DNA damage, in a cdc13-1 ts mutant, phosphorylates Bfa1 in a Rad53 and Dun1 dependent manner. This phosphorylation of Bfa1 could release Tem1 and prime cells to exit checkpoint arrest when cells pass through anaphase. Looking at Tem1 localization to spindle pole bodies and interactions with Bfa1/Bub2 in response to DNA damage might give insight into why cells don’t experience an anaphase-like arrest when they are released by either deactivation of the DNA damage checkpoint or SAC.

We have previously shown that a deletion of bub2 in a 1-DSB background shortens DSB-induced checkpoint arrest. Deletion of bfa1 in a 2-DSB background showed ~80-70% of cells stuck in a large-budded state as measured through an adaptation assay tracking the morphology of G1 cells on a YP-Gal plate and DAPI staining. Deletion or degradation of bfa1 might not release cells from arrest because the Mad2/Mad1 prevent cells from transitioning into anaphase. Our DAPI data for Bub2-AID shows an increase in cells with 2 DAPI signals (transition into anaphase) and small budded cells indicating that degradation of Bub2 is releasing cells into anaphase and allowing cells to complete mitosis.

Further suggestions:

It would be richer if authors could provide more than one experimental replicate in some panels (e.g., S1A,B; S4A; and S6B).

S1C confirms that Rad9-AID and Rad24-AID will adapt by 24 h even with the point mutant TIR1(F74G) which has lower basal degradation than TIR1. S4A has been updated with additional experimental replicates. The 48 h timepoint after DSB induction was to show the importance of Mad2 even when Ddc2 is overexpressed.

Figure 1: Rearrange figure panels when they are first mentioned in the text. For example, it makes more sense to have the plate adaptation assay as panel B for both 1-DSB and 2-DSB strains, budding plus DAPI as panel C, and Rad53 as panel D.

These figures have been rearranged in the order that they are mentioned in the paper.

Figure 5: Correct Ph-5-IAA in the Rad53 WBs (it should be 5-Ph-IAA).

This has been corrected.

Figure S2: The straight line under the "+IAA" text box is misleading. I think it should also cover the "-2" time point, right? Also, check the figure legend. Information is missing and does not correspond to the figure layout.

This has been corrected.

Figure S3: Perhaps "Cell cycle profile as determined by budding and DAPI staining" is a better and more accurate legend title.

The legend title has been updated to “Cell cycle profile as determined by budding and DAPI staining in Ddc2-AID and Rad53-AID mutants ± IAA 4 h after galactose.”

Figure S5: Detection of both Rad53 and Ddc2 in the same blot could lead to misinterpretation as hyperphosphorylated Rad53 appears to coincide with Ddc2 migration.

Figure S5A-B are representative western blots where Rad53 was probed to show activation of the DNA damage checkpoint by Rad53 phosphorylation. When measuring the relative abundance of Ddc2 we did not probe all blots for Rad53.

Table S1: Include the post-hoc test used for comparisons after ANOVA.

A Sidak post-hoc test was used in PRISM for the one-way ANOVA test. PRISM listed the Sidak post-hoc test as the recommended test to correct for multiple comparisons. A column has been added to S. Table 1 to show which post-hoc test was used.

Page 10, line 4: The putative additive effect of chk1 knockout with Dun1 depletion should also be compared to chk1 alone (in Figure 3A).

We address the additive effect of chk1 knockout with Dun1-AID depletion in a later section on Page 11, line 6. Since we had not explored possible effects from downstream targets of Rad53 for prolonging checkpoint arrest when Rad53 was depleted, we did not mention the effect of the chk1 knockout on Dun1 depletion.

Page 14, second paragraph, line 4: "Figure 6A-D", is it not?

Figure S6A is measuring checkpoint arrest in a deletion of mad2 in a 2-DSB strain. Figure 6A-D shows how degradation of Mad2-AID and Mad1-AID after the handoff of arrest causes cells to exit the checkpoint in a Rad53 independent manner.

Author response:

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

Reviewer #1 (Public Review):

Summary: The authors investigated the function of Microrchidia (MORC) proteins in the human malaria parasite Plasmodium falciparum. Recognizing MORC's implication in DNA compaction and gene silencing across diverse species, the study aimed to explore the influence of PfMORC on transcriptional regulation, life cycle progression and survival of the malaria parasite. Depletion of PfMORC leads to the collapse of heterochromatin and thus to the killing of the parasite. The potential regulatory role of PfMORC in the survival of the parasite suggests that it may be central to the development of new antimalarial strategies.

Strengths: The application of the cutting-edge CRISPR/Cas9 genome editing tool, combined with other molecular and genomic approaches, provides a robust methodology. Comprehensive ChIP-seq experiments indicate PfMORC's interaction with sub-telomeric areas and genes tied to antigenic variation, suggesting its pivotal role in stage transition. The incorporation of Hi-C studies is noteworthy, enabling the visualization of changes in chromatin conformation in response to PfMORC knockdown.

We greatly appreciate the overall positive feedback and cognisense of our efforts. Our application of CRISPR/Cas9 genome editing tools coupled with complementary cellular and functional approaches shed light on the importance of _Pf_MORC in maintaining chromatin structural integrity in the parasite and highlights this protein as a promising target for novel therapeutic intervention.

Weaknesses: Although disruption of PfMORC affects chromatin architecture and stage-specific gene expression, determining a direct cause-effect relationship requires further investigation.

Our conclusions were made on the basis of multiple, unbiased molecular and functional genomic assays that point to the relevance of the _Pf_MORC protein in maintaining the parasite’s chromatin landscape. Although we do not claim to have precise evidence on the step-by-step pathway to which _Pf_MORC is involved, we bring forth first-hand evidence of its role in heterochromatin binding, gene-regulation and its association with major TFs as well as chromatin remodeling and modifying enzymes. We however agree with the comment regarding the lack of direct effects of _Pf_MORC KD and have since provided additional evidence by performing ChIP-seq experiments against H3K9me3 and H3K9ac during KD. Our new results are presented in Fig. 5. We showed that the level of H3K9me3 decreased significantly during _Pf_MORC KD.

Furthermore, while numerous interacting partners have been identified, their validation is critical and understanding their role in directing MORC to its targets or in influencing the chromatin compaction activities of MORC is essential for further clarification. In addition, the authors should adjust their conclusions in the manuscript to more accurately represent the multifaceted functions of MORC in the parasite.

Validation of the identified interacting partners is indeed critical and essential to understanding their role in directing MORC to its targets. Our protein pull down experiments have been done using several biological replicates. Several of the interacting partners have also been identified and published by other labs and collaborators. To confirm our results, we completed a direct comparison of our work with previous published work. Results have now been incorporated into the revised manuscript to confirm the identified interacting partners and the accuracy of the data we obtained in our experiment. Molecular validation of novel proteins identified in our protein pull down requires generation of tagged lines and may take a few more years but will be submitted for publication in a follow up manuscript.

Reviewer #2 (Public Review):

Summary: This paper, titled "Regulation of Chromatin Accessibility and Transcriptional Repression by PfMORC Protein in Plasmodium falciparum," delves into the PfMORC protein's role during the intra-erythrocytic cycle of the malaria parasite, P. falciparum. Le Roch et al. examined PfMORC's interactions with proteins, its genomic distribution in different parasite life stages (rings, trophozoites, schizonts), and the transcriptome's response to PfMORC depletion. They conducted a chromatin conformation capture on PfMORC-depleted parasites and observed significant alterations. Furthermore, they demonstrated that PfMORC depletion is lethal to the parasite.

Strengths: This study significantly advances our understanding of PfMORC's role in establishing heterochromatin. The direct consequences of the PfMORC depletion are addressed using chromatin conformation capture.

We appreciate the Reviewer’s comments and reflection on the importance of our work.

Weaknesses: The study only partially addressed the direct effects of PfMORC depletion on other heterochromatin markers.

Here again, we agree with the reviewer’s comment and have performed additional experiments to delve deeper into the multifaceted roles of _Pf_MORC. We have performed additional ChIP-sequencing analysis on _Pf_MORC depleted conditions focusing on known heterochromatin and euchromatin markers H3K9me3 and H3K9ac respectively. We hope our new results presented in figure 5 will shed light on the more direct implications of _Pf_MORC on heterochromatin and gene silencing.

Reviewer #1 (Recommendations For The Authors):

Suggestions for improved or additional experiments, data or analyses.

• Why does MORC, which was used in the pull-down, seem to be only minimally enriched in the volcano plot, while a series of proteins (marked in red) and AP2 (highlighted in green) are enriched with log2 fold changes exceeding 15?

We apologize for the confusion. MORC was detected with the highest number of peptides (97 and 113) and spectra (1041 and 1177) confirming the efficiency of our pull-down. However, considering the relatively large size of the MORC protein (295kDa) and it weak detection in the control (5 and 7 peptides; 16 and 43 spectra), the Log2 FoldChange and Z-statistic after normalization are minimal compared to smaller proteins that were not identified in the control samples.

Additionally, can you explain why these proteins appear to be enriched at the same fold? 

We can postulate that these proteins form a complex with a ratio of 1:1. Two of these three proteins are described to interact with MORC in several publications, supporting a strong interaction between them.

Variations in the interactome could result from the washing buffer's stringency.

We agree that the IP conditions could affect the detection of the interactome as well as the parasite stage used. As indicated below, the overlap with previous publications and the presence of AP2 TFs and chromatin remodelers strongly support our results.

It would be highly appropriate for the authors, similar to the co-submitted article (Maneesh Kumar Singh et al.), to present their mass spectrometry data in relation to previous purifications in Plasmodium (Bryant et al. 2020; Subudhi et al. 2023; Hillier et al. 2019) and also in Toxoplasma (Farhat et al. 2020). It would be good if authors could also put their results into perspective in light of the following pre-prints:

We agree with the reviewer’s comment. In this revised manuscript, we compared our IP-MS data to previous published manuscripts. Key proteins including the AP2-P (PF3D7_1107800) and HDAC1 were indeed identified in several experiments validating our initial findings of the formation of large complexes with MORC. However, it’s important to highlight that the MORC protein was not used as the bait protein in previously published papers, and thus some discrepancies can be observed.

Given the tendency of MORCs to form multiple complexes with AP2 factors, have you explored whether specific AP2s are conserved between Plasmodium and Toxoplasma, within the phylum?

P. falciparum encodes for 27 putative AP2s, while T. gondii has over 60 AP2s, making direct comparison challenging. Some Plasmodium AP2s have multiple counterparts in T. gondii and typically conservation is limited to the AP2 binding domains. Attempts to identify sequence homology among AP2s and the regions of conservation have been performed (PMID: 30959972, PMID: 30959972, PMID: 16040597). Although this information would provide interesting insight, we believe exploring this topic at this time would diverge from our primary objectives. It would be more appropriate to address this in future studies.

Could this conservation be identified either through phylogenetic means or by using tools such as AlphaFold, especially considering not just the AP2 domains but also any existing ACDC domains?

Although this may reveal important information regarding the association between MORC proteins and AP2 domains, we believe investigating the conservation between AP2 across apicomplexan parasites may prove too challenging and is beyond the scope of this work.

Most of the genes are depicted without their immediate surroundings (Fig. 2d and Fig S2c, d). For instance, the promoter region of AP2g is not shown (Fig. 2d). It is therefore very challenging to determine the presence or absence of MORC upstream or downstream; considering that this factor, which can create DNA loop protrusions, might bind at a distance from the genes in question.

All gene coverage plots, including AP2-G, show 500 bp up- and downstream of the displayed gene. We have modified our figure legends to make sure that this information is provided.

Upon examining Figure S3, it is evident that the authors have indicated a decline in PfMORC expression, represented as percentages over two unique time frames. The methodology behind this quantification remains ambiguous. It's essential for the authors to specify whether normalization was done using a loading control. As a benchmark, Singh et al. (2021) in their Figure 4 transparently used GAPDH as a loading control and included an untreated sample in their western blot analysis.

We thank the Reviewer for bringing this to our attention. Our initial quantification was performed using ImageJ. To address the Reviewer’s comment, we have reperformed the experiment. Our quantitative analysis was performed through Bio-Rad ImageLab software using aldolase expression as a loading control (50% of the MORC loading). This information has now been incorporated into the supplementary figures (Figure S3).

There's a striking observation that, despite significant degradation of PfMORC (as depicted in Figures S1 and S3), only the upper band in the western blot diminishes. This inconsistency needs addressing, as it can raise questions about the interpretation of the results.

We agree with the reviewer's comment. We experienced some challenges upon performing a Western Blot on such a large protein (295kDa). Our initial attempts required long exposure that may have highlighted non-specific signals of smaller proteins. To address the reviewer’s comment, we have performed the experiment one more time and made necessary changes to our WB protocol. Our new result better reflects the expected down regulation of _Pf_MORC. These changes have been incorporated to our manuscript and Fig S3.

Recommendations for improving the writing and presentation.

MORC KD quantification and consistency with previous findings (Figure S3): When comparing their results with those from another study (Singh et al. 2021), it's critical to ensure that the experimental conditions, especially the methodology for KD and the quantification of protein levels, are similar. If not, a direct comparison might be misleading.

We greatly appreciate the suggestions and have made efforts to redesign the MORC KD quantifications according to the reviewer’s recommendations.

While the manuscript mentions the level of KD, it does not delve into the functional consequences of such a decrease in protein levels. It would be of interest to understand how this level of KD affects the parasite's biology, especially in the context of the paper's main findings.

We have addressed this question by looking at the changes in chromatin structure in WT versus KD parasites upon atc removal. We have also validated this initial result by designing an additional ChIP-seq experiment against histone marks in WT versus KD parasites upon atc removal. Our findings showed a significant downregulation in H3K9me coverage in heterochromatin regions, specifically in genes associated with antigenic variation and invasion genes. These findings suggest that PfMORC regulates at least partially gene silencing and chromatin arrangements. The manuscript has been edited accordingly. 

Concluding page 5, the authors present an interpretation of their findings that suggests a multi-faceted role of PfMORC in regulating stage-specific gene families, particularly the gametocyte-related genes and merozoite surface proteins. While the narrative they present is intriguing, several concerns arise:

Over-reliance on correlation: The authors draw a direct line between the levels of PfMORC binding and the function of these genes in the parasite's life cycle. However, a mere correlation between PfMORC binding and stage-specific gene activity does not necessarily imply causation. They would need to provide experimental evidence showing that manipulation of PfMORC levels directly impacts these genes' expression.

We agree with the reviewer's comment. We have however partially addressed this issue by comparing our ChIP-seq, RNA-seq and Hi-C experiments. We concluded that several of the transcriptional changes observed were due to an indirect effect of PfMORC KD and were most likely induced by a cell cycle arrest and partial collapse of the chromatin structure. The collapse of the heterochromatin structure was validated using our Hi-C experiment. To further address additional concerns the review’s had, we have included additional ChIP-seq experiments targeting histone marks to confirm our initial hypothesis. Result of this additional experiment has been incorporated in the revised version of the manuscript.

Ambiguity surrounding "low levels" and "high levels": The terms "low levels" and "high levels" of PfMORC binding are qualitative and could be subject to interpretation. Without quantification or a clear benchmark, these descriptions remain vague.

We agree with the reviewers that the terms "low levels" and "high levels" of PfMORC binding are qualitative and could be subject to interpretation. We have however quantified our change in DNA binding using normalized reads (RPKM). In trophozoite and schizont stages, most of the genes contain a mean of <0.5 RPKM normalized reads per nucleotide of Pf_MORC binding within their promoter region, whereas antigenic gene families such as _var and rifin contain ~1.5 and 0.5 normalized reads, respectively (Fig. 2b). Similar results are also obtained for the gametocyte-specific transcription factor AP2-G  that contains levels of Pf_MORC binding similar to what is observed in _var genes (Fig. 2c and S2c, d).

Shift in Binding Sites: The observed minor switch in PfMORC binding sites from gene bodies to intergenic and promoter regions is mentioned, but without context on how these shifts impact gene expression or any comparative analysis with other proteins showing similar shifts. The claim that this shift implicates PfMORC as an "insulator" is a leap without direct evidence.

We apologize for the confusion. We  have compared our ChIP-seq with RNA seq results at different time points of the cell cycle and demonstrated that the shift observed has an effect in gene expression. We have edit the manuscript to clarify these results.

Overextension of PfMORC's Role: The authors suggest that PfMORC moves to the regulatory regions around the TSS to guide RNA Polymerase and transcription factors. This is a substantial claim and would require additional experiments to validate. Simply observing binding in a region is insufficient to assign a specific functional role, especially one as critical as guiding RNA Polymerase. Historically, the MORC family has been primarily linked with gene silencing across Apicomplexan, plants, and metazoans. On page 7, the authors noted a minimal overlap between the ChIP-seq and RNA-seq signals (Fig. 4e). They also acknowledged that the pronounced gene expression shifts at schizont stages result from a combination of direct and indirect impacts of PfMORC degradation, which could cause cell cycle arrest and potential heterochromatin disintegration, rather than just decreased PfMORC binding. Therefore, the authors should adjust their conclusions in the manuscript to more accurately represent the multifaceted functions of MORC in the parasite.

We agree with the reviewer's comment and have edited the manuscript accordingly.  

DISCUSSION:

The authors concluded that "Using a combination of ChIP-seq, protein knock down, RNA-seq and Hi-C experiments, we have demonstrated that the MORC protein is essential for the tight regulation of gene expression through chromatin compaction, preventing access to gene promoters from TFs and the general transcriptional machinery in a stage specific manner."

Again, the assertion that MORC protein is essential for tight regulation of gene expression, based purely on correlational data (e.g., ChIP-seq showing binding doesn't prove functionality), assumes causality which might not be fully substantiated. The phrase "preventing access to gene promoters from TFs and the general transcriptional machinery in a stage-specific manner" needs also validation. Asserting that MORC is essential for this function might oversimplify the process and overlook other critical contributors.

We agree with the reviewer’s comments and the conclusion has since been edited accordingly.

The discussion is quite poor. It would be pertinent to put MORC in perspective within the broader picture of regulatory mechanisms of chromatin state at telomeres and var genes. For instance, how do SIR2 and HDAC1 (associated with MORC) divide the task of deacetylation? Or the contribution of HP1 and other non-coding RNAs.

We agree with the reviewer’s suggestion. However, in order to put MORC in perspective within a broader picture, we would need to measure changes in localization of several molecular components regulating heterochromatin in WT versus KD condition. This will require access to several molecular tools and specific antibodies that we do not currently have. We have addressed these issues in our discussion.  

Minor corrections to the text and figures.

Figure 1d: Could you provide the ID for each AP2 directly on the volcano plot? While some IDs are referenced in the manuscript, visual representation in the plot would facilitate a clearer understanding of their enrichment levels.

ID for unknown AP2 proteins have been added on the volcano plot.

I recommend presenting Figure S2b as a panel within a primary figure. This change would offer readers a more quantitative understanding of the distinct differences between developmental stages. Notably, there seems to be a limited number of genes in common when considering the total, and there is an apparent lack of enrichment in the ring stage.

This has been done.

The captions are very minimally detailed. An effort must be made to better describe the panels as well as which statistical tests were used. 

We have improved the figure legends and add the number of biological replicates as well as the statistic used in each figure legend.

Figure 1A: The protein diagram with its domains does not take scale into account.

The figure has been modified.

Reviewer #2 (Recommendations For The Authors):

(1) The study lacks a direct link between PfMORC's inferred function and the state of heterochromatin in the genome post-depletion.

We agree with the reviewer's comment and have included additional ChIP-seq experiments to measure changes in histone marks in PfMORC depleted parasite line. We show a significant decrease in histone H3K9me3 marks in PfMORC KD condition.

Conducting ChIP-seq on well-known heterochromatin markers such as H3K9me3, HP1, or H3K36me2/3 could shed light on the consequences of PfMORC depletion on global heterochromatin and its boundaries.

With no access to an anti-HP1 antibody with reasonable affinity, we have not been able to study the impact of MORC KD on HP1 but have successfully observed the impact on H3K9me3 marks. These results have been added to the revised manuscript in (Fig. 5).

(2) The authors should conduct a more comprehensive analysis of PfMORC's genomic localization, comparing it to ApiAP2 binding (interacting proteins) and histone modifications. This would provide valuable insights.

We have performed a more comprehensive genome wide analysis of MORC binding through ChIP-seq on WT and MORC-KD conditions. Our results show that Pf_MORC localizes to heterochromatin with significant overlap with H3K9-trimethylation (H3K9me3) marks, at or near _var gene regions. When downregulated, level of H3K9me3 was detected at a lower level, validating a possible role of _Pf_MORC in gene repression. Regarding the comparison with AP2 binding, our proteomics datasets have shown extensive MORC binding with several AP2 proteins.

(3) RNA-seq data reveals that only a few genes are affected after 24 hours of PfMORC depletion, with an equivalent number of up-regulated and down-regulated genes. The reasons behind down-regulation resulting from a heterochromatin marker depletion are not clearly established.

We agree with the reviewer’s comment. At this stage (24 hours), _Pf_MORC depletion is limited and the effects at the transcriptional level are quite restricted. Furthermore, it is highly probable that down-regulated genes are most likely due to an indirect effect of a cell cycle arrest. We have edited the manuscript to address this comment. 

The relationship between this data and the partial depletion of PfMORC needs further discussion.

We agree with the reviewers and have improved our discussion in the revised version of the manuscript.

(4) The authors did not compare their ChIP-seq data with the genes found downregulated in the RNA-seq data. Examining the correlation between these datasets would enhance the study.

We apologize for the confusion. We have compared ChIP-seq and RNA-seq data and identified a very limited number of overlapping genes indicating that most of the changes observed in gene expression are in fact most likely indirect due to a cell cycle arrest and a collapse of the chromatin. We have edited the manuscript to clarify this issue.

(5) The discussion section is relatively concise and does not fully address the complexity of the data, warranting further exploration.

We have improved the discussion section in the revised version of the manuscript.

Author response:

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

Reviewer #1 (Public Review):

The authors previously showed in cell culture that Su(H), the transcription factor mediating Notch pathway activity, was phosphorylated on S269 and they found that a phospho-deficient Su(H) allele behaves as a moderate gain of Notch activity in flies, notably during blood cell development. Since a downregulation of Notch signaling was proposed to be important for the production of a specialized blood cell types (lamellocytes) in response to wasp parasitism, the authors hypothesized that Su(H) phosphorylation might be involved in this cellular immune response.

Consistent with their hypothesis, the authors show that Su(H)S269A knock-in flies display a reduced response to wasp parasitism and that Su(H) is phosphorylated upon infestation. Using in vitro kinase assays and a genetic screen, they identify the PKCa family member Pkc53E as the putative kinase involved in Su(H) phosphorylation and they show that Pkc53E can bind Su(H). They further show that Pkc53E deficit or its knock-down in larval blood cells results in similar blood cell phenotypes as Su(H)S269A, including a reduced response to wasp parasitism, and their epistatic analyses indicate that Pkc53E acts upstream of Su(H).

Strengths

The manuscript is well presented and the experiments are sound, with a good combination of genetic and biochemical approaches and several clear phenotypes which back the main conclusions. Notably Su(H)S269A mutation or Pkc53E deficiency strongly reduces lamellocyte production and the epistatic data are convincing.

Weaknesses

The phenotypic analysis of larval blood cells remains rather superficial. Looking at melanized cells is a crude surrogate to quantify crystal cell numbers as it is biased toward sessile cells (with specific location) and does not bring information concerning the percentage of blood cells differentiated along this lineage.

In Su(H)S269A knock-in or Pkc53E zygotic mutants, the increase in crystal cells in uninfected conditions and the decreased capacity to induce lamellocytes following infection could have many origins which are not investigated. For instance, premature blood cell differentiation could promote crystal cell differentiation and reduce the pool of lamellocytes progenitors. These mutations could also affect the development and function of the posterior signaling center in the lymph gland, which plays a key role in lamellocyte induction.

Similarly, the mild decrease on resistance to wasp infestation (Fig. 2A) could reflect a constitutive reduction in blood cell numbers in Su(H)S269A larvae rather than a defective down-regulation of Notch activity.

We fully agree with the reviewer that sessile crystal cells counts are a coarse approach to capture hemocytes. However, they allowed the screening of numerous genotypes in the course of our kinase candidate screen. We recorded the hemocyte numbers in the various genetic backgrounds and with regard to wasp infestation. There was no significant difference between Su(H)S269A and Su(H)gwt control, independent of infection. This is in agreement with earlier observations of unchanged plasmatocyte numbers in N or Su(H) mutants compared to the wild type (Duvic et al., 2002). We noted, however, a small drop in hemocyte numbers in Su(H)S269D and a strong one in Pkc53ED28 mutants in both conditions relative to control. Presumably, Pkc53E has a more general role in blood cell development, which we have not further analysed. The results were included in new Figure 1_S1 and Figure 9_S1 supplements. Based on the link between hemocyte numbers and wasp resistance (e.g. McGonigle et al., 2017), we cannot exclude that the lowered resistance of Pkc53ED28 mutants regarding wasp attacks is partly due to reduced hemocyte numbers, albeit we did not see significant differences between either Su(H)S269A, nor Pkc53ED28 nor the double mutant. We have included this notion in the text.

Lamellocytes arise in response to external challenges like parasitoid wasp infestation by trans-differentiation from larval plasmatocytes, and by maturation of lamellocyte precursors in the lymph gland, yet barely in the Su(H)S269A and Pkc53ED28 mutants.

We find it hard to envisage, however, that a premature differentiation of plasmatocytes into crystal cells in our case could deplete the pool of lamellocyte progenitors in the hemolymph. (Is there a precedent?). Crystal cells make up about 5% of the hemocyte pool; they are increased max. 2 fold in the Su(H)S269A and Pkc53E mutants. Even if these extra crystal cells (now  ̴10%) had arisen by premature differentiation, there should be still enough plasmatocytes (̴ 80%) remaining with a potential to further divide and transdifferentiate into lamellocytes.

Indeed, we cannot exclude an effect of the Su(H)S269A mutant on the development and function of the posterior signaling center of the lymph gland. We noted, however, a slight but significant enlargement of the PS in the Su(H)S269A mutant, that to our understanding cannot explain the reduced lamellocyte numbers.

Whereas the authors also present targeted-knock down/inhibition of Pkc53E suggesting that this enzyme is required in blood cells to control crystal cell fate (Fig. 6), it is somehow misleading to use lz-GAL4 as a driver in the lymph gland and hml-GAL4 in circulating hemocytes as these two drivers do not target the same blood cell populations/steps in the crystal cell development process.

We fully agree with the reviewer that the two driver lines target different blood cell populations/ steps in hematopoiesis. The hml-Gal4 driver is regarded pan-hemocyte, common to both plasmatocytes and pre-crystal cells (e.g. Tattikota et al., 2020). It has been reported to drive specifically within differentiated hemocytes prior to or at the stage of crystal cells commitment (Mukherjee et al., 2011). Hence, hml-Gal4 appeared suitable to hit sessile and circulating hemocytes prior to final differentiation into crystal cells or lamellocytes, respectively.

In the lymph gland, however, hml is expressed within the cortical zone, where it appears specific to the plasmatocytes lineage, and not present in the crystal cell precursors (Blanco-Obregon et al., 2020). In contrast, lz-Gal4 is specific to the differentiating crystal cells in both lineages, i.e. in circulating and sessile hemocytes and in the lymph gland. Hence, we choose lz-Gal4 instead of hml-Gal4 at the risk of driving markedly later in the course of crystal cell differentiation. We included the reasoning in the text. Overall, we feel that this choice does not limit our conclusions.

In addition, the authors do not present evidence that Pkc55E function (and Su(H) phosphorylation) is required specifically in blood cells to promote lamellocyte production in response to infestation.

We have tried to address this interesting question by several means. Firstly, we show that Pkc53E is indeed expressed in the various cell types of larval hemocytes, shown in a new Figure 8 and Figure 8_S1 supplement. I.e., there is the potential of Pkc53E to promote lamellocyte formation. Moreover, RNAi-mediated downregulation of Pkc53E within hemocytes affected crystal cell formation similar to the Pkc53ED28 mutant, in agreement with a specific requirement within blood cells (Figure 6). Finally, we show a major drop in Notch target gene transcription (NRE-GFP) in response to wasp infestation within isolated hemocytes from Su(H)gwt in contrast to Su(H)S269A larvae (see new Figure 1 G). These data show that Su(H)-mediated Notch activity must be downregulated in hemocytes prior to lamellocyte formation in agreement with our hypothesis.

Finally, the conclusion that Pkc53E is (directly) responsible for Su(H) phosophorylation needs to be strengthened. Most importantly, the authors do not demonstrate that Pkc53E is required for Su(H) phosphorylation in vivo (i.e. that Su(H) is not phosphorylated in the absence of Pkc53E following infestation).

We would very much like to show respective results. Unfortunately, the low affinity of our pS269 antibody does not allow any in situ or in vivo experiments. We very much hope to obtain a more specific phosphoS269-Su(H) antibody allowing us further in situ studies, and show, for example co-localization with Pkc53E.

In addition, the in vitro kinase assays with bacterially purified Pkc53E (in the presence of PMA or using an activated variant of Pkc53E) only reveal a weak activity on a Su(H) peptide encompassing S269 (Fig. 4).

The reviewer correctly notes the poor activity of our purified Pkc53EEDDD kinase. This low activity also holds true for the standard peptide (PS), which in fact is even less well accepted than the Swt substrate. Indeed, the commercially available PKCα is a magnitude more active. Whether this reflects the poor quality of our isolated protein compared to the commercial PKCα, or whether it reflects a true biochemical property of Pkc53E remains to be shown in the future. We noted this observation in the manuscript.

Moreover, while the authors show a coIP between an overexpressed Pkc53E and endogenous Su(H) (Fig. 7) (in the absence of infestation), it has recently been reported that Pkc53E is a cytoplasmic protein in the eye (Shieh et al. 2023), calling for a direct assessment of Pkc53E expression and localization in larval blood cells under normal conditions and upon infestation.

Indeed, it is interesting that a Pkc53E-GFP fusion protein is cytoplasmic in the eye. The construct reported by Shieh et al. however, i.e. the B-isoform, is preferentially expressed in photoreceptors, where it regulates the de-polymerization of the actin cytoskeleton.

Due to the eye-specific expression, we unfortunately cannot use the Pkc53E-B-GFP construct to test for Pkc53E’s distribution in other tissues.

As this construct is of little use for studying hematopoiesis, we have instead used Pck53E-GFP (BL59413) derived from a protein trap: again, GFP is primarily seen in the cytoplasm of hemocytes, including lamellocytes of infected larvae. However, in a small number of hemocytes, GFP appears to be also nuclear (Fig. 8A), leaving the possibility that activated Pkc53E may localize to the nucleus, eventually phosphorylating Su(H) and downregulating Notch activity. As Su(H) enters the nucleus piggy-back with NICD, however, phosphorylation may as well occur at the membrane or within the cytoplasm. We note, however, that these hypotheses require a much more detailed analysis.

Furthermore, the effect of the PKCa agonist PMA on Su(H)-induced reporter gene expression in cell culture and crystal cell number in vivo is somehow consistent with the authors hypothesis, but some controls are missing (notably western blots to show that PMA/Staurosporine treatment does not affect Su(H)-VP16 level) and it is unclear why STAU treatment alone promotes Su(H)-VP16 activity (in their previous reports, the authors found no difference between Su(H)S269A-VP16 and Su(H)-VP16) or why PMA treatment still has a strong impact on crystal cell number in Su(H)S269A larvae.

We have added a Western blot showing that the treatment does not affect Su(H)-VP16 expression levels (Figure 5_supplement 1). As STAU is a general kinase inhibitor, it may obviate any inhibitory phosphorylation of Su(H)-VP16 in the HeLa cells, e.g. that by Akt1, CAMK2D or S6K which pilot T271, phosphorylation of which is expected to affect the DNA-binding of Su(H) as well (Figure 3_supplement 2). Moreover, in the previous report, we used different constructs with regard to the promoter, and we used RBPJ instead of Su(H), which may explain some of the discrepancies. As PMA is not specific to just Pkc53E, the altered crystal cell numbers may result from the influence on other kinases involved in blood cell homeostasis, as predicted by our genetic screen (Figure 3_supplement 1).

Reviewer #1 (Recommendations For The Authors):

(1) The authors should provide a more elaborate examination of larval blood cell types and blood cell counts under normal conditions and following infestation in the different zygotic mutants as well as upon Pkc53 knock-down. A thorough examination of PSC integrity should be performed and the maintenance of core blood cell progenitors examined. The authors should also clarify when after infestation the LG and larval bleeds are analyzed.

- a more elaborate examination of larval blood cell types:

- examination of larval blood cell counts under normal conditions: hemocyte # in gwt, SA, SD, & Pkc

- examination of larval blood cell counts after infestation: hemocyte # in gwt, SA, SD, & Pkc

- thorough examination  of PSC integrity: in gwt, SA, SD, & Pkc

- thorough examination of blood cell progenitors: in gwt, SA, SD, & Pkc

- clarify timing

Hemocyte numbers of the various genotypes and conditions were recorded and are presented in Figure 1_S1 and Figure 9_S1. Timing was elaborated in the text and the Methods section.

(2) The authors should clarify why they use lz-GAL4 or hml-GAL4 and what we can infer from using these different drivers.

See above. The reasoning was included in the text.

(3) The percentage of hatching of Su(H)S269A and Su(H)gwt flies in the absence of infestation should also be scored; a small decrease in Su(H)S269A viability might explain the observed differences in survival to wasp infestation. Absolute blood cell numbers (in the absence of infestation) have also been correlated with survival to infection and should be checked.

Percentage of the emerging flies and hemocyte numbers in the absence of infestation were recorded and included in Figure 2, Figure 1_S1, Figure 9_S1.

(4) Whereas the impact of Su(H)S269A or Pkc53E mutation on lamellocytes production is clear, there is still a substantial reduction in crystal cell production following infestation. So I wouldn't conclude that the Su(H) larvae are "unable" to detect this immune challenge or respond to it (line 116).

Thank you for the hint, we corrected the text.

(5) The expression and localization of Pkc53E in larval blood cells should be investigated, for instance using the Pkc53E-GFP line recently published by Shieh et al. (or at least at the RNA level).

Firstly, we confirmed expression of Pkc53E in hemocytes by RT-PCR (Figure 8_S1 supplement). Secondly, expression of Pkc53E-GFP was monitored in hemocytes (Figure 8). To this end, we used the protein trap (BL59413), since the one published by Shieh et al., 2023 is restricted to photoreceptors.

(6) It would be interesting to test the anti-pS269 antibody in immunostaining (using Su(H)S269A as negative control).

Unfortunately, the pS269 antiserum does not work in situ at all.

(7) The authors must perform a western blot with anti-pS269 in Pkc53e mutant to show that Su(H) is not phosphorylated anymore after wasp infestation.

The blot gives a negative result.

(8) It is surprising that no signal is seen in the absence of infestation with anti-pS269: the fact that Su(H)S269A have more crystal cells suggest that there is a constitutive level of phosphorylation of Su(H).

We fully agree: In the ideal world, we would expect a low level of S269 phosphorylation in the wild type as well. However, given the lousy specificity of our antibody, we were happy to see phospho-Su(H) in infected larvae. We are currently working hard to get a better antibody. 

(9) The authors should check Su(H)-VP16 levels and phosphorylation status after PMA and/or staurosporine treatment. Some clarifications are also needed to explain the impact of PMA in Su(H)S269 larvae (this clearly suggests that PKC has other substrates implicated in crystal cell development).

Su(H)-VP16 expression levels were monitored by Western blot and were not altered conspicuously (Figure 5_1 supplement). Presumably, Pkc53E is not the only kinase involved in Su(H) phosphorylation or the transduction of stress signals. Moreover, PMA may have a more general effect on larval development and hematopoiesis affecting both genotypes. We included this reasoning in the text.

(10) Concerning the redaction, the authors forgot to mention and discuss the work of Cattenoz et al. (EMBO J 2020). The presentation of the screen for kinase candidates could be streamlined and better illustrated (notably supplement table 4, which would be easier to grasp as a figure/graph). The discussion could be shortened (notably the part on T cells), and I don't really understand lines 374-376 (why is it consistent?).

We are sorry for omitting Cattenoz et al. 2020, which we have now included. We fully agree that this paper is of utmost importance to our work. We streamlined the screen and included a new figure in addition to table 4 summarizing the results graphically (Figure 3_S1 supplement). We cut on the T cell part and omitted the strange lines.

Reviewer #2 (Public Review):

Summary:

The current draft by Deischel et.al., entitled "Inhibition of Notch activity by phosphorylation of CSL in response to parasitization in Drosophila" decribes the role of Pkc53E in the phosphorylation of Su(H) to downregulate its transcriptional activity to mount a successful immune response upon parasitic wasp-infection. Overall, I find the study interesting and relevant especially the identification of Pkc53E in phosphorylation of Su(H) is very nice. However, I have a number of concerns with the manuscript which are central to the idea that link the phosphorylation of Su(H) via Pkc53E to implying its modulation of Notch activity. I enlist them one by one subsequently.

Strengths:

I find the study interesting and relevant especially because of the following:

(1) The identification of Pkc53E in phosphorylation of Su(H) is very interesting.

(2) The role of this interaction in modulating Notch signaling and thereafter its requirement in mounting a strong immune response to wasp infection is also another strong highlight of this study.

Weaknesses:

(1) Epistatic interaction with Notch is needed: In the entire draft, the authors claim Pkc53E role in the phosphorylation of Su(H) is down-stream of notch activity. Given the paper title also invokes Notch, I would suggest authors show this in a direct epistatic interaction using a Notch condition. If loss of Notch function makes many more lamellocytes and GOF makes less, then would modulating Pkc53E (and SuH)) in this manifest any change? In homeostasis as well, given gain of Notch function leads to increased crystal cells the same genetic combinations in homeostasis will be nice to see.

While I understand that Su(H) functions downstream of Notch, but it is now increasingly evident that Su(H) also functions independent of Notch. An epistatic relationship between Notch and Pkc will clarify if this phosphorylation event of Su(H) via Pkc is part of the canonical interaction being proposed in the manuscript and not a non-canoncial/Notch pathway independent role of Su(H).

This is important, as I worry that in the current state, while the data are all discussed inlight of Notch activity, any direct data to show this affirmatively is missing. In our hands we do find Notch independent Su(H) function in immune cells, hence this is a suggestion that stems from our own personal experience.

The role of Notch in Drosophila hematopoiesis, notably during crystal cell development in both hematopoietic compartments is well established; likewise the role of Su(H) as integral signal transducer in this context (e.g. Duvic et al., 2002). Not only promotes Notch activity crystal cell fate by upregulating target genes, at the same time it prevents adopting the alternative plasmatocyte fate (e.g. Terriente-Felix et al., 2013). We could confirm the downregulation of Notch target gene expression in response to wasp infestation by qRT-PCR, which was discovered earlier by Small et al. (2014). This is clearly in favor of a repression of Notch activity rather than a relief of inhibition by Su(H). A ligand-independent activation of Notch signaling has been uncovered in the context of crystal cell maintenance in the lymph gland involving Sima/Hif-α, including Su(H) as transcriptional mediator (Mukherjee et al., 2011). However, we are unaware of a respective Su(H) activity independent of Notch.

Certainly, Su(H) acts independently of Notch in terms of gene repression. Here, Su(H) forms a repressor complex together with H and co-repressors Groucho and CtBP to silence Notch target genes. Accordingly, loss of Su(H) or H may induce the upregulation of respective gene expression independent of Notch activity. This has been demonstrated, for example, during wing and heart development (Klein et al., 2000; Kölzer, Klein, 2006; Panta et al., 2020). Moreover, during axis formation of the early embryo, global repression is brought about by Su(H) and relieved by activated Notch (Koromila, Stathopolous, 2019). In all these instances, Su(H) is thought to act as a molecular switch, and the activation of Notch causes a strong expression of the respective genes. Likewise, the loss of DNA-binding resulting from the phosphorylation of Su(H) allows the upregulation of repressed Notch target genes in wing imaginal discs, e.g. dpn, as we have demonstrated before with overexpression and clonal analyses (Nagel et al. 2017; Frankenreiter et al., 2021). However, H does not contribute to crystal cell homeostasis, i.e. de-repression of Notch target genes does not appear to be a major driver in this context, asking for additional mechanisms to downregulate Notch activity. Our work provides evidence that these inhibitory mechanisms involves the phosphorylation of Su(H) by Pkc53E. Formally, we cannot exclude alternative mechanisms. Hence, we have tried to avoid the direct link between Su(H) phosphorylation and the inhibition of Notch activity throughout the text, including the title. Moreover, we have discussed the possible consequences of Su(H) lack of DNA binding, interfering either with the activation of Notch target genes or abrogating their repression.

In addition, we have performed new experiments addressing the epistasis between Notch and Su(H) during crystal cell formation (Figure 1_supplement 1). To this end, we knocked down Notch activity in hemocytes by RNAi (hml::N-RNAi) in the Su(H)gwt and Su(H)S269A background, respectively. Indeed, Notch downregulation strongly impairs crystal cell development independent of the genetic background as expected if Notch were epistatic to Su(H). We attribute the slightly elevated crystal cell numbers observed in the Su(H)S269A background to the increase in the embryonic precursors (see Fig. 4; Frankenreiter et al. 2021). Of note, the Notch gain of function allele Ncos479 also displayed a likewise increase in embryonic crystal cell precursors as well as in crystal cells within the lymph gland (Frankenreiter et al. 2021).

(2) Temporal regulation of Notch activity in response to wasp-infection and its overlapping dynamics of Su(H) phosphorylation via Pkc is needed:

First, I suggest the authors to show how Notch activity post infection in a time course dependent manner is altered. A RT-PCR profile of Notch target genes in hemocytes from infected animals at 6, 12, 24, 48 HPI, to gauge an understanding of dynamics in Notch activity will set the tone for when and how it is being modulated. In parallel, this response in phospho mutant of Su(H) will be good to see and will support the requirement for phosphorylation of Su(H) to manifest a strong immune response.

Indeed, it would be extremely nice to follow the entire processes in every detail, ideally at the cellular level. The challenge, however, is quantities. The mRNA isolated from hemocytes could be barely quantified, although the subsequent ct-values were ok. We quantified NRE-GFP expression, introduced into Su(H)gwt and Su(H)S269A, as well as atilla expression. We were able to generate data for two time slots, 0-6 h and 24-30 h post infection. The data are provided in the extended Figure 1G, and show a strong drop of NRE-GFP in the infected Su(H)gwt control compared to the uninfected animals, whereas expression in Su(H)S269A plateaus at around 60%-70% of the infected Su(H)gwt control. Atilla expression jumps up in the control, but stays low in Su(H)S269A hemocytes.

Second, is the dynamics of phosphorylation in a time course experiment is missing. While the increased phosphorylation of Su(H) in response to wasp-infestation shown in Fig.2B is using whole animal, this implies a global down-regulation of Su(H)/Notch activity. The authors need to show this response specifically in immune cells. The reader is left to the assumption that this is also true in immune cells. Given the authors have a good antibody, characterizing this same in circulating immune cells in response to infection will be needed. A time course of the phosphorylation state at 6, 12, 24, 48 HPI, to guage an understanding of this dynamics is needed.

We really would love to do these experiments. Unfortunately, our pS269 antibody is rather lousy. It does not allow to detect Su(H) protein in tissue or cells, nor does it work on protein extracts in Westerns or for IP. Hence, we have no way so far to demonstrate cell or tissue specificity of Su(H) phosphorylation. So far, we were lucky to detect mCherry-tagged Su(H) proteins pulled down in rather large amounts with the highly specific nano-bodies. We have tried very hard to repeat the experiment with hemolymph and lymph glands only, but we have failed so far. Hence, we have to state that our antibody is neither suitable for in vivo analyses, nor for a detection of phospho-Su(H) at lower levels.

The authors suggest, this mechanism may be a quick way to down-regulate Notch, hence a side by side comparison of the dynamics of Notch down-regulation (such as by doing RT-PCR of Notch target genes following different time point post infection) alongside the levels of pS269 will strengthen the central point being proposed.

We fully agree and hope to address these issues in the future by improving our tools.

Last, in Fig7. the authors show Co-immuno-precipitation of Pkc53EHA with Su(H)gwt-mCh 994 protein from Hml-gal4 hemocytes. I understand this is in homeostasis but since this interaction is proposed to be sensitive to infection, then a Co-IP of the two in immune cells, upon infection should be incorporated to strengthen their point.

We do not fully agree with the reviewer. Although we also think that the interaction between Pkc53E and Su(H) might occur more frequently upon infection, we propose that this is a transient process occurring in several but not all hemocytes at a given time. Moreover, in the described experiment, Pkc53E-HA was expressed in hemocytes via the UAS/Gal4 system. We cannot exclude that this approach causes an overexpression. Hence, we would not expect considerable differences between unchallenged and infested animals.

(3) In Fig 5B, the authors show the change in crystal cell numbers as read out of PMA induced activation of Pkc53E and subsequent inhibition of Su(H) transcriptional activity, I would suggest the authors use more direct measures of this read out. RT-PCR of Su(H) target genes, in circulating immune cells, will strengthen this point. Formation of crystal cells is not just limited to Notch, I am not convinced that this treatment or the conditions have other affect on immune cells, such as any impact on Hif expression may also lead to lowering of CC numbers. Hence, the authors need to strengthen this point by showing that effects are direct to Notch and Su(H) and not non-specific to any other pathway also shown to be important for CC development.

We agree with the Reviewer that the rather general influence of PMA on PKCs might present a systemic stress to the animal. For example, we observed a slight drop of crystal cell numbers also in Su(H)S269A, suggesting other kinases apart from Pkc53E were affected that are involved in crystal cell homeostasis. We have included this notion in the text. To provide more conclusive evidence we also fed Staurosporine to the larvae which reversed the PMA effect. In addition, we assayed the expression of NRE-GFP in hemocytes of infected animals by qRT-PCR, and observed a strong drop in the infected versus uninfected control but less so in Su(H)S269A. The new data are provided in extended Figures 1G and 5B.

(4) In addition to the above mentioned points, the data needs to be strengthened to further support the main conclusions of the manuscript. I would suggest the authors present the infection response with details on the timing of the immune response. Characterization of the immune responses at respective time points (as above or at least 24 and 48 HPI, as norms in the field) will be important. Also, any change in overall cell numbers, other immune cells, plasmatocytes or CC post infection is missing and is needed to present the specificity of the impact. The addition of these will present the data with more rigor in their analysis.

Total hemocyte numbers of the various genotypes, i.e. control, Su(H)S269A, Su(H)S269D, and Pkc53ED28 were included before and after wasp infestation in supplemental Figures 1_S1 and 9_S1. 

(5) Finally, what is the view of the authors on what leads to activation of Pkc53E, any upstream input is not presented. It will be good to see if wasp infection leads to increased Pkc53 kinase activity.

The analysis of the full process is an ongoing project. We propose that ROS is produced upon the wasps’ sting, which is to trigger the subsequent cascade of events. These have to end with activation of Pkc53E in the presumptive pre-lamellocyte pool of both lineages, i.e. in plasmatocyte of the hemolymph, presumably in the sessile compartment (Tattikotta et al., 2021) and at the same time in the lymph gland cortex harboring the LM precursors (Blanco-Obregon et al., 2020). One of the known upstream kinases, Pdk1 has a similar impact on crystal cell development as Pkc53E, making its involvement likely. Moreover, we think that other PKCs influence the process as well.

Without a good read out, e.g. a functional pSu(H) antiserum working in situ or a Pkc-activity reporter, it will be quite difficult to follow up this question. However, we already know that Pkc53E is expressed in hemocytes of all types independent of wasp infestation, in agreement with a role during lamellocyte differentiation. We hope to unravel the process in more of it in the future.

Overall, I think the findings in the current state are interesting and fill an important gap, but the authors will need to strengthen the point with more detailed analysis that includes generating new data and also presenting the current data with more rigor in their approach. The data have to showcase the relationship with Notch pathway modulation upon phosphorylation of CSL in a much more comprehensive way, both in homeostasis and in response to infection which is entirely missing in the current draft.

Reviewer #3 (Public Review):

Diechsel et al. provide important and valuable insights into how Notch signalling is shut down in response to parasitic wasp infestation in order to suppress crystal cell fate and favour lamellocyte production. The study shows that CSL transcription factor Su(H) is phosphorylated at S269A in response to parasitic wasp infestation and this inhibitory phosphorylation is critical for shutting down Notch. The authors go on to perform a screen for kinases responsible for this phosphorylation and have identified Pkc53E as the specific kinase acting on Su(H) at S269A. Using analysis of mutants, RNAi and biochemistry-based approaches the authors convincingly show how Pkc53E-Su(H) interaction is critical for remodelling hematopoiesis upon wasp challenge. The data presented supports the overall conclusions made by the authors. There are a few points below that need to be addressed by the authors to strengthen the conclusions:

(1) The authors should check melanized crystal cells in Su(H)gwt and Su(H)S269A in presence of PMA and Staurosporine?

Thank you for the suggestion. We included the results of PMA + Staurosporine feeding into an extended Fig. 5B; they match those from the HeLa cells. Unfortunately, Staurosporine alone was lethal for the larvae at various concentrations, presumably owing to the overarching inhibition of kinase activity. This global effect also explains the high crystal cell numbers in the control fed with PMA + STAU compared to the untreated animals, as the downregulation of many kinases results in higher crystal cell numbers, a fact uncovered in our genetic screen.

(2) Data for number of dead pupae, flies eclosed, wasps emerged post infestation should be monitored for the following genotypes and should be included:

Pkc53EΔ28_, Su(H)S269A,_ Pkc53EΔ28 Su(H)S269A, Su(H)S269D, Su(H)S269D Pkc53EΔ28

We extended the data with and without infection. The respective data are shown in a new Fig. 9 and an extended Fig. 2,  except for the Su(H)S269D allele. Su(H)S269D is larval lethal, i.e. dies too early for wasp development, and hence could not be included in the assay. Overall, Pkc53EΔ28 matched Su(H)S269A_._

(3) The exact molecular trigger for activation of Pkc53E upon wasp infestation is not clear.

Indeed, and we would love to know! Perhaps, the generation of Ca2+ by the wasp’s breach of the larval cuticle results in Pkc53E activation. The generation of ROS could be involved as well. At this point, we can only speculate. We hope to be able in the future to obtain direct experimental evidence for the one or the other hypothesis.

(4) The authors should check if activating ROS alone or induction of Calcium pulses/DUOX activation can mimic this condition and can trigger activation of Pkc53E and thereby cause phosphorylation of Su(H) at S269

The reviewer’s suggestions open up a new field of investigations, and are hence beyond of the scope of this article. However, we want to pursue the research in this direction, albeit we realize that counting crystal cells is too coarse but to give a first impression, and that lamellocytes may form already by breaching the larval cuticle. A major challenge shall be direct measurements of Pkc53E activation. To date, we have no tools for this, but ideally, we would like to have a direct, biochemical read out. Although we have been unsuccessful in the past, we want to develop a strong and specific phospho-S269 antibody that is also working in situ. Alternatively, we think of developing a PS-phosphorylation reporter, to allow reasonably addressing these questions.

(5) Does Pkc53E get activated during sterile inflammation?

We are in the process of addressing this issue, however, feel that his topic is beyond the scope of this paper. Our preliminary experiments, however, support the notion of a phospho-dependent regulation of Su(H) also in this context.

Reviewer #3 (Recommendations For The Authors):

The authors provide a graphical representation of major phenotypes that form the basis of their investigation and conclusions but have not supplemented the quantitation with images that represent these phenotypes. The authors need to include the following data to strengthen their conclusions:

(1) The authors should include representative images for each of the genotypes/conditions (in presence and absence of wasp infestation) based on which corresponding plots have been made in Figure 1. Please include this for both circulating lamellocytes in the hemolymph and in the lymph glands since this is one of the main figures presenting the key findings.

The data have been included in Figure 1-S2 supplement.

(2) Please include representative images of LG with Hnt staining and corresponding images for melanization for each of the genotypes used in the plots in Figure 6A and B.

The data have been included in Figure 6-S2 supplement.

(3) Representative images for each of the genotypes in Figure 7A & B should be included (circulating crystal cells and lymph gland crystal cell numbers).

Representative images for each of the genotypes for Fig. 7A have been included in Figure 7-S1 and for the old Fig. 7B in Figure 9-S2 supplement, respectively.

Author response:

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

Response to reviewers

We thank the Editor and the Reviewers for their constructure review. In the light of this feedback, we have made a number of changes and additions to the manuscript, that we think improved the presentation and hopefully address the majority of the concerns by the reviewers.

Main changes:

•   We added a new SI section (B1) with a population dynamics simulation in the high clonal interference regime and without expiring fitness (see R1: (1)).

•   We added a new SI section (A9) with the derivation of the equilibrium state of our SIR model in the case of 𝑀 immune groups and in the limit 𝜀 → 0 (see R1: (5)).

•   The text of the section Abstraction as “expiring” fitness advantage has been modified.

•   We added a new SI section (A4) describing the links between parameters of the “expiring fitness” and SIR models.

All three reviewers had concerns about the relation between our SIR model and the “expiring fitness” model, that we hope will be addressed by the last two items listed above. In particular, we would like to underline the following points:

•   The goal of our SIR model is to give a mechanistic explanation of partial sweeps using traditional epidemiological models. While ecological models (e.g. consumer resource) can give rise to the same phenomenology, we believe that in the context of host-pathogen interaction it is relevant to explicitely show that SIR models can result in partial sweeps.

•   The expiring fitness model is mainly an effective model: it reproduces some qualitative features of the SIR but does not quantitatively match all aspects of the frequency dynamics in SIR models.

•   It is possible to link the parameters of the SIR (𝛼,𝛾,𝑏,𝑓) and expiring fitness (𝑠,𝑥,𝜈) models at the beginning of the invasion of the variant (new SI section A4). However, the two models also differ in significant ways (the SIR model can for example oscillate, while the effective model can not). The correspondence of quantities like the initial invasion rate and the ‘expiration rate’ of fitness effects is thus only expected to hold for some time after the emergence of a novel variant.

Public reviews:

Reviewer 1:

Summary In this work, the authors study the dynamics of fast-adapting pathogens under immune pressure in a host population with prior immunity. In an immunologically diverse population, an antigenically escaping variant can perform a partial sweep, as opposed to a sweep in a homogeneous population. In a certain parameter regime, the frequency dynamics can be mapped onto a random walk with zero mean, which is reminiscent of neutral dynamics, albeit with differences in higher order moments. Next, they develop a simplified effective model of time dependent selection with expiring fitness advantage, and posit that the resulting partial sweep dynamics could explain the behaviour of influenza trajectories empirically found in earlier work (Barrat-Charlaix et al. Molecular Biology and Evolution, 2021). Finally, the authors put forward an interesting hypothesis: the mode of evolution is connected to the age of a lineage since ingression into the human population. A mode of meandering frequency trajectories and delayed fixation has indeed been observed in one of the long-established subtypes of human influenza, albeit so far only over a limited period from 2013 to 2020. The paper is overall interesting and well-written. Some aspects, detailed below, are not yet fully convincing and should be treated in a substantial revision.

We thank the reviewer for their constructive criticism. The deep split in the A/H3N2 HA segment from 2013 to 2020 is indeed the one of the more striking examples of such meandering frequency dynamics in otherwise rapidly adapting populations. But the up and down of H1N1pdm clade 5a.2a.1 in recent years might be a more recent example. We argue that such meandering dynamics might be a common contributor to seasonal influenza dynamics, even if it only spans 3-6 years.

(1) The quasi-neutral behaviour of amino acid changes above a certain frequency (reported in Fig, 3), which is the main overlap between influenza data and the authors’ model, is not a specific property of that model. Rather, it is a generic property of travelling wave models and more broadly, of evolution under clonal interference (Rice et al. Genetics 2015, Schiffels et al. Genetics 2011). The authors should discuss in more detail the relation to this broader class of models with emergent neutrality. Moreover, the authors’ simulations of the model dynamics are performed up to the onset of clonal interference 𝜌/ 𝑠0 \= 1 (see Fig. 4). Additional simulations more deeply in the regime of clonal interference (e.g. 𝜌/ 𝑠0 \= 5) show more clearly the behaviour in this regime.

We agree with the reviewer that we did not discuss in detail the effects of clonal interference on quasi-neutrality and predictability. As suggested, we conducted additional simulations of our population model in the regime of high clonal interference (𝜌/ 𝑠0 ≫ 1) and without expiring fitness effects. The results are shown in a new section of the supplementary information. These simulations show, as expected, that increasing clonal interference tends to decrease predictability: the fixation probability of an adaptive mutation found at frequency 𝑥 moves closer to 𝑥 as 𝜌 increases. However, even in a case of strong interference 𝜌/ 𝑠0 \= 32, 𝑝fix remains significantly different from the neutral expectation. We conclude from this that while it is true that dynamics tend to quasi-neutrality in the case of strong interference, this effect alone is unlikely to explain observations of H3N2 influenza dynamics. In our previous publication (BarratCharlaix et al, MBE, 2021) we have also investigated the effect of epistatic interactions between mutations, along side strong clonal interference. We concluded that, while most of these processes make evolution less predictable and push 𝑝fix towards the diagonal, it is hard to reproduce the empirical observations with realistic parameters. The “expiring fitness” model, however, produces this quite readily.

But there are qualitative differences between quasi-neutrality in traveling wave models and the expiring fitness model. In the traveling wave, a genotype carrying an adaptive mutation is always fitter than if it didn’t carry the mutation. Quasi-neutrality emerges from the accumulation of fitness variation at other loci and the fact that the coalescence time is not much bigger than the inverse selection coefficient of the mutation. In the expiring fitness model, the selective effect of the mutation itself goes away with time. We now discuss the literature on quasi-neutrality and cite Rice et al. 2015 and Schiffels et al. 2011.

In this context, I also note that the modelling results of this paper, in particular the stalling of frequency increase and the decrease in the number of fixations, are very similar to established results obtained from similar dynamical assumptions in the broader context of consumer resource models; see, e.g., Good et al. PNAS 2018. The authors should place their model in this broader context.

We thank the reviewer for pointing out the link between consumer resource models and our work. We further strengthened our discussion of the similarity of the phenomenology to models typically used in ecology and made an effort to highlight the link between consumer-resource models and ours in the introduction and in the part on the SIR model.

(2) The main conceptual problem of this paper is the inference of generic non-predictability from the quasi-neutral behaviour of influenza changes. There is no question that new mutations limit the range of predictions, this problem being most important in lineages with diverse immune groups such as influenza A(H3N2). However, inferring generic non-predictability from quasi-neutrality is logically problematic because predictability refers to individual trajectories, while quasi-neutrality is a property obtained by averaging over many trajectories (Fig. 3). Given an SIR dynamical model for trajectories, as employed here and elsewhere in the literature, the up and down of individual trajectories may be predictable for a while even though allele frequencies do not increase on average. The authors should discuss this point more carefully.

We agree with the reviewer that the deterministic SIR model is of course predictable. Similarly, a partial sweep is predictable. But we argue that expiring fitness makes evolution less predictable in two ways: (i) When a new adaptive mutation emerges and rises in frequency, we typically don’t know how rapidly its fitness effect is ‘expiring’. Thus even if we can measure its instantaneous growth rate accurately, we can’t predict its fate far into the future. (ii) Compared to the situation where fitness effects are not expiring, time to fixation is longer and there are more opportunities for novel mutations to emergence and change the course of the trajectory. We have tried to make this point clearer in the manuscript.

(3) To analyze predictability and population dynamics (section 5), the authors use a Wright-Fisher model with expiring fitness dynamics. While here the two sources of the emerging neutrality are easily tuneable (expiring fitness and clonal interference), the connection of this model to the SIR model needs to be substantiated: what is the starting selection 𝑠0 as a function of the SIR parameters (𝑓,𝑏,𝑀,𝜀), the selection decay 𝜈 = 𝜈(𝑓,𝑏,𝑀,𝜀,𝛾)? This would enable the comparison of the partial sweep timing in both models and corroborate the mapping of the SIR onto the simplified W-F model. In addition, the authors’ point would be strengthened if the SIR partial sweeps in Fig.1 and Fig.2 were obtained for a combination of parameters that results in a realistic timescale of partial sweeps.

We added a new section to the SI (A4) that relates the parameters of the SIR and expiring fitness models. In particular, we compute the initial growth rate 𝑠0 and a proxy for the fitness expiry rate 𝜈 as a function of the SIR parameters 𝛼,𝛾,𝑓,𝑏,𝑀, at the instant where the variant is introduced. The initial growth rate depends primarily on the degree of immune escape 𝑓, while the expiration rate 𝜈 is related to incidence 𝐼wt + 𝐼𝑚. However, as both models have fundamentally different dynamics, these relations are only valid on time scales shorter than potential oscillations of the SIR model. Beyond that, the connection between the models is mostly qualitative: both rely on the fact that growth rate of a strain diminishes when the strain becomes more frequent, and give rise to partial sweeps.

In Figure 1, the time it takes a partial sweep to finish is roughly 100− 200 generations (bottom right panel). If we consider H3N2 influenza and take one generation to be one week, this corresponds to a sweep time of 2 to 4 years, which is slightly slower but roughly in line with observations for selective sweeps. This time is harder to define if oscillatory dynamics takes place (middle right panel), but the time from the introduction of the mutant to the peak frequency is again of about 4 years. The other parameters of the model correspond to a waning time of 200 weeks and immune escape on the order of 20-30% change in susceptibility.

Reviewer 2:

Summary

This work addresses a puzzling finding in the viral forecasting literature: high-frequency viral variants evince signatures of neutral dynamics, despite strong evidence for adaptive antigenic evolution. The authors explicitly model interactions between the dynamics of viral adaptations and of the environment of host immune memory, making a solid theoretical and simulation-based case for the essential role of host-pathogen eco-evolutionary dynamics. While the work does not directly address improved data-driven viral forecasting, it makes a valuable conceptual contribution to the key dynamical ingredients (and perhaps intrinsic limitations) of such efforts.

Strengths

This paper follows up on previous work from these authors and others concerning the problem of predicting future viral variant frequency from variant trajectory (or phylogenetic tree) data, and a model of evolving fitness. This is a problem of high impact: if such predictions are reliable, they empower vaccine design and immunization strategies. A key feature of this previous work is a “traveling fitness wave” picture, in which absolute fitnesses of genotypes degrade at a fixed rate due to an advancing external field, or “degradation of the environment”. The authors have contributed to these modeling efforts, as well as to work that critically evaluates fitness prediction (references 11 and 12). A key point of that prior work was the finding that fitness metrics performed no better than a baseline neutral model estimate (Hamming distance to a consensus nucleotide sequence). Indeed, the apparent good performance of their well-adopted “local branching index” (LBI) was found to be an artifact of its tendency to function as a proxy for the neutral predictor. A commendable strength of this line of work is the scrutiny and critique the authors apply to their own previous projects. The current manuscript follows with a theory and simulation treatment of model elaborations that may explain previous difficulties, as well as point to the intrinsic hardness of the viral forecasting inference problem.

This work abandons the mathematical expedience of traveling fitness waves in favor of explicitly coupled eco-evolutionary dynamics. The authors develop a multi-compartment susceptible/infected model of the host population, with variant cross-immunity parameters, immune waning, and infectious contact among compartments, alongside the viral growth dynamics. Studying the invasion of adaptive variants in this setting, they discover dynamics that differ qualitatively from the fitness wave setting: instead of a succession of adaptive fixations, invading variants have a characteristic “expiring fitness”: as the immune memories of the host population reconfigure in response to an adaptive variant, the fitness advantage transitions to quasi-neutral behavior. Although their minimal model is not designed for inference, the authors have shown how an elaboration of host immunity dynamics can reproduce a transition to neutral dynamics. This is a valuable contribution that clarifies previously puzzling findings and may facilitate future elaborations for fitness inference methods.

The authors provide open access to their modeling and simulation code, facilitating future applications of their ideas or critiques of their conclusions.

We thank the reviewer for their summary, assessement, and constructive critique.

(1) The current modeling work does not make direct contact with data. I was hoping to see a more direct application of the model to a data-driven prediction problem. In the end, although the results are compelling as is, this disconnect leaves me wondering if the proposed model captures the phenomena in detail, beyond the qualitative phenomenology of expiring fitness. I would imagine that some data is available about cross-immunity between strains of influenza and sarscov2, so hopefully some validation of these mechanisms would be possible.

We agree with the reviewer that quantitatively confronting our model with data would be very interesting. Unfortunately, most available serological data for influenza and SARS-CoV-2 is obtained using post-infection sera from previoulsy naive animal models. To test our model, we would require human serology data, ideally demographically resolved, and a way to link serology to transmission dynamics. Furthermore, our model is mostly an explanation for qualitative features of variant dynamics and their apparent lack of predictability. We therefore considered that quantitative validation using data is out of scope of this work.

(2) After developing the SIR model, the authors introduce an effective “expiring fitness” model that avoids the oscillatory behavior of the SIR model. I hoped this could be motivated more directly, perhaps as a limit of the SIR model with many immune groups. As is, the expiring fitness model seems to lose the eco-evolutionary interpretability of the SIR model, retreating to a more phenomenological approach. In particular, it’s not clear how the fitness decay parameter 𝜈 and the initial fitness advantage 𝑠0 relate to the key ecological parameters: the strain cross-immunity and immune group interaction matrices.

The expiring fitness model emerges as a limiting case, at least qualitatively, of the SIR model when growth rate of the new variant is small compared to the waning rate and the SIR model does not oscillate. This can be readily achieved by many immune groups, which reconciles the large effect of many escape mutations and the lack of oscillation by confining the escape to some fraction of the population. Beyond that, the expiring fitness model is mainly an effective model that allows us to study the consequences of partial sweeps on predictability on long timescales. As stated in the “Main changes” section at the start of this reply, we added an SI section which links parameters of the two models. However, we underline the fact that beyond the phenomenon of partial sweeps, the dynamics of the two are different.

Reviewer 3:

Summary

In this work the authors start presenting a multi-strain SIR model in which viruses circulate in an heterogeneous population with different groups characterized by different cross-immunity structures. They argue that this model can be reformulated as a random walk characterized by new variants saturating at intermediate frequencies. Then they recast their microscopic description to an effective formalism in which viral strains lose fitness independently from one another. They study several features of this process numerically and analytically, such as the average variants frequency, the probability of fixation, and the coalescent time. They compare qualitatively the dynamics of this model to variants dynamics in RNA viruses such as flu and SARS-CoV-2.

Strengths

The idea that a vanishing fitness mechanisms that produce partial sweeps may explain important features of flu evolution is very interesting. Its simplicity and potential generality make it a powerful framework. As noted by the authors, this may have important implications for predictability of virus evolution and such a framework may be beneficial when trying to build predictive models for vaccine design. The vanishing fitness model is well analyzed and produces interesting structures in the strains coalescent. Even though the comparison with data is largely qualitative, this formalism would be helpful when developing more accurate microscopic ingredients that could reproduce viral dynamics quantitatively. This general framework has a potential to be more universal than human RNA viruses, in situations where invading mutants would saturate at intermediate frequencies.

We thank the reviewer for their positive remarks and constructive criticism below.

Weaknesses

The authors build the narrative around a multi-strain SIR model in which viruses circulate in an heterogeneous population, but the connection of this model to the rest of the paper is not well supported by the analysis. When presenting the random walk coarse-grained description in section 3 of the Results, there is no quantitative relation between the random walk ingredients importantly 𝑃(𝛽) - and the SIR model, just a qualitative reasoning that strains would initially grow exponentially and saturate at intermediate frequencies. So essentially any other microscopic description with these two features would give rise to the same random walk.

As also highlighted in the response to other reviewers, we now discuss how the parameter of the SIR model are related to the initial growth rate and the ‘expiration’ rate of the effective model. While the phenomenology of the SIR model is of course richer, this correspondence describes its overdamped limit qualitatively well.

Currently it’s unclear whether the specific choices for population heterogeneity and cross-immunity structure in the SIR model matter for the main results of the paper. In section 2, it seems that the main effect of these ingredients are reduced oscillations in variants frequencies and a rescaled initial growth rate. But ultimately a homogeneous population would also produce steady state coexistence between strains, and oscillation amplitude likely depends on parameters choices. Thus a homogeneous population may lead to a similar coarse-grained random walk.

The reviewer is correct that the primary effects of using many immune groups is to slow down the increase of novel variant, which in turn dampens the oscillations. Having multiple immune groups widens the parameter space in which partial sweeps without dramatic oscillations are observed. For slow sweeps, similar dymamics are observed in a homogeneous population.

Similarly, it’s unclear how the SIR model relates to the vanishing fitness framework, other than on a qualitative level given by the fact that both descriptions produce variants saturating at intermediate frequencies. Other microscopic ingredients may lead to a similar description, yet with quantitative differences.

Both of these points were also raised by other reviewers and we agree that it is worth discussing them at greater length. We now discuss how the parameters of the ‘expiring fitness’ model relate to those of the SIR. We also discuss how other models such as ecological models give rise to similar coarse grained models.

At the same time, from the current analysis the reader cannot appreciate the impact of such a mean field approximation where strains lose fitness independently from one another, and under what conditions such assumption may be valid.

In the SIR model, the rate at which strains lose fitness does depend on the precise state of the host population through the quantities 𝑆𝑚 and 𝑆wt , which is apparent in equation (A27) of the new SI section. The fact that a new variant shifts the equilibrium frequencies of previous strains in a proportional way is valid if the “antigenic space” is of very high dimensions, as explained in section Change in frequency when adding subsequent strains of the SI. It would indeed be interesting to explore relaxations of this assumption by considering a larger class of cross immunity matrices 𝐾. However, in the expiring fitness model, the fact that strains lose fitness independently from each ohter is a necessary simplification.

In summary, the central and most thoroughly supported results in this paper refer to a vanishing fitness model for human RNA viruses. The current narrative, built around the SIR model as a general work on host-pathogen eco-evolution in the abstract, introduction, discussion and even title, does not seem to match the key results and may mislead readers. The SIR description rather seems one of the several possible models, featuring a negative frequency dependent selection, that would produce coarse-grained dynamics qualitatively similar to the vanishing fitness description analyzed here.

We have revised the text throughout to make the connections between the different parts of the manuscript, in particular the SIR model and the expiring fitness model, clearer. We agree that the phenomenology of the expiring fitness model is more general than the case of human RNA viruses described by the SIR model, but we think this generality is an attractive feature of the coarse-graining, not a shortcoming. Indeed, other settings with negative frequency dependent selection or eco-systems that adapt on appropriate time scale generate similar dynamics.

Recommendations for the authors:

Reviewer 1:

(4) Line 74: what does fitness mean?

Many population dynamics models, including ones used for viral forecasting, attach a scalar fitness to each strain. The growth rate of each strain is then computed by substracting the average population fitness to the strain’s fitness. In this sentence, fitness is intended in this way.

(5) Fig. 1: The equilibrium frequency in the middle and bottom rows is hardly smaller than the equilibrium frequency in the top row for one immune group. This is surprising since for M=10, the variant escapes in only 1/10th of the population, which naively should impact the equilibrium frequency more strongly. Could the authors comment on this?

This is indeed non-trivial, and a hand-waving argument can be made by considering the extreme case 𝜀 = 0. The variant is then completely neutral for the immune groups 𝑖 > 1, and would be at equilibrium at any frequency in these immune groups. Its equilibrium frequency is then only determined by group 1, which is the only one breaking degeneracy. For 𝜀 > 0 but small, we naturally expect a small deviation from the 𝜀 = 0 case and thus 𝛽 should only change slightly.

A more rigorous argument with a mathematical proof in the case 𝜀 = 0 is now given in section A4 of the supplementary information.

(6) Fig. 1: In the caption, it is stated that the simulations are performed with 𝜀 = 0.99. Is this a typo? It seems that it should be 𝜀 = 0.01, as in and just below equation (7).

This was indeed a typo. It is now fixed.

(7) Fig. 3: The data analysis should be improved. In order to link the average frequency trajectories to standard population genetics of conditional fixation probabilities, the focal time should always be the time where the trajectory crosses the threshold frequency for the first time. Plotting some trajectories from a later time onwards, on their downward path destined to loss, introduces a systematic bias towards negative clonal interference (for these trajectories, the time between the first and the second crossing of the threshold frequency is simply omitted). The focal time of first crossing of the threshold frequency can easily be obtained, e.g., by linear interpolation of the trajectory between subsequent time points of frequency evalution. In light of the modified procedure, the statements on the on the inertia of the trajectories after crossing 𝑥⋆ (line 356) should be re-examined.

The way we process the data is already in line with the suggestions of the reviewer. In particular, we use as focal time the first time at which a trajectory is found in the threshold frequency bin. Trajectories that are never seen in the bin because of limited time-resolution are simply ignored.

In Fig. 3, there are no trajectories that are on their downward path at the focal time and when crossing the threshold frequency. Our other work on predictability of flu Barrat-Charlaix et. al. (2021) has a similar figure, which maybe created confusion.

(8) Fig. 4: authors write 𝛼/ 𝑠0 in the figure, but should be 𝜈/ 𝑠0.

Fixed.

(9) Line 420: authors refer to the blue curve in panel B as the case with strong interference. However, strong interference is for higher 𝜌/ 𝑠0, that is panel D (see point 1).

Fixed.

(10) Line 477: typo “there will a variety of mutations”.

Fixed.

Reviewer 2:

Should 𝛼 be 𝜈 in Figure 4 legends?

Thank you very much for spotting this error. We fixed it.

Equations 4-5 could be further simplified.

We factorised the 𝐼 term in equation 4. In equation 5, we prefered to keep the 1− 𝛿/ 𝛼 term as this quantity appears in different calculations concerning the model. For instance, 𝑆 = 𝛿/ 𝛼 at equilibrium.

The sentence before equation 8 references 𝑃𝛽(𝛽), but this wasn’t previously introduced.

We now introduce 𝑃𝑏𝜂 at the beginning of the section Ultimate fate of the variant.

In the last paragraph of page 12, “monotonously” maybe should be “monotonically”.

Fixed.

For the supplement section B, you might want a more descriptive title than “other”.

We renamed this section to Expiring fitness model and random walk.

Reviewer 3:

To expand on my previous comments, my main concerns regard the connection of section 2 and the SIR model with the rest of the paper.

In the first paragraph of page 9 the authors argue that a stochastic version of the SIR model would lead to different fixation dynamics in homogeneous vs heterogeneous populations due to the oscillations. This paragraph is quite speculative, some numerical simulations would be necessary to quantitatively address to what extent these two scenarios actually differ in a stochastic setting, and how that depends on parameters.

Likewise, the connection between the SIR model, the random walk coarse-grained description and the vanishing fitness model can be investigated through numerical simulations of a stochastic SIR given the chosen population and cross-immunity structures with i.e. 10-20 strains. This would allow for a direct comparison of individual strain dynamics rather than the frequency averages, as well as other scalar properties such as higher moments, coalescent, and fixation probability once reaching a given frequency. It would also be possible to characterize numerically the SIR P(beta) bridging the gap with the random walk description. It’s not obvious to me that the SIR P(beta) would not depend on the population size in the presence of birth-death stochasticity, potentially changing the moments scalings. I appreciate that such simulations may be computationally expensive, but similar numerical studies have been performed in previous phylodynamics works so it shouldn’t be out of reach.

An alternative, the authors should consider re-centering the narrative directly on the random walk of the vanishing fitness model, mentioning the SIR more briefly as a possible qualitative way to get there. Either way the authors should comment on other ways in which this coarse-grained dynamics could arise.

In the vanishing fitness model, where variants fitnesses are independent, is an infinite dimensional antigenic space implicitly assumed? If that’s the case, it should be explained in the main text.

A long simulation of the SIR model would indeed be interesting, but is numerically demanding and our current simulation framework doesn’t scale well for many strains and susceptibilities. We thus refrained from adding extensive simulations.

In Figure 2B of the main text, the simulation with 7 strains illustrates the qualitative match between the expiring fitness and the SIR model. However, it is clearly not long enough to discuss statistical properties of the corresponding random walk. Furthermore, we do not expect the individual strain dynamics of the SIR and expiring fitness models to match. The latter depends on few parameters (𝛼, 𝑠0), while the former depends on the full state of the host population and of the previous variants.

In the sectin linking the parameters of the two models, we now discuss the distribution 𝑃(𝛽) of the SIR model for two strains and a specific choice of distribution for the cross immunity 𝑏 and 𝑓.

Minor comments:

There is some back and forth in the writing. For instance, when introducing the model, 𝐶𝑖𝑗 is first defined as 1/ 𝑀, then a few paragraphs later the authors introduce that in another limit 𝐶𝑖𝑖 is just much higher than any 𝐶𝑖𝑗, and finally they specify that the former is the fast mixing scenario.

Another example is in section 2, in the first paragraph they put forward that heterogeneity and crossimmunity have different impacts on the dynamics, but the meaning attributed to these different ingredients becomes clear only a while later after the homogeneous population analysis. Uniforming the writing would make it easier for the reader to follow the authors’ train of thought.

We removed the paragraph below Equation (1) mentioning the 𝐶𝑖𝑗 \= 1/ 𝑀 case, which we hope will linearize the writing.

When mentioning geographical structure, why would geography affect how immunity sees pairs of viral strains (differences in 𝐾)?

Geographic structure could influence cross-immunity because of exposure histories of hosts. For instance in the case of influenza, different geographical regions do not have the same dominating strains in each season, and hosts from different regions may thus build up different immunity.

In the current narrative there are some speculations about non-scalar fitness, especially in section 2. The heterogeneity in this section does not seem so strong to produce a disordered landscape that defies the notion of scalar fitness in the same way some complex ecological systems do. A more parsimonious explanation for the coexistence dynamics observed here may be a negative frequency dependent selection.

Our language here was not very precise and we agree that the phenomenology we describe is related to that of frequency dependent selection (mediated by via immunity of the host population that integrates past frequencies). Traveling wave models typically use fitness function that are independent of the population distribution and only account for the evolution via an increasing average fitness. We have made discussion more accurate by stating that we consider a case where fitness depends explicitly on present and past population composition, which includes the case of negative frequency dependent selection.

I don’t understand the comparison with genetic drift (typo here, draft) in the last paragraph of section 3 given that there is no stochasticity in growth death dynamics.

We compare the random walk to genetic drift because of the expression of the second moment of the step size. The genetic draft has the same functional form. If one defines the effective population size as in the text, the drift due to random sampling of alleles (neutral drift) and the changes in strain frequency in our model have the same first and second moments. The stochasticity here does not come from the dynamics, which are indeed deterministic, but from the appearance of new mutations (variants) on backgrounds that are randomly sampled in the population. This latter property is shared with genetic draft.

In the vanishing fitness model, I think the reader would benefit from having 𝑃(𝑠) in the main text, and it should be made more clear what simulations assume what different choice of 𝑃(𝑠).

We added the expression of 𝑃(𝑠) in the main text. Simulations use the value 𝑠0 \= 0.03, which we added in the caption of Figure 4.

When comparing the model and data, is the point that COVID is not reproduced due to clonal interference? It seems from the plot that flu has clonal interference as well though. Why is that negligible?

A similar point has been raised by the first reviewer (see R1-(1)). Clonal interference is not negligible, but we find it to be insufficient to explain the observations made for H3N2 influenza, namely the lack of inertia of frequency trajectories or the probability of fixation. This is shown in the new section (B1) of the SI. Both SARS-CoV-2 and H3N2 influenza experience clonal interference, but the former is more predictable than the latter. Our point is that expiring fitness effects should be stronger in influenza because of the higher immune heterogeneity of the host population, making it less predictable than SARS-CoV-2.

Does the fixation probability as a function of frequency threshold match the flu data for some parameters sets?

For H3N2 influenza, the fixation probability is found to be equal to the threshold frequency (see Barrat-Charlaix MBE 2021, also indirectly visible from Fig. 3). In Figure 4, we obtain that either a high expiry rate or intermediate expiry rates and clonal interference regimes match this observation.

It would be instructive to see examples of the individual variant dynamics of the vanishing fitness model compared to the presented data.

We added an extra SI figure (S7) showing 10 randomly selected trajectories of individual variants in the case of H3N2/HA influenza and for the expiring fitness model with different parameter choices.

Figure 4E has no colorbar label. The reader shouldn’t have to look for what that means in the bottom of the SIs. In panels A and B the label should be 𝜈, not 𝛼. Same thing in most equations of page 42.

We added the colorbar label to the figure and also updated the caption: a darker color corresponds to a higher probability of sweeps to overlap. We fixed the 𝜈 – 𝛼 confusion in the SI and in the caption of the figure.

Author response:

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

eLife Assessment 

This study presents an important finding on the involvement of a Caspase 3-dependent pathway in the elimination of synapses for retinogeniculate circuit refinement and eye-specific territory segregation. This work fits well with the concept of "synaptosis" which has been proposed in the past but lacked in vivo support. Despite its elegant design and many strengths, the evidence supporting the claims of the authors is incomplete, particularly regarding whether Caspase-3 expression can really be isolated to synapses vs locally dying cells, whether microglia direct or instruct synapse elimination, and whether astrocytes are also involved. The work will be of interest to investigators studying cell death pathways, neurodevelopment, and neurodegenerative disease.

Regarding significance:

This study provides in vivo evidence that caspase-3 is important for synapse elimination in the visual pathway (Figure 3 and 4) and corroborates the previously proposed but not yet validated “synaptosis” hypothesis. But more significantly, we show that caspase-3 is activated in dLGN relay neurons in response to synapse inactivation (Figure 1) when synaptic competition is present (Figure 2), and that caspase-3 is important for efficient elimination of weakened synapses by microglia (Figure 5 and 6). We consider the causal link between synapse weakening/inactivation and caspase-3 activation to be the most important finding of this study and believe it is an error to not include this aspect of the study in the assessment. The mechanism by which neuronal activity influences synapse elimination is a fundamental question in neuroscience, and our study presents a significant advancement in understanding this problem.

Regarding strength of evidence:

We do not agree with the assessment that our evidence should be broadly labeled as “incomplete”. In fact, we argue that many concerns raised by the reviewers are not focused on the main claims made in this study.

(1) Regarding whether caspase-3 activation (not “expression”, which is the term used in the assessment) is isolated to synapses or occurs in entire cells, we show in Figure 1 that both types of signals can be present. The main concern of the reviewers seems to be that activated caspase-3 signals in apoptotic dLGN relay neurons are irrelevant to our analysis and confound interpretation. We argue that this is not the case.

In Figure 1, we have two sets of controls demonstrating that the observed apoptosis of dLGN relay neurons occurs specifically in response to synapse inactivation. For each animal that received TeTxLC injection in the right eye, activated caspase-3 signal is compared between the left dLGN, where most of the inactivated synapses are located, and the right dLGN, where the minority of the inactivated synapses are located (between Figure 1B and 1C, also between the first and second group of Figure 1E). We observed apoptotic neurons in the right dLGN with more inactivated synapses but not in the left dLGN with fewer inactivated synapses. The second control is between TeTxLC-injected animals (Figure 1B) and mock-injected animals (Figure 1D). We observed apoptotic relay neurons in the dLGN of TeTxLC-injected animals (Figure 1B) but not mock-injected animals (Figure 1D). Both these controls show that the observed apoptosis of dLGN relay neurons is caused by synapse inactivation.

In addition, in our synapse inactivation experiment (Figure 1), AAV-hSyn-TeTxLC is injected into the right eye and expressed only in RGCs, not in dLGN relay neurons. Since dLGN relay neurons in this experiment do not receive a perturbation that is independent of synaptic transmission, we conclude that their apoptosis occurs through synapse-dependent mechanisms.

Furthermore, if the apoptotic neurons are confounding the analysis (as implied by reviewers and editors) and do not occur through synapse-dependent mechanisms, then inhibiting both eyes with TeTxLC (Figure 2C, rightmost group) should cause high levels of caspase-3 activation, like that in the single-inhibition condition. Instead, we observe the opposite (Figure 2C, middle group) – overall caspase-3 activity goes down significantly in the dual-inhibition condition and is closer to the unperturbed condition, which can be explained by a loss of interaction between “strong” and “weak” synapses. Taken together, our data demonstrate that apoptosis of relay neurons in Figure 1 occurs specifically in response to synapse inactivation through synapse-dependent mechanisms, and the activated caspase-3 signal in the neurons should be included in our analysis.

Why does synaptic caspase-3 activation manifest in different forms: puncta, “blobs”, and cells?  This is not surprising when considering the mechanisms that neurons must utilize to spatially confine caspase-3 activation and the nature of the apoptotic signaling cascade. On one hand, it has been proposed that caspase-3 activity in dendrites can be locally confined by proteasomal degradation of cleaved caspase-3 (Erturk et al., DOI: 10.1523/JNEUROSCI.3121-13.2014 ). On the other hand, caspase-3 activation is known to trigger explosive feedback amplification of apoptotic signaling events (McComb et al., DOI: 10.1126/sciadv.aau9433 ). For caspase-3 activation to remain localized to dendrites, the negative regulation must outweigh the positive feedback amplification. By expressing TeTxLC in RGCs of one eye, we create a strong perturbation that silences a large fraction of the synapses in the retinogeniculate pathway, which likely shifts the balance between positive and negative regulation of caspase-3 activity in some relay neurons. To be more specific, if a given dLGN relay neuron receives too many inactivated synapses, which is likely the case in our perturbation, caspase-3 activity that is initially localized can overwhelm the physiological negative regulation mechanisms that act to spatially confine it, resulting in whole cell apoptosis. In fact, previous in vitro evidence (Enturk et al., DOI: 10.1523/JNEUROSCI.3121-13.2014 ) demonstrated that, while caspase-3 activation in a single distal dendrite can be locally contained, activating apoptosis signaling in dendrites proximal to the cell body can result in whole-cell apoptosis. Similarly, a few inactivated retinogeniculate synapses can elicit locally contained caspase-3 activity in dLGN relay neurons, but a large number of inactivated synapses on a single relay neuron may trigger sufficient caspase-3 activity that can lead to whole-cell apoptosis. We discussed how to interpret synapse inactivation-induced apoptosis in dLGN relay neurons both in the main text and in the discussion (line 123-132, and line 411-421).

(2) Regarding microglia, we did not claim that “microglia direct or instruct synapse elimination”. Our main claim is that caspase-3 activation is important for efficient elimination of weakened synapses by microglia. This claim emphasizes a regulatory role for caspase-3 activation in microglia-mediated synapse elimination, but not a regulatory role of microglia in synapse elimination. To be more specific, our data suggest that lack of synaptic activity induces caspase-3 activity, and caspase-3 activity in turn influences which synapses are preferentially eliminated by microglia. Therefore, the elimination specificity is fundamentally determined (i.e. instructed) by neuronal activity, not by microglia. We also did not presume the manner in which microglia engage in synapse elimination. We specifically address this point in the discussion at line 458 through 465 where we acknowledge that microglia may indirectly mediate synapse elimination by engulfing shed neuronal material. In our title and text, we use the phrase “microglia-mediated synapse elimination”, which is not the same as microglia-instructed synapse elimination and does not presume any instructive/directive role of microglia.

(3) Regarding whether astrocytes are involved, we did not challenge the notion that astrocytes play important roles in synapse elimination. Rather, our claim is that, unlike what we observed with microglia, the amount of synaptic material engulfed by astrocytes does not robustly depend on whether caspase-3 is present. We acknowledge that there might be a caspase-3 dependent phenotype that we were unable to detect (line 309-310), and that it is plausible that astrocytes mediate activity-dependent synapse elimination through other caspase-3-independent mechanisms. This claim is not central to our study, and we would like to qualify the statements in the manuscript. We will remove the phrase “but not astrocytes” in line 18 of the abstract.

In summary, using a state-of-the-art method to inactivate retinogeniculate synapses, we discovered a causal link between synapse weakening/inactivation and caspase-3 activation. Coupled with well-established in vivo assays (e.g., segregation analysis, electrophysiology, and engulfment analysis) that are used in many landmark studies we cite, we provide solid evidence supporting our claim that “caspase-3 is essential for synapse elimination driven by both spontaneous and experience-dependent neural activity”, and that “synapse weakening-induced caspase-3 activation determines the specificity of synapse elimination mediated by microglia”.

Public Reviews: 

Reviewer #1 (Public Review): 

In this manuscript, the authors study the effects of synaptic activity on the process of eye-specific segregation, focusing on the role of caspase 3, classically associated with apoptosis. The method for synaptic silencing is elegant and requires intrauterine injection of a tetanus toxin light chain into the eye. The authors report that this silencing leads to increased caspase 3 in the contralateral eye (Figure 1) and demonstrate evidence of punctate caspase 3 that does not overlap neuronal markers like map2. However, the quantifications showing increased caspase 3 in the silenced eye (done at P5) are complicated by overlap with the signal from entire dying cells in the thalamus. The authors also show that global caspase 3 deficiency impairs the process of eye-specific segregation and circuit refinement (Figures 3-4). 

The reviewer states: “this silencing leads to increased caspase 3 in the contralateral eye”. We observed increased caspase-3 activity, not protein levels, in the contralateral dLGN, not eye.

The reviewer states: “and demonstrate evidence of punctate caspase 3 that does not overlap neuronal markers like map2”. This is not accurate. We show that the punctate active caspase-3 signals overlap with the dendritic marker MAP2 (Figure S4A).

The reviewer states: “, the quantifications showing increased caspase 3 [activity] in the silenced [dLGN] (done at P5) are complicated by overlap with the signal from entire dying cells in the thalamus”. This is not accurate. The apoptotic neurons we observed are relay neurons located in the dLGN (confirmed by their morphology and positive staining of NeuN – Figure S4B-C), not “cells” of unknown lineage (as suggested by the reviewer) in the general “thalamus” area (as suggested by the reviewer). If the dying cells were non-neuronal cells, that would indeed confound our quantification and conclusions, but that is not the case.

We argue that the active caspase-3 signals in apoptotic dLGN relay neurons are not a confounding factor but a bona fide response to synaptic silencing and therefore should be included in the quantification. We have two sets of controls (please also see the general response above), one is between the strongly inactivated dLGN and the weakly inactivated dLGN in each TeTxLC-injected animal, second is between dLGN of TeTxLC-injected animals and mock-injected animals. In both controls, only the dLGN receiving strong synapse inactivation has these apoptotic dLGN relay neurons, demonstrating that these cells occur as a consequence of synapse inactivation. It is also unlikely that our perturbation is causing cell death through a non-synaptic mechanism. As mock injections do not cause apoptosis in dLGN neurons, this phenomenon is not related to surgical damage. TeTxLC is injected into the eyes and only expressed in presynaptic RGCs, not in postsynaptic relay neurons, so this phenomenon is also unlikely to be caused by TeTxLC-related toxicity. Furthermore, if apoptosis of dLGN relay neurons is not related to synapse inactivation, then when TeTxLC is injected into both eyes, one would expect to see either the same amount or more apoptotic relay neurons, but we instead observed a reduction in dLGN neuron apoptosis, suggesting a synapse-related mechanism must be responsible. Considering the above, apoptosis of relay neurons in TeTxLC-inactivated dLGN is causally linked to synapse inactivation, and active caspase-3 signals in these neurons are true signals that should be included in the quantification.

The authors also report that "synapse weakening-induced caspase-3 activation determines the specificity of synapse elimination mediated by microglia but not astrocytes" (abstract). They report that microglia engulf fewer RGC axon terminals in caspase 3 deficient animals (Figure 5), and that this preferentially occurs in silenced terminals, but this preferential effect is lost in caspase 3 knockouts. Based on this, the authors conclude that caspase 3 directs microglia to eliminate weaker synapses. However, a much simpler and critical experiment that the authors did not perform is to eliminate microglia and show that the caspase 3 dependent effects go away. Without this experiment, there is no reason to assume that microglia are directing synaptic elimination. 

The reviewer states: “microglia engulf fewer RGC axon terminals in caspase 3 deficient animals (Figure 5), and that this preferentially occurs in silenced terminals, but this preferential effect is lost in caspase 3 knockouts”. We are not sure what the reviewer means by “this preferentially occurs in silenced terminals”. Our results show that microglia preferentially engulf silenced terminals, and such preference is lost in caspase-3 deficient mice (Figure 6).

We do not understand the experiment where the reviewer suggested to: “eliminate microglia and show that the caspase 3 dependent effects go away”. To quantify caspase-3 dependent engulfment of synaptic material by microglia or preferential engulfment of silenced terminals by microglia, microglia must be present in the tissue sample. If we eliminate microglia, neither of these measurements can be made. What could be measured if microglia are eliminated is the refinement of retinogeniculate pathway. This experiment would test whether microglia are required for caspase-3 dependent phenotypes. This is not a claim made in the manuscript. Instead, we claimed caspase-3 is required for microglia to preferentially eliminate weak synapses.

We did not claim that “microglia are directing synaptic elimination”. Our claim is that synapse inactivation induces caspase-3 activity, and this caspase-3 activity in turn determines the substrate preference of microglia-mediated synapse elimination. Based on this model, it is the neuronal activity that fundamentally directs synapse elimination. Throughout the manuscript, we used the term “microglia-mediated synapse elimination”. This terminology does not assume a directive/instructive role of microglia in synapse elimination and only describes the observed engulfment of synaptic material by microglia. We also did not assume how microglia engage in synapse elimination. We acknowledge in the discussion (line 458 through 465) that microglia may mediate synapse elimination in an indirect, passive way by engulfing shed neuronal material. This topic is a matter of debate in the field (Eyo et al., DOI: 10.1126/science.adh7906 ).

Finally, the authors also report that caspase 3 deficiency alters synapse loss in 6-month-old female APP/PS1 mice, but this is not really related to the rest of the paper. 

We respectfully disagree that Figure 7 is not related to the rest of the paper. Many genes involved in postnatal synapse elimination, such as C1q and C3, have been implicated in neurodegeneration. It is therefore natural and important to ask whether the function of caspase-3 in regulating synaptic homeostasis extends to neurodegenerative diseases in adult animals. The answer to this question may have broad therapeutic impacts.

Reviewer #2 (Public Review): 

Summary: 

This manuscript by Yu et al. demonstrates that activation of caspase-3 is essential for synapse elimination by microglia, but not by astrocytes. This study also reveals that caspase 3 activation-mediated synapse elimination is required for retinogeniculate circuit refinement and eye-specific territories segregation in dLGN in an activity-dependent manner. Inhibition of synaptic activity increases caspase-3 activation and microglial phagocytosis, while caspase-3 deficiency blocks microglia-mediated synapse elimination and circuit refinement in the dLGN. The authors further demonstrate that caspase-3 activation mediates synapse loss in AD, loss of caspase-3 prevented synapse loss in AD mice. Overall, this study reveals that caspase-3 activation is an important mechanism underlying the selectivity of microglia-mediated synapse elimination during brain development and in neurodegenerative diseases. 

Strengths: 

A previous study (Gyorffy B. et al., PNSA 2018) has shown that caspase-3 signal correlates with C1q tagging of synapses (mostly using in vitro approaches), which suggests that caspase-3 would be an underlying mechanism of microglial selection of synapses for removal. The current study provides direct in vivo evidence demonstrating that caspase-3 activation is essential for microglial elimination of synapses in both brain development and neurodegeneration. 

The paper is well-organized and easy to read. The schematic drawings are helpful for understanding the experimental designs and purposes. 

Weaknesses: 

It seems that astrocytes contain large amounts of engulfed materials from ipsilateral and contralateral axon terminals (Figure S11B) and that caspase-3 deficiency also decreased the volume of engulfed materials by astrocytes (Figures S11C, D). So the possibility that astrocyte-mediated synapse elimination contributes to circuit refinement in dLGN cannot be excluded.

The experiments presented in Figure S11 aim to determine whether astrocyte-mediated synapse elimination depends on caspase- 3 signaling.  We do not claim that astrocytes are unimportant for synapse elimination or circuit refinement. We did observe a small decrease in synaptic material engulfed by astrocytes when caspase-3 is deficient, and we acknowledged that there could be defects that we were not able to detect (line 309-310). The claim that caspase-3 does not regulate astrocyte-mediated synapse elimination is not a central claim of the manuscript and we will qualify our statements in the text. We will remove the phrase “but not astrocytes” in the abstract (line 18).

Does blocking single or dual inactivation of synapse activity (using TeTxLC) increase microglial or astrocytic engulfment of synaptic materials (of one or both sides) in dLGN? 

We assume that by “blocking single or dual inactivation of synapse activity”, the reviewer refers to inactivating retinogeniculate synapses from one or both eyes.

We showed that inactivating retinogeniculate synapses from one eye (single inactivation) increases microglia-mediated engulfment of presynaptic terminals of inactivated synapses (Figure 6). We did not measure microglia-mediated engulfment of synaptic material while inactivating retinogeniculate synapses from both eyes (dual inactivation). However, based on the total active caspase-3 signal (Figure 2) in the dual inactivation scenario, we do not expect to see an increase in engulfment of synaptic material.

We did not measure astrocyte-mediated engulfment with single or dual inactivation, as we did not see a robust caspase-3 dependent phenotype in astrocyte-mediated engulfment.

Author response:

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

Public Review:

Summary:

Bursicon is a key hormone regulating cuticle tanning in insects. While the molecular mechanisms of its function are rather well studied--especially in the model insect Drosophila melanogaster, its effects and functions in different tissues are less well understood. Here, the authors show that bursicon and its receptor play a role in regulating aspects of the seasonal polyphenism of Cacopsylla chinensis. They found that low temperature treatment activated the bursicon signaling pathway during the transition from summer form to winter form and affect cuticle pigment and chitin content, and cuticle thickness. In addition, the authors show that miR-6012 targets the bursicon receptor, CcBurs-R, thereby modulating the function of bursicon signaling pathway in the seasonal polyphenism of C. chinensis. This discovery expands our knowledge of the roles of neuropeptide bursicon action in arthropod biology.

However, the study falls short of its claim that it reveals the molecular mechanisms of a seasonal polyphenism. While cuticle tanning is an important part of the pear psyllid polyphenism, it is not the equivalent of it. First, there are other traits that distinguish between the two morphs, such as ovarian diapause (Oldfield, 1970), and the role of bursicon signaling in regulating these aspects of polyphenism were not measured. Thus, the phenotype in pear psyllids, whereby knockdown bursicon reduces cuticle tanning seems to simply demonstrate the phenotypes of Drosophila mutants for bursicon receptor (Loveall and Deitcher, 2010, BMC Dev Biol) in another species (Fig. 2I, 4H). Second, the study fails to address the threshold nature of cuticular tanning in this species, although it is the threshold response (specifically, to temperature and photoperiod) that distinguishes this trait as a part of a polyphenism. Whereas miR-6012 was found to regulate bursicon expression, there no evidence is provided that this microRNA either responds to or initiates a threshold response to temperature. In principle, miR-6012 could regulate bursicon whether or not it is part of a polyphenism. Thus, the impact of this work would be significantly increased if it could distinguish between seasonal changes of the cuticle and a bona fide reflection of polyphenism.

Thanks for your valuable suggestion. We concur with the review’s comment that cuticle tanning does not equate to the C. chinensis polyphenism. To better reflect the core focus of our research, we have revised the title to "Neuropeptide Bursicon and its receptor mediated the transition from summer-form to winter-form of Cacopsylla chinensis".

In response to the reviewer's inquiry regarding the threshold nature of cuticular tanning in C. chinensis, we have included a detailed analysis of the phenotypic changes (including nymph phenotypes, cuticle pigment absorbance, and cuticle thickness) during the transition from summer-form to winter-form in C. chinensis at distinct time intervals (3, 6, 9, 12, 15 days) under different temperature conditions (10°C and 25°C). As shown in Figure S1, nymphs exhibit a light yellow and transparent coloration at 3, 6, and 9 days, while nymphs at 12 and 15 days display shades of yellow-green or blue-yellow under 25°C conditions. At 10°C conditions, the abdomen end turns black at 3, 6, and 9 days. By the 12 days, numerous light black stripes appear on the chest and abdomen of nymphs at 10°C. At 15 days, nymphs exhibit an overall black-brown appearance, featuring dark brown stripes on the left and right sides of each chest and abdominal section. Furthermore, the end of the abdomen and back display a large black-brown coloration at 10°C (Figure S1A). The UV absorbance of the total pigment extraction at a 300 nm wavelength markedly increases following 10°C exposure for 6, 9, 12, and 15 days compared to the 25°C treatment group (Figure S1B). Cuticle thicknesses also increased following 10°C exposure for 6, 9, 12, and 15 days compared to the 25°C treatment group (Figure S1C). The detailed results (L122-143), materials and methods (L647-652), and discussion (L319-322) have been added in our revised manuscript.

Regarding the response of miR-6012 to temperature, we have already determined its expression at 3, 6, 10 days under different temperatures in the previous Figure 5E. We now included additional time intervals (9, 12, 15 days) in the updated Figure 5E. Our results indicate a significant decrease in the expression levels of miR-6012 after 10°C treatment for 3, 6, 9, 12, 15 days compared to the 25°C treatment group. Detailed information regarding this has been integrated into the Materials and Methods (Line 608-610) of our revised manuscript.

Strengths:

This study convincingly identifies homologs of the genes encoding the bursicon subunits and its receptor, showing an alignment with those of another psyllid as well as more distant species. It also demonstrates that the stage- and tissue-specific levels of bursicon follow the expected patterns, as informed by other insect models, thus validating the identity of these genes in this species. They provide strong evidence that the expression of bursicon and its receptor depend on temperature, thereby showing that this trait is regulated through both parts of the signaling mechanism.

Several parallel measurements of the phenotype were performed to show the effects of this hormone, its receptor, and an upstream regulator (miR-6012), on cuticle deposition and pigmentation (if not polyphenism per se, as claimed). Specifically, chitin staining and TEM of the cuticle qualitatively show difference between controls and knockdowns, and this is supported by some statistical tests of quantitative measurements (although see comments below). Thus, this study provides strong evidence that bursicon and its receptor play an important role in cuticle deposition and pigmentation in this psyllid.

The study identified four miRNAs which might affect bursicon due to sequence motifs. By manipulating levels of synthetic miRNA agonists, the study successfully identified one of them (miR-6012) to cause a cuticle phenotype. Moreover, this miRNA was localized (by FISH) to the cuticle, body-wide. To our knowledge, this is the first demonstrated function for this miRNA, and this study provides a good example of using a gene of known function as an entry point to discovering others influencing a trait. Thus, this finding reveals another level of regulation of cuticle formation in insects.

Weaknesses:

(1) The introduction to this manuscript does not accurately reflect progress in the field of mechanisms underlying polyphenism (e.g., line 60). There are several models for polyphenism that have been used to uncover molecular mechanisms in at least some detail, and this includes seasonal polyphenisms in Hemiptera. Therefore, the justification for this study cannot be predicated on a lack of knowledge, nor is the present study original or unique in this line of research (e.g., as reviewed by Zhang et al. 2019; DOI: 10.1146/annurev-ento-011118-112448). The authors are apparently aware of this, because they even provide other examples (lines 104-108); thus the introduction seems misleading as framed.

Thanks for your excellent suggestion. We have added the paper of Zhang et al. 2019 which recommended by reviewer (DOI: 10.1146/annurev-ento-011118-112448) in Line 57 of our revised manuscript. The statement has been revised to “However, the specific molecular mechanism underling temperature-dependent polyphenism still require further clarification” in Line 60-61 of our revised manuscript.

(2) The data in Figure 2H show "percent of transition." However, the images in 2I show insects with tanned cuticle (control) vs. those without (knockdown). Yet, based on the description of the Methods provided, there appears to be no distinction between "percent of transition" and "percent with tanning defects". This an important distinction to make if the authors are going to interpret cuticle defects as a defect in the polyphenism. Furthermore, there is no mention of intermediate phenotypes. The data in 2H are binned as either present or absent, and these are the phenotypes shown in 2I. Was the phenotype really an all-or-nothing response? Instead of binning, which masks any quantitative differences in the tanning phenotypes, the authors should objectively quantify the degree of tanning and plot that. This would show if and to what degree intermediate tanning phenotypes occurred, which would test how bursicon affects the threshold response. This comment also applies to the data in Figures 4G and 6G. Since cuticle tanning is present in more insect than just those with seasonal polyphenism, showing how this responds as a threshold is needed to make claims about polyphenism.

We appreciate your insightful comments. As shown in Figure 1 of our published paper (Zhang et al., 2013; doi.org/10.7554/eLife.88744.3) and Figure 2C-2I of the current manuscript, the transition from summer-form to winter-form entails not only external cuticular tanning but also alterations in internal cuticular chitin levels and cuticle thickness. While external cuticular tanning serves as a prominent and easily observable indicator of this transition, it is crucial to acknowledge that internal changes also play a significant role and should be taken into consideration. Therefore, we propose that the term "percent of transition" may be more suitable than "percent with tanning defects" to describe this process accurately.

In order to provide a more visually comprehensive understanding of the phenotypic changes during the transition from summer-form to winter-form, we have included images at different time points (3, 6, 9, 12, 15 days) under different temperature conditions in Figure S1A of our revised manuscript. Specifically, under the 10°C condition, nymphs exhibit abdomen tanning after 6 and 9 days of treatment, while the thorax remains untanned. By days 12 to 15, both the abdomen and thorax of the nymphs show tanning, resulting in the majority of summer-form nymphs transitioning into winter-form, as depicted in Figure 2I for comparison. This observation indicates the presence of a critical threshold for cuticle tanning of C. chinensis following exposure to 10°C. Nymphs that did not undergo the transition to winter-form succumbed to the cold, highlighting the absence of intermediate phenotypes at 12-15 days under the 10°C condition. The UV absorbance of the total pigment extraction at a 300 nm wavelength markedly increases following 10°C exposure for 6, 9, 12, and 15 days compared to the 25°C treatment group (Figure S1B). Additionally, cuticle thickness shows an increase following 10°C exposure for 6, 9, 12, and 15 days compared to the 25°C treatment group (Figure S1C). These results highlight the relationship between the threshold of cuticular tanning and the transition process. The detailed description and information have been added in Results (L122-143), Materials and Methods (L647-652), and Discussion (L319-322) of our manuscript.

(3) This study also does not test the threshold response of cuticle phenotypes to levels of bursicon, its receptor, or miR-6012. Hormone thresholds are the most widespread and, in most systems where polyphenism has been studied, the defining characteristic of a polyphenism (e.g., Nijhout, 2003, Evol Dev). Quantitative (not binned) measurements of a polyphenism marker (e.g., chitin) should be demonstrated to result as a threshold titer (or in the case of the receptor, expression level) to distinguish defects in polyphenism from those of its component trait.

Thanks for your valuable feedback. We have supplemented additional data on the phenotypes (Figure S1A), cuticle pigment absorbance (Figure S1B), cuticle thickness (Figure S1C), expression levels of bursicon (Figure 1E and 1F), its receptors (Figure 3G), and miR-6012 (Figure 5E) corresponding to nymphs treated over different time periods (3, 6, 9, 12, 15 days) under both 10°C and 25°C conditions in our revised manuscript.

While all these identified markers exhibit a strong correlation with the transition from summer-form to winter-form, it is important to note that they are not suitable as definitive thresholds due to the nature of relative gene expression quantification and chitin content assessment, rather than absolute quantitation. Further, given that tanning hormones are neuropeptides present in trace amounts in insects, unlike steroid hormones, determining their titers poses a considerable challenge.

(4) Cuticle issue:

(a) Unlike Fig. 6D and F, Figs. 2D and F do not correspond to each other. Especially the lack and reduction of chitin in ds-a+b! By fluorescence microscopy there is hardly any signal, whereas by TEM there is a decent cuticle. Additionally, the dsGFP control cuticle in 2D is cut obliquely with a thick and a thin chitin layer. This is misleading.

Thanks for your insightful feedback. We have replaced the previous WGA chitin staining images in the dsCcbursα+β treatment of Figure 2D with new representative images aligning with Figure 2F. Furthermore, the presence of both thin and thick chitin layers observed in the dsEGFP treatment of Figure 2D could potentially be ascribed to the chitin content in the insect midgut or fat body as previously discussed (Zhu et al., 2016). It is notable that during the process of cuticle staining, the chitin located in the midgut and fat body of C. chinensis may exhibit green fluorescence, leading to the appearance of a thin chitin layer. A detailed analysis and elucidation of these observations have been added in the discussion section (Lines 347-352) of our revised manuscript.

Zhu KY, Merzendorfer H, Zhang W, Zhang J, Muthukrishnan S. Biosynthesis, Turnover, and Functions of Chitin in Insects. Annu Rev Entomol. 2016;61:177-196. doi:10.1146/annurev-ento-010715-023933.

(b) In Figs. 2F and 4F, the endocuticle appears to be missing, a portion of the procuticle that is produced post-molting. As tanning is also occurring post-molting, there seems to be a general problem with cuticle differentiation at this time point. This may be a timing issue. Please clarify.

Thank you for your suggestion. The insect cuticle typically comprises three distinct layers (endocuticle, exocuticle, and epicuticle), with the thickness of each layer varying among different insect species. Cuticle differentiation is closely linked to the molting cycle of insects (Mrak et al., 2017). In our study, nymphal cuticles exhibited normal differentiation patterns, characterized by a thin epicuticle and comparable widths of the endocuticle and exocuticle following dsEGFP treatment, as illustrated in Figure 2F and 4F. Conversely, nymphs treated with dsCcBurs-α, dsCcBurs-β, and dsCcburs-R displayed impaired development, manifesting only the exocuticle without a discernible endocuticle layer. These findings suggest that bursicon genes and their receptor play a pivotal role in regulating insect cuticle development (Costa et al., 2016). We have added some discussion about these results in Lines 356-367 of our revised manuscript.

Mrak, P., Bogataj, U., Štrus, J., & Žnidaršič, N. (2017). Cuticle morphogenesis in crustacean embryonic and postembryonic stages. Arthropod structure & development, 46(1), 77–95. https://doi.org/10.1016/j.asd.2016.11.001

Costa, C. P., Elias-Neto, M., Falcon, T., Dallacqua, R. P., Martins, J. R., & Bitondi, M. (2016). RNAi-mediated functional analysis of Bursicon genes related to adult cuticle formation and tanning in the Honeybee, Apis mellifera. PloS one, 11(12), e0167421. https://doi.org/10.1371/journal.pone.0167421

(c) To provide background information, it would be useful analyze cuticle formation in the summer and winter morphs of controls separately by light and electron microscopy. More baseline data on these two morphs is needed.

Thanks for your valuable feedback. To provide more background information about cuticle formation, we supplied the results of nymph phenotypes, cuticle pigment absorbance, and cuticle thickness at distinct time intervals (3, 6, 9, 12, 15 days) under different temperatures of 10°C and 25°C in Figure S1 of our revised manuscript. Hope these results can help better understand the baseline data on these two morphs.

(d) For the TEM study, it is not clear whether the same part of the insect's thorax is being sectioned each time, or if that matters. There is not an obvious difference in the number of cuticular layers, but only the relative widths of those layers, so it is difficult to know how comparable those images are. This raises two questions that the authors should clarify. First, is it possible that certain parts of the thoracic cuticle, such as those closer to the intersegmental membrane, are naturally thinner than other parts of the body? Second, is the tanning phenotype based on the thickness or on the number of chitin layers, or both? The data shown later in Figure 4I, J convincingly shows that the biosynthesis pathway for chitin is repressed, but any clarification of what this might mean for deposition of chitin would help to understand the phenotypes reported. Also, more details on how the data in Fig. 2G were collected would be helpful. This also goes for the data in Fig. 4 (bursicon receptor knockdowns).

Thanks for your great comment. The TEM investigation adhered to a standardized protocol was used as previous description (Zhang et al., 2023), Initially, insect heads were uniformly excised and then fixed in 4% paraformaldehyde. Subsequently, a consistent cutting and staining procedure was executed at a uniform distance above the insect's thorax. The dorsal region of the thorax was specifically chosen for subsequent fluorescence imaging or transmission electron microscopy assessments with the specific objective of quantifying cuticle thickness. Regarding the measurement of cuticle thickness, use the built-in measuring ruler on the software to select the top and bottom of the same horizontal line on the cuticle. Measure the cuticle of each nymph at two close locations. Six nymphs were used for each sample. Randomly select 9 values and plot them. The related description has been added in the Materials and Methods (Line 660-668) of our revised manuscript.

Zhang, S.D., Li, J.Y., Zhang, D.Y., Zhang, Z.X., Meng, S.L., Li, Z., & Liu, X.X. (2023). MiR-252 targeting temperature receptor CcTRPM to mediate the transition from summer-form to winter-form of Cacopsylla chinensis. eLife, 12. https://doi.org/10.7554/eLife.88744

(5) Tissue issue:

The timed experiments shown in all figures were done in whole animals. However, we know from Drosophila that Bursicon activity is complex in different tissues. There is, thus, the possibility, that the effects detected on different days in whole animals are misleading because different tissues--especially the brain and the epidermis, may respond differentially to the challenge and mask each other's responses. The animal is small, so the extraction from single tissue may be difficult. However, this important issue needs to be addressed.

Thanks for your excellent suggestion. We express our heartfelt appreciation to the reviewer for their valuable input regarding the challenges involved in dissecting various tissue sections from the diminutive early instar nymphs of C. chinensis. In light of the metamorphic transition of C. chinensis across developmental stages, this study concentrated on examining the extensive phenotypic alterations. Consequently, intact samples of C. chinensis were specifically chosen for for qPCR analysis. The related descriptions have been added in the Materials and Methods (Line 513, 517, 553, 555, and 613) and Discussion (Line 327-329) of our revised manuscript.

(6) No specific information is provided regarding the procedure followed for the rescue experiments with burs-α and burs-β (How were they done? Which concentrations were applied? What were the effects?). These important details should appear in the Materials and Methods and the Results sections.

Thanks for your excellent suggestion. For the rescue experiments, the dsRNA of CcBurs-R and proteins of burs α-α, burs β-β homodimers, or burs α-β heterodimer (200 ng/μL) were fed together. The concentration of heterodimer protein of CcBurs-α+β was 200 ng/μL. The heterodimer protein of CcBurs-α+β fully rescued the effect of RNAi-mediated knockdown on CcBurs-R expression, while α+α or β+β homodimers did not (Figure 3F). Feeding the α+β heterodimer protein fully rescued the defect in the transition percent and morphological phenotype after CcBurs-R knockdown (Figure 4G-4H). We have added the detailed methods of rescued experiments and specific concentrations in the Materials and Methods (Line 561-563), and Results (Line 263) of our revised manuscript.

(7) Pigmentation

(a) The protocol used to assess pigmentation needs to be validated. In particular, the following details are needed: Were all pigments extracted? Were pigments modified during extraction? Were the values measured consistent with values obtained, for instance, by light microscopy (which should be done)?

Thanks for your excellent comment. Our protocol for pigment extracted as detailed in Bombyx mori, the cuticles were pulverized in liquid nitrogen and then dissolved in 30 milliliters of acidified methanol (Futahashi et al., 2012; Osanai-Futahashi et al., 2012). Thus, all cuticle pigments were dissected and treated with acidified methanol. Pigments were not modified during extraction.. The details description have been integrated into the Materials and Methods (Line 630-633) of our revised manuscript.

Futahashi, R., Kurita, R., Mano, H., & Fukatsu, T. (2012). Redox alters yellow dragonflies into red. Proceedings of the National Academy of Sciences of the United States of America, 109(31), 12626–12631. https://doi.org/10.1073/pnas.1207114109

Osanai-Futahashi, M., Tatematsu, K. I., Yamamoto, K., Narukawa, J., Uchino, K., Kayukawa, T., Shinoda, T., Banno, Y., Tamura, T., & Sezutsu, H. (2012). Identification of the Bombyx red egg gene reveals involvement of a novel transporter family gene in late steps of the insect ommochrome biosynthesis pathway. The Journal of biological chemistry, 287(21), 17706–17714. https://doi.org/10.1074/jbc.M111.321331

(b) In addition, pigmentation occurs post-molting; thus, the results could reflect indirect actions of bursicon signaling on pigmentation. The levels of expression of downstream pigmentation genes (ebony, lactase, etc) should be measured and compared in molting summer vs. winter morphs.

Thanks for your valuable suggestion. Actually, we already studied the function of some downstream pigmentation genes, including ebony, Lactase, Tyrosine hydroxylase, Dopa decarboxylase, and Acetyltransferase. The variations in the expression patterns of these genes are closely tied to the molting dynamics of nymphs undergoing transitions between summer-form and winter-form. These findings will put in another manuscript currently being prepared for submission, thus detailed outcomes are not suitable for inclusion in the current manuscript.

(8) L236: "while the heterodimer protein of CcBurs α+β could fully rescue the effect of CcBurs-R knockdown on the transition percent (Figure 4G 4H)". This result seems contradictory. If CcBurs-R is the receptor of bursicon, the heterodimer protein of CcBurs α+β should not be able to rescue the effect of CcBurs-R knockdown insects. How can a neuropeptide protein rescue the effect when its receptor is not there! If these results are valid, then the CcBurs-R would not be the (sole) receptor for CcBurs α+β heterodimer. This is a critical issue for this manuscript and needs to be addressed (also in L337 in Discussion).

Thanks for your insightful suggestion. Following the administration of dsCcBur-R to C. chinensis, the expression of CcBurs-R exhibited a reduction of approximately 66-82% as depicted in Figure 4A, rather than complete suppression. Activation of endogenous CcBurs-R through feeding of the α+β heterodimer protein results in an increase in CcBurs-R expression, with the effectiveness of the rescue effect contingent upon the dosage of the α+β heterodimer protein. Consequently, the capacity of the α+β heterodimer protein to effectively mitigate the impacts of CcBurs-R knockdown on the conversion rate is clearly demonstrated. We have added additional discussion in Line 396-403 of our revised manuscript.

(9) Fig. 5D needs improvement (the magnification is poor) and further explanation and discussion. mi6012 and CcBurs-R seem to be expressed in complementary tissues--do we see internal tissues also (see problem under point 2)? Again, the magnification is not high enough to understand and appreciate the relationships discussed.

Thanks for your valuable suggestion. In order to enhance the resolution of the magnified images, we conducted FISH co-localization of miR-6012 and CcBurs-R in 3rd instar nymphs and obtained detailed zoomed-in images. As shown in the magnified view of Figure 5D, miR-6012 and CcBurs-R appear to exhibit complementary expression patterns in tissues. During the FISH assays, epidermis transparency of C. chinensis was achieved via decolorization treatment. Noteworthy observations from Figure 3G and Figure 5E reveal an inverse correlation in the expression profiles of CcBurs-R and miR-6012. Consequently, the FISH results distinctly highlight a significant disparity in the expression levels of CcBurs-R and miR-6012 within the same tissue. We have added related explanation and discussion in Line 291-293 of our revised manuscript.

(10) The schematic in Fig. 7 is a useful summary, but there is a part of the logic that is unsupported by the data, specifically in terms of environmental influence on cuticle formation (i.e., plasticity). What is the evidence that lower temperatures influence expression of miR-6012? The study measures its expression over life stages, whether with an agonist or not, over a single temperature. Measuring levels of expression under summer form-inducing temperature is necessary to test the dependence of miR-6012 expression on temperature. Otherwise, this result cannot be interpreted as polyphenism control, but rather the control of a specific trait.

Thanks for your great suggestion. We actually conducted the assessment of miR-6012 expression at specific time intervals (3, 6, 9, 12, 15 days) under different temperatures of 10°C and 25°C. As depicted in Figure 5E, the expression levels of miR-6012 were notably reduced at 10°C compared to 25°C. Additionally, the evaluation of agomir-6012 expression level of C. chinensis under 25°C conditions at various time points (3, 6, 9, 12, 15 days) revealed no significant changes. Hence, we suggest that the impact of miR-6012 on the seasonal morphological transition is influenced upon temperature.

Recommendations for the authors:

The authors report a novel role of Bursicon and its receptor in regulating the seasonal polyphenism of Cacopsylla chinensis. They found that low temperature treatment (10°C) activated the Bursicon signaling pathway during the transition from summer-form to winter-form, which influences cuticle pigment content, cuticle chitin content, and cuticle thickness. Moreover, the authors identified miR-6012 and show that it targets CcBurs-R, thereby modulating the function of Bursicon signaling pathway in the seasonal polyphenism of C. chinensis. This discovery expands our knowledge of multiple roles of neuropeptide bursicon action in arthropod biology. However, the m

anuscript does have several major weaknesses, described under "Public review", which the authors need to address.

Major issues:

(1) L152-154 Fig S2E and S2F: Bursicon has been shown to be expressed in the CNS in a specific set of neurons. For example, In the larval CNS of Manduca sexta, bursicon expression is restricted to the subesophageal ganglion (SG), thoracic ganglia, and first abdominal ganglion. Pharate pupae and pharate adults show expression of this heterodimer in all ganglia. In Drosophila larvae, expression of a bursicon heterodimer is confined to abdominal ganglia. The additional neurons in the ventral nerve cord express only burs. In pharate adults, bursicon is produced by neurons in the SG and abdominal ganglia. I am wondering where bursicon subunits are expressed in the C. chinensis CNS? Since the authors have the antibodies, it would be useful to include immunocytochemical staining of bursicon alpha and beta in the CNS. The qPCR results from head or other tissues (Fig S2E and S2F) is not the most informative way to document localization of gene expression. Regarding the qPCR results, they show that the cuticle and the fat body express CcBurs-α and CcBurs-β. Can the authors confirm this unexpected results independently?

Thanks for your insightful comment. In this study, we did not directly used antibodies targeting bursicon subunits, instead, the bursicon subunits along with a histidine tag were integrated into the expression vector pcDNA3.1 using homologous recombination. The experimental procedures were executed as follows: initially, the histidine tag was fused to the pcDNA3.1-mCherry vector through homologous recombination to generate the recombinant plasmid pcDNA3.1-his-mCherry. Subsequently, the amino acid sequences of the two bursicon subunits were introduced into the pcDNA3.1-his-mCherry vector via homologous recombination to produce the recombinant plasmids pcDNA3.1-CcBurs-α-his-mCherry and pcDNA3.1-CcBurs-β-his-mCherry. Finally, the P2A sequence was incorporated into the vector using reverse PCR to yield the recombinant plasmids pcDNA3.1-CcBurs-α-his-P2A-mCherry and pcDNA3.1-CcBurs-β-his-P2A-mCherry. Consequently, the bursicon subunits, along with the histidine tag, were capable of generating fusion proteins with the histidine tag. Western blot analysis was conducted using antibodies targeting the histidine tag, enabling the detection of histidine expression, which corresponds to the expression of the bursicon subunits. However, they are not suitable to conduct the in vivo immunocytochemical staining of bursicon alpha and beta in the CNS.

Due to the diminutive size of the C. chinensis nymphs, dissection of the central nervous system (CNS) was unfeasible, precluding specific assessment of bursicon expression in the CNS. Prior literature has documented the expression of bursicon subunits in the epidermis and fat body of C. chinensis. Studies suggest that bursicon subunits not only play a role in the melanization and sclerotization processes of insect epidermis but also have significant roles in insect immunity (An et al., 2012). The presence of bursicon subunits in the epidermis, gut, and fat body of C. chinensis may indicate their crucial roles in the immune functions of these tissues. Further investigation is required to elucidate the specific immune functions they perform, hinting at the potential expression of these bursicon subunits in these two tissues.

An, S., Dong, S., Wang, Q., Li, S., Gilbert, L. I., Stanley, D., & Song, Q. (2012). Insect neuropeptide bursicon homodimers induce innate immune and stress genes during molting by activating the NF-κB transcription factor Relish. PloS one, 7(3), e34510. https://doi.org/10.1371/journal.pone.0034510

(2) L222: "CcBurs-R is the Bursicon receptor of C. chinensis". Is this statement supported by affinity binding assay results?

Thanks for your excellent suggestion. We employed a fluorescence-based assay to quantify calcium ion concentrations and investigate the binding affinities of bursicon heterodimers and homodimers to the bursicon receptor across varying concentrations. Our findings suggest that activation of the receptor by the burs α-β heterodimer leads to significant alterations in intracellular calcium ion levels, whereas stimulation with burs α-α and burs β-β homodimers, in conjunction with Adipokinetic hormone (AKH), maintains consistent intracellular calcium ion levels. Consequently, this research definitively identifies CcBurs-R as the bursicon receptor. For further details, please refer to the Materials and Methods (Lines 493-504), Results (Lines 231-239), and Discussion (Lines 377-384) of our revised manuscript.

(3) L245 Figure 4I-4J: Since knockdown of bursicon and its receptor cause a decrease pigment accumulation in the cuticle, it would be useful to examine 1-2 rate limiting enzyme-encoding genes in the bursicon regulated cuticle darkening process if possible (as was done for genes involved in cuticle thickening).

Thanks for your excellent comment. Following the further study, a thorough analysis was conducted to evaluate the impact of bursicon and its receptor on the expression levels of Lactase, Tyrosine hydroxylase, Dopa decarboxylase, Acetyltransferase, and the effects of RNA interference targeting these genes on the seasonal morphological transition. The findings underscored their role in the bursicon-mediated cuticle darkening process. However, as this section is slated for inclusion in an upcoming manuscript intended for submission, it is deemed unsuitable for incorporation into the current manuscript.

Minor issues:

(1) L75 "stronger resistance (Ge et al., 2019; Tougeron et al., 2021)". Stronger resistance to what? Stronger resistance to environmental stress or weather condition? Please clarify.

Thanks for your excellent suggestion. We have changed the statement to “stronger resistance to weather condition” in Line 75 of our revised manuscript.

(2) L132 Figure 1A and 1B: Bursicon sequence was first identified and functionally characterized in Drosophila melanogaster: is there any reason why Drosophila bursicon sequences were not included in the comparison?

Thanks for your excellent comment. We have added the sequence of Burs-α and Burs-β of D. melanogaster in the sequence alignment results of Figure 1A and 1B of our revised manuscript.

(3) Although the authors clearly identify and validate the function for the bursicon genes and its receptor's, there is no mention of whether duplicates of this gene are also present in the pear psyllid. This has been known to happen in otherwise conserved hormone pathways (e.g., insulin receptor in some insects), so a formal check of this should be done.

Thanks for your excellent comment. As shown in Figure S2A-S2B and 3B, there are two bursicon subunit genes and only one bursicon receptor gene in our selected insect species, for examples Drosophila melanogaster, Diaphorina citri, Bemisia tabaci, Nilaparvata lugens, and Sogatella furcifera. In our transcriptome database of C. chinensis, we also only identified two bursicon subunit genes and only one bursicon receptor gene.

(4) Line 41: Here, as in the title, "fascinating" is a subjective judgement that does not improve a study's presentation.

Thanks for your great comment. We have changed "fascinating" to "transformation" in Line 41 and also revised the title of our revised manuscript.

(5) Line 44: What makes some fields "cutting-edge" and others not?

Thanks for your excellent suggestion. The expression of "in cutting-edge fields" has been deleted in Line 44 of our revised manuscript.

(6) Line 97: This is a peculiar choice of reference for the concept of slower development in cold temperatures. The concept of degree-days and growth rates is old and widespread in entomology.

Thanks for your insightful comment. The reference of Nyamaukondiwa et al., 2011 in Line 95 has been deleted in our revised manuscript.

(7) Lines 149-150: What justifies the assumption that higher levels of expression mean a more important role? This gene might be just as necessary for development of the summer form, even if expressed at lower levels.

Thanks for your excellent suggestion. This sentence has been revised to “Increased gene expression levels may potentially contribute to the transition from summer-form to winter-form in C. chinensis.” in Line 168-169 of our revised manuscript.

(8) The blue arrow in Fig. 7 is confusing.

Thanks for your excellent suggestion. In Figure 7, the blue arrow represents the down-regulated expression of miR-6012. We have added a description about the blue arrow in Figure 7 of our revised manuscript.

Author response:

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

Reviewer #1 (Public Review):

Over the last decade, numerous studies have identified adaptation signals in modern humans driven by genomic variants introgressed from archaic hominins such as Neanderthals and Denisovans. One of the most classic signals comes from a beneficial haplotype in the EPAS1 gene in Tibetans that is evidently of Denisovan origin and facilitated high altitude adaptation (HAA). Given that HAA is a complex trait with numerous underlying genetic contributions, in this paper Ferraretti et al. asked whether additional HAA-related genes may also exhibit a signature of adaptive introgression. Specifically, the authors considered that if such a signature exists, they most likely are only mild signals from polygenic selection, or soft sweeps on standing archaic variation, in contrast to a strong and nearly complete selection signal like in the EPAS1. Therefore, they leveraged two methods, including a composite likelihood method for detecting adaptive introgression and a biological networkbased method for detecting polygenic selection, and identified two additional genes that harbor plausible signatures of adaptive introgression for HAA.

Strengths: 

The study is well motivated by an important question, which is, whether archaic introgression can drive polygenic adaptation via multiple small effect contributions in genes underlying different biological pathways regulating a complex trait (such as HAA). This is a valid question and the influence of archaic introgression on polygenic adaptation has not been thoroughly explored by previous studies.

The authors reexamined previously published high-altitude Tibetan whole genome data and applied a couple of the recently developed methods for detecting adaptive introgression and polygenic selection. 

Weaknesses: 

My main concern with this paper is that I am not too convinced that the reported genomic regions putatively under polygenic selection are indeed of archaic origin. Other than some straightforward population structure characterizations, the authors mainly did two analyses with regard to the identification of adaptive introgression: First, they used one composite likelihood-based method, the VolcanoFinder, to detect the plausible archaic adaptive introgression and found two candidate genes (EP300 and NOS2). Next, they attempted to validate the identified signal using another method that detects polygenic selection based on biological network enrichments for archaic variants.

In general, I don't see in the manuscript that the choice of methods here are well justified. VolcanoFinder is one among the several commonly used methods for detecting adaptive introgression (eg. the D, RD, U, and Q statistics, genomatnn, maldapt etc.). Even if the selection was mild and incomplete, some of these other methods should be able to recapitulate and validate the results, which are currently missing in this paper. Besides, some of the recent papers that studied the distribution of archaic ancestry in Tibetans don't seem to report archaic segments in the two gene regions. These all together made me not sure about the presence of archaic introgression, in contrast to just selection on ancestral variation.

Furthermore, the authors tried to validate the results by using signet, a method that detects enrichments of alleles under selection in a set of biological networks related to the trait. However, the authors did not provide sufficient description on how they defined archaic alleles when scoring the genes in the network. In fact, reading from the method description, they seemed to only have considered alleles shared between Tibetans and Denisovans, but not necessarily exclusively shared between them. If the alleles used for scoring the networks in Signet are also found in other populations such as Han Chinese or Africans, then that would make a substantial difference in the result, leading to potential false positives.

Overall, given the evidence provided by this article, I am not sure they are adequate to suggest archaic adaptive introgression. I recommend additional analyses for the authors to consider for rigorously testing their hypothesis. Please see the details in my review to the authors. 

Reviewer #2 (Public Review):

In Ferrareti et al. they identify adaptively introgressed genes using VolcanoFinder and then identify pathways enriched for adaptively introgressed genes. They also use a signet to identify pathways that are enriched for Denisovan alleles. The authors find that angiogenesis and nitric oxide induction are enriched for archaic introgression.

Strengths: 

Most papers that have studied the genetic basis of high altitude (HA) adaptation in Tibet have highly emphasized the role of a few genes (e.g. EPAS1, EGLN1), and in this paper, the authors look for more subtle signals in other genes (e.g EP300, NOS2) to investigate how archaic introgression may be enriched at the pathway level.

Looking into the biological functions enriched for Denisovan introgression in Tibetans is important for characterizing the impact of Denisovan introgression.

Weaknesses: 

The manuscript lacks details or justification about how/why some of the analyses were performed. Below are some examples where the authors could provide additional details.

The authors made specific choices in their window analysis. These choices are not justified or there is no comment as to how results might change if these choices were perturbed. For example, in the methods, the authors write "Then, the genome was divided into 200 kb windows with an overlap of 50 kb and for each of them we calculated the ratio between the number of significant SNVs and the total number of variants." 

Additional information is needed for clarity. For example, "we considered only protein-protein interactions showing confidence scores {greater than or equal to} 0.7 and the obtained protein frameworks were integrated using information available in the literature regarding the functional role of the related genes and their possible involvement in high-altitude adaptation." What do the confidence scores mean? Why 0.7?

In the method section (Identifying gene networks enriched for Denisovan-like derived alleles), the authors write "To validate VolcanoFinder results by using an independent approach". Does this mean that for signet the authors do not use the regions identified as adaptively introgressed using volcanofinder? I thought in the original signet paper, the authors used a summary describing the amount of introgression of a given region.

Later, the authors write "To do so, we first compared the Tibetan and Denisovan genomes to assess which SNVs were present in both modern and archaic sequences. These loci were further compared with the ancestral reconstructed reference human genome sequence (1000 Genomes Project Consortium et al., 2015) to discard those presenting an ancestral state (i.e., that we have in common with several primate species)." It is not clear why the authors are citing the 1000 genomes project. Are they comparing with the reference human genome reference or with all populations in the 1000 genomes project? Also, are the authors allowing derived alleles that are shared with Africans? Typically, populations from Africa are used as controls since the Denisovan introgression occurred in Eurasia.

The methods section for Figures 4B, 4C, and 4D is a little hard to understand. What is the x-axis on these plots? Is it the number of pairwise differences to Denisovan? The caption is not clear here. The authors mention that "Conversely, for non-introgressed loci (e.g., EGLN1), we might expect a remarkably different pattern of haplotypes distribution, with almost all haplotype classes presenting a larger proportion of non-Tibetan haplotypes rather than Tibetan ones." There is clearly structure in EGLN1. There is a group of non-Tibetan haplotypes that are closer to Denisovan and a group of Tibetan haplotypes that are distant from Denisovan...How do the authors interpret this? 

In the original signet paper (Guoy and Excoffier 2017), they apply signet to data from Tibetans. Zhang et al. PNAS (2021) also applied it to Tibetans. It would be helpful to highlight how the approach here is different. 

We thank the Reviewers for having appreciated the rationale of our study and to have identified potential issues that deserve to be addressed in order to better focus on robust results specifically supported by multiple approaches.

First, we agree with the Reviewers that clarification and justification for the methodologies adopted in the present study should be deepened with respect to what done in the original version of the manuscript, with the purpose of making it more intelligible for a broad range of scientists. As reported thoroughly in the revised version of the text, the VolcanoFinder algorithm, which we used as the primary method to discover new candidate genomic regions affected by events of adaptive introgression, was chosen among several approaches developed to detect signatures ascribable to such an evolutionary process according to the following reasons: i) VolcanoFinder is one of the few methods that can test jointly events of both archaic introgression and adaptive evolution (e.g., the D statistic cannot formally test for the action of natural selection, having been also developed to provide genome wide estimates of allele sharing between archaic and modern groups rather than to identify specific genomic regions enriched for introgressed alleles); ii) the model tested by the VolcanoFinder algorithm remarkably differs from those considered by other methods typically used to test for adaptive introgression, such as the RD, U and Q statistics, which are aimed at identifying chromosomal segments showing low divergence with respect to a specific archaic sequence and/or enriched in alleles uniquely shared between the admixed group and the source population, as well as characterized by a frequency above a certain threshold in the population under study, thus being useful especially to test an evolutionary scenario conformed to that expected in the case that adaptation was mediated by strong selective sweeps rather than weak polygenic mechanisms (see answer to comment #1 of Reviewer #1 for further details); iii) VolcanoFinder relies on less demanding computational efforts respect to other algorithms, such as genomatnn and Maladapt, which also require to be trained on large genomic simulations built specifically to reflect the evolutionary history of the population under study, thus increasing the possibility to introduce bias in the obtained results if the information that guides simulation approaches is not accurate.

Despite that, we agree with Reviewer #2 that some criteria formerly implemented during the filtering of VolcanoFinder results (e.g., normalization of LR scores, use of a sliding windows approach, and implementation of enrichment analysis based on specific confidence scores) might introduce erratic changes, which depend on the thresholds adopted, in the list of the genomic regions considered as the most likely candidates to have experienced adaptive introgression. To avoid this issue, and to adhere more strictly to the VolcanoFinder pipeline of analyses developed by Setter et al. 2020, in the revised version of the manuscript we have opted to use raw LR scores and to shortlist the most significant results by focusing on loci showing values falling in the top 5% of the genomic distribution obtained for such a statistic (see Materials and methods for details). 

Moreover, to further reduce the use of potential arbitrary filtering thresholds we decided to do not implement functional enrichment analysis to prioritize results from the VolcanoFinder method. To this end, although a STRING confidence score (i.e., the approximate probability that a predicted interaction exists between two proteins belonging to the same functional pathway according to information stored in the KEGG database) above 0.7 is generally considered a high confidence score (string-db.org, Szklarczyk et al. 2014), we replaced such a prioritization criterion by considering as the most robust candidates for adaptive introgression only those genomic regions that turned out to be supported by all the approaches used (i.e., VolcanoFinder, Signet, LASSI and Haplostrips analyses).

According to the Reviewers’ comments on the use of the Signet algorithm, we realized that the rationale beyond such a validation approach was not well described in the original version of the manuscript. First and foremost, we would like to clarify that in the present study we did not use this method to test for the action of natural selection (as it was formerly used by Gouy et al. 2017), but specifically to identify genomic regions putatively affected by archaic introgression. For this purpose, we followed the approach described by Gouy and Excoffier 2020 by searching for significant networks of genes presenting archaic-derived variants observable in the considered Tibetan populations but not in an outgroup population of African ancestry. Accordingly, we used the Signet method as an independent approach to obtain a first validation of introgressed (but not necessarily adaptive) loci pointed out by VolcanoFinder results. 

In detail, in response to the question by Reviewer #2 about which genomic regions have been considered in the Signet analysis, it is necessary to clarify that to obtain the input score associated to each gene along the genome, as required by the algorithm, we calculated average frequency values per gene by considering all the archaic-derived alleles included in the Tibetan dataset but not in the outgroup one. Therefore, we did not take into account only those loci identified as significant by VolcanoFinder analysis, but we performed an independent genome scan. Then, we crosschecked significant results from VolcanoFinder and Signet approaches and we shortlisted the genomic regions supported by both. This approach thus differs from that of Zhang et al. 2021 in which the input scores per gene were obtained by considering only those loci previously pointed out by another method as putatively introgressed. Moreover, as mentioned in the previous paragraph, our approach differs also from that implemented by Guoy et al. 2017, in which the input scores assigned to each gene were represented by the variants showing the smallest P-value associated to a selection statistic, being thus informative about putative adaptive events but not introgression ones.

However, as correctly pointed out by both the Reviewers, we formerly performed Signet analysis by considering derived alleles shared between Tibetans and the Denisovan species, without filtering out those alleles that are observed also in other modern human populations. We agree with the Reviewers that this approach cannot rule out the possibility of retaining false positive results ascribable to ancestral polymorphisms rather than introgressed alleles. According to the Reviewers’ suggestion, we thus repeated the Signet analysis by removing derived alleles observed also in an outgroup population of African ancestry (i.e., Yoruba), by assuming that only Eurasian H. sapiens populations experienced Denisovan admixture. In detail, we considered only those alleles that: i) were shared between Tibetans and Denisovan (i.e., Denisovan-like alleles); ii) were assumed to be derived according to the comparison with the ancestral reconstructed reference human genome sequence; iii) were completely absent (i.e., present frequency equal to zero) in the Yoruba population sequenced by the 1000 Genomes Project. Despite the comment of Reviewer #1 seems to propose the possible use of Han Chinese as a further control population, we decided to do not filter out Denisovan-like derived alleles present also in this human group because evidence collected so far suggest that Denisovan introgression in the gene pool of East Asian ancestors predated the split between low-altitude and high-altitude populations (Lu et al. 2016; Hu et al. 2017) and, as mentioned before, we aimed at using the Signet algorithm to validate introgression events rather than adaptive ones (see the answer to comment #6 of Reviewer #1 for further details). Moreover, we would like to remark that we decided to maintain the Signet analysis as a validation method in the revised version of the manuscript because: i) comments from both the Reviewers converge in suggesting how to effectively improve this approach, and ii) it represents a method that goes beyond the simple identification of single putative introgressed alleles, by instead enabling us to point out those biological functions that might have been collectively shaped by gene flow from Denisovans.

In addition to validate genomic regions putatively affected by archaic introgression by crosschecking results from the VolcanoFinder and Signet analyses, according to the suggestion by Reviewer #1 we implemented a further validation procedure aimed at formally testing for the adaptive evolution of the identified candidate introgressed loci. For this purpose, we applied the LASSI likelihood haplotype based method (Harris & DeGiorgio 2020) to Tibetan whole genome data. Notably, we choose this approach mainly for the following reasons: i) because it is able to detect and distinguish genomic regions that have experienced different types of selective events (i.e. strong and weak ones); ii) it has been demonstrated to have increased power in identifying them with respect to other selection statistics (e.g., H12 and nSL) (Harris & DeGiorgio 2020). Again, we performed an independent genome scan using the LASSI algorithm and then we crosschecked the obtained significant results with those previously supported by VolcanoFinder and Signet approaches in order to shortlist genomic regions that have plausibly experienced both archaic introgression and adaptive evolution.

Moreover, we maintained a final validation step represented by Haplostrips analysis, which was instead specifically performed on chromosomal segments supported by results from both VolcanoFinder, Signet, and LASSI approaches. This enabled us to assess the similarity between Denisovan haplotypes and those observed in Tibetans (i.e., the population under study in which archaic alleles might have played an adaptive role in response to high-altitude selective pressures), Han Chinese (i.e., a sister group whose common ancestors with Tibetans have experienced Denisovan admixture, but have then evolved at low altitude), and Yoruba (i.e., an outgroup that is assumed to have not received gene flow from Denisovans). 

In conclusion, we believe that the substantial changes incorporated in the manuscript according to the Reviewers’ suggestions strongly improved the study by enabling us to focus on more solid results with respect to those formerly presented. Interestingly, although the single candidate loci supported by all the approaches now implemented for validating the obtained results have attained higher prioritization with respect to previous ones (which are supported by some but not all the adopted methods), angiogenesis still stands out as the one of the main biological functions that have been shaped by events of adaptive introgression in human groups of Tibetan ancestry. This provides new evidence for the contribution of introgressed Denisovan alleles other than the EPAS1 ones in modulating the complex adaptive responses evolved by Himalayan populations to cope with selective pressures imposed by high altitudes.

Responses to Recommendations For The Authors:

Reviewer #1:

The authors mainly relied on one method, VolcanoFinder (VF), to detect adaptive introgression signals. As one of the recently developed methods, VF indeed demonstrated statistical power at detecting mild selection on archaic variants, as well as detecting soft sweeps on standing variations. However, compared to other commonly used methods for detecting adaptive introgression, such as the U and Q stats (Racimo et al. 2017), genomatnn (Gower et al. 2021), or MaLAdapt (Zhang et al. 2023),

VF doesn't seem to have better power at capturing mild and incomplete sweeps. And it makes me wonder about the justification for choosing VF over other methods here, which is not clearly explained in the manuscript. If these adaptive introgression candidates are legitimate, even if the signals are mild, at least some of the other methods should be able to recapitulate the signature (even if they don't necessarily make it through the genome-wide significance thresholds). I would be more convinced about the archaic origin of these regions if the authors could validate their reported findings using some of the aforementioned other methods. 

According to the Reviewer’s suggestion, in the revised version of the manuscript we have expanded the considerations reported as concern the rationale that guided the choice of the adopted methods. In particular, in the Materials and methods section (see page 12) we have specificed the reasons for having used the VolcanoFinder algorithm. 

First, it represents one of the few approaches that relies on a model able to test jointly the occurrence of archaic introgression and the adaptive evolution of the genomic regions affected by archaic gene flow, without the need for considering the putative source of introgression. This was a relevant aspect for us, beacuse we planned to adopt at least two main independent (and possibly quite different in terms of the underlying approaches) methods to validate the identified candidate intregressed loci and the other algorithm we used (i.e., Signet) was explicitly based on the comparison of modern data with the archaic sequence. Accordingly, the model tested by VolcanoFinder differs from those considered by the RD, U and Q statistics. In fact, RD statistic is aimed at identifying regions of the genome with low divergence with respect to a given archaic reference, while the U/Q statistics can detect those chromosomal segments enriched in alleles that are i) uniquely shared between the admixed group (e.g., Tibetans) and the source population (e.g., Denisovans), and ii) that present a frequency above a specific threshold in the admixed population (Racimo et al. 2016). For instance, all the loci considered as likely involved in adaptive introgression events by Racimo et al. 2016 presented remarkable frequencies, with most of them showing values above 50%. That being so, we decided to do not implement these methods because we believe that they are more suitable for the detection of adaptive introgression events involving few variants with a strong effect on the phenotype, which comport a substantial increase in frequency in the population subjected to the selective pressure (i.e., cases such as that of  EPAS1), while it appears challenging to choose an arbitrary frequency threshold appropriate for the detection of weak and/or polygenic selective events. 

As regards the possible use of Maladapt or genomatnn approaches as validation methods, we believe that they rely on more demanding computational efforts with respect to the Signet algorithm and, above all, they have the disadvantage of requiring to be trained on simulated genomic data. This makes them more prone to the potential bias introduced in the obtained results by simulations that do not carefully reflect the evolutionary history of the population under study.

Overall, we do not agree with the Reviwer’s statement about the fact that we mainly relied on a single method to detect adaptive introgression signals because, as mentioned above, the Signet algorithm was specifically used to identify genomic regions putatively affected by introgression. This method relies on assumptions very similar to those described above for the U/Q statistics (e.g. it considers alleles uniquely shared between Tibetans and Denisovans), but avoids the necessity to select a frequency threshold to shortlist the most likely adaptive intregressed loci. In addition, according to another suggestion by the Reviewer we have now implemented a further approach to provide evidence for the adaptive evolution of the candidate introgressed loci (see response to comment #3).  

As regards the use of Signet, based on comments from both the Reviewers we realized that the rationale beyond such a validation approach was not well described in the original version of the manuscript. First and foremost, we would like to clarify that in the present study we did not use this method to test for the action of natural selection (as it was formerly used by Gouy et al. 2017), but specifically to identify genomic regions putatively affected by archaic introgression. For this purpose, we followed the approach described by Gouy and Excoffier (2020) by searching for significant networks of genes presenting archaic-derived variants observable in the considered Tibetan populations. That being so, we used the Signet method as an independent approach to obtain a first validation of VolcanoFinder results. However, by following suggestions from both the Reviweres, we modified the criteria adopted to filter for archaic-derived variants, by excluding those alleles in common between Denisovan and the Yoruba outgroup population (see response to comment #6 for further information regarding this aspect). 

To sum up, we think that the combination of VolcanoFinder and Signet+LASSI approaches offered a good compromise between required computational efforts to shortlist the most robust candidates of adaptive introgressed loci and the typologies of model tested (i.e. that does not diascard a priori genomic signatures ascribable to weak and/or polygenic selective events). Morevoer, we would like to remark that we decided to maintain the Signet method as a validation approach in the revised version of the manuscript because: i) comments from both the Reviewers converge in suggesting how to effectively improve this approach, and ii) it represents a method that can be used to perform both single-locus validation analysis and to search for those biological functions that have been collectively much more impacted by archaic introgression, allowing to test a more realistic approximation of the polygenic model of adaptation involving introgressed alleles. In fact, although the single candidate loci supported by all the approaches now implemented for validating the obtained results  (see responses to comments #3 and #7 for further details) have attained higher prioritization with respect to previous ones (i.e., EP300 and NOS2, which are now supported by some but not all the adopted methods), angiogenesis still stands out as one of the main biological functions that have been shaped by events of adaptive introgression in the ancestors of Tibetan populations. 

Besides, I am a little surprised to see that in Supplementary Figure 2, VF didn't seem to capture more significant LR values in the EPAS1 region (positive control of adaptive introgression) than in the negative control EGLN1 region. The author explained this as the selection on EPAS1 region is "not soft enough", which I find a bit confusing. If there is no major difference in significant values between the positive and negative controls, how would the authors be convinced the significant values they detected in their two genes are true positives? I would like to see more discussion and justification of the VF results and interpretations.

In the light of such a Reviewer’s observation and according to the Reviewer #2 overall comment on the procedures implemented for filtering VolcanoFinder results, we realized that both normalization of  LR scores and the use of a sliding windows approach might introduce erratic changes, which depend on the thresholds adopted, in the list of the genomic regions considered as the most likely candidates to have experienced adaptive introgression. To avoid this issue, and to adhere more strictly to the VolcanoFinder pipeline of analyses developed by Setter et al. 2020, in the revised version of the manuscript we have opted to use raw LR scores and to shortlist the most significant results by focusing on loci showing values falling in the top 5% of the genomic distribution obtained for such a statistic (see Materials and methods, page 13 lines 4 -16 for further details).

By following this approach, we indeed observed a pattern clearer than that previously described, in which the distribution of LR scores in the EPAS1 genomic region is remarkably different with respect to that obtained for the EGLN1 gene (Figure 2 – figure supplement 1). More in detail, we identified a total of 19 EPAS1 variants showing scores within the top 5% of LR values, in contrast to only three EGLN1 SNVs. Moreover, LR values were collectively more aggregated in the EPAS1 genomic region and showed a higher average value with respect to what observed for EGLN1. We reported LR values, as well as -log (a) scores calculated for these control genes in Supplement tables 3 and 4.

Nevertheless, we agree with the Reviewer that results pointed out by VolcanoFinder require to be confirmed by additional methods, which is was what we have done to define both new candidate adaptive intregressed loci and the considered positive/negative controls. In fact, validation analyses performed to confirm signatures of both archaic introgression and adaptive evolution (i.e., Signet, LASSI and Haplostrips) converged in indicating that Tibetan variability at the EGLN1 gene does not seem to have been shaped by archaic introgression events but only by the action of natural selection (see Results, page 5 lines 3-9, page 6 lines 23-25, page 7 lines 29-36; Discussion page 14 lines 33-36; Figure 2 – figure supplement 1B and Figure 4 – figure supplement 1B, 3B and 3D), also according to what was previously proposed (Hu et al., 2017). On the other hand, results from all validation analyses confirmed adaptive introgression signatures at the EPAS1 genomic region (see Results page 4 lines 32-37, page 5 lines 1-2 and 30-34, page 6 lines 23-29; Figure 3A, 3B and Figure 4 – figure supplement 1A, 3A and 3C). 

Finally, as already reported in the former version of the manuscript, our choice of considering EPAS1 and EGLN1 respectively as positive and negative controls for adaptive introgression was guided by previous evidence suggesting these loci as targets of natural selection in high-altitude Himalayan populations (Yang et al., 2017; Liu et al., 2022), although only EPAS1 was proved to have been involved also in an adaptive introgression event (Huerta-Sanchez et al., 2014; Hu et al., 2017). 

With that being said, I suggest the authors try to first validate the signal of positive selection in the two gene regions using methods such as H2/H1 (Garud et al. 2015), iHS (Voight et al. 2006) etc. that have demonstrated power and success at detecting mild sweeps and soft sweeps, regardless of if these are adaptive introgression.

According to the Reviewer’s suggestion, we validated the new candidate adaptive introgressed loci by using also a method to formally test for the action of natural selection. In particular, we decided to use the LASSI (Likelihood-based Approach for Selective Sweep Inference) algorithm developed by Harris & DeGiorgio (2020) mainly for the following reasons: i) it is able to identify both strong and weak genomic signatures of positive selection similarly to others approaches, but additionally it can distinguish these signals by explicitly classifying genomic windows affected by hard or soft selective sweeps; ii) when applied on simulated data generated under different demographic models and by setting a range of different values for the parameters that describe a selective event (e.g., the time at which the beneficial mutation arose, the selection coefficient s) it has been proved to have an increased power with respect to traditional selection scans, such as nSL, H2/H1 and H12 (see Harris & DeGiorgio 2020 for further details).  

According to such an approach, we were able to recapitulate signatures of natural selection previously observed in Tibetans for both EPAS1 and EGLN1 (Figure 4 – figure supplement 1 and 3C – 3D).  We also obtained comparable patterns for our previous candidate adaptive introgressed loci (i.e., EP300 and NOS2), as well as for the new ones that have been instead prioritized in the revised version of the manuscript according to consistent results also from VolcanoFinder, Signet and Haplostrips analyses (see Results, page 6 lines 30-35; Figure 4C, 4D, Figure 4 – figure supplement 2C and 2D).    

With regard to the plausible archaic origin of the haplotypes under selection in these gene regions, my concern comes from the fact that other recent studies characterizing the archaic ancestry landscape in Tibetans and East Asians (eg. SPrime reports from Browning et al. 2018, as well as ArchaicSeeker reports from Yuan et al. 2021) didn't report archaic segments in regions overlapping with EP300 and NOS2. So how would the authors explain the discrepancy here, that adaptive introgression is detected yet there is little evidence of archaic segments in the regions? 

We thank the Reviewer for the comment and the references provided. However, we read the suggested articles and in both of them it does not seem that genomes from individuals of Tibetan ancestry have been analysed. Moreover, in the study by Yuan et al. 2021 we were not able to find any table or supplementary table reporting the genomic segments showing signatures of Denisovan-like introgression in East Asian groups, with only findings from enrichment analyses performed on significant results being described for the Papuan population. Anyway, as reported below in the response to comment #5, in line with what observed by the Reviwer as concerns the original version of the manuscript, according to the additional validation analyses implemented during this revison EP300 and NOS2 received lower prioritization with respect to other loci showing more robust signatures supporting introgression of Denisovan alleles in the gene pool of Tibetan ancestors (i.e., TBC1D1, PRKAG2, KRAS and RASGRF2). Three out of four of these genes are in accordance also with previously published results supporting introgression of Denisovan alleles in the ancestors of present-day Han Chinese (Browning et al. 2018) or directly in the Tibetan genomes (Hu et al. 2017) (see Results, page 5 lines 10-21 and Supplement table 5). Despite that, the reason why not all the candidate adaptive introgression regions detected by our analyses are found among results from Browning et al. 2018 can be represented by the fact that in Han Chinese this archaic variation could have evolved neutrally after the introgression events, thus preventing the identification of chromosomal segments enriched in putative archaic introgressed variants according to VolcanoFinder and LASSI approaches (which consider also the impact of natural selection). In fact, the Sprime method implemented by Browning et al. 2018 focuses only on introgression events rather than adaptive introgression ones. For instance, the Denisovan-like regions identified with Sprime in Han Chinese by such a study do not comprise at all the EPAS1 region. 

Additionally, looking at Figure 4 and Supplementary Figure 4, the authors showed haplotype comparisons between Tibetans, Denisovan, and Han Chinese for EP300 and NOS2 regions. However, in both figures, there are about equal number of Tibetans and Han Chinese that harbor the haplotype with somewhat close distance to the Denisovan genotype. And this closest haplotype is not even that similar to the Denisovan. So how would the authors rule out the possibility that instead of adaptive introgression, the selection was acting on just an ancestral modern human haplotype?

We agree with the Reviewer that according to the analyses presented in the original version of the manuscript haplotype patterns observed at EP300 and NOS2 loci by means of the Haplostrips approach cannot ruled out the possibility that their adaptative evolution involved ancestral modern human haplotypes. In fact, after the modifications implemented in the adopted pipeline of analyses based on the Reviewers’ suggestions, their role in modulating complex adaptations to high-altitudes was confirmed also by results obtained with the LASSI algorithm (in addition to results from previous studies Bigham et al., 2010; Zheng et al., 2017; Deng et al., 2019; X. Zhang et al., 2020), but their putative archaic origin received lower prioritization with respect to other loci, being not confirmed by all the analyses performed.

Furthermore, I have a question about how exactly the authors scored the genes in their network analysis using Signet. The manuscript mentioned they were looking for enrichment of archaic-like derived alleles, and in the methods section, they mentioned they used SNPs that are present in both Denisovan and Tibetan genomes but are not in the chimp ancestral allele state. But are these "derived" alleles also present in Han Chinese or Africans? If so, what are the frequencies? And if the authors didn't use derived alleles exclusively shared between Tibetans and Denisovans, that may lead to false positives of the enrichment analysis, as the result would not be able to rule out the selection on ancestral modern human variation.

As mentioned in the response to comment #1, by following the suggestions of both the Reviewers we have modified the criteria adopted for filtering archaic derived variants exclusively shared between Denisovans and Tibetans. In particular, we retained as input for Signet analysis only those alleles that i) were shared between Tibetans and Denisovan (i.e., Denisovan-like alleles) ii) were in their derived state and iii) were completely absent (i.e., show frequency equal to zero) in the Yoruba population sequenced by the 1000 Genome Project and used here as an outgroup by assuming that only Eurasian H. sapiens populations experienced Denisovan admixture. We instead decided to do not filter out potential Denisovan-like derived alleles present also in the Han Chinese population because multiple evidence agreed at indicating that gene flow from Denisovans occurred in the ancestral East Asian gene pool no sooner than 48–46 thousand years ago (Teixeira et al. 2019; Zhang et al. 2021; Yuan et al. 2021), thus predating the split between low-altitude and high-altitude groups, which occurred approximately 15 thousand years ago (Lu et al. 2016; Hu et al. 2017). In fact, traces of such an archaic gene-flow are still detectable in the genomes of several low-altitude populations of East Asian ancestry (Yuan et al. 2021).

Concerning the above, I would also suggest the authors replot their Figure 4 and Figure S4 by adding the African population (eg. YRI) in the plot, and examine the genetic distance among the modern human haplotypes, in contrast to their distance to Denisovan.

According to the Reviewer’s suggestion, after having identified new candidate adaptive introgressed loci according to the revised pipeline of analyses, we run the Haplostrips algorithm by including in the dataset 27 individuals (i.e., 54 haplotypes) from the Yoruba population sequenced by the 1000 Genomes Project (Figure 4A, 4B, Figure 4 - figure supplement 2A, 2B, 3A).

Reviewer #2:

In the methods the authors write "Since composite likelihood statistics are not associated with pvalues, we implemented multiple procedures to filter SNVs according to the significance of their LR values." What does significance mean here?

After modifications applied to the adopted pipeline of analyses according to the Reviewers’ suggestions (see responses to public reviews and to comments #1, #3, #6, #7 of Reviewer #1), new candidate adaptive introgressed loci have been identified specifically by focusing on variants showing LR values falling in the top 5% of the genomic distribution obtained for such a statistic in order to adhere more strictly to the VolcanoFinder approach developed by Setter et al. 2020. Therefore, the related sentence in the materials and methods section was modified accordingly.

Signet should be cited the first time it appears in the manuscript. The citation in the references is wrong. It lists R. Nielsen as the last author, but R. Nielsen is not an author of this paper.

We thank the Reviewer for the comment. We have now mentioned the article by Gouy and Excoffier (2020) in the Results section where the Signet algorithm was first described and we have corrected the related reference.

I could not find Figure 5 which is cited in the methods in the main text. I assume the authors mean Supplementary Figure 5, but the supplementary files have Figure 4.

We thank the Reviewer for the comment. We have checked and modified figures included in the article and in the supplementary files to fix this issue.

I didn't see a table with the genes identified as adaptatively introgressed with VolcanoFinder. This would be useful as I believe this is the first time VolcanoFinder is being used on Tibetan data?

According to the Reviewer suggestion, we have reported in Supplement table 2 all the variants showing LR scores falling in the top 5% of the genomic distribution obtained for such a statistic, along with the associated α parameters computed by the VolcanoFinder algorithm.

It is easier for the reviewer if lines have numbers.

According to the Reviewer suggestion, we have included line numbers in the revised version of the manuscript.

Author response:

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

Reviewer #1 (Public Review):

Summary:

The authors aimed to elucidate the cytological mechanisms by which conjugated linoleic acids (CLAs) influence intramuscular fat deposition and muscle fiber transformation in pig models. Utilizing single-nucleus RNA sequencing (snRNA-seq), the study explores how CLA supplementation alters cell populations, muscle fiber types, and adipocyte differentiation pathways in pig skeletal muscles.

Thanks!

Strengths:

Innovative approach: The use of snRNA-seq provides a high-resolution insight into the cellular heterogeneity of pig skeletal muscle, enhancing our understanding of the intricate cellular dynamics influenced by nutritional regulation strategy.

Robust validation: The study utilizes multiple pig models, including Heigai and Laiwu pigs, to validate the differentiation trajectories of adipocytes and the effects of CLA on muscle fiber type transformation. The reproducibility of these findings across different (nutritional vs genetic) models enhances the reliability of the results.

Advanced data analysis: The integration of pseudotemporal trajectory analysis and cell-cell communication analysis allows for a comprehensive understanding of the functional implications of the cellular changes observed.

Practical relevance: The findings have significant implications for improving meat quality, which is valuable for both the agricultural and food industry.

Thanks!

Weaknesses:

Model generalizability: While pigs are excellent models for human physiology, the translation of these findings to human health, especially in diverse populations, needs careful consideration.

Thanks!

Reviewer #2 (Public Review):

Summary:

This study comprehensively presents data from single nuclei sequencing of Heigai pig skeletal muscle in response to conjugated linoleic acid supplementation. The authors identify changes in myofiber type and adipocyte subpopulations induced by linoleic acid at depth previously unobserved. The authors show that linoleic acid supplementation decreased the total myofiber count, specifically reducing type II muscle fiber types (IIB), myotendinous junctions, and neuromuscular junctions, whereas type I muscle fibers are increased. Moreover, the authors identify changes in adipocyte pools, specifically in a population marked by SCD1/DGAT2. To validate the skeletal muscle remodeling in response to linoleic acid supplementation, the authors compare transcriptomics data from Laiwu pigs, a model of high intramuscular fat, to Heigai pigs. The results verify changes in adipocyte subpopulations when pigs have higher intramuscular fat, either genetically or diet-induced. Targeted examination using cell-cell communication network analysis revealed associations with high intramuscular fat with fibro-adipogenic progenitors (FAPs).  The authors then conclude that conjugated linoleic acid induces FAPs towards adipogenic commitment. Specifically, they show that linoleic acid stimulates FAPs to become SCD1/DGAT2+ adipocytes via JNK signaling. The authors conclude that their findings demonstrate the effects of conjugated linoleic acid on skeletal muscle fat formation in pigs, which could serve as a model for studying human skeletal muscle diseases.

Thanks!

Strengths:

The comprehensive data analysis provides information on conjugated linoleic acid effects on pig skeletal muscle and organ function. The notion that linoleic acid induces skeletal muscle composition and fat accumulation is considered a strength and demonstrates the effect of dietary interactions on organ remodeling. This could have implications for the pig farming industry to promote muscle marbling. Additionally, these data may inform the remodeling of human skeletal muscle under dietary behaviors, such as elimination and supplementation diets and chronic overnutrition of nutrient-poor diets. However, the biggest strength resides in thorough data collection at the single nuclei level, which was extrapolated to other types of Chinese pigs.

Thanks!

Weaknesses:

While the authors generated a sizeable comprehensive dataset, cellular and molecular validation needed to be improved. For example, the single nuclei data suggest changes in myofiber type after linoleic acid supplementation, yet these data are not validated by other methodologies. Similarly, the authors suggest that linoleic acid alters adipocyte populations, FAPs, and preadipocytes; however, no cellular and molecular analysis was performed to reveal if these trajectories indeed apply. Attempts to identify JNK signaling pathways appear superficial and do not delve deeper into mechanistic action or transcriptional regulation. Notably, a variety of single cell studies have been performed on mouse/human skeletal muscle and adipose tissues. Yet, the authors need to discuss how the populations they have identified support the existing literature on cell-type populations in skeletal muscle.Moreover, the authors nicely incorporate the two pig models into their results, but the authors only examine one muscle group. It would be interesting if other muscle groups respond similarly or differently in response to linoleic acid supplementation.Further, it was unclear whether Heigai and Laiwu pigs were both fed conjugated linoleic acid or whether the comparison between Heigai-fed linoleic acid and Laiwu pigs (as a model of high intramuscular fat). With this in mind, the authors do not discuss how their results could be implicated in human and pig nutrition, such as desirability and cost-effectiveness for pig farmers and human diets high in linoleic acid. Notably, while single nuclei data is comprehensive, there needs to be a statement on data deposition and code availability, allowing others access to these datasets. Moreover, the experimental designs do not denote the conjugated linoleic acid supplementation duration. Several immunostainings performed could be quantified to validate statements. This reviewer also found the Nile Red staining hard to interpret visually and did not appear to support the conclusions convincingly. Within Figure 7, several letters (assuming they represent statistical significance) are present on the graphs but are not denoted within the figure legend.

Thanks for your suggestions! We accepted your suggestion to revised our manuscript.

For changes in myofiber type, we performed qPCR to verify the changes of muscle fiber type related gene expression after CLA treatment (Figure 2E); for changes of adipocyte and preadipocyte populations, we also performed immunofluorescence staining, qPCR, and western blotting in LDM tissues and FAPs to verify the alterations of cell types after feeding with CLA (Figure 3D, 3E, 6G, 7C, and 7D). Hence, we think these cellular and molecular results could support our conclusions.

For JNK signaling pathway, we selected this signaling pathway based on snRNA-seq dataset and verified by activator in vitro experiment. However, we did not explore the mechanistic action and the downstream transcriptional regulators need to be further discussed. We have added these in the discussion part (line 443-448).

We have added the comparation between different cell-type populations in skeletal muscles (line 362-368 and 385-390).

For changes in myofiber type of Laiwu pigs, we have discussed in our previous study(Wang et al., 2023). Interestingly, we also found in high IMF content Laiwu pigs, the percentage of type IIa myofibers had an increased tendency (29.37% vs. 23.95%) while the percentage of type IIb myofibers had a decreased tendency (38.56% vs. 43.75%) in this study. We also added this discussion in the discussion part (line 392-395).

We have supplied the information of treatment in the materials and methods part (line 469-478). We also added the discussion about significance of our study for human and pig nutrition in the discussion part (line 375-376 and 446-447).

Our data will be made available on reasonable request (line 574-576).

We have supplied the information of the CLA supplementation duration in the materials and methods part (line 465).

Porcine FAPs have little lipid droplets and we improved the image quality (Figure 7A). In Figure 7, the Nile Red staining could be quantified and we have the quantification of Oil Red O staining (Figure 7B and 7J). We also added the statistical significance in figure legend.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Suggestions for Improved or Additional Experiments, Data, or Analyses

Cross-species analysis: To strengthen the generalizability of the results, it would be beneficial to include a comparative analysis with other species, such as human, bovine, or rodent models, using publicly available snRNA-seq datasets.

Thanks! Our previous study has compared the conserved and unique signatures in fatty skeletal muscles between different species(Wang, Zhou, Wang, & Shan, 2024). We mainly focused on the regulatory mechanism of CLAs in regulating intramuscular fat deposition. However, there is still a blank in the snRNA-seq or scRNA-seq datasets about the effects of CLAs on regulating fat deposition in muscles across other species, including human, bovine or rodent models. Hence, we only analyze the regulatory mechanisms of CLAs influencing intramuscular fat deposition in pigs.

Functional link: the authors should discuss in the manuscript how the muscles differ in terms of texture, flavor, aroma, etc. before and after CLA administration or between Heigai and Laiwu to provide context and help readers better understand how the observed high-resolution cellular changes relate to these functional properties of meat.

Thanks! We have added these in the introduction part (line 90-98).

Improve figures: some figures, particularly those involving Oil Red O and Nail Red, could be improved by including higher magnification images to assess the organization of lipid droplets of individual adipocytes (Figure 7A, I, and K).

Thanks! Porcine FAPs have little lipid droplets and we improved the image quality (Figure 7A).

Reviewer #2 (Recommendations For The Authors):

All of my comments are above. However, I would recommend improving the writing as several areas throughout the results needed clarity.

Thanks! We have revised our manuscript carefully after accepting your revisions.

Wang, L., Zhao, X., Liu, S., You, W., Huang, Y., Zhou, Y., . . . Shan, T. (2023) Single-nucleus and bulk RNA sequencing reveal cellular and transcriptional mechanisms underlying lipid dynamics in high marbled pork NPJ Sci Food 7: 23. https://doi.org/10.1038/s41538-023-00203-4

Wang, L., Zhou, Y., Wang, Y., & Shan, T. (2024) Integrative cross-species analysis reveals conserved and unique signatures in fatty skeletal muscles Sci Data 11: 290. https://doi.org/10.1038/s41597-024-03114-5

Author response:

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

Reviewer #1 (Public Review):

Summary:

Moir, Merheb et al. present an intriguing investigation into the pathogenesis of Pol III variants associated with neurodegeneration. They established an inducible mouse model to overcome developmental lethality, administering 5 doses of tamoxifen to initiate the knock-in of the mutant allele. Subsequent behavioral assessments and histological analyses revealed potential neurological deficits. Robust analyses of the tRNA transcriptome, conducted via northern blotting and RNA sequencing, suggested a selective deleterious effect of the variant on the cerebrum, in contrast to the cerebellum and non-cerebral tissues. Through this work, the authors identified molecular changes caused by Pol III mutations, particularly in the tRNA transcriptome, and demonstrated its relative progression and selectivity in brain tissue. Overall, this study provides valuable insights into the neurological manifestations of certain genetic disorders and sheds light on transcripts/products that are constitutively expressed in various tissues.

Strengths:

The authors utilize an innovative mouse model to constitutively knock in the gene, enhancing the study's robustness. Behavioral data collection using a spectrometer reduces experimenter bias and effectively complements the neurological disorder manifestations. Transcriptome analyses are extensive and informative, covering various tissue types and identifying stress response elements and mitochondrial transcriptome patterns. Additionally, metabolic studies involving pancreatic activity and glucose consumption were conducted to eliminate potential glucose dysfunction, strengthening the histological analyses.

Weaknesses:

The study could have explored identifying the extent of changes in the tRNA transcriptome among different cell types in the cerebrum. Although the authors attempted to show the temporal progression of tRNA transcriptome changes between P42 and P75 mice, the causal link was not established. A subsequent rescue experiment in the future could address this gap.

Nonetheless, the claims and conclusions are supported by the presented data.

We thank Reviewer 1 for their thoughtful review and commentary.  We appreciate the reviewer’s finding that our “claims and conclusions are supported by the presented data.”   

We note that our findings on the temporal progression of transcriptional changes between P42 and P75 apply to both the Pol II and Pol III transcriptomes. Importantly, in the case of Pol III, only precursor and mature tRNAs are affected at P42 whereas at P75, numerous other Pol III transcripts are also changed.  We therefore attribute the changes in tRNA as being causal in disease initiation since this is the earliest  direct consequence of the Polr3a mutation.

To expand on the evidence demonstrating the progressive nature of Polr3-related disease in our mouse model, the revised manuscript includes new immunofluorescence data showing no change in microglial cell density in the cerebral cortex or the striatum at an early stage in the disease (Supplementary Fig. S6F, G).  This is in striking contrast to the findings at later times (P75) where the number of microglia increased significantly in the Polr3a mutant and exhibit an activated morphology (Fig. 4G,H).   

We agree with the reviewer that it will be interesting in the future to assess the impact of the Polr3a mutation in different neural cell types and to explore opportunities for suppressing disease phenotypes. 

Reviewer #2 (Public Review):

Summary:

The study "Molecular basis of neurodegeneration in a mouse model of Polr3 related disease" by Moir et.al. showed that how RNA Pol III mutation affects production, maturation and transport of tRNAs. Furthermore, their study suggested that RNA pol III mutation leads to behavioural deficits that are commonly observed in neurodegeneration. Although, this study used a mouse model to establish theses aspects, the study seems to lack a clear direction and mechanism as to how the altered level of tRNA affects locomotor behaviour. They should have used conditional mouse to delete the gene in specific brain area to test their hypothesis. Otherwise, this study shows a more generalized developmental effect rather than specific function of altered tRNA level. This is very evident from their bulk RNA sequencing study. This study provides some discrete information rather than a coherent story. My enthusiasm for publication of this article in eLife is dampened considering following reasons mentioned in the weakness.

Reviewer 2’s summary contains two misstatements: 

Moir et.al. showed that how RNA Pol III mutation affects production, maturation and transport of tRNAs.

Our experiments document the effect of a neurodegenerative disease-causing mutation in RNA polymerase III on the Pol III transcriptome with a particular focus on the tRNAome (i.e. the mature tRNA population). Experiments on the maturation and transport of tRNA were not performed as there was no indication that these processes might be negatively impacted at the earliest time point (P42). Additional comments about tRNA maturation and export are provided under points 8 and 9 (see below). 

The study seems to lack a clear direction and mechanism as to how the altered level of tRNA affects locomotor behaviour.

This comment misstates the purpose of our study while overlooking the important results. As stated in the abstract, our goal was to develop “a postnatal whole-body mouse model expressing pathogenic Polr3a mutations to examine the molecular mechanisms by which reduced Pol III transcription results primarily in central nervous system phenotypes.”

Accordingly, our work provides the first molecular analysis of RNA polymerase III transcription in an animal model of Polr3-related disease. The novelty and importance of the findings, as stated in the abstract, include the discovery that a global reduction in tRNA levels (and not other Pol III transcripts) at an early stage in the disease precedes the frank induction of integrated stress and innate immune responses, activation of microglia and neuronal loss at later times. These later events readily account for the observed neurobehavioral deficits that collectively include risk assessment, locomotor, exploratory and grooming behaviors. 

Strengths:

The study created a mouse model to investigate role of RNA PolIII transcription. Furthermore, the study provided RNA seq analysis of the mutant mice and highlighted expression specific transcripts affected by the RNA PolIII mutation.

Weaknesses:

(1) The abstract is not clearly written. It is hard to interpret what is the objective of the study and why they are important to investigate. For example: "The molecular basis of disease pathogenesis is unknown." Which disease? 4H leukodystrophy? All neurodegenerative disease?

We have modified the abstract to more clearly frame the objective of the study and its importance as reflected in the title “Molecular basis of neurodegeneration in a mouse model of Polr3-related disease”. We hope the reviewer will agree that the fourth sentence of the abstract, unchanged from the initial submission, clearly outlines the objective of the study.  

(2) How cerebral pathology and exocrine pancreatic atrophy are related? How altered tRNA level connects these two axes?

It is not known how cerebral pathology and exocrine pancreatic atrophy are related beyond their shared Pol III dysfunction in our mouse model of Polr3-related disease. We anticipate that altered tRNA levels connect these two axes. Indeed, the pancreas and the brain are both known to be highly sensitive to perturbations affecting translation (Costa-Mattioli and Walter, 2020 Science doi: 10.1126/science.aat5314). Changes to the tRNA population in the cerebrum and cerebellum of Polr3a mutant mice were extensively documented in the manuscript (e.g. Figs. 3, 5 and 6).  We also found reduced tRNA levels in the pancreas of the mutant mice but did not report these findings due to the absence of a stable reference transcript in total RNA from the atrophied pancreatic tissue, even at the earliest time point examined (P42). 

(3) Authors mentioned that previously observed reduction mature tRNA level also recapitulated in their study. Why this study is novel then?

Our study reports the novel finding that a pathogenic Polr3a mutation causes a global reduction in the steady state levels of mature tRNAs, i.e. the levels of all tRNA decoders were reduced with the vast majority these reaching statistical significance (Fig. 6D and 6F). In the introduction we refer to several studies that examined the effect of pathogenic Polr3 mutations on the levels of Pol III-derived transcripts. We noted that these studies examined only a small number of Pol III transcripts in CRISPR-Cas9 engineered cell lines, patient-derived fibroblasts and patient blood. Thus, no study until now has tested for or reported a global defect in the abundance of mature tRNAs in any model of Polr3-related disease. Moreover, no previous study of _Polr3_related disease has analyzed Pol III transcript levels in the brain or in any other tissue. 

(4) It is very intuitive that deficit in Pol III transcription would severely affect protein synthesis in all brain areas as well as other organs. Hence, growth defect observed in Polr3a mutant mice is not very specific rather a general phenomenon.

While we agree with the simple assumption that a “deficit in Pol III transcription likely would affect protein synthesis in all brain areas as well as other organs”, this turned out not to be the case. In fact, a novel finding of our study is that not all Polr3a mutant tissues show a translation stress response despite reduced Pol III transcription and reduced mature tRNA levels. This implies that in some tissues the reduction in tRNA levels caused by the Polr3a mutation is not sufficient to affect protein synthesis, at least to a point where the Integrated Stress Response is induced. The underlying basis for the growth deficit has not been defined in this work. However, we noted in the discussion that a growth defect was previously seen in mice where expression of the Polr3a mutation was restricted to the Olig2 lineage.  In the present postnatal whole-body inducible model, we anticipate that the diminished growth of the mice results from a combination of hormonal and nutritional deficits caused by cerebral and pancreatic dysfunction.

(5) Authors observed specific myelination defect in cortex and hippocampus but not in cerebellum. This is an interesting observation. It is important to find the link between tRNA removal and myelin depletion in hippocampus or cortex? Why is myelination not affected in cerebellum?

We agree that the specific myelin defect observed in the cortex and hippocampus, but not the cerebellum, is an interesting observation. Pol III dysfunction in this model and reduced tRNA levels are common to both cerebra and cerebella, yet the pathological consequences differ between these regions.  While we do not know why this is the case, the cells that oligodendrocytes support in these regions are functionally different. We suggest in the discussion that subtle defects in oligodendrocyte function in the cerebellum may be uncovered using more sensitive or specific assays than the ones we have employed to date.  In addition, consistent with our findings in other tissues where Pol III transcription and tRNA levels are reduced but phenotypes are lacking, we suggest that oligodendrocytes in the cerebellum may have a different minimum threshold for Pol III activity than in other regions of the brain. 

(6) How was the locomotor activity measured? The detailed description is missing. Also, locomotion is primarily cerebellum dependent. There is no change in term of growth rate and myelination in cerebellar neurons. I do not understand why locomotor activity was measured.

We used a behavioral spectrometer with video tracking and pattern-recognition software to quantify ~20 home cage-like behaviors, including locomotor activity, as part of our phenotypic characterization of the mice. This experimenter-unbiased approach reported several metrics of locomotion, specifically, total Track length (the total distance traveled in the instrument), Center Track length and the time spent running (Run Sum) and standing still (Still Sum) in a longitudinal study (Figs. 2A-C and Supplemental Fig. S3A-C). The Materials and Methods section on mouse behavior has been amended to provide a detailed description of these experiments. 

locomotion is primarily cerebellum dependen_t_

While we agree that the cerebellum plays a critical role in balance and locomotion, regions of the cerebrum that are affected in our mice, including the primary motor cortex and the basal ganglia (Fig. 4), also have important roles in locomotor activity and control. 

(7) The correlation with behavioural changes and RNA seq data is missing. There a number of transcripts are affected and mostly very general factors for cellular metabolism. Most of them are RNA Pol II transcribed. How a Pol III mutation influences RNA Pol II driven transcription? I did not find differential expression of any specific transcripts associated with behavioural changes. What is the motivation for transcriptomics analysis? None of these transcripts are very specific for myelination. It is rather a general cellular metabolism effect that indirectly influences myelination.

The differentially expressed mRNAs identified in our RNAseq analysis at P75 reflect both direct and secondary consequences of dysfunctional Pol III transcription on Pol II transcription. These effects can be achieved by multiple mechanisms. Induction of the Integrated Stress Response (ISR) due to insufficient tRNA can be considered a direct consequence of diminished Pol III transcription on Pol II transcription. An example of a secondary response is the activation of microglia and the innate immune response (which is known to accompany prolonged activation of the ISR), and the loss of neurons and oligodendrocytes. These changes are documented in Figs. 3 and 4. Importantly, loss of neurons, activated microglia and reduced oligodendrocyte numbers are each readily reconciled with changes in behavior.  

None of these transcripts are very specific for myelination 

The RNAseq data at P75 indicates only a modest reduction in oligodendrocyte-specific gene expression (as defined by single-cell RNAseq studies of purified cell populations, Mackenzie et al., 2018 Sci. Rep. doi: 10.1038/s41598-018-27293-5). Despite this, some oligodendrocytespecific transcripts with well-known roles in myelination were down-regulated in the Polr3a mutant (e.g. Plp1, Mog and Mobp). In addition, steroid synthesis pathway transcripts involved in the production of cholesterol, an abundant and essential component of myelin, were also downregulated (Supplementary Fig. S4E).

(8) What genes identified by transcriptomics analysis regulates maturation of tRNA? Authors should at least perform RNAi study to identify possible factor and analyze their importance in maturation of tRNA.

Of the many proteins involved in the maturation of tRNA (Phizicky and Hopper, 2023 RNA doi: 10.1261/rna.079620.123), RNAseq analysis at P75 identified only amino-acyl tRNA synthetases as being differentially-expressed (fold change >1.5, p adj. < 0.05, Table S1). These genes are canonical indicators of the ATF4-dependent Integrated Stress Response and their upregulation is widely interpreted as an attempt to restore efficient translation. In addition, our analysis of Pol III transcripts at P75 identified a reduction in the level of RppH1 (Fig. 3C), the RNA component of RNase P, which removes the 5’ leader of precursor tRNAs.  However, at P42, there was no effect on RppH1 abundance, or the expression of amino-acyl tRNA synthetase genes (Fig. 5C and Table S3).  Thus, an RNAi study to identify and analyze a possible factor involved in the maturation of tRNA is neither warranted nor relevant to the current body of work.

(9) What factors are influencing tRNA transport to cytoplasm? It may be possible that Polr3a mutation affect cytoplasmic transport of tRNA. Authors should study this aspect using an imaging experiment.

Our analysis of tRNA populations in this study employed total cellular RNA and thus reflect the abundance of mature tRNA from all cellular compartments. We have not assessed whether the reduction in tRNA abundance caused by the Polr3a mutation alters the dynamics of tRNA transport from the nucleus to the cytoplasm. However, we consider it highly unlikely that the Polr3a mutation would have a significant effect on cytoplasmic transport of tRNA. Imaging experiments along these lines are beyond the scope of the current study.

(10) Does alteration of cytoplasmic level of tRNA affects translation? Author should perform translation assay using bio-orthoganal amino acid (AHA) labelling.

It is not known whether the reduced tRNA levels affect translation globally in the Polr3a mutant, but we predict that this may not be the case. Since tissues (heart and kidney) and brain regions (cerebrum and cerebellum) that share a decrease in tRNA abundance do not share activation of the Integrated Stress Response (a reporter of aberrant translation), we anticipate that effects on translation may be limited to specific regions or cell populations and to specific mRNAs within these cells. The current study provides the foundation for further work to address these questions.

Reviewer #1 (Recommendations For The Authors):

Below are a few comments, mostly regarding typographical errors, presentation, and clarity, that we believe would enhance this manuscript:

On the heatmaps generated, it would be ideal to place "WT" before "KI," with "WT" on the left. This will maintain consistency with the rest of the manuscript, where "WT" conditions precede "KI" conditions, as observed in the bar graphs and dot plots.

All heatmaps have been remade with WT on the left and KI on the right to maintain consistency throughout the manuscript. 

Authors mentioned in several instances (Discussion Pg 19 Line 2, for instance) the analysis of changes in the "Pol II transcriptome." Is this a typographical error?

The reference to the Pol II transcriptome is not a typographical error (Discussion Pg 19 Line2). Here and elsewhere in the manuscript, we are distinguishing between changes to the Pol III transcriptome and the timing of subsequent changes to the Pol II transcriptome. The text has been edited to clarify this relationship in several places.   

(1) Introduction, Page 4, last paragraph.

Analysis of the Pol III transcriptome reveals a common decrease in pre-tRNA and mature tRNA populations and few if any changes among other Pol III transcripts across multiple tissues. Analysis of the Pol II transcriptome reveals activation of the integrated stress response in cerebra but not in other surveyed tissues.

(2) Results, page 8, 2nd paragraph

To investigate the molecular changes to Pol III transcript levels caused by the Polr3a mutation and any secondary effects on the Pol II transcriptome, we initially focused on the cerebra of adult mice at P75.

(3) Discussion, Page 19, second paragraph

Pol III dysfunction and the reduction in the cerebral tRNA population at P42 coincides with behavioral deficits and precedes substantial downstream alterations in the Pol II transcriptome, which include induction of an innate immune response (IR) and an ISR, and indicators of neurodegeneration (i.e., activation of cell death pathways and loss of mitochondrial DNA). These findings suggest a causal role for the lower tRNA abundance and/or altered tRNA profile in disease progression.

In supplementary figure 1, authors validated the expression of their systems using flow cytometry and observed a high level of recombination frequency in different tissue types. Can the flow cytometry data distinguish between cell types within the cerebrum (neurons/microglia/astrocytes)?

The flow cytometry experiments reported in Supplementary Fig. S1 used a dual tdTomato-EGFP reporter to assess recombination. The cerebral and cerebellar samples were gated on fluorescence from endogenous expression of tdTomato (red), EGFP (green) and DAPI (blue) staining. In principle, flow cytometry could be used to distinguish between cell types within the cerebrum (neurons/microglia/astrocytes). However,  this would require (i) an antibody to a cell surface marker on the cell type of interest and (ii) a fluorescent probe conjugated to the primary antibody or a fluorescent secondary antibody that is spectrally well resolved from the emission spectra of tdTomato, eGFP and DAPI.

Results section 1: Is there any particular reason why P28 was chosen as the commencement of tamoxifen injection?

P28 was chosen so that any effect of the Polr3a mutation on development and differentiation would be limited in the tissues we examined. 

Fig 1C: The number of asterisks does not match between the graph and the figure legend.

Fig. 1C has been corrected to match the number of asterisks in the graph and figure legend.

Results section 3:

This section seemed a little brief, especially when compared to the depth of the succeeding sections. Authors can state in greater detail which behaviors were quantified. In S3A-C, my understanding is that the animals were placed in an open-field test. This procedure can be briefly mentioned in the methods, as well as in the main manuscript text.

In the legends of S3, a bracket is missing for "(D-F)" on line 5. Additionally, the alignment of legends for each bar graph could be consistent for all graphs except under the condition of spatial constraint.

Detailed methods pertaining to the measurement and calculation of home cage-like behaviors reported by the behavioral spectrometer have been added to the Methods section on Mouse Behavior. 

In the Results, Figs. S3A-C show anxiety-like behaviors which measure the number and duration of visits and the distance traveled  in a 15 cm2  central area of the arena. Figs. 2A-C show locomotor behaviors including Tracklength, Run sum and Still sum. The open field-like behavior is reported as total Tracklength in the behavioral spectrometer, i.e. the total distance travelled in the arena. This is now more clearly described in  the main manuscript and the Methods section. “overall locomotor activity was decreased in Polr3a-tamKI mice as indicated by the reduced track length at P42, P49, P56 and P63 (Fig. 2A).” 

The legend of S3, now has the missing bracket "(D-F)" on line 5. 

The legends within each bar graph are now consistent and aligned as much as spatial constraints allow.

Results section 4:

Similar to our earlier questions for S1, is it possible to distinguish samples derived from different cell types (neurons/glia)? In figure 4, this is mainly done post-hoc, based on the known gene expression. Maybe the authors could discuss this small limitation? In Fig S4C, the color contrast for the heatmap legend needs to be corrected.

It is not possible to accurately distinguish different neural cell sub-types, such as different types of neurons, or different types of oligodendrocytes in bulk RNAseq. Hence, we have reported only high confidence correlations based on known gene expression signatures (Fig. 4). We discuss only the data for which we can draw confident conclusions. The heatmap and legend in Fig. S4C has been amended. 

Results section 5:

In figure S5A, the alignment of asterisk significance markers could be adjusted.

Asterisks have been realigned in Fig. S5A

Reviewer #2 (Recommendations For The Authors):

Methods Section should include detailed procedure.

A detailed description of the methods pertaining to the measurement and calculation of behaviors using the behavioral spectrometer has been added to the Methods section.

Statistical tests should have detailed information

Statistical tests are detailed in the Methods section “Statistical Analysis”. Additional details pertaining to calculations of behavioral data have been added to the “Mouse behavior” section of the Methods.

Author response:

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

Public Reviews:

Reviewer #2 (Public review):

Weaknesses:

The authors have clarified that the first features available for each patient have been used. However, they have not shown that these features did not occur before the time of post-stroke epilepsy. Explicit clarification of this should be performed.

The data utilized in our analysis were collected during the first examination or test conducted after the patients' admission. We specifically excluded any patients with a history of epilepsy, ensuring that all cases of epilepsy identified in our study occurred after admission. Therefore, the features we analyzed were collected after the patients' admission but prior to the onset of post-stroke epilepsy.

Reviewer #3 (Public review):

Weaknesses:

The writing of the article may be significantly improved.

Although the external validation is appreciated, cross-validation to check robustness of the models would also be welcome.

Thank you for your helpful advice.  Performing n-fold cross-validation is a crucial step to ensure the reliability and robustness of the reported results, especially when dealing with the datasets which don't have sufficient quantity.   We revised our code and did a 5 fold cross-validation version ,it didn’t have much promote（because our model has reach the auc of 0.99）.Considering that we have sufficient quantity of more than 20000 records, we think split the dataset by 7:3 and train the model is enough for us. We have uploaded the code of 5 fold cross-validation version and ploted the 5 fold test roc  on GitHub at https://github.com/conanan/lasso-ml/lasso_ml_cross_validation.ipynb as an external resource. We  trained the 5 fold average model and ploted the 5 fold test roc curves, the results show some improvement, but it is not substantial because the best model are still tree models in the end.

External validation results may be biased/overoptimistic, since the authors informed that "The external validation cohort focused more on collecting positive cases 80 to examine the model's ability to identify positive samples", which may result in overoptimistic PPV and Sensitivity estimations. The specificity for the external validation set has not been disclosed.

Thank you for your valuable feedback regarding the external validation results. We appreciate your concerns about potential bias and overoptimism in our estimations of positive predictive value (PPV) and sensitivity.

To clarify, we have uploaded the code for external validation on GitHub at https://github.com/conanan/lasso-ml. The results indicate that the PPV is 0.95 and the specificity is 0.98.

While we focused on collecting more positive cases due to their lower occurrence rate, this approach allows us to better evaluate the model's ability to predict positive samples, which is crucial in clinical settings. We believe that emphasizing positive cases enhances the model's utility for practical applications(So a little overoptimism is acceptable ).

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

Public Reviews:

Reviewer #1 (Public Review):

Weaknesses 1:

The methodology needs further consideration. The Discussion needs extensive rewriting.

Thanks for your advice, we have revised the Discussion

Reviewer #2 (Public Review):

Weaknesses 2:

There are many typos and unclear statements throughout the paper.

There are some issues with SHAP interpretation. SHAP in its default form, does not provide robust statistical guarantees of effect size. There is a claim that "SHAP analysis showed that white blood cell count had the greatest impact among the routine blood test parameters". This is a difficult claim to make.

Thank you for your suggestion that the SHAP analysis is really just a means of interpreting the model.  In our research, we compared the SHAP analysis with traditional statistical methods, such as regression analysis.  We found the SHAP results to be consistent with the statistical results from the regression for variables like white blood cell count (see Table 1). This alignment leads us to believe the SHAP analysis is providing reliable insights in this context

The Data Collection section is very poorly written, and the methodology is not clear.

Thanks for your advice, we have revised the Data Collection section.

There is no information about hyperparameter selection for models or whether a hyperparameter search was performed. Given this, it is difficult to conclude whether one machine learning model performs better than others on this task.

Thank you for the advices of performing hyperparameter. We used the package of sklearn, xgboost, lightgbm of python 3.10 to construct the model and  didn’t change the default settings before. It is not proper and may lead to  less certain conclusions. Now we carry out grid search to select and optimize hyperparameters and they make the model better. The best model is still RF.

The inclusion and exclusion criteria are unclear - how many patients were excluded and for what reasons?

The procedure of selection is in figure1. Total there are 42079 records from the stroke database, 24733 patients were diagnosed as ischemic stroke or lacular stoke with new onset. Then we excluded hemorrage stroke(4565),history of stroke(2154), TIA(3570), unclear cause stroke(561) and records who missed important data(6496). Then we excluded patients whose seizure might be attributed to other potential causes (brain tumor, intracranial vascular malformation, traumatic brain injury,etc)(865). Then we exclude patient who had a seizure history(152) or died in hospital (1444). Then we excluded patients who were lost in follow-up (had no outpatient records and can’t contact by phone )or died within 3 months of the stroke incident(813). Finally 21459 cases are involved in this research.

There is no sensitivity analysis of the SMOTE methodology: How many synthetic data points were created, and how does the number of synthetic data points affect classification accuracy?

Thanks for your remind, we have accept these advice and change the SMOTE to SMOTEENN (Synthetic Minority Over-sampling Technique combined with Edited Nearest Neighbors) technique to resample an imbalanced dataset for machine learning. The code is

smoteenn = SMOTEENN(samplingstrategy='auto', randomstate=42)

the SMOTEENN class comes from the imblearn library. The samplingstrategy='auto' parameter tells the algorithm to automatically determine the appropriate sampling strategy based on the class distribution. The randomstate=42 parameter sets a seed for the random number generator, ensuring reproducibility of the results.

Did the authors achieve their aims? Do the results support their conclusions?

Yes, we have achieve some of the aims of predicting PSE while still leave some problem.

The paper does not clarify the features' temporal origins. If some features were not recorded on admission to the hospital but were recorded after PSE occurred, there would be temporal leakage.

The data used in our analysis is from the first examination or test conducted after the patients' admission, retrieved from a PostgreSQL database. First, we extracted the initial admission date for patients admitted due to stroke. Then, we identified the nearest subsequent examination data for each of those patients.

The sql code like follows:

SELECT TO_DATE(condition_start_date, 'DD-MM-YYYY') AS DATE

FROM diagnosis

WHERE person_id ={} and (condition_name like '%梗死%' or condition_name like '%梗塞%') and(condition_name like '%脑%'or condition_name like '%腔隙%'))

order by DATE limit 1

The authors claim that their models can predict PSE. To believe this claim, seeing more information on out-of-distribution generalisation performance would be helpful. There is limited reporting on the external validation cohort relative to the reporting on train and test data.

Thank you for the advice. The external validation is certainly very important, but there have been some difficulties in reaching a perfect solution.  We have tried using open-source databases like the MIMIC database, but the data there does not fit our needs as closely as the records from our own hospital.  The MIMIC database lacks some of the key features we require, and also lacks the detailed patient follow-up information that is crucial for our analysis.   Given these limitations, we have decided to collect newer records from the same hospitals here in Chongqing.  We believe this will allow us to build a more comprehensive dataset to support robust external validation.  While it may not be a perfect solution, gathering this additional data from our local healthcare system is a pragmatic step forward.   Looking ahead, we plan to continue expanding this Chongqing-based dataset and report on the results of the greater external validation in the future.  We are committed to overcoming the challenges around data availability to strengthen the validity and generalizability of our research findings.

For greater certainty on all reported results, it would be most appropriate to perform n-fold cross-validation, and report mean scores and confidence intervals across the cross-validation splits

Thank you for your helpful advice. Performing n-fold cross-validation is a crucial step to ensure the reliability and robustness of the reported results, especially when dealing with the datasets which don't have sufficient quantity. While we have sufficient quantity of more than 20000 records, so we think split the dataset by 7:3 and train the model is enough for us. We revised our code and did a 5 fold cross-validation version ,it had little promote（because our model has reach the auc of 0.99）, we may use this great technique in our next study if there is not enough cases.

Additional context that might help readers

The authors show force plots and decision plots from SHAP values. These plots are non-trivial to interpret, and the authors should include an explanation of how to interpret them.

Thank you for your helpful advice. It is a great improve for our draft, we have added the explanation that we use the force plot of the first person to show the influence of different features of the first person, we can see that long APTT time contribute best to PSE, then the AST level and others, the NIHSS score may be low and contribute opposite to the final result. Then the decision plot is a collection of model decisions that show how complex models arrive at their predictions

Reviewer #3 (Public Review):

Weaknesses3:

There are issues with the readability of the paper. Many abbreviations are not introduced properly and sometimes are written inconsistently. A lot of relevant references are omitted. The methodological descriptions are extremely brief and, sometimes, incomplete.

Thanks for your advice, we have revised these flaws.

The dataset is not disclosed, and neither is the code (although the code is made available upon request). For the sake of reproducibility, unless any bioethical concerns impede it, it would be good to have these data disclosed.

Thank you for your recommendations. We have made the code available on GitHub at https://github.com/conanan/lasso-ml. While the data is private and belongs to the hospital. Access can be requested by contacting the corresponding author to apply from the hospitals and specifying the purpose of inquiry.

Although the external validation is appreciated, cross-validation to check the robustness of the models would also be welcome.

Thank you for your valuable advice. Performing n-fold cross-validation is crucial for ensuring the reliability and robustness of results, especially with limited datasets. However, since we have over 20,000 records, we believe that a 70:30 split for training and testing is sufficient.

We revised our code and implemented 5-fold cross-validation, which provided minimal improvement, as our model has already achieved an AUC of 0.99. We plan to use this technique in future studies if we encounter fewer cases.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

My comments include two parts:

(1) Methodology<br /> a-This study was based on multiple clinical indicators to construct a model for predicting the occurrence of PSE. It involved various multi-class indicators such as the affected cortical regions, locations of vascular occlusion, NIHSS scores, etc. Only using the SHAP index to explain the impact of multi-class variables on the dependent variable seems slightly insufficient. It might be worth considering the use of dummy variables to improve the model's accuracy.

Thank you for the detailed feedback on the study methodology. The SHAP analysis is really just a means of interpreting the model, which we compared with the combination of SHAP and traditional statistics, so we think SHAP analysis is reliable in this research. We have used the dummy variables, expecially when dealing with the affected cortical regions, locations of vascular occlusion, for example if frontal region is involved the variable is 1. But they have less impact in the machine learning model

b-The study used Lasso regression to select 20 features to build the model. How was the optimal number of 20 features determined?

Lasso regression is a commonly used feature screening method. Since we extract information from the database and try to include as many features as possible, the cross-verification curve of lasso regression includes 78 features best, but it will lead to too complex model. We select 10,15,20,25,30 features for modeling according to the experiment. When 20 features are found, the model parameters are good and relatively concise. Improve the number of features contribute little to the model effect, decrease the number of features influence the concise of model ,for example the auc of the model with 15 features will drop under 0.95. So we finally select 20 features.

c-The study indicated that the incidence rate of PSE in the enrolled patients is 4.3%, showing a highly imbalanced dataset. If singly using the SMOTE method for oversampling, could this lead to overfitting?

Thanks for your remind, singly using the SMOTE method for oversampling is inproper. Now we have find this improvement and change the SMOTE to SMOTEENN (Synthetic Minority Over-sampling Technique combined with Edited Nearest Neighbors) technique to resample an imbalanced dataset for machine learning. First, oversampling with SMOTE and then undersampling with ENN to remove possible noise and duplicate samples. The code is

smoteenn = SMOTEENN(sampling_strategy='auto', random_state=42)

the SMOTEENN class comes from the imblearn library. The sampling_strategy='auto' parameter tells the algorithm to automatically determine the appropriate sampling strategy based on the class distribution. The random_state=42 parameter sets a seed for the random number generator, ensuring reproducibility of the results.

(2) Clinical aspects:

Line 8, history of ischemic stroke, this is misexpression, could be: diagnosis of ischemic stroke.

Line 8, several hospitals, should be more exact; how many?

Line 74 indicates that the data are from a single centre, this should be clarified.

Line 4 data collection: The criteria read unclear; please clarify further.

Thanks for your remind, we have revised the draft and correct these errors.

Line 110, lab parameters: Why is there no blood glucose?

Because many patients' blood sugar fluctuates greatly and is easily affected by drugs or diet, we finally consider HBA1c as a reference index by asking experts which is more stable.

Line 295, The author indicated that data lost; this should be clarified in the results part, and further, the treatment of missing data should be clarified in the method part.

Thanks for your remind, we have revised the draft and correct these errors.

I hope to see a table of the cohort's baseline characters. The discussion needs extensive rewriting; the author seems to be swinging from the stoke outcome and the seizure, sometimes losing the target.

Figure1 is the procedure of the selection of patients. Table1 contains the cohort's baseline characters

For the swinging from the stoke outcome and the seizure, that is because there are few articles on predicting epilepsy directly by relevant indicators, while there are more articles on prognosis. So we can only take epilepsy as an important factor in prognosis and comprehensively discuss it, or we can't find enough articles and discuss them

Reviewer #2 (Recommendations For The Authors):

There are typos and examples of text that are not clear, including:

"About the nihss score, the higher the nihss score, the more likely to be PSE, nihss score has a third effect just below white blood cell count and D-dimer."

"and only 8 people made incorrect predictions, demonstratijmng a good predictive ability of the model."

"female were prone to PSE"

" Waafi's research"

"One-heat' (should be one-hot)

Thanks for your remind, we have revised the draft and correct these errors.

The Data Collection section is poorly written, and the methodology is not clear. It would be much more appropriate to include a table of all features used and an explanation of what these features involve. It would also be useful to see the mean values of these features to assess whether the feature values are reasonable for the dataset.

Thanks for your remind. All data are from the first examination or test after admission, presented through the postgresql database . First we extract the first date of the patients who was admitted by stroke ,then we extract informations from the nearest examination from the admission. We extract by the SQL code by computer instead of others who may extract data by manual so we get as much data as possible other than only get the features which was reported before .The table of all features used and their mean±std is in table1.

The paper does not clarify the features' temporal origins. If some features were not recorded on admission to the hospital but were recorded after PSE occurred, there would be temporal leakage. I would need this clarified before believing the authors achieved their claims of building a predictive model.

All relevant index results were from the first examination after admission, and the mean standard deviation was listed in the statistical analysis section in table1.

The authors claim that their models can predict PSE. To believe this claim, seeing more information on out-of-distribution generalisation performance would be helpful. There is limited reporting on the external validation cohort relative to the reporting on train and test data.

Thank you for the advice, the external validation is very important but there are some difficulties to reach a perfect one. We have tried some of the open source database like the mimic database ,but these data don't fit our request because they don't have as much features as our hospital and lack of follow-up of the relevant patients. In the end we collected the newer records in the same hospitals in Chongqing and we will collect more and report a greater external validation in the future.

For greater certainty on all reported results, It would be most appropriate to perform n-fold cross-validation, and report mean scores and confidence intervals across the cross-validation splits.

Thank you for your helpful advice. Performing n-fold cross-validation is a crucial step to ensure the reliability and robustness of the reported results, especially when dealing with the datasets which don't have sufficient quantity. While we have sufficient quantity of more than 20000 records, so we think split the dataset by 7:3 and train the model is enough for us. We revised our code and did a 5 fold cross-validation version ,it had little promote, we will use this great technique in our next study.

The authors show force plots and decision plots from SHAP values. These plots are non-trivial to interpret, and the authors should include an explanation of how to interpret them.

It is a great improve for our draft, we have added the explanation we use the force plot of the first person to show the influence of different features of the first person, we can see that long APTT time contribute best to PSE, then the AST level and others, the NIHSS score may be low and contribute lower to the final result. Then the decision plot is a collection of model decisions that show how complex models arrive at their predictions

Reviewer #3 (Recommendations For The Authors):

Abbreviations should not be defined in the abstract )or only in the abstract).

Please explicit what are the purposes of the study you are referring to in "Currently, most studies utilize clinical data to establish statistical models, survival analysis and cox regression."

Authors affirm: "there is still a relative scarcity of research 49 on PSE prediction, with most studies focusing on the analysis of specific or certain risk factors ." This statement is especially curious since the current study uses risk factors as predictors.

It is not clear to me what the authors mean by "No study has proposed or established a more comprehensive and scientifically accurate prediction model." The authors do not summarize the statistical parameters of previously reported model, or other relevant data to assess coverage or validity (maybe including a Table summarizing such information would be appropriate. In any case, I would try to omit statements that imply, to some extent, discrediting previous studies without sufficient foundation.

"antiepileptic drugs" is an outdated name. Please use "antiseizure medications"

Thanks for your remind, we have revised the draft and correct these errors.

The authors say regarding missing data that they "filled the data of the remaining indicators with missing values of more than 1000 cases by random forest algorithm". Please clarify what you mean by "of more than 1000 cases." Also, provide details on the RF model used to fill in missing data.

Thanks for your remind. "of more than 1000 cases" was a wrong sentence and we have corrected it. Here is the procedure, first we counted the values of all laboratory indicators for the first time after stroke admission( everyone who was admitted because of stroke would perform blood routine , liver and kidney function and so on), excluded indicators with missing values of more than 10%, and filled the data of the remaining indicators with missing values by random forest algorithm using the default parameter. First, we go through all the features, starting with the one with the least missing (since the least accurate information is needed to fill in the feature with the least missing). When filling in a feature, replace the missing value of the other feature with 0. Each time a regression prediction is completed, the predicted value is placed in the original feature matrix and the next feature is filled in. After going through all the features, the data filling is complete.

Please specify what do you mean by negative group and positive group, Avoid tacit assumptions.

Thanks for your remind, we have revised the draft and correct these errors.

Please provide more details (and references) on the smote oversampling method. Indicate any relevant parameters/hyperparameters.

Thanks for your remind, we have accept these advice and change the SMOTE to SMOTEENN (Synthetic Minority Over-sampling Technique combined with Edited Nearest Neighbors) technique to resample an imbalanced dataset for machine learning. The code is

smoteenn = SMOTEENN(sampling_strategy='auto', random_state=42)

the SMOTEENN class comes from the imblearn library. The sampling_strategy='auto' parameter tells the algorithm to automatically determine the appropriate sampling strategy based on the class distribution. The random_state=42 parameter sets a seed for the random number generator, ensuring reproducibility of the results.

The methodology is presented in an extremely succinct and non-organic manner (e.g., (Model building) Select the 20 features with the largest absolute value of LASSO." Please try to improve the narrative.

Lasso regression is a commonly used feature screening method. Since we extract information from the database and try to include as many features as possible, the cross-verification curve of lasso regression includes 78 features best, but it will lead to too complex model. We select 10,15,20,25,30 features for modeling according to the experiment. When 20 features are found, the model parameters are good and relatively concise. Improve the number of features contribute little to the model effect, decrease the number of features influence the concise of model ,for example the auc of the model with 15 features will drop under 0.95. So we finally select 20 features.

Many passages of the text need references. For example, those that refer to Levene test, Welch's t-test, Brier score, Youden index, and many others (e.g., NIHSS score). Please revise carefully.

Thanks for your remind, we have revised the draft and correct these errors.

"Statistical details of the clinical characteristics of the patients are provided in the table." Which table? Number?

Thanks for your remind, we have revised the draft and correct these errors， it is in table1.

Many abbreviations are not properly presented and defined in the text, e.g., wbc count, hba1c, crp, tg, ast, alt, bilirubin, bua, aptt, tt, d_dimer, ck. Whereas I can guess the meaning, do not assume everyone will. Avoid assumptions.

ROC is sometimes written "ROC" and others, "roc." The same happens for PPV/ppv, and many other words (SMOTE; NIHSS score, etc.).

Please rephrase "ppv value of random forest is the highest, reaching 0.977, which is more accurate for the identification of positive patients(the most important function of our models).". PPV always refer to positive predictions that are corroborated, so the sentences seem redundant.

Thanks for your remind, we have revised the draft and correct these errors.

What do you mean by "Complex algorithms". Please try to be as explicit as possible. The text looks rather cryptic or vague in many passages.

Thanks for your remind, "Complex algorithms" is corrected by machine learning.

The text needs a thorough English language-focused revision, since the sense of some sentences is really misleading. For instance "only 8 people made incorrect predictions,". I guess the authors try to say that the best algorithm only mispredicted 8 cases since no people are making predictions here. Also, regarding that quote... Are the authors still speaking of the results of the random forest model, which was said to be one of the best performances?

Thanks for your remind, we have revised the draft and correct these errors.

The authors say that they used, as predictors "comprehensive clinical data, imaging data, laboratory test data, and other data from stroke patients". However, the total pool of predictors is not clear to me at this point. Please make it explicit and avoid abbreviations.

Thanks for your remind, we have revised the draft and correct these errors.

Although the authors say that their code is available upon request, I think it would be better to have it published in an appropriate repository.

Thanks for your remind, we showed our code at  https://github.com/conanan/lasso-ml.

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

The authors investigated how the presence of interspecific introgressions in the genome affects the recombination landscape. This research was intended to inform about genetic phenomena influencing the evolution of introgressed regions, although it should be noted that the research itself is based on examining only one generation, which limits the possibility of drawing far-reaching evolutionary conclusions. In this work, yeast hybrids with large (from several to several dozen percent of the chromosome length) introgressions from another yeast species were crossed. Then, the products of meiosis were isolated and sequenced, and on this basis, the genome-wide distribution of both crossovers (COs) and noncrossovers (NCOs) was examined. Carrying out the analysis at different levels of resolution, it was found that in the regions of introduction, there is a very significant reduction in the frequency of COs and a simultaneous increase in the frequency of NCOs. Moreover, it was confirmed that introgressions significantly limit the local shuffling of genetic information, and NCOs are only able to slightly contribute to the shuffling, thus they do not compensate for the loss of CO recombination.

Strengths:

- Previously, experiments examining the impact of SNP polymorphism on meiotic recombination were conducted either on the scale of single hotspots or the entire hybrid genome, but the impact of large introgressed regions from another species was not examined. Therefore, the strength of this work is its interesting research setup, which allows for providing data from a different perspective.

- Good quality genome-wide data on the distribution of CO and NCO were obtained, which could be related to local changes in the level of polymorphism.

Weaknesses:

(1)  The research is based on examining only one generation, which limits the possibility of drawing far-reaching evolutionary conclusions. Moreover, meiosis is stimulated in hybrids in which introgressions occur in a heterozygous state, which is a very unlikely situation in nature. Therefore, I see the main value of the work in providing information on the CO/NCO decision in regions with high sequence diversification, but not in the context of evolution.

While we are indeed only examining recombination in a single generation, we respectfully disagree that our results aren't relevant to evolutionary processes. The broad goals of our study are to compare recombination landscapes between closely related strains, and we highlight dramatic differences between recombination landscapes. These results add to a body of literature that seeks to understand the existence of variation in traits like recombination rate, and how recombination rate can evolve between populations and species. We show here that the presence of introgression can contribute to changes in recombination rate measured in different individuals or populations, which has not been previously appreciated. We furthermore show that introgression can reduce shuffling between alleles on a chromosome, which is recognized as one of the most important determinants for the existence and persistence of sexual reproduction across all organisms. As we describe in our introduction and conclusion, we see our experimental exploration of the impacts of introgression on the recombination landscape as complementary to studies inferring recombination and introgression from population sequencing data and simulations. There are benefits and challenges to each approach, but both can help us better understand these processes. In regards to the utility of exploring heterozygous introgression, we point out that introgression is often found in a heterozygous state (including in modern humans with Neanderthal and/or Denisovan ancestry). Introgression will always be heterozygous immediately after hybridization, and depending on the frequency of gene flow into the population, the level of inbreeding, selection against introgression, etc., introgression will typically be found as heterozygous.

- The work requires greater care in preparing informative figures and, more importantly, re-analysis of some of the data (see comments below).

More specific comments:

(1) The authors themselves admit that the detection of NCO, due to the short size of conversion tracts, depends on the density of SNPs in a given region. Consequently, more NCOs will be detected in introgressed regions with a high density of polymorphisms compared to the rest of the genome. To investigate what impact this has on the analysis, the authors should demonstrate that the efficiency of detecting NCOs in introgressed regions is not significantly higher than the efficiency of detecting NCOs in the rest of the genome. If it turns out that this impact is significant, analyses should be presented proving that it does not entirely explain the increase in the frequency of NCOs in introgressed regions.

We conducted a deeper exploration of the effect of marker resolution on NCO detection by randomly removing different proportions of markers from introgressed regions of the fermentation cross in order to simulate different marker resolutions from non-introgressed regions. We chose proportions of markers that would simulate different quantiles of the resolution of non-introgressed regions and repeated our standard pipeline in order to compare our NCO detection at the chosen marker densities. More details of this analysis have been added to the manuscript (lines 188-199, 525-538). We confirmed the effect of marker resolution on NCO detection (as reported in the updated manuscript and new supplementary figures S2-S10, new Table S10) and decided to repeat our analyses on the original data with a more stringent correction. For this we chose our observed average tract size for NCOs in introgressed regions (550bp), which leads to a far more conservative estimate of NCO counts (As seen in the updated Figure 2 and Table 2). This better accounts for the increased resolution in introgressed regions, and while it's possible to be more stringent with our corrections, we believe that further stringency would be unreasonable. We also see promising signs that the correction is sufficient when counting our CO and NCO events in both crosses, as described in our response to comment 39 (response to reviewer #3).

(2) CO and NCO analyses performed separately for individual regions rarely show statistical significance (Figures 3 and 4). I think that the authors, after dividing the introgressed regions into non-overlapping windows of 100 bp (I suggest also trying 200 bp, 500 bp, and 1kb windows), should combine the data for all regions and perform correlations to SNP density in each window for the whole set of data. Such an analysis has a greater chance of demonstrating statistically significant relationships. This could replace the analysis presented in Figure 3 (which can be moved to Supplement). Moreover, the analysis should also take into account indels.

We're uncertain of what is being requested here. If the comment refers to the effect of marker density on NCO detection, we hope the response to comment 2 will help resolve this comment as well. Otherwise, we ask for some clarification so that we may correct or revise as appropriate.

(3) In Arabidopsis, it has been shown that crossover is stimulated in heterozygous regions that are adjacent to homozygous regions on the same chromosome (http://dx.doi.org/10.7554/eLife.03708.001, https://doi.org/10.1038/s41467-022-35722-3).

This effect applies only to class I crossovers, and is reversed for class II crossovers (https://doi.org/10.15252/embj.2020104858, https://doi.org/10.1038/s41467-023-42511-z). This research system is very similar to the system used by the authors, although it likely differs in the level of DNA sequence divergence. The authors could discuss their work in this context.

We thank the reviewer for sharing these references. We have added a discussion of our work in the context of these findings in the Discussion, lines 367-376.

Reviewer #2 (Public Review):

Summary:

Schwartzkopf et al characterized the meiotic recombination impact of highly heterozygous introgressed regions within the budding yeast Saccharomyces uvarum, a close relative of the canonical model Saccharomyces cerevisiae. To do so, they took advantage of the naturally occurring Saccharomyces bayanus introgressions specifically within fermentation isolates of S. uvarum and compared their behavior to the syntenic regions of a cross between natural isolates that do not contain such introgressions. Analysis of crossover (CO) and noncrossover (NCO) recombination events shows both a depletion in CO frequency within highly heterozygous introgressed regions and an increase in NCO frequency. These results strongly support the hypothesis that DNA sequence polymorphism inhibits CO formation, and has no or much weaker effects on NCO formation. Eventually, the authors show that the presence of introgressions negatively impacts "r", the parameter that reflects the probability that a randomly chosen pair of loci shuffles their alleles in a gamete.

The authors chose a sound experimental setup that allowed them to directly compare recombination properties of orthologous syntenic regions in an otherwise intra-specific genetic background. The way the analyses have been performed looks right, although this reviewer is unable to judge the relevance of the statistical tests used. Eventually, most of their results which are elegant and of interest to the community are present in Figure 2.

Strengths:

Analysis of crossover (CO) and noncrossover (NCO) recombination events is compelling in showing both a depletion in CO frequency within highly heterozygous introgressed regions and an increase in NCO frequency.

Weaknesses:

The main weaknesses refer to a few text issues and a lack of discussion about the mechanistic implications of the present findings.

- Introduction

(1) The introduction is rather long. | I suggest specifically referring to "meiotic" recombination (line 71) and to "meiotic" DSBs (line 73) since recombination can occur outside of meiosis (ie somatic cells).

We agree and have condensed the introduction to be more focused. We also made the suggested edits to include “meiotic” when referring to recombination and DSBs.

(2) From lines 79 to 87: the description of recombination is unnecessarily complex and confusing. I suggest the authors simply remind that DSB repair through homologous recombination is inherently associated with a gene conversion tract (primarily as a result of the repair of heteroduplex DNA by the mismatch repair (MMR) machinery) that can be associated or not to a crossover. The former recombination product is a crossover (CO), the latter product is a noncrossover (NCO) or gene conversion. Limited markers may prevent the detection of gene conversions, which erase NCO but do not affect CO detection.

We changed the language in this section to reflect the reviewer’s suggestions.

(3) In addition, "resolution" in the recombination field refers to the processing of a double Holliday junction containing intermediates by structure-specific nucleases. To avoid any confusion, I suggest avoiding using "resolution" and simply sticking with "DSB repair" all along the text.

We made the suggested correction throughout the paper.

(4) Note that there are several studies about S. cerevisiae meiotic recombination landscapes using different hybrids that show different CO counts. In the introduction, the authors refer to Mancera et al 2008, a reference paper in the field. In this paper, the hybrid used showed ca. 90 CO per meiosis, while their reference to Liu et al 2018 in Figure 2 shows less than 80 COs per meiosis for S. cerevisiae. This shows that it is not easy to come up with a definitive CO count per meiosis in a given species. This needs to be taken into account for the result section line 315-321.

This is an excellent point. We added this context in the results (lines 180-187).

(5) In line 104, the authors refer to S. paradoxus and mention that its recombination rate is significantly different from that of S. cerevisiae. This is inaccurate since this paper claims that the CO landscape is even more conserved than the DSB landscape between these two species, and they even identify a strong role played by the subtelomeric regions. So, the discussion about this paper cannot stand as it is.

We agree with the reviewer's point. We also found that the entire paragraph was unnecessary, so it and the sentence in question have been removed.

(6) Line 150, when the authors refer to the anti-recombinogenic activity of the MMR, I suggest referring to the published work from Martini et al 2011 rather than the not-yet-published work from Copper et al 2021, or both, if needed.

Added the suggested citation.

Results

(7) The clear depletion in CO and the concomitant increase in NCO within the introgressed regions strongly suggest that DNA sequence polymorphism triggers CO inhibition but does not affect NCO or to a much lower extent. Because most CO likely arises from the ZMM pathway (CO interference pathway mainly relying on Zip1, 2, 3, 4, Spo16, Msh4, 5, and Mer3) in S. uvarum as in S. cerevisiae, and because the effect of sequence polymorphism is likely mediated by the MMR machinery, this would imply that MMR specifically inhibits the ZMM pathway at some point in S. uvarum. The weak effect or potential absence of the effect of sequence polymorphism on NCO formation suggests that heteroduplex DNA tracts, at least the way they form during NCO formation, escape the anti-recombinogenic effect of MMR in S. uvarum. A few comments about this could be added.

We have added discussion and citations regarding the biased repair of DSB to NCO in introgression, lines 380-386.

(8) The same applies to the fact that the CO number is lower in the natural cross compared to the fermentation cross, while the NCO number is the same. This suggests that under similar initiating Spo11-DSB numbers in both crosses, the decrease in CO is likely compensated by a similar increase in inter-sister recombination.

Thank you to the reviewer for this observation. We agree that this could explain some differences between the crosses.

(9) Introgressions represent only 10% of the genome, while the decrease in CO is at least 20%. This is a bit surprising especially in light of CO regulation mechanisms such as CO homeostasis that tends to keep CO constant. Could the authors comment on that?

We interpret these results to reflect two underlying mechanisms. First, the presence of heterozygous introgression does reduce the number of COs. Second, we believe the difference in COs reflects variation in recombination rate between strains. We note that CO homeostasis need not apply across different genetic backgrounds. Indeed, recombination rate is appreciated to significantly differ between strains of S. cerevisiae (Raffoux et al. 2018), and recombination rate variation has been observed between strains/lines/populations in many different species including Drosophila, mice, humans, Arabidopsis, maize, etc. We reference S. cerevisiae strain variability in the Introduction lines 128-130, and have added context in the Results lines 180-187, and Discussion lines 343-350.

(10) Finally, the frequency of NCOs in introgressed regions is about twice the frequency of CO in non-introgressed regions. Both CO and NCO result from Spo11-initiating DSBs.

This suggests that more Spo11-DSBs are formed within introgressed regions and that such DSBs specifically give rise to NCO. Could this be related to the lack of homolog engagement which in turn shuts down Spo11-DSB formation as observed in ZMM mutants by the Keeney lab? Could this simply result from better detection of NCO in introgressed regions related to the increased marker density, although the authors claim that NCO counts are corrected for marker resolution?

The effect noted by the reviewer remains despite the more conservative correction for marker density applied to NCO counts (as described in the response to Reviewer 1, comment #2). Given that CO+NCO counts in introgressed regions are not statistically different between crosses, it is likely that these regions are simply predisposed to a higher rate of DSBs than the rest of the genome. This is an interesting observation, however, and one that we would like to further explore in future work.

(11) What could be the explanation for chromosome 12 to have more shuffling in the natural cross compared to the fermentation cross which is deprived of the introgressed region?

We added this text to the Results, lines 323-327, "While it is unclear what potential mechanism is mediating the difference in shuffling on chromosome 12, we note that the rDNA locus on chromosome 12 is known to differ dramatically in repeat content across strains of S. cerevisiae (22–227 copies) (Sharma et a. 2022), and we speculate that differences in rDNA copy number between strains in our crosses could impact shuffling."

Technical points:

(12) In line 248, the authors removed NCO with fewer than three associated markers.

What is the rationale for this? Is the genotyping strategy not reliable enough to consider events with only one or two markers? NCO events can be rather small and even escape detection due to low local marker density.

We trust the genotyping strategy we used, but chose to be conservative in our detection of NCOs to account for potential sequencing biases.

(13) Line 270: The way homology is calculated looks odd to this reviewer, especially the meaning of 0.5 homology. A site is either identical (1 homology) or not (0 homology).

We've changed the language to better reflect what we are calculating (diploid sequence similarity; see comment #28). Essentially, the metric is a probability that two randomly selected chromatids--one from each parent--will share the same nucleotide at a given locus (akin to calculating the probability of homozygous offspring at a single locus). We average it along a segment of the genome to establish an expected sequence similarity if/when recombination occurs in that segment.

(14) Line 365: beware that the estimates are for mitotic mismatch repair (MMR). Meiotic MMR may work differently.

We removed the citation that refers exclusively to mitotic recombination. The statement regarding meiotic recombination is otherwise still reflective of results from Chen & Jinks-Robertson

(15) Figure 1: there is no mention of potential 4:0 segregations. Did the authors find no such pattern? If not, how did they consider them?

The program we used to call COs and NCOs (ReCombine's CrossOver program) can detect such patterns, but none were detected in our data.

Reviewer #3 (Public Review):

When members of two related but diverged species mate, the resulting hybrids can produce offspring where parts of one species' genome replace those of the other. These "introgressions" often create regions with a much greater density of sequence differences than are normally found between members of the same species. Previous studies have shown that increased sequence differences, when heterozygous, can reduce recombination during meiosis specifically in the region of increased difference. However, most of these studies have focused on crossover recombination, and have not measured noncrossovers. The current study uses a pair of Saccharomyces uvarum crosses: one between two natural isolates that, while exhibiting some divergence, do not contain introgressions; the other is between two fermentation strains that, when combined, are heterozygous for 9 large regions of introgression that have much greater divergence than the rest of the genome. The authors wished to determine if introgressions differently affected crossovers and noncrossovers, and, if so, what impact that would have on the gene shuffling that occurs during meiosis.

(1) While both crossovers and noncrossovers were measured, assessing the true impact of increased heterology (inherent in heterozygous introgressions) is complicated by the fact that the increased marker density in heterozygous introgressions also increases the ability to detect noncrossovers. The authors used a relatively simple correction aimed at compensating for this difference, and based on that correction, conclude that, while as expected crossovers are decreased by increased sequence heterology, counter to expectations noncrossovers are substantially increased. They then show that, despite this, genetic shuffling overall is substantially reduced in regions of heterozygous introgression. However, it is likely that the correction used to compensate for the effect of increased sequence density is defective, and has not fully compensated for the ascertainment bias due to greater marker density. The simplest indication of this potential artifact is that, when crossover frequencies and "corrected" noncrossover frequencies are taken together, regions of introgression often appear to have greater levels of total recombination than flanking regions with much lower levels of heterology. This concern seriously undercuts virtually all of the novel conclusions of the study. Until this methodological concern is addressed, the work will not be a useful contribution to the field.

We appreciate this concern. Please see response to comments #2 and #38. We further note that our results depicted in Figure 3 and 4 are not reliant on any correction or comparison with non-introgressed regions, and thus our results regarding sequence similarity and its effect on the repair of DSBs and the amount of genetic shuffling with/without introgression to be novel and important observations for the field.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

(1) Line 149 - this sentence refers to a mixture of papers reporting somatic or meiotic recombination and as these processes are based on different crossover pathways, this should not be mixed. For example, it is known that in Arabidopsis MSH2 has a pro-crossover function during meiotic recombination.

Corrected

(2) What is unclear to me is how the crosses are planned. Line 308 shows that there were only two crosses (one "natural" and one "fermentation"), but I understand that this is a shorthand and in fact several (four?) different strains were used for the "fermentation cross". At least that's what I concluded from Fig. 1B and its figure caption. This needs to be further explained. Were different strains used for each fermentation cross, or was one strain repeated in several crosses? In Figure 1, it would be worth showing, next to the panel showing "fermentation cross", a diagram of how "natural cross" was performed, because as I understand it, panel A illustrates the procedure common to both types of crosses, and not for "natural cross".

We thank the reviewer for drawing our attention to confusion about how our crosses were created. We performed two crosses, as depicted in Figure 1A. The fermentation cross is a single cross from two strains isolated from fermentation environments. The natural cross is a single cross from two strains isolated from a tree and insect. Table S1 and the methods section "Strain and library construction" describe the strains used in more detail. We modified Figure 1 and the figure legend to help clarify this. See also response to comment #37.

(3) The authors should provide a more detailed characterization of the genetic differences between chromosomes in their hybrids. What is the level of polymorphism along the S. uvarum chromosomes used in the experiments? Is this polymorphism evenly distributed? What are the differences in the level of polymorphism for individual introgressions? Theoretically, this data should be visible in Figure 2D, but this figure is practically illegible in the present form (see next comment).

As suggested, we remade Figure 2D to only include chromosomes with an introgression present, and moved the remaining chromosomes to the supplements (Figure S11). The patterns of markers (which are fixed differences between the strains in the focal cross) should be more clear now. As we detail in the Methods line 507-508, we utilized a total of 24,574 markers for the natural cross and 74,619 markers for the fermentation cross (the higher number in the fermentation cross being due to more fixed differences in regions of introgression).

(4) Figure 2D should be prepared more clearly, I would suggest stretching the chromosomes, otherwise, it is difficult to see what is happening in the introgression regions for CO and NCO (data for SNPs are more readable). Maybe leave only the chromosomes with introgressions and transfer the rest to the supplement?

See previous comment.

(5) How are the Y scales defined for Figure 2D?

Figure 2D now includes units for the y-axis.

(6) Are increases in CO levels in fermentation cross-observed at the border with introgressions? This would indicate local compensation for recombination loss in the introgressed regions, similar to that often observed for chromosomal inversions.

We see no evidence of an increase in CO levels at the borders of introgressions, neither through visual inspection or by comparing the average CO rate in all fermentation windows to that of windows at the edges of introgressions. This is included in the Discussion lines 360-366, "While we are limited in our interpretations by only comparing two crosses (one cross with heterozygous introgression and one without introgression), these results are in line with findings in inversions, where heterozygotes show sharp decreases in COs, but the presence of NCOs in the inverted region (Crown et al., 2018; Korunes & Noor, 2019). However, unlike heterozygous inversions where an increase in COs is observed on freely recombining chromosomes (the inter-chromosomal effect), we do not see an increase in COs on the borders flanking introgression or on chromosomes without introgression."

(7) Line 336 - "We find positive correlations between CO counts..." - you should indicate here that between fermentation and natural crosses, it was quite hard for me to understand what you calculated.

We corrected the language as suggested.

(8) The term "homology" usually means "having a common evolutionary origin" and does not specify the level of similarity between sequences, thus it cannot be measured. It is used incorrectly throughout the manuscript (also in the intro). I would use the term "similarity" to indicate the degree of similarity between two sequences.

We corrected the language as suggested throughout the document.

(9) Paragraph 360 and Figure 3 - was the "sliding window" overlapping or non-overlapping?

We added clarifying language to the text in both places. We use a 101bp sliding window with 50bp overlaps.

(10) Line 369 - what is "...the proportion of bases that are expected to match between the two parent strains..."?

We clarified the language in this location, and hopefully changes associated with the comment about sequence similarity will make the comment even clearer in context.

(11) Line 378 - should it refer to Figure S1 and not Figure 4?

Corrected.

(12) Line 399 - should refer to Figure 4, not Figure 5.

Corrected

(13) Line 444-449 - the analysis of loss of shuffling in the context of the location of introgression on the chromosome should be presented in the result section.

We shifted the core of the analysis to the results, while leaving a brief summary in the discussion.

(14) The authors should also take into account the presence of indels in their analyses, and they should be marked in the figures, if possible.

We filtered out indels in our variant calling. However, we did analyze our crosses for the presence of large insertions and deletions (Table S2), which can obscure true recombination rates, and found that they were not an issue in our dataset.

Reviewer #2 (Recommendations For The Authors):

This reviewer suggests that the authors address the different points raised in the public review.

(1) This reviewer would like to challenge the relevance of the r-parameter in light of chromosome 12 which has no introgression and still a strong depletion in r in the fermentation cross.

We added this text to the Results, lines 377-381, "While it is unclear what potential mechanism is mediating the difference in shuffling on chromosome 12, we note that the rDNA locus on chromosome 12 is known to differ dramatically in repeat content across strains of S. cerevisiae (22–227 copies) (Sharma et a. 2022), and we speculate that differences in rDNA copy number between strains in our crosses could impact shuffling."

(2) This reviewer insists on making sure that NCO detection is unaffected by the marker density, notably in the highly polymorphic regions, to unambiguously support Figure 1C.

We've changed our correction for resolution to be more aggressive (see response to comment #2), and believe we have now adequately adjusted for marker density (see response to comment #38).

Reviewer #3 (Recommendations For The Authors):

I regret using such harsh language in the public review, but in my opinion, there has been a serious error in how marker densities are corrected for, and, since the manuscript is now public, it seems important to make it clear in public that I think that the conclusions of the paper are likely to be incorrect. I regret the distress that the public airing of this may cause. Below are my major concerns:

(1) The paper is written in a way that makes it difficult to figure out just what the sequence differences are within the crosses. Part of this is, to be frank, the unusual way that the crosses were done, between more than one segregant each from two diploids in both natural and fermentation cases. I gather, from the homology calculations description, that each of these four diploids, while largely homozygous, contained a substantial number of heterozygosities, so individual diploids had different patterns of heterology. Is this correct? And if so, why was this strategy chosen? Why not start with a single diploid where all of the heterologies are known? Why choose to insert this additional complication into the mix? It seems to me that this strategy might have the perverse effect of having the heterology due to the polymorphisms present in one diploid affect (by correction) the impact of a noncrossover that occurs in a diploid that lacks the additional heterology. If polymorphic markers are a small fraction of total markers, then this isn't such a great concern, but I could not find the information anywhere in the manuscript. As a courtesy to the reader, please consider providing at the beginning some basic details about the starting strains-what is the average level of heterology between natural A and natural B, and what fraction of markers are polymorphic; what is the average level of heterology between fermentation A and fermentation B in non-introgressed regions, in introgressed regions, and what fraction of markers are polymorphic? How do these levels of heterology compare to what has been examined before in whole-genome hybrid strains? It also might be worth looking at some of the old literature describing S. cerevisiae/S. carlsbergensis hybrids.

We thank the reviewer for drawing our attention to confusion about the cross construction. These crosses were conducted as is typical for yeast genetic crosses: we crossed 2 genetically distinct haploid parents to create a heterozygous diploid, then collected the haploid products of meiosis from the same F1 diploid. Because the crosses were made with haploid parents, it is not possible for other genetic differences to be segregating in the crosses. We have revised Figure 1 and its caption to clarify this. Further details regarding the crosses are in the Methods section "Strain and library construction" and in Supplemental Table S1. We only utilized genetic markers that are fixed differences between our parental strains to call CO and NCO. As we detail in the Methods line 507-508, we utilized a total of 24,574 markers for the natural cross and 74,619 markers for the fermentation cross (the higher number in the fermentation cross being due to more fixed differences in regions of introgression). We additionally revised Figure 2D (and Figure S11) to help readers better visualize differences between the crosses.

(2) There are serious concerns about the methods used to identify noncrossovers and to normalize their levels, which are probably resulting in an artifactually high level of calculated crossovers in Figure 2. As a primary indication of this, it appears in Figure 2 that the total frequency of events (crossovers + noncrossovers) in heterozygous introgressed regions are substantially greater than those in the same region in non-introgressed strains, while just shifting of crossovers to noncrossovers would result in no net increase. The simplest explanation for this is that noncrossovers are being undercounted in non-introgressed relative to introgressed heterozygous regions. There are two possible reasons for this: i. The exclusion of all noncrossover events spanning less than three markers means that many more noncrossovers in introgressed heterozygous regions than in non-introgressed. Assuming that average non-homology is 5% in the former and 1% in the latter, the average 3-marker event will be 60 nt in introgressed regions and 300 nt in non-introgressed regions - so many more noncrossovers will be counted in introgressed regions. A way to check on this - look at the number of crossover-associated markers that undergo gene conversion; use the fraction that involves < 3 markers to adjust noncrossover levels (this is the strategy used by Mancera et al.). ii. The distance used for noncrossover level adjustment (2kb) is considerably greater than the measured average noncrossover lengths in other studies. The effect of using a too-long distance is to differentially under-correct for noncrossovers in non-introgressed regions, while virtually all noncrossovers in heterozygous introgressed regions will be detected. This can be illustrated by simulations that reduce the density of scored markers in heterozygous introgressed regions to the density seen in non-introgressed regions. Because these concerns go to the heart of the conclusions of the paper, they must be addressed quantitatively - if not, the main conclusions of the paper are invalid.

We adjusted the correction factor (See also response to comment #2) and compared the average number of CO and NCO events in introgressed and non-introgressed regions between crosses (two comparisons: introgression CO+NCO in natural cross vs introgression CO+NCO in fermentation cross; non-introgression CO+NCO in natural cross vs non-introgression CO+NCO in fermentation cross). We found no significant differences between the crosses in either of the comparisons. This indicates that the distribution of total events is replicated in both crosses once we correct for resolution.

(3) It is important to distinguish the landscape of double-strand breaks from the landscape of recombination frequencies. Double-strand breaks, as measured by uncalibrated levels of Spo11-linked oligos, is a relative number - not an absolute frequency. So it is possible that two species could have a similar break landscape in terms of topography but have absolute levels higher in one species than in the other.

We agree with this statement, however, we have removed the relevant text to streamline our introduction.

(4) Lines 123-125. Just meiosis will produce mosaic genomes in the progeny of the F1; further backcrossing will reduce mosaicism to the level of isolated regions of introgression.

Adjusted the language to be more specific.

(5) Please provide actual units for the Y axes in Figure 2D.

We have corrected the units on the axes.

(6) Tables (general). Are the significance measures corrected for multiple comparisons?

In Table 3, the cutoff was chosen to be more conservative than a Bonferroni corrected alpha=0.01 with 9 comparisons (0.0011). In text, any result referred to as significant has an associated hypothesis test with a p-value less than its corresponding Bonferroni-corrected alpha of 0.05. This has been clarified in the caption for Table 3 and in the text where relevant.

Author response:

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

I have added a paragraph that addresses the issue of how landmarks might be used and why they are not. The suggestions made in the "Weaknesses" paragraph were concise and excellent and have directly incorporated them into my revised manuscript. This text appears on Page 21 and is shown below. I hope that this is what the editors and reviewers were looking.

The requested revision is the second paragraph.

The first paragraph was not written in response to reviews but inspired by a recent paper by Mahdev et al (2024) - https://doi.org/10.1038/s41593-024-01681-9.  I had already requested to add this reference and was encouraged to do so by the Editors. The Mahdev et al paper was very surprising in that it showed that path integration is not constant but that its "gain" can be recalibrated by selfmotion signals. I wondered whether this unexpected capacity extended to path integration also recalibrating the cognitive map and thereby generating the shortcutting behavior we observe. I suggested that, at an abstract level, this would correspond to "coordinate transformation" of the cognitive map. I realize that this is entirely speculative. If the Editors feel that it does not add much to the manuscript and that the speculation goes to far, I will remove the first paragraph and re-submit.

Added text. P21 and just before the heading: " Implications for theories of hippocampal representations of spatial maps" There were no other changes made in the paper.

"Path integration uses self-motion signals to update the animal's estimated location on its internal cognitive map. Path integration gain has been shown to be plastic and regulated by landmarks (52). Remarkably, a recent study has revealed that path integration gain can also be directly recalibrated by self-motion signals alone (53), albeit not as effectively as by landmarks (52, 53). An interesting question for future research is whether self-motion signals can also recalibrate the coordinates of a cognitive map. From this perspective, the Target B to Target A shortcut requires a transformation of the cognitive map coordinates so that the start point is now Target B.

Extensive research has shown that external cues can control hippocampal neuron place fields (11, 12, 54) and the gain of the path integrator (52), making the failure of mice in our study to use such cues puzzling. The failure to use landmarks may be related to our task being low stakes and our pretraining procedure teaching the mouse that such cues are not necessary. Our results may not generalize to more natural conditions where many reliable prominent cues are available, and where there is urgency to find food or water while avoiding predation (55). Under these more naturalistic conditions the use of distal cues to rapidly find a food reward is more likely to be observed."

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

In this work, the authors continue their investigations on the key role of glycosylation to modulate the function of a therapeutic antibody. As a follow-up to their previous demonstration on how ADCC was heavily affected by the glycans at the Fc gamma receptor (FcγR)IIIa, they now dissect the contributions of the different glycans that decorate the diverse glycosylation sites. Using a well-designed mutation strategy, accompanied by exhaustive biophysical measurements, with extensive use of NMR, using both standard and newly developed methodologies, they demonstrate that there is one specific locus, N162, which is heavily involved in the stabilization of (FcγR)IIIa and that the concomitant NK function is regulated by the glycan at this site.

Strengths:

The methodological aspects are carried out at the maximum level.

Weaknesses:

The exact (or the best possible assessment) of the glycan composition at the N162 site is not defined.

We revised the Introduction to include previous findings from our laboratory regarding processing on YTS cells:

“YTS cells, a key cytotoxic human NK cell line used for these studies, express FcγRIIIa with extensive glycan processing, including the N162 site with predominantly hybrid and complex-type glycoforms {Patel 2021}.” 

Reviewer #2 (Public Review):

Summary:

The authors set out to demonstrate a mechanistic link between Fcgamma receptor (IIIA) glycosylation and IgG binding affinity and signaling - resulting in antibody-dependent cellular cytotoxicity - ADCC. The work builds off prior findings from this group about the general impact of glycosylation on FcR (Fc receptor)-IgG binding.

Strengths:

The structural data (NMR) is highly compelling and very significant to the field. A demonstration of how IgG interacts with FcgRIIIA in a manner sensitive to glycosylation of both the IgG and the FcR fills a critical knowledge gap. The approach to demonstrate the selective impact of glycosylation at N162 is also excellent and convincing. The manuscript/study is, overall, very strong.

Weaknesses:

There are a number of minor weaknesses that should be addressed.

(1) Since S164A is the only mutant in Figure 1 that seems to improve affinity, even if minimally, it would be a nice reference to highlight that residue in the structural model in panel B.

We revised Figure 1B to include the S164 site.

(2) It is confusing why some of the mutants in the study are not represented in Figure 1 panel A. Those affinities and mutants should be incorporated into panel A so the reader can easily see where they all fall on the scale.

We thank the reviewer for this comment. We restructured the Results section to highlight that a primary outcome of the experiment referenced was to map the contribution of interface residues to antibody binding affinity. These data were not previously available, highlighting hotspots at the interface. Figure 1A and B report these results.

We then used a subset of mutations from this experiment, as well as a subset of mutations from an additional library containing mutations proximal to the interface, to build a small library for evaluation using ADCC. The complete binding data for all variants, binding to two different IgG1 Fc glycoforms, is presented in Supplemental Table 1. 

T167Y in particular needs to be shown, as it is one of few mutants that fall between what seems to be ADCC+ and ADCC- lines. Also, that mutant seems to have a stronger affinity compared to wt (judged by panel D), yet less ADCC than wt. This would imply that the relationship between affinity and activity is not as clean as stated, though it is clearly important. Comments about this would strengthen the overall manuscript.

We thank the reviewer for this particular insight. We agree that the lack of a clean correlation between ADCC potency and affinity implies additional factors that could have affected these experimental results. We added the following sentence to the discussion. 

“Notably, the ADCC potency for those high-affinity variants does not fall cleanly on a line, indicating that other factors affect our observations, which may include organization at the cell surface, changes to glycan composition, or receptor trafficking.”

(3) This statement feels out of place: "In summary, this result demonstrates that the sensitivity to antibody fucosylation may be eliminated through FcγRIIIa engineering while preserving antibody-binding affinity." In Figure 2, the authors do indeed show that mutations in FcgRIIIa can alter the impact of IgG core fucosylation, but implying that receptor engineering is somehow translatable or as impactful therapeutically as engineering the antibody itself deflates the real basic science/biochemical impact of understanding these interactions in molecular detail. Not everything has to be immediately translatable to be important. 

We agree and removed the highlighted sentence.   

(4) The findings reported in Figure 2, panel C are exciting. Controls for the quality of digestion at each step should be shown (perhaps in supplementary data).

We agree. We added an example of the digestions as Figure S2.  

(5) Figure 3 is confusing (mislabeled?) and does not show what is described in the Results. First, there is a F158V variant in the graph but a V158F variant in the text.

Please correct this. 

Thank you for identifying this typo. We corrected Figure 3.

Second, this variant (V158F/F158V) does not show the 2-fold increase in ADCC with kifunesine as stated. 

Thank you for drawing our attention to this rounding error. We revised the text to report a statistically significant 1.4-fold increase.

Finally, there are no statistical evaluations between the groups (+/- kif; +/- fucose). 

We provide the p values for +/-fuc and +/- Kifunensine for each YTS cell line in the figure. We did not provide a global comparison of p values that included all cell lines due to some cell lines experiencing a significant change and others not. However, we added the raw data as Supplemental Table 2 should readers wish to perform these analyses.

The differences stated are not clearly statistically significant given the wide spread of the data. This is true even for the wt variant.

We agree that there are points that overlap in this figure between the different treatments. However, our use of the students T-test (two tailed) using three experiments collected on three different days (each with three technical replicates) provides enough resolution to determine the significance of difference of the means for the different treatments. This is, by our estimation, a highly rigorous manner to collect and analyze the data.  

(6) The kifunensine impact is somewhat confusing. They report a major change in ADCC, yet similar large changes with trimming only occur once most of the glycan is nearly gone (Figure 2). Kifunensine will tend to generate high mannose and possibly a few hybrid glycans. It is difficult to understand what glycoforms are truly important outside of stating that multi-branched complex-type N-glycans decrease affinity.

Note that Figure 2 does not evaluate the kifunensine-treated glycan, which is mostly Man8 and Man9 structures. In our previous work, these structures likewise provide increased binding affinity (see pubmed ID 30016589). We believe the most important message is that composition of the N162 glycan (removed with the S164A mutation) regulates NK cell ADCC. On cells, we are not able to modulate N162 glycan composition without affecting potentially every other N-glycan on the surface, so we do not have an ADCC experiments that is directly comparable to Figure 2. Thus, this increased ADCC resulting from kifunensine treatment is consistent with previously observed increases in binding affinity measurement.  

(7) This is outside of the immediate scope, but I feel that the impact would be increased if differences in NK cell (and thus FcgRIIIA) glycosylation are known to occur during disease, inflammation, age, or some other factor - and then to demonstrate those specific changes impact ADCC activity via this mechanism.

We agree completely. As mentioned in the Introduction, we know that N162 glycan composition varies substantially from donor to donor based on previous work from our

lab. Curiously, little variability appeared between donors at the other four Nglycosylation sites. Thus, there is the potential that different NK cell N162 glycan compositions are coincident with different indications. This is an area we are quite interested in pursuing.

Author response:

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

Many thanks to the editors for the reviewing of the revised manuscript.

We are very grateful to the Reviewers for their time and for the appreciation of the revision.

We thank the Reviewer 3 for acknowledging the use of sulforhodamine B (SRB) fluorescence as a real-time readout of astrocyte volume dynamics. Experimental data in brain slices were provided to validate this approach.<br /> The incomplete matching of our observation with early reported data in cultured astrocytes (e.g., Solenov et al., AJP-Cell, 2004), might reflect certain of their properties differing from the slice/in vivo counterparts as discussed in the manuscript.<br /> The study (T.R. Murphy et al., Front Cell Neurosci., 2017) showed that AQP4 knockout increased astrocyte swelling extent in response to hypoosmotic solution in brain slices (Fig 9), and discussed '... AQP4 can provide an efficient efflux pathway for water to leave astrocytes.’ Correspondingly, our data suggest that AQP4 mediate astrocyte water efflux in basal conditions.<br /> We have discussed the study (Igarashi et al., NeuroReport 2013); our current data would help to understand the cellular mechanisms underlying the finding of Igarashi et al.

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

Pham and colleagues provide an illuminating investigation of aquaporin-4 water flux in the brain utilizing ex vivo and in vivo techniques. The authors first show in acute brain slices, and in vivo with fiber photometry, SRB-loaded astrocytes swell after inhibition of AQP4 with TGN-020, indicative of tonic water efflux from astrocytes in physiological conditions. Excitingly, they find that TGN-020 increases the ADC in DW-MRI in a region-specific manner, potentially due to AQP4 density. The resolution of the DW-MRI cannot distinguish between intracellular or extracellular compartments, but the data point to an overall accumulation of water in the brain with AQP4 inhibition. These results provide further clarity on water movement through AQP4 in health and disease.

Overall, the data support the main conclusions of the article, with some room for more detailed treatment of the data to extend the findings.

Strengths:

The authors have a thorough investigation of AQP4 inhibition in acute brain slices. The demonstration of tonic water efflux through AQP4 at baseline is novel and important in and of itself. Their further testing of TGN-020 in hyper- and hypo-osmotic solutions shows the expected reduction of swelling/shrinking with AQP4 blockade.

Their experiment with cortical spreading depression further highlights the importance of water efflux from astrocytes via AQP4 and transient water fluxes as a result of osmotic gradients. Inhibition of AQP4 increases the speed of tissue swelling, pointing to a role in the efflux of water from the brain.

The use of DW-MRI provides a non-invasive measure of water flux after TGN-020 treatment.

We thank the reviewer for the insightful comments.

Weaknesses:

The authors specifically use GCaMP6 and light sheet microscopy to image their brain sections in order to identify astrocytic microdomains. However, their presentation of the data neglects a more detailed treatment of the calcium signaling. It would be quite interesting to see whether these calcium events are differentially affected by AQP4 inhibition based on their cellular localization (ie. processes vs. soma vs. vascular end feet which all have different AQP4 expressions).

Following the suggestion, we provide new data on the effect of AQP4 inhibition on spontaneous calcium signals in perivascular astrocyte end-feet. As shown now in Fig.S2, acute application of TGN020 induced Ca2+ oscillations in astrocyte end-feet regions where the GCaMP6 labeling lines the profile of the blood vessel. It is noted that on average, the strength of basal Ca2+ signals in the end-feet is higher than that observed across global astrocyte territories (4.65 ± 0.55 vs. 1.45 ± 0.79, p < 0.01), as does the effect of TGN (8.4 ± 0.62 vs. 6.35 ± 0.97, p < 0.05; Fig S2 vs. Fig 2B). This likely reflects the enrichment of AQP4 in astrocyte end-feet. We describe the data in Fig.S2, and on page 8, line 20 – 23.  

We now use the transgenic line GLAST-GCaMP6 for cytosolic GCaMP6 expression in astrocytes. Spontaneous calcium signals, reflected by transient fluorescence rises, occur in discrete micro-domains whereas the basal GCaMP6 fluorescence in the soma is weak. In the present condition, it is difficult to unambiguously discriminate astrocyte soma from the highly intermingled processes. 

The authors show the inhibition of AQP4 with TGN-020 shortens the onset time of the swelling associated with cortical spreading depression in brain slices. However, they do not show quantification for many of the other features of CSD swelling, (ie. the duration of swelling, speed of swelling, recovery from swelling).

Regarding the features of the CSD swelling, we have performed new analysis to quantify the duration of swelling, speed of swelling and the recovery time from swelling in control condition and in the presence of TGN-020. The new analysis is now summarized in Fig. S5. Blocking AQP4 with TGN-020 increases the swelling speed, prolongs the duration of swelling and slows down the recovery from swelling, confirming our observation that acute inhibition of AQP4 water efflux facilitates astrocyte swelling while restrains shrinking. We describe the result on page 11, line 19-21. 

Significance:

AQP4 is a bidirectional water channel that is constitutively open, thus water flux through it is always regulated by local osmotic gradients. Still, characterizing this water flux has been challenging, as the AQP4 channel is incredibly water-selective. The authors here present important data showing that the application of TGN-020 alone causes astrocytic swelling, indicating that there is constant efflux of water from astrocytes via AQP4 in basal conditions. This has been suggested before, as the authors rightfully highlight in their discussion, but the evidence had previously come from electron microscopy data from genetic knockout mice.

AQP4 expression has been linked with the glymphatic circulation of cerebrospinal fluid through perivascular spaces since its rediscovery in 2012 [1]. Further studies of aging[2], genetic models[3], and physiological circadian variation[4] have revealed it is not simply AQP4 expression but AQP4 polarization to astrocytic vascular endfeet that is imperative for facilitating glymphatic flow. Still, a lingering question in the field is how AQP4 facilitates fluid circulation. This study represents an important step in our understanding of AQP4's function, as the basal efflux of water via AQP4 might promote clearance of interstitial fluid to allow an influx of cerebrospinal fluid into the brain. Beyond glymphatic fluid circulation, clearly, AQP4-dependent volume changes will differentially alter astrocytic calcium signaling and, in turn, neuronal activity.

(1) Iliff, J.J., et al., A Paravascular Pathway Facilitates CSF Flow Through the Brain Parenchyma and the Clearance of Interstitial Solutes, Including Amyloid β. Sci Transl Med, 2012. 4(147): p. 147ra111.

(2) Kress, B.T., et al., Impairment of paravascular clearance pathways in the aging brain. Ann Neurol, 2014. 76(6): p. 845-61.

(3) Mestre, H., et al., Aquaporin-4-dependent Glymphatic Solute Transport in the Rodent Brain. eLife, 2018. 7.

(4) Hablitz, L., et al., Circadian control of brain glymphatic and lymphatic fluid flow. Nature Communications, 2020. 11(1).

We thank the reviewer in acknowledging the significance of our study and the functional implication in brain glymphatic system. We have now highlighted the mentioned studies as well as the potential implication glymphatic fluid circulation (page 4, line 9-10; page 5, line 1-3; and page 19, line 3-10). 

Reviewer #2 (Public Review):

Summary:

The paper investigates the role of astrocyte-specific aquaporin-4 (AQP4) water channel in mediating water transport within the mouse brain and the impact of the channel on astrocyte and neuron signaling. Throughout various experiments including epifluorescence and light sheet microscopy in mouse brain slices, and fiber photometry or diffusion-weighted MRI in vivo, the researchers observe that acute inhibition of AQP4 leads to intracellular water accumulation and swelling in astrocytes. This swelling alters astrocyte calcium signaling and affects neighboring neuron populations. Furthermore, the study demonstrates that AQP4 regulates astrocyte volume, influencing mainly the dynamics of water efflux in response to osmotic challenges or associated with cortical spreading depolarization. The findings suggest that AQP4-mediated water efflux plays a crucial role in maintaining brain homeostasis, and indicates the main role of AQP4 in this mechanism. However authors highlight that the report sheds light on the mechanisms by which astrocyte aquaporin contributes to the water environment in the brain parenchyma, the mechanism underlying these effects remains unclear and not investigated. The manuscript requires revision.

Strengths:

The paper elucidates the role of the astrocytic aquaporin-4 (AQP4) channel in brain water transport, its impact on water homeostasis, and signaling in the brain parenchyma. In its idea, the paper follows a set of complimentary experiments combining various ex vivo and in vivo techniques from microscopy to magnetic resonance imaging. The research is valuable, confirms previous findings, and provides novel insights into the effect of acute blockage of the AQP4 channel using TGN-020.

We thank the reviewer for the constructive comments.

Weaknesses:

Despite the employed interdisciplinary approach, the quality of the manuscript provides doubts regarding the significance of the findings and hinders the novelty claimed by the authors. The paper lacks a comprehensive exploration or mention of the underlying molecular mechanisms driving the observed effects of astrocytic aquaporin-4 (AQP4) channel inhibition on brain water transport and brain signaling dynamics. The scientific background is not very well prepared in the introduction and discussion sections. The important or latest reports from the field are missing or incompletely cited and missconcluded. There are several citations to original works missing, which would clarify certain conclusions. This especially refers to the basis of the glymphatic system concept and recently published reports of similar content. The usage of TGN-020, instead of i.e. available AER-270(271) AQP4 blocker, is not explained. While employing various experimental techniques adds depth to the findings, some reasoning behind the employed techniques - especially regarding MRI - is not clear or seemingly inaccurate. Most of the time the number of subjects examined is lacking or mentioned only roughly within the figure captions, and there are lacking or wrongly applied statistical tests, that limit assessment and reproducibility of the results. In some cases, it seems that two different statistical tests were used for the same or linked type of data, so the results are contradictory even though appear as not likely - based on the figures. Addressing these limitations could strengthen the paper's impact and utility within the field of neuroscience, however, it also seems that supplementary experiments are required to improve the report.

The current data hint at a tonic water efflux from astrocyte AQP4 in physiological condition, which helps to understand brain water homeostasis and the functional implication for the glymphatic system. The underlying molecular and cellular mechanisms appear multifaceted and functionally interconnected, as discussed (page 14 line 8 –page 15, line 3). We agree that a comprehensive exploration will further advance our understanding.

The introduction and discussion are now strengthened by incorporating the important advances in glymphatic system while highlighting the relevant studies. 

The use of TGN-020 was based on its validation by wide range of ex vivo and in vivo studies including the use of heterologous expression system and the AQP4 KO mice. The validation of AER-270(271, the water soluble prodrug) using AQP4 KO mice is reported recently (Giannetto et al., 2024). AER-271 was noted to impact brain water ADC (apparent diffusion coefficient evaluated by diffusion-weighted MRI) in AQP4 KO mice ~75 min after the drug application (Giannetto et al., 2024). This likely reflects that AER270(271) is also an inhibitor for κΒ nuclear factor (NF-κΒ) whose inhibition could reduce CNS water content independent of AQP4 targeting (Salman et al., 2022). In addition, the inhibition efficiency of AER-270(271) seems lower than TGN-020 (Farr et al., 2019; Giannetto et al., 2024; Huber et al., 2009; Salman et al., 2022). We have now supplemented this information in the manuscript (page 7, line 1-6 and page15, line 7-17).

The description on the DW-MRI is now updated (page 4, line 10-14). 

We also performed new experiments and data analysis as described in a point-to-point manner below in the section ‘Recommendations For The Authors’.

Reviewer #3 (Public Review):

Summary:

In this manuscript, the authors propose that astrocytic water channel AQP4 represents the dominant pathway for tonic water efflux without which astrocytes undergo cell swelling. The authors measure changes in astrocytic sulforhodamine fluorescence as the proxy for cell volume dynamics. Using this approach, they perform a technically elegant series of ex vivo and in vivo experiments exploring changes in astrocytic volume in response to AQP4 inhibitor TGN-020 and/or neuronal stimulation. The key finding is that TGN-020 produces an apparent swelling of astrocytes and modifies astrocytic cell volume regulation after spreading depolarizations. Additionally, systemic application of TGN-020 produced changes in diffusion-weighted MRI signal, which the authors interpret as cellular swelling. This study is perceived as potentially significant. However, several technical caveats should be strongly considered and perhaps addressed through additional experiments.

Strengths:

(1) This is a technically elegant study, in which the authors employed a number of complementary ex vivo and in vivo techniques to explore functional outcomes of aquaporin inhibition. The presented data are potentially highly significant (but see below for caveats and questions related to data interpretation).

(2) The authors go beyond measuring cell volume homeostasis and probe for the functional significance of AQP4 inhibition by monitoring Ca2+ signaling in neurons and astrocytes (GCaMP6 assay).

(3) Spreading depolarizations represent a physiologically relevant model of cellular swelling. The authors use ChR2 optogenetics to trigger spreading depolarizations. This is a highly appropriate and much-appreciated approach.

We thank the reviewer for the effort in evaluating our work.

Weaknesses:

(1) The main weakness of this study is that all major conclusions are based on the use of one pharmacological compound. In the opinion of this reviewer, the effects of TGN-020 are not consistent with the current knowledge on water permeability in astrocytes and the relative contribution of AQP4 to this process.

Specifically: Genetic deletion of AQP4 in astrocytes reduces plasmalemmal water permeability by ~two-three-fold (when measured a 37oC, Solenov et al., AJP-Cell, 2004). This is a significant difference, but it is thought to have limited/no impact on water distribution. Astrocytic volume and the degree of anisosmotic swelling/shrinkage are unchanged because the water permeability of the AQP4null astrocytes remains high. This has been discussed at length in many publications (e.g., MacAulay et al., Neuroscience, 2004; MacAulay, Nat Rev Neurosci, 2021) and is acknowledged by Solenov and Verkman (2004).

Keeping this limitation in mind, it is important to validate astrocytic cell volume changes using an independent method of cell volume reconstruction (diameter of sulforhodamine-labeled cell bodies? 3D reconstruction of EGFP-tagged cells? Else?)

Solenov and coll. used the calcein quenching assay and KO mice demonstrating AQP4 as a functional water channel in cultured astrocytes (Solenov et al., 2004). AQP4 deletion reduced both astrocyte water permeability and the absolute amplitude of swelling over comparable time, and also slowed down cell shrinking, which overall parallels our results from acute AQP4 blocking. Yet in Solenovr’s study, the time to swelling plateau was prolonged in AQP4 KO astrocytes, differing from our data from the pharmacological acute blocking. This discrepancy may be due to compensatory mechanisms in chronic AQP4 KO, or reflect the different volume responses in cultured astrocytes from brain slices or in vivo results as suggested previously (Risher et al., 2009). 

Soma diameter might be an indicator of cell volume change, yet it is challenging with our current fluorescence imaging method that is diffraction-limited and insufficient to clearly resolve the border of the soma in situ. In addition, the lateral diameter of cell bodies may not faithfully reflect the volume changes that can occur in all three dimensions. Rapid 3D imaging of astrocyte volume dynamics with sufficient high Z-axis resolution appears difficult with our present tools. 

We have now accordingly updated the discussion with relevant literatures being cited (page 17 line 14 – page 18, line 3).

(2) TGN-020 produces many effects on the brain, with some but not all of the observed phenomena sensitive to the genetic deletion of AQP4. In the context of this work, it is important to note that TGN020 does not completely inhibit AQP4 (70% maximal inhibition in the original oocyte study by Huber et al., Bioorg Med Chem, 2009). Thus, besides not knowing TGN-020 levels inside the brain, even

"maximal" AQP4 inhibition would not be expected to dramatically affect water permeability in astrocytes.

This caveat may be addressed through experiments using local delivery of structurally unrelated AQP4 blockers, or, preferably, AQP4 KO mice.

It is an important point that TGN-020 partially blocks AQP4, implying the actual functional impact of AQP4 per se might be stronger than what we observed. TGN provides a means to acutely probe AQP4 function in situ, still we agree, its limitation needs be acknowledged. We mention this now on page 15, line 7-9 and 14-17.

We agree that local delivery of an alternative blocker will provide additional information. Meanwhile, local delivery requires the stereotaxic implantation of cannula, which would cause inflammations to surrounding astrocytes (and neurons). The recently introduced AQP4 blocker AER-270(271) has received attention that it influences brain water dynamics (ADC in DW-MRI) in AQP4 KO mice (Giannetto et al., 2024), recalling that AER-270(271) is also an inhibitor for κΒ nuclear factor (NF-κΒ). This pathway can potentially perturb CNS water content and influence brain fluid circulation, in an AQP4independent manner (Salman et al., 2022). The inhibition efficiency on mouse AQP4 of AER-270 (~20%, Farr et al., 2019; Salman et al., 2022) appears lower than TGN-020 (~70%, Huber et al., 2009).

We chose to use the pharmacological compound to achieve acute blocking of AQP4 thereby avoiding the chronic genetics-caused alterations in brain structural, functional and water homeostasis. Multiple lines of evidence including the recent study (Gomolka et al., 2023), have shown that AQP4 KO mice alters brain water content, extracellular space and cellular structures, which raises concerns to use the transgenic mouse to pinpoint the physiological functions of the AQP4 water channel. 

We have now mentioned the concerns on AQP4 pharmacology by supplementing additional literatures in the field (page 15, line 8-18). 

(3) This reviewer thinks that the ADC signal changes in Figure 5 may be unrelated to cellular swelling. Instead, they may be a result of the previously reported TGN-020-induced hyphemia (e.g., H. Igarashi et al., NeuroReport, 2013) and/or changes in water fluxes across pia matter which is highly enriched in AQP4. To amplify this concern, AQP4 KO brains have increased water mobility due to enlarged interstitial spaces, rather than swollen astrocytes (RS Gomolka, eLife, 2023). Overall, the caveats of interpreting DW-MRI signal deserve strong consideration.

The previous observation show that TGN-020 increases regional cerebral blood flow in wild-type mice but not in AQP4 KO mice (Igarashi et al., 2013). Our current data provide a possible mechanism explanation that TGN-020 blocking of astrocyte AQP4 causes calcium rises that may lead to vasodilation as suggested previously (Cauli and Hamel, 2018). We now add updates to the discussion on page 15, line 3-7.

We are in line with the reviewer regarding the structural deviations observed with the AQP4 KO mice

(Gomolka et al., 2023), now mentioned on page 19, line 3-5. Following the Reviewer’s suggestion, we have also updated the interpretation of the DW-MRI signal and point that in addition to being related to the astrocyte swelling, the ADC signal changes may also be caused by indirect mechanisms, such as the transient upregulation of other water-permeable pathways in compensating AQP4 blocking. We now describe this alternative interpretation and the caveats of the DW-MRI signals (page 20, line 1-8). 

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Private recommendations

My more broad experimental suggestions are in the "weaknesses" section. Some minor points that would improve the manuscript are included below:

(1) A more detailed explanation for why SRB fluorescence reflects the astrocyte volume changes, whereas typical intracellular GFP does not.

As an engineered fluorescence protein, the GFP has been used to tag specific type of cells. Meanwhile, as a relatively big protein (MW, 26.9 kDa), the diffusion rate of EGFP is expected to be much less than SRB, a small chemical dye (MW, 558.7 Da). Also, the IP injection of SRB enables geneticsless labeling of brain astrocytes, so to avoid the influence of protein overexpression on astrocyte volume and water transport responses. We have now stated this point in the manuscript (page 13, line 21 – page 14, line 4).

(2) Figure 1 panel B should have clear labels on the figure and a description in the legend to delineate which part of the panel refers to hyper- or hypo-osmotic treatment.

We have now updated the figure and the legend.  

(3) For Figure 2, what is the rationale for analyzing the calcium signaling data between the cell types differently?

We analyzed calcium micro-domains for astrocytes as their spontaneous signals occur mainly in discrete micro-domains (Shigetomi et al., 2013). While for neurons, we performed global analysis by calculating the mean fluorescence of imaging field of view, because calcium signal changes were only observed at global level rather than in micro-domains. This information is now included (page 24, line1820).

(4) For Figure 3, the authors mention that TGN-020 likely caused swelling prior to the hypotonic solution administration. Do they have any measurements from these experiments prior to the TGN-020 application to use as a "true baseline" volume?

The current method detects the relative changes in astrocyte volume (i.e., transmembrane water transport), which nevertheless is blind to the absolute volume value. We have no readout on baseline volumes.  

(5) For Figures 3 and 4, did the authors see any evidence for regulatory volume decrease? And is this impaired by TGN-020? It is a well-characterized phenomenon that astrocytes will open mechanosensitive channels to extrude ions during hypo-osmotic induced swelling. This process is dependent on AQP4 and calcium signaling [5]

Mola and coll. provided important results demonstrating the role of AQP4 in astrocyte volume regulation (Mola et al., 2016). In the present study in acute brain slices, when we applied hypotonic solution to induce astrocyte swelling, our protocol did not reveal rapid regulatory volume decrease (e.g., Fig. 3D). When we followed the volume changes of SRB-labeled astrocytes during optogenetically induced CSD, we observed the phase of volume decrease following the transient swelling (Fig. 4F), where the peak amplitude and the degree of recovery were both reduced by inhibiting AQP4 with TGN020. These data imply that regulatory astrocyte volume decrease may occur in specific conditions, which intriguingly has been suggested to be absent in brain slices and in vivo (e.g., Risher et al., 2009). We have not specifically investigated this phenomenon, and now briefly discuss this point on page18 line 6-14.

(6) Figure 5 box plots do not show all data points, could the authors modify to make these plots show all the animals, or edit the legend to clarify what is plotted?

We have now updated the plot and the legend. This plot is from all animals (n = 7 per condition).

(7) pg. 9 line 6, there is a sentence that seems incomplete or otherwise unfinished. "We first followed the evoked water efflux and shrinking induced by hypertonic solution while."

Fixed (now, page 9 line 17-18). 

(8)  During the discussion on pg 13 line 11, it may be more clear to describe this as the cotransport of water into the cells with ions/metabolites as reviewed by Macaulay 2021 [6].

We agree; the text is modified following this suggestion (now page14, line 12-13).  

(1) Iliff, J.J., et al., A Paravascular Pathway Facilitates CSF Flow Through the Brain Parenchyma and the Clearance of Interstitial Solutes, Including Amyloid β. Sci Transl Med, 2012. 4(147): p. 147ra111.

(2) Kress, B.T., et al., Impairment of paravascular clearance pathways in the aging brain. Ann Neurol, 2014. 76(6): p. 845-61.

(3) Mestre, H., et al., Aquaporin-4-dependent Glymphatic Solute Transport in the Rodent Brain. eLife, 2018. 7.

(4) Hablitz, L., et al., Circadian control of brain glymphatic and lymphatic fluid flow. Nature Communications, 2020. 11(1).

(5) Mola, M., et al., The speed of swelling kinetics modulates cell volume regulation and calcium signaling in astrocytes: A different point of view on the role of aquaporins. Glia, 2016. 64(1).

(6) MacAulay, N., Molecular mechanisms of brain water transport. Nat Rev Neurosci, 2021. 22(6): p. 326-344.

We thank the reviewer. These important literatures are now supplemented to the manuscript together with the corresponding revisions.

Reviewer #2 (Recommendations For The Authors):

In its concept, the paper is interesting and provides additional value - however, it requires revision.

Below, I provide the following remarks for the following sections/ pages/lines:

ABSTRACT/page 2 (remarks here refer to the rest of the manuscript, where these sentences are repeated):

- It seems that the 'homeostasis' provides not only physical protection, but also determines the diffusion of chemical molecules...' Please correct the sentence as it is grammatically incorrect.

It is now corrected (page 2, line 1).

- The term 'tonic water' is not clear. I understand, after reading the paper, that it is about tonicity of the solutes injected into the mouse.

We use the term ‘tonic’ to indicate that in basal conditions, a constant water efflux occurs through the APQ4 channel.

- 'tonic aquaporin water efflux maintains volume equilibrium' - I believe it is about maintaining volume and osmotic equilibrium?

This description is now refined (now page 2, line 10).

- It is not clear whether the tonic water outflow refers to the cellular level or outflow from the brain parenchyma (i.e., glymphatic efflux)

It refers to the cellular level. 

INTRODUCTION/page 3:

- 'clearance of waste molecules from the brain as described in the glymphatic system' - The original papers describing the phenomena are not cited: Iliff et al. 2012, 2013, Mestre et al. 2018, as well as reviews by Nedergaard et al.

Indeed. We have now cited these key literatures (now page 4, line 10).

- 'brain water diffusion is the basis for diffusion-weighted magnetic resonance imaging (DW-MRI)' - The statement is wrong. it is the mobility of the water protons that DWI is based on, but not the diffusion of molecules in the brain. This should be clarified and based on the DW-MRI principle and the original works by Le Bihan from 1986, 1988, or 2015.

This sentence is now updated (page 4, line10-14).

- Similarly, I suggest correcting or removing the citations and the sentence part regarding the clinical use of DWI, as it has no value here. Instead, it would be worth mentioning what actually ADC reflects as a computational score, and what were the results from previous studies assessing glymphatic systems using DWI. This is especially important when considering the mislocalization of the AQP4 channel.

We now states recent studies using DW-MRI to evaluate glymphatic systems (page 4, line16-17).  

- 'In the brain, AQP4 is predominantly expressed in astrocytes'-please review the citations. I suggest reading the work by Nielsen 1997, Nagelhus 2013, Wolburg 2011, and Li and Wang from 2017. To my best knowledge, in the brain AQP4 is exclusively expressed in astrocytes.

Thanks for the reviewer. It is described that while enriched in astrocytes, AQP4 is also expressed in ependymal cells lining the ventricles (e.g., (Mayo et al., 2023; Verkman et al., 2006)). ‘predominantly’ is now removed (page 4, line 21).

- The conclusion: ' Our finding suggests that aquaporin acts as a water export route in astrocytes in physiological conditions, so as to counterbalance the constitutive intracellular water accumulation caused by constant transmitter and ion uptake, as well as the cytoplasmic metabolism processes. This mechanism hence plays a necessary role in maintaining water equilibrium in astrocytes, thereby brain water homeostasis' seems to be slightly beyond the actual findings in the paper. I suggest clarifying according to the described phenomena.

We have now refined the conclusion sticking to the experimental observations (page 5, line16-18).

- The introduction lacks important information on existing AQP4 blockers and their effects, pros and cons on why to use TGN-020. Among others, I would refer to recent work by Giannetto et al 2024, as well as previous work of Mestre et al. 2018 and Gomolka et al. 2023.

We initiated the study by using TGN-020 as an AQP4 blocker because it has been validated by wide range of ex vivo and in vivo studies as documented in the text (page 7, line 1-6). We also update discussions on the recent advances in validating the AQP4 blocker AER-270(271) while citing the relevant studies (page 15, line 7-17).  

RESULTS:

- Page 5, lines 19-20: '...transport, we performed fluorescence intensity translated (FIT) imaging.' - this term was never introduced in the methods so it is difficult for the reader to understand it at first sight. -'To this end,' - it is not clear which action refers to 'this'. (is it about previous works or the moment that the brain samples were ready for imaging? Please clarify, as it is only starting to be clear after fully reading the methods.

We now refine the description give the principle of our imaging method first, then explain the technical steps. To avoid ambiguity, the term ‘To this end’ is removed. The updated text is now on page 6, line 1-3.  

- From page 6 onwards - all references to Figures lack information to which part of the figure subpanel the information refers (top/middle bottom or left/middle/right).

We apologize. The complementary indication is now added for figure citations when applicable.  

- 'whereas water export and astrocyte shrinking upon hyperosmotic manipulation increased astrocyte fluorescence (Figure 1B). Hence, FIT imaging enables real-time recording of astrocyte transmembrane water transport and volume dynamics.' - this part seems to be undescribed or not clear in the methods.

We have now refined this description (page 6, line 19-20).

- Page 6, lines 17-22: TGN-020. In addition to the above, I suggest familiarizing also with the following works by Igarashi 2011. doi: 10.1007/s10072-010-0431-1, and by Sun 2022. doi: 10.3389/fimmu.2022.870029.

These studies are now cited (page 7, line 3-4).

- Page 7: ' AQP4 is a bidirectional channel facilitating... ' - AQP4 water channel is known as the path of least resistance for water transfer, please see Manley, Nature Medicine, 2000 and Papadopoulos, Faseb J, 2004.

This sentence is now updated (page 7, line 12-13).

- ' astrocyte AQP4 by TGN-020 caused a gradual decrease in SRB fluorescence intensity, indicating an intracellular water accumulation' - tissue slice experiment is a very valuable method. However it seems right, the experiment does not comment on the cell swelling that may occur just due to or as a superposition of tissue deterioration and the effect of TGN-020. The AQP4 channel is blocked, and the influx of water into astrocytes should be also blocked. Thus, can swelling be also a part of another mechanism, as it was also observed in the control group? I suggest this should be addressed thoroughly.

We performed this experiment in acute brain slices to well control the pharmacological environment and gain spatial-temporal information. Post slicing, the brain slices recovered > 1hr prior to recording, so that the slices were in a stable state before TGN-020 application as evidenced by the stable baseline. The constant decrease in the control trace is due to photobleaching which did not change its curve tendency in response to vehicle. TGN-020, in contrast, caused a down-ward change suggesting intracellular water accumulation and swelling. 

The experiment was performed at basal condition without active water influx; a decrease in SRB fluorescence hints astrocyteintracellular water buildup. This result shows that in basal condition, astrocyte aquaporin mediates a constant (i.e., tonic) water efflux; its blocking causes intracellular water accumulation and swelling. 

We have accordingly updated the description of this part (page 7, line 15-20).

- From the Figure 1 legend: Only 4 mice were subjected to the experiment, and only 1 mouse as a control. I suggest expanding the experiment and performing statistics including two-way ANOVA for data in panels B, C, and D, as no results of statistical tests confirm the significance of the findings provided.

The panel B confirms that cytosolic SRB fluorescence displays increasing tendency upon water efflux and volume shrinking, and vice versa. As for the panel C, the number of mice is now indicated. Also, the downward change in the SRB fluorescence was now respectively calculated for the phases prior and post to TGN (and vehicle) application, and this panel is accordingly updated. TGN-020 induced a declining in astrocyte SRB fluorescence, which is validated by t-test performed in MATLAB. To clarify, we now add cross-link lines to indicate statistical significance between the corresponding groups (Fig 1C, middle). As for panel D, we calculated the SRB fluorescence change (decrease) relative to the photobleaching tendency illustrated by the dotted line. The significance was also validated by t-test performed in MATLAB.  

- Figure 1: Please correct the figure - pictures in panel A are low quality and do not support the specificity of SRB for astrocytes. Panels B-D are easier to understand if plotted as normal X/Y charts with associated statistical findings. Some drawings are cut or not aligned.

In GFAP-EGFP transgenic, astrocytes are labeled by EGFP. SRB labeling (red fluorescence) shows colocalization with EGFP-positive astrocytes, meanwhile not all EGFP-positive astrocytes are labeled by SRB. The PDF conversion procedure during the submission may also somehow have compromised image quality. We have tried to update and align the figure panels.  

- Page 12: ' TGN-020 increased basal water diffusion within multiple regions including the cortex,

hippocampus and the striatum in a heterogeneous manner (Figure 5C).'

This sentence is updated now (page 12, line 12 – page13, line 2). It reads ‘The representative images reveal the enough image quality to calculate the ADC, which allow us to examine the effect of TGN-020 on water diffusion rate in multiple regions (Fig. 5C).’

- The expression of AQP4 within the brain parenchyma is known to be heterogenous. Please familiarize yourself with works by Hubbard 2015, Mestre 2018, and Gomolka 2023. A correlation between ADC score and AQP4 expression ROI-wise would be useful, but it is not substantial to conduct this experiment.

We thank the reviewer. This point is stressed on page 19, line 12-14.

DISCUSSION:

- Most of the issues are commented on above, so I suggest following the changes applied earlier. -Page 16: 'We show by DW-MRI that water transport by astrocyte aquaporin is critical for brain water homeostasis.' This statement is not clear and does not refer to the actual impact of the findings. DWI is allowed only to verify the changes of ADC fter the application of TGN-020. I suggest commenting on the recent report by Giannetto 2024 here.

This sentence is now refined (page 19, line 1-2), followed by the updates commenting on the recent studies employing DW-MRI to evaluate brain fluid transport, including the work of (Giannetto et al., 2024) (page 19, line 3-10). 

METHODS:

- Page 18: no total number of mice included in all experiments is provided, as well as no clearly stated number of mice used in each experiment. Please correct.

We have now double checked the number of the mice for the data presented and updated the figure legends accordingly (e.g., updates in legends fig1, fig5, etc).

-  Page 18, line 7: 'Axscience' is not a producer of Isoflurane, but a company offering help with scientific manuscript writing. If this company's help was used, it should be stated in the acknowledgments section. Reference to ISOVET should be moved from line 15 to line 7.

We apologize. We did not use external writing help, and now have removed the ‘Axcience’. The Isoflurane was under the mark ‘ISOVET’ from ‘Piramal’. This info is now moved up (page 21, line 11). 

- Page 18, line 9: ' modified artificial cerebrospinal fluid (aCSF)'. Additional information on the reason for the modified aCSF would be useful for the reader.

In this modified solution, the concentration of depolarizing ions (Na+, Ca2+) was reduced to lower the potential excitotoxicity during the tissue dissection (i.e., injury to the brain) for preparing the brain slices. Extra sucrose was added to balance the solution osmolarity. This solution has been used previously for the dissection and the slicing steps in adult mice (Jiang et al., 2016). We now add this justification in the text and quote the relevant reference (page 21, line14-16). 

- Page 19, line 6: a reasoning for using Tamoxifen would be helpful for the reader.

The Glast-CreERT2 is an inducible conditional mouse line that expresses Cre recombinase selectively in astrocytes upon tamoxifen injection. We now add this information in the text (page 22, line 10-11). 

- Line 8 - 'Sigma'

Fixed.

- Line 7/8: It is not clear if ethanol is of 10% solution or if proportions of ethanol+tamoxifen to oil were of 1:9. The reasoning for each performed step is missing.

We have now clarified the procedure (page 22, line 11-15).

- Line 10: '/' means 'or'?

Here, we mean the bigenic mice resulting from the crossing of the heterozygous Cre-dependent GCaMP6f and Glast-CreERT2 mouse lines. We now modify it to ‘Glast-CreERT2::Ai95GCaMP6f//WT’, in consistence with the presentation of other mouse lines in our manuscript (page 22, line 16).

- Lines 22-23: being in-line with legislation was already stated at the beginning of the Methods so I suggest combining for clearance.

Done. 

- Page 21, line 4: it is good to mention which printer was used, but it would be worth mentioning the material the chamber was printed from - was it ABS?

Yes. We add this info in the text now (page 24, line 5).

- Line 9 -'PI' requires spelling out.

It is ‘Physik Instrumente’, now added (page 24, line 10).

- Line 11-12: What is the reason for background subtraction - clearer delineation of astrocytes/ increasing SNR in post-processing, or because SRB signal was also visible and changing in the background over time? Was the background removed in each frame independently (how many frames)? How long was the time-lapse and was the F0 frame considered as the first frame acquired? The background signal should be also measured and plotted alongside the astrocytic signal, as a reference (Figure 1). This should be clarified so that steps are to be followed easily.

We sought to follow the temporal changes in SRB fluorescence signal. The acquired fluorescent images contain not only the SRB signals, but also the background signals consisting of for instance the biological tissue autofluorescence, digital camera background noise and the leak light sources from the environments. The value of the background signal was estimated by the mean fluorescence of peripheral cell-free subregions (15 × 15 µm²) and removed from all frames of time-lapse image stack. The traces shown in the figures reflect the full lengths of the time-lapse recordings. F0 was identified as the mean value of the 10 data points immediately preceding the detected fluorescence changes. The text is now updated (page 24 line 21 - page 25 line 5).

- Line 15: Was astrocyte image delineation performed manually or automatically? Where was the center of the region considered in the reference to the astrocyte image? It would be good to see the regions delineated for reference.

Astrocytes labeled by SRB were delineated manually with the soma taken as the center of the region of interest. We now exemplify the delineated region in Fig 1A, bottom.

- Page 22, line 2: 'x4 objective'.

Added (now, page 25, line 16). 

- Line 3: 'barrels' - reference to publication or the explanation missing.

The relevant reference is now added on barrel cortex (Erzurumlu and Gaspar, 2020) (page 25, line 19-20). 

- Line 19: were the coordinates referred to = bregma?

Yes. This info is now added (page 26, line 12). 

- Line 20: was the habituation performed directly at the acquisition date? It is rather difficult to say that it was a habituation, but rather acute imaging. I suggest correcting, that mice were allowed to familiarize themselves with the setup for 30 minutes prior to the imaging start.

In this context, although it is a very nice idea and experiment, the influence of acute stress in animals familiar with the setup only from the day of acquisition is difficult to avoid. It is a major concern, especially when considering norepinephrine as a master driver of neuronal and vascular activity through the brain, and strong activation of the hypothalamic-adrenal axis in response to acute stress. It is well known, that the response of monoamines is reduced in animals subjected to chronic v.s acute stress, but still larger than that if the stressor is absent.

Major remark: The animals should, preferably, be imaged at least after 3 days of habituation based on existing knowledge. I suggest exploring the topic of the importance of habituation. It is difficult though, to objectively review these findings without considering stress and associated changes in vascular dynamics.

Many thanks for the reviewer to help to precise this information. The text is accordingly updated to describe the experiment (now page 26, line 14). 

- Page 23, line 17: number of animals included in experiments missing.

The number of animals is added in Methods (page 27, line 12) and indicated in the legend of Figure 5. 

- Line 18/19: were the respiratory effects observed after injection of saline or TGN-020? Since DWI was performed, the exclusion of perfusive flow on ADC is impossible.

I suggest an additional experiment in n=3 animals per group, verifying the HR (and if possible BP) response after injection of TGN-020 and saline in mice.

The respiratory rate has been recorded. We added the averaged respiratory rate before and after injection of TGN-020 or saline (now, Fig. S6; page 13, line 5-6).

- Line 22: Please, provide the model of the scanner, the model of the cryoprobe, as well as the model of the gradient coil used, otherwise it is difficult to assess or repeat these experiments.

We have now added the information of MRI system in Methods section (page 27, line17-21).

- Page 24: line 3/4: although the achieved spatial resolution of DWI was good and slightly lower than desired and achievable due to limitations of the method itself as well as cryoprobe, it is acceptable for EPI in mice.

Still, there is no direct explanation provided on the reasoning for using surface instead of volumetric coil, as well as on assuming an anisotropic environment (6 diffusion directions) for DWI measurements. This is especially doubtful if such a long echo-time was used alongside lower-thanpossible spatial resolution. Longer echo time would lower the SNR of the depicted signal but also would favor the depiction of signal from slow-moving protons and larger water pools. On the other hand, only 3 b-values were used, which is the minimum for ADC measurements, while a good research protocol could encompass at least 5 to increase the accuracy of ADC estimation and avoid undersampling between 250 and 1800 b-values. What was the reason for choosing this particular set of b-values and not 50, 600, and 2000? Besides, gradient duration time was optimally chosen, however, I have concerns about the decision for such a long gradient separation times.

If the protocol could have been better optimized, the assessment could have been also performed in respiratory-gated mode, allowing minimization of the effects of one of the glymphatic system driving forces.

Thus, I suggest commenting on these issues.

We chose the cryoprobe to increase the signal-to-noise ratio (SNR) in DW-MRI with long echo-time and high b-value. The volume coil has a more homogeneous SNR in the whole brain rather than the cryoprobe, but SNR should be reduced compared with cryoprobe. We confirmed that, even at the ventral part of the brain, the image quality of DW-MRI images was enough to investigate the ADC with cryoprobe (Fig. 5B-C). This is mentioned now in Methods (page 27, line 17-21).

We performed DW-MRI scanning for 5 min at each time-point using the condition of anisotropic resolution and 3 b-values, to investigate the time-course of ADC change following the injection of TGN020. Because the effect of TGN-020 appears about dozen of minutes post the injection (Igarashi et al., 2011), fast DW-MRI scanning is required. If isotropic DW-MRI with lower echo-time and more direction is used, longer scan time at each time point is required, maybe more than 1h. We agree that three bvalues is minimum to calculate the ADC and more b-values help to increase the accuracy. However, to achieve the temporal resolution so as to better catch the change of water diffusion, we have decided to use the minimum b-values. The previous study also validates the enough accuracy of DW-MRI with three b-values (Ashoor et al., 2019). Furthermore, previous study that used long diffusion time (> 20 ms) and long echo time (40 ms) shows the good mean diffusivity (Aggarwal et al., 2020), supporting that our protocol is enough to investigate the ADC. We have now updated the description (page 28 line 5-9).  The reason why we choose the b = 250 and 1800 s/mm² is that 2000 s/mm² seems too high to get the good quality of image. In the previous study, we have optimized that ADC is measurable with b = 0, 250, and 1800 s/mm² (Debacker et al., 2020). 

- Page 24, line 7: What was the post-processing applied for images acquired over 70 minutes? Did it consider motion-correction, co-registration, or drift-correction crucial to avoid pitfalls and mismatches in concluding data?

The motion correction and co-registration were explained in Methods (page 28, line 12-14).

Also, were these trace-weighted images or magnitude images acquired since DTI software was used for processing - while ADC fitting could be reliably done in Matlab, Python, or other software. Thus, was DSI software considering all 3 b-values or just used 0 and 1800 for the calculation of mean diffusivity for tractography (as ADC). The details should be explained.

DSIstudio was used with all three b values (b = 0, 250, and 1800 s/mm²) to calculate the ADC. We added the description in Methods (page 28, line 16-18).

To make sure that the results are not affected by the MR hardware, I suggest performing 3 control measurements in a standard water phantom, and presenting the results alongside the main findings.

Thanks for this suggestion. We have performed new experiments and now added the control measurement with three phantoms, that is water, undecane, and dodecane. These new data are summarized now in Fig. S7, showing the stability of ADC throughout the 70 min scanning. We have updated the description on Method part (page 28, line 9-11) and on the Results (page 13, line 6-8).  

- Line 13: were the ROI defined manually or just depicted from previously co-registered Allen Brain atlas?

The ROIs of the cortex, the hippocampus, and the striatum were depicted with reference to Allen mouse brain atlas (https://scalablebrainatlas.incf.org/mouse/ABA12). This is explained in Methods (page 28, line 14-16).

- Line 10: why the average from 1st and 2nd ADC was not considered, since it would reduce the influence of noise on the estimation of baseline ADC?

We are sorry that it was a typo. The baseline was the average between 1st and 2nd ADC. We corrected the description (page 28, line 20).

STATISTIC:

Which type of t-test - paired/unpaired/two samples was used and why? Mann-Whitney U-tets are used as a substitution for parametric t-tests when the data are either non-parametric or assuming normal distribution is not possible. In which case Bonferroni's-Holm correction was used? - I couldn't find any mention of any multiple-group analysis followed by multiple comparisons. Each section of the manuscript should have a description of how the quantitative data were treated and in which aim. I suggest carefully correcting all figures accordingly, and following the remarks given to the Figure 1.

We used unpaired t-test for data obtained from samples of different conditions. Indeed, MannWhitney U-test is used when the data are non-parametric deviating from normal distributions.  Bonferroni-Holm correction was used for multiple comparisons (e.g., Fig. 4D-E).

Reviewer #3 (Recommendations For The Authors):

I think that the following statement is insufficient: "The authors commit to share data, documentation, and code used in analysis". My understanding is eLife expects that all key data to be provided in a supplement.

We thank the reviewer; we follow the publication guidelines of eLife. 

References

Aggarwal, M., Smith, M.D., and Calabresi, P.A. (2020). Diffusion-time dependence of diffusional kurtosis in the mouse brain. Magn Reson Med 84, 1564-1578.

Ashoor, M., Khorshidi, A., and Sarkhosh, L. (2019). Estimation of microvascular capillary physical parameters using MRI assuming a pseudo liquid drop as model of fluid exchange on the cellular level. Rep Pract Oncol Radiother 24, 3-11.

Cauli, B., and Hamel, E. (2018). Brain Perfusion and Astrocytes. Trends in neurosciences 41, 409-413.

Debacker, C., Djemai, B., Ciobanu, L., Tsurugizawa, T., and Le Bihan, D. (2020). Diffusion MRI reveals in vivo and non-invasively changes in astrocyte function induced by an aquaporin-4 inhibitor. PLoS One 15, e0229702.

Erzurumlu, R.S., and Gaspar, P. (2020). How the Barrel Cortex Became a Working Model for Developmental Plasticity: A Historical Perspective. J Neurosci 40, 6460-6473.

Farr, G.W., Hall, C.H., Farr, S.M., Wade, R., Detzel, J.M., Adams, A.G., Buch, J.M., Beahm, D.L., Flask, C.A., Xu, K., et al. (2019). Functionalized Phenylbenzamides Inhibit Aquaporin-4 Reducing Cerebral Edema and Improving Outcome in Two Models of CNS Injury. Neuroscience 404, 484-498.

Giannetto, M.J., Gomolka, R.S., Gahn-Martinez, D., Newbold, E.J., Bork, P.A.R., Chang, E., Gresser, M., Thompson, T., Mori, Y., and Nedergaard, M. (2024). Glymphatic fluid transport is suppressed by the aquaporin-4 inhibitor AER-271. Glia.

Gomolka, R.S., Hablitz, L.M., Mestre, H., Giannetto, M., Du, T., Hauglund, N.L., Xie, L., Peng, W., Martinez, P.M., Nedergaard, M., et al. (2023). Loss of aquaporin-4 results in glymphatic system dysfunction via brain-wide interstitial fluid stagnation. eLife 12.

Huber, V.J., Tsujita, M., and Nakada, T. (2009). Identification of aquaporin 4 inhibitors using in vitro and in silico methods. Bioorg Med Chem 17, 411-417.

Igarashi, H., Huber, V.J., Tsujita, M., and Nakada, T. (2011). Pretreatment with a novel aquaporin 4 inhibitor, TGN-020, significantly reduces ischemic cerebral edema. Neurol Sci 32, 113-116.

Igarashi, H., Tsujita, M., Suzuki, Y., Kwee, I.L., and Nakada, T. (2013). Inhibition of aquaporin-4 significantly increases regional cerebral blood flow. Neuroreport 24, 324-328.

Jiang, R., Diaz-Castro, B., Looger, L.L., and Khakh, B.S. (2016). Dysfunctional Calcium and Glutamate Signaling in Striatal Astrocytes from Huntington's Disease Model Mice. J Neurosci 36, 3453-3470.

Mayo, F., Gonzalez-Vinceiro, L., Hiraldo-Gonzalez, L., Calle-Castillejo, C., Morales-Alvarez, S., Ramirez-Lorca, R., and Echevarria, M. (2023). Aquaporin-4 Expression Switches from White to Gray Matter Regions during Postnatal Development of the Central Nervous System. Int J Mol Sci 24.

Mola, M.G., Sparaneo, A., Gargano, C.D., Spray, D.C., Svelto, M., Frigeri, A., Scemes, E., and Nicchia, G.P. (2016). The speed of swelling kinetics modulates cell volume regulation and calcium signaling in astrocytes: A different point of view on the role of aquaporins. Glia 64, 139-154.

Risher, W.C., Andrew, R.D., and Kirov, S.A. (2009). Real-time passive volume responses of astrocytes to acute osmotic and ischemic stress in cortical slices and in vivo revealed by two-photon microscopy. Glia 57, 207-221.

Salman, M.M., Kitchen, P., Yool, A.J., and Bill, R.M. (2022). Recent breakthroughs and future directions in drugging aquaporins. Trends Pharmacol Sci 43, 30-42.

Shigetomi, E., Bushong, E.A., Haustein, M.D., Tong, X., Jackson-Weaver, O., Kracun, S., Xu, J., Sofroniew, M.V., Ellisman, M.H., and Khakh, B.S. (2013). Imaging calcium microdomains within entire astrocyte territories and endfeet with GCaMPs expressed using adeno-associated viruses. J Gen Physiol 141, 633-647.

Solenov, E., Watanabe, H., Manley, G.T., and Verkman, A.S. (2004). Sevenfold-reduced osmotic water permeability in primary astrocyte cultures from AQP-4-deficient mice, measured by a fluorescence quenching method. Am J Physiol Cell Physiol 286, C426-432.

Verkman, A.S., Binder, D.K., Bloch, O., Auguste, K., and Papadopoulos, M.C. (2006). Three distinct roles of aquaporin-4 in brain function revealed by knockout mice. Biochim Biophys Acta 1758, 10851093.

Author response:

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

Public Reviews:  

Reviewer #1 (Public Review):  

Summary:  

The authors have presented data showing that there is a greater amount of spontaneous differentiation in human pluripotent cells cultured in suspension vs static and have used PKCβ and Wnt signaling pathway inhibitors to decrease the amount of differentiation in suspension culture.  

Strengths:  

This is a very comprehensive study that uses a number of different rector designs and scales in addition to a number of unbiased outcomes to determine how suspension impacts the behaviour of the cells and in turn how the addition of inhibitors counteracts this effect. Furthermore, the authors were also able to derive new hiPSC lines in suspension with this adapted protocol.  

Weaknesses:  

The main weakness of this study is the lack of optimization with each bioreactor change. It has been shown multiple times in the literature that the expansion and behaviour of pluripotent cells can be dramatically impacted by impeller shape, RPM, reactor design, and multiple other factors. It remains unclear to me how much of the results the authors observed (e.g. increased spontaneous differentiation) was due to not having an optimized bioreactor protocol in place (per bioreactor vessel type). For instance - was the starting seeding density, RPM, impeller shape, feeding schedule, and/or any other aspect optimized for any of the reactors used in the study, and if not, how were the values used in the study determined?  

Thank you for your thoughtful comments. According to your comments, we have performed several experiments to optimize the bioreactor conditions in revised manuscripts. We tested several cell seeding densities and several stirring speeds with or without WNT/PKCβ inhibitors  (Figure 6—figure supplement 1). We found that 1 - 2 x 105 cells/mL of the seeding densities and 50 - 150 rpm of the stirring speeds were applicable in the proliferation of these cells. Also, PKCβ and Wnt inhibitors suppressed spontaneous differentiation in bioreactor conditions regardless with stirring speeds. As for the impeller shape and reactor design, we just used commonly-used ABLE's bioreactor for 30 mL scale and Eppendorf's bioreactors for 320 mL scale, which had been designed and used for human pluripotent stem cell culture conditions in previous studies, respectively (Matsumoto et al., 2022 (doi: 10.3390/bioengineering9110613); Kropp et al., 2016 (doi: 10.5966/sctm.2015-0253)). We cited these previous studies in the Results and Materials and Methods section. We believe that these additional data and explanation are sufficient to satisfy your concerns on the optimization of bioreactor experiments.

Reviewer #2 (Public Review):  

This study by Matsuo-Takasaki et al. reported the development of a novel suspension culture system for hiPSC maintenance using Wnt/PKC inhibitors. The authors showed elegantly that inhibition of the Wnt and PKC signaling pathways would repress spontaneous differentiation into neuroectoderm and mesendoderm in hiPSCs, thereby maintaining cell pluripotency in suspension culture. This is a solid study with substantial data to demonstrate the quality of the hiPSC maintained in the suspension culture system, including long-term maintenance in >10 passages, robust effect in multiple hiPSC lines, and a panel of conventional hiPSC QC assays. Notably, large-scale expansion of a clinical grade hiPSC using a bioreactor was also demonstrated, which highlighted the translational value of the findings here. In addition, the author demonstrated a wide range of applications for the IWR1+LY suspension culture system, including support for freezing/thawing and PBMC-iPSC generation in suspension culture format. The novel suspension culture system reported here is exciting, with significant implications in simplifying the current culture method of iPSC and upscaling iPSC manufacturing.  

Another potential advantage that perhaps wasn't well discussed in the manuscript is the reported suspension culture system does not require additional ECM to provide biophysical support for iPSC, which differentiates from previous studies using hydrogel and this should further simplify the hiPSC culture protocol.  

Interestingly, although several hiPSC suspension media are currently available commercially, the content of these suspension media remained proprietary, as such the signaling that represses differentiation/maintains pluripotency in hiPSC suspension culture remained unclear. This study provided clear evidence that inhibition of the Wnt/PKC pathways is critical to repress spontaneous differentiation in hiPSC suspension culture.  

I have several concerns that the authors should address, in particular, it is important to benchmark the reported suspension system with the current conventional culture system (eg adherent feeder-free culture), which will be important to evaluate the usefulness of the reported suspension system.  

Thank you for this insightful suggestion. In this revised manuscript, we have performed additional experiments using conventional media, mTeSR1 (Stem Cell Technologies, Vancouver, Canada), comparing with the adherent feeder-free culture system in four different hiPSC lines simultaneously. Compared to the adherent conditions, the suspension conditions without chemical treatment decreased the expression of self-renewal marker genes/proteins and increased the expression levels of SOX17, T, and PAX6 (Figure 4 - figure supplement 2). Importantly, the treatment of LY333531 and IWR-1-endo in mTeSR1 medium reversed the decreased expression of these undifferentiated markers and suppressed the increased expression of differentiation markers in suspension culture conditions, reaching the comparable levels of the adherent culture conditions. These results indicated that these chemical treatments in suspension culture are beneficial even when using a conventional culture medium.

Also, the manuscript lacks a clear description of a consistent robust effect in hiPSC maintenance across multiple cell lines.  

Thank you for this insightful suggestion. We have performed additional experiments on hiPSC maintenance across 5 hiPSC lines in suspension culture using StemFit AK02N medium simultaneously (Figure 3C - E). Overall, the treatment of LY333531 and IWR-1-endo in the StemFit AK02N medium reversed the decreased expression of these undifferentiated markers and suppressed the increased expression of differentiation markers in suspension culture conditions. Also as above, we have added results using conventional media, mTeSR1, in comparison to the adherent feeder-free culture system in four different hiPSC lines simultaneously. These results show that this chemical treatment consistently produced robust effects in hiPSC maintenance across multiple cell lines using multiple conventional media.

There are also several minor comments that should be addressed to improve readability, including some modifications to the wording to better reflect the results and conclusions.  

In the revised manuscript, we have added and corrected the descriptions to improve readability, including some modifications to the wording to better reflect the results and conclusions. 

Reviewer #3 (Public Review):  

In the current manuscript, Matsuo-Takasaki et al. have demonstrated that the addition of PKCβ and WNT signaling pathway inhibitors to the suspension cultures of iPSCs suppresses spontaneous differentiation. These conditions are suitable for large-scale expansion of iPSCs. The authors have shown that they can perform single-cell cloning, direct cryopreservation, and iPSC derivation from PBMCs in these conditions. Moreover, the authors have performed a thorough characterization of iPSCs cultured in these conditions, including an assessment of undifferentiated stem cell markers and genetic stability. The authors have elegantly shown that iPSCs cultured in these conditions can be differentiated into derivatives of three germ layers. By differentiating iPSCs into dopaminergic neural progenitors, cardiomyocytes, and hepatocytes they have shown that differentiation is comparable to adherent cultures.

This new method of expanding iPSCs will benefit the clinical applications of iPSCs.  

Recently, multiple protocols have been optimized for culturing human pluripotent stem cells in suspension conditions and their expansion. Additionally, a variety of commercially available media for suspension cultures are also accessible. However, the authors have not adequately justified why their conditions are superior to previously published protocols (indicated in Table 1) and commercially available media. They have not conducted direct comparisons.  

Thank you for this careful suggestion. In this revised manuscript, we have added results using a conventional medium, mTeSR1 (Stem Cell Technologies), which has been used for the suspension culture in several studies. Compared to the adherent conditions using mTeSR1 medium, the suspension conditions with the same medium decreased the ratio of TRA1-60/SSEA4-positive cells and OCT4positive cells and the expression levels of OCT4 and NANOG and decreased the expression levels of SOX17, T, and PAX6 in 4 different hiPSC lines simultaneously (Figure 4 - Supplement 2). Importantly, the treatment of LY333531 and IWR-1-endo in the mTeSR1 medium reversed the decreased expression of these undifferentiated markers. With these direct comparisons, we were able to justify why our conditions are superior to previously published protocols using commercially available media.

Additionally, the authors have not adequately addressed the observed variability among iPSC lines. While they claim in the Materials and Methods section to have tested multiple pluripotent stem cell lines, they do not clarify in the Results section which line they used for specific experiments and the rationale behind their choices. There is a lack of comparison among the different cell lines. It would also be beneficial to include testing with human embryonic stem cell lines.  

Thank you for this insightful suggestion. In this revised manuscript, we have added results on 5 different hiPSC lines at the same time (Figure 3 C-E). Excuse for us, but it is hard to use human embryonic stem cell lines for this study due to ethical issues in Japanese governmental regulations. The treatment of LY333531 and IWR-1-endo increased the expression of self-renewal marker genes/proteins and decreased the expression levels of SOX17, T, and PAX6 in these hiPSC lines in general. These results indicated that these chemical treatments in suspension culture were robust in general while addressing the observed variability among iPSC lines.

Additionally, there is a lack of information regarding the specific role of the two small molecules in these conditions.  

In this revised manuscript, we have added data and discussion regarding the specific role of the two small molecules in these conditions in the Results and Discussion section. For using WNT signaling inhibitor, we hypothesized that adding Wnt signaling inhibitors may inhibit the spontaneous differentiation of hiPSCs into mesendoderm. Because exogenous Wnt signaling induces the differentiation of human pluripotent stem cells into mesendoderm lineages (Nakanishi et al, 2009; Sumi et al, 2008; Tran et al, 2009; Vijayaragavan et al, 2009; Woll et al, 2008). Also, endogenous expression and activation of Wnt signaling in pluripotent stem cells are involved in the regulation of mesendoderm differentiation potentials (Dziedzicka et al, 2021). For using PKC inhibitors, "To identify molecules with inhibitory activity on neuroectodermal differentiation, hiPSCs were treated with candidate molecules in suspension conditions. We selected these candidate molecules based on previous studies related to signaling pathways or epigenetic regulations in neuroectodermal development (reviewed in (GiacomanLozano et al, 2022; Imaizumi & Okano, 2021; Sasai et al, 2021; Stern, 2024) ) or in pluripotency safeguards (reviewed in (Hackett & Surani, 2014; Li & Belmonte, 2017; Takahashi & Yamanaka, 2016; Yagi et al, 2017))." 

We also found that the expression of naïve pluripotency markers, KLF2, KLF4, KLF5, and DPPA3, were up-regulated in the suspension conditions treated with LY333531 and IWR-1-endo while the expression of OCT4 and NANOG was at the same levels (Figure 5—figure supplement 2). Combined with RT-qPCR analysis data on 5 different hiPSC lines (Figure 3E), these results suggest that IWRLY conditions may drive hiPSCs in suspension conditions to shift toward naïve pluripotent states.

The authors have not attempted to elucidate the underlying mechanism other than RNA expression analysis.  

Regarding the underlying mechanisms, we have added results and discussion in the revised manuscript.  For Wnt activation in human pluripotent stem cells, several studies reported some WNT agonists were expressed in undifferentiated human pluripotent stem cells (Dziedzicka et al., 2021; Jiang et al, 2013; Konze et al, 2014). In suspension culture, cell aggregation causes tight cell-cell interaction. The paracrine effect of WNT agonists in the cell aggregation may strongly affect neighbor cells to induce spontaneous differentiation into mesendodermal cells. Thus, we think that the inhibition of WNT signaling is effective to suppress the spontaneous differentiation into mesendodermal lineages in suspension culture.

For PKC beta activation in human pluripotent stem cells, we have shown that phosphorylated PKC beta protein expression is up-regulated in suspension culture than in adherent culture with western blotting (Figure 3 - figure supplement 1). The treatment of PKCβ inhibitor is effective to suppress spontaneous differentiation into neuroectodermal lineages. For future perspectives, it is interesting to examine (1) how and why PKCβ is activated (or phosphorylated), especially in suspension culture conditions, and (2) how and why PKCβ inhibition can suppress the neuroectodermal differentiation. Conversely, it is also interesting to examine how and why PKCβ activation is related to neuroectodermal differentiation.

For these reasons some aspects of the manuscript need to be extended:  

(1) It is crucial for authors to specify the culture media used for suspension cultures. In the Materials and Methods section, the authors mentioned that cells in suspension were cultured in either StemFit AK02N medium, 415 StemFit AK03N (Cat# AK03N, Ajinomoto, Co., Ltd., Tokyo, Japan), or StemScale PSC416 suspension medium (A4965001, Thermo Fisher Scientific, MA, USA). The authors should clarify in the text which medium was used for suspension cultures and whether they observed any differences among these media.  

Sorry for this confusion. Basically in this study, we use StemFit AK02N medium (Figure 1-5, 7-9). For bioreactor experiments (Figure 6), we use StemFit AK03N medium, which is free of human and animalderived components and GMP grade. To confirm the effect of IWRLY chemical treatment, we use StemScale suspension medium (Figure 4 - figure supplement 1) and mTeSR1 medium (Figure 4 - figure supplement 2 and Figure 8 - figure supplement 1). In the revised manuscript we clarified which medium was used for suspension cultures in the Results and Materials and Methods section.

Although we have not compared directly among these media in suspension culture (, which is primarily out of the focus of this study), we have observed some differences in maintaining self-renewal characteristics, preventing spontaneous differentiation (including tendencies to differentiate into specific lineages), stability or variation among different experimental times in suspension culture conditions. Overcoming these heterogeneity caused by different media, the IWRLY chemical treatment stably maintain hiPSC self-renewal in general. We have added this issue in the Discussion section.

(2) In the Materials and Methods section, the authors mentioned that they used multiple cell lines for this study. However, it is not clear in the text which cell lines were used for various experiments. Since there is considerable variation among iPSC lines, I suggest that the authors simultaneously compare 2 to 3 pluripotent stem cell lines for expansion, differentiation, etc.  

Thank you for this careful suggestion. We have added more results on the simultaneous comparison using StemFit AK02N medium in 5 different hiPSC lines (Figure 3 C-E) and using mTeSR1 medium in 4 different hiPSC lines (Figure 4 - figure supplement 2). From both results, we have shown that the treatment of LY333531 and IWR-1-endo was beneficial in maintaining the self-renewal of hiPSCs while suppressing spontaneous differentiation.

(3) Single-cell sorting can be confusing. Can iPSCs grown in suspensions be single-cell sorted?

Additionally, what was the cloning efficiency? The cloning efficiency should be compared with adherent cultures.  

Sorry for this confusion. With our method, iPSCs grown in IWRLY suspension conditions can be singlecell sorted. We have improved the clarity of the schematics (Figure 7A). Also, we added the data on the cloning efficiency, which are compared with adherent cultures (Figure 7B). The cloning efficiency of adherent cultures was around 30%. While the cloning efficiency of suspension cultures without any chemical treatment was less than 10%, the IWR-1-endo treatment in the suspension cultures increased the efficiency was more than 20%. However, the treatment of LY333531 decreased the efficiency. These results indicated that the IWR-1-endo treatment is beneficial in single-cell cloning in suspension culture.

(4) The authors have not addressed the naïve pluripotent state in their suspension cultures, even though PKC inhibition has been shown to drive cells toward this state. I suggest the authors measure the expression of a few naïve pluripotent state markers and compare them with adherent cultures  

Thank you for this insightful comment. In the revised manuscript, we have added the data of RT-qPCR in 5 different hiPSC lines and specific gene expression from RNA-seq on naïve pluripotent state markers (Figure 3E and Figure 5 - figure supplement 2), respectively. Interestingly, the expression of KLF2, KLF4, KLF5, and DPPA3 is significantly up-regulated in IWRLY conditions. These results suggested that IWRLY suspension conditions drove hiPSCs toward naïve pluripotent state.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):  

Overall, I feel that this study is very interesting and comprehensive, but has significant weaknesses in the bioprocessing aspects. More optimization data is required for the suspension culture to truly show that the differentiation they are observing is not an artifact of a non-optimized protocol.  

Thank you for your thoughtful comments. Following your comments, we have performed several experiments to optimize the bioreactor conditions in revised manuscripts. We tested several cell seeding densities and several stirring speeds with or without WNT/PKCβ inhibitors (Figure 6—figure supplement 1). From these optimization experiments, we found that 1 - 2 x 105 cells/mL of the seeding densities and 50 - 150 rpm of the stirring speeds were applicable in the proliferation of these cells. Also, PKCβ and Wnt inhibitors suppressed spontaneous differentiation in bioreactor conditions regardless with acceptable stirring speeds. As for the impeller shape and reactor design, we just used commonly-used ABLE's bioreactor for 30 mL scale and Eppendorf's bioreactors for 320 mL scale, which had been designed and used for human pluripotent stem cell culture conditions in previous studies, respectively (Matsumoto et al., 2022 (doi: 10.3390/bioengineering9110613); Kropp et al., 2016 (doi:10.5966/sctm.2015-0253). We cited these previous studies in the Results section. We believe that these additional data and explanation are sufficient to satisfy your concerns on the optimization of bioreactor experiments.

Reviewer #2 (Recommendations For The Authors):  

The following comments should be addressed by the authors to improve the manuscript:  

(1) Abstract: '...a scalable culture system that can precisely control the cell status for hiPSCs is not developed yet.' There were previous reports for a scalable iPSC culture system so I would suggest toning down/rephrasing this point: eg that improvement in a scalable iPSC culture system is needed.  

Thank you for this careful suggestion. Following this suggestion, We have changed the sentence as "the improvement in a scalable culture system that can precisely control the cell status for hiPSCs is needed."

(2) Line 71: please specify what media was used as a 'conventional medium' for suspension culture, was it Stemscale?  

As suggested, we specified the media as StemFit AK02N used for this experiment. 

(3) Fig 1E: It's not easy to see gating in the FACS plots as the threshold line is very faint, please fix this issue.  

As suggested, we used thicker lines for the gating in the FACS plots (Figure 1E).

(4) Fig 1G-J, Fig 2D-H: The RNAseq figures appeared pixelated and the resolution of these figures should be improved. The x-axis label for Fig 1H is missing.  

We have improved these figures in their resolution and clarity. Also, we have added the x-axis label as "enrichment distribution" for gene set enrichment analysis (GSEA) in Figures 1H, 5F, and 5- figure supplement 1B.

(5) Line 103-107: 'Since Wnt signaling induces the differentiation of human pluripotent stem cells into mesendoderm lineages, and is endogenously involved in the regulation of mesendoderm differentiation of pluripotent stem cells.....'. The two points seem the same and should be clarified.  

Sorry for this unclear description. We have changed this description as "Exogenous Wnt signaling induces the differentiation of human pluripotent stem cells into mesendoderm lineages (Nakanishi et al, 2009; Sumi et al, 2008; Tran et al, 2009; Vijayaragavan et al, 2009; Woll et al, 2008). Also, endogenous expression and activation of WNT signaling in pluripotent stem cells are involved in the regulation of mesendoderm differentiation potentials (Dziedzicka et al, 2021; Jiang et al, 2013)." With this description, we hope that you will understand the difference of two points.

(6) Line 113: 'In samples treated with inhibitors' should be 'In samples treated with Wnt inhibitors'.  

Thank you for this careful suggestion. We have corrected this. 

(7) Line 115: '....there was no reduction in PAX6 expression.' That's not entirely correct, there was a reduction in PAX6 in IWR-1 endo treatment compared to control suspension culture (is this significant?), but not consistently for IWP-2 treatment. Please rephrase to more accurately describe the results.  

Sorry for this inaccurate description. We have corrected this phrase as "there was only a small reduction in PAX6 expression in the IWR-1-endo-treated condition and no reduction in the IWP2-treated condition" as recommended.

(8) It's critical to show that the effect of the suspension culture system developed here can maintain an undifferentiated state for multiple hiPSC lines. I think the author did test this in multiple cell lines, but the results are scattered and not easy to extract. I would recommend adding info for the hiPSC line used for the results in the legend, eg WTC11 line was used for Figure 3, 201B7 line was used for Figure 2. I would suggest compiling a figure that confirms the developed suspension system (IWR-1 +LY) can support the maintenance of multiple hiPSC lines.  

Thank you for this insightful suggestion. We have added data on hiPSC maintenance across 5 hiPSC lines in suspension culture using StemFit AK02N medium simultaneously (Figure 3C - E) and on hiPSC maintenance across 4 hiPSC lines in suspension culture using mTeSR1 medium simultaneously  (Figure 4 - figure supplement 2). Together, the treatment of LY333531 and IWR-1-endo in these media reversed the decreased expression of these undifferentiated markers and suppressed the increased expression of differentiation markers in suspension culture conditions. These results show that these chemical treatment produced a consistent robust effect in hiPSC maintenance across multiple cell lines.

(9) Line 166: Please use the correct gene nomenclature format for a human gene (italicised uppercase) throughout the manuscript. Also, list the full gene name rather than PAX2,3,5.  

Sorry for the incorrectness of the gene names. We have corrected them.

(10) Please improve the resolution for Figure 4D.  

We have provided clearer images of Figure 4D.

(11) In the first part of the study, the control condition was referred to as 'suspension culture' with spontaneous differentiation, but in the later parts sometimes the term 'suspension culture' was used to describe the IWR1+LY condition (ie lines 271-272). I would suggest the authors carefully go through the manuscript to avoid misinterpretation on this issue.  

Thank you for this careful suggestion. To avoid this misinterpretation on this issue, we use 'suspension culture' for just the conventional culture medium and 'LYIWR suspension culture' for the culture medium supplemented with LY333531 and IWR1-endo in this manuscript.

(12) Figure 5: It is impressive to demonstrate that the IWR1+LY suspension culture enables large-scale expansion of a clinical-grade hiPSC line using a bioreactor, yielding 300 vials/passage. Can the author add some information regarding cell yield using a conventional adherent culture system in this cell line? This will provide a comparison of the performance of the IWR1+LY suspension culture system to the conventional method.  

Thank you for this valuable suggestion. We have provided information regarding cell yield using a conventional adherent culture system in this cell line in the Results as "Since the population doubling time (PDT) of this hiPSC line in adherent culture conditions is 21.8 - 32.9 hours at its production (https://www.cira-foundation.or.jp/e/assets/file/provision-of-ips-cells/QHJI14s04_en.pdf), this proliferation rate in this large scale suspension culture is comparable to adherent culture conditions."

(13) Line 273: For testing the feasibility of using IWR1+LY media to support the freeze and thaw process, the author described the cell number and TRA160+/OCT4+ cell %. How is this compared to conventional media (eg E8)? It would be nice to see a head-to-head comparison with conventional media, quantification of cell count or survival would be helpful to determine this.  

For this issue, we attempted a direct freeze and thaw process using conventional media, StemFit AK02N in 201B7 line (Figure 8) or mTeSR1 in 4 different hiPSC lines(Figure 8 - figure supplement 1) with or without IWR1+LY. However, since the hiPSCs cultured in suspension culture conditions without IWR1+LY quickly lost their self-renewal ability, these frozen cells could not be recovered in these conditions nor counted. Our results indicate that the addition to IWR1+LY in the thawing process support the successful recovery in suspension conditions.

(14) More details of the passaging method should be added in the method section. Do you do cell count following accutase dissociation and replate a defined density (eg 1x10^5/ml)?  

Yes. We counted the cells in every passage in suspension culture conditions. We have added more explanation in the Materials and Methods as below.

"The dissociated cells were counted with an automatic cell counter (Model R1, Olympus) with Trypan Blue staining to detect live/dead cells. The cell-containing medium was spun down at 200 rpm for 3 minutes, and the supernatant was aspirated. The cell pellet was re-suspended with a new culture medium at an appropriate cell concentration and used for the next suspension culture."

(15) The IWR1+LY suspension culture system requires passage every 3-5 days. Is there still spontaneous differentiation if the hiPSC aggregate grows too big?  

Thank you for this insightful question.

Yes. The size of hiPSC aggregates is critical in maintaining self-renewal in our method as previous studies showed. Stirring speed is a key to make the proper size of hiPSC aggregates in suspension culture. Also, the culture period between passages is another key not to exceed the proper size of hiPSC aggregates. Thus, we keep stirring speed at 90 rpm (135 rpm for bioreactor conditions) basically and passaging every 3 - 5 days in suspension culture conditions.

(16) Several previous studies have described the development of hiPSC suspension culture system using hydrogel encapsulation to provide biophysical modulation (reviewed in PMID: 32117992). In comparison, it seems that the IWR1+LY suspension system described here does not require ECM addition which further simplifies the culture system for iPSC. It would be good to add more discussion on this topic in the manuscript, such as the potential role of the E-cadherin in mediating this effect - as RNAseq results indicated that CDH1 was upregulated in the IWR1+LY condition).  

Thank you for this valuable suggestion. We have added more discussion on this topic in the Discussion section as below.

"Thus, our findings show that suspension culture conditions with Wnt and PKCβ inhibitors (IWRLY suspension conditions) can precisely control cell conditions and are comparable to conventional adhesion cultures regarding cellular function and proliferation. Many previous 3D culture methods intended for mass expansion used hydrogel-based encapsulation or microcarrier-based methods to provide scaffolds and biophysical modulation (Chan et al, 2020). These methods are useful in that they enable mass culture while maintaining scaffold dependence. However, the need for special materials and equipment and the labor and cost involved are concerns toward industrial mass culture. On the other hand, our IWRLY suspension conditions do not require special materials such as hydrogels, microcarriers, or dialysis bags, and have the advantage that common bioreactors can be used. "

"On the other hand, it is interesting to see whether and how the properties of hiPSCs cultured in IWRLY suspension culture conditions are altered from the adherent conditions. Our transcriptome results in comparison to adherent conditions show that gene expression associated with cell-to-cell attachment, including E-cadherin (CDH1), is more activated. This may be due to the status that these hiPSCs are more dependent on cell-to-cell adhesion where there is no exogenous cell-to-substrate attachment in the three-dimensional culture. Previous studies have shown that cell-to-cell adhesion by E-cadherin positively regulates the survival, proliferation, and self-renewal of human pluripotent stem cells (Aban et al, 2021; Li et al, 2012; Ohgushi et al, 2010). Furthermore, studies have shown that human pluripotent stem cells can be cultured using an artificial substrate consisting of recombinant E-cadherin protein alone without any ECM proteins (Nagaoka et al, 2010). Also, cell-to-cell adhesion through gap junctions regulates the survival and proliferation of human pluripotent stem cells (Wong et al, 2006; Wong et al, 2004). These findings raise the possibility that the cell-to-cell adhesion, such as E-cadherin and gap junctions, are compensatory activated and support hiPSC self-renewal in situations where there are no exogenous ECM components and its downstream integrin and focal adhesion signals are not forcedly activated in suspension culture conditions. It will be interesting to elucidate these molecular mechanisms related to E-cadherin in the hiPSC survival and self-renewal in IWRLY suspension conditions in the future."

Reviewer #3 (Recommendations For The Authors):  

(1) I am a bit confused about the passage of adherent cultures. The authors claim that they used EDTA for passaging and plated cells at a density of 2500 cells/cm2. My understanding is that EDTA is typically used for clump passaging rather than single-cell passaging.  

Sorry about this confusion. We routinely use an automatic cell counter (model R1, Olympus) which can even count small clumpy cells accurately. Thus, we show the cell numbers in the passaging of adherent hiPSCs.  

(2) Figure 2D- The authors have not directly compared IWR-1-endo with IWR-1-endo+Go6983 for the expression of T and SOX17, a simultaneous comparison would be an interesting data.  

As recommended, we have added the data that directly compared IWR-1-endo with IWR-1endo+Go6983 for the expression of T and SOX17 in Figure 2D. The addition of IWR-1-endo alone decreased the expression of T and SOX17, but not PAX6, which were similar to the data in Figure 2C.

(3) Oxygen levels play a crucial role in pluripotency maintenance. Could the authors please specify the oxygen levels used for culturing cells in suspension?  

Sorry for not mentioning about oxygen levels in this study. We basically use normal oxygen levels (i.e., 21% O2) in suspension culture conditions. We have explained this in the Materials and Methods section.

(4) Figure supplement 1 (G and H): In the images, it is difficult to determine whether the green (PAX6 and SOX17) overlaps with tdT tomato. For better visualization, I suggest that the authors provide separate images for the green and red colors, as well as an overlay.  

Sorry for these unclear images. We have provided separate images for the green and red colors, as well as an overlay in Figure 1- figure supplement 1 G and H.

(5) The authors have only compared quantitatively the expression of TRA-1-60 for most of the figures. I suggest that the authors quantitatively measure the expression of other markers of undifferentiated stem cells, such as NANOG, OCT4, SSEA4, TRA-1-81, etc.  

We have added the quantitative data of the expression of markers of undifferentiated hiPSCs including NANOG, OCT4, SSEA4, and TRA-1-60 on 5 different hiPSC lines in Figure 3 C-E.

(6) In Figure 2D, the authors have tested various small molecules but the rationale behind testing those molecules is missing in the text.  

These molecules are chosen as putatively affecting neuroectodermal induction from the pluripotent state.

We have added the rationale with appropriate references in the Results section as below.

"We have chosen these candidate molecules based on previous studies related to signaling pathways or epigenetic regulations in neuroectodermal development (reviewed in (Giacoman-Lozano et al, 2022; Imaizumi & Okano, 2021; Sasai et al, 2021; Stern, 2024) ) or in pluripotency safeguards (reviewed in (Hackett & Surani, 2014; Li & Belmonte, 2017; Takahashi & Yamanaka, 2016; Yagi et al, 2017)) (Figure 2A; listed in Supplementary Table 1). "

(7) In the beginning authors used Go6983 but later they switched to LY333531, the reasoning behind the switch is not explained well.  

To explain the reasons for switching to LY333531 from Go6983 clearly, we reorganized the order of results and figures. In short, we found that the suppression of PAX6 expression in hiPSCs cultured in suspension conditions was observed with many PKC inhibitors, all of which possessed PKCβ inhibition activity (Figure 2—figure supplement 2B-D). Also, elevated expression of PKCβ in suspension-cultured hiPSCs could affect the spontaneous differentiation (Figure 3—figure supplement 1A-C). To further explore the possibility that the inhibition of PKCβ is critical for the maintenance of self-renewal of hiPSCs in the suspension culture, we evaluated the effect of LY333531, a PKCβ specific inhibitor. The maintenance of suspension-cultured hiPSCs is specifically facilitated by the combination of PKCβ and Wnt signaling inhibition (Figure 3A and B; Figure 2—figure supplement 1). Last, we performed longterm culture for 10 passages in suspension conditions and compared hiPSC growth in the presence of LY333531 or Go6983. LY333531 was superior in the proliferation rate and maintaining OCT4 protein expression in the long-term culture (Figure 4). Thus, we used IWR-1-endo and LY333531 for the rest of this study.

(8) I suggest the authors measure cell death after the treatment with LY+IWR-1-endo.  

Thank you for this valuable suggestion. We have measured cell death after the treatment with LY+IWR1-endo and found that the chemical combination had no or little effects on the cell death. We have added data in Figure 3—figure supplement 2 and the description in the Results section as below. "We also examined whether the combination of PKCb and Wnt signaling inhibition affects the cell survival in suspension conditions. In this experiment, we used another PKC inhibitor, Staurosporine (Omura et al, 1977), which has a strong cytotoxic effect as a positive control of cell death in suspension conditions. The addition of IWR-1-endo and LY333531 for 10 days had no effects on the apoptosis while the addition of Staurosporine for 2 hours induced Annexin-V-positive apoptotic cells  (Figure 3—figure supplement 2). These results indicate that the combination of PKCb and Wnt signaling inhibition has no or little effects on the cell survival in suspension conditions."

(9) The authors have performed reprogramming using episomal vectors and using Sendai viruses. In both the protocols authors have added small molecules at different time points, for episomal vector protocol at day 3 and Sendai virus protocol at day 23. Why is this different?  

Thank you for this insightful question. We intended that these differences should be reflected in the degree of the expression from these reprogramming vectors. The expression of reprogramming factors from these vectors should suppress the spontaneous differentiation in reprogramming cells. Sendai viral vectors should last longer than episomal plasmid vectors. Thus, we thought that adding these chemical inhibitors for episomal plasmid vector conditions from the early phase of reprogramming and for Sendai viral vector conditions from the late phase of reprogramming. For future perspectives, we might further need to optimize the timing of adding these molecules.

(10) The protocol for three germ layer differentiation using a specific differentiation medium requires further elaboration. For instance, the authors mentioned that suspension cultures were transferred to differentiation media but did not emphasize the cell number and culture conditions before moving the cultures to the differentiation media.  

Sorry for this unclear description. We have added the explanation on the cell number and culture conditions before moving the cultures to the differentiation media in the Materials and Methods section as below.

"As in the maintenance conditions, 4 × 105 hiPSC were seeded in one well of a low-attachment 6-well plate with 4 mL of StemFit AK02N medium supplemented with 10 µM Y-27632. This plate was placed onto the plate shaker in the CO2 incubator. Next day, the medium was changed to the germ layer specific differentiation medium."

Author response:

Joint Public Reviews:

Here, the authors compare how different operationalizations of adverse childhood experience exposure related to patterns of skin conductance response during a fear conditioning task. They use a large dataset to definitively understand a phenomenon that, to date, has been addressed using a range of different definitions and methods, typically with insufficient statistical power. Specifically, the authors compared the following operationalizations: dichotomization of the sample into "exposed" and "non-exposed" categories, cumulative adversity exposure, specificity of adversity exposure, and dimensional (threat versus deprivation) adversity exposure. The paper is thoughtfully framed and provides clear descriptions and rationale for procedures, as well as package version information and code. The authors' overall aim of translating theoretical models of adversity into statistical models, and comparing the explanatory power of each model, respectively, is an important and helpful addition to the literature. However, the analysis would be strengthened by employing more sophisticated modelling techniques that account for between-subjects covariates and the presentation of the data needs to be streamlined to make it clearer for the broad audience for which it is intended.

Strengths

Several outstanding strengths of this paper are the large sample size and its primary aim of statistically comparing leading theoretical models of adversity exposure in the context of skin conductance response. This paper also helpfully reports Cohen's d effect sizes, which aid in interpreting the magnitude of the findings. The methods and results are generally thorough.

Weaknesses

Weakness 1: The largest concern is that the paper primarily relies on ANOVAs and pairwise testing for its analyses and does not include between-subjects covariates. Employing mixedeffects models instead of ANOVAs would allow more sophisticated control over sources of random variance in the sample (especially important for samples from multi-site studies such as the present study), and further allow the inclusion of potentially relevant between-subjects covariates such as age (e.g. Eisenstein et al., 1990) and gender identity or sex assigned at birth (e.g. Kopacz II & Smith, 1971) (perhaps especially relevant due to possible to gender or sex-related differences in ACE exposure; e.g. Kendler et al., 2001). Also, proxies for socioeconomic status (e.g. income, education) can be linked with ACE exposure (e.g. Maholmes & King, 2012) and warrant consideration as covariates, especially if they differ across adversity-exposed and unexposed groups. 

We appreciate the reviewer's suggestion and recognize the value of using (more) sophisticated statistical methods. However, we think that considerations which methods to employ should not only be guided by perceived complexity and think that the chosen ANOVA -based approach provides reliable and valid data. In our revision, we address the reviewer's suggestion by demonstrating that employing mixed models leaves the reported results unchanged (a). We would also like to refer the reviewer to the robustness analyses provided in the initial supplementary material (b).

a) Re-running analyses using mixed models

Based on the reviewers' suggestion, we repeated our main analyses (association between exposure to childhood adversity and SCRs, arousal, valence, and contingency ratings during fear acquisition and generalization) using linear mixed models, including age, sex, educational attainment, and childhood adversity as fixed effects, and site as a random effect. These analyses produced results similar to those in our manuscript, demonstrating a significant effect of childhood adversity on SCRs, as assessed by CS discrimination during both acquisition training and the generalization phase, and on general reactivity, but not on linear deviation scores (LDS). For the different rating types, we did not observe any significant effects of childhood adversity.

We would prefer to retain our main analyses as they are and report the linear mixed model results as additional results in the supplement. However, if the reviewer and editor have strong preferences otherwise, we are open to presenting the mixed models in the main manuscript and moving our previous analyses to the supplement.

We added the following paragraph to the main manuscript (page 25-26):

“At the request of a reviewer, we repeated our main analyses by using linear mixed models including age, sex, school degree (i.e., to approximate socioeconomic status), and exposure to childhood adversity as mixed effects as well as site as random effect. These analyses yielded comparable results demonstrating a significant effect of childhood adversity on CS discrimination during acquisition training and the generalization phase as well as on general reactivity, but not on the generalization gradients in SCRs (see Supplementary Table 2 A). Consistent with the results of the main analyses reported in our manuscript, we did not observe any significant effects of childhood adversity on the different types of ratings when using mixed models (see Supplementary Table 2 B-D). Some of the mixed model analyses showed significantly lower CS discrimination during acquisition training and generalization, and lower general reactivity in males compared to females (see Supplementary Table 2 for details).”

b) Additional robustness tests for the main analyses (already provided in the initial submission as supplementary material)

We would also like to refer the reviewer to the robustness analyses in the initial supplement to account for possible site effects. Adding site to the analyses affected the pvalue in only one instance: entering site as covariate in analyses of CS discrimination during acquisition training attenuated the p-value of the ACQ exposure effect from p = 0.020 to p = 0.089.

Further robustness checks involved repeating our main analyses while excluding (a) physiological non-responders (participants with only SCRs = 0) and (b) extreme outliers (data points ± 3 SDs from the mean) to ensure generalizable results. These repetitions of the analyses did not lead to any changes in the results.

We did not include age in our primary analyses due to the homogeneity of our sample and the lack of related hypotheses. Additionally, socio-economic status was assessed only crudely via the highest education level attained, rendering it of limited use.

Weakness 2: On a related methodological note, the authors mention that scores representing threat and deprivation were not problematically collinear due to VIFs being <10; however, some sources indicate that VIFs should be <5 (e.g. Akinwande et al., 2015).

We thank the reviewer for bringing different cut-offs to our attention. We have revised this section to highlight the arbitrary nature of their interpretation (page 33):

“Within the dimensional model framework, the issue of multicollinearity among predictors (i.e., different childhood adversity types) is frequently discussed (McLaughlin et al., 2021; Smith & Pollak, 2021). If we apply the rule of thumb of a variance inflation factor (VIF) > 10, which is often used in the literature to indicate concerning multicollinearity (e.g., Hair, Anderson, Tatham, & Black, 1995; Mason, Gunst, & Hess, 1989; Neter, Wasserman, & Kutner, 1989), we can assume that that multicollinearity was not a concern in our study (abuse: VIF = 8.64; neglect: VIF = 7.93). However, some authors state that VIFs should not exceed a value of 5 (e.g., Akinwande, Dikko, and Samson (2015)), while others suggest that these rules of thumb are rather arbitrary (O’brien, 2007).”

Weakness 3: Additionally, the paper reports that higher trait anxiety and depression symptoms were observed in individuals exposed to ACEs, but it would be helpful to report whether patterns of SCR were in turn associated with these symptom measures and whether the different operationalizations of ACE exposure displayed differential associations with symptoms.

We thank the reviewer for highlighting these relevant points. We have included additional analyses in the supplementary material in response to this comment. Figures and the corresponding text are also copied below for your convenience.

We added the following paragraphs to the main manuscript: Methods (page 21):

“Analyses of trait anxiety and depression symptoms

To further characterize our sample, we compared individuals being unexposed compared to exposed to childhood adversity on trait anxiety and depression scores by using Welch tests due to unequal variances.

On the request of a reviewer, we additionally investigated the association of childhood adversity as operationalized by the different models used in our explanatory analyses (i.e., cumulative risk, specificity, and dimensional model) and trait anxiety as well as depression scores (see Supplementary Figure 7). By using STAI-T and ADS-K scores as independent variable, we calculated a) a comparison of conditioned responding of the four severity groups (i.e., no, low, moderate, severe exposure to childhood adversity) using one-way ANVOAs and the association with the number of sub-scales exceeding an at least moderate cut-off in simple linear regression models for the implementation of the cumulative risk model, and b) the association with the CTQ abuse and neglect composite scores in separate linear regression models for the implementation of the specificity/dimensional models. On request of the reviewer, we also calculated the Pearson correlation between trait anxiety (i.e., STAI-T scores), depression scores (i.e., ADS-K scores) and conditioned responding in SCRs (see Supplementary Table 8).”

Results (page 38):

“Analyses of trait anxiety and depression symptoms

As expected, participants exposed to childhood adversity reported significantly higher trait anxiety and depression levels than unexposed participants (all p’s < 0.001; see Table 1 and Supplementary Figure 6). This pattern remained unchanged when childhood adversity was operationalized differently - following the cumulative risk approach, the specificity, and dimensional model (see methods). These additional analyses all indicated a significant positive relationship between exposure to childhood adversity and trait anxiety as well as depression scores irrespective of the specific operationalization of “exposure” (see Supplementary Figure 7).

CS discrimination during acquisition training and the generalization phase, generalization gradients, and general reactivity in SCRs were unrelated to trait anxiety and depression scores in this sample with the exception of a significant association between depression scores and CS discrimination during fear acquisition training (see Supplementary Table 8). More precisely, a very small but significant negative correlation was observed indicating that high levels of depression were associated with reduced levels of CS discrimination (r = -0.057, p =0.033). The correlation between trait anxiety levels and CS discrimination during fear acquisition training was not statistically significant but on a descriptive level, high anxiety scores were also linked to lower CS discrimination scores (r = -0.05, p = 0.06) although we highlight that this should not be overinterpreted in light of the large sample. However, both correlations (i.e., CS-discrimination during fear acquisition training and trait anxiety as well as depression, respectively) did not statistically differ from each other (z = 0.303, p = 0.762, Dunn & Clark, 1969). Interestingly, and consistent with our results showing that the relationship between childhood adversity and CS discrimination was mainly driven by significantly lower CS+ responses in exposed individuals, trait anxiety and depression scores were significantly associated with SCRs to the CS+, but not to the CS- during acquisition training (see Supplementary Table 8).”

Weakness 4: Given the paper's framing of SCR as a potential mechanistic link between adversity and mental health problems, reporting these associations would be a helpful addition. These results could also have implications for the resilience interpretation in the discussion (lines 481-485), which is a particularly important and interesting interpretation.

We have added a paragraph on this to the discussion (page 41):

“Interestingly, in our study, trait anxiety and depression scores were mostly unrelated to SCRs, defined by CS discrimination and generalization gradients based on SCRs as well as general SCR reactivity, with the exception of a significant - albeit minute - relationship between CS discrimination during acquisition training and depression scores (see above). Although reported associations in the literature are heterogeneous (Lonsdorf et al., 2017), we may speculate that they may be mediated by childhood adversity. We conducted additional mediation analyses (data not shown) which, however, did not support this hypothesis. As the potential links between reduced CS discrimination in individuals exposed to childhood adversity and the developmental trajectories of psychopathological symptoms are still not fully understood, future work should investigate these further in - ideally - prospective studies.”

Weakness 5: Given that the manuscript criticizes the different operationalizations of childhood adversity, there should be greater justification of the rationale for choosing the model for the main analyses. Why not the 'cumulative risk' or 'specificity' model? Related to this, there should also be a stronger justification for selecting the 'moderate' approach for the main analysis. Why choose to cut off at moderate? Why not severe, or low? Related to this, why did they choose to cut off at all? Surely one could address this with the continuous variable, as they criticize cut-offs in Table 2.

We thank the reviewers and editors for bringing to our attention that our reasoning for choosing the main model was not clear. As outlined in the manuscript, we chose the approach for the main analyses from the literature as a recent review on this topic (Ruge et al., 2023) has shown the moderate CTQ cut-off to be the most abundantly employed in the field of research on associations between childhood adversity and threat learning. We have made this rationale more explicit in our revised manuscript (page 15/21):

“Operationalization of "exposure"

We implemented different approaches to operationalize exposure to childhood adversity in the main analyses and exploratory analyses (see Table 2). In the main analyses, we followed the approach most commonly employed in the field of research on childhood adversity and threat learning - using the moderate exposure cut-off of the CTQ (for a recent review see Ruge et al. (2024)). In addition, the heterogeneous operationalizations of classifying individuals into exposed and unexposed to childhood adversity in the literature (Koppold, Kastrinogiannis, Kuhn, & Lonsdorf, 2023; Ruge et al., 2024) hampers comparison across studies and hence cumulative knowledge generation. Therefore, we also provide exploratory analyses (see below) in which we employ different operationalizations of childhood adversity exposure.”

“Exploratory analyses

Additionally, the different ways of classifying individuals as exposed or unexposed to childhood adversity in the literature (Koppold et al., 2023; for discussion see Ruge et al., 2024) hinder comparison across studies and hence cumulative knowledge generation. Therefore, we also conducted exploratory analyses using different approaches to operationalize exposure to childhood adversity (see Table 2 for details).”

Furthermore, as correctly noted, we fully agree that employing the moderate cut-off (or any cut-off in fact) is in principle an arbitrary decision - despite being guided by and derived from the literature in the field. However, we would like to draw the reviewers’ attention to Figure 5 in the initial submission (please see also below): Although the differences in SCR between severity groups were not significant, the overall pattern suggests at a descriptive level that the decline in CS discrimination, LDS and general reactivity in SCR occurs mainly when childhood adversity exceeds a moderate level. Thus, while we used the moderate cut-off as it was recently shown to be the most widely used approach in the literature (see Ruge et al., 2023), our exploratory analyses also seem to suggest on a descriptive level, that this cut-off may indeed “make sense”. We also refer to this in the results section (page 31-32) and discussion (page 43-44):

Results:

“However, on a descriptive level (see Figure 5), it seems that indeed exposure to at least a moderate cut-off level may induce behavioral and physiological changes (see main analysis, Bernstein & Fink, 1998). This might suggest that the cut-off for exposure commonly applied in the literature (see Ruge et al., 2024) may indeed represent a reasonable approach.”

Discussion:

“It is noteworthy, however, that this cut-off appears to map rather well onto psychophysiological response patterns observed here (see Figure 5). More precisely, our exploratory results of applying different exposure cut-offs (low, moderate, severe, no exposure) seem to indicate that indeed a moderate exposure level is “required” for the manifestation of physiological differences, suggesting that childhood adversity exposure may not have a linear or cumulative effect.”

Weakness 6: In the Introduction, the authors predict less discrimination between signals of danger (CS+) and safety (CS-) in trauma-exposed individuals driven by reduced responses to the CS+. Given the potential impact of their findings for a larger audience, it is important to give greater theoretical context as to why CS discrimination is relevant here, and especially what a reduction in response specifically to danger cues would mean (e.g. in comparison to anxiety, where safety learning is impacted).

We thank the reviewer for highlighting that this was not sufficiently clear. We revised the paragraph in the introduction as follows (page 7-8):

“Fear acquisition as well as extinction are considered as experimental models of the development and exposure-based treatment of anxiety- and stress-related disorders. Fear generalization is in principle adaptive in ensuring survival (“better safe than sorry”), but broad overgeneralization can become burdensome for patients. Accordingly, maintaining the ability to distinguish between signals of danger (i.e., CS+) and safety (i.e., CS-) under aversive circumstances is crucial, as it is assumed to be beneficial for healthy functioning (Hölzel et al., 2016) and predicts resilience to life stress (Craske et al., 2012), while reduced discrimination between the CS+ and CS- has been linked to pathological anxiety (Duits et al., 2015; Lissek et al., 2005): Meta-analyses suggest that patients suffering from anxiety- and stress-related disorders show enhanced responding to the safe CS- during fear acquisition (Duits et al., 2015). During extinction, patients exhibit stronger defensive responses to the CS+ and a trend toward increased discrimination between the CS+ and CS- compared to controls, which may indicate delayed and/or reduced extinction (Duits et al., 2015). Furthermore, meta-analytic evidence also suggests stronger generalization to cues similar to the CS+ in patients and more linear generalization gradients (Cooper, van Dis, et al., 2022; Dymond, Dunsmoor, Vervliet, Roche, & Hermans, 2015; Fraunfelter, Gerdes, & Alpers, 2022). Hence, aberrant fear acquisition, extinction, and generalization processes may provide clear and potentially modifiable targets for intervention and prevention programs for stress-related psychopathology (McLaughlin & Sheridan, 2016).”

Recommendations for the authors:

Abstract:

Comment 1:

(a) It does not succinctly describe the background rationale well (i.e. it tries to say too much). It should be streamlined. There is a lot of 'jargon', which muddies the results, and too many concepts are introduced at each part and assume knowledge from the reader. 

We thank the reviewer for providing constructive guidance for revisions. We have revised our abstract according to these suggestions.

(b) Multiple terms for childhood trauma are used: ACEs, early adversity, childhood trauma, and childhood maltreatment. Choose one term and stick to it to enhance clarity. Why not just use childhood adversity, as in the title? Related to this, the use of ACEs sets up an expectation that ACE questionnaire was used, so readers are then surprised to find they used the childhood trauma questionnaire.

We thank the reviewer for bringing this to our attention. As suggested by the reviewer, we use the term “childhood adversity” in our revised manuscript.

Introduction:

Comment 2:

The phrasing seems to 'exaggerate' the trauma problem and is too broad in the first paragraph - e.g., "two-thirds of people experience one or more traumatic events..." It is important to clarify that not all of these people will go on to develop behavioral, somatic, and psychopathological conditions. Could break this down more into how many people have low, moderate, or severe for clarity, as 1 childhood adversity is different to 5+, and the type.

We thank the reviewer for bringing this to our attention and have revised the first paragraph accordingly (page 6). Please note, however, that in the literature typically a specific cut-off (e.g. moderate) is used and the number of individuals that would meet different cut-offs (e.g., low and high) are not specifically reported.

“Exposure to childhood adversity is rather common, with nearly two thirds of individuals experiencing one or more traumatic events prior to their 18th birthday (McLaughlin et al., 2013). While not all trauma-exposed individuals develop psychopathological conditions, there is some evidence of a dose-response relationship (Danese et al., 2009; Smith & Pollak, 2021; Young et al., 2019). As this potential relationship is not yet fully clear, understanding the mechanisms by which childhood adversity becomes biologically embedded and contributes to the pathogenesis of stress-related somatic and mental disorders is central to the development of targeted intervention and prevention programmes.”

Comment 3:

The published cut-offs for exposed/unexposed should be indicated here.

We have included the published cut-offs as suggested (page 10):

We operationalize childhood adversity exposure through different approaches: Our main analyses employ the approach adopted by most publications in the field (see Ruge et al., 2024 for a review) - dichotomization of the sample into exposed vs. unexposed based on published cut-offs for the Childhood Trauma Questionnaire [CTQ; Bernstein et al. (2003); Wingenfeld et al. (2010)]. Individuals were classified as exposed to childhood adversity if at least one CTQ subscale met the published cut-off (Bernstein & Fink, 1998; Häuser, Schmutzer, & Glaesmer, 2011) for at least moderate exposure (i.e., emotional abuse  13, physical abuse  10, sexual abuse  8, emotional neglect  15, physical neglect  10).

Comment 4:

Please check for overly complex sentences, and reduce the complexity. For example: "In addition, we provide exploratory analyses that attempt to translate dominant (verbal) theoretical accounts (McLaughlin et al., 2021; Pollak & Smith, 2021) on the impact of exposure to ACEs into statistical tests while acknowledging that such a translation is not unambiguous and these exploratory analyses should be considered as showcasing a set of plausible solutions."

We have revised this section and carefully proofread our manuscript by paying attention to this (page 10):

“In addition, we provide exploratory analyses that attempt to translate dominant (verbal) theoretical accounts (McLaughlin et al., 2021; Pollak & Smith, 2021) on the impact of exposure to childhood adversity into statistical tests. At the same time, we acknowledge that such a translation is not unambiguous and these exploratory analyses should be considered as showcasing a set of plausible solutions”

Here is another example of reducing the complexity of our sentences (page 6):

“Learning is a core mechanism through which environmental inputs shape emotional and cognitive processes and ultimately behavior. Thus, learning mechanisms are key candidates potentially underlying the biological embedding of exposure to childhood adversity and their impact on development and risk for psychopathology (McLaughlin & Sheridan, 2016).”

Methods:

Comment 5:

Is this study part of a larger project? These outcomes were probably not the primary outcomes of this multicenter project. The readers need to understand how this (crosssectional?) analysis was nested in this larger trial.

We thank the reviewers and editor for bringing to our attention that this was not sufficiently clear. Thus far, we included the information that we used the participants recruited for large multicentric study in the main manuscript, but point to the inclusion of more information in the supplement (page 11):

“In total, 1678 healthy participants (age_M_ = 25.26 years, age_SD_ = 5.58 years, female = 60.10%, male = 39.30%) were recruited in a multi-centric study at the Universities of Münster, Würzburg, and Hamburg, Germany (SFB TRR58). Data from parts of the Würzburg sample have been reported previously (Herzog et al., 2021; Imholze et al., 2023; Schiele, Reinhard, et al., 2016; Schiele, Ziegler, et al., 2016; Stegmann et al., 2019). These previous reports, also those focusing on experimental fear conditioning (Schiele, Reinhard, et al., 2016; Stegmann et al., 2019), addressed, however, research questions different from the ones investigated here (see also Supplementary Material for details).”

Moreover, we have included additional information on the larger trial in our revised supplement (page 2):

“Participants of this study were recruited in a multi-centric collaborative research center “Fear, anxiety, anxiety disorders” joining forces between the Universities of Hamburg,

Würzburg, and Münster, Germany (SFB TRR58). During the second funding period of (20132016), all three sites recruited a large sample (N ~500) in the context of the Z project. All participants underwent the cross-sectional experimental paradigm reported here and were additionally extensively characterized to allow specific subprojects to recruit target subpopulations serving different aims with a focus on molecular genetic, epigenetic, or other research questions (see Herzog et al. (2021); Imholze et al. (2023); Schiele, Reinhard, et al. (2016); Schiele, Ziegler, et al. (2016); Stegmann et al. (2019)). The question on the association of exposure to childhood adversity and recent adversity was part of the primary research question of one subproject led by the senior author of this work (B07, TBL) and was hence a research question of primary interest also for this multicentric project.”

Comment 6:

Table 1 does not include percentages (a reader must calculate them: for example, 15% exposed?). These numbers belong in the results (i.e., it is confusing to read about the exposed/non-exposed before we know how it has been calculated).

We have added the percentages as suggested and have included information on how exposed and unexposed was calculated as a table caption. We have considered moving the table to the results section but find it more suitable here. 

Comment 7:

A procedure figure could be useful.

We thank the reviewer for this advice and have included a procedure figure in the supplementary material.

Comment 8:

Physiological data recordings and processing paragraph: The reasoning as to why the authors chose log transformation over square root transformation, or an approach that does not require transformation is not clear.

We thank the reviewer for notifying us that we did not make this point clear enough. We opted for a log-transformation and range-correction of the SCR data because we use these transformations consistently in our laboratory (e.g., Ehlers et al., 2020; Kuhn et al., 2016; Scharfenort & Lonsdorf, 2016; Sjouwerman et al., 2015; Sjouwerman et al. 2020). In addition, log-transformed and range-corrected data are assumed to be closer to a normal distribution, to have a lower error variance resulting in larger effect sizes (Lykken & Venables, 1971; Lykken, 1972; Sjouwerman et al., 2022), and appear to have - at least descriptively - higher reliability compared to raw data (Klingelhöfer-Jens et al., 2022). We added a sentence on this to the methods section (page 14):

Note that previous work using this sample (Schiele, Reinhard, et al., 2016; Stegmann et al., 2019) had used square-root transformations but we decided to employ a log-transformation and range-correction (i.e., dividing each SCR by the maximum SCR per participant). We used log-transformation and range-correction for SCR data because these transformations are standard practice in our laboratory and we strive for methodological consistency across different projects (e.g., Ehlers, Nold, Kuhn, Klingelhöfer-Jens, & Lonsdorf, 2020; Kuhn, Mertens, & Lonsdorf, 2016; Scharfenort, Menz, & Lonsdorf, 2016; Sjouwerman & Lonsdorf, 2020; Sjouwerman, Niehaus, & Lonsdorf, 2015). Additionally, log-transformed and rangecorrected data are generally assumed to approximate a normal distribution more closely and exhibit lower error variance, which leads to larger effect sizes (Lykken, 1972; Lykken & Venables, 1971; Sjouwerman, Illius, Kuhn, & Lonsdorf, 2022). Additionally, on a descriptive level, this combination of transformations appear to offer greater reliability compared to using raw data alone (Klingelhöfer-Jens, Ehlers, Kuhn, Keyaniyan, & Lonsdorf, 2022).

Ehlers, M. R., Nold, J., Kuhn, M., Klingelhöfer-Jens, M., & Lonsdorf, T. B. (2020). Revisiting potential associations between brain morphology, fear acquisition and extinction through new data and a literature review. Scientific Reports, 10(1), 19894. https://doi.org/10.1038/s41598-020-76683-1

Kuhn, M., Mertens, G., & Lonsdorf, T. B. (2016). State anxiety modulates the return of fear. International Journal of Psychophysiology: Official Journal of the International Organization of Psychophysiology, 110, 194–199. https://doi.org/10.1016/j.ijpsycho.2016.08.001

Scharfenort, R., & Lonsdorf, T. B. (2016). Neural correlates of and processes underlying generalized and differential return of fear. Social Cognitive and Affective Neuroscience, 11(4), 612–620. https://doi.org/10.1093/scan/nsv142

Sjouwerman, R., Niehaus, J., & Lonsdorf, T. B. (2015). Contextual Change After Fear Acquisition Affects Conditioned Responding and the Time Course of Extinction Learning—Implications for Renewal Research. Frontiers in Behavioral Neuroscience, 9. https://doi.org/10.3389/fnbeh.2015.00337

Sjouwerman, R., Scharfenort, R., & Lonsdorf, T. B. (2020). Individual differences in fear acquisition: Multivariate analyses of different emotional negativity scales, physiological responding, subjective measures, and neural activation. Scientific Reports, 10(1), 15283. https://doi.org/10.1038/s41598-020-72007-5

Comment 9:

There are 24 lines of text of R packages. I do not think this is necessary for the manuscript document and could be moved to the Supplement.

We thank the reviewer for this comment and understand that it may take a considerable amount of space to list all the references of the R packages. However, we think it is important to prominently credit the respective authors of the R packages. Yet, if this is an important concern of the reviewer and editor, we will reconsider this point.

Comment 10:

It is not clear why the authors chose to analyze summary scores across trials rather than including a time factor for the acquisition phase.

We would like to thank the reviewer for highlighting that the factor time may be interesting as well. However, we think that in our case the time factor is less interesting, as the acquisition effect itself is rather strong. Nevertheless, we have included a figure in the supplement that shows the time course of the SCR by displaying trial-by-trial data across the acquisition and generalization phase for transparency. This figure (Supplementary figure 4) shows that the trajectories appear to barely differ between individuals who were unexposed vs. exposed to moderate childhood adversity. Hence, we think that the analysis approach we have chosen is unlikely to overshadow central time-depending effects. However, if the reviewer and editor has strong feelings about this point, we will consider integrating additional analyses including the time factor in the supplement.

Results:

Comment 11:

The caption of Figure 3 does not match the figure. Please check this.

We thank the reviewers and editor for attentive reading and have revised this part.

References:

Comment 12:

The Ruge et al paper that is cited many times throughout does not have a valid DOI in the References section. Additionally, the author list on the preprint server is substantially different from that listed in the manuscript. Please correct this reference.

We thank the reviewers and editor for attentive reading and have corrected this reference. The provided doi was functioning at our end and we hope that this now also applies to the reviewers.

Author response:

Reviewer #1:

Response to Public Review

We thank the reviewer for taking the time to carefully read our paper and to provide helpful comments and suggestions, most of which we have incorporated in our revised manuscript.  One of this reviewer’s (and reviewer #2’s) main concerns was that the confocal images provided in some cases did not appear to reflect the quantitative data in the bar graphs.  These images were provided only for illustrative purposes, to give the reader a sense of what the primary data look like. The reviewer may not have appreciated that the quantitative data reflect counts of RNA smFISH signals (dots) in hundreds of cells collected through z-stacks comprising multiple optical sections in multiple flies for each condition  For example, in P1a control condition (in Figure 2A), we have analyzed 135 neurons from 8 individuals. There, the number of z-planes ranged from 3 to 8 per hemisphere. It is generally not possible to find a single confocal section that encompasses quantitatively the statistics that are presented in the graphs. Presenting the data as an MIP (Maximum Intensity Projection, i.e., collapsed z-stack) in a single panel would generate an image that is too cluttered to see any detail.  We have now included, for the reader’s benefit, additional example confocal sections in both a z-stack and from the opposite hemisphere, in Supplemental Figure S4D. We have also inserted clarifying statements in the text on p. 7 (lines 154-156).

Another suggestion from Reviewer #1 is that "it would be more informative to separate in the quantification between the GAL4-expressing neurons and the non-expressing ones" based on the presented pictures where more non-P1a neurons (that the reviewer speculates may be pC1-type neurons) are activated by a male-male encounter than by a male-female encounter, while the P1a-positive neurons seem to be more responsive during courtship behavior. In this paper, we were not looking at pC1 neurons and did not try to answer which neuronal population(s) outside of the P1a population is/are responsible for aggression and/or courtship. Rather, we focused on P1a neurons and addressed whether P1a neurons that induce both aggression and courtship behavior when they are artificially activated (Hoopfer et al. 2015) are also naturally activated during spontaneous performance of these two social behaviors. However, this result did not exclude the possibility that P1a neurons were inactive during naturalistic courtship or aggression. Our data in the current manuscript provide further experimental evidence in support of the idea that P1a neurons as a population play a role in both of these behaviors. Moreover, we provided data identifying P1a neurons activated only during aggression or during courtship (or both). However this does not exclude that pC1 or other neighboring populations are activated during aggression as well (See also the response to 'Recommendations For The Authors' and text lines 151-154).

In Figure 3, we used opto-HI-FISH to identify candidate downstream targets (direct or indirect) of P1a neurons. We used 50 Hz Chrimson stimulation to activate P1a neurons to induce expression of Hr38 and identified Kenyon cells in the mushroom body (MB) and PAM neurons (as well as pCd neurons) as potential downstream targets of P1a cells. In Figure 3 – supplement we performed calcium imaging of KCs and PAM neurons in response to P1a optogenetic stimulation to confirm independently our results from the Hr38 labeling experiments. That control was the purpose of that supplemental experiment.

Based on those imaging data, the reviewer asked the further question of which [natural] behavioral context induces Hr38 expression in these populations (i.e., mating or aggression). This question is reasonable because our calcium imaging data (Figure 3-supplement) showed that both Kenyon cells and PAM neurons are active only during photo-stimulation of P1a neurons.  Our previous behavioral studies (Inagaki et al., 2014; Hoopfer et al., 2015) showed that 50 Hz photo-stimulation of P1a neurons in freely moving flies induced unilateral wing extension during stimulation, while aggression was observed only after the offset of the stimulation (Hoopfer et.al., 2015). Based on the comparison of those behavioral data to the imaging results in this paper, the reviewer suggested that Kenyon cells and PAM neurons are activated during courtship rather than during aggression. This is certainly a possible interpretation. However it is difficult to extrapolate from behavioral experiments in freely moving animals to calcium imaging results in head-fixed flies, particularly with response to neural dynamics.  Furthermore, Hr38 expression, like that of other IEGs (e.g., c-fos), may reflect persistently activated 2nd messenger pathways (e.g., cAMP, IP3) in Kenyon cells and PAM neurons that are not detected by calcium imaging, but that nevertheless play a role in mediating its behavioral effects. We still do not understand the mechanisms of how optogenetic stimulation of P1a neurons in freely behaving flies induces aggression vs. courtship behavior. Although 50 Hz stimulation of P1a neurons does not induce aggressive behavior during photo-stimulation, it is possible that this manipulation activates both aggression and courtship circuits, but that the courtship circuit might inhibit aggressive behavior at a site downstream of the MB (e.g., in the VNC). Once stimulation is terminated and courtship stops the fly would show aggressive behavior, due to release of that downstream inhibition (see Models in Anderson (2016) Fig 2d, e). In that case, there would be no apparent inconsistency between the imaging data and behavioral data. We agree that the reviewer's question is interesting and important but we feel that answering this question with decisive experiments is beyond the scope of this manuscript.

Finally, Reviewer #1 suggested a method to evaluate the Hr38 signals in the catFISH experiment of Figure 4. We appreciate their suggestions, but the way that we evaluated the Hr38 signals was basically the same as the way the reviewer suggested. We apologize for the confusion caused by the lack of detailed descriptions in the original manuscript. We have now revised the methods section to explain more clearly how we define the cells as positive based on Hr38EXN and Hr38INT signals.

Response to Recommendations for the authors:

“To strengthen the author's argumentation, I would distinguish in their quantification between gal4+ from the other [classes of neighboring neurons]” (Fig. 2 and 4).”

Our focus in this paper was to ask simply whether P1a neurons are active or not active during natural occurrences of the social behaviors they can evoke when artificially activated. We did not claim that they are the only cells in the region that control the behaviors.  It is not possible to compare their activation to that of 'other' cells neighboring P1a neurons without a separate marker to identify those cells driven by a different reporter system (e.g., LexA). This in turn would require repeating all of the experiments in Figs 2 and 4 from scratch with new genotypes permitting dual-labeling of the two populations by different XFPs, and quantifying the data using 4-color labeling. We respectfully submit that such curiosity-driven experiments, while in principle interesting, are beyond the scope of the present manuscript.  However, we have inserted text to acknowledge the possibility that the aggression-activated Hr38 signals in P1a- cells neighboring P1a+ cells may correspond to other classes of P1 neurons (of which there are 70 in total) or to pC1 cells. Changes:  Text lines 151-154.

“if the magenta dot is outside of the nuclei I would not count this as positive also the size of the dot seems to be a good marker of the reality of the signal). I would measure the intensity of the hr38EXN. A high Hr38EXN level associated with the presence of hr38INT would indicate that the cell has been activated during both encounters, while a lower hr38EXN with no hr38INT would suggest only an activation during the 1st behavioural context. Finally, a lower hr38EXN associated with the presence of hr38INT would suggest the opposite, an activation only during the 2nd behaviour.”

We agree that there are some tiny dot signals with hr38 INT probe that are more likely the background signals. We only counted the INT probe signals as positive when the cells had a clearly visible dot and also co-localize with the exonic probe's signal, as primary (un-spliced) Hr38 transcripts in the nucleus should be positive for both EXN and INT probes. Regarding the reviewer’s latter comments, we agree with their interpretation of the catFISH results and that is how we interpreted them originally. We measured the intensity of hr38EXN expression and defined hr38EXN-labeled cells as “positive” when the relative intensity was 3σ >average, a stringent criterion. In the revised manuscript, we added more detailed information in the methods section regarding our criteria for defining cell types as positive.

“Knowing that the P1a neurons (using the split-gal4) can trigger only wing extension when activated by optogenetic 50Hz, I would test to which behavioral context the MB neurons and the PAM neurons positively respond to.”

As we answered in 'Response to Public Review,' our opto-HI-FISH experiments identified Kenyon cells in the mushroom body (MB) and PAM neurons (as well as pCd neurons) as potential downstream targets of P1a cells, using Hr38 labeling. The purpose of the calcium imaging experiment in Figure 3 – supplement was to confirm the P1a-dependent activation of KCs and PAM neurons using an independent method. In that respect this control experiment was successful in that methodological confirmation. The reviser raised an interesting question about how our calcium imaging experiments relate to our behavioral experiments, in terms of the dynamics of KC and PAM activation. A recent publication (Shen et al., 2023) revealed that courtship behavior has a positive valence and that activation of P1 neurons mimics a courtship-reward state via activation of PAM dopaminergic neurons. Therefore, it is reasonable to think that PAM neurons (and Kenyon cells as downstream of PAM neurons) are activated during female exposure. However those data do not exclude the possibility that inter-male aggression is also rewarding in Drosophila males, as it has shown to be in mice. This is an interesting curiosity-driven question that has yet to be resolved.  Therefore, as mentioned in the 'Response to Public Review,' we feel that the additional experiment the reviewer suggests is beyond the scope of our manuscript.

Changes: None.

Minor comments:

“Please provide different pictures from main fig2 and sup2 for the three common conditions (control, aggression, and courtship).” 

The data set for Figure 2 and Figure 2 supplement are from the same experiment. Because of the limited space, we just presented the selected key conditions ('Control', 'Aggression', and 'Courtship') in the main figure and put the complete data set (including these three key conditions) in the supplemental figure.

Changes: None

“Please, provide scale bars for the images.”

Also, Reviewer #2 commented, 'Scale bars are missing on all the images throughout the main and supplementary figures.'

We have now added scale bars for each figure. 

“Fig.1: “Is the chrimsonTdtom images from endogenous fluorescence? It is not said in the legend and anti-dsred is not provided in the material and method while anti-GFP is.”

We are sorry for the confusion and thank the reviewer for raising that question. The signals were native fluorescence, and we have now added that information to the figure legend.

P7: "As an initial proof-of-concept application of HI-FISH, we asked whether neuronal subsets initially identified in functional screens for aggression-promoting neurons (Asahina et al., 2014; Hoopfer et al., 2015; Watanabe et al., 2017) were actually active during natural aggressive behavior. These included P1a, Tachykinin-FruM+ (TkFruM), and aSP2 neurons". Please put the references to the corresponding group of neurons listed. For example: "These included P1a neurons [Hoopfer et al., 2015]". 

We have now added these references.

P9: "Optogenetic and thermogenetic stimulation experiments have shown that that P1a interneurons can promote both male-directed aggression and male- or female-directed courtship" typo

We appreciate the reviewer for catching this error and have corrected the text.

(P10:" To validate this approach, we first asked whether we could detect Hr38 induction in pCd neurons, which were previously shown by calcium imaging to be (indirect) targets of P1a neurons". Reference [Jung et al., 2020] 

We have now added this reference.

Fig. 4A: Put the time scale on the diagram (3h adaptation-20min-30min rest-20min-10min rest-collect) 

We have now added the time scale in Figure 4A.

Reviewer #2: 

Response to Public Review: 

We thank the reviewer for their helpful comments and suggestions. We have addressed most of them in our revised manuscript. The main concern of Reviewer #2 was the temporal resolution of the HI-catFISH experiment shown in Figure 4 and Figure 4-Supplement. Our original manuscript illustrated temporal patterns of Hr38EXN and Hr38ITN signals concomitant with different behavioral paradigms (Figure 4B). The reviewer pointed out that the illustrated experimental design does not reflect the actual data shown in Figure 4-Supplement A-C. We believe this issue was raised because we drew the temporal pattern of Hr38EXN signals in Figure 4B based on the intensity of Hr38EXN signals (Figure 4-Supplement B) rather than based on the % number of positive cells (Figure 4-Supplement C). We have now revised the schematic time course of Hr38EXN signals in Figure 4B using the % of positive cells. We believe this change will be helpful for readers to understand better the experimental design since we used the % of positive cells to identify patterns of P1a neuron activation during male-male vs. male-female social interactions in Figure 4D. Another suggestion from Reviewer #2 was to add additional controls, such as the quantification of the intronic and exonic Hr38 probes after either only the first or second social context exposure. In response, we have now added the data from only the first social context (Figure 4C, and 4D, right column). These new data provides evidence that there are essentially no detectable Hr38INT signals 60 minutes later without a second behavioral context, while Hr38EXN signals are still present at the time of the analysis.  Unfortunately, we are not able to provide the converse dataset with the second behavioral context only to show that Hr38 INT signals are detected. On this point, we call the reviewer’s attention to Figure 4-supplement-S4A-C, which show that the INT probe signals are detectable at 15 and 30 minutes following stimulation, but not at 60 minutes.  In the experiment of Fig. 4B, flies are fixed and labeled for Hr38 30 minutes after the beginning of the second behavior, conditions under which we should obtain robust INT signals (as observed).  EXN signals are also expected at 30 minutes because the primary (non-spliced) RNA transcript detected by the INT probe also contains exonic sequences.

Response to Recommendations for the authors:

Given that the development of in situ HCR for the adult fly brain is so central to the present manuscript, I think that the methods section describing the HCR protocol can be significantly improved. In particular, the authors should fully describe the in situ HCR protocol including the 'minor modifications' they refer to, and define how they calculate the 'relative intensity to the background'.

We appreciate the reviewer’s suggestion. We have now revised the methods section to describe the procedure in more detail. Also, we will submit a separate document describing the HI-FISH protocol.

Note: The authors refer to a recently published paper by Takayanagi-Kiya et al (2023) describing activity-based neuronal labeling using a different immediate early gene, stripe/egr-1. The authors state the following: 'That study used a GAL4 driver for the stripe/egr-1 gene to label and functionally manipulate activated neurons. In contrast, our approach is based purely on detecting expression of the IEG mRNA using..'. Takayanagi-Kiya et al. (2023) also use in situ mRNA detection of the IEG stripe/egr-1 and not only a GAL4 driver system. This claim should be modified and the paper should be cited in the introduction of the present paper.

We have now cited the paper in the Introduction and have modified and moved the description originally in 'Note' section to Discussion (text lines: 392-404) as the reviewer requested. We have emphasized the difference between the two approaches for comparing neuronal activities during two different behaviors within the same animal. Takayanagi-Kiya used GAL4/UAS and stripe protein expression with immunohistochemistry to analyze neuronal activities during two different behaviors, while we exclusively analyzed Hr38 mRNA expression for this purpose, using intronic and exonic Hr38 probes. This approach made it possible to perform catFISH with higher temporal resolution and also allows extension of our approach to other IEGs for which antibodies are not available.

Please specify the nature of the iron fillings in the methods section.

We added a detailed description in the methods section, including the catalog number.

In Figure 1B, the authors may add a dashed outline to the regions magnified in 1C so that readers can more easily follow the figures. Moreover, it would be informative to see a more detailed quantification of the number of Hr38-positive cells in different brain regions marked by Fru-GAL4.

We have now added the whole brain images for each condition in Figure 1C and also quantitative data in Figure 1-Supplement C, as the reviewer suggested.

In the middle right aggression panel of Figure 2A, it looks as if one P1a neuron is not outlined.

We have carefully examined other z-planes through this region and based on those data have concluded that the signals mentioned by the reviewer are neurites from neurons labeled in other z-planes.

Changes: None.

The images in Figure 2A can be again found in Figure Supplement 2A, yet the number of neurons analyzed suggests the quantification was performed from different samples. The images in Figure Supplement 2A should be either changed or it should be explained as to why the images are the same yet the numbers in the legend are different.

We apologize for the confusion. Figure 2 and Figure 2-Supplement are from the same experiment. To avoid clutter we illustrated three key conditions ('Control,' 'Aggression,' and 'Courtship') in the main figure. The reason why the numbers in the legend are different is that the purpose of presenting Figure 2-Supplement B-D was to determine whether there were differences in the intensity of Hr38 FISH signals in the neurons considered as 'positive' in different conditions. Therefore, the numbers described in Figure 2-Supplement legend are derived only from those neurons that were considered Hr38-positive, while the numbers in Figure 2 include all neurons analyzed. We have now added notes to explain this in the Figure 2 – supplement legend.

The panels of the quantification of the Hr38 relative intensity in Figure 2B/C/D are very difficult to read, ideally, they should be plotted as in Figure Supplement 2B/C/D.

The graphs in Figure 2B-D (upper) show data from all GFP-labeled cells scored, including cells defined as 'negative' or 'borderline.' In contrast, the graphs in Figure 2-supplement show the relative Hr38 signal intensity in those GFP neurons defined as positive based on the analysis in Fig. 2B. If we were to plot the data in Fig. 2B (upper) as box plots (like that in Figure-2-supplement), we would see either a skewed (only negative cells) or a bimodal distribution (one around the negative population and the other around the positive population); the shapes of these distributions would likely be hidden in the box-whisker plots format. Therefore, we prefer to plot all of the data points as we did in the original manuscript. However, we agree that the data points in the original manuscript were hard to read. We therefore changed the format of the datapoints from blurry dots to open circles with clear solid lines.

In Figure 2B/C/D, please specify in the figure legend what 'grouped in categories according to character' means. 

We used letters to mark statistically significant differences (or lack thereof) between conditions. Bars sharing at least one common letter are not significantly different.  If they do not share any letter, they are significantly different. For example, Aggression: bc vs. Dead: bc, means no difference. Aggression: bc vs. No Food: b, or Aggression: bc vs. Courtship: c also means no difference between Aggression and each of the two other conditions. However, 'No Food: b' and 'Courtship: c' have no common letter, meaning they are different. This is a standard method for showing statistically comparisons among multiple bars without lots of asterisks and horizontal bars cluttering the figure, and we have revised the legend to clarify what each letter means. We have also removed the color shading in Figure 2 B-D as it may have been confusing.

A quantification of the number of Hr38-positive neurons and Hr38 relative intensity during the entire time course would be informative in Figure 3D. 

Although the data set for this figure is different from that for Figure 4-Supplement A-C, the main claim is the same. Therefore, Figure 4 - Supplement essentially provides the information that the reviewer suggested. However, we also reanalyzed the data set used for the original Figure 3D and evaluated % positive cells at the 30-minute time point and have now added that number in the figure legend.

In the legend of Figure 3D, it says '..The expression level reaches its peak at 30-60min', yet I don't see timepoints beyond 60min. Please rephrase or add additional timepoints. 

We apologize for the error. We have rephrased the text.

Figure Supplement 3A/D: please add an outline or a schematic figure to better understand where the imaging is performed.

We added illustrated schemas next to the title of each experiment (P1->PAM neurons (bundle) and P1 -> Kenyon cells (bundle)).

Figure Supplement 3C/F: please add information about the statistical test to the corresponding figure legend.

We have added a phrase to describe the test used.

Figure Supplement 3G/H/I/J: motion artifacts can potentially strongly affect the performed analysis given that cell bodies are very small and highly subjected to motion. Can the authors comment on how they corrected for motion?

We have now described how we corrected for motion artifacts in the Methods section.

Figure 4C/D: It seems as if the representative images don't reflect the quantification, e.g., in the male -> female panel, close to 100% of the neurons are positive for the exonic probe as opposed to approx. 40% in the bar graph.

Please see our response to this issue in the 'Response to Public Review (Reviewer #1)'.

Additional controls should be included in Figure 4C in order to assess the temporal resolution of HI-CatFISH more in detail (see 'Weaknesses').

We have also answered this in the 'Response to Public Review'.

The authors should adjust the scheme in the main Figure 4B to reflect the data presented in Figure S4A and C. For instance, the peak for the intronic version is observed at 15 minutes, while at 30 minutes, both the exonic and intronic signals show an equal level of signal.

We have addressed this issue in the 'Response to Public Review'.

We thank the reviewers again for their helpful comments and hope that with these changes, the manuscript will now be acceptable for official publication in eLife.

Author response:

Reviewer #1 (Public Review): 

The manuscript entitled "A septo-hypothalamic-medullary circuit directs stress-induced analgesia" by Shah et al., showed that the dLS-to-LHA circuit is sufficient and necessary for stress-induced analgesia (SIA), which is mediated by the rostral ventromedial medulla (RVM) in a opioid-dependent manner. This study is interesting and important and the conclusions are largely supported by the data. I have a few concerns as follows:

We thank the reviewer for finding our study “interesting”, “important”, and “conclusions are largely supported by data”.

(1)  The present data show that activation of dLS neurons produces SIA, however, this manipulation is non-specific. It may be better to see the effect of specific manipulation of stress-activated c-Fos positive neurons in the dLS using a combination of the Tet-Off system and chemogenetic/optogenetic tools. 

We agree with the reviewer that activating the stress-“trapped” neurons will be more specific way to induce SIA through septal activation, compared to the activation of entire dLS strategy pursued by us. In most likelihood, we expect to see a robust SIA if specifically stress responsive dLS neurons are observed. We are in the process of acquiring the genetic tools required for “Trapping” stress neurons and expect to be able to perform the experiments suggested by the reviewers in the coming months. 

(2)  Depending on its duration, and intensity, stress can exert potent and bidirectional modulatory effects on pain, either reducing pain (SIA) or exacerbating it (stress-induced hyperalgesia, SIH). Is the circuit in the manuscript involved in SIH?

As mentioned by the reviewer, it would be reasonable to suspect that the dLS neurons are involved in SIH. However, we believe that the experiments to test this hypothesis is outside the scope of this paper, since here we have focused on the circuit mechanisms for SIA. However, in the revised discussion section, we have included the possibility of dLS neurons driving SIH. 

(3)  It is well-accepted that opioid and cannabinoid receptors participate in the SIA, and the evidence is especially strong for the RVM endocannabinoid system. Given this, why did the authors focus their study on the opioid system?

We agree with the reviewer that dLS-mediated SIA may work through neural circuits centered on RVM expressing receptors for either or both opioids and endocannabinoids. We primarily focused on the opioidergic system in the RVM as decades of mechanistic work has revealed how the ON, OFF, and neutral neurons modulate pain through the endogenous opioids and even mediate SIA. In the revised discussion, we have included the possibility of involvement of both pain modulatory systems. 

(4)  Does silencing of the dLS neurons affect stress-induced anxiety-like behaviors? Alternatively, what is the relationship between SIA and the level of stress-induced anxiety?

We did not test if the silencing of dLS would affect stress-induced anxiety, as our focus was on the pain modulatory effects of dLS activation. The relationships between levels of SIA and stress-induced anxiety will be interesting to explore in future. We believe we would need better behavioral assays compared to the existing ones to quantitatively measure levels of stress-induced anxiety and SIA levels.

(5)  Direct electrophysiological evidence should be provided to confirm the efficacy of the MP-CNO.

We agree with the reviewer that ex-vivo electrophysiology experiments will substantiate the effectiveness of the MP-CNO. However, we do not have the expertise, or the instrumentation required to perform these experiments in our laboratory.

(6)  Is the LHA a specific downstream target for SIA, and is the LHA involved in stressinduced anxiety-like behaviors?

Several lines of evidence points to the fact that LHA neurons are involved in stressinduced anxiety. We have also shown that the dLS downstream neurons in the LHA are activated by acute restraint by fiber photometry recordings. Thus, we expect activation of the LHA neurons will cause stress-induced anxiety. However, we wanted to focus on the pain modulation aspect of the dLS-LHA-RVM circuitry.

(7)  Do LHA neurons have direct projections to the RVM? If yes, what is its role in the SIA?

Our anatomical studies using transsynaptic anterograde and retrograde viral strategies in the Figure 6 shows that the LHA neurons have direct projections to the RVM, and these neurons are sufficient in driving hyperalgesia, as well as necessary for SIA. 

Reviewer #2 (Public Review): 

Summary: 

In this manuscript, Shah et al. explore the function of an understudied neural circuitry from the dLS -> LHA -> RVM in mediating stress-induced analgesia. They initially establish this neural circuitry through a series of intersectional tracings. Subsequently, they conduct behavioral tests, coupled with optogenetic or chemogenetic manipulations, to confirm the involvement of this pathway in promoting analgesia. Additionally, fiber photometry experiments are employed to investigate the activity of each brain region in response to stress and pain. 

Strengths: 

Overall, the study is comprehensive, and the findings are compelling. 

We appreciate the reviewer for finding our manuscript “comprehensive” and “compelling”.

Weaknesses: 

One noteworthy concern arises regarding the overarching hypothesis that restrainedinduced stress promotes analgesia. A more direct interpretation suggests that intense struggling, rather than stress per se, activates the dLS -> LHA -> RVM pathway that may drive analgesic responses. 

We agree with the reviewer that our data can be interpreted as “intense struggling”, rather than the “acute stress” might have altered the pain thresholds in mice. However, we would like to point out that the restraint induced stress model that we have used has been long regarded as a standard for inducing stress. Moreover, we have demonstrated that dLS activation results into acute stress by measuring the blood corticosterone levels, and showed that dLS activations caused stress-induced anxiety through lightdark box tests.

Reviewer #2 (Recommendations For The Authors): 

Please find below my other comments for improvements. 

Introduction: The authors claimed that "dLS neurons receive nociceptive inputs from the thalamus and somatosensory cortices." However, citations are missing.

We have added the citations.

Figure 1 B&C: Although this paper focuses on the dLS, it would be informative to also include vLS c-Fos images (maybe in a supplementary figure), given that these data appear to be already acquired. The inclusion of vLS data will provide critical information regarding potential specificity (or lack of) across LS subregions in stress responses.

In the revised manuscript we have added the vLS c-Fos images as suggested by the reviewer. 

Figure 1D: Quantification of Vgat vs. Vglut neurons is missing. It is unclear if the Vgat neurons are restricted to small clusters.

We did not add the Vglut vs, Vgat quantification since from both of our experiments and publicly available data from the Allen Brain Atlas show that almost all of the neurons in the LS are gabaergic. We found very rare,0-2 Vglut2 expressing neurons per section in the the LS of the mouse brain.

Figure 1G: The Y-axis label is missing. 

We have added the axis in the revised manuscript.

Figure 2: The authors claimed that dLS neurons are preferentially tuned to stress caused by physical restraint. However, it appears that these neurons are specifically tuned to intense struggle behavior (transient) rather than stress (prolonged).

We agree with the reviewer that the SIA observed in mice with dLS activation, can be interpreted as the effect of transient struggle behavior rather than the prolonged stress. However, we would like to point out that the acute restraint for one hour is known to produce prolonged stress, and is backed up by increased blood coticosterone levels and stress-induced anxiety (Fig1-Fig Supplementary 1).

Figure 4: The authors provided compelling evidence that dLS neurons synapse on LHA Vglut2 neurons. However, it is unclear if they exclusively target the Vglut2 neurons or also synapse on LHA Vgat neurons.

We agree with the reviewer that even though the majority of the dLS downstream neurons in the LHA are glutamatergic, as now shown in the Fig. 4D, few neurons do not express Vglut and thus must be Gabaergic. 

Figure 5D: It is unclear if the trace represents dLS or LHA calcium signal (in the main text, the authors claimed both).

Now, we have mentioned the neurons on the LHA we have recorded from at the top of Figure 5C, D. 

Figure 6 G&H: Presumably, ΔG-Rabies does not transmit across neurons due to the deletion of the glycoprotein (G) gene. Thus, it is unclear why dLS and LHA neurons express mCherry after injecting rabies into RVM.

The aim of the rabies experiment was to test that the cells in the LHA that receive inputs from the dLS are the same ones that send projections downstream to the RVM. To this end, we used a monosynaptic rabies virus that has retrograde properties. Hence, when injected into the RVM, it was taken up by the terminals of the LHA neurons in the RVM and traveled to the cell bodies in the LHA. We injected the AAV1-Transsyn-Cre in the dLS, so only the cells downstream of the dLS in the LHA can express the Credependent glycoprotein (G) gene. Thus, the rabies-mCherry virus infected the LHA neurons downstream of dLS specifically, and jumped a synapse, to label the upstream dLS neurons.

The authors claim that "RVMpost-LHA neurons may modulate nociceptive thresholds through their local synaptic connections within the RVM, recurrent connections with the PAG, or direct interactions with spinal cord neurons." It is unclear what the "local synaptic connections within the RVM" means. It is also unclear whether there is evidence of recurrent connections between the RVM and PAG.

We meant by local connections as intrinsic connections within the RVM, as in some or few of the RVM neurons, post LHA might be interneurons and mediating SIA by modulating the ON or OFF cells. There are some anatomical evidence for the ascending inputs from RVM to the PAG and the we have now included the citation in the mentioned section of the manuscript.

Author response:

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

Public Reviews:

Reviewer #1 (Public review):

Summary

In this manuscript, Day et al. present a high-throughput version of expansion microscopy to increase the throughput of this well-established super-resolution imaging technique. Through technical innovations in liquid handling with custom-fabricated tools and modifications to how the expandable hydrogels are polymerized, the authors show robust ~4-fold expansion of cultured cells in 96-well plates. They go on to show that HiExM can be used for applications such as drug screens by testing the effect of doxorubicin on human cardiomyocytes. Interestingly, the effects of this drug on changing DNA organization were only detectable by ExM, demonstrating the utility of HiExM for such studies.

Overall, this is a very well-written manuscript presenting an important technical advance that overcomes a major limitation of ExM - throughput. As a method, HiExM appears extremely useful and the data generally support the conclusions.

Strengths

Hi-ExM overcomes a major limitation of ExM by increasing the throughput and reducing the need for manual handling of gels. The authors do an excellent job of explaining each variation introduced to HiExM to make this work and thoroughly characterize the impressive expansion isotropy. The dox experiments are generally well-controlled and the comparison to an alternative stressor (H2O2) significantly strengthens the conclusions.

Weaknesses

(1) It is still unclear to me whether or not cells that do not expand remain in the well given the response to point 1. The authors say the cells are digested and washed away but then say that there is a remaining signal from the unexpanded DNA in some cases. I believe this is still a concern that potential users of the protocol should be aware of.

Although ProteinaseK digestion removes most of the unexpanded cells, DNA can sometimes persist. As such, we occasionally observe Hoechst signal underneath cells. The residual DNA is easily differentiated from nuclear Hoechst signal and does not confound interpretation of results. We have added a new supplementary figure that further clarifies this point.

(2) Regarding the response to point 9, I think this information should be included in the manuscript, possibly in the methods. It is important for others to have a sense of how long imaging may take if they were to adopt this method.

We have added detailed information to the methods section to address this point as shown below.  In general, we image HiExM samples on the Opera Phenix at 63x with the following parameters: 100% laser power for all channels; 200 ms exposure for Hoechst, 500-1000+ ms exposure for immunostained channels depending on the strength of the stain and the laser; 60 optical sections with 1 micron spacing; and 4-20 fields of view per well depending on the cell density and sample size requirements. Therefore, imaging one full 96-well plate (60 wells total as we avoid the outer wells) takes anywhere from 3 hr to 64 hr depending on the combination of parameters used.

Reviewer #2 (Public review):

Summary:

In the present work, the authors present an engineering solution to sample preparation in 96-well plates for high-throughput super resolution microscopy via Expansion Microscopy. This is not a trivial problem, as the well cannot be filled with the gel, which would prohibit expansion of the gel. They thus engineered a device that can spot a small droplet of hydrogel solution and keep it in place as it polymerises. It occupies only a small portion space at the center of each well, the gel can expand into all directions and imaging and staining can proceed by liquid handling robots and an automated microscope.

Strengths:

In contrast to Reference 8, the authors system is compatible with standard 96 well imaging plates for high-throughput automated microscopy and automated liquid handling for most parts of the protocol. They thus provide a clear path towards high throughput exM and high throughout super resolution microscopy, which is a timely and important goal.

Addition upon revision:

The authors addressed this reviewer's suggestions.

Reviewer #3 (Public review):

Summary:

Day et al. introduced high-throughput expansion microscopy (HiExM), a method facilitating the simultaneous adaptation of expansion microscopy for cells cultured in a 96-well plate format. The distinctive features of this method include: 1) the use of a specialized device for delivering a minimal amount (~230 nL) of gel solution to each well of a conventional 96-well plate, and 2) the application of the photochemical initiator, Irgacure 2959, to successfully form and expand toroidal gel within each well.

Addition upon revision:

Overall, the authors have adequately addressed most of the concerns raised. There are a few minor issues that require attention.

Minor comments:

Figure S10: There appears to be a discrepancy in the panel labeling. The current labels are EH, but it is unclear whether panels A-D exist. Also, this reviewer thought that panels G and H would benefit from statistical testing to strengthen the conclusions. As a general rule for scientific graph presentation, the y-axis of all graphs should start at zero unless there is a compelling reason not to do so.

We have revised Figure S10 to address your comments.

Author response:

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

Reviewer #1 (Public Review):

Summary:

By examining the prevalence of interactions with ancient amino acids of coenzymes in ancient versus recent folds, the authors noticed an increased interaction propensity for ancient interactions. They infer from this that coenzymes might have played an important role in prebiotic proteins.

Strengths:

(1) The analysis, which is very straightforward, is technically correct. However, the conclusions might not be as strong as presented.

(2) This paper presents an excellent summary of contemporary thought on what might have constituted prebiotic proteins and their properties.

(3) The paper is clearly written.

We are grateful for the kind comments of the reviewer on our manuscript. However, we would like to clarify a possible misunderstanding in the summary of our study. Specifically, analysis of "ancient versus recent folds" was not really reported in our results. Our analysis concerned "coenzyme age" rather than the "protein folds age" and was focused mainly on interaction with early vs. late amino acids in protein sequence. While structural propensities of the coenzyme binding sites were also analyzed, no distinction on the level of ancient vs. recent folds was assumed and this was only commented on in the discussion, based on previous work of others. 

Weaknesses:

(1) The conclusions might not be as strong as presented. First of all, while ancient amino acids interact less frequently in late with a given coenzyme, maybe this just reflects the fact that proteins that evolved later might be using residues that have a more favorable binding free energy.

We would like to point out that there was no distinction between proteins that evolved early or late in our dataset of coenzyme-binding proteins. The aim of our analysis was purely to observe trends in the age of amino acids vs. age of coenzymes. While no direct inference can be made from this about early life as all the proteins are from extant life (as highlighted in the discussion of our work), our goal was to look for intrinsic propensities of early vs. late amino acids in binding to the different coenzyme entities. Indeed, very early interactions would be smeared by the eons of evolutionary history (perhaps also towards more favourable binding free energy, as pointed out also by the reviewer). Nevertheless, significant trends have been recorded across the PDB dataset, pointing to different propensities and mechanistic properties of the binding events. Rather than to a specific evolutionary past, our data therefore point to a “capacity” of the early amino acids to bind certain coenzymes, and we believe that this is the major (and standing) conclusion of our work, along with the properties of such interactions. In our revised version, we will carefully go through all the conclusions and make sure that this message stands out, but we are confident that the following concluding sentences copied from the abstract and the discussion of our manuscript fully comply with our data:

“These results imply the plausibility of a coenzyme-peptide functional collaboration preceding the establishment of the Central Dogma and full protein alphabet evolution”

“While no direct inferences about distant evolutionary past can be drawn from the analysis of extant proteins, the principles guiding these interactions can imply their potential prebiotic feasibility and significance.”

“This implies that late amino acids would not be necessarily needed for the sovereignty of coenzyme-peptide interplay.”

We would also like to add that proteins that evolved later might not always have higher free energy of binding. Musil et al., 2021 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8294521/)  showed in their study on the example of haloalkane dehalogenase Dha A that the ancestral sequence reconstruction is a powerful tool for designing more stable, but also more active proteins. Ancestral sequence reconstruction relies on finding ancient states of protein families to suggest mutations that will lead to more stable proteins than are currently existing proteins. Their study did not explore the ligand-protein interactions specifically but showed that ancient states often show more favorable properties than modern proteins.

(2) What about other small molecules that existed in the probiotic soup? Do they also prefer such ancient amino acids? If so, this might reflect the interaction propensity of specific amino acids rather than the inferred important role of coenzymes.

We appreciate the comment of the reviewer towards other small molecules, which we assume points mainly towards metal ions (i.e. inorganic cofactors). We completely agree with the reviewer that such interactions are of utmost importance to the origins of life. Intentionally, they were not part of our study, as these have already been studied previously by others (e.g. Bromberg et al., 2022; and reviewed in Frenkel-Pinter et al., 2020) and also us (Fried et al., 2022). For example, it is noteworthy that prebiotically relevant metal binding sites (e.g. of Mg2+) exhibit enrichment in early amino acids such as Asp and Glu while more recent metal (e.g. Cu and Zn) site in the late amino acids His and Cys (Fried et al., 2022). At the same time, comparable analyses of amino acid - coenzyme trends were not available.

Nevertheless, involvement of metal ions in the coenzyme binding sites was also studied here and pointed to their bigger involvement with the Ancient coenzymes. In the revised version of the manuscript, we will be happy to enlarge the discussion of the studies concerning inorganic cofactors.

The following sentence was added in the discussion of the revised manuscript:

“This would also be true for direct interaction of early peptides/proteins and metal ions, independent of organic cofactor involvement, as discussed previously by us and others (Bromberg et al., 2022; Frenkel-Pinter et al., 2020; Fried et al., 2022).  For example, it has been observed that coordination of prebiotically most relevant metal ions (e.g., Mg2+) is more often mediated by early amino acids such as Asp and Glu, whereas metal ions of later relevance (e.g., Cu and Zn) bind more frequently via late amino acids like His and Cys (Fried et al. 2022). Similarly, ancient metal binding folds have been shown to be enriched in early amino acids (Bromberg et al., 2022).”

(3) Perhaps the conclusions just reflect the types of active sites that evolved first and nothing more.

We partly agree on this point with the reviewer but not on the fact why it is listed as the weakness of our study and on the “nothing more” notion. Understanding what the properties of the earliest binding sites is key to merging the gap between prebiotic chemistry and biochemistry. The potential of peptides preceding ribosomal synthesis (and the full alphabet evolution) along with prebiotically plausible coenzymes addresses exactly this gap, which is currently not understood.  

Reviewer #2 (Public Review):

I enjoyed reading this paper and appreciate the careful analysis performed by the investigators examining whether 'ancient' cofactors are preferentially bound by the first-available amino acids, and whether later 'LUCA' cofactors are bound by the late-arriving amino acids. I've always found this question fascinating as there is a contradiction in inorganic metal-protein complexes (not what is focused on here). Metal coordination of Fe, Ni heavily relies on softer ligands like His and Cys - which are by most models latecomer amino acids. There are no traces of thiols or imidazoles in meteorites - although work by Dvorkin has indicated that could very well be due to acid degradation during extraction. Chris Dupont (PNAS 2005) showed that metal speciation in the early earth (such as proposed by Anbar and prior RJP Williams) matched the purported order of fold emergence.

As such, cofactor-protein interactions as a driving force for evolution has always made sense to me and I admittedly read this paper biased in its favor. But to make sure, I started to play around with the data that the authors kindly and importantly shared in the supplementary files. Here's what I found:

Point 1: The correlation between abundance of amino acids and protein age is dominated by glycine.

There is a small, but visible difference in old vs new amino acid fractional abundance between Ancient and LUCA proteins (Figure 3, Supplementary Table 3). However, the bias is not evenly distributed among the amino acids - which Figure 4A shows but is hard to digest as presented. So instead I used the spreadsheet in Supplement 3 to calculate the fractional difference FDaa = F(old aa)-F(new aa). As expected from Figure 3, the mean FD for Ancient is greater than the mean FD for LUCA. But when you look at the same table for each amino acid FDcofactor = F(ancient cofactor) - F(LUCA cofactor), you now see that the bias is not evenly distributed between older and newer amino acids at all. In fact, most of the difference can be explained by glycine (FDcofactor = 3.8) and the rest by also including tryptophan (FDcofactor = -3.8). If you remove these two amino acids from the analysis, the trend seen in Figure 3 all but disappears.

Troubling - so you might argue that Gly is the oldest of the old and Trp is the newest of the new so the argument still stands. Unfortunately, Gly is a lot of things - flexible, small, polar - so what is the real correlation, age, or chemistry? This leads to point 2.

We truly acknowledge the effort that the reviewer made in the revision of the data and for the thoughtful, deeper analysis. We agree that this deserves further discussion of our data. 

As invited by the reviewer, we indeed repeated the analysis on the whole dataset. First, we would like to point out that the reviewer was most probably referring to the Supplementary Fig. 2 (and not 3, which concerns protein folds). While the difference between Ancient and LUCA coenzyme binding is indeed most pronounced for Gly and Trp, we failed to confirm that the trend disappears if those two amino acids are removed from the analysis (additional FDcofactors of 3.2 and -3.2 are observed for the early and late amino acids, resp.), as seen in Table I below. The main additional contributors to this effect are Asp (FD of 2.1) and Ser (FD of 1.8) from the early amino acids and Arg (FD of -2.6) and Cys (FD of -1.7) of the late amino acids. Hence, while we agree with the reviewer that Gly and Trp (the oldest and the youngest) contribute to this effect the most, we disagree that the trend reduces to these two amino acids.  

In addition, the most recent coenzyme temporality (the Post-LUCA) was neglected in the reviewer’s analysis. The difference between F (old) and F (new) is even more pronounced in Post-LUCA than in LUCA, vs. Ancient (Supplementary table 5A) and depends much less on Trp. Meanwhile, Asp, Ser, Leu, Phe, and Arg dominate the observed phenomenon (Supplementary table 5b). This further supports our lack of agreement with the reviewer’s point. Nevertheless, we remain grateful for this discussion and we will happily include this additional analysis in the Supplementary Material of our revised manuscript.

The following text (and the additional data) was included in the revised manuscript version:

“To explore the contribution of individual amino acids to this effect, fractional difference (FD) for early vs. late amino acids among the Ancient, LUCA, and Post-LUCA coenzyme binding was calculated (Supplementary Table 5). The mean FD revealed a similar trend to the amino acid composition analysis (Fig. 3). The amino acids most enriched in LUCA vs. Post-LUCA are Gly, Ser, and Leu (FD of 4.4, 4.3, and 4.1 respectively), while the most depleted include Phe, Arg, and His (FD of -11, -4.2, and -3.2) (Supplementary Table 5B).”

Point 2 - The correlation is dominated by phosphate.

In the ancient cofactor list, all but 4 comprise at least one phosphate (SAM, tetrahydrofolic acid, biopterin, and heme). Except for SAM, the rest have very low Gly abundance. The overall high Gly abundance in the ancient enzymes is due to the chemical property of glycine that can occupy the right-hand side of the Ramachandran plot. This allows it to make the alternating alphaleft-alpharight conformation of the P-loop forming Milner-White's anionic nest. If you remove phosphate binding folds from the analysis the trend in Figure 3 vanishes.

Likewise, Trp is an important functional residue for binding quinones and tuning its redox potential. The LUCA cofactor set is dominated by quinone and derivatives, which likely drives up the new amino acid score for this class of cofactors.

Once again, we are thankful to the reviewer for raising this point. The role of Gly in the anionic nests proposed by Milner-White and Russel, as well as the Trp role in quinone binding are important points that we would be happy to highlight more in the discussion of the revised manuscript. 

Nevertheless, we disagree that the trends reduce only to the phosphate-containing coenzymes and importantly, that “the trend in Figure 3 vanishes” upon their removal. Supplementary table 6A and 6B show the data for coenzymes excluding those with phosphate moiety and the trend in Fig. 3 remains, albeit less pronounced.

The following text was included in the revised manuscript version:

“Moreover, we investigated whether the observed trend in amino acid occurrence at the binding sites was dominated by the presence of phosphate groups, which are common in many ancient cofactors except for SAM, Tetrahydrofolic acid, Biopterin, and Heme. An additional analysis therefore excluded all phosphate-containing coenzymes indicating that while the trend is less pronounced, it remains even in the absence of phosphate groups (Supplementary Table 6).”

In summary, while I still believe the premise that cofactors drove the shape of peptides and the folds that came from them - and that Rossmann folds are ancient phosphate-binding proteins, this analysis does not really bring anything new to these ideas that have already been stated by Tawfik/Longo, Milner-White/Russell, and many others.

I did this analysis ad hoc on a slice of the data the authors provided and could easily have missed something and I encourage the authors to check my work. If it holds up it should be noted that negative results can often be as informative as strong positive ones. I think the signal here is too weak to see in the noise using the current approach.

We are grateful to the reviewer for encouraging further look at our data. While we hope that the analysis on the whole dataset (listed in Tables I - IV) will change the reviewer’s standpoint on our work, we would still like to comment on the questioned novelty of our results. In fact, the extraordinary works by Tawfik/Longo and Milner-While/Russel (which were cited in our manuscript multiple times) presented one of the motivations for this study.   We take the opportunity to copy the part of our discussion that specifically highlights the relevance of their studies, and points out the contribution of our work with respect to theirs.  

“While all the coenzymes bind preferentially to protein residue sidechains, more backbone interactions appear in the ancient coenzyme class when compared to others. This supports an earlier hypothesis that functions of the earliest peptides (possibly of variable compositions and lengths) would be performed with the assistance of the main chain atoms rather than their sidechains (Milner-White and Russel 2011). Longo et al., recently analyzed binding sites of different phosphate-containing ligands which were arguably of high relevance during earliest stages of life, connecting all of today’s core metabolism (Longo et al., 2020 (b)). They observed that unlike the evolutionary younger binding motifs (which rely on sidechain binding), the most ancient lineages indeed bind to phosphate moieties predominantly via the protein backbone.

Our analysis assigns this phenomenon primarily to interactions via early amino acids that (as mentioned above) are generally enriched in the binding interface of the ancient coenzymes. This implies that late amino acids would not be necessarily needed for the sovereignty of coenzyme-peptide interplay.”

Unlike any other previous work, our study involves all the major coenzymes (not just the phosphate-containing ones) and is based on their evolutionary age, as well as age of amino acids. It is the first PDB-wide systematic evolutionary analysis of coenzyme-amino acid binding. Besides confirming some earlier theoretical assertions (such as role of backbone interactions in early peptide-coenzyme evolution) and observations (such as occurrence of the ancient phosphate-containing coenzymes in the oldest protein folds), it uncovers substantial novel knowledge. For example, (i) enrichment of early amino acids in the binding of ancient coenzymes, vs. enrichment of late amino acids in the binding of LUCA and Post-LUCA coenzymes, (ii) the trends in secondary structure content of the binding sites of coenzyme of different temporalities, (iii) increased involvement of metal ions in the ancient coenzyme binding events, and (iv) the capacity of only early amino acids to bind ancient coenzymes. In our humble opinion, all of these points bring important contributions in the peptide-coenzyme knowledge gap which has been discussed in a number of previous studies.

Recommendations for the authors:

(1) By only focusing on coenzymes, the authors may have overestimated their importance. What about other small molecules that existed in the prebiotic soup? Do they also prefer such ancient amino acids? If so, this might reflect the interaction propensity of specific amino acids rather than some possible role in very ancient proteins. Or it might diminish the conjectured importance of coenzymes.

The following sentence was added in the discussion of the revised manuscript:

“This would also be true for direct interaction of early peptides/proteins and metal ions, independent of organic cofactor involvement, as discussed previously by us and others (Bromberg et al., 2022; Frenkel-Pinter et al., 2020; Fried et al., 2022).  For example, it has been observed that coordination of prebiotically most relevant metal ions (e.g., Mg2+) is more often mediated by early amino acids such as Asp and Glu, whereas metal ions of later relevance (e.g., Cu and Zn) bind more frequently via late amino acids like His and Cys (Fried et al. 2022). Similarly, ancient metal binding folds have been shown to be enriched in early amino acids (Bromberg et al., 2022).”

(2) The authors should analyze whether the interactions are with similar types of amino acids in ancient versus early proteins.

While we appreciate the interesting suggestion, we would like to clarify that we did not aim to elucidate the differences between early and late protein folds - we agree that this might add an interesting perspective to our work, but we feel that it is well beyond the scope of our current study.

(3) The authors might also wish to do sequence alignments to the structures in early versus late evolving proteins to see how general this pattern of residue usage is beyond the limited set of proteins found in the PDB.

This is an interesting suggestion but similar to the previous recommendation, it is not within the scope of this study where no distinction between early and late evolving proteins has been made.  

There has been a number of attempts to classify the folds as shared among Bacteria, Archea and Eukaryota or specific to  one or two of these groups of organisms (https://link.springer.com/article/10.1007/s00239-023-10136-x, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9541633/) - this does not however compare easily with our time scales - where ancient ligands occur well before the last common ancestor.

We also agree  the set of sequences present in the PDB is biased, but perhaps it is less biased than we have thought. The recent fantastic work https://www.biorxiv.org/content/10.1101/2024.03.18.585509v2)  from Nicola Bordin and his colleagues from Orengo group attempted to classify over 200 milion structures in Alphafold database in so called Encyclopedia of Domains and they found out that nearly 80% of detected domains can be assigned to already known superfamilies in CATH (https://www.biorxiv.org/content/10.1101/2024.03.18.585509v2).

(4) The authors might wish to consider the results in Skolnick, H. Zhou, and M. Gao. On the possible origin of protein homochirality, structure, and biochemical function. PNAS 2019: 116(52): 26571-26579.

Based on the editorial recommendation, the following sentence was added in the discussion:

“It has been implied by computer simulations that coenzymes could bind to proteins with similar propensity even before the onset of protein homochirality, despite lower structural stability and secondary structure content in heterochiral polypeptides (Skolnick et al., 2019).”

Author Response:

Reviewer #1 (Public Review):

This work makes several contributions: (1) a method for the self-supervised segmentation of cells in 3D microscopy images, (2) an cell-segmented dataset comprising six volumes from a mesoSPIM sample of a mouse brain, and (3) a napari plugin to apply and train the proposed method.

First, thanks for acknowledging our contributions of a new tool, new dataset, and new software.

(1) Method

This work presents itself as a generalizable method contribution with a wide scope: self-supervised 3D cell segmentation in microscopy images. My main critique is that there is almost no evidence for the proposed method to have that wide of a scope. Instead, the paper is more akin to a case report that shows that a particular self-supervised method is good enough to segment cells in two datasets with specific properties.

First, thanks for acknowledging our contributions of a new tool, new dataset, and new software. We agree we focus on lightsheet microscopy data, therefore to narrow the scope we have changed the title to “CellSeg3D: self-supervised 3D cell segmentation for light-sheet microscopy”.

To support the claim that their method "address[es] the inherent complexity of quantifying cells in 3D volumes", the method should be evaluated in a comprehensive study including different kinds of light and electron microscopy images, different markers, and resolutions to cover the diversity of microscopy images that both title and abstract are alluding to. The main dataset used here (a mesoSPIM dataset of a whole mouse brain) features well-isolated cells that are easily distinguishable from the background. Otsu thresholding followed by a connected component analysis already segments most of those cells correctly.

You have selectively dropped the last part of that sentence that is key: “.... 3D volumes, often in cleared neural tissue” – which is what we tackle. The next sentence goes on to say: “We offer a new 3D mesoSPIM dataset and show that CellSeg3D can match state-of-the-art supervised methods.” Thus, we literally make it clear our claims are on MesoSPIM and cleared data.

The proposed method relies on an intensity-based segmentation method (a soft version of a normalized cut) and has at least five free parameters (radius, intensity, and spatial sigma for SoftNCut, as well as a morphological closing radius, and a merge threshold for touching cells in the post-processing). Given the benefit of tweaking parameters (like thresholds, morphological operation radii, and expected object sizes), it would be illuminating to know how other non-learning-based methods will compare on this dataset, especially if given the same treatment of segmentation post-processing that the proposed method receives. After inspecting the WNet3D predictions (using the napari plugin) on the used datasets I find them almost identical to the raw intensity values, casting doubt as to whether the high segmentation accuracy is really due to the self-supervised learning or instead a function of the post-processing pipeline after thresholding.

First, thanks for testing our tool, and glad it works for you. The deep learning methods we use cannot “solve” this dataset, and we also have a F1-Score (dice) of ~0.8 with our self-supervised method. We don’t see the value in applying non-learning methods; this is unnecessary and beyond the scope of this work.

I suggest the following baselines be included to better understand how much of the segmentation accuracy is due to parameter tweaking on the considered datasets versus a novel method contribution:<br /> * comparison to thresholding (with the same post-processing as the proposed method)<br /> * comparison to a normalized cut segmentation (with the same post-processing as the proposed method)<br /> * comparison to references 8 and 9.

Ref 8 and 9 don’t have readily usable (https://github.com/LiangHann/USAR) or even shared code (https://github.com/Kaiseem/AD-GAN), so re-implementing this work is well beyond the bounds of this paper. We benchmarked Cellpose, StartDist, SegResNets, and a transformer – SwinURNet. Moreover, models in the MONAI package can be used. Note, to our knowledge the transformer results also are a new contribution that the Reviewer does not acknowledge.

I further strongly encourage the authors to discuss the limitations of their method. From what I understand, the proposed method works only on well-separated objects (due to the semantic segmentation bottleneck), is based on contrastive FG/BG intensity values (due to the SoftNCut loss), and requires tuning of a few parameters (which might be challenging if no ground-truth is available).

We added text on limitations. Thanks for this suggestion.

(2) Dataset

I commend the authors for providing ground-truth labels for more than 2500 cells. I would appreciate it if the Methods section could mention how exactly the cells were labelled. I found a good overlap between the ground truth and Otsu thresholding of the intensity images. Was the ground truth generated by proofreading an initial automatic segmentation, or entirely done by hand? If the former, which method was used to generate the initial segmentation, and are there any concerns that the ground truth might be biased towards a given segmentation method?

In the already submitted version, we have a 5-page DataSet card that fully answers your questions. They are ALL labeled by hand, without any semi-automatic process.

In our main text we even stated “Using whole-brain data from mice we cropped small regions and human annotated in 3D 2,632 neurons that were endogenously labeled by TPH2-tdTomato” - clearly mentioning it is human-annotated.

(3) Napari plugin

The plugin is well-documented and works by following the installation instructions.

Great, thanks for the positive feedback.

However, I was not able to recreate the segmentations reported in the paper with the default settings for the pre-trained WNet3D: segments are generally too large and there are a lot of false positives. Both the prediction and the final instance segmentation also show substantial border artifacts, possibly due to a block-wise processing scheme.

Your review here does not match your comments above; above you said it was working well, such that you doubt the GT is real and the data is too easy as it was perfectly easy to threshold with non-learning methods.

You would need to share more details on what you tried. We suggest following our code; namely, we provide the full experimental code and processing for every figure, as was noted in our original submission: https://github.com/C-Achard/cellseg3d-figures.

Reviewer #2 (Public Review):

Summary:

The authors propose a new method for self-supervised learning of 3d semantic segmentation for fluorescence microscopy. It is based on a WNet architecture (Encoder / Decoder using a UNet for each of these components) that reconstructs the image data after binarization in the bottleneck with a soft n-cuts clustering. They annotate a new dataset for nucleus segmentation in mesoSPIM imaging and train their model on this dataset. They create a napari plugin that provides access to this model and provides additional functionality for training of own models (both supervised and self-supervised), data labeling, and instance segmentation via post-processing of the semantic model predictions. This plugin also provides access to models trained on the contributed dataset in a supervised fashion.

Strengths:

(1) The idea behind the self-supervised learning loss is interesting.

(2) The paper addresses an important challenge. Data annotation is very time-consuming for 3d microscopy data, so a self-supervised method that yields similar results to supervised segmentation would provide massive benefits.

Thank you for highlighting the strengths of our work and new contributions.

Weaknesses:

The experiments presented by the authors do not adequately support the claims made in the paper. There are several shortcomings in the design of the experiment and presentation of the results. Further, it is unclear if results of similar quality as reported can be achieved within the GUI by non-expert users.

Major weaknesses:

(1) The main experiments are conducted on the new mesoSPIM dataset, which contains quite small and well separated nuclei. It is unclear if the good performance of the novel self-supervised learning method compared to CellPose and StarDist would hold for dataset with other characteristics, such as larger nuclei with a more complex morphology or crowded nuclei.

StarDist is not pretrained, we trained it from scratch as we did for WNet3D. We retrained Cellpose and reported the results both with their pretrained model and our best-retrained model. This is documented in Figure 1 and Suppl. Figure 1. We also want to push back and say that they both work very well on this data. In fact, our main claim is not that we beat them, it is that we can match them with a self-supervised method.

Further, additional preprocessing of the mesoSPIM images may improve results for StarDist and CellPose (see the first point in minor weaknesses). Note: having a method that works better for small nuclei would be an important contribution. But I am uncertain the claims hold for larger and/or more crowded nuclei as the current version of the paper implies.

Figure 2 benchmarks our method on larger and denser nuclei, but we do not intend to claim this is a universal tool. It was specifically designed for light-sheet (brain) data, and we have adjusted the title to be more clear. But we also show in Figure 2 it works well on more dense and noisy samples, hinting that it could be a promising approach. But we agree, as-is, it’s unlikely to be good for extremely dense samples like in electron microscopy, which we never claim it would be.

With regards to preprocessing, we respectfully disagree. We trained StarDist (and asked the main developer of StarDist, Martin Weigert, to check our work and he is acknowledged in the paper) and it does very well. Cellpose we also retrained and optimized and we show it works as-well-as leading transformer and CNN-based approaches. Again, we only claimed we can be as good as these methods with an unsupervised approach.

The contribution of the paper would be stronger if a comparison with StarDist / CellPose was also done on the additional datasets from Figure 2.

We appreciate that more datasets would be ideal, but we always feel it’s best for the authors of tools to benchmark their own tools on data. We only compared others in Figure 1 to the new dataset we provide so people get a sense of the quality of the data too; there we did extensive searches for best parameters for those tools. So while we think it would be nice, we will leave it to those authors to be most fair. We also narrowed the scope of our claims to mesoSPIM data (added light-sheet to the title), which none of the other examples in Figure 2 are.

(2) The experimental setup for the additional datasets seems to be unrealistic. In general, the description of these experiments is quite short and so the exact strategy is unclear from the text. However, you write the following: "The channel containing the foreground was then thresholded and the Voronoi-Otsu algorithm used to generate instance labels (for Platynereis data), with hyperparameters based on the Dice metric with the ground truth." I.e., the hyperparameters for the post-processing are found based on the ground truth. From the description it is unclear whether this is done a) on the part of the data that is then also used to compute metrics or b) on a separate validation split that is not used to compute metrics. If a): this is not a valid experimental setup and amounts to training on your test set. If b): this is ok from an experimental point of view, but likely still significantly overestimates the quality of predictions that can be achieved by manual tuning of these hyperparameters by a user that is not themselves a developer of this plugin or an absolute expert in classical image analysis, see also 3. Note that the paper provides notebooks to reproduce the experimental results. This is very laudable, but I believe that a more extended description of the experiments in the text would still be very helpful to understand the set-up for the reader. Further, from inspection of these notebooks it becomes clear that hyper-parameters where indeed found on the testset (a), so the results are not valid in the current form.

We apologize for this confusion; we have now expanded the methods to clarify the setup is now b; you can see what we exactly did as well in the figure notebook: https://c-achard.github.io/cellseg3d-figures/fig2-b-c-extra-datasets/self-supervised-extra.html#threshold-predictions. For clarity, we additionally link each individual notebook now in the Methods.

(3) I cannot obtain similar results to the ones reported in the manuscript using the plugin. I tried to obtain some of the results from the paper qualitatively: First I downloaded one of the volumes from the mesoSPIM dataset (c5image) and applied the WNet3D to it. The prediction looks ok, however the value range is quite narrow (Average BG intensity ~0.4, FG intensity 0.6-0.7). I try to apply the instance segmentation using "Convert to instance labels" from "Utilities". Using "Voronoi-Otsu" does not work due to an error in pyClesperanto ("clGetPlatformIDs failed: PLATFORM_NOT_FOUND_KHR"). Segmentation via "Connected Components" and "Watershed" requires extensive manual tuning to get a somewhat decent result, which is still far from perfect.

We are sorry to hear of the installation issue; pyClesperanto is a dependency that would be required to reproduce the images (sounds like you had this issue; https://forum.image.sc/t/pyclesperanto-prototype-doesnt-work/45724 ) We added to our docs now explicitly the fix: https://github.com/AdaptiveMotorControlLab/CellSeg3D/pull/90. We recommend checking the reproduction notebooks (which were linked in initial submission): https://c-achard.github.io/cellseg3d-figures/intro.html.

Then I tried to obtain the results for the Mouse Skull Nuclei Dataset from EmbedSeg. The results look like a denoised version of the input image, not a semantic segmentation. I was skeptical from the beginning that the method would transfer without retraining, due to the very different morphology of nuclei (much larger and elongated). None of the available segmentation methods yield a good result, the best I can achieve is a strong over-segmentation with watersheds.

- We are surprised to hear this; did you follow the following notebook which directly produces the steps to create this figure? (This was linked in preprint): https://c-achard.github.io/cellseg3d-figures/fig2-c-extra-datasets/self-supervised-extra .html

-  We have made a video demo for you such that any step that might be unclear is also more clear to a user: (https://youtu.be/U2a9IbiO7nE).

-  We also expanded the methods to include the exact values from the notebook into the text.

Minor weaknesses:

(1) CellPose can work better if images are resized so that the median object size in new images matches the training data. For CellPose the cyto2 model should do this automatically. It would be important to report if this was done, and if not would be advisable to check if this can improve results.

We reported this value in Figure 1 and found it to work poorly, that is why we retrained Cellpose and found good performance results (also reported in Figure 1). Resizing GB to TB volumes for mesoSPIM data is otherwise not practical, so simply retraining seems the preferable option, which is what we did.

(2) It is a bit confusing that F1-Score and Dice Score are used interchangeably to evaluate results. The dice score only evaluates semantic predictions, whereas F1-Score evaluates the actual instance segmentation results. I would advise to only use F1-Score, which is the more appropriate metric. For Figure 1f either the mean F1 score over thresholds or F1 @ 0.5 could be reported. Furthermore, I would advise adopting the recommendations on metric reporting from https://www.nature.com/articles/s41592-023-01942-8.

We are using the common metrics in the field for instance and semantic segmentation, and report them in the methods. In Figure 2f we actually report the “Dice” as defined in StarDist (as we stated in the Methods). Note, their implementation is functionally equivalent to F1-Score of an IoU >= 0, so we simply changed this label in the figure now for clarity. We agree this clarifies for the expert readers what was done, and we expanded the methods to be more clear about metrics. We added a link to the paper you mention as well.

(3) A more conceptual limitation is that the (self-supervised) method is limited to intensity-based segmentation, and so will not be able to work for cases where structures cannot be distinguished based on intensity only. It is further unclear how well it can separate crowded nuclei. While some object separation can be achieved by morphological operations this is generally limited for crowded segmentation tasks and the main motivation behind the segmentation objective used in StarDist, CellPose, and other instance segmentation methods. This limitation is only superficially acknowledged in "Note that WNet3D uses brightness to detect objects [...]" but should be discussed in more depth.

Note: this limitation does not mean at all that the underlying contribution is not significant, but I think it is important to address this in more detail so that potential users know where the method is applicable and where it isn't.

We agree, and we added a new section specifically on limitations. Thanks for raising this good point. Thus, while self-supervision comes at the saving of hundreds of manual labor, it comes at the cost of more limited regimes it can work on. Hence why we don’t claim this should replace excellent methods like Cellpose or Stardist, but rather complement them and can be used on mesoSPIM samples, as we show here.

Author Response:

We thank the reviewers for their thoughtful comments on our manuscript. In this provisional response, we aim to address the major concerns raised and outline a plan for a revised version of the manuscript. A more detailed point-by-point response will follow with the revision.

The reviewers appreciated our efforts to combine computational modelling with experimental work. However, they also expressed the need for more clarity in explaining how the model was set up, what was simulated, and what the insights and limitations are. In the revision, we plan to improve the discussion section to clarify all of these points. 

The reviewers also highlighted the need for more transparency regarding the code and the mathematical formulas used in this study. We agree that this is an important issue. While we have already made the software and code for our computational model, along with instructions on how to run it, available in Zenodo (see Ref. 1), and have extensively described the original computational model and formulas in a 13-page supplementary file in our previous study (see Ref. 2), we recognize from the reviewers’ comments that additional transparency is needed. To address this, we will provide an appendix in the revision that includes a full model description, covering the incorporation of cell differentiation and death, a list of parameters, and details on how parameter values were chosen.

Additionally, in the revised manuscript, we will add a paragraph to more thoroughly discuss the limitations of our approach, as well as avenues for future studies. We hope this will clarify both capabilities and limitations of our model in a way that is more  accessible to readers of eLife.

References:

1. Virtual Thymus Model (version 2.0). Published: Jun 14, 2024.  doi:10.5281/zenodo.11656320

2. Aghaallaei, Narges, et al. "αβ/γδ T cell lineage outcome is regulated by intrathymic cell localization and environmental signals." Science Advances 7.29 (2021): eabg3613.

Author response:

Reviewer #1:

Weaknesses:

(1) The activity of the dominant negatives lacks appropriate controls. This is crucial given that mouse mutants for PG5, PG6, PG7, and three of the four PG4 genes show no major effects on limb induction or growth. Understanding these discrepancies is essential.

Given the importance of the Loss of Function (LOF) experiments, we will provide additional evidence for the validity of the dominant-negative strategy and constructs used.

(2) The authors mention redundancies in Hox activity, consistent with numerous previous reports. However, they only use single dominant-negative versions of each Hox paralog gene individually. If Hox4 and Hox5 functions are redundant, experiments should include simultaneous dominant negatives for both groups.

To clarify redundancies in Hox activity, we will test whether simultaneous expression of dominant-negative forms of more than one Hox genes induces a stronger effect compared to the expression of a single dominant-negatives Hox genes.

(3) The main conclusion that Hox4 and Hox5 provide permissive cues on which Hox6/7 induce the forelimb is not sufficiently supported by the data. An experiment expressing simultaneous dnHox4/5 and Hox6/7 is needed. If the hypothesis is correct, this should block Hox6/7's capacity to expand the limb bud or generate an extra bulge.

We agree that this is an excellent additional experiment to corroborate our conclusion and will perform this experiment in our revision.

(4) The identity of the extra bulge or extended limb bud is unclear. The only marker supporting its identity as a forelimb is Tbx5, while other typical limb development markers are absent. Tbx5 is also expressed in other regions besides the forelimb, and its presence does not guarantee forelimb identity. For instance, snakes express Tbx5 in the lateral mesoderm along much of their body axis.

To date, Tbx5 is the best marker for the forelimb. While it is true that the Tbx5 expression is broader than the limb field, this occurs only at early stages before forelimb bud formation. We will work towards a further definition of this extra bulge.

(5) It is important to analyze the skeletons of all embryos to assess the effect of reduced limb buds upon dnHox expression and determine whether extra skeletal elements develop from the extended bud or ectopic bulge.

We have analysed the cartilage structure of operated embryos with GOF experiments and found no skeletal elements within the ectopic wing bud in the neck. Additionally, in our revision, we can further analyse the wing skeleton of operated embryos with LOF experiments, which would provide more detailed assessments of the impact of dominant-negative Hox genes on wing bud formation.

Reviewer #2:

Weaknesses

(1) By contrast to the GOF experiments that induce ectopic limb budding, the LOF experiments, which use dominant negative forms of Hoxa4, Hoxa5, Hoxa6, and Hoxa7, are more challenging to interpret due to the absence of data on the specificity of the dominant negative constructs. Absent such controls, one cannot be certain that effects on limb development are due to disruption of the specific Hox proteins that are being targeted.

We will revise our manuscript to clarify the specificity of the dominant-negative strategy used.

(2) A test of their central hypothesis regarding the necessity and sufficiency of the Hox genes under investigation would be to co-transfect the neck with full-length Hoxa6/a7 AND the dnHoxA4/a5. If their hypothesis is correct, then the dn constructs should block the limb-inducing ability of Hoxa6/a7 overexpression (again, validation of specificity of the DN constructs is important here).

This is an excellent idea and we will implement the experiment in our revision.

(3) The paper could be strengthened by providing some additional data, which should already exist in their RNA-Seq dataset, such as supplementary material that shows the actual gene expression data that are represented in the Venn diagram, heatmap, and GO analysis in Figure 3.

We will incorporate this suggestion and include additional data from our RNA-seq analysis.

(4) The results of these experiments in chick embryos are rather unexpected based on previous knockout experiments in mice, and this needs to be discussed.

In our revision, we will appropriately expand the discussion on the discrepancies observed between knockout mouse models and our chick embryo experiments.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

Fernandez et al. investigate the influence of maternal behavior on bat pup vocal development in Saccopteryx bilineata, a species known to exhibit vocal production learning. The authors performed detailed longitudinal observations of wild mother-pup interactions to ask whether non-vocal maternal displays during juvenile vocal practice or 'babbling', affect vocal production. Specifically, the study examines the durations of pup babbling events and the developmental babbling phase, in relation to the amount of female display behavior, as well as pup age and the number of nearby singing adult males. Furthermore, the authors examine pup vocal repertoire size and maturation in relation to the number of maternal displays encountered during babbling. Statistical models identify female display behavior as a predictor of i) babbling bout duration, ii) the length of the babbling phase, iii) song composition, and iv) syllable maturation. Notably, these outcomes were not influenced by the number of nearby adult males (the pups' source of song models) and were largely independent of general maturation (pup age). These findings highlight the impact of non-vocal aspects of social interactions in guiding mammalian vocal development.

We thank Reviewer 1 for the time and effort dedicated to the revision of our study. The suggestions for the revision of our manuscript were very helpful and will improve our manuscript. 

Strengths:

Historically, work on developmental vocal learning has focused on how juvenile vocalizations are influenced by the sounds produced by nearby adults (often males). In contrast, this study takes the novel approach of examining juvenile vocal ontogeny in relation to non-vocal maternal behavior, in one of the few mammals known to exhibit vocal production learning. The authors collected an impressive dataset from multiple wild bat colonies in two Central American countries. This includes longitudinal acoustic recordings and behavioral monitoring of individual mother-pup pairs, across development.

The identified relationships between maternal behavior and bat pup vocalizations have intriguing implications for understanding the mechanisms that enable vocal production learning in mammals, including human speech acquisition. As such, these findings are likely to be relevant to a broad audience interested in the evolution and development of social behavior as well as sensory-motor learning.

We thank reviewer 1 for this assessment. 

Weaknesses:

The authors qualitatively describe specific patterns of female displays during pup babbling, however, subsequent quantitative analyses are based on two aggregate measures of female behavior that pool across display types. Consequently, it remains unclear how certain maternal behaviors might differentially influence pup vocalizations (e.g. through specific feedback contingencies or more general modulation of pup behavioral states).

In analyzing the effects of maternal behavior on song maturation, the authors focus on the most common syllable type produced across pups. This approach is justified based on the syllable variability within and across individuals, however, additional quantification and visual presentation of categorized syllable data would improve clarity and potentially strengthen resulting claims.

We agree that our analysis of maternal behaviour does not investigate potential contingencies between particular maternal behavioural displays and pup vocalizations (e.g.

particular syllable types). Our data collected for this study on maternal behaviour includes direct observations, field notes and/or video recordings. In the future, it will be necessary to work with high-speed cameras for the analysis of potential contingencies between particular maternal behavioural displays and specific pup vocalizations, which allow this kind of fine-detailed analysis. We have planned future studies investigating whether pup vocalizations elicit contingent maternal responses or vice versa. In the revision of our manuscript, we will include a comment pointing out that this special behaviour will be investigated in greater detail in the future. 

As suggested by reviewer 1, in our revised manuscript we will include more information on methods to improve understandability. In particular, we will:

- present more information on different steps of our acoustic analyses

- provide additional and clearer spectrogram figures representing the different syllable types and categorizations 

- change the figures accompanying our GLMM analyses following the suggestion of Reviewer 1

Reviewer #2 (Public review):

Summary:

This study explores how maternal behaviors influence vocal learning in the greater sac-winged bat (Saccopteryx bilineata). Over two field seasons, researchers tracked 19 bat pups from six wild colonies, examining vocal development aspects such as vocal practice duration, syllable repertoire size, and song syllable acquisition. The findings show that maternal behaviors significantly impact the length of daily babbling sessions and the overall babbling phase, while the presence of adult male tutors does not.

The researchers conducted detailed acoustic analyses, categorizing syllables and evaluating the variety and presence of learned song syllables. They discovered that maternal interactions enhance both the number and diversity of learned syllables and the production of mature syllables in the pups' vocalizations. A notable correlation was found between the extent of acoustic changes in the most common learned syllable type and maternal activity, highlighting the key role of maternal feedback in shaping pups' vocal development.

In summary, this study emphasizes the crucial role of maternal social feedback in the vocal development of S. bilineata. Maternal behaviors not only increase vocal practice but also aid in acquiring and refining a complex vocal repertoire. These insights enhance our understanding of social interactions in mammalian vocal learning and draw interesting parallels between bat and human vocal development.

We thank reviewer 2 for his/her time and effort dedicated to the revision of our study. The suggestions were very helpful in improving our manuscript. 

Strengths:

This paper makes significant contributions to the field of vocal learning by looking at the role of maternal behaviors in shaping the vocal learning phenotype of Saccopteryx bilineata. The paper uses a longitudinal approach, tracking the vocal ontogeny of bat pups from birth to weaning across six colonies and two field seasons, allowing the authors to assess how maternal interactions influence various aspects of vocal practice and learning, providing strong empirical evidence for the critical role of social feedback in non-human mammalian vocal learners. This kind of evidence highlights the complexity of the vocal learning phenotype and shows that it goes beyond the right auditory experience and having the right circuitry.

The paper offers a nuanced understanding of how specific maternal behaviors impact the acquisition and refinement of the vocal repertoire, while showing the number of male tutors - the source of adult song - did not have much of an effect. The correlation between maternal activity and acoustic changes in learned syllable types is a novel finding that underscores the importance of non-vocal social interactions in vocal learning. In vocal learning research, with some notable exceptions, experience is often understood as auditory experience. This paper highlights how, even though that is one important piece of the puzzle, other kinds of experience directly affect the development of vocal behavior. This is of particular importance in the case of a mammalian species such as Saccopteryx bilineata, as this kind of result is perhaps more often associated with avian species.

Moreover, the study's findings have broader implications for our understanding of vocal learning across species. By drawing parallels between bat and human vocal development (and in some ways to bird vocal development), the paper highlights common mechanisms that may underlie vocal practice and learning in both humans and other mammals. This interdisciplinary perspective enriches the field and encourages further comparative studies, ultimately advancing our knowledge of the evolutionary and developmental processes that shape vocal productive learning in all its dimensions.

Weaknesses:

Some weaknesses can be pointed out, but in fairness, the authors acknowledge them in one way or another. As such, these are not flaws per se, but gaps that can be filled with further research.

Experimental manipulations, such as controlled playback experiments or controlled environments, could strengthen the causal claims by directly testing the effects of specific maternal behaviors on vocal development. Certainly, the strengths of the paper will be consolidated after such work is performed.

The reliance on the number of singing males as a proxy for social acoustic input. This measure does not account for the variability in the quality, frequency, or duration of the male songs to which the pups are exposed. A more detailed analysis of the acoustic environment, including direct measurements of song exposure and its impact on vocal learning, would provide a clearer understanding of the role of male tutors.

Finally, and although it would be unlikely that these results are unique to Saccopteryx bilineata, the study's focus on a single species limits at present the generalizability of some of its findings to other vocal learning mammals. While the parallels drawn between bat and human vocal development are intriguing, the conclusions will be more robust when supported by comparative studies involving multiple species of vocal learners. This will help to identify whether the observed maternal influences on vocal development reported here are unique to Saccopteryx bilineata or represent a broader phenomenon in chiropteran, mammalian, or general vocal learning. Expanding the scope of research to include a wider range of species and incorporating cross-species comparisons will significantly enhance the contribution of this study to the field of vocal learning.

Thank you for your suggestions and comments. 

Regarding your main comment 1: In the future, we plan to implement temporary captivity experiments to investigate how maternal behaviours affect pup vocal development. This study provides the necessary basis for conducting future playback studies investigating specific behaviours in a controlled environment.

Regarding your main comment 2: We completely agree that the number of singing males only represents a proxy for acoustic input that pups receive during ontogeny. In the future, we plan to investigate in detail how the acoustic landscape influences pup vocal development and learning. This will include quantifying how long pups are exposed to song during ontogeny and, assessing the influence of different tutors, including a detailed analysis of song syllables of the adult tutors to compare it to vocal trajectories of song syllables in pups. 

Regarding your main comment 3: We also fully agree that it is unlikely that these results are unique to Saccopteryx bilineata. We are certain that other mammalian vocal learners show parallels to the vocal development and learning processes of S. bilineata. Especially bats are a promising taxon for comparative studies because their vocal production and perception systems are highly sophisticated (due to their ability to echolocate). The high sociability of this taxon also includes a variety of social systems and vocal capacities (e.g. regarding vocal repertoire size, vocal learning capacities, information content, etc.) which support social learning and social feedback – as shown in our study. 

As suggested, in our revised manuscript we will include information on the validation of the ethogram. Furthermore, we will correct all the spelling mistakes – thank you very much for pointing them out!

Author response:

We appreciate all the reviewers for their encouraging comments and thoughtful feedback. We are confident that we can incorporate many of the suggestions to provide a clearer overall picture in the revised manuscript. In particular, we agree with the reviewers' concern that some of our methodological decisions, including our choice of metrics, require further clarification. We will focus on revising the methods section to make these decisions more transparent and to address any misunderstandings related to the analysis.

We also value the request to include more data, such as intermediate results and additional control analyses. We will carefully assess which results to include in the main manuscript and which to provide in an extended supplementary section.

To offer a more detailed understanding of our quantification of "prediction tendency," we refer to our previous work (Schubert et al., 2023, 2024), where we elaborate on our analytical choices in great detail and provide additional control analyses (e.g., ensuring that the relationship with speech tracking is not driven by participants' signal-to-noise ratio; Schubert et al., 2023).

Additionally, we would like to clarify that the aim of this manuscript is not to analyze viewing behavior in depth but to replicate the general finding of ocular speech tracking, as presented in Gehmacher et al. (2024). A thorough investigation of specific ocular contributions (e.g., microsaccades or blinks) would require a separate research question and distinct analysis approaches, given the binary nature of such events.

Nevertheless, we share the reviewers' interest in independent results from the current study, and we plan to carefully select and present the most relevant findings in the revised manuscript.

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

In this paper, Misic et al showed that white matter properties can be used to classify subacute back pain patients that will develop persisting pain.

Strengths:

Compared to most previous papers studying associations between white matter properties and chronic pain, the strength of the method is to perform a prediction in unseen data. Another strength of the paper is the use of three different cohorts. This is an interesting paper that provides a valuable contribution to the field.

We thank the reviewer for emphasizing the strength of our paper and the importance of validation on multiple unseen cohorts.

Weaknesses:

The authors imply that their biomarker could outperform traditional questionnaires to predict pain: "While these models are of great value showing that few of these variables (e.g. work factors) might have significant prognostic power on the long-term outcome of back pain and provide easy-to-use brief questionnaires-based tools, (21, 25) parameters often explain no more than 30% of the variance (28-30) and their prognostic accuracy is limited.(31)". I don't think this is correct; questionnaire-based tools can achieve far greater prediction than their model in about half a million individuals from the UK Biobank (Tanguay-Sabourin et al., A prognostic risk score for the development and spread of chronic pain, Nature Medicine 2023).

We agree with the reviewer that we might have under-estimated the prognostic accuracy of questionnaire-based tools, especially, the strong predictive accuracy shown by Tangay-Sabourin 2023.  In this revised version, we have changed both the introduction and the discussion to reflect the questionnaire-based prognostic accuracy reported in the seminal work by Tangay-Sabourin. 

In the introduction (page 4, lines 3-18), we now write:

“Some studies have addressed this question with prognostic models incorporating demographic, pain-related, and psychosocial predictors.1-4 While these models are of great value showing that few of these variables (e.g. work factors) might have significant prognostic power on the long-term outcome of back pain, their prognostic accuracy is limited,5 with parameters often explaining no more than 30% of the variance.6-8. A recent notable study in this regard developed a model based on easy-to-use brief questionnaires to predict the development and spread of chronic pain in a variety of pain conditions capitalizing on a large dataset obtained from the UK-BioBank. 9 This work demonstrated that only few features related to assessment of sleep, neuroticism, mood, stress, and body mass index were enough to predict persistence and spread of pain with an area under the curve of 0.53-0.73. Yet, this study is unique in showing such a predictive value of questionnaire-based tools. Neurobiological measures could therefore complement existing prognostic models based on psychosocial variables to improve overall accuracy and discriminative power. More importantly, neurobiological factors such as brain parameters can provide a mechanistic understanding of chronicity and its central processing.”

And in the conclusion (page 22, lines 5-9), we write:

“Integrating findings from studies that used questionnaire-based tools and showed remarkable predictive power9 with neurobiological measures that can offer mechanistic insights into chronic pain development, could enhance predictive power in CBP prognostic modeling.”

Moreover, the main weakness of this study is the sample size. It remains small despite having 3 cohorts. This is problematic because results are often overfitted in such a small sample size brain imaging study, especially when all the data are available to the authors at the time of training the model (Poldrack et al., Scanning the horizon: towards transparent and reproducible neuroimaging research, Nature Reviews in Neuroscience 2017). Thus, having access to all the data, the authors have a high degree of flexibility in data analysis, as they can retrain their model any number of times until it generalizes across all three cohorts. In this case, the testing set could easily become part of the training making it difficult to assess the real performance, especially for small sample size studies.

The reviewer raises a very important point of limited sample size and of the methodology intrinsic of model development and testing. We acknowledge the small sample size in the “Limitations” section of the discussion.   In the resubmission, we acknowledge the degree of flexibility that is afforded by having access to all the data at once. However, we also note that our SLF-FA based model is a simple cut-off approach that does not include any learning or hidden layers and that the data obtained from Open Pain were never part of the “training” set at any point at either the New Haven or the Mannheim site.  Regarding our SVC approach we follow standard procedures for machine learning where we never mix the training and testing sets. The models are trained on the training data with parameters selected based on cross-validation within the training data. Therefore, no models have ever seen the test data set. The model performances we reported reflect the prognostic accuracy of our model. We write in the limitation section of the discussion (page 20, lines 20-21, and page 21, lines 1-6):

“In addition, at the time of analysis, we had “access” to all the data, which may lead to bias in model training and development.  We believe that the data presented here are nevertheless robust since multisite validated but need replication. Additionally, we followed standard procedures for machine learning where we never mix the training and testing sets. The models were trained on the training data with parameters selected based on cross-validation within the training data. Therefore, no models have ever seen the test data set. The model performances we reported reflect the prognostic accuracy of our model”. 

Finally, as discussed by Spisak et al., 10 the key determinant of the required sample size in predictive modeling is the ” true effect size of the brain-phenotype relationship”, which we think is the determinant of the replication we observe in this study. As such the effect size in the New Haven and Mannheim data is Cohen’s d >1.

Even if the performance was properly assessed, their models show AUCs between 0.65-0.70, which is usually considered as poor, and most likely without potential clinical use. Despite this, their conclusion was: "This biomarker is easy to obtain (~10 min of scanning time) and opens the door for translation into clinical practice." One may ask who is really willing to use an MRI signature with a relatively poor performance that can be outperformed by self-report questionnaires?

The reviewer is correct, the model performance is fair which limits its usefulness for clinical translation.  We wanted to emphasize that obtaining diffusion images can be done in a short period of time and, hence, as such models’ predictive accuracy improves, clinical translation becomes closer to reality. In addition, our findings are based on older diffusion data and limited sample sizes coming from different sites and different acquisition sequences.  This by itself would limit the accuracy especially since the evidence shows that sample size affects also model performance (i.e. testing AUC)10.  In the revision, we re-worded the sentence mentioned by the reviewer to reflect the points discussed here. This also motivates us to collect a more homogeneous and larger sample.  In the limitations section of the discussion, we now write (page 21, lines 6-9):

“Even though our model performance is fair, which currently limits its usefulness for clinical translation, we believe that future models would further improve accuracy by using larger homogenous sample sizes and uniform acquisition sequences.”

Overall, these criticisms are more about the wording sometimes used and the inference they made. I think the strength of the evidence is incomplete to support the main claims of the paper.

Despite these limitations, I still think this is a very relevant contribution to the field. Showing predictive performance through cross-validation and testing in multiple cohorts is not an easy task and this is a strong effort by the team. I strongly believe this approach is the right one and I believe the authors did a good job.

We thank the reviewer for acknowledging that our effort and approach were useful.

Minor points:

Methods:

I get the voxel-wise analysis, but I don't understand the methods for the structural connectivity analysis between the 88 ROIs. Have the authors run tractography or have they used a predetermined streamlined form of 'population-based connectome'? They report that models of AUC above 0.75 were considered and tested in the Chicago dataset, but we have no information about what the model actually learned (although this can be tricky for decision tree algorithms). 

We apologize for the lack of clarity; we did run tractography and we did not use a pre-determined streamlined form of the connectome.

Finding which connections are important for the classification of SBPr and SBPp is difficult because of our choices during data preprocessing and SVC model development: (1) preprocessing steps which included TNPCA for dimensionality reduction, and regressing out the confounders (i.e., age, sex, and head motion); (2) the harmonization for effects of sites; and (3) the Support Vector Classifier which is a hard classification model11.

In the methods section (page 30, lines 21-23) we added: “Of note, such models cannot tell us the features that are important in classifying the groups.  Hence, our model is considered a black-box predictive model like neural networks.”

Minor:

What results are shown in Figure 7? It looks more descriptive than the actual results.

The reviewer is correct; Figure 7 and Supplementary Figure 4 were both qualitatively illustrating the shape of the SLF. We have now changed both figures in response to this point and a point raised by reviewer 3.  We now show a 3D depiction of different sub-components of the right SLF (Figure 7) and left SLF (Now Supplementary Figure 11 instead of Supplementary Figure 4) with a quantitative estimation of the FA content of the tracts, and the number of tracts per component.  The results reinforce the TBSS analysis in showing asymmetry in the differences between left and right SLF between the groups (i.e. SBPp and SBPr) in both FA values and number of tracts per bundle.

Reviewer #2 (Public Review):

The present study aims to investigate brain white matter predictors of back pain chronicity. To this end, a discovery cohort of 28 patients with subacute back pain (SBP) was studied using white matter diffusion imaging. The cohort was investigated at baseline and one-year follow-up when 16 patients had recovered (SBPr) and 12 had persistent back pain (SBPp). A comparison of baseline scans revealed that SBPr patients had higher fractional anisotropy values in the right superior longitudinal fasciculus SLF) than SBPp patients and that FA values predicted changes in pain severity. Moreover, the FA values of SBPr patients were larger than those of healthy participants, suggesting a role of FA of the SLF in resilience to chronic pain. These findings were replicated in two other independent datasets. The authors conclude that the right SLF might be a robust predictive biomarker of CBP development with the potential for clinical translation.

Developing predictive biomarkers for pain chronicity is an interesting, timely, and potentially clinically relevant topic. The paradigm and the analysis are sound, the results are convincing, and the interpretation is adequate. A particular strength of the study is the discovery-replication approach with replications of the findings in two independent datasets.

We thank reviewer 2 for pointing to the strength of our study.

The following revisions might help to improve the manuscript further.

- Definition of recovery. In the New Haven and Chicago datasets, SBPr and SBPp patients are distinguished by reductions of >30% in pain intensity. In contrast, in the Mannheim dataset, both groups are distinguished by reductions of >20%. This should be harmonized. Moreover, as there is no established definition of recovery (reference 79 does not provide a clear criterion), it would be interesting to know whether the results hold for different definitions of recovery. Control analyses for different thresholds could strengthen the robustness of the findings.

The reviewer raises an important point regarding the definition of recovery.  To address the reviewers’ concern we have added a supplementary figure (Fig. S6) showing the results in the Mannheim data set if a 30% reduction is used as a recovery criterion, and in the manuscript (page 11, lines 1,2) we write: “Supplementary Figure S6 shows the results in the Mannheim data set if a 30% reduction is used as a recovery criterion in this dataset (AUC= 0.53)”.

We would like to emphasize here several points that support the use of different recovery thresholds between New Haven and Mannheim.  The New Haven primary pain ratings relied on visual analogue scale (VAS) while the Mannheim data relied on the German version of the West-Haven-Yale Multidimensional Pain Inventory. In addition, the Mannheim data were pre-registered with a definition of recovery at 20% and are part of a larger sub-acute to chronic pain study with prior publications from this cohort using the 20% cut-off12. Finally, a more recent consensus publication13 from IMMPACT indicates that a change of at least 30% is needed for a moderate improvement in pain on the 0-10 Numerical Rating Scale but that this percentage depends on baseline pain levels.

- Analysis of the Chicago dataset. The manuscript includes results on FA values and their association with pain severity for the New Haven and Mannheim datasets but not for the Chicago dataset. It would be straightforward to show figures like Figures 1 - 4 for the Chicago dataset, as well.

We welcome the reviewer’s suggestion; we added these analyses to the results section of the resubmitted manuscript (page 11, lines 13-16): “The correlation between FA values in the right SLF and pain severity in the Chicago data set showed marginal significance (p = 0.055) at visit 1 (Fig. S8A) and higher FA values were significantly associated with a greater reduction in pain at visit 2 (p = 0.035) (Fig. S8B).”

- Data sharing. The discovery-replication approach of the present study distinguishes the present from previous approaches. This approach enhances the belief in the robustness of the findings. This belief would be further enhanced by making the data openly available. It would be extremely valuable for the community if other researchers could reproduce and replicate the findings without restrictions. It is not clear why the fact that the studies are ongoing prevents the unrestricted sharing of the data used in the present study.

We greatly appreciate the reviewer's suggestion to share our data sets, as we strongly support the Open Science initiative. The Chicago data set is already publicly available. The New Haven data set will be shared on the Open Pain repository, and the Mannheim data set will be uploaded to heiDATA or heiARCHIVE at Heidelberg University in the near future. We cannot share the data immediately because this project is part of the Heidelberg pain consortium, “SFB 1158: From nociception to chronic pain: Structure-function properties of neural pathways and their reorganization.” Within this consortium, all data must be shared following a harmonized structure across projects, and no study will be published openly until all projects have completed initial analysis and quality control.

Reviewer #3 (Public Review):

Summary:

Authors suggest a new biomarker of chronic back pain with the option to predict the result of treatment. The authors found a significant difference in a fractional anisotropy measure in superior longitudinal fasciculus for recovered patients with chronic back pain.

Strengths:

The results were reproduced in three different groups at different studies/sites.

Weaknesses:

- The number of participants is still low.

The reviewer raises a very important point of limited sample size. As discussed in our replies to reviewer number 1:

We acknowledge the small sample size in the “Limitations” section of the discussion.   In the resubmission, we acknowledge the degree of flexibility that is afforded by having access to all the data at once. However, we also note that our SLF-FA based model is a simple cut-off approach that does not include any learning or hidden layers and that the data obtained from Open Pain were never part of the “training” set at any point at either the New Haven or the Mannheim site.  Regarding our SVC approach we follow standard procedures for machine learning where we never mix the training and testing sets. The models are trained on the training data with parameters selected based on cross-validation within the training data. Therefore, no models have ever seen the test data set. The model performances we reported reflect the prognostic accuracy of our model. We write in the limitation section of the discussion (page 20, lines 20-21, and page 21, lines 1-6):

“In addition, at the time of analysis, we had “access” to all the data, which may lead to bias in model training and development.  We believe that the data presented here are nevertheless robust since multisite validated but need replication. Additionally, we followed standard procedures for machine learning where we never mix the training and testing sets. The models were trained on the training data with parameters selected based on cross-validation within the training data. Therefore, no models have ever seen the test data set. The model performances we reported reflect the prognostic accuracy of our model”. 

Finally, as discussed by Spisak et al., 10 the key determinant of the required sample size in predictive modeling is the ” true effect size of the brain-phenotype relationship”, which we think is the determinant of the replication we observe in this study. As such the effect size in the New Haven and Mannheim data is Cohen’s d >1.

- An explanation of microstructure changes was not given.

The reviewer points to an important gap in our discussion.  While we cannot do a direct study of actual tissue microstructure, we explored further the changes observed in the SLF by calculating diffusivity measures. We have now performed the analysis of mean, axial, and radial diffusivity. 

In the results section we added (page 7, lines 12-19): “We also examined mean diffusivity (MD), axial diffusivity (AD), and radial diffusivity (RD) extracted from the right SLF shown in Fig.1 to further understand which diffusion component is different between the groups. The right SLF MD is significantly increased (p < 0.05) in the SBPr compared to SBPp patients (Fig. S3), while the right SLF RD is significantly decreased (p < 0.05) in the SBPr compared to SBPp patients in the New Haven data (Fig. S4). Axial diffusivity extracted from the RSLF mask did not show significant difference between SBPr and SBPp (p = 0.28) (Fig. S5).”

In the discussion, we write (page 15, lines 10-20):

“Within the significant cluster in the discovery data set, MD was significantly increased, while RD in the right SLF was significantly decreased in SBPr compared to SBPp patients. Higher RD values, indicative of demyelination, were previously observed in chronic musculoskeletal patients across several bundles, including the superior longitudinal fasciculus14.  Similarly, Mansour et al. found higher RD in SBPp compared to SBPr in the predictive FA cluster. While they noted decreased AD and increased MD in SBPp, suggestive of both demyelination and altered axonal tracts,15 our results show increased MD and RD in SBPr with no AD differences between SBPp and SBPr, pointing to white matter changes primarily due to myelin disruption rather than axonal loss, or more complex processes. Further studies on tissue microstructure in chronic pain development are needed to elucidate these processes.”

- Some technical drawbacks are presented.

We are uncertain if the reviewer is suggesting that we have acknowledged certain technical drawbacks and expects further elaboration on our part. We kindly request that the reviewer specify what particular issues need to be addressed so that we can respond appropriately.

Recommendations For The Authors:

We thank the reviewers for their constructive feedback, which has significantly improved our manuscript. We have done our best to answer the criticisms that they raised point-by-point.

Reviewer #2 (Recommendations For The Authors):

The discovery-replication approach of the current study justifies the use of the terminus 'robust.' In contrast, previous studies on predictive biomarkers using functional and structural brain imaging did not pursue similar approaches and have not been replicated. Still, the respective biomarkers are repeatedly referred to as 'robust.' Throughout the manuscript, it would, therefore, be more appropriate to remove the label 'robust' from those studies.

We thank the reviewer for this valuable suggestion. We removed the label 'robust' throughout the manuscript when referring to the previous studies which didn’t follow the same approach and have not yet been replicated.

Reviewer #3 (Recommendations For The Authors):

This is, indeed, quite a well-written manuscript with very interesting findings and patient group. There are a few comments that enfeeble the findings.

(1) It is a bit frustrating to read at the beginning how important chronic back pain is and the number of patients in the used studies. At least the number of healthy subjects could be higher.

The reviewer raises an important point regarding the number of pain-free healthy controls (HC) in our samples. We first note that our primary statistical analysis focused on comparing recovered and persistent patients at baseline and validating these findings across sites without directly comparing them to HCs. Nevertheless, the data from New Haven included 28 HCs at baseline, and the data from Mannheim included 24 HCs. Although these sample sizes are not large, they have enabled us to clearly establish that the recovered SBPr patients generally have larger FA values in the right superior longitudinal fasciculus compared to the HCs, a finding consistent across sites (see Figs. 1 and 3). This suggests that the general pain-free population includes individuals with both low and high-risk potential for chronic pain. It also offers one explanation for the reported lack of differences or inconsistent differences between chronic low-back pain patients and HCs in the literature, as these differences likely depend on the (unknown) proportion of high- and low-risk individuals in the control groups. Therefore, if the high-risk group is more represented by chance in the HC group, comparisons between HCs and chronic pain patients are unlikely to yield statistically significant results. Thus, while we agree with the reviewer that the sample sizes of our HCs are limited, this limitation does not undermine the validity of our findings.

(2) Pain reaction in the brain is in general a quite popular topic and could be connected to the findings or mentioned in the introduction.

We thank the reviewer for this suggestion.  We have now added a summary of brain response to pain in general; In the introduction, we now write (page 4, lines 19-22 and page 5, lines 1-5):

“Neuroimaging research on chronic pain has uncovered a shift in brain responses to pain when acute and chronic pain are compared. The thalamus, primary somatosensory, motor areas, insula, and mid-cingulate cortex most often respond to acute pain and can predict the perception of acute pain16-19. Conversely, limbic brain areas are more frequently engaged when patients report the intensity of their clinical pain20, 21. Consistent findings have demonstrated that increased prefrontal-limbic functional connectivity during episodes of heightened subacute ongoing back pain or during a reward learning task is a significant predictor of CBP.12, 22. Furthermore, low somatosensory cortex excitability in the acute stage of low back pain was identified as a predictor of CBP chronicity.23”

(3) It is clearly observed structural asymmetry in the brain, why not elaborate this finding further? Would SLF be a hub in connectivity analysis? Would FA changes have along tract features? etc etc etc

The reviewer raises an important point. There is ground to suggest from our data that there is an asymmetry to the role of the SLF in resilience to chronic pain. We discuss this at length in the Discussion section. We have, in addition, we elaborated more in our data analysis using our Population Based Structural Connectome pipeline on the New Haven dataset. Following that approach, we studied both the number of fiber tracts making different parts of the SLF on the right and left side. In addition, we have extracted FA values along fiber tracts and compared the average across groups. Our new analyses are presented in our modified Figures 7 and Fig S11.  These results support the asymmetry hypothesis indeed. The SLF could be a hub of structural connectivity. Please note however, given the nature of our design of discovery and validation, the study of structural connectivity of the SLF is beyond the scope of this paper because tract-based connectivity is very sensitive to data collection parameters and is less accurate with single shell DWI acquisition. Therefore, we will pursue the study of connectivity of the SLF in the future with well-powered and more harmonized data.

(4) Only FA is mentioned; did the authors work with MD, RD, and AD metrics?

We thank the reviewer for this suggestion that helps in providing a clearer picture of the differences in the right SLF between SBPr and SBPp. We have now extracted MD, AD, and RD for the predictive mask we discovered in Figure 1 and plotted the values comparing SBPr to SBPp patients in Fig. S3, Fig. S4., and Fig. S5 across all sites using one comprehensive harmonized analysis. We have added in the discussion “Within the significant cluster in the discovery data set, MD was significantly increased, while RD in the right SLF was significantly decreased in SBPr compared to SBPp patients. Higher RD values, indicative of demyelination, were previously observed in chronic musculoskeletal patients across several bundles, including the superior longitudinal fasciculus14.  Similarly, Mansour et al. found higher RD in SBPp compared to SBPr in the predictive FA cluster. While they noted decreased AD and increased MD in SBPp, suggestive of both demyelination and altered axonal tracts15, our results show increased MD and RD in SBPr with no AD differences between SBPp and SBPr, pointing to white matter changes primarily due to myelin disruption rather than axonal loss, or more complex processes. Further studies on tissue microstructure in chronic pain development are needed to elucidate these processes.”

(5) There are many speculations in the Discussion, however, some of them are not supported by the results.

We agree with the reviewer and thank them for pointing this out. We have now made several changes across the discussion related to the wording where speculations were not supported by the data. For example, instead of writing (page 16, lines 7-9): “Together the literature on the right SLF role in higher cognitive functions suggests, therefore, that resilience to chronic pain is a top-down phenomenon related to visuospatial and body awareness.”, We write: “Together the literature on the right SLF role in higher cognitive functions suggests, therefore, that resilience to chronic pain might be related to a top-down phenomenon involving visuospatial and body awareness.”

(6) A method section was written quite roughly. In order to obtain all the details for a potential replication one needs to jump over the text.

The reviewer is correct; our methodology may have lacked more detailed descriptions.  Therefore, we have clarified our methodology more extensively.  Under “Estimation of structural connectivity”; we now write (page 28, lines 20,21 and page 29, lines 1-19):

“Structural connectivity was estimated from the diffusion tensor data using a population-based structural connectome (PSC) detailed in a previous publication.24 PSC can utilize the geometric information of streamlines, including shape, size, and location for a better parcellation-based connectome analysis. It, therefore, preserves the geometric information, which is crucial for quantifying brain connectivity and understanding variation across subjects. We have previously shown that the PSC pipeline is robust and reproducible across large data sets.24 PSC output uses the Desikan-Killiany atlas (DKA) 25 of cortical and sub-cortical regions of interest (ROI). The DKA parcellation comprises 68 cortical surface regions (34 nodes per hemisphere) and 19 subcortical regions. The complete list of ROIs is provided in the supplementary materials’ Table S6.  PSC leverages a reproducible probabilistic tractography algorithm 26 to create whole-brain tractography data, integrating anatomical details from high-resolution T1 images to minimize bias in the tractography. We utilized DKA 25 to define the ROIs corresponding to the nodes in the structural connectome. For each pair of ROIs, we extracted the streamlines connecting them by following these steps: 1) dilating each gray matter ROI to include a small portion of white matter regions, 2) segmenting streamlines connecting multiple ROIs to extract the correct and complete pathway, and 3) removing apparent outlier streamlines. Due to its widespread use in brain imaging studies27, 28, we examined the mean fractional anisotropy (FA) value along streamlines and the count of streamlines in this work. The output we used includes fiber count, fiber length, and fiber volume shared between the ROIs in addition to measures of fractional anisotropy and mean diffusivity.”

(7) Why not join all the data with harmonisation in order to reproduce the results (TBSS)

We have followed the reviewer’s suggestion; we used neuroCombat harmonization after pooling all the diffusion weighted data into one TBSS analysis. Our results remain the same after harmonization. 

In the Supplementary Information we added a paragraph explaining the method for harmonization; we write (SI, page 3, lines 25-34):

“Harmonization of DTI data using neuroCombat. Because the 3 data sets originated from different sites using different MR data acquisition parameters and slightly different recruitment criteria, we applied neuroCombat 29  to correct for site effects and then repeated the TBSS analysis shown in Figure 1 and the validation analyses shown in Figures 5 and 6. First, the FA maps derived using the FDT toolbox were pooled into one TBSS analysis where registration to a standard template FA template (FMRIB58_FA_1mm.nii.gz part of FSL) was performed.  Next, neuroCombat was applied to the FA maps as implemented in Python with batch (i.e., site) effect modeled with a vector containing 1 for New Haven, 2 for Chicago, and 3 for Mannheim originating maps, respectively. The harmonized maps were then skeletonized to allow for TBSS.”

And in the results section, we write (page 12, lines 2-21):

“Validation after harmonization

Because the DTI data sets originated from 3 sites with different MR acquisition parameters, we repeated our TBSS and validation analyses after correcting for variability arising from site differences using DTI data harmonization as implemented in neuroCombat. 29 The method of harmonization is described in detail in the Supplementary Methods. The whole brain unpaired t-test depicted in Figure 1 was repeated after neuroCombat and yielded very similar results (Fig. S9A) showing significantly increased FA in the SBPr compared to SBPp patients in the right superior longitudinal fasciculus (MNI-coordinates of peak voxel: x = 40; y = - 42; z = 18 mm; t(max) = 2.52; p < 0.05, corrected against 10,000 permutations).  We again tested the accuracy of local diffusion properties (FA) of the right SLF extracted from the mask of voxels passing threshold in the New Haven data (Fig.S9A) in classifying the Mannheim and the Chicago patients, respectively, into persistent and recovered. FA values corrected for age, gender, and head displacement accurately classified SBPr  and SBPp patients from the Mannheim data set with an AUC = 0.67 (p = 0.023, tested against 10,000 random permutations, Fig. S9B and S7D), and patients from the Chicago data set with an AUC = 0.69 (p = 0.0068) (Fig. S9C and S7E) at baseline, and an AUC = 0.67 (p = 0.0098)  (Fig. S9D and S7F) patients at follow-up,  confirming the predictive cluster from the right SLF across sites. The application of neuroCombat significantly changes the FA values as shown in Fig.S10 but does not change the results between groups.”

Minor comments

(1) In the case of New Haven data, one used MB 4 and GRAPPA 2, these two factors accelerate the imaging 8 times and often lead to quite a poor quality.<br /> Any kind of QA?

We thank the reviewer for identifying this error. GRAPPA 2 was in fact used for our T1-MPRAGE image acquisition but not during the diffusion data acquisition. The diffusion data were acquired with a multi-band acceleration factor of 4.  We have now corrected this mistake.

(2) Why not include MPRAGE data into the analysis, in particular, for predictions?

We thank the reviewer for the suggestion. The collaboration on this paper was set around diffusion data. In addition, MPRAGE data from New Haven related to prediction is already published (10.1073/pnas.1918682117) and MPRAGE data of the Mannheim data set is a part of the larger project and will be published elsewhere.

(3) In preprocessing, the authors wrote: "Eddy current corrects for image distortions due to susceptibility-induced distortions and eddy currents in the gradient coil"<br /> However, they did not mention that they acquired phase-opposite b0 data. It means eddy_openmp works likely only as an alignment tool, but not susceptibility corrector.

We kindly thank the reviewer for bringing this to our attention. We indeed did not collect b0 data in the phase-opposite direction, however, eddy_openmp can still be used to correct for eddy current distortions and perform motion correction, but the absence of phase-opposite b0 data may limit its ability to fully address susceptibility artifacts. This is now noted in the Supplementary Methods under Preprocessing section (SI, page 3, lines 16-18): “We do note, however, that as we did not acquire data in the phase-opposite direction, the susceptibility-induced distortions may not be fully corrected.”

(4) Version of FSL?

We thank the reviewer for addressing this point that we have now added under the Supplementary Methods (SI, page 3, lines 10-11): “Preprocessing of all data sets was performed employing the same procedures and the FMRIB diffusion toolbox (FDT) running on FSL version 6.0.”

(5) Some short sketches about the connectivity analysis could be useful, at least in SI.

We are grateful for this suggestion that improves our work. We added the sketches about the connectivity analysis, please see Figure 7 and Supplementary Figure 11.

(6) Machine learning: functions, language, version?

We thank the reviewer for pointing out these minor points that we now hope to have addressed in our resubmission in the Methods section by adding a detailed description of the structural connectivity analysis. We added: “The DKA parcellation comprises 68 cortical surface regions (34 nodes per hemisphere) and 19 subcortical regions. The complete list of ROIs is provided in the supplementary materials’ Table S7.  PSC leverages a reproducible probabilistic tractography algorithm 26 to create whole-brain tractography data, integrating anatomical details from high-resolution T1 images to minimize bias in the tractography. We utilized DKA 25 to define the ROIs corresponding to the nodes in the structural connectome. For each pair of ROIs, we extracted the streamlines connecting them by following these steps: 1) dilating each gray matter ROI to include a small portion of white matter regions, 2) segmenting streamlines connecting multiple ROIs to extract the correct and complete pathway, and 3) removing apparent outlier streamlines. Due to its widespread use in brain imaging studies27, 28, we examined the mean fractional anisotropy (FA) value along streamlines and the count of streamlines in this work. The output we used includes fiber count, fiber length, and fiber volume shared between the ROIs in addition to measures of fractional anisotropy and mean diffusivity.”

The script is described and provided at: https://github.com/MISICMINA/DTI-Study-Resilience-to-CBP.git.

(7) Ethical approval?

The New Haven data is part of a study that was approved by the Yale University Institutional Review Board. This is mentioned under the description of the data “New Haven (Discovery) data set (page 23, lines 1,2).  Likewise, the Mannheim data is part of a study approved by Ethics Committee of the Medical Faculty of Mannheim, Heidelberg University, and was conducted in accordance with the declaration of Helsinki in its most recent form. This is also mentioned under “Mannheim data set” (page 26, lines 2-5): “The study was approved by the Ethics Committee of the Medical Faculty of Mannheim, Heidelberg University, and was conducted in accordance with the declaration of Helsinki in its most recent form.”

(1) Traeger AC, Henschke N, Hubscher M, et al. Estimating the Risk of Chronic Pain: Development and Validation of a Prognostic Model (PICKUP) for Patients with Acute Low Back Pain. PLoS Med 2016;13:e1002019.

(2) Hill JC, Dunn KM, Lewis M, et al. A primary care back pain screening tool: identifying patient subgroups for initial treatment. Arthritis Rheum 2008;59:632-641.

(3) Hockings RL, McAuley JH, Maher CG. A systematic review of the predictive ability of the Orebro Musculoskeletal Pain Questionnaire. Spine (Phila Pa 1976) 2008;33:E494-500.

(4) Chou R, Shekelle P. Will this patient develop persistent disabling low back pain? JAMA 2010;303:1295-1302.

(5) Silva FG, Costa LO, Hancock MJ, Palomo GA, Costa LC, da Silva T. No prognostic model for people with recent-onset low back pain has yet been demonstrated to be suitable for use in clinical practice: a systematic review. J Physiother 2022;68:99-109.

(6) Kent PM, Keating JL. Can we predict poor recovery from recent-onset nonspecific low back pain? A systematic review. Man Ther 2008;13:12-28.

(7) Hruschak V, Cochran G. Psychosocial predictors in the transition from acute to chronic pain: a systematic review. Psychol Health Med 2018;23:1151-1167.

(8) Hartvigsen J, Hancock MJ, Kongsted A, et al. What low back pain is and why we need to pay attention. Lancet 2018;391:2356-2367.

(9) Tanguay-Sabourin C, Fillingim M, Guglietti GV, et al. A prognostic risk score for development and spread of chronic pain. Nat Med 2023;29:1821-1831.

(10) Spisak T, Bingel U, Wager TD. Multivariate BWAS can be replicable with moderate sample sizes. Nature 2023;615:E4-E7.

(11) Liu Y, Zhang HH, Wu Y. Hard or Soft Classification? Large-margin Unified Machines. J Am Stat Assoc 2011;106:166-177.

(12) Loffler M, Levine SM, Usai K, et al. Corticostriatal circuits in the transition to chronic back pain: The predictive role of reward learning. Cell Rep Med 2022;3:100677.

(13) Smith SM, Dworkin RH, Turk DC, et al. Interpretation of chronic pain clinical trial outcomes: IMMPACT recommended considerations. Pain 2020;161:2446-2461.

(14) Lieberman G, Shpaner M, Watts R, et al. White Matter Involvement in Chronic Musculoskeletal Pain. The Journal of Pain 2014;15:1110-1119.

(15) Mansour AR, Baliki MN, Huang L, et al. Brain white matter structural properties predict transition to chronic pain. Pain 2013;154:2160-2168.

(16) Wager TD, Atlas LY, Lindquist MA, Roy M, Woo CW, Kross E. An fMRI-based neurologic signature of physical pain. N Engl J Med 2013;368:1388-1397.

(17) Lee JJ, Kim HJ, Ceko M, et al. A neuroimaging biomarker for sustained experimental and clinical pain. Nat Med 2021;27:174-182.

(18) Becker S, Navratilova E, Nees F, Van Damme S. Emotional and Motivational Pain Processing: Current State of Knowledge and Perspectives in Translational Research. Pain Res Manag 2018;2018:5457870.

(19) Spisak T, Kincses B, Schlitt F, et al. Pain-free resting-state functional brain connectivity predicts individual pain sensitivity. Nat Commun 2020;11:187.

(20) Baliki MN, Apkarian AV. Nociception, Pain, Negative Moods, and Behavior Selection. Neuron 2015;87:474-491.

(21) Elman I, Borsook D. Common Brain Mechanisms of Chronic Pain and Addiction. Neuron 2016;89:11-36.

(22) Baliki MN, Petre B, Torbey S, et al. Corticostriatal functional connectivity predicts transition to chronic back pain. Nat Neurosci 2012;15:1117-1119.

(23) Jenkins LC, Chang WJ, Buscemi V, et al. Do sensorimotor cortex activity, an individual's capacity for neuroplasticity, and psychological features during an episode of acute low back pain predict outcome at 6 months: a protocol for an Australian, multisite prospective, longitudinal cohort study. BMJ Open 2019;9:e029027.

(24) Zhang Z, Descoteaux M, Zhang J, et al. Mapping population-based structural connectomes. Neuroimage 2018;172:130-145.

(25) Desikan RS, Segonne F, Fischl B, et al. An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. Neuroimage 2006;31:968-980.

(26) Maier-Hein KH, Neher PF, Houde J-C, et al. The challenge of mapping the human connectome based on diffusion tractography. Nature Communications 2017;8:1349.

(27) Chiang MC, McMahon KL, de Zubicaray GI, et al. Genetics of white matter development: a DTI study of 705 twins and their siblings aged 12 to 29. Neuroimage 2011;54:2308-2317.

(28) Zhao B, Li T, Yang Y, et al. Common genetic variation influencing human white matter microstructure. Science 2021;372.

(29) Fortin JP, Parker D, Tunc B, et al. Harmonization of multi-site diffusion tensor imaging data. Neuroimage 2017;161:149-170.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

The authors intended to investigate the earliest mechanisms enabling self-prioritization, especially in the attention. Combining a temporal order judgement task with computational modelling based on the Theory of Visual Attention (TVA), the authors suggested that the shapes associated with the self can fundamentally alter the attentional selection of sensory information into awareness. This self-prioritization in attentional selection occurs automatically at early perceptual stages. Furthermore, the processing benefits obtained from attentional selection via self-relatedness and physical salience were separated from each other.

Strengths:

The manuscript is written in a way that is easy to follow. The methods of the paper are very clear and appropriate.

Thank you for your valuable feedback and helpful suggestions. Please see specific answers below.

Weaknesses:

There are two main concerns:

(1) The authors had a too strong pre-hypothesis that self-prioritization was associated with attention. They used the prior entry to consciousness (awareness) as an index of attention, which is not appropriate. There may be other processing that makes the stimulus prior to entry to consciousness (e.g. high arousal, high sensitivity), but not attention. The self-related/associated stimulus may be involved in such processing but not attention to make the stimulus easily caught. Perhaps the authors could include other methods such as EEG or MEG to answer this question.

We found the possibility of other mechanisms to be responsible for “prior entry” interesting too, but believe there are solid grounds for the hypothesis that it is indicative of attention:

First, prior entry has a long-standing history as in index of attention (e.g., Titchener, 1903; Shore et al., 2001; Yates and Nicholls, 2009; Olivers et al. 2011; see Spence & Parise, 2010, for a review.) Of course, other factors (like the ones mentioned) can contribute to encoding speed. However, for the perceptual condition, we systematically varied a stimulus feature that is associated with selective attention (salience, see e.g. Wolfe, 2021) and kept other features that are known to be associated with other factors such as arousal and sensitivity constant across the two variants (e.g. clear over threshold visibility) or varied them between participants (e.g. the colours / shapes used).

Second, in the social salience condition we used a manipulation that has repeatedly been used to establish social salience effects in other paradigms (e.g., Li et al., 2022; Liu & Sui, 2016; Scheller et al., 2024; Sui et al., 2015; see Humphreys & Sui, 2016, for a review). We assume that the reviewer’s comment suggests that changes in arousal or sensitivity may be responsible for social salience effects, specifically. We have several reasons to interpret the social salience effects as an alteration in attentional selection, rather than a result of arousal or sensitivity:

Arousal and attention are closely linked. However, within the present model, arousal is more likely linked to the availability of processing resources (capacity parameter C). That is, enhanced arousal is typically not stimulus-specific, and therefore unlikely affects the *relative* advantage in processing weights/rates of the self-associated (vs other-associated) stimuli. Indeed, a recent study showed that arousal does not modulate the relative division of attentional resources (as modelled by the Theory of Visual Attention; Asgeirsson & Nieuwenhuis, 2017). As such, it is unlikely that arousal can explain the observed results in relative processing changes for the self and other identities.

Further, there is little reason to assume that presenting a different shape enhances perceptual sensitivity. Firstly, all stimuli were presented well above threshold, which would shrink any effects that were resulting from increases in sensitivity alone. Secondly, shape-associations were counterbalanced across participants, reducing the possibility that specific features, present in the stimulus display, lead to the measurable change in processing rates as a result of enhanced shape-sensitivity.

Taken together, both, the wealth of literature that suggests prior entry to index attention and the specific design choices within our study, strongly support the notion that the observed changes in processing rates are indicative of changes in attentional selection, rather than other mechanisms (e.g. arousal, sensitivity).

(2) The authors suggested that there are two independent attention processes. I suspect that the brain needs two attention systems. Is there a probability that the social and perceptual (physical properties of the stimulus) salience fired the same attention processing through different processing?

We appreciate this thought-provoking comment. We conceptualize attention as a process that can facilitate different levels of representation, rather than as separate systems tuned to specific types of information. Different forms of representation, such as the perceptual shape, or the associated social identity, may be impacted by the same attentional process at different levels of representation. Indeed, our findings suggest that both social and perceptual salience effects may result from the same attentional system, albeit at different levels of representation. This is further supported by the additivity of perceptual and social salience effects and the negative correlation of processing facilitations between perceptually and socially salient cues. These results may reflect a trade-off in how attentional resources are distributed between either perceptually or socially salient stimuli.

Reviewer #2 (Public review):

Summary:

The main aim of this research was to explore whether and how self-associations (as opposed to other associations) bias early attentional selection, and whether this can explain well-known self-prioritization phenomena, such as the self-advantage in perceptual matching tasks. The authors adopted the Visual Attention Theory (VAT) by estimating VAT parameters using a hierarchical Bayesian model from the field of attention and applied it to investigate the mechanisms underlying self-prioritization. They also discussed the constraints on the self-prioritization effect in attentional selection. The key conclusions reported were:

(1) Self-association enhances both attentional weights and processing capacity

(2) Self-prioritization in attentional selection occurs automatically but diminishes when active social decoding is required, and

(3) Social and perceptual salience capture attention through distinct mechanisms.

Strengths:

Transferring the Theory of Visual Attention parameters estimated by a hierarchical Bayesian model to investigate self-prioritization in attentional selection was a smart approach. This method provides a valuable tool for accessing the very early stages of self-processing, i.e., attention selection. The authors conclude that self-associations can bias visual attention by enhancing both attentional weights and processing capacity and that this process occurs automatically. These findings offer new insights into self-prioritization from the perspective of the early stage of attentional selection.

Thank you for your valuable feedback and helpful suggestions. Please see specific answers below.

Weaknesses:

(1) The results are not convincing enough to definitively support their conclusions. This is due to inconsistent findings (e.g., the model selection suggested condition-specific c parameters, but the increase in processing capacity was only slight; the correlations between attentional selection bias and SPE were inconsistent across experiments), unexpected results (e.g., when examining the impact of social association on processing rates, the other-associated stimuli were processed faster after social association, while the self-associated stimuli were processed more slowly), and weak correlations between attentional bias and behavioral SPE, which were reported without any p-value corrections. Additionally, the reasons why the attentional bias of self-association occurs automatically but disappears during active social decoding remain difficult to explain. It is also possible that the self-association with shapes was not strong enough to demonstrate attention bias, rather than the automatic processes as the authors suggest. Although these inconsistencies and unexpected results were discussed, all were post hoc explanations. To convince readers, empirical evidence is needed to support these unexpected findings.

Thank you for outlining the specific points that raise your concern. We were happy to address these points as follows:

a. Replications and Consistency: In our study, we consistently observed trends (relative reduction in processing speed of the self-associated stimulus) in the social salience conditions across experiments. While Experiment 2 demonstrated a significant reduction in processing rate towards self-stimuli, there was a notable trend in Experiment 1 as well.

b. Condition-specific parameters: The condition-specific C parameters, though presenting a small effect size, significantly improved model fit. Inspecting the HDI ranges of our estimated C parameters indicates a high probability (85-89%) that processing capacity increased due to social associations, suggesting that even small changes (~2Hz) can hold meaningful implications within the context attentional selection.

Please also note that the main conclusions about relative salience (self/other, salient/non-salient) are based on the relative processing rates. Processing rates are the product of the processing capacity (condition- but not stimulus dependent) and the attentional weight (condition and stimulus dependent). The latter is crucial to judge the *relative* advantage of the salient stimulus. Hence, the self-/salient stimulus advantage that is reflected in the ‘processing rate difference’ is automatically also reflected in the relative attentional weights attributed to the self/other and salient/non-salient stimuli. As such, the overall results of an automatic relative advantage of self-associated stimuli hold, independently of the change in overall processing capacity.

c. Correlations: Regarding the correlations the reviewer noted, we wish to clarify that these were exploratory, and not the primary focus of our research. The aim of these exploratory analyses was to gauge the contribution of attentional selection to matching-based SPEs. As SPEs measured via the matching task are typically based on multiple different levels of processing, the contribution of early attentional selection to their overall magnitude was unclear. Without being able to gauge the possible effect sizes, corrected analyses may prevent detecting small but meaningful effects. As such, the effect sizes reported serve future studies to estimate power a priori and conduct well-powered replications of such exploratory effects. Additionally, Bayes factors were provided to give an appreciation of the strength of the evidence, all suggesting at least moderate evidence in favour of a correlation. Lastly, please note that effects that were measured within individuals and task (processing rate increase in social and perceptual decision dimensions in the TOJ task) showed consistent patterns, suggesting that the modulations within tasks were highly predictive of each other, while the modulations between tasks were not as clearly linked. We will add this clarification to the revised manuscript.

d. Unexpected results: The unexpected results concerning the processing rates of other-associated versus self-associated stimuli certainly warrant further discussion. We believe that the additional processing steps required for social judgments, reflected in enhanced reaction times, may explain the slower processing of self-associated stimuli in that dimension. We agree that not all findings will align with initial hypotheses, and this variability presents avenues for further research. We have added this to the discussion of social salience effects.

e. Whether association strength can account for the findings: We appreciate the scepticism regarding the strength of self-association with shapes. However, our within-participant design and control matching task indicate that the relative processing advantage for self-associated stimuli holds across conditions. This makes the scenario that “the self-association with shapes was not strong enough to demonstrate attention bias” very unlikely. Firstly, the relative processing advantage of self-associated stimuli in the perceptual decision condition, and the absence of such advantage in the social decision condition, were evidenced in the same participants. Hence, the strength of association between shapes and social identities was the same for both conditions. However, we only find an advantage for the self-associated shape when participants make perceptual (shape) judgements. It is therefore highly unlikely that the “association strength” can account for the difference in the outcomes between the conditions in experiment 1. Also, note that the order in which these conditions were presented was counter-balanced across participants, reducing the possibility that the automatic self-advantage was merely a result of learning or fatigue. Secondly, all participants completed the standard matching task to ascertain that the association between shapes and identities did indeed lead to processing advantages (across different levels).

In summary, we believe that the evidence we provide supports the final conclusions. We do, of course, welcome any further empirical evidence that could enhance our understanding of the contribution of different processing levels to the SPE and are committed to exploring these areas in future work.

(2) The generalization of the findings needs further examination. The current results seem to rely heavily on the perceptual matching task. Whether this attentional selection mechanism of self-prioritization can be generalized to other stimuli, such as self-name, self-face, or other domains of self-association advantages, remains to be tested. In other words, more converging evidence is needed.

The reviewer indicates that the current findings heavily rely on the perceptual matching task, and it would be more convincing to include other paradigm(s) and different types of stimuli. We are happy to address these points here: first, we specifically used a temporal order paradigm to tap into specific processes, rather than merely relying on the matching task. Attentional selection is, along with other processes, involved in matching, but the TOJ-TVA approach allows tapping into attentional selection specifically.  Second, self-prioritization effects have been replicated across a wide range of stimuli (e.g. faces: Wozniak et al., 2018; names or owned objects: Scheller & Sui, 2022a, or even fully unfamiliar stimuli: Wozniak & Knoblich, 2019) and paradigms (e.g. matching task: Sui et al., 2012; cross-modal cue integration: e.g. Scheller & Sui, 2022b; Scheller et al., 2023; continuous flash suppression: Macrae et al., 2017; temporal order judgment: Constable et al., 2019; Truong et al., 2017). Using neutral geometric shapes, rather than faces and names, addresses a key challenge in self research: mitigating the influence of stimulus familiarity on results. In addition, these newly learned, simple stimuli can be combined with other paradigms, such as the TOJ paradigm in the current study, to investigate the broader impact of self-processing on perception and cognition.

To the best of our knowledge, this is the first study showing evidence about the mechanisms that are involved in early attentional selection of socially salient stimuli. Future replications and extensions would certainly be useful, as with any experimental paradigm.

(3) The comparison between the "social" and "perceptual" tasks remains debatable, as it is challenging to equate the levels of social salience and perceptual salience. In addition, these two tasks differ not only in terms of social decoding processes but also in other aspects such as task difficulty. Whether the observed differences between the tasks can definitively suggest the specificity of social decoding, as the authors claim, needs further confirmation.

Equating the levels of social and perceptual salience is indeed challenging, but not an aim of the present study. Instead, the present study directly compares the mechanisms and effects of social and perceptual salience, specifically experiment 2. By manipulating perceptual salience (relative colour) and social salience (relative shape association) independently and jointly, and quantifying the effects on processing rates, our study allows to directly delineate the contributions of each of these types of salience. The results suggest additive effects (see also Figure 7). Indeed, the possibility remains that these effects are additive because of the use of different perceptual features, so it would be helpful for future studies to explore whether similar perceptual features lead to (supra-/sub-) additive effects. In either case, the study design allows to directly compare the effects and mechanisms of social and perceptual salience.

Regarding the social and perceptual decision dimensions, they were not expected to be equated. Indeed, the social decision dimension requires additional retrieval of the associated identity, making it likely more challenging. This additional retrieval is also likely responsible for the slower responses towards the social association compared to the shape itself. However, the motivation to compare the effects of these two decisional dimensions lies in the assumption that the self needs to be task relevant. Some evidence suggests that the self needs to be task-relevant to induce self-prioritization effects (e.g., Woźniak & Knoblich, 2022). However, these studies typically used matching tasks and were powered to detect large effects only (e.g. f = 0.4, n = 18). As it is likely that lacking contribution of decisional processing levels (which interact with task-relevance) will reduce the SPE, smaller self-prioritization effects that result from earlier processing levels may not be detected with sufficient statistical power. Targeting specific processing levels, especially those with relatively early contributions or small effect sizes, requires larger samples (here: n = 70) to provide sufficient power. Indeed, by contrasting the relative attentional selection effects in the present study we find that the self does not need to be task-relevant to produce self-prioritization effects. This is in line with recent findings of prior entry of self-faces (Jubile & Kumar, 2021)

Author response:

Reviewer #1 (Public review):

Summary:

Liu et al., present an immersion objective adapter design called RIM-Deep, which can be utilized for enhancing axial resolution and reducing spherical aberrations during inverted confocal microscopy of thick cleared tissue.

Strengths:

RI mismatches present a significant challenge to deep tissue imaging, and developing a robust immersion method is valuable in preventing losses in resolution. Liu et al., present data showing that RIM-Deep is suitable for tissue cleared with two different clearing techniques, demonstrating the adaptability and versatility of the approach.

Greetings, we greatly appreciate your feedback. In truth, we have utilized three distinct clearing techniques, including iDISCO, CUBIC, and MACS, to substantiate the adaptability and multifunctionality of the RIM-Deep adapter.

Weaknesses:

Liu et al., claim to have developed a useful technique for deep tissue imaging, but in its current form, the paper does not provide sufficient evidence that their technique performs better than existing ones.

We are in complete agreement with your recommendation, and the additional experiments will conduct a thorough comparison of the efficacy between the RIM-deep adapter and the official adapter in the context of fluorescence bead experiments, along with their performance in cubic and MASC tissue clearing techniques.

Reviewer 2 (Public review):

The authors used different clearing methods to demonstrate the suitability of RIM-Deep for various sample preparation protocols with clearing solutions of different refractive indices. They clearly demonstrate that the RIM-Deep chamber is compatible with all three methods. Brain samples are characterized by complex networks of cells and are often hard to visualize. Despite the dense, complex structure of brain tissue, the RIM-Deep method generated high-quality images of all three samples. As the authors stated, increasing imaging depth often goes hand in hand with purchasing expensive new equipment, exchanging several microscopy parts, or purchasing a new microscopy setup. Innovations like the RIM-Deep chamber might pave the way for cost-effective imaging and expand the applicability of inverted confocal microscopy.

Weeknesses:

(1) However, since this study introduces a novel imaging technique aiming to revolutionize imaging of large samples, additional control experiments would strengthen the data. From the three clearing protocols used (CUBIC, MACS, and iDISCO), only the brain section from Macaca fascicularis cleared with iDISCO was imaged with the standard chamber and the RIM-Deep method. This comparison indeed shows a more than 2-fold increase in imaging depth, a significant enhancement in microscopy. However, it would have been important to evaluate and show the imaging depth differences in the other two samples, as they were cleared with different protocols and treated with clearing solutions of different refractive indices compared to iDISCO.

Thank you for your suggestion. We will investigate the imaging performance of brain tissue using the other two clearing protocols with both the official adapter and the RIM-deep method.

(2) The description of the figures and figure panels should be improved for a better understanding of the experiments performed and the resulting images/data.

Thank you for your suggestion. We will revise the figure legends in detail.

(3) While the authors used a Nikon AX inverted laser scanning confocal microscope, the study would benefit from evaluating the performance of the RIM-Deep method using other inverted confocal microscopes or even wide-field microscopes.

Thank you for your suggestion. We also recognize that evaluating the performance of the RIM-Deep method on other inverted confocal microscopes will help further validate its applicability and robustness. We will supplement these experiments to expand the scope and reliability of RIM-Deep.

Author response:

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

Public Reviews: 

Reviewer #1 (Public Review): 

In their manuscript, the authors propose a learning scheme to enable spiking neurons to learn the appearance probability of inputs to the network. To this end, the neurons rely on error-based plasticity rules for feedforward and recurrent connections. The authors show that this enables the networks to spontaneously sample assembly activations according to the occurrence probability of the input patterns they respond to. They also show that the learning scheme could explain biases in decision-making, as observed in monkey experiments. While the task of neural sampling has been solved before in other models, the novelty here is the proposal that the main drivers of sampling are within-assembly connections, and not between-assembly (Markov chains) connections as in previous models. This could provide a new understanding of how spontaneous activity in the cortex is shaped by synaptic plasticity. 

The manuscript is well written and the results are presented in a clear and understandable way. The main results are convincing, concerning the spontaneous firing rate dependence of assemblies on input probability, as well as the replication of biases in the decision-making experiment. Nevertheless, the manuscript and model leave open several important questions. The main problem is the unclarity, both in theory and intuitively, of how the sampling exactly works. This also makes it difficult to assess the claims of novelty the authors make, as it is not clear how their work relates to previous models of neural sampling. 

We agree with the reviewer that our previous manuscript was not clear regarding the mechanism of the model. We have performed additional simulations and included a derivation of the learning rule to address this, which we explain below.

Regarding the unclarity of the sampling mechanism, the authors state that withinassembly excitatory connections are responsible for activating the neurons according to stimulus probability. However, the intuition for this process is not made clear anywhere in the manuscript. How do the recurrent connections lead to the observed effect of sampling? How exactly do assemblies form from feedforward plasticity? This intuitive unclarity is accompanied by a lack of formal justification for the plasticity rules. The authors refer to a previous publication from the same lab, but it is difficult to connect these previous results and derivations to the current manuscript. The manuscript should include a clear derivation of the learning rules, as well as an (ideally formal) intuition of how this leads to the sampling dynamics in the simulation. 

We have included a derivation of our plasticity rules in lines 871-919 in the revised manuscript. Consistent with our claim that predictive plasticity updates the feedforward and the recurrent synapses to predict output firing rates, we have shown that the corresponding cost function measures the discrepancy among the recurrent prediction, feedforward prediction, and the output firing rate. The resultant feedforward plasticity is the same with our previous rule (Asabuki and Fukai, 2020), which segments the salient patterns embedded in the input sequence. The recurrent plasticity rule suggests that the recurrent prediction learns the statistical model of the evoked activity, enabling the network to replay the learned internal model.  

Similarly, for the inhibitory plasticity, we defined a cost function that evaluates the difference between the firing rate and inhibitory potential within each neuron. This rule is crucial for maintaining balanced network dynamics. See our response below for more details on the role of inhibitory plasticity.

Some of the model details should furthermore be cleared up. First, recurrent connections transmit signals instantaneously, which is implausible. Is this required, would the network dynamics change significantly if, e.g., excitation arrives slightly delayed? Second, why is the homeostasis on h required for replay? The authors show that without it the probabilities of sampling are not matched, but it is not clear why, nor how homeostasis prevents this. Third, G and M have the same plasticity rule except for G being confined to positive values, but there is no formal justification given for this quite unusual rule. The authors should clearly justify (ideally formally) the introduction of these inhibitory weights G, which is also where the manuscript deviates from their previous 2020 work. My feeling is that inhibitory weights have to be constrained in the current model because they have a different goal (decorrelation, not prediction) and thus should operate with a completely different plasticity mechanism. The current manuscript doesn't address this, as there is no overall formal justification for the learning algorithm. 

First, while the reviewer's suggestion to test with delayed excitation is intriguing and crucial for a more biologically detailed spiking neuron model, we have chosen to maintain the current model configuration. Our use of Poisson spiking neurons, which generate spikes based on instantaneous firing rates, does not heavily depend on precise spike timing information. Therefore, to preserve the simplicity of our results, we kept the model unchanged.

Second, we agree that our previous claim regarding the importance of the memory trace h for sampling may have been confusing. As shown in Supplementary Figure 7b in the revised manuscript, when we eliminated the dynamics of the memory trace, sampling performance did indeed decrease. However, we also observed that the assembly activity ratio continued to show a linear relationship with stimulus probabilities. Based on these findings, we have revised our claim in the manuscript to clarify that the memory trace is primarily critical for firing rate homeostasis, rather than directly influencing sampling within the learned network. We have explained this in ll. 446-448 in the revised manuscript.

Third, we explored a new architecture where all recurrent connections are either exclusively excitatory or inhibitory, keeping their sign throughout the learning process. This change addresses the reviewer's concern about our initial assumption that only the inhibitory connection G was constrained to non-negative values. We found that inhibition plays a crucial role in decorrelation and prediction, helping activate specific assemblies through competition while preventing runaway excitation within active assemblies. We have explained this in ll.560-593 in the revised manuscript.

Finally, the authors should make the relation to previous models of sampling and error-based plasticity more clear. Since there is no formal derivation of the sampling dynamics, it is difficult to assess how they differ exactly from previous (Markov-based) approaches, which should be made more precise. Especially, it would be important to have concrete (ideally experimentally testable) predictions on how these two ideas differ. As a side note, especially in the introduction (line 90), this unclarity about the sampling made it difficult to understand the contrast to Markovian transition models. 

As the reviewer pointed out, previous computational models have demonstrated that recurrent networks with Hebbian-like plasticity can learn appropriate Markovian statistics (Kappel et al., 2014; Asabuki and Clopath, 2024). However, our model differs conceptually from these previous models. While Kappel et al. showed that STDP in winner-take-all circuits can approximate online learning of hidden Markov models (HMMs), a key difference with our model is that their neural representations acquire sequences using Markovian sampling dynamics, whereas our model does not depend on such ordered sampling. Specifically, in their model, sequential sampling arises from learned structures in the off-diagonal elements of the recurrent connections (i.e., between-assembly connections). In contrast, our network learns to stochastically generate recurrent cell assemblies by relying solely on within-assembly connections. A similar argument can be made for Asabuki and Clopath paper as well. Further, while our model introduced plasticity rule for all types of connections, Asabuki and Clopath paper introduced plasticity for recurrent synapses projecting on the excitatory neurons only and the cell assembly memberships were preconfigured unlike our model. We have added additional clarifying sentences in ll. 757-772 of the revised manuscript to elaborate on this point.

There are also several related models that have not been mentioned and should be discussed. In 663 ff. the authors discuss the contributions of their model which they claim are novel, but in Kappel et al (STDP Installs in Winner-Take-All Circuits an Online Approximation to Hidden Markov Model Learning) similar elements seem to exist as well, and the difference should be clarified. There is also a range of other models with lateral inhibition that make use of error-based plasticity (most recently reviewed in Mikulasch et al, Where is the error? Hierarchical predictive coding through dendritic error computation), and it should be discussed how the proposed model differs from these. 

We have clarified the difference from previously proposed recurrent network model to perform Markovian sampling. Please see our reply above.

We have also included additional sentence in ll. 704-709 in the revised manuscript to discuss how our model differs from similar predictive learning models: “It should be noted that while several network models that perform errorbased computations like ours exploit only inhibitory recurrent plasticity (Mikulasch et al., 2021; Mackwood et al., 2021; Hertäg and Clopath., 2022; Mikulasch et al., 2023), our model learns the structured spontaneous activity to reproduce the evoked statistics by modifying both excitatory and inhibitory recurrent connections.”

Reviewer #2 (Public Review):

Summary: 

The paper considers a recurrent network with neurons driven by external input. During the external stimulation predictive synaptic plasticity adapts the forward and recurrent weights. It is shown that after the presentation of constant stimuli, the network spontaneously samples the states imposed by these stimuli. The probability of sampling stimulus x^(i) is proportional to the relative frequency of presenting stimulus x^(i) among all stimuli i=1,..., 5. 

Methods: 

Neuronal dynamics: 

For the main simulation (Figure 3), the network had 500 neurons, and 5 nonoverlapping stimuli with each activating 100 different neurons where presented. The voltage u of the neurons is driven by the forward weights W via input rates x, the inhibitory recurrent weights G, are restricted to have non-negative weights (Dale's law), and the other recurrent weights M had no sign-restrictions. Neurons were spiking with an instantaneous Poisson firing rate, and each spike-triggered an exponentially decaying postsynaptic voltage deflection. Neglecting time constants of the postsynaptic responses, the expected postsynaptic voltage reads (in vectorial form) as 

u = W x + (M - G) f (Eq. 5) 

where f =; phi(u) represents the instantaneous Poisson rate, and phi a sigmoidal nonlinearity. The rate f is only an approximation (symbolized by =;) of phi(u) since an additional regularization variable h enters (taken up in Point 4 below). The initialisation of W and M is Gaussian with mean 0 and variance 1/sqrt(N), N the number of neurons in the network. The initial entries of G are all set to 1/sqrt(N). 

Predictive synaptic plasticity: 

The 3 types of synapses were each adapted so that they individually predict the postsynaptic firing rate f, in matrix form 

ΔW ≈ (f - phi( W x ) ) x^T 

ΔM ≈ (f - phi( M f ) ) f^T 

ΔG ≈ (f - phi( M f ) ) f^T but confined to non-negative values of G (Dale's law). 

The ^T tells us to take the transpose, and the ≈ again refers to the fact that the ϕ entering in the learning rule is not exactly the ϕ determining the rate, only up to the regularization (see Point 4). 

Main formal result: 

As the authors explain, the forward weight W and the unconstrained weight M develop such that, in expectations, 

f =; phi( W x ) =; phi( M f ) =; phi( G f ) , 

consistent with the above plasticity rules. Some elements of M remain negative. In this final state, the network displays the behaviour as explained in the summary. 

Major issues: 

Point 1: Conceptual inconsistency 

The main results seem to arise from unilaterally applying Dale's law only to the inhibitory recurrent synapses G, but not to the excitatory recurrent synapses M. 

In fact, if the same non-negativity restriction were also imposed on M (as it is on G), then their learning rules would become identical, likely leading to M=G. But in this case, the network becomes purely forward, u = W x, and no spontaneous recall would arise. Of course, this should be checked in simulations. 

Because Dale's law was only applied to G, however, M and G cannot become equal, and the remaining differences seem to cause the effect. 

Predictive learning rules are certainly powerful, and it is reasonable to consider the same type of error-correcting predictive learning rule, for instance for different dendritic branches that both should predict the somatic activity. Or one may postulate the same type of error-correcting predictive plasticity for inhibitory and excitatory synapses, but then the presynaptic neurons should not be identical, as it is assumed here. Both these types of error-correcting and error-forming learning rules for same-branches and inhibitory/excitatory inputs have been considered already (but with inhibitory input being itself restricted to local input, for instance). 

The model presented above lacked biological plausibility in several key aspects. Specifically, we assumed that the recurrent connection M could change sign through plasticity and be either excitatory or inhibitory, while the inhibitory connection G was restricted to being inhibitory only. This initial setting does not reflect the biological constraint that synapses typically maintain a consistent excitatory or inhibitory type. Furthermore, due to this unconstrained recurrent connectivity M, the original model had two types of inhibitory connections (i.e., the negative part of M and the inhibitory connection G) without providing a clear computational role for each type of inhibition.

To address these limitations and to understand the role of the two types of inhibition, we explored a new architecture where all recurrent connections are either exclusively excitatory or inhibitory, keeping their sign throughout the learning process. This change addresses the reviewer's concern about our initial assumption that only the inhibitory connection G was constrained to non-negative values. We found that inhibition plays a crucial role in prediction and decorrelation, helping activate specific assemblies through competition while preventing runaway excitation within active assemblies. We have explained this in ll. 561593 in the revised manuscript.

Point 2: Main result as an artefact of an inconsistently applied Dale's law? 

The main result shows that the probability of a spontaneous recall for the 5 nonoverlapping stimuli is proportional to the relative time the stimulus was presented. This is roughly explained as follows: each stimulus pushes the activity from 0 up towards f =; phi( W x ) by the learning rule (roughly). Because the mean weights W are initialized to 0, a stimulus that is presented longer will have more time to push W up so that positive firing rates are reached (assuming x is non-negative). The recurrent weights M learn to reproduce these firing rates too, while the plasticity in G tries to prevent that (by its negative sign, but with the restriction to non-negative values). Stimuli that are presented more often, on average, will have more time to reach the positive target and hence will form a stronger and wider attractor. In spontaneous recall, the size of the attractor reflects the time of the stimulus presentation. This mechanism so far is fine, but the only problem is that it is based on restricting G, but not M, to non-negative values. 

As mentioned above, we have included an additional simulation where all weights are non-negative. We have demonstrated the new results in Figure 6 before presenting the two-population model in the revised manuscript (Figure 7), so that readers can follow the importance of two pathways of inhibitory connections.

Point 3: Comparison of rates between stimulation and recall. 

The firing rates with external stimulations will be considerably larger than during replay (unless the rates are saturated). 

This is a prediction that should be tested in simulations. In fact, since the voltage roughly reads as  u = W x + (M - G) f,  and the learning rules are such that eventually M =; G, the recurrences roughly cancel and the voltage is mainly driven by the external input x. In the state of spontaneous activity without external drive, one has  u = (M - G) f ,  and this should generate considerably smaller instantaneous rates f =; phi(u) than in the case of the feedforward drive (unless f is in both cases at the upper or lower ceiling of phi). This is a prediction that can also be tested. 

Because the figures mostly show activity ratios or normalized activities, it was not possible for me to check this hypothesis with the current figures. So please show non-normalized activities for comparing stimulation and recall for the same patterns. 

We agree with the reviewer that the activity levels of spontaneous and induced activity should be compared. We have shown the distributions of activity level of these activities in our new Figure 2d. As expected, we found that the evoked activity showed stronger activity compared to the spontaneous activity.  

Point 4: Unclear definition of the variable h. 

The formal definition of h = hi is given by (suppressing here the neuron index i and the h-index of tau) 

tau dh/dt = -h if h>u, (Eq. 10)  h = u otherwise. 

But if it is only Equation 10 (nothing else is said), h will always become equal to u, or will vanish, i.e. either h=u or h=0 after some initial transient. In fact, as soon as h>u, h is decaying to 0 according to the first line. If u is >0, then it stops at u=h according to the second line. No reason to change h=u further. If u<=0 while h>u, then h is converging to 0 according to the first line and will stay there. I guess the authors had issues with the recurrent spiking simulations and tried to fix this with some regularization. However as presented, it does not become clear how their regulation works. 

We apologize for the reviewer that our definition of h was unclear. As the reviewer pointed out, since the memory trace is always positive and larger than (or equal to) the membrane potential, it is possible that the membrane potential becomes always negative and the memory trace reach to 0 constantly. However, since the network is always balanced between excitatory and inhibitory inputs, and it does not happen that the membrane potential always diverges negatively. In fact, we trained without any manipulations other than the memory trace described in the manuscript, and the network was able to learn the assembly structure stably. 

BTW: In Eq. 11 the authors set the gain beta to beta = beta0/h which could become infinite and, putatively more problematic, negative, depending on the value of h. Maybe some remark would convince a reader that no issues emerge from this. 

We have mentioned in ll. 864-866 in the revised manuscript that no issues emerge from the slope parameter.

Added from discussions with the editor and the other reviewers: 

Thanks for alerting me to this Supplementary Figure 8. Yes, it looks like the authors did apply there Dale's law for both the excitatory and inhibitory synapses. Yet, they also introduced two types of inhibitory pathways converging both to the excitatory and inhibitory neurons. For me, this is a confirmation that applying Dale's law to both excitatory and inhibitory synapses, with identical learning rules as explained in the main part of the paper, does not work. 

Adding such two pathways is a strong change from the original model as introduced before, and based on which all the Figures in the main text are based. Supplementary Figure 8 should come with an analysis of why a single inhibitory pathway does not work. I guess I gave the reason in my Points 1-3. Some form of symmetry breaking between the recurrent excitation and recurrent inhibition is required so that, eventually, the recurrent excitatory connection will dominate. 

Making the inhibitory plasticity less expressive by applying Dale's law to only those inhibitory synapses seems to be the answer chosen in the Figures of the main text (but then the criticism of unilaterally applying Dale's law). 

Applying Dale's law to both types of synapses, but dividing the labor of inhibition into two strictly separate and asymmetric pathways, and hence asymmetric development of excitatory and inhibitory weights, seems to be another option. However, introducing such two separate inhibitory pathways, just to rescue the fact that Dale's law is applied to both types of synapses, is a bold assumption. Is there some biological evidence of such two pathways in the inhibitory, but not the excitatory connections? And what is the computational reasoning to have such a separation, apart from some form of symmetry breaking between excitation and inhibition? I guess, simpler solutions could be found, for instance by breaking the symmetry between the plasticity rules for the excitatory and inhibitory neurons. All these questions, in my view, need to be addressed to give some insights into why the simulations do work. 

The reviewer’s intuition is correct. To effectively learn cell assembly structures and replay their activities, our model indeed requires two types of inhibitory connections. Please refer to our response above for further details. 

Overall, Supplementary Figure 8 seems to me too important to be deferred to the Supplement. The reasoning behind the two inhibitory pathways should appear more prominently in the main text. Without this, important questions remain. For instance, when thinking in a rate-based framework, the two inhibitory pathways twice try to explain the somatic firing rate away. Doesn't this lead to a too strong inhibition? Can some steady state with a positive firing rate caused by the recurrence, in the absence of an external drive, be proven? The argument must include the separation into Path 1 and Path 2. So far, this reasoning has not been entered. 

In fact, it might be that, in a spiking implementation, some sparse spikes will survive. I wonder whether at least some of these spikes survive because of the other rescuing construction with the dynamic variable h (Equation 10, which is not transparent, and that is not taken up in the reasoning either, see my Point 4)

Perhaps it is helpful for the authors to add this text in the reply to them. 

We have moved the former Supplemental Figure 8 to the main Figure 7. Please see our response above about the role of dual inhibitory connection types.

Reviewer #3 (Public Review): 

Summary: 

The work shows how learned assembly structure and its influence on replay during spontaneous activity can reflect the statistics of stimulus input. In particular, stimuli that are more frequent during training elicit stronger wiring and more frequent activation during replay. Past works (Litwin-Kumar and Doiron, 2014; Zenke et al., 2015) have not addressed this specific question, as classic homeostatic mechanisms forced activity to be similar across all assemblies. Here, the authors use a dynamic gain and threshold mechanism to circumnavigate this issue and link this mechanism to cellular monitoring of membrane potential history. 

Strengths: 

(1) This is an interesting advance, and the authors link this to experimental work in sensory learning in environments with non-uniform stimulus probabilities. 

(2) The authors consider their mechanism in a variety of models of increasing complexity (simple stimuli, complex stimuli; ignoring Dale's law, incorporating Dale's law). 

(3) Links a cellular mechanism of internal gain control (their variable h) to assembly formation and the non-uniformity of spontaneous replay activity. Offers a promise of relating cellular and synaptic plasticity mechanisms under a common goal of assembly formation. 

Weaknesses: 

(1) However, while the manuscript does show that assembly wiring does follow stimulus likelihood, it is not clear how the assembly-specific statistics of h reflect these likelihoods. I find this to be a key issue. 

We agree that our previous claim regarding the importance of the memory trace h for sampling may have been confusing. As shown in Supplementary Figure 7b, when we eliminated the dynamics of the memory trace, sampling performance did indeed decrease. However, we also observed that the assembly activity ratio continued to show a linear relationship with stimulus probabilities. Based on these findings, we revised our claim in the manuscript to clarify that the memory trace is primarily critical for learning to avoid trivial solutions, rather than directly influencing sampling within the learned network. We have explained this in ll. 446-448 in the revised manuscript.

(2) The authors' model does take advantage of the sigmoidal transfer function, and after learning an assembly is either fully active or nearly fully silent (Figure 2a). This somewhat artificial saturation may be the reason that classic homeostasis is not required since runaway activity is not as damaging to network activity. 

The reviewer's intuition is correct. The saturating nonlinearity is important for the network to form stable assembly structures. We have added an additional sentence in ll. 866-868 to mention this.

(3) Classic mechanisms of homeostatic regulation (synaptic scaling, inhibitory plasticity) try to ensure that firing rates match a target rate (on average). If the target rate is the same for all neurons then having elevated firing rates for one assembly compared to others during spontaneous activity would be difficult. If these homeostatic mechanisms were incorporated, how would they permit the elevated firing rates for assemblies that represent more likely stimuli? 

LIF neurons) may solve this problem by utilizing spike-timing statistics.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors): 

Minor issues: 

Figure 1: It would be helpful to display the equation for output rate here as well. 

We have included the equation in the revised Figure 1a.

Figure 3c: Typo "indivisual neurons". 

We have modified the typo. We thank the reviewer for their careful review.

Line 325: Do you mean Figure 3f,g? 

We repeated the task with different numbers of stimuli in Supplementary Figure 1c,d.

Line 398: Winner-take-all can be misunderstood, as it typically stands for competition in inference, not in learning. 

We have rephrased it as “unstable dynamics” in l. 400

Line 429: Are intra-assembly and within-assembly the same? If so please use consistent terminology. 

We have made the terminology consistent.

Line 792 ff.: Please mention that (t) was left away. 

We have included a sentence to mention it in ll. 847-848 in the revised manuscript.

Line 817: Should u_i be v_i? 

We have modified the term.

Methods: What is the value of tau_h? 

We have used 𝜏! \=10 s, which is mentioned in l. 853

Author response:

We are appreciative of the editors’ and reviewers’ positive comments and constructive suggestions, which will help us to improve our manuscript. We will make changes as required by the reviewers. Our primary focus will be on revising and clarifying certain aspects:

First, recent research has revealed a strong correlation between brain synchronization and group decision-making, a key neural marker. We aim to bolster our hypothesis by reviewing additional literature, ensuring accuracy in terminology and appropriateness in phrasing.

Second, it is crucial to note that we will include additional methodological details, such as the details of the experiment, the significance of individual difference variables, and the details of the data analyses.

Third, despite introducing a novel perspective in our study, we acknowledge the utilization of the conventional fNIRS hyperscanning analyses, which are widely accepted within the research community. Our methodology entails the identification of significant channels via one-sample t-tests, subsequently complemented by either ANOVAs or independent sample t-tests, without the need for double dipping.

We will address all the issues raised by the reviewers.We believe that the manuscript will significantly benefit from the insightful suggestions and invaluable contributions made by the editors and reviewers.

Author response:

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

In the second round of reviews, Reviewer 2 made three specific comments. The first comment criticises us for not including a set of equations they had requested in their first review. We did, in fact, include the requested equations in our revised submission, which were in the Supplementary Information, and were also cited in the main text of our revised manuscript and our changes were made clear in our response to the reviewer. The second comment, the reviewer suggested adding one word to a sentence in the abstract. We have made this change (line 23). The third comment, the reviewer highlights a sentence where we agree we could have been more clear. The sentence can be rectified by adding one word to the current sentence, which we have done (line 232). We believe the changes required to our manuscript are very minor, and we have implemented these two suggested changes, which are highlighted in the revised manuscript.

Author response:

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

Reviewer 1

Overall, this work is quite comprehensive and is logically and rigorously designed. The phenotypic and functional data on 2C are strong.

Thank you for your positive feedback on our findings!

(1) Comment from Reviewer 1 suggesting the mechanistic insights of 2C are primarily derived from transcriptomic and genomic datasets without experimental verification. 

Thank you for emphasizing the importance of experimental validation to support our transcriptomic and genomic findings. We acknowledge the gap in direct experimental evidence for the mechanistic insights of section 2C and recognize the value of such validation in strengthening our conclusions. While we recognize the importance of such validation, our current dataset lacks the comprehensive preliminary results necessary for inclusion in the supplemental material. We believe that the mechanistic insights presented offer a substantial foundation for the future research, where we aim to explore these aspects in depth with targeted experimental approaches.

Reviewer 2

Together their data may suggest a regenerative effect of 2C both in vitro and in vivo settings. If confirmed, this study might unlock therapeutic strategy for cardiac regeneration.

Thank you for your positive comment on the significance of our findings and the valuable therapeutic potential of 2C in cardiac regeneration!

(1) Comment from Reviewer 2 pointing out the the main hypothesis (line 50) that Isl1 cells have regenerative properties is not extremely novel. 

We agree with the reviewer that Isl1-positive cells possess regenerative properties. Following the reviewer’s suggestion, we have revised the original wording (line 46 in the revised manuscript).

(2) Comment from Reviewer 2 asking for providing a rationale for this 20x reduction of A-485 concentration? It would be useful to get a titration of this compound for the effects tested. 

As suggested by the reviewer, we have added the titration results of A-485 in Figure 1—figure supplement 1F-G.

(3) Comment from Reviewer 2 confusing to clearly understand what proportion of CMs dedifferentiate to become RCCs. The lineage tracing data suggests only 0.6%-1.5% of cells undergo this transition. It is difficult to understand how such a small fraction can have wide effects in their different experimental settings. This is specifically true when the author quantified nuclear and cytosolic area on brightfield pictures - would the same effect on nuclear/cytosolic area be observed in Isl1 KO cells. 

We appreciate the reviewer's insightful observation on the proportion of CMs undergoing dedifferentiation into RCCs and the potential impact of this subset on our experimental outcomes. The lineage tracing data indicating that only 0.6%-1.5% of CMs transition to RCCs indeed reflects a modest proportion. This observation raises valid questions regarding the broader implications of such a limited fraction in the context of cardiac regeneration and the experimental effects reported. It's important to note that while the proportion of CMs dedifferentiating into RCCs is small, the biological significance and potential impact of these RCCs could be disproportionately large. Emerging evidence suggests that even a minimal number of stem or progenitor cells can exert significant effects on tissue repair and regeneration, possibly through paracrine mechanisms or by acting as key signaling centers within the tissue microenvironment (Fernandes et al., 2015). Regarding the specific question about 2C’s effects on nuclear/cytosolic area in Isl1 knockout (KO) cells, we appreciate the suggestion and consider that such comparative studies would provide valuable insights for future comprehensively understanding the significant impact of 2C-induced RCCs in future search. In addition, ISL1 KO cells are also described in detail in the article published in eLife in 2018 by Quaranta et al.

(4) Comment from Reviewer 2 asking for the effect of CHIR + I-BET-762 alone. 

As suggested by the reviewer, we have added the results of CHIR + T-BET-762 in Figure 1—figure supplement 1H.

(5) Comment from Reviewer 2 suggesting a transparent explaination about the effects of A-485 on acetylation status.

We thank the reviewer for highlighting the confusion regarding the effects of A-485 on the acetylation status of H3K27Ac and H3K9Ac. Upon re-examination of our data and statements, we recognize the need for clarity in our explanation and the inconsistency it may have caused (lines 223-231 on page 8).

Initially, our observations suggested a selective effect of A-485 on H3K27Ac based on early experimental results (Figure 7—figure supplement 1). This conclusion was drawn from preliminary analyses that focused predominantly on this specific histone mark. However, upon further comprehensive examination of our data, including additional replicates and more sensitive detection methods, we observed that A-485 also impacts H3K9Ac levels (Figure 7—figure supplement 1F). This latter finding emerged from expanded datasets that were not initially considered in our preliminary conclusions.

The "further analyses" mentioned referred to these subsequent experimental investigations, which included chromatin immunoprecipitation (ChIP) assays and extended sample sizes, providing a more robust dataset for evaluating the effects of A-485. We understand the importance of transparency and rigor in scientific communication. To address this, we have revised the manuscript to clearly delineate the progression of our analyses and the evidential basis for our revised understanding of A-485's effects. This includes a detailed description of the methodologies employed in our follow-up experiments (line 537 on page 27), the statistical approaches for data analysis (lines 226-227 in supporting information), and how these led to the updated interpretation regarding A-485's impact on histone acetylation (lines232-269).

(6) Comment from Reviewer 2 asking for the difference in the ChIP peaks representation of the y-axis on the ChIP traces.

Thank you for raising this quest. Actually, we did not normalise the sequencing depth and the y-axis represents the number of counts (line 537 on page 27 and lines 226-227 in supporting information).

(7) Comment from Reviewer 2 suggesting the possibility of testing this 2C protocol on mESCs to see if similar changes are subject to and how these mouse RCCs differ transcriptionally from Isl1+ progenitor cells isolated from neonatal mice (P1-P5)?

Thank you for your insightful questions. Testing the 2C protocol on mouse embryonic stem cells (mESCs) to observe if similar changes occur presents an excellent opportunity to further validate the versatility and applicability of our findings across different stem cell models. We agree that such experiments would not only strengthen the current study but also provide valuable insights into the conservation of mechanisms across species. We are currently in the process of setting up experiments to address this very question and anticipate that the results will significantly contribute to our understanding of cardiomyocyte differentiation processes. Regarding the transcriptional comparison between mouse regenerative cardiac cells (RCCs) induced by our 2C protocol and Isl1+ progenitors isolated from neonatal mice (P1-P5), this comparison is indeed crucial for delineating the specific identity and developmental potential of the RCCs generated. However, a comprehensive side-by-side transcriptomic analysis is required to systematically identify these differences and understand their biological implications. We plan to undertake this analysis as part of our future studies, which will include detailed RNA sequencing and comparative gene expression profiling to elucidate the transcriptional similarities and differences between these cell populations. These future directions will enhance our current findings, provide a deeper mechanistic understanding, and confirm the potential of the 2C protocol in regenerative medicine applications. We appreciate the reviewer's suggestions and acknowledge the importance of these experiments in advancing the field.

(8) Comment from Reviewer 2 with a suggestion to have a precise clarification of statistics & data acquisition.

As suggested by the reviewer, we have revised clarifications to make them clearer (lines 228-233 in supporting information and a precise description of each paragraph involving statistical analyses).

Reviewer 3

The findings may have a translation potential. The idea of promoting the regenerative capacity of the heart by reprogramming CMs into RCCs is interesting.

Thank you for your appreciation of the significance and translational potential of our findings!

(1) Comment from Reviewer 3 suggesting the mechanism involved in the 2C-mediated generation of RCCs is unclear and the lead found in the RAN-seq and ChIP-seq are not experimatally validated.

We acknowledge the reviewer's concern regarding the lack of experimental validation for the mechanisms identified through RNA-seq and ChIP-seq analyses in the generation of RCCs from the 2C state. We understand the importance of substantiating these molecular leads with empirical data to strengthen our conclusions. Currently, our findings are based on in-depth bioinformatic analyses, which have provided us with valuable insights and a strong basis for hypothesis generation. Moving forward, we plan to prioritize experimental validation of key pathways and targets identified in our study. This will include designing targeted experiments to elucidate the functional roles of these mechanisms in the 2C-mediated generation of RCCs. We appreciate the opportunity to clarify our approach and future directions, and we are committed to addressing this gap in subsequent work.

(2) Comment from Reviewer 3 considering the very low number of RCCs (0.6%-1.5% of cells) generated cannot protect the heart from MI, and whether 2C affects the the survival or metabolism of existing CM under hypoxia conditions, and what percentage of cells are regenerated by 2C treatment post-MI?

We appreciate the reviewer's insightful queries regarding the protective effects of 2C treatment against myocardial infarction (MI) given the low percentage of RCCs generated. It is our hypothesis that the benefits of 2C treatment extend beyond mere cell numbers. We propose that 2C may enhance the survival and metabolic resilience of existing CMs under hypoxic conditions, thereby contributing to cardiac protection post-MI. Our future investigations will aim to quantify the precise percentage of cells regenerated by 2C treatment post-MI and explore its broader impacts on cardiac tissue survival and repair mechanisms.

(3) Comment from Reviewer 3 suggesting the administration of 2C in mice, as well as whether 2C affects cardiac function under basal conditions and any physiology in mice, and the need to examine cardiac structural and functional parameters after administration of 2C.

We appreciate the reviewer's interest in the potential effects of 2C administration on cardiac function and overall physiology in mice. While we observed a decrease in body weight at P5 compared to controls, our immunofluorescence staining did not indicate any changes in cardiac structure (Figure 4— figure supplement 1E). This suggests that while 2C administration impacts neonatal rat physiology, it does not adversely affect cardiac structure under basal conditions. Further investigations are planned to assess the functional parameters of the heart post-2C administration to comprehensively understand its effects.

(4) Comment from Reviewer 3 suggesting the potential effects of 2C on other cell types of the heart, including fibroblasts and endothelial cells, in vitro and in vivo.

We value the reviewer's suggestion to explore the effects of 2C on various cardiac cell types, including fibroblasts and endothelial cells, both in vitro and in vivo. We acknowledge the importance of understanding the broader impact of 2C treatment across different cell populations within the heart, given its potential protective effects. To address this, we are designing a series of experiments to assess 2C's influence on these cell types, aiming to elucidate any changes in their behavior, proliferation, and function following treatment. This comprehensive approach will allow us to better understand the mechanistic basis of 2C's cardioprotective effects.

(5) Comment from Reviewer 3 suggesting validation the effect of 2C in a dose-dependent manner.

As suggested by the reviewer, we have supplemented the effect of 2C in dose-dependent (Figure 1— figure supplement 1F-G).

(6) Comment from Reviewer 3 suggesting an explanation of how A-485 affects H3K27Ac and H3K9Ac.

We appreciate the reviewer pointing out the discrepancy regarding the effects of A-485 on H3K27Ac and H3K9Ac. Upon re-examination of our data, we realize that our initial interpretation may have overlooked the broader impact of A-485 on histone acetylation patterns. It appears that A-485 does indeed influence both H3K27Ac and H3K9Ac, contrary to our initial statement. This oversight will be corrected in our revised manuscript, where we will provide a more detailed analysis and discussion of A-485's impact on these histone marks, alongside an explanation for the observed effects (lines 223-269 across page 8-9).

(7) Comment from Reviewer 3 with a correction to use "regeneration" at the screeing stage.

As suggested by the reviewer, we have amended the wording in the text (line 66 on page 3).

Reviewer 4

Comment from Reviewer 4 suggesting more information that clarifies and justifies the hypothesis.

As suggested by the reviewer, we added more information to clarify and justify the hypothesis (lines 39-47 on page 3).

(1) Comment from Reviewer 4 pointing out the story line is not well developed.