eLife Assessment

This valuable study builds on previous work by the authors by presenting a potentially key method for correcting optical aberrations in GRIN lens-based micro endoscopes used for imaging deep brain regions. By combining simulations and experiments, the authors show that the obtained field of view is significantly increased with corrected, versus uncorrected microendoscopes. The evidence supporting the claims of the authors is solid, although some aspects of the manuscript should be clarified and missing information provided. Because the approach described in this paper does not require any microscope or software modifications, it can be readily adopted by neuroscientists who wish to image neuronal activity deep in the brain.

Reviewer #1 (Public review):

Summary:

Sattin, Nardin, and colleagues designed and evaluated corrective microlenses that increase the useable field of view of two long (>6mm) thin (500 um diameter) GRIN lenses used in deep-tissue two-photon imaging. This paper closely follows the thread of earlier work from the same group (e.g. Antonini et al, 2020; eLife), filling out the quiver of available extended-field-of-view 2P endoscopes with these longer lenses. The lenses are made by a molding process that appears practical and easy to adopt with conventional two-photon microscopes.

Simulations are used to motivate the benefits of extended field of view, demonstrating that more cells can be recorded, with less mixing of signals in extracted traces, when recorded with higher optical resolution. In vivo tests were performed in the piriform cortex, which is difficult to access, especially in chronic preparations.

The design, characterization, and simulations are clear and thorough, but not exhaustive (see below), and do not break new ground in optical design or biological application. However, the approach shows much promise, including for applications not mentioned in the present text such as miniaturized GRIN-based microscopes. Readers will largely be interested in this work for practical reasons: to apply the authors' corrected endoscopes.

Strengths:

The text is clearly written, the ex vivo analysis is thorough and well-supported, and the figures are clear. The authors achieved their aims, as evidenced by the images presented, and were able to make measurements from large numbers of cells simultaneously in vivo in a difficult preparation.

Weaknesses:

(1) The novelty of the present work over previous efforts from the same group is not well explained. What needed to be done differently to correct these longer GRIN lenses?

(2) Some strong motivations for the method are not presented. For example, the introduction (page 3) focuses on identifying neurons with different coding properties, but this can be done with electrophysiology (albeit with different strengths and weaknesses). Compared to electrophysiology, optical methods more clearly excel at genetic targeting, subcellular measurements, and molecular specificity; these could be mentioned. Another example, in comparing microfabricated lenses to other approaches, an unmentioned advantage is miniaturization and potential application to mini-2P microscopes, which use GRIN lenses.

(3) Some potentially useful information is lacking, leaving critical questions for potential adopters:

How sensitive is the assembly to decenter between the corrective optic and the GRIN lens? What is the yield of fabrication and of assembly?

Supplementary Figure 1: Is this really a good agreement between the design and measured profile? Does the figure error (~10 um in some cases on average) noticeably degrade the image? How do individual radial profiles compare to the presented means?<br /> What is the practical effect of the strong field curvature? Are the edges of the field, which come very close to the lens surface, a practical limitation?

The lenses appear to be corrected for monochromatic light; high-performance microscopes are generally achromatic. Is the bandwidth of two-photon excitation sufficient to warrant optimization over multiple wavelengths?

GRIN lenses are often used to access a 3D volume by scanning in z (including in this study). How does the corrective lens affect imaging performance over the 3D field of view?

(4) The in vivo images (Figure 7D) have a less impressive resolution and field than the ex vivo images (Figure 4B), and the reason for this is not clear. Given the difference in performance, how does this compare to an uncorrected endoscope in the same preparation? Is the reduced performance related to uncorrected motion, field curvature, working distance, etc? Regarding Figure 7, there is no analysis of the biological significance of the calcium signals or even a description of where olfactory stimuli were presented. The timescale of jGCaMP8f signals in Figure 7E is uncharacteristically slow for this indicator (compared to Zhang et al 2023 (Nature)), though perhaps this is related to the physiology of these cells or the stimuli.

(5) The claim of unprecedented spatial resolution across the FOV (page 18) is hard to evaluate and is not supported by references to quantitative comparisons. The promises of the method for future studies (pages 18-19) could also be better supported by analysis or experiment, but these are minor and to me, do not detract from the appeal of the work.

(6) The text is lengthy and the material is repeated, especially between the introduction and conclusion. Consolidating introductory material to the introduction would avoid diluting interesting points in the discussion.

Reviewer #2 (Public review):

In this manuscript, the authors present an approach to correct GRIN lens aberrations, which primarily cause a decrease in signal-to-noise ratio (SNR), particularly in the lateral regions of the field-of-view (FOV), thereby limiting the usable FOV. The authors propose to mitigate these aberrations by designing and fabricating aspherical corrective lenses using ray trace simulations and two-photon lithography, respectively; the corrective lenses are then mounted on the back aperture of the GRIN lens.

This approach was previously demonstrated by the same lab for GRIN lenses shorter than 4.1 mm (Antonini et al., eLife, 2020). In the current work, the authors extend their method to a new class of GRIN lenses with lengths exceeding 6 mm, enabling access to deeper brain regions as most ventral regions of the mouse brain. Specifically, they designed and characterized corrective lenses for GRIN lenses measuring 6.4 mm and 8.8 mm in length. Finally, they applied these corrected long micro-endoscopes to perform high-precision calcium signal recordings in the olfactory cortex.

Compared with alternative approaches using adaptive optics, the main strength of this method is that it does not require hardware or software modifications, nor does it limit the system's temporal resolution. The manuscript is well-written, the data are clearly presented, and the experiments convincingly demonstrate the advantages of the corrective lenses.

The implementation of these long corrected micro-endoscopes, demonstrated here for deep imaging in the mouse olfactory bulb, will also enable deep imaging in larger mammals such as rats or marmosets.

Reviewer #3 (Public review):

Summary:

This work presents the development, characterization, and use of new thin microendoscopes (500µm diameter) whose accessible field of view has been extended by the addition of a corrective optical element glued to the entrance face. Two micro endoscopes of different lengths (6.4mm and 8.8mm) have been developed, allowing imaging of neuronal activity in brain regions >4mm deep. An alternative solution to increase the field of view could be to add an adaptive optics loop to the microscope to correct the aberrations of the GRIN lens. The solution presented in this paper does not require any modification of the optical microscope and can therefore be easily accessible to any neuroscience laboratory performing optical imaging of neuronal activity.

Strengths:

(1) The paper is generally clear and well-written. The scientific approach is well structured and numerous experiments and simulations are presented to evaluate the performance of corrected microendoscopes. In particular, we can highlight several consistent and convincing pieces of evidence for the improved performance of corrected micro endoscopes:<br /> a) PSFs measured with corrected micro endoscopes 75µm from the centre of the FOV show a significant reduction in optical aberrations compared to PSFs measured with uncorrected micro endoscopes.<br /> b) Morphological imaging of fixed brain slices shows that optical resolution is maintained over a larger field of view with corrected micro endoscopes compared to uncorrected ones, allowing neuronal processes to be revealed even close to the edge of the FOV.<br /> c) Using synthetic calcium data, the authors showed that the signals obtained with the corrected microendoscopes have a significantly stronger correlation with the ground truth signals than those obtained with uncorrected microendoscopes.

(2) There is a strong need for high-quality micro endoscopes to image deep brain regions in vivo. The solution proposed by the authors is simple, efficient, and potentially easy to disseminate within the neuroscience community.

Weaknesses:

(1) Many points need to be clarified/discussed. Here are a few examples:

a) It is written in the methods: « The uncorrected microendoscopes were assembled either using different optical elements compared to the corrected ones or were obtained from the corrected probes after the mechanical removal of the corrective lens. »<br /> This is not very clear: the uncorrected microendoscopes are not simply the unmodified GRIN lenses?

b) In the results of the simulation of neuronal activity (Figure 5A, for example), the neurons in the center of the FOV have a very large diameter (of about 30µm). This should be discussed. Also, why is the optical resolution so low on these images?

c) It seems that we can't see the same neurons on the left and right panels of Figure 5D. This should be discussed.

d) It is not very clear to me why in Figure 6A, F the fraction of adjacent cell pairs that are more correlated than expected increases as a function of the threshold on peak SNR. The authors showed in Supplementary Figure 3B that the mean purity index increases as a function of the threshold on peak SNR for all micro endoscopes. Therefore, I would have expected the correlation between adjacent cells to decrease as a function of the threshold on peak SNR. Similarly, the mean purity index for the corrected short microendoscope is close to 1 for high thresholds on peak SNR: therefore, I would have expected the fraction of adjacent cell pairs that are more correlated than expected to be close to 0 under these conditions. It would be interesting to clarify these points.

e) Figures 6C, H: I think it would be fairer to compare the uncorrected and corrected endomicroscopes using the same effective FOV.

f) Figure 7E: Many calcium transients have a strange shape, with a very fast decay following a plateau or a slower decay. Is this the result of motion artefacts or analysis artefacts? Also, the duration of many calcium transients seems to be long (several seconds) for GCaMP8f. These points should be discussed.

g) The authors do not mention the influence of the neuropil on their data. Did they subtract the neuropil's contribution to the signals from the somata? It is known from the literature that the presence of the neuropil creates artificial correlations between neurons, which decrease with the distance between the neurons (Grødem, S., Nymoen, I., Vatne, G.H. et al. An updated suite of viral vectors for in vivo calcium imaging using intracerebral and retro-orbital injections in male mice. Nat Commun 14, 608 (2023). https://doi.org/10.1038/s41467-023-36324-3; Keemink SW, Lowe SC, Pakan JMP, Dylda E, van Rossum MCW, Rochefort NL. FISSA: A neuropil decontamination toolbox for calcium imaging signals. Sci Rep. 2018 Feb 22;8(1):3493. doi: 10.1038/s41598-018-21640-2. PMID: 29472547; PMCID: PMC5823956)<br /> This point should be addressed.

h) Also, what are the expected correlations between neurons in the pyriform cortex? Are there measurements in the literature with which the authors could compare their data?

(2) The way the data is presented doesn't always make it easy to compare the performance of corrected and uncorrected lenses. Here are two examples:

a) In Figures 4 to 6, it would be easier to compare the FOVs of corrected and uncorrected lenses if the scale bars (at the centre of the FOV) were identical. In this way, the neurons at the centre of the FOV would appear the same size in the two images, and the distances between the neurons at the centre of the FOV would appear similar. Here, the scale bar is significantly larger for the corrected lenses, which may give the illusion of a larger effective FOV.

b) In Figures 3A-D it would be more informative to plot the distances in microns rather than pixels. This would also allow a better comparison of the micro endoscopes (as the pixel sizes seem to be different for the corrected and uncorrected micro endoscopes).

(3) There seems to be a discrepancy between the performance of the long lenses (8.8mm) in the different experiments, which should be discussed in the article. For example, the results in Figure 4 show a considerable enlargement of the FOV, whereas the results in Figure 6 show a very moderate enlargement of the distance at which the person's correlation with the first ground truth emitter starts to drop.

a) There is also a significant discrepancy between measured and simulated optical performance, which is not discussed. Optical simulations (Figure 1) show that the useful FOV (defined as the radius for which the size of the PSF along the optical axis remains below 10µm) should be at least 90µm for the corrected microendoscopes of both lengths. However, for the long microendoscopes, Figure 3J shows that the axial resolution at 90µm is 17µm. It would be interesting to discuss the origin of this discrepancy: does it depend on the microendoscope used? Are there inaccuracies in the construction of the aspheric corrective lens or in the assembly with the GRIN lens? If there is variability between different lenses, how are the lenses selected for imaging experiments?

Reviewer #2 (Public review):

Summary:

This manuscript builds on the authors' 2020 study by combining tissue expansion with light sheet microscopy to quantify the organism-wide spatial distribution of various cell types in the planarian.

Strengths:

(1) The quantification of cell types as a function of animal size and regeneration stages could be a useful resource for the planarian research community.

(2) The high-quality images can help clarify some anatomical structures within the planarian tissues.

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.

(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.

(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).

(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.

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

(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.

(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.

Reviewer #3 (Public review):

Summary:

In this manuscript, the authors apply tissue expansion and tiling light sheet microscopy to study allometric growth and regeneration in planaria. They developed image analysis pipelines to help them quantify different neuronal subtypes and muscles in planaria of different sizes and during regeneration. Among the strengths of this work, the authors provide beautiful images that show the potential of the approaches they are taking and their ability to quantify specific cell types in relatively large numbers of whole animal samples. Many of their findings confirm previous results in the literature, which helps validate the techniques and pipelines they have applied here. Among their new observations, they find that the body wall muscles at the anterior and posterior poles of the worm are organized differently and show that the muscle pattern in the posterior head of beta-catenin RNAi worms resembles the anterior muscle pattern. They also show that glial cell processes appear to be altered in beta-catenin or insulin receptor-1 RNAi worms. Weaknesses include some over-interpretation of the data and lack of consideration or citation of relevant previous literature, as discussed below.

Strengths:

This method of tissue expansion will be useful for researchers interested in studying this experimental animal. The authors provide high-quality images that show the utility of this technique. Their analysis pipeline permits them to quantify cell types in relatively large numbers of whole animal samples.

The authors provide convincing data on changes in total neurons and neuronal sub-types in different-sized planaria. They report differences in body wall muscle pattern between the anterior and posterior poles of the planaria, and that these differences are lost when a posterior head forms in beta-catenin RNAi planaria. They also find that glial cell projections are reduced in insulin receptor-1 RNAi planaria.

Weaknesses:

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.

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.

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.

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.

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.

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 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.

eLife Assessment

The study provides valuable insight into the biological significance of SARS-CoV-2 by using a series of computational analyses of viral proteins. While evidence is solid, it is obscured by a lack of clarity about the objectives of the analyses and in the overall writing of the article. The study will be impactful to the researchers in the field but will benefit from improved presentation.

Reviewer #1 (Public review):

Summary:

Park et al. conducted various analyses attempting to elucidate the biological significance of SARS-CoV-2 mutations. However, the study lacks a clear objective. The specific goals of the analyses in each subsection are unclear, as is how the results from these subsections are interconnected. Compiling results from unrelated analyses into a single paper can be confusing for readers. Clarifying the objective and narrowing down the topics would make the paper's purpose clearer.

The logic of the study is also unclear. For instance, the authors developed an evaluation score, APESS, for analyzing viral sequences. Although they state that the APESS score correlates with viral infectivity, there is no explanation in the results section about why this is the case.

The structure of the paper should be reconsidered.

Reviewer #2 (Public review):

Summary:

The authors have developed a machine learning tool AIVE to predict the infectivity of SARS-CoV-2 variants and also a scoring metric to measure infectivity. A large number of virus sequences were used with a very detailed analysis that incorporates hydrophobic, hydrophilic, acid, and alkaline characteristics. The protein structures were also considered to measure infectivity and search for core mutations. The study especially focused on the S protein of SARS-CoV-2. The contents of this study would be of interest to many researchers related to this area and the web service would be helpful to easily analyze such data without in-depth bioinformatics expertise.

Strengths:

- Analysis of large-scale data.

- Experimental validation on a partial set of searched mutations.

- A user-friendly web-based analysis platform that is made public.

Weaknesses:

- Complexity of the research.

eLife Assessment

This important paper employs multiple experimental approaches and presents evidence that changes in membrane voltage directly affect ERK signaling to regulate cell division. This result is relevant because it supports an ion channel-independent pathway by which changes in membrane voltage can affect cell growth. The reviewers point out that some experimental results and interpretations are compelling, but the strength of evidence is incomplete and additional experiments are needed to rule out other possible interpretations of the data.

Reviewer #1 (Public review):

Summary:

This is a contribution to the field of developmental bioelectricity. How do changes of resting potential at the cell membrane affect downstream processes? Zhou et al. reported in 2015 that phosphatidylserine and K-Ras cluster upon plasma membrane depolarization and that voltage-dependent ERK activation occurs when constitutive active K-RasG12V mutants are overexpressed. In this paper, the authors advance the knowledge of this phenomenon by showing that membrane depolarization up-regulates mitosis and that this process is dependent on voltage-dependent activation of ERK. ERK activity's voltage-dependence is derived from changes in the dynamics of phosphatidylserine in the plasma membrane and not by extracellular calcium dynamics.

Strengths:

Bioelectricity is an important field for areas of cell, developmental, and evolutionary biology, as well as for biomedicine. Confirmation of ERK as a transduction mechanism, and a characterization of the molecular details involved in control of cell proliferation, is interesting and impactful.

Weaknesses:

The functional cell division data need to be stronger. They show that increasing K+ increases proliferation and argue that since a MEK inhibitor (U0126) reduces proliferation in K+ treated cells, K+ induces cell division via ERK. But I don't see statistics to show that the rescue is significant, and I don't see a key U0126-only control. If the U0126 alone reduces proliferation, the combined effect wouldn't prove much.

Also, unless I'm missing something, it looks like every sample in their control has exactly the same number of mitotic cells. I understand that they are normalizing to this column, but shouldn't they be normalizing to the mean, with the independent values scattering around 1? It doesn't seem like it can be paired replicates since there are 6 replicates in the control and 4 replicates in one of the conditions?

Reviewer #2 (Public review):

Sasaki et al. use a combination of live-cell biosensors and patch-clamp electrophysiology to investigate the effect of membrane potential on the ERK MAPK signaling pathway, and probe associated effects on proliferation. This is an effect that has long been proposed, but convincing demonstration has remained elusive, because it is difficult to perturb membrane potential without disturbing other aspects of cell physiology in complex ways. The time-resolved measurements here are a nice contribution to this question, and the perforated patch clamp experiments with an ERK biosensor are fantastic - they come closer to addressing the above difficulty of perturbing voltage than any prior work. It would have been difficult to obtain these observations with any other combination of tools.

However, there are still some concerns as detailed in specific comments below:

Specific comments:<br /> (1) All the observations of ERK activation, by both high extracellular K+ and voltage clamp, could be explained by cell volume increase (more discussion in subsequent comments). There is a substantial literature on ERK activation by hypotonic cell swelling (e.g. https://doi.org/10.1042/bj3090013, https://doi.org/10.1002/j.1460-2075.1996.tb00938.x, among others). Here are some possible observations that could demonstrate that ERK activation by volume change is distinct from the effects reported here:<br /> (i) Does hypotonic shock activate ERK in U2OS cells?<br /> (ii) Can hypotonic shock activate ERK even after PS depletion, whereas extracellular K+ cannot?<br /> (iii) Does high extracellular K+ change cell volume in U2OS cells, measured via an accurate method such as fluorescence exclusion microscopy?<br /> (iv) It would be helpful to check the osmolality of all the extracellular solutions, even though they were nominally targeted to be iso-osmotic.

(2) Some more details about the experimental design and the results are needed from Figure 1:<br /> (i) For how long are the cells serum-starved? From the Methods section, it seems like the G1 release in different K+ concentration is done without serum, is this correct? Is the prior thymidine treatment also performed in the absence of serum?<br /> (ii) There is a question of whether depolarization constitutes a physiologically relevant mechanism to regulate proliferation, and how depolarization interacts with other extracellular signals that might be present in an in vivo context. Does depolarization only promote proliferation after extended serum starvation (in what is presumably a stressed cell state)? What fraction of total cells are observed to be mitotic (without normalization), and how does this compare to the proliferation of these cells growing in serum-supplemented media? Can K+ concentration tune proliferation rate even in serum-supplemented media?

(3) In Figure 2, there are some possible concerns with the perfusion experiment:<br /> (i) Is the buffer static in the period before perfusion with high K+, or is it perfused? This is not clear from the Methods. If it is static, how does the ERK activity change when perfused with 5 mM K+? In other words, how much of the response is due to flow/media exchange versus change in K+ concentration?<br /> (ii) Why do there appear to be population-average decreases in ERK activity in the period before perfusion with high K+ (especially in contrast to Fig. 3)? The imaging period does not seem frequent enough for photobleaching to be significant.

(4) Figure 3 contains important results on couplings between membrane potential and MAPK signaling. However, there are a few concerns:<br /> (i) Does cell volume change upon voltage clamping? Previous authors have shown that depolarizing voltage clamp can cause cells to swell, at least in the whole-cell configuration: https://www.cell.com/biophysj/fulltext/S0006-3495(18)30441-7 . Could it be possible that the clamping protocol induces changes in ERK signaling due to changes in cell volume, and not by an independent mechanism?<br /> (ii) Does the -80 mV clamp begin at time 0 minutes? If so, one might expect a transient decrease in sensor FRET ratio, depending on the original resting potential of the cells. Typical estimates for resting potential in HEK293 cells range from -40 mV to -15 mV, which would reach the range that induces an ERK response by depolarizing clamp in Fig. 3B. What are the resting potentials of the cells before they are clamped to -80 mV, and why do we not see this downward transient?

(5) The activation of ERK by perforated voltage clamp and by high extracellular K+ are each convincing, but it is unclear whether they need to act purely through the same mechanism - while additional extracellular K+ does depolarize the cell, it could also be affecting function of voltage-independent transporters and cell volume regulatory mechanisms on the timescales studied. To more strongly show this, the following should be done with the HEK cells where there is already voltage clamp data:<br /> (i) Measure resting potential using the perforated patch in zero-current configuration in the high K+ medium. Ideally this should be done in the time window after high K+ addition where ERK activation is observed (10-20 minutes) to minimize the possibility of drift due to changes in transporter and channel activity due to post-translational regulation.<br /> (ii) Measure YFP/CFP ratio of the HEK cells in the high K+ medium (in contrast to the U2OS cells from Fig. 2 where there is no patch data).<br /> (iii) The assertion that high K+ is equivalent to changes in Vmem for ERK signaling would be supported if the YFP/CFP change from K+ addition is comparable to that induced by voltage clamp to the same potential. This would be particularly convincing if the experiment could be done with each of the 15 mM, 30 mM, and 145 mM conditions.

(6) Line 170: "ERK activity was reduced with a fast time course (within 1 minute) after repolarization to -80 mV." I don't see this in the data: in Fig. 3C, it looks like ERK remains elevated for > 10 min after the electrical stimulus has returned to -80 mV

Reviewer #3 (Public review):

Summary:

This paper demonstrates that membrane depolarization induces a small increase in cell entry into mitosis. Based on previous work from another lab, the authors propose that ERK activation might be involved. They show convincingly using a combination of assays that ERK is activated by membrane depolarization. They show this is Ca2+ independent and is a result of activation of the whole K-Ras/ERK cascade which results from changed dynamics of phosphatidylserine in the plasma membrane that activates K-Ras. Although the activation of the Ras/ERK pathway by membrane depolarization is not new, linking it to an increase in cell proliferation is novel.

Strengths

A major strength of the study is the use of different techniques - live imaging with ERK reporters, as well as Western blotting to demonstrate ERK activation as well as different methods for inducing membrane depolarization. They also use a number of different cell lines. Via Western blotting the authors are also able to show that the whole MAPK cascade is activated.

Weaknesses

A weakness of the study is the data in Figure 1 showing that membrane depolarization results in an increase of cells entering mitosis. There are very few cells entering mitosis in their sample in any condition. This should be done with many more cells to increase confidence in the results. The study also lacks a mechanistic link between ERK activation by membrane depolarization and increased cell proliferation.

The authors did achieve their aims with the caveat that the cell proliferation results could be strengthened. The results for the most part support the conclusions.

This work suggests that alterations in membrane potential may have more physiological functions than action potential in the neural system as it has an effect on intracellular signalling and potentially cell proliferation.

Author response:

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

As you can see from the assessment (which is unchanged from before) and the reviews included below, the reviewers felt that the revisions did not yet address all of the major concerns. There was agreement that the strength of evidence would be upgraded to "solid" by addressing, at minimum, the following: 

(1) Which of the results are significant for individual monkeys; and 

(2) How trials from different target contrasts were analyzed 

In this revision, we have addressed the two primary editorial recommendations:

(1) We apologize if this information was not clear in the previous version. We have updated Table 1 to highlight clearly the significant results for individual monkeys. Six of our key results – pupil diameter (Fig 2B), microsaccades (Fig 2D), decoding performance for narrow-spiking units (Fig 3A), decoding performance for broad-spiking units (Fig 3B), target-evoked firing rate for all units (Fig 3E) and target-evoked firing rate for broad-spiking units (Fig 3F) – are significant for individual animals and therefore gives us high confidence regarding our results. Please also note that we present all results for individual animals in the Supplementary figures accompanying each main figure.

(2) We have updated the manuscript and methods to explain how trials of each contrast were included in each analysis, and how contrast normalization was performed for the analysis in Figure 3. In addition, we discuss this point in the Discussion section, which we quote below:

“Non-target stimulus contrasts were slightly different between hits and misses (mean: 33.1% in hits, 34.0% in misses, permutation test, 𝑝 = 0.02), but the contrast of the target was higher in hits compared to misses (mean: 38.7% in hits, 27.7% in misses, permutation test, 𝑝 = 1.6   𝑒 − 31). To control for potential effects of stimulus contrast, firing rates were first normalized by contrast before performing the analyses reported in Figure 3. For all other results, we considered only non-target stimuli, which had very minor differences in contrast (<1%) across hits and misses. In fact, this minor difference was in the opposite direction of our results with mean contrast being slightly higher for misses. While we cannot completely rule out any other effects of stimulus contrast, the normalization in Figure 3 and minor differences for non-target stimuli should minimize them.”

Reviewer #1 (Public Review): 

Summary: 

In this study, Nandy and colleagues examine neural, physiological and behavioral correlates of perceptual variability in monkeys performing a visual change detection task. They used a laminar probe to record from area V4 while two macaque monkeys detected a small change in stimulus orientation that occurred at a random time in one of two locations, focusing their analysis on stimulus conditions where the animal was equally likely to detect (hit) or not-detect (miss) a briefly presented orientation change (target). They discovered two behavioral and physiological measures that are significantly different between hit and miss trials - pupil size tends to be slightly larger on hits vs. misses, and monkeys are more likely to miss the target on trials in which they made a microsaccade shortly before target onset. They also examined multiple measures of neural activity across the cortical layers and found some measures that are significantly different between hits and misses. 

Strengths: 

Overall the study is well executed and the analyses are appropriate (though several issues still need to be addressed as discussed in Specific Comments). 

Thank you.

Weaknesses: 

My main concern with this study is that, with the exception of the pre-target microsaccades, the correlates of perceptual variability (differences between hits and misses) appear to be weak, potentially unreliable and disconnected. The GLM analysis of predictive power of trial outcome based on the behavioral and neural measures is only discussed at the end of the paper. This analysis shows that some of the measures have no significant predictive power, while others cannot be examined using the GLM analysis because these measures cannot be estimated in single trials. Given these weak and disconnected effects, my overall sense is that the current results provide limited advance to our understanding of the neural basis of perceptual variability. 

Please see our response above to item #1 of the editorial recommendation. Six of our key results are individually significant in both animals giving us high confidence about the reliability and strength of our results. 

Regarding the reviewer’s comment about the GLM, we note (also stated in the manuscript) that among the measures that we could estimate reliably on a single trial basis, two of these – pre-target microsaccades and input-layer firing rates – were reliable signatures of stimulus perception at threshold. This analysis does not imply that the other measures – Fano Factor, PPC, inter-laminar population correlations, SSC (which are all standard tools in modern systems neuroscience, and which cannot be estimated on a single-trial basis) – are irrelevant. Our intent in including the GLM analyses was to complement the results reported from these across-trial measures (Figs 4-7) with the predictive power of single-trial measures.

While no study is entirely complete in itself, we have attempted to synthesize our results into a conceptual model as depicted in Fig 8.

Reviewer #2 (Public Review): 

Strengths: 

The experiments were well-designed and executed with meticulous control. The analyses of both behavioural and electrophysiological data align with the standards in the field. 

Thank you.

Weaknesses: 

Many of the findings appear to be subtle differences and incremental compared to previous literature, including the authors' own work. While incremental findings are not necessarily a problem, the manuscript lacks clear statements about the extent to which the dataset, analysis, and findings overlap with the authors' prior research. For example, one of the main findings, which suggests that V4 neurons exhibit larger visual responses in hit trials (as shown in Fig. 3), appears to have been previously reported in their 2017 paper. 

We respectfully disagree with the assessment that the findings reported here are incremental over the results reported in our prior study (Nandy et al,. 2017). In the previous study, we compared the laminar profile of neural modulation due to the deployment of attention i.e. the main comparison points were the attend-in and the attend-away conditions while controlling for visual stimulation. In this study, we go one step further and home in on the attend-in condition and investigate the differences in the laminar profile of neural activity (and two additional physiological measures: pupil and microsaccades) when the animal either correctly reports or fails to report a stimulus with equal probability. We thus control for both the visual stimulation and the cued attention state of the animal. While there are parallels to our previous results (as the reviewer correctly noted), the results reported here cannot be trivially predicted from our previous results. Please also note that we discuss our new results in the context of prior results, from both our group and others, in the manuscript (lines 310-332).

Furthermore, the manuscript does not explore potentially interesting aspects of the dataset. For instance, the authors could have investigated instances where monkeys made 'false' reports, such as executing saccades towards visual stimuli when no orientation change occurred, which allows for a broader analysis that considers the perceptual component of neural activity over pure sensory responses. Overall, lacking broad interest with the current form.

We appreciate the reviewer’s feedback on analyzing false alarm trials. Our focus for this study was to investigate the behavioral and neural correlates accompanying a correct or incorrect perception of a target stimulus presented at perceptual threshold. False alarm trials, by definition, do not include a target presentation. Moreover, false alarm rates rapidly decline with duration into a trial, with high rates during the first non-target presentation and rates close to zero by the time of the eighth presentation (see figure). Investigating false alarms will thus involve a completely different form of analysis than we have undertaken here. We therefore feel that while analyzing false alarm trials will be an interesting avenue to pursue in the future, it is outside the scope of the present study.

Author response image 1.

eLife Assessment

This useful study by Nandy and colleagues examined relationships between behavioral state, neural activity in cortical area V4, and trial-by-trial variability in the ability to detect weak visual stimuli. They present solid evidence indicating that certain changes in arousal and eye-position stability, along with patterns of synchrony in the activity of neurons in different layers of V4, can show modest correspondences to changes in the ability to correctly detect a stimulus. These findings are likely to be of interest to those who seek a deeper understanding of circuit mechanisms that underlie perception.

Reviewer #1 (Public review):

Summary:

In this study, Nandy and colleagues examine neural, physiological and behavioral correlates of perceptual variability in monkeys performing a visual change detection task. They used a laminar probe to record from area V4 while two macaque monkeys detected a small change in stimulus orientation that occurred at a random time in one of two locations, focusing their analysis on stimulus conditions where the animal was equally likely to detect (hit) or not-detect (miss) a briefly presented orientation change (target). They discovered two behavioral and physiological measures that are significantly different between hit and miss trials - pupil size tends to be slightly larger on hits vs. misses, and monkeys are more likely to miss the target on trials in which they made a microsaccade shortly before target onset. They also examined multiple measures of neural activity across the cortical layers and found some measures that are significantly different between hits and misses.

Strengths:

Overall the study is well executed and the analyses are appropriate (with some possible caveats discussed below).

Weaknesses:

I have two remaining concerns. First, with the exception of the pre-target microsaccades, the correlates of perceptual variability (differences between hits and misses) appear to be weak and disconnected. The GLM analysis of the predictive power of trial outcome based on the behavioral and neural measures is only discussed at the end of the paper. This analysis shows that some of the measures have no significant predictive power, while others cannot be examined using the GLM analysis because these measures cannot be estimated in single trials. Given these weak and disconnected effects, my overall sense is that the current results provide a limited advance to our understanding of the neural basis of perceptual variability.

In addition, because the authors combine data across stimulus contrasts, I am somewhat uneasy about the possible confounding effect of contrast. As expected, stimulus contrast affected the probability of hits vs. misses. Independently, contrast may have affected some of the physiological measurements. Therefore, showing that contrast is not the source of the covariations between the physiological/behavioral measurements and perception can be challenging, and I am not convinced that the authors have ruled this out as a possible confound. It is unclear why the authors had to vary contrast in the first place, and why the analyses had to be done by combining the data across contrasts or by ignoring contrast as a variable (e.g., in the GLM analysis).

eLife Assessment

This study presents a valuable finding on the identification of a complex consisting of NHE1, hERG1, β1/integrin and NaV1.5 on the membrane of breast cancer cells. The evidence supporting the claims of the authors is somewhat incomplete. The inclusion of clarification of some experimental design and the amendment of cropping Western blot data would have strengthened the study. The work will be of interest to scientists working on breast cancer.

Reviewer #1 (Public Review):

This manuscript by Capitani et al. extends previous studies of ion channel expression in triple-negative breast cancer cell lines. Probing four phenotypically different breast cancer cell lines, they used co-IP and confocal immunofluorescence (IF) colocalization to reveal that beta1 integrin forms a complex with the neonatal form of the Na+ channel NaV1.5 (nNaV1.5) and the Na+/H+ antiporter NHE1 in addition to previously reported hERG1. They used siRNA to show that silencing beta1 results in a co-depletion of hERG and Nav1.5, further supporting the conclusion that they form a complex; a complementary enhancement of Na current with increased hERG expression was also demonstrated. These data compellingly describe a complex of membrane proteins unregulated in breast cancer and thus present novel potential targets for treatment.

There are several concerns with experimental approaches. How fluorescence measurements were compared and controlled among experiments was not described, and masks drawn to define membrane expression seemed arbitrary, and included in some cases large sections of cytoplasm. There are issues associated with the use of channel blocking agents and a bifunctional small-chain antibody that are not well rationalized. Why are they being used, to test what hypotheses or disrupt what processes? The extremely high concentrations of E-4031 (4000x IC50 for block), e.g., are not expected to have selective actions. The effects of E-4031 at high concentrations altering cytoskeleton properties associated with invasiveness (and thus cancer progression) are questionable. There are numerous problems with co-IPs together carried out together with knock-down, which in one case depleted the protein targeted by the primary IP antibody. Western blots (WB) were quantified by comparing treatment to control, which does not control for loading errors. The control and treated signals should be divided by the respective tubulin signals to control for loading errors. Then the treated value can be compared with the control.

Reviewer #2 (Public Review):

The manuscript by Chiara Capitani and Annarosa Arcangeli reports the identification of a complex comprising NHE1,hERG1, β1 integrin, and NaV1.5 on the plasma membrane of breast cancer cells. The authors further investigated the mutual regulatory interactions among these proteins using Western blotting and co-immunoprecipitation assays. They also examined the downstream signaling pathways associated with this complex and assessed its impact on the malignant behavior of breast cancer cells.

Strengths

The manuscript used different breast cancer cell lines and combined Western blot, immunostaining, and electrophysiology to provide evidence for the proposed complex. The inhibitors are also used to test the requirement of channel activity to function in the development of breast cancer cells with in-vitro studies.

Weaknesses

The data shown in this manuscript include the western blots that are cropped and imaged separately to draw conclusions about protein levels and changes in immunoprecipitation. These cannot be done on separate, cropped blots but must be imaged together to make these comparisons.

Antibodies used for hERG, NaV1.5 and β1 integrin must be validated to work for IP using KO or KD cell lines for the respective proteins to demonstrate specificity. The same goes for all the immunofluorescence imaging shown in the manuscript as these are all key pieces of data to support the conclusions.

eLife Assessment

SGLT2 inhibitors (SGLT2i) have assumed important roles in reducing cardiovascular risk, particularly in those with diabetes. It has become appreciated that its protective effects are likely beyond their ability to lower blood sugar levels. This research presents a novel approach to studying the SGLT2i mechanism of action which is yet to be fully elucidated.

Reviewer #1 (Public Review):

The authors examined the hypothesis that plasma ApoM, which carries sphingosine-1-phosphate (S1P) and activates vascular S1P receptors to inhibit vascular leakage, is modulated by SGLT2 inhibitors (SGLTi) during endotoxemia. They also propose that this mechanism is mediated by SGLTi regulation of LRP2/ megalin in the kidney and that this mechanism is critical for endotoxin-induced vascular leak and myocardial dysfunction. The hypothesis is novel and potentially exciting. However, the author's experiments lack critical controls, lack rigor in multiple aspects, and overall does not support the conclusions.

Reviewer #2 (Public Review):

Apolipoprotein M (ApoM) is a plasma carrier for the vascular protective lipid mediator sphingosine 1-phospate (S1P). The plasma levels of S1P and its chaperones ApoM and albumin rapidly decline in patients with severe sepsis, but the mechanisms for such reductions and their consequences for cardiovascular health remain elusive. In this study, Ripoll and colleagues demonstrate that the sodium-glucose co-transporter inhibitor dapagliflizin (Dapa) can preserve serum ApoM levels as well as cardiac function after LPS treatment of mice with diet-induced obesity. They further provide data to suggest that Dapa preserves serum ApoM by increasing megalin-mediated reabsorption of ApoM in renal proximal tubules and that ApoM improves vascular integrity in LPS treated mice. These observations put forward a potential therapeutic approach to sustain vascular protective S1P signaling that could be relevant to other conditions of systemic inflammation where plasma levels of S1P decrease. However, although the authors are careful with their statements, the study falls short of directly implicating megalin in ApoM reabsorption and of ApoM/S1P depletion in LPS-induced cardiac dysfunction and the protective effects of Dapa.

The observations reported in this study are exciting and potentially of broad interest. The paper is well written and concise, and the statements made are mostly supported by the data presented. However, the mechanism proposed and implied is mostly based on circumstantial evidence, and the paper could be substantially improved by directly addressing the role of megalin in ApoM reabsorption and serum ApoM and S1P levels and the importance of ApoM for the preservation for cardiac function during endotoxemia. Some observations that are not necessarily in line with the model proposed should also be discussed.

The authors show that Dapa preserves serum ApoM and cardiac function in LPS-treated obese mice. However, the evidence they provide to suggest that ApoM may be implicated in the protective effect of Dapa on cardiac function is indirect. Direct evidence could be sought by addressing the effect of Dapa on cardiac function in LPS treated ApoM deficient and littermate control mice (with DIO if necessary).

The authors also suggest that higher ApoM levels in mice treated with Dapa and LPS reflect increased megalin-mediated ApoM reabsorption and that this preserves S1PR signaling. This could be addressed more directly by assessing the clearance of labelled ApoM, by addressing the impact of megalin inhibition or deficiency on ApoM clearance in this context, and by measuring S1P as well as ApoM in serum samples.

Methods: More details should be provided in the manuscript for how ApoM deficient and transgenic mice were generated, on sex and strain background, and on whether or not littermate controls were used. For intravital microscopy, more precision is needed on how vessel borders were outland and if this was done with or without regard for FITC-dextran. Please also specify the type of vessel chosen and considerations made with regard to blood flow and patency of the vessels analyzed. For statistical analyses, data from each mouse should be pooled before performing statistical comparisons. The criteria used for choice of test should be outlined as different statistical tests are used for similar datasets. For all data, please be consistent in the use of post-tests and in the presentation of comparisons. In other words, if the authors choose to only display test results for groups that are significantly different, this should be done in all cases. And if comparisons are made between all groups, this should be done in all cases for similar sets of data.

Reviewer #3 (Public Review):

The authors have performed well designed experiments that elucidate the protective role of Dapa in sepsis model of LPS. This model shows that Dapa works, in part, by increasing expression of the receptor LRP2 in the kidney, that maintains circulating ApoM levels. ApoM binds to S1P which then interacts with the S1P receptor stimulating cardiac function, epithelial and endothelial barrier function, thereby maintaining intravascular volume and cardiac output in the setting of severe inflammation. The authors used many experimental models, including transgenic mice, as well as several rigorous and reproducible techniques to measure the relevant parameters of cardiac, renal, vascular, and immune function. Furthermore, they employ a useful inhibitor of S1P function to show pharmacologically the essential role for this agonist in most but not all the benefits of Dapa. A strength of the paper is the identification of the pathway responsible for the cardioprotective effects of SGLT2is that may yield additional therapeutic targets. There are some weaknesses in the paper, such as, studying only male mice, as well as providing a power analysis to justify the number of animals used throughout their experimentation. Overall, the paper should have a significant impact on the scientific community because the SGLT2i drugs are likely to find many uses in inflammatory diseases and metabolic diseases. This paper provides support for an important mechanism by which they work in conditions of severe sepsis and hemodynamic compromise.

eLife Assessment

This study provides a valuable new perspective on how motor learning occurring in one state generalizes to new states (for example, a different limb posture). The proposed model improves upon previous theories in its ability to predict patterns of generalization, but evidence supporting this specific proposed model over possible alternatives is incomplete. The newly proposed theory appears promising but would be more convincing if its conceptual and theoretical basis were clearer and more rigorously derived.

Reviewer #1 (Public Review):

This paper proposes a novel framework for explaining patterns of generalization of force field learning to novel limb configurations. The paper considers three potential coordinate systems: cartesian, joint-based, and object-based. The authors propose a model in which the forces predicted under these different coordinate frames are combined according to the expected variability of produced forces. The authors show, across a range of changes in arm configurations, that the generalization of a specific force field is quite well accounted for by the model.

The paper is well-written and the experimental data are very clear. The patterns of generalization exhibited by participants - the key aspect of the behavior that the model seeks to explain - are clear and consistent across participants. The paper clearly illustrates the importance of considering multiple coordinate frames for generalization, building on previous work by Berniker and colleagues (JNeurophys, 2014). The specific model proposed in this paper is parsimonious, but there remain a number of questions about its conceptual premises and the extent to which its predictions improve upon alternative models.

A major concern is with the model's premise. It is loosely inspired by cue integration theory but is really proposed in a fairly ad hoc manner, and not really concretely founded on firm underlying principles. It's by no means clear that the logic from cue integration can be extrapolated to the case of combining different possible patterns of generalization. I think there may in fact be a fundamental problem in treating this control problem as a cue-integration problem. In classic cue integration theory, the various cues are assumed to be independent observations of a single underlying variable. In this generalization setting, however, the different generalization patterns are NOT independent; if one is true, then the others must inevitably not be. For this reason, I don't believe that the proposed model can really be thought of as a normative or rational model (hence why I describe it as 'ad hoc'). That's not to say it may not ultimately be correct, but I think the conceptual justification for the model needs to be laid out much more clearly, rather than simply by alluding to cue-integration theory and using terms like 'reliability' throughout.

A more rational model might be based on Bayesian decision theory. Under such a model, the motor system would select motor commands that minimize some expected loss, averaging over the various possible underlying 'true' coordinate systems in which to generalize. It's not entirely clear without developing the theory a bit exactly how the proposed noise-based theory might deviate from such a Bayesian model. But the paper should more clearly explain the principles/assumptions of the proposed noise-based model and should emphasize how the model parallels (or deviates from) Bayesian-decision-theory-type models.

Another significant weakness is that it's not clear how closely the weighting of the different coordinate frames needs to match the model predictions in order to recover the observed generalization patterns. Given that the weighting for a given movement direction is over-parametrized (i.e. there are 3 variable weights (allowing for decay) predicting a single observed force level, it seems that a broad range of models could generate a reasonable prediction. It would be helpful to compare the predictions using the weighting suggested by the model with the predictions using alternative weightings, e.g. a uniform weighting, or the weighting for a different posture. In fact, Fig. 7 shows that uniform weighting accounts for the data just as well as the noise-based model in which the weighting varies substantially across directions. A more comprehensive analysis comparing the proposed noise-based weightings to alternative weightings would be helpful to more convincingly argue for the specificity of the noise-based predictions being necessary. The analysis in the appendix was not that clearly described, but seemed to compare various potential fitted mixtures of coordinate frames, but did not compare these to the noise-based model predictions.

Reviewer #2 (Public Review):

Leib & Franklin assessed how the adaptation of intersegmental dynamics of the arm generalizes to changes in different factors: areas of extrinsic space, limb configurations, and 'object-based' coordinates. Participants reached in many different directions around 360{degree sign}, adapting to velocity-dependent curl fields that varied depending on the reach angle. This learning was measured via the pattern of forces expressed in upon the channel wall of "error clamps" that were randomly sampled from each of these different directions. The authors employed a clever method to predict how this pattern of forces should change if the set of targets was moved around the workspace. Some sets of locations resulted in a large change in joint angles or object-based coordinates, but Cartesian coordinates were always the same. Across three separate experiments, the observed shifts in the generalized force pattern never corresponded to a change that was made relative to any one reference frame. Instead, the authors found that the observed pattern of forces could be explained by a weighted combination of the change in Cartesian, joint, and object-based coordinates across test and training contexts.

In general, I believe the authors make a good argument for this specific mixed weighting of different contexts. I have a few questions that I hope are easily addressed.

Movements show different biases relative to the reach direction. Although very similar across people, this function of biases shifts when the arm is moved around the workspace (Ghilardi, Gordon, and Ghez, 1995). The origin of these biases is thought to arise from several factors that would change across the different test and training workspaces employed here (Vindras & Viviani, 2005). My concern is that the baseline biases in these different contexts are different and that rather the observed change in the force pattern across contexts isn't a function of generalization, but a change in underlying biases. Baseline force channel measurements were taken in the different workspace locations and conditions, so these could be used to show whether such biases are meaningfully affecting the results.

Experiment 3, Test 1 has data that seems the worst fit with the overall story. I thought this might be an issue, but this is also the test set for a potentially awkwardly long arm. My understanding of the object-based coordinate system is that it's primarily a function of the wrist angle, or perceived angle, so I am a little confused why the length of this stick is also different across the conditions instead of just a different angle. Could the length be why this data looks a little odd?

The manuscript is written and organized in a way that focuses heavily on the noise element of the model. Other than it being reasonable to add noise to a model, it's not clear to me that the noise is adding anything specific. It seems like the model makes predictions based on how many specific components have been rotated in the different test conditions. I fear I'm just being dense, but it would be helpful to clarify whether the noise itself (and inverse variance estimation) are critical to why the model weights each reference frame how it does or whether this is just a method for scaling the weight by how much the joints or whatever have changed. It seems clear that this noise model is better than weighting by energy and smoothness.

Are there any force profiles for individual directions that are predicted to change shape substantially across some of these assorted changes in training and test locations (rather than merely being scaled)? If so, this might provide another test of the hypotheses.

I don't believe the decay factor that was used to scale the test functions was specified in the text, although I may have just missed this. It would be a good idea to state what this factor is where relevant in the text.

Reviewer #3 (Public Review):

The author proposed the minimum variance principle in the memory representation in addition to two alternative theories of the minimum energy and the maximum smoothness. The strength of this paper is the matching between the prediction data computed from the explicit equation and the behavioral data taken in different conditions. The idea of the weighting of multiple coordinate systems is novel and is also able to reconcile a debate in previous literature.

The weakness is that although each model is based on an optimization principle, but the derivation process is not written in the method section. The authors did not write about how they can derive these weighting factors from these computational principles. Thus, it is not clear whether these weighting factors are relevant to these theories or just hacking methods. Suppose the author argues that this is the result of the minimum variance principle. In that case, the authors should show a process of how to derive these weighting factors as a result of the optimization process to minimize these cost functions.

In addition, I am concerned that the proposed model can cancel the property of the coordinate system by the predicted variance, and it can work for any coordinate system, even one that is not used in the human brain. When the applied force is given in Cartesian coordinates, the directionality in the generalization ability of the memory of the force field is characterized by the kinematic relationship (Jacobian) between the Cartesian coordinate and the coordinate of interest (Cartesian, joint, and object) as shown in Equation 3. At the same time, when a displacement (epsilon) is considered in a space and a corresponding displacement is linked with kinematic equations (e.g., joint displacement and hand displacement in 2 joint arms in this paper), the generated variances in different coordinate systems are linked with the kinematic equation each other (Jacobian). Thus, how a small noise in a certain coordinate system generates the hand force noise (sigma_x, sigma_j, sigma_o) is also characterized by the kinematics (Jacobian). Thus, when the predicted forcefield (F_c, F_j, F_o) was divided by the variance (F_c/sigma_c^2, F_j/sigma_j^2, F_o/sigma_o^2, ), the directionality of the generalization force which is characterized by the Jacobian is canceled by the directionality of the sigmas which is characterized by the Jacobian. Thus, as it has been read out from Fig*D and E top, the weight in E-top of each coordinate system is always the inverse of the shift of force from the test force by which the directionality of the generalization is always canceled. Once this directionality is canceled, no matter how to compute the weighted sum, it can replicate the memorized force. Thus, this model always works to replicate the test force no matter which coordinate system is assumed. Thus, I am suspicious of the falsifiability of this computational model. This model is always true no matter which coordinate system is assumed. Even though they use, for instance, the robot coordinate system, which is directly linked to the participant's hand with the kinematic equation (Jacobian), they can replicate this result. But in this case, the model would be nonsense. The falsifiability of this model was not explicitly written.

eLife Assessment

This important work advances our understanding of factors influencing early childhood development. The large sample size and methodology applied make the findings of this study convincing; however, support for some of the claims made by the authors is incomplete. The work will be of interest to researchers in developmental science and early childhood pediatrics.

Reviewer #1 (Public Review):

Padilha et al. aimed to find prospective metabolite biomarkers in serum of children aged 6-59 months that were indicative of neurodevelopmental outcomes. The authors leveraged data and samples from the cross-sectional Brazilian National Survey on Child Nutrition (ENANI-2019), and an untargeted multisegment injection-capillary electrophoresis-mass spectrometry (MSI-CE-MS) approach was used to measure metabolites in serum samples (n=5004) which were identified via a large library of standards. After correlating the metabolite levels against the developmental quotient (DQ), or the degree of which age-appropriate developmental milestones were achieved as evaluated by the Survey of Well-being of Young Children, serum concentrations of phenylacetylglutamine (PAG), cresol sulfate (CS), hippuric acid (HA) and trimethylamine-N-oxide (TMAO) were significantly negatively associated with DQ. Examination of the covariates revealed that the negative associations of PAG, HA, TMAO and valine (Val) with DQ were specific to younger children (-1 SD or 19 months old), whereas creatinine (Crtn) and methylhistidine (MeHis) had significant associations with DQ that changed direction with age (negative at -1 SD or 19 months old, and positive at +1 SD or 49 months old). Further, mediation analysis demonstrated that PAG was a significant mediator for the relationship of delivery mode, child's diet quality and child fiber intake with DQ. HA and TMAO were additional significant mediators of the relationship of child fiber intake with DQ.

Strengths of this study include the large cohort size and study design allowing for sampling at multiple time points along with neurodevelopmental assessment and a relatively detailed collection of potential confounding factors including diet. The untargeted metabolomics approach was also robust and comprehensive allowing for level 1 identification of a wide breadth of potential biomarkers. Given their methodology, the authors should be able to achieve their aim of identifying candidate serum biomarkers of neurodevelopment for early childhood. The results of this work would be of broad interest to researchers who are interested in understanding the biological underpinnings of development and also for tracking development in pediatric populations, as it provides insight for putative mechanisms and targets from a relevant human cohort that can be probed in future studies. Such putative mechanisms and targets are currently lacking in the field due to challenges in conducting these kind of studies, so this work is important.

However, in the manuscript's current state, the presentation and analysis of data impede the reader from fully understanding and interpreting the study's findings. Particularly, the handling of confounding variables is incomplete. There is a different set of confounders listed in Table 1 versus Supplementary Table 1 versus Methods section Covariates versus Figure 4. For example, Region is listed in Supplementary Table 1 but not in Table 1, and Mode of Delivery is listed in Table 1 but not in Supplementary Table 1. Many factors are listed in Figure 4 that aren't mentioned anywhere else in the paper, such as gestational age at birth or maternal pre-pregnancy obesity.

The authors utilize the directed acrylic graph (DAG) in Figure 4 to justify the further investigation of certain covariates over others. However, the lack of inclusion of the microbiome in the DAG, especially considering that most of the study findings were microbial-derived metabolite biomarkers, appears to be a fundamental flaw. Sanitation and micronutrients are proposed by the authors to have no effect on the host metabolome, yet sanitation and micronutrients have both been demonstrated in the literature to affect microbiome composition which can in turn affect the host metabolome.

Additionally, the authors emphasized as part of the study selection criteria the following,<br /> "Due to the costs involved in the metabolome analysis, it was necessary to further reduce the sample size. Then, samples were stratified by age groups (6 to 11, 12 to 23, and 24 to 59 months) and health conditions related to iron metabolism, such as anemia and nutrient deficiencies. The selection process aimed to represent diverse health statuses, including those with no conditions, with specific deficiencies, or with combinations of conditions. Ultimately, through a randomized process that ensured a balanced representation across these groups, a total of 5,004 children were selected for the final sample (Figure 1)."

Therefore, anemia and nutrient deficiencies are assumed by the reader to be important covariates, yet, the data on the final distribution of these covariates in the study cohort is not presented, nor are these covariates examined further.

The inclusion of specific covariates in Table 1, Supplementary Table 1, the statistical models, and the mediation analysis is thus currently biased as it is not well justified.

Finally, it is unclear what the partial-least squares regression adds to the paper, other than to discard potentially interesting metabolites found by the initial correlation analysis.

Reviewer #2 (Public Review):

A strength of the work lies in the number of children Padilha et al. were able to assess (5,004 children aged 6-59 months) and in the extensive screening that the Authors performed for each participant. This type of large-scale study is uncommon in low-to-middle-income countries such as Brazil.<br /> The Authors employ several approaches to narrow down the number of potentially causally associated metabolites.<br /> Could the Authors justify on what basis the minimum dietary diversity score was dichotomized? Were sensitivity analyses undertaken to assess the effect of this dichotomization on associations reported by the article? Consumption of each food group may have a differential effect that is obscured by this dichotomization.<br /> Could the Authors specify the statistical power associated with each analysis?<br /> Could the Authors describe in detail which metric they used to measure how predictive PLSR models are, and how they determined what the "optimal" number of components were?<br /> The Authors use directed acyclic graphs (DAG) to identify confounding variables of the association between metabolites and DQ. Could the dataset generated by the Authors have been used instead? Not all confounding variables identified in the literature may be relevant to the dataset generated by the Authors.<br /> Were the systematic reviews or meta-analyses used in the DAG performed by the Authors, or were they based on previous studies? If so, more information about the methodology employed and the studies included should be provided by the Authors.<br /> Approximately 72% of children included in the analyses lived in households with a monthly income superior to the Brazilian minimum wage. The cohort is also biased towards households with a higher level of education. Both of these measures correlate with developmental quotient. Could the Authors discuss how this may have affected their results and how generalizable they are?<br /> Further to this, could the Authors describe how inequalities in access to care in the Brazilian population may have affected their results? Could they have included a measure of this possible discrepancy in their analyses?<br /> The Authors state that the results of their study may be used to track children at risk for developmental delays. Could they discuss the potential for influencing policies and guidelines to address delayed development due to malnutrition and/or limited access to certain essential foods?

Reviewer #3 (Public Review):

The ENANI-2019 study provides valuable insights into child nutrition, development, and metabolomics in Brazil, highlighting both challenges and opportunities for improving child health outcomes through targeted interventions and further research.

Strengths of the methods and results:<br /> (1) The study utilizes data from the ENANI-2019 cohort, which was already existing. This cohort choice allows for longitudinal assessments and exploration of associations between metabolites and developmental outcomes. In addition, there was conservation of resources which are scanty in all settings in the current scenario.<br /> (2) The study aims to investigate the relationship between circulating metabolites (exposure) and early childhood development (outcome), specifically developmental quotient (DQ). The objectives are clearly stated, which facilitates focused research questions and hypotheses. The population that is studied is clearly mentioned.<br /> (3) The study accessed a large number of children under five years, with blood collected from a final sample size of 5,004 children. The exclusion of infants under six months due to venipuncture challenges and lack of reference values highlights practical considerations in research design.<br /> The study sample reflects a diverse range of children in terms of age, sex distribution, weight status, maternal education, and monthly family income. This diversity enhances the generalizability of findings across different sociodemographic groups within Brazil.<br /> (4) The study uses standardized measures (e.g., DQ assessments) and chronological age. Confounding variables, such as child's age, diet quality, and nutritional status, are carefully considered and incorporated into analyses through a Directed Acyclic Graph (DAG). The mean DQ of 0.98 indicates overall developmental norms among the studied children, with variations noted across different demographic factors such as age, region, and maternal education. The prevalence of Minimum Dietary Diversity (MDD) being met by 59.3% of children underscores dietary patterns and their potential impact on health outcomes. The association between nutritional status (weight-for-height z-scores) and developmental outcomes (DQ) provides insights into the interplay between nutrition and child development.<br /> The study identified key metabolites associated with developmental quotient (DQ):<br /> Component 1: Branched-chain amino acids (Leucine, Isoleucine, Valine).<br /> Component 2: Uremic toxins (Cresol sulfate, Phenylacetylglutamine).<br /> Component 3: Betaine and amino acids (Glutamine, Asparagine).<br /> The study focused on several serum metabolites like PAG (phenylacetylglutamine), CS (p-cresyl sulfate), HA (hippuric acid), TMAO (trimethylamine-N-oxide), MeHis (methylhistidine), and Crtn (creatinine). These metabolites are implicated in various metabolic pathways linked to gut microbiota activity, amino acid metabolism, and dietary factors.<br /> These metabolites explained a significant portion of both metabolite variance (39.8%) and DQ variance (4.3%). The study suggests that these metabolites can be used as proxy measures of the gut microbiome in children.<br /> (5) The use of partial least square regression (PLSR) with cross-validation (80% training, 20% testing) which is a robust approach to identify metabolites predictive of DQ, which minimizes overfitting. This model allows for outliers to remain outliers for transparency.<br /> The Directed Acyclic Graph (DAG) identifies and adjusts for confounding variables (e.g., child's diet quality, nutritional status) and strengthens the validity of findings by controlling for potential biases. Developmental and gender differences were studied by testing interactions with the age of the child and the sex.<br /> Mediation analysis exploring metabolites as potential mediators provides insights into underlying pathways linking exposures (e.g., diet, microbiome) with DQ.<br /> The use of Benjamini-Hochberg correction for multiple comparisons and bootstrap tests (5,000 iterations) enhances the reliability of results by controlling false discovery rates and assessing significance robustly.

Significant correlations between serum metabolites and DQ, particularly negative associations with certain metabolites like PAG and CS, suggest potential biomarkers or pathways influencing developmental outcomes. Notably, these associations varied with age, suggesting different metabolic impacts during early childhood development.

Weaknesses:<br /> (1) The data collected was incomplete especially those related to breastfeeding history and birth weight. These have been mentioned in the limitations of the study but yet might have been potential confounders or even factors leading to the particular identified metabolite state of the population.<br /> (2) Other tests than mediation analysis might have been used to ensure reliability and robustness of the data. How data was processed, data cleaning methods, how outliers were handled and sensitivity analyses would ensure robustness of the findings.<br /> (3) The generalizability of the data is not sound especially considering the children mostly belonged to a higher socioeconomic group in Brazil with mother or caregiver education being above a certain level. Comparative studies with children from other socio-economic groups and other cohorts might have been useful. Consideration of sample size adequacy and power analysis might have helped in generalizing the findings.<br /> (4) Caution is needed in interpreting causality from this data because of the nature of the study design Discussing alternative explanations and potential confounding factors in more depth could strengthen the conclusions.

Appraisal<br /> (1) The aims of the study were to identify associations between children's serum metabolome and Early Childhood development. This aim was met. The results do confirm their conclusions.<br /> Impact of the work on the field

(1) Unless actual gut microbiome of children in this age group from gut bacteria examination or gastrointestinal examination of the gut of children, the causality of gut metabolome on early childhood development cannot be established with certainty. Because this may not be possible in every situation, proxy methods such as the one elucidated here might be useful, considering the risk-benefit ratio.<br /> (2) More research is needed on this theme through longitudinal studies to validate these findings and explore underlying pathways involving gut-brain interactions and metabolic dysregulation.<br /> Other readings: Readers are advised to read other research from other countries and other languages to understand the connection between gut microbiome, metabolite spectra, and child development. In addition to study the effect of these factors on child mental development too.

Readers might consider the following questions:<br /> (1) Should investigators study the families through direct observation of diet and other factors to look for a connection between food taken in and gut microbiome and child development?<br /> (2) Can an examination of the mother's gut microbiome influence the child's microbiome? Can the mother or caregiver's microbiome influence early childhood development?<br /> (3) Is developmental quotient enough to study early childhood development? Is it comprehensive enough?

eLife Assessment

This important work addresses the role of Marcks/Markcksl during spinal cord development and regeneration. The study is exceptional in combining molecular approaches to understand the mechanisms of tissue regeneration with behavioural assays, which is not commonly employed in the field. The data presented is convincing and comprehensive, using many complementary methodologies.

Reviewer #1 (Public Review):

In this manuscript, El Amri et al. are exploring the role of Marcks and Marcksl1 proteins during spinal cord development and regeneration in Xenopus. Using two different techniques to knockdown their expressions, they argue that these proteins are important for neural progenitors proliferation and neurites outgrowth in both contexts. Finally, using a pharmalogical approach, they suggest that Marcks and Marcksl1 work by modulating the activity of PLD and the levels of PIP2 whilst PKC could modulate Marcks activity.<br /> The strength of this manuscript resides in the ability of the authors to knockdown the expression of 4 different genes using 2 different methods to assess the role of this protein family during early development and regeneration at the late tadpole stage. This has always been a limiting factor in the field as the tools to perform conditional knockouts in Xenopus are very limited. However, this will not really be applicable to essential genes as it relies on the general knockdown of protein expression. The generation of antibodies able to detect endogenous Marcks/Marcksl1 is also a powerful tool to assess the extent to which the expression of these proteins is down-regulated.<br /> Whilst there is a great amount of data provided in this manuscript and there is strong evidence to show that Marcks are important for spinal cord development and regeneration, their roles in both contexts is not explored fully. The description of the effect of knocking down Marcks/Marcksl1 on neurons and progenitors is rather superficial and the evidence for the underlying mechanism underpinning their roles is not very convincing.

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.<br /> 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.<br /> 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,<br /> 1) Why did the authors choose Marcks and Marcksl1?<br /> 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.

2) Gene knock-out experiments with F0 crispants,<br /> 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.

3) Limitations of pharmacological compound rescue<br /> 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.

Reviewer #3 (Public Review):

El Amri et al conducted an analysis on the function of marcks and marcksl in Xenopus spinal cord development and regeneration. Their study revealed these proteins are crucial for neurite outgrowth and cell proliferation, including Sox2+ progenitors. Furthermore, they suggested these genes may act through the PLD pathway. The study is well-executed with appropriate controls and validation experiments, distinguishing it from typical regeneration research by including behavioral assays. The manuscript is commendable for its quantifications, literature referencing, careful conclusions, and detailed methods. Conclusions are well-supported by the experiments performed in this study. Overall, this manuscript contributes to the field of spinal cord regeneration and sets a good example for future research in this area.

eLife Assessment

This important paper demonstrates that different PKA subtypes exhibit distinct subcellular localization at rest in CA1 neurons. The authors provide compelling evidence that when all tested PKA subtypes are activated by norepinephrine, catalytic subunits translocate to dendritic spines but regulatory subunits remain unmoved. Furthermore, PKA-dependent regulation of synaptic plasticity and transmission can be supported only by wildtype, dissociable PKA, but not by inseparable PKA.

Reviewer #1 (Public review):

Summary:

This is a short self-contained study with a straightforward and interesting message. The paper focuses on settling whether PKA activation requires dissociation of the catalytic and regulatory subunits. This debate has been ongoing for ~ 30 years, with renewed interest in the question following a publication in Science, 2017 (Smith et al.). Here, Xiong et al demonstrate that fusing the R and C subunits together (in the same way as Smith et al) prevents the proper function of PKA in neurons. This provides further support for the dissociative activation model - it is imperative that researchers have clarity on this topic since it is so fundamental to building accurate models of localised cAMP signalling in all cell types. Furthermore, their experiments highlight that C subunit dissociation into spines is essential for structural LTP, which is an interesting finding in itself. They also show that preventing C subunit dissociation reduces basal AMPA receptor currents to the same extent as knocking down the C subunit. Overall, the paper will interest both cAMP researchers and scientists interested in fundamental mechanisms of synaptic regulation.

Strengths:

The experiments are technically challenging and well executed. Good use of control conditions e.g untransfected controls in Figure 4.

Weaknesses:

The novelty is lessened given the same team has shown dissociation of the C subunit into dendritic spines from RIIbeta subunits localised to dendritic shafts before (Tillo et al., 2017). Nevertheless, the experiments with RII-C fusion proteins are novel and an important addition.

Reviewer #2 (Public review):

Summary:

PKA is a major signaling protein which has been long studied and is vital for synaptic plasticity. Here, the authors examine the mechanism of PKA activity and specifically focus on addressing the question of PKA dissociation as a major mode of its activation in dendritic spines. This would potentially allow to determine the precise mechanisms of PKA activation and address how it maintains spatial and temporal signaling specificity.

Strengths:

The results convincingly show that PKA activity is governed by the subcellular localization in dendrites and spines and is mediated via subunit dissociation. The authors make use of organotypic hippocampal slice cultures, where they use pharmacology, glutamate uncaging, and electrophysiological recordings.

Overall, the experiments and data presented are well executed. The experiments all show that at least in the case of synaptic activity, distribution of PKA-C to dendritic spines is necessary and sufficient for PKA mediated functional and structural plasticity.<br /> The authors were able to persuasively support their claim that PKA subunit dissociation is necessary for its function and localization in dendritic spines. This conclusion is important to better understand the mechanisms of PKA activity and its role in synaptic plasticity.

Weaknesses:

While the experiments are indeed convincing and well executed, the data presented is similar to previously published work from the Zhong lab (Tillo et al., 2017, Zhong et al 2009). This reduces the novelty of the findings in terms of re-distribution of PKA subunits, which was already established, at least to some degree.

Reviewer #3 (Public review):

Summary:

Xiong et al. investigated the debated mechanism of PKA activation using hippocampal CA1 neurons under pharmacological and synaptic stimulations. Examining all major PKA-R isoforms in these neurons, they found that a portion of PKA-C dissociates from PKA-R and translocate into dendritic spines following norepinephrine bath application. Additionally, their use of a non-dissociable form of PKA demonstrates its essential role in structural long-term potentiation (LTP) induced by two-photon glutamate uncaging, as well as in maintaining normal synaptic transmission, as verified by electrophysiology. This study presents a valuable finding on the activation-dependent re-distribution of PKA catalytic subunits in CA1 neurons, a process vital for synaptic functionality. The robust evidence provided by the authors makes this work particularly relevant for biologists seeking to understand PKA activation mechanisms, its downstream effects, and synaptic plasticity.

Strengths:

The study is methodologically robust, particularly in the application of two-photon imaging and electrophysiology. The experiments are well-designed with effective controls and a comprehensive analysis. The credibility of the data is further enhanced by the research team's previous works in related experiments. The study provides sufficient evidence to support the classical model of PKA activation via dissociation in neurons.

Weaknesses:

No specific weaknesses are noted in the current study; future research could provide additional insights by exploring PKA dissociation under varied physiological conditions, particularly in vivo, to further validate and expand upon these findings.

Author response:

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

New Experiments

(1) Activation-dependent dynamics of PKA with the RIα regulatory subunit, adding to the answer to Reviewers 1 and 2. To determine the dynamics of all PKA isoforms, we have added experiments that used PKA-RIα as the regulatory subunit. We found differential translocation between PKA-C (co-expressed with PKA-RIα) and PKA-RIα (Figure 1–figure supplement 3), similar to the results when PKA-RIIα or PKA-RIβ was used.

(2) PKA-C dynamics elicited by a low concentration of norepinephrine, addressing Reviewer 3’s comment. We have found that PKA-C (co-expressed with RIIα) exhibited similar translocation into dendritic spines in the presence of a 5x lowered concentration (2 μM) of norepinephrine, suggesting that the translocation occurs over a wide range of stimulus strengths (Figure 1-figure supplement 2).

Reviewer #1 (Public Review):

Summary:

This is a short self-contained study with a straightforward and interesting message. The paper focuses on settling whether PKA activation requires dissociation of the catalytic and regulatory subunits. This debate has been ongoing for ~ 30 years, with renewed interest in the question following a publication in Science, 2017 (Smith et al.). Here, Xiong et al demonstrate that fusing the R and C subunits together (in the same way as Smith et al) prevents the proper function of PKA in neurons. This provides further support for the dissociative activation model - it is imperative that researchers have clarity on this topic since it is so fundamental to building accurate models of localised cAMP signalling in all cell types. Furthermore, their experiments highlight that C subunit dissociation into spines is essential for structural LTP, which is an interesting finding in itself. They also show that preventing C subunit dissociation reduces basal AMPA receptor currents to the same extent as knocking down the C subunit. Overall, the paper will interest both cAMP researchers and scientists interested in fundamental mechanisms of synaptic regulation.

Strengths:

The experiments are technically challenging and well executed. Good use of control conditions e.g untransfected controls in Figure 4.

We thank the reviewer for their accurate summarization of the position of the study in the field and for the positive evaluation of our study.

Weaknesses:

The novelty is lessened given the same team has shown dissociation of the C subunit into dendritic spines from RIIbeta subunits localised to dendritic shafts before (Tillo et al., 2017). Nevertheless, the experiments with RII-C fusion proteins are novel and an important addition.

We thank the reviewer for noticing our earlier work. The first part of the current work is indeed an extension of previous work, as we have articulated in the manuscript. However, this extension is important because recent studies suggested that the majority of PKA-RIIβ are axonal localized. The primary PKA subtypes in the soma and dendrite are likely PKA-RIβ or PKA-RIIα. Although it is conceivable that the results from PKA-RIIβ can be extended to the other subunits, given the current debate in the field regarding PKA dissociation (or not), it remains important to conclusively demonstrate that these other regulatory subunit types also support PKA dissociation within intact cells in response to a physiological stimulant. To complete the survey for all PKA-R isoforms, we have now added data for PKA-RIα (New Experiment #1), as they are also expressed in the brain (e.g., https://www.ncbi.nlm.nih.gov/gene/5573). Additionally, as the reviewer points out, our second part is a novel addition to the literature.

Reviewer #2 (Public Review):

Summary:

PKA is a major signaling protein that has been long studied and is vital for synaptic plasticity. Here, the authors examine the mechanism of PKA activity and specifically focus on addressing the question of PKA dissociation as a major mode of its activation in dendritic spines. This would potentially allow us to determine the precise mechanisms of PKA activation and address how it maintains spatial and temporal signaling specificity.

Strengths:

The results convincingly show that PKA activity is governed by the subcellular localization in dendrites and spines and is mediated via subunit dissociation. The authors make use of organotypic hippocampal slice cultures, where they use pharmacology, glutamate uncaging, and electrophysiological recordings.

Overall, the experiments and data presented are well executed. The experiments all show that at least in the case of synaptic activity, the distribution of PKA-C to dendritic spines is necessary and sufficient for PKA-mediated functional and structural plasticity.

The authors were able to persuasively support their claim that PKA subunit dissociation is necessary for its function and localization in dendritic spines. This conclusion is important to better understand the mechanisms of PKA activity and its role in synaptic plasticity.

We thank the reviewer for their positive evaluation of our study.

Weaknesses:

While the experiments are indeed convincing and well executed, the data presented is similar to previously published work from the Zhong lab (Tillo et al., 2017, Zhong et al 2009). This reduces the novelty of the findings in terms of re-distribution of PKA subunits, which was already established. A few alternative approaches for addressing this question: targeting localization of endogenous PKA, addressing its synaptic distribution, or even impairing within intact neuronal circuits, would highly strengthen their findings. This would allow us to further substantiate the synaptic localization and re-distribution mechanism of PKA as a critical regulator of synaptic structure, function, and plasticity.

We thank the reviewer for noticing our earlier work. The first part of the current work is indeed an extension of previous work, as we have articulated in the manuscript. However, this extension is important because recent studies suggested that the majority of PKA-RIIβ are axonal localized. The primary PKA subtypes in the soma and dendrite are likely PKA-RIβ or PKA-RIIα. Although it is conceivable that the results from PKA-RIIβ can be extended to the other subunits, given the current debate in the field regarding PKA dissociation (or not), it remains important to conclusively demonstrate that these other regulatory subunit types also support PKA dissociation within intact cells in response to a physiological stimulant. To complete the survey for all PKA-R isoforms, we have now added data for PKA-RIα (New Experiment #1), as they are also expressed in the brain (e.g., https://www.ncbi.nlm.nih.gov/gene/5573). Additionally, as Reviewer 1 points out, our second part is a novel addition to the literature.

We also thank the reviewer for suggesting the experiments to examine PKA’s synaptic localization and dynamics as a key mechanism underlying synaptic structure and function. We agree that this is a very interesting topic. At the same time, we feel that this mechanistic direction is open ended at this time and beyond what we try to conclude within this manuscript: prevention of PKA dissociation in neurons affects synaptic function. Therefore, we will save the suggested direction for future studies. We hope the reviewer understand.

Reviewer #3 (Public Review):

Summary:

Xiong et al. investigated the debated mechanism of PKA activation using hippocampal CA1 neurons under pharmacological and synaptic stimulations. Examining the two PKA major isoforms in these neurons, they found that a portion of PKA-C dissociates from PKA-R and translocates into dendritic spines following norepinephrine bath application. Additionally, their use of a non-dissociable form of PKC demonstrates its essential role in structural long-term potentiation (LTP) induced by two-photon glutamate uncaging, as well as in maintaining normal synaptic transmission, as verified by electrophysiology. This study presents a valuable finding on the activation-dependent re-distribution of PKA catalytic subunits in CA1 neurons, a process vital for synaptic functionality. The robust evidence provided by the authors makes this work particularly relevant for biologists seeking to understand PKA activation and its downstream effects essential for synaptic plasticity.

Strengths:

The study is methodologically robust, particularly in the application of two-photon imaging and electrophysiology. The experiments are well-designed with effective controls and a comprehensive analysis. The credibility of the data is further enhanced by the research team's previous works in related experiments. The conclusions of this paper are mostly well supported by data. The research fills a significant gap in our understanding of PKA activation mechanisms in synaptic functioning, presenting valuable insights backed by empirical evidence.

We thank the reviewer for their positive evaluation of our study.

Weaknesses:

The physiological relevance of the findings regarding PKA dissociation is somewhat weakened by the use of norepinephrine (10 µM) in bath applications, which might not accurately reflect physiological conditions. Furthermore, the study does not address the impact of glutamate uncaging, a well-characterized physiologically relevant stimulation, on the redistribution of PKA catalytic subunits, leaving some questions unanswered.

We agreed with the Reviewer that testing under physiological conditions is critical especially given the current debate in the literature. That is why we tested PKA dynamics induced by the physiological stimulant, norepinephrine. It has been suggested that, near the release site, local norepinephrine concentrations can be as high as tens of micromolar (Courtney and Ford, 2014). Based on this study, we have chosen a mid-range concentration (10 μM). At the same time, in light of the Reviewer’s suggestion, we have now also tested PKA-RIIα dissociation at a 5x lower concentration of norepinephrine (2 μM; New Experiment #2). The activation and translocation of PKA-C is also readily detectible under this condition to a degree comparable to when 10 μM norepinephrine was used.

Regarding the suggested glutamate uncaging experiment, it is extremely challenging because of finite signal-to-noise ratios in our experiments. From our past studies, we know that activated PKA-C can diffuse three dimensionally, with a fraction as membrane-associated proteins and the other as cytosolic proteins. Although we have evidence that its membrane affinity allows it to become enriched in dendritic spines, it is not known (and is unlikely) that activated PKA-C is selectively targeted to a particular spine. Glutamate uncaging of a single spine presumably would locally activate a small number of PKA-C. It will be very difficult to trace the 3D diffusion of these small number of molecules in the presence of surrounding resting-state PKA-C molecules. Finally, we hope the reviewer agrees that, regardless of the result of the glutamate uncaging experiment, the above new experiment (New Experiment #2) already indicate that certain physiologically relevant stimuli can drive PKA-C dissociation from PKA-R and translocation to spines, supporting our conclusion.

Reviewer #2 (Recommendations For The Authors):

It was a pleasure reading your paper, and the results are well-executed and well-presented.

My main and only recommendations are two ways to further expand the scope of the findings.

First, I believe addressing the endogenous localization of PKA-C subunit before and after PKA activation would be highly important to validate these claims. Overexpression of tagged proteins often shows vastly different subcellular distribution than their endogenous counterparts. Recent technological advances with CRISPR/Cas9 gene editing (Suzuki et al Nature 2016 and Gao et al Neuron 2019 for example) which the Zhong lab recently contributed to (Zhong et al 2021 eLife) allow us to tag endogenous proteins and image them in fixed or live neurons. Any experiments targeting endogenous PKA subunits that support dissociation and synaptic localization following activation would be very informative and greatly increase the novelty and impact of their findings.

We agreed that addressing the endogenous PKA dynamics is important. However, despite recent progress, endogenous labeling using CRISPR-based methods remains challenging and requires extensive optimization. This is especially true for signaling proteins whose endogenous abundance is often low. We have tried to label PKA catalytic subunits and regulatory subunits using both the homologous recombination-based method SLENDR and our own non-homologous end joining-based method CRISPIE. We did not succeed, in part because it is very difficult to see any signal under wide-field fluorescence conditions, which makes it difficult to screen different constructs for optimizing parameters. It is also possible that, at the endogenous abundance, the label is just not bright enough to be seen. Nevertheless, for both PKA type Iβ and type IIα that we studied in this manuscript, we have correlated the measured parameters (specifically, Spine Enrichment Index or SEI) with the overexpression level (Figure 1-figure supplement 1). We found that they are not strongly correlated with the expression level under our conditions. By extrapolating to non-overexpression conditions, our conclusion remains valid.

To overcome the inability to label endogenous PKA subunits using CRISPR-based methods, we have also attempted a conditional knock-in method call ENABLED that we previously developed to label PKA-Cα. In preliminary results, we found that endogenously label PKA were very dim. However, in a subset of cells that are bright enough to be quantified, the PKA catalytic subunit indeed translocated to dendritic spines upon stimulation (see Additional Fig. 1 in the next page), corroborating our results using overexpression. These results, however, are not ready to be published because characterization of the mouse line takes time and, at this moment, the signal-to-noise ratio remains low. We hope that the reviewer can understand.

Author response image 1.

Endogeneous PKA-Cα translocate to dendritic spines upon activation.

Second, experiments which would advance and validate these findings in vivo would be highly valuable. This could be achieved in a number of ways - one would be overexpression of tagged PKA versions and examining sub-cellular distribution before and after physiological activation in vivo. Another possibility is in vivo perturbation - one would speculate that disruption or tethering of PKA subunits to the dendrite would lead to cell-specific functional and structural impairments. This could be achieved in a similar manner to the in vitro experiments, with a PKA KO and replacement strategy of the tethered C-R plasmid, followed by structural or functional examination of neurons.

I would like to state that these experiments are not essential in my opinion, but any improvements in one of these directions would greatly improve and extend the impact and findings of this paper.

We thank the reviewer for the suggestion and the understanding. The suggested in vivo experiments are fascinating. However, in vivo imaging of dendritic spine morphology is already in itself challenging. The difficulty greatly increases when trying to detect partial, likely transient translocation of a signaling protein. It is also very difficult to knock down endogenous PKA while simultaneously expressing the R-C construct in a large number of cells to achieve detectable circuit or behavioral effect (and hope that compensation does not happen over weeks). We hope the reviewer agrees that these experiments would be their own project and go beyond the time and scope of the current study.

Reviewer #3 (Recommendations For The Authors):

Please elaborate on the methods used to visualize PKA-RIIα and PKA-RIβ subunits.

As suggested, we have now included additional details for visualizing PKA-Rs in the text. Specifically, we write (pg. 5): “…, as visualized using expressed PKA-R-mEGFP in separate experiments (Figs. 1A-1C).”.

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.

eLife Assessment

In this potentially important study, the authors conducted atomistic simulations to probe the salt-dependent phase separation of the low-complexity domain of hnRN-PA1 (A1-LCD). The authors have identified both direct and indirect mechanisms of salt modulation, provided explanations for four distinct classes of salt dependence, and proposed a model for predicting protein properties from amino acid composition. There is a range of opinions regarding the strength of evidence, with some considering the evidence as incomplete due to the limitations in the length and statistical errors of the computationally intense atomistic MD simulations.

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:

Based on the reviewer's assessment of the manuscript, the following points were raised:

(1) The simulation duration is too short to draw comprehensive conclusions about phase separation.<br /> (2) There are concerns regarding the convergence of the simulations, particularly as highlighted in Figure 2A.<br /> (3) The simulation begins with a protein concentration of 3.5 mM ("we built an 8-copy model for the dense phase (with an initial concentration of 3.5 mM)"), which is high for phase separation studies. The reviewer questions the use of the term "dense phase" and suggests that the authors conduct a clearer analysis depicting the coexistence of both the dilute and dense phases to represent a steady state. Without this, the realism of the described phenomena is doubtful. Commenting on phase separation under conditions that don't align with typical phase separation parameters is not acceptable.<br /> (4) The inference that "Each Arg sidechain often coordinates two Cl- ions simultaneously, but each Lys sidechain coordinates only one Cl- ion" is questioned. According to Supplementary Figure 2A, Lys seems to coordinate with Cl- ions more frequently than Arg.<br /> (5) The authors are requested to update the figure captions for Supplementary Figures 2 and 3, specifying which system the analyses were performed on.<br /> (6) It is difficult to observe a clear trend due to irregularities in the data. Although the authors have included a red dotted line in the figures, the trend is not monotonic. The reviewer expresses concerns about significant conclusions drawn from these figures (e.g., Figure 2C, Figure 5A, Supplementary Figure 1).<br /> (7) Given the error in the radius of gyration (Rg) calculations, the reviewer questions the validity of drawing conclusions from this data.<br /> (8) The pair correlation function values in Figure 5E and supplementary figure 4 show only minor differences, and the reviewer questions whether these differences are significant.<br /> (9) Previous reports suggest that, upon self-assembly, protein chains extend within the condensate, leading to a decrease in intramolecular contacts. However, the authors show an increase in intramolecular contacts with increasing salt concentration (Figure 2C), which contradicts prior studies. The reviewer advises the authors to carefully review this and provide justification.<br /> (10) A systematic comparison of estimated parameters with varying salt concentrations is required. Additionally, the authors should provide potential differences in salt concentrations between the dilute and condensed phases.<br /> (11) The reviewer finds that the majority of the data presented shows no significant alteration with changes in salt concentration, yet the authors have made strong conclusions regarding salt activity.

The manuscript lacks sufficient scientific details of the calculations.

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.).

Comments on revised version:

The authors have adequately addressed my previous concerns and suggestions. The manuscript is now significantly improved. The new results and analyses provided by the authors represent a substantial advance in our understanding of the role of electrostatics in the assembly of biomolecular condensates.

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.<br /> • Rigorous Methodology: The authors utilize all-atom MD simulations, a powerful computational tool, to investigate the molecular details of salt-protein interactions.<br /> • Comprehensive Analysis: The study systematically explores a wide range of salt concentrations, revealing a nuanced picture of salt effects on phase separation.<br /> • 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. However, given the much larger size of the full-length protein, it is acceptable to omit it given the current computing resources available.

Overall, this manuscript represents a significant contribution to the field of IDP phase separation. The authors' findings provide valuable insights into the molecular mechanisms by which salt modulates this process, with potential implications for understanding and treating neurodegenerative diseases.

eLife Assessment

This manuscript presents a valuable new quantitative crosslinking mass spectrometry approach using novel isobaric crosslinkers. The data are solid and the method has potential for a broad application in structural biology if more isobaric crosslinking channels are available and the quantitative information of the approach is exploited in more depth.

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.

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.

Comments on latest version:

I raised only two points which they have not addressed: Higher multiplexing of Qlinkers (1) and experiments to assess the statistical power of their quantitation strategy (2).

I can see that point (1) requires substantial experimental efforts and synthesis of novel Qlinkers would be months of work. This is an editorial decision if the limited quantitative power of the "2-plex" approach they have right now is sufficient to support publication in eLife. While I like the approach, I feel it falls short of its potential in its current form.

For point (2), the authors did not do any supporting experiments. They claim "higher plex Qlinkers" would need to be available, but I suggested experiments that can be done even with Q2linkers: Using one of the two channels as a reference channel (similar the Super-SILAC strategy published in 2010 by Geiger et al; using an isotope-labeled channel as a stable reference channel between different experiments and LC-MS runs), they could do time-courses or ligand-concentration-series with the other channel and then show that Qlinkers allow quantitative monitoring of the different populations (e.g. conformations or ligand-bound proteins).

As an additional point, I was a bit surprised to read that the quantitation evaluation in Figure 1 is based on a single experiment (reviewer response document page 6, line 2 in the authors' reply). I strongly suggest this to be repeated a few times so a proper statistical test on experimental reproducibiltiy of Qlinkers can be conducted.

In summary, the authors declined to do any experimental work to address my concerns.

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.

Comments on latest version:

The authors have not really addressed most of the concerns. They have added minimal discussion points to the text. This is okay from my perspective as eLife's policy is to leave it up to the authors of how strongly to consider the reviewers' comments. I should add that I do fully agree with the other reviewer that the quantitative assessment from Figure 1 should have been done in triplicates at least and that this would actually be essential.

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.

eLife Assessment

This useful manuscript addresses some key molecular mechanisms on the neuroprotective roles of soluble TREM2 in neurodegenerative diseases. Thw study will advance our understanding of TREM2 mutations, particularly on the damaging effect of known TREM2 mutations, and also explain why soluble TREM2 can antagonize Aβ aggregation. However, the primary experimental method, MD simulations, suffers from limited sampling, rendering the results incomplete for definite conclusions.

Reviewer #1 (Public review):

In this manuscript, Saeb et al reported the mechanistic roles of the flexible stalk domain in sTREM2 function using molecular dynamics simulations. They have reported some interesting molecular bases explaining why sTREM2 shows protective effects during AD, such as partial extracellular stalk domain promoting binding preference and stabilities of sTREM2 with its ligand even in the presence of known AD-risk mutation, R47H. Furthermore, they found that the stalk domain itself acts as the site for ligand binding by providing an "expanded surface", known as 'Expanded Surface 2' together with the Ig-like domain. Also, they observed no difference in the binding free energy of phosphatidyl-serine 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).

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.

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.

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".

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.

Reviewer #2 (Public review):

Significance:

TREM2 is an immunomodulatory receptor expressed on myeloid cells and microglia in the brain. TREM2 consists of a single immunoglobular (Ig) domain that leads into a flexible stalk, transmembrane helix, and short cytoplasmic tail. Extracellular proteases can cleave TREM2 in its stalk and produce a soluble TREM2 (sTREM2). TREM2 is genetically linked to Alzheimer's disease (AD), with the strongest association coming from an R47H variant in the Ig domain. Despite intense interest, the full TREM2 ligand repertoire remains elusive, and it is unclear what function sTREM2 may play in the brain. The central goal of this paper is to assess the ligand-binding role of the flexible stalk that is generated during the shedding of TREM2. To do this, the authors simulate the behavior of constructs with and without stalk. However, it is not clear why the authors chose to use the isolated Ig domain as a surrogate for full-length TREM2. Additionally, experimental binding evidence that is misrepresented by the authors contradicts the proposed role of the stalk.

Summary and strengths:

The authors carry out MD simulations of WT and R47H TREM2 with and without the flexible stalk. Simulations are carried out for apo TREM2 and for TREM2 in complex with various lipids. They compare results using just the Ig domain to results including the flexible stalk that is retained following cleavage to generate sTREM2. The computational methods are well-described and should be reproducible. The long simulations are a strength, as exemplified in Figure 2A where a CDR2 transition happens at ~400-600 ns. The stalk has not been resolved in structural studies, but the simulations suggest the intriguing and readily testable hypothesis that the stalk interacts with the Ig domain and thereby contributes to the stability of the Ig domain and to ligand binding. I suspect biochemists interested in TREM2 will make testing this hypothesis a high priority.

Weaknesses:

Unfortunately, the work suffers from two fundamental flaws.

(1) 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.

(2) 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.

eLife Assessment

The study provides a valuable showcase of a workflow to perform large-scale characterization of drug mechanisms of action using proteomics in which on-target and off-targets of 166 compounds using proteome solubility analysis in living cells and cell lysates were determined. The evidence supporting the claims of the authors is solid, however, the inclusion of more replicate experiments and more statistical rigor would have strengthened the study. This will be of broad interest to medicinal chemists, toxicologists, computational biologists and biochemists.

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.”

eLife Assessment

This landmark study elucidates the intricate structural mechanisms by which both covalent and non-covalent synthetic ligands can co-occupy the binding pocket of the nuclear receptor transcription factor PPARγ. Through a compelling integration of structural, biochemical, and biophysical evidence, the authors challenge the reliability of two commonly used covalent inhibitors. These findings have far-reaching implications for the broader field of nuclear receptor research. This work will be of high interest to structural biologists and biochemists exploring ligand interactions within the nuclear receptor superfamily.

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 has 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 wide-spread 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.<br /> - Multiple orthogonal approaches are used to provide high resolution information on ligand binding poses and protein dynamics.<br /> - The large number of x-ray crystal structures solved for this manuscript should be applauded along with their rigorous validation and interpretation.

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 cobind to the receptor. However, whether a covalent inhibitor functions by blocking the binding of a non-covalent ligand, or cobind 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 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 as presented is strong and convincing.

Additional comments:

The co-crystal studies were performed by soaking a non-covalent ligand 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, one might ask if the sequence was reversed (i.e., soaking a covalent inhibitor to LBD pre-crystalized with a non-covalent ligand), would similar conclusion be drawn? The authors have reasonably speculated that it might be difficult to soak a covalent inhibitor into preformed crystals where the PPARγ LBD is already bound to a non-covalent ligand, because the larger non-covalent ligand could block the covalent inhibitor to gain access to the region of the orthosteric pocket required for covalent modification.

eLife Assessment

The manuscript presents an important model for the field of endosome maturation, providing perspective on the role of the deubiquitinating enzyme UPS-50/USP8 in the process. The evidence presented in the paper is clear, incorporating well-designed experiments that suggest the dual actions of UPS-50 and USP8 in the conversion of early endosomes into late endosomes. Overall, the work is convincing and centers on an intriguing subject.

Reviewer #1 (Public review):

Summary:

The manuscript focuses on the role of the deubiquitinating enzyme UPS-50/USP8 in endosome maturation. The authors aimed to clarify how this enzyme drives the conversion of early endosomes into late endosomes. Overall, they did achieve their aims in shedding light on the precise mechanisms by which UPS-50/USP8 regulates endosome maturation. The results support their conclusions that UPS-50 acts by disassociating RABX-5 from early endosomes to deactivate RAB-5 and by recruiting SAND-1/Mon1 to activate RAB-7. This work is commendable and will have a significant impact on the field. The methods and data presented here will be useful to the community in advancing our understanding of endosome maturation and identifying potential therapeutic targets for diseases related to endosomal dysfunction. It is worth noting that further investigation is required to fully understand the complexities of endosome maturation. However, the findings presented in this manuscript provide a solid foundation for future studies.

Strengths:

The major strengths of this work lie in the well-designed experiments used to examine the effects of UPS-50 loss. The authors employed confocal imaging to obtain a picture of the aftermath of USP-50 loss. Their findings indicated enlarged early endosomes and MVB-like structures in cells deficient in USP-50/USP8.

Weaknesses:

Specifically, there is a need for further investigation to accurately characterize the anomalous structures detected in the ups-50 mutant. Also, the correlation between the presence of these abnormal structures and ESCRT-0 is yet to be addressed, and the current working model needs to be revised to prevent any confusion between enlarged early endosomes and MVBs.

Reviewer #2 (Public review):

Summary:

In this study, the authors study how the deubiquitinase USP8 regulates endosome maturation in C. elegans and mammalian cells. The authors have isolated USP8 mutant alleles in C. elegans and used multiple in vivo reporter lines to demonstrate the impact of USP8 loss-of-function on endosome morphology and maturation. They show that in USP8 mutant cells, the early endosomes and MVB-like structures are enlarged while the late endosomes and lysosomal compartments are reduced. They elucidate that USP8 interacts with Rabx5, a guanine nucleotide exchange factor (GEF) for Rab5, and show that USP8 likely targets specific lysine residue of Rabx5 to dissociate it from early endosomes. They also find that localization of USP8 to early endosomes are disrupted in Rabx5 mutant cells. They observe that in both Rabx5 and USP8 mutant cells, the Rab7 GEF SAND-1 puncta which likely represents late endosomes are diminished, although that Rabex5 are accumulated in USP8 mutant cells. The authors provide evidence that USP8 regulates endosomal maturation in a similar fashion in mammalian cells. Based on their observations they propose that USP8 dissociates Rabex5 from early endosomes and enhances the recruitment of SAND-1 to promote endosome maturation.

Strengths:

The major highlights of this study include the direct visualization of endosome dynamics in a living multi-cellular organism, C. elegans. The high-quality images provide clear in vivo evidences to support the main conclusions. The authors have generated valuable resources to study mechanisms involved in endosome dynamics regulation in both the worm and mammalian cells, which would benefit many members in the cell biology community. The work identifies a fascinating link between USP8 and the Rab5 guanine nucleotide exchange factor Rabx5, which expands the targets and modes of action of USP8. The findings make a solid contribution toward the understanding of how endosomal trafficking is controlled.

Weaknesses:

- The authors utilized multiple fluorescent protein reporters, including those generated by themselves, to label endosomal vesicles. Although these are routine and powerful tools for studying endosomal trafficking, these results cannot tell that whether the endogenous proteins (Rab5, Rabex5, Rab7, etc.) are affected in the same fashion. Note that the authors have provided convincing evidence about the effects on Rab proteins in the revised manuscript.<br /> - The authors clearly demonstrated a link between USP8 and Rabx5, and they showed that cells deficient of both factors displayed similar defects in late endosomes/lysosomes. But the authors didn't confirm whether and/or to which extent that USP8 regulates endosome maturation through Rabx5. Additional genetic and molecular evidence might be required to better support their working model. Note that the authors have provided convincing evidence about the role of USP8-Rabx5 axis in the revised manuscript.

Reviewer #3 (Public review):

Summary:

The authors elucidated the role of USP8 in the endocytic pathway. Using C. elegans epithelial cells as a model, they observed that when USP8 function is lost, the cells have a decreased number and size in lysosomes. Since USP8 was already known to be a protein linked to ESCRT components, they looked into what role USP8 might play in connecting lysosomes and multivesicular bodies (MVB). They observed fewer ESCRT-associated vesicles but an increased number of abnormal enlarged vesicles (aberrant early endosomes) when USP8 function was lost. They showed that USP8 interacts with Rabx5 to dissociate it from early endosomes promoting the recruitment of the Rab7 GEF SAND-1/Mon1 and the maturation of the endosomes. The authors provided evidence that USP8 regulates endosomal maturation in a similar fashion in mammalian cells.

Strengths:

The use of two models, C. elegans and a mammalian cell line to describe a similar mechanism.

Author response:

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

Public Reviews:

Reviewer #1 (Public Review): 

Summary: 

The manuscript focuses on the role of the deubiquitinating enzyme UPS-50/USP8 in endosome maturation. The authors aimed to clarify how this enzyme drives the conversion of early endosomes into late endosomes. Overall, they did achieve their aims in shedding light on the precise mechanisms by which UPS-50/USP8 regulates endosome maturation. The results support their conclusions that UPS-50 acts by disassociating RABX-5 from early endosomes to deactivate RAB-5 and by recruiting SAND-1/Mon1 to activate RAB-7. This work is commendable and will have a significant impact on the field. The methods and data presented here will be useful to the community in advancing our understanding of endosome maturation and identifying potential therapeutic targets for diseases related to endosomal dysfunction. It is worth noting that further investigation is required to fully understand the complexities of endosome maturation. However, the findings presented in this manuscript provide a solid foundation for future studies. 

We thank this reviewer for the instructive suggestions and encouragement.

Strengths: 

The major strengths of this work lie in the well-designed experiments used to examine the effects of UPS-50 loss. The authors employed confocal imaging to obtain a picture of the aftermath of the USP-50 loss. Their findings indicated enlarged early endosomes and MVB-like structures in cells deficient in USP-50/USP8. 

We thank this reviewer for the instructive suggestions and encouragement.

Weaknesses: 

Specifically, there is a need for further investigation to accurately characterize the anomalous structures detected in the usp-50 mutant. Also, the correlation between the presence of these abnormal structures and ESCRT-0 is yet to be addressed, and the current working model needs to be revised to prevent any confusion between enlarged early endosomes and MVBs. 

Excellent suggestions. USP8 has been identified as a protein associated with ESCRT components, which are crucial for endosomal membrane deformation and scission, leading to the formation of intraluminal vesicles (ILVs) within multivesicular bodies (MVBs). In usp-50 mutants, we observed a significant reduction in the punctate signals of HGRS-1::GFP and STAM-1 (Figure 1G and H; and Figure1-figure supplement 1B), indicating a disruption in ESCRT-0 complex localization (Author response image 1). Additionally, lysosomal structures are markedly reduced in these mutants. In contrast, we found that early endosomes, as marked by FYVE, RAB-5, RABEX5, and EEA1, are significantly enlarged in usp-50 mutants. Electron microscopy (EM) imaging further revealed an increase in large cellular vesicles containing various intraluminal structures. Given the reduction in lysosomal structures and the enlargement of early endosomes in usp-50 mutants, these enlarged vesicles are likely aberrant early endosomes rather than late endosomal or lysosomal structures. To address potential confusion, we have revised the manuscript according to the reviewer's comments and updated the model to accurately reflect these observations.

Reviewer #2 (Public Review): 

Summary: 

In this study, the authors study how the deubiquitinase USP8 regulates endosome maturation in C. elegans and mammalian cells. The authors have isolated USP8 mutant alleles in C. elegans and used multiple in vivo reporter lines to demonstrate the impact of USP8 loss-of-function on endosome morphology and maturation. They show that in USP8 mutant cells, the early endosomes and MVB-like structures are enlarged while the late endosomes and lysosomal compartments are reduced. They elucidate that USP8 interacts with Rabx5, a guanine nucleotide exchange factor (GEF) for Rab5, and show that USP8 likely targets specific lysine residue of Rabx5 to dissociate it from early endosomes. They also find that the localization of USP8 to early endosomes is disrupted in Rabx5 mutant cells. They observe that in both Rabx5 and USP8 mutant cells, the Rab7 GEF SAND-1 puncta which likely represents late endosomes are diminished, although Rabex5 is accumulated in USP8 mutant cells. The authors provide evidence that USP8 regulates endosomal maturation in a similar fashion in mammalian cells. Based on their observations they propose that USP8 dissociates Rabex5 from early endosomes and enhances the recruitment of SAND-1 to promote endosome maturation. 

We thank this reviewer for the instructive suggestions and encouragement.

Strengths: 

The major highlights of this study include the direct visualization of endosome dynamics in a living multi-cellular organism, C. elegans. The high-quality images provide clear in vivo evidence to support the main conclusions. The authors have generated valuable resources to study mechanisms involved in endosome dynamics regulation in both the worm and mammalian cells, which would benefit many members of the cell biology community. The work identifies a fascinating link between USP8 and the Rab5 guanine nucleotide exchange factor Rabx5, which expands the targets and modes of action of USP8. The findings make a solid contribution toward the understanding of how endosomal trafficking is controlled. 

We thank this reviewer for the instructive suggestions and encouragement.

Weaknesses: 

- The authors utilized multiple fluorescent protein reporters, including those generated by themselves, to label endosomal vesicles. Although these are routine and powerful tools for studying endosomal trafficking, these results cannot tell whether the endogenous proteins (Rab5, Rabex5, Rab7, etc.) are affected in the same fashion. 

Good suggestion. Indeed, to test whether the endogenous proteins (Rab5, Rabex5, Rab7, etc.) are affected in the same fashion as fluorescent protein reporters, we supplemented our approach with the utilization of endogenous markers. These markers, including Rab5, RAB-5, Rabex5, RABX-5, and EEA1 for early endosomes, as well as RAB-7, Mon1a, and Mon1b for late endosomes, were instrumental in our investigations (refer to Figure 3, Figure 6, Figure 5-figure supplement 1, Figure 5-figure supplement 2, and Figure 6-figure supplement 1). Our comprehensive analysis, employing various methodologies such as tissue-specific fused proteins, CRISPR/Cas9 knock-in, and antibody staining, consistently highlights the critical role of USP8 in early-to-late endosome conversion.

- The authors clearly demonstrated a link between USP8 and Rabx5, and they showed that cells deficient in both factors displayed similar defects in late endosomes/lysosomes. However, the authors didn't confirm whether and/or to which extent USP8 regulates endosome maturation through Rabx5. Additional genetic and molecular evidence might be required to better support their working model. 

Excellent point. To test whether USP-50 regulates endosome maturation through RABX-5, we performed additional genetic analyses. In rabx-5(null) mutant animals, the morphology of 2xFYVE-labeled early endosomes is comparable to that of wild-type controls (Figure 4H and I). Introducing the rabx-5(null) mutation into usp-50(xd413) backgrounds resulted in a significant suppression of the enlarged early endosome phenotype characteristic of usp-50(xd413) mutants (Figure 4H and I). These findings suggest that USP-50 may modulate the size of early endosomes through its interaction with RABX-5.

Reviewer #3 (Public Review): 

Summary: 

The authors were trying to elucidate the role of USP8 in the endocytic pathway. Using C. elegans epithelial cells as a model, they observed that when USP8 function is lost, the cells have a decreased number and size in lysosomes. Since USP8 was already known to be a protein linked to ESCRT components, they looked into what role USP8 might play in connecting lysosomes and multivesicular bodies (MVB). They observed fewer ESCRT-associated vesicles but an increased number of "abnormal" enlarged vesicles when USP8 function was lost. At this specific point, it's not clear what the objective of the authors was. What would have been their hypothesis addressing whether the reduced lysosomal structures in USP8 (-) animals were linked to MVB formation? Then they observed that the abnormally enlarged vesicles, marked by the PI3P biosensor YFP-2xFYVE, are bigger but in the same number in USP8 (-) compared to wild-type animals, suggesting homotypic fusion. They confirmed this result by knocking down USP8 in a human cell line, and they observed enlarged vesicles marked by YFP-2xFYVE as well. At this point, there is quite an important issue. The use of YFP-2xFYVE to detect early endosomes requires the transfection of the cells, which has already been demonstrated to produce differences in the distribution, number, and size of PI3P-positive vesicles (doi.org/10.1080/15548627.2017.1341465). The enlarged vesicles marked by YFP-2xFYVE would not necessarily be due to the loss of UPS8. In any case, it appears relatively clear that USP8 localizes to early endosomes, and the authors claim that this localization is mediated by Rabex-5 (or Rabx-5). They finally propose that USP8 dissociates Rabx-5 from early endosomes facilitating endosome maturation. 

Weaknesses: 

The weaknesses of this study are, on one side, that the results are almost exclusively dependent on the overexpression of fusion proteins. While useful in the field, this strategy does not represent the optimal way to dissect a cell biology issue. On the other side, the way the authors construct the rationale for each approximation is somehow difficult to follow. Finally, the use of two models, C. elegans and a mammalian cell line, which would strengthen the observations, contributes to the difficulty in reading the manuscript. 

The findings are useful but do not clearly support the idea that USP8 mediates Rab5-Rab7 exchange and endosome maturation, In contrast, they appear to be incomplete and open new questions regarding the complexity of this process and the precise role of USP8 within it. 

We thank this reviewer for the insightful comments. Fluorescence-fused proteins serve as potent tools for visualizing subcellular organelles both in vivo and in live settings. Specifically, in epidermal cells of worms, the tissue-specific expression of these fused proteins is indispensable for studying organelle dynamics within living organisms. This approach is necessitated by the inherent limitations of endogenously tagged proteins, whose fluorescence signals are often weak and unsuitable for live imaging or genetic screening purposes. Acknowledging concerns raised by the reviewer regarding potential alterations in organelle morphology due to overexpression of certain fused proteins, we supplemented our approach with the utilization of endogenous markers. These markers, including Rab5, RAB-5, Rabex5, RABX-5, and EEA1 for early endosomes, as well as RAB-7, Mon1a, and Mon1b for late endosomes, were instrumental in our investigations (refer to Figure 3, Figure 6, Figure 5-figure supplement 1, Figure 5-figure supplement 2, and Figure 6-figure supplement 1). Our comprehensive analysis, employing various methodologies such as tissue-specific fused proteins, CRISPR/Cas9 knock-in, and antibody staining, consistently highlights the critical role of USP8 in early-to-late endosome conversion. Specifically, we discovered that the recruitment of USP-50/USP8 to early endosomes is depending on Rabex5. However, instead of stabilizing Rabex5, the recruitment of USP-50/USP8 leads to its dissociation from endosomes, concomitantly facilitating the recruitment of the Rab7 GEF SAND-1/Mon1. In cells with loss-of-function mutations in usp-50/usp8, we observed enhanced RABX-5/Rabex5 signaling and mis-localization of SAND-1/Mon1 proteins from endosomes. Consequently, this disruption impairs endolysosomal trafficking, resulting in the accumulation of enlarged vesicles containing various intraluminal contents and rudimentary lysosomal structures.

Through an unbiased genetic screen, verified by cultured mammalian cell studies, we observed that loss-of-function mutations in usp-50/usp8 result in diminished lysosome/late endosomes. Electron microscopy (EM) analysis indicated that usp-50 mutation leads to abnormally enlarged vesicles containing various intraluminal structures in worm epidermal cells. USP8 is known to regulate the endocytic trafficking and stability of numerous transmembrane proteins. Given that lysosomes receive and degrade materials generated by endocytic pathways, we hypothesized that the abnormally enlarged vesicular structures observed in usp-50 or usp8 mutant cells correspond to the enlarged vesicles coated by early endosome markers. Indeed, in the absence of usp8/usp-50, the endosomal Rab5 signal is enhanced, while early endosomes are significantly enlarged. Given that Rab5 guanine nucleotide exchange factor (GEF), Rabex5, is essential for Rab5 activation, we further investigated its dynamics. Additional analyses conducted in both worm hypodermal cells and cultured mammalian cells revealed an increase of endosomal Rabex5 in response to usp8/usp-50 loss-of-function. Live imaging studies further demonstrated active recruitment of USP8 to newly formed Rab5-positive vesicles, aligning spatiotemporally with Rabex5 regulation. Through systematic exploration of putative USP-50 binding partners on early endosomes, we identified its interaction with Rabex5. Comprehensive genetics and biochemistry experiments demonstrated that USP8 acts through K323 site de-ubiquitination to dissociate Rabex5 from early endosomes and promotes the recruitment of the Rab7 GEF SAND-1/Mon1. In summary, our study began with an unbiased genetic screen and subsequent examination of established theories, leading to the formulation of our own hypothesis. Through multifaceted approaches, we unveiled a novel function of USP8 in early-to-late endosome conversion.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

(1) Within Figures 1K-N, diverse anomalous structures were detected in the usp-50 mutant. Further scrutiny is needed to definitively characterize these structures, particularly as the images in Figures 1M and 1L exhibit notable similarities to lamellar bodies.

We thank the reviewer for the insightful question regarding the resemblance between the vesicles observed in our study and lamellar bodies (LBs). Lamellar bodies are specialized organelles involved in lipid storage and secretion1, prominently studied in keratinocytes of the skin and alveolar type II (ATII) epithelial cells in the lung2. These organelles contain not only lipids but also cell-type specific proteins and lytic enzymes. Due to their acidic pH and functional similarities, LBs are classified as lysosome-related organelles (LROs) or secretory lysosomes3,4. In usp-50 mutants, we observed a considerable number of abnormal vesicles, some of which contain threadlike membrane structures and exhibit morphological similarities to LBs (Figure 2O). However, further analysis with a comprehensive panel of lysosome-related markers demonstrated a significant reduction in lysosomal structures within these mutants. In contrast, vesicles marked by early endosome markers, such as FYVE, RAB-5, RABX-5, and EEA1, were notably enlarged. These results suggest that the enlarged vesicles observed in usp-50 mutants are more likely aberrant early endosomes rather than true lamellar bodies. We have revised the manuscript to reflect these findings and to clearly differentiate between these structures and lysosome-related organelles.

(2) The correlation between the presence of these abnormal structures and ESCRT-0 remains unaddressed, thus the assertion that UPS-50 regulates endolysosome trafficking in conjunction with ESCRT-0 lacks empirical support.

We thank the reviewer for the valuable suggestions. We apologize for any confusion and appreciate the opportunity to clarify our findings. The ESCRT machinery is essential for driving endosomal membrane deformation and scission, which leads to the formation of intraluminal vesicles (ILVs) within multivesicular bodies (MVBs). Recent research has shown that the absence of ESCRT components results in a reduction of ILVs in worm gut cells5. In wild type animals, the ESCRT-0 components HGRS-1 and STAM-1 display a distinct punctate distribution (Figure 1G and H). However, in usp-50 mutants, the punctate signals of HGRS-1::GFP and STAM-1::GFP are significantly reduced (Figure 1G and H; and Figure 1-figure supplement 1B), indicating a role for USP-50 in stabilizing the ESCRT-0 complex. Our TEM analysis revealed an accumulation of abnormally enlarged vesicles containing intraluminal structures in usp-50 mutants. When we examined a panel of early endosome and late endosome/lysosome markers, we found that early endosomes are significantly enlarged, while late endosomal/lysosomal structures are markedly reduced in these mutants. This suggests that the abnormal structures observed in usp-50 mutants are likely enlarged early endosomes rather than classical MVBs. To further investigate whether the reduction in ESCRT components contributes to the late endosome/lysosome defects, we analyzed stam-1 mutants. In these mutants, the size of RAB-7-coated vesicles was reduced (Author response image 1C), and the lysosomal marker LAAT-1 indicated a reduction in lysosomal structures (Author response image 1B). These results highlight the importance of the ESCRT complex in late endosome/lysosome formation. However, the morphology of early endosomes, as marked by 2xFYVE, remained similar to that of wild type in stam-1 mutants (Author response image 1A). Therefore, while reduced ESCRT-0 components may contribute to the late endosome/lysosome defects observed in usp-50 mutants, the enlargement of early endosomes in these mutants may involve additional mechanisms. We have revised the manuscript to incorporate these insights and to address the reviewer's comments more comprehensively.

Author response image 1.

(A) Confocal fluorescence images of hypodermis expressing YFP::2xFYVE to detect EEs in L4 stage animals in wild type and stam-1(ok406) mutants. Scale bar: 5 μm. (B) Confocal fluorescence images of hypodermal cell 7 (hyp7) expressing the LAAT-1::GFP marker to highlight lysosome structures in 3-day-old adult animals. Compared to wild type, LAAT-1::GFP signal is reduced in stam-1(ok406) animals. Scale bar, 5 μm. (C) The reduction of punctate endogenous GFP::RAB-7 signals in stam-1(ok406) animals. Scale bar: 10 μm.

(3) Endosomal dysfunction typically leads to significant alterations in the spatial arrangement of marker proteins across distinct endosomes. In the manuscript, the authors examined the distribution and morphology of early endosomes, multivesicular bodies (MVBs), late endosomes, and lysosomes in a usp-50 deficient background primarily through single-channel confocal imaging. By employing two color images showing RAB-5 and RAB-7, in conjunction with HGRS-1, a more comprehensive picture of the aftermath of USP-50 loss can be obtained.

Good suggestions. We have conducted a double-labeling analysis to examine the distribution of RAB-5 and RAB-7 in conjunction with HGRS-1. In wild type animals, HGRS-1 exhibits a punctate distribution that is partially co-localized with both RAB-5 and RAB-7. In contrast, in usp-50 mutants, the punctate signal of HGRS-1 is significantly reduced, along with its co-localization with RAB-5 and RAB-7 (Author response image 2). These results suggest that, in the absence of USP-50, the stabilization of ESCRT-0 components on endosomes is compromised.

Author response image 2.

ESCRT-0 is adjacent to both early endosomes and late endosomes. (A) Confocal fluorescence images of wild-type and usp-50(xd413) hypodermis at L4 stage co-expressing HGRS-1::GFP (hgrs-1 promoter) and endogenous wrmScarlet::RAB-5. (B) HGRS-1 and RAB-5 puncta were analyzed to produce Manders overlap coefficient M1 (HGRS-1/RAB-5) and M2 (RAB-5/HGRS-1) (N=10). (C) Confocal fluorescence images of wild-type and usp-50(xd413) hypodermis at L4 stage co-expressing HGRS-1::GFP (hgrs-1 promoter) and endogenous wrmScarlet::RAB-7. (D) HGRS-1 and RAB-7 puncta were analyzed to produce Manders overlap coefficient M1 (HGRS-1/RAB-7) and M2 (RAB-7/HGRS-1) (N=10). Scale bar: 10 μm for (A) and (C).

(4) The authors observed enlarged early endosomes in cells depleted of usp-50/usp8, along with enlarged MVB-like structures identified through TEM. The potential identity of these structures as the same organelle could be determined using CLEM.

We thank the reviewer for the valuable suggestion. Our TEM analysis identified a large number of abnormally enlarged vesicles with various intraluminal structures accumulated in usp-50 mutants. As the reviewer correctly noted, CLEM (correlative light and electron microscopy) would be an ideal approach to further characterize these structures. We have been attempting to implement CLEM in C. elegans for a few years. Given that CLEM relies on fluorescence markers, in this study we focused on two tagged proteins, RAB-5 and RABX-5, which show enlargement in their vesicles in usp-50 mutants. Unfortunately, we encountered significant challenges with this approach, as the GFP-tagged RAB-5 and RABX-5 signals did not survive the electron microscopy procedure. Attempts to align EM sections with residual GFP signaling yielded results that were not convincing. Consequently, we concentrated our analysis on a panel of molecular markers, including 2xFYVE, RAB-5, RABX-5, RAB-7, and LAAT-1. These markers consistently indicated that early endosomes are specifically enlarged in usp-50 mutants, while late endosomal/lysosomal structures are notably reduced. Thus, the abnormal structures identified in usp-50 mutants via TEM are likely to be enlarged early endosomes rather than the classical view of MVBs. We have revised the manuscript to reflect these findings and to clarify this point.

(5) The working model depicted in Figure 6 Y (right) requires revision, as it has the potential to mislead authors into mistaking enlarged early endosomes for multivesicular bodies (MVBs).

We thank the reviewer for the excellent suggestion. We have revised the model to clarify that it is the enlarged early endosomes, rather than MVBs, that are observed in usp-50 mutants.

Reviewer #2 (Recommendations For The Authors):

(1) Is there any change of Rabx5 protein level in USP8/USP50 mutant cells?

Good question. In the absence of usp-50/usp8, we indeed observed a noticeable increase in the signal of Rabex5 on endosomes. To determine whether usp-50/usp8 affects the protein level of Rabex5, we investigated the endogenous levels of RABX-5 using the RABX-5::GFP knock-in line. Compared to wild-type controls, we found an elevated protein level of RABX-5::GFP in the knock-in line (Author response image 3). This suggests that USP-50 may play a role in the destabilization of RABX-5/Rabex5 in vivo.

Author response image 3.

The endogenous RABX-5 protein level is increased in usp-50 mutants. (A) The RABX-5::GFP KI protein level is increased in usp-50(xd413). (B) Quantification of endogenous RABX-5::GFP protein level in wild type and usp-50(xd413) mutant animals.

(2) It is interesting that "The rabx-5(null) animals are healthy and fertile and do not display obvious morphological or behavioral defects.", which seems contrary to its role in regulating USP8 localization and endosome maturation.

It has been previously documented that rabx-5 functions redundantly with rme-6, another RAB-5 GEF in C. elegans, to regulate RAB-5 localization in oocytes6. RNA interference (RNAi) targeting rabx-5 in a rme-6 mutant background results in synthetic lethality, whereas neither rabx-5 nor rme-6 single mutants are essential for worm viability. RME-6 co-localizes with clathrin-coated pits, while Rabex-5 is localized to early endosomes. Rabex-5 forms a stable complex with Rabaptin-5 and is part of a large EEA1-positive complex on early endosomes, whereas RME-6 does not interact with Rabaptin-5 (RABN-5) or EEA-1. These findings suggest that while RME-6 and RABX-5 may function redundantly, they likely play distinct roles in regulating intracellular trafficking processes. In the absence of RABX-5, USP-50 appears to lose its endosomal localization, although the size of the early endosome remains comparable to that of wild type. This observation contrasts with the phenotype associated with USP-50 loss-of-function, in which the early endosome is notably enlarged. These results suggest that residual USP-50 present in the endosomes is sufficient to maintain its role in the endocytic pathway. Conversely, the complete absence of USP-50 likely disrupts the transition of early endosomes to late endosomes, indicating a crucial role of USP-50 in this conversion process. It is also noteworthy that, although loss-of-function of rabx-5 does not result in obvious changes to early endosomes, increasing the gene expression level of rabx-5/Rabex-5 alone is sufficient to cause enlargement of early endosomes (Author response image 4) . Indeed, we observed that loss-of-function mutations in u_sp-50/usp_8 lead to abnormally enlarged early endosomes, accompanied by an enhanced signal of endosomal RABX-5. When the rabx-5(null) mutation was introduced into usp-50 mutant animals, the enlarged early endosome phenotype seen in usp-50 mutants was significantly suppressed (Figure 4H and I). This implies that maintaining a lower level of Rab5 GEF may be crucial for endolysosomal trafficking.

(3) Does Rabx5 mutation has any impact on early endosomes?

To address the question, we utilized the CRISPR/Cas9 technique to create a molecular null for rabx-5 (Figure 4E). In the rabx-5(null) mutant animals, we found that the 2xFYVE-labeled early endosomes are indistinguishable from wild type (Figure 4H and 4I). Given that r_abx-5_ functions redundantly with rme-6, another RAB-5 GEF in C. elegans, it is likely that the regulation of early endosome size involves a cooperative interaction between RABX-5 and RME-6.

(4) The authors observed a reduction of ESCRT-0 components in USP8 mutant cells, could this contribute to the late endosome/lysosome defects?

Good suggestion. In wild-type animals, the two ESCRT-0 components, HGRS-1 and STAM-1, exhibit a distinct punctate distribution (Figure 1G and H). However, in usp-50 mutants, the punctate signals of HGRS-1::GFP and STAM-1::GFP are significantly diminished (Figure 1G and H; and Figure 1-figure supplement 1B), which aligns with the role of USP-50 in stabilizing the ESCRT-0 complex. To investigate whether the reduction in ESCRT components might contribute to defects in late endosome/lysosome formation, we examined stam-1 mutants. In stam-1 mutants, we observed a reduction in the size of RAB-7-coated vesicles (Author response image 1). Further, when we introduced the lysosomal marker LAAT-1::GFP into stam-1 mutants, we found a substantial decrease in lysosomal structures compared to wild-type animals (Author response image 1). This suggests that the ESCRT complex is essential for proper late endosome/lysosome formation. In contrast, the morphology of early endosomes, as indicated by the 2xFYVE marker, appeared normal in stam-1 mutants, similar to wild-type animals (Author response image 1). This implies that while a reduction in ESCRT-0 components may contribute to the late endosome/lysosome defects observed in usp-50 mutants, the early endosome enlargement phenotype in _usp-5_0 mutants may involve additional mechanisms.

(5) Rabx5 is accumulated in USP8 mutant cells, I am very curious about the phenotype of USP8-Rabx5 double mutants. Could over-expression of Rabx5 (wild type or mutant forms) cause any defects?

Excellent suggestions. To address the question, we employed the CRISPR/Cas9 technique to create a molecular null for rabx-5 (Figure 4E). In the rabx-5(null) mutant animals, we observed that the punctate USP-50::GFP signal became diffusely distributed (Figure 4F and G). This suggests that rabx-5 is necessary for the endosomal localization of USP-50. Interestingly, in rabx-5(null) mutant animals, the 2xFYVE-labeled early endosomes appeared similar to those in wild-type animals (Figure 4H and I). When rabx-5(null) was introduced into usp-50 mutant animals, the enlarged early endosome phenotype observed in usp-50 was significantly suppressed (Figure 4H and I). This finding indicates that usp-50 indeed functions through rabx-5 to regulate early endosome size. Additionally, we constructed strains overexpressing either wild-type or K323R mutant RABX-5. Our results showed that overexpression of wild-type RABX-5 led to early endosome enlargement (as indicated by YFP::2xFYVE labeling) (Author response image 4A, B and D). In contrast, overexpression of the K323R mutant RABX-5 did not result in noticeable early endosome enlargement (Author response image 4A, C and D). Together, these data are in consistent with our model that USP-50 may regulate RABX-5 by deubiquitinating the K323 site.

Author response image 4.

(A-C) Over-expression wild type RABX-5 causes enlarged EEs (labeled by YFP::2xFYVE) while RABX-5(K323R) mutant form does not. (D) Quantification of the volume of individual YFP::2xFYVE vesicles. Data are presented as mean ± SEM. ****P<0.0001. ns, not significant. One-way ANOVA with Tukey’s test.

(6) Rabx5 could be ubiquitinated at K88 and K323, and Rabx5-K323R showed different activity when compared with the wild-type protein in USP8 mutant cells. Could the authors provide evidence that USP8 could remove the ubiquitin modification from K323 in Rabx5 protein?

We appreciate the reviewer's insightful suggestions. To explore the potential of USP-50 in removing ubiquitin modifications from lysine 323 on the RABX-5 protein, we undertook a series of experiments. Initially, we sought to determine whether USP-50 influences the ubiquitination level of RABX-5 in vivo. However, due to the low expression levels of USP-50, we encountered challenges in obtaining adequate amounts of USP-50 protein from worm lysates. To overcome this, we expressed USP-50::4xFLAG in HEK293 cells for subsequent affinity purification. Concurrently, we utilized anti-GFP agarose beads to purify RABX-5::GFP from worms expressing the rabx-5::gfp construct. We then incubated RABX-5::GFP with USP-50::4xFLAG for varying durations and performed immunoblotting with an anti-ubiquitin antibody. As shown in Author response image 5A, our results revealed a decrease in the ubiquitination level of RABX-5 in the presence of USP-50, suggesting that USP-50 directly deubiquitinates RABX-5. Previous studies have indicated that only a minor fraction of recombinant RABX-5 undergoes ubiquitination in HeLa cells, which is believed to have functional significance7. Our findings are consistent with this observation, as only a small fraction of RABX-5 in worms is ubiquitinated. Rabex-5 is known to interact with both K63- and K48-linked poly-ubiquitin chains. To further elucidate whether USP-50 specifically targets K48 or K63-linked ubiquitination at the K323 site of RABX-5, we incubated various HA-tagged ubiquitin mutants with either wild-type or K323R mutant RABX-5 protein. Our results indicated that the K323R mutation reduces K63-linked ubiquitination of RABX-5 (Author response image 5). This experiment was repeated multiple times with consistent results. Additionally, while overexpression of wild-type RABX-5 led to an enlargement of early endosomes, as evidenced by YFP::2xFYVE labeling, overexpression of the K323R mutant did not produce a noticeable effect on endosome size (Author response image 4). Collectively, this finding indicates that RABX-5 is subject to ubiquitin modification in vivo and that USP-50 plays a significant role in regulating this modification at the K323 site.

Author response image 5.

(A) RABX-5::GFP protein was purified from worm lysates using anti-GFP antibody. FLAG-tagged USP-50 was purified from HEK293T cells using anti-FLAG antibody. Purified RABX-5::GFP was incubated with USP-50::4FLAG for indicated times (0, 15, 30, 60 mins), followed by immunoblotting using antibody against ubiquitin, FLAG or GFP. In the presence of USP-50::4xFLAG, the ubiquitination level of RABX-5::GFP is decreased. (B) Quantification of RABX-5::GFP ubiquitination level from three independent experiments. (C) HEK293T cells were transfected with HA-Ub or indicated mutants and 4xFLAG tagged RABX-5 or RABX-5 K323R mutant for 48h. The cells were subjected to pull down using the FLAG beads, followed by immunoblotting using antibody against HA or Flag.

(7) The authors described "the almost identical phenotype of usp-50/usp8 and sand-1/Mon1 mutants", found protein-protein interaction between USP8 and sand-1, and showed that sand1-GFP signal is diminished in USP8 mutant cells. These observations fit with the possibility that USP8 regulates the stability of sand-1 to promote endosomal maturation. Could this be tested and integrated into the current model?

are grateful for the insightful comments provided by the reviewer. Rab5, known to be activated by Rabex-5, plays a crucial role in the homotypic fusion of early endosomes. Rab5 effectors also include the Rab7 GEF SAND-1/Mon1–Ccz1 complex. Rab7 activation by SAND-1/Mon1-Ccz1 complex is essential for the biogenesis and positioning of late endosomes (LEs) and lysosomes, and for the fusion of endosomes and autophagosomes with lysosomes. The Mon1-Ccz1 complex is able to interact with Rabex5, causing dissociation of Rabex5 from the membrane, which probably terminates the positive feedback loop of Rab5 activation and then promotes the recruitment and activation of Rab7 on endosomes. In our study, we identified an interaction between USP-50 and the Rab5 GEF, RABX-5. In the absence of USP-50, we observed an increased endosomal localization of RABX-5 and the formation of abnormally enlarged early endosomes. This phenotype is reminiscent of that seen in sand-1 loss-of-function mutants, which also exhibit enlarged early endosomes and a concomitant reduction in late endosomes/lysosomes. Notably, USP-50 also interacts with SAND-1, suggesting a potential role in regulating its localization. We could propose several models to elucidate how USP-50 might influence SAND-1 localization, including:

(1) USP-50 may stabilize SAND-1 through direct de-ubiquitination.

(2) In the absence of USP-50, the sustained presence of RABX-5 could lead to continuous Rab5 activation, which might hinder or delay the recruitment of SAND-1.

(3) USP-50 could facilitate SAND-1 recruitment by promoting the dissociation of RABX-5.

We are actively investigating these models in our laboratory. Due to space constraints, a more detailed exploration of how USP-50 regulates SAND-1 stability will be presented in a separate publication.

References:

(1) Schmitz, G., and Müller, G. (1991). Structure and function of lamellar bodies, lipid-protein complexes involved in storage and secretion of cellular lipids. J Lipid Res 32, 1539-1570.

(2) Dietl, P., and Frick, M. (2021). Channels and Transporters of the Pulmonary Lamellar Body in Health and Disease. Cells-Basel 11. https://doi.org/10.3390/cells11010045.

(3) Raposo, G., Marks, M.S., and Cutler, D.F. (2007). Lysosome-related organelles: driving post-Golgi compartments into specialisation. Current opinion in cell biology 19, 394-401. https://doi.org/10.1016/j.ceb.2007.05.001.

(4) Weaver, T.E., Na, C.L., and Stahlman, M. (2002). Biogenesis of lamellar bodies, lysosome-related organelles involved in storage and secretion of pulmonary surfactant. Semin Cell Dev Biol 13, 263-270. https://doi.org/10.1016/s1084952102000551.

(5) Ott, D.P., Desai, S., Solinger, J.A., Kaech, A., and Spang, A. (2024). Coordination between ESCRT function and Rab conversion during endosome maturation. bioRxiv, 2024.2005.2014.594104. https://doi.org/10.1101/2024.05.14.594104.

(6) Sato, M., Sato, K., Fonarev, P., Huang, C.J., Liou, W., and Grant, B.D. (2005). Caenorhabditis elegans RME-6 is a novel regulator of RAB-5 at the clathrin-coated pit. Nature cell biology 7, 559-569. https://doi.org/10.1038/ncb1261.

(7) Mattera, R., Tsai, Y.C., Weissman, A.M., and Bonifacino, J.S. (2006). The Rab5 guanine nucleotide exchange factor Rabex-5 binds ubiquitin (Ub) and functions as a Ub ligase through an atypical Ub-interacting motif and a zinc finger domain. The Journal of biological chemistry 281, 6874-6883. https://doi.org/10.1074/jbc.M509939200.

eLife Assessment

This study demonstrates mRNA-specific regulation of translation by subunits of the eukaryotic initiation factor complex 3 (eIF3) using convincing methods, data, and analyses. The investigations have generated important information that will be of interest to biologists studying translation regulation. However, the physiological significance of the gene expression changes that were observed is not clear.

Reviewer #1 (Public review):

Summary:<br /> In this manuscript, Herrmannova et al explore changes in translation upon individual depletion of three subunits of the eIF3 complex (d, e and h) in mammalian cells. The authors provide a detailed analysis of regulated transcripts, followed by validation by RT-qPCR and/or Western blot of targets of interest, as well as GO and KKEG pathway analysis. The authors confirm prior observations that eIF3, despite being a general translation initiation factor, functions in mRNA-specific regulation, and that eIF3 is important for translation re-initiation. They show that global effects of eIF3e and eIF3d depletion on translation and cell growth are concordant. Their results support and extend previous reports suggesting that both factors control translation of 5'TOP mRNAs. Interestingly, they identify MAPK pathway components as a group of targets coordinately regulated by eIF3 d/e. The authors also discuss discrepancies with other reports analyzing eIF3e function.

Strengths:<br /> Altogether, a solid analysis of eIF3 d/e/h-mediated translation regulation of specific transcripts. The data will be useful for scientists working in the Translation field.

Weaknesses:<br /> The authors could have explored in more detail some of their novel observations, as well as their impact on cell behavior.

The manuscript has improved with the new corrections. I appreciate the authors' attention to the minor comments, which have been fully solved. The authors have not, however, provided additional experimental evidence that uORF-mediated translation of Raf-1 mRNA depends on an intact eIF3 complex, nor have they addressed the consequences of such regulation for cell physiology. While I understand that this is a subject of follow-up research, the authors could have at least included their explanations/ speculations regarding major comments 2-4, which in my opinion could have been useful for the reader.

Reviewer #2 (Public review):

Summary:<br /> mRNA translation regulation permits cells to rapidly adapt to diverse stimuli by fine tuning gene expression. Specifically, the 13-subunit eukaryotic initiation factor 3 (eIF3) complex is critical for translation initiation as it aids in 48S PIC assembly to allow for ribosome scanning. In addition, eIF3 has been shown to drive transcript-specific translation by binding mRNA 5' cap structures through the eIF3d subunit. Dysregulation of eIF3 has been implicated in oncogenesis, however the precise eIF3 subunit contributions are unclear. Here, Herrmannová et al. aim to investigate how eIF3 subcomplexes, generated by knock down (KD) of either eIF3e, eIF3d or eIF3h, affect the global translatome. Using Ribo-seq and RNA-seq, the authors identified a large number of genes that exhibit altered translation efficiency upon eIF3d/e KD, while translation defects upon eIF3h KD were mild. eIF3d/eKD share multiple dysregulated transcripts, perhaps due to both subcomplexes lacking eIF3d. Both eIF3d/e KD increase translation efficiency (TE) of transcripts encoding lysosomal, ER and ribosomal proteins, suggesting a role of eIF3 in ribosome biogenesis and protein quality control. Many transcripts encoding ribosomal proteins harbor a TOP motif, and eIF3d KD and eIF3e KD cells exhibit a striking induction of these TOP-modified transcripts. On the other hand, eIF3d KD and eIF3e KD leads to a reduction of MAPK/ERK pathway proteins. Despite this downregulation, eIF3d KD and eIF3e KD activates MAPK/ERK signaling as ERK1/2 and c-Jun phosphorylation was induced. Finally, in all three knockdowns, MDM2 and ATF4 protein levels are reduced. This is notable because MDM2 and ATF4 both contain short uORFs upstream of the start codon, and further supports a role of eIF3 in reinitiation. Altogether, Herrmannová et al. have gained key insights to precise eIF3-mediated translational control as it relates to key signaling pathways implicated in cancer.

Strengths:<br /> The authors have provided a comprehensive set of data to analyze RNA and ribosome footprinting upon perturbation of eIF3d, eIF3e, and eIF3h. As described above in the summary, these data present many interesting starting points to understand additional roles of the eIF3 complex and specific subunits in translational control.

Weaknesses:<br /> - The differences between eIF3e and eIF3d knockdown are difficult to reconcile, especially since eIF3e knockdown leads to reduction in eIF3d levels.<br /> - The paper would be strengthened by experiments directly testing what RNA determinants allow for transcript-specific translation regulation by the eIF3 complex. This would allow the paper to be less descriptive.<br /> - The paper would have more biological relevance if eIF3 subunits were perturbed to mimic naturally occurring situations where eIF3 is dysregulated. For example, eIF3e is aberrantly upregulated in certain cancers, and therefore an overexpression and profiling experiment would have been more relevant than a knockdown experiment.

The first review is unchanged as no additional experiments were provided to address the first review.

Reviewer #3 (Public review):

Summary:<br /> In this article, Hermannova et al catalog the changes in ribosome association with mRNAs when the multisubunit eukaryotic translation initiation factor 3 is disrupted by knocking down individual subunits. They find that RNAs relying on TOP motifs for translation, such as ribosomal protein RNAs, and RNAs encoding modification enzymes in the ER and components of the lysosome are upregulated. In contrast, proteins encoding components of MAP kinase cascades are downregulated when subunits of eIF3 are knocked down, but retain elevated levels of activity.

Strengths:<br /> The authors use ribosome profiling of well-characterized mutants lacking subunits of eIF3 and assess the changes in translation that take place. They supplement the ribosome association studies with western blotting to determine protein level changes of affected transcripts. They analyze what transcripts undergo translation changes, which is important for understanding more broadly how translation initiation factor levels affect cancer cell translatomes. Changes observed by both ribosome profiling and western blotting supports their claims that eIF3 functions in mRNA-specific control of translation.

Weaknesses:<br /> (1) The paper would be strengthened if there were a clear model tying the various effects together or linking individual subunit knockdown to cancerous phenotypes. It is noted that the authors plan to address such outcomes of eIF3 dysregulation in future work, which will be of interest.

(2) The paper could also be strengthened if some of the experiments were performed in at least one other cell type to determine whether changes observed are general or cell-type specific. The authors discuss this issue and provide a literature citation to support a more general mechanism.

eLife Assessment

The study is considered important with solid evidence that demonstrates the impact of plasma membrane nano-domains and protein interactions in the plant defence response to viruses. It includes a molecular understanding of the role of a calcium dependent kinase (CPK3) and a remorin protein in the cell-to-cell spread of viruses and cytoskeletal dynamics demonstrating, conclusively, the role of CPK3 with multiple lines of evidence. The work opens avenues to investigate different viruses and other plasma membrane proteins to gain a fuller picture of the involvement of plasmodesmata and other nanodomains in virus spreading.

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:

I would like to thank the researchers for including some additional work suggested in the previous round of peer review. However, I still have concerns over this work which are two fold.

(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. In addition, the quality of lots of the confocal based imaging (including new figure on colocalisation) is simply not sufficient. The images are too noisy and no clear conclusions can be made. The point made previously, the system this data was collected on has an Airyscan detector capable of 120nm resolution and as such NDs can be resolved. The sptPALM data conclusions are nice and fit the narrative. The inclusion of sptPALM movies is useful for the reader and tracks numbers is highly beneficial. But they do not show a high signal to noise ratio compared to other work in the field (see work from Alex Martineire) and the mEOS prticles are only just observable over the detector noise in some videos. As such, I worry about the data quality on which the analysis is based on. In addition, in some of the videos the conversion laser seems too high as it is difficult to separate some of the single particles as they emerge which would again, hinder the analysis.

(2) Secondly, remorins are involved in a lot of nano domain 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 and the inclusion of CPK3-CA also strengthens the authors claims. But to propose a kiss and go model bearing in mind the differences in diffusion between CPK3 and REM3 and differential changes to diffusion between the two proteins after PIAMV infection without two colour imaging of both proteins at the same time, the claims are much stronger than the evidence. Negative control experiments are required here utilising other PM localised proteins which have no role during viral infection (such as Lti6B).

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

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, including examination of their roles in cytoskeletal dynamics.

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 limited discussion or exploration of the role that CPK3 has in cytoskeletal organisation and whether this may play a role in the plant's defence against viral propagation. Further. although the authors show that there is no accumulation of CPK3/Rem1.2 at plasmodesmata, it would be interesting to investigate whether the demonstrated reduction of viral propagation is due to changes in PD permeability.

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

eLife Assessment

This valuable work describes results from a set of simulation and empirical studies of a set-up assessing exploratory behavior in a potentially rewarding environment that contains danger. The core idea is that an instrumental agent can be helped to be both effective and safe, thus avoiding excessive danger, during exploratory behavior, if its influence is flexibly gated by an independent Pavlovian fear learning system. The conclusion that safe, but effective exploration can be achieved based on a flexibly weighted combination of a Pavlovian and an instrumental agent is solid.

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.

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.

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.

Strengths:

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

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.

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?

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).

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.

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.

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.

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).

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.

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.

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

eLife Assessment

The authors identify a novel relationship between exosome secretion and filopodia formation that has implications for cancer cell metastasis and neuronal synapse formation. Further, they identify the exosomal cargo, THSD7A, as a regulator of this process. The data presented is convincing, and represents an important advancement in our understanding of how these two biological processes are linked and play roles in regulating cell migration and cell-cell communication.

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.

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

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.

(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.

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.

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.

(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.

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?

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.

eLife Assessment

This valuable study investigates how the size of an LLM may influence its ability to model the human neural response to language recorded by ECoG. Overall, solid evidence is provided that larger language models can better predict the human ECoG response. Further discussion would be beneficial as to how the results can inform us about the brain or LLMs, especially about the new message that can be learned from this ECoG study beyond previous fMRI studies on the same topic. This study will be of interest to both neuroscientists and psychologists who work on language comprehension and computer scientists working on LLMs.

Reviewer #1 (Public review):

Summary:

The authors perform an analysis of the relationship between the size of an LMM and the predictive performance of an ECoG encoding model made using the representations from that LMM. They find a logarithmic relationship between model size and prediction performance, consistent with previous findings in fMRI. They additionally observe that as the model size increases, the location of the "peak" encoding performance typically moves further back into the model in terms of percent layer depth, an interesting result worthy of further analysis into these representations.

Strengths:

The evidence is quite convincing, consistent across model families, and complementary to other work in this field. This sort of analysis for ECoG is needed and supports the decade-long enduring trend of the "virtuous cycle" between neuroscience and AI research, where more powerful AI models have consistently yielded more effective predictions of responses in the brain. The lag analysis showing that optimal lags do not change with model size is a nice result using the higher temporal resolution of ECoG compared to other methods like fMRI.

Weaknesses:

I would have liked to have seen the data scaling trends explored a bit too, as this is somewhat analogous to the main scaling results. While better performance with more data might be unsurprising, showing good data scaling would be a strong and useful justification for additional data collection in the field, especially given the extremely limited amount of existing language ECoG data. I realize that the data here is somewhat limited (only 30 minutes per subject), but authors could still in principle train models on subsets of this data.

Separately, it would be nice to have better justification of some of these trends, in particular the peak layerwise encoding performance trend and the overall upside-down U-trend of encoding performance across layers more generally. There is clearly something very fundamental going on here, about the nature of abstraction patterns in LLMs and in the brain, and this result points to that. I don't see the lack of justification here as a critical issue, but the paper would certainly be better with some theoretical explanation for why this might be the case.

Lastly, I would have wanted to see a similar analysis here done for audio encoding models using Whisper or WavLM as this is the modality where you might see real differences between ECoG and other slower scanning approaches. Again, I do not see this omission as a fundamental issue, but it does seem like the sort of analysis for which the higher temporal resolution of ECoG might grant some deeper insight.

Reviewer #2 (Public review):

Summary:

This paper investigates whether large language models (LLMs) of increasing size more accurately align with brain activity during naturalistic language comprehension. The authors extracted word embeddings from LLMs for each word in a 30-minute story and regressed them against electrocorticography (ECoG) activity time-locked to each word as participants listened to the story. The findings reveal that larger LLMs more effectively predict ECoG activity, reflecting the scaling laws observed in other natural language processing tasks.

Strengths:

(1) The study compared model activity with ECoG recordings, which offer much better temporal resolution than other neuroimaging methods, allowing for the examination of model encoding performance across various lags relative to word onset.

(2) The range of LLMs tested is comprehensive, spanning from 82 million to 70 billion parameters. This serves as a valuable reference for researchers selecting LLMs for brain encoding and decoding studies.

(3) The regression methods used are well-established in prior research, and the results demonstrate a convincing scaling law for the brain encoding ability of LLMs. The consistency of these results after PCA dimensionality reduction further supports the claim.

Weaknesses:

(1) Some claims of the paper are less convincing. The authors suggested that "scaling could be a property that the human brain, similar to LLMs, can utilize to enhance performance", however, many other animals have brains with more neurons than the human brain, making it unlikely that simple scaling alone leads to better language performance. Additionally, the authors claim that their results show 'larger models better predict the structure of natural language.' However, it remains unclear to what extent the embeddings of LLMs capture the "structure" of language better than the lexical semantics of language.

(2) The study lacks control LLMs with randomly initialized weights and control regressors, such as word frequency and phonetic features of speech, making it unclear what the baseline is for the model-brain correlation.

(3) The finding that peak encoding performance tends to occur in relatively earlier layers in larger models is somewhat surprising and requires further explanation. Since more layers mean more parameters, if the later layers diverge from language processing in the brain, it raises the question of what aspects of the larger models make them more brain-like.

Reviewer #3 (Public review):

This manuscript studies the connection between neural activity collected through electrocorticography and hidden vector representations from autoregressive language models, with the specific aim of studying the influence of language model size on this connection. Neural activity was measured from subjects who listened to a segment from a podcast, and the representations from language models were calculated using the written transcription as the input text. The ability of vector representations to predict neural activity was evaluated using 10-fold cross-validation with ridge regression models.

The main results are that (as well summarized in section headings):

(1) Larger models predict neural activity better.

(2) The ability of language model representations to predict neural activity differs across electrodes and brain regions.

(3) The layer that best predicts neural activity differs according to model size, with the "SMALL" model showing a correspondence between layer number and the language processing hierarchy.

(4) There seems to be a similar relationship between the time lag and the ability of language model representations to predict neural activity across models.

Strengths:

(1) The experimental and modeling protocols generally seem solid, which yielded results that answer the authors' primary research question.

(2) Electrocorticography data is especially hard to collect, so these results make a nice addition to recent functional magnetic resonance imaging studies.

Weaknesses:

(1) The interpretation of some results seems unjustified, although this may just be a presentational issue.

a) Figure 2B: The authors interpret the results as "a plateau in the maximal encoding performance," when some readers might interpret this rather as a decline after 13 billion parameters. Can this be further supported by a significance test like that shown in Figure 4B?

b) Figure S1A: It looks like the drop in PCA max correlation is larger for larger models, which may suggest to some readers that the same trend observed for ridge max correlation may not hold, contra the authors' claim that all results replicate. Why not include a similar figure as Figure 2B as part of Figure S1?

(2) Discussion of what might be driving the main result about the influence of model size appears to be missing (cf. the authors aim to provide an explanation of what seems to drive the influence of the layer location in Paragraph 3 of the Discussion section). What explanations have been proposed in the previous functional magnetic resonance imaging studies? Do those explanations also hold in the context of this study?

(3) The GloVe-based selection of language-sensitive electrodes (at least to me) isn't explained/motivated clearly enough (I think a more detailed explanation should be included in the Materials and Methods section). If the electrodes are selected based on GloVe embeddings, then isn't the main experiment just showing that representations from larger language models track more closely with GloVe embeddings? What justifies this methodology?

(4) (Minor weakness) The main experiments are largely replications of previous functional magnetic resonance imaging studies, with the exception of the one lag-based analysis. Is there anything else that the electrocorticography data can reveal that functional magnetic resonance imaging data can't?

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:

The Notch signaling pathway plays an important role in many developmental and disease processes. Although well-studied there remain many puzzling aspects. One is the fact that as well as activating the receptor through trans-activation, the transmembrane ligands can interact with receptors present in the same cell. These cis-interactions are usually inhibitory, but in some cases, as in the assays used here, they may also be activating. With a total of 6 ligands and 4 receptors, there is potentially a wide array of possible outcomes when different combinations are co-expressed in vivo. Here the authors set out to make a systematic analysis of the qualitative and quantitative differences in the signaling output from different receptor-ligand combinations, generating sets of "signaling" (ligand expressing) and "receiving" (receptor +/- ligand expressing cells).

The readout of pathway activity is transcriptional, relying on the fusion of GAL4 in the intracellular part of the receptor. Positive ligand interactions result in the proteolytic release of Gal4 that turns on the expression of H2B-citrine. As an indicator of ligand and receptor expression levels, they are linked via TA to H2B mCherry and H2B mTurq expression respectively. The authors also manipulate the expression of the glycosyltransferase Lunatic-Fringe (LFng) that modifies the EGF repeats in the extracellular domains impacting their interactions. The testing of multiple ligand-receptor combinations at varying expression levels is a tour de force, with over 50 stable cell lines generated, and yields valuable insights although as a whole, the results are quite complex.

Strengths:

Taking a reductionist approach to testing systematically differences in the signaling strength, binding strength, and cis-interactions from the different ligands in the context of the Notch1 and Notch 2 receptors (they justify well the choice of players to test via this approach) produces a baseline understanding of the different properties and leads to some unexpected and interesting findings. Notably:

-                Jag1 ligand expressing cells failed to activate Notch1 receptor although were capable of activating Notch2. Conversely, Jag2 cells elicited the strongest activation of both receptors. The results with

Jag1 are surprising also because it exhibits some of the strongest binding to plate-bound ligands. The failure to activate Notch1 has major functional significance and it will be important in the future to understand the mechanistic basis.

-                Jagged ligands have the strongest cis-inhibitory effects and the receptors differ in their sensitivity to cis-inhibition by Dll ligands. These observations are in keeping with earlier in vivo and cell culture studies. More referencing of those would better place the work in context but it nicely supports and extends previous studies that were conducted in different ways.

-                Responses to most trans-activating ligands showed a degree of ultrasensitivity but this was not the case for cis-interactions where effects were more linear. This has implications for the way the two mechanisms operate and for how the signaling levels will be impacted by ligand expression levels.

-                Qualitatively similar results are obtained in a second cell line, suggesting they reflect fundamental properties of the ligands/receptors.

We appreciate the positive and constructive feedback.

Weaknesses:

One weakness is that the methods used to quantify the expression of ligands and receptors rely on the co-translation of tagged nuclear H2B proteins. These may not accurately capture surface levels/correctly modified transmembrane proteins. In general, the multiple conditions tested partly compensate for the concerns - for example, as Jag1 cells do activate Notch2 even if they do not activate Notch1 some Jag1 must be getting to the surface. But even with Notch2, Jag1 activities are on the lower side, making it important to clarify, especially given the different outcomes with the plated ligands. Similarly, is the fact that all ligands "signalled strongest to Notch2" an inherent property or due to differences in surface levels of Notch 2 compared to Notch1? The results would be considerably strengthened by calibration of the ligand/receptor levels (and ideally their sub-cellular localizations). Assessing the membrane protein levels would be relatively straightforward to perform on some of the basic conditions because their ligand constructs contain Flag tags, making it plausible to relate surface protein to H2B, and there are antibodies available for Notch1 and Notch2.

We agree that mCherry fluorescence does not provide a direct readout of active surface ligand levels. As the reviewer points out, the ability of Jag1 to activate Notch2 demonstrates that expressed Jag1 is competent for signaling. Further, in some cases, Jag1-Notch2 activation can be comparable to Dll1-Notch2 activation (Figure 2A). Following the reviewer’s suggestion, we performed a Western blot for multiple expression levels for each of three surface ligands (Dll1, Dll4, Jag1) (Figure 2—figure supplement 2). This blot revealed a signal for surface expression of Jag1. Interpretation is complicated by the expected dependence of the efficiency of surface protein purification on the number of primary amines in the protein, which varies among these ligands, and qualitatively correlates with the staining intensity. While this makes quantitative interpretation difficult, this result further supports the notion that Jag1 is present on the cell surface. Finally, we note that high signaling activity need not, in general, directly correlate with surface expression levels. In fact, one study showed an example in which increased ligand activity occurred with decreased basal ligand surface levels (Antfolk et al., 2017). While one would ideally like to know all parameters of the system, including surface protein levels, rates of recycling, etc. the perspective taken here is that the net effect of these many post-translational processing steps can be subsumed into the overall relationship between the expression of the protein (which, in our case, is read out by the co-translational reporter) and its activity, which is relevant for the behavior of developmental circuits, among other systems. To address this comment, we now explicitly mention the limitation of mCherry as a proxy for surface protein, and add a reference to previous work highlighting the relationship between surface levels and ligand activity.

In terms of the dependence of signaling on Notch levels, the metric of signaling activity used here is explicitly normalized by the mTurquoise co-translational reporter of Notch expression to account for differences in receptor expression across receiver clones. We have added a new figure to show the variation in expression (Figure 1—figure supplement 1A) and to demonstrate this normalization (Figure 1—figure supplement 5). Having said that, as the reviewer correctly points out, we cannot directly address the dependence on surface receptor levels with mTurquoise alone. To address this comment, we have added a figure that shows cotranslational and surface receptor expression for a subset of our receiver clones (Figure 1—figure supplement 1B). Although antibody binding strengths may vary, it appears unlikely that higher surface levels could explain most ligands’ preferential activation of Notch2 over Notch1, since Notch2 levels were lower than Notch1 levels in both surface expression and cotranslational expression.

Cis-activation as a mode of signaling has only emerged from these synthetic cell culture assays raising questions about its physiological relevance. Cis-activation is only seen at the higher ligand (Dll1, Dll4) levels, how physiological are the expression levels of the ligands/receptors in these assays? Is it likely that this would make a major contribution in vivo? Is it possible that the cells convert themselves into "signaling" and "receiving" sub-populations within the culture by post-translational mechanism? Again some analysis of the ligand/receptors in the cultures would be a valuable addition to show whether or not there are major heterogeneities.

The cis-activation results in this paper are, as the reviewer points out, conducted in synthetic cell culture assays. Cis-activation is observed across a large dynamic range of ligand expression, possibly including non-physiologically high levels. However, our previous work (Nandagopal et al, eLife 2019) showed that cis-activation does not require over-expression, as it occurred in unmodified Caco-2 and NMuMG cells with their endogenous ligand and receptor expression levels. As shown here in Figure 4B, cis-activation for Notch2 increases monotonically and is substantial even at intermediate ligand concentrations. In other cases, cis-activation is maximal at intermediate concentrations. We agree that the in vivo role remains unclear, and is difficult to determine due to the typical close contacts among cells in tissues. Therefore, these assays do not speak to in vivo relevance. Note that we can, however, rule out the possibility of trans signaling between well-mixed cell populations at these densities (Figure 4A).

It is hard to appreciate how much cell-to-cell variability in the "output" there is. For example, low "outputs" could arise from fewer cells becoming activated or from all cells being activated less. As presented, only the latter is considered. That may be already evident in their data, but not easy for the reader to distinguish from the way they are presented. For example, in many of the graphs, data have been processed through multiple steps of normalization. Some discussion/consideration of this point is needed.

We agree that in different experiments changes in a mean response can reflect changes in fraction of activated cells, or level of activation or some combination of both. In this work, most assays were conducted by flow cytometry, which provides a full distribution of cellular responses. We provided distributions for some experiments in the supplementary figures (i.e., Figure 4—figure supplement 1, and Figure 5—figure supplement 4). The sheer number of experiments and samples prevents us from displaying all underlying histograms. Therefore, we have provided all flow data sets in an extensive archive that is publicly available on data.caltech.edu (https://doi.org/10.22002/gjjkn-wrj28).

Impact:

Overall, cataloging the outcomes from the different ligand-receptor combinations, both in cis and trans, yields a valuable baseline for those investigating their functional roles in different contexts. There is still a long way to go before it will be possible to make a predictive model for outcomes based on expression levels, but this work gives an idea about the landscape and the complexities. This is especially important now that signaling relationships are frequently hypothesized based on single-cell transcriptomic data. The results presented here demonstrate that the relationships are not straightforward when multiple players are involved.

We appreciate this concise impact summary, and agree with its conclusions.

Reviewer #2 (Public Review):

Summary:

In this manuscript, the authors extend their previous studies on trans-activation, cis-inhibition (PMID: 25255098), and cis-activation (PMID: 30628888) of the Notch pathway. Here they create a large number of cell lines using CHO-K1 and C2C12 cells expressing either Notch1-Gal4 or Notch2-Gal4 receptors which express a fluorescent protein upon receptor activation (receiver cells). For cis-inhibition and cis-activation assays, these cells were engineered to express one of the four canonical Notch ligands (Dll1, Dll4, Jag1, Jag2) under tetracycline control. Some of the receiver cells were also transfected with a Lunatic fringe (Lfng) plasmid to produce cells with a range of Lfng expression levels. Sender cells expressing all of the canonical ligands were also produced. Cells were mixed in a variety of co-culture assays to highlight trans-activation, cis-activation, and cis-inhibition. All four ligands were able to trans-activate Notch1 and Notch 2, except Jag1 did not transactivate Notch1. Lfng enhanced trans-activation of both Notch receptors by Dll1 and Dll2, and inhibited Notch1 activation by Jag2 and Notch2 activation by both Jag 1 and Jag2. Cis-expression of all four ligands was predominantly inhibitory, but Dll1 and Dll4 showed strong cis-activation of Notch2. Interestingly, cis-ligands preferentially inhibited trans-activation by the same ligand, with varying effects on other trans-ligands.

Strengths:

This represents the most comprehensive and rigorous analysis of the effects of canonical ligands on cis- and trans-activation, and cis-inhibition, of Notch1 and Notch2 in the presence or absence of Lfng so far. Studying cis-inhibition and cis-activation is difficult in vivo due to the presence of multiple Notch ligands and receptors (and Fringes) that often occur in single cells. The methods described here are a step towards generating cells expressing more complex arrays of ligands, receptors, and Fringes to better mimic in vivo effects on Notch function.

In addition, the fact that their transactivation results with most ligands on Notch1 and 2 in the presence or absence of Lfng were largely consistent with previous publications provides confidence that the author's assays are working properly.

We appreciate the thoughtful comments and feedback.

Weaknesses:

It was unusual that the engineered CHO cells expressing Notch1-Gal4 were not activated at all by co-culture with Jag1-expressing CHO cells. Many previous reports have shown that Jag1 can activate Notch1 in co-culture assays, including when Notch1 was expressed in CHO cells. Interestingly, when the authors used Jag1-Fc in a plate coating assay, it did activate Notch1 and could be inhibited by the expression of Lfng.

In our assays, we do in fact also see some signaling of Jag1 to Notch1, especially when dLfng is coexpressed (Figure 2—figure supplement 4, formerly Figure 2—figure supplement 3). While these levels are lower than those observed for other ligand-receptor combinations, they are significantly elevated compared to baseline. In specific natural contexts, it will be important to determine whether the weak but non-zero Jag1-Notch1 signaling acts negatively to suppress signaling from other ligands, or provides weak but potentially functionally important levels of signaling. Evidence for both modes exists in the literature. To address this, we have expanded the discussion of Jag1-Notch1 signaling and added references to other work on Jag1-Notch1 signaling to the Discussion section.

The cell surface level of the ligands was determined by flow cytometry of a co-translated fluorescent protein. Some calibration of the actual cell surface levels with the fluorescent protein would strengthen the results.

This issue was also raised by Reviewers #1 and #3. Please see responses to Reviewer #1, above.

Reviewer #3 (Public Review):

Summary:

This manuscript reports a comprehensive analysis of Notch-Delta/Jagged signaling inclusive of the human Notch1 and Notch2 receptors and DLL1, DLL4, JAG1, and JAG2 ligands. Measurements

encompassed signaling activity for ligand trans-activation, cis-activation, cis-inhibition, and activity modulation by Lfng. The most striking observations of the study are that JAG1 has no detectable activity as a Notch1 ligand when presented on a cell (though it does have activity when immobilized on a surface), even though it is an effective cis-inhibitor of Notch1 signaling by other ligands, and that DLL1 and DLL4 exhibit cis-activating activity for Notch1 and especially for Notch2. Notwithstanding the artificiality of the system and some of its shortcomings, the results should nevertheless be a valuable resource for the Notch signaling community.

Strengths:

(1)  The work is systematic and comprehensive, addressing questions that are of importance to the community of researchers investigating mammalian Notch proteins, their activation by ligands, and the modulation of ligand activity by LFng.

(2)  A quantitative and thorough analysis of the data is presented.

Weaknesses:

(1) The manuscript is primarily descriptive and does not delve into the underlying, mechanistic origin or source of the different ligand activities.

We agree that the goals of this paper were largely to discover the range of signaling modes that occur. A mechanistic analysis would be beyond the scope of this work, but we agree it is an important next step.

(2) The amount of ligand or receptor expressed is inferred from the flow cytometry signal of a co-translated fluorescent protein-histone fusion, and is not directly measured. The work would be more compelling if the amount of ligand present on the cell surface were directly measured with anti-ligand antibodies, rather than inferred from measurements of the fluorescent protein-histone fusion.

This issue was also raised by Reviewers #1 and #2. Please see responses to Reviewer #1, above.

(3) It would be helpful to see plots of the raw activity data before transformation and normalization, because the plots present data after several processing steps, and it is not clear how the processed data relate to the original values determined in each measurement.

We included examples showing how raw data is processed in Figure 4—figure supplement 1 and Figure 5—figure supplement 4. The sheer number of experiments precludes including similar figures for all data sets. However, all raw and processed data and data analysis code is publicly available at (https://doi.org/10.22002/gjjkn-wrj28).

(4) The authors use sparse plating of engineered cells with parental (no ligand or receptor-expressing cell to measure cis activation). However, the cells divide within the cultured period of 22-24 h and can potentially trans-activate each other.

If measured cis-activation signal arises solely from trans-activation, then the measured cis-activation signal per cell should increase with cell density, since trans-activation per cell does depend on cell density (Figure 4A). However, for the strongest cis-activators (Dll1- and Dll4-Notch2), signaling magnitude is similar when these cells are cultured sparsely or at confluence, which would otherwise allow efficient trans signaling (Figure 5A). Thus, for Dll1- and Dll4-Notch2 receivers, total signaling strength per cell depends little or not at all on the opportunity to signal intercellularly. Moreover, cis-activation signal for the Dll1- and Dll4-Notch2 combinations exceeded the maximum trans-signaling levels we could achieve for the same receivers when cis-ligand was suppressed (Figure 4B). These results argue that cis interactions dominate signaling in this context. However, we have not ruled out the possibility that trans-signaling between sister cells after division contributes to the comparatively weak cis-activation observed for Notch1 receivers.

Reviewer #1 (Recommendations For The Authors):

As outlined in the public review, there is a question of whether the nuclear H2B accurately reflects the surface levels of the transmembrane proteins (ligand and receptor). Clearly, it would not be feasible to check levels in all of the experimental conditions, but some baseline conditions should be analyzed.

We addressed this above.

Reviewer #2 (Recommendations For The Authors):

(1)  As mentioned above, it was unusual that Jag1 did not activate Notch1 in co-culture assays, but did activate Notch1 in plate-coating assays. The authors should add some text to the Discussion to explain why they think this is happening in their engineered cells. One possibility is that the CHO cells express Manic fringe (Mfng) which is known to reduce Jag1-Notch1 activation. Data for Mfng levels in CHO cells were not included in Supplemental Table 2. Knocking down all three Fringes in CHO cells might increase Jag1-Notch1 activation.

This is already addressed in a sentence in the results: “Strikingly, while Jag1 sender cells failed to activate Notch1 receivers above background (Figure 2D), plate-bound Jag1-ext-Fc activated Notch1 only ~3-fold less efficiently than it activated Notch2 (Figure 3B-D). This suggests that the natural endocytic activation mechanism, or potential differences in tertiary structure between the expressed and recombinant Jag1 extracellular domains, could play roles in preventing Jag1-Notch1 signaling in coculture.” Regarding the point about Mfng, we added a note to Supplementary Table about other CHO-K1 expression data.

(2) Figure 1-supplemental figure 1: Both the Notch1-Jag1 and Notch1-Jag2 cells show high expression of Jag1 in low 4epi, but any higher concentration reduces to control levels. How much of a problem is this for interpreting your data?

This was not the ideal behavior, but by binning cells by co-translational reporters for ligand expression, we were able to obtain enough cells in intermediate bins. (Note: Figure 1—figure supplement 1 is now Figure 1—figure supplement 2.)

(3)  Figure 1C legend: Are these stably-expressing cells or Tet-off cells? Please state in legend.

The figure legend has been updated.

(4)  Figure 1E: How long is the knockdown of Rfng and Lfng effective? Does it affect the expression of Lfng later?

siRNA effects generally last for at least 72-96 hours, so we do not anticipate this being an issue.

(5) Page 9: "Lfng significantly decreased trans-activation of both receptors by Jag1 (>2.5-fold)". If there is no Jag1-Notch1 activation, how can Lfng decrease trans-activation?

We added a note in the main text to clarify that while Jag1-Notch1 signaling is relatively low, it can still be detectably decreased.

(6) Figure 4A legend: Please define what "2.5k ea senders and Rec" means. In the text, it says "To focus on cis-interactions alone, we then cultured receiver cells at low density, amid an excess of wildtype CHO-K1 cells" (page 14).

This was clarified in the text.

(7)  Page 14: "By contrast, Notch2 was cis-activated by both Dll1 and Dll4, to levels exceeding those produced by trans-activation by high-Dll1 senders (Figure 4B, lower left)." Where is the trans-activation data? 4B, lower right?

We updated this reference in the main text.

(8)  Page 16: "For Notch2-Dll1 and Notch2-Dll4, single cell reporter activities correlated with cis-ligand expression, regardless of whether cells were pre-induced at a high or low culture density (Figure 4D)." It appears that Notch2-Dll1 has lower Notch activation at sparse culture than confluent.

We agree that the level signaling is lower in sparse compared to confluent on average. This is explained by the sensitivity of the Tet-OFF promoter to culture density (Figure 4—figure supplement 2). However, the key point of this experiment is the positive correlation, which is consistent with cis-activation, and inconsistent with the pre-generation of NEXT hypothesis diagrammed in Figure 4C, which would not be expected to produce such a correlation.

(9a) For the creation of the C2C12-Nkd cells: Has genomic sequencing been done to confirm editing of Notch2 and Jag1 loci?

We confirmed the knockdown but did not do genomic sequencing.

(9b) The gel in Figure 7-Supplement 1C is not adequate for showing loss of Jag1. It should be repeated.

In this case, we have only the single gel. We added a note in figure legend that no duplicate was performed.

(10) Figure 7A: Which Fringes are expressed in C2C12 cells? You should provide a rationale for knocking down just Rfng.

Figure 7—figure supplement 1A shows the levels of expression in C2C12. Note that Mfng is not highlighted because its levels were undetectable.

(11) Figure 7-Supplement 1D: This is confusing. Notch2 levels are not reduced in the left panel, and Notch1 and Notch2 levels are not reduced in the right panel?

C2C12-Nkd cells exhibit reduced levels of Notch1 and Notch3. This can be seen in Figure 7—figure supplement 1A. Panel D presents the results of additional siRNA knockdown, performed to prevent subsequent up-regulation of Notch1 and Notch3 during the assay. These knockdown results were variable, as shown. The Notch2 siRNA knockdown was not essential for these experiments, but performed despite very low levels of Notch2 to begin with. In the revision, we have added this note to the Methods.

Reviewer #3 (Recommendations For The Authors):

(1) The results section of the manuscript is very dense and difficult to follow, as are the figure legends.

We appreciate the criticism, and regret that it is not easier to read in its current form.

(2) The authors could emphasize areas of concordance with published results (where available) to place their artificial, engineered system into a better biological context. Are there any examples of studies in whole organisms where cis-activation plays a role?

We are not aware of examples of cis-activation in whole organisms at this point.

(3) How do the authors rationalize the different responses of Notch1 to cell-presented Jag1 as opposed to immobilized Jag1, where its signal strength is second in rank order on a molar basis?

This comment was addressed above in response to the first recommendation from Reviewer #2.

It is also difficult to understand Figure 2_—_figure Supplement 3B, in which it appears that Jag1 induces a Notch1 reporter response when LFng is knocked down (dLfng), and how those data relate to the inactive response to Jag1 shown in the main figures.

The issue here is a difference of normalization. Figure 2A in the main text is normalized to the sender expression level, i.e. relative signaling strength. By contrast, Figure 2—figure supplement 4B (previously Figure 2—figure supplement 3B) shows absolute signaling activity, which can appear higher because it does not normalize for ligand expression. For Jag1-Notch1 signaling in particular, substantial signaling required very high levels of Jag1. We have added a new figure to demonstrate these two types of normalization (Figure 2—figure supplement 1A).

See the Authr response image 1 below for a direct comparison of these two normalization modes using data from both Figure 2A and Figure 2—figure supplement 4B. Note how the Jag1-Notch1 signaling activities that are nonzero in the top plot go to zero in the bottom plot as a result of normalizing the values to ligand expression.

Author response image 1.

Comparison of normalization modes in Figure 2A and Figure 2—figure supplement 4B (formerly 3B).

Normalized trans-activation signaling activities for different ligand-receptor combinations (with dLfng only), either with further normalization to ligand expression (bottom row) or without further normalization (top row). Normalized signaling activity is defined as reporter activity (mCitrine, A.U.) divided by cotranslational receptor expression (mTurq2, A.U.), normalized to the strongest biological replicate-averaged signaling activity across all ligand-receptor-Lfng combinations in this experiment. Saturated data points, defined here as those with normalized signaling activity over 0.75 in both dLfng and Lfng conditions, were excluded. Colors indicate the identity of the trans-ligand expressed by cocultured sender cells. Error bars denote bootstrapped 95% confidence intervals (Methods), in this case sampled from the number of biological replicates given in the legend—n1 (for Notch1) or n2 (for Notch2). See Methods and Figure 2A caption for more details. Note that the only difference between this figure and the new Figure 2—figure supplement 1A is that this figure additionally includes the Jag1-high data from Figure 2—figure supplement 4B.

script>

eLife Assessment

This valuable study significantly enhances our understanding of how various ligands and receptors interact within the Notch signaling pathway. By developing novel cell-based assay systems, the authors systematically analyzed the effects of different ligand-receptor combinations on pathway activation. The convincing data reveal intriguing and unexpected differences and provide a foundation for interpreting Notch signalling in both normal and disease-related contexts.

Reviewer #1 (Public review):

Summary:

The Notch signaling pathway plays important roles in many developmental and disease processes. Although well-studied there remain many puzzling aspects. One is the fact that as well as activating the receptor through a trans-activation, the transmembrane ligands can interact with receptors present in the same cell. These cis-interactions are usually inhibitory, but in some cases, as in the assays used here, they may also be activating. With a total of 6 ligands and 4 receptor there are potentially a wide array of possible outcomes when different combinations are co-expressed in vivo. Here the authors set out to make a systematic analysis of the qualitative and quantitative differences in the signaling output from different receptor ligand combinations, generating sets of "signaling" (ligand expressing) and "receiving" (receptor +/- ligand expressing cells).

The readout of pathway activity is transcriptional, relying on the fusion of GAL4 in the intracellular part of the receptor. Positive ligand interactions result in proteolytic release of Gal4 that turns on expression of H2B-citrine. As an indicator of ligand and receptor expression levels, they are linked via TA to H2B mCherry and H2B mTurq expression respectively. The authors also manipulate expression of the glycosyltransferase Lunatic-Fringe (LFng) that modifies the EGF repeats in the extracellular domains impacting on their interactions. The testing of multiple ligand receptor combinations at varying expression levels is a tour de force, with over 50 stable cell lines generated, and yields valuable insights although as a whole, the results are quite complex.

Strengths:

Taking a reductionist approach to test systematically differences in the signaling strength, binding strength and cis-interactions from the different ligands in the context of the Notch1 and Notch 2 receptors (they justify well they choice of players to test via this approach) produces a baseline understanding of the different properties and leads to some unexpected and interesting findings. Notably:<br /> - Jag1 ligand expressing cells failed to activate Notch1 receptor although were capable of activating Notch2. Conversely, Jag2 cells elicited the strongest activation of both receptors. The results with Jag1 are surprising also because it exhibits some of the strongest binding to plate bound ligands. The failure to activate Notch1 has major functional significance and it will be important in future to understanding the mechanistic basis.<br /> - Jagged ligands have the strongest ciis-inhibitory effects and the receptors differ in their sensitivity to cis-inhibition by Dll ligands. These observations are in keeping with earlier in vivo and cell culture studies. More referencing of those would better place the work in context but it nicely supports and extends previous studies that were conducted in different ways.<br /> - Responses to most trans-activating ligands showed a degree of ultrasensitivity but this was not the case for cis-interactions where effects were more linear. This has implications for the way the two mechanisms operate and for how the signaling levels will be impacted by ligand expression levels.<br /> - Qualitatively similar results are obtained in a second cell line, suggesting they reflect fundamental properties of the ligands/receptors.

Weaknesses:

One weakness is that the methods used to quantify the expression of ligands and receptors rely on co-translation of tagged nuclear H2B proteins. These may not accurately capture surface levels/correctly modified transmembrane proteins. In general, the multiple conditions tested partly compensate for the concerns - for example as Jag1 cells do activate Notch2 even if they do not activate Notch1 some Jag1 must be getting to the surface. But even with Notch2, Jag1 activities are on the lower side, making it important to clarify, especially given the different outcomes with the plated ligands. Similarly, is the fact that all ligands "signalled strongest to Notch2" an inherent property or due to differences in surface levels Notch 2 compared to Notch1?.. The results would be considerably strengthened by calibration of the ligand/receptor levels (and ideally their sub-cellular localizations). Assessing the membrane protein levels would be relatively straightforward to perform on som eof the basic conditions because their ligand constructs contain Flag tags, making it plausible to relate surface protein to H2B, and there are antibodies available for Notch1 and Notch2

In the revised version this has been addressed to some extent. A figure showing the relationship between co-translated mTurquiose and surface receptor expression for some clones (Figure 1-figure supplement 1B) goes some way to address the concerns that differences in Notch1 and Notch 2 could be due to the receptor levels. The data analyzing surface ligand levels is more equivocal, (a Western blot for biotinylated surface proteins), as the levels detected vary substantially between Dll1 and Dll4 (the latter barely detectable). But as a signal for surface expression of Jag1 was obtained this rules-out one concern that this ligand was failing to reach the surface. A discussion of the caveats of the approach is warranted, to make clear the limitations.

Cis-activation as a mode of signaling has only emerged from these synthetic cell culture assays raising questions about its physiological relevance. Cis-activation is only seen at the higher ligand (Dll1, Dll4) levels, how physiological are the expression levels of the ligands/receptors in these assays? Is it likely that this would make a major contribution in vivo? Is it possible that the cells convert themselves into "signaling" and "receiving" sub-populations within the culture by post-translational mechanism. Again some analysis of the ligand/receptors in the cultures would be a valuable addition to show whether or not there are major heterogeneities.

It is hard to appreciate how much cell to cell variability in the "output" there is. For example, low "outputs" could arise from fewer cells becoming activated or from all cells being activated less. As presented, only the latter is considered. That maybe already evident in their data, but not easy for the reader to distinguish from the way they are presented. For example, in many of the graphs, data have been processed through multiple steps of normalization. Some discussion/consideration this point is needed.

Impact:<br /> Overall, cataloguing of the outcomes from the different ligand-receptor combinations, both in cis and trans, yields a valuable baseline for those investigating their functional roles in different contexts. There is still a long way to go before it will be possible to make a predictive model for outcomes based on expression levels, but this work gives an idea about the landscape and the complexities. This is especially important now that signaling relationships are frequently hypothesised based on single cell transcriptomic data. The results presented here demonstrate that the relationships are not straightforward when multiple players are involved.

Reviewer #2 (Public review):

Summary:

In this manuscript the authors extend their previous studies on trans-activation, cis-inhibition (PMID: 25255098) and cis-activation (PMID: 30628888) of the Notch pathway. Here they create a large number of cell lines using CHO-K1 and C2C12 cells expressing either Notch1-Gal4 or Notch2-Gal4 receptors which express a fluorescent protein upon receptor activation (receiver cells). For cis-inhibition and cis-activation assays, these cells were engineered to express one of the four canonical Notch ligands (Dll1, Dll4, Jag1, Jag2) under tetracycline control. Some of the receiver cells were also transfected with a Lunatic fringe (Lfng) plasmid to produce cells with a range of Lfng expression levels. Sender cells expressing all of the canonical ligands were also produced. Cells were mixed in a variety of co-culture assays to highlight trans-activation, cis-activation, and cis-inhibition. All four ligands were able to trans-activate Notch1 and Notch 2, although Jag1 transactivated Notch1 weakly. Lfng enhanced trans-activation of both Notch receptors by Dll1 and Dll4, and inhibited both receptors by Jag 1 and Jag2. Cis-expression of all four ligands were predominantly inhibitory, but Dll1 and Dll4 showed strong cis-activation of Notch2. Interestingly, cis-ligands preferentially inhibited trans-activation by the same ligand, with varying effects on other trans-ligands.

Strengths:

This represents the most comprehensive and rigorous analysis of the effects of canonical ligands on cis- and trans-activation, and cis-inhibition, of Notch1 and Notch2 in the presence or absence of Lfng so far. Studying cis-inhibition and cis-activation is difficult in vivo due to the presence of multiple Notch ligands and receptors (and Fringes) that often occur in single cells. The methods described here are a step towards generating cells expressing more complex arrays of ligands, receptors and Fringes to better mimic in vivo effects on Notch function.

In addition, the fact that their transactivation results with most ligands on Notch1 and 2 in the presence or absence of Lfng were largely consistent with previous publications provides confidence that the author's assays are working properly.

Weaknesses:

In the original version, there was a major concern about quantifying the amount of Notch receptors and ligands on the cell surface (especially Jag1) based on total fluorescence. The authors have added data to demonstrate that most of the receptors and ligands are on the cell surface, allaying most of these concerns.

eLife Assessment

This important study uses in vitro and in vivo methods to identify HpARI proteins from H. polygyrus as modulators of the host immune system. The data from comprehensive approaches for investigating differential roles of HpARI proteins are convincing. This paper is relevant to those who investigate host-pathogen interactions at the systems and molecular levels.

Reviewer #1 (Public Review):

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.

Reviewer #2 (Public Review):

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.

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).

eLife Assessment

This is an important study on the damage-induced checkpoint maintenance and termination in budding yeast that provides novel and convincing evidence for a role of the spindle assembly checkpoint and mitotic exit network in halting the cell cycle after prolonged arrest in response to irreparable DNA double strand breaks (DSBs). The study identifies particular components from these checkpoints that are specifically required for the establishment and/or the maintenance of a cell cycle block triggered by such DSBs. The authors propose an interesting model for how these different checkpoints intersect and crosstalk for timely resumption of cell cycling even without repairing DNA damage that has been revised by addressing the bulk of the reviewers' comments to the first version of the manuscript.

Reviewer #1 (Public review):

Summary:

In their manuscript, Zhou et al. analyze the factors controlling the activation and maintenance of a sustained cell cycle block in response to persistent DNA DSBs. By conditionally depleting components of the DDC using auxin-inducible degrons, the authors verified that some of them are only required for the activation (e.g., Dun1) or the maintenance (e.g., Chk1) of the DSB-dependent cell cycle arrest, while others such as Ddc2, Rad24, Rad9 or Rad53 are required for both processes. Notably, they further show that after a prolonged arrest (>24 h) in a strain carrying two DSBs, the DDC becomes dispensable and the mitotic block is then maintained by SAC proteins such as Mad1, Mad2 or the mitotic exit network (MEN) component Bub2.

Strengths:

The manuscript dissects the specific role of different components of the DDC and the SAC during the induction of a cell cycle arrest induced by DNA damage, as well as their contribution for the short-term and long-term maintenance of a DNA DSB-induced mitotic block. Overall, the experiments are well described and properly executed, and the data in the manuscript are clearly presented. The conclusions drawn are generally well supported by the experimental data. Their observations contribute to drawing a clearer picture of the relative contribution of these factors to the maintenance of genome stability in cells exposed to permanent DNA damage.

Weaknesses:

The main weakness of the study is that it is fundamentally based on the use of the auxin-inducible degron (AID) strategy to deplete proteins. This widely used method allows an efficient depletion of proteins in the cell. However, the drawback is that a tag is added to the protein, which can affect the functionality of the targeted protein or modify its capacity to interact with others. In fact, three of the proteins that are depleted using the AID systems are shown to be clearly hypomorphic, and hence their capacity to induce a strong checkpoint response might be compromised. A corroboration of at least some of the results using an alternative manner to eliminate the proteins would help to strengthen the conclusions of the manuscript.

Reviewer #2 (Public review):

Summary:

The manuscript analyzes and attempts to discriminate genetic requirements for DNA damage-induced cell cycle checkpoint induction, maintenance, and adaptation in budding yeast bearing one or two unrepairable DNA double strand breaks using auxin-induced degradation (AID) of key DNA damage response (DDR) factors. The study paid particular attention to solving a puzzle regarding how yeasts bearing two unrepaired DNA breaks fail to engage in "adaptation" whereas those with a single unrepairable break eventually resume cell cycling after a prolonged (up to 12 h) G2 arrest.

The key findings are: 1. Genetic requirements for the entry and the maintenance of DDC are separable. For instance, Dun1 is partially required for the entry but not the DDC maintenance whereas Chk1 is only required for maintenance. 2. Cells with two unrepairable breaks respond to DDR only up to a certain time (~12-15 h post damage) and beyond this point, depend on spindle assembly checkpoint (SAC) and mitotic exit network (MEN) to halt cell cycling. 3. The authors also propose an interesting concept that the location of DNA breaks and their distance to centromeres are important factors dictating the effect of SAC/MEN on the duration of cell cycle arrest after prolonged arrest (and cells become "deaf" to persistent arrest signals) and yeast's adaptability following DNA damage. The results provide most compelling evidence to date on the role of SAC/MEN in DNA damage response and cell cycle arrest albeit its impact might be limited to the handful of model systems due to the vastly different centromeric elements and far larger chromosome sizes in metazoan cells. The study albeit briefly discussed the basis of transitions from entry, maintenance, and adaptation ( ex. changes in centromeric architectures), it does not offer detailed explanations or a testable hypothesis to this topic.

Overall, the conclusion of the study is well supported by the elegant set of genetic experimental data and employed multiple readouts on DDC factor depletion on checkpoint integrity and cell cycle status. Although the study simply measures Rad53 phosphorylation as the primary metric to assess checkpoint status, it successfully demonstrated how the signaling is modified through the different stages and that eventually cells become recalcitrant to DDC signaling after a prolonged arrest. The results are clear, and rigorously tested and carefully interpreted with good discussion on the possible limitations. The revision provided detailed responses to the reviewers' comments and addressed a few key concerns, one of which is universally raised by the reviewers on the full functionality of AID tagged DDC factors, by simply expressing excess Rad9-AID to restore more normal looking checkpoint response. It will be interesting if the excess expression of other DDC factors could overcome suboptimal checkpoints in cells after 24 h post damage.

Reviewer #3 (Public review):

Summary:

The DNA damage checkpoint (DDC) inhibits the metaphase-anaphase transition to repair various types of DNA damage, including DNA double strand breaks (DSBs). One irreparable DSB can maintain the DDC for 12-15 hours in yeast, after which the cells resume the cell cycle. If there are two DSBs, the DDC is maintained for at least 24 hours. In this study, the authors take advantage of this tighter DDC to investigate whether the best-known proteins involved in establishing the DDC are also responsible for its long-term maintenance during irreparable DSBs. They do this by cleverly degrading such proteins after DSB formation. They show that most, but not all, DDC proteins maintain the cell cycle block. Interestingly, DDC proteins become dispensable after 15 hours and the block is then maintained by spindle assembly checkpoint (SAC) proteins.

Strengths:

The authors have engineered a tight yeast system to study DDC shutdown after irreparable DSBs and used it to address whether checkpoint proteins (DDC and SAC) contribute to the long-term maintenance of DSB-mediated G2/M block. The different roles of Ddc2, Chk1 and Dun1 are interesting, while the fact that SAC overtakes DDC after 15 hours is intriguing and highlights how DSBs near and far from centromeres can have a profound impact on cell adaptation to DSBs. In their revision, the authors have now improved the Rad9-AID methodology to place Rad9 in the context of DDC adaptation, as well as widening the association between adaptation and proximity to centromeres.

Weaknesses:

Some of the results they present essentially confirm their own previous findings, albeit with a tighter strain design for long-term arrest. Conclusions about the maintenance of G2/M in several mutant combinations could have been strengthened by adding simple microscopy experiments with DAPI staining. No clear mechanism for how depletion of Bub2, but not Bfa1, can relieve the G2/M (metaphase) block is given.

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.

eLife Assessment

This important study examines the role of Microrchidia (MORC) proteins in the human malaria parasite Plasmodium falciparum. Solid experimental results, including genome editing and chromatin profiling methods (ChIP-seq and Hi-C), provide a comprehensive picture of the critical role MORC plays in shaping parasite chromatin. Depletion of MORC results in a lethal collapse of heterochromatin and parasite death, nominating the factor as a new target of antimalarial therapies.

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.

Weaknesses:

Although disruption of PfMORC affects chromatin architecture and stage-specific gene expression, determining a direct cause-effect relationship requires further investigation. 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.

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.

eLife Assessment

This valuable study focuses on the regulation of Notch signaling during the immune response in Drosophila. The authors provide solid evidence in support of roles for Su(H) and Pkc53E-induced phosphorylation in Drosophila immunity. The work will be of interest to colleagues in immunity and receptor signaling.

Reviewer #1 (Public review):

The authors previously showed in cell culture that Su(H), the transcription factor mediating Notch pathway activity in Drosophila, 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 downregulation of Notch signaling is important for the production of 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.<br /> Consistent with their hypothesis, the authors now 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 and their epistatic analyses indicate that Pkc53E acts upstream of Su(H). Finally, they show that Pkc53E mutants aslo display a compromised immune response to wasp parasitism.

Strengths

The manuscript is well presented and the experiments are sound, with a good combination of genetic and biochemical approaches and several clear phenotypes backing the main conclusions. Notably Su(H)S269A mutation strongly reduces lamellocyte production. Moreover, the epistatic data are convincing, notably concerning the relationship between Notch/Su(H) and Pkc53E for crystal cell production.<br /> Even though it is not fully established, the overall model is credible and interesting. In addition, it opens further avenues of research to study the activation of Pkc in response to an immune challenge.

Weaknesses

Apparently, the hypothesis that Pkc53E is required for Su(H) phosphorylation in vivo could not be directly tested due to the lack of an appropriate tool (the specificity and sensitivity of the current anti-pS269 antibody was insufficient).<br /> Also, the poor immune response of Pkc53E mutant might rather be linked to their constitutively reduced circulating blood cell number than to a deficit in Notch/Su(H) down-regulation following wasp infestation.

Reviewer #2 (Public review):

The current draft by Deischel et.al., describes the role of Pkc53E in the phosphorylation of Su(H) to down regulate 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. The authors have proved the central idea linking phosphorylation of Su(H) via Pkc53E to implying its modulation of Notch activity to mount a robust immune response is now well addressed in its entirety and I find the paper indeed very interesting.

Comments on revised version:

The authors have addressed all pending concerns and I have no further comments. I indeed complement the authors for their wonderful piece of work.

Reviewer #3 (Public review):

Diechsel et al. provide important and valuable insights into how Notch signaling is shut down in response to parasitic wasp infestation in order to suppress crystal cell fate and favor 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 remodeling hematopoiesis upon wasp challenge. I find the study interesting, and the data presented supports the overall conclusions made by the authors. The authors have addressed all my comments satisfactorily in the revised submission.

Strengths:

The manuscript is well presented, and the conclusions made are backed by genetic, biochemical and molecular biology-based approaches. Overall, the authors convincingly demonstrate how Pkc53E mediated phosphorylated of Su(H) shuts down Notch signaling during wasp infestation in Drosophila.

Weaknesses:

The exact molecular trigger for activation of Pkc53E is still uncharacterized and it would be interesting to know how Pkc53E gets activated during wasp infestation and whether Pkc53E gets activated turning down Notch in other stress induced scenarios.

The authors have addressed comments satisfactorily. Overall, I think the findings are interesting and would be useful to the field of developmental biology and immunology and address an important gap in the field. The most significant conclusion from the work is how Notch acts as a molecular switch during parasitic wasp infestation.

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.

eLife Assessment

This important study provides a new perspective on how human immunity shapes the antigenic evolution of pathogens. By combining theory and simulation the authors make a compelling case for the importance of eco-evolutionary interactions in population-level virus-host dynamics, which arise due to coupling between the dynamics of immune memories and viral variants. Although the work does not propose improved data-driven viral forecasting methods, it makes a conceptual contribution that advances the field's understanding of this problem's intrinsic difficulty.

Reviewer #1 (Public review):

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.

In the revised version, the authors have addressed questions on the role of clonal interference by new simulations in the SI, clarified the connection between the SIR model and vanishing-fitness models, and placed their analysis into the broader context of consumer resource dynamics.

However, the general conclusion, as stated in the abstract, that variant trajectories become unpredictable as a consequence of the SIR dynamics remains somewhat misleading. Two aspects contribute to this problem. (1) The empirical observation of ``quasi-neutrality', i.e. the absence of a net frequency increase inferred as an average of many trajectories at intermediate frequencies, does not imply that individual trajectories are neutral (i.e., fully stochastic and unpredictable) over the time span of observation. Rather, it just says that some have a positive and some have a negative selection coefficient over that time span. (2) As stated by the authors, the observation of average quasi-neutrality is indeed incompatible with the travelling wave model, where initially successful new variants are assumed to retain a fixed, positive selection coefficient from origination to fixation. This observation also limits predictions by extrapolation, where a positive selection coefficient inferred at small frequency is assumed to remain the same at later times and higher frequencies. However, predictions derived from Gog and Grenfell's multi-strain SIR model, as used by several authors, do not make the assumption of fixed selection coefficients and incorporate trajectory-specific, time-dependent expiration effects into their model predictions. This distinction remains blurred throughout the text of the paper.

Reviewer #3 (Public review):

In this work the authors present a multi-strain SIR model in which viruses circulate in a heterogeneous population with different groups characterized by different cross-immunity structures. They reformulate the qualitative features of these SIR dynamics 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

The idea that 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.<br /> This general framework has the potential to be more universal than human RNA viruses, in situations where invading mutants would saturate at intermediate frequencies.

The qualitative connection between the coarse-grained features of these vanishing fitness dynamics and structured SIR processes offers additional intuition relevant to host-pathogens interactions, although as noted by the authors other ecological processes could drive similar evolutionary patterns. The additions in the revised manuscript, substantiating more thoroughly the connection between the SIR and the vanishing fitness description, are important to better appreciate the scope of the work.

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.

Recommendations For The Authors:

Reviewer #1:

●      It might help the reader if you make it explicit that mDES allows you to create an approximate amalgam of different kinds of experiences by assuming that, across individuals, there is a general consensus of experiences at particular points in the movie. Whether this assumption is an accurate reflection of the way in which each individual's brain is an important, testable prediction that could be discussed/examined in different projects. For instance, in other projects there are clear idiosyncratic responses to the same naturalistic stimuli: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8064646/.

Thank you, this is an excellent point. We have included this article in our revision and expanded on the introduction to emphasize how this study relates to our work. Additionally, we have included an additional figure that helps illustrate how mDES can be used to evaluate the idiosyncrasy for each respective thought component to visually display the variance across moments in the film:

Page 6-7 [137-148] In our study, we used multi-dimensional experience sampling (mDES) to describe ongoing thought patterns during the movie-watching experience [8]. mDES is an experience sampling method that identifies different features of thought by probing participants about multiple dimensions of their experiences. mDES can provide a description of a person’s thoughts, generating reliable thought patterns across laboratory cognitive tasks [22, 32, 33] and in daily life [34, 35], and is sensitive to accompanying changes in brain activity [24, 36]. Studies that use mDES to describe experience ask participants to provide experiential reports by answering a set of questions about different features of their thought on a continuous scale from 1 (Not at all) to 10 (Completely) [24, 32-41]. Each question describes a different feature of experience such as if their thoughts are oriented in the future or the past, about oneself or other people, deliberate or intrusive in nature, and more (See methods for a full list of questions used in the current study).

●      A cartoon describing the mDES technique could be helpful for uninitiated readers.

Thank you for your suggestion, we have added an additional figure (Figure 3) that illustrates the process of mDES in the laboratory during this experiment, clarifying that participants answer mDES items using a slider to indicate their score (rather than expressing it verbally).

●      Did the authors check for any measures of reliability across mDES estimates other than split-half reliability? For instance, the authors could demonstrate construct validity by showing that engagement with certain features of the thought-sampling space aligned with specific points in the movies. If so, the start of the Results section would be a great place to demonstrate the reliability of the approach. For instance, did any two participants sample the same 15-second window of time in a particular stimulus? If so, you could compare their experience samples to determine whether the method was extensible across subjects.

This is a great point, thank you very much for highlighting this. We have eight individuals at each time point in our analysis, which is probably not enough to calculate meaningful reliability measures. However, we have added a time series analysis of experience in each clip to our revision (Figure 3). In these time plots, it is possible to see clear moments in the film in which scores do not straddle 0 (using 95% CI), and often, these persist across successive moments (Figure 3; see time-series plot four for the clearest example).  When the confidence intervals of a sampling epoch do not overlap with zero, this suggests a high degree of agreement in thought content across participants. At the same time, our analysis shows that individual differences do exist since the relative presence of each component for each participant was linked to objective measures of movie watching (in this case, comprehension). In this revision we have specifically addressed this question by conducting ANOVAs to determine how scores on each component across the clip (See also supplementary table 11). This additional analysis shows that mDES effectively captures shared aspects of movie-watching and is also sensitive to individual variation (since it can describe individual differences).

Page 15 [304-323]: Next, we examined how each pattern of thought changes across each movie clip. For this analysis, we conducted separate ANOVA for each film clip for the four components (see Table 1 and Figure 3). Clear dynamic changes were observed in several components for different films. We analyzed these data using an Analysis of Variance (ANOVA) in which the time in each clip were explanatory variables of interest. This identified significant change in “Episodic Social Cognition” scores across Little Miss Sunshine, F(1, 712) = 10.80, p = .001, , η2 = .03, and Citizenfour, F(1, 712) = 5.23, p = .023, , η2 = .02. There were also significant change in “Verbal Detail” scores across Little Miss Sunshine, F(1, 712) = 31.79, p <.001, η2 = .09. Lastly, there were significant changes in “Sensory Engagement” scores for both Citizenfour, F(1, 712) = 6.22, p = .013, η2 = .02, and 500 Days of Summer, F(1, 706) = 80.41, p <.001, η2 = .18. These time series are plotted in Figure 3 and highlight how mDES can capture the dynamics of different types of experience across the three movie clips. Moreover, in several of these time series plots, it is clear that thought patterns reported extend beyond adjacent time periods (e.g. scores above zero between time periods 150 to 400 for Sensory Engagement in 500 days of Summer and for time periods between 175 and 225 for Verbal Detail in Little Miss Sunshine). It is important to note that no participant completed experience sampling reports during adjacent sampling points (see Supplementary Figure 7), so the length of these intervals indicates agreement in how specific scenes within a film were experienced and conserved across different individuals. Notably, the component with the least evidence for temporal dynamics was “Intrusive Distraction.”

●      P10: "Generation of the thought-space" - how stable are these word clouds to individual subjects? If there are subject-specific differences, are there ways to account for this with some form of normalization?

Thank you for bringing up this point. Our current goal was to show how the average experience of one group of participants relates to the brain activity of a second group. In this regard it is important to seek the patterns of similarity across individuals in how they experience the film. However, as is normal in our studies using mDES, we can also use the variation from the mean to predict other cognitive measures and, in this way, account for the variability that individuals have in their movie-watching experience. In other words, the word clouds reflect the mean of a particular dimension, so when an individual score is close to 0, their thought content does not align with this dimension -- however, deviating scores, positive or negative, indicating that this dimension provides meaningful information about the individual's experience. Evidence of the meaningful nature of this variation can be seen in the links between the reported thoughts and the individuals’ comprehension (e.g. individuals whose thoughts do not contain strong evidence of “Intrusive Distraction”, or in other words, a negative score, tended to do better on comprehension tests of information in the movies they watched).

●      P11: "Variation in thought patterns" - can the authors use a null model here to demonstrate that the associations they've observed would occur above chance levels (e.g., for a comparison of time series with similar temporal autocorrelation but non-preserved semantic structure)? Further, were there any pre-defined hypotheses over whether any of the three different movies would engage any of the 4 observed dimensions?

This is a great point. We chose to sample from three distinctly different films to help us understand if mDES was sensitive to different semantic and affective features of films. Our analysis, therefore, shows that at a broad level, mDES is able to discriminate between films, highlighting its broad sensitivity to variation in semantic or affective content. Armed with this knowledge, researchers in the future could derive mechanistic insights into how the semantic features may influence the mDES data. For example, future studies could ask participants to watch movies in a scrambled order to understand how varying the structure of semantics or information breaks the mapping between brains and ongoing experience. In this revision we have amended the text to reflect this possibility:

Page 34 [674-679]. Our analysis shows that mDES is able to discriminate between films, highlighting its broad sensitivity to variation in semantic or affective content. Armed with this knowledge, we propose that in the future, researchers could derive mechanistic insights into how the semantic features may influence the mDES data. For example, it may be possible to ask participants to watch movies in a scrambled order to understand how the structure of semantic or information influences the mapping between brains and ongoing experience as measured by mDES.

●      P14: "Brain - Thought Mappings: Voxel-space Analysis" - this is a cool analysis, and a nice validation of the authors' approach. I would personally love to see some form of reliability analysis on these approaches - e.g., do the same locations in the cerebral cortex align with the four features in all three movies? Across subjects?

This is another great point, and we thank you for your enthusiasm. The data we have has only sampled mDES during a relatively short period of brain activity which we suspect would make an individual-by-individual analysis underpowered. In the future, however, it may be possible to adopt a precision mapping approach in which we sample mDES during longer periods of movie watching and identify how group-level mappings of experience relate to brain activity within a single subject. To reflect this possibility, we have amended the text in this revision in the following way:

Page 34-35 [672-687]: In addition, our study is correlational in nature, and in the future, it could be useful to generate a more mechanistic understanding of how brain activity maps onto the participants' experience. Our analysis shows that mDES is able to discriminate between films, highlighting its broad sensitivity to variation in semantic or affective content. Armed with this knowledge, we propose that in the future, researchers could derive mechanistic insights into how the semantic features may influence the mDES data. For example, it may be possible to ask participants to watch movies in a scrambled order to understand how the structure of semantic or information influences the mapping between brains and ongoing experience as measured by mDES. Finally, our study focused on mapping group-level patterns of experience onto group-level descriptions of brain activity. In the future, it may be possible to adopt a “precision-mapping” approach by measuring longer periods of experience using mDES and determining how the neural correlates of experience vary across individuals who watched the same movies while brain activity was collected [1]. In the future, we anticipate that the ease with which our method can be applied to different groups of individuals and different types of media will make it possible to build a more comprehensive and culturally inclusive understanding of the links between brain activity and movie-watching experience

Reviewer #2:

(1) The three-dimensional scatter plot in Figure 2 does not represent "Intrusive Distraction." Would it make sense to color-code dots by this important dimension?

Thank you for this suggestion. Although it could be possible to indicate the location of each film in all four dimensions, we were worried that this would make the already complex 3-D space confusing to a naive reader. In this case, we prefer to provide this information in the form of bar graphs, as we did in the previous submission.

(2) The coloring of neural activation patterns in Figure 3 is not distinct enough between the different dimensions of thought. Please reconsider color intensities or coding. The same applies to the left panel in Figure 4.

Thanks for this comment; we found it quite difficult to find a colour mapping that allows us to show the distinction between four states in a simple manner, yet we believe it is valuable to show all of the results on a similar brain. Nonetheless, to provide a more fine-grained viewing of our results in this revision we have provided a supplementary figure (Supplementary Figure 6) that shows each of the observed patterns of activity in isolation.

(3) The new method (mDES) is mentioned too often without explanation, making it hard to follow without referring to the methods section. It would be helpful to state prominently that participants rated their thoughts on different dimensions instead of verbalizing them.

Thank you for this point, we have adjusted the Introduction to clarify and expand on the mDES method. We have also included an example of the mDES method in an additional figure that we have now included to visually express how participants respond to mDES probes (Figure 3).

Page 6-7 [136-148]: In our study, we used multi-dimensional experience sampling (mDES) to describe ongoing thought patterns during the movie-watching experience [2]. mDES is an experience sampling method that identifies different features of thought by probing participants about multiple dimensions of their experiences. mDES can provide a description of a person’s thoughts, generating reliable thought patterns across laboratory cognitive tasks [3-5] and in daily life [6, 7], and is sensitive to accompanying changes in brain activity when reports are gained during scanning [8, 9]. Studies that use mDES to describe experience ask participants to provide experiential reports by answering a set of questions about different features of their thought on a continuous scale from 1 (Not at all) to 10 (Completely) [3, 5-14]. Each question describes a different feature of experience, such as if their thoughts are oriented in the future or the past, about oneself or other people, deliberate or intrusive in nature, and more (See Methods for a full list of questions used in the current study).

Author response image 1.

(4) Reporting of single-movie thought patterns seems quite extensive. Could this be condensed in the main text?

Thank you for this point, upon re-visiting the manuscript, we have adjusted the text to be more concise.

Reviewer #3:

●      This is a very elegant experiment and seems like a very promising approach. The text is currently hard to read.

Thank you for this point, we have since revisited the text and adjusted the manuscript to be more concise and add more clarity.

●      The introduction (+ analysis goals) fails to explain the basic aspects of the analysis and dataset. It is not clear how many participants and datapoints were used to establish the group-level thought patterns, nor is it entirely clear that the fMRI data is a separate existing dataset. Some terms are introduced and highlighted and never revisited (e.g decoupled states and the role of the DMN).

Thank you for this critique, we have since adjusted the introduction to clearly explain the difference between Sample 1 and Sample 2 and further clarify that the fMRI data is an entirely separate, independent sample compared to the laboratory mDES sample:

Page 7-8 [158-174]: Thus, to overcome this obstacle, we developed a novel methodological approach using two independent sample participants. In the current study, one set of 120 participants was probed with mDES five times across the three ten-minute movie clips (11 minutes total, no sampling in the first minute). We used a jittered sampling technique where probes were delivered at different intervals across the film for different people depending on the condition they were assigned. Probe orders were also counterbalanced to minimize the systematic impact of prior and later probes at any given sampling moment. We used these data to construct a precise description of the dynamics of experience for every 15 seconds of three ten-minute movie clips. These data were then combined with fMRI data from a different sample of 44 participants who had already watched these clips without experience sampling [15]. By combining data from two different groups of participants, our method allows us to describe the time series of different experiential states (as defined by mDES) and relate these to the time series of brain activity in another set of participants who watched the same films with no interruptions. In this way, our study set out to explicitly understand how the patterns of thoughts that dominate different moments in a film in one group of participants relate to the brain activity at these time points in a second set of participants and, therefore, better understand the contribution of different neural systems to the movie-watching experience.

Page 8-9 [177-188] The goal of our study, therefore, was to understand the association between patterns of brain activity over time during movie clips in one group of participants and the patterns of thought that participants reported at the corresponding moment in a different set of participants (see Figure 1). This can be conceptualized as identifying the mapping between two multi-dimensional spaces, one reflecting the time series of brain activity and the other describing the time series of ongoing experience (see Figure 1 right-hand panel). In our study, we selected three 11-minute clips from movies (Citizenfour, Little Miss Sunshine and 500 Days of Summer) for which recordings of brain data in fMRI already existed (n = 44) [15] (Figure 1, Sample 1). A second set of participants (n = 120) viewed the same movie clips, providing intermittent reports on their thought patterns using mDES (Figure 1, Sample 2). Our goal was to understand the mapping between the patterns of brain activity at each moment of the film and the reports of ongoing thought recorded at the same point in the movies.

●      It is unclear what the utility of the method is - is it meant to be done in fMRI studies on the same participants? Or is the idea to use one sample to model another?

Great point, thank you for highlighting this important question. This paper aimed to interrogate the relationship between experience and neural states while preserving the novelty of movie-watching. Although it could be done in the same sample, it may be difficult to collect frequent reports of experience without interrupting the dynamics of the brain. However, in the future it could be possible to collect mDES and brain activity in the same individuals while they watched movies. For example, our prior studies (e.g. [9]) where we combined mDES with openly-available brain data activity during tasks. In the future, this online method could also be applied during movie watching to identify direct mapping between brain activity and films. However, this online approach would make it very expensive to produce the time series of experience across each clip given that it would require a large number of participants (e.g. 200 as we used in our current study). The following has been included in our manuscript:

Page 7 [149-159] One challenge that arises when attempting to map the dynamics of thought onto brain activity during movie watching is accounting for the inherently disruptive nature of experience sampling: to measure experience with sufficient frequency to map the dynamics of thoughts during movies would disrupt the natural dynamics of the brain and would also alter the viewer’s experience (for example, by pausing the film at a moment of suspense). Therefore, if we periodically interrupt viewers to acquire a description of their thoughts while recording brain activity, this could impact capturing important dynamic features of the brain. On the other hand, if we measured fMRI activity continuously over movie-watching (as is usually the case), we would lack the capacity to directly relate brain signals to the corresponding experiential states. Thus, to overcome this obstacle, we developed a novel methodological approach using two independent sample participants

●      The conclusions currently read as somewhat trivial (e.g "Our study, therefore, establishes both sensory and association cortex as core features of the movie-watching experience", "Our study supports the hypothesis that perceptual coupling between the brain and external input is a core feature of how we make sense of events in movies").

Thank you for this comment. In this revision we have attempted to extend the theoretical significance of our work in the discussion (for example, in contrasting the links between Intrusive distraction and the other components). To this end we have amended the text in this revision by including the following sections:

Page 33-35 [654-687]: Importantly, our study provides a novel method for answering these questions and others regarding the brain basis of experiences during films that can be applied simply and cost-effectively. As we have shown mDES can be combined with existing brain activity allowing information about both brain activity and experience to be determined at a relatively low cost.  For example, the cost-effective nature of our paradigm makes it an ideal way to explore the relationship between cognition and neural activity during movie-watching during different genres of film. In neuroimaging, conclusions are often made using one film in naturalistic paradigm studies [16]. Although the current study only used three movie clips, restraining our ability to form strong conclusions regarding how different patterns of thought relate to specific genres of film, in the future, it will be possible to map cognition across a more extensive set of movies and discern whether there are specific types of experience that different genres of films engage. One of the major strengths of our approach, therefore, is the ability to map thoughts across groups of participants across a wide range of movies at a relatively low cost.

Nonetheless, this paradigm is not without limitations. This is the first study, as far as we know, that attempts to compare experiential reports in one sample of participants with brain activity in a second set of participants, and while the utility of this method enables us to understand the relationship between thought and brain activity during movies, it will be important to extend our analysis to mDES data during movie watching while brain activity is recorded. In addition, our study is correlational in nature, and in the future, it could be useful to generate a more mechanistic understanding of how brain activity maps onto the participants experience. Our analysis shows that mDES is able to discriminate between films, highlighting its broad sensitivity to variation in semantic or affective content. Armed with this knowledge, we propose that in the future, researchers could derive mechanistic insights into how the semantic features may influence the mDES data. For example, it may be possible to ask participants to watch movies in a scrambled order to understand how the structure of semantic or information influences the mapping between brains and ongoing experience as measured by mDES. Finally, our study focused on mapping group-level patterns of experience onto group-level descriptions of brain activity. In the future it may be possible to adopt a “precision-mapping” approach by measuring longer periods of experience using mDES and determining how the neural correlates of experience vary across individuals who watched the same movies while brain activity was collected [1]. In the future, we anticipate that the ease with which our method can be applied to different groups of individuals and different types of media will make it possible to build a more comprehensive and culturally inclusive understanding of the links between brain activity and movie-watching experience

●      The beginning of the discussion is very clear and explains the study very well. Some of it could be brought up in the intro/analysis goal sections.

Thank you for this comment, this is an excellent idea. We have revisited the introduction and analysis goals section to mirror this clarity across the manuscript.

●      The different components are very interesting, and not entirely clear. Some examples in the text could help. Especially regarding your thought that verbal components would refer to a "decoupled" mental verbal analysis participants might be performing in their thoughts.

Thank you for this point. We would prefer not to elaborate on this point since, at present, it would simply be conjecture based on our correlational design. However, we have included a section in the discussion which explains how, in principle, we would draw more mechanistic conclusions (for example, by shuffling the order of scenes in a movie as suggested by another reviewer). In the current revision, we have amended the text in the following way:

Page 34 [674-679]: Our analysis shows that mDES is able to discriminate between films, highlighting its broad sensitivity to variation in semantic or affective content. Armed with this knowledge, we propose that in the future, researchers could derive mechanistic insights into how the semantic features may influence the mDES data. For example, it may be possible to ask participants to watch movies in a scrambled order to understand how the structure of semantic or information influences the mapping between brains and ongoing experience as measured by mDES

●      The reference to using neurosynth as performing a meta-analysis seems a little stretched.

We have adjusted the manuscript to remove ‘meta-analysis’ when referring to the analysis computed with neurosynth. Thank you for bringing this to our attention.

●      State-space is defined as brain-space in the methods.

Thank you, we have since updated this.

●      It could be useful to remind the reader what thought and brain spaces are at the top of the state-space results section.

This is an excellent point, and it has since been updated to remind the reader of thought- and brain-space. Thank you for this comment.

Page 24 [458-467]: Our next analysis used a “state-space” approach to determine how brain activity at each moment in the film predicted the patterns of thoughts reported at these moments (for prior examples in the domain of tasks, see [12, 17], See Methods). In this analysis, we used the coordinates of the group average of each TR in the “brain-space” and the coordinates of each experience sampling moment in the “thought-space.”. To clarify, the location of a moment in a film in “brain-space” is calculated by projecting the grand mean of brain activity for each volume of each film against the first five dimensions of brain activity from a decomposition of the Human Connectome Project (HCP) resting state data, referred to as Gradients 1-5. “Thought-space” is the decomposition of mDES items to create thought pattern components, referred to as “Episodic Knowledge”, “Intrusive Distraction”, “Verbal Detail” and “Sensory Engagement.”

●      DF missing from the t-test for episodic knowledge/grad 4.

Thank you for catching this, the degrees of freedom has since been included in this revision.

Page 24 [474-476]: First, we found a significant main effect of Gradient 4 (DAN to Visual), which predicted the similarity of answers to the “Episodic Knowledge” component, t(2046) = 2.17, p = .013, η2 = .01.

Public Reviews:

Reviewer #1:

●      The lack of direct interrogation of individual differences/reliability of the mDES scores warrants some pause.

Our study's goal was to understand how group-level patterns of thought in one group of participants relate to brain activity in a different group of participants. To this end, we decomposed trial-level mDES data to show dimensions that are common across individuals, which demonstrated excellent split-half reliability. Then we used these data in two complementary ways. First, we established that these ratings reliably distinguished between the different films (showing that our approach is sensitive to manipulations of semantic and affective features in a film) and that these group-level patterns were also able to predict patterns of brain activity in a different group of participants (suggesting that mDES dimensions are also sensitive to broad differences in how brain activity emerges during movie watching). Second, we established that variation across individuals in their mDES scores predicted their comprehension of information from the films. This establishes that when applied to movie-watching, mDES is sensitive to individual differences in the movie-watching experience (as determined by an individual's comprehension). Given the success of this study and the relative ease with which mDES can be performed, it will be possible in the future to conduct mDES studies that hone in on the common and distinct features of the movie-watching experience.

Reviewer #2:

(1) The dimensions of thought seem to distinguish between sensory and executive processing states. However, it is unclear if this effect primarily pertains to thinking. I could imagine highly intrusive distractions in movie segments to correlate with stagnating plot development, little change in scenery, or incomprehensible events. Put differently, it may primarily be the properties of the movies that evoke different processing modes, but these properties are not accounted for. For example, I'm wondering whether a simple measure of engagement with stimulus materials could explain the effects just as much. How can the effects of thinking be distinguished from the perceptual and semantic properties of the movie, as well as attentional effects? Is the measure used here capturing thought processes beyond what other factors could explain?

Our study used mDES to identify four distinct components of experience, each of which had distinct behavioural and neural correlates and relationships to comprehension. Together this makes it unlikely that a single measure of engagement would be able to capture the range of effects we observed in our study. For example, “Intrusive Distraction” was associated with regions of association cortex, while the other three components highlighted regions of sensory cortex. Behaviorally, we found that some components had a common effect on comprehension (e.g. “Intrusive distraction” was related to worse comprehension across all films), while others were linked to clear benefits to comprehension in specific films (e.g. “Episodic Knowledge” was associated with better comprehension in only one of the films). Given the complex nature of these effects, it would be difficult for a single metric of engagement to explain this pattern of results, and even if it did, this could be misleading because our analysis implies that they are better explained by a model of movie-watching experience in which there are several relatively orthogonal dimensions upon which our experience can vary.

At the same time, we also found that films vary in the general types of experience they can engender. For example, Citizenfour was high on “Intrusive Distraction” and participants performed relatively low on comprehension. This shows that manipulations of the semantic and affective content of films also have implications for the movie-watching experience. This pattern is consistent with laboratory studies that applied mDES during tasks and found that different tasks evoke different types of experience (for example, patterns of ‘intrusive’ thoughts were common in movie clips that were suspenseful, [18]). At the same time, in the same study, patterns of intrusive thought across the tasks were also associated with trait levels of dysphoria reported by participants. Other studies using mDES in daily life have shown that the data can be described by multiple dimensions and that each of these types of thought is more prevalent in certain activities than others ([19]). For example, in daily life, patterns of ‘intrusive distraction’ thoughts were more prevalent when individuals were engaged in activities that were relatively unengaging (such as resting). Collectively, therefore, studies using mDES suggest that is likely that human thought is multidimensional in nature and that these dimensions vary in a complex way in terms of (a) the contexts that promote them, and (b) how they are impacted by features of the individual (whether they be traits like anxiety or depression or memory for information in a film).

(2) I'm skeptical about taking human thought ratings at face value. Intrusive distraction might imply disengagement from stimulus materials, but it could also be an intended effect of the movie to trigger higher-level, abstract thinking. Can a label like intrusive distraction be misleading without considering the actual thought and movie content?

Our method uses a data-driven approach to identify the dimensions that best describe the range of answers that our participants provided to describe their experience. We use these dimensions to understand how these patterns of thought emerge in different contexts and how they vary across individuals (in this case, in different movies, but in other studies, laboratory tasks [3, 8, 9, 12, 20-22] or activities in daily life[6, 7]). These context relationships help constrain interpretations of what the components mean. For example, “Intrusive Distraction” scores were highest in the film with the most real-world significance for the participants (Citizenfour) and were associated with worse comprehension. In daily life, however, patterns of “Intrusive Distraction” thoughts tend to occur when activities engage in non-demanding activities, like resting. Psychological perspectives on thoughts that arise spontaneously occur in this manner since there is evidence that they occur in non-demanding tasks with no semantic content (when there is almost no external stimulus to explain the occurrence of the experience, see [23]), however, other studies have shown that specific cues in the environment can also cue the experience (see [23]). Consistent with this perspective, and our current data, patterns of ‘Intrusive Distraction’ thought are likely to arise for multiple reasons, some of which are more intrinsic in nature (the general association with poor comprehension across all films) and others which are extrinsic in nature (the elevation of intrusive distraction in Citizenfour).

It is also important to note that our data-driven approach also found patterns of experience that provide more information about the content of their experience, for example, the dimension of “Episodic Knowledge” is characterized by thoughts based on prior knowledge, involving the past, and concerning oneself, and was most prevalent in the romance film (500 Days of Summer). Likewise, “Sensory Engagement” was associated with experiences related to sensory input and positive emotionality and occurred more during the romance movie (500 Days of Summer) than in the documentary (Citizenfour) and was linked to increased brain activity across the sensory systems. This shows that mDES can also provide information about the content of that experience, and discriminate between different sources of experience. In the future, it will be possible to improve the level of detail regarding the content of experiences by changing the questions used to interrogate experience.     

(3) A jittered sampling approach is used to acquire thought ratings every 15 seconds. Are ratings for the same time point averaged across participants? If so, how consistent are ratings among participants? High consistency would suggest thoughts are mainly stimulus-evoked. Low consistency would question the validity of applying ratings from one (group of) participant(s) to brain-related analyses of another participant.

In this experiment, we sampled experience every 15 seconds in each clip, and in each sampling epoch, we gained mDES responses from eight participants. Furthermore, no participant was sampled at an adjacent time point, as our approach jittered probes approximately 2 minutes apart (See Supplementary Figure 7). To illustrate the consistency of mDES data, we have included an additional figure (Figure 3) highlighting how experience varies over time in each clip. It is evident from these plots that there are distinct moments in which group-averaged reported thoughts across participants are stable and that these can extend across adjacent sampling points (i.e. when the confidence intervals of the score at a timepoint do not overlap with zero). Therefore, in some cases, adjacent sampling points, consisting of different sets of eight participants, describe their experiences as having similar positions on the same mDES dimension. This suggests that there is agreement among individuals regarding how they experienced a specific moment in a film, and in some cases, this agreement was apparent in successive sets of eight participants. Together, our findings indicate a conservation of agreement across participants that spans multiple moments in a film. A clear example of agreement on experience across multiple sets of 10 participants can be seen between 150-400 seconds in the clip from 500 Days of Summer for the dimension of “Sensory Engagement” (time series plot 4 in Figure 3).

(4) Using three different movies to conclude that different genres evoke different thought patterns (e.g., line 277) seems like an overinterpretation with only one instance per genre.

We found that mDES was able to distinguish between each film on at least one dimension of experience. In other words, information encoded in the mDES dimensions was sensitive to variation in semantic and affective experiences in the different movie clips. This provides evidence that is necessary but not sufficient to conclude that we can distinguish different genres of films (i.e. if we could not distinguish between films, then we would not be able to distinguish genres). However, it is correct that to begin answering the broader question about experiences in different genres then it would be necessary to map cognition across a larger set of movies, ideally with multiple examples of each genre.

(5) I see no indication that results were cross-validated, and no effect sizes are reported, leaving the robustness and strength of effects unknown.

Thank you for drawing this to our attention. We have re-run the LMMs and ANOVA models to include partial eta-squared values to clarify the strength of the effects in each of our reported outcomes.

Reviewer #3:

●      What are the considerations for treating high-order thought patterns that occur during film viewing as stable enough to be used across participants? What would be the limitations of this method? (Do all people reading this paper think comparable thoughts reading through the sections?)

It is likely, based on our study, that films can evoke both stereotyped thought patterns (i.e. thoughts that many people will share) and others that are individualistic. It is clear that, in principle, mDES is capable of capturing empirical information on both stereotypical thoughts and idiosyncratic thoughts. For example, clear differences in experiences across films and, in particular, during specific periods within a film, show that movie-watching can evoke broadly similar thought patterns in different groups of participants (see Figure 3 right-hand panel). On the other hand, the association between comprehension and the different mDES components indicate that certain individuals respond to the same film clip in different ways and that these differences are rooted in objective information (i.e. their memory of an event in a film clip). A clear example of these more idiosyncratic features of movie watching experience can be seen in the association between “Episodic Knowledge” and comprehension. We found that “Episodic Knowledge” was generally high in the romance clip from 500 Days of Summer but was especially high for individuals who performed the best, indicating they remembered the most information. Thus good comprehends responded to the 500 Days of Summer clip with responses that had more evidence of “Episodic Knowledge” In the future, since the mDES approach can account for both stereotyped and idiosyncratic features of experience, it will be an important tool in understanding the common and distinct features that movie watching experiences can have, especially given the cost effective manner with which these studies can be run.   

●      How does this approach differ from collaborative filtering, (for example as presented in Chang et al., 2021)?

Our study is very similar to the notion of collaborative filtering since we can use an approach that is similar to crowd-sourcing as a tool for understanding brain activity. One of its strengths is its generalizability since it is also a method that can be used to understand cognition because it is not limited to movie-watching. We can use the same mDES method to sample cognition in multiple situations in daily life ([6, 19]), while performing tasks in the behavioural lab [18, 24], and while brain activity is being acquired [8, 25, 26]. In principle, therefore, we can use mDES to understand cognition in different contexts in a common analytic space (see [27] for an example of how this could work)

Page 5 [106-110]: In our study, we acquired experiential data in one group of participants while watching a movie clip and used these data to understand brain activity recorded in a second set of participants who watched the same clip and for whom no experiential data was recorded. This approach is similar to what is known as “collaborative filtering” [28].

●      In conclusion, this study tackles a highly interesting subject and does it creatively and expertly. It fails to discuss and establish the utility and appropriateness of its proposed method.

Thank you very much for your feedback and critique. In our revision and our responses to these questions, we provided more information about the method's robustness utility and application to understanding cognition.

References

(1) Gordon, E.M., et al., Precision Functional Mapping of Individual Human Brains. Neuron, 2017. 95(4): p. 791-807.e7.

(2) Smallwood, J., et al., The neural correlates of ongoing conscious thought. Iscience, 2021. 24(3).

(3) Konu, D., et al., Exploring patterns of ongoing thought under naturalistic and conventional task-based conditions. Consciousness and Cognition, 2021. 93.

(4) Smallwood, J., et al., The default mode network in cognition: a topographical perspective. Nature Reviews Neuroscience, 2021. 22(8): p. 503-513.

(5) Turnbull, A., et al., Age-related changes in ongoing thought relate to external context and individual cognition. Consciousness and Cognition, 2021. 96: p. 103226.

(6) McKeown, B., et al., The impact of social isolation and changes in work patterns on ongoing thought during the first COVID-19 lockdown in the United Kingdom. Proceedings of the National Academy of Sciences, 2021. 118(40): p. e2102565118.

(7) Mulholland, B., et al., Patterns of ongoing thought in the real world. Consciousness and Cognition, 2023. 114: p. 103530.

(8) Konu, D., et al., A role for the ventromedial prefrontal cortex in self-generated episodic social cognition. NeuroImage, 2020. 218: p. 116977.

(9) Turnbull, A., et al., Left dorsolateral prefrontal cortex supports context-dependent prioritisation of off-task thought. Nature Communications, 2019. 10.

(10) Ho, N.S.P., et al., Facing up to the wandering mind: Patterns of off-task laboratory thought are associated with stronger neural recruitment of right fusiform cortex while processing facial stimuli. NeuroImage, 2020. 214: p. 116765.

(11) Karapanagiotidis, T., et al., Tracking thoughts: Exploring the neural architecture of mental time travel during mind-wandering. NeuroImage, 2017. 147: p. 272-281.

(12) McKeown, B., et al., Experience sampling reveals the role that covert goal states play in task-relevant behavior. Scientific Reports, 2023. 13(1): p. 21710.

(13) Vatansever, D., et al., Distinct patterns of thought mediate the link between brain functional connectomes and well-being. Network Neuroscience, 2020. 4(3): p. 637-657.

(14) Wang, H.-T., et al., Dimensions of Experience: Exploring the Heterogeneity of the Wandering Mind. Psychological Science, 2017. 29(1): p. 56-71.

(15) Aliko, S., et al., A naturalistic neuroimaging database for understanding the brain using ecological stimuli. Scientific Data, 2020. 7(1).

(16) Yang, E., et al., The default network dominates neural responses to evolving movie stories. Nature Communications, 2023. 14(1): p. 4197.

(17) Turnbull, A., et al., Reductions in task positive neural systems occur with the passage of time and are associated with changes in ongoing thought. Scientific Reports, 2020. 10(1): p. 9912.

(18) Konu, D., et al., Exploring patterns of ongoing thought under naturalistic and conventional task-based conditions. Consciousness and cognition, 2021. 93: p. 103139.

(19) Mulholland, B., et al., Patterns of ongoing thought in the real world. Consciousness and cognition, 2023. 114: p. 103530.

(20) Christoff, K., et al., Experience sampling during fMRI reveals default network and executive system contributions to mind wandering. Proc Natl Acad Sci U S A, 2009. 106(21): p. 8719-24.

(21) Zhang, M., et al., Perceptual coupling and decoupling of the default mode network during mind-wandering and reading. eLife, 2022. 11: p. e74011.

(22) Zhang, M.C., et al., Distinct individual differences in default mode network connectivity relate to off-task thought and text memory during reading. Scientific Reports, 2019. 9.

(23) Smallwood, J. and J.W. Schooler, The science of mind wandering: Empirically navigating the stream of consciousness. Annual review of psychology, 2015. 66(1): p. 487-518.

(24) Turnbull, A., et al., The ebb and flow of attention: Between-subject variation in intrinsic connectivity and cognition associated with the dynamics of ongoing experience. Neuroimage, 2019. 185: p. 286-299.

(25) Turnbull, A., et al., Left dorsolateral prefrontal cortex supports context-dependent prioritisation of off-task thought. Nature communications, 2019. 10(1): p. 3816.

(26) Mckeown, B., et al., Experience sampling reveals the role that covert goal states play in task-relevant behavior. Scientific reports, 2023. 13(1): p. 21710.

(27) Chitiz, L., et al., Mapping cognition across lab and daily life using experience-sampling. 2023.

(28) Chang, L.J., et al., Endogenous variation in ventromedial prefrontal cortex state dynamics during naturalistic viewing reflects affective experience. Science Advances, 2021. 7(17): p. eabf7129.

eLife Assessment

This study presents a valuable methodological advancement in quantifying thoughts over time. A novel multi-dimensional experience-sampling approach is presented, identifying data-driven patterns that the authors use to interrogate fMRI data collected during naturalistic movie-watching. The experimentation is inventive and the analyses carried out are convincing.

Reviewer #1 (Public review):

Summary:

The authors used a novel multi-dimensional experience sampling (mDES) approach to identify data-driven patterns of experience samples that they use to interrogate fMRI data collected during naturalistic movie-watching data. They identify a set of multi-sensory features of a set of movies that delineate low-dimensional gradients of BOLD fMRI signal patterns that have previously been linked to fundamental axes of cortical organization.

Strengths:

* The novel solution to challenges associated with experience sampling offer potential access to aspects of experience that have been challenging to assess.

Weaknesses:

* The lack of direct interrogation of individual differences/reliability of the mDES scores warrants some pause.

Reviewer #2 (Public review):

Summary:

The present study explores how thoughts map onto brain activity, a notoriously challenging question because of the dynamic, subjective, and abstract nature of thoughts. To tackle this question, the authors collected continuous thought ratings from participants watching a movie, and additionally made use of an open-source fMRI dataset recorded during movie watching as well as five established gradients of brain variation as identified in resting state data. Using a voxel-space approach, the results show that episodic knowledge, verbal detail, and sensory engagement of thoughts commonly modulate visual and auditory cortex, while intrusive distraction modulates the frontoparietal network. Additionally, sensory engagement mapped onto a gradient from primary to association cortex, while episodic knowledge mapped onto a gradient from the dorsal attention network to visual cortex. Building on the association between behavioral performance and neural activation, the authors conclude that sensory coupling to external input and frontoparietal executive control are key to comprehension in naturalistic settings.

The manuscript stands out for its methodological advancements in quantifying thoughts over time and its aim to study the implementation of thoughts in the brain during naturalistic movie watching. However, the conceptualization of thoughts remains vague, limiting the study's insights into brain function.

Strengths:

(1) The study raises a question that has been difficult to study in naturalistic settings so far but is key to understanding human cognition, namely how thoughts map onto brain activation.<br /> (2) The thought ratings introduce a novel method for continuously tracking thoughts, promising utility beyond this study.<br /> (3) The authors used diverse data types, metrics, and analyses to substantiate the effects of thinking from multiple perspectives.

Weaknesses:

(1) The distinction between thinking and stimulus processing (in the sense of detecting and assigning meaning to features, modulated by factors such as attention) remains unclear. Is "thinking" a form of conscious access or a reportable read-out from sensory and higher-level stimulus processing? Or does it simply refer to the method used here to identify different processing states?<br /> (2) The dimensions of thought appear to be directly linked to brain areas traditionally associated with core faculties of perception and cognition. For example, superior temporal cortex codes for speech information, which is also where thought reports on verbal detail localize in this study. This raises the question of whether the present study truly captures mechanisms specific to thinking and distinct from processing, especially given that individual variations in reports were not considered and movie-specific features were not controlled for.

Reviewer #3 (Public review):

This study attempted to investigate the relations between processing in the human brain during movie watching and corresponding thought processes. This is a highly interesting question, as movie watching presents a semi-constrained task, combining naturally occurring thoughts and common processing of sensory inputs across participants. This task is inherently difficult because in order to know what participants are thinking at any given moment, one has to interrupt the same thought process which is the object of study.

This study attempts to deal with this issue by aggregating staggered experience sampling data across participants in one behavioral study and using the population level thought patterns to model brain activity in different participants in an open access fMRI dataset.

The behavioral data consist of 120 participants who watched 3 11-minute movie clips. Participants responded to the mDES questionnaire: 16 visual scales characterizing ongoing thought 5 times, two minutes apart, in each clip. The 16 items are first reduced to 4 factors using PCA, and their levels are compared across the different movies. The factors are "episodic knowledge", "intrusive distraction", "verbal detail", and "sensory engagement". The factors differ between the clips, and distraction is negatively correlated with movie comprehension and sensory engagement is positively correlated with comprehension.

The components are aggregated across participants (transforming single subject mDES answers into PCA space and concatenating responses of different participants) and are used as regressors in a GLM analysis. This analysis identifies brain regions corresponding to the components. The resulting brain maps reveal activations that are consistent with the proposed mental processes (e.g. negative loading for intrusion in frontoparietal network, positive loadings for visual and auditory cortices for sensory engagement).

Then, the coordinates for brain regions which were significant for more than one component are entered into a paper search in neurosynth. It is not clear what this analysis demonstrates beyond the fact that sensory engagement contained both visual and auditory components.

The next analysis projected group-averaged brain activation onto gradients (based on previous work) and used gradient timecourses to predict the behavioral report timecourses. This revealed that high activations in gradient 1 (sensory→association) predicted high sensory engagement, and that "episodic knowledge" thought patterns were predicted by increased visual cortex activations. Then, permutation tests were performed to see whether these thought pattern related activations corresponded to well defined regions on a given cluster.

This paper is framed as presenting a new paradigm but it does little to discuss what this paradigm serves, what are its limitations and how it should have been tested. The novelty appears to be in using experience sampling from 1 sample to model the responses of a second sample.

What are the considerations for treating high-order thought patterns that occur during film viewing as stable enough to use across participants? What would be the limitations of this method? (Do all people reading this paper think comparable thoughts reading through the sections?) This is briefly discussed in the revised manuscript and generally treated as an opportunity rather than as a limitation.

In conclusion, this study tackles a highly interesting subject and does it creatively and expertly. It fails to discuss and establish the utility and appropriateness of its proposed method.

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.

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 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.

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.

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.

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?

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.

eLife Assessment

This valuable paper provides an unbiased landscape for the cerebellar cortical outputs to the brainstem nuclei. By conducting anatomical and physiological analyses of the axonal terminals of Purkinje cells, the data provide convincing evidence that Purkinje cells innervate brainstem nuclei directly. The results show that in addition to previously known inputs to vestibular and parabrachial nuclei, Purkinje cells synapse onto the pontine central grey nucleus but have little effect on the locus coeruleus and mesencephalic trigeminal neurons.

Reviewer #1 (Public review):

Summary:

This paper is an incremental follow-up to the authors' recent paper which showed that Purkinje cells make inhibitory synapses onto brainstem neurons in the parabrachial nucleus which project directly to the forebrain. In that precedent paper, the authors used a mouse line that expresses the presynaptic marker synaptophysin in Purkinje cells to identify Purkinje cell terminals in the brainstem and they observed labeled puncta not only in the vestibular and parabrachial nuclei, as expected, but also in neighboring dorsal brainstem nuclei, prominently the central pontine grey. The present study, motivated by the lack of thorough characterization of PC projections to the brainstem, uses the same mouse line to anatomically map the density and a PC-specific channelrhodopsin mouse line to electrophysiologically assess the strength of Purkinje cell synapses in dorsal brainstem nuclei. The main findings are (1) the density of Purkinje cell synapses is highest in vestibular and parabrachial nuclei and correlates with the magnitude of evoked inhibitory synaptic currents, and (2) Purkinje cells also synapse in the central pontine grey nucleus but not in the locus coeruleus or mesencephalic nucleus.

Strengths:

The complementary use of anatomical and electrophysiological methods to survey the distribution and efficacy of Purkinje cell synapses on brainstem neurons in mouse lines that express markers and light-sensitive opsins specifically in Purkinje cells is the major strength of this study. By systematically mapping presynaptic terminals and light-evoked inhibitory postsynaptic currents in the dorsal brainstem, the authors provide convincing evidence that Purkinje cells do synapse directly onto pontine central grey and nearby neurons but do not synapse onto trigeminal motor or locus coeruleus neurons. Their results also confirm previously documented heterogeneity of Purkinje cell inputs to the vestibular nucleus and parabrachial neurons.

Weaknesses:

Although the study provides strong evidence that Purkinje cells do not make extensive synapses onto LC neurons, which is a helpful caveat given previous reports to the contrary, it falls short of providing the comprehensive characterization of Purkinje cell brainstem synapses which seemed to be the primary motivation of the study. The main information provided is a regional assessment of PC density and efficacy, which seems of limited utility given that we are not informed about the different sources of PC inputs, variations in the sizes of PC terminals, the subcellular location of synaptic terminals, or the anatomical and physiological heterogeneity of postsynaptic cell types. The title of this paper would be more accurate if "characterization" were replaced by "survey".

Several of the study's conclusions are quite general and have already been made for vestibular nuclei, including the suggestions in the Abstract, Results, and Discussion that PCs selectively influence brainstem subregions and that PCs target cell types with specific behavioral roles.

Reviewer #2 (Public review):

Summary:

While it is often assumed that the cerebellar cortex connects, via its sole output neuron, the Purkinje cell, exclusively to the cerebellar nuclei, axonal projections of the Purkinje cells to dorsal brainstem regions have been well documented. This paper provides comprehensive mapping and quantification of such extracerebellar projections of the Purkinje cells, most of which are confirmed with electrophysiology in slice preparation. A notable methodological strength of this work is the use of highly Purkinje cell-specific transgenic strategies, enabling selective and unbiased visualization of Purkinje terminals in the brainstem. By utilizing these selective mouse lines, the study offers compelling evidence challenging the general assumption that Purkinje cell targets are limited to the cerebellar nuclei. While the individual connections presented are not entirely novel, this paper provides a thorough and unambiguous demonstration of their collective significance. Regarding another major claim of this paper, "characterization of direct Purkinje cell outputs (Title)", however, the depth of electrophysiological analysis is limited to the presence/absence of physiological Purkinje input to postsynaptic brainstem neurons whose known cell types are mostly blinded. Overall, conceptual advance is largely limited to confirmatory or incremental, although it would be useful for the field to have the comprehensive landscape presented.

Strengths:

(1) Unsupervised comprehensive mapping and quantification of the Purkinje terminals in the dorsal brainstem are enabled, for the first time, by using the current state-of-the-art mouse lines, BAC-Pcp2-Cre and synaptophysin-tdTomato reporter (Ai34).

(2) Combinatorial quantification with vGAT puncta and synaptophysin-tdTomato labeled Purkinje terminals clarifies the anatomical significance of the Purkinje terminals as an inhibitory source in each dorsal brainstem region.

(3) Electrophysiological confirmation of the presence of physiological Purkinje synaptic input to 7 out of 9 dorsal brainstem regions identified.

(4) Pan-Purkinje ChR2 reporter provides solid electrophysiological evidence to help understand the possible influence of the Purkinje cells onto LC.

Weaknesses:

(1) The present paper is largely confirmatory of what is presented in a previous paper published by the author's group (Chen et al., 2023, Nat Neurosci). In this preceding paper, the author's group used AAV1-mediated anterograde transsynaptic strategy to identify postsynaptic neurons of the Purkinje cells. The experiments performed in the present paper are, by nature, complementary to the AAV1 tracing which can also infect retrogradely and thus is not able to demonstrate the direction of synaptic connections between reciprocally connected regions. Anatomical findings are all consistent with the preceding paper. The likely absence of robust physiological connections from the Purkinje to LC has also been evidenced in the preceding paper by examining c-Fos response to Purkinje terminal photoinhibition at the PBN/LC region.

(2) Although the authors appear to assume uniform cell type and postsynaptic response in each of the dorsal brainstem nuclei (as noted in the Discussion, "PCs likely function similarly to their inputs to the cerebellar nuclei, where a very brief pause in firing can lead to large and rapid elevations in target cell firing"), we know that the responses to the Purkinje cell input are cell type dependent, which vary in neurotransmitter, output targets, somata size, and distribution, in the cerebellar and vestibular nuclei (Shin et al., 2011, J Neurosci; Najac and Raman, 2015, J Neurosci; Özcan et al., 2020, J Neurosci). This consideration impacts the interpretation of two key findings: (a) "Large ... PC-IPSCs are preferentially observed in subregions with the highest densities of PC synapses (Abstract)". For example, we know that the terminal sparse regions reported in the present paper do contain Floccular Targeted Neurons that are sparse yet have dense somatic terminals with profound postinhibitory rebound (Shin et al.). Despite their sparsity, these postsynaptic neurons play a distinct and critical role in proper vestibuloocular reflex. Therefore, associating broad synaptic density with "PC preferential" targets, as written in the Abstract, may not fully capture the behavioral significance of Purkinje extracerebellar projections. (b) "We conclude ... only a small fraction of cell. This suggests that PCs target cell types with specific behavioral roles (Abstract, the last sentence)". Prior research has already established that "PCs target cell types with specific behavioral roles in brainstem regions". Also, whether 23 % (for PCG), for example, is "a small fraction" would be subjective: it might represent a numerically small but functionally important cell type population. The physiological characterization provided in the present cell type-blind analysis could, from a functional perspective, even be decremental when compared to existing cell type-specific analyses of the Purkinje cell inputs in the literature.

(3) The quantification analyses used to draw conclusions about<br /> (a) the significance of PC terminals among all GABAergic terminals and<br /> (b) the fractions of electrophysiologically responsive postsynaptic brainstem neurons may have potential sampling considerations:.<br /> (b.i) this study appears to have selected subregions from each brainstem nucleus for quantification (Figure 2). However, the criteria for selecting these subregions are not explicitly detailed, which could affect the interpretation of the results.<br /> (b.ii) the mapping of recorded cells (Figure 3) seems to show a higher concentration in terminal-rich regions of the vestibular nuclei.

Reviewer #3 (Public review):

Summary:

The manuscript by Chen and colleagues explores the connections from cerebellar Purkinje cells to various brainstem nuclei. They combine two methods - presynaptic puncta labeling as putative presynaptic markers, and optogenetics, to test the anatomical projections and functional connectivity from Purkinje cells onto a variety of brainstem nuclei. Overall, their study provides an atlas of sorts of Purkinje cell connectivity to the brainstem, which includes a critical analysis of some of their own data from another publication. Overall, the value of this work is to both provide neural substrates by which Purkinje cells may influence the brainstem and subsequent brain regions independent of the deep cerebellar nuclei and also, to provide a critical analysis of viral-based methods to explore neuronal connectivity.

Strengths:

The strengths lie in the simplicity of the study, the number of cells patched, and the relationship between the presence of putative presynaptic puncta and electrophysiological results. This type of study is important and should provide a foundation for future work exploring cerebellar inputs and outputs. Overall, I think that the critique of viral-based methods to define connectivity, and a more holistic assessment of what connectivity is and how it should be defined is timely and warranted, as I think this is under-appreciated by many groups and overall, there is a good deal of research being published that do not properly consider the issues that this manuscript raises about what viral-based connectivity maps do and do not tell us.

Weaknesses:

While I overall liked the manuscript, I do have a few concerns that relate to interpretation of results, and discussion of technological limitations. The main concerns I have relate to the techniques that the authors use, and an insufficient discussion of their limitations. The authors use a Cre-dependent mouse line that expresses a synaptophysin-tomato marker, which the authors confidently state is a marker of synapses. This is misleading. Synaptophysin is a vesicle marker, and as such, labels axons, where vesicles are present in transit, and likely cell bodies where the protein is being produced. As such, the presence of tdtomato should not be interpreted definitively as the presence of a synapse. The use of vGAT as a marker, while this helps to constrain the selection of putative pre-synaptic sites, is also a vesicle marker and will likely suffer the same limitations (though in this case, the expression is endogenous and not driven by the ROSA locus). A more conservative interpretation of the data would be that the authors are assessing putative pre-synaptic sites with their analysis. This interpretation is wholly consistent with their findings showing the presence of tdtomato in some regions but only sparse connectivity - this would be expected in the event that axons are passing through. If the authors wish to strongly assert that they are specifically assessing synapses, a marker better restricted to synapses and not vesicles may be more appropriate.

Similarly, while optogenetics/slice electrophysiology remains the state of the art for assessing connectivity between cell populations, it is not without limitations. For example, connections that are not contained within the thickness of the slice (here, 200 um, which is not particularly thick for slice ephys preps) will not be detected. As such, the absence of connections is harder to interpret than the presence of connections. Slices were only made in the coronal plane, which means that if there is a particular topology to certain connections that is orthogonal to that plane, those connections may be under-represented. As such, all connectivity analyses likely are under-representations of the actual connectivity that exists in the intact brain. Therefore, perhaps the authors should consider revising their assessments of connections, or lack thereof, of Purkinje cells to e.g., LC cells. While their data do make a compelling case that the connections between Purkinje cells and LC cells are not particularly strong or numerous, especially compared to other nearby brainstem nuclei, their analyses do indicate that at least some such connections do exist. Thus, rather than saying that the viral methods such as rabies virus are not accurate reflections of connectivity - perhaps a more circumspect argument would be that the quantitative connectivity maps reported by other groups using rabies virus do not always reflect connectivity defined by other means e.g., functional connections with optogenetics. In some cases, the authors do suggest this (e.g."Together, these findings indicate that reliance on anatomical tracing experiments alone is insufficient to establish the presence and importance of a synaptic connection"), but in other cases, they are more dismissive of viral tracing results (e.g. "it further suggests that these neurons project to the cerebellum and were not retrogradely labeled"). Furthermore, some statements are a bit misleading e.g., mentioning that rabies methods are critically dependent on starter cell identity immediately following the citation of studies mapping inputs onto LC cells. While in general, this claim has merit, the studies cited (19-21) use Dbh-Cre to define LC-NE cells which does have good fidelity to the cells of interest in the LC. Therefore, rewording this section in order to raise these issues generally without proximity to the citations in the previous sentence may maintain the authors' intention without suggesting that perhaps the rabies studies from LC-NE cells that identified inputs from Purkinje cells were inaccurate due to poor fidelity of the Cre line. Overall, this manuscript would certainly not be the first report indicating that the rabies virus does not provide a quantitative map of input connections. In my opinion, this is still under-appreciated by the broad community and should be explicitly discussed. Thus, an acknowledgment of previous literature on this topic and how their work contributes to that argument is warranted.

eLife Assessment

This important study reports that the neurohormone, bursicon, and its receptor, play a role in the seasonal polyphenism of the bug Cacopsylla chinensis. Low temperature activates the bursicon signaling pathway during the transition from the summer to the winter form, affecting cuticle pigment and thickness as well as chitin content. The solid experiments reveal how bursicon signaling, which is modulated by the microRNA miR-6012, regulates features of polyphenism related to the exoskeleton, although it is less clear what the upstream regulatory events are.

Reviewer #1 (Public Review):

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.

Reviewer comments on revised version

(a) Major concerns<br /> (1) The revision did not respond to the major concern regarding the threshold response that defines polyphenism. Therefore, it still falls short of the claims made, since the claims were not revised either. Specifically, the authors now include a time series of tanning at two different temperatures, demonstrating the time points at which the induced tanning proceeds (Fig. S1). However, the appropriate response to that comment would have temperatures on the x-axis, not time. Intermediate temperatures are needed to test whether the induction is a threshold response or simply a continuous norm of reaction.<br /> (2) The authors also did not respond to the major comment regarding environmental induction of miR-6012 expression. Rather, Fig. 5E shows a time series under two temperatures, similar to the tanning time series. To test whether its induction is a threshold response (again, what defines polyphenism), a series of induction conditions is needed. Fig. 5E simply shows changes in expression over time under one induction temperature (25 ºC).<br /> (3) Although the manuscript title has been changed, little to nothing else in the revised text addresses the concern that this study is about tanning in psyllids, not seasonal polyphenism. The other traits making up the polyphenism, as well as their threshold response, were not measured.

In summary, this revision failed to address most of the chief concerns of the review summary. This manuscript should be reframed as a study of tanning in a species other than Drosophila, and any claims about polyphenism (that is, an environmentally induced threshold trait) still need to be tested.

Regarding the other concerns raised by the reviewers:

(4) Issues related to the assignment of the receptor used as a bursicon receptor were satisfactorily addressed.<br /> (5) Experiments regarding the timing of cuticle production presented in Supplementary Figure 1 are valuable, albeit, there are still some inaccuracies: i) the layering of the cuticle is not given accurately as there are more than the 3 layers indicated in the manuscript; ii), the reduced endocuticle in all relevant dsRNA cases suggests a massive molting defect that may underline the involvement of bursicon in molting in general, potentially masking its effect on morph transition. In other words, the phenotype is too strong to allow for the interpretation of its function with respect to morph transition. It would have been necessary to apply different concentrations of dsRNA in order to address this point. iii) The developmental timing at 10oC vs. 25oC seem to be similar, although duration would be expected to be longer at 10oC; iv) It would have been nice to see the days of development also for dsRNA injected animals.<br /> (6) Another unresolved point regards the source and target tissue of bursicon signaling. Admittedly, this problem is difficult to solve in a small insect species.

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.

eLife Assessment

This study presents valuable findings on what networks of genes were impacted by introgression from Denisovans, to identify the biological functions involved in high-altitude adaptation in Tibet. This study applies solid and previously validated methodology to identify genes with signatures of both introgression and positive selection. This paper would be of interest to population geneticists, anthropologists, and scientists studying the genetic basis underlying high-altitude adaptation.

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 Denisovan introgression facilitated HAA in other ways by contributing to additional HAA-related genetic variants. Specifically, the authors considered that if such 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 the EPAS1. They leveraged a few recently developed methods, including a composite likelihood method for detecting adaptive introgression and a biological network-based method for detecting polygenic selection, and identified compelling evidence of additional genes that exhibit Denisovan-like adaptive introgression signature and contributed to the polygenic adaptation at high altitude in Tibetans.

Strength:

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 detected new evidence of adaptive introgression and polygenic selection. Specifically, by applying VolcanoFinder, they confirmed previously identified adaptive introgression alleles such as EPAS1 and PPARA. By applying signet, they identified subsets of biological pathways enriched for archaic variants that contributed to HAA polygenic selection. They also leveraged additional methods such as LASSI and haplotype plotting to help confirm the signature of natural selection on their newly discovered adaptive introgression candidate genes.

Weakness:

The manuscript also improved substantially since the initial review, and the new candidate genes presented here now harbor compelling and convincing evidence of both adaptive introgression and HAA polygenic selection. There are no notable weaknesses in the revised manuscript and updated results.

Reviewer #2 (Public review):

Summary:

In Ferrareti et al. they identify adaptively introgressed genes using VolcanoFinder and then identify pathways enriched for adaptively introgressed genes. They use signet to identify pathways that are enriched for Denisovan alleles. The authors find that angiogenesis is one of the biological functions enriched for 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 of selection in other genes to investigate how archaic introgression may be enriched at the pathway level. A couple of methods are used to confirm the consistency of the results.

Looking into the biological functions enriched for Denisovan introgression in Tibetans is important for characterizing the impact of Denisovan introgression in facilitating high altitude adaptations.

Weaknesses:

I thank the authors for providing an improved version of their manuscript.

eLife Assessment

ProtSSN is a valuable approach that generates protein embeddings by integrating sequence and structural information, demonstrating improved prediction of mutation effects on thermostability compared to competing models. The evidence supporting the authors' claims is solid, with well-executed comparisons. This work will be of particular interest to researchers in bioinformatics and structural biology, especially those focused on protein function and stability.

Reviewer #1 (Public review):

After revisions:

My concerns have been addressed.

Prior to revisions:

Summary:<br /> The authors introduce a denoising-style model that incorporates both structure and primary-sequence embeddings to generate richer embeddings of peptides. My understanding is that the authors use ESM for the primary sequence embeddings, take resolved structures (or use structural predictions from AlphaFold when they're not available), then develop an architecture to combine these two with a loss that seems reminiscent of diffusion models or masked language model approaches. The embeddings can be viewed as ensemble-style embedding of the two levels of sequence information, or with AlphaFold, an ensemble of two methods (ESM+AlphaFold). The authors also gather external datasets to evaluate their approach and compare it to previous approaches. The approach seems promising, and appears to out-compete previous methods at several tasks. Nonetheless, I have strong concerns about a lack of verbosity as well as exclusion of relevant methods and references.

Advances:<br /> I appreciate the breadth of the analysis and comparisons to other methods. The authors separate tasks, models, and sizes of models in an intuitive, easy-to-read fashion that I find valuable for selecting a method for embedding peptides. Moreover, the authors gather two datasets for evaluating embeddings' utility for predicting thermostability. Overall, the work should be helpful for the field as more groups choose methods/pretraining strategies amenable to their goals, and can do so in an evidence-guided manner.

Considerations:<br /> Primarily, a majority of the results and conclusions (e.g., Table 3) are reached using data and methods from ProteinGym, yet the best-performing methods on ProteinGym are excluded from the paper (e.g., EVE-based models and GEMME). In the ProteinGym database, these methods outperform ProtSSN models. Moreover, these models were published over a year---or even 4 years in the case of GEMME---before ProtSSN, and I do not see justification for their exclusion in the text.

Secondly, related to comparison of other models, there is no section in the methods about how other models were used, or how their scores were computed. When comparing these models, I think it's crucial that there are explicit derivations or explanations for the exact task used for scoring each method. In other words, if the pre-training is indeed the important advance of the paper, the paper needs to show this more explicitly by explaining exactly which components of the model (and previous models) are used for evaluation. Are the authors extracting the final hidden layer representations of the model, treating these as features, then using these features in a regression task to predict fitness/thermostability/DDG etc.? How are the model embeddings of other methods being used, since, for example, many of these methods output a k-dimensional embedding of a given sequence, rather than one single score that can be correlated with some fitness/functional metric. Summarily, I think the text is lacking an explicit mention of how these embeddings are being summarized or used, as well as how this compares to the model presented.

I think the above issues can mainly be addressed by considering and incorporating points from Li et al. 2024[1] and potentially Tang & Koo 2024[2]. Li et al.[1] make extremely explicit the use of pretraining for downstream prediction tasks. Moreover, they benchmark pretraining strategies explicitly on thermostability (one of the main considerations in the submitted manuscript), yet there is no mention of this work nor the dataset used (FLIP (Dallago et al., 2021)) in this current work. I think a reference and discussion of [1] is critical, and I would also like to see comparisons in line with [1], as [1] is very clear about what features from pretraining are used, and how. If the comparisons with previous methods were done in this fashion, this level of detail needs to be included in the text.

To conclude, I think the manuscript would benefit substantially from a more thorough comparison of previous methods. Maybe one way of doing this is following [1] or [2], and using the final embeddings of each method for a variety of regression tasks---to really make clear where these methods are performing relative to one another. I think a more thorough methods section detailing how previous methods did their scoring is also important. Lastly, TranceptEVE (or a model comparable to it) and GEMME should also be mentioned in these results, or at the bare minimum, be given justification for their absence.

[1] Feature Reuse and Scaling: Understanding Transfer Learning with Protein Language Models Francesca-Zhoufan Li, Ava P. Amini, Yisong Yue, Kevin K. Yang, Alex X. Lu bioRxiv 2024.02.05.578959; doi: https://doi.org/10.1101/2024.02.05.578959<br /> [2] Evaluating the representational power of pre-trained DNA language models for regulatory genomics Ziqi Tang, Peter K Koo bioRxiv 2024.02.29.582810; doi: https://doi.org/10.1101/2024.02.29.582810

Reviewer #2 (Public review):

Summary:

To design proteins and predict disease, we want to predict the effects of mutations on the function of a protein. To make these predictions, biologists have long turned to statistical models that learn patterns that are conserved across evolution. There is potential to improve our predictions however by incorporating structure. In this paper the authors build a denoising auto-encoder model that incorporates sequence and structure to predict mutation effects. The model is trained to predict the sequence of a protein given its perturbed sequence and structure. The authors demonstrate that this model is able to predict the effects of mutations better than sequence-only models.

As well, the authors curate a set of assays measuring the effect of mutations on thermostability. They demonstrate their model also predicts the effects of these mutations better than previous models and make this benchmark available for the community.

Strengths:

The authors describe a method that makes accurate mutation effect predictions by informing its predictions with structure.

Weaknesses:

In the review period, the authors included a previous method, SaProt, that similarly uses protein structure to predict the effects of mutations, in their evaluations.<br /> They see that SaProt performs similarly to their method.

Readers should note that methods labelled as "few-shot" in comparisons do not make use of experimental labels, but rather use sequences inferred as homologous; these sequences are also often available even if the protein has never been experimentally tested.

ProteinGym is largely made of deep mutational scans, which measure the effect of every mutation on a protein. These new benchmarks contain on average measurements of less than a percent of all possible point mutations of their respective proteins. It is unclear what sorts of protein regions these mutations are more likely to lie in; therefore it is challenging to make conclusions about what a model has necessarily learned based on its score on this benchmark. For example, several assays in this new benchmark seem to be similar to each other, such as four assays on ubiquitin performed in pH 2.25 to pH 3.0.

The authors state that their new benchmarks are potentially more useful than those of ProteinGym, citing Frazer 2021; readers should be aware that the mutations from the later source are actually mutations whose impact on human health has been determined through multiple sources, including population genetics, clinical evidence and some experiment.

Author response:

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

Response to Reviewer 1

Summary:

The authors introduce a denoising-style model that incorporates both structure and primary-sequence embeddings to generate richer embeddings of peptides. My understanding is that the authors use ESM for the primary sequence embeddings, take resolved structures (or use structural predictions from AlphaFold when they're not available), and then develop an architecture to combine these two with a loss that seems reminiscent of diffusion models or masked language model approaches. The embeddings can be viewed as ensemble-style embedding of the two levels of sequence information, or with AlphaFold, an ensemble of two methods (ESM+AlphaFold). The authors also gather external datasets to evaluate their approach and compare it to previous approaches. The approach seems promising and appears to out-compete previous methods at several tasks. Nonetheless, I have strong concerns about a lack of verbosity as well as the exclusion of relevant methods and references.

Thank you for the comprehensive summary. Regarding the concerns listed in the review below, we have made point-to-point response. We also modified our manuscript in accordance. 

Advances:

I appreciate the breadth of the analysis and comparisons to other methods. The authors separate tasks, models, and sizes of models in an intuitive, easy-to-read fashion that I find valuable for selecting a method for embedding peptides. Moreover, the authors gather two datasets for evaluating embeddings' utility for predicting thermostability. Overall, the work should be helpful for the field as more groups choose methods/pretraining strategies amenable to their goals, and can do so in an evidence-guided manner.

Thank you for recognizing the strength of our work in terms of the notable contributions, the solid analysis, and the clear presentation.

Considerations:

(1) Primarily, a majority of the results and conclusions (e.g., Table 3) are reached using data and methods from ProteinGym, yet the best-performing methods on ProteinGym are excluded from the paper (e.g., EVEbased models and GEMME). In the ProteinGym database, these methods outperform ProtSSN models. Moreover, these models were published over a year---or even 4 years in the case of GEMME---before ProtSSN, and I do not see justification for their exclusion in the text.

We decided to exclude the listed methods from the primary table as they are all MSA-based methods, which are considered few-shot methods in deep learning (Rao et al., ICML, 2021). In contrast, the proposed ProtSSN is a zero-shot method that makes inferences based on less information than few-shot methods. Moreover, it is possible for MSA-based methods to query aligned sequences based on predictions. For instance, Tranception (Notin et al., ICML, 2022) selects the model with the optimal proportions of logits and retrieval results according to the average correlation score on ProteinGym (Table 10, Notin et al., 2022).

With this in mind, we only included zero-shot deep learning methods in Table 3, which require no more than the sequence and structure of the underlying wild-type protein when scoring the mutants. In the revision, we have added the performance of SaProt to Table 3, and the performance of GEMME, TranceptEVE, and SaProt to Table 5. Furthermore, we have released the model's performance on the public leaderboard of ProteinGym v1 at proteingym.org.

(2) Secondly, related to the comparison of other models, there is no section in the methods about how other models were used, or how their scores were computed. When comparing these models, I think it's crucial that there are explicit derivations or explanations for the exact task used for scoring each method. In other words, if the pre-training is indeed an important advance of the paper, the paper needs to show this more explicitly by explaining exactly which components of the model (and previous models) are used for evaluation. Are the authors extracting the final hidden layer representations of the model, treating these as features, and then using these features in a regression task to predict fitness/thermostability/DDG etc.? How are the model embeddings of other methods being used, since, for example, many of these methods output a k-dimensional embedding of a given sequence, rather than one single score that can be correlated with some fitness/functional metric? Summarily, I think the text lacks an explicit mention of how these embeddings are being summarized or used, as well as how this compares to the model presented.

Thank you for the suggestion. Below we address the questions in three points. 

(1) The task and the scoring for each method. We followed your suggestion and added a new paragraph titled “Scoring Function” on page 9 to provide a detailed explanation of the scoring functions used by other deep learning zero-shot methods.

(2) The importance of individual pre-training modules. The complete architecture of the proposed ProtSSN model has been introduced on page 7-8. Empirically, the influence of each pre-training module on the overall performance has been examined through ablation studies on page 12. In summary, the optimal performance is achieved by combining all the individual modules and designs.

(3) The input of fitness scoring. For a zero-shot prediction task, the final score for a mutant will be calculated by wildly-used functions named log-odds ratio (for encoder models, including ours) or loglikelihood (for autoregressive models or inverse folding models. In the revision, we explicitly define these functions in sections “Inferencing” (page 7) and “Scoring Function” (page 9). 

(3) I think the above issues can mainly be addressed by considering and incorporating points from Li et al. 2024[1] and potentially Tang & Koo 2024[2]. Li et al.[1] make extremely explicit the use of pretraining for downstream prediction tasks. Moreover, they benchmark pretraining strategies explicitly on thermostability (one of the main considerations in the submitted manuscript), yet there is no mention of this work nor the dataset used (FLIP (Dallago et al., 2021)) in this current work. I think a reference and discussion of [1] is critical, and I would also like to see comparisons in line with [1], as [1] is very clear about what features from pretraining are used, and how. If the comparisons with previous methods were done in this fashion, this level of detail needs to be included in the text.

The initial version did not include an explicit comparison with the mentioned reference due to the difference in the learning task. In particular, [1] formulates a supervised learning task on predicting the continuous scores of mutants of specific proteins. In comparison, we make zero-shot predictions, where the model is trained in a self-supervised learning manner that requires no labels from experiments. In the revision, we added discussions in “Discussion and Conclusion” (lines 476-484):

Recommendations For The Authors:

Comment 1

I found the methods lacking in the sense that there is never a simple, explicit statement about what is the exact input and output of the model. What are the components of the input that are required by the user (to generate) or supply to the model? Are these inputs different at training vs inference time? The loss function seems like it's trying to de-noise a modified sequence, can you make this more explicit, i.e. exactly what values/objects are being compared in the loss?

We have added a more detailed description in the "Model Pipeline" section (page 7), which explains the distinct input requirements for training and inference, as well as the formulation of the employed loss function. To summarize:

(1) Both sequence and structure information are used in training and inference. Specifically, structure information is represented as a 3D graph with coordinates, while sequence information consists of AA-wise hidden representations encoded by ESM2-650M. During inference, instead of encoding each mutant individually, the model encodes the WT protein and uses the output probability scores relevant to the mutant to calculate the fitness score. This is a standard operation in many zero-shot fitness prediction models, commonly referred to as the log-odds-ratio.

(2) The loss function compares the differences between the noisy input sequence and the output (recovered) AA sequence. Noise is added to the input sequences, and the model is trained to denoise them (see “Ablation Study” for the different types of noise we tested). This approach is similar to a one-step diffusion process or BERT-style token permutation. The model learns to recover the probability of each node (AA) being one of 33 tokens. A cross-entropy loss is then applied to compare this distribution with the ground-truth (unpermuted) AA sequence, aiming to minimize the difference.

To better present the workflow, we revised the manuscript accordingly.

Comment 2

Related to the above, I'm not exactly sure where the structural/tertiary structure information comes from. In the methods, they don't state exactly whether the 3D coordinates are given in the CATH repository or where exactly they come from. In the results section they mention using AlphaFold to obtain coordinates for a specific task---is the use of AlphaFold limited only to these tasks/this is to show robustness whether using AlphaFold or realized coordinates?

The 3D coordinates of all proteins in the training set are derived from the crystal structures in CATH v4.3.0 to ensure a high-quality input dataset (see "Training Setup," Page 8). However, during the inference phase, we used predicted structures from AlphaFold2 and ESMFold as substitutes. This approach enhances the generalizability of our method, as in real-world scenarios, the crystal structure of the template protein to be engineered is not always available. The associated descriptions can be found in “Training Setup” (lines 271-272) and “Folding Methods” (lines 429-435).

Comment 3

Lines 142+144 missing reference "Section establishes", "provided in Section ."

199 "see Section " missing reference

214 missing "Section"

Thank you for pointing this out. We have fixed all missing references in the revision.

Comment 4

Table 2 - seems inconsistent to mention the number of parameters in the first 2 methods, then not in the others (though I see in Table 3 this is included, so maybe should just be omitted in Table 2).

In Table 2, we present the zero-shot methods used as baselines. Since many methods have different versions due to varying hyperparameter settings, we decided to list the number of parameters in the following tables.

We have double-checked both Table 3 and Table 5 and confirm that there is no inconsistency in the reported number of parameters. One potential explanation for the observed difference in the comment could be due to the differences in the number of parameters between single and ensemble methods. The ensemble method averages the predictions of multiple models, and we sum the total number of parameters across all models involved. For example, RITA-ensemble has 2210M parameters, derived from the sum of four individual models with 30M, 300M, 680M, and 1200M parameters.

Comment 5

In general, I found using the word "type" instead of "residue" a bit unnatural. As far as I can tell, the norm in the field is to say "amino acid" or "residue" rather than "type". This somewhat confused me when trying to understand the methods section, especially when talking about injecting noise (I figured "type" may refer to evolutionarily-close, or physicochemically-close residues). Maybe it's not necessary to change this in every instance, but something to consider in terms of ease of reading.

Thank you for your suggestion. The term "type" we used is a common expression similar to "class" in the NLP field. To avoid further confusion to the biologists, we have revised the manuscript accordingly. 

Comment 6

197 should this read "based on the kNN "algorithm"" (word missing) or maybe "based on "its" kNN"?

We have corrected the typo accordingly. It now reads “the 𝑘-nearest neighbor algorithm (𝑘NN)” (line 198).

Comment 7

200 weights of dimension 93, where does this number come from?

The edge features are derived by Zhou et al., 2024. We have updated the reference in the manuscript for clarity (lines 201-202).

Comment 8

210-212 "representations of the noisy AA sequence are encoded from the noisy input" what is the "noisy AA sequence?" might be helpful to exactly defined what is "noisy input" or "noisy AA sequence". This sentence could potentially be worded to make it clearer, e.g. "we take the modified input sequence and embed it using [xyz]."

We have revised the text accordingly. In the revised see lines 211-212:

Comment 9

In Table 3

Formatting, DTm (million), (million) should be under "# Params" likely?

Also for DDG this is reported on only a few hundred mutations, it might be worth plotting the confidence intervals over the Spearman correlation (e.g. by bootstrapping the correlation coefficient).

We followed the suggestion and added “million” under the "# Params". We have added the bootstrapped results for DDG and DTm to Table 6. For each dataset, we randomly sampled 50% of the data for ten independent runs. ProtSSN achieves the top performance with a considerably small variance.

Comment 10

The paragraph in lines 319 to lines 328 I feel may lack sufficient evidence.

"While sequence-based analysis cannot entirely replace the role of structure-based analysis, compared to a fully structure-based deep learning method, a protein language model is more likely to capture sufficient information from sequences by increasing the model scale, i.e., the number of trainable parameters."

This claim is made without a citation, such as [1]. Increasing the scale of the model doesn't always align with improving out-of-sample/generalization performance. I don't feel fully convinced by the claim that worse prediction is ameliorated by increasing the number of parameters. In Table 3 the performance is not monotonic with (nor scales with) the number of parameters, even within a model. See ProGen2 Expression scores, or ESM-2 Stability scores, as a function of their model sizes. In [1], the authors discuss whether pretraining strategies are aligned with specific tasks. I think rewording this paragraph and mentioning this paper is important. Figure 3 shows that maybe there's some evidence for this but I don't feel entirely convinced by the plot.

We agree that increasing the number of learnable parameters does not always result in better performance in downstream tasks. However, what we intended to convey is that language models typically need to scale up in size to capture the interactions among residues, while structure-based models can achieve this more efficiently with lower computational costs. We have rephrased this paragraph in the paper to clarify our point in lines 340-342.

Comment 11

Line 327 related to my major comment, " a comprehensive framework, such as ProtSSN, exhibits the best performance." Refers to performance on ProteinGym, yet the best-performing methods on ProteinGym are excluded from the comparison.

The primary comparisons were conducted using zero-shot models for fairness, meaning that the baseline models were not trained on MSA and did not use test performance to tune their hyperparameters. It's also worth noting that SaProt (the current SOTA model) had not been updated on the leaderboard at the time of submitting this paper. In the revised manuscript, we have included GEMME and TranceptEVE in Table 5 and SaProt in Tables 3, 5, and 6. While ProtSSN does not achieve SOTA performance in every individual task, our key argument in the analysis is to highlight the overall advantage of hybrid encoders compared to single sequence-based or structure-based models. We made clearer statement in the revised manuscript (line 349):

Comment 12

Line 347, line abruptly ends "equivariance when embedding protein geometry significantly." (?).

We have fixed the typo, (lines 372-373): 

Comment 13

Figure 3 I think can be made clearer. Instead of using True/false maybe be more explicit. For example in 3b, say something like "One-hot encoded" or "ESM-2 embedded".

The labels were set to True/False with the title of the subfigures so that they can be colored consistently.

Following the suggestion, we have updated the captions in the revised manuscript for clarity.

Comment 14

Lines 381-382 "average sequential embedding of all other Glycines" is to say that the score is taken as the average score in which Glycine is substituted at every other position in the peptide? Somewhat confused by the language "average sequential embedding" and think rephrasing could be done to make things clearer.

We have revised the related text accordingly a for clearer presentation (lines 406-413). 

Comment 15

Table 5, and in mentions to VEP, if ProtSSN is leveraging AlphaFold for its structural information, I disagree that ProtSSN is not an MSA method, and I find it unfair to place ProtSSN in the "non-MSA" categories. If this isn't the case, then maybe making clearer the inputs etc. in the Methods will help.

Your response is well-articulated and clear, but here is a slight revision for improved clarity and flow:

We respectfully disagree with classifying a protein encoding method based solely on its input structure. While AF2 leverages MSA sequences to predict protein structures, this information is not used in our model, and our model is not exclusive to AF2-predicted structures. When applicable, the model can encode structures derived from experimental data or other folding methods. For example, in the manuscript, we compared the performance of ProtSSN using proteins folded by both AF2 and ESMFold.

However, we would like to emphasize that comparing the sensitivity of an encoding method across different structures or conformations is not the primary focus of our work. In contrast, some methods explicitly use MSA during model training. For instance, MSA-Transformer encodes MSA information directly into the protein embedding, and Tranception-retrieval utilizes different sets of MSA hyperparameters depending on the validation set's performance.

To avoid further confusion, we have revised the terms "MSA methods" and "non-MSA methods" in the manuscript to "zero-shot methods" and "few-shot methods."

Comment 16

Table 3 they're highlighted as the best, yet on ProteinGym there's several EVE models that do better as well as GEMMA, which are not referenced.

The comparison in Table 3 focuses on zero-shot methods, whereas GEMME and EVE are few-shot models. Since these methods have different input requirements, directly comparing them could lead to

unfair conclusions. For this reason, we reserved the comparisons with these few-shot models for Table 5, where we aim to provide a more comprehensive evaluation of all available methods.            

Response to Reviewer 2

Summary:

To design proteins and predict disease, we want to predict the effects of mutations on the function of a protein. To make these predictions, biologists have long turned to statistical models that learn patterns that are conserved across evolution. There is potential to improve our predictions however by incorporating structure. In this paper, the authors build a denoising auto-encoder model that incorporates sequence and structure to predict mutation effects. The model is trained to predict the sequence of a protein given its perturbed sequence and structure. The authors demonstrate that this model is able to predict the effects of mutations better than sequence-only models.

Thank you for your thorough review and clear summary of our work. Below, we provide a detailed, pointby-point response to each of your questions and concerns. 

Strengths:

The authors describe a method that makes accurate mutation effect predictions by informing its predictions with structure.

Thank you for your clear summary of our highlights.

Weaknesses:

Comment 1

It is unclear how this model compares to other methods of incorporating structure into models of biological sequences, most notably SaProt.

(https://www.biorxiv.org/content/10.1101/2023.10.01.560349v1.full.pdf).

In the revision, we have updated the performance of SaProt single models (with both masked and unmasked versions with the pLDDT score) and ensemble models in the Tables 3, 5, and 6.

In the revised manuscript, we have updated the performance results for SaProt's single models (both masked and unmasked versions with the pLDDT score) as well as the ensemble models. These updates are reflected in Tables 3, 5, and 6.

Comment 2

ProteinGym is largely made of deep mutational scans, which measure the effect of every mutation on a protein. These new benchmarks contain on average measurements of less than a percent of all possible point mutations of their respective proteins. It is unclear what sorts of protein regions these mutations are more likely to lie in; therefore it is challenging to make conclusions about what a model has necessarily learned based on its score on this benchmark. For example, several assays in this new benchmark seem to be similar to each other, such as four assays on ubiquitin performed at pH 2.25 to pH 3.0.

We agree that both DTm and DDG are smaller datasets, making them less comprehensive than ProteinGym. However, we believe DTm and DDG provide valuable supplementary insights for the following reasons:

(1) These two datasets are low-throughput and manually curated. Compared to datasets from highthroughput experiments like ProteinGym, they contain fewer errors from experimental sources and data processing, offering cleaner and more reliable data.

(2) Environmental factors are crucial for the function and properties of enzymes, which is a significant concern for many biologists when discussing enzymatic functions. Existing benchmarks like ProteinGym tend to simplify these factors and focus more on global protein characteristics (e.g., AA sequence), overlooking the influence of environmental conditions.

(3) While low-throughput datasets like DTm and DDG do not cover all AA positions or perform extensive saturation mutagenesis, these experiments often target mutations at sites with higher potential for positive outcomes, guided by prior knowledge. As a result, the positive-to-negative ratio is more meaningful than random mutagenesis datasets, making these benchmarks more relevant for evaluating model performance.

We would like to emphasize that DTm and DDG are designed to complement existing benchmarks rather than replace ProteinGym. They address different scales and levels of detail in fitness prediction, and their inclusion allows for a more comprehensive evaluation of deep learning models.

Recommendations For The Authors:

Comment 1

I recommend including SaProt in your benchmarks.

In the revision, we added comparisons with SaProt in all the Tables (3, 5 and 6). 

Comment 2

I also recommend investigating and giving a description of the bias in these new datasets.

The bias of the new benchmarks could be found in Table 1, where the mutants are distributed evenly at different level of pH values.

In the revision, we added a discussion regarding the new datasets in “Discussion and Conclusion” (lines 496-504 of the revised version).

Comment 3

I also recommend reporting the model's ability to predict disease using ClinVar -- this experiment is conspicuously absent.

Following the suggestion, we retrieved 2,525 samples from the ClinVar dataset available on ProteinGym’s website. Since the official source did not provide corresponding structure files, we performed the following three steps:

(1) We retrieved the UniProt IDs for the sequences from the UniProt website and downloaded the corresponding AlphaFold2 structures for 2,302 samples.

(2) For the remaining proteins, we used ColabFold 1.5.5 to perform structure prediction.

(3) Among these, 12 proteins were too long to be folded by ColabFold, for which we used the AlphaFold3 server for prediction.

All processed structural data can be found at https://huggingface.co/datasets/tyang816/ClinVar_PDB. Our test results are provided in the following table. ProtSSN achieves the top performance over baseline methods.

Author response table 1.

eLife assessment

This fundamental work provides creative and thoughtful analysis of rodent foraging behavior and its dependence on body reference frame-based vs world reference frame-based cues. Compelling evidence demonstrates that a robust map, capable of supporting taking novel shortcuts, is learned primarily if not exclusively based on self-motion cues, which has rarely if ever been reported outside of the human literature. The work, which will be of interest to a broad audience of neuroscientists, provides a rich discussion about the role of the hippocampus in supporting the behavior that should guide future neurophysiological investigations.

eLife Assessment

This fundamental work provides creative and thoughtful analysis of rodent foraging behavior and its dependence on body reference frame-based vs world reference frame-based cues. Compelling evidence demonstrates that a robust map, capable of supporting taking novel shortcuts, can be learned primarily if not exclusively based on self-motion cues, which has rarely if ever been reported outside of the human literature. The work, which will be of interest to a broad audience of neuroscientists, provides a rich discussion about the role of the hippocampus in supporting the behavior that should guide future neurophysiological investigations.

Reviewer #1 (Public review):

Assessment:

This fundamental work advances our understanding of navigation and path integration in mammals by using a clever behavioral paradigm. The paper provides compelling evidence that mice are able to create and use a cognitive map to find "short cuts" in an environment, using only the location of rewards relative to the point of entry to the environment and path integration, and need not rely on visual landmarks.

Summary:

The authors have designed a novel experimental apparatus called the 'Hidden Food Maze (HFM)' and a beautiful suite of behavioral experiments using this apparatus to investigate the interplay between allothetic and idiothetic cues in navigation. The results presented provide a clear demonstration of the central claim of the paper, namely that mice only need a fixed start location and path integration to develop a cognitive map. The experiments and analyses conducted to test the main claim of the paper -- that the animals have formed a cognitive map -- are conclusive and include many thoughtfully designed control experiments to eliminate alternatives.

Strengths:

The 90 degree rotationally symmetric design and use of 4 distal landmarks and 4 quadrants with their corresponding rotationally equivalent locations (REL) lends itself to teasing apart the influence of path integration and landmark-based navigation in a clever way. The authors use a complete set of experiments and associated controls to show that mice can use a start location and path integration to develop a cognitive map and generate shortcut routes to new locations.

Weaknesses:

There were no major weaknesses identified that were not addressed during revisions.

Reviewer #3 (Public review):

Summary:

How is it that animals find learned food locations in their daily life? Do they use landmarks to home in on these learned locations or do they learn a path based on self-motion (turn left, take ten steps forward, turn right, etc.). This study carefully examines this question in a well-designed behavioral apparatus. A key finding is that to support the observed behavior in the hidden food arena, mice appear to not use the distal cues that are present in the environment for performing this task. Removal of such cues did not change the learning rate, for example. In a clever analysis of whether the resulting cognitive map based on self-motion cues could allow a mouse to take a shortcut, it was found that indeed they are. The work nicely shows the evolution of the rodent's learning of the task, and the role of active sensing in the targeted reduction of uncertainty of food location proximal to its expected location.

Strengths:

A convincing demonstration that mice can synthesize a cognitive map for the finding of a static reward using body frame-based cues. Showing that uncertainty of final target location is resolved by an active sensing process of probing holes proximal to the expected location. Showing that changing the position of entry into the arena rotates the anticipated location of the reward in a manner consistent with failure to use distal cues.

Weaknesses:

Weaknesses: The Reviewing Editor felt that previously identified weaknesses from Reviewer #3 were adequately addressed in the final manuscript.

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.

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

eLife Assessment

This revised study provides valuable information on the single nucleus RNA sequencing transcriptome, pathways, and cell types in pig skeletal muscle in response to conjugated linoleic acid (CLA) supplementation. Based on the comprehensive data analyses, the data are considered compelling and provide new insight into the mechanisms underlying intramuscular fat deposition and muscle fiber remodeling. The revised study clarifies major aspects of its methodology and analysis, addresses previous reviewer concerns, and contributes significantly to the understanding of nutritional strategies for fat infiltration in pig muscle.

Reviewer #1 (Public review):

In this revised manuscript, the authors aim to elucidate the cytological mechanisms by which conjugated linoleic acids (CLAs) influence intramuscular fat deposition and muscle fiber transformation in pig models. They have utilized single-nucleus RNA sequencing (snRNA-seq) to explore the effects of CLA supplementation on cell populations, muscle fiber types, and adipocyte differentiation pathways in pig skeletal muscles. Notably, the authors have made significant efforts in addressing the previous concerns raised by the reviewers, clarifying key aspects of their methodology and data analysis.

Strengths:

(1) Thorough validation of key findings: The authors have addressed the need for further validation by including qPCR, immunofluorescence staining, and western blotting to verify changes in muscle fiber types and adipocyte populations, which strengthens their conclusions.

(2) Improved figure presentation: The authors have enhanced figure quality, particularly for the Oil Red O and Nile Red staining images, which now better depict the organization of lipid droplets (Figure 7A). Statistical significance markers have also been clarified (Figure 7I and 7K).

Weaknesses:

(1) Cross-species analysis and generalizability of the results: Although the authors could not perform a comparative analysis across species due to data limitations, they acknowledged this gap and focused on analyzing regulatory mechanisms specific to pigs. Their explanation is reasonable given the current availability of snRNA-seq datasets on muscle fat deposition in other human and mouse.

(2) Mechanistic depth in JNK signaling pathway: While the inclusion of additional experiments is a positive step, the exploration of the JNK signaling pathway could still benefit from deeper analysis of downstream transcriptional regulators. The current discussion acknowledges this limitation, but future studies should aim to address this gap fully.

(3) Limited exploration of other muscle groups: The authors did not expand their analysis to additional muscle groups, leaving some uncertainty regarding whether other muscle groups might respond differently to CLA supplementation. Further studies in this direction could enhance the understanding of muscle fiber dynamics across the organism.

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.

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.

Weaknesses:

Although the authors compiled a substantial and comprehensive dataset, the scope of cellular and molecular-level validation still needs to be expanded. For instance, the single nuclei data suggest changes in myofiber type after linoleic acid supplementation, but these findings need more thorough validation. Further histological and physiological assessments are necessary to address fiber types and oxidative potential. Similarly, the authors propose that linoleic acid alters adipocyte populations, FAPs, and preadipocytes; however, there are limited cellular and molecular analyses to confirm these findings. The identified JNK signaling pathways require additional follow-ups on the molecular mechanism or transcriptional regulation. However, these issues are discussed as potential areas for future exploration. While various individual studies have been conducted on mouse/human skeletal muscle and adipose tissues, these have only been briefly discussed, and further investigation is warranted. Additionally, the authors incorporate two pig models into their results, but they only examine one muscle group. Exploring whether other muscle groups respond similarly or differently to linoleic acid supplementation would be valuable. Furthermore, the authors should discuss how their results translate to human and pig nutrition, such as the desirability and cost-effectiveness for pig farmers and human diets high in linoleic acid. Notably, while the single nuclei data is comprehensive, there needs to be a statement on data deposition and code availability, allowing others access to these datasets.

eLife Assessment

This valuable study addresses potential roles of the master regulator of X chromosome inactivation, the Xist long non-coding RNA, in autosomal gene regulation. Using data from mouse cells, the authors propose that Xist can coat specific autosomal promoters, which in turn leads to the attenuation of their transcriptional activity, complementing recently published results from humans. While the evidence from individual genes is suggestive, shortcomings in the data and statistical analyses leave the evidence currently incomplete. The work would be of interest to anyone studying gene regulation and noncoding RNAA

Reviewer #1 (Public review):

Summary:

The manuscript by Yao S. and colleagues aims to monitor the potential autosomal regulatory role of the master regulator of X chromosome inactivation, the Xist long non-coding RNA. It has recently become apparent that in the human system, Xist RNA can not only spread in cis on the future inactive X chromosome but also reach some autosomal regions where it recruits transcriptional repression and Polycomb marking. Previous work has also reported that Xist RNA can show a diffused signal in some biological contexts in FISH experiments.

In this study, the authors investigate whether Xist represses autosomal loci in differentiating female mouse embryonic stem cells (ESCs) and somatic mouse embryonic fibroblasts (MEFs). They perform a time course of ESC differentiation followed by Capture Hybridization of Associated RNA Targets (CHART) on both female and male ESCs, as well as pulldowns with sense oligos for Xist. The authors also examine transcriptional activity through RNA-seq and integrate this data with prior ChIP-seq experiments. Additional experiments were conducted in MEFs and Xist-ΔB repeat mutants, the latter fails to recruit Polycomb repressors.

Based on this experimental design, the authors make several bold claims:

(1) Xist binds to about a hundred specific autosomal regions.<br /> (2) This binding is specific to promoter regions rather than broad spreading.<br /> (3) Xist autosomal signal is inversely correlated with PRC1/2 marks but positively correlated with transcription.<br /> (4) Xist targeting results in the attenuation of transcription at autosomal regions.<br /> (5) The B-repeat region is important for autosomal Xist binding and gene repression.<br /> (6) Xist binding to autosomal regions also occurs in somatic cells but does not lead to gene repression.

Together, these claims suggest that Xist might play a role in modulating the expression of autosomal genes in specific developmental and cellular contexts in mice.

Strengths:

This paper deals with an interesting hypothesis that Xist ncRNA can also function at autosomal loci.

Weaknesses:

The claims reported in this paper are largely unsubstantiated by the data, with multiple misinterpretations, lacking controls, and inadequate statistics. Fundamental flaws in the experimental design/analysis preclude the validity of the findings. Major concerns are listed below:

(1) The entire paper is based on the CHART observation that Xist is specifically targeted to autosomal promoters. Overall, the data analysis is flawed and does not support such conclusions. Importantly the sense WT and the 0h controls are not used, nor are the biological replicates. Data is typically visualized without quantification, and when quantified, control loci/gene sets are erroneously selected. Firstly, CHART validation on the X in FigS1 is misleading and not based on any quantifications (e.g., see the scale on Kdm6a (0-190) compared to Cdkl5 (0-40)). If scaled appropriately, there is Xist signal on the escapee. All X-linked loci should have been quantified and classified based on escape status; sense control should also be quantified, and biological replicates should be shown separately. Secondly, and most importantly, Figure 1 does not convincingly show specific Xist autosomal binding. Panel A quantification is on extremely variable y-scales and actually shows that Xist is recruited globally to nearly all autosomal genes, likely indicating an unspecific signal. Again, the sense and 0h controls should have been quantified along with biological replicates. Upon inspecting genome browser tracks of all regions reported in the manuscript (Rbm14, Srp9, Brf1, Cand2, Thra, Kmt2c, Kmt2e, Stau2, and Bcl7b), the signal is unspecific on all sites with the possible exception of Kmt2e. On all other loci, there is either a strong signal in the 0h ESC controls or more signal in some of the sense controls. This implies that peak calling is picking up false positive regions. How many peaks would have been picked up if the sense or the 0h controls were used for peak calling? It is likely that there would be a lot since there are also possible "peaks" (e.g., Fzd9) in control tracks. Further inspection of the data was not possible as the authors did not provide access to the raw fastq files. When inspecting results from past published experiments {Engreitz, 2013 #1839} reported regions were not bound by Xist. Thirdly, contrary to the authors' claim, deleting the B repeat does not lead to a loss of autosomal signal. Indeed, comparing Fig1A and Fig2B side by side clearly shows no difference in the autosomal signal, likely because the autosomal signal is CHART background. Properly quantifying the signal with separate replicates as well as the sense and 0h controls is vital. Overall current data together with published results indicate that CHART peak calling on autosomes is due to technical noise or artefacts.

(2) The RNA-seq analysis is also flawed and precludes strong statements. Firstly, the analysis frequently lacks statistical analysis (Fig3B, FigS2B-C) and is often based on visualizations (Fig 3D-G) without quantifications. Day 4 B-repeat deletion does not lead to a significant change in the expression of genes close to Xist signal (Fig3H, d14 does not fully show). Secondly, for all transcriptional analysis, it is important to show autosomal non-target genes, which is not always done. Indeed, both males and B repeat deletion will lead to transcriptional changes on autosomes as a secondary effect from different X inactivation status. The control set, if used, is inappropriate as it compares one randomly selected set of ~100 genes. This introduces sampling error and compares different classes of genes. Since Xist signal targets more active genes, it is important to always compare autosomal target genes to all other autosomal genes with similar basal expression patterns.

(3) The ChIP-seq analysis also has some problems. The authors claim that there is no positive correlation between genes close to Xist autosomal binding (10kb) compared to those 50kb away (Fig 3C, S2D); however, this analysis is based entirely on metagene visualization. Signal within the Xist binding sites should be quantified (not genes close by) and compared to other types of genomic loci and promoters. Focusing on the 50kb group only as controls is misleading. Secondly, the authors only look at PRC mark signal upon differentiation; what about the 0h timepoint, i.e., is there pre-marking? Most worryingly, the data analysis is not consistent between figures (see Fig3C vs 5H-I). In Fig5, the group of Xist targets was chosen as those within 100kb of Xist binding, which would encompass all the control regions from Fig3C. In this analysis, the authors report that there is Xist-dependent H3K27me3 deposition, and in fact, here the Xist autosomal targets have more of it than the controls. Overall, all of this analysis is misleading, and clear conclusions cannot be made.

All in all, because the fundamental observation is not robust (see point 1), all subsequent analyses are also affected. There are also multiple other inconsistencies within the analysis; however, they have not been included here for brevity.

Reviewer #2 (Public review):

Summary:

To follow-up on recent reports of Xist-autosome interaction the authors examine female (and male transgenic) mESCs and MEFs by CHARTseq. Upon finding that only 10% of reads map to X, they sought to identify reproducible alternative sites of Xist-binding, and identify ~100 autosomal Xist-binding sites and show a transient impact on expression.

Strengths:

The authors address a topical and interesting question with a series of models including developmental timepoints and utilize unbiased approaches (CHARTseq, RNAseq). For the CHARTseq they have controls of both sense probes and male cells; and indeed do detect considerable background with their controls. The use of deletions emphasizes that intact functional Xist is involved. The use of 'metagene' plots provides a visual summation of genic impact.

Weaknesses:

Overall, the result presentation has many 'sample' gene presentations (in contrast to the stronger 'metagene' summation of all genes). The manuscript often relies on discussion of prior X chromosomal studies, while the data generated would allow assessment of the X within this study to confirm concordance with prior results using the current methodology/cell lines. Many of the 'follow-up' analyses are in fact reprocessing and comparison of published datasets. The figure legends are limited, and sample size and/or source of control is not always clear. While similar numbers of autosomal Xist-binding sites were often observed, the presented data did not clarify how many were consistent across time-points/cell types. While there were multiple time points/lines assessed, only 2 replicates were generally done.

Aim achievement:

The authors do identify autosomal sites with enrichment of chromatin marks and evidence of silencing. More details regarding sample size and controls (both treatment, and most importantly choice of 'non-targets' - discussed in comments to authors) are required to determine if the results support the conclusions.

Specific scenarios for which I am concerned about the strength of evidence underlying the conclusion:

I found the conclusion "Thus, RepB is required not only for Xist to localize to the X- chromosome but also for its localization to the ~100 autosomal genes " (p5) in constrast to the statement 2 lines prior: "A similar number of Xist peaks across autosomes in ΔRepB cells was observed and the autosomal targets remained similar". Some quantitative statistics would assist in determining impact, both on autosomes and also X; perhaps similar to the quintile analysis done for expression.

It is stated that there is a significant suppression of X-linked genes with the autosomal transgenes; however, only an example is shown in Figure 4B. To support this statement, a full X chromosomal geneset should be shown in panels F and G, which should also list the number of replicates. As these are hybrid cells, perhaps allelic suppression could be monitored? Is Med14 usually subject to X inactivation in the Ctrl cells, and is the expression reduced from both X chromosomes or preferentially the active (or inactive) X chromosome?

The expression change for autosomes after transgene induction is barely significant; and it was not clear what was used as the Ctrl? This is a critical comparator as doxycycline alone can change expression patterns.

In the discussion there is the statement. "Genetic analysis coupled to transcriptomic analysis showed that Xist down-regulates the target autosomal genes without silencing them. This effect leads to clear sex difference - where female cells express the ~100 or so autosomal genes at a lower level than male cells (Figure 7H)." This sweeping statement fails to include that in MEFs there is no significant expression difference, in transgenics only borderline significance, and at d14 no significant expression difference. The down-regulation overall seems to be transient during development while targeting is ongoing?

Finally, I would have liked to see discussion of the consistency of the identified genes to support the conclusion that the autosomal sites are not merely the results of Xist diffusion.

The impact of Xist on autosomes is important for consideration of impact of changes in Xist expression with disease (notably cancers). Knowing the targets (if consistent) would enable assessment of such impact.

Reviewer #3 (Public review):

Summary:

Yao et al use CHART to identify chromatin associated with Xist in female mouse ESCs, and, as control, male ESCs at various timepoints of differentiation. Besides binding of Xist to X chromosome regions they found significant binding to autosomes, concentrating mostly on promoter regions of around 100 autosomal genes, as elucidated by MACS. The authors went on to show that the RepB repeat is mostly responsible for these autosomal interactions using a female ESC line in which RepB is deleted. Evidence is provided that Xist interacts with active autosomal genes containing lower coverage of repressive marks H3K27me3 and H2AK119ub and that RepB dependent Xist binding leads to dampening of expression, but not silencing of autosomal genes. These results were confirmed by overexpression studies using transgenic ESCs with doxycycline-inducible Xist as well as via a small molecule inhibitor of Xist (X1), inducing/inhibiting the dampening of autosomal genes, respectively. Finally, using MEFs and Xist mutants RepB or RepE the authors provide evidence that Xist is bound to autosomal genes in cells after the XCI process but appears not to affect gene expression. The data presented appear generally clear and consistent and indicate some differences between human and mouse autosomal regulation by Xist.

Strengths:

Regulation of autosomal gene expression by Xist is a "big deal" as misregulation of this lncRNA causes developmental defects and human disease. Moreover, this finding may explain sex-specific developmental differences between the sexes. The results in this manuscript identify specific mouse autosomal genes bound by Xist and decipher critical Xist regions that mediate this binding and gene dampening. The methods used in this study are appropriate, and the overall data presented appear convincing and are consistent, indicating some differences between human and mouse autosomal regulation by Xist.

Weaknesses:

(1) The figure legends and/or descriptions of data are often very short lacking detail, and this unnecessarily impedes the reading of the manuscript, in particular the figures would benefit not only from more detailed descriptions/explanations of what has been done but also what is shown. This will facilitate the reading and overall comprehension by the reader. One out of many examples: In Fig S1B in the CHART data at d4 and d7 there is not only signal in female WT Xist antisense but also in female sense control. For a reader that is not an expert in XCI it would be helpful to point out in the legend that this signal corresponds to the lncRNA Tsix (I suppose), that is transcribed on the other strand.

(2) Different scales are used in the lower panels of Figures 1A and 2A, which makes it difficult to directly compare signals between the different differentiation stages.

(3) In this study some of the findings on mouse cells contrast previously published results in human ESCs: 1) Xist binding occurs preferentially to promoters in mice, not in human. 2) Binding of Xist is mostly detected in polycomb-depleted regions in mice but there is a positive correlation between Xist RNA and PRC2 marks in human ESCs. These differences are surprising but may be very interesting and relevant. While I am aware that this might be a difficult task, it would be helpful to experimentally address this issue in order to distinguish whether species specific and/or methodological differences between the studies are responsible for these differences.

Joint Public Review:

Summary:

The authors 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.

Comments on revised version from expert Editor #1:

The authors in the revised manuscript have effectively responded to all of the comments and suggestions raised by both reviewers. Overall, I find the revised version to be an important contribution to the field and the strength of evidence supporting the work's claims to be compelling.

Comments on revised version from expert Editor #2:

The authors have responded constructively to all the comments in the first round of reviews and clarified many issues in the manuscript. The current report represents a significant advance.

Comments on revised version from Reviewer #2:

The authors should include their clarifications of all concern raised by reviewer #2 (mentioned in the previous weaknesses) in the main text. They should consider including point #2 to point #10 in the main text (discussion section). The should highlight limitations of this study in discussion.

Also, they should clearly state that deciphering brain area specific behavioural deficits is beyond the scope of the manuscript with appropriate justification mentioned in the rebuttal letter.

I still do not agree with the author to state that "brain region-specific sensitivities to a defect in Pol III transcription". The changes are global and also not restricted to brain. Authors may consider restating this sentence. It is obvious that transcription defects related to tRNA production will lead to alteration in whole body physiology.

eLife Assessment

This study provides important insights into the mechanistic basis of neurological manifestations of RNA polymerase III-related disease by creating a mutant mouse to dissect transcriptional changes. The data provide compelling evidence for disease progression initiated by a global reduction in tRNA levels leading to integrated stress and innate immune responses and neuronal loss. The work will be of interest to those engaged in the study of chromosome biology, developmental biology and neurodegeneration.

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.