Author Response:

Reviewer #3 (Public Review):

The paper contains a substantial amount of novel experimental work, the experiments appear well done, and the analysis of the data makes sense. Raw data and analysis scripts have been made fully available.

I have two specific comments:

• While the paper talks extensively about deep mutational scanning, I don't think this is a deep mutational scanning study. In deep mutational scanning, we usually make every possible single-point mutation in a protein. This is not what was done here, as far as I can tell.

In the revised manuscript, we have avoided using deep mutational scanning to describe our experimental design. Instead, we described our approach as “a high-throughput experimental approach that coupled combinatorial mutagenesis and next-generation sequencing”

• For the analysis of epistasis vs distance (Fig 4d, e, f), it would be better to look at side-chain distances rather than C_alpha distances. In covariation analyses, it can be seen that C_alpha distances are not a good predictor of pairwise interactions. Similar patterns may be observable here.

See e.g.: A. J. Hockenberry, C. O. Wilke (2019). Evolutionary couplings detect side-chain interactions. PeerJ 7:e7280.

Thank you for the suggestion. In the revised manuscript, we replaced the Cα analysis by a side-chain analysis according to Hockenberry and Wilke (see response to Essential Revisions above).

Author Response:

Reviewer #1 (Public Review):

This manuscript describes single molecule measurements of rotation of the C10 ring of E. coli ATP synthase in intact complexes embedded in lipid nanodiscs. The major point of the work is to identify the mechanisms by which protonation/deprotonation steps produce torque between the a-subunit and the C10 ring, which is subsequently conveyed to F1 to couple to ATP synthesis. The work explores the pH-dependent of the "transient dwell" (TD) phenomenon of rotation motion to identify likely intermediates, showing a likely step of 11o in the "clockwise" (ATP synthase-related) direction. The results are then interpreted in the context of detailed structural information from previous cryo-EM and X-ray crystallographic reports, to arrive at a more detailed model for the partial steps for coupling of proton translocation to motion. The effects of site-specific mutations in the c-subunits appears to support the overall model.

While the detailed structural arguments seem, at least to this reader, to be plausible, the text is not structured for any hypothesis testing, and one might imagine that alternative models are possible. No alternative models were presented, so it is not clear to what extent the 11o rotation step rules out such possibilities. This leaves the reader with the feeling that a lot of speculation occurs in the Discussion, but it is very difficult to figure out which parts are solid and which parts are speculation.

We have now presented the results in terms of hypothesis testing specifying alternative hypotheses that exist in the literature. We then specify results presented in the manuscript that discriminate between alternate hypotheses.

The Discussion also tries to pack in too many concepts, going well beyond the advances enabled by the TD results themselves. For example, the proton "funnel" concept is quite interesting, but it is not easy to see how the TD leads up to it. This overpacking makes it difficult to pinpoint the real advances, and dilutes the message sets the reader up to ask for more support for such extensive modeling. Do the mechanistic details set up good testable hypothesis for future experimental tests?

It is clear that the pKa values we determined are the result of multiple residues involved in the proton transfer process. It is currently not possible to determine where the input channel starts. The recent structures now show that the residues that were thought to define the input channel are far from the surface and must communicate with the periplasm via the funnel. Residues in the funnel likely impact the pKa values that we have measured. The proton transfer-dependent 11 degree step that we measured must also depend upon the funnel. Our results clearly show that this 11 degree step depends upon the correct protonation states of both the input and output channels, and that this depends on the differences in the high and low pKa values. The possibility also exists that this funnel that is absent in the output channel may provide a proton reservoir to supply the input channel, which promotes the ability of input and output channels to drive the 11 degree synthase steps. We have now included this information in the manuscript, and for these reasons, we decided that discussion of the funnel must remain.

Overall, the text would be far more impactful if it focused more tightly on the implications of the TD results themselves, testing specific sets of models, and taking more care to guide the readers through the interpretation.

We have extensively rewritten the entire manuscript to address these issues. To help guide the readers, we added more background information to the introduction and pose alternate hypotheses. In the results, we now guide the readers by restating how the experiments can test a given hypothesis, and include brief conclusions that explain why a hypothesis is eliminated or favored based on the results. We shortened the Discussion to make it more focused, with the exception that we provided additional information that has been requested by the reviewers. We also tie each point in the discussion to the results presented. Of course, a good discussion is meant to put the results and conclusions of the manuscript into the context of results and conclusions from other laboratories, which we have done.

Reviewer #2 (Public Review):

This brilliant, beautiful and important study provides the essential kinetic framework for the recent, static high-resolution cryo-EM structures of F1FO ATPases from bacteria, chloroplasts and mitochondria. The elegantly conducted single-molecule work is necessarily complex, and its analysis is difficult to follow, even for someone who is intimately familiar with F1FO ATPases. Some more background and better explanations would help.

We added additional background information to the Introduction, and we now periodically explain the reasoning and conclusions in the Results to help guide the readers.

For F1FO ATPases, CCW rotation has little if any biological relevance, whereas CW rotation is centrally important. Evidently, the CW ATP synthesis mode is not accessible to the approach taken in this manuscript, since the ATP synthase is reconstituted into lipid nanodiscs rather than liposomes. This critical fact should be stated more clearly in the introduction.

We now state explicitly that net rotation was observed as the result of F1-ATPase activity as requested. We also note that E. coli does sometimes use F1Fo as an ATPase-dependent proton pump to maintain a pmf across the membrane depending upon metabolic conditions.

The central concepts of "transient dwells", "dwell times" and "power strokes" need to be introduced more fully for a general, non-expert audience.

We added this information to the Introduction as requested.

The manuscript describes the power stroke and dwell times in CCW ATP hydrolysis mode in unprecedented detail. Presumably the dwell times and power strokes apply equally to the physiologically relevant CW ATP synthesis mode, but are they actually the exact reverse? Is there evidence for transient dwells and 36{degree sign} power strokes divided into 11{degree sign}+25{degree sign} substeps during ATP synthesis?

The 36° subunit-c stepping that contain 11° synthase-direction steps is a novel observation first reported in this study. To date, single-molecule studies of rotation during net ATP synthesis have been carried out using single-molecule FRET that have been able to observe only 3 or 4 consecutive synthase steps for a given F1Fo molecule (Dietz et al. ((2004) Proton-powered subunit rotation in single membrane-bound FoF1-ATP synthase. Nature Struct & Mol Bio). The FRET measurements do not have the time resolution to resolve sub-steps. Whether or not continuous rotation in the synthesis direction is the exact reverse of ATPase-dependent rotation is an important question that remains to be answered.

The meaning of low, medium and high efficiency of transient dwell formation (Figure legend 2; lines 189/190; Figure 3; line 365) is not obvious and not well explained. How are these efficiencies defined? Why are they important? What would be 100% efficiency? And what would be 0%?

Background information concerning the three efficiencies of transient dwell formation has been added to the Introduction, and we now explain their importance. We also now explain what 100% and 0% efficiency is in the Results.

Why is it important whether transition dwells do or do not contain a synthase step? Is this purely stochastic? If not, what does it depend on?

We added a paragraph to the discussion to explain that their formation depends on the kinetics of the rate of formation of the interaction between subunit-a and the c-ring versus the velocity of ATPase-depending rotation in the opposite direction, and that it depends on the energy that can drive the synthase-direction step relative to the energy that drives the ATPase direction power stroke. More work is required to define the energetic parameters of these opposing rotations that is beyond the scope of the work presented here.

The formation of a salt bridge between aR210 of subunit-a and cD61 of the c-ring rotor would seem to be counter-productive for unhindered rotary catalysis. What is the evidence for such a salt bridge from the cryo-EM structures or molecular dynamics simulations?

This is an excellent question, especially since the distances between aR210 and cD61 are more consistent with intervening water molecules. We revised the paragraph in the Discussion describing this point and have been more explicit about the importance of the aqueous vestibule between the output channel and aR210 must play during rotation, which includes the impact of the dielectric constant inside the vestibule. As a direct answer to the reviewer’s question, in the absence of water, a salt bridge between aR210 and cD61 in such a hydrophobic environment would be so strong that the energy of a proton from the input channel would never be able to dislodge them.

Reviewer #3 (Public Review):

Yanagisawa and Frasch utilise a gold nanorod single molecule method to probe the pH dependency of F1FO rotation. The experimental setup has been previously used to investigate both F1-ATPase and FO function in multiple studies. In this study, clockwise rotations are observed in transient dwells which may correlate to synthesis sub-steps. Mutations along the proposed proton path modify the pH dependency of the transient dwells.

The strength of this manuscript can be seen in the rigorous way in which the problem has been explored. Testing the pH dependence of mutants along the proposed proton path and linking this to potential sub-steps using the known atomic structure.

In my view, the main weakness of this study is the experimental design (shown in Fig. 1C). Strictly, the measurements show rotation of the c-ring relative to subunit-Beta rather than relative to subunit-a. Recent structures of E. coli F1FO ATP synthase inhibited by ADP (doi: 10.1038/s41467-020-16387-2) have shown that the peripheral stalk is flexible and can accommodate movements of the c-ring relative to the F1 (AlphaBeta)3-subunit ring. For example, comparison of PDB entries 6PQV and 6OQS shows that FO (the c-ring and subunit-a) can rotate 10 degrees as a rigid body relative to the F1 (AlphaBeta)3-subunit ring - with no relative rotation between the c-ring and subunit-a, or rotation of subunit-gamma. The authors discuss structures from this study related by a 25 degree rotation of the c-ring relative to subunit-a, but I do not believe they have ruled out the possibility that their observations show rotation of the FO as a rigid body. A preprint investigating E. coli F1FO ATP synthase in the presence of ATP has proposed that the complex becomes more flexible during ATP hydrolysis (doi: 10.1101/2020.09.30.320408), with the central stalk twisting by up to 65 degrees. The small CW movements seen in the transient dwells in this study could be attributed to 36 degree FO sub steps, facilitated by central stalk flexibility, with counter rotation facilitated by peripheral stalk flexibility.

The data clearly show that the mutations of subunit-a residues in the input or output channels significantly change the pKa values of TD formation (Figs 2B and 2C), and can dramatically change the occurrence of the synthase-direction steps (see Figs 4D and 4E). These results clearly indicate that the rotational events observed in this study do not result from rotation of subunit-a and the c-ring as a unit.

With regard to the recent structures that the reviewer refers to, we report differences in the efficiency of TD formation that are consistent with torsion induced by rotation of the c-ring relative to the beta subunit, which we reported previously (Yanagisawa and Frasch, JBC 2017), and which has been confirmed by independent single-single molecule studies by the Junge lab and by the Boersch lab using different approaches to our own (Sielaff et al., Molecules 2019). Both papers are cited in the manuscript. We have now expanded the introduction to include these results describing the impact of central stalk flexibility on the ability to form synthase-direction steps, and how these results are consistent with E. coli cryo-EM structures similar to those referred to (27).

It is also unclear what causes the stochastic nature of transient dwells. Are these related to inhibition of F1-ATPase? Could increased drag in FO increase the likelihood of F1-ATPase inhibition?

We now include background information from our prior publications that characterizes the kinetic component that affects the ability to form a transient dwell. Ishmukhametov et al. EMBO J (2010), reported an increase in TDs upon increasing the drag on the nanorod that slowed the power stroke angular velocity. We decribed the kinetics of TD formation in that paper. In the Discussion, we also now provide information concerning how the bioenergetics can impact the probability of TD formation.

Author Response:

Reviewer #3 (Public Review):

[...]Overall, the quality of the RNAseq data seems sound, and the conclusions presented seem mostly supported by the data. Additionally, the manuscript is well written and easy to read.

We are thankful to hear that the quality of the paper convinces the reviewer.

The spatial profiling of Physarum by physically segregating it by centrifuging it into a 384-well plate is clever. While the approach is probably cannot be generalized to most organisms, it still provides a nice example of creative experimental design that is somewhat lacking in the single-cell genomics field at the moment. Moreover, given that there seem to be no/few published studies with RNA in situ hybridization gene expression patterns in this animal, it probably provides a wealth of information to Physarum researchers.

We are glad to hear that the reviewer finds our experimental design clever and we also believe that our manuscript is a great resource for the Physarum but potentially also the whole protozoan and fungi community.

Some aspects of the experimental design potentially limit the conclusions that can be drawn from the data. The authors find that plasmodia in distinct states of life (mitotic, non-mitotic, chemotaxing, contacting food) have broad syncytium-wide transcriptional differences. A major caveat of this finding is each separate condition was only profiled once without replicates, which makes it more difficult to tie which of these transcriptional differences are related to the samples' biological differences and which might be a batch effect.

We agree with the reviewer on that point and try in our revised manuscript to also highlight similarities between replicate samples (SM1/2) and similar plasmodium parts (fans) in Fig. 2- Fig. supp. 1D-F and not only the differences in order to show that there are also overlaps suggesting that differences are not driven by batch. However, some of our observations might have some effect due to batch, and we have been careful in the revised manuscript to point this out. We feel that the methods that we have established in this manuscript can be employed in higher throughput in the future in order to sample multiple environmental conditions across multiple batches. Before our work, it had been entirely unclear if there is transcriptional heterogeneity within the syncytium and this is one of our major findings, which we feel is a very robust, interesting, and important finding.

Additionally, it's not clear why the authors profiled different timepoints via snRNAseq (1 week with oat flake) and spatial RNAseq (only a few hours with oat flakes) in their experiment to assess feeding behavior.

We thank the reviewer for the comment. The reason for not having the same time point is indeed not ideal but happened for technical reasons. The main goal for the snRNA-seq data acquisition was to obtain a rough spatial separation of fan and network parts which we believe is important to link the nuclei data to the spatially resolved data as shown in Fig. 3 (see point below). In order to obtain enough material, we needed to grow the plasmodium longer, whereas we performed all the Spatial Transcriptomics experiments in a more condensed time frame to keep at least these experiments as comparable as possible.

While the work identifies spatial heterogeneity and nuclear heterogeneity, they are not directly compared (how much of the nuclear heterogeneity is explained by spatial heterogeneity?) perhaps because different timepoints were used with the two approaches?

We apologize that this point was not conveyed clearly in our manuscript. We do try to draw a direct link between the nuclear and the spatial heterogeneity in Fig. 3G where we correlate the transcriptomes of the clusters of SM4 with the clusters identified for the secondary plasmodium. Strikingly, the fan specific cluster in SM4 is highest correlated with the nuclei cluster that is strongly enriched with nuclei from the fan sample allowing to draw a direct link between the nuclei and the spatially resolved data. We rewrote the corresponding paragraph accordingly. This also highlights that age has obviously rather a minor impact for the comparison. However, we agree that there are technically more direct techniques available like measuring the mRNA in situ (e.g. SeqFish or OsmFish) which are, unfortunately, being far away from being routinely established in Physarum.

Additionally, some aspects of the analysis seem to miss opportunities. A significant portion of the presentation of the gene expression results discovered by the authors is focused on the cell cycle, which seems less exciting than perhaps other biological phenomena related to structural specialization within different parts of the organism or related to its feeding and metabolic behaviors might be.

We are thankful for the reviewer’s suggestions and added new panels to Fig. 2- Fig. supp. 1D-F that focus on similarities between and within samples to emphasize other biological phenomena like common transcriptomic fan signatures. In line with this, we performed additional GO enrichment analyses (Fig. 2- Fig. supp. 3 and Fig. 3-Fig. supp. 1) which, for instance, also highlight GO enrichments in dependence on nutrient interaction in addition to our findings regarding cell cycle progression in Physarum. We agree that there are a lot of very interesting biological phenomena that can be explored with Physarum, however, we also think that regulation of cell cycle within the syncytium and in free-living cells is also interesting and important.

Also, while multiple classes of nuclei (stationary and mobile) are identified, it's unclear how those relate to the different transcriptional states identified through the snRNAseq.

We thank the reviewer for making this point and we would also like to understand differences between stationary and mobile nuclei. We agree that the next step is to find ways how to specifically label, track and isolate moving/stuck nuclei in order to understand how the different classes of nuclei are linked to the transcriptomic differences. Unfortunately, this is not yet possible but has to be the goal for future research.

Lastly, some aspects of the presentation detract from the work. Some of the results and discussion focus on 'coordinated intra-syncytial behaviors', and a major one of focus is that a wave of mitosis seems to proceed across the organism. However, to my knowledge, many syncytial systems (e.g. Xenopus or Drosophila embryos) exhibit synchronized mitosis, so I would have expected this to be the default state, rather than an exciting finding. Is this result unexpected? If so, it would be helpful if better contextualized. One aspect that would likely improve this manuscript would be to place it more firmly within the larger context of well-studied syncytial cells that exhibit specialization. For instance, a major example of a well-studied syncytium that exhibits spatial gene expression and nuclear specialization is the Drosophila embryo, which undergoes much of its early patterning while syncytial. Furthermore, muscle cells are typically syncytial, and some exciting recent studies have similarly used snRNAseq to observe heterogeneity and specialization of particular nuclei.

We are thankful for that comment and contextualize this finding now more clearly in our Discussion. Briefly, synchronous cell division is not the default state in acellular slime molds and also not for similarly functioning systems like fungi hyphae. However, the reviewer is right that the synchronized division is not a new finding we made neither for Physarum nor for other syncytial systems in general. Our research, however, adds to this view that such a synchronization wave can be established over a large distance in Physarum (much larger than in ‘standard’ developmental biology systems like frogs and flies) while nuclei are in addition in a continuous shuttle flow which makes it a unique model system compared to the different nuclei states that are established in more rigid systems like muscles and embryos. In addition, a link between mode of nuclei division and plasmodial size and age exists but is not yet understood emphasizing its uniqueness as model system.

Lastly, while not meant to diminish the contributions of this work, it does seem that given the diversity of syncytia that are well studied and exhibit nuclear specification, perhaps the title "Nuclei are mobile processors enabling specialization in a gigantic single-celled syncytium" oversells the results presented in this work.

We thank the reviewer for this point and changed the title of the manuscript to avoid overselling and to more accurately reflect the research presented in the manuscript. The new title is: “Spatial transcriptomic and single-nucleus analysis reveals heterogeneity in a gigantic single-celled syncytium.” We agree that more work is needed to finally prove that mobile nuclei can be used to more quickly ‘seed’ new transcriptomic states in other plasmodial parts, thereby acting as mobile processors.

Author Response:

Reviewer #1 (Public Review):

The authors extend their previous work on thymus epithelial cells (TECs) and antigen presenting cells (APCs), which focused on TLR signaling in TECs and monocyte-derived dendritic cells (mDCs), and now focus on medullary (m) TECs and ask whether different subsets of DCs can uniquely serve as APCs for tissue-restricted antigens (TRAs) expressed by mTECs.

The approach used makes use several reporter transgenic mouse models to conditionally express a fluorescent reporter gene in mTECs (Defa6-cre) or in TECs more broadly (Foxn1-cre or Csnb-cre). Their findings show that restricted expression of reporter genes in a subset of mTECs, which typically expressed AIRE-dependent TRAs, are typically presented by a subset of DCs that express XCR1 and CCR7, which, together with mDCs, are able to perform cooperative antigen transfer (CAT) effectively. The authors also examined the ability of different DC subsets to acquire antigens from other DCs, and show that mDCs excel in this ability.

Overall, the work is clearly presented and makes use of several elegant mouse models to further delineate the role of different DC subsets in T cell selection. However, as pointed out by the authors, it remains to be determined whether the differences in the ability of different DC subsets to perform CAT has any fundamental impact in the establishing self-tolerance, either by negative selection or induction of Treg differentiation.

We thank Reviewer #1 for this positive comment. We concur that describing a direct effect of the preferential pairing in CAT between TECs and DCs on the mechanisms of central immune tolerance would be very insightful and beneficial for the scientific community. However, a better understanding of the rules underpinning the interactions of this dynamic thymic cellular network in its complex and physiological outcomes will require the development of novel cellular tools and organismal genetic models, which is beyond the scope of this current work.

Reviewer #3 (Public Review):

In this manuscript, Voboril et al. address the question of whether specific dendritic cell (DC) types in the thymus acquire antigens from distinct thymic epithelial cell subsets. It is well-documented that both medullary thymic epithelial cells (mTECs) and thymic DCs present self-antigens to thymocytes to induce central tolerance. mTECs express the majority of the proteome, including AIRE-dependent tissue restricted antigens (TRAs), and thymocytes must be tolerized against this diverse set of self-antigens to prevent autoimmunity. While some self-antigens are expressed abundantly in an AIRE-independent manner by mTECs, a given AIRE-dependent TRA is expressed at low levels by only 1-3% of mTECs, raising the question of how thymocytes encounter rare, sparse self-antigens during their residence in the medulla. Part of the answer to this comes from the fact that thymic DCs can acquire self-antigens from mTECs to improve the efficiency of display to thymocytes. As the authors point out, several recent studies have utilized single-cell transcriptional profiling to identify multiple distinct medullary thymic epithelial cell subsets. Furthermore, work from the authors' lab and others has demonstrated heterogeneity within the thymic DC compartment. Thus, the authors set out to address whether DC interactions with mTECs are promiscuous or whether specific DC cell types interact with different mTEC subsets to acquire self-antigens to induce tolerance to different types of self-antigens, such as Aire-dependent versus Aire-independent TRAs or ubiquitous antigens.

Using mouse strains that express a fluorescent protein in TECs under the control of different promoters, the authors use flow cytometry to determine the relative ability of thymic DC subsets to acquire self-antigens, TdTomato in this case, from TEC subsets. Using linear regression modeling to compare the frequency of TEC subsets expressing TdTomato in the different reporter strains to the frequency of DC subsets that acquired TdTomato, the authors conclude that there is specificity to the interactions of different DC subsets with distinct mTECs, resulting in antigen acquisition by the interacting DCs. Specifically, they conclude that pDCs and macrophages acquire antigens from mTEClow cells, cDC1 and activated XCR1+ DCs acquire antigen from mTEChigh cells, activated XCR1+ and activated XCR1- DCs acquire antigen from pre-post-Aire mTECs, cDC2 exclusively acquire antigen from Post-Aire mTECs, and activated XCR1+ DCs acquire antigen from Tuft cells. It is well documented that activated XCR1+ DCs (Ardouin et al. 2016, Oh et al. 2018, Perry et al. 2018) and activated XCR1- DCs (Leventhal 2016) acquire and present Aire-dependent self-antigens from mTECs to induce thymic central tolerance. Thus, the most novel claim is that cDC2 acquire antigens from Post-Aire mTECs. However, given that the model is derived from the linear regression analysis of fluorescent reporter mice in which multiple TEC subsets express each reporter, and there are some known caveats to these analyses, this interesting conclusion is not adequately supported by the data. The authors cleverly use Foxn1-cre ConfettiBrainbow2.1 reporters to conclude that moDCs are particularly adept at acquiring antigen serially from different mTECs. Furthermore, they use mixed bone marrow congenic/reporter mice to demonstrate that moDC are particularly good at acquiring antigens from other DC subsets. These two conclusions are well-supported by the data, although it is notable that while moDC are efficient at these two processes, other DC subsets acquire antigens from DCs, and fluorescent reporter acquisition does not indicate the ability to process and present antigens to developing T cells to promote central tolerance, somewhat reducing the impact of the findings. Altogether, this is a promising study that cleverly uses a variety of mouse models with flow cytometric analysis of recently identified TEC and DC subsets to delve into whether specificity of interactions between TEC and DC subsets could enable distinct DC subsets to contribute differentially to central tolerance induction against distinct types of self-antigens. However, the conclusions could be strengthened by additional analyses.

We thank the Reviewer for the many comments and suggestions. We are well aware of the fact that our model is based on the linear regression analysis that brings about some known caveats, which have been now newly described in the Discussion section. As suggested by the Reviewer, to better resolve some discrepancies and strengthen our conclusions, we conducted a novel analysis of our data and made some vital changes to the manuscript.

Author Response:

Reviewer #1 (Public Review):

Mulholland et al show that there is a very close relationship between the development of excitatory and inhibitory networks in the developing cortex. This paper makes an important contribution to our understanding of the structure of inhibition during an early stage in cortical development. The work has been carefully performed and analysed. The changes I suggest are principally to improve clarity in some places.

Lines 48-55: express these possibilities more didactically as individual items in a list, rather than grouping the first two together.

We thank the reviewer for this suggestion and have revised our manuscript accordingly.

Fig 2: Show a raw image of the imaging window, with a scale bar.

We thank the reviewer for the suggestion, and we have added a raw image of the imaging window to Figure 2.

Line 81: clarify the type of imaging used (both wide-field and 2-photon are mentioned in the Introduction).

The experiments in Figure 2 were performed using wide-field calcium imaging. We have now clarified this point in the main text and figure legend.

Line 89: This is presumably in a different set of animals: make this clearer. n value?

The reviewer is correct that the animals in Figure 3 are a different experiment from those in Figure 2. We have clarified the text, and now more clearly state the number of animals in these experiments.

Fig 3b: "Example mean trace…"

We have edited the figure legend to incorporate this suggestion.

Fig 3e: Is this for one animal or averaged over all? Why not show negative correlations as well?

Figure 3e shows the values of correlation maxima for a single representative experiment shown in panels (b-d, f). We quantified the spatial extent of correlations by measuring the amplitude of positively correlated peaks in the correlation pattern as a function of distance from the seed point, and therefore only plotted the values of these peaks in Figure 3e.

We have revised the axis label for this panel to more clearly indicate that we are plotting the values of correlations at local maxima (New label: “Correlation at maxima”).

Fig 3f: This appears to be the same image as S1c?

The reviewer is correct. Our original Figure S1 (now Figure S2) aims to elaborate on the description of correlation fractures shown in Figure3f, and thus reproduces the fracture image shown previously with a more detailed explanation. We have revised the legend for Figure S2c to clearly indicate that this panel is reproduced from Figure 3f.

Line 119-120: Suggest you explain more clearly that this is a preliminary step towards the later simultaneous E-I imaging, and why it's still useful given that it's somewhat indirect compared to the later simultaneous imaging.

We thank the reviewer for the suggestion, and have revised the manuscript to better convey the rationale behind imaging spontaneous inhibitory and excitatory activity in separate animals.

Line 124: Why tenth of maximum? Can you confirm that the main results are not highly sensitive to this choice?

We selected the tenth of the maximum of the fitted two-dimensional gaussian as a way to estimate the diameter of the active pixels within an event domain. When using an alternative threshold, such as the full-width at half maximum, the values for inhibitory and excitatory domain size scale together, indicating that the relationship between the relative size of inhibitory and excitatory active domains is not sensitive to this threshold. However, we found that full-width half max underestimated the diameter of the active region, leaving out active pixels. We have now included the values for full-width half max in the text of the results, to show that the threshold does not impact the relationship between inhibitory and excitatory domain size.

Fig 4: It would be clearer to give the graph parts of a and d their own panel labels. Is it worth explicitly flagging that none of the E-I distributions are significantly different? It would be useful to explicitly define pink and blue in the caption. Does panel e have units? In panel a I'm confused about what is being referred to as "circles" and what as "lines". Panel f caption: "principal components"

We have incorporated these suggestions into a revised figure. We have included bars to indicate non-significance of E-I distributions, updated panel e and f figure axes, and edited the figure legend to provide more explicitly define inhibitory and excitatory data color scheme.

Line 143: suggest "similarity in the statistical structure", to make it clearer what you mean by "structure".

We have revised the text to explicitly refer to “similarity in the spatial structure” in this paragraph.

Line 159: state the n value here.

We have updated the text to include the number of animals used in this particular experiment.

Fig 5: It would be interesting to speculate about the significance of the correlation fractures. In the rhs of panel b the red seed points are almost impossible to see, since they occur on top of a red patch (perhaps include a black or white border to these circles?). a caption: "-2-2 z-score" is confusing; perhaps say "-2 to +2 z-score". g caption: Is this correlation similarity for E, I or both?

Fractures show areas in the network where there are sharp discontinuities in the spatial patterns of the correlations. As discussed in the results, mature functional maps in the visual cortex also exhibit smooth variation punctuated by abrupt discontinuities, such as pinwheels and fractures in orientation preference maps (Bonhoeffer and Grinvald, 1991), or direction reversals in direction preference maps (Okai et al., 2005). Previous work demonstrated that correlation fractures derived from spontaneous activity after eye opening precisely coincided with the highrate-of-change regions in the orientation preference map (Smith et al., 2018). Therefore, the correlation fractures observed from spontaneous inhibitory activity prior to eye opening reveal that the underlying network is already precisely organized, and while the fractures are refined over development (Smith et al 2018), they may provide insight into the future structure of the mature orientation preference map, although this remains to be directly tested.

We thank the reviewer for the suggestions to improve the clarity of this figure. We have changed the color of the seed points, to make them easier to see, and have adjusted the figure legends for Panel a. In Panel g, the correlation similarity is comparing E vs I networks as a function of distance, and the figure legend has been updated to more explicitly state this.

Author Response:

Reviewer #1 (Public Review):

The manuscript is clearly written, the data are largely of sufficient quality, and the findings are certainly of interest to the endosomal research community. I also agree with the model the authors propose.

We thank the reviewer for their appreciation of this study.

One weak point of the study is the dependence on microscopic techniques to analyze integrin surface levels and endosomal recruitment of the retriever complex. The study would have benefited from additional methods to confirm the microscopy data.

We agree. We have now added an experiment showing the changes in the surface levels of β1-integrins by flow-cytometry. This technique allowed us to easily sample 10,000 cells per experiment (New Figure 1G).

It would also be good if the authors could confirm their inhibitor studies with genetic suppression/deletion of PIKfyve, ideally followed by rescues with a kinase deficient mutant.

We agree. We now include experiments where we depleted PIKfyve with siRNA and rescued PIKfyve by exogenous expression of siRNA resistant PIKfyve. Depletion of PIKfyve caused a decrease in surface levels of β1- and α5-integrin by 21% and 17% respectively (Figure S3), which is very similar to what we observed by immunofluorescence.

Later in the manuscript, we also tested changes in the endosomal localization of COMMD1 due to PIKfyve inhibition (Figure 7), which resulted in a 22% decrease. We now include new experiments showing PIKfyve depletion. For the depletion studies, we also tested rescue by PIKfyve expression. PIKfyve depletion resulted in a statistically significant but modest lowering of COMMD1 endosomal localization by 15%. This decrease was rescued back to basal levels on endosomes by re-expression of PIKfyve (Figure S8).

Since PIKfyve has so many roles, and since the PIKfyve inhibitor has been shown to be specific by so many labs, we consider acute PIKfyve inhibition preferable. Long-term depletion studies are more likely to result in effects are indirect or provide cells with a chance to adapt.

The authors rely on microscopy of integrin beta 1 for most of their data. However, SNX17 and the retriever complex are not required for the recycling of all beta 1 integrins (Steinberg et al., 2012). In HeLa cells, it is mainly integrin alpha 5/beta1 that is recycled by SNX17. Therefore, the other beta 1 integrins that recycle SNX17 independently tend to mask the recycling phenotypes caused by the loss of SNX17/retriever. I think that the authors could detect a much more pronounced recycling phenotype upon PIKfyve inhibition if they stained integrin alpha 5 instead of integrin beta 1. Does integrin alpha 5 "get stuck" in a LAMP1 positive compartment similar to what Steinberg et al., 2012 or McNally et al., 2017 describe in their studies? These two studies clearly show that almost all integrin alpha 5/beta 1 accumulates in a LAMP1 or LAMP2 positive compartment upon loss of retriever/SNX17 function. If the authors are correct in their assumptions, this should be happening upon loss of PIKfyve activity. One could use the Abcam antibody against integrin alpha 5 that was used in the McNally et al. study as it works very well.

We did not see pronounced differences in surface levels of α5-integrin vs. β1-integrin upon inhibition or depletion of PIKfyve. For example, PIKfyve siRNA treatment lowered the surface levels of α5-integrin and β1-integrin in the similar range by 21% and 17 % respectively, as measured by immunofluorescence microscopy.

Author Response:

Reviewer #1:

Summary:

Moody et al. presented a comprehensive investigation into the choice of marker genes and its impact on the reconstruction of the early evolution of life, especially regarding the length of the branch that separates domains Bacteria and Archaea in the phylogenetic tree. Specifically, this work attempts to resolve a debate raised by a previous work: Zhu et al. Nat Commun. 2019, that the evolutionary distance between the two domains is short as estimated using an expanded set of marker genes, in contrast to conventional strategies which involve a small number of "core" genes and indicate a long branch.

Through a series of analyses on 1000 genomes, Moody et al. defended the use of core genes, and reinforced the conventional notion that the inter-domain branch (the AB branch) is long, as inferred by the core gene set. They proposed that with the 381 marker genes (the "expanded" set) used by Zhu et al., the observed short branch length is an artifact due to inter-domain gene transfer and hidden paralogy. Through topology tests, they ranked the markers by "verticality", and showed that it is positively correlated with the AB branch length. They also conducted divergence time estimation and showed that even the most vertical genes led to an implausible estimate of the origin of life.

In parallel, Moody et al. surveyed the best marker genes using a set of 700 genomes. They recovered 54 markers, and demonstrated that ribosomal markers do not indicate a longer AB branch than non-ribosomal markers do. With the better half (27) of these marker genes, they conducted further phylogenetic analyses, which shows that potential substitutional saturation and the use of site-homogeneous models could contribute to the underestimation of the AB branch. Using this taxon set and marker set, they reconstructed the prokaryotic tree of life, which revealed a long AB branch, a basal placement of DPANN in Archaea, and a derived placement of CPR in Bacteria.

Prokaryotic tree of life:

The scope(s) of the manuscript is somehow split. First, it is posed as a point-to-point rebuttal to the Zhu et al. paper, on the long vs. short AB branch question. Second, it introduces a new phylogeny of prokaryotes using 27 "good" marker genes, and demonstrates that DPANN is basal to Archaea, and CRP is derived within Bacteria.

Thanks for the summary. The two aspects of the manuscript identified by the reviewer are closely related, because the different issues boil down to the same underlying question: which genes should we use to infer the deep structure of the tree of life? The provocative work of Zhu et al. acted as an impetus to compare and evaluate the properties of several published marker gene sets, and then to identify (what our analyses suggest are) the subset best-suited for deep phylogeny, which we then use to infer an updated tree of life. We have clarified this logical structure in the revised manuscript, writing (at the end of the Introduction):

“Here, we investigate these issues in order to determine how different methodologies and marker sets affect estimates of the evolutionary distance between Archaea and Bacteria. First, we examine the evolutionary history of the 381 gene marker set (hereafter, the expanded marker gene set) and identify several features of these genes, including instances of inter-domain gene transfers and mixed paralogy, that may contribute to the inference of a shorter AB branch length in concatenation analyses. Then, we re-evaluate the marker gene sets used in a range of previous analyses to determine how these and other factors, including substitutional saturation and model fit, contribute to inter-domain branch length estimations and the shape of the universal tree. Finally, we identify a subset of marker genes least affected by these issues, and use these to estimate an updated tree of the primary domains of life and the length of the stem branch that separates Archaea and Bacteria.”

The second scope has inadequate novelty. A recent paper (Coleman et al. Science. 2021), which was from a partially overlapping group of authors, was dedicated to the topic of CPR placement, and indicated the same conclusion (CPR being derived and sister to Chloroflexi) as the current work does, albeit using more sophisticated approaches. The paper also addressed the debate of CPR placement (including citing the Zhu et al. paper). Additionally, the basal placement of DPANN has also been suggested by previous works (such as Castelle and Banfield. Cell. 2018). Therefore, re-addressing these two topics using a largely well-established and repeatedly adopted method on a relatively small taxon set does not constitute a significant extension of current knowledge.

We disagree. Resolving the deep structure of the tree of life is an important topic --- this is what we, Zhu et al. (2019), and of course many others have been trying to achieve, in different and sometimes conflicting ways. Most of the published work is based on limited or biased taxon sampling (see Figure 1 Figure Supplement 14,15,16) or else focused on just one of the two prokaryotic domains of life. Furthermore, deep phylogeny is uncertain, and new results become convincing only when they receive support from multiple datasets and approaches. For instance, Coleman et al. (2021) recently found support for the placement of CPR as a sister clade to Chloroflexota rather than as a basal branch within the Bacteria. Notably, this work focused only on Bacteria, and made use of a different rooting method (with its own strengths and limitations) and taxon sampling. Most previous analyses using Archaea as an outgroup to root the bacterial tree recovered CPR as a deeply branching lineage within Bacteria, a placement likely resulting from LBA. In turn, our present findings represent an important confirmation of the CPR+Chloroflexi clade. Similarly, the basal placement of DPANN within Archaea remains controversial despite a number of studies on the topic, and our study also contributes to that ongoing debate.

The debate:

The first scope appears to be the more important goal of this manuscript, as it extensively discusses the claims made by Zhu et al. and presents a point-to-point rebuttal, including counter evidence. This may narrow the interest of this work to a small audience of specialists. Nevertheless, to best evaluate the current work, it is necessary to review the Zhu et al. paper and compare individual analyses and conclusions of the two studies.

In doing so, I found that the two articles have distinct scopes that appear similar but not actually inline. To a large extent, the current work does not constitute actual rebuttal to the points made by Zhu et al. In contrast, some of the analyses presented in the current work support those by Zhu et al., despite being interpreted in a different way. For the claims that directly contest Zhu et al., I do not see sufficient evidence that they are supported by the analyses.

Below is a summary of the comparison, which I will explain point-by-point in later paragraphs.

• Moody et al. assessed AB branch length, while Zhu et al. assessed AB evolutionary distance (which is different).
• Moody et al. evaluated the phylogeny indicated by a small number of core markers, while Zhu et al. evaluated the genome average using hundreds of global markers.
• Zhu et al.'s results also showed that gene non-verticality, substitutional saturation, and site-homogeneous models shorten the AB distance, which is consistent with Moody et al.'s.
• However, Zhu et al. found that some core markers are outliers in the genome-wide context, and the long AB distance indicated by them cannot be compensated for by the aforementioned effects. Moody et al. hasn't addressed this. Therefore, the novelty and potential impact of the current work is less compelling: It used a classical method (a few dozen core genes) and found a pattern that has been found many times by some of the same authors and others (including Zhu et al., who also analyzed core genes).

Thanks for this detailed comparison of the two studies --- the points raised here and elaborated on below have prompted us to perform additional analyses which provide further insight into the properties and behaviour of the various marker gene sets analyzed. We nonetheless disagree that “the current work does not constitute actual rebuttal to the points made by Zhu et al.”: our finding that ribosomal and other “core” proteins are among the best phylogenetic markers for resolving both within- and between-domain relationships, estimating the length of the AB stem, and performing divergence time estimation, challenges an important claim of Zhu et al.’s study, and will be of broad interest to the community of researchers working on early life/early evolution.

That said, we do also agree that one aspect of the disagreement between our study and that of Zhu et al. has to do with what is meant by evolutionary distance, and we have now discussed these issues in detail in the revised manuscript (as detailed below). In revising the manuscript, we have also sought to avoid a reductive focus on rebuttal, have revised the text to acknowledge important strengths and interesting features of the Zhu et al. analyses, and have made text revisions to ensure a consistent constructive tone: these are fundamental and challenging questions, and different perspectives and analyses are valuable in making progress. We also note that there has been an ongoing debate about the suitability of ribosomal genes for deep phylogeny in the literature (e.g. Petitjean et al. 2014, discussed in more detail below). Our analyses, and those of Zhu et al. (2019) previously, contribute to that broader discussion.

Detailed responses to each of the above points follow below.

AB distance metric:

There is a subtle but critical difference between the scopes of the two papers: The Zhu et al. paper "reveals evolutionary proximity between domains Bacteria and Archaea". By stating "evolutionary proximity", it investigated two metrics: The length of the branch separating Archaea from Bacteria in the phylogenetic tree, i.e., the "AB branch". This was the main focus of the current work.

The average tip-to-tip distance (sum of branch lengths) between pairs of Archaea and Bacteria taxa in the tree. A significant proportion of the Zhu et al. work was discussing this metric, and it led to several important conclusions (e.g., Figs. 4F, 5). The current work has not explored this metric.

Thanks for raising the point about relative AB distance. In our revised manuscript, we have expanded Figure 1 and the associated analyses to include this metric. These analyses demonstrate that relative AB distance behaves similarly to AB branch length: they are positively correlated with each other; both are reduced by inter-domain HGT, and both are negatively correlated with ΔLL and with split score, an additional metric of within- and between-domain marker gene verticality which we have included in the revised Figure 1. Taken together, these results suggest that high-verticality marker genes (as judged both by the recovery of reciprocal AB monophyly, and of established within-domain relationships) support a longer AB branch and show a higher relative AB distance.

These two metrics implicate distinct research strategies: For 1), HGTs and paralogy are usually considered problematic (as the current and many previous works argued). However, 2) is naturally compatible with the presence (and prevalence) of HGTs and paralogy.

Authors of the current work equate "genetic distance" to "branch length" (line 70), and only investigated the latter. This equation is misleading. If organism groups A and B diverged early, but then exchanged many genes post-divergence, then this is indisputable evidence that their "genetic distance" is close. This point needs to be clearly explained in the manuscript.

We agree with the reviewer that various definitions of evolutionary distance are possible, and some may be more useful than others for particular applications. The reviewer’s argument that “If organism groups A and B diverged early, but then exchanged many genes post-divergence, then this is indisputable evidence that their "genetic distance" is close” makes the case for a kind of phenetic distance: a distance based on overall similarity, regardless of how that similarity was brought about in terms of evolutionary process. We appreciate the democratic appeal of such a metric, and we have no desire to impose any particular philosophy of classification on the reader. However, the key point here is that methods that rely on concatenation for branch length or divergence time estimation (as used by Zhu et al., and in our current study) make the assumption that all of the sites in the concatenate evolved on the same underlying tree and if this assumption is not met, analyses can be misled. Thus, the shorter AB branch length and the more recent Archaea-Bacteria divergence times estimated from concatenations of incongruent marker genes result from unmodelled gene transfers which are misinterpreted as evidence for more recent common ancestry. Gene transfer is an important aspect of genome evolution, but none of the currently available methods, including those used by Zhu et. al., allow for genome-scale comparisons to be made in a way that accounts for our understanding of the underlying evolutionary processes.

The point about different possible definitions of evolutionary distance made by the reviewer is valid, and we have now revised the opening of our conclusion to discuss these issues in more detail, writing:

“We note that alternative conceptions of evolutionary distance are possible; for example, in a phenetic sense of overall genome similarity, extensive HGT will increase the evolutionary proximity (Zhu et al., 2019) of the domains so that Archaea and Bacteria may become intermixed at the single gene level. While such data can encode an important evolutionary signal, it is not amenable to concatenation analysis.”

Core vs genome:

This difference between "AB distance" and "AB branch length" is relevant to a more fundamental question: What defines the "evolutionary distance" between two groups of organisms? Both papers did not explicitly discuss this topic. It likely cannot be resolved in one article (as many scholars have continuously attempted on related topics in the past decades). But the discordance in understanding led to very different research strategies in the two papers, and rendering them incongruent in methodology.

Specifically, the current work (and multiple previous works) based phylogenetic inference on only genes that demonstrate a strong pattern of vertical evolution. HGTs were considered deleterious, and needed to be excluded from the analysis. This left a few dozen genes at most, and many are spatially syntenic and functionally related (e.g. ribosomal proteins). In this work, the final number is 27. Previous critiques of this methodology have suggested that this is not a tree of life, but a "tree of one percent" (Dagan and Martin, Genome Biol. 2006).

In contrast, Zhu et al. (and related previous works) attempted to evaluate the evolution of whole genomes by "maximizing the included number of loci.". They used a "global" set of 381 genes. They faced the challenge of "reconciling discordant evolutionary histories among different parts of the genome", because "HGT is widespread across the domains". To resolve this, they adopted the gene tree summary method ASTRAL.

Therefore, the "AB distance" estimated by Zhu et al. is a genome-level distance, calculated by merging conflicting gene evolutions (which itself can be disputed, see below). Whereas the "AB branch" evaluated in this work is strictly the branch length in the core gene evolution. Therefore, the results presented in the two papers do not necessarily conflict, because of the different scopes.

This point is closely related to the previous one, and the new section (final paragraphs of the Conclusion, quoted directly above) goes some way to addressing this comment. Regarding the issue of a focus on just a small proportion of vertically-evolving genes, the critical point is as above: current methods for branch length and divergence time estimation (including those used by Zhu et al.) require such vertically-evolving genes, because they make the assumption that all of the sites evolve on the same tree, i.e. trace back to the same origin via vertical evolution. We agree that most prokaryotic gene families do not evolve under these restrictive assumptions and therefore cannot be analysed using concatenation methods for branch length estimation. Indeed, one of the main points of our study is that most of the genes in the 381-gene set of Zhu et al. do not meet these assumptions and are thus unsuited for estimating evolutionary distance and divergence times.

There is much ongoing method development which will allow more of the genome to be used in deep-time comparative analyses; Astral-Pro, FastMulRFS and SpeciesRax, among others, are recent promising steps in this direction. However, our central critique of Zhu et al. is that inferences under concatenation-based methods can be misled by HGT and other sources of incongruence, and indeed our analyses show that these unmodelled signals underlie the difference between the conclusions of Zhu et al. and other studies (e.g. (Liu et al., 2021; Spang et al., 2015; Williams et al., 2020) that have instead supported a deep divergence between Archaea and Bacteria. In our revised manuscript, we have shown that the relative AB distance, like the AB branch length, is shortened by unmodelled gene transfers (Figure 1), and that estimates of the AB stem length from different studies are similar when the congruent subset of the data is analysed with the best available substitution models (Figure 6). We therefore disagree that the scopes are distinct: richer, broader measures of genomic diversity can be proposed and, with the development of new methods, estimated; but so far, the vertical signal is the only signal that can be harnessed to infer divergence times using concatenations.

The expanded marker set:

The authors made a valid critique (line 121-135) that many of the 381 genes in the "expanded marker set" adopted by Zhu et al., are under-represented in Archaea. According to the PhyloPhlAn paper (Segata et al. Nat Commun. 2013) which originally developed the 400 markers (a superset of the 381 markers), these genes were selected from ~3,000 bacterial and archaeal genomes available in IMG at that time time (note that it was 2013). Zhu et al. also admitted, in the discussion section, that this marker set falls short in addressing some questions (such as the placement of DPANN). What is important in the current context, is that they were not specifically selected to address the AB distance question.

We agree that the taxon sampling of archaea and the choice of marker genes in the Zhu et al. study were not ideal for estimating the evolutionary distance between the domains. However, we note that this distance (or proximity), and the hypothesis that traditional core genes over-estimate the Archaea-Bacteria divergence, was one of the main results of the paper (c.f. the title of that paper, “Phylogenomics of 10,575 genomes reveals evolutionary proximity between domains Bacteria and Archaea”).

However, note that Zhu et al.'s Fig. 5A, B presented the AB distance informed by 161 out of the 381 genes. These genes have at least 50% taxa represented in both domains - the same threshold discussed in the current work (line 132).

While the 50% sampling criterion indeed enriches for the genes of the expanded set that were present in LUCA and on the AB branch, we note that the 50% criterion represents a minimum of 4953 bacteria and 335 archaea; that is, it still reflects the unbalanced sampling of the dataset overall. For example, 30 of the genes had fewer than two archaeal homologues, and in 100 of the trees there were fewer than 50 archaea reflecting the large disparity in taxon sampling (Supplementary Information Table S1). The phylogenetic signal in these genes is discussed in more detail below. Looking at the subsampled versions of these 161 genes, we found the majority of these genes (123/161) to have no discernible AB branch length. The 38/161 genes which had an arguable AB branch length (but still with transfers/paralogs) possessed a range of AB lengths: 0.0814:5.26, with a mean AB length of 1.03 and a median of 0.635.

Even with those sufficiently represented genes, they still found that ribosomal proteins and a few other core genes are "outliers" in the far end of the AB distance spectrum.

The reviewer raises an interesting point about outliers with high relative AB distances, which gets to the heart of the debate about how best to estimate the evolutionary distance between Archaea and Bacteria. The new analyses of relative AB distance introduced in our revised manuscript (Figure 1) demonstrate that this metric is affected by HGT in a similar manner to AB branch length (that is, high-verticality marker genes have a greater relative AB distance (relative AB vs ΔLL: p = 0.0001051 & R = -0.2213292, relative AB vs between-domain split score: p = 2.572e-06 & R = -0.2667739). Thus, core genes can be viewed as “outliers” compared to other prokaryotic genes in the sense that they have experienced an unusually low amount of HGT. This high verticality makes them among the few prokaryotic gene families that can be analysed by concatenation methods, which make the assumption that all sites evolve on the same underlying tree topology.

Domain monophyly in gene trees:

The authors' efforts in manually checking the gene trees are appreciable (Table S1), considering the number and size of those trees. They found (line 147) "Archaea and Bacteria are recovered as reciprocally monophyletic groups in only 24 of the 381 published (Zhu et al., 2019) maximum likelihood (ML) gene trees of the expanded marker set."

The domain monophyly check was valid, however the result could be misleading because any sporadical A/B mixture was considered evidence of non-monophyly for the entire gene tree. As the taxon sampling grows, the opportunity of observing any A/B mixture also increases. For example, in Puigbò et al. J. Biology. 2009, 56% (a much higher ratio) of nearly universal genes trees had perfect domain monophyly based on merely 100 taxa. This is because even the "perfect" marker genes (such as ribosomal proteins) are not completely free from HGTs (e.g., Creevey et al. Plos One. 2011), let alone the fact that there are many artifacts in the published reference genomes (Orakov et al. Genome Biol. 2021).

Therefore, to have an objective assessment of this topic, it would be better to have a metric that allows some imperfection and reports an overall "degree" of separation (also see below).

We agree that complementing the monophyly check with a more nuanced metric is useful. In our revised manuscript, we now also evaluate the split score (Dombrowski et al. (2020) Nat Commun) of each marker, which reflects the degree to which a gene recovers the monophyly of established taxonomic ranks (a higher score reflects the splitting of monophyletic groups into a number of smaller clades in the gene tree, and so the metric permits a degree of “imperfection”, as suggested; in addition, the metric is averaged over bootstrap replicates, so that lack of resolution or poorly-supported disagreements with the reference taxonomy do not disproportionately affect the score). This expanded analysis (Figure 1) indicates that both within- and between-domain split score and ΔLL are significantly positively correlated (R = 0.836679, p < 2.2✕10-16), and that phylogenetic markers that more strongly reject domain monophyly (higher delta-LL) also perform worse at recovering between-domain (and within-domain) relationships (higher split score) and support a shorter AB branch length.

AB branch by gene: correlation and outliers

Figure 1 is the single most important result in this work, because it argues that the short AB branch observed in Zhu et al. is an artifact due to "inter-domain gene transfer and hidden paralogy" (line 202). This argument is based on the observation that the indicated AB branch length is negatively correlated with "verticality" (measured by ΔLL and split score) of the gene.

Our argument that the short AB branch results from inter-domain gene transfer and hidden paralogy is based on three main lines of evidence: (i) documentation of extensive transfers and intermixing of paralogues in the gene trees for the 381 gene set; (ii) the analyses in Figure 1, which demonstrate that verticality positively correlates with AB branch length and AB distance; (iii) the demonstration that the incremental addition of low-verticality markers to a concatenate results in a concomitant decrease in AB branch length.

However, Zhu et al. also investigated the impact of verticality on AB distance, and they also found that they are negatively correlated (Fig. 5E). Therefore, the current result does not appear to deliver new information (as do multiple other analyses, see below).

Zhu et al. indeed identified a weak positive relationship between gene verticality and AB distance. Our analyses go beyond that work by showing, using a variety of complementary metrics of verticality, that AB branch length and relative AB distance are strongly positively correlated with verticality (see Figure 1), and that the low verticality of the genes in the 381 gene set largely explains the difference in stem length inference between that dataset and earlier analyses (Figure 6). An additional factor not considered in the analyses of Zhu et al. was the question of whether a gene was present in LUCA, and so can provide information on the AB branch length. Our analyses (detailed below) suggest that the majority of genes (317) in the 381 gene set do not contain an unambiguous AB branch, and so do not contribute interpretable signal to estimates of the AB branch length.

An important finding in Zhu et al., which is largely not discussed in the current work, is that a handful of "core" genes are outliers in the spectrum of AB distance, as compared to the majority of the genome (Fig. 5A). The AB distance indicated by these core genes is so long compared with the genome average that it cannot be compensated for by the impact of non-verticality, substitutional saturation, site-homogeneous model, etc (see below).

Fig. 1A of the current work also clearly shows that many long-AB branch genes are outliers compared with the majority of the genome (the bottom of the blue bar).

Figs. 3 and 4 attempted to show that ribosomal proteins are not outliers, but that analysis was based on a very small set of core genes, and the figures clearly show that there are outliers even in this small set (to be further discussed below).

This comment re-iterates the reviewer’s earlier points about “core” genes as outliers compared to the majority of the genome. The key issue is that “most of the genome”, and a significant portion (317 genes) of the 381-gene set, contain features that make them unsuitable for estimation of AB branch length by concatenation, or indeed estimation of an interpretable relative AB distance. We have documented the cases of HGT and mixing of paralogues in the 381-gene dataset; this information is summarised in the main text and presented in more detail in Supplementary Information Table S1.

Focusing on the 161 genes with >50% representation in both Archaea and Bacteria, manual inspection of gene trees inferred on the 1000-species subsample under the LG+G+F model indicate that 123/161 do not have a clear AB branch (that is, a branch that separates most or all Archaea from Bacteria). While distinguishing such cases from early gene transfers is not straightforward, there is no compelling reason to think that these genes were present in LUCA. The simplest explanation for these gene phylogenies is instead an origin within Bacteria and subsequent transfer on one or multiple occasions into Archaea. As a result, estimates of AB branch length or relative AB distance inferred from these genes cannot be straightforwardly compared to those of the traditional “core” or other genes for which the evidence of a pre-LUCA origin is stronger. Considering only the 38/161 genes for which a LUCA origin appears, from the gene phylogeny, to be likely, the mean AB branch length is 1.03, greater than that estimated from the concatenation of the most vertical genes in the expanded set (0.56), and suggesting that phylogenetic incongruence, combined with (for some families) a more recent origin explains the shorter AB distances inferred from the 381 gene set. Thus, it is not the case that the AB branch lengths (or relative distances) estimated from the majority of genes form a null distribution” against which “core” genes can be seen as outliers; instead, our analyses suggest that “core” genes are among the limited number of genes that trace vertically to LUCA.

Regarding Figures 3 and 4, see the more detailed discussion below.

Verticality is not causative of short AB branch:

In spite of the outlier question, there is an important logic problem in these analyses: The authors observed that gene verticality (measured by negative ΔLL) is correlated with AB branch length (Fig. 1), and concluded that HGTs and paralogy shortened the AB branch (line 202). However, they did not directly assess the rate of evolution in this model. It is totally possible that the most vertical genes happen to be those that evolved faster at the AB split. In order to support the claim made in this work, it is important to separate the effect of the rate of evolution from the effect of HGT / paralogy.

The ideal solution would be to include ALL genes (not just "good" ones), build gene trees, identify parts of the gene trees that once experienced HGT or paralogy, and prune off these PARTS, instead of excluding the entire gene tree. The remaining data are thus free of HGT / paralogy, based on which one can quantify the "true" AB branch length, and further assess how much it is correlated with "verticality", and whether there are still "outliers". This solution is likely not trivial in implementation, though. However, without such assessment, the observed short AB branch still only applies to the "tree of one percent", not the "tree of life".

Thanks for this comment --- the reviewer raises a subtle and valid point. Our analyses indicate that vertically-evolving genes have longer AB branch lengths, but in the first version of our manuscript we did not test the alternative hypothesis that this relationship might simply result from a faster rate of evolution in vertically-evolving genes. To evaluate the relationship between evolutionary rate, verticality, AB branch length and relative AB distance on as broad a set of genes as possible, we took the 302 genes from the 381-gene expanded set, excluding 56 genes for which the 1000-species subsample included no archaea, and another 23 which included only 1 archaeon. To estimate per-gene evolutionary rate, we rooted each gene tree using MAD (Tria et al. 2017) and calculated the mean root- to-tip distance on the MAD-rooted gene tree, then evaluated the relationship between rate and verticality. This analysis indicated that vertically-evolving genes evolve more slowly (have shorter mean root-to-tip distances) than less vertical genes (using deltaLL and between-domain split score as proxies for marker verticality, with a Pearson’s product- moment correlation: MAD rooted mean root-to-tip distance against deltaLL: R = 0.1397803, p = 0.01506 or against split score: R = 0.1902056 p = 0.000893), despite having longer AB branches and relative AB distances (using a Pearson’s product moment correlation of MAD rooted mean root-to-tip distance against AB length: p = 0.2025, R= 0.1143076, or against relative AB distance p = 0.007435, R=0.1537479). Thus, the longer AB branches of vertically evolving genes do not appear to be the indirect result of faster evolution of those genes. These analyses are reported in the main text, where we write:

An alternative explanation for the positive relationship between marker gene verticality and AB branch length could be that vertically-evolving genes experience higher rates of sequence evolution. For a set of genes that originate at the same point on the species tree, the mean root-to-tip distance (measured in substitutions per site, for gene trees rooted using the MAD method (Tria et al., 2017)) provides a proxy of evolutionary rate. Mean root-to-tip distances were significantly positively correlated with ∆LL and between-domain split score (∆LL: R = 0.1397803, p = 0.01506, split score: R = 0.1705415 p = 0.002947; Figure 1 Figure Supplement 5,6, indicating that vertically-evolving genes evolve relatively slowly (note that large values of ∆LL and split score denote low verticality)). Thus, the longer AB branches of vertically-evolving genes do not appear to result from a faster evolutionary rate for these genes. Taken together, these results indicate that the inclusion of genes that do not support the reciprocal monophyly of Archaea and Bacteria, or their constituent taxonomic ranks, in the universal concatenate explain the reduced estimated AB branch length.

Differential metric for verticality:

In spite of the similarity between the current result and Zhu et al.'s (see above), the two works approached this goal using different metrics.

First, the authors attempted to quantify the AB branch length in individual gene trees, including those that do not have Archaea and Bacteria perfectly separated. To do so they performed a constrained ML search (line 210). I am wary of this treatment because it could force distinct sequences (due to HGT or paralogy) to be grouped together, and the resulting branch length estimates could be highly inaccurate.

We agree with the reviewer that estimating AB branch lengths in this way might lead to inaccuracy. We note that this is, in effect, what was done in the published analysis (Zhu et al. 2019): a topology in which Archaea and Bacteria were reciprocally monophyletic was inferred using ASTRAL (a reasonable analysis, given the robustness of ASTRAL to some degree of HGT/gene tree incongruence), and then the AB branch length was estimated from the concatenation of these 381 genes, fixing on the ASTRAL topology. We performed this experiment (inferring AB branch length on constrained trees) in order to evaluate how incongruence between the gene and species trees might affect AB branch length inference.

In contrast, Zhu et al. quantifies the average taxon-to-taxon phylogenetic distance between the two domains, regardless of the overall domain monophyly. That method was free of this concern, although it computed a different metric.

Thanks for raising this point. As described above, in the revised manuscript we have also evaluated the relative AB distance metric used by Zhu et al., and show that it behaves similarly to the AB branch metric we evaluated in the first version of the manuscript (see revised Figure 1).

Second, the authors assessed "marker gene verticality" using two metrics: a) AU test result (rejected or not) (Fig. 1A), c) ΔLL, the difference in log likelihood between the constrained ML tree and ML gene tree (line 222, Fig. 1B, C). I am concerned that they are sensitive to taxon sampling and stochastic events, as I explained above regarding domain monophyly. It is possible that a single mislabeling event would cause the topology test to report a significant result. In addition, they evaluate how severely domain monophyly is violated, but they do not account for intra-domain HGTs and other artifacts, which are also part of "verticality", and they can potentially distort the AB branch as well.

In the revised manuscript, we also evaluate a complementary metric for marker gene verticality, the split score (see above), which measures the extent to which marker genes recover established relationships at a given taxonomic level (we computed both within-domain and between-domain split scores). The split score is a more granular measure than ΔLL and, by summing over bootstrap replicates, it also better accommodates phylogenetic uncertainty. The two metrics (ΔLL and split score) are positively correlated and the analyses come to the same conclusions regarding the impact of HGT and other sources of incongruence on estimates of the AB branch length and relative AB distance.

I did not find the ΔLL values of individual markers in any supplementary table. I also did not find any correlation statistics associated with Fig. 1B.

The ΔLL values for individual markers can be found in the data supplement in: “Expanded_Bacterial_Core_Nonribosomal_analyses/Individual_gene_tree_analyses/Expanded//Expanded_AB_AU.csv”

We have now updated the readme.txt file for clarity and included all the new results from the analyses which we have undertaken as part of the review process in the latest version of the supplemental available on figshare (10.6084/m9.figshare.13395470) as well as updated the directory and file names for clarity. We also have added the statistics associated with the correlations in Figure 1 to the Figure Legend.

Statistical test:

Line 157: "For the remaining 302 genes, domain monophyly was rejected (p < 0.05) for 232 out of 302 (76.8%) genes." Did the authors perform multiple hypothesis correction? If not, they probably should.

Thanks for this suggestion. We have now used a Bonferonni correction to account for multiple testing. As a result, fewer marker genes are rejected at the 5% level (151/302), although the overall conclusions are unaffected.

Line 217: "This result suggests that inter-domain gene transfers reduce the AB branch length when included in a concatenation." and Fig. 1A. If I understand correctly, this analysis was performed on individual gene trees, rather than in a concatenated setting. Therefore, the result does not directly support this conclusion.

Thanks for pointing this out. The reviewer is correct that this inference depends not only on the single gene analyses, but also on the subsequent concatenation results presented in this section. We have therefore moved this sentence later in the section, after the concatenation analysis.

Line 224: "Furthermore, AB branch length decreased as increasing numbers of low-verticality markers were added to the concatenate (Figure 1(c))". While this statement is likely true, Zhu et al. also presented similar results (Fig. 5) despite using a different metric, and they concluded that the impact is moderate and cannot explain the status of some core genes as outliers.

Zhu et al. did identify some of these trends, as we acknowledge in our manuscript ("The original study investigated and acknowledged (Zhu et al., 2019) the varying levels of congruence between the marker phylogenies and the species tree, but did not investigate the underlying causes.“ --- line 178; “These results are consistent with (Zhu et al., 2019), who also noted that AB branch length increases as model fit improves for the expanded marker dataset.” --- line 337) and as discussed above. Our analysis (Figure 6, Table 1) goes further in showing that the most vertical subset of the 381-gene set supports an inter-domain branch length closely similar (2.4 subs/site compared to e.g. 2.5 subs./site for the 27-gene dataset) to analyses using the traditional marker gene set.

Concatenation and branch length:

The authors pointed out that "Concatenation is based on the assumption that all of the genes in the supermatrix evolve on the same underlying tree; genes with different gene tree topologies violate this assumption and should not be concatenated because the topological differences among sites are not modelled, and so the impact on inferred branch lengths is difficult to predict." (line 187).

This argument is valid. In my opinion, this is the one most important potential issue of Zhu et al.'s analysis. In that work, they inferred genome tree topology through ASTRAL, which resolves conflicting gene evolutions. However ASTRAL does not report branch lengths in the unit of number of mutations. Therefore, they plugged the concatenated alignment into this topology for branch length estimation, hoping that it will "average out" the result. That workaround was apparently not ideal.

Yes, we agree --- this is our main critique of the Zhu et al. analyses.

However, the practice of molecular phylogenetics is complicated. Theoretically, every gene, domain, codon position and site may have its unique evolutionary process, and there have been efforts to develop better partition and mixture models to address these possibilities. But there is a trade off; these technologies are computationally demanding and have the risk of overfitting. It is plausible that in some scenarios, the gain of concatenating many loci (despite conflicting phylogeny) may outweigh the cost of having unpredictable effects.

But this dilemma needs to be analyzed rather than just being discussed. The Zhu et al. paper did not assess the impact of such concatenation on branch length estimation. The best answer is to conduct an analysis to show that concatenating genes with conflicting phylogeny would result in an AB branch that is shorter than the mean of those genes, and the reduction of AB branch length is correlated with the amount of conflict involved. The current work has not done this.

Thanks for raising this point. We agree that phylogenetics is complicated and that

we lack methods that can account for all possible factors. With respect to the impact of gene transfers on the AB branch length, and as touched on above, there are two issues here.

The first is with the analysis actually performed by Zhu et al: of the 381 extended set genes, 79 have one or no archaea in the 1000-taxon subsample, and a further 176 have an AB branch length close to 0 (<0.00001) in the constrained analyses. To investigate further, we manually inspected ML gene trees for the 381 genes (1000 taxon subsample). Allowing for recent gene transfer, we nevertheless identified only 64 genes with an unambiguous branch separating most Archaea from most Bacteria that might correspond to the ancestral AB divergence (Supplementary File 1).

Taken together, these analyses suggest that there is no strong evidence that these genes were present in LUCA or evolved along the AB branch, and so they do not provide information on its length. Since the branch length in the concatenation is an average over the branch length per site, the inclusion of this set of genes in the analysis did reduce the AB branch length, as demonstrated by our analyses (Figure 1(H)).

The second issue is: for genes which were likely present in LUCA and evolved on the AB branch, does gene transfer cause a reduction in the AB branch length inferred from their concatenation? To test this, we initially tested iterative concatenations of increasing numbers of non-vertical markers (Figure 1H), as well as a comparison of the most vertical genes to the whole expanded marker set (Figure 3 Figure Supplement 2). This revealed that as more markers were added (with lower verticality), the inferred AB branch length from the concatenate was reduced. We also found an increased AB branch length when only the 20 most vertical markers were used as opposed to the whole (381 marker) dataset (0.56 vs 0.16 substitutions/site, Figure 6).

The reviewer proposes an additional test of the impact of marker gene incongruence on branch length inference from concatenations: to compare the AB branch length before and after pruning of HGTs from individual marker gene alignments. To do this, we took the 54 marker genes from our new dataset and concatenated them before and after pruning of unambiguous HGTs. The AB branch length inferred from the concatenation with HGTs removed was 1.946 substitutions/site, compared to 1.734 substitutions/site without pruning HGTs, demonstrating the impact of even a relatively small number of HGTs on branch length estimation from concatenates.

Divergence time estimation:

The manuscript dedicates one section (line 230-266) to argue that the divergence time estimation analysis performed by Zhu et al. was not good evidence for marker gene suitability. Zhu et al. showed congruence of the expanded marker set with geological records whereas ribosomal proteins were conflicting with the geologic record.To support their argument, the authors estimated divergence times using the top 20 most "vertical" genes measured by ΔLL.

It would be good to clarify which genes they are, and it would be important to check whether they include some of the most "AB-distant" ones found by Zhu et al. Their Fig. 5A shows that there are genes that divide the two domains several folds further than the ribosomal proteins (such as rpoC). If they are among the 20 genes, it will not be surprising that the estimated AB split is older than it should be.

We now include the annotations for these 20 genes in Supplementary File 5a. The 20 most vertical genes include two of the “AB-distant” outliers identified by Zhu et al., tuf and infB, and one ribosomal marker, rpsG.

Overall, I think this section is logically questionable. Zhu et al. suggested that "They show the limitation of using core genes alone to model the evolution of the entire genome, and highlight the value in using a more diverse marker gene set.". The current work showed that using another set of a few genes (I do not know if they include multiple "core" genes, as discussed above, but it is plausible) also did not work well. This does not refute Zhu et al.'s claim.

What's important in Zhu et al.'s analysis is this: they demonstrated that using a small set of genes in DTE may cause artifacts due to them significantly violating the molecular clock at certain stages of evolution. Instead, using a larger set of markers that represent a portion of the entire genome would help to "smooth out" these artifacts. This of course is not the ideal solution, likely because concatenating conflicting genes and modelling them uniformly is not the best idea (see above). But as an operational workaround, it was not challenged by the analysis in the current work.

Finally, I agree with the authors' statement that more and reliable calibrations are the best way to improve divergence time estimation.

The dating analyses presented in the first version of our manuscript demonstrated that the apparent agreement between molecular clock estimates using the 381-gene set and the fossil record was the result of artifactual shortening of the AB branch, as discussed in detail above. Once the subset of the data least affected by these issues (that is, the most vertical subset) was used, the limitations of current clock methods, particularly with few calibrations, for dating deep nodes became clear.

That said, we agree with the reviewer (and also R3) that the dating section in the first version of our manuscript was somewhat unsatisfactory: it identified an important limitation of the published analysis, but did not explore the underlying question of why molecular clock methods infer unrealistically old divergence times from vertically-evolving genes. In the revised manuscript we have reworked and improved this section extensively, including new analyses on the 27-gene dataset, with more fossil calibrations, that help to diagnose how and why clocks struggle to date the archaeal and bacterial stems from the available data. We now show that the old ages result from a combination of low rates of molecular evolution across the tree inferred from “shallow” calibrations, combined with a lack of age maxima for nodes other than the root of the tree; when the rate distribution is informed in this way, the long AB branch is interpreted as representing a long period of time and estimates of LUCA age are strongly influenced by prior assumptions about root maximum age. These analyses now suggest how the difficulties might be overcome in the future, for example using better calibrations (particularly maximum ages, and indeed any fossil calibrations within the Archaea), or alternatively other sources of time information such as from gene transfers. Reflecting the new, broader focus, we have moved this section to the end of the manuscript.

AB branch by ribosomal and non-ribosomal genes:

Two figures (Figs. 3 and 4) are two sections (line 270-303) dedicated to the argument that ribosomal markers do not indicate a longer AB branch than a non-ribosomal one. However, this is a small scale test (38 ribosomal markers vs. 16 non-ribosomal markers) compared with the similar analysis in Zhu et al. (30 ribosomal markers vs. 381 global markers). A closer look at Figs. 3 and 4 suggests that while the AB lengths indicated by the ribosomal markers are within a relatively narrow range, those by the non-ribosomal ones are very diverse, including ones that are several folds longer than the ribosomal average. This result is in accordance with that of Zhu et al.'s Fig. 5A, although the latter was describing a different metric. Do these genes also overlap the ones found by Zhu et al.?

Nevertheless, this analysis does not falsify Zhu et al.'s, because it compared a different, much smaller, and deliberately chosen group of genes.

As the reviewer indicates, the purpose of the analyses presented in Figures 3-4 is to evaluate the hypothesis of accelerated ribosomal protein evolution: that is, the idea that ribosomal proteins over-estimate the AB branch length due to accelerated evolution during the divergence of Archaea and Bacteria. Although this hypothesis was independently proposed in Zhu et al., to our knowledge it actually originates with Petitjean et al. (2014) GBE (https://academic.oup.com/gbe/article/7/1/191/601621; see their Figure 2), and has been at play in analyses of deep evolution and in particular the position of DPANN Archaea in the phylogeny since that time. Thus, this section of our manuscript (indeed, all but the first section) is not a critique of Zhu et al.’s work, but a contribution to the broader ongoing discussion about which marker genes are best to use in deep phylogeny. We compare only vertically-evolving genes in Figures 3-4 so as to distinguish the impact of gene function (ribosomal versus non-ribosomal) from confounding factors such as HGT, paralogy, and gene origination time.

To clarify this point, we have modified our main text discussion to make it clear that we are making a comparison between ribosomal genes and other vertically-evolving members of the traditional “core” gene set, rather than a broader genome-wide claim. We now write:

“If ribosomal proteins experienced accelerated evolution during the divergence of Archaea and Bacteria, this might lead to the inference of an artifactually long AB branch length (Petitjean et al., 2014; Zhu et al., 2019). To investigate this, we plotted the inter-domain branch lengths for the 38 and 16 ribosomal and non-ribosomal genes, respectively, comprising the 54 marker genes set. We found no evidence that there was a longer AB branch associated with ribosomal markers than for other vertically-evolving “core” genes (Figure 2(b); mean AB branch length for ribosomal proteins 1.35 substitutions/site, mean for non-ribosomal 2.25 substitutions/site).”

Substitutional saturation:

The comparative analysis of slow- and fast-evolving sites is interesting. The result (Fig. 5) is visually impactful. In my view, this analysis is valid, and the conclusion is supported. It would be better to explain the rationale with more detail to facilitate understanding by a general audience.

Thanks for this assessment. We have now expanded on the rationale of this analysis in the main text, writing:

“It is interesting to note that the proportion of inferred substitutions that occur along the AB branch differs between the slow-evolving and fast-evolving sites. As would be expected, the total tree length measured in substitutions per site is shorter from the slow-evolving sites, but the relative AB branch length is longer (1.2 substitutions/site, or ~2% of all inferred substitutions, compared to 2.6 substitutions/site, or ~0.04% of all inferred substitutions for the fastest-evolving sites). Since we would not expect the distribution of substitutions over the tree to differ between slow-evolving and fast-evolving sites, this result suggests that some ancient changes along the AB branch at fast-evolving sites have been overwritten by more recent events in evolution --- that is, that substitutional saturation leads to an underestimate of the AB branch length.”

Zhu et al. also tested the impact of substitution saturation on the AB branch, using a more traditional approach (Fig. S19). They also found that the inter-domain distance is more influenced by potential substitution saturation, but the difference is minor. They concluded that (AB distance) "is not substantially impacted by saturation."

Like other analyses, these two analyses involved very different locus sampling (27 most "vertical" genes vs. 381 expanded genes). They also differ by the metric being measured (AB branch length vs. average distance between AB taxa). Therefore, the analysis in the current work does not falsify the analysis by Zhu et al. In contrast, it is inline with (though not in direct support of) Zhu et al. and others' suggestion that there was "accelerated evolution of ribosomal proteins along the inter-domain branch" (line 25) in the 27 core genes (of which 15 are ribosomal proteins).

We disagree that our analysis is consistent with the hypothesis of accelerated ribosomal protein evolution. The analysis that directly addresses this point is Figure 3, where we show that the distributions of AB branch lengths in single gene trees are not significantly different between ribosomal and non-ribosomal datasets (Figure 3; mean AB branch length for ribosomal proteins 1.35 substitutions/site, mean for non-ribosomal 2.25 substitutions/site).

Evolutionary model fit:

The authors compared the AB branch length indicated by the standard, site-homogeneous model LG+G4+F vs. the site-heterogeneous model LG+C60+G4+F, and found that the latter recovered a longer AB branch (2.52 vs. 1.45). The author's reasoning for using a site-heterogeneous model is valid, and this analysis is sound.

However, Zhu et al. also analyzed their data using the site-heterogeneous model C60 -- the same as in this work, but through the PMSF (posterior mean site frequency) method. Zhu et al. also compared it with two site-homogeneous models (Gamma and FreeRate). The results were extensively presented and discussed (Figs. 3, 4E, F, S23, S24, Note S2). They also found that C60+PMSF elongated the AB branch compared with the site-homogeneous models (Fig. S24A). As for the average AB distance (another metric evaluated by Zhu et al., as discussed above), C60+PMSF increased this metric when using ribosomal proteins, but not much when using the expanded marker set (Fig. S25A). And overall, the elongation by C60+PMSF with the expanded markers cannot compensate for the long branch indicated by the ribosomal proteins.

Therefore, similar to the point I made above, this analysis is sound but it does not logically falsify the conclusion made by Zhu et al., as it only concerns a small set of markers, and it recovered a previously described pattern.

Thanks for this comment. As above, note that the second part of our manuscript presents a general analysis of the issues around marker gene and model selection using our meta-analysis and new dataset, and is not a direct response to Zhu et al’s work. On reflection, we agree that this was not sufficiently clear in the first version of the paper, and we have now modified the text to acknowledge the model fitting analyses of Zhu et al.

The manuscript also did not clarify what the phrase "poor model fit" refers to (line 34 and line 304). If this is addressing the Gamma model evaluated by the authors, then this claim is valid though not novel (but see my previous comment on the trade-off). If that is a general reference to Zhu et al.'s methodology, then the authors should at least include the C60+PMSF model in the analysis, and show that C60 indicates a significantly longer AB branch than C60+PMSF does (if that's the case, which is doubtful). Admittedly, C60+PMSF is cheaper than the native C60 in computation, but "In some empirical and simulation settings PMSF provided more accurate estimates of phylogenies than the mixture models from which they derive." (Wang et al. Syst Biol. 2018).

Thanks for this comment. We did not intend the phrase “poor model fit” to imply a critique of Zhu et al.’s work; as the reviewer notes, those authors carried out a range of analyses to investigate the impact of model choice on their inferences. Rather, the title of the section is intended to summarise its main conclusion, which is that substitutional saturation and poor model fit (on any dataset, and even with the best available models) can lead to under-estimation of the AB branch length. Note that the analyses in Table 1 illustrating the impact of model fit are from the new dataset that is assembled and analysed in the second part of the manuscript. As above, we agree that this was not sufficiently clear in the first version of the paper. We think the title of this section is accurate and so we have not changed it, but we have changed the final two paragraphs of the section (as quoted immediately above) so as to acknowledge the model fitting analyses of Zhu et al., and to clarify that the results are general (and based on our new dataset), rather than a critique of Zhu et al’s work.

Finally, Zhu et al. also performed an analysis using the native C60 model on a further reduced taxon set. That result was not presented in the published paper, but it can be found in the "Peer Review File" posted on the Nature Communications website. That tree also recovered a short AB distance, and placed CPR at the base of Bacteria, and showed that this placement was not impacted by the removal of Archaea.

Thanks for pointing us to this additional analysis. The unrooted, bacteria-only tree referred to by the reviewer (panel B) recovers a clan (that is, a cluster of branches on the unrooted tree) comprising CPR+Chloroflexi, in agreement with the analysis on the new marker dataset we present here (Figure 6). The disagreement between that analysis and the new tree presented here relates to the position of the archaeal outgroup, which in the Peer Review File panel A connects to the bacterial tree between CPR and Chloroflexi. If, as recently suggested, the bacterial root lies between Gracilicutes and Terrabacteria (Coleman et al. 2021), then CPR and Chloroflexi represent monophyletic sister lineages. We note that the CPR+Chloroflexi relationship recovered here and in Peer Review File Panel (B) has also been obtained in several other recent analyses (Taib et al. 2020, Coleman et al. 2021, Martinez-Gutierrez and Aylward 2021), as cited in the main text.

Taxon sampling:

My final comment is about taxon sampling. Zhu et al. developed an algorithm for less biased taxon sampling, and they argued that extensive taxon sampling is important in resolving the early evolution of life. They presented evidence showing that reduced taxon sampling changed overall topology and basal relationships (Figs. S13, S14, S23, Note S2). The analyses were performed in combination with the assessment of site sampling, locus sampling, substitution model and other factors. The importance of less biased and/or extensive taxon sampling was also noted by previous works, especially in a phylogenomic framework (e.g., Hedtke et al. Syst Biol. 2006; Wu and Eisen. Genome Biol. 2008; Beiko. Biol Direct. 2011). The current work is based on a smaller set of taxa, and it has not addressed the impact of taxon sampling. As I suggested above, some results may be sensitive to taxon sampling.

We agree that taxon sampling is important for phylogenetics. While the analyses of Zhu et al. (2019) included a very large number of genomes, sampling of genomes (and indeed marker genes) was biased, both towards Bacteria compared to Archaea, and also within Bacteria. In our revised manuscript, we now compare the taxon sampling between Zhu et al.’s work and our new analyses (see Figure 1 Figure Supplements 13,14,15 and Figure 4 Figure Supplements 1,2). Balanced sampling is important for phylogenetic inference (Heath et al., 2008; Hillis, 1998) and, by this criteria, the taxon sampling in the analyses of Zhu et al. was not ideal. Our new analyses made use of fewer genomes (700), but these sample the known diversity of Archaea and Bacteria in a more representative way (Figure 4 Figure Supplement 1,2).

Reviewer #3:

Moody and coworkers principally address a recent paper presented by Zhu et al. (Nature Communications, 2019). In their paper, Zhu and coworkers claim that (i) ribosomal protein genes, commonly used in resolving deep phylogenies, have experienced an increased rate of evolution right after LUCA, and (ii) that an expanded set of markers show that the branch separating archaea from bacteria (AB-branch) is 10-fold shorter than previously thought. Moody et and coworkers first demonstrate flaws in the Zhu et al. analysis: first, the expanded gene set is biased towards bacteria, with 25% of the single-gene trees having very few archaeal counterparts. Second, that over 75% of the single-gene trees from Zhu et al are not monophyletic at domain level, suggesting a large influence of horizontal gene transfers (HGT), inter-domain exchanges, and inclusion of paralogous sequences in the original datasets. Third, they show that genes with fewer HGT display longer AB-branches. Fourth, they show that the argument by Zhu et al. that the longer AB-branch yields absurd LUCA datation is not relevant. Fifth, and maybe most important, they show that the shorter AB-branches recovered by Zhu et al in their expanded dataset result from inadequate substitution models, which lead to underestimating rates and thus branch lengths.

Going further, they select a set of 54 manually curated markers (showing mostly monophyletic archaea and bacteria), both from ribosomal proteins (36) and non-ribosomal proteins (18) and retrieve these in a balanced set of 350 archaea and 350 bacteria. With this set, they show that ribosomal protein markers do not display longer AB-branches than non-ribosomal ones. They also show that diversity among Archaea and Bacteria, as measured as the total tree length within each domain, is very similar, when sampling equal number of genomes in both domains.

Strengths:

The paper is well-written and well structured. In general, the methodology chosen here is adapted to the question at hand and very rigorously followed. The balanced dataset (with equal amounts of bacteria and archaea) of 54 carefully selected genes is also appropriate to explore diversity differences between the two domains of life.

Although all arguments presented in Zhu et al are carefully re-evaluated, the part where Moody et al show that substitutional saturation and poor model fit is artifactually producing short AB-branches is quite compelling and elegantly presented.

Weaknesses:

One potential weakness, more in terms of significance than in terms of scientific soundness is that the paper is mostly "reactive", responding to a single other paper. The authors might have used the data and methodology presented here to give the paper a broader scope. An example would be to provide the audience with a solid protocol or general guidelines on how to avoid artifacts in making deep phylogenies. I believe that the authors have demonstrated that they have the authority to do that.

Thanks for this suggestion. We considered including guidelines of this type in the first version of the manuscript, but we were --- and remain --- wary of attempting to promote one particular way of doing deep phylogeny over others. These are difficult and slippery questions, and different approaches and perspectives (including ones we might disagree with) are, in a broader sense, useful in refining ideas and helping the field to make progress as a whole. That said, a recurring issue appears to be the question of the fit between model and data, both in terms of substitution model fit (as with the impact of site-heterogeneous models on branch length inferences) and the broader issue of using models that, for example, account for gene duplication or transfer. There are several recent reviews (including one by some of us) which treat these topics in detail and provide detailed advice. We have now raised and discussed these issues in our conclusion. We have also updated Figure 6 to illustrate the approach we used in assembling the new 27-gene dataset, which may be of use to others, and goes some way towards the suggestion of providing guidelines for future analyses. We now write:

“Our analysis of a range of published marker gene datasets (Petitjean et al., 2014; Spang et al., 2015; Williams et al., 2020; Zhu et al., 2019) indicates that the choice of markers and the fit of the substitution model are both important for inference of deep phylogeny from concatenations, in agreement with an existing body of literature (reviewed in (Kapli et al., 2021, 2020; Williams et al., 2021). We established a set of 27 highly vertically evolving marker gene families and found no evidence that ribosomal genes overestimate stem length; since they appear to be transferred less frequently than other genes, our analysis affirms that ribosomal proteins are useful markers for deep phylogeny. In general, high-verticality markers, regardless of functional category, supported a longer AB branch length. Furthermore, our phylogeny was consistent with recent work on early prokaryotic evolution, resolving the major clades within Archaea and nesting the CPR within Terrabacteria. Notably, our analyses suggested that both the true Archaea-Bacteria branch length (Figure 6A), and the phylogenetic diversity of Archaea, may be underestimated by even the best current models, a finding that is consistent with a root for the tree of life between the two prokaryotic domains.”

In the figure 6 legend, we also expand on guidelines for future analyses, writing:

“(B) Workflow for iterative manual curation of marker gene families for concatenation analysis. After inference and inspection of initial orthologue trees, several rounds of manual inspection and removal of HGTs and distant paralogues were carried out. These sequences were removed from the initial set of orthologues before alignment and trimming. For a detailed discussion of some of these issues, and practical guidelines on phylogenomic analysis of multi-gene datasets, see (Kapli et al., 2020) for a useful review.”

The authors use the difference in log-likelihood between the constrained and unconstrained gene trees as a proxy for verticality and thus marker gene quality (Figure 1b). However, they don't demonstrate that that metric is actually appropriate. Could the monophyly (or split score) be also involved here? The authors might want to comment on that.

Thanks for this suggestion, which has substantially improved our analysis of Archaea-Bacteria distance and marker gene verticality (see the revised Figure 1 and associated text). We have now evaluated the relationship between AB branch length and split score (both within- and between-domain level relationships) for the expanded marker set and have updated our results and discussion accordingly. We found that deltaLL and split score (both within- and between-domains) are positively correlated with each other, and negatively correlated with AB length (that is, high-verticality markers have longer AB branch lengths). These analyses also revealed that within-domain and between-domain split scores are strongly positively correlated, implying that genes that recover domain monophyly also do better at resolving within-domain relationships.

The argument about the age of LUCA an ad absurdum one, showing that using better suited genes one gets impossible time estimates. However, the argument presented by Zhu et al is also a "just so" argument (if we get a time estimate that doesn't make sense then the phylogeny must be wrong), which doesn't give it much weight. The authors themselves note well that this part of the paper is more revealing of the limitations of the strict clock method, or of the relaxed clock with one single calibration point, than of the quality or appropriateness of the dataset.

We agree that the dating section in the first version of our manuscript was somewhat unsatisfactory. We have now expanded it to include new analyses on our 27-gene dataset, using more fossil calibrations, in order to diagnose why current clock methods struggle to estimate evolutionary rate near the root of the tree, and how this impacts on the age of LUCA and other deep nodes. These analyses add substantial value to this section, which has been moved to the end of the manuscript to reflect its expanded focus.

Another small weakness (or loose end) is that manual curation of the 95 genes dataset is not consistently reducing the percentage of non-monopyhletic genes (e.g. 62 to 69% from the 95 to the 54 genes dataset for non-ribosomal genes; 21 to 33% from the 95 to the 27 genes dataset for ribosomal genes). The author don't discuss how this impacts the manual curation they perform on the datasets; however, they state that "manual curation of marker genes is important". The authors might want to discuss that aspect further.

Thanks for raising this point. We were not sufficiently clear in describing the logic of our approach in the first version of the manuscript, and have now revised the text to clarify. In this analysis, we used a strict binary definition of monophyly --- that is, even a single inter-domain transfer leads to non-monophyly (note that this is in contrast to the re- analysis of the expanded set, where we considered whether each marker statistically rejected domain monophyly). For some genes scored as non-monophyletic in this way, manual removal of a small number of unambiguous recent transfers is sufficient restore domain monophyly; for others, HGT is extensive and it is difficult to know how to filter the sequences so as to obtain a reliable marker gene alignment; it was these latter cases that we set aside. We have now revised this section to make the logic of the approach clear, writing:

“Prior to manual curation, non-ribosomal markers had a greater number of HGTs and cases of mixed paralogy. In particular, for the original set of 95 unique COG families (see ‘Phylogenetic analyses’ in Methods), we rejected 41 families based on the inferred ML trees, either due to a large degree of HGT, paralogous gene families or LBA. For the remaining 54 markers, the ML trees contained evidence of occasional recent HGT events. Strict monophyly was violated in 69% of the non-ribosomal and 29% of the ribosomal families. We manually removed the individual sequences which violated domain monophyly before re-alignment, trimming, and subsequent tree inference (see Methods). These results imply that manual curation of marker genes is important for deep phylogenetic analyses, particularly when using non-ribosomal markers. Comparison of within-domain split scores for these 54 markers indicated that markers that better resolved established relationships within each domain also supported a longer AB branch length (Figure 2A).”

In summary and despite the small weaknesses listed above, my opinion is that the authors reach their goal of showning that the AB-branch is indeed a long one, and that the results support the conclusion.

Impact:

The main point addressed by the authors here, the time of divergence between Archaea and Bacteria, is crucial to our understanding of early evolution. The long branch separating Bacteria and Archaea has long been thought to be a long one, and the paper by Zhu et al casted a doubt about the validity of this long-standing hypothesis. Here, Moody et al convincingly establish that the divergence between archaea and bacteria is a profound one. The paper also has profound implications on the validity of the commonly used core-gene phylogenies, particularly those based on ribosomal protein genes. Indeed, it shows that the these proteins are appropriate for deep phylogenies. They also show the impact of model violations on deep phylogenies, and how to avoid them.

We thank the reviewer for this positive assessment of impact.

Author Response:

Reviewer #1 (Public Review):

This is a well-executed study looking at the association of urinary metabolites to the types of diets consumed by European children. They focus on four analytes that have opposing patterns from a "good" KIDMED Mediterranean style diet versus a "bad" diet with processed foods and high sugars. They then create an association with levels of C-peptide, which has in turn been linked to health outcomes.

Overall there is extensive data provided in the supplementary data to justify their findings. The one omission is the effects of activity levels and total caloric consumption. There is an attempt to link body weight to C-peptide associations, but in a minor revision, it would be nice to also include MBI as a parameter for the concentrations of metabolites.

We thank the reviewer for his/her positive feedback. We agree that inclusion of information on physical activity levels and total caloric consumption would strengthen our study. Unfortunately, we do not have available data on these variables. To counteract this, we adjusted all our models for child sedentary behavior (minutes/day of time spent watching TV, playing computer games or other sedentary games) which has been shown to associate to physical activity levels - this association could be due to the fact that the time devoted to sedentary screen-time activities affects availability of time devoted for exercise, or vice versa.(1-3) Further, we adjusted our models for child body mass index (BMI), a measure that strongly correlates to energy intake, and assessed ultra-processed food intake as proportion of total food intake in order to take into account inter-individual differences in total food (and hence caloric) consumption. We clarify these points in the discussion section.

Regarding the second part of reviewer’s comment to consider BMI as a parameter for the concentrations of metabolites, we considered, and controlled for, any potential influence of BMI on the associations of both diet and C-peptide with the urinary metabolites as all our models were adjusted for this measure. We have previously reported the associations of children’s BMI with the urinary metabolome in the same study population (Lau CHE et al (4)) and hence we did not repeat this analysis in our manuscript. In our previous HELIX study by Lau CHE at al, we found significant associations between children’s BMI z-score and three urinary metabolites; positive associations with valine and 4-deoxyerythronic acid and a negative association with pantothenic acid. Of these metabolites, we found that KIDMED score was positively associated with pantothenic acid after adjustment for BMI (Supplementary Table 5), suggesting that Mediterranean diet adherence could affect urinary levels of this metabolite independently of BMI. Further, in our analysis, we found that UPF intake was negatively associated with the branched-chain amino acid (BCAA) valine after adjustment for BMI. Even though the importance of the BCAAs in adiposity has been reported previously, our findings provide an important foundation for future research to better understand the role of UPF intake on BCAA metabolism. Throughout the discussion section, we now discuss our results in context with our previous HELIX analysis examining associations of BMI with the urinary metabolites.

Modified manuscript text:

"We did not have data available on children’s physical activity. Nevertheless, we adjusted all our models for sedentary behavior (including time spent in front of the screen) which has been shown to associate to physical activity levels, as the time devoted to sedentary screen-time activities might affect availability of time devoted for exercise, or vice versa.(1-3) Further, we did not have data available to control for energy intake. However, in all our models, we included BMI of the children, a measure strongly correlated to energy intake,(5) and assessed ultra-processed food intake as proportion of total food intake."

"In addition, adherence to the Mediterranean diet was also positively associated with urinary levels of pantothenic acid and acetate. Both compounds have a central role in human biochemistry and the metabolism and synthesis of carbohydrates, proteins, and fats. Pantothenic acid (vitamin B5, necessary to form coenzyme-A) is present in many foods, and we have previously reported a positive association between consumption of dairy products and urinary pantothenic acid in the same study population.(4) Further, we have previously shown that BMI is negatively associated with urinary levels of this metabolite,(4) and our results suggest that higher adherence to the Mediterranean diet associates with pantothenic acid independently of the potential influence of BMI."

"Moreover, we found that UPF intake was negatively associated with two urinary amino acids, valine and tyrosine. Tyrosine is regarded as a conditionally essential amino acid in adults and essential in children. Foods high in dietary tyrosine include dairy, meat, eggs, beans, nuts, grains. Tyrosine is a precursor for neurotransmitters and hormones, increases dopamine availability which in turn could enhance cognitive performance.(6) Valine is an essential branch chain amino acid (BCAA) critical to energy homeostasis, protein and muscle metabolism.(7,8) In many studies, it has been observed that elevated BCAAs are associated with insulin resistance and diabetes.(9) Also, in our previous HELIX study,(4) we found that urinary valine was associated with higher children’s BMI. However, it remains to be eludicated whether these associations are causal (e.g. via mTOR activation) or consequential (e.g. due to reduced mitochondrial oxidation) in metabolic disease,(9) and whether UPF intake plays a role in the etiology of the association of BCAAs with metabolic health."

References:

1. Serrano-Sanchez JA, Marti-Trujillo S, Lera-Navarro A, Dorado-Garcia C, Gonzalez-Henriquez JJ, Sanchis-Moysi J. Associations between screen time and physical activity among Spanish adolescents. PLoS One. 2011;6(9):e24453.
2. Pearson N, Braithwaite RE, Biddle SJ, van Sluijs EM, Atkin AJ. Associations between sedentary behaviour and physical activity in children and adolescents: a meta-analysis. Obes Rev. 2014;15(8):666-675.
3. Aira T, Vasankari T, Heinonen OJ, et al. Physical activity from adolescence to young adulthood: patterns of change, and their associations with activity domains and sedentary time. Int J Behav Nutr Phys Act. 2021;18(1):85.
4. Lau CE, Siskos AP, Maitre L, et al. Determinants of the urinary and serum metabolome in children from six European populations. BMC Med. 2018;16(1):202.
5. Jakes RW, Day NE, Luben R, et al. Adjusting for energy intake--what measure to use in nutritional epidemiological studies? Int J Epidemiol. 2004;33(6):1382-1386.
6. Kühn S, Düzel S, Colzato L, et al. Food for thought: association between dietary tyrosine and cognitive performance in younger and older adults. Psychological Research. 2019;83(6):1097-1106.
7. Brosnan JT, Brosnan ME. Branched-Chain Amino Acids: Enzyme and Substrate Regulation. The Journal of Nutrition. 2006;136(1):207S-211S.
8. Nie C, He T, Zhang W, Zhang G, Ma X. Branched Chain Amino Acids: Beyond Nutrition Metabolism. Int J Mol Sci. 2018;19(4).
9. Lynch CJ, Adams SH. Branched-chain amino acids in metabolic signalling and insulin resistance. Nat Rev Endocrinol. 2014;10(12):723-736.

Author Response:

Reviewer #1 (Public Review):

5.The reported data point to an important role of the premotor and parietal regions of the left as compared to the right hemisphere in the control of ipsilateral and contralateral limb movements. These are also the regions where the electrodes were primarily located in both subgroups of patients. I have 2 concerns in this respect. The first concern refers to the specific locus of these electrodes. For premotor cortex, the authors suggest PMd as well as PMv as potential sites for these bilateral representations. The other principal site refers to parietal cortex but this covers a large territory. It would help if more specific subregions for the parietal cortex can be indicated, if possible. Do the focal regions where electrodes were positioned refer to the superior vs inferior parietal cortex (anterior or posterior), or intra-parietal sulcus. Second, the manuscript's focus on the premotor-parietal complex emerges from the constraints imposed by accessible anatomical locations in the participants but does not preclude the existence of other cortical sites as well as subcortical regions and cerebellum for such bilateral representations. It is meaningful to clarify this and/or list this as a limitation of the current approach.

On the first issue, we have updated the manuscript to specify the subregion within the parietal cortex in which we see stronger across-arm generalization - namely, the superior parietal cortex. On the second issue, we have added text in the Discussion that reference subcortical areas shown to exhibit laterality differences in bimanual coordination, providing a more holistic picture of bimanual representations across the brain. In addition, we acknowledge that with our current patient population we are limited to regions with substantial electrode coverage, which does not include all areas of the brain.

6.The evidence for bilateral encoding during unilateral movement opens perspectives for a better understanding of the control of bimanual movements which are abundant during every day life. In the discussion, the authors refer to some imaging studies on bimanual control in order to infer whether the obtained findings may be a consequence of left hemisphere specialization for bimanual movement control, leading to speculations about the information that is being processed for each of both limb movements. Another perspective to consider is the possibility that making a movement with one limb may require postural stabilization in the trunk and contralateral body side, including a contribution from the opposite limb that is supposedly resting on the start button. Have the authors considered whether this postural mechanism could (partly) account for this bilateral encoding mechanism, in particular, because it appears more prominent during movement execution as compared to preparation. Furthermore, could the prominence of bilateral encoding during movement execution be triggered by inflow of sensory information about both limbs from the visual as well as the somatosensory systems.

Thank you for these comments. We have added a paragraph to the Discussion to address the hypothesis that some component of ipsilateral encoding may be related to postural stabilization.

In response to the final point in this comment, we agree that bilateral information during execution could be reflective of afferent inputs (somatosensory and/or visual). However, the encoding model shows that activity in premotor and parietal regions are well predicted based on kinematics during the task. While visual and somatosensory system information are likely integrated in these areas, the kinematic encoding would point to a more movement-based representation.

Reviewer #2 (Public Review):

Weaknesses:

1. Although the current human ECoG data set is valuable, there is still large variability in electrode coverage across the patients (I fully acknowledge the difficulty). This makes statistical assessment a bit tricky. The potential factors of interest in the current study would be Electrode (=Region), Subject, Hemisphere, and their interactions. The tricky part is that Electrode is nested within Subject, and Subject is nested within Hemisphere. Permutation-based ANOVA used for the current paper requires proper treatment of these nested factors when making permutations (Anderson and Braak, 2003). With this regard, sufficient details about how the authors treated each factor, for instance, in each pbANOVA, are not provided in the current version of the manuscript. Similarly, the scope of statistical generalizability, whether the inference is within-sample or population-level, for the claims (e.g., statement about the hemispheric or regional difference) needs to be clarified.

We discuss at length the issue of electrode variability and have addressed this in the revised manuscript. Graphically, we have added a Supplemental Figure (S2). Statistically, we appreciate the point about the need for the analysis to address the nested structure of the data. We have redone all of the statistics, now using a permutation-based linear mixed effects model with a random effect of patient. This approach did not change any of the findings.

As to the comment about hemispheric or regional differences, the data show that both are important factors. Our hemispheric effect is characterized by stronger ipsilateral encoding in the left hemisphere and subsequently better across-arm generalization (Figures 2-4). We then examine the spatial distribution of electrodes that generalized well or poorly and found clusters in both hemispheres of electrodes that generalize poorly. In contrast, only in the left hemisphere did we find clusters of electrodes that generalize well. These electrodes were localized to PMd, PMv and superior parietal cortex (Fig 5D). In summary, we argue that activity patterns in M1 are similar in the left and right hemispheres, but there is a marked asymmetry for activity patterns over premotor and parietal cortices.

Additional contexts that would help readers interpret or understand the significance of the work: The greater amount of shared movement representation in the left hemisphere may imply the greater reliance of the left arm on the left hemisphere. This may, in turn, lead to the greater influence of the ongoing right arm motion on the left arm movement control during the bimanual coordination. Indeed, this point is addressed by the authors in the Discussion (page 15, lines 26-41). One critical piece of literature missing in this context is the work done by Yokoi, Hirashima, and Nozaki (2014). In the experiments using the bimanual reaching task, they in fact found that the learning by the left arm is to the greater degree influenced by the concurrent motion of the right arm than vice versa (Yokoi et al., J Neurosci, 2014). Together with Diedrichsen et al. (2013), this study will strengthen the authors' discussion and help readers interpret the present result of left hemisphere dominance in the context of more skillful bimanual action.

The Yokoi paper is a very important paper in revealing hemispheric asymmetries during skilled bimanual movements. However, we think it is problematic to link the hemispheric asymmetries we observe to the behavioral effects reported in the Yokoi paper (namely, that the nondominant, left arm was more strongly influenced by the kinematics of the right arm). One could hypothesize that the left hemisphere, given its representation of both arms, could be controlling both arms in some sort of direct way (and thus the action of the right arm will have an influence on left arm movement given the engagement of the same neural regions for both movements). It is also possible that the left hemisphere is receiving information about the state of both the right and left arms, and this underlies the behavioral asymmetry reported in Yokoi.

Reviewer #3 (Public Review):

In the present work, Merrick et al. analyzed ECoG recordings from patients performing out-and-back reaching movements. The authors trained a linear model to map kinematic features (e.g., hand speed, target position) to high frequency ECoG activity (HFA) of each electrode. The two primary findings were: 1) encoding strength (as assessed by held-out R2 values) of ipsilateral and contralateral movements was more bilateral in the left hemisphere than in the right and 2) across-arm generalization was stronger in the left hemisphere than in the right. As the authors point out in the Introduction, there are known 'asymmetries between the two hemispheres in terms of praxis', so it may not be surprising to find asymmetries in the kinematic encoding of the two hemispheres (i.e., the left hemisphere contributes 'more equally' to movements on either side of the body than the right hemisphere).

There is one point that I feel must be addressed before the present conclusions can be reached and a second clarification that I feel will greatly improve the interpretability of the results.

First, as is often the case when working with patients, the authors have no control over the recording sites. This led to some asymmetries in both the number of electrodes in each hemisphere (as the authors note in the Discussion) and (more importantly) in the location of the recording electrodes. Recording site within a hemisphere must be controlled for before any comparisons between the hemispheres can be made. For example, the authors note that 'the contralateral bias becomes weaker the further the electrodes are from putative motor cortex'. If there happen to be more electrodes placed further from M1 in the left hemisphere (as Supplementary Figure 1 seems to suggest), than we cannot know whether the results of Figures 2 and 3 are due to the left hemisphere having stronger bilateral encoding or simply more electrodes placed further from M1.

The reviewer makes a very valid point and this comment has led to our inclusion of a new Supplementary Figure, S2, in which we quantify the percentage of electrodes in each subregion.

Second, it would be useful if the authors provided a bit of clarification about what type of kinematic information the linear model is using to predict HFA. I believe the paragraph titled 'Target modulation and tuning similarity across arms' suggests that there is very little across-target variance in the HFA signal. Does this imply that the model is primarily ignoring the Phi and Theta (as well as their lagged counterparts) and is instead relying on the position and speed terms? How likely is it that the majority of the HFA activity around movement onset reflects a condition-invariant 'trigger signal' (Kaufman, et al., 2016). This trigger signal accounts for the largest portion of neural variance around movement onset (by far), and the weight of individual neurons in trigger signal dimensions tend to be positive, which means that this signal will be strongly reflected in population activity (as measured by ECoG). This interpretation does not detract from the present results in any way, but it may serve to clarify them.

To address this comment, we have added a new figure (Fig 6) which shows the relative contribution of each kinematic feature as well as their average weights across time for both contralateral and ipsilateral movements. This figure also addresses the reviewer’s question about the contribution of the target position to the model. As can be seen, features that reflect timing/movement initiation (position, speed) make a larger contribution compared to the two features which capture directional tuning (theta, phi). As the reviewer suggested, this result is in line Kaufman et al. (2016) which reported that a condition-invariant ‘trigger signal’ comprises the largest component of neural activity. We note that the target dependent features theta and phi still make a substantial contribution to the model (relative contribution: contra = 32%, ipsi = 37%). Previously, we have tested the contribution of the theta and phi features by comparing two models, one that only used position and speed (Movement model) and one that also included the two angular components phi and theta (Target Model). For a subset of electrodes, the held-out predictions were significantly better using the Target Model, a result we take as further evidence of electrode tuning within our dataset.

The figure below shows an electrode located in M1 that is tuned to targets when the patient reached with their contralateral arm as an example. We believe that having an explicit depiction of how the four features contribute to the HFA predictions will help the reader evaluate the model. These points are now addressed in the text in the results section discussing Figure 6.

Author Response:

Evaluation Summary:

This manuscript addresses outstanding questions about the molecular mechanisms by which the two types of arginine-methylating enzymes affect the processing and fate of transcripts in mammalian cells. This work makes important inroads into these questions, uncovering an inverse effect of the two types of enzymes on intron retention during post-transcriptional splicing, linking the effects to specific target proteins. With better support of some key claims , the paper will provide a lot of new information about the functional consequences of asymmetric and symmetric demethylation.

We thank the reviewers for their support of the study and hope that the described revisions better support the key claims.

Reviewer #2 (Public Review):

Previous work has established that inhibition or knockdown of the Type II (symmetric) arginine methyl-transferase PRMT5 has global effects on splicing, and although it is less well-characterized, loss of the major Type I (asymmetric) enzyme PRMT1 affects both splicing and RNA export activities. In both cases, inhibition or depletion of the enzymatic activities of these proteins has been shown to negatively impact cancer cells, but the specific targets that are required for the RNA processing effects have not been identified. The key findings here are that levels of transcripts containing unspliced introns were inversely affected by the two classes of inhibitor, with intron retention increasing upon PRMT5 inhibition, and decreasing relative to the control in the case of PRMT1 inhibition. The affected introns were shown to be localized to the nucleus, indicating that they belong to the class of 'detained' introns (DI). Using kinetic assays to measure transcriptional elongation and splicing rates, the authors concluded that PRMT inhibition affects DI levels post-transcriptionally. They found that spliceosome component SNRPB and nuclear RNA export factor CHTOP were both enriched in chromatin-associated, poly(A) RNA fractions, that SNRPB was specifically demethylated by PRMT5 inhibitors while PRMT1 inhibition demethylated CHTOP in the chromatin associated fractions, and that both knockdown of the methyltransferases as well as replacement of the modified arginine residues in each protein recapitulated the effects of the inhibitors. Together, these experiments provide strong evidence supporting a coherent mechanism of differential arginine methylation on RNA processing. They support and significantly extend previously published observations implicating the PRMT enzymes in gene expression. These findings are of broad interest to those who study RNA processing and transcription, cancer biology, and signaling through post-translational modifications.

We thank the reviewer for support of our main findings and that they are of broad interest.

Author Response:

Evaluation Summary:

The present work aims to increase our understanding of marine epizootics caused by the dinoflagelate parasite Hematodinium sp. in crabs. The work includes a large data set of field collected specimens from a wide geographical area. The authors have evaluated presence or absence of this parasite as well as co-infections by several other groups of pathogens and model the main factors that shape crab community structure. The topic of study is very important in the context of current marine pandemics and, therefore, adequate examination of this data set may lead to significant advances in the field. Refinement of the approaches to produce quantitative data is needed in order to reach to more solid conclusions.

We are grateful to the reviewers and editorial members of eLife for their evaluation of our submission and recognising the importance/rarity of our dataset for advancing our understanding of marine epizootics. In our revised text, we have included further quantitative data in the context of Hematodinium-parasitised crabs and re-run all our analyses.

Reviewer #1 (Public Review):

The study would have benefited from qPCR instead of presence or absence. It is interesting to note that differences were observed when CFUs were compared. I understand that for most organisms, in the absence of a draft genome, assigning copy numbers to cells is not yet possible; however, it would have provided a more robust dataset for the statistical analysis.

Similarly, histology relies upon one of a hundred possible for every single specimen. With this caveat, scoring the presence of the parasites rather than (+) (-) would also have been more informative. Without knowing the intensity of the infection for each of the other pathogens makes it more difficult to reject the hypothesis. The authors discuss other papers with cellular immune data, which is lacking in this manuscript. Observing fresh hemolymph and histology are just not enough. The data strongly support that parasite community composition is affected by the location. Smaller crabs also appear to be more likely to display co-infections compared to disease-free crabs.

We thank reviewer 1 for spending their time considering our submission.

We used population genetic markers for Hematodinium and crabs, and PCR-based molecular diagnostics, in addition to freshly withdrawn haemolymph and multi-tissue histology in our original submission. In our revised text we have incorporated additional quantitative Hematodinium loads in haemolymph, gill and hepatopancreas in crabs.

In our original submission, we presented data that discriminated between Hematodinium-positive crabs (n = 162) and Hematodinium-negative crabs (n = 162) based on targeted PCR. Rather than use qPCR – due to the absence of sufficient molecular information to ensure single-copy gene targets per genome for Hematodinium spp. and the diversity of ecotypes presented in our study – we have incorporated count data (actual number of parasites per mL haemolymph) from the original liquid tissue screens using haemocytometry, and grade data (0-4) from solid (gill and hepatopancreas) tissues assessed using histology. In our view, these quantitative data are more powerful than those that we could have achieved via qPCR.

We have analysed these additional data in the context of co-infection presence. Indeed, when we restricted the Hematodinium-positive crabs to n = 111 based on individuals that were positive for the parasite in all three diagnostic methods across key tissues (haemolymph via haemocytometry and PCR, and gill/hepatopancreas via histology), and re-ran all our analyses/models, the outcomes did not yield contradictory conclusions. Hematodinium-positive crabs were no more likely to contain co-infections when compared to Hematodinium-negative crabs, AND, location remains the determining factor for pathogen diversity among crabs.

Reviewer #2 (Public Review):

Strength:

1. The authors looked into the prevailing idea that parasitic infection make crab immune-compromised although evidence to support this idea is lacking except one study by where immune gene expression was found to be modulated upon Hematodinium infection in Japanese blue crab. Apparently, the authors studying gene expression in Japanese blue crab upon parasitic infection did not identify any effort molecules of parasitic origin.
2. It was interesting to note that haplotype diversity analysis revealed higher genetic diversity in the host compared to the Hematodinium parasite in two locations examined.

We thank Reviewer 2 for their time and insight.

Weakness:

1. The authors should clarify how the sample sizes were selected.

We apologise for the unintentional ambiguity. The initial survey covered ~50 crabs per location per month for 1 year (n = 1191) - using an alpha value of 0.05 and desired power >80% indicated a minimum of 38 (1-sided test) or 48 (2-sided test) crabs were needed based on an a priori prediction of 15% Hematodinium prevalence in the C. maenas population. The 162 Hematodinium-positive crabs (PCR) were size/sex/location matched to 162 Hematodiniumnegative crabs.

The methods section has been edited to ensure that this information is clear.

1. It is also recommended that details on sample collections are included (for example the times in which samples from the surrounding waters of infected crabs)

More details on sample collection have been included in our revision

Reviewer #3 (Public Review):

This manuscript aims to assess the main factors that drive disease in crabs across Europe. Crabs are critical members of the trophic chain in marine environments and also serve as food source for human consumption. Therefore, understanding the health of this group of organisms is very important for ecosystem health as well as human health. This is a well written manuscript that identifies important knowledge gaps in the field. The authors evaluate whether crabs are infected by an important parasitic dinoflagelate, Hematodinium, and question several current dogmas in the field. The first one is whether or not presence of Hematodinium infection is a significant factor driving co-infections by other pathogens including trematodes, bacteria or fungi. The authors, based on their data set, conclude that this is not the case, since co-infections are observed in similar proportions in Hematodinium free animals. The authors perform in depth assessment of other drivers of infection and identify geographical location as the main factor driving community structure.

We thank Reviewer 3 for their time and insight.

A second important aspect of Hematodinium biology is the previous notion that this parasite immunosuppresses the crab host (and therefore co-infections may be found). However, there is a paucity of studies to confirm or reject this hypothesis, and therefore, the current study is important to expand our current knowledge of marine invertebrate immunobiology and specifically in the mechanisms by which Hematodinium is harmful to the host. The current study provides some evidence that, in fact, this parasite is not an immune suppressor but rather evades the cellular immune response of the host. In particular, the authors claim that the host hemocytes (immune cells) fail to phagocytose (engulf) and encapsulate (surround and fend off) this parasite. These two cellular immune responses are the most common in invertebrates such as crabs.

Overall, this study is important because very few studies evaluate wild marine diseases, especially in invertebrate hosts and because it contributes with a large data set from a wide geographical range.

We thank Reviewer 3 for their time and insight.

Author Response:

Reviewer #1:

This manuscript explores the role of ER and PR in the endometrial cancer cell model called the Ishikawa cell line. The authors conduct a series of detailed experiments to assess the estrogen (E2) and progesterone (R5020) response in this model and show that E2 can promote cell growth which is subsequently inhibited by co-treatment with R5020. RNA-seq revealed E2 or R5020 gene targets with most differential genes being unique to the treatment condition. The functional role of PR was assessed and confirmed on a specific locus of interest and using a reporter assay. ChIP-seq was conducted revealing gained PR binding events following R5020 and some that were already present prior to treatment, as well as sites that were lost. Substantially more PR binding sites were observed in the T47 breast cancer model and the authors mimic the elevated levels of PR, by expressing exogenous PR in Ishikawa cells and conducting a series of ChIP-seq experiments. Analysis of specific binding regions revealed the enrichment of motifs for the Pax family of transcription factors and the authors assess the hormonal regulation of PAX2 cellular expression. PAX2 ChIP-seq was conducted, revealing very few binding peaks and these were partially integrated with the PR and ER binding peaks. Finally, a Hi-C experiment was conducted, revealing that ER, PR and PAX2 binding occurs in genomic compartments and specific gene signatures were derived from this analysis.

This is a topical area and the work is of potential interest, but several key issues need to be addressed:

• The role for PAX2 (over other family members) is inferred by the enrichment for motifs specific to that family member, but the motif enrichments are not good as defining individual family members that share a common motif. What are the expression levels of the PAX family members in the Ishikawa cell line and in primary endometrial cancers, to support the role for PAX2 over other family members? This wouldn't require any experiments and could simply involve analysis of public expression datasets.

We agree with the reviewer that “the motif enrichments are not good as defining individual family members that share a common motif”. Besides choosing PAX2 because its motif was the most represented in PR and ERbs, we evaluated expression of PAX2 and other family members in publicly available normal and endometrial cancer samples. The expression levels of the PAX family members in Ishikawa cells and in primary endometrial cancer samples have been detailed in Figure 5, Figure 5-Figure Supplement 1, Figure 7 and Figure 7- Figure Supplement 1 (see new version of the manuscript). Although other PAX family members seem to be more expressed than PAX2 in tumors (like PAX8), many reports link the loss of PAX2 in endometrial tissue to bad prognosis of tumors (EIN) (see Sanderson et al, 2017 in revised manuscript). This was included in the discussion and now it has been included in the corresponding result argument connected to Figure 5 and Figure 7 of the revised manuscript.

• The authors conclude that PAX2 binding overlaps with PR in pre-treated cells, but data in Figure 5C and 5D could simply represent co-binding at open enhancers, which are notorious for recruiting many transcription factors that are expressed in that cell type. What is the overlap in peaks between PAX2 and PR/ER, ideally via a Venn diagram or some visual that allows for a comparison of the total number of peaks for each factor and the common ones? Is there a statistically enriched co-binding of PAX2 to PR/ER sites, at levels that are more than expected?

The fraction of overlap between sets of binding sites was studied in a pairwise fashion and is graphically represented in Figure 5C of the revised manuscript. Also, we performed a Fisher test to statistically evaluate the significance of the co-localization between PAXbs and hormone receptors which were included in Figure 5E and F. Results were included in text of revised manuscript (line 314).

• The link with PAX2 is potentially exciting but is not convincing in its current form. The only real evidence linking PAX2 to ER/PR is that PAX2 cellular location can be altered and there are some binding peaks where there is co-enrichment, but this would likely happen with any transcription factor expressed in that cell line model. What is currently missing (and essential), is some evidence providing a functional link between ER/PR and PAX2. Is PAX2 required for PR and/or ER function, either ER/PR binding or induction of target genes? If not, then the data on PAX2 is circumstantial and isn't really relevant to the transcriptional pathways regulated by PR or ER.

We completely agree with the reviewer. Functional data linking PAX2 to ER/PR signaling were studied using depletion of PAX2 with specific siRNA, showing clear effects on PR binding and hormone-regulated genes. These effects were observed in both E2 pretreated and non-pretreated cells, indicating that PAX2 is involved in many aspects of PR regulatory action and its interplay with ERalpha. Results on PR binding and gene expression with or without siRNA against PAX2 have been included in new Figure 5 (line 329 of revised manuscript). M&M and discussion were also revised accordingly.

• There is no explanation put forward to the 307 lost PR sites.

The sentence has been removed because we do not have replicates of 30min time point ChIPseq samples.

• The GEO dataset indicates that only one replicate was conducted for the ChIP-seq experiments. This does not meet the minimum ENCODE requirement and many of the differential peaks (i.e. the 307 lost peaks) are potentially false positives that result from having only one replicate.

We processed data from replicates and found similar results between them (see Figure 2 -Figure Supplement 2). While PRbs in E2-pretreated cells and ERbs were mostly reproduced (80% and 70%, respectively), PRbs in non- pretreated cells showed 40% of common peaks. This is mainly due to the fact that one of the replicates produced a higher number of peaks above threshold during peak calling, forcing a lower percentage of common peaks. However, we have demonstrated that almost all reported PRbs were independently reproduced by PR ChIP-seq data in other conditions, namely E2-pretreated PRbs and PRbs from FPR cells (Figure 2 and Figure 2-Figure Supplement 2). It is important to note that only PRbs from E2-pretreated cells and ERbs promoted the main conclusions of our manuscript. We will upload new data to GEO as soon as possible.

• The authors claim that the motif enrichment supports a conclusion where monomer PR could bind at 30 min and dimers at 60 min, but there is no direct evidence that this is the case. Unless the authors plan to pursue this functionally and can show dimeric vs monomeric binding, this statement should be removed, as it is not backed up by data and the presence of a half site vs a full palindromic motif does not provide evidence for the genuine mode of binding.

The statement has been removed.

• All the work is conducted in a single cell line model. I understand that there are few endometrial cancer cell line models and I also acknowledge that the authors have conducted a complicated series of genomic experiments and it would be unrealistic to repeat these in another model. However, the findings from this one model should reveal new insight that can be validated in either another model or in a cohort of clinical samples of the cancer types. But, in its current state, neither are done. The authors attempt to extract gene signatures from the genomic data to assess in patient cohorts, but the data (see my next comment) is not compelling or convincing and the only conclusion I can make, is that out of the hundreds of somatic mutations and hundreds of PgCR genes, only a handful of genes correlate with outcome. I suspect the same conclusion could be made with a random set of several hundred genes.

We reformulated the analysis of tumor samples to avoid biased conclusions. Results are part of the new version of Figure 7.

Author Response:

Reviewer #2 (Public Review):

Yu et al provide a comprehensive set of experiments to determine that bradyzoites have much slower cytosolic Ca2+ parameters, which impact on gliding motility, a key process of Toxoplasma spread and persistence.

The only main criticism that I have is the use of the MIC2-GLuc reporter to measure microneme secretion in bradyzoites. Do bradyzoites have any appreciable level of MIC2 and its associated protein M2AP?? This is important that may affect the outcome. If bradyzoites do not, then the MIC2-GLuc reporter might not have appropriate levels of M2AP to correctly traffic to the micronemes. I recommend that the authors quantitate, either by western blot or IFA, the levels of MIC2 and M2AP in bradyzoites versus tachyzoites and also show that M2AP co-localises with MIC2-GLuc to give confidence that MIC2-GLuc is trafficked correctly and thus the low readings of secretion are not just a result of the reporter mistrafficked. It would also be pleasing to see, that 1hr incubation leads to restoration of MIC2-GLuc secretion.

We acknowledge that the expression and localization of MIC2-Gluc reporter is a potential concern. We performed western blotting (Figure 2C) and IFA (Figure 2 supplement 1A) to confirm that bradyzoites express MIC2-Gluc and M2AP albeit at lower levels compared with tachyzoites. Moreover, MIC2-GLuc and M2AP were properly co-localized to the apical end in bradyzoites, ruling out the possibility of mis-localization of the MIC2-GLuc reporter. Based on these results, we believe that MIC2-GLuc provides a reliable read-out for microneme secretion in in vitro differentiated bradyzoites. Additionally, the conclusion that MIC secretion is dampened in bradyzoites is also supported by the studies using the FNR-Cherry reporter in Figure 2E,F,G.

Reviewer #3 (Public Review):

This is a first study that looks in detail at Ca-controlled gliding motility and ATP supply in bradyzoites. A comparison of such different parasite stage by manipulating Ca and ATP metabolism is challenging. Intervention by chemical compounds needs to overcome a prominent cyst wall and the usage of genetic tools needs to consider the broad changes in protein expression between tachyzoites and bradyzoites as well as a heterology between individual bradyzoites. The authors used excysted bradyzoites to exclude the cyst wall as a diffusion barrier as a major factor in the efficacy of different Ca agonists. To address differences in expression levels between tachyzoites and bradyzoite stages the authors developed a ratiometric Ca sensor based upon an autocleaved GCaMP6f-BFP dimer protein.

Overall the conclusions are well supported but there are methodological questions that need to be addressed.

Bradyzoites show a heterogenous expression of Bag1 / Sag1 markers as well as heterologous proteins. This is shown in Fig 1A and Fig 2b for example. However, in most time-dependent measurements of Ca-dependent fluorescence (Fig 2G, 3D the authors only average three cells. This appears to be insufficient to represent the bradyzoite population. How is the variance between the three measured cells?

We have quantified more cells in all figures related to fluorescence measurements. For measurements of single parasites in Figure 5B, 5D, 5E, 6F, 8A, 8B and Figure 7 supplement 1A, we have now quantified 10 parasites for each condition and plotted the data as means ±S.D. to show the variance. For in vitro induced cysts or ex vivo cysts in Figure Fig 2G, 3D, 3E, 4C,4G, 6E, 7B and Figure 4 supplement 1A, we measured 5 cysts or vacuoles per condition. Because these samples contain many parasites within each vacuole or cyst, they represent a greater sample size. The data are also plotted a means ±S.D.

In addition, the Mic2 promoter driven Gluc-myc protein is not expressed in all bradyzoites. This is perhaps not suprising as Mic2 seems to be downregulated in bradyzoites according to Pittman and Bucholz et al dataset in ToxoDB. If interpreted correctly the lower expression of Gluc in some bradyzoites would favour an underestimation of the RLUs in Fig 2D.

We acknowledge that the expression and localization of MIC2-Gluc reporter is a potential concern. We performed western blotting (Figure 2C) and IFA (Figure 2 supplement 1A) to confirm that bradyzoites express MIC2-Gluc and M2AP albeit at lower levels compared with tachyzoites. Moreover, MIC2-GLuc and M2AP were properly co-localized to the apical end in bradyzoites, ruling out the possibility of mis-localization of the MIC2-GLuc reporter. Based on these results, we believe that MIC2-GLuc provides a reliable read-out for microneme secretion in in vitro differentiated bradyzoites. Additionally, the conclusion that MIC secretion is dampened in bradyzoites is also supported by the studies using the FNR-Cherry reporter in Figure 2E,F,G.

The maturation of bradyzoite takes several weeks. This cannot be accomplished with currently available system in vitro and the authors use 1 week matured bradyzoites. To facilitate comparability to data from other manuscripts it would be helpful if the authors could quantify the differentiation stage of the in vitro bradyzoites. This could be done by measuring the fractions of Bag1-positive and Sag1-negative bradyzoites.

We thank the reviewer for this useful comment. We have quantified the percentage of BAG1-positive SAG1-negative bradyzoites within each cyst induced for 3, 5 or 7 days by IFA and spinning disc confocal microscopy (Figure 3 supplement 1A). This analysis demonstrated that the percentage of BAG1-positive and SAG1-negative bradyzoites reached ~70% at day 7 after induction (Figure 3 supplement 1B). For this reason, we used a 7 day induction treatment for the majority of experiments. Also, where imaging was used in the analysis, we focused on regions of in vitro differentiated cysts that expressed high levels of BAG1-mCherry.

The mcherry and GCaMP6f signal in fig 3B seem mutually exclusive. This may be due to difference in calcium signalling between Bag1 pos or neg parasites or due to expression differences of GCaMP6f.

To test the possibility of expression differences in GCaMP6f, we quantified the fluorescence of BAG1-mCherry and GCaMP6f in different bradyzoites within the cyst shown in Figure 3B. At time 0 prior to stimulation, we observed heterogenous expression of BAG1- mCherry while the signal for GCaMP6f expression was relatively constant (Figure 3B supplement 1C and 1D). In contrast, when in vitro differentiated bradyzoites were stimulated with A23187, they showed reduced levels of GCaMP expression in cells that were strongly positive for BAG1-mCherry (Figure 3B). Collectively, these findings are consistent with the difference in GCaMP fluorescence being due to dampened calcium responses in bradyzoites rather than expression differences. This conclusion is supported by studies on GCaMP responses in cells where we normalized for expression level using a dual-expression BFP reporter in Figure 6. Therefore, we do not think that heterogeneity in the expression of GCaMP is responsible for the observed dampened response in bradyzoites.

The authors use syringe, trypsin-released and FACS sorted bradyzoites in multiple Ca assays. How can it be excluded that this procedure affects (depletes) Ca stores?

In all the figures except Figure 2C-2D, we did not use FACS to sort bradyzoites. Instead, we scraped cells cultured at pH 8.2, used syringe passage through 25g needle followed by centrifugation. Cyst pellets were resuspended and digested with trypsin to liberate bradyzoites. For tachyzoites, all procedures were similar except that we did not use trypsin digestion. As a control, we have now treated tachyzoites similarly with trypsin and monitored the calcium stores using ionomycin. We found that trypsin digestion did not affect the calcium stores or response as shown in Figure 7 figure supplement 1A.

In my opinion several experiments in this manuscript would benefit from clarification of this point. For example: In Fig 7A Fu et al measure Ca for 5min during trypsin digestion, however, for gliding assays cysts are digested for 10min. The Ca monitoring should cover the complete 10min off trypsin digest.

We understand the concern but there were practical reasons for the slightly different times used. In panel A where we are monitoring calcium during trypsin digestion, the majority of cysts are dispersed after 5 min resulting the parasites being out of focus. As such, it is not practical to monitor beyond this time point. In the panel C, we were interested in observing parasites after the cysts where fully digested and hence we used a slightly longer time period to allow complete digestion and for the parasites to settle to the bottom of the dish before further recording. In this instance, similar to the result in A, most parasites remained dormant and did not show elevated calcium levels. In the figure, we are selectively showing a rare example where calcium signaling was observed in order to compare the patterns to what is normally observed with tachyzoites. These combined panels are not meant to be a comparison of kinetics, as this aspect is tested more directly in later experiments. We have modified the text to make the rationale for this experiment clear.

In Fig 2B Fu et al digest infected monolayers with trypsin to release mcherry from cysts matrices. How can the authors exclude that trypsin is not digesting mCherry protein in this assay?

I think the reviewer means 2F as in 2B we are using BAG1 mCherry to visualize bradyzoites – but they are not being liberated in this image. In 2F we use a different construct, FnR-mCherry that directs the reporter to be constitutively secreted to either the PV (surrounding tachyzoites) or the cyst matrix (surrounding bradyzoites). When the cysts are disrupted with trypsin, the mCherry is likely to disperse and may also be digested. However, this would not happen if it remains inside the parasite. This control is provided to show that the protein is secreted into the matrix. We have revised the text to clarify the use of this control.

Fig 7 E,F: the authors measure shorter gliding distances of bradyzoite as compared to tachyzoites. Trails of both parasites however, are detected by visualizing using different antigens that may have different shedding behavior on the FBS-coated glass surface. The Bag1 trail also depends on Bag1 expression, which is shown in numerous images to not be equal among individual bradyzoites. This point is very challenging to address but should at least be discussed.

BAG1 is used here to discern the bradyzoites, not to detect the trail. Trails are stained with either SAG1 or SRS9 – corresponding to the most abundant surface GPI anchored antigen in each stage. Since these proteins are part of the same C-C fold family and are similarly anchored, we feel they are comparable. We have added the following statement to the results: “These two surface markers are both members of the cysteine rich SRS family that are tethered to the surface membrane by a GPI anchor, thus they represent comparable reporters for each stage.”

Fig 7E: Bradyzoites are considered to satisfy their ATP needs mostly via glycolysis and the data shown do support this capability. I find the ability of OligomycinA to block glucose-dependent gliding surprising as this suggests a necessary mitochondrial transport chain for ATP-production from glucose. This result should be mentioned clearly in the text and its implications discussed.

The Discussion has been revised as suggested.

Figure 8: The authors claim a recovery of bradyzoite ATP and Ca levels after 1hr incubation with carbon sources and Ca, that together enable efficient gliding. However, the elevation of bradyzoite ATP occurs after the parasites spend 2 hours in glucose-free and Ca-free conditions, whereas gliding assays are done after a short 10min trypsin digest. I am not entirely convinced that low ATP levels post-egress are responsible for the low gliding activity. Ideally gliding assays should be done after a similar purification procedure to correlate the two experiments.

We have repeated the gliding assays using bradyzoites purified in the same manner as for the ATP measurements and found the same result that a combination of exogenous calcium and glucose enhance recovery of gliding motility (Figure 8D, 8F). In addition, we used the same time point to purify bradyzoites for MIC2-Gluc secretion and found exogenous calcium and glucose also led to an increase in MIC2-GLuc secretion, indicative of the recovery of microneme secretion (Figure 8C).

Author Response:

Reviewer #3 (Public Review):

Miskolci et al have investigated if it is possible to measure the natural fluorescence of two important co-enzymes (NADH/NADPH and FAD) in living cells to determine their metabolic status. This tests the hypothesis that changes to the relative ratio of NADH/NADPH to FAD+ reflect a shift between glycolytic and oxidative phosphorylation in living macrophages. To investigate this they have used 2-photon FLIM to measure intensity and fluorescence lifetime of NAD/NADPH and FAD+ in mouse macrophages in vitro and zebrafish macrophages in vivo in a tail injury model. By comparing their measures of NAD(P)H and FAD+ from macrophages responding to different injury or infection cues and comparing this to a maRker of inflammation (TNF-alpha) they argue that there is a reduced redox state indicative of glycolytic metabolism in pro-inflammatory macrophages.

The adoption of label free imaging techniques to measure metabolic processes in cells in vivo is a valuable and important development that, although not novel to this work, will help researchers to probe cell biology in situ. FLIM using time correlated single photon counting (TCSPC) allows an accurate and robust measure of the relative state of a molecule that shows changes in its fluorescent lifetime as a consequence of changing chemical state. Although Stringari et al (doi.org/10.1038/s41598-017-03359-8) were the first to describe the utility of wavelength mixing FLIM for measuring NAD(P)H and FAD+ levels in zebrafish, they did not focus on macrophages which is the focus of this work.

The results from this work are are interesting, as they argue that it is possible to determine cell metabolism in cells within living animals without a need to use a genetically encoded sensor and they argue that pro-inflammatory macrophages in zebrafish appear to have a lower redox state, which may reflect a more glycolytic metabolism. This assumption is not tested but rather inferred based on the measures of fluorescence intensity and lifetime of endogenous NADH/NADPH and FAD coupled with a small metabolic sampling of injured tissue. This lack of corroboration for a the supposed difference in metabolism between pro-inflammatory and non-inflammatory macrophages is a weakness of the paper and makes it hard to accept the conclusion that the redox state may reflect different metabolic profiles. A biosensor for NADH/NADPH (iNap) has been demonstrated to be a sensitive tool for measuring NADPH concentration in vivo in zebrafish during the injury response (Tao et al (doi: 10.1038/nmeth.4306) and it would be intriguing to know how similar the response is of this biosensor to the label free measurements described using FLIM. This is additionally relevant as the authors also note that in mouse macrophages cultured in vitro, they observe an opposite redox response which is well supported by the literature and a variety of different methods. Why the zebrafish macrophages should show a different redox state to mouse macrophages is not clear and an alternative explanation is that the measures used do not directly reflect the metabolic profile of the cells. One further caveat to the chosen method of using fluorescence lifetime to measure the redox state of NADH/NADPH is that lifetime of NADH is affected by which proteins it is bound to. This is not accounted for in the method used for calculating the redox ratio used for defining the redox state and could potentially alter the interpretations of relative NADH/NADPH levels in a cell. The authors acknowledge this, but do not consider whether this would affect the conclusions they arrive at from their measures of NAD(P)H intensity and fluorescence lifetime in macrophages.

We thank the reviewer for their comments. We have added additional data that indicate that the imaging does indeed reflect the metabolic profile of the cells (see Metformin and STAT6 data).

Author Response:

Reviewer #1 (Public Review):

The reviewer believes that there is a fundamental problem with the approach of the current MS. Dense reconstruction from serial EM images is a powerful tool for revealing the connectivity matrix in many brain area, where the majority of synaptic connections are made by glutamatergic pyramidal and GABAergic interneurons. Many studies have convincingly demonstrated that the site of synaptic communications among these cells is the well-known EM defined synapses with a presynaptic cloud of vesicles, a rigid presynaptic active zone membrane facing a rigid postsynaptic membrane that either has or does not have a pronounced postsynaptic density. We know from many EM localization results that e.g. the active zone contains the essential molecules of the release sites, the presynaptic Ca2+ channels and the PSD contains the appropriate receptors. Thus, with this information, when the connectome is created from serial EM sections, the sites of communication can be defined based on the EM images. To the knowledge of the reviewer, such pre-existing information is lacking for the DA varicosities. The authors argue that almost all varicosities lack synapses. Verification of such a statement would require the molecular characterization of these varicosities, demonstrating that the molecules essential for vesicle docking/priming/release are lacking. However, if these molecules are present in these varicosities without forming an apparent active zone, then the conclusion of the MS is misleading.

We thank the reviewer for this point and now believe we have addressed it in the discussion.

The authors demonstrate the clear labeling of DA neuronal processes using the cytoplasmic-targeted Apex2. However, due to the well-known masking effect of DAB precipitate in the cytoplasm, which prevent the unequivocal identification of vesicles, the authors decided to use the mitochondria targeted Apex2 in the first half of the MS. However, for the cocaine part, they turned to the cytoplasmic version for some reasons. They then analyzed the axonal branching structure and the varicosities/contact points. The reviewer cannot see how this later was achieved with densely filled DAB containing structures.

We apologize for the confusion. There are several reasons we turned to cyto-Apex for the cocaine part of the MS. We have added a broader discussion of this topic and reproduced below.

"we used cyto-Apex2 for reconstructing axons and their contact points for a several reasons. First, we found that tracing axons in low resolution EM data sets, for both controls and experimental groups, was substantially easier in cytoplasmic Apex2 axons, thus increasing our tracing throughput. In addition, we found that contact points, (e.g., spinules), were also easier to detect. Detecting either a darkly Apex2 labeled cytoplasmic process in an unlabeled structure, or vice versa, was easier because of the stark contrast between Apex2 labeled and unlabeled processes. In Figure 6 and Figure 6-Fig Supplement 2 an example of this difference is shown. While it is possible that cytoplasmic Apex2 expression could potentially obscure the internal contents of varicosities (e.g., synaptic vesicles or endoplasmic reticulum), we found little evidence that cyto-Apex2 obscured the relevant ultra-structural features that were investigated here."

The evidence for DA varicosities making synapses (Fig4) is not convincing. The presented EM images does not have the quality/resolution to see the opposing rigid pre- and postsynaptic membranes and the widening of extracellular space in the cleft.

To make the results clearer, we now show an 8-panel montage of each putative DA synapse as Figure 4- Figure Supplement 1 and 2. We hope that showing the serial sections that span the synapse will make details of the pre- and post-synaptic membranes more convincing. All the images in the manuscript are SEM, not TEM.

Analyzing the structures immediately next to DA varicosities is questionable. If DA is indeed a volume transmitter, how would the authors know how far it can exert its effect. 1 or 5 microns? If 5 microns, there are many structures of all kinds (axon, glia, spine, dendrite) and only their DA receptor content will tell whether they are sensitive or not (and not necessarily their physical distance) to the released DA.

We agree with the reviewer that this analysis distracts from the main finding of the story and is better investigated with a more accurate model of which types of cellular processes are sensitive to released DA. We have removed these experiments and their discussion from the manuscript.

Author Response:

Reviewer #1 (Public Review):

In their manuscript, Sengupta et al. describe a developmental mechanism that positions a single neuron across multiple layers in the hierarchical C. elegans nerve ring. The authors show that neighborhood placement of the interneuron AIB is established during embryogenesis and is maintained throughout development. AIB is one of the few C. elegans neurons that are divided into distinct pre- and post-synaptic regions, and its axons curiously occupy two physically separated neighborhoods or layers. How this occurs is not known. This study uses time-lapse imaging to show that unlike canonical axon tip outgrowth mediating fasciculation in a target region, AIB's axon occupies two neighborhoods by first growing completely into one, and then gradually unzippering from the first, switching, and zippering onto the second neighborhood. Importantly, axon outgrowth and neighborhood choice are continuously visualized during embryogenesis, an impressive experiment typically constrained by lack of cell-specific reporters during early development as well as the struggle of imaging embryos.

The authors posit that zippering is mediated by temporally regulated differential adhesive forces between AIB's neighboring pre- and post-synaptic neurons. How this differs from differential adhesion in classic fasciculating neurons is described but could be made much clearer. They proceed to identify the immunoglobulin syg-1/syg-2 receptor-ligand pair to be necessary and sufficient for AIB's axon switch; in syg-1/syg-2 mutants, AIB is not able to position itself in the second neighborhood and remains fasciculated with the first one, suggesting that adhesive forces are dampened in syg-1/syg-2 mutants. Lastly, the authors show that pre-synapse assembly follows zippering, linking AIB axon placement with synaptogenesis, and that this is also compromised in syg mutants.

The pipeline used to study axon outgrowth at a single-cell level in the embryo at relevant time points is commendable and will be useful to people studying C. elegans nervous system establishment. Although the overall manuscript and data are well-presented, we think the mechanism of retrograde zippering could be better described. Also, syg-1/syg-2 expression needs to be delineated to support the notion of differential adhesion between neighborhoods.

We have further clarified the novelty of the zippering mechanism, contrasting it with tip-directed outgrowth. We have also performed a thorough analysis of syg-1 expression.

Reviewer #2 (Public Review):

A large amount of data is presented in this paper. The experiments are carefully documented and support the conclusions. Of particular importance is the live imaging of the outgrowth of the AIB neurite in the embryo. This is challenging and required the development of a new marker for labelling and the adaptation of a new type of microscope. This enabled the initial and surprising observation that part of the neurite relocates after outgrowth. I'm not sure that the mathematical modeling adds much. The main conclusion is that the modeling is consistent with a "net increase of adhesive forces in the anterior neighbourhood", which is to be expected. The authors then try to identify the relevant adhesion molecules and find that a pair of IgCAMs (syg-1 and syg-2), which are known to act as receptor-ligand pair, are involved. A series of experiments establishes that syg-2 act in the AIB neurons, whereas syg-1 does not. The neurite positioning defects in syg-1 and syg-2 mutants are partially penetrant, suggesting that other adhesion molecules must be involved. While a large percentage of mutant animals show defects, the defects within an individual animal are surprisingly low with only 21.5% +/- 4% of the neurite detached. This would suggest that syg-1/syg-2 aren't even the major adhesion molecules involved here. In further studies, where the authors ablate the RIM neurons (which express syg-1), the authors use a different measure to quantify the defects (minimal distance between neurite segments, Suppl Figure 7). This makes it difficult to compare the results to those of the syg-1 mutants. For the ectopic expression experiments with syg-1 the authors only report the percentage of animal with defects and not the extent of the defects (how much of the neurite was in an abnormal position).

Overall, this is a very detailed study describing an important novel mechanism for neurite positioning within an nerve bundle.

We have added in this revised version additional quantifications, including ‘the minimal distance between neurite segments’ measure (the one used for the RIM ablation experiments) in Figure 4-figure supplement 1 for the placement defects in syg-1(ky652) and syg-2(ky671) mutants, allowing direct comparisons between the phenotypes. We have also added a measure for the percentage of the distal neurite that is mispositioned in the ectopic expression experiments (Figure 6-figure supplement 1A). We do not claim that SYG-1/SYG-2 are the only adhesion molecules involved in AIB neurite placement, but that they are required for complete and proper placement of the neurite. We clarify this in the text.

Reviewer #3 (Public Review):

This is a very interesting manuscript describing the changes of neurite position in a complex neuropil during development. The experimental system is well chosen because AIB's function within the circuit requires its neurite to be in two different neuropil "neighborhoods". The manuscript included some technically difficult experiments of imaging neurite outgrowth in C. elegans embryos which are very hard to do. The surprising finding here is that neurite position is not sole dependent on its growth cone navigation. In the case of the AIB neuron, the growth cone is anchored after it reaches its destination point and then a segment of the neurite shift direction towards its final position through a zippering action. They also show that this shift in position is driven by adhesion molecules SYG-1 and SYG-2. Overall, I think this is a strong candidate for eLife. I have one main point and a few minor points.

My main point is about the relationship between synapse formation and neurite zippering. In my opinion, this is an interesting point because it would tell us if the zippering behavior is a consequence of synapse formation or it is a distinct specificity step before synapse formation. From the time course that was described in the paper, it seems that the accumulation of RAB-3 only starts after the zippering has completed. I would suggest the authors to examine at least another synaptic marker like SNB-1 or SYD-2. We have created cell specific endogenous labeling of several active zone markers that can be used for these experiments. If the results hold, then, I think the authors should make it clear in the text that the zippering takes place before synapse formation and serves as a distinct step in achieving the neighborhood specificity.

We thank the reviewer for generously sending us strains for cell-specific endogenous active zone protein labeling (McDonald et al., 2021) using the SapTrap method (Schwartz and Jorgensen, 2016). We made constructs expressing FLP recombinase downstream of the inx-1 and unc-42 promoters for cell-specific labeling of these active zone proteins in AIB and injected them into the strains. Although we observed cell-specific synaptic signal in larvae with both inx-1p and unc-42p-driven FLP, we were not able to observe signal during embryogenesis, probably due to cell-specific synaptic protein expression levels being low.

Therefore and to address the reviewer’s question about the temporal order of zippering and synapse formation, we have cell-specifically expressed two active zone proteins in AIB (CLA-1 and SYD-2) and measured their intensities over time in AIB in embryos (Figure 7, Figure 7-figure supplement 1A-F). We find that similar to RAB-3, synaptic signal is not visible until after the end of zippering, and progressively increases over time following zippering. These observations suggest that synapses do not initiate retrograde zippering. We added the time-course of active zone protein localization in AIB in the context of the time course of retrograde zippering (Figure 7J). Consistent with these observations, in a syd-2(ola341) allele identified in our screen, we find that although synapses are mislocalized, AIB neurite placement is unaffected, consistent with the idea that synapse formation is not upstream of zippering-mediated placement (Figure 7-figure supplement 1G-K).

We acknowledge, however, that our studies are limited by detection of the synaptic proteins, and that, while it does not appear that synaptogenesis leads to zippering, a cooperative and synergistic relationship might exist between the process of zippering and synaptogenesis to hold the neurite position in place. We have added text to better discuss this relationship between zippering and synaptogenesis in light of these findings.

Minor points

1. The schematic diagram is somewhat misleading because in the axial view, the anterior and posterior segment of the nerve ring should appear on top of each other. The lateral view is the right view to show the anterior and posterior segments.

This is a valid point - if we look at the worm directly head-on, the two segments of the neurite would be on top of one another. A slight tilt of the worm head enables visualization of the parts of the neurite in the two neighborhoods. We have now clarified this in the figure legends.

1. Describe the screen that led to the mutant alleles of syg-1 and syg-2 better. Any other mutants?

We have described the screen further in Methods and now include another allele, corresponding to syd-2 (Figure 7-figure supplement 1), isolated from the screen.

1. "Consistent with the importance of adhesion-based mechanisms in the observed phenotypes, ectopic expression of the SYG-1 endodomain in the posterior neighborhood did not result in mislocalization of AIB (Figure 6-figure supplement 1A,B). " This statement is wrong. I suspect the authors meant in syg-2 mutants.

The statement might have been confusing, but it is not wrong. We have found that ectopic expression of the SYG-1 endodomain, which lacks SYG-1’s extracellular domains, does not cause ectopic placement of the AIB neurite, which is what we described in that statement. We have edited to make it more clear.

1. For Fig. 7-figure supplement 1, please quantify this phenotype.

We have now included quantifications for this in Figure 7-figure supplement 3.

References

MCDONALD, N. A., FETTER, R. D. & SHEN, K. 2021. Author Correction: Assembly of synaptic active zones requires phase separation of scaffold molecules. Nature, 595, E35. SCHWARTZ, M. L. & JORGENSEN, E. M. 2016. SapTrap, a Toolkit for High-Throughput CRISPR/Cas9 Gene Modification in Caenorhabditis elegans. Genetics, 202, 1277-88.

Author Response:

Reviewer #1 (Public Review):

Summary

Moncunill et al set out to investigate a very important question: why are half of children vaccinated three times with RTS,S AS01 protected from clinical malaria - and half not? To do so they isolated PBMCs before vaccination and one month after third vaccination and stimulated them in vitro with DMSO (vehicle control), two malaria antigens (CSP (part of RTS,S) & AMA1) or HBS (hepatitis B antigen - part of RTS,S).

They then assessed their transcriptional response by blood transcriptional module analysis and correlated those results with previous published data on antibody titers and T cell cytokine production to find associations. To assess risk of clinical malaria, responses were compared between RTS,S vaccinated children who developed clinical malaria in the one year follow-up (cases) and those who received RTS,S or a comparator vaccine and did not (controls). They found that responses after RTS,S vaccination did not predict protection from clinical malaria. Instead a blood transcriptional module signature related to dendritic cells, inflammation, and monocytes before vaccination may be associated with clinical malaria risk.

Strengths

Immune correlates of protection are evaluated in African children (who are the RTS,S target population) in a natural transmission setting.

Excellent set of controls: children (same age) vaccinated with RTS,S or comparator vaccine alongside each other -> retrospectively stratified by whether the did or did not develop clinical malaria : controls for the effect of a developing immune system and would allow to disentangle RTS,S specific and clinical malaria specific response patterns.

Weaknesses

RTS,S is composed of CSP & HBS. yet when PBMCs from children vaccinated three time with RTS,S are stimulated with these peptides no transcriptional differences compared to children receiving a rabies or meningitis vaccine were detected (Figure 2). this lack of recall response impacts all downstream conclusions and comparisons made in the paper.

The fact that bulk transcriptional profiling of Ag-stimulated PBMCs (and specifically to CSP) did not identify large significant differences in BTM expression between the RTS,S vs. comparator group could be due to several factors. First of all, the frequency of antigen-specific CD4+ T cells was very low among CD4+ T cells (Figure 4 of the manuscript shows that CSP-specific CD4+ T cells comprise < 0.004% of all CD4+ T cells). This low frequency of CSP-specific T cells is consistent with other RTS,S studies [e.g. as we state on line 330, we have previously found that CSP-specific T-cells in RTS,S/AS01 vaccines comprise < 0.10% of all CD4+ T cells (1)]. Moreover, CD4+ T cells themselves comprise approximately 45-57% of all PBMCs (2). Thus, finding an expression signal between the RTS,S vs. comparator group would require the signal to be high enough to be detected in only 0.002% of all PBMCs [0.004% (% CSP-specific CD4+ T cells out of total CD4+ T cells) x 51% (average % of CD4+ T cells out of all PBMCs) = 0.002%]. Thus, lack of detectable recall response does not mean lack of recall response. Moreover, as suggested below, we opted not to focus the rest of the manuscript on the Ag-stimulation results.

Second of all, the PBMCs were stimulated on site for 12h and then cryopreserved. This stimulation time was chosen based on the kinetics of IFN- and IL-2 mRNA response (3), but other responses may have had different kinetics and thus have already resolved or have not yet occurred by the 12-h cryopreservation. We have added text in the manuscript to discuss these caveats (“Another potential reason for why no BTMs were found to associate with the response to RTS,S/AS01 vaccination or with protection when analyzing CSP-stimulated PBMC is that all PBMC were stimulated on site for 12 hours (this stimulation time was chosen based on the kinetics of the IFN-gamma transcriptional response) and then cryopreserved. Thus, we were unable to detect earlier transient responses that had already resolved by 12 hours, as well as more delayed response that had not yet initiated by 12 hours, if such responses occurred”).

It should be noted that in all our analyses, the stimulated results were adjusted for DMSO to focus on the antigen-specific response only. This would explain why we detect signal in the DMSO samples but not in response to stimulation. We have realized that this was not very well described in the figure captions and the Methods section and have added more details, including the model description in Methods section. As such, we do not believe that these results impact all downstream conclusions. We believe that the unstimulated results provide significant new insights into the immune and molecular mechanisms of RTS,S vaccine efficacy, not necessarily directly related to the RTS,S-specific acquired immune response. Finally, we would like to highlight the fact that we have improved our model specification to directly account for the pairing of some of the samples using a random effect using the limma package. This has slightly increased statistical power, and as such the number of significantly differentially expressed BTMs in response to stimulation is a bit higher (but still much less than that for the DMSO). Originally, we had decided against the use of a random effect due to the computational cost of estimating the random effect.

Transcriptional responses 1 month after the final RTS,S vaccination do not predict clinical malaria risk (Figure 3) - this is a key finding, which should be central to the conclusion of this paper.

Considering that Kazmin et al. (4) showed that the transcriptional response to the third RTS,S/AS01 dose peaks at Day 1 post-injection, with some decline by Day 6 and approximately 90% of the response having waned by Day 21 (with the caveat that Kazmin et al.’s study population was malaria-naïve adults), we do not find it surprising that there were only a few BTMs whose 1 month post-final RTS,S dose associated with clinical malaria risk. However, the point is well-taken about the relative merits of the baseline. We have edited the Discussion to include discussion of the Month 3 correlates results:

“Compared to the 45 BTMs whose baseline levels significantly associated with clinical malaria risk in RTS,S/AS01-vaccinated African children, fewer BTMs (seven) had levels at one month post-final RTS,S/AS01 dose that significantly associated with clinical malaria risk. Moreover, if a more stringent FDR cutoff had been used (i.e. 5%), six of these seven BTMs would not have been identified. Thus it is entirely possible that, at one month post-final RTS,S/AS01 dose, there is no circulating immune transcriptomic signature predictive of risk. Such a conclusion would not be surprising, given that in malaria-naïve adults, the transcriptional response to the third RTS,S/AS01 dose has been shown to peak at Day 1 post-injection, with some decline by Day 6 and approximately 90% of the response having waned by Day 21 (17). Therefore, it is likely that the sampling scheme in this study (one month post-final dose) misses the majority of the transcriptional response to RTS,S/AS01.”

The take-home message put forward in the title/abstract (that a monocyte and DC related pre-vaccination signature predicts risk of clinical malaria in RTS,S vaccinated children) is not strongly supported by the data. It is based on blood transcriptional modules related to monocytes being picked out when comparing RTS,S vaccinated cases and controls.

Thank you for giving us the opportunity to provide further rationale for our focus on the 7 monocyte-related and 4 DC-related BTMs shown in Figure 6B (MAL067 column) out of the 45 total BTMs whose baseline expression associated with clinical malaria risk in RTS,S/AS01-vaccinated children. The reviewer implies that these modules were chosen for focus somewhat randomly or without justification (or, even worse, “cherry picked”), which we would agree would be an imperfect method for drawing conclusions.

First, we have always ensured to mention that the 45 baseline modules that correlated with risk in RTS,S recipients (Fig 6B, MAL067 column) belonged to many functional annotations, including DC cells and monocytes. (Abstract: “In contrast, baseline levels of BTMs associated with dendritic cells and with monocytes (among others) correlated with malaria risk”) (Main text, lines 519-522: “Compared to the results from the month 3 analysis (7 BTMs), the baseline correlates analysis of MAL067 revealed a larger number (45) of BTMs, spanning many functional categories, whose month 0 levels in vehicle-stimulated PBMC nearly all associated with clinical malaria risk in RTS,S/AS01 recipients .”

The focus on DC cells and monocytes is due to two reasons: 1) the fact that the DC-related modules and the monocyte-related modules were some of the most significant correlations (lines 522-524: “The BTM with the most significant association with risk was “enriched in monocytes (II) (M11.0)” (FDR = 1.80E-14), followed by “inflammatory response (M33)” (FDR = 2.45E-07) and “resting dendritic cell surface signature (S10)” (FDR = 6.03E-07).”

Second, the baseline association of DC- and monocyte-related modules appeared to generalize across populations: (Abstract: “A cross-study analysis supported generalizability of the baseline dendritic cell- and monocyte-related BTM correlations with malaria risk to healthy, malaria-naïve adults, suggesting that certain monocyte subsets may inhibit protective RTS,S/AS01-induced responses.”; Main text: “BTMs related to dendritic cells and to monocytes were most consistently associated with risk across these three studies [“resting dendritic cell surface signature (S10)”, “DC surface signature (S5)”, “enriched in dendritic cells (M168)”, “enriched in monocytes (I) (M4.15)”, “enriched in monocytes (II) (M11.0)”, “enriched in monocytes (IV) (M118.0)”, and “monocyte surface signature (S4)” significantly correlated with risk in all three studies].”

The first two sentences of the Discussion (lines 577-580) explain our focus on monocytes and DCs:

“Our main finding is the identification of a baseline blood transcriptional module (BTM) signature that associates with clinical malaria risk in RTS,S/AS01-vaccinated African children. In a cross-study comparison, much of this baseline risk signature – specifically, dendritic cell- and monocyte-related BTMs – was also recapitulated in two of the three CHMI studies in healthy, malaria-naïve adults.”

Finally, we note that the title (“A baseline transcriptional signature associates with clinical malaria risk in RTS,S/AS01-vaccinated African children”) does not restrict to DC-related or monocyte-related BTMs, rather, we chose this title based on the larger number of BTMs, and higher correlations with risk, in the baseline analysis compared to the Month 3 analysis.

We have revised all instances where we have communicated this less clearly, e.g. “for why we identified a baseline monocyte transcriptional signature of risk” has been changed to “for why we identified monocyte-related BTMs in our transcriptional signature of risk”.

Many other modules are picked out as well e.g. cell cycle (Figure 6B). An in-depth analysis of the genes in these module and what their up and downregulation can tell us about their function is warranted to support the conclusions.

Thank you for the suggestion to look at the cell cycle module in Figure 6B. You make a good point that this module is the only module to show a significant association with clinical malaria risk across all 4 of the RTS,S studies and should therefore be further examined. First, we have added this to the text:

“Only one BTM, “cell cycle and transcription (M4.0)”, was significantly associated with risk across all four studies. Of the 335 genes in this module (M4.0), 130 were also present in one or more of the six “monocyte-related” BTMs shown in Figure 6B (297 genes total across all six BTMs), suggesting that the “cell cycle” and “monocyte” results may actually be picking up the same signal.”

We have done the gene-level analysis as suggested, resulting in 8 new supplemental figures (Figure 6-figure supplements 1-8) and one new supplemental table (S5). We have also made the following revisions to the text:

In Results: “To gain insight into specific module-member genes that may be involved in the RTS,S/AS01 baseline risk signature, we performed the same analysis on the gene level, i.e. examined associations with clinical malaria risk for each of the constituent genes in the 45 BTMs shown in Figure 6B. Figure 6-figure supplements 1-8 show the gene-level association results within the eight BTMs that were significantly associated with clinical malaria risk in MAL067 and at least two of the three CHMI studies, and had at least one gene in MAL067 that was significantly associated with risk (these eight correspond to M4.0, S10, S5, M168, M4.3, M11.0, M4.15, and S4). Within MAL067, 35 unique genes were shown to significantly associate with malaria risk (Supplementary Table 5); 9 of these genes (CCNF, MK167, KIF18A, NPL, RBM47, CFD, MAFB, IL13RA1, and CCR1) also had significant association with non-protection in one of the CHMI studies. Although no individual gene was significantly associated with risk across >2 studies, many showed consistent effect (direction and magnitude) across 3 studies. This further supports our choice to focus on modules instead of individual genes as GSEA increases power to detect more subtle but coordinated changes in gene expression data that would be missed otherwise. For this same reason, GSEA has been shown to enhance cross-study comparisons (45).”

In Discussion: “Our gene-level correlates analyses suggest an alternative hypothesis, however. With the caveat that the gene-level analyses were performed post hoc, high baseline expression of STAB1 (which is present in DC-related, monocyte-related, and cell cycle-related modules) was found to positively associate with clinical malaria risk (Figure 6-figure supplements 1, 2, and 6). STAB1 encodes stabilin-1 (also called Clever-1), a transmembrane glycoprotein scavenger receptor that links extracellular signals to intracellular vesicle trafficking pathways (58). Interestingly, stabilin-1high monocytes show downregulation of proinflammatory genes, and T cells co-cultured with stabilin-1high monocytes showed decreased antigen recall, suggesting that monocyte stabilin-1 suppresses T cell activation (56). Thus one possibility is that stabilin-1high immunosuppressive monocytes circulating at baseline could decrease protective RTS,S-induced T-cell responses, or inhibit another aspect of adaptive immunity. Single-cell transcriptomic profiling of PBMC or purified monocyte subsets in future RTS,S trials in African children in malaria-endemic areas could help test this hypothesis.”

Impact

This paper will inform future studies looking for correlates of RTS,S induced protection from clinical malaria in a variety of ways:

It validates the blood transcriptional module approach (as published by Li S, Rouphael N, Duraisingham S, Romero-Steiner S, Presnell S, Davis C, Schmidt DS, Johnson SE, Milton A, Rajam G, et al: Molecular signatures of antibody responses derived from a systems biology study of five human vaccines. Nat Immunol 2014) to find target cell populations which can then be investigated in much more detail.

It shows that studying PBMC recall responses after peptide stimulation post final vaccination is not the way forward, since no response is detected (Figure 2). future studies can now take an alternative approach. e.g. since unstimulated PBMCs (vehicle control) from RTS,S vaccinated children were different from those who received a comparator vaccine (Figure 2) RTS,S vaccine signatures could be picked up much more easily by whole blood RNAseq.

It implicates innate immune cells in shaping an individuals response to a vaccine - an exciting basis for future functional and mechanistic studies.

We are glad the reviewer appreciates the value of the study.

Reviewer #2 (Public Review):

This paper reports a sub-study of the RTS,S/AS01 malaria vaccine Phase 3 trial, which aimed to identify groups of genes (blood transcriptional modules, BTMs) for which expression in DMSO or antigen-stimulated PBMCs was associated with clinical malaria during a 12-month follow-up period. Study subjects were infants and children who received either RTS,S/AS01 or comparator vaccines (meningococcal C for infants, rabies vaccine for children), enrolled in the study in Tanzania and Mozambique (with some additional analyses using samples from Gabonese infants).

Using PBMCs collected at baseline before vaccination and 3 months later (a month after the third vaccine dose), stimulated with DMSO or parasite antigens, the authors used RNA-sequencing to identify BTMs which were different between recipients of RTS,S/AS01 vs comparator vaccines; which were different between baseline and month 3 in RTS,S/AS01 recipients; and which differed between RTS,S/AS01 recipients with a malaria episode and those without a malaria episode during the follow-up period. This combination of analyses might help to distinguish BTMs specifically associated with RTS,S/AS01 vaccine efficacy from those associated with other factors influencing susceptibility to malaria. To further aid mechanistic understanding the authors examined correlations between BTMs and measures of cellular and humoral immune responses. To try to establish generalisability the authors examined whether BTMs identified in African children were also associated with developing malaria in RTS,S/AS01-vaccinated malaria-naïve adults in the United States who underwent controlled human malaria infection (CHMI).

Strengths of the study include:

1) The relatively large number of subjects, the large amount of transcriptomic and immunological data which has been generated (and made publicly accessible), and the extensive analysis to evaluate associations between BTMs and numerous immunological variables.

2) Clear explanation of both the rationale and methods for most of the analyses

3) The attempt to validate findings in the CHMI studies

4) Matching of subjects to try to eliminate the confounding effects of age, study site, and time of vaccination

Weaknesses of the study include:

1) Despite the relatively large size of the study, it is hard to know whether it had sufficient power to achieve its main objective, and we are not presented with data to demonstrate how successfully the authors managed to match subjects for age, timing of vaccination and follow-up duration

We have added the following to our “limitations” paragraph in the Discussion: “Fourth, despite the relatively large size of the study, our statistical power was limited by the number of malaria cases with available samples; sampling additional controls would not have increased our statistical power.”

Moreover, we now also provide the new Supplementary Table 1, which provides complete information on participant match ID, site, age cohort, sex assigned at birth, and time of vaccination.

2) The comparator group to the RTS,S/AS01 vaccine is not a single vaccine, but two vaccines, but the presentation of the data makes it difficult to identify what effect this may have had on the results

Indeed, comparators received a rabies vaccine or the meningococcal C conjugate depending on the age cohort. However, we think that the impact on the study results and conclusions is minimal since the main results are based on baseline gene expression and its association with malaria risk within RTS,S vaccinees. Correlates of malaria risk in comparators are done separately. Comparator vaccination may be a confounding factor for age cohort, but we are not analyzing the effect of age cohort on the transcriptional profile. Comparators are only included in the analysis of RTS,S immunogenicity at post-vaccination (RTS,S vs Comparators, Fig 2A, Comparison (1)) and we have adjusted analyses by age cohort and hence by comparator vaccine. The fact that the comparators received different control vaccines only stresses that the BTMs found to be associated with RTS,S vaccination are specific to the RTS,S vaccine.

Moreover, as an alternative way to identify RTS,S-specific transcriptional responses, we also include Comparison (2), which compares Month 3 to Month 0 transcription levels within RTS,S vaccinees. We include in the text extensive discussion of the merits and drawbacks of each comparison:

“Two comparisons were done to characterize the transcriptional response to RTS,S/AS01 vaccination: Comparison (1): comparing gene expression in month 3 samples from RTS,S/AS01 vs comparator recipients (month 3 RTS,S/AS01 vs comparator); and Comparison (2): comparing gene expression in month 3 vs month 0 from RTS,S/AS01 recipients (RTS,S/AS01 month 3 vs month 0). Each comparison has its own advantages: Comparison (1) allows the identification of RTS,S/AS01-specific responses while taking into account other environmental factors to which the children are exposed, such as malaria exposure (albeit malaria transmission intensity was low during the study at both sites). Moreover, the very young ages of the trial participants mean that RTS,S/AS01-induced changes may be confounded with normal developmental changes in participant immune systems, further underscoring the value of Comparison (1), as it does not involve comparison across two different time points. On the other side, an advantage of Comparison (2) is that it takes into consideration each participant’s intrinsic baseline gene expression. Comparison (1) uses data from both infants and children, whereas Comparison (2) can only yield insight into RTS,S/AS01 responses in children (as baseline samples were not collected from infants).”

3) A very "liberal" false-discovery rate (FDR) threshold has been used throughout to define significant associations. An FDR of 0.2 indicates that 20% (or 1 in 5) results which are considered significant will be false-discoveries. This means that the "significant" results must be interpreted with a high degree of caution. Typically researchers use lower FDR thresholds, like 0.05 or 0.01, although one may argue for different thresholds under different circumstances

While it is not uncommon to use a threshold of 20% for immune correlates studies [e.g. (5-10)], we agree with you that it is important to clearly state the chosen FDR rate and to discuss conclusions in the context of the FDR rate used. We see we could improve our manuscript in this respect. We have added the following:

Results: “Compared to the 45 BTMs whose baseline levels significantly associated with clinical malaria risk in RTS,S/AS01-vaccinated African children, fewer BTMs (seven) had levels at one month post-final RTS,S/AS01 dose that significantly associated with clinical malaria risk. Moreover, if a more stringent FDR cutoff had been used (i.e. 5%), six of these seven BTMs would not have been identified. Thus it is entirely possible that, at one month post-final RTS,S/AS01 dose, there is no circulating immune transcriptomic signature predictive of risk…”

Discussion: “Finally, while it is not uncommon to use an FDR cutoff of 20% in high-dimensional immune correlates studies [e.g. (65-70)], our results should be interpreted with the requisite level of caution. However, we do note that many of our significant modules in the baseline risk analysis would have survived even lower FDR cutoffs (in many cases even a 1% cutoff), giving us a fair degree of confidence in our results. For example, of the seven monocyte-related BTMs whose baseline levels associated with risk, all would have survived a 5% FDR cut-off, and three even a 1% cut-off; likewise, of the four dendritic cell-related BTMs whose baseline levels associated with risk, all would have survived a 5% FDR cut-off, and three even a 1% cut-off.”

Moreover, we have revised Figures 2, 3, and 6 so that it is easy to discern whether a specific BTM correlation would also pass more stringent FDR cutoffs, through the addition of 1, 2, or 3 asterisks where appropriate: “|FDR| < 0.2 (), < 0.05 (), < 0.01 ().” Note that, most central to the key message of the paper, many of the monocyte-related, DC-related, and cell cycle-related BTMs would have passed more stringent FDR cutoffs, with many even passing a 1% FDR cutoff (as discussed above).

4) A perplexing finding, which is not addressed in detail, is the large number of BTMs which differ between RTS,S and comparator vaccine groups after DMSO stimulation of PBMCs, but these are not seen when PBMCs are stimulated with parasite antigens in DMSO (and a similar finding for month 3 vs month 0 samples from RTS,S recipients). This raises some concern about the stimulation experiments, because one might expect that the DMSO vehicle in the antigen preparations would trigger a similar response to DMSO alone.

It should be noted that in all our analyses, the stimulated results were adjusted for DMSO to focus on the antigen-specific response only. This would explain why we detect signal in the DMSO samples but not in response to stimulation. We have realized that this was not very well described in the figure captions and the Methods section and have added more details, including the model description in Methods section. As such, we do not believe that these results impact all downstream conclusions. We believe that the unstimulated results provide significant new insights into the immune and molecular mechanisms of RTS,S vaccine efficacy, not necessarily directly related to the RTS,S-specific acquired immune response. We would also like to highlight the fact that we have improved our model specification to directly account for the pairing of some of the samples using a random effect using the limma package. This has slightly increased statistical power, and as such the number of significantly differentially expressed BTMs in response to stimulation is a bit higher (but still much less than that for the DMSO). Originally, we had decided against the use of a random effect due to the computational cost of estimating the random effect.

The fact that bulk transcriptional profiling of Ag-stimulated PBMCs did not identify almost any significant differences in BTM expression between the RTS,S vs. comparator group could be due to several factors. First of all, the frequency of antigen-specific CD4+ T cells was very low among CD4+ T cells (Figure 4 of the manuscript shows that CSP-specific CD4+ T cells comprise < 0.004% of all CD4+ T cells). This low frequency of CSP-specific T cells is consistent with other RTS,S studies [e.g. as we state on line 330, we have previously found that CSP-specific T-cells in RTS,S/AS01 vaccinees comprise < 0.10% of all CD4+ T cells (1)].

Moreover, CD4+ T cells themselves comprise approximately 45-57% of all PBMCs (2). Thus, finding an expression signal between the RTS,S vs. comparator group would require the signal to be high enough to be detected in only 0.002% of all PBMCs [0.004% (% CSP-specific CD4+ T cells out of total CD4+ T cells) x 51% (average % of CD4+ T cells out of all PBMCs) = 0.002%]. Thus, lack of detectable recall response does not mean lack of recall response. Moreover, as suggested below, we opted not to focus the rest of the manuscript on the Ag-stimulation results.

Second of all, the PBMCs were stimulated on site for 12h and then cryopreserved. This stimulation time was chosen based on the kinetics of IFN-g and IL-2 mRNA response (3), but other responses may have had different kinetics and thus have already resolved or have not yet occurred by the 12-h cryopreservation. We have added text in the manuscript to discuss these caveats (“Another potential reason for why no BTMs were found to associate with the response to RTS,S/AS01 vaccination or with protection when analyzing CSP-stimulated PBMC is that all PBMC were stimulated on site for 12 hours (this stimulation time was chosen based on the kinetics of the IFN-g transcriptional response) and then cryopreserved. Thus, we were unable to detect earlier transient responses that had already resolved by 12 hours, as well as more delayed response that had not yet initiated by 12 hours, if such responses occurred.”.

The authors partly achieved their aims. They identified BTMs differentially expressed between RTS,S/AS01 and the comparator vaccines, and between baseline and month 3 in RTS,S/AS01 recipients. They also identified BTMs at month 3 associated with developing malaria, and BTMs at baseline associated with developing malaria. These latter BTMs were partly replicated in the CHMI study subjects. Higher expression of BTMs associated with monocytes and dendritic cells were most consistently identified across the different analyses and their expression in stimulated baseline samples was most consistently associated with development of clinical malaria in RTS,S/AS01 recipients. However there were inconsistencies in associations between some of the studies, and it is possible that the "consistent" monocyte and dendritic cell BTMs would not be so consistent if a more stringent FDR threshold was used. However the authors conclusions are largely quite measured and for the most part they do not over-interpret the significance of their findings.

We have added the following to the Discussion: “Finally, while it is not uncommon to use an FDR cutoff of 20% in high-dimensional immune correlates studies [e.g. (65-70)], our results should be interpreted with the requisite level of caution. However, we do note that many of our significant modules in the baseline risk analysis would have survived even lower FDR cutoffs (in many cases even a 1% cutoff), giving us a fair degree of confidence in our results. For example, of the seven monocyte-related BTMs whose baseline levels associated with risk, all would have survived a 5% FDR cut-off, and three even a 1% cut-off; likewise, of the four dendritic cell-related BTMs whose baseline levels associated with risk, all would have survived a 5% FDR cut-off, and three even a 1% cut-off.”

Overall the work provides some evidence that baseline immunological status, particularly related to monocyte and dendritic cell responses and possibly their role in or response to baseline inflammation, may be a determinant of how well the RTS,S vaccine works to prevent malaria. This provides a basis for further work to optimise the effectiveness of the vaccine. The usefulness of PBMC stimulation to predict an individual's response to vaccination will be limited because this is not a method which can be used at scale in resource limited settings, but the concept that vaccine response could be enhanced by modifying pre-vaccine immunological or inflammatory status is potentially important. The data published with this study will be a valuable resource and will undoubtedly be used by others to address similar questions. Increasing the efficacy of malaria vaccines remains an extremely important goal, and identifying possible mechanisms which restrict the efficacy of RTS,S is important.

References:

1. Moncunill G, De Rosa SC, Ayestaran A, Nhabomba AJ, Mpina M, Cohen KW, Jairoce C, Rutishauser T, Campo JJ, Harezlak J, Sanz H, Diez-Padrisa N, Williams NA, Morris D, Aponte JJ, Valim C, Daubenberger C, Dobano C, McElrath MJ. RTS,S/AS01E Malaria Vaccine Induces Memory and Polyfunctional T Cell Responses in a Pediatric African Phase III Trial. Front Immunol. 2017;8:1008.
2. Kleiveland CR. Peripheral Blood Mononuclear Cells. In: Verhoeckx K, Cotter P, López-Expósito I, Kleiveland C, Lea T, Mackie A, et al., editors. The Impact of Food Bioactives on Health: in vitro and ex vivo models. Cham: Springer International Publishing; 2015. p. 161-7.
3. Schultz-Thater E, Frey DM, Margelli D, Raafat N, Feder-Mengus C, Spagnoli GC, Zajac P. Whole blood assessment of antigen specific cellular immune response by real time quantitative PCR: a versatile monitoring and discovery tool. J Transl Med. 2008;6:58.
4. Kazmin D, Nakaya HI, Lee EK, Johnson MJ, van der Most R, van den Berg RA, Ballou WR, Jongert E, Wille-Reece U, Ockenhouse C, Aderem A, Zak DE, Sadoff J, Hendriks J, Wrammert J, Ahmed R, Pulendran B. Systems analysis of protective immune responses to RTS,S malaria vaccination in humans. Proc Natl Acad Sci U S A. 2017;114(9):2425-30.
5. Liu C, Martins AJ, Lau WW, Rachmaninoff N, Chen J, Imberti L, Mostaghimi D, Fink DL, Burbelo PD, Dobbs K, Delmonte OM, Bansal N, Failla L, Sottini A, Quiros-Roldan E, Han KL, Sellers BA, Cheung F, Sparks R, Chun TW, Moir S, Lionakis MS, Consortium NC, Clinicians C, Rossi C, Su HC, Kuhns DB, Cohen JI, Notarangelo LD, Tsang JS. Time-resolved systems immunology reveals a late juncture linked to fatal COVID-19. Cell. 2021;184(7):1836-57 e22.
6. Andersen-Nissen E, Fiore-Gartland A, Ballweber Fleming L, Carpp LN, Naidoo AF, Harper MS, Voillet V, Grunenberg N, Laher F, Innes C, Bekker LG, Kublin JG, Huang Y, Ferrari G, Tomaras GD, Gray G, Gilbert PB, McElrath MJ. Innate immune signatures to a partially-efficacious HIV vaccine predict correlates of HIV-1 infection risk. PLoS Pathog. 2021;17(3):e1009363.
7. Lu P, Guerin DJ, Lin S, Chaudhury S, Ackerman ME, Bolton DL, Wallqvist A. Immunoprofiling Correlates of Protection Against SHIV Infection in Adjuvanted HIV-1 Pox-Protein Vaccinated Rhesus Macaques. Front Immunol. 2021;12:625030.
8. Haynes BF, Gilbert PB, McElrath MJ, Zolla-Pazner S, Tomaras GD, Alam SM, Evans DT, Montefiori DC, Karnasuta C, Sutthent R, Liao HX, DeVico AL, Lewis GK, Williams C, Pinter A, Fong Y, Janes H, DeCamp A, Huang Y, Rao M, Billings E, Karasavvas N, Robb ML, Ngauy V, de Souza MS, Paris R, Ferrari G, Bailer RT, Soderberg KA, Andrews C, Berman PW, Frahm N, De Rosa SC, Alpert MD, Yates NL, Shen X, Koup RA, Pitisuttithum P, Kaewkungwal J, Nitayaphan S, Rerks-Ngarm S, Michael NL, Kim JH. Immune-correlates analysis of an HIV-1 vaccine efficacy trial. N Engl J Med. 2012;366(14):1275-86.
9. Fletcher HA, Snowden MA, Landry B, Rida W, Satti I, Harris SA, Matsumiya M, Tanner R, O'Shea MK, Dheenadhayalan V, Bogardus L, Stockdale L, Marsay L, Chomka A, Harrington-Kandt R, Manjaly-Thomas ZR, Naranbhai V, Stylianou E, Darboe F, Penn-Nicholson A, Nemes E, Hatherill M, Hussey G, Mahomed H, Tameris M, McClain JB, Evans TG, Hanekom WA, Scriba TJ, McShane H. T-cell activation is an immune correlate of risk in BCG vaccinated infants. Nat Commun. 2016;7:11290.
10. Young WC, Carpp LN, Chaudhury S, Regules JA, Bergmann-Leitner ES, Ockenhouse C, Wille-Reece U, deCamp AC, Hughes E, Mahoney C, Pallikkuth S, Pahwa S, Dennison SM, Mudrak SV, Alam SM, Seaton KE, Spreng RL, Fallon J, Michell A, Ulloa-Montoya F, Coccia M, Jongert E, Alter G, Tomaras GD, Gottardo R. Comprehensive Data Integration Approach to Assess Immune Responses and Correlates of RTS,S/AS01-Mediated Protection From Malaria Infection in Controlled Human Malaria Infection Trials. Front Big Data. 2021;4:672460.

Author Response:

Reviewer #1:

In this study, the authors seek to understand the target and mechanism of action of two structurally related orally available antibiotic drug candidates active against Neisseria gonorrhoeae (Ng). The experimental approach involves a detailed investigation of drug efficacy in bacterial culture experiments and a mouse model for gonorrhea infections, along with biochemical experiments to identify the drug target. The latter experiments include discovery of resistance-inducing mutations in class Ia ribonucleotide reductase (RNR), in vitro validation of the ability of the Ng inhibitors to diminish enzyme activity, and structural studies to evaluate the effects of the compounds on Ng RNR structure. The work succeeds in providing convincing evidence for inhibition of the RNR but it does not fully explain how the drug candidates bind to the enzyme. Although the findings represent an important advance that could motivate other work in exploiting bacterial RNRs for antibiotic drug development, conclusions about the mechanism of action could be better supported by more thorough understanding of inhibitor-enzyme interaction. This insight would be important for improving drug design and broad expansion of the approach to other pathogens. Additionally, the inhibitors are billed as narrow-spectrum antibiotic candidates, but these claims are based on analysis of a small and specialized group of bacteria that are not likely to contain or exclusively rely on a class Ia RNR. It is not clear from this study if the inhibitors could affect growth of commensal organisms that contain aerobic RNRs.

Table S7 has been included to highlight the known forms of RNRs present in various organisms. The selective inhibition of Neisseria gonorrhoeae and N. meningitidis is corroborated by the presence of only a single RNR, type Ia. Other aerobic Gram-negative organisms, including E. coli and K. pneumoniae, have type Ia, Ib and III, and are not inhibited by the PTC compounds. The presence or multiple forms of RNR types may be the explanation for other organisms lacking sensitivity to the PTC compounds, but the data are not conclusive. Obtaining conclusive data will require additional experimentation with engineered knock-out isogenic strains to fully elucidate why other organisms are not inhibited.

Reviewer #2:

This study by Narasimhan et al. describes the identification of ribonucleotide reductase (RNR), a critical enzyme in all organisms, as a new target for treatment of antibiotic-resistant gonorrhea, via a novel mechanism for RNR inhibition. The authors begin with the identification of two inhibitors that selectively target Neisseria gonorrhoeae, including multidrug resistant strains, over other pathogens and microbiota. They then show that these inhibitors target the synthesis of DNA, but not by the mechanism of other members of this class of compounds; instead, isolation of resistant mutants indicates that the class Ia ribonucleotide reductase of this organism is the target of the molecules. These results are supported by in vitro activity assays of the RNR, along with electron microscopy characterization of the RNR, showing that the resistance mutations disrupt the protein's ability to form a ring-like inhibited state, implying that the mechanism of action of the compounds involves that ring-like state. While other compounds have been shown to induce formation of the inhibited state of the human RNR, the mechanism of inhibition of the gonorrheal RNR evidenced here is distinct. Finally, the authors present data from a mouse infection model showing efficacy of the compounds. The comprehensive nature of this study, from small molecule to in vitro analysis to in vivo efficacy, is compelling, and the results are of interest both from an enzymological perspective and from the development of new strategies to combat important pathogens. There are two issues that I believe the authors should address to support two important aspects of their work, the mechanisms of inhibition and resistance:

1) The major question that sticks in my mind after reading this manuscript is the mechanism by which the molecules inhibit the RNR. They act as potent inhibitors in vitro, and the identification of the resistance mutations, H25R and S41L, which interfere with the ability of the RNR to form the inactive a4b4 form, are strong pieces of evidence in favor of the authors' proposal that the inhibitors "potentiate conversion of its active a2b2 state to an inactive a4b4." The clincher of this argument would be EM evidence that the presence of the inhibitors leads to a4b4 formation, just as the H25R and S41L variants do, or (perhaps simpler) size exclusion chromatography or direct evidence from another analytical method pointing to formation of the a4b4 species.

Relatedly, it would also be helpful if the dependency of inhibition on dATP would be clarified. Figure S6 suggests that dATP is not required for this state, but on p. 14, line 12, the authors write "the dependency on a dATP-induced inactive a4b4 state also explains why the Ng and Ec Ia RNRs are both sensitive to these inhibitors…" and on p. 20, line 27, it is also implied that dATP could be involved. Please clarify this point.

We have carried out additional experiments to probe the mechanism of inhibition of PTC compounds, and our results suggest that there is likely to be more than one mode of enzyme inhibition. One mode of inhibition does appear to be related to dATP-inhibition and thus α4β4 ring formation. In particular, we now present data that show that PTC-672 and PTC-846 potentiate the inhibitory effects of dATP (Figure 4). However, the PTC compounds themselves are not dATP mimics; they do not appear to be able to substantially increase the amount of α4β4 ring formation in vitro. This finding is based on new mass photometry data that we now present (Supplemental data Figure S8, S9). We also now present data that show that PTC-672 can inhibit (to some degree) variants of Ng RNR that can’t form rings (H25L and S41L) (Figure S7), providing evidence of a second mode of inhibition. Collectively, these results indicate that the mode(s) of inhibition of PTC compounds are complex. We don’t know why E. coli is not less effected than Ng by these compounds. It could be that the class Ia RNR in E. coli is equally inhibited but that the presence of multiple RNRs in E. coli is protective. It is possible that the synergistic effect of the PTC compounds with dATP is less dramatic for the E. coli enzyme or that the second (unknown) mode of inhibition is not in play in E. coli. Much more work will need to be done to answer this question and that work is out of the scope of this paper. We are pleased, however, to present several new pieces of data, including the mass photometry results, to fill out this story.

2) The observation that the resistance mutant strains have lower fitness is an interesting and important one. I suggest that the authors determine whether this decreased fitness might be the result of the mutations in the RNR leading to lower activity - the authors should give the activities of the H25R and S41L with the normal substrate, vs. wild-type alpha. If the activities are similar to wild-type, perhaps the authors could suggest another potential explanation. One that seems possible to me is that the loss of dATP inhibition (see Fig S5) might lead to loss of fitness via misregulation of (deoxy)nucleotide pools.

H25R and S41L Ng RNR variants are impaired in their ability to be down-regulated by dATP but are not otherwise impaired. Our hypothesis is as the reviewer suggests that loss of fitness is due to the inability to down-regulate RNR, which misregulates nucleotide pools. This phenomenon was observed for E. coli RNR class Ia. In the case of E. coli class Ia RNR, S39F and E42K variants were shown only to be impaired in their ability to be down-regulated by dATP, and yet, these mutations were linked to a mutator phenotype (see Chen 2018 JBC 293, 10404 and Ahluwalia 2012 DNA Repair 11, 480). Not being able to turn off an RNR is problematic for the cell.

Author Response:

We thank the reviewers and editor for the feedback on our manuscript. We now included high-magnification images of somite-like structures, which clearly show the shape of the somitic cells and the polarity of these epithelial cells by staining for multiple polarity markers (new Figure 3-Supplemental Fig 1; new Figure 3-Video 1). Consistent with the morphology and polarity of in vivo somites, we observe PAX3+/TCF15+ bottle-shaped cells radially arranged around a central cavity, which form rosette-like structures that are approximately the same size as the Carnegie stage 11 in vivo somites (Fig 5B). This can be observed in multiple new images added to the manuscript (Fig 3-Supplemental Fig 1A-A’’, Fig 3-Supplemental Fig 1B-B’’, Fig 2-Supplemental Fig 3A,B). Additionally, apical surface markers F-ACTIN (Fig 5A) and N-CADHERIN (Fig 3-Supplemental Fig 1) are both expressed around the central cavity, suggesting that the apical side of the somitic cells is facing the inside of the somite structure, again consistent with their in vivo counterparts. This is particularly evident in the newly added high-magnification images (Figure 3-Supplemental Fig 1) and accompanying movie showing a full confocal z-stack through the in vitro somites (Figure 3-Video 1). Together with our protein (Fig 3B,C and Fig 5A) and gene expression data (both in bulk (Fig 1C, Fig 5C, Fig 1-Supplemental Fig 1B) and at the single-cell level (Fig 4, Fig 4-Supplemental Fig 1-5)), and directed differentiation experiments of Somitoid-derived cells towards sclerotome (Fig 5C) and dermomyotome (newly added Figure 5-Supplemental Fig 1), we conclude that our in vitro somites are molecularly, morphologically, and functionally equivalent to in vivo somites.

Reviewer #1:

The foundational differentiation protocol up until day3 (formation of PSM) has been published previously in Diaz-Cuadros et al., 2020; Matsuda et al., 2020. The main difference between this manuscript and published protocol being the 2D (published) vs 3D differentiation. In this manuscript the authors were able to generate Pax3+ Somites (day4-5) from PSM. Both Diaz-Cuadros et al., 2020; Matsuda et al., 2020 were unable to generate Pax3+ somite in their 2D culture system but instead could only obtain a TCF15+ somatic mesoderm intermediate state. Moreover, the somites obtained in this manuscript could be further differentiated to sclerotome.

The experiments across the paper were validated using three repeats using appropriate quantitative microscopy. Imaging data are high quality and mostly presented in a clear manner. However, it is unclear exactly what the authors are scoring as a somite. Moreover, for each figure it is not clear whether technical or biological replicates are presented. Similarly, the heatmap block presented in Fig2C,D, 3C,D apparently represents just one organoid/replicate. Authors should comment on the efficiency of the protocol over the different cell lines used. Transcriptome data presented strongly support the reproducibility and accuracy of the 3D differentiation, although comparison with the in vivo situation is highly limited - in part due to lack of availability of human in vivo data at these developmental stages.

We thank the reviewer for the insightful comments and questions. Following the suggestion of the reviewer, we generated new organoids with somite structures and imaged them at a higher magnification (shown in the newly added Figure 3-Supplement Figure 1). The higher magnification clearly shows the ‘bottle-shaped’ morphology of the cells that comprise the somites in that the apical surface of these columnar cells is typically smaller than their basal side. These higher-magnification images also more clearly show the rosette structure (radial arrangement of columnar cells) with a central cavity in which NCAD is highly expressed. Nuclear expression of PAX3 is also clearly visible in these cells. These empirical observations were the criteria used to manually identify somites in the images acquired from the organoid screen. Additionally, we generated movies of z-stacks acquired on a confocal microscope showing examples of organoids with and without somite-like structures based on our scoring criteria using PAX3 and NCAD staining (Newly added Figure 3-Video 1).

Regarding the number of replicates used and the questions related to variability/efficiency of our optimized protocol, we have made the following changes and performed new experiments to address these questions. To clarify which cell lines were used for each figure, we have added the information to each figure caption in the manuscript. We have also specified the number of organoids used to quantify variation for each experiment in the figure captions. As described below, we have conducted new experiments to further quantify the technical variability across experiments, as well as quantify the efficacy of our optimized differentiation protocol when applied to genetically independent cell lines.

For our initial screens (Figures 2 and 3), we only used one cell line (NCRM1 hiPSCs). As mentioned by the reviewer, we measured variability across individual organoids for each condition of the screen to help identify the conditions that minimized variability and produced the most reproducible organoids (Figure 2-Supplemental Fig 2B; Figure 3-Supplemental Fig 1B). To analyze technical variability of our optimized protocol, we have repeated our optimized protocol two more times and quantified the number of somites across 10 organoids for each experiment (see new Figure 3-Supplemental Fig 3). Inter-organoid variability was comparable to our initial results from the secondary screen (CV for Experiment 1 = 17% and CV for Experiment 2 = 9.8%; see also Fig 3-Supplemental Fig 2B). Furthermore, mean and median are very comparable between the two additional experiments (Experiment 1: 39+/-8 (mean+/-std), median=40; Experiment 2: 43+/-4 (mean+/-std), median=41, p-val = 0.16). Please note that the absolute number of somites in our new experiments has increased compared to our initial screen (see Figure 3). This is a result of improvements in both our immunostaining protocol as well as the image acquisition workflow. The two new experiments both used the same improved staining and image acquisition workflow and are therefore comparable with each other.

To extend our variability analysis to other cell lines, we tested the following cell lines: our original cell line (NCRM1), the WTC cell line released by the Conklin Laboratory at the Gladstone Institute, and a reporter cell line, ACTB-GFP, from the Allen Cell Collection. We tested our optimized protocol alongside another high scoring condition in all three cell lines:

• CL+FGF2 for 24h, basal media for 24h (our optimized protocol)
• WNTihi(C59, 2 µM) for 48h

Applying our optimized protocol to two other cell lines confirmed that our protocol is reproducible across different genetic backgrounds/cell lines. NCRM1, average somite number = 43+/-4 (mean+/-std); ACTB-GFP, average somite number = 40+/-6; WTC, average somite number = 33+/-4. We also tested one additional top scoring condition (C59, 2µM for 48 hours) in all three cell lines. Notably, for this condition, the ACTB-GFP derived Somitoids (32+/-4 (mean+/-std)) showed a higher average number of somites compared with the Somitoids derived from the other cell lines (NCRM1, 20+/-3; WTC, 17+/-4). However, our optimized protocol resulted in the highest number of somites across all cell lines.

We agree with the reviewer that our transcriptome analysis strongly supports the reproducibility and accuracy of our optimized 3D differentiation protocol. We unfortunately were not able to obtain human in vivo data at the relevant developmental stages to make a direct comparison. However, we did extend our analysis to compare our single-cell RNA-seq dataset with the previously published transcriptome data from 2D paraxial mesoderm differentiation protocols (Diaz-Cuadros et al., 2020, Matsuda et al., 2020), which we have included in our updated Discussion section.

We compared the single-cell RNA-seq data from Diaz-Cuadros et al. with our own single-cell RNA-seq data. The 2D differentiated cells in Diaz-Cuadros et al. at the final time point of their experiment do not show a clear somitic cell state signature. PAX3, MEOX1, MEOX2, FOXC2, UNCX, and TBX18 are not expressed (compared for example to FOXC1). FOXC1 does not appear to be a somite-specific marker as it is expressed in early and late PSM-like cells as seen in our own data, starting on day 2 (Figure 4-Supplemental Fig 3). In the Diaz-Cuadros et al. dataset, TCF15 is not expressed uniformly in all cells nor specifically in the late-stage cells. Conversely, TCF15 is specifically expressed in day 5 somitic cells in our dataset both in a uniform manner and at high levels (as shown in the figure below and Figure 4C).

We also analyzed the bulk RNA-seq data in Matsuda et al. 2020 as shown in Figure 1B of their paper. They show the expression of 4 somite markers (TCF15, MEOX1, PAX3, and RIPPLY1). As can be seen in the figure below (and Figure 4-Supplemental Fig 3), all markers highlighted in their paper are specifically expressed in our day 5 somitic cell population, including RIPPLY1 (see also updated Figure 4-Supplemental Fig 3). Beyond comparison of marker gene expression, it is difficult to assess the similarities and differences with the Matsuda dataset since their data lacks single-cell resolution. Thus, heterogeneity and efficiency of somitic fate induction within their cell population is unclear. Finally, neither of these two papers report the formation of somite-like structures.

Reviewer #2:

This study is interesting in the sense that it brings us one step closer to the formation of complex structures (such as somites) from human iPSCs. Markers that are typical of somites are indeed present at the end of the (rather complex) culture protocol. There is also quite a lot of work involved and the illustrations are of good quality.

However, the spatial organization that is typical of somites is lacking in a number of important ways. Early somites in amniotes are epithelial (in fact it is a pseudo-stratified epithelium made of bottle shaped cells), apical facing the somitocoele (a cavity filled with a loose mesenchyme) and basal to the outside. Quite similar to the organization of the neural tube. This organization is initiated in the anterior last third of the PSM and amplified concomitant to segmentation. It would be important to show that somitoids display such structure. It does not seem that there is a central cavity. It is also unclear from the picture whether somitoid cells are bottle-shaped. Importantly, one sees (Figure 5a) that F-actin (which labels the apical side of epithelial cells) is facing the outside of the somitoid, and not the inside as it should. In this condition, the term "somitoid" seems quite inaccurate in comparison to other organoid systems that faithfully reproduce their in vivo counterparts, not only the "classical" intestinal crypts but also the more recently published neural tube organoids. Aggregates of somite-like cells may be more accurate.

We are glad the reviewer found our study interesting and a step towards formation of complex structures in vitro. We believe that the reviewer has misunderstood the structure of the somites that we observe in vitro. We now include high magnification images that more clearly show the shape of the cells and the locations at which the epithelial polarity markers are expressed (new Figure 3-Supplemental Fig 1; new Figure 3-Video 1). Consistent with in vivo somites, we do observe bottle-shaped cells radially arranged around a central cavity forming rosette-like structures that are the same size as their in vivo counterparts (Fig 5B). This can be observed in multiple images shown in our manuscript (Fig 3-Supplemental Fig 1A-A’’, Fig 3-Supplemental Fig 1B-B’’, Fig 2-Supplemental Fig 3A,B). Importantly, both F-ACTIN (Fig 5A) and N-CADHERIN (Fig 3-Supplemental Fig 1) are indeed expressed around the central cavity, suggesting that the apical side of the PAX3+/TCF15+ somitic cells is facing the inside of the somite structures. This is especially evident in the newly added high-magnification images (Figure 3-Supplemental Fig 1) and accompanying movie showing a full confocal z-stack through the in vitro somites (Figure 3-Video 1). Taken together with our protein (Fig 3B,C and Fig 5A) and gene expression data (both in bulk (Fig 1C, Fig 5C, Fig 1-Supplemental Fig 1B) and at the single-cell level (Fig 4, Fig 4-Supplemental Fig 1-5) and our directed differentiation experiments of our Somitoid cells to dermomyotome (Fig 5-Supplemental Fig 1) and sclerotome fates (Fig 5C), we believe that the somite structures generated by our optimized in vitro protocol are indeed equivalent to their in vivo counterparts.

Author Response:

Reviewer #1 (Public Review):

Previous studies have provided crystallographic snapshots of the autokinase domains of several sensor histidine kinases (HK) involved in signal transduction in bacteria. Nevertheless, the lack of a full-length structure of these HK hampered the understanding of the molecular mechanism of signaling. Moreover, how a stimuli perceived by the membrane-bound sensor domain is transmitted to the catalytic cytoplasmic domain of an HK, to modulate its activity is poorly understood. To probe the coupling between the sensor and autokinase domains Mensa et al. used cysteine cross linking and reporter gene assay to probe the signaling state of E. coli PhoQ in a set of several point mutations. Using these data they developed a 3-domain model in which the sensor, HAMP and catalytic domain are in allosteric communication to interconvert the kinase state in an "on" or "off" conformation. The authors conclude that signals transmit to the catalytic domain through intradomain allosteric transitions, rather than through a concerted conformational change.

This work represents an important and novel attempt to understand the mechanism of signal transduction by sensor kinases in two-component systems. This contribution challenges the concept that signal transmit via propagation of single concerted conformational changes of the sensor kinase. The authors, instead, propose that signal is transmitted by the sensor via an interdomain allosteric mechanism.

The way in that the paper is presented appears to be directed to enzymologist working in enzyme kinetics models, rather than to a wide audience. For example, the paper starts saying that "Fully cooperative two-state models are unable to explain the gamut of activities of mutants". This affirmation seems too abrupt without defining what kind of model they are talking about. Is a molecular model, a kinetic model or a thermodynamic model? They should explain these concepts before to show the results. After we start to read it seems clear that they propose a thermodynamic model to explain the coupling of the different domains.

We have consolidated writing on modeling to the latter half of the manuscript. The introduction to this section now clearly states that we are exploring several thermodynamic allosteric coupling models.

For an enzymologist should be quite worrying to interpret data of activity assays with gene reporters without knowing the answers to the following: Do the mutations affect the PhoQ protein levels in cells? How accurate are the Western blots to quantify dimer formation in Y60C and establish the kinase "on" or kinase "off" states of PhoQ? The error bar of the of crosslinking experiments, shown in the different Figs, seems quite small for a Western blot quantification. Nevertheless, in Figure 8-figure supplement 2 panel, in the mutant I221F is obtained a poor fit which is not taken into account. Is it because the error is dismissed? Same for panels A and D. Is missing something the proposed model?

We have included Figure 2-figure supplement 1, which gives examples of crosslinking western blots and quantification (also “ Figure 2-figure supplement 1 - Source data 1”). These western blots also allow for the evaluation of total protein expression. In our model fitting, each individual replicate experiment was treated separately to give data with more replicates increased statistical weight as discussed in Results and Methods. The I221F mutation was measured in a single experiment. We also have conducted the assays under conditions where the concentration is approximately linear with respect to receptor occupancy as discussed previously by Goulian et al (ref 41).

Reviewer #2 (Public Review):

This manuscript characterized the effects of 35 mutational substitutions in three domains the bacterial transmembrane two-component sensor protein for Mg2+, PhoQ, on the signaling state of the periplasmic sensor domain and the cytoplasmic histidine kinase domain. Signaling state was assayed by a diagnostic cysteine cross-link for the sensor domain and the expression of a coupled beta-galactosidase reporter for the kinase domain. The results of those characterizations were used to develop an allosteric coupling model of conformational signaling from sensor domain to kinase domain, with a key role played by the HAMP domain that connects sensor to kinase. Single-site mutational substitutions were at positions expected to be in the interior of the protein structure in the periplasmic, HAMP and the S-helix regions of the protein as well as at boundaries of transmembrane segments. In addition, the connections between the second transmembrane helix and the HAMP domain, and between the HAMP domain and the S-helix were disrupted by introduction of a sequence of seven glycines. Each mutant protein was assayed for the signaling states of the sensor and kinase domains at five different concentrations of Mg2+. Some of the resulting dose-response curves showed patterns much like that for the wild-type receptor in which the signaling state of the sensor and kinase domain were correlated. However, a majority of the curves exhibited a variety of altered relationships between patterns for the two domains. Importantly, the effects of the glycine insertions before and after the HAMP domain indicated that this domain reduced the native "on" signaling state of both the sensor and kinase domain to be less extreme and thus in a more balanced state between on and off. Examination of the effects of similar glycine substitutions in two related two-component sensor kinases showed a similar negative influence of HAMP domain coupling on kinase domain signaling state. A global fitting using the allosteric coupling model between the three domains was performed for all 35 pairs of dose-response curves for PhoQ, allowing variation of one or a few individual parameters relevant to the position of the particular mutational substitution. An important validation of the resulting global parameters was a reasonable fit of the wild-type dose-response curves. The global parameters fit the experimental data for most but not all mutant receptors. Overall, the allosteric coupling model performed well, providing support for its validity.

Thus, this work provides support for concept of intra-receptor signaling via allosteric coupling between independent domains that each have their own intrinsic equilibrium between the "on" and "off" state. This allosteric coupling model introduces a third way of thinking about how ligand occupancy of a transmembrane receptor site facing the cell's exterior generates altered activity of a cytoplasmic domain inside the cell. Instead of considering that ligand binding "sends a signal" by sequential conformational changes that travel through the receptor structure or that ligand binding shifts a conformational equilibrium of the entire receptor in a concerted manner, the allosteric model suggests that signaling occurs by allosteric coupling between relatively independent domains of a multi-domain receptor protein. This constitutes an important contribution to our concepts of receptor conformational signaling.

However, the impact of this contribution is likely to be less than it could be because of the way the manuscript is written. Specifically, the devotion of a majority of the Results section to consideration of models for signaling obscures the most compelling parts of the work, the experimental observations of the striking effects of mutational substitutions throughout PhoQ on the signaling state of the sensor and kinase domains and the explanation of those disparate effects by the allosteric coupling model of conformational signaling. For many experimental scientists interested in mechanisms of signaling, this work would be much more accessible if the experimental results were presented first, the allosteric coupling model was introduced as a way to explain the results, and much of the consideration of other models and the development and details of the allosteric model were shifted to the Materials and Methods or provided as part of supplementary materials.

We thank the reviewer for this useful criticism. We have made several changes to bring forward and emphasize the experimental observations in our data. 1) We have moved concerted signaling model from Figure 2 (formerly Figure 2B, 2D) to Figure 5. New Figure 2 now contains an expanded set of experimentally generated data only. 2) We have supplemented Figure 2C with additional representatives of our diverse experimentally generated functional data. 3) We have added Figure 2-figure supplement 1 with examples of western blots and crosslinking quantifications. 4) We have moved the former Figure 4 forward to Figure 3. 5) We have moved the former Figure 5 forward to Figure 4. 6) We have added Figure 7-figure supplement 1 to further highlight and discuss rationale for choice of point mutations and Gly7 insertions. 7) We have made several changes in the Results section that mirror the above changes.

We have also made several changes to consolidate the discussion of the modeling work. 1) The concerted signaling model from the former Figure 2 (Figure 2B, 2D) and the 2-domain signaling model from former Figure 3 have been moved to a new Figure 5. 2) Former Figure 3D has been moved to Figure 5 figure supplement 1. 3) Population fraction equations in new Figure 5 (formerly Figure 2B, 2D, 3A) have been moved to Materials and Methods 4) Text discussing alternate allosteric models has been consolidated into one section.

Reviewer #3 (Public Review):

This manuscript describes a comprehensive study of kinase activation and allosteric coupling in the sensor histidine kinase (SHKs) PhoQ. Quantitative assays for sensor domain activation and kinase response are used to evaluate a large number of variant proteins that display a range of properties with respect to ligand binding, interdomain coupling and kinase activity. The data is used to construct and fit a conceptually elegant model that provides a thermodynamic explanation for domain interactions, allostery and sensing responses in SHKs. The experiments also demonstrate that sensor kinase domains intrinsically favor their "on" states and that HAMP domains act to deactivate both the sensor and the kinase units. In all it is a very impressive study that sets the bar for enzymatic approaches aimed at understanding signaling by multidomain transmembrane kinases. Generality of key principles are explored by examining several SHKs related to PhoQ. The paper is well written and the complex data and their interpretation are for the most part clearly discussed. That said, there are some issues the authors should address:

The model applied for Mg binding should be described to a greater extent. The equations of Figs. 1, 2,3,6,7 represent a situation more complex than the accompanying schematics portray. Even the simplest equation of 1B implies sequential binding of 2 Mg ions to one PhoQ dimer (presumably 1 site per subunit). Furthermore, the binding sites are assumed to be independent and, importantly, there are no intermediate states in the model in which one subunit is "on" (in either its sensor or kinase domain) and the other is "off". Is it known experimentally that the two subunits act independently and what is the consequence of not allowing for hybrid activation states within the dimer?

We discuss how we handle Mg2+ binding and has been elaborated based on this feedback. Given the data, we are unable to distinguish between a cooperative 2-state model versus a sequential binding model. We have 3 options: 1) 1-Mg2+/dimer creates an asymmetric signal state. This is well precedented in the review article cited (ref 17). 2) Binding occurs at both subunits in independent, unlinked events, or 3) binding occurs with negative cooperativity (first site higher affinity) or negative cooperativity (second site higher affinity). These alternatives differ only subtly in the steepness of transition from low to high signaling states. Unfortunately, our data are not sufficiently precise to distinguish between these options.

In addition, there may be a factor of 2 missing in treating the relative dissociation constants for Mg binding to an empty PhoQ or to a singly Mg-occupied PhoQ. Because the multiplicity changes by a factor of 2 in going from both the empty to the half-occupied state and again by 2 in going from the half-occupied to the fully occupied states, the effective Kd for binding to the singly occupied state is 4x larger than for binding to the empty state. It appears that all of the models accommodate only a factor of 2. This issue affects the (1 + [Mg]/Kd)2 term, likely to a minor extent.

Due to the difficulties in explicitly handling ligand binding in PhoQ as discussed above and in text, we report an overall ‘observed’ Kd for Mg2+. This observed Kd represents the true Kd for Mg2+ binding is implicitly scaled by the statistical factor of 2 (dimeric ligand binding), which we now state in lines 266-267. However, it is noteworthy that our purpose is to determine how mutants alter the energetic landscape, so differences such as multiplication of both the WT and mutant equilibrium by a constant factor (of 2.0) cancel out when comparing mutants.

In a similar vein, for the final models of Fig. 6,7, why is Mg binding only considered to selected states (SenOFF/HAMP1/Akon/off, for example)? And in Fig. 6A what does AK "on/off" signify?

All species are allowed to bind Mg2+, but only 2 such species are shown for clarity. Figure 6 legend has been modified to state this explicitly.

Line 549 - Discussion of the setpoint of the autokinase domain depends on the "reference point" given that KAK and alpha2 are correlated parameters. For example, one could view the intrinsic activity of the autokinase as being the fully uncoupled state, with KAK defined closed to 1.0 and alpha2 having a smaller value that currently modeled in the case of the Y60C (WT) protein. Could one fix KSen and KAK at the values for the Gly-decoupled systems and allow the shifts in equilibrium owing to HAMP coupling to be compensated solely for by alpha1 and alpha2? This framing might be more straightforward for understanding the HAMP coupling.

While some HAMP coupling mutations are adequately compensated by changes in α1/ α2, we adopted a universal standard state for local parameter variation, described in lines 420-425, 866-877. Thus, coupling mutations within the HAMP primary sequence were also allowed to alter KHAMP. In cases where α1/ α2 modulation is sufficient for fitting, the KHAMP value was found to be close to the global fit parameter value (which we have highlighted in green in Table 2). HAMP mutations near the autokinase junction also necessitated floating the KAK parameter for adequate fitting; therefore, we cannot fix KAK across the board for these coupling mutants. Furthermore, Gly7 insertions do not relieve the restraints associated with PhoQ being membrane-localized, and it is hard to consider the sensor and autokinase domains as fully uncoupled.

Although the reference position is largely arbitrary and in any given fitting scheme likely depends on the choice of constraining and fixing parameters, it does alter how one views the role of kinase-activating mutations. i.e. with the fully decoupled state as the reference, the HAMP is always deactivating, with different variants (including the WT) deactivated to varying extents. Some additional comments on this issue may help readers understand the range of kinase behavior and how it is influenced by HAMP.

Based on this comment, we have added lines 695-698 in the Discussion.

Related to the previous point, in Fig. 7 the alpha2 parameter seems to have a large amount of uncertainty, and appears biphasic in the fits, this behavior deserves a comment as to its impact in the model. How much would the interpretations change if alpha2 is considered to hold its extreme values?

We have added Figure 7-figure supplement 3 to show the effect of holding α2 at one of the 2 parameter value peaks, and have made additional comments in text (lines 468-473). There is no change in fitting quality at all values of α2 < 0.1, and the biphasic behavior appears artificial in the sense that it does not appreciably change the fit so long as α2 < 0.1. The bottom line is α2 for WT is consistent with strong negative coupling between the HAMP and catalytic domain.

p. 30 line 587 - It's unclear what is meant by the statement that the HAMP domain "serves to tune the ligand-sensitivity amplitude of the response" (p. 30 line 587). In this model, the HAMP domain does alter the sensitivity of the sensor domain by favoring the sensor OFF state (even though it does not directly modulate KdOFF), but what is meant by "sensitivity amplitude".

We have clarified this phrase on lines 665-667.

Author Response:

Evaluation Summary:

The authors studied the neural correlates of planning and execution of single finger presses in a 7T fMRI study focusing on primary somatosensory (S1) and motor (M1) cortices. BOLD patterns of activation/deactivation and finger-specific pattern discriminability indicate that M1 and S1 are involved not only during execution, but also during planning of single finger presses. These results contribute to a developing story that the role of primary somatosensory cortex goes beyond pure processing of tactile information and will be of interest for researchers in the field of motor control and of systems neuroscience.

We thank all reviewers and the editor for their assessment of our paper. We acknowledge that our description of the methods and some interpretation of the results can be clarified and expanded. We address every comment and proposed suggestion in the following below.

Reviewer #1 (Public Review):

This is a very important study for the field, as the involvement of S1 in motor planning has never been described. The paradigm is very elegant, the methods are rigorous and the manuscript is clearly written. However, there are some concerns about the interpretation of the data that could be addressed.

We thank Reviewer #1 for the positive evaluation of our study. We clarify our methodological choices and interpretation of the data in the following response.

• The authors claim that planning and execution patterns are scaled version of each other, and that overt movement during planning is prevented by global deactivation. This is an interesting perspective, however the presented data are not fully convincing to support this claim:

(1) the PCM analysis shows that correlation models ranging from 0.4 to 1 perform similarly to the best correlation model. This correlation range is wide and suggests that the correspondence between execution/planning patterns is only partial.

The reviewer is correct that the current data leaves us with a specific amount of uncertainty. However, it should be noted that the maximum-likelihood estimates of correlations between noisy patterns are biased, as they are constrained to be smaller or equal to 1. Thus, we cannot test the hypothesis that the correlation is 1 by just comparing correlation estimates to 1 (for details on this, see our recent blog on this topic: http://www.diedrichsenlab.org/BrainDataScience/noisy_correlation/). To test this idea, we therefore use a generative approach (the PCM analysis). We find that no correlation model has a higher log-likelihood than the 1-correlation model, therefore we cannot rule out that the underlying true correlation is actually 1. In other words, we have as much evidence that the correspondence is only partial as we do that the correspondence is perfect. The ambiguity given by the wide correlation range is due to the role of measurement noise in the data and should not be interpreted as if the true correlation was lower than 1. What we can confidently conclude is that activity patterns have a substantial positive correlation between planning and execution. We take this opportunity to clarify this point in the results section.

(2) in Fig.4 A-B, the distance between execution/planning patterns is much larger than the distance between fingers. How can such a big difference be explained if planning/execution correspond to scaled versions of the same finger-specific patterns? If the scaling is causing this difference, then different normalization steps of the patterns should have very specific effects on the observed results: 1) removing the mean value for each voxel (separately for execution and planning conditions) should nullify the scaling and the planning/execution patterns should perfectly align in a finger-specific way; 2) removing the mean pattern (separately for each finger conditions) should effectively disturb the finger-specific alignment shown in Fig.4C. These analyses would corroborate the authors' conclusion.

The large distance between planning and execution patterns (compared to the distance between fingers) is caused by the fact that the average activity pattern associated with planning differs substantially from the average activity pattern during execution. Such a large difference is of course expected, given the substantially higher activity during execution. However, here we are testing the hypothesis that the pattern vectors that are related to a specific finger within either planning or execution are scaled version of each other. Visually, this can be seen in Figure 4B (bottom), where the MDS plot is rotated, such the line of sight is in the direction of the mean pattern difference between planning and execution—such that it disappears in the projection. Relative to the baseline mean of the data (cross), you can see that arrangement of the fingers in planning (orange) is a scaled version of the arrangement during execution (blue). The PCM model provides a likelihood-based test for this idea. The model accounts for the overall difference between planning and execution by including (and estimating) model terms related to the mean pattern of planning and execution, respectively, therefore effectively removing the mean activation of planning and execution. We have now explained this better in the results and methods sections, also referring to a Jupyter notebook example of the correlation model used (https://pcm-toolbox-python.readthedocs.io/en/latest/demos/demo_correlation.html).

Regarding your analysis suggestions, removing the mean pattern for planning and execution across fingers as a fixed effect (suggestion 1) leads to the distance structure shown in Fig 4B (bottom)—showing that the finger-specific patterns during planning are scaled versions of those during execution (also see Fig. R1 below). On the other hand, subtracting the mean finger pattern across planning and execution (suggestion 2) will not fully remove the finger specific activation as the finger-specific patterns are differently scaled in planning and execution. Furthermore, neither of these subtraction analyses allows for a formal test of the hypotheses that the data can be explained by a pure scaling of the finger-specific patterns.

Figure R1. RDM of left S1 activity patterns evoked by the three fingers (1, 3, 5) during no-go planning (orange) and execution (blue) after removing the mean pattern across fingers (separately for planning and execution). The bottom shows the corresponding multidimensional scaling (MDS) projection of the first two principal components. Black cross denotes mean pattern across conditions.

• A conceptual concern is related to the task used by the authors. During the planning phase, as a baseline task, participants are asked to maintain a low and constant force for all the fingers. This condition is not trivial and can even be considered a motor task itself. Therefore, the planning/execution of the baseline task might interfere with the planning/execution of the finger press task. Even more controversial, the design of the motor task might be capturing transitions between different motor tasks (force on all finger towards single-finger press) rather than pure planning/execution of a single task. The authors claim that the baseline task was used to control for involuntary movements, however, EMG recordings could have similarly controlled for this aspect, without any confounds.

Participants received training the day before scanning, which made the “additional” motor task very easy, almost trivial. In fact, the system was calibrated so that the natural weight of the hand on the keys was enough to bring the finger forces within the correct range to be maintained. Thus, very little planning/online control was required by the participants before pressing the keys. As for the concern of capturing transitions between different motor tasks, that it is indeed always the case in natural behavior. Arguably there is no such thing as “pure rest” in the motor system, active effort has to be made even to maintain posture. Furthermore, if the motor system considers the hold phase as a simultaneous movement phase, it should have prevented M1 and S1 to participate in the planning of upcoming movements, as it would be busy with maintaining and controlling the pre-activation. Having found clear planning related signals in M1 and S1 in this situation makes our argument, if anything, stronger.

Finally, we specifically chose not to do EMG recordings because finger forces are a more sensitive measure of micro movements than EMG. Extensive pilot experiments for our papers studying ipsilateral representations and mirroring (e.g., Diedrichsen et al., 2012; Ejaz et al., 2018) have shown that we can pick up very subtle activations of hand muscles by measuring forces of a pre-activated hand, signals that clearly escape detection when recording EMG in the relaxed state. Based on these results, we actually consider the recording of EMG during the relaxed state as an insufficient control for the absence of cortical-spinal drive onto hand muscles. This is especially a concern when recording EMG during scanning, due to the decreased signal-to-noise ratio.

• In Fig.2F, the authors show no-planning related information in high-order areas (PMd, aSPL), while such information is found in M1 and S1. This null result from premotor and parietal areas is rather surprising, considering current literature, largely cited by the authors, pointing to high-order motor or parietal areas involved in action planning.

We agree with the reviewer that, to some extent, the lack of involvement of high-order areas in planning is surprising. However, we believe that task difficulty (i.e., planning demands) plays a role in the amount of observed planning activation. In other words, because participants were only asked to plan repeated movements of one finger, there was little to plan. The fact that this may have contributed to the null result in premotor and parietal areas was further confirmed by the second half of the dataset, which is not reported in the current paper. Here, we investigated the planning of multi-finger sequences, where planning demands are certainly higher. We found that high-order areas such as PMd and SPL were indeed active and involved in the planning of those, as expected. We decided to split the dataset across two publications as the multi-finger sequences have their own complexities, which would have distracted from the main finding of planning related activity in M1 and S1.

Reviewer #3 (Public Review):

I found the manuscript to be well written and the study very interesting. There are, however, some analytical concerns that in part arise because of a lack of clarity in describing the analyses.

1) Some details regarding the methods used and results in the figures were missing or difficult to understand based on the brief description in the Methods section or figure legend.

We thank Reviewer #3 for pointing out some lack of clarity in our description of the methods. We now expanded both the methods section and the figure captions (Fig. 2-3-4).

2) I think the manuscript would benefit from a more balanced description on the role of S1. As the authors state, S1 is traditionally thought to process afferent tactile and proprioceptive input. However, in the past years, S1 has been shown to be somatopically activated during touch observation, attempted movements in the absence of afferent tactile inputs, and through attentional shifts (Kikkert et al., 2021; Kuehn et al., 2014; Puckett et al., 2017; Wesselink et al., 2019). Furthermore, S1 is heavily interconnected with M1, so perhaps if such activity patterns are present in M1, they could also be expected in S1?

To better characterize the role of S1 during movement planning, we now include recent research showing that S1 can be somatotopically recruited even in the absence of tactile inputs.

3) Related to the previous comment: If attentional shifts on fingers can activate S1 somatotopically, could this potentially explain the results? Perhaps the participants were attending to the fingers that were cued to be moved and this would have led to the observed activity patterns. I don't think the data of the current study allows the authors to tease apart these potential contributions. It is likely that both processes contribute simultaneously.

We agree that our results could also be explained by attentional shifts on the fingers. It is very likely that, during planning, participants were specifically focusing on the cued finger. However, as the reviewer points out, our current dataset cannot distinguish between planning and attention as voluntary planning requires attention. We expanded the discussion section to include this possibility.

4) The authors repeatedly interpret the absences of significant differences as indicating that the tested entities are the same. This cannot be concluded based on results of frequentist statistical testing. If the authors would like to make such claims, then they I think they should include Bayesian analysis to investigate the level of support for the null hypothesis.

We have now clarified the parts in the manuscript that sounded as if we were interpreting the absence of significant difference (null results) as significant absence of differences (equivalence).

Author Response:

Reviewer #2 (Public Review):

Chan et al set out to assess the transcriptomic (bulk and single cell), proteomic and metabolic changes that occur as primary WI38 human lung fibroblasts progress from early proliferative stages through to replicative senescence (RS) in vitro, as well as using ATAC-seq to assess changes in chromatin accessibility in senescence. The authors compare findings from RS in primary WI38 cells with immortalised cells of the same lineage expressing hTERT, cells that are quiescent through contact inhibition and cells with radiation-induced DNA damage. The data presented confirm findings in the literature from individual -omics studies; what makes this work novel and provides new insight is the combination of a range of -omics techniques, including time resolved scRNAseq, to provide deep molecular profiling across the cell lifespan. This indicates that senescence is a process of gradual onset throughout the proliferative lifecourse, and that a few key pathways are strongly associated with (and probably drive) replicative senescence, particularly a fibroblast to mesenchymal transition (FMT) akin to the epithelial-mesenchymal transition (EMT) observed in cancer development. The identification of changes that occur at different stages along the senescence trajectory is important in that it may allow tailored interventions. Moreover, their finding of nicotinamide N-methyltransferase (NNMT) upregulation in senescence provides an explanation for the greater chromatin accessibility observed in senescence as well as NAD+ depletion.

The reliance on -omics techniques is also to some extent a weakness - no attempt is made to orthogonally validate the findings e.g. by qRT-PCR for transcripts, or western blotting for proteins identified to change on senescence. While the data on replicative senescence appear mostly robust, there are potential weaknesses in comparisons with DNA damage-induced senescence, as the early time points analysed may reflect more the acute DNA damage response rather than senescence. While it is sensible to conduct the full range of analyses on the same cell line to identify degree of concordance between gene expression control at RNA and protein levels, and correlate with metabolic consequences, there is only a cursory attempt to compare with other senescence models (a single published dataset on oxidative stress induced senescence in astrocytes) so the findings are at this stage confined to senescence in the WI38 cell line studied, though it is likely they will have much wider applicability.

We apologize for the lack of clarity in communicating that we did in fact compare our data to a model composed of multiple replicative senescence studies from different labs and different fibroblast cell lines compiled by Judith Campisis’s lab [Hernandez-Segura et al., 2017]. This comparison is the subject of what is now manuscript Figure 1D (1C previously). We report a very high correlation (r2=0.92) between our data and a large compendium of replicative senescence data. We apologize that this was not clear and have added clarifying text.

The comparison with oxidative stress induced senescence in astrocytes was used specifically to show that we observe similar putative regulatory TFs enriched in a senescence context far removed from replicative senescence in WI38 fibroblasts to suggest the possibility of wider applicability outside of WI-38 cells as mentioned in discussion. Although interesting and suggestive, this is outside the main thrust of the paper which was to generate a high resolution molecular description of replicative senescence in WI-38 cells.

A larger, comprehensive analysis to determine to what extent the TFs we highlight could be master regulators of senescence across many different senescence models and tissue cell lines is useful for the field, but it is outside the scope of this paper given the size of the literature. That analysis would be a publication in its own right similar to Hernandez-Segura et. al. If the reviewer finds the astrocyte comparison misleading or unhelpful we are happy to remove.

Author Response:

Reviewer #1 (Public Review):

The study aims to investigate the role of A11 neurons in courtship behavior and vocalizations. In particular, the authors determine the inputs/outpus of A11 neurons and uncover that the outputs are both dopamine and glutamate positive. They then lesion A11 cell bodies and terminals in the songbird song-motor nucleus HVC and find that these lesions affect song production, especially, though not exclusively, of courtship song. They also measure the location and movement of lesioned birds and find that birds with lesions of A11 cell bodies show less engagement with a female. Finally, they use fiber photometry to study the activity of A11 terminals in HVC during singing. While this is an interesting question supported by novel data, and I appreciate the diverse and creative approaches employed in this study, the role of A11 in courtship behavior appears complicated and does not easily fit into the framework proposed by the authors. In particular, the authors argue that A11 is important for coordinating innate and learned aspects of courtship, however, their data fall short of supporting this idea.

Strengths This is an impressive data set with considerable attention to detail.

The tracing and histology data identify some novel connections not previously described in songbirds as well as the potential of A11 neurons to co-release of glutamate and dopamine.

Photometry provides real-time monitoring of A11 and HVC neuron activity during singing.

In principle, targeting both HVC terminals and A11 cell bodies has the potential to lend insight into the role of HVC terminals vs. the role of projections to other areas (see below for caveats).

We appreciate the reviewer’s efforts and attention in evaluating our manuscript. We are grateful that the reviewer recognizes strengths in our study, which we agree provides novel insights into the brain circuits that enable a fully integrated courtship display comprising learned and innate behaviors.

Weaknesses 1) While I find the overall question and the data interesting, I am not convinced that they demonstrate that A11 is important for "coordinating innate and learned aspects of courtship". In general, birds with A11 lesions appear less motivated to perform female-directed song, however, it's not clear that this is a consequence of a lack of coordination between innate/learned aspects of behavior. Rather, perhaps A11 neurons are important to instigate or drive courtship behavior, or to relay signals from the POA or other regions important for courtship. Because the lesions abolish behavior, it is difficult to discern the role of these neurons in courtship.

We agree that discerning the precise role of A11 is tricky. It could be acting to gate a (motivational) drive from another source, providing a primary source of this drive, and/or performing a more intricate role in coordinating the various aspects of the courtship display. The reviewer is correct that the current experiments do not allow us to clearly distinguish between these possibilities, and we have revised the manuscript accordingly, first by replacing “coordinate” with “gate” in the title and introduction and including a more thorough treatment of gating and other possible roles for A11 in lines 258-262 of the discussion. That said, we lean towards the latter possibility - a coordinating role for A11 - because of its location immediately proximal to regions that drive learned (HVC) and innate (ICo, RPgc) aspects of behavior and because A11 neurons can contain synthetic enzymes for a fast acting neurotransmitter (glutamate) in addition to DA. But, ultimately, we acknowledge that future experiments are needed to more completely answer this question.

In addition, I disagree with the innate vs. learned distinction as recent data indicate that introductory notes, which the authors treat as innate, are actually learned (e.g. Kalra et al., 2021). Further, there is also no quantification of the effects of lesions on female-directed calls and little analysis of the activity during call production. This would seem to further complicate the overall interpretation. Overall, it's difficult to make sense of how A11 activity relates to vocalizations, especially given the innate/learned framework that they focus on.

We thank the reviewers for drawing our attention to the recent Kalra 2021 paper, which we now cite while also making sure to emphasize that introductory notes may have learned features (lines 194-195 and 278-279). However, even that recent study concluded that males raised without a tutor or tutored on recorded songs that lack introductory notes altogether still developed songs that include introductory notes. Nonetheless, we include citation of this recent study and qualify our characterization of introductory notes as being shaped by innate predispositions and experience. Furthermore, we conducted additional analyses to quantify female-directed calling before and after 6-OHDA lesions in either HVC or A11 (results can be found in lines 164-165 and Figure 4C). In line with the divergent effects of these two types of lesions on the production of introductory notes, lesions in HVC did not affect female-directed calling whereas lesions in A11 largely abolished these vocalizations. While we acknowledge that the fiber photometry data on female-directed calling was limited, it nonetheless reinforces the conclusion that A11 transmits information to HVC about innate vocalizations, and it also transmits information to HVC about introductory notes. Along with the loss of introductory note production following A11 lesions, we do believe that our findings support the idea that A11’s role is essential to female-directed vocalizations generally, regardless of whether they are learned or innate, and of of somehow enabling the transition from production of female-directed calling and introductory notes to motifs. We have done our best to draw out these points in the revised discussion.

2) The HVC lesions appear to create damage/necrosis (Fig 3-suppl 2) and this raises the question of the degree to which the HVC lesion effects are the result of dopamine/glutamate depletion or local damage. In particular, it is surprising that syllable structure and stereotypy show such a dramatic breakdown with HVC A11 input lesions and effectively no change with lesions of the cell bodies, even though both treatments lead to effectively similar reductions in song production.

We appreciate that 6-OHDA lesions are not highly specific and can introduce unwanted effects on non-TH+ cells and processes. To further quantify the effects of 6-OHDA lesions on HVC cells, we conducted additional 6-OHDA injections in HVC and TUNEL staining studies in addition to the preliminary efforts we had made in the original manuscript. Quantification of these data confirmed our original impression that 6-OHDA treatment in HVC increased HVC cell death (these data are shown in Figure 3-figure supplement 2J, K). To further address this issue, we also added an analysis of song structure when D1 receptor blockers were dialyzed into HVC. No changes in song morphology were detected, similar to the lack of effects on song morphology following A11 cell body lesions (Figure 3 - figure supplement 3). Taken together, these additional experiments and analyses indicate that the changes in song morphology following 6-OHDA treatment in HVC may arise from local damage to HVC cell bodies. In contrast, the reduction in singing following A11 terminal or cell body lesions is likely to reflect diminished DA signaling from A11. However, as the reviewer notes, our primary finding is the differential effects on female-directed singing, and the distinction between more purely singing-related effects following 6-OHDA treatment in HVC and a broad effect on all courtship behaviors following 6-OHDA treatment in A11.

3) If the idea is that A11 is important for coordinating innate and learned movements, it seems that a detailed analysis of the movements would be important. As is, the movement data provide further support of a decrease in either the motivation or ability to perform female-directed song, but they do not speak to a more specific role for A11 in coordinating innate and learned movements.

We maintain that we did provide a detailed analysis of a number of important nonsong behaviors, including changes in head orientation and translational movements that the male makes towards the female, both of which are major appetitive features of courtship in songbirds and other vertebrates. We also appreciate that these analyses do not allow us to say much about precisely how movements are being coordinated during courtship, and we have changed language throughout the manuscript to emphasize a gating rather than coordinating role for A11. Furthermore, in response to the reviewer’s concern, we performed additional analyses of the male’s movements during courtship, including beak wipes, vertical changes in posture (“standing tall”), which are finer components of female-directed displays. Notably, this new analysis reveals that all of these behavioral components are abolished by A11 cell body lesions, but not by A11 terminal lesions in HVC (lines 168-190 and Figure 4I, J). We appreciate the reviewer’s suggestion, as we believe these additional analyses strengthen our core finding, namely that A11 functions as a hub to gate, recruit and possibly coordinate innate and learned movements to generate a complete courtship display. These different roles are more fully considered in the revised discussion (lines 256-262).

Reviewer #2 (Public Review):

Ben-Tov et al. investigate function of midbrain region A11 and provide evidence that it plays a role in promoting and coordinating a variety of motor responses to sexually or socially salient stimuli. They show lesions of A11 cell bodies abolish female directed calling, orienting and singing, while lesions of terminals in the song premotor nucleus HVC prevent female directed singing, but leave female directed calling and orienting intact. Together with anatomical data indicating projections from A11 to multiple downstream targets associated with song (HVC), calling (DM/ICO) and locomotion, these data support the authors' idea that A11 forms a 'hub' that drives and 'coordinates' multiple different aspects of behavioral responses to social (here female/sexual) stimuli. The results are intriguing and begin to reveal how a single social context can elicit and coordinate multiple coordinated responses. However, as outlined below, I think that some of the specific stronger claims would benefit from additional data, discussion or moderation.

The authors also provide compelling support for the idea that A11 plays a differential role in female-directed versus undirected song. This is especially underpinned by the observations that 1) A11 afferent activity in HVC appears to differ between directed and undirected signing, with increases in activity preceding song motifs only during directed song, and 2) lesions of A11 cell bodies or inputs to HVC have a dramatic suppressive effect on directed singing, but can leave undirected song largely unchanged. These observations that A11 differentially contributes to socially elicited versus spontaneous singing seem especially interesting and merit further highlighting and discussion as one of the especially striking aspects of the study that seems distinct from the thesis of a role in coordinating learned and unlearned behaviors.

We appreciate the reviewer’s efforts and attention in evaluating our manuscript. We are grateful that the reviewer recognizes strengths in our study, which we agree provides novel insights into the differential contribution of A11 to socially elicited versus spontaneous singing. We also agree that this point should be highlighted and we expanded our treatment of this point in the discussion section of the revised manuscript (lines 296-309).

Specific comments

A central idea around which the results are discussed is that A11 plays a particular role in coordinating learned versus innate behaviors. I have several questions around this thesis where further guidance from the authors about both technical points and interpretation would be helpful.

First is the question of how specific are the manipulations and conclusions to A11 itself versus other neighboring midbrain dopaminergic regions within which it is embedded. The authors show histology of lesions, injection sites and retrograde labelling in supplementary figures, but do not provide enough guidance for me to understand the strength of the argument that manipulations are restricted to A11 and/or its afferents. Can the boundaries between A11 and neighboring regions be better demarcated? What are the neighboring regions to which there might have been spillover? For lesions of A11 axons within HVC, wouldn't 6-OHDA also damage any other dopaminergic afferent to HVC, including those coming from regions such as VTA? Some discussion of these and related points regarding the specificity of manipulations to A11 would be helpful, especially in light of the literature that points to potential roles of neighboring dopaminergic regions in contributing to motivated behaviors and song more specifically.

We appreciate that the definition and boundaries of A11 might be confusing. We demarcated A11 and neighboring regions in the relevant figures to better define A11’s boundaries. The reviewer is correct in surmising that the VTA is fairly close to A11 and hence a reasonable concern is that 6-OHDA treatment in A11 could spill over to the VTA and possibly the SNc. To address the concern that 6-OHDA lesions in HVC might cause cell damage to other DA sources to HVC, we quantified the number of VTA/SNc cells following HVC DA lesions. This additional analysis, provided in Figure 3-figure supplement 1D-F, shows that the number of VTA/SNc cells following 6-OHDA injections into either A11 or HVC is comparable to that of intact birds. These additional analyses support the conclusion that the behavioral deficits that emerge following 6-OHDA treatments reflect damage to A11 or A11 terminals in HVC.

These points also relate to the general question of what is meant by A11 being a 'hub for coordination of learned and innate courtship behaviors'. Ultimately, it seems likely that many regions must work together to orchestrate these behaviors, and it is not clear from the present results how much I should view A11 as having a more specific role than other neighboring dopaminergic regions (or hypothalamic regions such as POA) that are interconnected and seem likely to also play critical roles. As the authors note, many of the relevant structures, including A11 and song system structures, are recurrently connected, further complicating interpretation of any one area as a hub. In this respect, I am not sure how much the authors are intending to argue that A11 is both necessary and sufficient for driving each of the studied behaviors in a courtship context, and it would be helpful to discuss this more specifically - does 'coordination' as used here imply that A11 is capable of triggering these behaviors - an interesting possibility raised by the current results but that does not yet seem to be demonstrated - or something else?

As we noted in our response to a similar point made by the first reviewer, we agree that discerning the precise role of A11 is tricky. As we commented in that earlier response, A11 could gating a (motivational) drive from another source, providing a primary source of this drive, and/or performing a more intricate role in coordinating the various aspects of the courtship display. We agree that the current experiments do not allow us to make a clear distinction between these possibilities, and we have revised the manuscript accordingly, including a more thorough treatment of these various roles for A11 in the discussion (lines 256-262). That said, we lean towards the latter possibility - a coordinating role for A11 - because of its location immediately proximal to regions that drive learned (HVC) and innate (ICo, RPgc) aspects of behavior and because A11 neurons can release a fast acting neurotransmitter (glutamate) in addition to DA. But, ultimately, we acknowledge that future experiments are needed to more completely answer this question. In the revised manuscript, we emphasize a gating role for A11 in the title and introduction, and then in the discussion expand to encompass the possibility of a coordinating or timing role for A11.

One additional question regarding the framework for interpreting the function of A11 as coordinating 'learned and innate' courtship behaviors, is for some further clarification and citations regarding what is learned versus innate, especially as it relates to song. The authors characterize introductory notes as 'innate', but previous work from Rajan and colleagues has demonstrated that aspects of introductory notes including acoustic structure and patterning are influenced by learning, and I am not sure what the literature says about orienting and calling to females.

We thank the reviewer for drawing our attention to this recent study from the Rajan group which indeed concluded that some aspects of the introductory notes are learned. We also note that this study showed that juvenile males tutored on song playbacks that lacked introductory notes or that were raised without a tutor still produced introductory notes. Nonetheless, we include a citation of this recent study and qualify our characterization of introductory notes as being shaped by innate predispositions as well as through experience and learning (lines 194-195 and 278-279). Furthermore, our original analyses of birds with 6-OHDA treatment in HVC revealed that introductory note morphology was unchanged, whereas syllable morphology was degraded. Therefore, even if certain features of introductory notes are influenced by tutor experience, they apparently do not depend on HVC in the same manner as do the learned syllables in the motif. Lastly, we conducted additional analyses to quantify female-directed calling and other movements, before and after 6-OHDA lesions in either HVC or A11. In line with the divergent effects of 6-OHDA treatment in these two regions on the production of introductory notes, lesions in HVC did not affect female-directed calling, beak wipes or changes in male’s posture, whereas lesions in A11 largely abolished all of these behaviors (Figure 4C, I, J). While we agree with the reviewer that a distinction between innate and learned behaviors may not be straightforward, the more fundamental observation is that we can dissociate different aspects of the courtship display and that A11 is situated in a position to drive, gate or coordinate a unified display that involves a variety of learned and innate vocal and non-vocal movements.

I also would find it helpful to have some further clarification in this context about what it means to coordinate learned and innate aspects of song. The authors indicate that undirected song is largely unaffected by A11 lesions while directed song is largely eliminated, leaving only innate calls or introductory notes. I think it would be helpful to see here a more complete characterization of the nature of vocalizations that remain following A11 lesions in the female directed context. While I understand that no recognizable 'learned motifs' are produced, it is unclear from the example that is shown how much the residual vocalizations should be construed as 'severely disrupted songs' versus strings of calls that resemble innate calls that were present prior to lesions, versus 'normal' patterns of introductory notes that resemble in acoustic structure what the birds produced prior to lesions, but that never proceed to song motifs, etc. A better understanding of the nature of these residual vocalizations might also help to interpret what A11 is doing. Do these birds seem motivated to 'sing' in terms of their posture? Do the authors think that HVC is engaged or that the same residual vocalizations would be produced in a bird that had HVC lesions? How do the authors interpret these data in terms of how learned and unlearned vocalizations are normally coordinated in the context of directed singing?

We performed additional analyses of the male’s vocalizations and movements during courtship, including female-directed calls, beak wipes, vertical changes in posture (“standing tall”), all of which are components of female-directed courtship displays. Notably, this new analysis reveals that all of these behavioral components are abolished by A11 cell body lesions, but not by A11 terminal lesions in HVC (lines 168-190 and Figure 4C, I, J). Along with our prior report that males with A11 cell body lesions do not sing female-directed motifs, the additional analysis indicates that these males produce little or no female-directed vocalizations or non-vocal behaviors of any kind.

We previously reported that males with A11 terminal lesions produced only introductory notes but not motifs but realize that this observation would benefit from more quantification. As noted in the previous response, we previously established that introductory note morphology was unchanged by 6-OHDA treatment in HVC (Figure 4 - figure supplement 1A-D). To extend this analysis further in this revised manuscript, we built on the observation that males with 6-OHDA treatment in HVC produce only introductory notes to females, with no song motif, whereas they produce a series of introductory notes followed by motifs comprising distorted syllables when alone (Figure 3K, Figure 3 - figure supplement 2, Figure 4B). To confirm that the directed introductory notes and undirected syllables were indeed distinct vocalizations, we computed their durations and spectral similarity scores (using Sound Analysis Pro). The introductory notes produced during directed conditions differed markedly in their durations from distorted syllables produced during undirected conditions, and these two types of vocalizations had very low similarity scores, indicating that they were cleanly separable vocal behaviors (Figure 4 - figure supplement E, F). Given that introductory notes are unchanged by 6-OHDA treatment in HVC, these analyses support the idea that males treated in this manner can still produce motifs, albeit distorted ones, when alone but not when in the company of a female.

These questions relate in part to that of how much is the trigger to sing eliminated by A11 afferent lesions versus the ability to produce the relevant song output? It seems like there may still be a trigger to sing - short latency vocal response to female - but inability to produce motif. One point that may be interesting to note in this regard is that this seems somewhat opposite of observations made in other contexts about the effects of directed versus undirected context on song - for example, juveniles can produce better song when it is directed (Kojima), and deafened birds that are beginning to exhibit song deterioration can exhibit normalization of song structure during directed conditions (Nordeen).

We agree with the reviewer’s point that birds with 6-OHDA lesions in HVC may still be triggered to sing, but are unable to produce a motif, given that they still produce introductory notes and seem to have the right posture, orientation and proximity to the female. We appreciate the reviewer’s comment regarding changes in song that can be elicited by females in either juvenile males or adult males that are deaf, although these additional contexts fall outside of the current study, which focused on adult male finches with normal hearing.

Reviewer #3 (Public Review):

The authors use a combination of quantitative acoustic and other behavioral analyses to evaluate the role of the midbrain dopaminergic area A11 in the production of female-directed song in adult male zebra finches. They show that female-directed courtship displays, which consist of song and the production of female-directed displacement behaviors, are dependent on A11 because targeted chemical lesions of this structure, using 6-hydroxydopamine (6-OHDA), permanently (i.e. for at least several months) eliminate both the vocal and non-vocal elements of this behavior. Destruction of A11 axons that directly target HVC, by administering 6-OHDA into HVC, only eliminates female-directed singing without causing any change in the other observed female-directed behaviors. Because these same lesions only temporarily (5-10 days) abolish undirected song, these findings suggest that A11 is not directly involved in song production but acts instead as a gate for the production of female directed courtship behaviors. The authors follow these lesion studies with fiber photometry-based calcium imaging of A11 axons that target HVC to show that A11 activation patterns precede activity in HVC during female-directed singing and that calcium elevation is primarily elevated during the production of the many introductory notes (a component of song that is primarily observed during female-directed singing) that precede the production of the learned song motif. These findings suggest that A11 inputs to HVC likely play a role in triggering and/or activating HVC to synchronize the production of introductory notes (which are likely produced by midbrain circuits) with the learned song component that immediately follows them. In contrast, activation of A11 axons during undirected song (which contain few to no introductory notes) do not precede HVC activation patterns. Consistent with the rapid transmission of A11 neurons, the authors also confirm, as has been suggested for A11 in mammals, that A11 dopaminergic neurons co-release glutamate.

The findings of this study are of significant interest to our understanding of the neural mechanisms by which these complex behaviors are synchronized and open up a new way of thinking about how learned behavioral motifs can be synchronized with non-learned (e.g. female displacement behavior) behaviors. The study is rigorous, with many different experimental approaches being used to examine the proposed hypotheses, and the findings are convincing. Particularly impressive is the complete elimination of female-directed courtship behaviors following targeted elimination of A11. The primary weaknesses of the manuscript lie (1) in the way they present their anatomical findings and (2) how the authors discuss their findings in the discussion. In the discussion, which is very short (~750 words), the authors miss the opportunity to draw parallels with similar studies in drosophila (they only provide a cursory statement with a few references). In the discussion, the authors propose a model that seems quite oversimplified and lacks, in fact, many of the anatomical connectivity that they show in the first part of their study (for example A11 is only shown having a unidirectional connection to ICo/DM when in fact the connections are bidirectional). The model is also presented in simple hierarchical fashion with many connections omitted. Perhaps these omissions were made to simplify the model but in my opinion such simplification possibly misrepresents the actual mechanisms involved in the coordinated control of courtship song.

We thank the reviewer for their careful reading of the manuscript and his supportive and constructive comments. We agree that the loss of all female-directed behaviors (which we now extend to female-directed calling and other non-vocal behaviors, such as beakwipes and postural changes) following A11 cell body lesions is especially intriguing. Further, the different effects of A11 cell body lesions and A11 terminal lesions in HVC, along with the connectivity of A11, indicate that A11 acts via a range of downstream sites to gate these various female-directed behaviors. We have done our best to address the two primary weaknesses identified by the reviewer. First, we have done our best to provide a more detailed accounting of the anatomical findings. Second, we have expanded the discussion to address parallels with other studies, as in the fly, and to provide a more nuanced and complete consideration of how A11 may function to facilitate male courtship behaviors.

Author Response:

Reviewer #1:

Reviewer 1 expresses some concerns regarding concentrations of soluble proteins during our experiments. This is a good point and, in response, we are rewriting the manuscript to more clearly describe the metastable nature of the soluble protein pool. The key feature of our reaction mixture is that it contains both profilin and capping protein, which work together to suppress filament assembly. Spontaneously nucleated filaments are rapidly capped at their barbed ends. Profilin then effectively prevents elongation from the pointed ends of these filaments and they disassemble. We will cite relevant work that establishes and discusses these synergistic activities. Young et al. (1990) found that factors that cap >90% of filament barbed ends increase the critical concentration from that of the barbed end to that of the pointed end, and several groups demonstrated that profilin and barbed-end capping proteins work together to suppress filament assembly and promote disassembly of filaments with free pointed ends (e.g. DeNubile, 1985; Blanchoin, 2000; Pernier, 2016). This combination produces a large pool of monomeric actin that is capable of transiently elongating any newly formed barbed ends. We previously described this pool as ‘metastable’ (Pollard, 2000) while others have described it as ‘dynamically stable’ (Pernier, 2016). Only the branched actin networks formed by the micro-patterned nucleation promoting factors have an appreciable lifetime and consume a significant fraction of the soluble proteins, because filaments can only be formed by continual rounds of nucleation and only remain stable when their pointed ends are capped by the Arp2/3 complex (Blanchoin, 2000). In addition, the total amount of protein incorporated into the micro-patterned branched networks is only a small fraction of the total protein present in the reaction mix. This is demonstrated by the fact that the network growth rate is constant over the course of each experiment. We will mention this in the revised manuscript and provide the following simple calculation to emphasize this point: The concentration of actin in our reaction mixes is 5 µM, with a total volume of 150 µl. The maximum concentration of actin in our networks is 1.25 mM, but the maximum total volume of these networks is only <0.002 µl (based on a total of 400 WAVE1 patches with an average area of 50 µm^2, generating networks with a maximum height of <100 µm). The fraction of actin used up during an experiment, therefore, is less than 0.3%.

REFERENCES:

Blanchoin L, Pollard TD, Mullins RD. (2000) Interactions of ADF/cofilin, Arp2/3 complex, capping protein and profilin in remodeling of branched actin filament networks. Curr Biol. 10(20):1273-82.

DeNubile MJ, Southwick FS. (1985)Effects of Macrophage Profilin on Actin in the Presence and Absence of Acumentin and Gelsolin J. Biol. Chem. 260(12):7402-7409.

Pernier J, Shekhar S, Jegou A, Guichard B, Carlier MF. (2016) Profilin Interaction with Actin Filament Barbed End Controls Dynamic Instability, Capping, Branching, and Motility. Dev Cell. 36(2):201-14.

Young CL, Southwick FS, Weber A. (1990) Kinetics of the Interaction of a 41-Kilodalton Macrophage Capping Protein with Actin: Promotion of Nucleation during Prolongation of the Lag Period. Biochemistry, 29:2232-2240.

Reviewer #2:

Reviewer #2’s comments about the molecular mechanism underlying the force-induced increase in free barbed ends make it clear that our explanation was not as clear as it should have been. We will provide more detailed derivations for our mathematical methods, but in the meantime, we hope that the following explanation will clear up any misunderstanding.

Reviewer 2 rightly notes that “…for both capping and branching, the authors find that they decrease the same way with increasing loads - as they should: this is imposed by their being at steady state, where the birth rate of growing barbed ends (branching) must match their death rate (capping).” This steady state condition is actually the starting point for our analysis. At steady state the overall rates of nucleation and capping must be equal (Rcapping = Rnucleation). Importantly, the overall rate of nucleation is a complicated function that depends on the occupancy of the WH2 domains, the surface-associated Arp2/3 complex, and the local density of polymeric actin. On the other hand filament capping in our system appears to be a simple bimolecular interaction between soluble capping protein and free barbed ends. We demonstrated this by showing that the average filament length (i.e. the ratio of polymeric actin to capping protein in the growing network) varies as a simple inverse function of the capping protein concentration. This means that the overall rate of nucleation (Rnucleation) must equal the product of the capping protein concentration ([CP]), the surface density of free barbed ends (E), and an appropriate capping rate constant (kc). This yields,

kc[CP]E = Rnucleation

Which can be rearranged to give the density of free barbed ends,

E = Rnucleation/(kc*[CP])

As the reviewer notes, this equation describes a density, not a unitless number. Note that a sudden decrease in per-filament capping rate (e.g. a decrease in the rate constant, kc) with no change in overall nucleation rate will cause the number of free barbed ends to increase until the overall rate of capping (kc[CP]E) once again matches the overall rate of nucleation. This equation is an “iron law” imposed by the steady-state (or quasi-steady state) character of the system, and it means that any increase in the density of free barbed ends must reflect EITHER an increase in the overall rate of nucleation OR a decrease in the per-filament capping rate (or possibly both). Our direct measurements of the overall nucleation rate (the quantity in the numerator) rule out the first possibility, meaning that the per-filament capping rate MUST go down with applied force. Furthermore, our measurements demonstrate that this capping rate displays the same force sensitivity as actin filament elongation. The best explanation for this phenomenon is that the insertion of a capping protein onto a filament barbed end is subject to the same constraints as the insertion of an actin monomer. This could have been predicted from Brownian Ratchet theory, but as the reviewer points out, it was not. Our “bulky capping protein” experiments are a direct test of whether Brownian Ratchet theory can account for the force sensitivity of filament capping, and they demonstrate that it can.

In summary, we stand by our original explanation, namely that applied forces cause a decrease in the rate at which individual filaments are capped (via a change in the rate constant for filament capping, kc). This decrease, which can be explained by Brownian Ratchet Theory, leads directly to an increase in the steady-state barbed end density.

Author Response:

Public Review:

This manuscript from Pacheco-Moreno et al. compares the microbiome of potato fields with and without irrigation. Irrigation is known to control potato scab caused by Streptomyces scabies and the authors hypothesized that changes in the microbiome may contribute to disease suppression after irrigation. Using 16S rRNA sequencing, they identified a number of taxa, including Pseudomonas that are enriched after irrigation. They went on to isolate and sequence the genomes of many Pseudomonas strains. By correlating the ability of Pseudomonas to suppress Streptomyces growth in vitro with genomic data, the authors identified a novel group of cyclic lipopeptides (CLPs) that can inhibit Streptomyces in vitro and in planta.

This work provides a substantial contribution that advances our understanding of disease suppressive soil mechanisms. It is novel in scope in that it focuses on suppression of a bacterial pathogen, while many prior studies focus on suppression of fungal pathogens. Additionally, the large-scaled comparative genomics is a useful resource, and the identification of CLPs that inhibit Streptomyces is novel. Importantly, the authors provide in planta data to show role a for CLPs in disease suppression in vivo. The manuscript is well written and the data are well presented. The analyses are quite thorough and I appreciate the extensive use of genetics and metabolomics to support the genomic predictions. The main weakness is a lack of data the conclusively links the change in microbiome function to disease suppression after irrigation in the field. However, I think the data they've presented, combined with those in the drought literature, might suggest that an increase in total Pseudomonas (and the corresponding disease-suppressive genes) in well-watered soil might contribute to suppression, rather than a change in function of Pseudomonas.

While the reviewer is correct that we cannot conclusively link disease suppression to a change in microbiome function after irrigation, we are confident that our results demonstrate a real and repeatable phenomenon that must be considered in future studies of soil scab suppression. Independent field experiments conducted two years apart both show a decrease in the proportion of suppressive pseudomonads associated with potato roots. The first experiment (Figures 1 & 2) contained too few sequenced isolates to draw statistically robust conclusions, therefore we designed the second experiment (Figure 8) to investigate this phenomenon further. This experiment showed highly significant differences in the proportion of suppressive isolates on irrigated and non-irrigated roots. The alternative hypothesis presented by the reviewer; that relative Pseudomonas and Streptomyces abundance are affected by irrigation and this may be a factor in scab suppression, is also a valid possibility, although relatively small abundance changes were observed in the data reported in Figure 1. We have amended the discussion to include this as an alternative explanation for our results.

Author Response:

Reviewer #1 (Public Review):

The authors conducted an extensive analysis of transcription termination in Methanococcus maripaludis, which is an archael species. Using a combination of functional genomics, statistical analyses, in vitro/biochemical approaches, and reporters; the authors explore mechanistic aspects of termination. The authors determine that the archael CPSF protein binds an upstream uridine tract using a KH domain. The RNA binding activity of aCPSF is not present in eukaryotes and is shown to be important for termination in a uridine tract dependent manner.

Overall, the work is well-conceived and, in general, the conclusions by the authors are supported by the data. Investigation of archael species is not mainstream but still has significant potential to impact the field of transcription as the process of termination is still be unraveled. Moreover, several aspects of the methodology would benefit other researchers - notably, the development of Term-seq. The readers would benefit from consulting early structural studies of aCPSF to more fully gain a perspective on the interesting aspects of KH domain presence in this homology of CPSF.

Immensely thank you for the positive comments.

Reviewer #2 (Public Review):

Transcription termination defines accurate transcript 3'-ends and ensures programmed transcriptomes, making it critical to gene expression regulation. Our understanding of archaeal transcription termination mechanisms is still limited. Li et al present new evidences that support their model of aCPSF1-dependent transcription termination in Archaea (Yue et al NAR 2020). Importantly, they show that aCSPF1 recognize U-rich terminator signals with its KH-domain, using the methanogen Methanococcus maripaludis, as study model. The reported results of the manuscript are the continuity of their work published in Nucleic Acid Research journal in 2020 (Yue et al NAR 2020). Previously, the authors demonstrated the importance of the absolutely conserved ribonuclease aCPSF1 to terminate transcription by cleaving the transcripts at their 3' ends. They proposed a model for transcription termination in Archaea in which they defined aCPSF1 as general transcription termination factor. In here, by reinvestigating Term-seq data and by using RNA-protein binding assays (EMSA & SPR) and genetics experiments, Li et al argue that (i) PolyU-tracts are signals for transcription termination in M. maripaludis, (ii) aCPSF1 binds to PolyU-tract signals in transcript 3'-ends through its KH domain, and (iii) transcription termination is more effective in presence of aCPSF1. In general, the experiments and analyses that are shown are well-conducted. A major criticism is the lack (i) of quantification of RNA-protein binging assays which will allow going deeper in the understanding of how aCPSF1 specifically recognized PolyU-tract signals and (ii) of data on the oligomerisation status of aCPSF1. It is important to decipher if aCPSF1 is acting as a monomer or a dimer. Both will be helpful for proposing a more precise model for transcription termination mechanism in Archaea.

Immensely thank you for the positive comments and the valuable suggestions. According to your comments and suggestions, we have supplemented the following experimental data in the revision as (i) the aCPSF1-U rich RNA binding association contents (Kd) have been quantified for the binding assays; (ii) truncation of the C-terminal 13 residues (C13) that are essential for aCPSF1 dimerization leads to the loss of the in vivo termination function of aCPSF1, demonstrating that the dimerization of aCPSF1 is essential to its routine function; (iii) Footprint assay also determined that aCPSF1 binds to the U-tract region of the tested terminator. All the related results and discussions have been supplemented in the revised manuscript.

Reviewer #3 (Public Review):

In this manuscript by Li et al. examine U-rich motifs enriched for the transcript end sites in archaea M. marpaludis and analyze the role of aCPSF1 and its KH domains in binding these U-rich motifs. Their data indicate aCPSF1 binding to U-rich motifs is necessary for efficient 3' end definition of transcripts. Overall this work is well carried out, but there are also several key issues that should be addressed in order to more fully support the conclusions of the authors.

Immensely thank you for the positive comments.

Major:

1. The conclusion that aCPSF1 functions as a back-up termination is not supported by their data. As far as I can tell, back-up termination should happen at non-primary sites, which have a lower frequency of U quadruplex. It is not clear how aCPSF1 functions at those sites.

Combination of the comments of you and the reviewer 1, this related inappropriate conclusion has been removed in the revised manuscript.

1. The authors appear to indicate that aCPSF1 is the sole factor for 3' end cleavage of archaeal transcripts. But this is not supported by the data. Their data indicate both U-rich motif and aCPSF1 are necessary for cleavage. No data are shown to indicate that these two alone are sufficient for 3' end cleavage.

The obscure related description has been revised as “the only trans-action factor”. (Line 36 in the Abstract and Lines 476-477).

1. The U-rich motif (U quadruplex) was recently reported in their NAR paper (Yue et al.). There appears to be limited additional information on motifs in this work. It would be useful to readers to know if U-rich motifs are the only type for 3' end cleavage. The authors may want to examine motifs beyond single nucleotide models (which is what they are doing in this work). For example, are there any k-mer enrichments besides U quadruplex?

According to your advice, we searched RNA sequences proximal to TTSs; however, no other motif in addition the U-rich sequence is found.

Minor:

1. Strictly speaking, they are measuring transcript ending sites, not polymerase termination sites. The Term-seq data do not map the polymerase termination site. The authors should make this distinction in their writing. Likewise, it is better to rename transcription termination efficiency (TTE) to transcript termination frequency, because they are measuring steady state RNAs which embody both 3' end cleavage efficiency and RNA stability. The efficiency implies termination kinetics, which is not what they are measuring.

Thank you for the advice. To deliver the information more clearly, description of Term-seq method has been added as “Recently, Term-seq, an approach that enables accurate mapping of all exposed RNA 3′-ends in prokaryotes and determines the transcription termination sites (TTSs) at the genome-wide level in representative bacteria and archaea (Dar et al., 2016b; Porrua et al., 2016; Yue et al., 2020), has been developed. ” (Lines 82-85). In addition, “transcription termination efficiency (TTE)” has been revised as “Transcription termination efficacy” throughout the manuscript according to the suggestion.

1. Figure 3 needs better quantification. Binding curves showing binding constant are needed to make quantitative conclusions.

Equilibrium dissociation constants (Kd) were calculated based on the binding curves, which is generated by quantifying the unbound and bound substrates in Figures 3, 4, and 6A, and the Kd values are indicated in respective figures.

1. Figure 7. the eukaryotic model seems based on budding yeast. This should be noted. 3' end motifs in other eukaryotes are quite different than those in the budding yeast.

Thank you for the point. Budding yeast has been indicated in Figure 7 legend (Lines 1200).

1. There are numerous grammatical errors, which the authors should address.

English language and grammar have revised by a professional English language editing company. We hope the language and grammar have been improved.

Author Response:

Reviewer #1:

This work significantly advances our understanding of the role of Strongyloides stercoralis nuclear receptor Ss-DAF-12 in determining parasite life cycle and infection outcomes. Strongyloidiasis is a facultative soil-transmitted nematode that infects hundreds of millions of humans. While infections are typically subclinical, 'hyperinfections' associated with host immunosuppression bring about severe morbidity and mortality that are not well-controlled by existing anthelmintics. DAF-12 signaling in free-living nematodes is known to promote growth and development, lifespan extension, and avoidance of the dauer dormancy state. Some of these findings have been shown to be conserved in parasitic nematodes, where infective stages are argued to be akin to dauer, motivating the study of the orthologous DAF-12 pathway as an antiparasitic substrate.

Through a lipid fractionation strategy, the authors first show that Δ7-DA is a potent endogenous ligand for Ss-DAF-12 and that Δ7-DA abundance throughout the life cycle is correlated with states of parasite reproductive development. The authors show that S. stercoralis possess the machinery to synthesize Δ7-DA from dietary cholesterol and use an insect cell expression system to identify biosynthetic enzymes in this pathway: Ss-DAF-35, Ss-DHS-16, and the cytochrome P450 Ss-CYP-22a9. CRISPR/Cas9-mediated disruption of Ss-CYP-22a9 is shown to near-completely inhibit cGMP-mediated Δ7-DA production and activation of infective larvae as measured by feeding, which can be rescued by exogenous Δ7-DA. Finally, a gerbil infection model is used to show that Δ7-DA treatment drastically reduces fecal output of larvae in uncomplicated strongyloidiasis and that a combined Δ7-DA/ivermectin regimen can decrease the burden of autoinfective intestinal larvae and prevent death in hyperinfection induced by immune compromise.

Overall, the conclusions are well-supported and this work provides important new insights on a pathway that is of broad relevance to the regulation of the complex and diverse life cycles of parasitic nematodes. The discovery that the combinatorial action of a DAF-12 agonist and ivermectin can synergistically control hyperinfective strongyloidiases is a major and impactful finding. This work will be of great interest to the larger parasitology and nematode biology community. My enthusiasm is only slightly tempered by the acknowledged caveats that currently limit the therapeutic outlook of this approach. The eventual development of therapeutics targeting this pathway could aid the treatment of uncontrolled strongyloides infections and be of potential value for the treatment and control of other parasitic worm infections.

Strengths:

• Experiments are generally well-designed and rigorous, clearly establishing that Δ7-DA is a primary ligand for Ss-DAF-12 and resolving the primary biosynthetic pathway for the production of Δ7-DA in S. stercoralis. While Δ7-DA is the known ligand of C. elegans DAF-12, significant difference in primary sequence justified caution and experimental validation. Similarly, while two of the Δ7-DA biosynthetic pathway enzymes have one-to-one C. elegans orthologs, the role of Ss-CYP-22a9 as the DAF-9 isoenzyme could not have been bioinformatically inferred.

• CRISPR-based knockdowns are not trivial in this system and both the HDR and NHEJ pathways were elegantly leveraged to confirm the in vitro activity of Ss-CYP-22a9 and its essentiality to the life stage responsible for infection.

• Animal studies convincingly reveal the synergistic effect of Δ7-DA and ivermectin in disseminated strongyloidiases. The burdens of intestinal larvae and adults in both uncomplicated and disseminated infection after treatment align with the standing model that Δ7-DA is acting against the intestinal larval stages (L3a) that are naturally deficient in the hormone.

• While there is precedent for combinatorial drug therapies in antifilarial control, there is great novelty in combining drugs that are known to target different stages as opposed to just different molecular targets. This provides the first clear demonstration of this as far as parasitic nematodes are concerned.

Thank you for your enthusiastic and supportive comments.

Weaknesses: Weaknesses are categorically minor.

• As the authors recognize, the animal studies required daily Δ7-DA dosing over a two-week period. While some of this is explained by a short half-life and poor pharmacokinetics, drug was continually delivered directly to the gut at high (uM) concentrations. This is a major hurdle to surmount and it is entirely possible that even if drugs with equivalent potency and more favorable pharmacokinetics were discovered, they would have to be administered in multiple dosing regimens or at prohibitively high concentrations to achieve a curative effect.

We did address this potential limitation in the second to the last paragraph of the Discussion and believe that it should not be an insurmountable task to develop such drugs. The repeated dosing that we used was necessary because the endogenous ligand has poor pharmacokinetic properties, which is a relatively common characteristic of other endogenous nuclear receptor ligands as well. Notably, however, this hurdle has been overcome successfully with the design of potent, long-acting agonists and antagonists (the depot drugs used in reproductive medicine are great examples of this strategy).

• While there is some evidence in other nematode clades that modulation of the DAF-12 pathway can affect developmental phenotypes, many of these parasites have huge phylogenetic separation from strongyloides and lack dormant stages requiring 'activation' or free-living stages. Given the independent evolution of parasitism across the phylum, it is just as likely that drugs acting on DAF-12 will have subtle (and not curative) effects in these other parasite systems.

This will be an important point to address with other parasites. However, we note that a large number of them (including all of the soil-borne nematode parasites that have been surveyed) have a DAF-12 and a similar L3i stage. In fact, the so-called “rule of the infective third stage” is a paradigm that is broadly conserved throughout the nematode phylum.

Reviewer #2:

Mangelsdorf, Kliewer and colleagues here identified the endogenous ligand in Strongyloides stercoralis that governs that parasitic nematode's capacity to autoinfect its mammalian hosts. The ligand, [Delta]7-DA, interacts with nuclear receptor Ss-DAF-12, just as it does with its C. elegans ortholog, Ce-DAF-12, governing transcription of genes essential for metabolism and reproductive growth. Specifically, [Delta]7-DA appears to mediate a switch in DAF-12 function: in unfavorable conditions, [Delta]7-DA is absent and unliganded DAF-12 is said to arrest growth in both species and produces developmentally quiescent infective S. stercoralis larvae; in favorable conditions, the ligand is synthesized and the liganded DAF-12 triggers infection by S. stercoralis and subsequent development and reproduction in both species. The authors determined the ligand's biosynthetic pathway and showed by mutating the rate-limiting enzyme in the pathway that [Delta]7-DA is essential for parasite reproduction, whereas its absence is required for infectious larval development. In an animal model, they demonstrated that administration of [Delta]7-DA suppresses autoinfection and host lethality. Given in combination with an existing drug targeted to actively developing stages, [Delta]7-DA virtually cures the disease.

This work establishes a finding and an implication. The finding: gene circuits for growth and reproduction in nematode species with distinct life cycles -- parasitic vs free-living -- are regulated by a hormonal signal and cognate receptor that are structurally and functionally conserved. This is evolutionarily unremarkable, made mildly surprising because S. stercoralis lacks a cytochrome P450 with strong sequence identity to C. elegans DAF-9, which catalyzes the rate-limiting step in [Delta]7-DA synthesis in C. elegans; a screen of S. stercoralis P-450s demonstrated that Ss-CYP22a9 is the DAF-9 isozyme. The implication: targeting DAF-12 function (either agonism to block lethal hyperinfection or antagonism to prevent development of adult worms) or ligand synthesis may offer a therapeutic route to treating nematode parasitism. This is a valuable implication, identifying three therapeutic target approaches -- Ss-DAF-12 agonism or antagonism, and Ss-CYP22a9 inhibition -- that potentially might be be advanced from these pre-clinical observations. Overall, the manuscript makes a modest yet significant contribution.

Typical of the work from Mangelsdorf and Kliewer, the research plan and experiments are rigorously designed, executed and interpreted. My only quibble is that the requirement for Ss-DAF-12 (unliganded) to produce infectious L3i larvae is claimed (lines 165-167) but not directly demonstrated here. Instead, the authors depend on, and eventually cite in the Discussion (line 340) their nice PNAS paper earlier this year, which makes this case. Because of its importance in rounding out the implication noted above, my preference would be for the authors to add an experiment to this work that documents that Ss-DAF-12 is essential both pre- and post-ligand production.

Thank you for your positive comments and enthusiasm for our work. Regarding your comment on the requirement of Ss-DAF-12 for L3i, in the text we have updated the citation of our previous published results to support our finding that unliganded Ss-DAF-12 is required for L3i formation. Importantly, we also showed this requirement using a different strategy in our present study: we demonstrated that the enzyme that makes the ligand is required for recovery from L3i and that knocking it out results in worms that develop into L3i but do not progress unless exogenous ligand is added back.

Regarding the preference for an experiment that details Ss-DAF-12’s requirement for pre-ligand production, this was shown in our 2021 PNAS paper by Cheong et al. using the Ss-DAF-12 loss of function worms. For post-ligand production, these Ss-DAF-12 KO worms could not be used for the following reason. Recall that the apo-receptor functions as a transcriptional repressor and so loss of Ss-DAF-12 results in de-repression of its targets, thereby phenocopying the presence of the ligand. Thus, adding the ligand to the Ss-DAF-12 KO has no easily discernable phenotype. However, in our PNAS paper we did knock out the essential coactivator (DIP-1) for Ss-DAF-12, which is required for post-ligand activity. This knockout eliminates Ss-DAF-12 ligand transactivation (but not transrepression) and shows the expected phenotype of not being able to recover from L3i.

Author Response:

Reviewer #1:

For this manuscript, I focused on the metabolite analysis. The data is presented as supporting a common response based on shared selective histories if I'm understanding properly. However, primary metabolite data is hard to interpret in the same fashion as genetic data. This arises because of the high degree of pleiotropy wherein it is very hard to find a mutant or variant that doesn't alter primary metabolism. As such, it is possible that there is a common response less because of shared history and more because there is constraining selection that shapes what is the optimal primary metabolite response to cold in photosynthetic organisms. For example, in Arabidopsis, it has been found that accessions tend to have a highly similar primary metabolism but when they are crossed, the progeny have a vastly wider array of primary metabolism phenotypes, suggesting that the similarity in accessions is not shared genetics but constraining selection that forces compensatory variants. I don't think this detracts from the utility of including the primary metabolism but it would help to have more clarity in the strengths and weaknesses in using metabolite data to track theories and arguments that are largely genetic based.

We fully agree with the reviewer. The idea of constraining selection is at least as interesting as our explanation, and should be in the forefront. Given this interesting idea of compensatory mutations that are private to each accession (or ‘lineage’ or ‘line’), in principle this idea also hints towards the parallel/convergent evolution (‘constraining selection’ in the reviewer’s words) of this important trait or trait complex. We re-phrased this within the manuscript and considered this comment seriously throughout. We also incorporate into our manuscript this interesting compensatory variant notion and metabolic network pleiotropy.

One difference we would like to highlight still is that in our study (compared to Arabidopsis thaliana studies) we are comparing across many different species, ploidal levels, and varying species-level evolutionary histories. This makes our experiment different from Arabidopsis thaliana ecotype experiments and crossings; but indeed the reviewer is fully right that our results may also follow a similar evolutionary path as for Arabidopsis thaliana.

Reviewer #2:

Cochlearia, and other species that have rapidly evolved new ecological niches, represent excellent systems to study adaptation to past, present, future and changing environments. Furthermore, reticulate evolution within such groups offers a natural experiment to test hypotheses about the roles of hybridization, introgression, etc. on evolutionary dynamics, including pre-adaptation. However, there are also several significant challenges to using such systems, most crucially separating adaptation as the causal mechanism from the wide array of non-adaptive processes that could also cause the observed patterns. Overall, Wolf and colleagues do a nice job describing this complex taxonomic system and provide multiple lines of inquiry into how observed patterns may align with various adaptive scenarios. Despite the strong descriptive framework, I had trouble understanding exactly how causality could be assigned. Thus, the interpretation and discussion of the results felt speculative.

Thank you for the encouraging comments. Yes, we agree: the points towards an important aspect of this kind of phylogenetic-systematic-evolutionary research, namely demonstrating causality. Honestly speaking, in such studies we are not able to show causality in its strict sense, and we think that the reviewer wants to claim this without using quite so strong wording. We considered this while re-phrasing respective paragraphs and also town down some speculative conclusion.

Reviewer #3:

There has been intense interest in how plants have responded during periods of rapid climate change in the past. Understanding these responses can increase our understanding of how plants might respond to rapidly accelerating anthropogenic climate shifts and help set conservation priorities. Many paleoecological studies have provided insight on how plants have migrated and persisted in suitable climate refugia (i.e. pockets of suitable habitat that exist even if regional climate is unfavorable for the persistence of a species) throughout glacial cycles, however there has been considerably less work that details the evolutionary dynamics of plants during these periods. This piece provides timely and valuable analyses illustrating the potential influence of pronounced climate change on the evolutionary dynamics of the genus Cochlearia.

Thank you for the encouraging comments.

The authors' use of cytogenetic analyses, organellar phylogenies, and demographic modeling allows for insights into the geographic patterns of diversity, speciation rates, and postglacial expansion scenarios of Cochlearia. Drawing unique conclusions from these different lines of evidence provides new understandings into the putative role of Pleistocene glacial cycles in driving evolutionary processes such as speciation. The study also aims to provide insight into the origins of the stated putative cold tolerance exhibited by Cochlearia by using a metabolomics approach; however, the framing and use of a single related outgroup (sister genus Ionopsidium) obfuscate the link between the results and stated conclusions.

We appreciate this point, but indeed there is no other outgroup to be used. In this study we included all (both) genera with most of its species of tribe Cochlearieae. Within a family- wide phylogenetic context this tribe is placed along a polytomy (together with not well resolved other tribes) and stem group age of Cochlearieae is of appr. 18.9 million years ago (Walden et al., 2020). Therefore, for our research question additional outgroups from other tribes will not contribute any further information, because more basal splits are then nearly 20 million years ago (Early Miocene) with no biogeographic and environmentally defined scenarios that can be compared. 16-23 million years ago most tribes of evolutionary lineage II underwent an early radiation with highest net diversification rates (Walden et al. 2020) during this time. We included some of this information into the introduction.

Specifically, regarding the approach that resulted in figure 4 which encompassed the metabolomics and related analyses, the initial climate groupings into 'climate ecotypes' would benefit from clarification and consideration of assignment methods. Typically, using the term ecotype invokes the idea of distinct forms of a species with phenotypic differences adapted to local conditions rather than groupings to those under broad climate regimes. While grouping populations according to climate origin can be useful, it is not clear how the final 9 WorldClim bioclimatic variables were selected (e.g. it is not apparent how importance of or correlations between climate variables, etc. were considered). Consequently, knowing this information would help understand the patterns in figure 4b, which seems to indicate that geographically distant populations experience very similar climate conditions (understanding that similarities can exist but variable selection can greatly influence these patterns).

Thanks for this reminder to explaining selection and analyses of BioClim variables.

As for the term “ecotype”: In plant taxonomy ecotypes are often referred to on subspecies level, in particular if environmental conditions are extremely different (e.g. heavy metal contaminated versus not-contaminated soils) and often these subspecies do not significantly differ in morphology (Noccaea caerulescens, Minuartia verna). In Cochlearia morphology is at best a morphospace which is more or less shared between all species in different ways. Species definition and taxonomy is based on a combination of largely overlapping morphospace, cytotype, ecotype and habitat types (bedrock; arctic, lowland to alpine; soil type and salt, life cycle) and distribution – often sole morphology is a bad species predictor (morphologically cryptic species – this is well-known also for some other arctic species such from the genus Draba). But the reviewer is fully right, that using the term ecotype here is somehow misleading. Our idea was to highlight that groups of taxa are combined by bioclimatic variables (and biomes or habitat types) while spanning the entire species/ecotype space of the genus – and this grouping follows also evolutionary meaningful cluster. We clarified this.

As for selection of BioClim variables: we agree, indeed selection might have appeared arbitrary to the reader. Our original selection followed our field and cultivation experiences. However, structuring into four clusters as originally shown with the first submission is robust also when including all 19 BioClim variables. The same four cluster are retained in PCA, when temperature related BioClim variables are used only.

Therefore, we added a Principal Component Analysis as starting point for Bioclim variable selection, secondly we added a PCA using temperature related BioClim variables 1-11 only. Built upon this we added a sentence why our nine selected variables were used to highlight the four groups in Fig. 4. The two PCA scree plots (including vector data) plus the correlation matrix and the results of a KMO test (Kaiser-Meyer-Olkin test: testing significant difference between the correlation matrix of variables and an identity matrix) are additionally provided with the Suppl. Material.

The other concern is in regards to the framing and interpretation of these results. For instance, in the results (lines 329-330) and discussion (lines 419-423), the impression is given that experimental results here match those found in plants belonging to a different genus (i.e. Arabidopsis). However, rather than attributing this to more generally conserved mechanisms in response to considerable cold stress, the authors relate this to the unique history of Cochlearia (and its relationship to the drought adapted sister genus). The authors also note that surprisingly there was no demarcation of cold responses between the climate-defined groupings. Detailing why this is surprising given some of the other conclusion statements would be helpful. Some targeted revision to strengthen this link would be useful to bolster the inference of about the origins of cold tolerance in Cochlearia, rather than making it seem like this result could be expected in other taxa.

Thank you for this. We agree that we did not explain our reasoning as well as we could and we now have reworked this. Original lines 329-330 simply refers to the (expected) and obvious general response to cold – some explanatory text has been added, e.g. such as at the end of the discussion and directly with the above-mentioned lines.

Lastly, another area that would benefit from some clarification and tightening is revisiting the connection between the results and stated conclusions. For instance, some of the statements from the introduction and conclusions indicate the reader might expect explicit niche exploration analyses and more detailed genomic approaches. It is not abundantly clear for a general audience how these results definitively demonstrate how genetic diversity was rescued in reticulate and polyploid gene pools or species barriers were torn down. These are very specific, strong claims that do not appear to be explicitly discussed outside of the introduction/discussion or directly related to the results presented in this manuscript.

Thank you for pointing out how this could be read in this way. We have revised this to indicate that agree: we do not think our data ‘definitively demonstrate’ (in the reviewer’s words, not our) this. We modify the text to avoid such interpretation.

This is no way diminishes the considerable effort of the authors to conduct the informative array of presented analyses, but more closely aligning the conclusions within the scope of presented results (or providing direct links on how the results provide these insights) would help increase the effectiveness of this manuscript.

Many thanks for this very encouraging note. We have worked to incorporate these thoughtful comments.

Author Response:

Reviewer #1 (Public Review):

Molecular probes that respond to disease-specific activities to produce a diagnostic readout have had a major impact in the clinical management of cancer. The current study extends the teams previous work on the development of a molecular sensor for the cancer-associated, fibroblast activation protein (FAP). The molecular sensor is based on CGRP (Calcitonin gene-related peptide), a potent vasodilator of human arteries which mediates relaxation of arteries via activation of the CGRP(1)-type receptor. The sensor is fused to biotin with a linker sequence that contains FAP cleavage sites. In its intact form, the sensor fails to activate the CGRP-receptor, however, in the presence of FAP, proteolytic release of CGRP from inhibition, leads to CGRP-receptor engagement, which is then detected by changes in MRI contrast. Receptor activation on the vasculature, provides a diagnostic readout via local changes in hemodynamic image contrast for MRI. This is a technical report that provides a proof of principle evidence that a sensor for FAP proteolytic activity can be used in rodent models, with a robust signal to noise. However, the discussion and abstract overstate the clinical impact of the findings.

We are grateful for the Reviewer’s overall assessment and have made a number of additions and edits to the manuscript to characterize and clarify the clinical potential.

Reviewer #2 (Public Review):

In this manuscript, the authors create an imaging probe for magnetic resonance imaging (MRI) that is based on triggering vasodilation in a protease-dependent manner. The imaging probe is a steric blocking domain fused to the N-terminus of a vasoactive peptide connected by a linker that is sensitive to proteolytic cleavage by fibroblast activation protein (FAP). The linker design was optimized for FAP-mediated cleavage and led to a 34-fold increase in activity when there was FAP present. The imaging probe detected cells overexpressing FAP implanted into the rat striatum when infused directly at the transplantation site or into the cerebrospinal fluid. The authors also create a kinetic model to determine FAP catalysis rate, k, from temporal MRI signal. Lastly, the authors demonstrate in a proof-of-concept experiment that the vasoactive peptide is able to create imaging contrast in a nonhuman primate brain.

This group has previously described using peptide-mediated vasodilation as a method for image contrast in MRI. In this work, they advance this concept to make the peptide activity triggered by protease cleavage, thus creating an activity-based molecular imaging probe. The design presented in this work could likely be adapted to a wide range of proteases through substitution of the substrate that links the steric blocking domain and the vasoactive peptide allowing for the study of a wide range of protease activity in diseases that affect the brain. The creation of activity-based imaging probes is an important area of study for advancing precision medicine because the imaging signal may more accurately represent disease prognosis and stratification over conventional imaging probes.

A claim made in the abstract that is provided with limited support is the whether the probe allows for quantitative analysis of FAP activity. A useful measure for a diagnostic would be whether the imaging signal can quantitate the amount of enzyme activity. In this study, all in vivo experiments were conducted with the injection of a single concentration of cells overexpressing FAP transgene.

We thank the Reviewer for this considered assessment. In this paper, we use kinetic modeling to quantify FAP activity over animals and over voxels in the experiments of Figures 2 and 3. Although these experiments followed implantation of a fixed number of FAP-expressing or control cells per animal, the three-dimensional distribution of FAP activity produced in the brain is heterogeneous, thus allowing for spatially resolved quantification of enzyme activity that addresses the Reviewer’s point. To this revision, we further add an analysis of enzymatic constants obtained on a per-voxel level from the data in Figure 4, indicating approximate linearity of the relationship between maximum signal change and the probe cleavage rate constant k. Values of k that range from 0 to 0.03 s^-1 are obtained from our analysis, corresponding to estimated localized FAP concentrations ranging from 0 to 20 nM in rat brains. These values are physically plausible and consistent with in vitro quantification of FAP activity over a range of transfected and patient-derived tumor cell numbers in data added to Figure 4 of the revised manuscript.

Author Response:

Reviewer #1 (Public Review):

In their manuscript, Urtatiz and colleagues propose that gain-of-function mutations affecting the G-alpha-q signaling pathway are not tolerated in melanocytes residing in the interfollicular epidermis because of paracrine signals from neighboring keratinocytes. This is an interesting and important hypothesis that would explain a mystery to the melanoma field - i.e. why are GNAQ/11 mutations common in uveal melanoma, among other rare subtypes, yet are exceedingly rare in cutaneous melanoma.

Specific comments on experimental work:

Previously, this group showed that forcible expression of oncogenic GNAQ during embryogenesis depletes melanocytes in the interfollicular epidermis. This paper offers an advance because TYR-cre mouse is inducible at later points in life, which also permits in vitro explant cultures to perform more in-depth studies. This is a major strength of the manuscript.

Major concerns:

The most significant issue with the experimental results is that in most of the explant cultures, the melanocytes are not proliferating. Instead, the authors are observing which melanocytes are dying slower than others. This seems a bit strange because there are countless examples of laboratories who have established healthy murine melanocytes in tissue culture, and it raises the question that there is something off with the culture conditions.

Thank you to the reviewer for their thorough analysis and for raising interesting questions.

Yes, it is true that the wildtype and GNAQ^Q209L expressing melanocytes in the survival curves (in Figure 2C, for example) did not increase above the number plated during the experiment. However, the Braf^V600E expressing cells did increase, which suggests that the culture conditions were not fundamentally off. It is important to note that the melanocytes in our study were sorted from mouse tail IFE. As far as we know, other labs have used mouse trunk skin or neonatal skin, which is populated by immature migrating melanocytes. These tissues have stem cells, for example in the hair follicle niche, and highly proliferative cell populations. We chose tail skin because it has a permanent population of IFE melanocytes.

Our goal in these studies was a short observation with minimal interference. Encouraging the melanocytes to behave in a certain way in culture could change the results of the experiment or mask the differences between wildtype and GNAQ^Q209L melanocytes resulting from the microenvironment. We don't agree that studying the rate of dying off is not applicable, when the relevant in vivo phenotype is that GNAQ^Q209L melanocytes experience gradual attrition from the IFE.

In addition, when sorting the GNAQ-mutant melanocytes, there is a selection for a subpopulation that did not die. This introduces a bias and seems, at minimum, worthy of discussion. One potential experiment to remove these doubts would be to isolate the GNAQ-mutant melanocytes prior to tamoxifen treatment and then induce the mutation formation in vitro.

If we understand this correctly, the concern is that since some melanocytes have been lost in the GNAQ^Q209L IFE by the time we do the FACS, this approach may have selected for a subpopulation of melanocytes that are more resistant to the effects of GNAQ^Q209L. Maybe the GNAQ^Q209L cells survived better than wildtype in culture on fibronectin because they were naturally more robust. How then do we explain why the GNAQ^Q209L melanocytes survived less well than wildtype when cultured with IFE? If there is a subpopulation at play, it is behaving exactly as one would expect from the mice, so we don't see how it changes the conclusions.

We have added this sentence to the manuscript: "Note, since some melanocytes have been lost in the GNAQ^Q209L IFE, this approach may have selected for melanocytes that are more resistant to the effects of GNAQ^Q209L.".

The suggested experiment is difficult to do because Cre is required to turn on both Tomato and GNAQ^Q209L. To be able to do the suggested experiment, we would have to first obtain a different transgenic mouse line that constitutively expresses a fluorescent marker in melanocytes and cross it into our mouse model. Alternatively, we could try to sort melanocytes following antibody staining against a cell surface marker. In fact, neither of these approaches is free from the possibility of selecting subpopulations that vary in expression of the markers.

Specific comments on bioinformatic work:

Major issues:

The evidence that there is selection against PLCB4 in cutaneous melanoma is weak. It is true that mutations frequently affect PLCB4, but this is equally true for a great number of genes in melanoma because of the high mutation burden in this cancer type. Following their lines of reasoning, the authors could make an equally compelling case that TTN, the gene encoding the muscle fiber titin which is the largest gene in the genome, is under selection. Unfortunately, the ratio of nonsynonymous to synonymous mutations is not supportive of the authors' argument that PLCB4 is under selection. It is somewhat bizarre that the authors' entirely disregarded synonymous mutations, but this reviewer looked them up in the TCGA study, and they are abundant. To identify genes under selection, there are much more sophisticated strategies that take into account the trinucleotide context of mutations, the ratio of nonsynonymous to synonymous mutations, and/or the ratio of exonic to intronic mutations. To be sure, the authors correctly point out these sophisticated algorithms have missed driver mutations in the past, but the missed mutations tend to be exceedingly rare, hotspot mutations. If the authors are going to make the case that loss-of-function PLCB4 mutations are under selection in melanoma, then the onus is on them to explain why the much more sophisticated strategies, previously invoked, have missed this finding, and they should employ even better methods to make their point. Unfortunately, the strategies that the authors do employ fall for the same traps that many older papers in the field stumbled upon. For example, they make the case that PLCB4 mutation are more frequent in melanomas with high mutation burdens. While this seems to denote a biological signal, it is exactly what one would expect to observe if a gene only harbored passenger mutations.

We understand the concerns of the reviewer. To explain, we were led to the question of whether the frequent non-synonymous mutations in PLCB4 could play a role in cutaneous melanoma from a logical deduction based on biological insight, which we found quite convincing:

1) We had found that oncogenic Gq inhibits melanocyte growth/survival when the cells are located in the epidermis.

2) Phospholipase C beta (PLC-Beta) is the immediate effector of heterotrimeric G protein alpha subunits of the q class, GNAQ and GNA11. It is very likely to be involved in the pathway inhibiting melanocytes in the IFE.

3) PLCB4 was already identified as a significant melanoma gene in uveal and CNS melanomas, through a recurrent hotspot mutation. Furthermore, the hotspot mutation in PLCB4 is mutually exclusive with the hotspot mutations in GNAQ or GNA11 in these cancers.

We agree that more could have been done to describe the synonymous mutations and examine the ratio of non-synonymous (missense, nonsense) mutations to synonymous mutations. These studies can be found in the manuscript and in Supplementary File 1 Table 1f, 1g and Supplementary File 3. Using the approach described in Van den Eynden and Larsson (1), which takes into account the mutational signature of melanoma, we found that the synonymous mutations were significantly less frequent than expected based on the expected ratio (p = 7.35 x 10^-4; one-tailed binomial test). This indicates that there is positive selection for the non-synonymous mutations in PLCB4 in cutaneous melanoma.

(1)Van den Eynden and Larsson (2017) Front Genet Jun 8;8:74 [https://doi.org/10.3389/fgene.2017.00074]

In addition, not included in the above analysis were 13 more mutations having to do with splicing: 7 splice region, 3 splice donor site and 3 splice acceptor site mutations in the melanomas. While reviewing the literature, we noticed that PLCB4 was already identified in Chen et al. (2) as a gene subject to positive selection for splice-site mutations (p=7.86Ex10^-4; q=5.77Ex10^-2) in a genome wide analysis of melanoma.

(2)Chen et al (2015) Mol Biol Evol. 2015 Aug;32(8):2181-5 (mentioned in the Supplement).

These findings strengthen our hypothesis. We propose that loss-of-function mutations in PLCB4 simultaneously reduce Galpha_q and Galpha_11 signaling, releasing some of the inhibitory effects of the pathway caused by the epidermal microenvironment.

As you know, due to a number of factors related to the process of mutagenesis, gene size and accessibility, different genes have different rates of mutation. PLCB4 might be a gene that is more likely to be mutated. It is the 99th most frequently mutated gene in the SKCM dataset. PLCB4 has more synonymous mutations (34 affected cases in total) than NF1 (14 cases), TP53 (0 cases) or PTEN (0 cases). Looking at the problem from our perspective, it seems that biological insight and hypothesis driven inquiry could help determine whether some of the genes with a higher inherent rate of mutation are playing a role in melanoma.

We also would like to clarify the statement that "...PLCB4 mutations are more frequent in melanomas with high mutation burdens." What we saw was that the melanomas with 2000 to 3000 mutations overall had the highest rate of PLCB4 non-synonymous mutations (Figure 7E). Melanomas with more than 3000 mutations actually had a reduced rate of non-synonymous mutations in PLCB4. It was not simply that more overall mutations equaled more mutations in PLCB4.

The Semaphorin mechanism is interesting but remains too speculative to warrant so much attention and space.

We have reduced the emphasis on semaphorin signaling.

In addition, there was too much speculation regarding signaling mechanisms in the apoptosis section of the manuscript. Generally speaking, RNA-sequencing is a powerful tool, but when there are >1000 differentially expressed genes, it is too easy to construct "in silico" mechanistic stories centered around a handful of genes. These would need be backed up with biological data, but in this case, additional mechanistic studies would go beyond the scope of the manuscript. Instead, this reviewer suggests removing these points.

We have reduced the discussion of particular genes that supported increased apoptosis in GNAQ^Q209L melanocytes in our RNAseq analysis.

Author Response:

Reviewer #1 (Public Review):

Reichardt M. et al investigated cardiac tissue of Covid-19 samples compared to other infections (influenza and coxsackie) and control. Using X-ray phase-contrast techniques, they provide interesting results on the microstructure description. For instance, the orientation of the cardiomyocytes, their degrees of anisotropy, their shapes (obtained via structure tensor analysis). They also present interesting findings thanks to the segmentation of the vascular network analysis (via deep learning method).

This paper is using state of the art techniques. The experiments use the latest development of X-ray phase-contrast techniques (at laboratory and synchrotron). Analysis are using machine learning approach. This paper will therefore serve as a reference for future analysis on cardiac tissue. Furthermore most of the tools used are publicly available.

The results presented show that X-ray imaging is providing more information than standard histology by accessing 3D information. This is illustrated for example by the 3D vasculature tree to assess intussusceptive angiogenesis.

In overall the paper is well written and giving clear background and explanation that people outside the fields can follow. The findings and conclusions of this paper are mostly well supported by data and analysis, but some aspects in the image acquisition, sample choice and data analysis need to be clarified.

We are very motivated by this positive assessment of the reviewer, regarding the methodology, the information gain by the 3d approach, and in particular by his/her judgment of the work to be a reference for future analysis.

1/ In the abstract, the authors don't mention the laboratory setup which is a big part of their results and actually the technique that could pave the way to a clinical translation of the technique. A sentence mentioning it is necessary.

We have now added the information on the laboratory data in the abstract.

2/ The authors present their results as "fully quantified". It would be nice to see how the results presented in this paper are comparable to the gold standard analysis used so far (i.e. conventional histology analysis). This technique seems to provide more or different information not accessible by 2D slices analysis as done in histology.

The structural parameters obtained by shape measure analysis introduced in this manuscript as well as the segmentation of the vasculature extend conventional 2d histological investigations by a third dimension. Figure 1 of the appendix shows the HE stain which is gold standard in clinical routine of all samples.

3/ A major drawback of the technique presented here is that it is necessary to make a biopsy punch on the initial paraffin block. It means that the original sample is destroyed (which also goes again a bit the "non-destructive" claim of the method). For the high resolution acquisition done with the Wave-Guide setup, a second biopsy punch is even done. Several questions can then be raised: How those biopsy punches have been selected, how is this representative compare to the entire samples, etc.

In general X-ray phase contrast tomography is a destruction-free imaging technique. However, in order to reduce absorption, we chose to take biopsies from the paraffin blocks. In view of the desired resolution and image quality our intention was to suppress artifacts of local (interior) tomography, and hence we did not record scans on large tissue blocks, but chose the approach of biopsy punching. Importantly, the structure of these biopsy cores is still intact, and does not suffer from cutting and staining artifacts as in conventional histology. Further, the biopsy cores taken can either be used separately for additional histological slices or reconstructed into the existing core leaving near to no trace of the procedure itself and not hampering clinical diagnoses. We agree that for future work, larger tissue pieces up to an entire organ would be an interesting option.

4/ The statistics obtained are based on sparse data with large error bars. Only 26 samples have been used. For instance, the parameters obtained for the shape of the cardiac tissue represented in a ternary diagram in Figure 5, present tendencies but it would need more statistics to clearly affirm that there is a clear difference between the groups.

At this point, the main limitation with respect to increasing the cohort is actually to obtain more samples from post mortem autopsies as this is not always possible in clinical routine. For this reason, we think that 26 samples is already a good start, in particular since the shape measure analysis described in this manuscript is intended as a proof-of-concept pipeline for future investigations of the 3d cardiac tissue structure. Each tomographic reconstruction yields a volume of approximately 8x10^9 voxels, hence there is certainly no sparsity for each patient. At the same time –we fully agree – that increasing the cohort size is the next important step. Once that the potential gain of these investigation is made clear, it will be easier to convince the medical community that this needs to be done.

5/ Concerning the sample selections, several samples have been taken for control (2 patients and 6 samples) and coming from 2 young patients while for the diseased samples only one sample per patient have been collected, on older patients. Furthermore, the majority of the patients are men. How the authors are sure for instance that the control patients didn't have another disease that could have affected the cardiac structure? Could they see any differences between gender, as it has been shown in other COVID-19 studies? Could the difference in age also have an impact?

The medical background of all patients is provided in Appendix 2 Table 1. All samples have been investigated by pathologists before we investigated the structure of the tissue using X-ray tomography. Effects of gender and age were not taken into account in this study since the focus of this manuscript is on the introduction of the shape measure analysis pipeline and more importantly the evidence of the presence of intra-luminar pillars in cardiac tissue of Covid-19 patients. However, the background information is publicly available and can be further investigated.

Reviewer #2 (Public Review):

The works aim to characterize cardiac tissues from patients which have succumbed to Covid-19. The authors studied pathological and normal tissues using microtomography scans performed at different resolution scales. Starting on the reconstructed volumes, special automatic analytical procedures were developed to extract some quantitative structural parameters about the samples themselves. This characterization method was used previously in the study of murine heart models. The main outcome of the research is that there are some well defined characteristics found in Covid tissues that are not revealed in other pathological and normal samples. The authors achieved the proposed aims and their conclusions are supported by the obtained data. The samples statistics should be further improved but it is already enough significant to validate the outcomes.

We are grateful for this precise and positive assessment.

Author Response:

Reviewer #1:

This paper uses intracellular patch-clamp recordings of hippocampal CA1 pyramidal cells in awake mice running on a cued treadmill to investigate whether dendritic plateau potentials, which can induce place fields in silent cells through Behavioral Time Scale Plasticity (BTSP), could also modify the spatial modulation of existing place cells. They report that plateau potentials can lead to the formation of a secondary place field by synaptic potentiation while reducing the primary place field by synaptic depression. As for place fields induced in silent cells, the spatial extend of this bi-directional plasticity depends on the speed of the animal during induction suggesting a fixed time course. Further analysis revealed that the sign and magnitude of Vm changes varied in a distance/time -dependent manner from the location/time of plateau induction such that Vm tended to increase at plateau location and to decrease away from the plateau in both directions adding a bidirectional property to previously described BTSP. The sign and magnitude of the plasticity also depends on the value of Vm at the time/location of plateau induction such that if Vm is more hyperpolarized than -55 mV the plasticity induces a depolarization at the plateau location and less hyperpolarization away from that location as observed in silent cells while if Vm is more depolarized than -55 mV the plasticity induces a smaller depolarization and more hyperpolarization further away. The authors then used Vm manipulations and computational modeling to show that the critical factor was not the absolute Vm level but instead the level of potentiation of activated synapses. Finally, the authors used a network model of CA1 to show that BTSP can account for over-representation at reward location within a familiar environment. Altogether, this work represents a nice combination of cutting-edge experimental work and modeling that shed new lights on cellular mechanisms allowing experience -dependent modifications of spatial maps in familiar environments.

Major comments:

1) The procedure for BTSP induction is described without much details in the text and method. According to the text from 1 to 8 laps/ stimulation were used to induce the secondary place field. Why is there such variability? I guess that sometimes one stimulation is not enough to induce the place field but this should be clearly stated in the introduction. Also how does one know if a new place field is induced? Sometimes the new and old place field strongly overlap (e.g. Fig. 1C, blue trace) and it must be difficult "by eye" to decide if a new place field was induced. For readers to get a better idea of the all process could you report the average number of laps/stimulations used to induce the secondary place field? Could you mention how many laps/stimulations were used for the example traces shown in Figure 1C. Was the success rate 100%? How many laps were recorded after induction and used to compute the average traces shown in Figure 1C? Could you please report those numbers in the text and the figure legend? Also, in case a primary place field was induced by BTSP how many laps/stimulations were used? Was it more difficult to induce a secondary place field compared to a primary one?

We have now included additional details in a new Supplementary Figure S1, and have added the following text to the Materials and Methods:

P25, L664: “Plasticity was induced in vivo by injecting current (700 pA, 300 ms) intracellularly into recorded CA1 neurons to evoke dendritic plateau potentials at the same position on the circular treadmill for multiple consecutive laps. In most cases, plateaus were evoked on five consecutive laps (Figure S1D, left). However, during some experiments, large changes in the spatial Vm ramp depolarization could be observed to develop after as few as one plateau (consistent with the observation that plasticity could be induced by a single spontaneously-occurring plateau), and so fewer induction laps were used. In other experiments, plateaus were induced on more than five consecutive laps if place field expression remained weak after the first five trials (Figure S1D, left). The source of this variability across cells/animals is not yet clear, and requires future investigation. Overall, this procedure induced changes in spatial Vm ramp depolarization in 100% of cells in which it was attempted by three investigators. In some cells, the initial place field was first induced by this procedure, and then the procedure was repeated a second or third time in the same cell with plateaus induced at different locations. In those cases, there was no systematic difference in the number of plateaus required to induce the first place field compared to subsequent fields (Figure S1D, right).”

We also now report in the legend of Figure 1 and in the Materials and Methods (P27, L716) that spatial Vm ramps before and after plasticity were computed by averaging across 10 laps.

2) Along the same line the behavior of the mice is not extensively described. However, behavior and notably running speed seems to have a major impact on the spatial extend of the plasticity. Could you plot the speed profile of mice below voltage traces for example in Figure 1B, S1I and so on. Also could you show lap by lap speed profiles superimposed for one recording session to have an idea of the variability and overall stereotypy of behavior. It appears that the place fields can span the border between laps (i.e. start before and end after the reward zone). How is it possible if animal stops at the end of a lap to get reward? Usually these stops induce a state change with reduced theta oscillations, which is less favorable to place cell coding. Is it the case that some animals do not stop at reward location? Could you give more details here?

We have now included additional details in a new Supplementary Figure S1, and have added the following text to the Materials and Methods:

P26, L679: “Since the time window for plasticity induction by BTSP extends for seconds around each plateau, and plateaus were typically evoked on multiple consecutive laps, the changes in synaptic weights induced by BTSP depended on the run behavior of the animals across all induction laps. We showed in Figure 3D that the spatial width of place fields induced by BTSP varied with the average velocity of animals across all plasticity induction laps. Another factor that contributed to the spatial width of induced fields is the proximity of the evoked plateaus to the reward site, as animals tended to stop running briefly to lick near the fixed reward site. Variability across laps in either the run velocity or the duration of pauses could pose a challenge in trying to relate spatial changes in Vm ramp depolarization to the time delay to the plateau (see below). Figure S1 shows the full run trajectories of animals during all plasticity induction laps for the five example cells shown in Figure 1. While some variability across induction laps was observed, each animal tended to run consistently at similar velocities across laps.”

Please also note that, on the circular treadmill, the place fields of presynaptic neurons in CA3 can “wrap around” the track (e.g. see presynaptic firing rates schematized in Figure 3J). In some cases, this meant that the same synapse that generated an eligibility trace and underwent plasticity before the animal stopped to lick for reward, was also activated and contributed to Vm ramp depolarization once the animal continued running, since the spatial positions were traversed contiguously. In the model, spatial CA3 inputs were considered to be silent during pauses in running.

3) The rationale behind the analysis of delta Vm against time from plateau induction shown in Figure 2E, 3 and 4 and associated supplementary figures is not clear from the text and method sections. If I understood well this analysis uses the difference between average Vm of several laps after the induction laps minus the average Vm of several laps before the induction laps but then uses the speed of the animal during the induction laps to convert this delta Vm trace in the temporal domain. But this assumes a relatively constant behavior of the animal during induction. If induction is performed over 1 or 2 laps the chance of a constant speed are probably higher than if it is performed over say 7-8 laps. If the animal slows down consistently or even stops during induction laps 6-8 but runs fast during induction laps 1-5 how does one interpret the DeltaVm over time representation? Authors should report the number of laps used for induction for the traces illustrated in Fig. 2 and the time against position traces for all individual induction laps superimposed on top of the average in Fig. 2C and delta Vm against time traces for all individual induction laps superimposed on top of the average in Fig. 2E.

We have now included additional details in a new Supplementary Figure S1, and have added the following text to the Materials and Methods:

P27, L732: “In order to relate spatial changes in Vm ramp depolarization to the time delay to a plateau (e.g. Figures 2E, 2F, 3A – 3F, 3I, 4B, 4C, 4E, 4F and 6E), we assigned to each spatial position the shortest time delay to plateau that occurred across multiple induction laps (Figure S1). This is a conservative estimate, as the shortest delay between presynaptic activity and postsynaptic plateau will generate the largest overlap between eligibility traces (ET) and instructive signals (IS), and will result in the largest changes in synaptic weight. While this method is imperfect and did discard variability in running behavior across laps, it enabled direct comparison of the time-course of BTSP across neurons. We also note that, to generate the modeling results shown in Figure 6, the full run trajectory of each animal during all induction laps, including pauses, was provided as input to the model (see details below). This resulted in good quantitative agreement between experimentally-recorded and modeled spatial Vm ramps (Figure 6D).”

4) In Figure 3D it is unclear where the PCs within-field data comes from. The n = 26 suggests that this data includes all stimulation but in most cases induction was performed by stimulating outside place field location (as shown in Figure 1D) and induction is done by stimulating always in the same position (except for 2/24 cells were there was a third induction). Could you please specify?

We acknowledge this it was confusing how data points were selected for inclusion in the category “PCs (within-field)” in Figures 3D and 4C. For each of the 26 inductions performed in cells with pre-existing place fields, cells were most depolarized at spatial bins within their place fields, but were also relatively hyperpolarized at positions outside their place fields. On this background, we induced plasticity by evoking plateau potentials at a fixed location, which was at a different distance from the initial place field in different cells, as highlighted in Figure 1D. In Figure 3D, we sought to determine if changes in Vm ramp were different at positions that were depolarized, compared to positions that were hyperpolarized. To do this, we pooled data from all 26 inductions, selecting only the spatial bins in each recording where the Vm ramp depolarization exceeded a threshold of -56 mV.

We have now revised the text to clarify this point:

P8, L184: “In Figures 3D – F, we examined this further by comparing data from initially hyperpolarized silent cells (black; n=29 inductions, see Figure S3 and Materials and Methods) to data from place cells (dark red; n=26 inductions). Place cells were on average more depolarized before plasticity than silent cells (Figure 3D), and more depression occurred in place cells compared to silent cells (Figure 3E). However, each place cell had both spatial positions where it was depolarized within its place field, and positions where it was hyperpolarized out-of-field. To determine if spatial positions that were initially depolarized were associated with larger depression, we grouped Vm ramp data from all place cells, considering only spatial bins where each cell was more depolarized than a threshold of -56 mV (light red traces labeled “PCs (within-field)” in Figures 3D – F). Indeed, more depression and less potentiation was induced in place cells at those spatial positions that were initially most depolarized (Figure 3E).”

5) In the modeling experiment (Figure 6 A) it is unclear why the dV/dt trace show no change for synapses activated before the plateau (unlike what is illustrated in Figure 5C). In my understanding the eligibility trace of these synapses shown in green allow them to be potentiated by a certain amount that depends on the overlap of their eligibility trace with the instructive signal. Maybe to facilitate understanding authors could show the post-synaptic potentials before and after plateau induction in Figure 5A.

The Reviewer pointed out that in Figure 5, changes in synaptic weight occur for inputs activated before a plateau, but this appeared to not occur in Figure 6A. This is not the case – in both Figures, an eligibility trace (ET) is generated at the time of presynaptic activity, but changes in synaptic weight do not occur until later when the plateau arrives and an instructive signal (IS) is generated. In the example shown in Figure 6A, nonzero changes in synaptic weight occur for all presynaptic inputs (each row in the bottom panel labeled dW/dt). However, these changes do not begin until after the plateau is initiated. We have revised the text to clarify this point:

P20, L498: “Note that, at inputs activated before the onset time of the plateau, changes in synaptic weight (bottom row) do not begin until after plateau onset when the instructive signal IS and the signal overlap ET*IS are nonzero.”

Author Response:

Reviewer #1 (Public Review):

This manuscript integrates conditional mouse models for TRAP, PAPERCLIP and FMRP-CLIP together with compartment specific profiling of mRNA in hippocampal CA1 neurons. Previously, similar approaches have been used to interrogate mRNA localization, differential regulation of 3'UTR isoforms, their local translation, and FMRP-dependent mRNA regulation. This study builds on these previous findings by combining all three approaches, together with analysis of mRNA dysregulation in Fmr1 KO neuron model of FXS. The strengths of the paper are the rich data sets and innovative integration of methods that will provide a valuable technical resource for the field. The weakness of the paper is the limited conceptual advance as well as lack of deeper mechanistic insights on FMRP biology over previous studies, although the present study validates and integrates past studies, adding some new information on 3'UTR isoforms.

We appreciate the Reviewer’s recognition that “the present study validates and integrates past studies, adding some new information on 3'UTR isoforms”. We also appreciate the Reviewer’s recognition that “The strengths of the paper are the rich data sets and innovative integration of methods that will provide a valuable technical resource for the field.”

We differ, however, with the concern that the work presents a “limited conceptual advance.” Specifically, we find, for the first time, that FMRP regulates two different biologically coherent sets of mRNAs in CA1 neuronal cell bodies and neurites. This provides a profound new insight into FMRP-RNA regulation, including the fact that these two different sets of mRNA targets (encoding chromatin-associated proteins and synaptic proteins, respectively) are both translationally regulated by FMRP and transcribed from genes implicated in autism.

We recognize that FMRP was known, by our own work and that of others (as noted by the Reviewer) to regulate specific targets “in bulk” in neuronal cell types, brain and even in CA1 neurons. What is most unexpected here? Among directly bound FMRP mRNAs in brain CA1 neurons, there is subcellular compartmentalization of this regulation. This is new for FMRP, and in fact is new for RNA binding proteins more generally (recognizing of course the extensive work on RNA localization in different compartments previously discovered by others, beginning with Rob Singer’s work on actin localization and up to the present in work on neurons).

We also think it is also important for readers to understand up-front the novelty in “combining approaches” referred to. We use cell-specific (cTag) CLIP to define direct FMRP interactions in subcompartments--dendrites vs cell bodies--of CA1 neurons within mouse brain hippocampus. We also normalize this data to ribosome-bound mRNAs in CA1 neurons, and validate observations by studying WT and FMRP-null brains. This set of complex mouse models and methods is completely new, and its application is what allowed us to make robust conclusions about FMRP translational regulation of different mRNAs in different cellular compartments.

We strongly disagree with the Reviewer’s comment that FMRP directly interacts with functional classes of mRNAs in different cellular compartments “has previously been shown in the field.” Compartment-specific FMRP-CLIP has not been reported that we’re aware of, much less in a cell-type specific manner. Our previous cell-type specific FMRP-CLIP experiments have been on bulk neuronal material (Sawicka et al. 2019; Van Driesche et al., n.d.). Although cell-type specific TRAP-seq has been performed on microdissected CA1 compartments (Ainsley et al. 2014), investigators were unable to isolate significant amounts of RNA from resting neurons, and degradation of the isolated RNAs did not allow the types of 3’UTR and alternative splicing analyses that were performed here. The Schuman group has performed extensive analysis of mRNAs from microdissected CA1 compartments (Cajigas et al. 2012a; Tushev et al. 2018), but have not performed FMRP-CLIP or any experiments using cell-type specific or direct protein-RNA regulatory methods. In vitro systems have been used to analyze mRNA localization in FMRP KO systems (i.e. (Goering et al. 2020)), but in vitro systems are unable to fully recapitulate the complexities of in vivo brain regions, and did not analyze direct RNA-protein interactions. As our work is on in vivo brain slices, is cell-type specific, and integrates TRAP-seq, PAPERCLIP and CLIP-seq datasets, we believe that our work is novel and will be of great interest to the field.

Despite the fact that FMRP targets are overrepresented in the dendritic transcriptome, it does not appear from this study that FMRP plays an active role in the mechanism of dendritic mRNA localization, at least under steady state conditions. One goal of the manuscript is to address a major question in the mRNA localization field, which is how FMRP may differentially modulate "localization" of functional classes of mRNAs such as those encoding transcriptional regulators and synaptic plasticity genes (Line 78-90). The data here indicate that FMRP directly interacts with functional classes of mRNAs in different cellular compartments, which has previously been shown in the field. However, no evidence is provided that mechanistically reveal a role for FMRP to promote subcellular localization of different functional classes of mRNAs. The correlative evidence presented in this manner does not add mechanistic insight.

We do recognize that the question of what localizes FMRP mRNA targets differentially in the dendrite (and cell body) is of great interest, and remains unanswered. We also appreciate that, despite the Reviewer’s comment above, they also recognize “it does not appear from this study that FMRP plays an active role in the mechanism of dendritic mRNA localization, at least under steady state conditions.”

We believe that some of the confusion here lies in the Reviewer’s comment “One goal of the manuscript is to address a major question in the mRNA localization field, which is how FMRP may differentially modulate "localization" of functional classes of mRNAs such as those encoding transcriptional regulators and synaptic plasticity genes (Line 78-90).” While this is a question of interest that has been studied, we think there is a major disconnect here in the Reviewer’s comments and our findings. To be clear, in the original manuscript, we did not find evidence, in WT vs KO CA1 neurons, that FMRP was acting to differentially localize mRNAs, including those mentioned by the Reviewer.

Nonetheless, to further address the issue of a possible role for FMRP in localizing the transcripts it regulates, we have now performed quantitative analysis of FMRP target mRNA localization in dendrites from WT vs. Fmr1 KO mice. These results are now presented in Supplemental Figures 9 and 10 of the manuscript, and which we present and summarize below.

Supplemental Figure 9. FMRP is not required for localization of its targets into the dendrites of CA1 neurons. A) Dendrite-enriched mRNAs were defined in FMRP KO mice (red) in the same manner as in Figure 1 for FMRP WT animals using bulk RNA-seq and TRAP-seq data. Overlap with dendrite-enriched mRNAs in WT (Figure 1, shown here in green) and CA1 FMRP targets (blue) in shown. 95.6% of dendrite-enriched FMRP targets in the WT were also found to be enriched in the dendrites of FMRP KO animals. B) Dendrite-present mRNAs were defined in FMRP KO. Overlap with dendrite-present mRNAs in WT (Figure 1) and CA1 FMRP targets is shown. 95.7% of dendrite-present FMRP targets in WT are also to be found as dendrite-present in KO animals. C-E) FISH was performed to assess FMRP target localization (Kmt2d (C) , Lrrc7 (D) and Map2 (E)) in FMRP KO mouse brain slices. Left panel shows the proportion of detected mRNAs that were detected in the neuropil (> 10 um from the predicted Cell bodies layer) in WT and KO animals. Wilcoxon ranked sum was performed to detect significance. Middle panel shows densitometry of 1000 spots samples from each picture analyzed. Distance from the CB was determined as described in methods and Figure 1. In the right panel, spots were binned into 15 groups according to the distance traveled from the CB, and the fraction of spots in each genotype in this range was analyzed by t-test to determined differences in the fraction of spots at each location in FMRP WT and KO animals (* indicates p-value < .05, ** is < .01).

Supplemental Figure 10. FMRP is not required for differential localization of 3’UTR isoforms of its targets. A) Differential 3’UTR usage was analyzed using DEXseq as described in Figure 2 to identify 3’UTRs whose ratio of usage between neuropil and CB in FMRP WT and KO animals were altered. Shown is results from DEXseq analysis showing the log2foldChange (neuropil vs cell bodies, KO vs WT) and -log10(p-value) of each 3’UTR. Gray spots indicate that all 3’UTRs analyzed have an FDR > .05, indicating no significant change in usage between FMRP KO and WT animals. B and C) FISH analysis of localization of 3’UTR isoforms of Cnksr2 (B) and Anks1b (C ) isoforms in FMRP WT and KO animals. These genes were found in Figure 2 to express 3’UTR isoforms that are differentially localized to dendrites. Sequestered isoforms are those that are significantly localized to cell bodies in FMRP WT, and Localized are those that are significantly used in the dendrites of WT CA1 neurons. Left panel, the fraction of spots that are found to be localized to the neuropil (> 10 um from the cell body layer) are shown for each isoform in FMRP WT and KO animals. Differences were assessed by wilcoxon ranked sum tests. Middle panel, densitometry of the distance traveled from the cell bodies for a representative 1000 spots from each picture that was analyzed. Right panel, as described in Supplemental Figure 9, detected mRNAs were binned into 15 bins according to the distance traveled from the cell bodies, and differences in the fractions of spots in each bin in FMRP WT and KO slices were analyzed. Significance indicates results of t-tests (* indicates p-value < .05).

In summary, we characterized the dendritic transcriptome in FMRP KO animals, and compared it to the FMRP WT results presented in Figures 1 and 2, as suggested by the Reviewers. We find that the dendritic transcriptome of FMRP KO animals is extremely similar to that of FMRP WT animals, with ~95% of mRNAs found to be dendrite-present or dendrite-enriched in WT also being found in FMRP KO animals (Figure S9). We validated these results with FISH and found no evidence for significant disruption in the localization of FMRP targets Kmt2d (Figure S9C), Lrrc7 (Figure S9D) or Map2 (Figure S9E) to the CA1 neuropil.

To detect FMRP-dependent changes in distribution of 3’UTR isoforms of FMRP targets, we first performed global analysis of 3’UTR usage in TRAP from FMRP KO animals, using the expressed 3’UTR isoforms that were found in Figure 2. DEXseq analysis on 3’UTR expression in CA1 neuropil vs cell bodies TRAP showed no significant instances of altered 3’UTR usage ratios in FMRP KO animals (Figure S10A). We validated these results by performing FISH on the sequestered and localized 3’UTR isoforms of Cnksr2 and Anks1b genes and show no significant changes in the localization of the 3’UTR isoforms in FMRP KO animals (Figure S10B-C). Taken together, this data suggests that FMRP is not significantly involved in localization of its targets in resting CA1 neurons, but rather shows remarkable selection for localized mRNA isoforms. Instead, we find evidence that FMRP regulates the ribosome association of its targets in a compartment-specific manner by showing an increase in ribosome association of a subset of FMRP targets in the dendrites of CA1 neurons (see Figure 7E).

Besides the addition of the figures described above, we have also now made corrections to the text of the manuscript, enumerated below, to address this.

First, we have, as much as possible, reduced our emphasis throughout the manuscript on the “localization” of mRNAs and rather point out that the study seeks to characterize the differences between the regulated transcriptomes in CA1 cell bodies and dendrites. For example, for Figure 4, instead characterizing the log2FoldChange (neuropil vs CA1 cell bodies) as “dendritic localization”, we change the wording to “relative dendritic abundance” to focus on changes in the abundance of these transcripts in the dendrite vs the cell bodies. We also changed the section heading in the results that describes analysis in the FMRP KO animal from “Dysregulation of mRNA localization in FMRP KO animals” to “FMRP regulates the ribosome association of its targets in dendrites”. We believe that these changes will help to clear up this confusion for the reader.

Second, we reformatted the model in Figure 7F. The new version of the model (shown here) emphasizes the point that our study reveals compartment-specific FMRP regulation of a subset of its targets without implying a role for FMRP in the mRNA localization of these transcripts. The text of the manuscript and figure legends have been updated accordingly.

Figure 7F Distinct, compartment-specific FMRP regulation of functionally distinct subsets of mRNAs in CA1 cell bodies and dendrites. In dendrites, the absence of FMRP increases the ribosome association of its targets; this finding is consistent with a model in which FMRP inhibits ribosomal elongation and thereby translation (J. C. Darnell et al. 2011). In resting neurons, the translation of FMRP-bound mRNAs encoding synaptic regulators (FM2 and FM3 mRNAs) is repressed. When FMRP is absent, due to either genetic alteration (FMRP KO or FXS) or neuronal activity-dependent regulation (e.g. FMRP calcium-dependent dephosphorylation (Lee et al. 2011; Bear, Huber, and Warren 2004), ribosome association and translation of targets are increased. In cell bodies, FMRP binds mRNAs that encode for chromatin regulators (the FM1 cluster of FMRP targets), as well as FM2/3 mRNAs (consistent with synapses forming on the cell soma). FM1 targets show patterns of mRNA regulation similar to what our group observed in bulk CA1 neurons: FMRP target abundance is decreased in FMRP KO cells, perhaps due to loss of FMRP-mediated block of degradation of mRNAs with stalled ribosomes (Sawicka et al. 2019; R. B. Darnell 2020).

Third, we have revised the Discussion in order to more completely discuss the model above and also emphasize the finding that FMRP was not found to be involved in the localization of its mRNA targets, but rather in the regulation of the local translation of its targets in a compartment-specific manner. We further speculate on the roles of FMRP in regulation of mRNA abundance and translation in these compartments.

We hope that these changes better reflect the interpretation and novelty of our findings for both the Reviewers and the readers.

Further related to a role of FMRP in mRNA localization, a recent paper in eLife reports that FMRP RGG box promotes mRNA localization of a set of FMRP targets through G-quadruplexes (Goering et al 2020). This relevant paper needs to be cited and discussed.

We apologize for this omission, and have now cited and discussed this paper in the Results and Discussion of the manuscript. Importantly, we find that dendrite-enriched mRNAs have high GC content (see figure below, which is now Supplemental Figure 5). This complicates the discovery of potential G-quadruplexes; put another way, G-rich mRNAs will therefore be enriched when compared to not-localized mRNAs, and this is also true for C-rich mRNAs. Dendrite-enriched FMRP directly-bound CA1 neuronal targets (defined by CLIP) are actually G-poor when compared to dendrite-enriched FMRP non-targets (see new Figure S5 and below).

Supplemental Figure 5A-D: Dendrite-enriched are GC rich and dendrite-enriched FMRP targets are GC poor compared to dendrite-enriched non FMRP targets. A) Schematic of the overlap between CA1 FMRP targets and dendrite-enriched mRNAs (defined in Main Figure 1) B) GC content, as defined by percent G + C for all CA1 mRNAs, dendrite enriched mRNAs (1211), dendrite-enriched FMRP targets (413), and dendrite-enriched non-FMRP targets (798, see A). Stars indicate significance in wilcoxon rank sum tests ( is p < .05, ** is p < .0001). C) G content, as defined by percent G, D) C content, as defined by percent C.

In light of these observations, analysis of G- or C- containing motifs needs to be examined in this context. To this end, we performed the experiments suggested here, but did so by searching for the prevalence of G-quadruplexes in dendrite-enriched FMRP targets versus dendrite-enriched FMRP non-targets (Figure S5A). To do this, we used both experimentally-defined G-quadruplexes (described in (Guo and Bartel 2016), Figure S5E), as well as motifs (described in (Goering et al. 2020), Figure S5F). We include the results below, and in a new Figure S5 in the paper.

Supplemental Figure 5E-F: mRNAs containing G-quadruplexes are not enriched in dendritic FMRP targets vs dendrite-enriched non-FMRP targets. E) The percent of all CA1 mRNAs, all dendrite-enriched mRNAs, dendrite-enriched FMRP-bound targets (413), and dendrite-enriched non-FMRP targets (798) that contain experimentally-defined G-quadruplexes is plotted. Shown are the results of chi-squared analysis comparing the enrichment of G-quadruplex containing mRNAs in dendrite-enriched FMRP targets vs dendrite-enriched non-FMRP targets. F) As in E, except looking for the presence of mRNAs with G-quadruplex motifs in 3’UTRs as described in (Goering et al. 2020)

Interestingly, we found no difference in the presence of G-quadruplex motifs in the 3’UTRs of these two sets (above and new Supplemental Figure 5). For example, of 413 dendrite-enriched FMRP targets, 100 (24%) had experimentally defined G-quadruplexes in the 3’UTRs, while 159 (22.5%) dendrite-enriched non-FMRP targets had experimentally defined G-quadruplexes. These differences were not significant (by chi-square test).

Searching the 3’UTR sequences of 413 dendrite-enriched FMRP targets above for G-quadruplex motifs (as described in (Goering et al. 2020), which searched for an empirically derived specific motif: GW--G, separated by 7nt), we only found 3 instances in dendrite-enchriched FMRP-bound target mRNAs. Similarly, we found out of 798 non-FMRP targets, only a small subset (6) contained this specific motif in their 3’UTRs. These results were not significant (chi-square test).

In summary, we do not find evidence in our data of G-quadruplexes playing a role in determination of FMRP binding in CA1 dendrites. This data is now included in the results and discussed in the Discussion of the paper.

Reviewer #2 (Public Review):

The authors performed transcriptomic analyses from compartment-specific, micro-dissected hippocampal CA1 region tissue from transgenic mice. One feature that distinguishes this work from previous studies is the use of conditional knock-in of tags (GFP or HA) and tissue specific expression of the Cre recombinase to target a very specific population of pyramidal neurons in the CA1 region--as well as the combined use of TRAPseq, PAPERCLIP and FMRP-CLIP. Also, central to this work are the analysis pipelines that look at large populations of mRNA with the goal of finding features shared by those mRNA that bind FMRP.

First, they established the identity of mRNAs that are dendritically enriched or/and alternatively polyadenylated (APA) by sequencing; followed by validation of a few candidates using smFISH. Next, the APA data was filtered through the rMATS statistical program to identify alternatively spliced (AS) mRNA variants within the APA population. The authors concluded that the majority of splicing events were of the exon-skipping type with NOVA2 as the likely culprit leading to this differential localization of AS isoforms. The authors then proceeded to perform FMRP-CLIP which was analyzed against the TRAP dataset. The (413) mRNAs that were shared by the two experiments (TRAP and FMRP-CLIP) exhibited two notable features: dendrite-enrichment and longer average transcript length. More importantly, They demonstrated that FMRP can preferentially bind to an AS isoform that is enriched in dendrites. Further analyses of FMRP CLIP targets showed that they shared a significant level of genes designated by gene set enrichment analysis (GSEA) as involved in ion transport and receptor signaling and similarly for ASD-related candidate genes.

Strengths: -The combined use of tissue-specific Cre and conditional tags for RPL22, PABPC1 and FMRP help make these pull-downs highly specific and robust. -RNA sequencing approach allows for identification and comparison of populations of ribosome-, PABPC1- and FMRP-associated mRNAs. -Preferential binding of FMRP to AS or APA isoforms in dendrites is an impactful and significant finding.

Weaknesses: -A caution in interpreting comparative or differential RNA-sequencing results as some are correlative.

We appreciate this concern, and agree that RNA-seq analysis alone can be difficult to interpret. However, we feel that our unique approach of combining multiple cell-type specific approaches, including CLIP-seq and PAPERCLIP along with TRAP-seq and RNA-seq result in stronger conclusions that are supported by multiple lines of evidence.

-Validation of FMRP interaction with AS or APA isoforms or ASD candidates by smFISH-IF is lacking.

We find that smFISH-IF in the CA1 neuropil is difficult to interpret in mouse brain slices due to dense networks of processes in addition to contaminating cell types, making IF signals dense, noisy and difficult to quantitate. Although we could theoretically attempt these experiments using an in vitro cell culture model, we believe that the novelty of our work is in a) the cell-type specific nature of our analyses and in b) the fact that our analysis and validation is all performed in vivo. We do not feel confident that in vitro systems are similar enough to our in vivo system to be relevant for this work. This is due not only to differences in their transcriptomes, but also due to the limited number of synapses in vitro cells make with other neurons when compared to CA1 neurons in the brain. Instead, we validate the interactions between FMRP and AS and APA isoforms by isolating junction reads among FMRP-CLIP tags isolated in a cell-type specific manner from intact mouse brains (Figure 5). In this manner, we find direct evidence of FMRP selectively binding to dendritic mRNA isoforms in vivo.

-Although hippocampal CA1 region is an excellent site to study FMRP-RNA interactome, are there other projection systems where altered FMRP-RNA interaction may lead to greater dysfunction?

We appreciate this point and now include this in the revised Discussion.

Author Response:

Reviewer #1 (Public Review):

This study set to test the hypothesis that alleles of paired NLRs Pik-1 and Pik2 have co-evolved to prevent premature inactivation and enable strong activation in response to matching effectors. They show that co-expression of Pikm-1 and Pikp-2 allowed weaker HR in response to Avr proteins compared to co-expression of Pikm-1 and Pikm-2, and this is attributed to a single amino acid residue at 230 in Pik-2. Most interestingly, they found that co-expression of Pikp-1 and Pikm-2 led to Avr-independent HR and this is also determined by polymorphism at residue 230. This HR requires Pikm-2 P-loop and MHD domains, which are known to be required for Pik-2 function. The authors reconstructed phylogenetic tree to trace the evolutionary history of the polymorphism of residue 230. The data showed that Gly230 is the ancestral residue and Pik-2 carrying this residue is functional in working with Pik-1 for Avr-D recognition, whereas Asp230 (Pikp-2) arose from Gly230. Glu230 (Pikm-2) likely resulted from a further mutation from Asp230. The authors further provided evidence that, while both matching and mismatching alleles of Pik-1 and Pik-2 can generally interact, there seems to be a preference for matching pairs, supporting the possibility for the co-evolution of Pik-1 and Pik-2. Other interesting results include greater accumulation of Pik-2 protein associated with autoimmunity. Overall, the study provides insight into the co-evolution of paired NLRs which has important implications in hybrid necrosis. However, the work could be further strengthened if the author could address the following issues:

1) The study relies solely on transient expression in Nb plants. It will be more convincing if the authors could show whether combination of Pikp-1 and Pikm-2 in rice by either crossing or transgenics leads to autoimmunity.

2) The observation that autoimmunity caused by Pikp-1 and Pikm-2 can be strengthened by extending to additional alleles. How general is this phenomenon? Do other combinations of mismatched pairs also show autoimmunity?

Interestingly, all known sensor Pik alleles distribute in two clades according to the HMA domain: Pikp/Pikh form one clade and Pik+/Piks/Pikm form a second (Białas et al., 2021; De la Concepcion et al., 2021). Within clades, variation is very limited. For example, Pikp and Pikh differ in a single amino acid only (De la Concepcion et al., 2021)). Further, the Pikp-1 and Pikh-1 sensor NLRs are linked to a helper NLR with 100% amino acid sequence identity to Pikp-2, while Piks-1, Pikm-1 and Pik+-1 are linked to a helper identical to Pikm-2. Therefore, it is reasonable to assume that the autoimmune phenotypes shown between Pikp and Pikm in this paper are representative of what would occur in other mismatches between clades i.e. Pikh/Pikm, Pikp/Piks, etc.

We have included a paragraph in the discussion to clarify this.

3) Figure 2A, Pikp-2D230E showed stronger HR compared to Pikm-2 when co-expressed with Pikm-1 and Avrs. What about co-exressing Pikp-2D230E and Pikm-1 without Avr? Does it show autoimmunity?

The experiment co-expressing Pikp-2D230E and Pikm-1 without Avr-PikD was shown in Figure 5 of the original manuscript. The Pikm-1/Pikp-2D230E combination shows a low level of autoimmunity in some repeats, although very reduced overall when compared with Pikp-1 (Figure 5). We suggest this may be because Pikm-1 has also evolved to suppress unregulated activation by the D230E polymorphism. We are currently investigating this and, as it was also pointed by other reviewer, we have expanded our discussion to comment on it.

Why is it stronger than Pikm-2?

As we lack a full mechanistic understanding of Pik NLR activation we cannot explain why the cell death responses triggered by Pikp-2 D230E are stronger than Pikm-2 WT at this time. However, as noted in the manuscript, polymorphisms in position 434 and 627 seem to harbour a negative contribution towards cell death (Figure 2-Figure supplement 2, 3 and 4). Although is tempting to speculate that these polymorphisms harbour a potential regulatory role to tame the increased activation of D230E, we preferred to only state this and not overspeculate.

4) Figure S5A, why Pikp-2T434S showed weaker HR compared to Pikp-2 in Figure 1A? It is necessary to compare Pikp-2 and Pikp-2T434S side-by-side.

As mentioned above, we suggest that polymorphisms 434 and 627 may have a negative contribution towards cell death. Therefore, when introduced in Pikp-2, they lead to a lowered response compared to WT. Indeed, the reverse mutations in Pikm-2 increase the HR level over WT Pikm-2 (Figure 2-Figure supplement 2, 3 and 4). We appreciate the point that to directly compare Pikp-2 with the mutants ideally these experiments would be performed side-by-side, but we have not done this as our strategy was to look for mutations that altered cell death outcomes compared to Pikm-2.

5) Is the autoactivation associated with oligomerization? A blue-native gel assay would do.

We currently have no evidence for this. We continue to perform experiments to detect oligomerization in Pik NLRs using WT, inactive and autoactive mutants. However, these have proven challenging and did not yield a conclusive result to date. We continue to work towards optimising these assays.

6) It needs to be cautious to draw conclusion from the competition experiments where different ODs are compared, as these may not guarantee correlation with protein concentrations. For example, in Figure 9C, a OD of 0.1 gave stronger Pikp-2 band in co-IP compared to higher ODs.

We fully agree with this cautionary note. As stated in the manuscript, we could not obtain even inputs. Therefore, we chose to report this phenomenon, including a paragraph highlighting to readers about the limitations of this assay, and avoided drawing strong conclusions based on this assay alone.

Author Response:

Reviewer #2:

Weaknesses:

• The priority given to metagenomic protein sequences over reference genome sequences in the clustering pipeline is not sufficiently justified. Indeed, the metagenomic coding sequences are notably more likely to be fragmented due to challenges in assembly. A combined clustering of both would present a conceptually simpler and potentially less biased workflow. Likewise, the conceptual division between genomic and environmental genes belies their mutual importance in discovering unknown functions.

We explained better in the text the rationale for the different decisions we took. Briefly, by using metagenomic data instead of references as initial data, we can show the robustness of AGNOSTOS to deal with noisy and incomplete data. Most of the studies that will use our methods will use data derived from metagenomes (contigs or MAGs), and it is crucial to show that our validation and refinement steps perform as expected. Later, we added the GTDB sequences to show the capabilities of AGNOSTOS to enrich already processed data. The results of clustering both data sets together or updating the existing gene clusters will be almost identical, but by doing it in two steps, one can track the dynamics of the singletons, the stability of the gene clusters and many other interesting processes that can provide a better understanding of the data. Also one can “paint” the gene clusters by integrating other sources of data, like enzymatic sequences like we did in Dittmar et al., 2021 where we integrated CAZymes, KEGG and other data sources in the seed database used in this manuscript.

• The authors do not compare their methods to other possible ways to identify the unknown fraction. It is therefore unclear how much better than a naive approach it might be. Likewise, it is worthwhile to question the sensitive of their results to analysis parameters. As a suggestive example, in the one case where they did compare possible parameter values-the systematic selection of the inflation parameter for MCL clustering of gene clusters into super-clusters (Supplemental Figure 7-1)-the selected values resulted in distinctly different super-cluster properties compared to all other assessed parameter values. The manuscript would be strengthened by highlighting how the chosen parameters maximize sensitivity to remote homology.

We moved and expanded from the supplementary our comparison against FunkFams and show that many of the FunkFam families belong to the known coding sequence space. In addition, we expanded the section where we reanalyze the data from Salazar et al., 2019 where they used eggNOG to explore the known and unknown fractions of the OM-RGCv2. Here we show how a large proportion of the genes classified under the category [S] by eggnog-mapper correspond to our known fraction. For the remote homology searches we used the cut-offs recommended by the authors of HHblits (https://github.com/soedinglab/hh-suite/wiki#how-can-i-verify-if-a-database-match-is-homologous)

• It is not clear why super-clusters ("cluster communities") are identified within each of the cluster classifications (Known / Genomic Unknown / etc.) instead of across all four groups. Intuitively, this would present the opportunity to detect distant homology between clusters with known and unknown function.

We improved the part where we explain why we perform the identification of the gene cluster communities by category and not combining all gene clusters. Briefly, as we are dealing with the unknown, we need to have a reference to evaluate the quality of the MCL clustering. The reference is the known fraction, as we can exploit the information related to the domain architectures to fine tune the parameter selection and avoid over splitting or lumping gene clusters together. Then we use the learnt parameters to partition the other categories where we lack the domain architecture information. One can identify the relationships between known and unknown gene clusters a posteriori, combining the sequences of a gene cluster community and creating a new HMM profiles that can be used to link known with unknowns.

• It is not clear why small clusters and those with many fragmented members are removed entirely from downstream analyses, given that the inclusion of additional sequences in later steps would presumably improve the quality of these clusters by adding new representatives.

We included in the main text parts of the Supp. Note 11 to improve the explanation of how we handle singletons. Singletons or those gene clusters that didn’t pass the validation process are not removed but flagged, so the user is aware that these gene clusters might be problematic. In the manuscript, depending on the downstream analysis, we keep or remove them, it depends on the question we want to answer, the user can decide what to do. For example, for the collector curves shown in Figure 3, we use a subset of the singletons. These singletons are selected based on the information we gathered by integrating GTDB in the metagenomic dataset combined with the inferred gene abundances from the metagenomes. As explained in the methods section, we removed the singletons from the metagenomic dataset with an abundance smaller than the modal abundance of the singletons that got reclassified as good-quality clusters after integrating the GTDB data to minimize the impact of potential spurious singletons. We clarified this point in the main text to avoid confusion

• While maximizing sensitivity to remote homology is appropriate for the overarching goal of characterizing entirely unknown protein clusters, the likely decrease in specificity means that the accuracy of functional annotations and the shared function of all sequences in a cluster are suspect (as the authors are aware). It would have been interesting and valuable to extend the hierarchical clustering framework, already partially developed here, to enable both sensitive and specific annotations.

Now, we explicitly stated in the main that we don’t perform any functional annotation besides PFAM as this is not our purpose. We believe that predicting function from sequence similarity methods is not a trivial task and, in many cases, it might be wrong. Related to the last comment of the reviewer, as shown in Figure 1C, this is possible with the current implementation of AGNOSTOS, as one can use the most adequate level of resolution. Our gene clustering and gene cluster community inference are highly constrained to preserve domain architectures. With this approach we have a good balance between sensitivity and specificity, although some noise is expected as we show in Figure2C-D. To support this, we included a new table and figure in Supplementary, where we evaluated the entropy of the eggNOG annotations at the gene cluster and gene cluster community level. In both cases the entropy values are very low.

Author Response:

Reviewer #1 (Public Review):

Sherrard and colleagues investigated the dynamics of stress fibers of the follicular epithelia during the migratory stages of egg chamber development. During the migratory stages of egg chamber development the follicular epithelium forms actin-based protrusions at the leading edge followed by a parallel array of stress fibers. As the follicular epithelia migrates along the basement membrane the encapsulates the egg chamber, the egg chamber rotates. At later stages of development the follicle cells stop migrating and the stress fibers are then used to form a contractile network which needed to help make the elongated shape of the egg. Using near total internal reflection microscopy, Sherrard and colleagues show that during this migratory period these stress fibers showed tread-milling behavior (growing in the front and shrinking in the back) and persisted for longer than the time for cell to travel at least one cell length (62%). Concomitant with these treadmilling stress fibers was the appearance of adhesions at the front of these stress fibers and their disappearance at the rear. This means that rather than just observing adhesions at the termini of the stress fibers they can be found all along their lengths. These dynamics were captured by the use of fluorescently tagged adhesion proteins (Paxillin,Talin) but could also been visualized by staining for endogenous proteins (betaPS integrin). The authors go on to correlate the treadmilling stress fibers with adhesion dynamics and observed that new adhesions are pulled reward, consistent with their maturing under tension. Blocking cell migration (with CK-666 or genetically with depletion of Abi) resulted in a conversion of these migratory (modular), tread-milling stress fibers to the conical stress fibers with adhesions on either termini rather than along the length. The authors go on to demonstrate the modular migratory stress fibers are dependent on the formin, DAAM, as depletion of this formin specifically led to 30% reduction in F-actin levels without affecting other actin-based structures. Furthermore, as the migratory phase of these follicle cells subsides, DAAM expression decreases with the more canonical stress fibers being dependent on the formin Dia.

Strengths:

The claims of the manuscript are well supported by the data presented. The analyses rely on the dynamics of stress fibers being observed in the context of organism in vivo. This is in contrast to other studies where stress fiber dynamics have been limited to tissue culture cells in 2D on a single extracellular matrix protein. The finding that stress fibers a) treadmill in this context and b) form adhesions along their length and not just their termini will likely have a big impact on our understanding of cell migration and the role of stress fibers in general. This study also elegantly capitalizes on Drosophila genetics in order to identify which formins involved in stress fiber formation and dynamics and clearly shows that there is developmental shift from DAAM to Dia as the follicular epithelium stops migrating. They also show that modular stress fibers are also dependent on behavior, and when cell migration is inhibited there is shift to canonical stress fibers, thus modular stress fiber formation is both DAAM-dependent and dependent on migration.

Weaknesses:

The 30% reduction in F-actin in the DAAM amorphic allele does suggest that there is something else contributing to the formation of modular stress fibers which is not surprising but could also be more thoroughly addressed in the manuscript. While the authors went through great lengths to quantify their observations. Many of their claims are based on qualitative observations in particular the descriptions of adhesion dynamics when correlated to modular stress fiber dynamics (Figure 4) and the shift from modular stress fibers to canonical stress fibers (Figure 5). The near total internal reflection microscopy offers them an opportunity to derive rate constants to focal adhesion dynamics, and quantify the change in stress fiber type would have strengthen the manuscript.

We measured the lifetimes of both stationary and sliding adhesions and reported these data in Figure 4E. We also related these lifetimes to the distance required for the cell to travel one cell length within the text to provide the reader with additional context for these values. Although it would be possible to derive rate constants for focal adhesion assembly and disassembly, the relative lifetimes of the two adhesion types seemed to be a more pertinent piece of information for this first description of treadmilling stress fiber assembly.

Author Response:

Reviewer #1 (Public Review):

Munc13 is a key regulator of synaptic vesicle (SV) fusion that is thought to mediate SV tethering and regulate SNARE assembly. Based primarily on Munc13 crystal structures, the authors design a set of four charge reversal mutations in the C1C2B region that are predicted to affect the interaction of Munc13 with the plasma membrane (PM). Various in vivo and in vitro consequences of these mutations are studied, leading to two main conclusions: (1) an interaction between the PM and a polybasic surface of Munc13 is likely important for SV tethering, and (2) two residues in the Ca2+-binding loops of the C2B domain are important for SV fusion.

So far, so good - I think the data strongly support the two main conclusions noted above. It is less clear that these studies support (or could falsify) the main hypothesis, stated in the title, that re-orientation of membrane-bound Munc13 controls neurotransmitter release. Primed vesicles appear to exist in dynamic equilibrium between two states, one of them "loosely" primed (LS) and the other "tightly" primed (TS). Inasmuch as this simple model is correct, one could characterize the various players - SNAREs, synaptotagmin, complexin and of course Munc13 - in terms of their ability to influence the LS/TS equilibrium, perhaps in response to Ca2+ or other small molecules. This manuscript postulates that the orientation of Munc13 relative to the membrane has a major impact on the LS/TS equilibrium, with a perpendicular orientation favoring LS and a slanted orientation favoring TS.

The authors' previous structure (Xu et al., 2017) suggested that two partially-discrete faces of C1C2B, one polybasic and the other centered around the Ca2+-binding loops of C2B, are likely involved in PM binding. In that paper they hypothesized that the polybasic face would dominate in the absence of Ca2+ whereas the 'Ca2+-binding face' [not a very good name, but the authors haven't suggested a better one] would dominate in the presence of Ca2+. Binding to the PM via the polybasic face would yield a more erect or 'perpendicular' binding orientation, whereas binding to the PM by the Ca2+-binding face would yield a more tilted or 'slanted' binding orientation.

In revised manuscript we use the term DAG/Ca2+/PIP2-binding face, or Ca2+-dependent face when we discuss the effects of Ca2+ in particular.

Here the authors performed two molecular dynamics simulations, one without and one with bound Ca2+. In the Results section, they correctly point out that their findings cannot be used to support their hypothesis because, in each case, Munc13 was placed in the hypothesized orientation - perpendicular for minus Ca2+, slanted for plus Ca2+ - at the beginning of the simulation. In the Discussion however the authors argue that the MD simulations support their model. I disagree because the simulations needed to falsify the model have not yet been conducted. In addition, an opportunity was seemingly missed by not doing MD simulations on the mutants.

We have removed the sentence stating that the MD simulations support the model in the corresponding paragraph of the discussion (page 22). A meaningful analysis of the effects of the mutations would have required much longer simulations of this large system, which would take several months for each mutant in the UT Southwestern BioHPC facility or acquisition of a dedicated allocation at the Texas Advanced Computing Center.

Of the four mutations studied, two (K603E and K720E) should specifically destabilize PM binding by the polybasic face, one (K706E) should destabilize binding by the Ca2+-binding face, and one (R769E) is expected to destabilize both. Two of the mutants (K603E and R769E) in fact abrogate priming. This result, along with biochemical experiments, implicates the polybasic face in SV tethering and thus represents the main evidence supporting the first of the main conclusions (see Evaluation Summary above). However, since an unprimed vesicle does not participate in the LS/TS equilibrium, these mutants are in this respect uninformative. Only the remaining mutants, K720E and K706E, would therefore appear to have the potential of yielding information about the LS/TS equilibrium and its relationship to Munc13 orientation.

Although we understand the concern expressed by the reviewer, we do not fully agree with the last sentence. If we accept that the K603E and R769E mutations impair priming, this result implies that binding through the polybasic face occurs for WT Munc13-1. This conclusion does not demonstrate the LS/TS equilibrium, but it does support the notion that one of the proposed states exists.

Both K720E and K706E support normal priming but have opposite effects on vesicular release probability and evoked release. These results can be rationalized in terms of an LS/TS equilibrium. The K720E mutation, which selectively destabilizes binding by the polybasic face, would shift the equilibrium toward TS and thereby increase the release probability. Conversely the K706E mutation, which destabilizes binding by the Ca2+-binding face, would shift the equilibrium toward LS and thereby reduce the release probability.

However, the authors themselves cast serious doubt on this straightforward interpretation. In the case of K720E, they point out that the other 'polybasic mutant', K603E, has no effect of release probability. (I argued above that, perhaps, K603E is best viewed as uninformative about the LS/TS equilibrium owing to its strong upstream priming defect.) In the case of K706E, the authors point out that phorbol ester potentiation was similar for K706E and wild-type, suggesting to them "that the effects of the K706E mutation might not be related to the transition to slanted orientations but rather to another mechanism that directly influences fusion. For instance, the Munc13-1 C2B domain might cause membrane perturbations analogous to those that are believed to underlie the function of the Syt1 C2 domains in triggering release (Fernandez-Chacon et al., 2001; Rhee et al., 2005). It is also possible that the phenotypes caused by the K706E mutation and other mutations studied here reflect effects of Munc13-1 in more than one step leading to release, which complicates the interpretation of the data." If this is indeed the case, we are down to one mutant - K720E - that can be informative about the LS/TS equilibrium. (For the most part, I did not find the double and quadruple mutants informative, especially because each of them contains at least one mutation that strongly abrogates priming.)

We again understand the concerns expressed by the reviewer but do not agree that the K706E mutant does not provide any information on the LS/TS equilibrium. If we accept that the K706E mutation does have an effect on evoked release and that K603E has an effect on priming, these results support the notion that both proposed binding modes occur and are functionally relevant. We do agree however that this conclusion does not prove that there is an equilibrium between two primed states.

It looks like K720E is right in the center of the polybasic surface (although it's hard to tell from a single 'projection' image) so it would have been expected to impair Ca2+-independent liposome binding, and it does. However the liposome clustering effects are very weird, displaying a much broader distribution than any other experiment, an observation which the authors disregard. However, overall, I would say that the authors' K720E findings offer modest support for their overall main hypothesis. But for me it's not enough to justify making that hypothesis the title of the paper.

We agree with the reviewer that it is a stretch to include the hypothesis in the title of the paper. We have changed the title to: ‘Control of neurotransmitter release by two distinct membrane-binding faces of the Munc13-1 C1C2B region’, which emphasizes the notion that there are two functional membrane-binding faces of the Munc13-1 C1C2B region without making a specific claim on a role of two faces on presynaptic plasticity. We note that the notion that the Ca2+- and DAG-dependent face of the C1C2B region is functionally relevant was already supported by previous studies (Rhee et al. 2002; Shin et al. 2010), which we now cite in additional sentences to emphasize this point (e.g. pages 22, 23). Hence, we believe that, together with the previous data, our results strongly support the conclusion that two faces of the C1C2B region are functionally important. We still present the LS/TS model and use it often to interpret our results, but have tried to be careful throughout the manuscript to not overstate our conclusions and point out when our results are consistent with the model without concluding that they prove it.

For the most part I could not follow the discussion of figures 4 and 5. But I am struck by strong similarity between the data for K603E and K706E (comparing Fig. 4B/C to Fig. 4H/I). How can these results be reconciled with the opposite roles predicted for K603 and K706?

The normalized data obtained for K603E and K706E mutants do look similar (new Fig. 8C,I), but the absolute amplitudes are lower for the former (new Fig. 8B,H). Nevertheless, we agree that it not straightforward to interpret some of the data obtained in the repetitive stimulation experiments. To acknowledge this difficulty, we have included the following sentence at the end of the first paragraph of the corresponding section (line 416): ‘Nevertheless, interpretation of some of the data was not straightforward, and there may be alternative explanations to those offered below, which are based in part on the proposed LS-TS equilibrium.’

I'm not sure how the results of the PDBu experiments contribute to the conclusion that "two faces of the C1-C2B region are critical for Munc13-1-dependent short-term plasticity" (p. 15), since the only mutant that selectively affects one of the faces, K706E, has no impact (Fig. 6).

We have toned down the sentence at the end of the section describing the PDB data, which now reads (line 475): ‘Overall, these results show that basic residues in the Munc13-1 C1-C2B region influence the potentiation of synaptic responses by PDBu and, together with the data obtained with repetitive stimulation, they support the notion that two faces of the C1-C2B region are involved in Munc13-1-dependent short-term plasticity.’

Why are the liposome-binding assays in Fig. 7 done with C2C present - isn't that just a confounding factor? And if Ca2+-independent binding by C2C is as weak as suggested by the results in Fig. 7, how do any of the Munc13 constructs cluster liposomes in Fig. 8? (Note that, according to my reading of the methods, V-type liposomes are simply T-type liposomes without the DAG and PIP2.)

Binding of the C2C domain to liposomes is indeed weak but still can contribute to liposome clustering because multiple C1C2BMUNC2C molecules can cooperate in this activity (see Quade et al. 2019). We used C1C2BMUNC2C mutants in the binding assays because they were also employed for the liposome clustering and fusion assays, in which C1C2BMUN is much less active (see Quade et al., 2019). We agree that having the C2C domain present could be a confounding factor, but we included the binding results because the effects of the mutations did correlate, albeit qualitatively, with those of the clustering and fusion assays.

What is the basis for the claim (p. 22) that "the perpendicular orientation of Munc13-1 is expected to facilitate initiation of SNARE core complex assembly"?

The perpendicular orientation may hinder the initiation of SNARE complex assembly if Munc13-1 is located between the vesicle and the plasma membrane, but can facilitate initiation of assembly if the bridging Munc13-1 molecules are located further from the center of the vesicle-plasma membrane interface (e.g. as in Fig. 1D; see also cryo electron tomography images of Quade et al. 2019). We agree that the term ‘expected’ is too strong and now state that the perpendicular orientation ‘may facilitate initiation of SNARE complex assembly’ (line 523).

Reviewer #2 (Public Review):

In this manuscript, Rosenmund and colleagues describe new results regarding the mode of action of Munc13 in neurotransmitter release. Based on molecular dynamics simulations of Munc13 (C1C2BMun) with phospholipid membranes, the authors selected promising point mutations and comparatively investigated their functional impact with electrophysiological experiments in hippocampal neurons and with a variety of in vitro experiments (lipid binding assay, liposome clustering and fusion). The results show that specific mutations in the C1C2B-domain (also referred to as polybasic face) of Munc13 (K603E, R769E) strongly inhibit vesicle priming, a property that correlates well with their re duced ability to bind to phospholipid membranes in a calcium-independent manner.

The manuscript describes comprehensive electrophysiological and biochemical experiments that are complemented and extended by thoughtful analyses. The direct combination of electrophysiological and biochemical expertise from the Rosenmund and Rizo laboratories, respectively, represents a particular strength of this study, allowing the authors to develop new insights into the function of the Munc13 protein. A welcome (but not necessary) extension of the data presented would be the demonstration that the mutants in question (K603E, R769E) also show altered phospholipid binding in the MD simulations. In any case, the presentation of the data is clear and the authors' conclusions are convincing.

Taken together, the manuscript and the results represent a significant advance in the understanding of the molecular mechanisms underlying synaptic vesicle priming.

We thank the reviewer for the very positive evaluation of our study. As mentioned above, a meaningful analysis of the effects of the mutations would have required much longer MD simulations of this large system.

Author Response:

Reviewer #1 (Public Review):

This ms targets an interesting question, whether changes of feedforward inhibition at the DG-CA3 synapses regulate the representational capabilities of contextual fear memory at CA1 and the anterior cingulate cortex (ACC). The paper exploits a recent tool developed by the group (viral-mediated shRNA interference of Ablim3 in DG), to enhance PV+ mediated inhibition of CA3 pyramidal cells by increasing both their recruitment by DG cells and their number of contacts over postsynaptic cells. Using micro-endoscopic imaging of mice experiencing contextual fear conditioning, the authors nicely evaluate the effect of feedforward inhibitory control of CA3 outputs in the formation, stabilization and specificity of contextual fear memory representations in the CA1 and ACC. Data is relevant to understand how specific microcircuit motifs can influence representational dynamics in downstream regions. I have some methodological comments and recommendations for authors to improve their presentation and to exclude potential confounding factors.

1- Since imaging is performed in CA1 and ACC separately, the study design entails 4 groups: shNT vs shRNA which is the main experimental manipulation, plus CA1 vs ACC. While data is in general carefully presented, some analysis may require additional validation to discard whether some regional effects caused by manipulation may actually reflect group differences. This is important because there may be some differences between ACC and CA1 groups in some behavioral readout (e.g. Fig.2c; Fig.S2b) which may actually explains different effect of manipulation. Formal comparisons of behavior in ACC and CA1 shNT groups may be required to discard this effect.

We compared behavior data in the control groups across brain region to test if our calcium imaging findings are driven by differences in groups rather than virus manipulation. We did not find a significant difference for any of the data sets (see figure legend Rebuttal Figure 1 a-d for details). In general, we tried to avoid presenting the same (or part of the same) dataset in multiple figures. An alternative would be to plot all 4 groups in 1 graph and test as such but that would decrease readability in our opinion. Therefore, we are happy to provide the additional graphs and analysis but prefer not to include them in the main manuscript. (Rebuttal Figure 1a-d).

<a href= “https://imgur.com/Y4mLzrg"></a>

2- Differences of activity level (calcium rate) are examined using bins of 5 seconds for a total of 360 sec of exploratory activity. To discard motility effects an analysis is implemented using 1 sec bins. Thus, the two data samples are not commensurate. Also, an ANOVA on calcium rate is applied over uneven multiple comparisons to account for statistical effects of region x time or context x time. This is relevant for fig.1g vs 1i and Fig.S2j,l and may require correction.

We assume you mean “1 minute” and not “1 sec” here. We presented the two datasets (calcium event rate) and moving index indeed using different time bins (5 sec and 1 minute respectively). It is true that a difference in binning and therefore different sample size in one factor (time) could affect the result of the ANOVA. Rebuttal Figure 1 e-f shows the behavior comparison made in Suppl.Figure 2b in the original manuscript with a 5 second bin. A 2-Way ANOVA with repeated measurements reveals no main virus effect [Two-way repeated measures ANOVA, ACC (e): virus x time effect 0.0113; virus main effect N.S., time main effect N.S., n=5 per group; CA1 (f): virus x time effect N.S.; virus main effect N.S., time main effect N.S., n=5 shNT, n=6 shRNA]. In ACC, we find a significant interaction effect but a posthoc Sidak test did not reveal a difference between virus groups at any time point. This confirms our previous findings that differences in movement do not seem to drive the differences between virus groups.

<a href= “https://imgur.com/0x0xqqw"></a>

3- Fig.3 nicely show accurate context classification based on calcium activity from A&C contexts neurons using support-vector machine. The authors report very interesting representational effects for shNT vs shRNA manipulations. Is prediction accuracy of the SVM classifier correlated with behavioral discrimination? That would reinforce conclusions.

Thank you for raising this very interesting point and indeed, we found a positive correlation between the discrimination ratio and the accuracy of the SVM classifier (Pearson’s r, shNT: R2 = 0.5794, p= 0.0282, n=4; shRNA: R2= 0.5771, p= 0.0288 , n=4. We added these data in Figure 4 (Figure 4c) and in Rebuttal Figure 1g.

<a href= “https://imgur.com/J8mZhfJ"></a>

Regarding conclusions and physiological relevance, the authors may need to discuss why enhanced feedforward inhibition at DG-CA3 synapses is not naturally established given the beneficial effect in context discrimination.

We apologize that we did not make that aspect of our manipulation clearer in our discussion. We edited the introduction and discussion (LL 65, LL 365) to clearly convey that FFI in DG-CA3 is naturally temporarily increased following learning (Ruediger 2011, Ruediger 2012, Guo et al 2018).

Reviewer #3 (Public Review):

In this study, Twarkowski et al. aim to understand the role of a specific circuit motif, dentate gyrus (DG) to CA3 feed-forward inhibition (FFI), for memory encoding and consolidation. FFI is a ubiquitous circuit motif in the brain. As a result, providing insights on its function is an interesting and a potentially very impactful contribution to neuroscience.

To tackle this issue, the authors describe how increasing DG-CA3 FFI impacts the ensemble activity in hippocampal area CA1 and the anterior cingulate cortex (ACC) in mice undergoing a contextual fear conditioning paradigm. To selectively increase FFI onto CA3 neurons, the study uses a molecular tool (downregulation of Ablim3 using virally mediated expression of shRNA), which has been developed by the same group (Guo et al, 2018, Nature Medicine). The impact of this manipulation is assessed via chronic in vivo one-photon Ca2+ imaging of dorsal CA1 and ACC neurons on the day of fear conditioning, one day after (recent recall), and 16 days after (remote recall) the fear conditioning. During and after fear conditioning, the results show in both experimental groups (shRNA and control) various population activity changes in both CA1 and ACC. Furthermore, the study finds improved context discrimination in the shRNA group only at the remote recall timepoint. The authors' conclusion is that increasing FFI enhances the formation of learning-specific ensembles, first in CA1 and later in ACC, which is associated with an improved memory recall. The experiments presented here were very technically challenging and produced a comprehensive and valuable dataset describing the parallel ensemble activity changes in CA1 and ACC after fear conditioning, with or without increasing DG-CA3 FFI. However, a causal relationship between the manipulation of DG-CA3 FFI, the network activity changes in CA1 and ACC, and the behavioral improvement is, in my opinion, not fully demonstrated. This is for a couple of reasons:

1) The magnitude of the effect of the shRNA manipulation on the immediate downstream area CA3 remains unclear. Therefore, the findings in the downstream areas CA1 or even ACC (which is at least three synapses removed from CA3) are, in my opinion, difficult to interpret. This uncertainty includes (1) the extent of the virus injection in the dentate gyrus and the extent of subsequent changes in CA3, and (2) the effect of the manipulation on CA3 pyramidal cell activity in vivo. The original paper (Guo et al, 2018) uses in vitro voltage-clamp recordings to record EPSCs/IPSCs in CA3, but does not exclude possible compensatory changes in vivo, e.g., in the excitability of CA3 neurons, which could result from increasing FFI chronically over a few weeks. The data in Figures 1f and g seems to suggest that there are baseline activity changes in CA1, which might be caused by changes in the upstream CA3 network activity. Along the same lines, I am unsure how to interpret the comparisons between CA1 and ACC in Figure 1; within brain region comparisons are more relevant and should be shown instead.

This is a great point and was raised by all reviewers. We acknowledge the weakness of this comparison, apologize for this misstep in our analysis and have accordingly, removed this dataset from our manuscript. Instead, we performed new experiments using in vivo electrophysiology to allow for cross-region comparison of LFPs in CA1 and ACC within the same animal. We removed data from Figure 1 e-i and added new, simultaneous electrophysiological LFP recordings (Figure 5 and supplementary Figure 4 in revised manuscript).

We found an increased number of CA1 ripples that are coupled with ACC spindles (“coupled ripples”) in shRNA mice compared to control mice prior to a learning event (Figure 5c, two-tailed unpaired student’s t-test with Welch’s correction, p=0.0499, n=5) with no difference in time spend in slow-wave sleep (SWS) (supplementary Figure 4a) or total numbers of spindles or ripples (supplementary Figure 4b-c). Control mice show a learning-dependent increase in coupled ripples (Figure 5f, two-tailed paired student’s t-test, p=0.019, n=5) to a similar level as seen in shRNA mice prior to learning. No further increase is seen in shRNA mice indicating a saturation of circuit changes that cannot be further amplified following learning.

2) Several parameters are used in this study to describe the network activity in CA1 and ACC. These include the number of correlated neuron pairs, the number of neurons active in both the training context and a neutral context (so-called A-C neurons), or the event rate observed in these A-C neurons. Most of the activity changes observed do not appear specific to the shRNA group and occur also under control condition, suggesting that they are not caused by an increase in DG-CA3 FFI. It would be helpful to clarify the sequence, how increasing FFI onto CA3 is hypothesized to cause the changes in CA1 or even ACC.

We apologize for failing to make this clearer. Prior work has shown that learning increases FFI in DG-CA3 and downregulates Ablim3 in DG (Ruediger 2011, 2012, Guo et al 2018). Therefore, it is not surprising that we observe similar changes in the control (shNT) group as shRNA group.

From previous work we know that shNT mice show increased DG-CA3 FFI following learning (training day) for approximately 24 hours (Guo et al, 2018). Thus, our manipulation allows us to mimic and boost a naturally occurring learning-induced synaptic modification in an inhibitory microcircuit in DGCA3 and examine the impact on network mechanisms underlying systems consolidation. Importantly, enhanced feedforward inhibition at the DG-CA3 synapses is naturally established for several hours following a spatial learning event (see Ruediger et al, 2011, Guo et al, 2018). Leveraging a molecular tool to enhance FFI prior to learning, we were able to reveal that DG-CA3 FFI plays a role in tuning the circuit towards cross-regional long-term storage of precise neuronal representations. (see also edits in text, LL 365).

Author Response:

Reviewer #1 (Public Review):

Short chain fatty acids produced by gut microbiota interact with the short chain fatty acid receptors FFA2 and FFA3 (formerly GPR43 and GPR41). Barki and colleagues report the results of studies designed to define the roles of FFA2 and FFA3.

Using a Designer Receptor Exclusively Activated by Designer Drugs (DREADD) derived from human FFA2 modified to allow a BRET signal in the presence of agonists, 1210 compounds with structural similarity to the known agonist sorbic acid were screened in an appropriately validated assay. Reconfirmation screens identified sorbic acid and MOMBA. Assessment of 320 additional compounds identified chemicals related to MOMBA.

MOMBA did not activate human FFA2 in interaction assays. However, interaction assays could not be performed with mouse FFA2, so Gi inhibition of cAMP assays were pursued. Neither sorbic acid nor MOMBA inhibited cAMP levels using human or mouse FFA2, and the same lack of effect was seen using human and mouse FFA3. MOMBA was shown to be an orthosteric agonist of hFFA2-DREADD.

FFA2 and 3 were shown to be expressed in myenteric neurons. MOMBA increased GI transit time in hFFA2-DREADD-HA mice but not control mice. MOMBA also increased transit in animals expressing only FFA2 or only FFA3. MOMBA also increased GLP-1 release from colonic cells and tissues, an effect already reported by this group for sorbic acid. MOMBA also promoted release of PYY.

Vagal afferents in the colon were stimulated by activation of FFA3 but not FFA2. Cells from the nodose ganglion were stimulated by C3 and to a lesser extent by MOMBA. FFA2 and FFA3 also appeared to be active and functional in DRG cells.

Using wild type mice, C3 administered to the rectum activated spinal cord neurons. C3 and MOMBA activated c-Fos in hFFA2-DREADD mice. The authors conclude that there is a SCFA-gut-brain axis.

1. This paper extends findings from this group published in a very strong paper in 2019 (Nat Chem Biol 15:489-498) using knockin receptor hFFA2-DREADD mice and showing that activation of FFA2 promotes GLP-1 release, accelerates gut transit, and promotes lipolysis in adipocytes. The GLP-1 observations are confirmed here, and a new agonist for FFA2 is identified, but it is difficult to appreciate how these studies "define" a short chain fatty acid receptor gut-brain axis.

Answer: to address whether we have ‘defined’ a short chain fatty acid receptor-gut-brain axis, we have slightly altered the title to more correctly reflect the work performed. It is now ‘Chemogenetics defines the roles of short chain fatty acid receptors within the gut-brain axis’

Why did the authors choose to pursue detailed studies of the myenteric neurons, nodose ganglion and DRG?

Answer: That there is connectivity between gut and brain and therefore a ‘gut-brain axis’ is now well established. Many products of both anabolic and catabolic metabolism by the gut microbiota have been suggested to play roles in this connection. Chief amongst these are the short chain fatty acids (SCFAs). The hypothesis explored herein is that SCFAs derived from the microbiota will impact on the gut-brain axis at many levels, hence regulating physiology and function. The FFA2 and FFA3 receptors are major targets for the SCFAs and the animal models and novel ligands we have developed allow us to define explicitly whether function at key points in this axis involve activation of FFA2, FFA3, both receptors or neither, and this is the first large scale study to have undertaken such mapping at multiple different levels. This is why, among other tissues and cells types we have explored myenteric neurons, nodose ganglion (NG) and DRG. Each of these tissues showed expression of FFA2 and/or FFA3. This naturally lead to the question of whether afferents from the gut innervating DRG/NG were activated by FFA2/3 stimulation – this was the case. It was then entirely rational to assess whether FFA2/3 were functionally expressed in the DRG and NG – this would give information regarding the possibility that there is receptor expression in nerve fibers innervating the gut from the DRG/NG. Through this investigation we present a previously unappreciated profile of FFA2/3 expression and function in neurons associated with the gut.

Other mediators of this purported axis could also be involved. What is the purpose of this axis? GI motility?

Answer: Other mediators certainly will play roles and co-ordinate with the actions of the SCFAs. We had shown previously that FFA2 in the gut controls aspects of gut motility and we now show that FFA3 also plays a role. We hypothesised that the roles of SCFAs might also invove control of sensory function. The results provided herein<br> confirm this and dissect the specific roles of each of FFA2 and FFA3.

If that is the focus, there are technical issues that limit the conclusions that can be drawn.

Answer: As noted above, the work and the conclusions are not restricted to gut motility but include actions from the gut to the spinal cord.

1. A strength of the paper is the elegant and rigorous screening strategy and validation of receptor agonists.

Thank you for this comment: it was indeed a substantial task and component of the work because without rigorous assessment of the selective actions of MOMBA and TUG1907 it would have been impossible to fully dissect the specific contributions of FFA2<br> and FFA3.

1. A weakness is that expression of hFFA2-DREADD is induced by use of a whole-body Cre mouse. Given the broad distribution of FFA2, it is likely that this receptor is being activated in multiple tissues when MOMBA is administered. How can the authors be sure that observed effects after administration of agonist in drinking water are due to local expression as opposed to an effect mediated at a distant site removed from the myentery?

Answer: We can understand where the concern of the reviewer comes from but our aim was to understand the pathway in as close as possible to whole animal and physiologically normal conditions. Hence, we wanted to maintain all elements of FFA2/3 expression whilst investigating the effects of introducing ligands into the gut, which is where the endogenous SCFAs largely originate. So the question we wanted to ask was by introducing FFA2/3 ligands into the gut (thereby mimicking the release of SCFA by the microbiota) what were the physiological consequences of this at levels of the gut, enteric neurons, DRG/NG and spinal cord.

Reviewer #2 (Public Review):

Strengths. Barki et al. report extensive and rigorous studies which convincingly establish that FFA2 and FFA3 are functionally expressed in dorsal root ganglia and nodose ganglia where they signal through different G proteins and mechanisms that regulate intracellular calcium concentrations. The authors further demonstrate that activation of both FFA2 and FFA3 within the gut lumen stimulates spinal cord activity and that activation of gut FFA3 directly regulates sensory afferent neuronal firing. These data support the authors' contention that their investigations define a SCFA-gut-brain axis.

The authors have employed a number of complimentary pharmacological, genetic, cell culture, and ex vivo approaches to obtain their data. The use of these diverse methodological approaches is a key strength of the work. They have employed transgenic mice where FFA2 was replaced by an altered form of FFA2 referred to as FFA2-DREADD (Designer Receptor Exclusively Activated by Designer Drugs), which can be activated by novel ligands but not SCFAs. They have further identified through screens of chemical libraries a novel FFA2-DREADD agonist, referred to as MOMBA, for use in their investigations. Using the FFA-DREADD mice, which are also HA tagged to allow for immunologic detection, and other related transgenic lines, they have been able to establish and identify distinct roles for FFA2 and FFA3 in signaling SCFA production and presence in the gut (specifically the colon) to neuronal pathways that communicate directly with the brain. Using isolated cells, the authors further establish roles for different G proteins and mechanisms that affect cellular calcium levels. Collectively, the data obtained using these diverse experimental approaches support the existence of a SCFA-gut-brain axis.

The authors' new findings significantly extent understanding of the molecular actions of SCFA's produced in the colon through bacterial fermentation of dietary fiber. Multiple publications have identified altered SCFA levels in the gut as a significant contributor to dysregulated metabolism and metabolic disease. Often such studies fail to provide insight into the molecular basis for the observed linkages between SCFAs and altered metabolic states, only reporting the association. The new data being reported by Barki et al. provide new possibilities for understanding these associations.

Answer: We thank the reviewer for these very positive comments.

Weaknesses. The perceived weaknesses of the work are minor compared to the strengths. Although the authors provide a description of the general characteristics of the chemical libraries they screened to identify MOMBA, sparse other information is provided. This is especially true for the second screen of 320 compounds where the data provided indicate a number of compounds may be equally potent agonists. What similarities were there in these structures?

Answer: We state specifically in the text that compound 132 was 4-methoxy-3-chlorobenzoic acid and compound 235 4-methoxy-3-hydroxy-benzoic acid. These are indeed closely related to MOMBA. We did not test the other hits from the secondary screen for their broader selectivity as none of them were substantially more potent than the MOMBA-related compounds and the others did not provide the close structure-activity relationship of MOMBA, compound 132 and compound 235, so there is no useful information we can provide.

Were any of these compounds naturally occurring or resemble naturally occurring molecules?

Answer: No, they were not, although as small molecule carboxylates they can of course be considered to ‘resemble’ certain naturally occurring compounds. For example they ‘resemble’ short chain fatty acids.

The issue here is one of trying to understand whether endogenous substances exist that might have influenced experimental outcomes and/or data interpretation. This is far from being transparent.

Answer: We have no evidence to suggest that an endogenous FFA2-DREADD activator exists – in fact our data supports the opposite, that the FFA2-DREADD is a true DREADD receptor i.e. is only activated by a synthetic ligand.

In the authors' first report of the FFA2-DREADD mice (reference 15), they establish that sorbic acid is an agonist for FFA2-DREADD. Many of the studies being reported in the present manuscript are repetitive of this earlier work but using MOMBA in place of sorbic acid. The authors provide some justification for this, but this reviewer does not find these very compelling. Seemingly, sorbic acid stimulated FFA2-DREADD signaling could have been studied without the need to screen libraries and characterize MOMBA actions. This issue detracts from the overall positive feelings towards the work. Why was the identification of MOMBA needed prior to undertaking the present studies? The authors note greater activity of MOMBA over sorbic acid, but why was this important for establishing FFA2 presence and actions in signaling to the brain?

Answer: Although this is a good point one of the important outcomes of this study is that we have generated a novel FFA2-DREADD agonist, and in so doing further validated the FFA2-DREADD model – i.e. it is not dependent on just one ligand:receptor pairing but in fact we are now able to employ two chemically distinct ligands. Interestingly,<br> reviewer 2 found the identification of a novel FFA2-DREADD ligand an important component of the paper.

Author Response:

Reviewer #2 (Public Review):

In this manuscript, Markello and colleagues exhaustively characterize the impact and relative importance of the many data-processing decisions that go into constructing whole-brain transcriptomic maps from microarray data in the Allen Human Brain Atlas. The authors motivate the need for and have developed an open-source toolbox, abagen, for standardizing workflows in imaging transcriptomics. The authors propose a taxonomy of analyses commonly performed on these data in the literature; they then use abagen to compute the distributions of statistical outcomes for three prototypical analyses across 750,000 combinatorial choices of end-to-end data-processing pipelines. Informed by these findings, the authors then place into context several specific pipelines reported in recent and influential studies.

The paper is well-written and the authors are successful in illustrating and attempting to address the need for standardized and systematic research in the burgeoning field of imaging transcriptomics. The abagen toolbox is an important contribution and is to my knowledge the current state-of-the-art. The code is clean, flexible, and very well-documented. The chief weakness of this paper is the lack of clear guidance on best practices. Readers should, however, be sympathetic to the fact that there is currently a lack of ground-truth data against which to benchmark different data-processing pipelines.

Even after reading the paper thoroughly, it's still not completely clear to me whether the analyses in this study are performed for cortex only, or at the whole-brain level (or bi- or uni-laterally for that matter). I'm assuming this study is cortex-only as you say in the methods that "the brain atlas used in the current manuscript represents only cortical parcels." But abagen supports joint cortical+subcortical atlases too. It'd be helpful to readers to make this explicit.

To ensure comparability across both the volumetric and surface-based versions of the Desikan-Killiany parcellation examined in our analyses, we investigated bilateral cortical samples (i.e., we omitted samples from the cerebellum, subcortex, and brainstem). We have clarified this in the manuscript (“Materials and Methods” section, “Data” subsection, “Parcellations” subsubsection):

“To facilitate comparison between volumetric- and surface-based parcellations, samples from the cerebellum, subcortex and brainstem were omitted.”

Along similar lines, do you expect any of the main findings of this study to change when deriving whole-brain maps?

We anticipate that examining whole-brain gene expression—rather than just cortical expression as in the current manuscript—would likely strengthen the primary findings of our analyses for several reasons. Primarily, there are known differences in gene expression values between cortical and subcortical / brainstem / cerebellar tissue samples in the AHBA (Arnatkevic̆ iūtė et al., 2019). We expect that differentially normalizing these samples across pipelines would therefore result in greater differences between effect estimates for the three examined analyses. In a similar vein, we expect that the rankings of parameter importance would likely remain stable, especially at the extremes. It is possible that some parameters related to normalization (e.g., normalize matched, normalize structures) may move up in rankings; however, overall, the qualitative interpretation of these results is likely to remain unchanged.

We have revised the Discussion to highlight this consideration (paragraph #4):

"Although we only considered cortical tissue samples in the current analyses, we expect that including non-cortical samples would further reinforce these results (Arnatkevic̆ iūtė et al., 2019) as known differences in microarray expression values between cortex and subcortical structures will likely emphasize the impact of different normalization procedures across pipelines."

Arnatkevic̆ iūtė, A., Fulcher, B. D., & Fornito, A. (2019). A practical guide to linking brain-wide gene expression and neuroimaging data. Neuroimage, 189, 353-367.

Would it make sense to use PET maps or another type of neuroimaging data as a (pseudo-)benchmark in a future study?

This is a great question and an area of ongoing research, including in our own group. The few studies that have compared PET data with the AHBA have shown that the spatial correlation between gene expression and receptor density is highly variable, with correspondence strongly dependent on the genes and receptors being considered (Beliveau et al., 2017, Martins et al., 2021). This is likely due to the fact that gene expression (as measured by mRNA) is not equivalent to protein synthesis and that PET tracers vary in their specificity and sensitivity for specific receptors. Our group is currently collating a large sample of PET datasets from multiple tracers to demonstrate this lack of correspondence (work forthcoming; presented by Hansen et al., 2021). Given this, we would be hesitant to suggest such a comparison as a benchmark.

Comparisons of microarray expression data with the RNAseq data also included in the AHBA (as performed in Arnatkeviciute et al., 2019) are also feasible; however, given that some of the pipelines in the current manuscript utilize the RNAseq data to determine probe selection we felt that using this as a benchmark would be biased. Alternatively, a different dataset (e.g., PsychENCODE) could be used; unfortunately, in these datasets the precise spatial location of collected samples are uncertain, and for that reason we would also hesitate to use them as a reference.

Martins, D., Giacomel, A., Williams, S. C., Turkheimer, F. E., Dipasquale, O., Veronese, M., & PET templates working group. (2021). Imaging transcriptomics: Convergent cellular, transcriptomic, and molecular neuroimaging signatures in the healthy adult human brain. bioRxiv. Beliveau, V., Ganz, M., Feng, L., Ozenne, B., Højgaard, L., Fisher, P. M., ... & Knudsen, G. M. (2017). A high-resolution in vivo atlas of the human brain's serotonin system. Journal of Neuroscience, 37(1), 120-128. Hansen, J. Y., Markello, R. D., Palomero-Gallagher, N., Dagher, A., Misic, B. (2021). Correspondence between gene expression and neurotransmitter receptor and transporter density in the human cortex. In 13th International Symposium of Functional Neuroreceptor Mapping of the Living Brain.

What about a cross-validation strategy where data are selectively withheld during processing and then predicted after the fact? This may only be possible for a subset of genes and/or pipelines, but it could nonetheless be informative.

A cross-validation strategy is feasible; however, it will depend on what exactly you are trying to assess. What features are being omitted (i.e., samples or genes) will be strongly influenced by the research question and null hypothesis being tested. For example, when examining the distance-dependent relationship of correlated gene expression, you could leave some tissue samples out and "predict" the fit of these samples (e.g., as in Hansen et al., 2021). As the reviewer suggests, a cross-validation strategy will thus only be possible for some specific research questions, but not generally for entire pipelines.

One alternative that would be applicable in many cases would be to examine the robustness of the observed effects via a leave-one-donor-out strategy, whereby analyses are repeated six times, omitting one donor each time, to ensure that none of the donors are unduly influencing analytic estimates (Vogel et al., 2020; Arnatkevic̆ iūtė et al., 2019). This may require careful interpretation, however, as different donors contribute variable numbers of samples, and so gene expression estimates will have variable spatial coverage across folds.

We have added the following text to the Discussion to expand on these points (paragraph #9):

"One potential solution to this could be to examine the robustness of pipelines based on a leave-one-donor-out strategy (e.g., Vogel et al., 2020; Arnatkevic̆ iūtė et al., 2019), wherein analyses are repeated six times, omitting one donor each time, to ensure that none of the donors are unduly influencing analytic estimates. This approach is likely to become more useful as data from more individuals becomes available, but at present may be a worthwhile approach for assessing whether chosen processing parameters are appropriate."

In the discussion, you claim that "the optimal set of processing parameters will very likely vary based on research question." I'd like to see this elaborated on a bit further, at least for the most important parameters. For example, when would it make more sense to use one form of gene normalization over the other? What are the implicit assumptions underlying each choice?

This is an important aspect of processing for the AHBA data: not only do we believe that the optimal set of processing parameters will vary based on research questions, but which processing parameters are most important may also be influenced.

Gene normalization is a great example. Some genes have very low expression values whereas others have very high expression, and this variability can influence downstream analysis. For example, consider the distance dependent correlated gene expression (CGE) analysis shown in the manuscript: CGE values derived from non-normalized gene expression values will be high because the correlation will be driven by these differences in expression levels across genes rather than common patterns of expression. Normalizing expression values will therefore result in CGE values being more broadly distributed and better capturing shared spatial expression patterns.

More generally, gene-expression values in the AHBA are imperfect; it is an open problem in transcriptomics to obtain measures of expression that are comparable across genes. Throughout the literature, research has shown that the binding strength of in situ hybridization depends on properties of the RNA sequence used in the binding process, making it difficult to compare "raw" values across different genes. As such, gene normalization allows for a more fair comparison of expression patterns across probes.

However, even if we were able to obtain perfect measurements that were comparable across genes, there are contexts where researchers may want to retain the variance contributed by genes to accurately reflect their relative expression levels. For example, since many genes measured in the AHBA are not brain-specific, normalization will amplify their noisy expression patterns, potentially obscuring more relevant expression information. This can be avoided by sub-selecting genes in a hypothesis-driven manner, but, as before, this will depend on the research question.

Within the forms of gene normalization examined (i.e., z-scoring, scaled robust sigmoid normalization), we believe that scaled robust sigmoid is the optimal choice as it is less sensitive to outliers, which are known to exist in the imaging microarray-based transcriptomics data (Fulcher et al., 2019; Arnatkevic̆ iūtė et al., 2019).

We have added text to the Discussion to expand on these points (paragraph #9):

“For instance, in most applications gene normalization is appropriate, as it ensures that downstream analyses are not driven by a small subset of highly expressed genes. However, in other applications it may be desirable to retain the variance contributed by genes to accurately reflect their relative expression levels. For example, many genes in AHBA are not brain-specific, so normalization will amplify their expression patterns, potentially obscuring more relevant expression information. This can be avoided by sub-selecting genes in a hypothesis-driven manner and skipping the normalization step altogether.”

Is there anything to be said about the order of operations? There seem to be several steps in Table 1 which could conceivably be interchanged. If nothing else, this procedural ambiguity is yet another good reason to standardize workflows.

We believe that the importance of processing order is strongly dependent on which processing steps are being considered. For example, intensity-based filtering of probes must always be performed before probe selection—reversing the order of these operations would, in the majority of cases, be problematic because it would potentially result in the selection of noisy probes to be carried through to analysis. However, the order of other steps (i.e., sample versus gene normalization) could arguably be reversed with no ostensible detriment. We agree with the reviewer that this ambiguity is a good reason to standardize these workflows, and believe that the order of operations implemented in abagen and described in the manuscript is a principled solution to this problem.

We have added text to the Discussion to clarify this point (paragraph #5):

“Note that there are some processing steps that should be performed in a specific sequence, and others whose order could potentially be interchanged. For example, intensity-based filtering of probes must always be performed before probe selection—reversing the order of these operations would, in the majority of cases, be problematic because it would potentially result in the selection of noisy probes to be carried through to analysis. However, the order of other steps (e.g., sample versus gene normalization) could arguably be reversed with no ostensible detriment. This procedural ambiguity is a salient example of the need to standardize workflows.”

I particularly liked the analysis in Figure 2A and thought it made a nice contribution to the paper.

We appreciate the reviewer's kind words, especially given their extensive foundational work in this field.

Reviewer #3 (Public Review):

The work Standardizing workflows in imaging transcriptomics with the Abagen toolbox is a major meta analysis pipeline workflow for comparing and integrating parameter choices in imaging transcriptomics using the Allen Human Brain Atlas (AHBA). The release of the AHBA has strongly increased the interest in determining transcriptomic associations in brain imaging studies, yet there is much variability in the analysis, methods used, and subsequent interpretation.

This work is illustrative of an important trend in informatics analysis allowing strong metadata control by users so as to access, and implement optimal choices of parameters and to study there distribution. The work implemented as an open source Python toolkit is likely to be of importance to analysts working in these areas.

It would be helpful to clarify and specifically define the term pipeline as a specific set of parameter, normalization, and other choices that are selected. Whereas this term is in common use in the field, in the present work the meaning is specific to a set of selectable options. Of course the any number of such variable selections could be implemented in the Abagen toolbox, it will help for clarity to more clearly define this term up front.

We have added text to the Results clarifying what we mean when we refer to "pipeline" (“Results” section):

“We refer to each unique set of processing choices and parameters as a “pipeline”.

Similarly, we have added text to the Methods to clarify this as well:

“Each unique set of these 17 processing choices and parameters constitutes a pipeline, yielding 746,946 unique pipelines."

My major consideration in this work concerns are two issues. The first is how to characterize and summarize the results of pipeline output produced by Abagen. The manuscript illustrates the workflows and various means of summarizing results but does not offer guidance into preferred interpretation of relative value of the results. Whereas we may argue that the primary purpose of Abagen is to run the various pipelines, allowing downstream interpretation to the user, it would be helpful to understand how the Abagen toolbox organizes, summarizes, and sets this output options up for interpretation. This appears to be only weakly addressed in the present manuscript.

The primary output of abagen is a single brain region x gene expression matrix based on a researcher-specified atlas. We believe this is the simplest and most fundamental output object of the AHBA that can facilitate a range of analyses, including those we examined in the paper (i.e., correlated gene expression, gene co-expression, and regional gene expression or gene-ofinterest analyses).

In the manuscript, we examined the outputs of various pipelines only to highlight the potential variability of results as a function of parameter selection; however, in most use cases, we would recommend that researchers only use abagen to run a single pipeline, yielding one brain region x gene expression matrix that they can carry forward to their desired analyses. Selecting different parameters when using abagen will modify the shape or values of this matrix, but not the structure.

To clarify this we have added the following text to the Results (section "Standardized processing and reporting with the abagen toolbox"):

"The main output of abagen is a single brain region (or tissue sample) x gene expression matrix. Changing the parameters may modify the shape of the matrix (e.g., different atlases will yield different numbers of regions or samples) or different values (e.g., different processing choices may yield different numbers of genes), but not the structure."

The second point I of importance I believe is more description of the available functionality in the toolbox, perhaps as more of a specific use case analysis. The authors provide substantial documentation on installing and working with Abagen, and but some more direct indication of how the toolkit would be used would be valuable.

We agree that it is important to clearly lay out the functionality of the toolbox in the manuscript. We have modified the following paragraph to the Results (Standardized processing and reporting with the abagen toolbox) to elaborate on the tools made available to researchers in abagen:

“The abagen toolbox supports two use-case driven workflows: (1) a workflow that accepts an atlas and returns a parcellated, preprocessed regional gene expression matrix (Fig. 4a); and, (2) a workflow that accepts a mask and returns preprocessed expression data for all tissue samples within the mask (Fig. 4b). Workflows can be called via a single line of code from either the command line or Python terminal, and take approximately one minute to run with default settings using the Desikan-Killiany atlas. The main output of abagen is a single brain region (or tissue sample) x gene expression matrix. Changing the parameters may modify the shape of the matrix (e.g., different atlases will yield different numbers of regions or samples) or different values (e.g., different processing choices may yield different numbers of genes), but not the structure. The outputs of these workflows can be used generally to examine the three prototypical research questions enabled by the AHBA: correlated gene expression, gene co-expression, and regional expression of genes of interest more broadly (Fornito et al., 2019, Trends Cogn Sci). Beyond its primary workflows, abagen has additional functionality for post-processing the AHBA data (e.g., removing distance-dependent effects from expression data, calculating differential stability estimates; Hawrylycz et al., 2015, Nat Neuro), and for accessing data from the companion Allen Mouse Brain Atlas (e.g., providing interfaces for querying the Allen Mouse API; https://mouse.brain-map.org/; Lein et al., 2007, Nature).”

As we envision the abagen software to continue to develop in the coming years, we have purposefully omitted the inclusion of code examples in the current manuscript as the API is liable to change over time. To ensure that these examples stay up-to-date with the abagen API, we only include code in the online abagen documentation (https://abagen.readthedocs.io; citable via Zenodo; https://zenodo.org/record/3726257), which can be continuously updated along with the software package.

Author Response:

Reviewer #1:

This study largely confirms prior observations, and the strength of the study is in its comprehensive nature rather than in shedding new insight into the effects of either FH or SDH loss. Nevertheless, there are some somewhat unexpected observations including a defect in proline synthesis, and changes in glutathione and NADPH metabolism that are interesting and incompletely explained. Some suggestions to strengthen the study include:

1) Exactly how each perturbation affects cell proliferation is not clear. This should be considered, as whether some of the differences are a result in changes in growth or proliferation rate is possible, and will affect how they normalize their data.

We thank the referee for raising a critical point and allowing us to clarify how we normalize metabolomics experiments. All the metabolomics data takes into consideration the cell number. Indeed, prior to metabolite extraction, cells are counted using a separate counting plate prepared in parallel and treated exactly like the experimental plate. In this way, differences in cell number are accounted for and the efficiency of metabolic extraction is preserved. Consumption release (CoRe) metabolomics also takes into consideration the proliferation rate during normalization since we normalise data to the final cell number. We have expanded the description in the relevant sections in the Methods.

2) It is unclear why FH loss is different than SDH loss, and it is also somewhat surprising that the effects of acute and chronic loss of either enzyme are not that different. While explaining this is too much to ask, some additional speculation might be warranted.

We postulate that FH loss is different to SDH loss for several reasons:

A. FH is localized to mitochondria (specifically in the mitochondrial matrix) and the cytosol (cytFH). cytFH can translocate to the nucleus to regulate the DNA damage response (PMID: 26237645). In contrast, the SDH complex is only localized to mitochondria. As such, a loss of FH function is likely to have mitochondrial and extramitochondrial consequences.

B. SDH is also the only TCA cycle enzyme that's physically associated with the electron transport chain (ETC) and tethered to the inner mitochondrial membrane, where it also regulates the ubiquinone pool. The different distribution of both enzymes within mitochondria is likely to influence their impact mitochondrial bioenergetics and on the overall metabolic profile.

C. A major difference between FH and SDH loss is the accumulation of fumarate. As discussed in the manuscript, fumarate is a mildly electrophilic metabolite that can succinate GSH and protein cysteine residues to form a post-translational modification termed succination. Fumarate-mediated succination is known to impair iron-sulphur cluster metabolism and perturb aconitase and Complex I function (PMID: 29069586). This is just one example of how succination can affect cellular function. In contrast, SDH loss results in a decrease of fumarate and an accumulation of succinate.Unlike fumarate, succinate is not an electrophilic compound that can modify cystine residues, and so differences between FH and SDH loss are likely owed to succination, at least in part.

D. While it hasn't been investigated in this study, succinate released from cells can bind to the succinate receptor SUCNR1, which is expressed in the kidney (PMID: 21803970). Autocrine and paracrine ligation of SUCNR1 by high levels succinate accumulation and release is likely to alter the metabolic and transcriptional landscape of the cells.

Based on these observations, we also argue that the effect of acute and chronic enzyme inhibition is expectedly different regarding how the key metabolic and signalling hallmarks of FH and SDH loss develop and interact with each other over time. This hypothesis is part of work currently undergoing in our laboratory. For example, chronic SDH loss led to a significant increase in 20 metabolites however, acute SDH inhibition with TTFA and AA5 led to an increase in 60 and 50 metabolites, respectively. Chronic FH loss also led to a significant increase in 92 metabolites, whereas acute FH inhibition led to a significant increase in 49. The fact that only 2 metabolites overlap between all conditions indicates apparent differences between the loss of both enzymes on the metabolome and whether the loss is acute or chronic in nature. There are also notable differences between chronic FH loss and acute FH inhibition in relation to reductive carboxylation. Chronic FH loss triggers a higher accumulation of fumarate and succination of aconitase that impairs reductive carboxylation (PMID: 21849978); however, acute inhibition facilitates reductive carboxylation (Figure S2), likely due to lower levels of succination given the acute treatment. As such, we feel there are notable differences in the metabolite profiles and rewiring events associated with acute versus chronic enzyme inhibition. We have discussed these important points in the discussion section of the manuscript.

3) The increase in glutathione and GSSG is interpreted as a consequence of increased oxidative stress, but that will not necessarily affect total levels.

We agree that oxidative stress will not necessarily affect total glutathione levels and this finding is likely a time-dependent phenomenon due to persistent redox signalling. In this instance, the alterations in total glutathione levels are likely linked to transcriptional and post- transcriptional changes in GSH biosynthetic enzymes and the observed metabolic reprogramming. While ATF4 regulates the glutathione redox state and glutathione levels, it's not entirely clear if it is solely responsible for increasing total glutathione levels with TCA cycle inhibition. One possibility is that there is simultaneous activation of the transcription factor NRF2, which is a crucial regulator of glutathione synthesis and is known to be regulated by FH loss and succination (PMID: 22014567) and reactive oxygen species (ROS). ATF4 and NRF2 may cooperate to transactivate glutathione-related metabolic enzymes upon TCA cycle inhibition, as previously reported in other contexts (PMID: 23618921). Further investigation of the crosstalk between these two transcription factors is warranted in this context.

4) The text in the Figure S1 PCA plots have legends is too small to read. This should be corrected.

We thank you for this note and apologize for this oversight. We have now corrected the figure legends for the PCA plots.

Author Response:

We have now revised the manuscript to address the helpful comments and criticisms from the reviewers. The revised manuscript includes additional experiments demonstrating that inclusion of Csn2/Cas9 in the in vitro assays does not suppress the disintegration activity of Cas1-Cas2 to favor integration. These additional factors do not confer strand selectivity on integration either. Furthermore, the results of integration reactions using substrates mimicking PAM-containing pre- spacers have also been added.

New figures and figure modifications at a glance:

1) The new Figure 2 shows Cas1-Cas2 reactions in a linear target site and the effects of Csn2 and/or Cas9 on proto-spacer insertion into this target (Reviewer 1).

The original Figure 2 (with slight modifications) is now moved to ’Supplementary Data’ as Figure 2-figure supplement 2, and shows proto-spacer insertion by Cas1-Cas2 into a nicked linear target site (Reviewer 2). Figure 2 is the only one in the main set of figures that has been extensively modified.

2) The new Figure 2-figure supplement 1 (under ‘Supplementary Data’) shows the effects of Csn2, Cas9 or both on proto-spacer integration-disintegration by Cas1-Cas2 when the target site is present in a supercoiled plasmid (Reviewer 1).

3) The new Figure 4-figure supplement 1 lists the sequences of the full- and half-target sites used for the reactions shown in Figure 4 (Reviewer 2).

4) The new Figure 2-figure supplement 3 shows the insertion properties of PAM-containing pre- spacer mimics in reactions with Cas1-Cas2 alone or supplemented with Csn2, Cas9 or both (Reviewer 1).

5) The new Figure 6-figure supplement 1 gives a structural perspective of the trombone substrates used for the reactions shown in Figure 6B, C (Reviewer 1).

6) The original Supplementary Figure S8 showing assays for PAM-specific cleavage by Cas1- Cas2 has been removed (Reviewer 1).

7) There are no changes in the other figures under ‘Supplementary Data’, although several have new numbers consistent with the revisions made.

Public Review (Reviewers #1 and #2):

The present work is a critical extension of the in vitro biochemical activities of the Cas1- Cas2 complex described by Wright and Doudna (Nat Struct Mol Biol, 2016; 23: 876-883). We have kept all experimental conditions nearly identical to those used by these authors to make the results from the two studies directly comparable. Importantly, we now show that the prior model for proto-spacer integration into the CRISPR locus by Cas1-Cas2 is an oversimplification of a much more nuanced mechanism.

While both reviewers recognize the importance of our findings in challenging the current thinking on the adaptation mechanism of CRISPR immunity, they express reservations as to whether the in vitro results recapitulate the in vivo mechanism of spacer acquisition. This seems to us to be too broad a criticism from which few (if any) biochemical experiments can be immune.

Our key finding is that disintegration during the second step of proto-spacer integration generates a DNA structure that has all the hallmarks of a DNA damage intermediate that the bacterial repair machinery can readily process into an authentic integration product. We invoke no new or ad hoc mechanisms, and the model we propose fits neatly into the DNA gap-filling mechanisms known to operate in DNA transposition pathways.

The proto-spacer is functionally a ‘micro-transposon’, whose shortness imposes severe torsional strain on the transposition intermediate that precedes the final integration product. In vitro experiments suggest that transcription is potentially capable of resolving this intermediate (Budhathoki et al., Nat Struct Mol Biol, 2020, 27: 489-99). In principle, replication can also accomplish this task. Our study now demonstrates that simply nicking the DNA (disintegration) is an equally effective solution for relieving the topological stress accompanying integration. DNA loose ends can then be readily tied up by the bacterial repair machinery.

We concur with the concluding sentence of reviewer 2, “The simple conclusion that Cas1- Cas2 catalyzed hydrolysis of a phosphodiester may relieve strain and allow productive transposition to occur doesn’t get emphasized enough in my opinion.” We have now expanded on this point in the revised ‘Discussion’.

Reviewer #1:

In addition, the in vitro system used here is only partially reconstituted. The substrates lack a PAM sequence, which is necessary for protospacers to be incorporated in the correct orientation and may help direct the first integration event to the L-R junction. Presumably because of this all the reactions presented do not analyze the orientation of the incorporated prespacer sequence. Cas9 and Csn2 are also absent (as are other potentially required host factors), which are necessary for correct integration in vivo.

1A. Strand specificity: The in vitro integration reactions with the Cas1-Cas2 complex were done using a protospacer of the optimal size (26 nt on each strand with the four 3’- proximal bases on each strand as unpaired). Either proto-spacer strand is equally competent to initiate the strand transfer reaction, as could be inferred from Figure 3 of the original submission. Here, reactions utilized modified proto-spacers that differed in their top and bottom strand lengths. They gave two insertion products (IP) each at the L-R (leader-repeat) and R-S (repeat-spacer) junctions of a normal target site. In modified targets in which integration was limited to just the L- R junction, two insertion products were formed. One panel of Figure 3 (which is retained in the revised manuscript) showing the four insertion products from the normal target (lane 10) and two from the modified targets (lanes 11-13) for a protospacer with 26 nt and 31 nt long strands is displayed below.

The ability of either proto-spacer strand to initiate integration is now more directly shown in Figure 2 (new) of the revised manuscript. Here the labeled top or bottom strand of the proto- spacer (PS) gave insertion products (IP) at the L-R and R-S junctions of the target site. Panel B of Figure 2 (pasted below) demonstrates this result.

1B. Cas9, Csn2 included reactions: The data for reactions containing Csn2 or Cas9 or both were not shown previously, as they did not alter Cas1-Cas2 activity by promoting strand specificity of integration or suppressing disintegration. These results are now shown in the revised Figure 2 (linear target) and the new Figure 2-figure supplement 1 (supercoiled target). Portions of these figures are shown below.

The relevant revised text describing the lack of strand specificity to proto-spacer integration by Cas1-Cas2 and the Csn2/Cas9 effects on integration is pasted below.

Page 15, lines 229-235.

"Unlike orientation-specific proto-spacer integration in vivo, Cas1- Cas2 reactions in vitro showed no strand-specificity (Figure 2B). This bias-free insertion of the top or bottom strand from the proto-spacer was unchanged by the addition of Csn2 or Cas9 or both to the reactions (Figure 2C-E). These proteins, singly or in combiantion, also failed to stabilize proto-spacer integrations in the supercoiled plasmid target (Figure 2-figure supplement 1). Instead, they inhibited plasmid relaxation. Inhibition could occur at the level of integration per se or strand rotation during integration-disintegration"

1C. PAM-containing substrates: We have now tested Cas1-Cas2 activity (with and without added Csn2 or Cas9 or both) on PAM-containing substrates that mimic ‘pre-spacers’, Figure 2- figure supplement 3 (new).

In these substrates, a proto-spacer strand of the standard length (26 nt; lacking PAM or its complement) is inserted at the L-R junction with higher efficiency than the longer strand (containing PAM or its complement). Following the first integration at L-R, the pre-spacer mimics containing > 26 nt in one strand or both strands are inhibited in the second strand transfer to the R-S junction. A portion of Figure 2-figure supplement 3 illustrating theses points is shown below.

The revised ‘Results’ section has the following added description of the activities of PAM- containing pre-spacer mimics.

Pages 16-19, lines 265-297. Cas1-Cas2 activity on pre-spacer mimics carrying the PAM sequence

"The strand cleavage and strand transfer steps of proto-spacer insertion at the CRISPR locus must engender safeguards against self-targeting of the inserted spacer as well as its non-functional orientation. However, no strand selectivity is seen in the in vitro Cas1-Cas2 reactions with already processed proto-spacers lacking the PAM sequence (Figures 2 and 3). By coordinating PAM- specific cleavage of a pre-spacer with transfer of this cleaved strand to the L-R junction, the inserted spacer will be in the correct orientation to generate a functional crRNA. To examine this possibility, we tested the integration characteristics of pre-spacer mimics containing the PAM sequence.

The inclusion of PAM or PAM and its complement in the integration substrates (Figure 2- figure supplement 3A) did not confer strand specificity on reactions with Cas1-Cas2 alone or with added Csn2, Cas9 or both (Figure 2-figure supplement 3B-E). Optimal integration by Cas1-Cas2 occurred with the 26 nt strands of the native protospacer with their 4 nt 3’-overhangs (Figure 2- figure supplement 3B-E; lanes 2). The pre-spacer mimics containing one or both > 26 nt strands had reduced integration competence (Figure 2-figure supplement 3B-E; lanes 4). Even here, the 26 nt strand with the 4 nt overhang (Figure 2-figure supplement 3C; lane 4) was preferred in integration over the longer 29nt PAM-containing strand (Figure 2-figure supplement 3D; lane 4) or the 33 nt PAM complement-containing strand (Figure 2-figure supplement 3E; lane 4). In contrast to the processed proto-spacer that gave nearly equal integration at L-R and R-S, IP(L- R) ≈ IP(R-S) (Figure 2-figure supplement 3B-E; lanes 2), the longer pre-spacer mimics were inhibited in integration at R-S, IP(L-R) > IP(R-S) (Figure 2-figure supplement 3B-E lanes 4). This is the expected outcome if the initial strand transfer occurs at L-R, and a ruler-like mechanism orients the reactive 3’-hydroxyl for the second strand transfer at R-S. This sequential two-step scheme for proto-spacer integration is consistent with the results shown in Figure 3 as well. These reaction features were not modulated by Csn2 or Cas9 (Figure 2-figure supplement 3B-E; lanes 6 and 8), although Csn2 plus Cas9 was inhibitory (Figure 2-figure supplement 3B-E; lanes 10).

There is no evidence for integration accompanying PAM-specific cleavage in our in vitro reactions. In the E. coli CRISPR system, Cas1-Cas2 is apparently sufficient for PAM-specific cleavage in vitro (22). By contrast, in the S. pyogenes system, cleavage is attributed to Cas9 or as yet uncharacterized bacterial nuclease(s) (35). The mechanism for generating an integration- proficient and orientation-specific proto-spacer, which may not be conserved among CRISPR systems, is poorly understood at this time."

Author Response:

Reviewer #1 (Public Review):

1. The authors use a surrogate marker for "slow" vs "fast" MNs: immediate vs delayed firing in response to rectangular current injection. They switch their language to call these motoneurons slow and fast, but they should be more cautious about doing so, given that firing is a surrogate marker that has not been fully studied and characterised across this time period. We do not know if developmental changes in Kv1 might also contribute to the effects on spiking seen here. I agree, there is delayed firing from the outset, which is interesting in itself given the rather homogenous other properties. But that doesn't mean that Kv1 expression is stable and thus non-contributory to the changes in firing described.

As suggested in the editorial summary, we have changed all reference to fast and slow motoneurons to delayed and immediate firing motoneurons up until the discussion. Furthermore, we have expanded our discussion highlighting potential roles for Kv1 in shaping motoneuron rheobase across development as an interesting direction for future study. Page 19, Lines 630-639.

1. The focus of the manuscript is on MN recruitment, but recruitment is never defined, despite being used in the title, through the abstract and key points, as well as throughout the manuscript. What they are looking at is response to current injected at the soma vs recruitment during a behaviour when synaptic inputs are bombarding an extensive dendritic tree. Thus, this manuscript does not look at recruitment per se, but rather activation of action potentials in response to intra-somatic stimulation. Accordingly, the term "recruitment" would be best kept to the Discussion.

We have changed the wording in the title from ‘recruitment’ to ‘active properties’ in the title, changed references to recruitment current to rheobase in the results section, and re-addressed recruitment in the discussion. Further, we have highlighted the importance of where synaptic inputs terminate on motoneurons and implications for recruitment with emphasis placed on the compartmental localization between synaptic terminals and compartmental clustering of voltage-sensitive ion channels on Page 21, Lines 688-692.

I also note that the "size" principle relates more to electrical than physical size in the 21st century; I agree that the two are correlated, but the authors may not want to stick to arguments about physical size.

We have tried to clarify when we are referring to electrical or physical size within the revised text.

1. Figure 6 is problematic, possibly in the way it's presented. It seems to me, but not clear, that the authors suggest that Ih is active at rest, more so in "F" than "S," and that therefore "F" are more depolarised and have smaller mAHPs. So (a) how come RMP didn't seem to come out in the PC analysis earlier? (b) would that not suggest that "F" would be "recruited" earlier? (Or is it that there is reduced sodium channel availability because of the depolarisation - what are the differences in spike amplitude and rate of rise?) (c) shouldn't the mAHP be larger if the cell is more depolarised and further from the potassium equilibrium potential? On the last note, maybe it's that Ih makes the mAHP smaller, but with the kinetics of Ih, wouldn't the decay be faster, but in Fig 6 the decay seems faster when Ih is blocked? Finally, Fig 6E suggests changes in the fAHP and delayed depolarisation (and spike width?) - how to these come into the picture? If the fAHP is thought to result from a high conductance state, and if therefore one were to align the voltages based on this potential, then the mAHPs would be about the same amplitude? The authors likely have explanations, but I'm afraid that I can't follow it.

We agree that some aspects of Figure 6 were not ideal, particularly those related to RMP and mAHP. For example, we had attempted to utilise measures of mAHP to help provide evidence that Ih was active at rest. However, upon reflection, the data provided in the original manuscript did not achieve this as clearly as the new data we have added, where we now measure Ih at resting membrane potential. Furthermore, the subset of data utilised in the previous Figure 6B misleadingly suggested a slightly more depolarised RMP in delayed MNs. However, as noted in supplementary table 1, delayed firing MNs in fact have a more hyperpolarised RMP compared to immediate firing MNs. We have therefore removed some of these problematic aspects of the figure that lacked clarity and have become less important towards the main points we were attempting to make, given the new data we have added. The simple explanation for RMP not coming out as a strong contributor in the PCA is likely because this parameter is not a prime contributor of variance across MN type or through development. We did not find any difference in spike amplitude or rise time between subtypes (these data are summarized in Supplementary File 1). As is common practice, we injected bias current to measure parameters included in the overall PCA from approximately -60mV. However, in hindsight, this could have reduced potential effects of RMP on some of the properties related to activation.

1. A number of labs have looked at development of MN properties, and it would be useful to compare properties seen across different labs, for example Quinlan (e.g. PMID: 21486770) and Whelan (e.g. PMID: 20457856, which is only mentioned in the manuscript) (and for that matter, mine - e.g. PMID 10564356, PMID: 32851667 although I don't want to self-promote).

We have highlighted the consistencies between our results and those reported by others, including Whelan, Heckman, Zytnicki, and Brownstone Labs. This statement can be found on page 3, Lines 104-106.

We have also included a comparison table in the supplemental information (Supplementary File 3) and referenced this table in the discussion on Line 528.

1. In the Discussion, the authors might want to discuss propensity of F vs S MNs to express PICs / sustained firing as described in the Heckman lab (albeit particularly in cat; see PMID 9705452, for e.g.). How do these data correspond?

It is interesting to note that the properties of PICs that we measured (ie. onset and amplitude) in immediate and delayed firing motoneurons are similar to those of fully and partially bistable motoneurons described by Lee and Heckman. In particular, we find that higher input resistance, immediate firing motoneurons have smaller PIC amplitude than delayed firing motoneurons but their PICs are activated at more hyperpolarized membrane potentials. This is consistent with fully bistable motoneurons, which are higher input resistance, and have smaller PICs that are activated at more hyperpolarized voltages compared to the partially bistable motoneurons. While we did not see bistability in the samples of motoneurons that we studied, a key difference is that we studied intrinsic properties in the absence of neuromodulation, which is a key factor for promoting bistability in motoneurons. Modulation of PICs might contribute to this propensity for bistability; however, modulation of outward currents is also very likely. We have highlighted these similarities and differences in our discussion on PICs and can be found on Page 19, Line 591-596.

Reviewer #4 (Public Review):

1) During the 1st postnatal week, authors suggest that fast and slow MNs cannot be distinguished neither on their passive properties nor the rheobase and therefore their recruitment is mainly based on the size. The conclusion that the recruitment is not linked to MN functional differences is difficult to follow since the main distinction between MN subtypes is based upon the presence of a delayed firing, an active property that regulates the recruitment of MNs (Leroy et al., 2014). However, the square current pulse adopted to discriminate between delayed and immediate firing in Figure 1 was replaced with a ramped depolarization protocol on which the authors measured the rheobase (Figure 3A1). This suggests that the slow depolarization in immature motoneurons might minimize the activation of ionic conductance(s) responsible for the delayed firing and thus may bias the measure of the rheobase (the minimal current amplitude of infinite duration). In line with this, the recruitment of a motoneuron has been shown to depend on the rate of membrane potential depolarization preceding a spike (Krawitz et al., 2001). Rather than using a slow ramp depolarization, it therefore seems more appropriate to assess MN rheobase with the current pulse protocol used to distinguish between MN subtypes. With this kind of measure, differences in the excitability of MN subtypes related to active conductances may come out earlier during development.

The reviewer raises an interesting point, which is consistent with one also raised by Reviewer 1. We have now included additional analysis in which we calculated rheobase values from the long (5 s) square current steps that were used to identify delayed and immediate firing MNs. These rheobase measurements made using current steps correlate strongly with rheobase measured from slow ramps (W1: r=0.87 p < 1.0 e-15; W2: r = 0.95 p < 1.0 e-15; W3: r= 0.92 p < 1.0 e-15). This is consistent with findings from Leroy et al. (2014) and Buisas et al., (2012), who both also demonstrated similar rheobase values in response to ramps and current steps. Importantly, we also find similar developmental changes in rheobase across motoneuron subtypes when assessed with a current step - showing no difference in rheobase values between delayed and immediate firing cells at week 1, with differences emerging in week 2 due to a progressive increase in rheobase in delayed firing motoneurons at 2nd and 3rd weeks. These findings are included as a new Supplementary Figure (Figure 3 – Figure Supplement 1) and summarized in the results on Page 6, lines 173-178.

2) During the 2nd postnatal week, the study suggests that "PICs contribute to the emergence of orderly recruitment amongst MN subtypes". This interpretation appears too definitive because the study did not provide direct evidence for that.

This is a good point. We did not directly test the contribution of PIC maturation to the staggering of rheobase currents between weeks 1 and 2. We have revised this statement to soften our claim, restricting differences in PIC activation to the second week. This revision can be found on Page 9, Line 319-329.

The authors describe a more hyperpolarized activation of PICs in slow MNs suggesting that the early recruitment of PICs in slow MNs may help them to fire before fast MNs. By a pharmacological approach the authors show that the sodium PIC mediated by Nav1.6 channels sets the activation threshold of PICs and that their blockade increases the rheobase (recruitment current). However, since pharmacological investigations have been done only in fast MNs, it is not very informative on the putative role of the sodium PIC (and PICs in general) on the orderly recruitment of MN subtypes. Similar experiments should be extended to slow MNs to compare the effects with those observed on fast MNs. If sodium PIC plays a significant role in the differential recruitment of MN subtypes, its blockade should induce an overlap in the recruitment of slow and fast MNs.

We initially focused specifically on the roles of PICs in shaping recruitment of delayed firing motoneurons during weeks 2 and 3, because we were trying to account for changes in rheobase that occur within delayed firing motoneurons during this period of postnatal development. However, in response to the useful comments here and above, we have now conducted an additional set of experiments to determine the relative contribution of NaV1.6 (n = 12 MNs, 7 animals) and L-type calcium channels (n = 10 MNs, 7 animals) to PIC and rheobase in immediate firing motoneurons. These results have been integrated into Figure 4, the results section, and discussion.

Furthermore, voltage clamp recordings to characterize PICs in MN subtypes have been done without blocking potassium conductances. Therefore it is difficult to determine if differences in PICs between MN subtypes are related to inward currents or opposing outward currents.

We agree that we cannot rule out contributions of other currents to our measures of PIC in voltage clamp given that we did not block potassium conductances. This is indeed an interesting point. However, our approaches are consistent with previously published approaches (Quinlan, et al., 2011, Verneuil, et al., 2020), which may be useful for the purposes of cross-study comparisons. Of note, Quinlan, et al., (2011), did measure PICs in the presence and absence of TEA (n = 18), with no differences in PIC amplitude or onset found. However, recent work from the Brocard Lab has found opposing contributions of M-currents to measures of PICs in voltage clamp in Hb9 interneurons. This would therefore be an interesting direction for future study amongst motoneuron subtypes. As highlighted below, and in our revised manuscript, it is quite possible that outward currents may oppose and diminish the actions of Ih. This is also likely true for PICs and would be an interesting direction for future study. We have included an additional statement in our discussion on Page 19, Lines 630-639 to acknowledge this potential interaction and contribution to maturation of motoneuron recruitment.

3) During the 3rd postnatal week, the authors suggest that fast MNs display a prominent Ih current at rest that provides a depolarizing shunt delaying their recruitment compared to slow MNs. However, data appear not enough conclusive for such interpretation. First, the relationship between the resting membrane potential (RMP) and the amplitude of Ih (the larger the Ih, the more depolarized the RMP) depicted in Figure 6B from a small sample of MNs is not consistent with values reported in the supplementary table 1. Indeed, fast motoneurons supposed to have a prominent Ih current display a more hyperpolarized RMP compared to slow MNs. The opposite would be expected according to the authors' hypothesis. A similar concern can be raised regarding the strong relationship between the amplitude of Ih and that of the AHP illustrated in Figure 6F, which is not in line with the lack of difference in the amplitude of the AHP between slow and fast MNs in week 3 (see supplementary table 1).

As discussed above in response to similar comments from another reviewer, we realise that the data presented in our original manuscript (Figure 6) regarding the relationships between RMP, Ih and mAHP lacked clarity and perhaps depicted a subset of recordings that was not representative of the complete dataset reported in our supplementary table (as pointed out by the current reviewer). Furthermore, these data did not achieve our main objective of supporting the existence of a resting Ih current as well as the new data we have included - where we directly measure Ih at resting membrane potentials. Given the addition of these new data, and the potential oversimplification of our attempts to relate RMP, Ih and mAHP (e.g. Ih is unlikely to be the main contributor to RMP), the latter has been removed from the revised manuscript.

In addition, we have expanded our discussion to highlight potential roles for Ih in shaping recruitment of fast motoneurons during periods of inhibition, such as during rhythmic activity, where the membrane potential often dips below -70 mV. Fast fatigable (MMP9+) motoneurons have been shown to receive a greater density of inhibitory synaptic inputs, particularly those derived from V1 interneurons, compared to slow (ERRB+) motoneurons (Allodi et al., 2021), and this differential synaptic weighting may create greater opportunity for Ih to be engaged and contribute to staggering recruitment as our pharmacology data suggests. This addition can be found on Page 20 Lines 657-663.

Second, the inward current recorded in fast MNs to hyperpolarization at -70 mV appears not significantly affected by the Ih blocker ZD7288 (Figure 5J, and 5L) suggesting that Ih is not recruited at rest in this class of MNs.

We realize that displaying the full IV plots for Ih at weeks 2 and 3 in addition to before and after application of ZD7288 (at week 3) may not have effectively illustrated the magnitude of Ih measured at -70 mV given the wide range of current values, which vary 10-fold between measures made at -70 and -110mV. We have modified our graphs in Figure 5 to better illustrate the magnitude of Ih measured at -70 mV and -110 mV. We hope that these modified graphs better capture our observations. We have further simplified Figure 5 to reduce redundancy in results that are summarized in Supplementary File 2 and the Results text. We have also included additional traces and analysis in Figure 6 that highlight a significant ZD-sensitive sag potential detected in delayed but not immediate firing motoneurons when hyperpolarizing the membrane potential from -60 mV to resting potential. These data can be found on Page 13, Lines 434-439, and in figure 6A, B. These results have been further supported by an additional set of recordings of delayed (n= 13) and immediate (n = 11) firing motoneurons obtained from week 3 animals, where Ih was measured during a voltage step (in VC) from a holding potential of -50 mV down to the respective resting potential of each cell (Del: IhRMP = -96 ± 60 pA; Imm: IhRMP: -0.13 ± 11 pA; t(23) = 4.8, p = 8.7e-5). These results have been included on Page 13, Lines 431-434.

On the other hand, ZD7288 hyperpolarizes the RMP in fast MNs (Figure 6A) and reduces the amplitude of their sags recorded at -70mV (Figure 5M). Similar discrepancies are more striking for slow MNs. Slow MNs did not display inward current sensitive to ZD7288 above -80 mV (Figure 5N). However, ZD7288 unexpectedly hyperpolarizes their RMP (Figure 6C). How the authors can explain such discrepancies? An interestinsting, but unexpected observation is the hyperpolarization of the RMP by ZD7288 in immediate firing motoneurons, even though, we were unable to detect measurable Ih or sag at potentials at resting membrane potential in immediate firing motoneurons. We have two explanations for these observations.

1.) One possibility is that, as pointed out above regarding our measures of PICs, we did not block other conductances during our voltage clamp protocols for measuring Ih. It is therefore possible that immediate firing (and even delayed firing) motoneurons express other ion channels that oppose Ih and may mask its true magnitude and effects on membrane potential. Indeed, this has previously been demonstrated for a variety of currents (eg. Kjaerulff and Kiehn, 2001; MacLean et al., 2003; Picton et al., 2018; Buskila et al., 2019). If this is true, then it is possible that the relatively smaller Ih in immediate firing motoneurons may have been masked, whereas the relatively larger Ih in delayed firing motoneurons may have been more apparent. In support of this possibility, we note that many of the immediate firing motoneurons demonstrate a slow hyperpolarization of the membrane potential during current steps intended to measure sag. Interestingly we also find this phenomenon in delayed firing motoneurons (that demonstrated depolarizing sag potentials at baseline), in the presence of ZD7288. We have included an example in the modified figure 6 to highlight this phenomenon. We have included additional discussion to highlight these caveats and possibilities (Page 20 Lines 664-675).

2.) Alternatively, voltage and space clamp errors, may have caused an underestimate of Ih. While we expect space clamp errors to be greater in the largest motoneurons, such as the delayed firing motoneurons, it is possible that such errors may have been sufficient to mask the small, albeit significant Ih in immediate firing motoneurons. It is possible that blockade of this small Ih by ZD7288 in immediate firing motoneurons may have hyperpolarized their RMP due to their high input resistance. This has been highlighted on Page 20 Lines675-679

Finally, there is a mismatch in values reported in supplementary table 2 and figures 5G and 5I. In the table, both Ih amplitude and Ih density (at -70mV) appear significantly different between slow and fast MNs in week 3, but not in figures 5G and 5I. Altogether, these results appear inconsistent.

We have modified our graphs to better illustrate the range of inward currents measured at -70 and -110 mV, which due to such high variance (in some cases 10 fold comparing those measured at -70 and -110 mV), were not as apparent when showing the full IV plot. We have modified our graphs in Figure 5 to better reflect the data summarized in Supp Table 2, and capture our observations. Both the data in the table and the plots have been analyzed using the same 2 way anova with cell type and age as factors.

Regardless of inconsistencies, data should be replicated at least with a second Ih blocker such as Ivabradine hydrochloride or Zatebradine hydrochloride.

We have performed an additional set of experiments in delayed (n = 8) and immediate (n = 3) firing motoneurons with ivabradine. These new results are included in text on Page 14, Lines 457-465. 10 µM Ivabradine produced a 45% reduction in Ih, and consistent with ZD7288, hyperpolarized the RMP of both delayed and immediate firing motoneurons and caused a significant decrease in rheobase of delayed firing motoneurons.

Minor concerns: 1) Does the pharmacological blockade of Nav1.6 channels 4,9-AH-TTX induce changes in the spiking threshold as already reported in cortical neurons (Hargus et al., 2013)? Such an effect may contribute to the higher rheobase observed in fast MNs under 4,9-AH-TTX (Figure 4M).

We have included an analysis of spike threshold before and after application of 4,9-AH-TTX. In line with previous reports from cortical neurons (Hargus et al., 2013), 4,9-AH-TTX significantly depolarized the spike threshold of delayed and immediate firing motoneurons and could contribute to the higher rheobase observed in delayed firing motoneurons following blockade of Nav1.6. The results from this analysis have been included in text and can be found on page 9, Line 306-310.

2) The study reported a more depolarized PIC in fast MNs during the 2nd postnatal week but the acceleration onset voltage in response to a current ramp depolarization (attributed to the activation of PICs), is similar between slow and fast MNs at the same age (Figure 4C). This is in discrepancy with figure 4G, where a significant effect on PIC onset voltage is shown within the same time points.

Indeed, there are differences between delayed and immediate firing motoneurons in the onset voltage of the PIC measured in voltage clamp at week 2 and these findings are not mirrored by differences in the accelerating phase of the membrane potential depolarization as measured from depolarizing current ramps in current clamp mode. We believe that this difference likely reflects that measurements made in current clamp are indirect estimates of PICs, whereas voltage clamp protocols provide more direct and likely more sensitive measurements. This is highlighted on Page 8, Line 263-265.

Author Response:

Reviewer #1:

Cesanek et al. performed a series of experiments designed to reveal whether or not the encoding of motor memories for novel object weight carries categorical structure. That is, given a set of objects of varying sizes and with weights that must be learned through experience, are objects grouped into categories such that the learned weight of one object generalises to objects within the same category, but not to objects outside of that category. Their results convincingly demonstrate the presence of such a categorical encoding. They show the following:

1) The weight of an outlier object is not learned if its weight is near the weight predicted by category membership.

2) The weight of an outlier object is learned if its weight is far from the value predicted by category membership.

3) The weight of an outlier object is learned if there is no category structure binding the remaining objects (i.e., there is no category against which an outlier can be defined).

4) If an outlier object is learned, then it influences the estimated weight of the other category members.

5) If the weight of an outlier object is learned first in isolation, it is unlearned when the remaining objects are introduced if and only if its weight is near the value predicted by category membership.

6) The threshold that constitutes "near" or "far" from the category boundary depends on recent sensorimotor experience.

7) Learning of the outlier is all-or-nothing on a per participant basis.

The major strength of the paper is the persuasiveness of the behavioural experiments, which were designed soundly and yielded clear results. There is little doubt that the some motor memories carry the type of categorical encoding detailed by the authors.

Thank you for this clear summary of our main findings and the positive evaluation of the persuasiveness of our study.

The major weakness of the paper is that it does not make strong contact with the relevant existing literature to clearly show that categorical encoding is (1) a truly novel behavioural observation in the motor learning literature, and (2) that it is inconsistent with the predictions of common motor learning theories and models. In particular, the authors own prior work frames motor learning as being governed by multiple internal models that can be switched between depending on contextual cues and environmental demands (see reference below). Insofar as contextual effects can drive similar results as categorical memory encoding, it is unclear how and why this and related models would fail to account for the present data.

Wolpert, D. M., & Kawato, M. (1998). Multiple paired forward and inverse models for motor control. Neural Networks, 11(7-8), 1317-1329.

With respect to the first issue, we do not know of any literature that discusses categorical effects in motor control. In our revision, we have added text emphasizing that existing models of motor control cannot account for categorical effects because they do not contain any form of categorical encoding. With respect to the second issue, please see our response to Essential Revision #2, where we explain that the MOSAIC model of Wolpert & Kawato (1998) is effectively an associative learning model, not a categorical learning model, and hence could not explain our data.

Reviewer #2:

Using a novel 3D robotic device, the authors had participants learn to lift four training similar-looking objects whose weights were linearly correlated with the sizes and then tested how this training influenced the later memory formation for the middle-sized object with different densities. When the difference between the actual weight and the weight estimated from the linear relationship was not so large (i.e., within a family boundary), surprisingly, the training for lifting the test object was ineffective: The estimation of the object weight was constrained by the linear relationship of the family between size and the weight. The memory specific to the new object could be developed only when the difference was large enough.

The results were unexpected from the conventional idea that object properties are encoded in an "associative map." The authors interpreted these results as evidence that the motor memory for lifting objects with different sizes and weights could be formed according to the "object family effect." All results of other control experiments were consistent with this interpretation.

I was intrigued by the counter-intuitive results that the motor system sticks to estimate the object weight based on the family property even though this estimation is incorrect and induces the error. Although it remains unclear how such a memory for multiple objects is integrated from memory for each object, it is sure that this study has demonstrated a new aspect of motor memory while manipulating the objects with different sizes and weights.

Thank you for this positive summary of the impact of our study.

Reviewer #3:

In this paper, Cesanek et al. use a novel object lifting task to investigate the "format" of memories for object dynamics. Namely, they ask if those memories are organized according to a smooth, local map, or discrete categories ('families'). They pit these competing models against one another across several experiments, asking if subjects' predicted weights of objects follow the family model or a smooth map. This was tested by having people train on objects of varying volumes/masses that were either consistent with a linear mapping between volume/mass, or where those dimensions were uncorrelated. This training phase was either preceded by, or preceded, a testing phase where a novel object with a deviant mass (but a medium-size volume) was introduced. As the authors expected, individuals trained on the linear mappings treated novel objects that were relatively close to the "family" average mass as a member of the family, and thus obligatorily interpolated to compute the expected mass of that object (i.e., under-predicting its true mass); conversely, when a novel object's mass was a substantial outlier w/r/t the training items, it was treated as a singleton and thus lifted with close to the correct force. Additional variations of this experiment provided further evidence that people tend to treat an object's dynamical features as a category label, rather than simply forming local associative representations. These findings offer a novel perspective on how people learn and remember the dynamics of objects in the world.

Overall, I found this study to be both rigorous and creative. The experimental logic is refreshingly clear, and the results, which are replicated and extended several times in follow-up experiments, are rather convincing. I do think some additional analyses could be done, and data presentation could be improved. I also thought the generalization analysis, as I interpreted it, was difficult to align with the initial predictions.

Thank you for this assessment of our work. In particular, please note that we have modified the generalization analysis based on some of your recommendations, and we feel that it is now more convincing and easier to understand.

Author Response:

Reviewer #1 (Public Review):

In this manuscript, Khan et al. investigated the roles of SARS-CoV-2 proteins on activation of immune cells. The authors found that the macrophage cell lines such as human THP-1 cells and mouse RAW 274.7 cells with recombinant viral proteins, and found that only spike proteins (S1 and S2) could potently activated macrophages to produce pro-inflammatory cytokines and chemokines.

Strengths:

It was Intriguing that only spike proteins (S1 and S2) could potently activate macrophages to produce pro-inflammatory cytokines and chemokines. The authors also observed that direct contact of macrophages with spike protein transfected epithelial cells, that mimic viral infection, resulted in the activation of macrophages. Detail analyses showed that spike proteins were recognized by Toll-like receptor 2 to activate NF-kB signaling. In vivo mouse experiments further supported the in vitro experiments. This study revealed a pathogenic of the SARS-CoV-2 spike proteins that is directly activating the host inflammatory responses, which therefore may have a profound impact in understanding a novel aspect of the cytokine signaling that is involved critically in the COVID-19 pathogenesis unveiled for the first time.

Weaknesses:

Recent report by Shirato and Kizaki (Heliyon 7(2021) e06187: 10.1016/j.heliyon.2021.e06187) showed that SARS-CoV-2 spike protein can stimulate macrophages (RAW 264.7 cells and THP-1 cells) to produce pro-inflammatory cytokines via TLR4-dependent manner. This is likely to contradict this study. The authors must thoroughly argue these controversial observations.

We are grateful to the reviewer for finding our study impactful. We also appreciate the reviewer’s suggestion to discuss why our findings are inconsistent with other studies. In fact, three recent studies attempted to explore the role of SARS-CoV-2 structural proteins in inflammatory responses (Shirato K, Heliyon, 2021; Zhao Y, Cell Res, 2021; Zheng M, Nat Immunol, 2021). While two studies demonstrated that the Spike protein is responsible for triggering inflammation (Shirato K, Heliyon, 2021; Zhao Y, Cell Res, 2021), the other study found that the E protein is inflammatory (Zheng M, Nat Immunol, 2021). Further, as they tried to identify the cellular sensor for S and E proteins, the first two studies demonstrated that S protein is sensed by TLR4, which contradicts our findings. Notably, Dr. Kanneganti’s group nicely showed that Tlr2-/- macrophages do not produce cytokines during SARS-CoV2 infection, while Tlr4-/- macrophages are responsive. These data are consistent with our finding that TLR2 but not TLR4 is the sensor for S protein. It is intriguing that the findings of all these studies are partially consistent and partially contradictory. We believe that the discrepant results of other studies are possibly due to contamination of bacterial pattern molecules. Indeed, we noticed that other studies used recombinant proteins generated in E. coli. Thus, it is possible that those recombinant proteins were contaminated by LPS and other bacterial PAMPs, which may activate the TLR pathways. Since the inception of our studies, we were concerned about the possible contamination of recombinant SARS-CoV2 proteins. Therefore, throughout the study, we used recombinant proteins generated in mammalian cells (HEK293T cells). We used three different S proteins – S1, S2, and S tri from two different commercial sources (RayBiotech and R&D), and we found that all S proteins triggered the inflammatory response (Figure 1- Figure supplement 1E). Further, both S subunit and S-tri are sensed by TLR2, but not TLR4 (Figure 5E and 5F). We have included a discussion on this concern in Page 16.

Author Response:

Reviewer #1 (Public Review):

Rarely do I read a paper and have so little to criticize. The paper is very well written and the studies have been conducted carefully and described fully. The only comment that I would make is that the divergence of quaternary structure in CTPS filaments brings to mind some well-known parallels that should be cited and discussed. Perhaps the most prominent one is hemoglobin, where this protein can form tetramers in some species that are very different from the vertebrate tetramers. In plants, I believe that dimeric hemoglobins have been described. Another example would be the diversity of filaments formed by actin-like proteins in bacteria, while in eukaryotes the actin filament architecture has been extremely conserved.

We thank the reviewer for pointing out the evolutionary parallels with well-known polymers, and have updated our discussion with a comparison to diverse actin architecture.

Author Response:

Reviewer #1 (Public Review):

This paper seeks to address how animals use sensory feedback from multiple sensory modalities during the navigation of complex environments. In particular, the authors are concerned with the use of odor-based signals for search behaviors, as odors are sparse and do not provide reliable directional information. An additional goal of this paper is to model the general principles they extract from their experiments so that they can be applied to engineered systems. Through the use of a virtual reality (VR) system that combines visual, wind, and odor input in a controlled manner, the authors investigate the search behavior of the silkworm moth, Bombyx mori. The females of the species emit a pheromone that triggers both a search behavior as well as mating behavior in males. Whereas previous work has taken advantage of the highly stereotyped nature of this behavior, it is unclear how these animals integrate other sensory cues such as wind and vision to locate an odor source. By presenting these stimuli in different combinations where the odor is always present, the authors find that the presence of a wind source strongly influences the animal's behavior-specifically its ability to locate the target-although the direction from which it is presented is crucial when all three are together. Multimodal input also affects the moths' walking speed and turning behavior, and an updated model that incorporates wind direction outperforms other models of this behavior.

The conclusions of the paper are generally supported by the presented data, although there are some issues with the framing of the paper and the model that I feel should be addressed.

1) The introduction and discussion go to great lengths to emphasize the technical advance of the constructed VR apparatus. Although it is certainly an impressive achievement, the use of VR for exploring insect behaviors like search and navigation is hardly a new development at this point in time. Indeed, work on flies, Drosophila and otherwise, going back decades has used VR to extract principles regarding issues such as visually mediated flight control or walking. More pertinent to the ideas of multimodal sensory integration explored here, over the past decade numerous researchers have combined visual input with odor and other cues to discern the relative importance of each of these modalities during search behavior. For example, see Duistermars and Frye, 2008. Generally, I feel that the paper overemphasizes the technical advance without providing sufficient biological context. So much work has been done on Bombyx that a paper using these methods has the ability to address, but much of that literature is absent from the paper. I think focusing more on the behavior will broaden the appeal of the paper by putting it in conversation with a well-established phenomenon

Thank you for providing these insights. As you pointed out, we improved the quality and discussion of the paper by adding literature to the introduction because there were few descriptions of the multisensory integration of other insects. Regarding VR research, the recent definition of VR is that "virtual reality creates a physical and a mental space for people." In other words, we need to connect a device that can provide multisensory stimuli to organisms (physical space) and a mental space where avatars reflect their actions. Based on this definition, we claim that most biological experiments with insects have been conducted in physical space (just utilizing multisensory stimulators) and have not measured the purposeful behaviors that result from connection to the mental space. Behavioral experiments using a multi-sensory stimulator have revealed how each modality is utilized, but it has not been directly investigated how other modalities affect actual navigation predominantly utilizing an odor, such as in the current study. In fact, the previously proposed insect-inspired algorithm for navigation was modeled based on experimental data using a multisensory stimulus device, but as a result of simulation, it behaves differently to the actual insect. On the other hand, the model derived from the experimental results using the VR system can reproduce the behavior of the insect (silkmoth). In other words, we claim that in order to model the superior functions of an organism, it is necessary to do more than provide multisensory stimuli, and it is important to obtain the relationship between multisensory stimuli and behavioral output when the organism performs the actual purposeful behavior (e.g., navigation behavior). However, as you pointed out, we necessarily revised the manuscript including those comments because it is a fact that wonderful biological findings have been clarified by multisensory stimulation experiments.

2) I think that the model is well-done and fits with the goals of the paper. Asking about the role of wind direction in this behavior is an important step given the behavioral data presented. However, I am not convinced based on the data that the new model developed by the authors is much better than the surge-zigzag model. The success rates are slightly different (is the statistical difference a function of the number of runs or timesteps?) and both models search about the same amount of time before finding the source. Finally, the migration probability maps are rather similar, so it is hard for me to conclude that factoring in wind direction is necessary to get good performance out of the model.

Thank you for your valuable comments. Because a surge-zigzagging algorithm utilizes only odor stimuli to determine actions, an agent moves to the edge where the change in odor information occurs most. The behavior of MiM2 is modulated by information on wind direction and it moves into the areas with a high probability of odor reach. Hence, we thought that the effectiveness of the model could be shown by starting the search for the agent near the edge. Consequently, we set the initial position (x, y) = (300 ± 100) in the area where the odor is hard to reach and carried out the simulation. As a result, the MiM2 algorithm, which actively moves in the area where the probability of odor reach is high, was significantly better than other conventional algorithms. As for the search time, because MiM2 is based on the surge-zigzagging, the movement speeds were almost the same; therefore, both algorithms (MiM2 and surge-zigzagging) require almost the same search time. However, we found that there is a difference in the search success rate because there is a difference in the selected route.

Revised part: Lines 282–311 (red letters) at “Modeling and validation of behavioral modulation mechanisms”.

Reviewer #2 (Public Review):

This paper uses a multi-model virtual reality system to assess which combinations of visual, wind, and olfactory information male silk moths rely on to find a female. The overall conclusion is that for the moths to search effectively, wind direction information is an important input. Vision, on the other hand, while it is used to control angular velocity, does not appear to be important for the moths to search effectively. Given what is known about other walking and flying insects these results are not surprising. Although the virtual reality system is advertised as being able to provide naturalistic and complex stimuli, the visual stimuli are limited to a traditional LED array, and the olfactory stimulus is created by projecting a 3-dimensional plume into two dimensions. The analyses of the data are rather simplistic and do not provide a mechanistic description of what the moths are actually doing on a moment-by-moment basis. The authors then proceed to construct a model for search behavior that uses frequency and relative timing information for the odor stimulus in conjunction with the wind direction information. This search algorithm is ever so slightly better than the prior surge-zigzagging algorithm. The role of relative timing information in the olfactory signal supplied to the moths in the virtual reality experiments, however, was not investigated.

Thank you for your valuable comments. We can respond to the following points:

Effects of vision: Our biological experiments demonstrated that the visual stimulus maintains the balance of the angular velocities of the left and right rotational movements. When the effect of this balance on the function of an odor search was confirmed by simulation, the path became longer due to the bias in the search trajectory, and the search time became significantly longer. In addition, the success rate decreased as in the biological experiments. This suggests that an efficient odor-source search can be performed by maintaining the balance between the left and right angular velocities.

Limitation of visual stimulus: The main purpose of our study was the integration of other sensory information during navigation using odor. Therefore, the problem becomes complicated when cognition, i.e., visual object identification, is included. For that reason, we adopted optical flow as a visual stimulus, which does not involve cognition but affects behavior. Because there were several studies using LED arrays to provide optical flow to an insect with compound eyes, we conducted experiments using LED arrays this time as well. We had already tested whether the optical flow by the LED array could be provided correctly (see Supplementary Material).

The plume model: We utilized plume diffusion, which was observed in two dimensions as a virtual odor field. We captured the movement of particles on a two-dimensional plane using particle image velocimetry technology. Please refer to the Supplementary Material for details.

The behavior analysis: By plotting each trajectory in the behavior analysis and calculating a histogram of the change in heading angle, we investigated what kind of feature difference occurs during the odor search depending on the conditions of sensory stimulation. Regarding the odor, which is a cue/stimulus, we analyzed the behavioral change with respect to the "frequency", instead of in response to a single stimulus. The reason for this is that recent studies have shown that while searching the environment (1) the frequency of the odor changes with distance from the odor source, and (2) the behavior of Drosophila in walking also changes with frequency. By analyzing the behavioral changes with respect to the odor detection frequency, we modeled the behavior during the search based on the odor frequency information. As a result, we found that not only the behavior close to the movement of a silkmoth could be reproduced compared to the conventional model, but the search success rate could also be increased.

Simulation: As you pointed out, we increased the conditions for verification of simulation because the simulation conditions were limited. As a result, we found that the proposed model has a higher odor tracking performance than the other conventional models even if the initial position is different. However, as you pointed out, this is a relatively simple condition of simulation; therefore, in the future, we plan to implement the algorithm in an autonomous robot system and verify the performance of the algorithm in more complicated spaces, such as obstacle environments or an outdoor environment.

Author Response:

Reviewer #1 (Public Review):

This work presents a new Bayesian method for detecting those patterns of neural responses that are connected to behavioral output and are worth investigating further. The manuscript contains the derivation of the approach and its test on synthetic and real neural data.

The derivation should be improved by providing additional steps. For example, it was not clear how Eq. 5 was derived and why the double derivative with respect to parameters theta_mu and theta-nu are present ( these terms appear to be missing in the definition of log-likelihood).

Thank you for the suggestion. We have added steps in the derivation leading to the Ising equation for the indicator variables, now in Eq. (8). These intermediate steps corresponds to the two main approximations of the BIA method, namely, the saddle point approximation for the posterior (Eq. (5)) and the Taylor expansion in the inverse regularization strength (Eq. (6)). We hope that these changes improved the readability of the derivation.

Parameter M should be more clearly defined as the number of samples. It is briefly mentioned on line 170, but it was difficult to connect this to equation (7) and those following that use M explicitly.

We thank the Reviewer for the suggestion. We have clarified the definition of M just after Eq. 11.

Is it possible to include multiple binary quantifications of behavior, similarly to how words are constructed from neural spike trains? For example, one can envision describing a particular song segment with respect to multiple binary features simultaneously.

We explicitly examine this question in “Dictionaries for exploratory vs. typical behaviors” and the corresponding Figure 6, which repeats our analysis for different binary discretizations of our behavioral data.

Reviewer #2 (Public Review):

Summary:

Hernandez et al propose a new statistical tool for identifying codeword in multivariate binary data (for instance neural activity patterns), with a small number of measurements. It demonstrates the utility of the approach on neural responses to analyzing the statistical structure of songbird responses and how they change in different contexts (during exploration vs typical song production).

Strengths:

• The approach is innovative, in that it takes advantage of clever tools from sparse linear regression, in particular a method termed Bayesian Ising Approximation (BIA), to be able to identify codewords individually, rather than directly estimating a model of their joint statistics, by comparing to a null model that assumes independence across dimensions. This approach has the advantage of resulting in a very flexible model, with very few assumptions about the statistical structure of the data, that is applicable for a range of datasets sizes; the more data is available, more of the structure underlying it can be revealed .
• The strong mathematical foundations provides clear bounds on data regimes in which the approximation is theoretically well justified and reasons to expect that the estimated models are minimal and interpretable.
• The numerical estimation procedures are fast, and computationally efficient (for a reasonably sized neural dataset, can be run on a regular laptop).
• The code is available on github for quick community dissemination.
• Application to identification of behaviorally relevant patterns of co-activity goes beyond previous Ising-based models used in neuroscience.
• When applied to songbird data, it reveals that the variability in neural responses during exploration has much more structure than previously thought.

Weaknesses:

• Although the paper is written as a methods paper, emphasizing the technical contributions and promising wide applicability to a range of different types of datasets, the numerical validation of the method is very much restricted to the statistical regime of the songbird dataset. From the perspective of a potential future user of the tool it's less clear how the method would behave on different datasets, and what needs to happen in practice for adopting the tool to data with different statistics.

We have edited second half of the abstract and a few sentences in the Introduction (see latexdiff file) to make it clear that our main applications to date have been to songbird data.

• The numerical comparison to other existing methods is minimal.

We have argued in our previous submission that there really are no other methods to compare to, designed to work in the regime similar to uBIA. It seemed to us that it would be unfair to run other methods on our datasets, see them not work well (as expected – because they make assumptions that are invalid in our regime), and then claim success. However, since the concern has been raised again, we really have to address it. To do this, we added a section in the Online Methods “Direct application of MaxEnt methods to synthetic and experimental data”, in which we compare uBIA to the relevant interactions model of Ganmor et al., with which uBIA has the highest similarity. The results are as expected – a method not designed for our data regime fails. We emphasize here again that the relative superiority of uBIA on these data should not be taken as a slight directed at other methods, but rather as an indication than, to cover different data regimes, multiple methods should be combined. We emphasized this in the “Overview of prior related methods in the literature” supplemental section.

• The songbird analysis already reveals some challenges with respect to interpretability: in particular it is not clear how much information about the underlying neural processes can be revealed by summary statistics generated by the method, such as the number of codewords and their length distribution.

The reviewer is correct that our analysis of the songbird data raises a number of important questions for future studies. Although these remain to be answered, we emphasize that before the biological interpretation of over/underrepresented neural patterns can be attempted, such patterns must first be identified. uBIA therefore represents a crucial advance in our ability to address these questions.

Most conclusions are reasonably supported by the data. The analysis of the irreducibility of the codewords has insufficient support based on the numerical simulations. Moreover, the generality of the tool and comparison to other methods are discussed in almost entirely theoretical terms, which makes the claim on immediate utility for other datasets less convincing, especially outside the neuroscience community.

We hope that the addition of the new comparison figure partially alleviates these concerns. Additionally, we point out that 3rd and 4th order words are long, as most others deal with just pairs, as illustrated in the new Figure 7. Indeed, it is not easy to fit an N = 20 Ising model with 4th order terms, because there are 20 ∗ 19 ∗ 18 ∗ 17/(4 ∗ 3 ∗ 2 ∗ 1) = 4845 terms in this model, which cannot be fit from just a few hundred samples, which is precisely why the Ganmor model fails in this case (Fig. 7).

Nonetheless, the idea is quite interesting and likely of broad interest for theorists interested in the development of unsupervised statistical tools for neural data analysis, with practical applicability for a range of modern systems neuroscience experiments that involve task specific ensembles as the building block of circuit computation.

Author Response:

We thank the reviewers for highlighting the importance and potential clinical significance of our findings. The concentration of bisphosphonate drug in tissues outside the skeleton, such as lung, are unknown and we recognise that further studies are needed to determine whether the effects of a bisphosphonate that we describe in mice also occur in humans with standard clinical doses of these drugs. Nonetheless, our findings add further weight to the view that bisphosphonate therapy has benefits beyond just preventing bone loss and could be considered as prophylactic agents to reduce the risk of pneumonia in individuals with osteopenia or osteoporosis, who are already eligible for treatment under standard clinical guidelines.

Author Response:

Thank you to the reviewers and editors at eLife for the comments on our manuscript. We believe we can address or rebut all of the reviewer comments, and our responses are included below.

Reviewer #1 (Public Review):

In this manuscript Bigge et al. use chemical inhibitors and a mutant in ARPC4arpc4 mutant to investigate the role of the Arp2/3 complex in regulating cilia length and assembly in Chlamydomonas. The authors have previously shown that the actin cytoskeleton is required for ciliary assembly and maintenance in this organism, but the precise mechanism(s) involved were unclear. Furthermore, while previous studies targeted the actin cytoskeleton in general, the current study focuses on branched actin networks regulated by the Arp2/3 complex. The authors first demonstrate that chemical inhibition of the Arp2/3 complex leads to shortening of existing cilia, a phenotype that is recapitulated in the arpc4 mutant and which can be rescued be reintroducing V5-tagged ARPC4 in the latter mutant. Next, using similar approaches they show that initial stages of cilium biogenesis are also impaired upon Arp2/3 complex inhibition. They next use a variety of approaches, mostly involving chemical inhibitors and F-actin- or membrane dyes, to assess the mechanism by which Arp2/3 complex affects ciliary biogenesis. They provide evidence indicating that Arp2/3 complex specifically promotes endocytosis at the plasma membrane to support lipid and protein for the growing ciliary membrane. This is an interesting discovery that advances our understanding of how ciliary membrane biogenesis is regulated, especially in Chlamydomonas.

Thank you for these comments and the accurate summary of our findings.

Reviewer #2 (Public Review):

Previous studies have demonstrated that interfering with actin polymerization leads to the shortening of flagella in Chlamydomonas cells, indicating that F-actin is important for ciliary/flagellar elongation. However, the precise roles of F-actin, and especially of branched actin networks nucleated by the Arp2/3 complex in cilia formation have not been elucidated. Here, the authors aimed to examine one of the mechanisms that may be involved in F-actin-dependent cilia elongation/formation, namely, actin- and clathrin-dependent endocytosis of membrane proteins.

The authors used both pharmacological inhibition and genetic disruption of the Arp2/3 complex to demonstrate that interfering with the activity of the Arp2/3 complex reduces the ability of Chlamydomonas to form or elongate cilia. The authors then showed that incorporation of existing (not newly synthesized) proteins into cilia is perturbed by the genetic or pharmacological inhibition of the Arp2/3 complex, and that new membrane for building cilia may be derived via an Arp2/3-mediated membrane retrieval pathway.

These experiments are carefully performed and convincing. A description of the Arp2/3 complex components and clathrin-containing structures in Chlamydomonas is novel and will be of interest to the cell biologists working on this model organism.

The authors propose that the Arp2/3-mediated actin assembly promotes cilia elongation/formation due to the contribution of the branched actin to clathrin-dependent endocytosis. This is an intriguing idea that ties together previous findings showing that F-actin is needed for cilia elongation and that some ciliary proteins are internalized from the plasma membrane and then redistributed to cilia. One concern regarding this portion of the manuscript in that the experiments addressing the role of endocytosis rely solely on the use of a pharmacological inhibitor of clathrin-dependent endocytosis, PitStop2. PitStop2 was previously shown to have non-specific effects on endocytosis, suggesting that its mechanism of action may not be directly related to disrupting clathrin heavy chain interactions (see, for example, Willox et al., 2014).

We thank this reviewer for their helpful comments regarding the work presented in our manuscript. We understand the concerns regarding the off-target effects of PitStop2. However, the PitStop2 data is only one of multiple lines of evidence that we provide supporting endocytosis occurring in these cells (internalization of membrane dye in Figure 5, internalization and relocalization of a ciliary membrane protein in Figure 6, actin patches in Figures 4 & 7, the presence of plasma membrane proteins in the cilia in Figure 8). Further, the requirement for Arp2/3 in the phenotypes listed points to an endocytic mechanism as the Arp2/3 complex has repeatedly been shown to be involved in actin dynamics in endocytosis in other systems. To determine the mechanism of endocytosis most likely occurring in these cells, we searched the Chlamydomonas genome for proteins commonly thought to be involved in endocytic pathways. We propose clathrin mediated endocytosis is occurring in these cells because this is the only pathway where the most important components of the mechanism are present in Chlamydomonas. This is opposed to every other form of endocytosis which is lacking some major component. For example, Chlamydomonas do not contain endophilin for endophilin-mediated endocytosis, flotillin for flotillin-mediated endocytosis, or caveolin for caveolin-mediated endocytosis. Therefore, our conclusions regarding clathrin mediated endocytosis do not rely entirely on PitStop2 data. The PitStop2 data is merely present to support the findings of our comparative analysis and other endocytosis assays. Further, mutants of clathrin heavy or light chain does not exist and cannot be reliably generated in this organism.

We now have further evidence that endocytosis is occurring in these cells using dynamin inhibitors. Specifically, using the dynamin inhibitor, Dynasore, reduces membrane uptake in a dye internalization assay. We plan to include this data in the next version of the paper.

Intriguingly, a previous paper by Kim et al., 2010 demonstrated that interfering with the Arp2/3-dependent actin assembly resulted in cilia elongation in mammalian cells, suggesting that branched actin assembly was counteracting growth of cilia. Similarly, actin depolymerization is known to promote ciliogenesis in mammalian cells. The difference between mammalian actin organization/cilia growth regulation vs. the new observations in the Chlamydomonas system should be discussed by the authors to help the readers understand whether the findings can be generalized to the diverse ciliated cell types or are unique to algae.

While Chlamydomonas is an excellent model for ciliary studies due to the structural and mechanistic conservation of the cilia in relation to mammalian cilia, there are several important differences between mammalian cells and Chlamydomonas cells that might influence ciliary dynamics. These differences have been reported on by others in the field (Jack et al. 2019). Briefly, one difference is that Chlamydomonas cells have a cell wall, which means they have no need for a cortical actin network as mammalian cells do. This cortical actin network in mammalian cells is thought to potentially block access of basal bodies and ciliary proteins from the cortex of the cell. This in turn blocks ciliary formation and assembly.

We now have data that a mechanism of ciliary elongation mediated by lithium, which is conserved between mammalian cells and Chlamydomonas, is blocked by loss of the Arp2/3 complex, suggesting that ciliary/membrane growth not just from zero length but also from steady state requires the Arp2/3 complex. This data will be reported in a subsequent publication. While we could speculate on the conservation of ciliary growth between mammalian cells and Chlamydomonas in this paper, a more well-supported speculation would be in the subsequent paper where we directly discuss the role of the Arp2/3 complex in a conserved ciliary growth process.

Reviewer #3 (Public Review):

Wild-type Chlamydomonas cells possess a pool of ciliary precursors that, in conjunction with newly synthesized precursors, are used to build a new cilium after de-ciliation. The cellular and molecular mechanisms the underly retrieval of the pool are unknown. Given the role of the Arp2/3 complex in endocytosis in other systems, it was reasonable to test whether it also functions in reclaiming of the pool of precursors. The central finding is that cells bearing a mutation in an essential Arp2/3 component, ARPC4, fail to assemble cilia in a timely fashion after de-ciliation. Although the failure of assembly is well-documented here, the manuscript lacks evidence that cells missing ARPC4 actually establish a pool of ciliary precursors in the first place. Without such information it is not possible to determine the cellular function that fails to occur after de-ciliation in the mutant: Retrieval of a pool of ciliary precursors? Establishing the pool during the ciliary assembly that occurs after cell division? Synthesis of the precursors as new cilia are formed? Sensing loss of cilia and activating the events needed for re-ciliation?

Upon discovering that there was almost no initial ciliary assembly in mutants of the Arp2/3 complex caused by loss of the functional Arp2/3 complex, we immediately suspected that either it lacked a precursor pool, as the reviewer suggests, or that mutants were unable to incorporate it. To discriminate between these possibilities, you typically need to be able to regenerate cilia in cycloheximide. However, the inability of genetic mutants to regenerate in cycloheximide prevents us from being able to do the typical studies testing new protein synthesis, precursor pool size, and new protein incorporation as they all require regeneration in cycloheximide. Therefore, to get around this roadblock, we used an acute perturbation through chemical inhibition on wild-type cells that have a normal ciliary precursor pool (as evidenced by their ability to grow to half-length in cycloheximide). These cells were deciliated and then the Arp2/3 complex inhibitor CK-666 (in addition to cycloheximide) was added only for the regrowth, and thus it was not able to affect the size of the precursor pool. Cells treated with CK-666 and cycloheximide could not incorporate the precursor pool we know exists in these wild-type cells (Figure 2B). We referred to these results in the text saying: “This can be further dissected because we know in the case of cells treated with cycloheximide and CK-666, the protein pool is available but is still not incorporated, suggesting a problem with membrane incorporation or delivery as opposed to a problem with protein availability” (Lines 195-200). We understand that this piece of data may be confusing the way it is displayed (only the final growth length after 2 hours), so we plan to provide the full regeneration graphs in the supplement.

While we are not able to concretely rule out a problem with signal transduction alerting cells that they have lost their cilia or with protein synthesis, we believe these are lesser effects because some arpc4 mutant cells do grow immediately following deciliation and because eventually cells reach near full length in the absence of cycloheximide (Figure 1C). However, we plan to test protein synthesis immediately following deciliation using semiquantitative PCR for the next iteration of this manuscript.

Finally, we provide several pieces of orthogonal data to suggest that membrane incorporation is a problem in these cells. First, we show that blocking membrane delivery from the Golgi causes increased ciliary shortening in arpc4 mutant cells compared to wild-type cells. This suggests there is an alternate source of membrane that is defective in the arpc4 mutant cells (Figure 3). We go on to show that actin structures that form near the membrane at the base of the cilia are absent in the arpc4 mutant but present in wild-type cells, and that these structures increase following deciliation (Figures 4 and 7). We show that aprc4 mutants are less capable of internalizing membrane than wild-type cells (Figure 5) and less capable of internalizing a ciliary membrane protein for mating (Figure 6). Lastly, we show that the pattern of ciliary membrane proteins that came from the cell body plasma membrane is altered in arpc4 mutants suggesting defects in at least one pathway (Figure 8). Based on this entire body of data together, in addition to our CK-666 data on wild-type cells with an existing precursor pool, we feel that the best-supported model is one in which a defect in membrane delivery may be responsible for the impact we see on ciliary assembly in cells lacking a functional Arp2/3 complex.

Other conclusions also were not sufficiently supported by the experimental results, including the following: Clathrin function in endocytosis: Pitstop2 was described as specifically blocking clathrin-mediated endocytosis, but reports in the literature indicate that Pitstop2 also blocks nuclear functions and clathrin-independent endocytosis. Experiments with an anti-clathrin antibody were interpreted as showing mislocalization of clathrin in the arpc4 mutants. The antibody was raised against a peptide near the N-terminus of human clathrin, but the antibody was not validated, and it was not reported whether Chlamydomonas clathrin even has that peptide.

We understand the concerns regarding the off-target effects of PitStop2, and as this concern was shared by reviewer 2 we have addressed this above.

If further evidence is needed of endocytosis occurring in these cells, we also have new data suggesting that dynamin is also involved in these processes.

We thank the reviewers for this comment, and plan to add a western blot confirming the clathrin light chain antibody works in Chlamydomonas in the next version. It is true that mammalian antibodies do not always work in Chlamydomonas. This particular antibody was selected based on the immunogen sequence. The immunogen of this antibody was the most similar we could find to the Chlamydomonas sequence (61.5% similar). Our intent with the included data was to highlight the similarities between both membrane internalization and clathrin behavior whether endocytosis or the Arp2/3 complex are inhibited.

Internalization of a protein from the plasma membrane: In the experiments to use protease-sensitivity to examine SAG1-HA relocalization induced by db-cAMP, the authors assumed that all of the SAG1-HA was on the cell surface in untreated cells and the 2 chemically treated cells, but they never experimentally documented this assumption.

We respectfully disagree that our assumption is that all of the SAG1-HA is on the surface prior to induction. In fact, our data support the opposite, that there is indeed some percentage of SAG1-HA on the surface and some percentage of SAG1-HA already inside the cell even in uninduced cells given that some portion is protected from trypsin treatment. Regardless of whether the full population of SAG1-HA is on the surface, we can still draw conclusions about the amount of SAG1-HA internalized under the conditions through quantification of the changes in band intensity. This specific assay can only report on whether there is an internalization defect for a plasma membrane protein destined for the cilium.

Arp2/3 relation to actin dots: Structures termed actin dots that stained with Phalloidin were reported to undergo changes after de-ciliation of wild-type cells and were missing in the arpc4 mutant. The conclusion that the properties of the dots in the wild-type cells changed with de-ciliation were not supported by statistical analysis. Also, without experiments showing localization of Arp2/3-V5 at the dots, it was not possible to assess whether, as the text asserts, Arp2/3 functioned at the dots.

Statistics were not done for Figure 4 where we show dots are present in wild-type cells and absent in the arpc4 mutant cells because it is clear even without statistics. We do not feel that including an infinitesimally small p value added any substantive information to this piece of data. However, statistics can and should be included in the next version for Figure 7 which shows an increase in dots (both percentage of cells with dots and number of dots per cell) following deciliation. To fully convince readers, we also plan to include images with a larger field of view showing a population of cells, which will make it clear that dots increase following deciliation.

We attempted to look at ARPC4-V5 localization in Figure 1 Supplemental Figure 2, but saw mostly diffuse localization with high noise. The ARPC4 component of the Arp2/3 complex may be diffusely localized with only some proportion being incorporated into the complex at sites where it is active, or the diffuse nature could be due to overexpression, or both. While we do not definitively show that overexpressed ARPC4 is at the dots in addition to its diffuse localization, we do not believe that this prevents us from drawing conclusions about Arp2/3 complex function in the dots. We do show that the Arp2/3 complex is required for dot formation as cells lacking a functional Arp2/3 complex do not have dots while cells that contain the Arp2/3 complex do have dots. Further, expression of the ARPC4-V5 construct rescues the presence of the dots.

Cilia resorption induced by BFA in arpc4 mutants: In the experiments with the Golgi-active agent BFA, the percent ciliation in the arpc4 mutants, but not wild type, fell rapidly after drug addition. The authors concluded that BFA induced ciliary resorption, but they did not determine whether the lack of cilia on cells was a consequence of cilia resorption or cilia detachment.

As the reviewer points out, the percent ciliation of the arpc4 mutant cells treated with BFA fell rapidly after drug addition. We believe this is due to resorption and not ciliary detachment because we see ciliary length decrease over time (not all at once). Additionally, in Figure 3B, we only measure cells that have cilia, and we know that arpc4 mutant cells cannot regrow their cilia to half-length in 1 hour (Figure 1C). Therefore, we know that the cilia we are measuring are in fact shortening. Further, there was no increase in free cilia in the samples. Images can be provided in the supplement for the next version.

Author response

Reviewer #3 (Public Review)

Major concerns:

1. The most substantive concern is that there is an alternative explanation for the data which must be ruled out in order to conclude that mutations are occurring during the study period. Consider the following scenario. Suppose that B cell clones expanded and diversified through somatic hypermutation prior to the study period (that is, prior to the secondary vaccination event which is the focus of the study). It seems that preferential expansion of highly mutated subclones during the study period could bias detected sequences towards more divergent sequences, even without ongoing somatic mutation during the study period. Preferential expansion of divergent sequences would give rise to higher average divergence as the study period goes on, giving the appearance of accumulation of additional mutations, but in fact these mutations had occurred prior to the study period and are simply more readily detected in the sparsely sampled repertoire sequencing data after their expansion. Far from being simply a pathological counter-example, this scenario seems biologically plausible, given that B cells harboring more divergent, affinity-matured sequences should generally have higher affinity antibodies that allow them to better compete for limited antigen and thus provide stronger division stimulus. This model predicts that some highly divergent sequences exist at early timepoints and would occasionally be detected. An example of this is seen in Fig 3C, where a divergent tip from an early sample time is present (labeled PB and colored blue, in the middle of the diagram), indicating that this divergent sequence was present early. While the authors' model of ongoing evolution is supported, this alternative model also appears to be consistent with the data and must be ruled out in order to conclude that clones are accumulating mutations during the study period, which is the central claim and most interesting and impactful finding of the work. The authors must provide evidence that their approach can distinguish between these scenarios. This could potentially be accomplished using simulations of the two scenarios to determine the power of the approach to distinguish between them.

We agree that this is a potential alternative explanation for a significant positive correlation between divergence and time, and have now addressed this as a possibility in the Results and Discussion sections. We have also removed explicit references to detecting “ongoing SHM” in the text, in favor terms that more directly reflect what our test detects such as “B cell evolution” or “increasing SHM frequency” which do not imply novel SHM over the sampling interval. Nevertheless, we believe our results are more easily explained as a result of ongoing SHM, and have added some text making that point. In the context of influenza vaccination, day 5 plasmablasts represent the breadth of the B cell memory pool. If measurable evolution were due solely from preferential recall, we would expect the divergences of sequences at later timepoints to fall within the range of day 5 plasmablasts. Instead, in the high-GC influenza binding lineages we identified (Fig 3B/C), many late-sampled GC sequences are clearly more diverged from the day 5 plasmablast response. Further, if measurable evolution from influenza vaccination were due simply to preferential re-stimulation of highly mutated B cells, we would expect influenza binding lineages without any GC sequences to be measurably evolving. To test this, we repeated the analysis in Fig 3A using only lineages containing influenza-binding monoclonal antibodies (mAbs). Results were highly consistent with Fig 3A: influenza-binding lineages without GC sequences were less likely to be evolving than those with high proportions of GC sequences (Figure 3 – figure supplement 3). Thus, significant GC involvement, rather than simply binding to influenza, is more predictive of measurable evolution. All of these points are more easily explained if measurable evolution is the result of additional SHM. Nonetheless, we cannot definitely rule out this alternative explanation, we have highlighted both possible mechanisms of B cell evolution. We have included descriptions of this new analysis in the Results (pp. 14-15) and Discussion (pp. 20-21).

1. Statistical support for measurable evolution appears to be lacking in several key examples. The reported percentages of measurably evolving lineages in several scenarios (7.2% for primary hepatitis B vaccination; 6.5% for allergen-specific immunotherapy; 5.9% for HIV infection) are near the false positive rate of the test (5% of lineages measurably evolving). The authors have performed this test on datasets from ~21 studies, raising a concern that multiple hypothesis testing could give rise to false positives in some of the datasets. These results are interpreted as evidence of measurable evolution, even though they could seemingly be explained by the false discovery rate combined with multiple hypothesis testing. The authors should clarify how these results can be interpreted in light of the false positive rate of their test and multiple hypothesis testing, and must consider whether more conservative conclusions are warranted in these scenarios.

We appreciate the reviewer’s concern and have added a new section on multiple hypothesis testing to the Discussion detailing these caveats (pp. 19-20), as well as additional details to the relevant Results sections (p. 10). We also repeated our initial germline divergence analysis that used “adjusted” measurably evolving lineages without the multiple testing correction, and found similar results (Figure 2 – figure supplement 4). We also repeated these analyses using a more strict cutoff (adjusted p < 0.05), which also yielded similar results (Figure 2 – figure supplement 4). These are discussed in the main text

Minor comments

1. The authors define measurably evolving populations as systems undergoing mutation and selection rapidly enough to be detected. Despite this definition, the test employed for measuring evolution appears to focus purely on accumulation of mutations without examining selection per se. Mutations accumulating neutrally could be detected as measurable evolution. For the sake of clarity, the authors should explain more clearly whether their test examines selection and ensure that initial definitions are consistent with later usage. It may be interesting to further examine whether the mutations detected as measurable evolution in antibody lineages are neutral or selected using classical tests for selection, such as the dN/dS statistic or summary statistics of the site frequency spectrum.

We have now made it clearer in the initial definition that measurable evolution does not necessarily require selection. This definition is more in line with the original definition of measurably evolving populations from Drummond et al 2003: “We define measurably evolving populations (MEPs) as populations from which molecular sequences can be taken at different points in time, among which there are a statistically significant number of genetic differences.

1. Statistical support for the association between GC B cells and measurable evolution should be clarified. On p14 L7-8, it is reported that "6.5% of lineages containing sequences from GC B cells were measurably evolving, compared to only 3.7% of lineages with no identified GC sequences." However, this does not constitute convincing evidence for the association because this difference of proportions is not significant. The proportions are 3 lineages measurably evolving among 46 lineages containing sequences from GC B cells, and 4 lineages measurably evolving among 107 lineages not containing sequences from GC B cells. Applying Fisher's exact test for a difference of proportion yields P = 0.43. While the evidence based on the trend in Fig 3A (of fraction measurably evolving against GC sequence percentage) is compelling, the authors should clarify whether each difference of proportions is significant. Providing statistical support for the trend itself, such as through bootstrapping or simulation, would seem most direct.

We have now included a bootstrap analysis of this relationship to demonstrate its significance.

Author Response:

Reviewer #2 (Public Review):

The authors benchmark their workflow and analyses using fairly well characterized compounds that are relatively potent against established targets. However, the authors appear to use significantly higher concentrations than the reported activity for these inhibitors and observe relatively few stabilized targets. Similarly, the corresponding measured induced-stabilization fold change at these concentrations often appear to be 1.5-2 fold. For example, SCIO-469 has reported in vitro potencies of ~10nM against MAPK14, ~100nM against MAPK11, with ~1000-fold selectivity over other kinases (including other MAPKs), and cell-based IC50s of ~300nM. However, the authors use 100 micromolar of SCIO-469 in their solvent-PISA profiling experiments, where they observe ~2-fold change for MAPK14 and ~1.5 fold changes for MAPK12 and MAPK9, and MAPK11 does not appear to be detected. This might suggest that solvent-PISA might not be sensitive to detecting stabilization to less-well developed compounds, decreasing its utility to identify targets of bioactive compounds that are less characterized/developed. It would be informative if the authors provided context for the concentrations of the small molecules that they use and provided some assessment of the sensitivity of this approach in regard to required compound potencies/target affinities.

We thank the reviewer for the rigorous assessment of our manuscript. We agree that the concentrations of compounds used in our experiments are much higher than the respective IC50s. Because any thermal or chemical stability measurement represents the average denaturation point of every copy of an individual protein in a proteome, the ability to detect a meaningful curve shift depends on saturating the target with the ligand. This means that the small molecule concentrations generally need to be high. This is even more important in lysate-based experiments, in which the cellular architecture is lost and the proteome is massively diluted. Moreover, the concentrations we employ are in line with similar studies (Zhang et al., Anal. Chem., 2020, 92, 1363-1371, Gaetani et al., J. Proteome Res., 2019, 18, 4027-4037, Savitski et al., 2014, Science, 346). We also agree that overall the fold changes of the PISA assay are relatively small (1.5-2 fold). In solvent-PISA experiments, the magnitude of the fold changes are not only affected by the ligand concentration and the shift in denaturation point, but also the denaturation point of the protein and the concentrations selected. We have investigated the relationship between the fold changes and these factors in the PISA assay in a former paper (J. Proteome Res. 2020, 19, 5, 2159–2166). We showed that the fold changes are inherently small in the classic PISA assay, which is determined by the nature of sigmoidal curves. We also showed that optimization of the selected temperatures ameliorated the issue (J. Proteome Res. 2020, 19, 5, 2159–2166). In this manuscript, we also showed in Figure 4C-E that a careful selection of the concentrations improved the magnitude of the fold changes in the solvent-PISA assay. Importantly, although the fold changes are relatively small (1.5-2 fold), TMT-based multiplexed quantitative proteomics is capable of analyzing both control and treated samples in multiple replicates in one experiment, which minimizes the variance and has robust sensitivity in detecting these fold changes.

Reviewer #3 (Public Review):

This is a highly interesting work providing an alternative method for drug target deconvolution for thermal proteome profiling. The experiments are thoroughly performed, and the conclusions are mostly supported by the obtained data. The only conclusion that needs further support is the one of the complementarity of SPP and TPP (as in "these two approaches share much in common, they remain distinct and likely serve to complement one another").

We appreciate that reviewer raised this point and would be happy to clarify this statement. Commonality – Both approaches rely of protein denaturation to determine target engagement. Furthermore, either approach could be used to screen ~70% of the proteome for ligand-induced changes in thermal- or chemical-stability. Lastly, both approaches follow a similar workflow, take approximately the same amount of time to prep and measure, and in the end generate similar data (Figure 3). Complementarity – We note that there are certain proteins that denature well in only a single condition and that combining the two approaches allows one to cover a greater fraction of the proteome than either approach, individually (Figure 5). Furthermore, the two approaches can also be used to corroborate one another. If one observes a ligand-induced thermal shift, for example, then SPP could be used to provide greater confidence in this hit and vice versa. It seems that our original statement was far too general and made with knowledge of data that had not yet been presented in the manuscript (mainly figure 5). We modified our conclusion in the main text to more accurately reflect the data presented specifically in Figure 3. The new conclusion is stated below and can be found on pages 10 and 11 of the revised manuscript: “Overall these data suggest that while these two approaches are capable of generating a similar set of putative targets, these lists are not completely identical. Therefore, SPP and TPP appear to be complimentary, not only because they can provide independent corroboration but because one method could potentially identify a target that the other might miss.”

Author Response:

Reviewer #1 (Public Review):

This manuscript begins with the larger notion that comparing item similarity is an important principle that guides human behavior, and that these similarity representations can have both a general and an idiosyncratic component. While individual-specific representations have been identified in some visual processing areas using personally meaningful object stimuli or simple stimulus features, this study looks at complex real-world stimuli with no personal meaning. They observe that even using the same stimuli across people, there are differences in how people judge the similarity of same-category items, and these differences correlate with performance on comparing the identities of images presented in sequence. Further, they examine the visual stream – specifically early visual areas like the early visual cortex (EVC) and lateral occipital complex (LOC), and late visual areas like the perirhinal cortex (PRC) and the anterior lateral entorhinal cortex (alErC), to see how representations in the brain relate to these behavioral representations. They observe that while EVC and LOC show correlations with behavior, PRC and alErC really show the strongest links to the individual-specific representation and to fine-grained differences across stimuli.

The analyses in this paper are very methodologically sound. They rely on well-controlled and well-tested analyses (e.g., testing if representational similarity is indeed higher for comparisons with one own's behavior, versus someone else's). They also replicate their results using classification-based analyses. My only key methodological question is more about experimental design. Given that participants performed the object arrangement task right before entering the scanner, I wonder if similarities in the brain to their own behavior could be due to memory for the representation they created just prior (especially given the role of PRC and alErC in memory). So, if instead, participants were shown and interacted with someone else's similarity arrangement, I wonder if these regions would show more similarity to that other person's arrangement, or still show similarity with one's own representations. It is thus currently unclear if the current findings are due to some deep-seated individual, internal representations, or memory for a recently performed task.

We thank the reviewer for highlighting numerous strengths in our methodological approach. We note that our experimental design was intentionally designed to have participants complete the iMDS sorting task prior to completion of the 1-Back task in the scanner. This ensured that all participants had the same exemplar familiarity during scanning. We cannot rule out the possibility that this order led to priming effects, as suggested by the Reviewer, which may have facilitated the emergence of observer-specific effects. Notably, however, such priming effects could be expected to affect all exemplars equally whose neural representations were probed during scanning. Critically, we also obtained new behavioural evidence for this resubmission, now included in Supplementary Materials, revealing that reports of perceptual similarity in our iMDS task reflect an observer characteristic that is temporally stable rather than just a situational idiosyncrasy. In the follow-up behavioural experiment (Supplementary Figure 2; page 49), “a distinct group of 30 participants completed two sessions of the iMDS task for the 10 object categories separated by 7 days +/- 1 day later. Correlations were computed between each participants’ perceived similarity Representational Dissimilarity Matrices (RDMs) from Session 1 and from Session 2. The mean within-subject correlation across the two sessions was 0.84, indicating high stability of participant’s perceived similarity ratings one week apart. Intersubject correlations for perceived similarity ratings across all exemplars and categories. Correlations were computed between each participant’s RDM Session 1 with the mean RDM (excluding the participant) in Session 2. Mean inter-subject correlation was 0.68. Critically, a paired t-test(intra-subject>inter-subject correlation) confirmed that intra-subject correlations were significantly higher than inter-subject correlations (p<0.0001). This pattern of results indicates that the perceived similarity structure that is unique to the individual observer is a stable characteristic.” This new behavioural experiment provides critical support for the perspective that our findings elucidate individual differences equivalent to those discussed as observer-specific effects in the vision literature more broadly (Mollon, Boston, Peterzell, & Webster, 2017, Individual differences in visual science: What can be learned and what is good experimental practice? Vision Research).

The results presented in this work are very clear and fit in well with previous findings on idiosyncracies in visual areas (Charest et al., 2014), and various work on the PRC as it relates to oddball tasks and object representational similarity. One question I am stuck with in this work is whether these current results show us something surprising or new. I'm unsure if we would have expected anything different for these generic real-world stimuli (versus the personally meaningful stimuli, or limited visual features tested previously).

We acknowledge that the paper by Charest et al., 2014 was an important stepping stone for the work presented in our manuscript. We have modified the framing of our main research questions so as to place more emphasis on levels (or grain) of perceived similarity among category exemplars that are reflected in subjective reports and object representations in different VVS regions. We also place more emphasis on the representational-hierarchical model as the theoretical framework that allows for related predictions and that guides our interpretation. In the interpretation of our results in the Discussion, we also speculate that observer-specific effects in fine-grained similarity perception, and in corresponding representations in PrC and alErC, may reflect interindividual differences in category expertise. Here we make reference to recent behavioural findings (i.e., Collins & Behrmann, 2020; Minos, Ferko, & Kohler, 2021) and present hypotheses about neural representations that can be directly tested in future fMRI studies with training paradigms. Indeed, we are in the process of planning to conduct such follow-up work in our laboratory. Inasmuch as the current study (i) revealed a mapping of perceived similarities among exemplars of object categories to representations in PrC and alErC (regions traditionally not considered to be part of the VVS and not included in analyses reported by Charest and colleagues); (ii) was guided by the representational-hierarchical model of VVS organization for interpretation of findings in PrC/alErC vs more posterior regions; (iii) showed an impact of observer-specific perceived similarities on behaviour that was most pronounced for fine-grained discrimination; and (iv) involved computational modeling to help interpret differences between observer-specific and observer-general (i.e., averaged) representations in different VVS regions, we feel that the contribution of our study clearly goes beyond replicating idiosyncrasies in VVS object representations as previously reported in Charest and colleagues’ pioneering work.

The manuscript frames the study around the idea that similarities are an important guiding principle of behavior. But this statement is not necessarily so obvious to me – is judging similarity itself an important ecological behavior, or is it just that looking at similarity structures can give insight into underlying relationships in how we represent information? (The latter is how I often see these sorts of representational similarity analyses.) What is this similarity task really capturing about our representations for these objects, and why do these idiosyncrasies emerge? My main hesitation about the current work is that I struggle with seeing a scope beyond a replication (e.g., finding behavior-correlated idiosyncracies in the brain, but with a different stimulus set, and in a slightly similar but expected region). I really want to know what factors are driving these idiosyncracies (e.g., is it visual? mnemonic? semantic?), and what this implies about the mechanisms of the PRC and ERC.

This is a really interesting point to consider in the context of prior lesion research on the role of PrC in perceptual discrimination. In this rich literature (see Murray et al., 2007 for review) emphasis has been placed on the perception of similarities between objects as probed with oddity-discrimination tasks, which require observers (human or nonhuman) to judge perceived similarities among multiple objects. Findings of behavioural impairments on this task after medial temporal-lobe lesions that included PrC and ErC have played a key role in the development of the representational-hierarchical model that guides the interpretation of our research. While the results of this lesion research have critically informed theoretical arguments that PrC plays a role in perceptual discrimination of objects (see Bonnen et al., Neuron, 2021, for a recent computationally focused review), it is important to recognize that they do not provide a characterization of similarity structure of neural representations in PrC and alErC, nor a characterization of the transformation of representations from more posterior VVS regions to these regions in the medial temporal lobe. Moreover, lesion findings do not address whether neural representations in the medial temporal lobe capture the perceived similarity structure that is unique to individual observers. Critically, our behavioural results also directly reveal effects of the perceived similarities that observers reported on discrimination performance in the 1-back task we employed during scanning. Inasmuch as this task taps discrimination between exemplars of real-world categories we would argue that the examination of representational similarity structure in our study also sheds lights on ecologically relevant behaviour.

Author Response:

Reviewer #1 (Public Review):

This study sought to systematically identify the components and driving forces of transcriptome evolution in fungi that exhibit complex multicellularity (CM). The authors examined a series of parameters or expression signatures (i.e. natural antisense transcripts, allele-specific expression, RNA-editing) concluding that the best predictor of a gene behavior in the CM transcriptome was evolutionary age.

Thus, the transcriptomes of fruiting bodies showed a distinct gene-age-related stratification, where it was possible to sort out genes related to general sexual processes from those likely linked to morphogenetic aspects of the CM fruiting bodies. Notably, their results did not support a developmental hourglass, which is the rather predominant hypothesis in metazoans, including some analysis in fungi.

The studies involved analyses of new transcriptomic datasets for different developmental stages (and tissue types in some cases) of Pleurotus ostreatus and Pterula gracilis, as well as the analyses of existing datasets for other fungi.

There are diverse interesting observations such as ones regarding Allele Specific Expression (ASE), suggesting that in P. ostreatus ASE mainly occurs due to cis-regulatory allele divergence, possibly in fast evolving genes that are not under strong selection constraints, such as ones grouped in youngest gene ages categories. In addition, a large number of conserved unannotated genes among CM-specific orthogroups highlights the rather cryptic nature of CM in fungi and raises as an important area for future research.

Some of the key aspects of the analyses would need to be better exemplified such as:

– Providing a better description of the developmentally expressed TFs only in CM species

– Providing clear examples of the promoter divergence that could be the underlying mechanism behind ASE. In particular, for some cases, there may be enough information in the literature/databases to predict the appearance or disappearance of relevant cis-elements in the promoters showing the highest divergence in genes depicting the highest levels of ASE.

We appreciate the constructive comments of the Reviewer and have revised the ms in accordance with the suggestions. In particular, we link different parts of the ms better to each other, provided a more detailed discussion of developmentally expressed TFs (lines 615-621). We also provide case studies of ASE genes with cis-regulatory divergence (Figure 5 and see below), although we note that these analyses are based on inferred and not directly determined motifs, so they should be considered as preliminary.

We had considered using TF binding motifs previously, and now we gave a try to analyzing potential transcription factor binding sites in divergent promoters. We find that there are no P. ostreatus transcription factors for which motifs based on direct evidence are available; rather, all P. ostreatus motifs are based on extrapolations from experimentally determined motifs (typically in Neurospora crassa). Therefore, to avoid too general motifs, we used only those where at least 5 nucleotides show at least 80% expected frequency in the PWM-s. This left us with 158 motifs (126 excluded). High motif binding score (>=4) and self-rate (>=0.9) were also required to ignore false positive hits. Different binding ability and lack of binding in one of the parental genomes were counted for each promoter. We found that genes with allele specific expression (ASE S2 and S4) show significantly higher differences in motif binding (lacking motifs, or different binding ability) than non-ASE genes (Fig. A1). These observations show that, not only promoter divergence, but differential predicted TF binding ability is also more common among ASE genes than among non-ASE genes. This supports our conjecture that ASE arises from cis-regulatory divergence.

*Fig A1: The left plot below shows the number of cases when the promoter of one allele of an allele pair in the two parent genomes has, but the other lacks a motif. The right plot shows the same in terms of difference in binding score.*

We could find examples, such as the allele specific expression of PleosPC15_2_1031042, a Hemerythrin-like (IPR012312) protein which might be regulated by the conserved c2h2 transcription factor, containing zinc finger domain of the C2H2 type (Fig. A2). C2h2 has already been proved to be important during the initiation of primordia formation with targeted gene inactivation (Ohm et al 2011, https://pubmed.ncbi.nlm.nih.gov/21815946/). A binding site of c2h2 was detected in the upstream region of PleosPC15_2_1031042. There is a mismatch in the inferred binding motif which causes a reduced binding score in PC15 (Fig. A2/c). Indeed the PC9 nuclei contribute better to the total expression of this gene.

Despite this, and other (not shown) examples that we have found, we were not convinced about the reliability of this approach. There are many assumptions in this analysis, the positional weight matrices (PWM) that we used, are all based on indirect evidence, high number of loci these PWMs identify, uncertainty in the position of binding site from transcriptional start site, relation of difference in binding motif and expressional changes. We consider these factors to potentially contribute too much noise to the analyses for these to be robust, therefore, we are hesitant to include these results in the ms.

*Fig A2: An example for promoter divergence a) expression of c2h2 transcription factor (TF) in P. ostreatus. b) allele-specific expression pattern of PleosPC15_2_1031042 from the two parental genomes. c) inferred binding motif of c2h2 TF and a detected potential binding site in the upstream region of PleosPC15_2_1031042 gene.*

Reviewer #2 (Public Review):

The evolution of complex multicellularity represents a major developmental reprogramming, and comparing related species which differ in multicellular structures may shed light on the mechanisms involved. Here, the authors compare species of Basidiomycete fungi and focus on analyzing developmental transcriptomes to identify patterns across species. Deep RNA-Seq data is generated for two species, P. ostreatus and Pt. gracilis, sampling different developmental stages. The authors report conflicting evidence for a "developmental hourglass" using a weighted transcription index vs gene age categories. There is substantial allele-specific expression in P. ostreatus, and these genes tend to have a more recent origin, have more divergent upstream regions and coding sequences, and are enriched for developmentally regulated transcripts. Antisense transcripts have low overlap with coding regions and low conservation, and a subset show a positive or negative correlation with the overlapping gene. Comparison to a species without complex multicellular development is used to further classify the developmental program.

Overall the new transcriptional data and extensive analysis provide a thorough view of the types of transcripts that appear differentially regulated, their age, and associated gene function enrichment. The gene sets identified from this analysis as well as the potential to re-analyze this data will be useful to the community studying multicellularity in fungi. The primary insights drawn in this study relate to the dating of the developmental transcriptome, however some patterns observed with young genes and noncoding transcripts are primarily reflective of expected patterns of evolutionary time.

We appreciate the Reviewer’s nice words on our ms, we think the revised version has substantial improvements in many aspects listed above.

Reviewer #3 (Public Review):

Fungi are unique in forming complex 3D multicellular reproductive structures from 2D mycelium filaments, a property used in this paper to study the genetic changes associated with the evolution of complex 3D multicellularity. The manuscript by Merenyi et al. investigates the evolution of gene expression and genome regulation during the formation of reproductive structures (fruiting bodies) in the Agaricomycetes lineage of Fungi. Transcriptome and multicellularity evolution are very exciting fundamental questions in biology that only become accessible with recent technological developments and the appropriate analysis framework. Important perspectives include understanding how genes acquire new functions and what role plays transcriptional regulation in adaptation. The study gathers a very useful dataset to this end, and relies on generally relevant hypotheses-driven analyses.

Analysis of fruiting body transcriptome in nine species revealed that prediction from the development hourglass model (that young genes are expressed in early and late but not intermediate phases of development) verified only in a few species, including Pleurotus ostreatus. An allele-specific expression (ASE) analysis in P. ostreatus showed that young genes frequently show ASE during fruiting body development. A comparative analysis with C. neoformans, which reproduces sexually without forming fruiting body, indicates that young and old (but not intermediate) genes are likely involved specifically in fruiting body morphogenesis. A number of underlying hypothesis could be presented better, and importantly the connection between the various analyses did not appear obvious to me. Some hypotheses and reasoning therefore need clarification. Some important results from the analyses are not provided and not commented, although they are required to fully meet the manuscript's objectives.

We appreciate the Reviewer’s suggestions and have revised the ms as explained below.

1. I do not clearly see the connection between the developmental hourglass model studied in the first part of the ms, and the allele-specific expression patterns in the second half of the ms. Which "phase" of the hourglass is expected to contain true CM-related genes (by contrast to general sexual processes)? Was P. ostreatus chosen for the ASE analysis because evidence for a developmental hourglass pattern was detected in this species? The conclusion that "evolutionary age predicts, to a large extent, the behaviour of a gene in the CM transcriptome" was established thanks to ASE in P. ostreatus, which was also found to be rather an exception for conforming to the hourglass model of developmental evolution. To what extent is this conclusion transferable to other Agaricomycete/fungal species?

We chose P. ostreatus because this is the only species for which the genomes of both parental strains (PC9 and PC15) are available. Although the hourglass concept is indeed a central hypothesis in animal developmental biology (though see recent critiques some (Piasecka et al 2013), our results suggest that it simply does not generally apply to fungal development. This may be due to the unique developmental mechanisms of fungi, or the independent origin(s) of CM in fungi. Our ms might have been misleading in this respect, in the revision we clarify that the ASE and hourglass analyses are independent of each other. Our interpretation of the hourglass results is that this model is not or hardly applicable for fungal development and the fact that P. ostreatus was the only species that in fact showed an hourglass did not drive our selection of this species. We inserted a note on this in the ms.

1. The authors acknowledge that fruiting body-expressed genes may relate either to CM or to more general sexual functions, and that disentangling these functions is a major challenge in their study. An overview of which gene was assigned to which function is not explicit in the ms (proposed to be described in a separate publication). Do these functional gene classes show distinct transcriptome evolution patterns (hourglass model, ASE...)?

We made accessible the complete list of CM-related genes and genes with more general sexual functions in Table S2/b-c. Due to length restrictions, we do not discuss many or each of these genes here, but provided gene ontology-based overviews (Fig 8/c-d, from lines 631). To answer the question whether CM vs shared genes show distinct transcriptomic patterns, we analyzed ASE, NATs and the hourglass model separately for CM-specific and shared genes. as follows:

-hourglass: We calculated and visualised the TAI for CM-specific and Shared gene sets of P. ostreatus separately. The average value of TAI decreased a lot in Shared genes possibly due to the overrepresentation of ancient genes here, but the patterns remained similar to the original, which imply that not simply one or the other gene set drives these patterns (Fig A3).

*Fig A3: Transcriptome Age Index for CM-specific and Shared gene sets of P. ostreatus separately*

-ASE: As we detailed in the ms, allele specific expression occurs mainly in young genes. Indeed, only 13.1% of ASE genes belong to the conserved gene sets (CMspecific: 200 and Shared: 144). Although there are more ASE genes (>2FC) among CM-specific genes, they are still underrepresented compared to young genes that are neither shared, nor CM-specific. This indicates that ASE is generally a feature of non-conserved genes and is not particularly characteristic for either conserved or CM-specific genes.

-NAT: We found that 17.3% of CM-specific (141 genes) and 18.3% of Shared genes (165 genes) overlap with antisense transcripts. Since these numbers don't differ substantially from 17.6%, which is the proportion of NATs corresponding to all protein coding genes, it implies an independent occurrence between NATs and these gene conservation groups.

3.) As far as I understand, major functions of the fruiting body transcriptome are either CM or general sexual functions. Could these genes, notably those showing ASE, play a role in general processes other than sexual development (hyphal growth, environment sensing, cell homeostasis, pathogenicity)?

Certainly, ASE might also occur in genes related to these processes. However, the processes mentioned by the Reviewer are likely associated with very conserved genes (except pathogenicity, which we can’t examine here) and our results suggest that ASE is more typical of young genes that are under weak selection. We detected ASE in 931/343 (S2/S4 genes) genes expressed in the vegetative mycelium stage of P. ostreatus. We also note that by the definition of developmentally regulated genes, we do not expect very basic fungal processes, like hyphal growth to be among the functions of the genes we identified. Genes related to such basic (housekeeping) processes usually (exceptions exist) show flat expression profiles (because they are equally important in mycelia and all fruiting body stages) and will not be picked up by our pipelines for identifying shared developmentally regulated genes.

1. As stated by the authors, "the goal of this study was to systematically tease apart the components and driving forces of transcriptome evolution in CM fungi". What drives the interesting ASE pattern discovered however remains an open question at the end of the ms. The authors appropriately discuss that these patterns could be either adaptive or neutral but there is no direct evidence for any scenario in P. ostreatus. Is the expression of (some of) the young genes showing ASE required for CM? one or two case studies would allow providing support for such scenarios.

We respectfully disagree. We provide evidence that the driving force of ASE is promoter divergence (and consequently differential transcription factor binding) in genes in which it is tolerated (see conclusions, lines 708-712). Our results suggest that ASE is mostly a neutrally arising phenomenon. To get to the mechanistic bases of how promoter divergence can cause ASE (following the suggestion of Reviewer 1), we analysed putative, inferred transcription factor binding motifs in P. ostreatus and found that ASE genes had more divergence in putative TF binding sites. However, it is important to emphasise that all TF motifs we analyzed are inferred motifs and therefore these results are indicative at best.

Reviewer #4 (Public Review):

This work develops a comparative framework to test genes which support complex morphological structures and complex multicelluarity. This expands beyond simple gene sharing and phylogenomics by incorporating comparison of gene expression profiling of development of multicellular structures during sexual reproduction. This approach tests the hypothesis that genes underlying sexual reproductive structure formation are homologous and the molecular evolutionary processes that control transcriptome evolution which underlie complex multicellularity.

The approaches used are appropriate and employ modern comparative and transcriptome analyses to example allele specific expression, and evaluate an age of the evolutionary ages of genes. This work produced additional new RNAseq to examine developmental processes and combined it with existing published data to contrast fungi with either complex morphologies or yeast forms.

The strengths of work are well selected comparison organisms and efforts to have developmental stages which are appropriate comparisons.

We appreciate the Reviewer’s positive comments.

Weakness could be pointed to in how the NAT descriptions are interesting it isn't clear how they link directly to morphology variation or development. I am unclear if these are arising from new de novo promotors, are ferried by transposable elements, or if any other understanding of their genesis indicates they are more than very recent gains in a species for the most part and not part of any conserved developmental process (outside a few exemplars).

Originally, we assayed natural antisense transcripts (NAT) based on the assumption that they regulate developmental processes (e.g. Kim et al 2018 https://doi.org/10.1128/mBio.01292-18). Our analyses showed that although NATs are abundant in CM transcriptomes of fungi, they show no homology across species and so are unlikely to drive conserved developmental processes, which we are after in this ms. Rather, our data are compatible with most (but likely not all) NATs being transcriptional noise, arising from novel or random promoters. We therefore shortened this section and moved much of it to the Appendix 1.

The impact of this work will reside in how gene age intersects with variability and relative importance in CM. it will be interesting to see future work examine the functions of these genes and test how allele specific expression and specific alleles are contributing to the formation of these tissues and growth forms. I am still not sure if molecular mechanisms of how high variability in gene expression is still producing relatively uniform morphologies, or if it isn't quantification of morphological variation would be nice to link to whether ASE underlie that.

We agree that allele specific expression could influence morphologies significantly, but investigating that is beyond the scope of the current work (it would require a population genomics project). More direct evidence on allelic differences can be seen in monokaryon phenotypes, which only express one of the parental alleles. Phenotypic differences are obvious in the mycelium of the two parental monokaryons : the mycelium of PC9 is more fluffy and grows faster than that of PC15. This was reported recently by Lee et al 2021 (https://doi.org/10.1093/g3journal/jkaa008). We agree with the Reviewer that this is a very exciting future research direction.

To my read of the work, the authors achieved their goals and confirmed hypothesis about the age of genes and the variability of gene expression. I still feel there is some clarity lost in whether the findings across the large number of species compared here help inform predictions or classifications of types of genes which either have ASE or are implicated in CM. This is really work for the future as the authors have provided a detailed analysis and approach that can fuel further direction in this research area.

To address this issue we reworked the ms to make connections between ASE and CM clearer. Because ASE appears based on our results to (mostly) arise neutrally, predictions for other species are expected to be hard. On the other hand, we think we can make confident predictions on what types of genes are implicated in CM in other species, at least for conserved aspects of fruiting body development.

Author Response:

Reviewer #2 (Public Review):

[...] The strength of such an impressive and comprehensive analysis of large collection of nanobodies lies in the comparison with existing nanobodies. To fully benefit from the publication of this latest collection of nanobodies, the authors should publish all the sequences and have to make the best efforts to provide comparisons with existing nanobodies described in the literature:

-Values obtained for neutralization potency differ substantially between different techniques and labs. A good reference point for neutralization data is the use of ACE2-Fc, which is commercially available and widely used in earlier publications. It is hard to compare the described nanobodies with the control nanobodies from the literature mentioned (Wrapp et al., Xiang et al.), as they are not identified in detail. Differences between the potency described in the original description and the values determined here are not discussed.

-Epitope mapping defines a new list of epitope groups, although similar efforts had been undertaken earlier. It would make the comprehensive list of SARS-COV-2 nanobodies even more helpful if information about representative existing nanobodies targeting the same epitopes is included throughout the paper (in particular taking advantage of determined nanobody-Spike structures). Comparison to nanobodies with structural information would permit more meaningful predictions with regards to the mechanism of action and help explain the synergistic behavior observed.

-The manuscript substantially refers to previous antibody publications (with important contributions from the institution of the last author). Yet, important earlier publications of SARS-CoV-2 nanobody are barely mentioned/discussed. Many of these publications defined structures and epitopes, contributed to an understanding of spike activation, and determined modes of neutralization. These publications should be appropriately acknowledged.

A comprehensive review of this topic would certainly be extremely valuable, but is beyond the scope of this work. However, we have added additional nanobody citations as requested where appropriate throughout the text.

Author Response:

Reviewer #1:

The manuscript entitled, "Early evolution of beetles regulated by the end-Permian deforestation" by Zhao et al. is a strong, interesting, and well-written study worthy of publication after revision.

The authors met their goal of documenting and analyzing the diversity of Paleozoic beetle taxonomy, morphological disparity, ecosystem roles, and phylogeny. This, in my opinion, is the strongest portion of the paper as it brings several lines of evidence to show the high diversity of xylophagous beetles, up until the EPME, followed by a distinct extinction of xylophagous beetles and the expansion of ecological roles into a more modern component of beetles.

A distinct weakness of the paper is the reliance of correlation between biochemical cycling and the evolution of beetles. To address this, we would ideally see isotopic data associated with these statements. My overall suggestion is to make clear that this is speculative and bring other hypotheses to the table, and hopefully rule them out. It isn't very helpful to say something like xylophagous beetles were the main source of nutrient cycling in the Permian without discussing fungus at greater length. Or similarly, implying a drop in O2 was caused by beetles, without describing any of the other biotic/abiotic things going on at that time.

We really appreciate those insightful comments, and completely agreed that the correlation between carbon cycling and Permian beetles is speculative. We followed the reviewer’s suggestion and toned down the discussion about this correlation. The role of ancient insects in deep-time forest carbon cycle is unclear, partly because the contribution of extant insects to the decomposition of deadwood is poorly understood. Fortunately, a paper published last month reveals the functional importance of insects in the decomposition of deadwood and the forest carbon cycle (Seibold et al., 2021), and thus provide a further support for our conclusion. We added this reference to our paper. Moreover, regarding the Permian biochemical cycling change, we also have added an introduction about other two hypotheses (reduction in the extent of coal swamps and the evolution of lignin-consuming fungi) and ruled them out in the Discussion.

Please see lines 237–249:

“The oxygen concentration of the atmosphere began to rise in the early Palaeozoic, probably with a peak in the Carboniferous and large decline from the beginning of the Permian (Dahl et al., 2010; Berner, 2009; Krause et al, 2018). The reason for this plunge was attributed to a tectonic- or climate-driven reduction in the extent of coal swamps (Berner and Canfield, 1989) or to the evolution of lignin-consuming fungi (Floudas et al., 2012). However, global recoverable coal is only equivalent to a few percent of the oxygen budget in the atmosphere, and thus cannot account for the large drop of atmospheric oxygen (Nelsen et al., 2016). Furthermore, lignin-consuming fungi may have been present before the Carboniferous (Nelsen et al., 2016). Recently, a new geochemical model proposed that the development of Permian terrestrial herbivores may have limited transport and long-term burial of terrestrial organic compounds in marine sediments, resulting in less organic carbon burial and attendant declines in atmospheric oxygen (Laakso et al., 2020).”

Please see lines 261–264:

“In extant forest ecosystems, insects may account for 29 percent of the total carbon flux from deadwood and thus they have a functional importance in the decomposition of deadwood and the carbon cycle (Seibold et al., 2021).”

Please see lines 266–269:

“Permian beetles had probably evolved close interactions with various microorganisms especially lignin-consuming fungi (Nelsen et al., 2016), which also accelerated the decomposition of deadwood.”

We have added 7 references.

Berner RA. 2009. Phanerozoic atmospheric oxygen: new results using the GEOCARBSULF model. American Journal of Science 309: 603–606.

Dahl TW, Hammarlund EU, Anbar AD, Bond DPG, Gill BC, Gordon GW, Knoll AH, Nielsen AT, Schovsbo NH, and Canfield DE. 2010. Devonian rise in atmospheric oxygen correlated to the radiations of terrestrial plants and large predatory fish. PNAS 107: 17911–17915.

Krause AJ, Mills BJW, Zhang S, Planavsky NJ, Lenton TM, Poulton SW. 2018. Stepwise oxygenation of the Paleozoic atmosphere. Nature Communications 9: 4081.

Berner RA, Canfield DE. 1989. A new model for atmospheric oxygen over Phanerozoic time. American Journal of Science 289: 333–361.

Nelsen MP, DiMichele WA, Peters SE, Boyce CK. 2016. Delayed fungal evolution did not cause the Paleozoic peak in coal production. PNAS 113: 2442–2447.

Floudas TD, Binder M, Riley R, Barry K, Blanchette RA, Henrissat B, Martínez AT, Otillar R, Spatafora JW, Yadav JS, Aerts A, Benoit I, Boyd A, Carlson A, Copeland A, Coutinho PM, Vries RPD, Ferreira P, Findley K, Foster B, Gaskell J, Glotzer D, Górecki P, Heitman J, Hesse C, Hori C, Igarashi K, Jurgens JA, Kallen N, Kersten P, Kohler A, Kües U, Kumar TKA, Kuo A, Labutti K, Larrondo LF, Lindquist E, Ling A, Lombard V, Lucas S, Lundell T, Martin R, Mclaughlin DJ, Morgenstern I, Morin E, Murat C, Nagy LG, Nolan M, Ohm RA, Patyshakuliyeva A, Rokas A, Ruiz-Dueñas FJ, Sabat G, Salamov A, Samejima M, Schmutz J, Slot JC, John FSt, Stenlid J, Sun H, Sun S, Syed K, Tsang A, Wiebenga A, Young D, Pisabarro A, Eastwood DC, Martin F, Cullen D, Grigoriev IV, Hibbett DS. 2012. The Paleozoic origin of enzymatic lignin decomposition reconstructed from 31 fungal genomes. Science 336: 1715–1719.

Seibold S, Rammer W, Hothorn T, Seidl R, Ulyshen MD, Lorz J, Cadotte MW, Lindenmayer DB, Adhikari YP, Aragón R, Bae S, Baldrian P, Varandi HB, Barlow J, Bässler C, Beauchêne J, Berenguer E, Bergamin RS, Birkemoe T, Boros G, Brandl R, Brustel H, Burton PJ, Cakpo-Tossou YT, Castro J, Cateau E, Cobb TP, Farwig N, Fernández RD, Firn J, Gan KS, González G, Gossner MM, Habel JC, Hébert C, Heibl C, Heikkala O, Hemp A, Hemp C, Hjältén J, Hotes S, Kouki J, Lachat T, Liu J, Liu Y, Luo YH, Macandog DM, Martina PE, Mukul SA, Nachin B, Nisbet K, O’Halloran J, Oxbrough A, Pandey JN, Pavlíček T, Pawson SM, Rakotondranary JS, Ramanamanjato JB, Rossi L, Schmidl J, Schulze M, Seaton S, Stone MJ, Stork NE, Suran B, Thygeson AS, Thorn S, Thyagarajan G, Wardlaw TJ, Weisser WW, Yoon S, Zhang NL, Müller J. 2021. The contribution of insects to global forest deadwood decomposition. Nature 597: 77–81.

Below are some more detailed suggestions:

-It would be helpful to address the evolution of lignin-consuming fungi. Whether or not you can tie fungal symbiosis into the evolution of these beetles, fungal decomposition may (or may not) have accelerated in the Early Permian due to the timeline of particular clades of fungi. Worth a quick sentence or two. See relevant references below.

Nelsen, M.P., DiMichele, W.A., Peters, S.E. and Boyce, C.K., 2016. Delayed fungal evolution did not cause the Paleozoic peak in coal production. Proceedings of the National Academy of Sciences, 113(9), pp.2442-2447.

Floudas D, et al. (2012) The Paleozoic origin of enzymatic lignin decomposition reconstructed from 31 fungal genomes. Science 336(6089):1715-1719.Abstract/FREE Full TextGoogle Scholar

Kohler A, et al., Mycorrhizal Genomics Initiative Consortium (2015) Convergent losses of decay mechanisms and rapid turnover of symbiosis genes in mycorrhizal mutualists. Nat Genet 47(4):410-415.CrossRefPubMedGoogle Scholar

Thank you very much for pointing out this issue. We complete agreed with the reviewer that Permian beetles had probably evolved close interactions with fungi, which also accelerated the decomposition of deadwood. We have revised the text and added the references based on the reviewer’s comments and suggestions. Please see comment 1.

For the concluding paragraph in the Discussion, there is no acknowledgment to modern studies of xylophagous beetles in relation to climate change. There are many studies of the effects on climate change and xylophagous beetles, ex the North American pine bark beetles. Might be worth saying that the diversity and abundance of xylophagous beetles are extremely sensitive to climate change and can cause forest collapse too.

Thank you. We have added a sentence and two new references about extant xylophagous beetles and climate change in the concluding paragraph. Please see lines 328–330:

“In particular, the diversity and abundance of xylophagous beetles are extremely sensitive to climate change and can also cause forest collapse and carbon cycle disturbance (Kurz et al., 2008; Fei et al., 2019; Šamonil et al., 2020).”

Kurz WA, Dymond CC, Stinson G, Rampley GJ, Neilson ET, Carroll AL, Ebata T, Safranyik L. 2008. Mountain pine beetle and forest carbon feedback to climate change. Nature 452: 987–990.

Šamonil P, Phillips JD, Pawlik Ł. 2020. Indirect biogeomorphic and soil evolutionary effects of spruce bark beetle. Global and Planetary Change 195: 103317.

Author Response:

Reviewer #1:

This study aims to find the genetic mechanisms underlying sex-ratio distortion through male-killing in Drosophila melanogaster flies infected with the endosymbiont Wolbachia. The endosymbiont carries the prophage WO, which is in the center of interested in this study. The key result of this study is that a synonymous mutation in a prophage gene can explain the differences between sex-ratio distorting and not distorting symbionts. The study uses transgene technology to modify phage genes and to investigate which changes in the gene is involved in the phenotype. The finding, that a synonymous SNP plays a key role is not entirely novel in biology, but there are only few examples known of this type of genotype - phenotype associations. The study does not include experiments to show that the main finding is not limited to one particular background of the fly line used. An experiment including multiple genotypes would be needed to show this.

We agree that recapitulating the results in other backgrounds is intriguing and important for establishing a broader role of these findings. We thank the Reviewers and Editor for allowing us to pursue this line of investigation separately from this work, and we now discuss what experiments can be completed to answer these and other questions. We also edited the manuscript to tone down any conclusions that would imply generalizability of the findings at this point. For example:

"For example, we cannot conclude that the particular codon tested here is responsible for phenotype alterations in other host genetic backgrounds or species. It is possible that this codon plays a functional role only in a singular host genetic context. Here, we changed wmk sequences while holding the host genetic background fixed, but the reverse is required to conclude whether or not the particular codon plays a general role in other genotypes or natural contexts. Second, due to possible coevolution, various codons may or may not yield similar functional effects across different host backgrounds, and additional synonymous sites may contribute to the male-killing phenotype. Thus, the results here illuminate a previously unrecognized need for future research on the functional impacts of synonymous substitutions in endosymbionts. Future work may focus on determining if there is one specific synonymous codon that affects the male-killing function in all cases, if a more general feature exists where alteration of any or a subset of N-terminal or other wmk codons affects function, or if the effect of synonymous changes is specific to this background.”

Text summarizing the 06/21/2021 query to the Editor and Reviewers for further clarification: We believe there are several reasons why the results can stand on their own, while appropriately acknowledging caveats. First, we note the lack of genetic background testing on previous transgene experiments driving the major discoveries of Wolbachia genes involved in reproductive parasitism. This requirement would therefore hold the current work to a novel bar not previously applied by the field. In addition, the genetic background here is the same as used in previous work on these phenotypes, making it the most pertinent to test and inform previous and ongoing studies by many research groups. Second, the results shown here would still stand no matter the results of genetic background testing and would demonstrate that it is possible for synonymous changes to have functional relevance in the transgenic wmk phenotype. The major findings are still novel in the field, relevant to ongoing studies of reproductive parasitism, and informative regarding one of the most common genetic backgrounds. Finally, we note that two different lines with unique synonymous codon changes (the final experiment) independently created the same result that a synonymous codon change ablates phenotype, providing additional robustness to our findings. Doing additional experiments would be logistically difficult. Barriers include the relocation of the first author of the work to another lab for a postdoctoral position, completion of the funding for the project, remaining institutional COVID-19 restrictions, and lack of replacement personnel in the lab to continue the work. Notably, there is also the non-trivial requirement to create and test new transgene lines that would be costly and take nearly a year to complete (the experiments in the manuscript already took several years and the new fly lines would cost thousands to make).

The study is mostly clear and easy to follow, but requires a lot of attention. The authors choose to build up the story as I guess it was carried out in the lab. Thus, the reader is guided through every step of the process. While I see that this is appealing from the way the study was carried out, it results in a very long manuscript with a lot of material that would be much better placed in a supplement.

We thank the reviewer for pointing this out. We shortened the manuscript by removing redundant information and transferring some parts of the results to the supplement. We also removed about three pages of text from the discussion (before adding in new sections as requested by reviewers).

The introduction seems unfocused. It meanders around, jumping from topic to topic and does not give the reader a sense of where things will go.

We added a few topics into the Introduction as recommended in other comments, and we edited various portions of the Introduction to connect the ideas together more clearly. We hope the changes are now satisfactory, and we are of course happy to consider further feedback.

Fig. 1 gives an overview about the different aspects addressed here, but it is not used to guide the reader through the different lines of thought addressed in the introduction. If Fig. 1 will stay (I actually think it is not needed) it should be introduced earlier and used as a road map for the paper. Alternatively, the introduction could stay more general and only in the last paragraph the different ways the system is studied will be summarized.

We edited the final paragraph of the Introduction to more comprehensively cover the content of the figure and full direction of the paper. For readers not familiar with the biological system or questions, we believe this figure will serve as a gateway to the genetic alterations conducted in the experiments.

Along these lines, it would be good to have a better reasoning for the combination of experiments conducted. It is left to the reader to understand why certain types of experiments have been done.

It was not clear to us at the outset of these experiments what results would ultimately emerge and what follow-up experiments would be necessary as our initial hypotheses were proven wrong with many of the surprises from the work. So, there was no a priori reasoning for why experiments were done until we had the results of the previous experiments. We agree that this makes the reading a bit confusing. As such, we clarified the logic flow in the results section as the narrative progresses from experiment to experiment, and we reorganized some of the introduction to improve transition statements and offer a roadmap to readers earlier on.

On the other hand, the introduction misses a section on the biology of the phage and its interaction with the host(s). It is hard to understand the biology of the system without getting an understanding of the insect - Wolbachia - phage interactions. For non-specialist, understanding the role of the three players is essential for the system.

Thank you for the suggestion. We now add a section introducing phage WO and its relevance to the phenotypes tested here.

“The wmk gene and two cytoplasmic incompatibility factor (cif) genes that underlie cytoplasmic incompatibility (a parasitism phenotype whereby offspring die in crosses between infected males and uninfected females) occur in the eukaryotic association module (EAM) of prophage WO, which refers to the phage WO genome that is inserted into the bacterial chromosome. The EAM is common in WO phages across several Wolbachia strains and is rich in genes that are homologous to eukaryotic genes or annotated with eukaryotic functions. As such, the expression of reproductive parasitism genes from the EAM and tripartite interactions between phage WO, Wolbachia, and eukaryotic hosts are central to Wolbachia’s ability to interact with and modify host reproduction.”

The result section could be easily shortened by focusing on the essential experiments. Experiments that do not contribute to the final result can go into the supplement.

We removed redundant sentences and made some figures supplemental.

Also the discussion is much too long. I suggest to reduce it to half and focus on the important points and the take-home messages. Currently the discussion follows the way the results are presented in the result section. However, this is not needed. The important finding should be discussed first. Findings that are important in the development of the project, may not be important for the biology of the system overall. And they may not be important for the reader.

We reordered the discussion to cover the biggest findings first, and removed about a third of the original writing in the discussion.

Reviewer #2:

This study aims to unravel the genomic basis to wmk-induced male killing by transgenically expressing homologs of varying relatedness, with synonymous nucleotide changes, and predicted alternative start codons in D. melanogaster flies. The study builds on previous work showing that expression of wmk in fly embryos recapitulates several aspects of male killing. While more distantly related homologs did not induce male killing when expressed in D. melanogaster, more closely related wmk homologs induce either killing of both sexes or male killing only. However, the male-killing phenotype was not due to amino acid differences, but associated with RNA structural differences of the different wmk homologs. In addition, only one synonymous nucleotide change was sufficient to ablate the killing phenotype. These findings suggests that minor and even silent nucleotide differences impact on the expression of male killing in D. melanogaster. It is concluded that a new model incorporating the impacts of RNA structure and post-transcriptional processes in wmk-induced male killing needs to be developed.

The strength of the study lies in the systematic and carefully controlled approach to quantify the phenotypic effects of both sequence and structural changes to various wmk homologs for inducing the male-killing phenotype. Detailed dissection of the phenotypic impact of minor changes to the wmk homologs including sequence variation, silent nucleotide changes, and RNA structural differences was quantified. This approach reveals a complex genotype-phenotype relationship, but highlights the importance of including post-translational processes. The data is novel in that previous work have largely ignored structural changes and assumed that synonymous differences in codons has no effect on protein function, whereas the current study based on updated codon optimization algorithms reveal that this assumption is incorrect. The finding highlights the importance of considering also structural genetic variation for phenotypic expression differences. This suggestion is further corroborated by the lack of difference in wmk homologue expression levels, indicating that the functional differences are due to post-translational effects.

We thank the reviewer for the thoughtful comments.

There are limitations to the findings of this complex genotype-phenotype relationship. The current study only examined the phenotypic impact by expressing the different homologs in one D. melanogaster genetic background. Given the variability of the phenotypic pattern revealed based on minor changes to the wmk homologs, it will be critical to repeat some of the main findings in other D. melanogaster genotypes to determine the importance of the variation in the wmk homologs more generally. It is entirely plausible that the observed changes in the effect and strength of killing is due to an interaction between host and wmk genotype. This has implications for unravelling the underlying genetic basis to the male-killing phenotype more widely. It is as yet to be demonstrated whether wmk is involved in male killing in natural population, and to what extent there are shared patterns and mechanisms of male killing induced by other bacterial endosymbionts such as Spiroplasma.

We addressed this point in more detail above in the first response to the comments from Reviewer 1.

Author Response:

Reviewer #1 (Public Review):

Garcia-Souto, Bruzos, and Diaz et al. analyzed hemic neoplasia in warty venus clams at multiple sites throughout Europe. They identified cases of disease in two locations, in Galicia and in the Mediterranean. They then use Illumina sequencing to discover that the samples with cancer DNA had reads which mapped to the mtDNA reference sequences from a different clam species in the same family, suggesting a cross-species transmissible cancer. By mapping reads to both the V. verrucosa and C. gallina mitogenomes they showed that more reads mapped to C. gallina in cancer samples compared to matched host tissue samples, and this was consistent across the whole mitogenome. Phylogenetic analysis of mtDNA genes of the host and cancer samples as well as identification of SNVs at a short region of one single-copy nuclear locus suggest that all cancer samples come from a single C. gallina transmissible cancer clone. All data agree that a single lineage of cancer from C. gallina is responsible for all identified cancers in V. verrucosa.

There are a few sections where there are either unclear methods or the methods do not quite match the descriptions of the results.

1. Regarding mapping of reads to different reference Cox1 sequences (for Figure 2a): "Then, we mapped the paired-end reads onto a dataset containing non-redundant mitochondrial Cytochrome C Oxidase subunit 1 (Cox1) gene references from 137 Vererid clam species." I do not see where this is explained anywhere in the methods, where this list of references comes from, or what is in it.

Answer: We retrieved a dataset of 3,745 sequences comprising all the barcode-identified venerid clam Cox1 fragments available from the Barcode of Life Data System (BOLD, http://www.boldsystemns.org/). Redundancy was removed using CD-HIT (Fu, et al. 2012), applying a cut-off of 0.9 sequence identity, and sequences were trimmed to cover the same region. Whole-genome sequencing data from both healthy and tumoral warty venus clams was mapped onto this dataset, containing 118 venerid species-unique sequences, using BWA-mem, filtering out reads with mapping quality below 60 (-q60) and quantifying the overall coverage for each sequence with samtools idxstats. PCR primers were designed with Primer3 v2.3.7 (Koressaar, et al. 2018) to amplify a fragment of 354 bp from the Cox1 mitochondrial gene of V. verrucosa and C. gallina (F: CCT ATA ATA ATT GGK GGA TTT GG, R: CCT ATA ATA ATT GGK GGA TTT GG). PCR products were purified with ExoSAP-IT and sequenced by Sanger sequencing.

Action: We have included this new information in the methods section.

1. Regarding de novo assembly of mitogenomes: "Hence, we employed bioinformatic tools to reconstruct the full mitochondrial DNA (mtDNA) genomes in representative animals from the two species involved....Then, we mapped the paired-end sequencing data from the six neoplastic specimens with evidence of interspecies cancer transmission onto the two reconstructed species-specific mtDNA genomes." In contrast to this, the methods say, "Then, we run MITObim v1.9.1 (Hahn, Bachmann, & Chevreux, 2013) to assemble the full mitochondrial genome of all sequenced samples, using gene baits from the following Cox1 and 16S reference genes to prime the assembly of clam mitochondrial genomes." It is unclear which method was used.

Answer: In total, we performed whole-genome sequencing on 23 samples from 16 clam specimens, which includes eight neoplastic and eight non-neoplastic animals by Illumina pairedend libraries of 350 bp insert size and reads 150 bp long. First we assembled the mitochondrial genomes of one V. verrucosa (FGVV18_193), one C. gallina (ECCG15_201) and one C. striatula (EVCS14_02) specimens with MITObim v1.9.1 (Hahn, et al. 2013), using gene baits from the 7 following Cox1 and 16S reference genes to prime the assembly of clam mitochondrial genomes: V. verrucosa (Cox1, with GenBank accession number KC429139; and 16S: C429301), C. gallina (Cox1: KY547757, 16S: KY547777) and C. striatula (Cox1: KY547747, 16S: KY547767). These draft sequences were polished twice with Pilon v1.23 (Walker, et al. 2014), and conflictive repetitive fragments from the mitochondrial control region were resolved using long read sequencing with Oxford Nanopore technologies (ONT) on a set of representative samples from each species and tumours. ONT reads were assembled with Miniasm v0.3 (Li 2016) and corrected using Racon v1.3.1 (Vaser, et al. 2017). Protein-coding genes, rDNAs and tRNAs were annotated on the curated mitochondrial genomes using MITOS2 web server (Bernt, et al. 2013), and manually curated to fit ORFs as predicted by ORF-FINDER (Rombel, et al. 2002). Then, we employed the entire mitochondrial DNAs of V. verrucosa (FGVV18_193) and C. gallina (ECCG15_201) as “references” to map reads from individuals with neoplasia, filter reads matching either mitogenome and assemble and polish their two (healthy and tumoral) mitogenomes individually as above. Further healthy individuals were later sequenced and their mitogenomes assembled, to further investigate the geographic and taxonomic spread of this neoplasia.

Action: We have included this information in the methods section (page 21-22), and in the results (pages 7 and 8). mtDNA annotations are now shown in Supplementary Figure 3. Nucleotide data for the mitochondrial DNA assemblies has been uploaded to GenBank under accession numbers MW662590-MW662611 and will be released upon publication or request.

There is one minor claim which may not be fully supported by the data: the statement that, "The analysis of mitochondrial and nuclear gene sequences revealed no nucleotide divergence between the seven tumours sequenced." If I am understanding the filtering of the SNVs from the nuclear gene correctly, only the presence or absence of the 14 SNVs that were fixed within each of the two species were analyzed. Therefore, it is unclear whether the authors looked for any additional somatic mutations within the cancer lineage that would have occurred at other positions. For mitochondria, the authors state that sequences were "extracted from paired-end sequencing data," but it is not explained how this was done. The data suggest that there are no differences between cancer samples in the 13 coding genes and 2 rDNA genes, but data on possible SNVs in the intergenic regions is not shown.

Answer: We obtained a preliminary nuclear assembly using short-reads only. Obviously, the resulting assemblies are fragmented and incomplete. This has limited the identification of candidate regions shared by the three genomes (V. verrucosa and both Chamelea clams). Out of the 44 candidate nuclear fragments we tested, only two (DEAH12 and TFHII) turned out to give good PCR products, adequate for Sanger sequencing. As mentioned above, we now provide additional data on a second gene (TFIIH), identified and selected on the same basis as DEAH12. We find 14 and 15 sites, respectively, for the DEAH12 and the TFIIH loci, with fixed SNVs (allele frequency >95%) that allowed to discriminate between the three relevant species (V. verrucosa, C. gallina and C. striatula) and the tumour. These diagnostic nucleotides were then used to filter the reads from individuals with neoplasia harbouring both DNA’s. Variation within the host lineage but not within the tumour was found along the nuclear DNA fragments employen in the ML phylogenies (see figure below).

Figure. Molecular phylogenies based on the two selected nuclear markers. (a) DEAH12 gene and (b) TFIIH gene, and diagnostic loci discriminating among species and tumour. Bootstrap support values (500 replicates) from ML analyses above 50 are shown above the corresponding branches. Note all diagnostic nucleotides are identical between tumours (black dots).

Regarding the mtDNA, firstly, we assembled the mitochondrial genomes of one V. verrucosa (FGVV18_193), one C. gallina (ECCG15_201) and one C. striatula (EVCS14_02) specimens with MITObim v1.9.1 (Hahn, et al. 2013). Then, we employed the entire mitochondrial DNAs from V. verrucosa (FGVV18_193) and C. gallina (ECCG15_201) as “references” to map reads from individuals with neoplasia, filter reads matching either mitogenome and assemble and polish their two (healthy and tumoral) mitogenomes individually as above. Further healthy individuals were later sequenced and their mitogenomes assembled, to further investigate the geographic and taxonomic spread of this neoplasia. Despite the usefulness of the mitochondrial control region (CR) to detect differences among lineages, we refrained from using it for two reasons. (1) The CR shows considerable variation in both length and sequence among the three species, making their alignment difficult (in fact, previous phylogenetic studies based on whole mitochondrial DNA sequences in Veneridae excluded the CR: https://doi.org/10.1111/zsc.12454), and (2) the CR contains quasi-but-not-identical tandem repeats, as a other mollusks (i.e., the Venerid Dosinia clams https://doi.org/10.1371/journal.pone.0196466 or the Littorina marine snails https://doi.org/10.1016/j.margen.2016.10.006). In our case, repeats are larger than the short-reads insert size, and even though we could infer them by means of long read sequencing, polishing the resulting consensus sequences to overcome the intrinsic error rate of those lectures would yield inconclusive results, hindering the comparison between normal and tumoral haplotypes.

Action: We updated the methods for the mitochondrial DNA analyses (pages 21-22, 24) and the nuclear DNA analyses (page 23). We now include new data in the results and discussion (pages 9-10).

Reviewer #2 (Public Review):

In rare but well-documented instances, certain types of cancers can transmit horizontally. These transmissible cancers have a clonal origin and have adapted to bypass allorecognition. A form of marine leukemia (hemic neoplasia or HM) belongs to this class of transmissible cancers and has been detected in several bivalve species (oysters, mussels, cockles and clams). Although HM mostly propagates within the same bivalve species, instances of cross-species transmission have been reported. To better understand the mode of transmission of HM, Garcia-Souto et al. analysed mitochondrial DNA (mtDNA) by next generation sequencing in different bivalve species collected in the Mediterranean Sea and the Atlantic Ocean. The authors found that HM isolated in Venus verrucosa contained mtDNA that actually matched Chamelea gallina. Analysis of the nuclear gene DEAH12 also showed single nucleotide polymorphisms (SNPs) matching C. gallina DNA. Based on mtDNA and DEAH12 sequences, the authors use Bayesian inference to generate phylogenetic trees showing that HM found in V. verrucosa is much closer to C. gallina than the host species. They conclude that HM propagated from C. gallina to V. verrucosa.

Overall, the study is well performed with enough samples analysed. The results are quite convincing but there are also some concerns.

1. Transmissible cancers are known to split into clades based on mtDNA differential rate of evolution and also to incorporate mtDNA from exogenous sources, so one has to be extra careful that the results prove cross-species transmission and not HM divergence into two clades and/or exogenous acquisition. Samples HM ERVV17-2997 and EMVV18-376, both at the N1 stage, appear devoid of C. gallinae mtDNA and do not appear to have been screened for DEAH12. One explanation for this result is that there are too few HM cells in the samples (but supplementary Figure 1 shows some HM cells in ERVV17-2997. However, a different explanation is that these samples contain V. verrucosae mtDNA. ERVV17-2997 and EMVV18-376 could have been analysed in greater depth to verify that they also contained C. gallinae mtDNA and typical DEAH12 SNPs.

Answer: Despite the high sequencing coverage obtained for the sequenced individuals, we did not find foreign reads in the N1 tumours (ERVV17-2997 and EMVV18-73) to mitochondrial nor nuclear (i.e., DEAH12, TFHII) level. This is most likely due to a very low proportion of neoplastic cells in their tissues.

Action: We have added a sentence on page 8 that discuss this issue.

1. To strengthen their argument, the authors could have analysed a few more nuclear genes for specific SNPs, although the sensitivity of this approach will depend on the depth of sequencing.

Answer: We obtained a preliminary nuclear assembly using short-reads only. Obviously, the resulting assemblies are fragmented and incomplete. This has limited the identification of candidate regions shared by the three genomes (V. verrucosa and both Chamelea clams). Out of the 44 candidate nuclear fragments we tested, only two (DEAH12 and TFHII) turned out to give good PCR products, adequate for Sanger sequencing. As mentioned above, we now provide additional data on a second gene (TFIIH), identified and selected on the same basis as DEAH12. Individual ML phylogenies for these two fragments evidenced that tumours cluster together and separately from the host species and, in the case of DEAH12, closer to C. gallina. The MSC phylogeny was rebuilt including this new nuclear fragment. 12 In addition, we conducted a comparative screening of tandem repeats on the genomes of C. gallina and V. verrucosa. Two DNA satellites, namely CL4 and CL17, of, respectively, 332 and 429 bp monomer size, were very abundant in C. gallina and in the tumoral animals, but absent from all healthy V. verrucosa specimens. FISH probes designed for these satellites mapped on the heterochromatic regions, mainly in subcentromeric and subtelomeric positions, of both C. gallina and the neoplastic metaphases found in V. verrucosa, but were absent from the normal metaphases of the host species V. verrucosa. These results were consistent with the genomic abundance of these satellites in the NGS data and strongly suggest that these chromosomes derive from C. gallina.

Action: We include the analysis of one additional nuclear locus, TFIIH (pages 9-10). We have obtained new ML and MSC phylogenies including this new locus (pages 9-10, figures 3b-c). Additional FISH approach looking for satellite DNA CL4 and CL11 was performed (page 10, figure 3d, supplementary figure 5). The methods section has been updated accordingly (pages 20- 21, 23-24).

1. It would have been interesting to have more information in the Discussion on the potential immunological barriers that this tumour needs to overcome for cross-species transmission.

Answer: At a glance, we could argue/discuss that this transmissibility, inside or cross-species, is prone to occur in bivalves due to their filtering feeding system and the fact that their immune system is not entirely developed and yet to be completely understood, as the reviewer may know. Also, it would be tempting to suggest that some genetic restrictions allowing for cancer contagion happening only between close taxa might be in place, but, unfortunately we do not have the means to state that with our current data.

Action: At this point, no specific action has been taken for this query. However, we are happy to include something in the discussion if the reviewer still thinks this is relevant for improving the manuscript.

Author Response:

Reviewer #3 (Public Review):

INaR is related to an alternative inactivation mode of voltage activated sodium channels. It was suggested that an intracellular charged particle blocks the sodium channel alpha subunit from the intracellular space in addition to the canonical fast inactivation pathway. Putative particles revealed were sodium channel beta4 subunit and Fibroblast growth factor 14. However, abolishing the expression of neither protein does eliminate INaR. Therefore as recently suggested by several authors it is conceivable that INaR is not mediated by a particle driven mechanism at all. Instead, these and other proteins might bind to the pore forming alpha subunit and endow it with an alternative inactivation pathway as envisioned in this paper by the authors.

The main experimental findings were (1) The amplitude of INaR is independent of the voltage of the preceding step. (2) The peak amplitudes of INaR are dependent on the time of the depolarizing step but independent of the sodium driving force. (3) INaT and INaR are differential sensitive to recovery from inactivation. According to their experimental data the authors put forward a kinetic scheme that was fitted to their voltage-clamp patch-clamp recordings of freshly isolated Purkinje cells. The kinetic model proposed here has one open state and three inactivated states, two states related to fast inactivation (IF1, IF2) and one state related to a slower process (IS). Notably IS and IF are not linked directly in the kinetic scheme.

In my humble opinion, the proposed kinetic model fails to explain important experimental aspects and falls short to be related to the molecular machinery of sodium channels as outlined below. Still it is due time to advance the concepts of INaR. The new experimental findings of the authors are important in this respect and some ideas of the new model might be integrated in future kinetics schemes. In addition, the framework of INaR is not easy to get hold on with lots of experimental findings in the literature. Likely, my review falls also short in some aspects. Discussion is much needed and appreciated.

INaT & INaR decay The authors stated that decay speed of INaT and INaR is different and hence different mechanisms are involved. However at a given voltage (-45 mV) they have nicely illustrated (Fig. 2D and in the simulation Fig. 3H) that this is not the case. This statement is also not compatible with the used Markov model. That is because (at a given voltage) the decay of both current identities proceed from the same open state. Apparent inactivation time constants might be different, though, due to the transition to the on state.

We apologize that the language used was confusing. Our suggestion that there is more than one pathway for inactivation (from an open/conducting state) is the observation that the decay of INaT being biexponential at steady-state voltages. In the revised manuscript, we point out (lines 546-549) that, at some voltages, the slower of the two decay time constants (of INaT) is identical to the time constant of INaR decay. We also discuss how this observation was previously (Raman and Bean, 2001) interpreted.

Accumulation in the IS state after INaT inactivation in IF1 and IF2 has to proceed through closed states. How is this compatible with current NaV models? The authors have addressed this issue in the discussion. The arguments they have brought forward are not convincing for me since toxins and mutations are grossly impairing channel function.

Thank you for this comment. We would like to point out that, in our Markov model, Nav channels may accumulate in IS through either the closed state or open state. This requires, of course, that Nav channels can recover from inactivation prior to deactivation. While we agree that toxins and mutations can grossly impair channel function, we think these studies remain crucial in revealing the potential gating mechanisms of Nav channel pore-forming subunits, and how these mechanisms may vary across cell types that express different combinations of accessory proteins.

Fast inactivation - parallel inactivation pathways Related to the comment above the motivation to introduce a second fast-inactivated state IF2 is not clear. Using three states for inactivation would imply three inactivation time constants (O->IF1, IF1->IF2, O->IS) which are indeed partially visible in the simulation (Fig. 3). However, experimental data of INaT inactivation seldom require more than one time constant for fast inactivation. Importantly the authors do not provide data on INaT inactivation of the model in Fig. 3. Fast Inactivation is mapped to the binding of the IFM particle. In this model at slightly negative potential IF1 and IF2 reverse from absorbing states to dissipating states. How is this compatible with the IFM mechanism? Additionally, the statements in the discussion are not helpful, either a second time constants is required for IF (two distinct states, with two time constants) or not.

We thank this Reviewer for this comment. We tried to developed the model based on previous data on Nav channel inactivation. Indeed, much experimental data exists for the fast inactivation pathway (O -> IF1). As we noted in the discussion, without the inclusion of the IF2 state, we were unable to fully reproduce our experimental data, which led us to add the IF2 state. As with all model development, we balanced the need to faithfully reproduce the experimental data with efforts to limit the complexity of the model structure. In addition, as noted in the Methods section, our routine is an automatic parameter optimization routine that seeks to minimize the error between simulation and experiments. We can never be sure that we have found an absolute minimum, or that the optimization got stuck at a local minimum when simulating without inclusion of IF2. In other words, there may be a parameter set that sufficiently fits the data without inclusion of IF2, but we were unable to find it. As a safeguard against local minima, we used multistarts of the optimization routine with different initial parameter sets. In each case, we were unable to find a sufficiently acceptable parameter set.

We agree with this Reviewer that at slightly negative potentials (compared to strong depolarizations), channels exit the IF1 state at different rates, although we would point out that channels dissipate from the IF1 state (accumulating into IS1) under both conditions (see Figure 8B-C). This requires the binding and unbinding of the IFM motif to occur with some voltagesensitivity. We believe this to be a possibility in light of evidence that suggests IFM binding (and fast-inactivation) is an allosteric effect (Yan et al., 2017) and evidence showing that mutations in the pore-lining S6 segments can give rise to shifts of the voltage-dependence of fast inactivation without correlated shifts in the voltage-dependence of activation (Cervenka et al., 2018). However, it remains unclear how voltage-sensing in the Nav channel interact with fast- and slow-inactivation processes.

Due to space constraints in Figure 3, we did not show a plot of INaT voltage dependence. However, below, please find the experimental data (points), and simulated (line) INaT in our model.

Differential recovery of INaT & INaR Different kinetics for INaR and INaR are a very interesting finding. In my opinion, this data is not compatible with the proposed Markov model (and the authors do not provide data on the simulation). If INaT1 and INaT2 (Fig. 5 A) have the same amplitude the occupancy of the open state must be the same. I think there is no way to proceed differentially to the open state of INaR in subsequent steps unless e.g. slow inactivated states are introduced.

Thank you for bringing up this important point. The differential recovery of INaT and INaR indicates there are distinct Nav channel populations underlying the Nav currents in Purkinje neurons. We make this point on lines 632-635 of the revised manuscript. Because our Markov model is used to simulate a single channel population, we do not expect the model to reproduce the results shown in Figure 5. We have now added this point to the Discussion section on lines 637-640.

Kinetic scheme Comparison with the Raman-Bean model is a bit unfair unless the parameters are fitted to the same dataset used in this study. However, the authors have an important point in stating that this model could not reproduce all aspects of INaR. A more detailed discussion (and maybe analysis) of the states required for the models would be ideal including recent literature (e.g., J Physiol. 2020 Jan;598(2):381-40). Could the Raman-Bean model perform better if an additional inactivated state is introduced? Are alternative connections possible in the proposed model? How ambiguous is the model? Is given my statements above a second open state required? Finally, a better link of the introduced states to NaV structure-function relationship would be beneficial.

These are all excellent points. We absolutely agree; it was/is not our intention to “prove” that the Raman-Bean model does not fit our dataset (as you mention, with proper refinement of the parameters, some of the data may be well fit). In fact, qualitatively we found the Raman-Bean model quite consistent with our dataset (which is an excellent validation of both the model, and our data). It was our intention to show (in Figure 7) that there is good agreement between the Raman-Bean model and our experimental data for steady state inactivation (C), availability (D), and recovery from inactivation (E). While we find the magnitude of the resurgent current (F) to be markedly different than the Raman-Bean data, we now note this to likely be due to the large differences in the extracellular Na+ concentrations used in voltage-clamp experiments (lines 440-444). Our models, however, specifically differ in our parallel fast and slow inactivation pathways (Figure 7H). As seen in the Raman-Bean model, in response to a prolonged depolarizing holding potential, there is negligible inactivation, as the OB state remains absorbent until the channel is repolarized. This is primarily because the channel must transit through the Open state on repolarization. We find distinctly different behavior in our data. As seen in the experimental data shown in 7H, despite a prolonged depolarization, Nav channels begin to inactivate and accumulate in the slow inactivated state without prerequisite channel opening. This behavior is impossible to fit in the Raman-Bean model, given the topological constraint of the model requiring a single pathway through the open state from the OB state.

To that point, it is also unlikely that the addition of inactivated states to the Raman-Bean model would help fit this new dataset. Indeed, the Raman-Bean model contains 7 inactivated states. If there were a connection between OB ->I6, it is possible that direct inactivation (bypassing the O state) may help. Again, however, it is not our intention to discredit the Raman-Bean model, nor is it our intention to improve the Raman-Bean model. With new datasets, a fresh look at model topology was undertaken, which is how we developed our proposed model.

This Reviewer astutely points out a known limitation of Markov (state-chain) modeling; it is impossible to tell uniqueness, or ambiguity of the model (both with parameters as well as model topology). Following the results of Menon et al. 2009 (PNAS vol. 106 / #39 / 16829 – 16834), in which they used a state mutating genetic algorithm to vary topologies of a Markov model, our group (Mangold et al. 2021, PLoS Comp Bio) recently published an algorithm to distinctly enumerate all possible model structures using rooted graph theory (e.g. all possible combinations of models, rooted around a single open state). What we found (which is not entirely surprising) is that there are many model structures and parameter sets that adequately fit certain datasets (e.g., cardiac Nav channels).

Therefore, the goal is never to find the model (indeed we don’t propose that we have done so), but rather to find a model with acceptable fits to the data and then use that model to hypothesize why that model structure works, as well as to hypothesize higher dimensional dynamics. We make these points in the revised manuscript (lines 591-597).

We did not specifically explore the impact of a second open state in our modeling and simulation studies, but we would certainly agree that a model with a second open state may recapitulate the dataset.

Author Response:

Reviewer #1:

The study is elegantly done, with the outstanding questions clearly laid out and the results presented in a clear and informative fashion. I have only a few suggestions to strengthen some of the results.

1) Determination of layers: The CSD based method used to determine the layers seems a bit ad hoc, although other studies have often used a similar approach. Some histological evidence would be great. If that is not possible, the authors should provide some more details to determine the layer specificity. For example, where were the supragranular-granular and granular-infragranular borders for different penetrations (i.e., which electrode(s) marked these boundaries)? These could be expressed as fractions of the shaft length, and from that, we would approximately know the depth. Also, were these results affected by how the CSDs were smoothed?

We thank the Reviewer for the suggestions. Using CSD from individual penetrations to define the position of laminar compartments is a strategy has been used by several different laboratories, not only in primary visual cortex (e.g. Mitzdorf, 1985; Poort et al., 2016; Bijanzadeh et al., 2018), but also in extrastriate areas, such as other studies in V4 (Nandy et al., 2017; Lu et al., 2018, Pettine et al., 2019; Ferro et al., 2021), and in other cortical areas, for example in medial temporal cortex (Takeuchi et al., 2011). We include a new figure that shows the similarities between the CSD profiles from different studies (Figure 1-figure supplement 3). Figure 1-figure supplement 3A shows the population average of the CSD in our data. The similarity between this panel and the individual examples shown in Figure 1 and Figure 1-figure supplements 1 and 2 further highlights the fairly consistent sink-source patterns observed across individual penetrations in our data. Figure 1-figure supplement 3B shows that this pattern is also consistent with that found in macaque V4 in another laboratory (Pettine et al., 2019). Figure 1-figure supplement 3C shows that this pattern is not specific for V4, but also occurs in other areas, such as the medial temporal cortex (Takeuchi et al., 2011).

Importantly, in the latter study Takeuchi et al. were able to verify histologically, after applying electrolytic marks, that the prominent current sink with short latency (white star in Figure 1-figure supplement 3C) corresponds to the granular layer, thus consistent with CSD analysis in V1 (Mitzdorf, 1985). Thus although we are not able to perform such histological verification in our study (one animal has been euthanized, and the other animal is destined to take part in another project), the very similar sink-source patterns between these different studies (Figure 1-figure supplement 3A-C), including with work from others that has been verified histologically, gives us confidence that we can meaningfully use them to assign electrode contacts to granular, superficial and deep layers, as other laboratories have done (e.g. Lu et al., 2018). In addition to the new Figure 1-figure supplement 3, we added language in the corresponding paragraph in Results to explain this better. We clarified in Results that interpretable CSD maps were found in 81 out of 88 penetrations and that only those were used in the laminar analysis (which was already indicated in Methods in the previous version of the manuscript).

As suggested by the Reviewer, we have added the positions of the electrode contacts to the CSD maps in Figure 1 and Figure 1-figure supplement 2 (labels along ordinate on the right of the panels). The electrode contacts on the probe covered 3.1 mm (32 contacts with 0.1 mm distance between adjacent contacts), thus the full depth of the cortex was covered even though the vertical position of the probe varied between penetrations (also because a layer of granulation tissue develops over time between the artificial dura and the pial surface). Therefore, to aid in estimating the depth of the individual penetrations, we indicate the position of the most superficial contact on which multiunit activity was recorded (solid black triangles in Figure 1; Figure 1-figure supplements 1,2; for all cases where this contact could be identified, i.e. if the most proximal contact on the probe did not show multiunit activity). The average position of this contact is shown on the population CSD map (solid black triangle on Figure 1-figure supplement 3). The advantage of a method that assigns layers for a penetration based on data from the same penetration, as we have used here, as opposed to a method that assigns compartments based on depth derived from a population average, is that the former helps to avoid errors due to variations in probe position, and due to a variable degree of tissue compression that may occur for different penetrations.

To test whether the results were affected by how the CSDs were smoothed, we performed the layer assignment separately using different methods of smoothing. To ensure that we did not bias the results we blinded ourselves to the original layer assignment when applying each method. We find that the laminar position of >97% of well-isolated units was identical as that obtained when using the standard procedure of smoothing the CSDs. This resulted in robust latency differences of border ownership signals between layers, irrespective of which smoothing method was used. These results are presented in a new supplementary figure (Figure 2-figure supplement 3).

2) Another important factor is the orthogonality of the penetrations. This can also be better quantified based on the variation of the RF centers with depth.

We followed the Reviewer’s suggestion and evaluated orthogonality of the penetrations by computing the distance D between receptive field centers along the probe (Methods). We show this metric for the vertical positions of receptive field contours shown for the penetrations in Figure 1H,I and Figure 1-figure supplement 1, and describe the population data in the first paragraph of Results (median (IQR) 0.83 o/mm (1.00 o/mm), for all 81 penetrations that were included in the laminar analyses). This indicates that the variation is small relative to the average diameter of the receptive fields (7.36 o), and that the deviation from orthogonality is limited.

Author Response:

Reviewer #2:

The manuscript by Podinovskaia focuses on a new method to visualize and measure endosome maturation in common cell lines by enlarging early endosomes. This was achieved by producing acute insult to the cells by ionophore treatment, leading to budding of abnormally large post Golgi vesicles that fuse with early endosomes. Endosome maturation of these enlarged endosomes containing Golgi-derived cargo (GalT) proceeding with apparently normal kinetics, ultimately leading to lysosomal delivery. Taking advantage of this assay, the authors investigate Rab5-to-Rab7 conversion, acquisition and loss of PI3P, acquisition and loss of Snx1 on apparent endosomal subdomains, interaction of early and late endosomes with Rab11-positive recycling endosomes, and lumenal pH changes. The new maturation model presented here will likely be quite useful to the field with continuing impact. The current state of the endosome field in many ways remains fragmentary, with various processes studied extensively in isolation, but with little information on their relative timing and potential interactions as endosomes mature. This new assay should help understand the relationships between these processes, some of which are investigated in this manuscript.

Concerns:

1) The data and conclusions related to Rab11 interaction with early endosomes in Fig 8 are not convincing. There are simply too many Rab11 endosomes in the cell to know if their short term proximity indicates meaningful interaction with the early endosomes, or if the data simply reflects random collisions of small recycling endosomes with the enlarged early endosomes. No data is presented to show that the interactions are meaningful, e.g. that recycling cargo transfer occurs during these interactions. Conclusions from this analysis are overstated.

We now provide more evidence for the interaction of Rab11 vesicles with the enlarged endosomes. We made movies with shorter intervals (2 sec instead of 1 min) between the individual frames. These data clearly show that this is not an accidental bumping into an endosome but rather that Rab11 vesicles can circle around endosomes and stay for several minutes (Video Fig. 8A, supplement 2 and 3).

In addition, we imaged TfR-GFP together with mApple-Rab5. These data show that TfR-GFP positive vesicles bud off from mApple-Rab5 positive endosomes and that the GFP fluorescence intensity goes down over time in enlarged endosomes. These data are consistent with recycling of TfR to the plasma membrane. Moreover, CDMPR-GFP, which cycles between the TGN and endosomes was found to be present on Rab5 negative enlarged structure, which then turned Rab5 positive, and subsequently lost the CDMPR signal. Importantly those endosomes could regain CDMPR, which we interpret as acquisition from the TGN. These data may indicate that the TGN-endosome shuttle is intact after nigericin washout (Fig. 9).

That the TfR and CDMPR are really transported out of the enlarged endosome is also contrasted by our finding that GalT-GFP stayed in the enlarged endosome and the signal intensity did not significantly drop.

2) Lack of information on endocytic cargo acquisition by the enlarged early endosomes: to really establish this endosome maturation model the authors would need to establish if the enlarged endosomes contain endocytosed cargo, as opposed to Golgi-derived cargo, and determine how long it takes to acquire such cargo. This could be accomplished using Tf, EGF, or perhaps dextran at early timepoints after nigericin washout.

As described above, we now show that TfR-GFP is present in enlarged endosomes and is lost from these endosomes over time (Fig. 9A,D,G).

Additionally, we performed experiments with dextran-Alexa647 and nanobody-tagged surface TfR to show that endocytosed material from the plasma membrane indeed reached the enlarged endosomes (Fig. 3, figure supplement 1 and 2). Quantification of TfR signal at the enlarged endosomes demonstrates that TfR acquisition by the enlarged endosome takes place as soon as the enlarged compartment becomes Rab5-positive. This was also observed with the nanobody-tagged surface TfR and endocytosed Dextran-AF647, representative examples of which are provided (Fig 3, figure supplement 1 and 2). The quantification for the latter experiments was not carried out due to the very short time range during which asynchronous Rab5 recruitment events needed to be captured after addition of nanobody/Dextran pulse-and-chase.

3) Figure 7 - It was not convincing that data in panels F and G are different from each other.

We agree with the reviewer that the difference between the data presented in panel F and G is not very big. These panels represent the average of many endosomes and with the averaging the differences from the individual traces get cancelled out. The process is asynchronous and thus in this case the individual traces are more telling than the averaged traces. Nevertheless, we decided to keep the average traces in the manuscript because the highlight the asynchronous nature of the process. We modified the text to make this point clear.

4) Figure 11 - it is unclear how we can interpret this as connected to Rab conversion when even the labeled compartments at the earliest time point in the czz1 knockout have abnormally high pH, and during the time-course even the last timepoint for czz1 KO is higher than that of the earliest timepoint for WT.

We agree that the ccz1 KO cells display higher endosomal pH than WT cells throughout the time-course.

However, the cells in which we express the rescue plasmid of Ccz1 also have apparently less acidified endosomes, even though Ccz1 can still drive Rab conversion, and the pH dropped at an intermediate rate, when comparing rescued cells to control and ccz1 KO cells. Even in ccz1 KO cells endosomal traffic down the degradation pathway is not completely blocked, similarly to what we observed for sand-1 (-/-) in C. elegans and Mon1a/b knockdown in mammalian cells (Poteryaev et al. 2010). Acidification eventually will occur, but it is massively slowed down; the molecular basis of which is still under investigation in our lab.

We think that in the absence of Ccz1, a condition under which Rab conversion is severely impaired, acidification cannot occur at normal rate. As pointed out by reviewers 1 and 2, the pH is already higher in the ccz1 KO cells than in the control condition. However, in the rescue condition, the YFP/CFP ratio is not that different from the knockout and yet acidification can occur at an intermediate rate. Why under rescue conditions, the YFP/CFP ratio is at a similar level compared to the KO is not entirely clear. It is conceivable that too much Ccz1 has also a negative effect. Moreover, recently it has been shown that ccz1 KO cells accumulate free cholesterol in the enlarged endosomes (Van den Boomen Nat Comm., 2020). The transient expression might be not sufficient to rescue this accumulation phenotype or other secondary effects. Nevertheless, the v-ATPase appears to maintain its function because lysosomes can acidify in ccz1 KO cells, albeit with a delay (Figure 13).

5) Figure 12 - The criteria used to determine which GalT structures are Golgi or lysosomes seems questionable. Morphology alone is not sufficient to identify the compartments with high accuracy, especially after perturbation. Also, it is unclear to what extent GalT-CFP labels lysosomes without nigericin treatment.

To address these issues, we co-labelled cells with lysotracker. GalT-CFP (pHlemon) and lysotracker showed a very high degree of co-localization. These data are included in the manuscript (Fig. 10B).

Author Response:

Reviewer #1:

In this manuscript, the authors make use of next-generation sequencing to provide a preliminary inventory of tribe Metriorrhynchini, a hyperdiverse group of beetles with intricate systematics mainly due to likely morphological convergence of their Millerian rings. The authors provide an admirable sampling within Africa, Asia and Oceania, with about 700 successfully sampled localities and thousands of specimens.

The main result of the manuscript is the curated database of Metriorrhynchini that will be useful in future research. In addition, different statistical methods are used to provide an idea of the undescribed species within the tribe, the astonishing species richness in New Guinea or the use of phylogenomic data to explore major phylogenetic relationships. However, some of the author's claims should be questioned:

• Surprisingly, the authors rely on a very low threshold to identify mOTUs (2% in the manuscript). The authors refer to Hebert et al. (2003) and Eberle et al. (2020) to justify the threshold, but still, they are likely overestimating the number of mOTUs and thus, considering putative species what it may be different populations. Figure S17 provide estimates of mOTUs with different thresholds (1 to 10%), which rapidly decrease their estimates (a decrease of 25% mOTUs is found when 6% was considered). Still, an overwhelming sampling effort but a more realistic estimate.

• I think the phylogenomic tree did not receive the required attention (for example, the FcLM analysis is barely mentioned).

• It is not clear why should be important to mention the "person-months of focused field research" across the manuscript. Each study group has a unique sampling technique (also not found in the manuscript), preferred localities or traits, which make comparisons impossible. The authors' effort is remarkable, but it is not an important result/finding to be highlighted all over the manuscript.

Many thanks for all comments and suggestions that pointed to the weak parts of our argumentation. We modified the manuscript accordingly and added some references that can be used for the justification of some claims.

We addressed the question of thresholds for species number estimations. Now, two thresholds are considered in the manuscript as relevant for discussion: 2% and 5%. We added further information on our previous studies dealing with integrative species delimitation in Metriorrhynchini. Some of them were not referred in the earlier version (to avoid self-citations) and we also expanded information on the evidence given in the study which we have already referenced (Bocek et al. 2019). The earlier comparison of nextRAd, mtDNA and moprhology-based delimitation of species in Eniclases (the trichaline clade in the present study) showed that many well defined species (nextRAD and morphology) have highly similar mtDNA and they split only recently, eventually some introgresion or incomplete lineage sorting affect mtDNA signal. If we apply 5% threshold for this group, we would delimit as a single species two entities which differ in the body size, coloration and the relative size of male eyes (diurnal and nocturnal activity in putative sister species). In such a way, we would decrease the number of species in our analyzed sample of Eniclases by 40% in clear contrast with the number based on morphology and nextRADs. We found similar rapid morphological diversification also in other metriorrhynchines (Jiruskova et al., 2019,; Kalousova & Bocak, 2017) and other not referenced taxonomic studies that have shown that closely related species have well diversified male genitalia and often belong to different mimetic rings). To limit our discussion, we do not reference our earlier nextRAD study showing the speciation in other subfamily of net-winged beetles within a single mountain range (Bray & Bocak 2016). Also this study supports morphological differentiation in species with highly similar mtDNA. Now, we noted in the manuscript that before taxonomic revisions are produced, our claim is provisional and therefore we modified the text as proposed and present the lower numbers of species as a realistic possibility.

Phylogenetic relationships: We added additional information on the congruence with earlier studies to Results and Discussion, but we still do not describe details. The main reason is that morphology must be studied to delimit and formally name new taxa and that the morphology is out of scope of this work (except some information provided in Supplementary Text – description of delimited generic groups and subtribes). The FcLM analysis addressed only the relative position of the leptotrichaline and procautirine clade. Both clades are monophyletic, morphologically distinct and no conclusion is based on their relative position. We noted that without further data we are unable to robustly solve their positions. Provisionally, we prefer the deeper postion of the leptotrichalines (61%, a not very convincing phylogenenomic signal).

Quantification of sampling effort: As proposed, we excluded the consideration of person months as a measure of relative collecting effort in various regions and add justification for field research methods.

Reviewer #2:

Conservation efforts must be evidence-based, so rapid and economically feasible methods should be used to quantify diversity and distribution patterns. The principal objective of this study is to demonstrate how biodiversity information for a hyperdiverse tropical group can be rapidly expanded via targeted field research and large-scale sequencing. The authors have attempted to overcome current impediments to the gathering of biodiversity data by using integrative phylogenomic and three mtDNA fragment analyses. As a model, they sequenced the Metriorrhynchini beetle fauna, sampled from ~700 localities in three continents. The species-rich dataset included ~6,500 terminals, >2,300 putative species, more than a half of them unknown to science. It is an amazing finding. Their information and phylogenetic hypotheses can be a resource for higher-level phylogenetics, population genetics, phylogeographic studies, and biodiversity estimation. At the same time, they want to show how limited the taxonomical knowledge is and how this lack is hindering biodiversity research and management.

Thanks for your comments on our study. We agree with your specific recommendations and modify the manuscript accordingly.

Author Response:

Reviewer #1 (Public Review):

The authors propose several ways of leveraging single-particle tracking experiments to distinguish between intracellular phase separation and an alternative model of clustered binding sites. The first proposed scheme is particularly intuitively appealing: in the binding site scenario, the local density of binding sites both increases particle density and slows effective particle diffusion, leading to a definite relationship between these two quantities, while the phase separation scenario would not necessarily couple these two quantities. The additional schemes based on particle movement near a cluster boundary, angles between consecutive steps, and search times add to the arsenal of potential analysis tools. Overall, the work is timely, rigorous, and generally clearly presented and given the growing list of reported observations of phase separation, will appeal to a broad audience.

We thank the referee for the positive attitude towards our work and are happy for the insightful comments that have increased the value of this paper.

Reviewer #3 (Public Review):

Membraneless condensates have recently become a central focus of the molecular and cellular biophysics communities. While the dominant paradigm for their formation, liquid-liquid phase separation (LLPS), has been well established in a number of cases for large, optically resolved droplets, there are significant concerns regarding the generality of this mechanism for smaller foci or puncta, and other mechanisms have been proposed to explain their formation. The problem is that it is very difficult to distinguish experimentally between these mechanisms for sub-optical resolution condensates. In this article, Heltberg et al propose a novel method, based on the analysis of single molecule tracks, that allows discriminating between the liquid phase model (LPM) and one of the challenger mechanisms, the "polymer bridging model" (PBM). This method relies on the statistics of individual displacements - diffusion, radial displacements, angular changes - which are showed theoretically to exhibit different signatures for the two models. With realistic data this is sufficient to discriminate between the models: for instance in the case of double strand break foci (DSB), building on a recent work by some of the same authors, this article convincingly rules out the PBM in favor of the LPM. The author also investigate the influence on these two models on the search time to reach a specific small target - a commonly invoked role of condensates - and show that only the LPM substantially accelerates this, which could provide additional means to experimentally discriminate between the mechanisms, on top of the intrinsic interest of this finding.

This article is a welcome addition to the literature in this field, as it will help clarify the nature of these condensates, in particular below the optical resolution. It is well-written, interesting and the conclusions are justified. I particularly appreciate the effort to employ simulated data that are realistic for actual experiments, which strengthens the claims of applicability. Some aspects of the data analysis and of the modeling, however, are insufficiently discussed and would need to be precised / expanded.

1) The modeling is made under the assumption of thermal equilibrium, without further discussion. The authors should comment on why this is reasonable, in particular in view of the presence of active fluctuations and of chemical reactions in these condensates.

First of all, the experimental measurements are carried out after the formation of the foci, and the time of observation (tens of seconds) is small compared the the lifetime of foci (tens of minutes). Therefore we can assume that these measurements are not affected by the effects of formation and disruption of foci. Secondly, the data extracted to compute the results of Figure 2 (in particular for Figure 2H) are not very sensitive to the active fluctuations, since we derive an average diffusion coefficient inside and outside of the focus as well as a free energy difference between an inside and outside level. It is indeed very likely that the soup of proteins that forms the focus is active, however Rad52 is not involved in chemical reactions at the timescales we are looking at may be considered passive. This is supported by our investigations of the experimental results, where we have not seen any statistical differences as a function of the time of measurements, and we have no reason to believe that active fluctuations affect the diffusivity of Rad52 on the observed timescales. Regarding binding sites, they may also diffuse actively along with the genome and chromatin, but we describe this by an effective description of the motion of Rad52 on short time scales, so that active effects are folded into an effective diffusivity (left as a free parameter).

We want to highlight this issue as well as present our arguments of why this description is valid for the experiments considered in this work. We have added text between Eq. 6 and Eq. 7 summarizing the arguments outlined above.

2) How is the diffusivity measured? Are these measures corrected for experimental error (e.g. using three-point estimators)?

Estimates of the diffusion coefficients in Min´e-Hattab et al. 2021 were obtained in different ways. Our main method is to generate the displacement histogram, and then estimate the number of different diffusion coefficients in the population based on likelihood fitting and KStesting. Then we take for all the traces, and find the ones that we are certain belong to the slowest diffusion coefficient. These traces are the ones in the focus, but by doing it this way, we are not vulnerable to the position of the boundary and to determine which are in the focus based on their position. Then we compute the MSD curves for this distribution of slowly diffusing molecules, and fit the diffusion coefficient based on a confined fit (which has a good p-value). This method is strong since we are fitting a slow diffusion population and typically can reject traces belonging to the fast diffusion coefficient. We also include the possibility of separating traces if the molecule goes from inside the focus to outside or the other way round. The alternative way we calculated the diffusion coefficient, was based on the microscopy data, where we “cropped” all the traces that could be visibly identified as being inside the focus. This method had the strength that we could visibly follow all traces, but the drawback that we could mistakenly identify molecules as being inside the focus, then they could be under or above the focus, as discussed in the section above. However both method yielded similar results. It is also based on these methods that we extract the size of the focus.

In order to clarify this important point, we have added two sentences in the caption to Table I, describing how the diffusion was measured in the experimental paper, and added a new paragraph about experimental measurements in Materials and Methods. In addition, we have clarified in the caption of Fig. 2F that we extract the maximum likelihood value of D˜(r) in each radial segment.

3) The conditioning of the averages should be discussed, e.g. in Eq. 13: I assume that it is in the Ito convention? Similarly for the angle changes.

We assume that the density of the binding sites, follow a radial distribution, with no significant angular dependency. Thus the average displacement hdri is computed as a function of the initial position of the particle and averaged over all initial displacements with similar radial positions. It is indeed formulated in the Ito convention, which is why the “spurious” term appear in the first term of the second line of equation 13.

To clarify that we are using Ito convention, we have stated that we are using Ito convention for this paper, just before the introduction of eq. 1. We have furthermore clarified in the section related to eq. 13 and the section related to the distribution of the angles that we use the initial position when calculating the difference between the two connected points.

Author Response:

Reviewer #1:

George Elias et al investigated the response of a cohort of individuals to Hepatitis B vaccination and analysed the role of preexisting vaccine-reactive CD4+ memory T cell receptors in the immune response. They found that the presence of these cross-reactive receptors elicits a faster and stronger response in the vaccinees. This is an extremely interesting result, as it suggests that a better understanding of the immune receptor repertoire of an individual can be used to predict and analyse its response to vaccination.

Strengths:

The study presents a detailed experimental analysis of the role of CD4+ T cells in the immune response to vaccination.

The authors show clearly that the dynamics of expansion of memory CD4+ vaccine-specific clones follows the immune response, corroborating the results of previous studies that analysed effector CD4+ cells.

The authors asked also whether the presence of preexisting vaccine-specific clones impacts the response to vaccination. They found that this is the case. They defined an estimator of a normalized number of putative vaccine-specific clones and showed that can be used to classify individuals into early or late responders. This result has the potential to be extremely impactful in the way we understand immune response to vaccination.

We thank the reviewer for their kind comments.

Weaknesses:

This central result follows the definition of the R{hbs} measure. It is not completely clear how much the numerator and denominator of R{hbs} contribute to the results and how those bystander and putative receptor sequences have been chosen. Some additional explanations could help reinforce the trust to this specific analysis.

As indicated by this and other reviewers, the original definition R_{hbs} measure was confusing for readers. We have attempted to clarify the definition, through changes in the methods, results, figure legends and code base.

In order to specifically address this comment, we have extended the discussion on what makes up the numerator and denominator, and how each contribute to the final metric (which is visualized in figure 3b).

The new text reads:

"Rhbs is the ratio of the frequency of putative peptide-specific TCRβ divided by a normalization term for putative false positive predictions due to bystander activations in the training data set. This model applied to the memory repertoire at day 60 shows that early-converters tend to have a higher frequency of putative HBsAg peptide-specific TCRβ, while late-converters tend to have relatively more putative false positives as per the normalization term (Fig. 3b)."

It is also not clear if multiple testing correction has been performed in the presentation of the results of Fig5.

Multiple testing correction is implemented throughout the manuscript. This has now been clarified in the manuscript text.

The correlation between the number of putative vaccine-reactive CD4+ T cells at day 60 and antibody titers is an interesting and robust result. This however does not support the claim of the authors that preexisting vaccine-specific CD4+ memory cells are associated to stronger immune response. This could be the case only if a similar correlation would be observed at day 0.

The reviewer raises an important point. The relationship between the vaccine-reactive CD4+ T-cells and the antibody titer is already a very interesting finding. However, we do already confirm that this signal is also present at day 0 in the CD4+ T-cell memory repertoire. Thus, our results seem to indicate the presence of preexisting vaccine-specific CD4+ memory cells.

We have clarified this section in the manuscript in the hopes of avoiding similar confusion in other readers. The new section reads:

"Furthermore, searching for HBsAg peptide-specific clonotypes in the memory repertoires prior to vaccination (day 0) results in a Rhbs with a similar difference (one-sided Wilcoxon-test P value= 0.0010, Fig. 3d). In this manner, the presence of HBsAg peptide-specific clonotypes as represented by the ratio Rhbs can be used as a classifier to distinguish early from late-converters prior to vaccination (Fig. 3e), with an AUC of 0.825 (95% CI: 0.657 – 0.994) in a leave-one-out cross validation setting."

Reviewer #3:

This manuscript presents a comprehensive study of CD4 memory T cell receptor beta repertoire response to hepatitis B vaccination, including repertoire correlates of early, late, and non seroconversion, identification of antigen specific and epitope specific clones, and a statistical classifier to potentially predict early Vs late seroconverters based on their pre-vaccination bulk repertoire. The major strengths are a unified experimental and computational analysis of bulk TCR repertoire data with antigen and epitope specific sorted T cells from the same individuals, allowing them to track personalized dynamics of vaccine specific clones, as well as translate across individuals to predict vaccine-induced seroconversion outcomes from pre-vaccination repertoires. The experimental data and reproducible analysis code are publicly accessible, and represent a useful resource that will likely be of interest beyond this study to other immune repertoire researchers.

The results seem to support the authors conclusions, however several reported findings based on statistical analysis are less convincing, and would benefit from improved validation, clarification, or reworking. I next detail these aspects ordered by results sections.

Section beginning line 128:

The reported finding of this section is that early-converters (and not late-converters) undergo repertoire remodeling by day 60 post vaccination that decreases repertoire clonality. The evidence presented to support this is a computation of Shannon entropy for day 60 Vs day 0 in each individual, and a paired sample statistical test that is nominally significant for early and not for late converters. However, this nominally significant p value 0.042 is quite marginal, and the associated plots (Fig 2a) indicate only a very modest visual difference, and the presence of a distant outlier. The p value for late converters is not shown, however the marginally significant p value for early converters may not be nominally significant (at alpha 0.05) after multiple test correction (two tests). Additionally, the range of possible entropy values depends on the total sample size, so part of this difference may be driven by sample size. It may be more appropriate to use the Shannon equitability index (normalizing by the maximum possible entropy given the sample size, which is the log of the richness).

We wish to thank the reviewer for the suggestion of using the Shannon equitability index, which we had not considered before and is indeed highly appropriate for the analysis. This analysis has therefore been rerun with the Shannon equitability index, which has indeed resolved the visual outliers that appeared before. As could be expected, the marginal result that was found before was not sufficient robust, and the P-value for the late converter increase in Shannon equitability index was now found to be 0.0822 with a Wilcoxon test. These results have therefore been removed and they had no impact on the main conclusions of the paper.

Section beginning line 147:

Ag specific T cells were isolated from day 60 samples and sequenced, allowing the authors to track the dynamics of these clones in the bulk repertoire data across time points. In all vaccinee groups these Ag specific clones are found to increase from day 0 to day 60 in the bulk repertoires. A marginal p value (0.04909) is presented to support early-converters showing more increase in these Ag specific clones. However, statistics comparing early to non or late to non converters are not mentioned (and these would require a multiple test correction on the p value that is discussed).

This is a valid concern raised by the reviewer. This specific analysis as performed in the original study was lacking power, in large part due to the lack of a concrete null model. Due to this lack in power, we had opted not to include the non-converters in this analysis (as these are only three samples). However, thanks to this reviewer’s suggestion, we were able to rework this analysis with a novel null model (detailed in the next response). This section has therefore been removed and replaced with the new analysis (where a Bonferroni multiple testing correction has been applied for the three responder categories).

A more general difficulty I have with this section is that the null hypothesis isn't made clear, and is probably more subtle and complicated than it appears. Cells are sorted from day 60, then their prevalence is compared between day 0 and day 60. Don't we expect to see more of them in day 60 even if there is no specific expansion for these clonotypes, but just random repertoire churn? My concern here is that double dipping from day 60 is affecting the analysis, since this time point is initially used to define the marker clones in the first place. If you take a random set of day 60 TCRs as null marker clones do you also see they are more prevalent in day 60 Vs day 0, or are you assuming that there should be no difference under the null?

This is a valid point of criticism raised by the reviewer, and one that was not adequately explored in the previous version of the manuscript. In the prior version, the assumption was that there would be no difference under the null, despite the vaccine-specific clonotypes are derived from samples taken from day 60. While these do originate from a different cellular compartment (memory versus activated cells) and there are several weeks of stimulation experiments that separate the two, it can be argued that the clonotypes will always have more in common with the day 60 samples than the day 0 samples.

The establishment of a null model is difficult in this sense. Selecting random clonotypes from day 60 as the reviewer suggests, would establish a baseline based on the overlap between the two time points. This would not be a good comparison as the impact of day 60 clonotypes would be overinflated (as no experimental steps separate it) and any increase in epitope-specific clonotypes could never exceed this value.

Therefore, we have added additional experimental data to establish a null model against which to compare. Samples from day 60 were treated in identical manner as described before but in the presence of epitope lysates derived from the varicella zoster virus (VZV) instead of the hepatitis B surface antigen. The prevalence of VZV in our study population is near absolute, thus it can be expected that all individuals have built up a T-cell immunity against the virus. Moreover, as this is a childhood disease, one can expect that the T-cell immunity against varicella on average not to change between day 0 and day 60 in our cohort. We postulate that this additional experiment is perfectly suited to establish a null model for the vaccine-specific expansion.

Using the VZV-specific clonotypes in the comparisons between day 60 and day 0 show a difference of 1.021 [95% CI: 0.934-1.124]. Thus our assumption that the null model should show no increase seems to be valid, as here too the clonotypes were derived from the day 60 samples. In addition, all reported increases were found to be highly significant when compared to the VZV-derived baseline (e.g. P-value of 6.3e-05 for the HbsAg-specific increase of 2.080).

The last analysis in this section (presented in Fig 2d) does group-level comparisons of Ag specific clone fractions at day 60. I don't follow why normalizing by the number of Ag-specific clones detected in each individual is correct (i.e. would result in no differences under the null). Here again it could be helpful to see if null marker clone sets (the same size as the true Ag specific sets for each individual) indeed show no significant differences between groups.

This is an important remark made by the reviewer. In essence, we need to compare the set of marker TCR clonotypes identified in the expansion experiment with the TCR clonotypes found in the CD4+ memory compartment. Thus, we are considering the overlap between two sets of TCR clonotypes for each individual.

When one has two sets A and B, and wish to compare the overlap across different comparisons, one has to consider the impact of the size of A and B. The larger either A and B are, the larger the expected overlap (even by chance).

As a specific example tuned to the problem we are addressing in this study: Imagine two donor D1 and D2. Each donor has a sequenced TCR repertoire with 50,000 unique clones (which is on par with the observed values). D1 has 10 Ag-specific clonotypes derived from the stimulation and D2 has 200 Ag-specific clonotypes. These number can vary widely due to sampling bias, differences in clonal expansion and difference in immunoprevalence or immunodominance. This is also not considering any bystander (false positive) clonotypes. In any case, consider that the overlap in both cases is 5. This means that we find half of the Ag-specific clonotypes in the memory repertoire of D1, but only less than 3% for D2. Thus despite having equal overlap, we would argue that the overlap for D1 is more relevant than the overlap for D2. Thus when comparing between individuals, we argue that one must take into account the size of the TCR sets that are being used for the overlap calculation.

As our sets are not equal in size, i.e. |A| >> |B|, we applied the Szymkiewicz–Simpson coefficient (also known as the Overlap Coefficient), wherein one divides by the size of the smallest set. In our case, the smallest set is always the set of the Ag-specific marker TCR clonotypes. Therefore, in practice, we always normalize by the number of Ag-specific clones.

This has now been clarified in the paper and the new text reads:

"In this case, to allow for a between-vaccinees comparison (in contrast to the within-vaccinees timepoint comparison), we calculate the Overlap Coefficient, where HBsAg-specific sequences in the CD4 T-cell memory repertoire are normalized by the number of HBsAg-specific TCRβ found for each vaccinee."

Section beginning line 185:

In this section, a peptide pool approach is used to identify epitope specific TCRs from each individual at day 60, and a classifier is constructed to discriminate between early and late converter bulk repertoires, using a quantity R_hbs that measures the relative fraction of peptide specific TCRs in the repertoire according to Hamming distance similarity to the peptide specific TCRs. Importantly (as stated in the methods) a cross validation procedure is employed where TCRs from a given individual are not used for classification of that same individual. Since d is Hamming distance on CDR3 sequences, presumably comparisons are only made for TCRs with identical CDR3 length differences. This seems like a limitation, since clones with identical V and J gene, and CDR3 that differ by only one in CDR3 length could very well bind the same epitope. A more TCR-specific distance function, such as the TCRdist of Dash et al., may significantly increase classifier performance.

The suggestion raised by the reviewer is valid, and one that we had considered when designing the study. There are several methods available for making epitope-specific TCR annotations and our choice was informed by several considerations:

• Our data is primarily beta-chain TCR sequences. Despite the high performance of TCRdist on paired alpha-beta chain TCR data, it has been shown that these approaches do not outperform hamming distances on beta-chain only (Meysman et al., Bioinformatics, 2019). Not that the length-restriction may seem like a large restriction, but in practice length of the CDR3 sequence is known as a strong predictor for epitope preference (De Neuter et al., Immunogenetics, 2018; Meysman et al., Bioinformatics, 2019; Valkiers et al. Bioinformatics, 2021).

• The TCR repertoire data and the epitope-specific TCR data were extracted using different kits (Adaptive vs QIAGEN) due to difference in starting samples (millions of cells vs thousands of cells) and used different processing pipelines. It is well established that these processing can induce a bias in the TCRs that are reported. Thus we opted for the simplest method as it was deemed to be most robust against any such bias. Methods such as TCRdist have been designed to translate findings from one samples derived from an experimental setup to another sample with the same setup.

• The epitope-specific sequences are few in number. We wished to have as high a coverage of the HBs antigen as possible, so opted not to use more advanced methods which usually place a restriction on the minimum number of input TCRs that are required to build an annotation model. In addition, we wished to use an equivalent model for the bystander sequences (the denominator of the Rhbs metric), likely involve a set of TCRs targeting multiple epitopes. This invalidates many approaches which require the assumption of an epitope-specific data set.

Thus we opted for the straight-forward hamming distance approach, which has been applied in several prior studies (notably the look-up functionality of VDJdb uses hamming distance).

The origin of this choice was indeed obscure in the previous version of the paper, and has now been clarified.

There is a distance cutoff parameter c required to define R_hbs. How was this parameter chosen? In particular, if it was tuned to produce the best AUROC, then the cross validation procedure is not legitimate (nested cross validation would be needed, or separate held out test set).

The reviewer is correct. The distance cutoff c cannot be informed by tuning the AUROC as it would invalidate the cross validation procedure due to information bleed.

The cutoff was set based on prior research done several years ago on several independent data sets (Meysman et al., Bioinformatics, 2019). This cutoff was kept to not bias the current results. Furthermore the functions used within the code were already highly optimized towards this cutoff (as it allowed a hashing dictionary to be constructed for fast look-up). As the reviewer rightfully points out, the prior version of the code base did not even allow c to be set and was already baked-into the search algorithm. This has been clarified in the revision.

Author Response:

Reviewer #1 (Public Review):

Sanchez et al investigate how the complex morphogenetic movements that drive epithelial tube formation are patterned to occur with the correct spatiotemporal dynamics by upstream transcription factor expression, a fundamental question in developmental biology. In prior work, the authors defined the cell behaviors that drive tube formation in the Drosophila salivary gland, demonstrating that localized apical constriction induces epithelial bending to form a pit and circumferential cell intercalations narrow the primordium. In this study, they show that as more peripheral cells flow into the deepening pit they switch behaviors to constrict apically, promoting continued morphogenesis of the structure. These behaviors, as well as correlated patterns of myosin pathway activation, require transcription factors Fkh and Hkb suggesting the expression pattern of these TFs may drive the dynamic changes in cell behavior. Using endogenously tagged protein reporters of Fkh and Hkb, they show the two TFs display dynamic expression patterns that initiate roughly where apical constriction will predominate and spread outward to cells that will later constrict into the pit. Fog appears to be a key downstream target of Fkh and Hkb, and interfering with the radial pattern Fog expression disrupts tube formation. Strengths of the study include the high quality, quantitative morphometric analysis in both time and space, the use of endogenously tagged reporters of Fkh and Hkb together with time-lapse analysis, and the multiscale nature of the study encompassing and connecting upstream patterning events to intermediate regulators of cell shape to downstream cell and tissue-level behaviors. A minor weakness of the study in its current form is a lack of cell tracking data that connect cell identity with the stated changes in behavior, something that should be straightforward to address.

These data are inherent in our quantitative analysis, but were not explicitly shown.

Reviewer #2 (Public Review):

The authors analyze the cellular dynamics responsible for the formation of a tubular structure, taking as a model the formation of salivary gland in Drosophila embryo, which is initiated by the asymmetric invagination of cells from a circular placode. They observed a regionalized cellular behavior, dependent on Hkb and Fkh dynamic expression in the placode. Both these transcription factors are required to ensure the correct expression of Fog, which drives localized apical constriction and ensures correct morphogenesis of the salivary gland.

Strengths: This is a detailed analysis of cellular dynamics during salivary gland morphogenesis. This study highlights the regionalized behavior of cells from the presumptive gland (or placode) with a region close to the invagination pit where Hbk and Fkh drive Fog expression, leading to medio-apical myosin accumulation and apical constriction and a more distant region where cells mostly intercalate, a process driven by a junctional Myosin polarity. Although mainly descriptive, these data are precise and convincing. The conclusions fit with the observations.

Weaknesses: Although this work is interesting, it raises a lot of unanswered questions. How is the timing of apical constriction in the placode controlled?

The question about the timing of apical constriction and how it evolves across the placode is exactly what we are trying to address in our study. As we show and discuss, it is the patterned and dynamic expression of the transcription of Fkh and Hkb, starting at the future pit position, that initiates the apical constriction (via downstream expression and action of Fog,) and then part of the later maintenance and continuation of apical constriction as cells move into a position in proximity to the pit is through the continued expansion of the Fkh expression.

What is responsible for the delay in apical constriction observed in Hkb mutant?

We explain this in the discussion of the manuscript (line 570 onwards): We conclude from our data that the initial constriction at the eccentric position where the pit forms depends on both Fkh and Hkb. In the hkb-/- mutant the central cells in the placode manage to constrict in the absence of Hkb likely only require Fkh activity to initiate Fog expression and function (and hence constriction). These cells therefore undergo their normal apical constriction once Fkh expression has reached the central position (as Fkh expression is unaffected in the hkb-/- mutant as we show). As they now are the first cells to constrict and invaginate, the hkb-/- mutants show a central and delayed constriction and invagination.

How does apical constriction propagate? How is the switch between apical contraction and intercalating domains regulated?

Once apical constriction is initiated at the pit position in the dorsal-posterior corner due to the action of both Hkb and Fkh (and downstream Fog), the constriction spreads across the placode, as we describe and analyse, mostly due to expanding expressing of Fkh, driving the expansion of Fog expression in its wake. The cell intercalation behaviour we observe and describe here and in (Sanchez-Corrales et al., 2018) leads to a convergence and extension of the tissue that feeds more cells into a position near the pit. Once Fkh expression has reached cells now in a closer position to the pit they also start to apically constrict.

Author Response

Reviewer #1 (Public Review):

This manuscript provides an interesting analysis of the evolution of the IR75a protein across the Drosophila phylogeny. As reported by the authors previously, IR75a in D melanogaster and several related species preferentially recognize the odor acetic acid, while IR75a in D sechellia instead prefers butyric acid. Here, the authors report that more distant Drosophila species, as well as the putative ancestral IR75a, also prefer butyrate, suggesting that IR75a changed its preference from butyrate to acetate within the melanogaster/obscura group, with reversion to butyrate subsequently occurring in D sechellia. Moreover, the authors identify a key site (position 538) whose identity as Phe or Leu tracks with odor preference. They also identify other secondary lineage-specific mutations that presumably provide structural support to help optimize ligand preference. Interestingly, different solutions for secondary optimization were observed across lineages, suggesting multiple evolutionary and structural paths for tweaking the ligand pocket. These data are generally solid and expertly generated, but I do note that there is substantial speculation based on molecular modeling (which the authors acknowledge) as well as speculation of mutational timeline which should be trimmed or removed. There are many ways to extend these findings, such as linking odor recognition properties to behavior, which would substantially increase the impact of this study.

We fully agree that understanding the behavioral significance of the changes in odor recognition of these receptors is of high interest, but as we do not yet clearly know the behavioral function(s) of these receptors even in D. melanogaster, such an effort is likely to take months (if not years) and goes well beyond the current study. We hope that our identification of tuning changes of these Irs across the Drosophila genus might provide some additional clues to the contributions of these receptors to the odor-guided behaviors of these species.

Reviewer #2 (Public Review):

[...] Weaknesses: The authors attempt to link the characterized molecular events to the ecological needs that might have played a significant role in the linage's evolution, and to the structural aspects of receptor-ligand interaction. These two aspects are on the more speculative side, which the authors themselves acknowledge as limitations of the study.

1. In terms of the ecological context, this study focuses on a narrow set of ligands that are probed at concentrations that are quite high and thus of unclear physiological relevance. While their results are very exciting, they are restricted to the inverse relationship of the responses to C2-C4 carboxylic acids. Although data in the literature shows that these ligands, in particular acetic acid, are likely relevant, other data support that ligands from the same series may also be relevant. In particular, noni volatiles are dominated by octanoic acid (Auer et al., 2020; Pino et al., 2010), and Ir75a responds very strongly to propionic acid as well as acetic acid (Pietro Godino et al., 2016, Silbering et al., 2011). While the C2-C4 relationship captures most of the variance in PCA (which leads the authors to focus on these ligands in the first place), perhaps at other more physiological concentrations the relationship between other ligands in the series becomes more prominent, which would be interesting to explore.

It seems unlikely that odors that do not evoke robust responses at 10^-2 will stimulate biologically meaningful responses if tested at higher concentrations. Regarding the abundant noni volatile octanoic acid, we have previously shown that this odor (surprisingly) does not evoke strong responses in any acid-sensing Ir neuron, nor does it provoke strong olfactory behavioral attraction of D. sechellia (PMID 28111079, Figure 1). These results contrast with the physiological and behavioral responsiveness to hexanoic acid, and suggest octanoic acid may act principally via the gustatory system.

The universe of odors is enormous, so for practical reasons for this study we focused on a simple series of ligands to extract principles of molecular evolution of olfactory receptor specificity. Acknowledging the limitations of our study, we have taken care to state in the Discussion that our findings may only capture part changes of the response properties of individual receptors, and that we can only speculate on the behavioral/ecological significance of at least some particular odor tuning properties.

Regarding the quantitative analyses in our manuscript, the vast majority of the results are not restricted to C2-C4, but rather the whole linear series (C1-C6). In Figure 1E, we present the clustering of species on a C2 vs C4 response plot because we show that these two odors are the major contributors of variance in the data by PCA (Figure 1D) and because this allows easy visualization in two dimensions. When we analyze the statistical significance of the epistatic interactions between different mutations (Figure 3F), we use PC1 from the dataset, to encompass all of the main variance in the data and avoid biasing towards specific odors.

1. The authors don't discuss whether the proposed polymorphisms are found in population genomic data, which is available at least for Dmel and Dsec. Mining these datasets and looking at intraspecific variation (or lack thereof) has the potential to support their speculations on the evolutionary trajectories of mutations with empirical data and offer complementary insight.

We mentioned this briefly in our original submission (there is no informative population genetic variation) and have now added additional information to the manuscript on the results of our survey of intraspecific genetic variation in these drosophilids.

1. A more critical limitation is the use of docking onto homology models. Modeling techniques are incredibly powerful as they can provide solid hypotheses for how protein-ligand interactions might occur. However, much caution should be taken to interpreting modeling results without experimental validation. The reliability of homology models scales substantially with sequence identity, which turns this protein family into rather poor substrates for extracting atomic-scale conclusions from these models. In this case, homology models are combined with docking and very little support is offered. For example, the docking scores presented for the homology models are relatively low, and there is no significant difference between the docking score of acetic and butyric acid onto Dmel Ir75a. Although it is well known that docking results will in general only qualitatively match the behavior of a receptor-ligand pair, in the absence of alternative validation of the modeling procedures, these results fail to convince the reader that the homology models and docking results are reasonably likely. It should be noted that the proposed mode of action is entirely plausible and an interesting possibility, but as it is it appears too speculative and without validation.

We now present what we consider are better protein models to indicate the relative position of different residues in the LBD but have removed all docking analyses.

Author Response:

Reviewer #1 (Public Review):

[...] Some weaknesses of the manuscript design include the prolonged continuous optical inhibition, lack of optical control and the possibility of a state-dependent effect on neural manipulation.

The prolonged optical inhibition would conceivably produce tissue damage at the optic fiber tip. Accordingly, controls for the optic stimulation experiments comprised the experimental groups injected with AAV5-hSyn-mCherry—not coding halorhodopsin (HR-; controls)—that were continuously stimulated for 5 min with a yellow laser (589 nm) during either the acquisition or the expression phase. Moreover, histological analysis of the tissue at the optic fiber’s tip after the behavioral procedure showed no tissue damage in the areas surrounding the optic fiber’s tips (see Appendix Figure 11).

The possibility of a state-dependent effect on neural manipulation was also discussed. It is important to note that for a given pathway to generate a condition to produce a state-dependent effect on memory, the pathway would be expected to be activated to the same degree during both encoding and retrieval. However, in our case, all ACA paths involved in the acquisition or expression of contextual fear responses showed a differential activation between acquisition and expression of fear memory and are unlikely to provide a condition to produce a state-dependent effect on contextual fear memory (see Discussion).

It is unclear if the similarity in the effects obtained in some cases are due to similarity in the underlying behavioral process that is disrupted (e.g. failing to acquire an accurate representation of the context vs. associating the context with the threat).

Our findings collectively support the idea that the ACA integrates contextual and predator-related cues during the acquisition phase to provide predictive relationships between the context and the threatening stimuli and influence memory storage. As discussed in the manuscript, silencing the AM > ACA pathway during the acquisition phase of contextual fear memory would disrupt ACA integration of the predator-related cues and impair the ability of this cortical field to provide predictive relationships between the context and the predatory threat. Conversely, if the ACA outputs were silenced, this cortical region would nonetheless be able to generate predictive relationships between predatory threats and contextual landmarks but would be unable to influence memory storage sites in the BLA and ventral hippocampus or in brain regions involved in the expression of contextual fear responses, such as the PAGdl.

Reviewer #2 (Public Review):

The authors' main goals were to identify key circuits for the acquisition and expression of contextual fear conditioning in a paradigm that uses a live cat as the unconditioned stimulus. They used neuroanatomical tracing, opto-, and chemo-genetic techniques to observe and manipulate activity in anterior cingulate area (ACA) afferents and efferents at either the acquisition or retrieval stages. The strengths include a thorough characterization of multiple circuits and robust behavioral effects. Weaknesses include the confound of the experimenter being in the room for the acquisition, but not retrieval phase, a lack of characterization of escape-like behaviors, and the exclusive use of male animals, which reduces the potential impact.

In the revised version, we clarified that the experimenter was present in the experimental room during the habituation phase, cat exposure, and the context phase; in all conditions, the experimenter’s position in the experimental room remained consistent (see Methods).

We also clarified that our protocol conditions yielded only occasional escape responses. Thus, only a few escape episodes were noted during cat exposure—mostly when the animals noticed the cat and fled back to Box 1—and during exposure to the predatory context, the animals presented only occasional escape episodes (see Appendix Behavioral Protocol).

Finally, we discussed that the basic circuits organizing these responses should be similar in both genders. However, gender differences are expected in terms of the responsivity of the components within these circuits, and further studies are needed to investigate gender differences in the responsivity of circuits mediating contextual fear memory to predatory threat (see Discussion).

Reviewer #3 (Public Review):

In this study, de Lima et. al. examined the function of ACA in acquisition and expression of contextual fear memory to predator threat. The authors found that ACA is necessary for both processes. Using optogenetic terminal inactivation, the authors further demonstrate a necessary role of AM input to ACA in the contextual fear acquisition phase. At the output level, the projections from ACA to BLA and PERI are necessary for contextual fear acquisition while the projection from ACA to PAG is essential for contextual fear expression. Overall, the study is interesting and the results are straightforward. The presented data largely support the conclusions. The paper will provide new insight into the neural circuit underlying contextual fear learning and expression to predator threat. Some limitations of the study include the lack of controls to demonstrate how specific is the ACA response to threat and threat paired context is and validation of the terminal inhibition. Further characterization of the projections would also help to understand these results.

In the revised manuscript, to demonstrate how specific the ACA response is to threat and a threat-paired context, we performed in the ACA, DAPI and Fos quantification in four conditions: exposure to the cat, the predatory context, the novel non-threat stimulus (plush cat), and the context paired with non-threat stimulus. We were able to show that ACA Fos expression in response to the cat and cat-related context was significantly higher relative to its expression level in response to exposure to a novel non-threat stimulus and to the context paired with a non-threat stimulus (see in Appendix, Comparison of ACA activity and Figure 3; and Discussion).

We have also validated the terminal inhibition and performed patch clamp experiments that tested the efficiency of pre-synaptic inhibitions and showed clear inhibition of EPSC of postsynaptic cells after illumination at 585 nm light on halorhodopsin positive fibers, and no rebound excitation after halorhodopsin activation (see Appendix Figure 6).

Finally, to further characterize the ACA projections, first we performed the quantification of ACA terminal fields in the BLA, PERI, PAGdl and POST, revealing the densest ACA projection field in the BLA, followed by the projections to the PERI, PAGdl, and POST, which contained the weakest projections from the ACA in all cases examined (see in Appendix, Quantification of ACA terminal fields and Figure 8).

Next, we performed triple retrograde tracing in the same animal to investigate the distribution of ACA neurons projecting to the BLA, PERI and PAGdl, the ACA projections involved in the acquisition or expression of contextual fear memory (see in Appendix, Triple retrograde tracing and Figure 9). The results revealed a layering pattern of the ACA neurons projecting to BLA, PERI and PAGdl. ACA projecting neurons to PAGdl are located in infragranular layers (layers V and VI) and do not overlap with the neurons projecting to the BLA or the PERI. Conversely, neurons projecting to PERI and BLA are located in the supragranular layers (layers II and III) and present a nearly 50% overlap.

Author Response:

Reviewer #2 (Public Review):

The authors used an operant task in voles to assess preferences for social relationships and whether these preferences differed by sex and species. They also correlated some outcomes with oxytocin receptor binding.

A strength of the paper is the use of the vole models which allow comparisons between socially monogamous (prairie voles) vs. promiscuous breeders (meadow voles). Because prairie voles show a stronger preference for peers and mates than many other rodent species, they are a great model to assess selective relationships. The other major strength of this paper is the use of the operant procedure to assess preference. To my knowledge, this is the first time this procedure has been used to assess social reward in a monogamous species, and it has advantages over place preference procedures.

A weakness of the paper is the omission of groups from certain manipulations. Male meadow voles were excluded from the operant procedure and all male data was excluded from oxytocin receptor analysis with little rationale. This limits some of the conclusions that can be made about males in the study. Additionally, some experimental design details were missing. A subset of rats went through extinction. However, it was unclear whether those used for oxytocin receptor analysis went through extinction or not (or were from both conditions). The biochemical analysis also assessed the effect of oxytocin receptor genotype, replicating the effect that C allele carriers have higher oxytocin receptor binding in specific brain regions. However, the analysis of this genotype's effect on behavior was limited due, presumably, to power issues with most measures. Thus, conclusions regarding genotype were limited. Finally, there was a missed opportunity to do a progressive ratio test to better assess motivation for the partner rat.

The operant task and the subsequent behavioral results will be useful for the field, but the design issues somewhat limit impact. However, assessing the formation of selective relationships and using the vole model is innovative.

Regarding testing protocol, we exclusively used a progressive ratio schedule (PR-1) in all testing phases of our study. This has been emphasized throughout the manuscript, and hopefully obviates any concerns about limits to the interpretation of fixed ratio studies.

Excellent point about the need for more detail on inclusion of each sex. Information has now been added as to why males were not tested in two aspects of the study. Males were not included in the meadow vole portion of the study because only females of this species show pronounced seasonal differences in affiliative behavior. We have added the text: “Because the seasonal transition from solitary to social is most pronounced in female meadow voles in the field and laboratory (Madison and McShea, 1987; Beery et al., 2009), only females of this species were used.” We initially planned to include males in oxytocin receptor assays, but when early results suggested males did not work harder to access familiar females, we retained later study males to test a two-choice variant of the social operant setup (instead of collecting their brains at the conclusion of this study). That pilot led us to conduct a new study using the two-choice operant apparatus (in review), and the tissues from that second study could enable a similar analysis in both male and female brains.

Additional details have been added regarding which voles underwent extinction, and we have removed the genotype data from the manuscript as requested.

Author Response:

Reviewer #1 (Public Review):

In this manuscript by Gilbert et al., the authors found that MLC1 is required for postnatal maturation of perivascular astrocyte coverage. Through various detailed experiments, they further found that Mlc1 KO mice showed a number of defects, including the reduced VSMC contractility, neurovascular coupling and parenchymal CSF flow. The data is well presented, however there are several points that need to be addressed to strengthen the manuscript.

1) Since many PvAP proteins showed the normal expression after P60 in Mlc1 KO mice, it is also possible that many of the phenotype that the authors presented, such as the reduced VSMC contractility, neurovascular coupling and parenchymal CSF flow, can be recovered after P60. This would be important as well to understand the prime pathological cause of megalencephalic leukoencephalopathy induced by MLC1 deletion.

Although some protein levels are normal in Mlc1 KO mice, PvAP coverage and gliovascular unit morphology are altered at P60. We also now show (using TEM) that PvAPs are swollen in 1-year-old Mlc1 KO mice (the new Fig. S4) - indicating that edema develops progressively in Mlc1 KO mice. Myelin vacuolation starts at 3 months and progressively worsens (Dubey et al., 2015). Taken as a whole, these data show that MLC is a degenerative disease. Under these conditions, the recovery of gliovascular unit function after P60 is very unlikely.

All the molecular morphological and functional changes in gliovascular unit described in our manuscript precede myelin degradation, which starts at the age of 3 months in Mlc1 KO mice (Dubey et al., 2015). Moreover, our previous study (Gilbert et al., 2019) demonstrated that MLC1 expression starts around P5 and that the MLC1/GlialCAM complex is only mature at P15. Taken together, these results strongly suggest that gliovascular unit alterations are the primary pathological events in MLC.

Dubey M, Bugiani M, Ridder MC, Postma NL, Brouwers E, Polder E, Jacobs JG, Baayen JC, Klooster J, Kamermans M, et al. 2015. Mice with megalencephalic leukoencephalopathy with cysts: a developmental angle. Ann Neurol 77: 114-131.10.1002/ana.24307. Gilbert A, Vidal XE, Estevez R, Cohen-Salmon M, and Boulay AC. 2019. Postnatal development of the astrocyte perivascular MLC1/GlialCAM complex defines a temporal window for the gliovascular unit maturation. Brain Struct Funct 224: 1267-1278.10.1007/s00429-019-01832-w.

2) Does CBF get differed only after neuronal stimulation in Mlc1 KO mice? It is unclear whether the basal CBF/neurovascular coupling level is disrupted as well in Mlc KO brains and how this defect is related to the reduced vasoconstriction in these mice.

The baselines of the functional ultrasound experiments were aligned prior to stimulation. This technique measures the percentage increase in blood flow after neuronal stimulation (here, whisker movement) but does not measure the basal flow and does not enable one to distinguish between an abnormal basal cerebral blood flow (as suggested by the reduction of the arterial diameter) and the loss of vascular contractility - both of which probably contribute to the defect in neurovascular coupling.

3) The reduced cohesiveness of PvAPs and the associated neuronal fibers to the vessel in Mlc1 KO brains should be validated with additional experimental approach.

To strengthen our analysis, we now give the results of a parallel quantitative immunofluorescence analysis of purified brain vessels (presented in a new figure, Fig.5). The results show that part of the Aqp4 and NF-M perivascular immunolabeling is absent in the Mlc1 KO. Taken as a whole, our data demonstrate that PvAPs and the associated neuronal fibers (which normally remain attached to brain vessels during mechanical purification) are lost during the purification process in Mlc1 KO mice but not in the WT. In conclusion, the absence of MLC1 reduces the mechanical cohesiveness of PvAPs and the associated neuronal fibers.

4) The defective polarity of astrocytes should be better described by using other markers other than GFAP. The distribution of Aquaporin4, Cx43 or several glutamate transporters in the specific compartment of astrocytes can be examined.

GFAP is the marker typically used to analyze the astrocytes’ overall morphology and polarity. Nevertheless, we agree that it is of interest to study the molecular polarity of PvAPs. Indeed, morphological changes in the PvAPs and astrocytes and changes in polarity in Mlc1 KO might all influence the localization of molecules in PvAPs. To address this question, we performed a quantitative stimulated emission depletion (STED) analysis of protein localization in PvAPs. Our results indicate that the perivascular localization of aquaporin 4 was not affected. However, the density and size of Cx43 puncta were greater - indicating that the gap junctions in PvAPs are not organized in the same way in the Mlc1 KO as in the WT. This observation is consistent with our electron microscopy observations of perivascular astrocytic processes stacked on the top of each other and linked by extended gap junctions.

We have also added results for Kir4.1, a potassium channel that is expressed preferentially in PvAPs. The Kir4.1 expression level in Mlc1 KO was lower at all stages of development, indicating that perivascular potassium homeostasis was probably perturbed. These results are interesting because (i) epilepsy is a significant component of megalencephalic leukoencephalopathy (Dubey et al., 2018; Yalcinkaya et al., 2003), and (ii) Kir4.1 deletion or downregulation is associated with greater susceptibility to epilepsy (Sibille et al., 2014). These points are now discussed.

Dubey M, Brouwers E, Hamilton EMC, Stiedl O, Bugiani M, Koch H, Kole MHP, Boschert U, Wykes RC, Mansvelder HD, et al. 2018. Seizures and disturbed brain potassium dynamics in the leukodystrophy megalencephalic leukoencephalopathy with subcortical cysts. Ann Neurol 83: 636- 649.10.1002/ana.25190. Sibille J, Pannasch U, and Rouach N. 2014. Astroglial potassium clearance contributes to short-term plasticity of synaptically evoked currents at the tripartite synapse. J Physiol 592: 87- 102.jphysiol.2013.261735 [pii] 10.1113/jphysiol.2013.261735. Yalcinkaya C, Yuksel A, Comu S, Kilic G, Cokar O, and Dervent A. 2003. Epilepsy in vacuolating megalencephalic leukoencephalopathy with subcortical cysts. Seizure 12: 388-396.10.1016/s1059-1311(02)00350-3.

5) The authors provide interesting observations such that the formation of perivascular astrocyte coverages is required for the dissociation of the contacts between neuronal components and the vessel during development. The authors need to discuss more about potential regulation and implication of this phenomenon.

This is indeed a fascinating phenomenon. The postnatal period is also an intense synaptogenic phase in the mouse brain (Chung et al., 2015), during which astrocytes and neurons might compete for the perivascular space. In the absence of MLC1 and thus PvAPs, the neurons might expand into the free space. We now comment on this point.

Chung WS, Allen NJ, and Eroglu C. 2015. Astrocytes Control Synapse Formation, Function, and Elimination. Cold Spring Harb Perspect Biol 7: a020370.cshperspect.a020370 [pii] 10.1101/cshperspect.a020370.

6) It is interesting that DOTA-Gd tracer shows different traces in Mlc1 KO brains. However, it is unclear how MLC1 deletion affects glymphatic system. Does the tracer normally enter to the perivascular spaces in Mlc KO brains? Does the tracer leak out more from the perivascular spaces in Mlc1 KO mice? Is the general clearance or drainages of the tracer impaired in Mlc1 KO mice? Would these defects be originated by the reduced perivascular astrocyte coverage or the reduced vasoconstriction itself?

Paravascular transport (as revealed by the injection of a tracer into the CSF) depends mainly on dispersion of the tracer in the subarachnoid space (SAS), the cisternae, and the parenchyma (including the interstitial and perivascular spaces). The uneven, slow dispersion of the tracer within the SAS (compared with dispersion in the blood) means that the tracer’s kinetics in the parenchyma are regiondependent. These differences can be accentuated by regional differences in the anatomy of the brain’s vasculature, i.e. the presence or absence of a perivascular space and the vessel’s topology. Lastly, the amount of DOTA-Gd available for diffusion within the parenchyma depends directly on its local concentration in the SAS. This can be seen on our contrast concentration maps (see Fig. 8), where the highest DOTA-Gd concentrations are found near the injection site (the cisterna magna), in line with previous reports (Iliff et al., 2012). In the Mlc1 KO model, dispersion of DOTA-Gd is presumably affected in the SAS and the parenchyma.

With regard to tracer dispersion in the SAS and the cisternae, our anatomical MRI showed that the brain volume is greater in the Mlc1 KO mouse than in the WT (see Fig. 1). These variations in the geometry of the SAS may account for much of the difference between the Mlc1 KO mice and WT mice. Although tracer concentrations appear to be similar in the cerebellum (close to the injection site), they are much lower in the more distant septal area of Mlc1 KO mice - suggesting that tracer transport within the SAS is restricted.

With regard to parenchymal dispersion, we showed that MLC1 is essential for the position of the astrocytes’ perivascular endfeet. Thus, in Mlc1 KO mice, the formation of the perivascular space (as a conduit for solute distribution) is likely to be deficient. This aspect is revealed by the slope of the tracer’s concentration-time curve, which indicate slower kinetics in Mlc1 KO mice; this might be due to poor integrity of the perivascular space. The higher volume of fluid in the Mlc1 KO parenchyma (reflected by the increased apparent diffusion coefficient (ADC); Fig. 1 and S1) might also be involved in this phenotype.

The heart beat is the main driver of CSF circulation in the perivascular space (Iliff et al., 2013). The heart rate is very rapid and so the heart exerts a much greater driving force on the CSF than the vasodilation of the vessels induced by neuronal activity. Alterations in vascular contractility observed in Mlc1 KO mice might be involved in the impaired CSF flux but this is unlikely.

All these points are now discussed in the revised version of the manuscript.

Iliff JJ, Lee H, Yu M, Feng T, Logan J, Nedergaard M, and Benveniste H. 2013. Brain-wide pathway for waste clearance captured by contrast-enhanced MRI. Journal of Clinical Investigation 123: 1299-1309.10.1172/jci67677. Iliff JJ, Wang M, Liao Y, Plogg BA, Peng W, Gundersen GA, Benveniste H, Vates GE, Deane R, Goldman SA, et al. 2012. A paravascular pathway facilitates CSF flow through the brain parenchyma and the clearance of interstitial solutes, including amyloid beta. Sci Transl Med 4:147ra111.10.1126/scitranslmed.3003748.

Reviewer #2 (Public Review):

This very interesting manuscript by Gilbert and colleagues uncovers that the astrocyte specific membrane protein MLC1, the mutation of which causes a rare disease called megalencephalic leukoencephalopathy with subcortical Cysts (MLC), plays a fundamental role in the postnatal development of the gliovascular unit and the organization of the perivascular astrocyte processes, in particular. To reach this conclusion, the authors used an elegant multiscale approach including in vivo MRI, in vivo functional ultrasound, ex vivo analysis of vascular constriction, anatomical approaches at the light and electron microscopic level, and molecular characterization of the gliovascular unit from isolated microvessels. The manuscript is very well-written although it uses too many (unnecessary) abbreviations, which prevents a fluid reading of the manuscript, results are well illustrated and convincing and the discussion is reasonable.

I have a major concern regarding the results reported in Figure 4D, which seem somewhat contradictory to those shown in Figure 6A-F. Indeed, the authors report in Figure 4D that there is less Neurofilament-M protein around isolated microvessels in MLC1 KO mice, whereas Figure 6A-F shows that these animals have more neuronal processes in contact with the vessels than in wiltypes. How can the authors explain this?

The two situations are not comparable. On one hand, we observed the structure of the gliovascular unit in situ in fixed tissues. On the other, we mechanically purified microvessels. The detachment of astrocytic processes and associated neuronal fibers (linked to the mechanical dissociation of microvessels in the Mlc1 KO mouse) was clearly not counterbalanced by the presence of neuronal fibers contacting the vessels.

Author Response:

Reviewer #2 (Public Review):

The study by Ranzano et al. set out to reveal if the spinal cord contain motor circuits that can support co-activation and co-inhibition of diverse flexor and extensor motor neuron pools in the mouse spinal cord. For this they use modified rabies virus in a mouse model and with a set-up that will allow selective mono-synaptically restricted labeling of premotor neurons projecting to functional synergist or antagonist ankle motor neuron pools along the entire spinal cord. They show that a minor percentage of premotor motor neurons projecting to either synergist or antagonist pair of motor neuron pools diverge. Divergent premotor neurons were seen both close and rostral to the target lumbar motor neuron pools, but with an increased proportion with distance from the lumbar cord. In the cervical spinal cord the largest proportion of divergent neurons where commissural excitatory neurons with molecular characteristics of the V0 class. The study provide important, new and convincing data on the spinal anatomical landscape of distributed motor networks that may coordinate synergistic activity as well as mediate co-contraction of antagonist muscles across multiple motor pools in the same limb or across limbs. Overall the claims are well supported by the data. Some aspects of methods could need clarification and some aspects of the claims are weakened by lack of identification of premotor neuron populations. The discussion of the data could perhaps be made stronger by linking the present data to functional studies.

1) Transsynaptic method. The authors use a different model for trans-synaptic tracing than most previous studies in the spinal cord: namely the RGT mouse line crossed with ChAT-cre mice and combined with a retrograde labelling of motor neurons. The distribution of transsynaptic flexor and extensor related premotor neurons in this model is different from previously reported. Data for this are presented in (Ronzano et al. 2021 BioRxiv). But it will be useful to mention this here as well.

In this manuscript we show maps of single and double labelled neurons throughout the thoracic and cervical cord, for which, to the best of our knowledge, there are no published studies. Within the lumbar cord, we have focused only on double labelled neurons and we refer to our preprint for maps of single labelled interneurons. The issue of the different distributions obtained using different tracing methods is currently being addressed more extensively in the work that will follow that preprint. We feel that its details (some of them contentious with existing literature) would deviate the attention from the main point of the current manuscript, which is the presence and distribution of divergent premotor interneurons.

The authors discuss why the labeling is not caused by a second jump from V0C neurons or motor neurons labelled by collaterals. Another course of contamination is at the level of the muscle. All injections are done in newborn mice with tiny muscle. It would be useful to know how the authors secured that there is no viral spread peripherally.

A paragraph describing how we assured that there was no viral spread to other muscle(s) than the targeted one has been added in the methods. Briefly, before processing the tissue, the injected leg was dissected and muscles below and above the knees were examined under a fluorescence microscope to make sure the virus had not spread to any of the adjacent muscles. This was particularly critical for injection of synergist muscles, that are located side by side along the leg and are not in separate anatomical compartments separated by the fibula, as is the case for the antagonist GS and TA.

2) Identity of neurons. A limitation of the study is that there is no transmitter phenotyping of the divergent premotor neurons in the lumbar and thoracic region. The divergent neurons can be either excitatory or inhibitory and cause coactivation or co-inhibition, respectively of synergist or antagonists. Except for the cervical CNs there is no evidence for transmitter-phenotype in the data. Perhaps the authors could just mention that this would have required in situ hybridization in the double injected animals or a third color RabV togheter with the GlyT2-GFP mice. The identification of V0 neruons interneurons is suboptimal. The use of specific AB (Evx) for the V0 population could have provided a better characterization (see Crone et al. 2008). Maybe also mention that the commissural neurons could be SIM1.

We agree with the referee that the neurotransmitter phenotype is a crucial point. Our data show that while the cervical divergent interneurons are certainly not glycinergic or cholinergic (and therefore presumably excitatory), many of the thoracic and lumbar divergent interneurons are indeed inhibitory. Detailed analysis of divergent interneurons by phenotype would have required an a priori different study design and we agree with the reviewer #1 and #3 comment that the present work will drive further studies aimed at understanding the nature and function of divergent interneurons. In the new supplementary figures, we show examples of both glutamatergic and non-glutamatergic terminals apposed to motoneurons in the lumbar and thoracic region to highlight the finding that divergent premotor interneurons can be inhibitory or excitatory (as well as cholinergic, see new supplementary figure S3 and S6, in agreement with Stepien et al, 2011). This is now explicitly mentioned in the corresponding text (line 206-208). We agree that the identification of cervical divergent interneurons as belonging to the V0 family is not definitive. We have made a further attempt at identification by using an Evx1 antibody, as suggested by both reviewer #1 and #2, but unfortunately the downregulation of Evx1 prevented reliable labelling in test tissue of the same age as the one used for the paper, a problem we had also using the Lhx1 antibody, though to a lesser extent (see Supplementary figure S7). Another approach would be to use genetic labelling by crossing the ChAT-Cre mice with a (non-cre) reporter line for V0 interneurons and perform injections in their progeny with the RGT mice. This would require acquiring/rederiving a colony of reporter mice and performing the experiments on third generation animals. Also, the breeding of the RGT mouse line that we had at UCL had to be suspended during the closure of the labs due to the pandemic, and they would have to be re-derived from frozen embryos. Thus, these would be prohibitively long experiments. We agree that we cannot conclude that the cervical divergent interneurons belong to the V0 class and, in line with the similar concern of Reviewer 1 we have further toned down the part of our manuscript related to the identification of descending interneurons (see above).

3) The discussion is insightful but should perhaps link the data more directly to functional studies for example by considering how synergies are bound together across limb during locomotion which could both involve co-activation of synergies or co-inhibition of antagonists from commissural neurons (see Bellardita and Kiehn 2015; Butt and Kiehn 2003) or ipsilateral neurons (Levine et al. 2014; among others). It would also be useful to discuss how co-inhibition of synergists leads to functional movements.

As mentioned also by the other reviewers, this manuscript does not have any functional data. Determining the function of divergent interneurons is our next project, one that could not have been conceived without the findings we present here. Co-inhibition of synergists (Ia inhibitory interneurons for instance) is a well described phenomenon. On the other hand, co-inhibition of antagonists has received little to no attention. We could speculate on its functional role, for instance in movements in which limbs need to passively follow inertia (arm follow-through after the throw of an object, or foot movements during flamenco dancing) rather than maintain fine control of joint angles, but we feel that in the absence of functional data, such considerations would add too much speculation to this manuscript.

Reviewer #3 (Public Review):

The study by Ronzano et al uses monosynaptically-restricted transsynaptic labeling to reveal populations of "divergent" premotor interneurons, neurons that are presynaptic to multiple motor neuron pools. Various pairings of muscle injections are analyzed, including hindlimb synergists and antagonists, and hindlimb-forelimb combinations. They show that divergent neurons innervating both synergist and antagonist motor pools are similarly located and found in similar numbers. These neurons exist throughout the cord, in decreasing number but higher proportions of the total labeled neurons with more rostral locations (lumbar to thoracic to cervical cord). A subset of the population of the long descending divergent neurons is identified as part of the V0 class, with some similarly located (and possibly overlapping) neurons projecting the forelimb muscles in addition to hindlimb muscles. Other studies have shown distributions of premotor interneurons from single motor neuron pools. The novelty here is the focus on interneurons with divergent innervations of multiple motor pools.

One of the major differences (and advantages) of this study from prior muscle injections of transsynaptic virus is that rather than co-injecting the AAV containing the G glycoprotein, necessary for transsynaptic transfer, into the muscle with the modified Rabies virus, they genetically specify it to motor neurons (and other cholinergic neurons) using Chat-Cre. The advantage of this is that the potential confounds of transfer via the sensory neuron and/or selective AAV transfection/G expression by a particular subpopulation of motor neurons is removed. Although the possibility of 'double jumps' (MNs to ChAT interneurons to upstream neurons) is possible, it is less likely to occur in sufficient numbers in the time of the experiment.

The data presented are comprehensive for the motor pools examined. The location analyses are extensive. Divergent neurons demonstrated by the dual virus strategy are further supported by the demonstration of terminals of hindlimb premotor interneurons synapsing on thoracic and cervical motor neurons. The manuscript is well written. Overall, data are clearly presented and limitations are fully discussed. This study can form the basis for future studies regarding specific identity and function of the divergent populations. Main limitations of the study are related to the tracing technique used. The authors are fully transparent about these limitations in the Discussion. As pointed out, this technique is not optimal for quantifying double labeled neurons. Conclusions regarding the existence and location of neurons projecting to at least two motor pools can be made. However, these are likely to be severely underestimated and it is not possible to determine if these neurons are more broadly presynaptic to other motor pools in the limb (or beyond). The reduced efficiency of infection by two viruses due to viral interference is mentioned and relevant sources demonstrating the limitation are cited. Therefore, the data are solid but the functionally-related interpretations and conclusions are somewhat limited and speculative. The other related limitation inherent in the technique is the efficiency of transfer of even a single virus. Analysis is presented regarding a comparison with a more efficient virus, suggesting transsynaptic efficiency may be ~25%, but this does not fully get at the issue. The efficiency of the starter or seed cell labeling is not mentioned. Quantification of motor neurons taking up the RabV would be helpful as this will be directly related to the potential number of presynaptic neurons. This is especially crucial with the forelimb injections, in which 0-6 muscles were injected.

The MNs infected from the injection have been quantified in every second section across the lumbar cord; these data are now reported in table S3, together with the number of infected interneurons. However, these numbers are most likely to be underestimated due to the toxicity of RabV. For the forelimb injections, the aim of the experiment was to address whether at least some of long descending premotor interneurons were also premotor to any of the forelimb muscles. We therefore performed the injections without aiming at muscle selectivity (these are much smaller muscles), but at widespread infection from multiple muscles. As a consequence, we did not count and map cervical motoneurons or interneurons, since this measure would not have been a reflection of the premotor interneuron population at the level of the single muscle, contrary to what we achieved for the hindlimbs injections.

The percentage of divergent interneurons is also underestimated in that the denominator is single labeled neurons presynaptic to either motor neuron pool. Information may be gained by determining whether different portions of premotor neurons to one over another are more likely to be divergent. This is particularly the case with antagonist injections but also for comparisons of relative proportions of dual labeled neurons premotor to synergists and antagonists. This would likely need to be combined or controlled by the percent (or number) of motor neurons from each pool that are labeled to indicate potential differences in starter cells as mentioned above. Counterbalancing is mentioned in the Methods but it is not clear that is fully possible with an n=2 or 3.

As shown in our supplementary figure 10, working out the proportion of double infected neurons is affected by many confounding factors, not only the (unknown) efficiency of transsynaptic labelling and the viral interference, but also the inevitable differences in the number of starter cells within and across experiments. It is difficult to draw conclusions with so many underlying unknowns, and we agree that the confounding factors would cause an underestimate of the level of divergence, In this new version of the manuscript, we have added Venn diagrams (Figure S1-3.1C-D, S1-3.2C-D, S1-3.3C-E) with the raw numbers of single and double infected cells per muscle in each experiment. Note that despite the large (up to 5-fold) differences in the number of single infected neurons, the proportion of divergent cells is very similar across experiments and muscle pairs.

Author Response:

Reviewer #1 (Public Review):

This research follows up on prior work showing that human visual pattern discriminability is closely related to the statistical features of natural scenes. The present work developed a behavioral choice paradigm to test whether rats could discriminate between patterns, and then measured their sensitivity to different spatial correlation structures. This allowed them to test whether rats possess the same sensitivity to spatial correlation patterns as had been observed in human psychophysical experiments. The experiments found that the ordering of the sensitivity to spatial correlation patterns matched that measured in humans and follows the frequency of different structures in natural scenes, in accordance with an efficient coding hypothesis.

This work has a strong theoretical grounding. The behavioral experiments are well executed and the results show convincingly that rat behavior follows the same pattern as measured in humans. One strength of this data is that is shows that the order of presenting of these patterns during training did not matter for the eventual relative sensitivity measured in the rats.

This research opens up the opportunity to test whether the correlational sensitivities can be altered by changing visual environments, and if so, what neural substrates might be plastic in these cases.

We thank the reviewer for their enthusiastic support. In our revised manuscript, we have redesigned Figure 4 (i.e., our former Figure 3) to highlight even better the fact that the relative sensitivity to different patterns was largely independent of the training of the animal.

Reviewer #2 (Public Review):

Caramellino et al., investigated whether rat sensitivity to multipoint correlations show a similar rank order as observed in humans. They show that rat sensitivity to multipoint correlations exhibits a rank order similar to what was shown in Hermundstad et al., 2014. Interestingly, they also show that such rank order is robust within-group and within-subject. The authors further claim that this similarity indicates that rat sensitivity to multipoint correlations follows efficient coding of natural scenes.

The main conclusion of the paper is mostly supported by the data. However, the presentation of results may benefit from some points of clarification:

1) In Hermundstad et al., 2014, the degree of variation in images themselves (Figure 1E) as well as perceptual sensitivity comparison (Figure 2D and 3A) was done among: 2nd-order horizontal/vertical edges, 2nd-order diagonal edges,-shaped 3rd-order correlations, and 4th-order correlations. However, the comparisons here are: first-order, all 2nd-order correlations including ALL horizontal/vertical/diagonal edges, L-shape 3rd-order correlations, and 4th-order correlations. It is unclear how these two rank-order results are parallel given that Hermundstad et al., 2014 did not include 1st-order at all.

First of all, let us clarify some confusion that seems to exist about the identity of the texture patterns that we tested in the experiment. We did not, as the Reviewer seems to imply, test (and pool over) ALL 2nd order and ALL 3rd order correlations. What we did was to choose one pattern for each order, namely the horizontal 2-point pattern and the “bottom right pointing” 3-point pattern, to build our stimulus set. We apologize for not explaining this clearly enough in the initial version of our text. We now state this more explicitly (lines 88-114), and we discuss at length the rationale behind our choice of experimental patterns (beyond the aforementioned passage in the Results, also in the Discussion, lines 190-256, and the Methods, lines 304-333).

Regarding the inclusion of the 1st order statistic, is true that it was only studied in Victor and Conte 2012, which only contained phychophysics data, and not in Hermundstad et al 2014, which connected psychophysics with natural image statistics. Indeed, it is not possible to analyze the variability of γ in natural images with the method established by Hermundstad et al 2014, because each image is binarized in such a way to guarantee that γ=0 by construction. In this sense, like the use of qualitative ranking discussed more at length below, gamma was included to better reflect the approach in Victor and Conte. Moreover, we wanted to include a sensory stimulus condition that we were sure the animals could detect well, in order to ensure that any failure to learn or perform the task was due to limitations in sensory processing and not in the learning or decision-making process. Before performing our experiments, the only statistic that we were confident the rats could be trained to distinguish from noise was gamma [Tafazoli et al 2017, Vascon et al 2019], and therefore it made sense to include it in the experimental design. We have modified the Results (lines 90-93, 104-108), the Methods (316-321) and the Discussion (212-218) to express this point more clearly.

2) In Hermundstad et al., 2014, the paper emphasized that the difference of perceptual sensitivity between horizontal/vertical edges and diagonal edges is not merely an "oblique effect": Horizontal and vertical pairwise correlation share an edge, while pixels involved in diagonal pairwise correlation only share a corner. One wonders whether rats show any sensitivity difference between horizontal/vertical edges and diagonal edges. The manuscript in its current form misses this important comparison. Without showing this, the rat sensitivity does not fully reproduce the trend previously observed in humans.

When designing our experiment, we prioritized collecting data for the other statistics as they were closer to the extremes of the measured sensitivity values, therefore offering a clearer signal for a comparison with rat data. For instance, had we found better sensitivity to 3- or 4-point statistics than to (horizontal) 2-point statistics, this would have been a very clear sign that perceptual sensitivity in rat is organized differently than in humans. Conversely, we reasoned that a comparison based on 2-point diagonal instead of 2-point horizontal would have been more easily muddled and made inconclusive by the experimental noise that we expected to observe in rats. We agree that, given the high precision of the quantitative match between rats, humans and image statistics now highlighted by the new Fig. 3, it would be interesting to test rats also for their sensitivity to diagonal 2-point correlations and check whether they matched the pattern exhibited by humans. However, as the editor rightly surmises, acquiring new data at this stage would indeed be exceedingly time consuming. Therefore, we have modified the text to better highlight that we did not seek to replicate this particular result in Hermundstad et al 2014 (as well as that we could not test as many correlation patterns as in Hermundstad et al 2014 more generally, due to practical and ethical constraints). We also note that, since we did not test 2-point diagonal, we can’t draw conclusions similar to those in Hermundstad 2014 about the difference of an effect due to efficient coding and one due to a hypothetical oblique effect for the specific 2-point horizontal vs. diagonal comparison. These points are now all brought up in the Discussion of our revised manuscript (lines 189-208). It is also worth noting that the oblique effect was a minor point of the Hermundstad et al. paper and the main arguments did not hinge on it.

3) Combining 1) and 2), it is unclear why the ranking in rat sensitivity is evidence for efficient coding. In Hermundstad et al., 2014, efficient coding was established by comparing the image-based precision matrix with the human perceptual isodiscrimination contours. There is no such comparison here.

4) If the authors would like to hypothesize that the rat sensitivity shows efficient coding simply because its ranking is similar to humans, more needs to be done to shore up the quantitative comparison between the two.

In response to points 3 and 4: We thank the reviewer for underscoring the difference between, on the one hand, a quantitative comparison of the sensitivity to the variance of the statistics in natural images, and on the other hand a more qualitative comparison of their rank ordering. Besides our answers to point 1 and 2 above, we wish now to address specifically this important distinction. In our initial submission, we built our argument based on the rankings in order to better connect not only with Hermundstad et al 2014, but also with earlier human psychophysics results on the same task (Victor and Conte 2012), where there was no comparison with natural image statistics and therefore only the qualitative ranking among sensitivities was examined. We also note that Hermundstad et al do, in fact, make ample use of the rank-ordering agreement between natural image statistics and human sensitivity in order to support their argument (“rank-order”, or similar locutions, are used six times between the results and the discussion). In this sense, while it is true that “In Hermundstad et al., 2014, efficient coding was established by comparing the image-based precision matrix with the human perceptual isodiscrimination contours”, it is also true that the rank ordering was presented as part of the evidence for efficient coding.

Having said this, we nevertheless agree that our argument can be strengthened by presenting both approaches, qualitative and quantitative. We have now added a new figure (Fig. 3), where we compare our estimates of psychophysical sensitivity in rats with the corresponding values for human psychophysics and natural image statistics reported in Hermunstad 2014 (note that we were only able to compare three out of the four statistics that we tested, because – as the reviewer themselves noted in a previous comment – Hermunstad et al. did not consider 1-point correlations). The comparisons in Fig. 3 (and the related quantitative measures reported in the text, lines 146-160) reveal a strong quantitative match, similar to that between the human psychophysics and the image statistic data.

Author Response:

Reviewer #1 (Public Review):

Gastfriend et al present an analysis of the properties and gene expression patterns elicited by pharmacologically activating the Wnt signaling pathway in human pluripotent stem cells (hPSCs) that have been cultured under conditions that favor their differentiation to naïve endothelial cell (EC) progenitors. The result of Wnt activation is partial induction of blood-brain barrier-like properties, as judged by immunostaining for several well characterized marker proteins, trans-epithelial resistance, and RNAseq. Most of the experiments, and the largest effects, were obtained with CHIR99021, a small molecular weight inhibitor of GSK3-beta, the kinase that phosphorylates beta-catenin, leading to its ubiquitinylation and proteosomal degradation. Interestingly, low-passage ("naïve") hPSC-derived ECs were more susceptible to CHIR99021-induced BBB-like conversion than higher passage ("mature") hPSC-derived ECs. The experimental data is of uniformly high quality.

By way of context, several publications describe the use of hPSC or other starting cell sources for the generation of ECs with BBB-like properties, and several have described the BBB-enhancing effects of activating different signaling pathways (retinoic acid, TGF-beta, Wnt). The effect on BBB properties of manipulations that would be predicted to increase Wnt signaling varied among published studies - it was detectable in some studies (e.g. Paolinelli et al., 2013, Laksitorini et al., 2019) and it was undetectable in another (Sabbagh and Nathans, 2020). The present work adds to this literature and presents additional information on this attractive experimental system for dissecting Wnt signaling and potentially other signaling systems that affect the BBB-like differentiation of CNS ECs.

One potential question raised by the present study is the interpretation of the response to CHIR99021. GSK3-beta has many substrates, not just beta-catenin. Could the bioactivity of CHIR99021 for the BBBlike conversion reflect a combination of beta-catenin stabilization and reduced phosphorylation of other GSK3-beta substrates?

We thank the reviewer for this important comment and have addressed this possibility with expanded analysis of RNA-seq data and expanded discussion.

A minor point on page 8, lines 180-185. It might be appropriate to note that neural-rosette-conditioned and astrocyte-conditioned media may contain factors in addition to Wnt7a.

Because we have removed the neural rosette- and astrocyte-CM approaches from the Results section of the manuscript, we have instead included the following statement in the Discussion:

“Importantly, neural progenitor cells and astrocytes likely would also contribute other yet-unidentified ligands important for acquisition of CNS EC phenotype.”

Author Response:

Reviewer #1:

Chen et al. trained male and female animals on an explore/exploit (2-armed bandit) task. Despite similar levels of accuracy in these animals, authors report higher levels of exploration in males than in females. The patterns of exploration were analyzed in fine-grained detail: males are less likely to stop exploring once exploring is initiated, whereas female mice stop exploring once they learn. Authors find that both learning rate (alpha) and noise parameter (beta) increase in exploration trials in a hidden Markov model (HMM). When reinforcement learning (RL) models were fitted to animal data, they report females had a higher learning rate and over days of testing, suggesting higher meta-learning in females. They also report that of the RL models they fit, the model incorporating a choice kernel updating rule was found to fit both male and female learning. The results do suggest one should pay greater attention to the influence of sex in learning and exploration. Another important takeaway from this study is that similar levels of accuracy do not imply similar strategies. Essential revisions include a request to show more primary behavioral data, to provide a rationale for the different RL models and their parameters, to clarify the difference between learning and 'steady state,' and to qualify how these experiments uniquely identify latent cognitive variables not previously explored with similar methods.

We appreciate the reviewer’s thorough reading of the paper and hope that the changes we detail below will address these concerns.

Reviewer #2:

The authors investigated sex differences in explore-exploit tradeoff using a drifting binary bandit task in rodents. The authors tried to claim that males and females use different means to achieve similar levels of accuracy in making explore-exploit decisions. In particular, they argue that females explore less but learn more quickly during exploration. The topic is very interesting, but I am not yet convinced on the conclusions.

Here are my major points:

1) This paper showed that males explore more than females, and through computational modeling, they showed that females have a higher learning rate compared to males. The fact that males explore more and have lower learning rates compare to females, can be an interesting finding as the paper tried to claim, but it can also be that female rats simply learn the task better than male rats in the task used.

We have revised the manuscript to better demonstrate that male mice did not acquire fewer rewards than females, and included all analyses and plots requested in this review. Ultimately, there was no evidence that they learned the task any less well than the females did. We appreciated this comment because it has strengthened the evidence we were able to present that males and females take different paths to the same outcome. Completing these analyses has also allowed us to clarify the relationship between RL learning rates and performance in this classic dynamic decision-making task.

(a) First, from Figure 1B, it looks like p(reward, chance) are similar between sex, but visually the female rats' performances, p(reward, obtained), look slight better than males. It would be nice if the authors could show a bar plot comparison like in Figure 1C and 1E. A non-significant test here only fails to show sex differences in performance, but it cannot be concluded that there are no sex differences in performance here. Further evidence needs to be reported here to help readers see whether there are qualitative differences in performances at all.

The requested bar plot has been added in as Figure 1C and illustrates our central point: male mice did not acquire fewer rewards than females, so there is no evidence that they learned the task any less well than the females did. The t-test result we originally reported suggests that we can discard the hypothesis that males and females have different mean levels of percent reward obtained, but we take the reviewer’s point that the male and female distributions may differ in other, more subtle ways. Therefore, we conducted a better statistical test here. The Kolmogorov-Smirnov (KS) test takes into account not only the means of the distributions but also the shapes of the distributions. The null hypothesis is that both groups were sampled from populations with identical distributions. It tests for any violation of that null hypothesis -- different medians, different variances, or different distributions. The KS test suggested that males and females are not just not significantly different in their reward acquisition performance (Kolmogorov-Smirnov D = 0.1875, p = 0.94), but that males and females have the same distribution of performance.

New text from the manuscript (page 5, line 119-128):

“There was no significant sex difference in the probability of rewards acquired above chance (Figure 1C, main effect of sex, F(1, 30) = 0.05, p = 0.83). While the mean of percent reward obtained did not differ across sexes, we consider the possibility that the distribution of reward acquisition in males and females might be different. We conducted the Kolmogorov-Smirnov (KS) test, which takes into account not only the means of the distributions but also the shapes of the distributions. The KS test suggested that males and females are not just not significantly different in their reward acquisition performance (Kolmogorov-Smirnov D = 0.1875, p = 0.94), but that males and females have the same distributions for reward acquisition. This result demonstrates equivalently strong understanding and performance of the task in both males and females.”

(b) The exploration and exploitation states are defined by fitting a hidden Markov model. In the exploration phase, the agent chooses left and right randomly. From Figure 1E and 1F, it looks like for male rats, they choose completely randomly 70% of the times (around 50% for females). The exploration state here is confounded with the state of pure guessing (poor performance).

This comment seems to confuse our descriptive HMM with a generative model. The HMM does not imply that choices are being made randomly. Instead, exploratory choices are modeled as a uniform distribution over choices. This was done only because this is the maximum entropy distribution for a categorical variable -- the distribution that makes the fewest assumptions about the true underlying distribution and thus does not bias the model towards or away from any particular pattern of choices during exploration. For example, (Ebitz et al., 2019) have shown that the HMM can recover periods of exploration that are highly structured and information- maximizing, despite being modeled in exactly this way.

Because the model does not imply or require that exploratory choices are random, we could, in the future, ask whether these choices reflect random exploration or instead more directed forms of exploration. However, for various reasons, this task is not the ideal testbed for isolating random and directed exploration, though this is a direction we hope to go in the future. To clarify our model and address these issues for future research, we have added the following text (page 31, line 745-756):

“The emissions model for the explore state was uniform across the options. The emissions model for the explore state was uniform across the options:

This is simply the maximum entropy distribution for a categorical variable - the distribution that makes the fewest number of assumptions about the true distribution and thus does not bias the model towards or away from any particular type of high-entropy choice period. This doesn’t require, imply, impose, or exclude that decision-making happening under exploration is random. Ebitz et al. 2019 have shown that exploration was highly structured and information-maximizing, despite being modeled as a uniform distribution over choices (Ebitz et al., 2020, 2019). Because exploitation involves repeated sampling of each option, exploit states only permitted choice emissions that matched one option.”

(c) Figure 2 basically says that you can choose randomly for two reasons, to be more "noisy" in your decisions (have a higher temperature term), or to ignore the values more (by having a learning rate of 0, you are just guessing). It would be nice to show a simulation of p(reward, obtained) by learning rate x inverse temperature (like in Figure 2C). From Figure 2B, it looks like higher learning rates means better value learning in this task. It seems to me that it's more likely the male rats simply learn the task more poorly and behave more randomly which show up as more exploration in the HMM model.

This is an important comment and addressing it gave us a chance to show the complicated, nonlinear relationship between learning rate term and performance in this task. Per the reviewer’s request, we now include a plot showing how learning rate (ɑ) and inverse temperature (β)affect reward acquisition (Figure 3F). However, this figure demonstrates that higher learning rate does not mean better performance in this task. Performing well in this task requires both the ability to learn new information and the ability to hang onto the information that has already been learned. That can only happen when learning rates are moderate, not maximal. When the learning rate is maximal, behavior is reduced to a win-stay lose-shift policy, where only the outcome of the previous trial is taken into account for choice. This actually results in a lower percent of the reward obtained. We have addressed the difference between the learning rate parameter in the reinforcement learning (RL) model and actual learning performance in the comment above. We believe that this new figure illustrates an essential point that different strategies could result in the same learning performance.

This result shows that the male strategy was a valid one that doesn’t perform worse than the female strategy. Not only did they have identical performance (Figure 1C), but their optimized RL parameters put them both within the same predicted performance gradient in this new plot (Figure 3F). That’s exactly why we believe that it is important to understand differences in how individuals approach the same task, even as they may achieve the same overall levels of performance.

New text from the manuscript (page 14, line 368-385):

“While females had significantly higher learning rate (α) than males, they did not obtain more rewards than males. This is because the learning rate parameter in an RL model does not equate to the learning performance, which is better measured by the number of rewards obtained. The learning rate parameter reflects the rate of value updating from past outcomes. Performing well in this task requires both the ability to learn new information and the ability to hang onto the previously learned information. That occurs when the learning rate is moderate but not maximal. When the learning rate is maximal (α = 1), only the outcome of the immediate past trial is taken into account for the current choice. This essentially reduces the strategy to a win-stay lose-shift strategy, where choice is fully dependent on the previous outcome. A higher learning rate in a RL model does not translate to better reward acquisition performance. To illustrate that different combinations of learning rate and decision noise can result in the same reward acquisition performance. We conducted computer simulations of 10,000 RL agents defined by different combinations of learning rate (α) and inverse temperature (β) and plotted their reward acquisition performance for the restless bandit task (Figure 3F). This figure demonstrates that 1) different learning rate and inverse temperature combinations can result in similar performance, 2) the optimal reward acquisition is achieved when learning rate is moderate. This result suggested that not only did males and females had identical performance, their optimized RL parameters put them both within the same predicted performance gradient in this plot.”

(d) From figure 3E, it looks like female rats learn better across days but male rats do not, but I am not sure. If you plot p(reward, obtained) vs times(days), do you see an improvement in female rats as opposed to males? Figure 4 also showed that females show more win-stay-lose-shift behavior and use past information more, both are indicators of better learning in this task.

Taken the above together, I am not convinced about the strategic sex differences in exploration, it looks more like that the female rats simply learn better in this task.

Unfortunately, there was no change in performance across days in either males or females. Per request by the reviewer, we now included a new plot illustrating p (reward,obtained) over days in Supplemental Figure 1. Ultimately, this resonated with the points we clarified above and demonstrated in this figure: males and females had identical performance in this task.

To the other points raised here, about sex differences in win-stay lose-shift and mutual information: these are the strategic differences at the heart of the paper, but again did not alter overall performance for the reasons detailed above. Figure 4 did show that females were doing more win-stay. However, after further examining win-stay behavior by explore-exploit states, we found that females were only doing more win stay during exploratory trials (Figure 5E). There was no difference in win-stay during the exploitative trials. Figure 5F also demonstrated that females did more win-stay lose- shift in the exploration state, indicating that females only learned better during exploration. Although males learned slower during exploration, they compensated that by exploring for longer. Both male and female strategies are equally effective and may be differentially advantageous in different tasks.

Finally, to address the meta-learning: in developing our response to this comment and looking for any other signs of adaptation across days (sex differenced or not), we did revisit this results and decided to rewrite some passages to be more circumscribed about our interpretations. Figure 3E showed increased learning rate parameters across days in females. We were initially excited about this idea of meta-learning, however we find no other evidence of adaptation over time in multiple behavioral measures, including reward acquisition, response time, and retrieval time (Supplemental Figure 1). Changes in learning rate parameters over sessions from the RL model were marginally significant and we feel that it’s worth mentioning for completeness, but it was only a small contributor to the overall sex differences in the behavioral profile. As a result we have toned down the conclusion we drew from this result accordingly.

New text from the manuscript (page 4, line 93-113):

“It is worth noting that unlike other versions of bandit tasks such as the reversal learning task, in the restless bandit task, animals were encouraged to continuously learn about the most rewarding choice(s). There is no asymptotic performance during the task because the reward probability of each choice constantly changes. The performance is best measured by the amount of obtained reward. Prior to data collection, both male and female mice had learned to perform this task in the touchscreen operant chamber. To examine whether mice had learned the task, we first calculated the average probability of reward acquisition across sessions in males and females (Supplemental Figure 1A). There was no significant changes in the reward acquisition performance across sessions in both sexes, demonstrating that both males and females have learned to perform the task and had reached an asymptotic level of performance across sessions (two-way repeated measure ANOVA, main effect of session, p = 0.71). Then we examine two other primary behavioral metrics across sessions that are associated with learning: response time and reward retrieval time (Supplemental Figure 1B, C). Response time was calculated as the time elapsed between the display onset and the time when the nose poke response was completed. Reward retrieval time was measured as the time elapsed between nose-poke response and magazine entry for reward collection. There was no significant change in response time (two-way repeated measure ANOVA, main effect of session, p = 0.39) and reward retrieval time (main effect of session, p = 0.71) across sessions in both sexes, which again demonstrated that both sexes have learned how to perform the task. Since both sexes have learned to perform the task prior to data collection, variabilities in task performance are results of how animals learned and adapted their choices in response to the changing reward contingencies.”

page 14, line 386-390:

“One interesting finding is that, when compared learning rate across sessions within sex, females, but not males, showed increased learning rate over experience with task (Figure 3G, repeated measures ANOVA, female: main effect of time, F (2.26,33.97) = 5.27, p = 0.008; male: main effect of time, F(2.5,37.52) = 0.23, p = 0.84). This points to potential sex differences in meta-learning that could contribute to the differential strategies across sexes.”

2) I do like how the authors define exploration states vs exploitation states via HMM using choices alone. It would be interesting to see how the sex differences in reaction time are modulated by exploration vs exploitation state. As the authors showed, RT in exploration state is longer. Hence, it would make a conceptual difference whether the sex difference in reaction times is due to different proportions of time spent on exploration vs exploitation across sex.

That is a very interesting idea. We tested for this possibility by calculating a two-way ANOVA (with interaction) between explore-exploit state and sex in predicting RT. There was a significant main effect of state (RT is longer in explore state than exploit state, main effect of state: F (1,30) = 13.07, p = 0.0011), but males were slower during females during both exploitation and exploration (main effect of sex, F(1,30) = 14.15, p = 0.0007) and there was no significant interaction (F (1,30) = 0.279, P = 0.60). Unfortunately, this means that we cannot interpret the response time difference between males and females as a consequence of the greater male tendency to explore. Response time is a fairly noisy primary behavior metric, especially in the males, and a lot of other factors might be at play here, some of which we plan to follow up on in the future. We report this result as follows (page 10, line 248-254):

“Since males had more exploratory trials, which took longer, we tested the possibility that the sex difference in response time was due to prolonged exploration in male by calculating a two- way ANOVA between explore-exploit state and sex in predicting response time. There was a significant main effect of state (main effect of state: F (1,30) = 13.07, p = 0.0011), but males were slower during females during both exploitation and exploration (main effect of sex, F(1,30) = 14.15, p = 0.0007) and there was no significant interaction (F (1,30) = 0.279, P = 0.60).”

Reviewer #3:

In the manuscript 'Sex differences in learning from exploration', Chen and colleagues investigated sex differences in decision making behavior during a two-armed spatial restless bandit task. Sex differences and exploration dysregulation has been observed in various neuropsychiatric disorders. Yet, it has been unclear whether sex differences in exploration and exploitation contributes to sex-linked vulnerabilities in neuropsychiatric disorders.

Chen and colleagues applied comprehensive modeling (model free Hidden Markov model (HMM), and various reinforcement learning (RL) models) and behavioral analysis (analysis of choice behavior using the latent variables extracted from HMM), to answer this question. They found that male mice explored more than female mice and were more likely to spend an extended period of their time exploring before committing to a favored choice. In contrast, female mice were more likely to show elevated learning during the exploratory period, making exploration more efficient and allowing them to start exploiting a favored choice earlier.

Overall, I find the question studied in this work interesting, and compelling. Also, the results were convincing and the analysis through. However, assumptions in the proposed HMM is not fully justified and additional analyses are needed to strengthen authors' claims. To be more specific, the effect of obtained reward on state transitions, and biased exploitations should be further explored.

Thank you for your feedback. We have included two more complex versions of the Hidden Markov models (HMMs) that account for the effect of obtained reward on state transitions and biased exploitations. Although the additional parameters slightly improve the model fit, model comparison tests suggested that such improvement was not significant. We decided to use the original HMM from the original manuscript because it’s the simplest and best fit model that provides the best parameter estimation with the amount of data we have. We do appreciate the comments and believe that the inclusion of two new HMMs and justification of the original HMM has strengthened our claims.

Author Response:

Reviewer #2 (Public Review):

1. Presentation, analysis, and discussion of calcium imaging results

a) As the authors correctly pointed out, having a water control is indeed essential for interpreting calcium imaging results. As such, I would recommend having a water control panel (currently in Figure S3) in the main figure.

Thank you for this suggestion. We now provide a new figure, which shows the change in fluorescence of lratd2a right dHb neurons first exposed to water (control) and then to cadaverine or alarm substance. (refer to Figure 2A-D).

b) The current presentation and analysis of calcium imaging data in Figure 2B does not seem appropriate and can be improved - since the dynamics of olfactory responses are likely highly variable across neurons and fish, rather than comparing responses across time, it would be better to compare the summed response over a longer time window (as already done in Figure S2, but also including water flow control data). Do also mention the time window over which the calcium responses were integrated.

As noted above, we have added a new figure to represent the change in fluorescence of lratd2a right dHb neurons averaged over a 5 min time window. This trace also includes the standard error of the mean (shaded) to represent the variability in responses among all neurons that were imaged (refer to Figure 2A- D).

c) Discussion Line 393: "From calcium imaging, we validated that the right dHb appears more responsive than the left when larval zebrafish are exposed to aversive odors such as cadaverine or chondroitin sulfate" - this conclusion cannot be drawn from the existing presented data, unless calcium imaging was also performed in the left habenula.

Thank you for pointing out this error. Indeed, we were unable to monitor calcium signaling in the left dHb owing to barely detectable levels of GCaMP labeling in the lratd2a cells on the left. We have now corrected this and added the following sentence to the Results:

“We monitored the responses of individual cells in the right dHb, as GCaMP6f labeling was weakly or not detected in neurons on the left (data not shown).”

d) It would be good to include in the methods section more detail on how the odor was delivered, volume delivered etc, and whether control experiments were done on the same day / clutch of fish etc.

We added the requested information in the “Calcium imaging in larval zebrafish” section of the Materials and Methods.

1. Presentation, analysis, and discussion of c-Fos results and comparison with calcium imaging

a) Figure 2C-D: The difference / overlap between blue and brown are difficult to make out in the images, especially at this resolution and magnification. Is there a way to specifically quantify the % of lratd2a neurons that are activated by c-fos, rather than just neurons in the dorsal habenula as a whole? This would be necessary to support the claim in line 278: "Thus, the lratd2a subpopulation in the right dHb responds to cadaverine in both larvae and adults".

We have shown magnified images of double-labeling with fos and lratd2a probes in Figure 2E to G’ to help with visualization of the overlap in colorimetric double in situ hybridization. We quantified the % of the lratd2a expression domain where fos is expressed and provided this information in the results (page 7, lines 128-129).

b) In larvae (using calcium imaging), the effects of cadaverine to chondroitin sulfate were compared, whereas in adults (using c-Fos), the comparison is between cadaverine and alarm substance. Is there a reason why the alarm substance was not used in larvae, or chondroitin on adult fish? Perhaps the authors can elaborate on their rationale.

We were basing our experimental approach on results that had been published by others. For example, Krishnan et al. (2014) had shown that dHb neurons respond to chondroitin sulfate in 6-9 day old larvae. Previous studies had reported that the earliest responses to alarm substance can be seen around 42 days post fertilization in zebrafish (Waldman, 1982) and 48-57 days post-hatching in fathead minnows (Carreau-Green et al., 2008). Jetti et al. (2014) examined the response to alarm substance at 25 dpf zebrafish.

1. Presentation, analysis, and discussion of behavioral results a) The presentation of the alarm substance behavior results could be improved. The authors could include the words "alarm substance" somewhere in the panels so it is clear to the readers that they are looking at responses to that rather than to cadaverine which is described in the preceding panels. Similarly, to avoid confusion and to facilitate comparison, the same parameters should be presented for Figures 3-5 (currently distance in top is not shown in Figure 3 or 5, onset of fast swim and interval time not shown in Figures 4-5).

As suggested, we added the words “Cadaverine” and “Alarm substance” to Figures 3, 4, 5 and Supplementary figure 4. We also now show the same parameters for the response to alarm substance in BoTx-GFP and intersectional BoTx-GFP transgenic lines, and in tcf7l2 and bsx mutants (refer to Figures 3, 4, 5 and Supplementary figure 4).

b) Does cadaverine induce changes in swim speed and other kinematic parameters? Similarly, does the alarm substance induce avoidance of one side of the tank like cadaverine? It can be difficult for the reader to compare the effects of genetic manipulations on responses to both odorants since different behavioral parameters are being quantified, hence some means of direct comparison could be helpful.

As we had described in the discussion (page 14, lines 279-280): “Despite both being aversive cues (Hussain et al., 2013; Mathuru et al., 2012), cadaverine and alarm substance elicit different behavioral responses by adult zebrafish.” Alarm substance triggers immediate (within 1 min) erratic behavior such as rapid swimming, darting and prolonged freezing. Thus, it is not feasible to measure the same behavioral responses to the two aversive cues.

c) The effect of cadaverine on control groups seems to be quite variable. In Figures 3 and 4 the avoidance effect persists the entire duration of the experiment. In Figures 5 and S4 the effect is only significant in 2 time bins. The authors' conclusions are still valid since the correct comparisons are indeed to their respective sibling controls, however it does make it a bit difficult to compare results across genotypes. For example, non-botox-expressing lratd2a:QF2 fish appear to have about the same degree of cadaverine avoidance as lratd2a:QF2, scl5a7a:Cre, QUAS:Botox fish. Similar to point (b), are there other parameters that can be measured that are more consistent in controls across genotypes? Or at the least, some discussion of the behavioral variability in the text.

As correctly pointed out by the reviewer, different fish with different genetic backgrounds demonstrate different degrees in their response to odorants. However, behavioral measurements were reproducible over 2 to 3 trials testing the same groups of fish on different days. We now show the aversion index for all individuals tested for the response to cadaverine in a new figure (refer to Supplementary figure 7).

d) tcf7l2 mutants (like bsx mutants) also have a significantly lower swim speed than controls, this is also worth mentioning / discussing in the text.

We have now mentioned this in the results section of the main text (Page 10, lines 202-203).

e) The link between habenular LR asymmetry and aversive behavior is indeed interesting - in the discussion, one proposal was that this asymmetry could promote directed turning and escape. From the existing data (particularly for the lratd2a:QF2, scl5a7a:Cre, QUAS:Botox fish), is there any evidence of differences in turning behavior (LR asymmetry, or probability of turns in general)?

We did not observe any correlation between habenular L-R asymmetry and the direction of turning in response to alarm substance, although this is an interesting point. We added a sentence to the discussion (page 16, lines 323-324) to reference a recent study on the neural basis of this lateralized behavior.

f) As a related point, it is not clear to me that one would expect an enhancement of cadaverine avoidance in bsx mutants, especially if the argument is that asymmetry is important for aversive behavior. Perhaps the discussion on this point could be framed less as a negative result but as a notable observation.

We agree with the reviewer’s interpretation and have added the following sentences to the Results (page 11, lines 218-220): “Despite the symmetric activation of dHb neurons, bsx homozygotes and heterozygotes both showed reduced responsiveness to cadaverine,” and to the Discussion (page 15, lines 303-304): “We did not observe enhanced or prolonged aversion to cadaverine in bsxm1376 homozygotes relative to controls.”

1. Statistical analyses: Unless data is normally-distributed, non-parametric tests should be used to compare on calcium and behavioral imaging data (such as Kruskal-Wallis for time course of the calcium / behavioral data, Wilcoxon Rank-Sum Test for others).

As correctly pointed out by the reviewer, we have corrected our statistical analyses (refer to Materials and Methods and the Figure legends). We used two-way ANOVA followed by Bonferroni's post hoc test and an unpaired t-test for analyzing calcium imaging data. For analyzing the response to cadaverine within groups, we used the Wilcoxon signed-rank test and cited publications using comparable approaches (Koide et al., 2009; Wakisaka et al., 2017). For analyzing the data between groups, we used two-way ANOVA followed by Bonferroni's post hoc test.

Author Response:

Reviewer #1:

In this manuscript titled "The LRRK2 G2019S mutation alters astrocyte-to-neuron communication via extracellular vesicles and induces neuron atrophy in a human iPSC-derived model of Parkinson's disease", Jacquet and colleagues investigated the role of Parkinsonism gene mutation LRRK2 G2019S in hiPSC-differentiated astrocytes. By isolating extracellular vesicles from ACM and examining astrocytes with various electron microscopy techniques, the authors found that LRRK2 G2019S affects the morphology and distribution of MVBs and the morphology of secreted EVs in hiPSC-differentiated astrocytes. Furthermore, the authors observed that astrocyte-derived EVs can be internalized by dopaminergic neurons and such EVs support neuronal survival. However, LRRK2 G2019S EVs lost the ability of promoting neuronal survival. This is an interesting study showing a non-cell autonomous contribution to dopaminergic neuron loss in PD.

The proposed idea of how LRRK2 G2019S dysregulates EV-mediated astrocyte-to-neuron communication is novel and exciting. However, the authors present some conflicting data that is not addressed during the discussion: they first conclude upregulated exosome biogenesis by RNAseq in G2019S vs WT astrocytes, but later show a decrease in the number of <120nm particles in G2019S mutants suggesting a decrease in the classical exosome-sized vesicle secreted compared to WT. Lastly, their MVB images show less CD63 gold particles in G2019S compared to WT control (though this was not quantified). Do the authors suggest and increase or decrease in exosome biogenesis in G2019S mutants? How do they reconcile these seemingly contradicting data? Several experiments, controls and additional analyses are needed to fully demonstrate the validity of the proposed mechanism.

The RNA-sequencing data of LRRK2 G2019S astrocytes showed an enrichment in genes associated with the “extracellular exosome” gene ontology term but not with the MVB/EV trafficking or secretion pathways. While we found CD82 and Rab27b to be upregulated, the classical biogenesis markers of MVB/EV trafficking and secretion (e.g. VTA1, VPS4, ALIX) were not dysregulated. Instead, the gene list shows an overwhelming dysregulation of genes coding for EV-enclosed proteins which do not have known roles in MVB/EV biogenesis or function (we now discuss this point in the main text, see highlights in italics below). As a result, we do not believe that exosome biogenesis is upregulated but instead propose the working hypothesis that the EV pathway may contribute to LRRK2 G2019S astrocyte dysfunction. To complement the sequencing data, our study provides a characterization of this pathway by (i) describing the cellular distribution of CD63+ structures in astrocytes, (ii) measuring the size of secreted EVs, and (iii) analyzing the neurotrophic potential of control and LRRK2 G2019S astrocyte-secreted EVs. We have not characterized the cellular biology of exosome/EV biogenesis in depth, and we do not propose a mechanism by which the LRRK2 G2019S mutation dysregulates these pathways. These questions are beyond the scope of our study, which is focused on the role of astrocytes in neurodegeneration.

The reviewer also referred to the CD63 immunogold staining used in Figures 4C and 6A to localize MVBs. After careful quantification of the number of CD63 gold particles in WT and LRRK2 G2019S MVBs, we conclude that there are no differences between the two genotypes and we apologize for selecting non-representative images. We have now replaced these with representative images. Regarding the shift in the size of WT vs. LRRK2 G2019S vesicles, we complemented our cryo-EM analysis with new data generated using Nanoparticle Tracking Analysis (NTA) (Figure 3C,D). The NTA analysis enabled the quantification of a greater number of particles, and we found that both WT and LRRK2 G2019S astrocytes secrete a significant number of particles in the 0-120 nm range. The cryo-EM data suggested that mutant astrocytes secreted fewer particles in this size range, but this is not observed in the NTA analysis. This discrepancy could be explained by the following: (i) in contrast to cryo-EM, NTA does not distinguish EVs from cell debris, which could bias the quantification and increase the number of small particles quantified (Noble et al., 2020), and (ii) studies showed that the size distributions between NTA and cryo-EM differ, the latter enabling the identification of larger particles (Noble et al., 2020). These two techniques are therefore complementary in the study of secreted EV and our manuscript now presents data generated using these two approaches (Figure 3C-G) (see italicised text below).

Results

Expression of exosome components in iPSC-derived astrocytes is altered by the LRRK2 G2019S mutation Gene ontology (GO) analysis revealed that components of the extracellular compartment are up-regulated in LRRK2 G2019S astrocytes – these include GO terms corresponding to the extracellular region, extracellular matrix and extracellular exosomes (Figure 1D,F). The exosome component is one of the most significantly up-regulated GO terms in both isogenic and non-isogenic astrocytes, and is comprised of a total of 67 (isogenic pair) or 95 (non isogenic pair) genes (Supplementary Tables 1 and 2). The large majority (~ 98 %) of these gene products are described to be enclosed in exosomes (e.g. CBR1) but do not perform specific functions related to EV formation or secretion. Only a few genes are associated with exosome biogenesis (e.g. CD82) and trafficking (e.g. Rab27b) (Andreu & Yanez-Mo, 2014; Chiasserini et al., 2014; Ostrowski et al., 2010) and we did not detect differences in the expression of canonical factors that regulate MVB formation (e.g. VTA1, VPS4 or ALIX).

Profiling WT and LRRK2 G2019S EVs secreted by iPSC-derived astrocytes

The astrocyte-derived EV pellet is enriched in exosomes, as demonstrated by the expression of 8 exosomal markers and the absence of cellular contamination (Supplementary Figure 3D). NTA quantification showed that the number of secreted EVs does not differ between LRRK2 G2019S and isogenic control (Figure 3C), and it appears that LRRK2 G2019S particles have a slightly different size distribution compared to WT particles (Figure 3D). It should be noted that TEM and NTA are methods traditionally used to estimate the size distribution of EVs, but their accuracy is often challenged by sample processing artifacts and technical biases (Pegtel & Gould, 2019). To overcome these limitations, we complemented the NTA results with cryo-EM analysis of the size of EVs secreted by WT and LRRK2 G2019S isogenic astrocytes. EVs mostly displayed a circular morphology (as opposed to the cup-shaped morphology observed by TEM) (Figure 3E), but a variety of other shapes were also observed (Supplementary Figure 3E). Cryo-EM analysis confirmed that WT astrocyte-secreted EVs display a large range of sizes, from 80 nm to greater than 600 nm in diameter, with differences between WT and mutant populations (Figure 3F). The cryo-EM data suggested that mutant astrocytes secreted fewer particles in the 0-120 nm size range, and the discrepancy with the NTA results could be explained by the following: (i) in contrast to cryo-EM, NTA does not discriminate EVs from cell debris, which could bias the quantification and increase the number of small particles quantified (Noble et al., 2020), and (ii) studies showed that the size distributions between NTA and cryo EM differ, the latter enabling the identification of larger particles (Noble et al., 2020). However, cryo-EM is a low throughput methodology that limits data collection to a small sample size and has therefore a lower statistical power than NTA. Quantification of the number of simple vs. multiple EV structures did not reveal differences between the two lines, and represent up to 16% of the EV population (Figure 3G). We then sought to complement our EV profiling experiments with an analysis of secreted CD63+ particles, which form one of the known exosomal sub-populations. We previously showed that WT and LRRK2 G2019S MVBs contain similar levels of the CD63 tetraspanin (Figure 2E, Supplementary Figure 3A,B), and an ELISAbased quantification confirmed that the number of CD63+EVs remained unchanged between the two genotypes (Supplementary Figure 3F). We conclude from these results that the total number and morphology of EVs produced by WT and LRRK2 G2019S astrocytes are similar, but mutant EVs may have a different size distribution compared to WT vesicles.

Major concerns:

1) In figure 1 A authors demonstrate iPSC-derived astrocytes characterization. Since there is no one unified and validated method for astrocytes differentiation, there is a need for more accurate characterization of iPSC-derived astrocytes. Authors should demonstrate the percentage of cells positive to astrocytic markers and to prove that obtained astrocytes are functional (able to promote synaptogenesis and uptake glutamate). I would also recommend analyzing the iPSC-derived astrocyte cultures for expression of more specific astrocytic markers as GLT1, SOX9 in addition to those which have been analyzed. Moreover, it is highly important to know what is the proportion of astrocytes derived from LRRK2 G2019S line and its isogenic control in order to be able to compare their effect on neurons.

We thank the reviewer for these suggestions. It is true that there exist many different astrocyte differentiation protocols, and this study uses a protocol developed by TCW et al. that has been further optimized by our lab to derive astrocytes from a midbrain-patterned population of neural progenitor cells (NPCs) (de Rus Jacquet, 2019; Tcw et al., 2017). The protocol is published, and shows that these astrocytes are functional – they respond to inflammatory factors and alter secretion of the IL-6 cytokine. Furthermore, Supplementary Figure 2D shows a whole transcriptome analysis (by RNA-seq) of the cell populations produced for this study and demonstrates that iPSC-derived astrocytes cluster with human primary midbrain astrocytes and away from iPSCs or NPCs in an unsupervised cluster analysis. However, we agree that in-depth characterization of iPSC-derived astrocytes is essential, and the updated manuscript now shows that (i) the astrocyte differentiation protocol yields 100 % GFAP+ cells with both WT and mutant lines (Supplementary Figure 2B), (ii) expression of six astrocyte markers (GLT1, SOX9, APOE, BHLHE41, CD44, GLUD1) (Supplementary Figure 2Aii, B), as well as (iii) transient intracellular calcium signaling (Supplementary Figure 2E), and (iv) synaptosome uptake (Supplementary Figure 2F) in both WT and LRRK2 G2019S astrocytes. We also updated the text as follows (italicised):

Results section

Midbrain-patterned NPCs carrying the LRRK2 G2019S mutation or its isogenic control were differentiated into astrocytes as described previously (de Rus Jacquet, 2019; Tcw et al., 2017). As expected, astrocytes expressed the markers GFAP, vimentin, and CD44 as demonstrated by immunofluorescence (Figure 1A) and flow cytometry analyses (Supplementary Figure 2A). Differentiation was equally effective in WT and LRRK2 G2019S cells, with 100 % of the differentiated astrocytes expressing GFAP (Supplementary Figure 2Bi). To further demonstrate the successful differentiation of iPSCs into astrocytes, we analyzed gene expression using RNA-sequencing analysis (RNA-seq), including primary human midbrain astrocyte samples in the RNA-seq study to serve as a positive control for human astrocyte identity. iPSC-derived and human midbrain astrocytes expressed similar levels of genes markers of astrocyte identity, including SOX9 and GLUT1 (Supplementary Figure 2B). In addition, principal component and unsupervised cluster analyses separated undifferentiated iPSCs, iPSC-derived NPCs and iPSC-derived astrocytes into independent clusters, demonstrating that our differentiation strategy produces distinct cell types (Supplementary Figure 2C-D). Importantly, the transcriptome of iPSC-derived astrocytes showed more similarities to fetal human midbrain astrocytes than to NPCs or iPSCs, further validating their astrocyte identity (Supplementary Figure 2D). Lastly, control and LRRK2 G2019S astrocytes showed classic astrocytic functional phenotypes such as spontaneeous and transient calcium signaling and synaptosome uptake (Supplementary Figure 2E-F).

2) In Figure 1, the authors found a significant upregulation of exosome components in astrocytes, demonstrating an important role of LRRK2 G2019S in EV signaling pathway. In the discussion, the authors briefly mentioned 'sub-populations of CD63- EVs may be differentially secreted in mutant astrocytes'. Since the authors have obtained the RNA-seq data, it would be nice to dig deep into the data and comment on potential EV sub-populations which can be differentially secreted. This information can be very beneficial for follow-up studies in the PD and LRRK2 field. Furthermore, the authors should assess the expression of Rab27a and CD82 in WT and LRRK2 G2019S astrocytes by western blots to verify RT-qPCR data. Furthermore, the authors should present specifically exosome biogenesis or secretion genes are altered to provide further insight into the stage of exosome biogenesis that is affected (ESCRT0-3, VPS4, ALIX, etc).

In the first comment, the reviewer refers to the observation that the number of total and CD63-positive EVs secreted by astrocytes is unchanged between the WT and LRRK2 G2019S genotypes. The classification of different EV sub-populations based on marker proteins is an evolving field of research, and an important study by Kowal et al. defined generic and sub population-specific EV markers (Kowal et al., 2016). Our RNA-seq dataset revealed five upregulated genes identified in the Kowal study, namely actin, GAPDH, actinin, complement and fibronectin, but unfortunately there is no clear pattern correlated with specific EV sub populations. For example, actin and GAPDH are two upregulated proteins that can be found in multiple types of EVs, actinin is enriched in large and medium-sized EVs, and complement and fibronectin are enriched in high density but small EVs (Kowal et al., 2016). The majority of dysregulated genes identified in our sequencing experiment are not proteins classically used to categorize EVs, so unfortunately our data does not allow us to address the reviewer’s question. To make sure that the data is readily accessible to the scientific community, we have prepared a supplementary table with a list of extracellular exosome-related genes identified in the RNA sequencing study. To respond to the reviewer’s comment on a specific stage of EV biogenesis/secretion altered in LRRK2 G2019S, the sequencing data presented in this manuscript does not allow to conclude that there is such a dysregulation. Our gene list corresponding to the “extracellular exosome” gene ontology term contains a large majority of genes coding for proteins enclosed within EVs that do not play a role in biogenesis/secretion. For example, the gene list does not contain ESCRT0-3, VPS4, ALIX or other classical markers involved in EV biogenesis and we cannot conclude anything about the alteration of MVB/EV biogenesis or defects in specific stages of MVB trafficking or EV secretion. In addition, we thank the reviewer for suggesting the validation of RT-qPCR data by western blot. The purpose of the RT-qPCR experiment was to validate the gene expression data collected by RNA-seq. Given that our objective was to confirm gene expression levels, and that we do not further study CD82 and Rab27b, we think that collecting protein expression levels is not necessary in the context of this study.

We agree with the reviewer’s suggestion, and we added images showing the subcellular localization of CD63 in both WT and LRRK2 G2019S MVBs by immunogold staining (Figure 2E). Validation experiments available in Figure 1A and Supplementary Figure 2A and 2B confirm that our astrocytes express CD44 as well as markers of mature astrocytes (BHLHE41, SOX9, GLUT1, APOE, GLUD1). The reason for showing CD44 instead of a more mature marker such as GFAP in Figure 2 of the manuscript is because CD44 is a membrane marker, and it therefore enables a clear visualization of the astrocyte surface area. We also note that, as shown in Supplementary Figure 2Aii, iPSC-derived and human fetal astrocytes express CD44, but iPSCs and NPCs do not significantly express this marker gene. In addition, as suggested by the reviewer, we added information related to MVB maturation in the introduction and the new text reads as follows (changes in italics):

Introduction section The sorting and loading of exosome cargo is an active and regulated process (Temoche-Diaz et al., 2019), and the regulatory factors involved in EV/exosome biogenesis are just beginning to be identified. Among the well-known factors, Rab proteins are essential mediators of MVB trafficking and they regulate endosomal MVB formation/maturation as well as microvesicle budding directly from the plasma membrane (Pegtel & Gould, 2019; T. Wang et al., 2014). In addition, membrane remodeling is an essential aspect of MVB/EV formation that appears to be regulated, at least in part, by the endosomal sorting complex required for transport (ESCRT) machinery (Pegtel & Gould, 2019; Schoneberg, Lee, Iwasa, & Hurley, 2017).

3) In Figure 2A and B, data shows that both WT and LRRK2 G2019S astrocytes produce MVBs and MVBs in LRRK2 G2019S astrocytes is smaller than in WT astrocytes. In Figure 2E, the authors showed the abundance of CD63 localized within MVBs in WT astrocytes but did not show the CD63 localization in MVBs in G2019S astrocytes. However, it is important to show CD63 localization in MVBs in G2019S astrocytes to fully support the conclusion that CE63+ MVBs are present in LRRK2 G2019S astrocytes. In addition, CD44 is a marker for astrocyte-restricted precursor cells. Although CD44+ positive cells are committed to give rise to astrocytes, it is crucial to include another astrocyte marker to ensure these cells are indeed mature astrocytes. -Related, authors should consider citing some of the MVB maturation literature to guide the readers.

We agree with the reviewer’s suggestion, and we added images showing the subcellular localization of CD63 in both WT and LRRK2 G2019S MVBs by immunogold staining (Figure 2E). Validation experiments available in Figure 1A and Supplementary Figure 2A and 2B confirm that our astrocytes express CD44 as well as markers of mature astrocytes (BHLHE41, SOX9, GLUT1, APOE, GLUD1). The reason for showing CD44 instead of a more mature marker such as GFAP in Figure 2 of the manuscript is because CD44 is a membrane marker, and it therefore enables a clear visualization of the astrocyte surface area. We also note that, as shown in Supplementary Figure 2Aii, iPSC-derived and human fetal astrocytes express CD44, but iPSCs and NPCs do not significantly express this marker gene. In addition, as suggested by the reviewer, we added information related to MVB maturation in the introduction and the new text reads as follows (changes in italics):

Introduction section

The sorting and loading of exosome cargo is an active and regulated process (Temoche-Diaz et al., 2019), and the regulatory factors involved in EV/exosome biogenesis are just beginning to be identified. Among the well-known factors, Rab proteins are essential mediators of MVB trafficking and they regulate endosomal MVB formation/maturation as well as microvesicle budding directly from the plasma membrane (Pegtel & Gould, 2019; T. Wang et al., 2014). In addition, membrane remodeling is an essential aspect of MVB/EV formation that appears to be regulated, at least in part, by the endosomal sorting complex required for transport (ESCRT) machinery (Pegtel & Gould, 2019; Schoneberg, Lee, Iwasa, & Hurley, 2017).

4) In Figure 3, it is impressive that the authors are able to image EVs using cyro-EM approach and analyze their sizes. The authors also observed different shapes of EVs. Is there any shape difference between WT EVs and G2019S EVs? Is there a way that the authors could categorize these shapes and do a detailed analysis in EV shapes? Also, In Figure 3D, both WT EV and G2019S EV images should present side by side for comparison. -Related, the size frequencies of EVs presented suggest a difference in the types of EV's released. Interestingly, exosomes are classically known to range from ~50-120nm and this population is significantly decreased in G2019S compared to WT. What does this suggest?

As suggested by the reviewer, we classified the two main EV shapes as “simple” and “multiple” EVs, and found no quantitative differences between WT and LRRK2 G2019S. This new data and side-by-side images of WT and LRRK2 G2019S EV images are available in Figure 3E-G, and the text has been updated accordingly (see text in italics below). One of the observations of Figure 3 is that there exist genotype-specific differences in the size distribution of EVs, which suggests that different classes of vesicles may be preferably produced by WT vs. LRRK2 G2019S astrocytes. This could be the result of differences in dynamics related to cargo loading, or a shift from MVB-released exosomes to membrane budding and microvesicle production. These observations are of great interest and we added a short discussion (in italics below) but they are beyond the scope of this study focused on EV neurotrophic properties, and we do not currently have evidence to support these hypotheses.

Results - LRRK2 G2019S affects the size of EVs secreted by iPSC-derived astrocytes

EVs mostly displayed a circular morphology (as opposed to the cup-shaped morphology observed by TEM) (Figure 3E), but a variety of other shapes were also observed (Supplementary Figure 3C). (…) Quantification of the number of simple vs. multiple EV structures did not differ between the two lines, and represent up to 16 % of the EV population (Figure 3G).

Discussion – Dysregulation of iPSC-derived astrocyte-mediated EV biogenesis in Parkinson’s disease

The observation that LRRK2 G2019S MVBs are less frequently located in the perinuclear area suggests that they may spend less time loading cargo at the Trans-Golgi network, which could in turn produce smaller MVBs and EVs with a different size range compared to WT (Edgar, Eden, & Futter, 2014; Pegtel & Gould, 2019). We did not observe a difference in the number of secreted EVs (total and CD63+ subpopulation) between WT and LRRK2 G2019S astrocytes (Figure 3C,H), suggesting that the secretion of at least one population of EVs is independent of the astrocyte genotype.

5) In figure 3c, SBI ELISA claims to quantify CD63+ vesicles, the authors should present more standardized particle quantification data (either by CD63 FACs for isolated EVs in WT vs G2019S or ZetaView/QNano particle tracking). The authors should also directly quantify the total number of EVs secreted in WT vs G2019S conditions (not only CD63+).

The updated manuscript now contains the NTA analysis of WT and LRRK2 G2019S EVs (Figure 3C,D) which provides the total number of EVs secreted by WT and LRRK2 G2019S astrocytes.

6) In Figure 4, the authors quantify LRRK2+/CD63+ particles by imaging. Importantly, it appears that there are less CD63 "large gold" particles in MVB of G2019S compared to control. This CD63 baseline quantification in MVB of WT vs. G2019S should be presented in this figure. These data are not convincing and should be quantified by FACS in secreted EV. Supplementary figure 3 should be brought into this figure.

As suggested by the reviewer, we quantified the number of CD63 large gold particles per MVB in WT and LRRK2 G2019S lines (Supplementary Figure 3A,B), and we re-introduced Supplementary Figure 3 into the main text (Figure 4E). We also updated the text (see in italics below). Additionally, we present extensive quantification of LRRK2 levels in MVBs and secreted EVs via imaging and biochemical analysis (ELISA), two different but complementary analytical methods.

Results - LRRK2 G2019S affects the size of MVBs in iPSC-derived astrocytes

Tetraspanins are transmembrane proteins, and the tetraspanin CD63 is enriched in exosomes and widely used as an exosomal marker (Escola et al., 1998; Men et al., 2019). However, cell type specificities in the expression of exosomal markers such as CD63 have been documented (Jorgensen et al., 2013; Yoshioka et al., 2013). We therefore confirmed the presence of CD63- positive MVBs in iPSC-derived isogenic astrocytes by immunofluorescence (Figure 2D) and immunogold electron microscopy (IEM) (Figure 2E). Analysis of IEM images showed an abundance and similar levels of CD63 localized within MVBs in WT and LRRK2 G2019S astrocytes (Figure 2E, Supplementary Figure 3A,B), confirming that CD63 can be used as a marker of MVBs and exosomes in iPSC-derived astrocytes.

7) In Figure 5, using CD63 as a MVB marker is not the most accurate approach. ESCRT markers should be co-stained with these experiments to truly show MVB localization (CD63 can localize to MVBs but is known to have a wider distribution throughout the cell compared to TSG1010 or other ESCRT complex proteins). Additionally, the authors must show their Supplemental Figure 3 ELISA quantification of p-aSyn in this main figure, and comment on why they conclude higher p-aSyn content in MVBs based on their IEM but then find no differences in aSyn in secreted EVs in WT vs. G2019S by ELISA.

We thank the reviewer for the suggestion to use ESCRT proteins as MVB markers. We decided to use CD63 because it is recognized in the literature as an MVB and EV marker (Beatty, 2008; Edgar et al., 2014), and we now refer to these two studies in the manuscript to support this choice (see text in italics below). Using ESCRT complex proteins as MVB markers is an interesting alternative, but we note that proteins associated with this complex are also found to regulate other biological processes such as autophagy (Takahashi et al., 2018) and plasma membrane repair (Jimenez et al., 2014), and so they can co-localize to non-MVB structures (e.g. autophagosomes or plasma membrane). Similarly, TSG101 can also localize to non-MVB structures such as the nucleus and Golgi complex (Xie, Li, & Cohen, 1998), and also lipid droplet (LD) membranes where it promotes LD-mitochondria contact (J. Wang et al., 2021). As suggested by the reviewer, Supplemental Figure 3 has been re-introduced into the main text (Figure 6C). Regarding αSyn, the immunogold staining specifically detects the phosphorylated form of αSyn (p-αSyn), while the ELISA detects all forms of αSyn (total αSyn). We observed increased p-αSyn in LRRK2 G2019S MVBs, but similar levels of total αSyn in WT vs LRRK2 G2019S EVs. This observation suggests that the phosphorylated form of αSyn, but not the total amount of αSyn, is affected by the experimental conditions. The text has been updated and reads as follows (changes in italics).

Results - LRRK2 is associated with MVBs and EVs in iPSC-derived astrocytes

In light of our observations that mutations in LRRK2 result in altered astrocytic MVB and EV phenotypes, we asked if LRRK2 is directly associated with MVBs in astrocytes and if this association is altered by the LRRK2 G2019S mutation. We analyzed and quantified the co localization of LRRK2 with CD63 (Figure 4A), a marker for MVBs (Beatty, 2008; Edgar et al., 2014), and found that the proportion of LRRK2+ /CD63+ structures remains unchanged between WT and LRRK2 G2019S isogenic astrocytes (Figure 4B).

Results - The LRRK2 G2019S mutation increases the amount of phosphorylated alpha synuclein (Ser129) in MVBs

Since the MVB/EV secretion pathway is altered in our LRRK2 G2019S model of PD, we reasoned that mutant astrocytes might produce αSyn-enriched EVs by accumulating the protein in its native or phosphorylated form in MVBs or EVs. IEM analysis revealed an abundance of p-αSyn (small gold) inside and in the vicinity of MVBs of LRRK2 G2019S iPSC-derived astrocytes, but not isogenic control astrocytes (Figure 6A). We observed that 55 % of the CD63+ (large gold) MVBs in LRRK2 G2019S astrocytes are also p-αSyn+ (small gold), compared to only 16 % in WT MVBs. LRRK2 G2019S astrocytes contained on average 1.3 p-αSyn small gold particles per MVB compared to only 0.16 small gold particles in isogenic control astrocytes, and MVB populations containing more than 3 p-αSyn small gold particles were only observed in LRRK2 G2019S astrocytes (Figure 6B). When we analyzed the content of EVs by ELISA, we found that total αSyn levels (phosphorylated and non-phosphorylated) in EV-enriched fractions are similar between isogenic controls and LRRK2 G2019S (Figure 6C). These results suggest that astrocytes secrete αSyn-containing EVs, and the LRRK2 G2019S mutation appears to alter the ratio of p-αSyn/total αSyn in MVB-related astrocyte secretory pathways.

8) In figure 6, it is even more clear that there is a stark difference between the CD63 presence in/near MVBs between WT and G2019S conditions. Since the authors normalize several pieces of data to CD63 (MVB localization, LRRK2 co-localization, etc), it is critical to quantify the number of baseline CD63 gold particles in MVBs in WT vs G2019S.

After careful quantification of the number of CD63 gold particles in WT and LRRK2 G2019S MVBs (available in Supplementary Figure 3A,B), we conclude that there are no significant differences between the two genotypes, and the MVB images initially selected in Figure 6 are not representative. We therefore replaced Figure 6A with new images.

9) In Figure 7, the authors used the co-culture of astrocytes and neurons to assess astrocyte-derived EV uptake by dopaminergic neurons. Although 3D reconstitution of neurons and exosomes can be precise, the data may not be 100% clean. It would be better if the authors collect ACM containing EV fraction from WT astrocyte and G2019S astrocytes and then incubate dopaminergic neurons with ACM containing EV fraction. In this way, only dopaminergic neurons are in the culture and there will be no CD63-GFP expressed astrocytes to contaminate the CD63-GFP signal in neurons.

We understand the concerns raised by the reviewer, and we can ensure that state-of-the-art imaging technologies and image post-processing techniques have been used to prevent astrocytic CD63 signal from contaminating the neuronal signal. We performed confocal microscopy with a 63X oil objective lens (numerical aperture = 1.4), and the images were processed with a Gaussian Filter (0.18 μm filter width) to reduce background noise in the MAP2 channel, and deconvolved (10 iteration) to enhance confocal image resolution in the CD63 channel. Furthermore, CD63-positive structures were detected with background subtraction enabled.

10) In Figure 9, the authors must show their ACM control. They show untreated, EV-free, and EV-rich ACM, but do not show unmanipulated ACM control.

The results of dendrite length analysis for unmanipulated ACM was initially available in Figures 8E and 8F. For clarity, we prepared a new Figure 9 that shows treatment with unmanipulated ACM, EV-free ACM, and EV-enriched fractions.

Reviewer #2:

In this manuscript by de Rus Jacquet et al., authors present an interesting study to detect changes in extracellular vesicles in human PD patient derived iPSC-derived astrocytes carrying the LRRK2 G2019S mutation. Isogenic gene corrected iPSCs were used as controls in all experiments. Authors first performed RNA-Seq for global gene expression changes between G2019S and "WT" gene corrected astrocytes. GO analysis showed an upregulation of extracellular compartments (including exosome compartments) in LRRK2 astrocytes. Subsequent experiments focusing on extracellular vesicles (EVs) and multivesicular bodies (MVBs), showed specific differences of MVB area and the size of secreted EVs. Secreted EVs from G2019S astrocytes also contained more LRRK2 particles and G2019S EVs contained more phosphorylated aSyn particles. Co-culture of LRRK2 astrocytes with human dopamine neurons showed accumulation of CD63+ exosomes in neurites, compared to co-culture with WT astrocytes. Co-culture with LRRK2 astrocytes decreased viability of TH+ neurons and LRRK2 dendrites/neurites were also shorter. These co-culture findings were replicated using EV-enriched conditioned media. Finally, authors showed that the trophic effect of astrocytes on neurons was due both to soluble factors released into the media, and production and release of EVs. Overall, this is a well-written and systematically performed study. This reviewer has several comments as detailed below.

1) Based on their data, authors conclude that astrocyte-to-neuron signaling and trophic support mediated by EVs is disrupted in LRRK2 G2019S astrocytes. Have authors measured the differences in trophic factors released by LRRK2 astrocytes in EVs and in conditioned media?

This is an important question, and we have not measured the levels of various neurotrophic factors in the medium. We concluded that LRRK2 G2019S astrocytes failed to secrete neurotrophic factors based on the neuron viability data. Healthy neurons cultured with disease astrocytes displayed dendrite shortening equivalent to that of neurons cultured in basal medium lacking neurotrophic factors. Furthermore, the morphological alterations occurred over a long period of time (2 weeks) and did not recapitulate the rapid and high level of neuron death and neurite fragmentation typically observed as a result of exposure to neurotoxins (Liddelow et al., 2017). However, we performed a new analysis of our RNA-seq data and identified dysregulated trophic processes of interest in LRRK2 G2019S astrocytes.

2) Authors differentiate cells (astrocytes and neurons) from midbrain lineage NPCs. The data show convincing effects of the LRRK2 derived astrocytes on neurons, but one question is whether this is specific to dopaminergic cells. Would this genotype specific effect also be expected in other lineages, e.g. cortical neurons? Authors should discuss this point.

The reviewer is making an excellent point. We prepared mouse primary midbrain cultures, and co-cultured WT midbrain neurons with WT or LRRK2 G2019S astrocytes. We found that the survival of WT midbrain dopaminergic neurons was significantly affected by LRRK2 G2019S astrocytes, but the viability of non-dopaminergic midbrain neurons was not changed when co cultured with WT or disease astrocytes. A previous study by di Domenico et al. also showed that dopaminergic neurons are more sensitive to the effect of LRRK2 G2019S astrocytes compared to non-dopaminergic cell types (di Domenico et al., 2019).

3) Prior work has demonstrated reductions in neurite length in neurons derived from LRRK2 G2019S iPSCs (not specific to dopaminergic neurons in LRRK2 cells) (for example Reinhard et al 2013). It is curious that the LRRK2 G2019S mutation itself can cause such a phenotype in neurons mono-cultures, and as shown in the current study, that LRRK2 G2019S astrocytes also induce a similar effect on WT neurons in co-culture. Can authors expand on this point in the Discussion?

We thank the reviewer for this question, and we added a new point of discussion in our manuscript, which reads as follows (changes in italics):

Evidence from this study and previous reports indicates that the LRRK2 G2019S mutation affects neurons through a variety of mechanisms. Here, we show a non-cell autonomous effect on neuronal viability via impairment of essential astrocyte-to-neuron trophic signaling, but the LRRK2 G2019S mutation can also mediate cell-autonomous dopaminergic neurodegeneration (Reinhardt et al., 2013). These observations support the idea that the LRRK2 kinase may be involved in a large number of pathways essential to maintain cellular function, cell-cell communication and brain homeostasis, and disruption of LRRK2 in one cell type has cascading effects on other neighboring cell types. In conclusion, our study suggests a novel effect of the PD-related mutation LRRK2 G2019S in astrocytes, and in their ability to support dopaminergic neurons. This study supports a model of astrocyte-to-neuron signaling and trophic support mediated by EVs, and dysregulation of this pathway contributes to LRRK2 G2019S astrocyte mediated dopaminergic neuron atrophy.

4) Authors should provide data on % dopaminergic neurons generated in the cultures.

We agree that this is important information, and we updated the latest version of the manuscript with this information (see below in italics). We estimate that the neuron cultures consist of 50 to 70 % dopaminergic neurons, and they are depleted of non-neuronal cells as explained in Material and Methods.

Material and Methods - Preparation and culture of iPSC-derived NPCs, dopaminergic neurons and astrocytes

To isolate a pure neuronal population, the cells were harvested in Accumax medium, diluted to a density of 1 × 106 cells in 100 µl MACS buffer (HBSS, 1 % v/v sodium pyruvate, 1 % GlutaMAX, 100 U/ml penicillin/streptomycin, 1 % HEPES, 0.5 % bovine serum albumin) supplemented with CD133 antibody (5 % v/v, BD Biosciences, San Jose, CA, cat. # 566596), and the CD133+ NPCs were depleted by magnetic-activated cell sorting (MACS) using an LD depletion column (Miltenyi Biotech, San Diego, CA), as described previously (de Rus Jacquet, 2019). The final cultures are depleted of non-neuronal cells and contain approximately 70 % dopaminergic neurons, the remaining neurons consisting of uncharacterized non-dopaminergic populations.

5) p7. Authors refer to phosphorylated a-synuclein as accelerating PD pathogenesis, but the references cited do not show this. In fact, Gorbatyuk et al 2008, showed that overexpression of S129 with constitutive phosphorylation eliminated a-synuclein induced nigrostriatal degeneration. The Fujiwara et al 2002 reference showed the presence of phospho a-syunclein in Lewy bodies and neurites. Authors should revise their statement that phospho a-synuclein is associated with accelerated pathology.

The reviewer is correct. We meant to highlight that there is a correlation between phosphorylated αSyn levels and PD pathogenesis, not that phosphorylated αSyn causes an acceleration of PD pathogenesis. We rephrased the sentence as follows, and replaced the study by Gorbatyuk et al. with a study by Anderson et al. that shows presence of phosphorylated αSyn in Lewy bodies (new text in italics):

EVs isolated from the biofluids of PD patients exhibit accumulation of αSyn (Lamontagne Proulx et al., 2019; Shi et al., 2014; Zhao et al., 2018), a hallmark protein whose phosphorylation at the serine residue 129 (p-αSyn) is correlated with PD pathogenesis (Anderson et al., 2006; Fujiwara et al., 2002).

6) Please provide details on the number of iPSC lines used for these experiments.

Experiments in the first version of this manuscript were performed using a single LRRK2 G2019S iPSC line and its gene-corrected control. The manuscript now presents the results collected using a second, independent non-isogenic iPSC line, as well as mouse primary cultures.

7) Clarify whether the WT neurons used for co-culture were derived from the isogenic human neurons?

We confirm that the WT neurons used for co-culture experiments were derived from isogenic controls. We added subtitles to our figures to clarify when data show results from isogenic or non-isogenic iPSC-derived cells.

Author Response:

Reviewer #1 (Public Review):

The authors make juxtacellular recordings on awake mice, which should yield clear responses of actions potentials, and employ a number of manipulations to silence pathways. They also record from a "non"-whisker secondary thalamic region, LP, as a null hypothesis to establish if certain effects are related to "behavior" - read arousal or saliency". I have no major qualms.

In light of Petersen's paper (Cell Reports 2014) on cholinergic effects on spike rates in primary whisker somatosensory cortex, I can imagine that the authors considered measuring from cholinergic neurons in nucleus basalis during whisking. I'll assume that this is easier said than done. As such, the current manuscript passes my threshold for publication modulo issues raised below that are related to anatomy.

The cholinergic experiments are an interesting idea. However, inactivation of S1 did not change the relationship between POm and whisking, suggesting that cholinergic modulation of S1 and thereby corticothalamic output are not the key mechanism. It is conceivable that acetylcholine modulates POm directly, but the critical experiments would involve extensive manipulations of POm (a whole additional study). Nevertheless, we have added a reference to Eggermann & Petersen and discussed this issue further in the revision.

I provide a figure-by-figure critique:

(1) Recent work from Deschênes et al (Neuron 2016) points to a description of whisking in terms of Angle = Set-point_angle - Whisking-amplitude [1 + cosine(Phase - Phase_0)], where Phase is a rapidly varying, typically rhythmic function of time. Why not use this notation as opposed to yet another descriptive statistic and report the kinetics as the time averaged parameters , i.e., the most forward position, and ,Whisking-amplitude>, i.e., the half-amplitude of the average whisk?

We are not entirely sure what the reviewer means by “another descriptive statistic” as we do not introduce new approaches for analyzing whisking in this paper. (Perhaps the reviewer refers to “median angle”, which is an average of all the whisker positions on a single frame. We use this measurement because our videos contain the entire whisker field rather than just a single whisker as in our other studies, e.g. Hong et al 2018, Rodgers et al 2021). We based our parameterization of median angle on two publications: Hill et al (2011 Neuron) and Moore et al (2015 PLoS Biology). Moore et al describes whisking as a function of phase, amplitude, and midpoint:

where 𝜃(t) is the median whisker angle at time 𝑡 , 𝜙 is the phase as computed by the Hilbert transform of the filtered whisker angle, 𝜃^Amplitude is the difference between the most protracted and retracted whisker positions over a single cycle, and 𝜃^midpoint is the central angle of a single whisk cycle. As we understand the reviewer, we are using the formulation they describe. We are happy to consider alternate formulations if we are missing something.

A critical issue is to confirm where the recording were made. This the authors should supply at least a typical record of anatomy from their POm as well as VPM and LP recording. The beauty of the juxtacellular technique is that neurons can be labeling after the recording

We used the juxtacellular recording technique for its superior recording quality. We did not label individual cells after recording because we recorded multiple cells per animal over several days. The number of cells would complicate matching of filled cells to recorded physiological data, and biotin filling is not stable over multiple days (beyond 36 hours). Instead, as described in the original manuscript, we tracked the relative locations of all inserted pipettes and labeled the final track with DiI. Cells were roughly localized along the tracks using relative microdrive depths. Due to the morphological homogeneity of thalamic neurons, filling individual cells would not be more informative than labelling the recording site with DiI. New Figure 1 – figure supplement 1 includes representative histology images from our recordings in POm, VPM, LP, and M1.

(2) Did the authors make sure that the mystacial pad is not moving by imaging the pad as opposed to just the shaft of the whiskers? The top view in Figure 1A makes this hard to check.

To address this concern, we provide new data, in which both the cut and uncut sides of the face of mice were imaged. We measured the movement of the mystacial pads as motion energy – the mean absolute difference in pixel values across video frames. The motor nerve surgery almost completely abolished movement of the mystacial pad. A new figure panel (Figure 2B) demonstrates the movement of the normal and paralyzed mystacial pads.

Further, did the authors perform post-hoc anatomy to insure that both the ramus buccolabialis inferior and ramus buccolabialis superior muscles were cut? This is critical; it is also easy to leave the maxillolabialis (external retractor) innervated if the cut is too far rostral.

We did not attempt to cut muscles. We only cut the motor nerve. We did not examine the face post mortem, as it was obvious that both whisker and mystacial pad movement were absent (as in new Figure 2B).

(3/4) As relevant background, the text should note that whisker primary motor cortex maintains a copy of the envelope of the whisking, i.e., an ill-defined summation of set-point and amplitudes, even if the sensory input (Ahrens & Kleinfeld J Neurophysiol 2004) or motor output (Fee et al. J Neurophysiol 1997) in the periphery are cut.

The Results text now cites these papers as motivation for the experiments of Figure 3.

(6/7) Same comments in (1) in whisking parameters and anatomy.

As we discussed in (1), we are using the conventional parameterization of others. Histological examples are now included in Figure 1 – figure supplement 1.

Reviewer #3 (Public Review):

Previous studies in urethane-anesthetized rats (PMID 16605304) proposed that POm cells code whisker movements. This was observed using "artificial whisking" procedures (stimulating the motor nerve to produce a whisking-like movement). It has been clear for some time now that there are substantial (obvious) differences between this procedure and natural whisking. In addition, under urethane-anesthesia animals are in a sleep-like state that is very dissimilar to waking (although some work has tested the effect of network state on artificial whisking responses in both primary thalamus and cortex; see 25505118). In the present study, the authors measured activity in POm cells during whisking in awake (head-fixed) mice to determine if they code whisking movement. However, this seems to have already been done previously. For instance, Moore et al (2015; 26393890) found that coding of whisking in the ascending paralemniscal pathway, including POm, is "relatively poor" (as stated in the abstract), which is the same conclusion reached in the present study. The authors should clarify the main differences observed in whisking coding between their study and previous work.

The authors then focused on the idea that POm codes behavioral state. However, many studies have previously determined that state has a great impact on thalamocortical dynamics; thalamic cells are very sensitive to state including cells in primary whisker thalamic nuclei, such as VPM, and these effects can be produced by neuromodulators (see work by Castro-Alamancos' group, for example, 16306412). There is nothing special about VPM in this regard; other thalamic sensory nuclei are also sensitive to behavioral state and neuromodulators. Therefore, the observation that POm and LP cells are sensitive to state is unsurprising. It is also known that these thalamic state changes have a great impact on the state of the cortex (see 20053845), which seems very relevant to the main conclusion. The POm has to be doing something different than coding behavioral state since most thalamic nuclei do this. The study did not identify the role of POm, which certainly has to be different from LP (otherwise, why would these nuclei be differentiated?). POm is unlikely to be specialized for monitoring state since this is done by most of the thalamus -including VPM, which projects to the same cortical region. Thus, while it is interesting that most of the whisker-related activity in POm is state-dependent, the study does not clarify the role of POm.

We have added the references we did not already include to our text and improved our discussion.

Prior studies (such as Moore et al 2015 and Urbain et al 2015) have previously characterized the encoding of whisker motion in POm. Indeed, we note the consistency between our results and such studies in both the introduction and conclusion. Here we expand upon prior studies to directly test two prominent hypotheses about the role of the paralemniscal pathway: that it encodes sensory reafference, and that it inherits a motor efference copy from cortical and subcortical regions. We present the impact of several manipulations of the vibrissal system (facial paralysis, cortical silencing, and lesion of superior colliculus) on thalamic activity that, to our knowledge, have not been previously reported. Moreover, we leveraged a novel comparison of POm and LP to test whether movement‐correlations of POm reflected true motor modulation or rather state dependency. We have provided evidence that the coupling of POm activity to whisking reflects state rather than motor signals. We never suggested that POm is a unique monitor of behavioral state. We suggest instead that secondary thalamic nuclei may be state‐modulated and have specific impacts on response gain and plasticity in their respective cortical areas. While our work is consistent with previous studies, we believe these results are novel extensions of past work.

The main strength of the study is that it was performed in awake mice with behavioral state monitoring, which contributes to the current understanding of active whisking coding in the complex network of the vibrissa system.

In our opinion, the main strength of our study is its multiple manipulations to test the sources of modulation and the leveraging of a POm‐LP comparison. We have revised the text to reinforce these points.

Author Response:

Reviewer #3 (Public Review):

The manuscript of Price et al. used TURBO-ID of multiple P granule components to identify new factors required for their assembly and function. They identify over 75 shared proteins, including 2 related TUDOR-domain proteins that they analyze in further detail through mutation and localization studies. The two proteins are necessary for the localization of several P granule components to the nuclear periphery, and they show in an ectopic tissue that EGGD-1 is sufficient to localize GLH-1 to the nuclear periphery.

While the paper could go further to test physical interactions between EGGD-1 (and EGGD-2) with GLH-1 and the nuclear pore protein, this is not critical, although the authors should be cautious in their model that they have not proven that these associations are direct.

We agree with the reviewer. We revised the statement and added references that support the model.

It would be important to provide evidence that the STOP-IN cassette in eggd-2 is a true null. There may be downstream methionines that can be used. And the phenotype is weaker than eggd-1 so this could be an issue.

This is a good point. Using two CRISPR guide RNAs, we generated a second allele of eggd-2 by deleting its full open reading frame. This allele and the allele bearing 17-nt insertion exhibited similar phenotypes: 1). No noticeable change in PGL-1::TagRFP localization in the pachytene region, 2). PGL-1::TagRFP failed to concentrate in the germ lineage in the embryos. We updated the information in the Figure 3—figure supplement 1 and text (line 292-297).

All of the TURBO-ID strains have reduced viability. Since there is some concern that P granule composition could be affected in the tagged strains, showing the localization of other known components of P granules in these mutants (GLH-1, PGL-1) would be critical.

Thanks for this suggestion. Please see the responses above.

Author Response:

Reviewer #3 (Public Review):

This is a well-written paper describing the OpenNeuro data archive. OpenNeuro covers diverse mesoscale brain imaging data (fMRI, PET, etc) from many projects and including multiple experimental paradigms. The focus is on human imaging, but primate and rodent data are also represented. The project rests on a standardized and relatively mature Data format (BIDS) for imaging data. This is key to implementing FAIR principles and automated checking for compliance. OpenNeuro is already producing discoveries that could not have been made without meta-analysis across diverse data sets. OpenNeuro is also used to improve data analysis pipelines. By its nature, this kind of paper always reads a bit like an advertisement.

The paper makes a compelling case for domain-specific repositories supported by a modern cloud architecture and sound computer science. I appreciate the discussion of the challenges that have been overcome (e.g. versioning of data sets; privacy and consent) and others that are looming (e.g. long-term maintenance in the absence of obvious commercial drivers).

I would like to see a bit more comparison to other archives, including some mature projects in other fields (e.g. astronomy, IVOA; structural molecular biology; CCDC), as well and more nascent efforts in brain research (e.g. DANDI; BIL).

Rather than focus on direct comparisons to other archives, we have focused on the FAIR principles for data management. We also feel that, in an article highlighting OpenNeuro, it could be seen as bad form to compare critically to other databases, rather than letting the merits of OpenNeuro stand on their own.

Author Response:

Reviewer #1 (Public Review):

Lenz et al have shown that IP injection of atRA does not affect sEPSC amplitude, sIPSC amplitude and frequency in the denta gyrus of both ventral and dorsal hippocampus. Interestingly, they observed a strong promoting effect of atRA on sEPSC frequency in the denta gyrus of dorsal, but not ventral, hippocampus. Lastly, they did not observe an difference in I/O in vivo, but did observe enhanced in vivo LTP in denta gyrus of mice injected with atRA which is abolished in the synaptopodin KO mice. The effect of atRA on LTP is very interesting as on sEPSC frequency in dorsal denta gyrus.

1) I do not agree with the authors' claim that atRA does not have a major effect on excitatory synaptic transmission. It seems that the sEPSC frequency increase by ~100%. Even if the 4 outlier points are excluded, the rest of the data points still clearly indicate an increase of sEPSC frequency.

We agree with the reviewer, that this point warrants further discussion. We have highlighted out findings in the revised version of the manuscript.

The abstract (lines 30-32) of the revised manuscript now read: “No major changes in synaptic transmission were observed in the ventral hippocampus while a significant increase in both sEPSC frequencies and synapse numbers were evident in the dorsal hippocampus 6 hours after atRA administration."

Lines 392-395: “Nevertheless, the results of the present study demonstrate increased sEPSC frequencies and synapse numbers in the dorsal hippocampus of atRA-treated animals, thereby confirming that atRA targets excitatory synapses in the dorsal hippocampus.”

What is the possible explanation of increased sEPSC frequency by atRA in dorsal region? Increased excitability of presynaptic neurons? (use TTX to decipher this?) Increased spine density? It seems that the authors did dye fill already… Count spine density? AND/OR increased glutamate release probability? (PPR measurement?) Did the authors perform I/O measurement in slice?

It is imperative that the authors tackle this issue head on.

We thank the reviewer for these important comments. To further address this issue, additional experiments were performed and structural properties of asymmetric synapses were assessed in the molecular layer of the dorsal hippocampus using transmission electron microscopy (see new Figure 6). Indeed, these experiments revealed no significant difference in the morphological properties of individual asymmetric synapses, i.e., PSD length and presynaptic vesicle numbers. However, an increase in the number of PSDs per area was observed, which may reflect -at least in part- increased sEPSC frequencies in our experiments.

Lines 302-316: “Next, transmission electron microscopy was used to assess the structural properties of excitatory synapses in the outer two thirds of the molecular layer in the dorsal hippocampus which is the layer of the major excitatory input from the entorhinal cortex (Figure 6). Cross sections of asymmetric synapses, i.e., the numbers and length of postsynaptic densities (PSD) and presynaptic vesicle counts, were quantified in control and atRA-treated mice (Figure 6A). It is well-established that PSD length in synaptic cross sections correlates to synaptic strength [43]. In agreement with our electrophysiological recordings, which showed no significant difference in the sEPSC amplitude between the groups (c.f., Figure 1D), PSD lengths did not significantly change in the atRA-treated group (Figure 6B). However, a robust increase in the number of PSDs per area was detected, and presynaptic vesicle counts were not significantly different between the two groups (Figure 6B, C). These results indicate that the structural properties of synapses are not affected by atRA, and that increased synapse numbers may explain the increased sEPSC frequencies in the dorsal hippocampus of atRA-treated mice.”

Lines 373-379: “In the present study, however, we did not observe changes in excitatory synaptic strength in dentate granule cells in either the ventral or the dorsal hippocampus. Specifically, no changes in the sEPSC amplitudes were observed [12]. Consistent with these findings no major changes in the ultrastructural properties of excitatory synapses, i.e., PSD lengths and presynaptic vesicle counts of asymmetric synapses, were observed between the two groups. Interestingly, ultrastructural analysis revealed an increase in the number of asymmetric synapses in the dorsal hippocampus of atRA-treated animals.”

2) The author need to specify which part of the denta gyrus for their in vivo study, as they discovered difference between ventral and dorsal in sEPSC frequency in slice preparation.

Done! Thank you!

Lines 183-186 now read: “Then a tungsten recording electrode (TM33B01KT, World Precision Instruments) was lowered in 0.1 mm increments while monitoring the waveform of the field excitatory postsynaptic potential (fEPSP) in response to 500 µA test pulses until the granule cell layer in the dorsal part of the hippocampus was reached (1.7-2.2 mm below the surface).”

Lines 324-326 now read: “To test the effects of atRA on the ability of neurons to express synaptic plasticity, long-term potentiation (LTP) experiments on perforant path synapses to dentate granule cells were carried out in the dorsal hippocampus of anesthetized mice (Figure 7A).”

Author Response:

Reviewer #2 (Public Review):

Valentini et al. explore the contribution of inexperienced homing pigeons in a pair, while finding the most efficient route back home. My comments below mostly concern the need of broadening the scope of the introduction and discussion by discussing and citing literature beyond homing pigeons as at the moment the manuscript could be characterized as too specific for the readership.

We thank the reviewer for their suggestions which allowed us to expand the focus of our manuscript. Our answers to the reviewer’s comments are reported below together with modifications done on the revised manuscript.

The authors use and present transfer entropy methods which regard the transmission of information from one individual to the other and effect of this information on behaviour. I haven't used such methods myself, but I think the methodology is nicely explained and easy to follow as it's written here. However, I would still encourage the authors to avoid jargon and un-introduced terms while first presenting their methods and results in the introduction and results sections. I also think that the paragraph in the introduction (L92-104) that refers to transfer entropy (TE) has to be extended and also direct readers to reviews such as [1] that attempt to make TE accessible to a broad audience of non-physicists. Behavioural ecologists and primatologists that study leadership and influence in animals, using less data hungry methods than TE, will probably be interested in reading this manuscript. Because eLife is a journal that attracts a very broad audience I would suggest investing more on better introducing TE to biologist and anthropologists.

We thank the reviewer for their suggestions. In the revised version of our manuscript, we clarified the meaning of symbols and unintroduced terms and extended the introduction paragraph about transfer entropy to provide more information. We now discuss data requirements of information-theoretic approaches, point the reader towards recent literature reviews aimed at introducing these (and similar) approaches to the community of behavioural ecology, and better introduce the advantages of transfer entropy with respect to methods based on models of alignment, attraction, and repulsion.

“Leader–follower interactions of this sort can be accurately captured using information-theoretic measures that quantify causal relations in terms of predictive information (Butail, Mwaffo, and Porfiri 2016; Kim et al. 2018; Crosato et al. 2018; Ray et al. 2019; Valentini et al. 2020). This methodological approach, which generally requires large amounts of data (but see (Porfiri and Ruiz Marín 2020)), is gaining popularity among behavioural ecologists (Strandburg-Peshkin et al. 2018; Pilkiewicz et al. 2020) as tools for automatic monitoring and extraction of the necessary volumes of behavioural data become increasingly available (Egnor and Branson 2016). One of these measures, transfer entropy, quantifies information about the future behaviour of a focal individual that can be obtained exclusively from knowledge of the present behaviour of another subject (Schreiber 2000). Transfer entropy measures information transferred from the present of the sender to the future of the receiver (Lizier and Prokopenko 2010). It explicitly accounts for autocorrelations characteristic of individual birds’ trajectories (Mitchell et al. 2019) by discounting predictive information available from the sender’s present that is already included in the receiver’s past (see Figure 1). Furthermore, it does not require a model of how sender and receiver interact, and it is well suited to study social interactions both over space and time (Lizier, Prokopenko, and Zomaya 2008; Strandburg- Peshkin et al. 2018). This aspect of transfer entropy encompasses traditional methods to quantify collective movement that are based on modelling an individual’s behaviour as a combination of three motional tendencies (Couzin et al. 2002) – alignment of direction to nearby group members, attraction towards sufficiently distant members, and repulsion from sufficiently close members – that allow an individual to maintain proximity to the group. In this context, transfer entropy is advantageous as it can capture causal interactions due not only to alignment forces (Nagy et al. 2010) but also to attraction and repulsion forces that result in temporarily unaligned states (Pettit, Perna, et al. 2013).”

A thought I had while reviewing this work regards the theory of the wisdom of the crowd [2]. This indicates that when a group or a collective averages the different estimates of its members, they reach a more accurate collective estimate. Studies have also shown that animals can average their movement directions to resolve conflicts of interest [3,4]. The current manuscript also shows that pooling infomration leads to better movement decisions. Would it thus make sense for this manuscript to discuss how its findings may support the wisdom of the crowd theory?

We thank the reviewer for the suggestion. In the revised version of out manuscript, we included a new paragraph where we discuss a possible connection with the phenomenon of the wisdom of crowds as well as how our results might generalize to flocks of larger size.

“The ability of groups to outperform single individuals by pooling information across their members is an aspect of collective intelligence that has long intrigued researchers. One potential mechanism underlying this phenomenon, popularly known as the wisdom of crowds (Surowiecki 2005), is averaging many individuals’ estimates independent from each other. Averaging individual decisions is expected to provide a more accurate group estimate than any individuals’ guess. Previous studies have also shown that animals can average their movement decisions to reach a compromise (Biro et al. 2006; Strandburg-Peshkin et al. 2015). Although the mechanisms by which experienced and naïve individuals pool information during route development remain unknown, our study points to the importance of naïve group members within the information-pooling process. Moreover, the wisdom of crowds is known to require personal information to be independent among group members (Couzin 2018) otherwise group performance can degrade quickly for increasing group size (Kao and Couzin 2014). Experimental pairs could thus benefit from pooling information with naïve individuals that, at least at the beginning of each generation, likely provide a source of information independent from that of the experienced bird. The potentially deleterious effects of losing independence may provide another pressure to shift over time from innovative exploration to route6 preserving exploitation. It remains to be explored how our results generalize to larger flock sizes. Previous experiments without generational replacement showed that, even in larger flocks, birds flying ahead of the flock had a tendency to assume leadership positions (Nagy et al. 2010). However, the repeated introduction of naïve individuals into larger flocks might complicate the dichotomy between leaders and followers by inducing turnover dynamics between the front and the back of the flock.”

As briefly mentioned earlier, I think that the cited literature in this manuscript (especially in L58-138 and throughout the discussion) includes mostly studies on homing pigeons whereas relevant studies to the current manuscript have been performed on other species and by discussing and citing relevant studies on various species the manuscript would become more attractive to a broader audience and wouldn't read as homing-pigeon specific.

We thank the reviewer for pointing us towards additional literature related to our study. We included the suggestions from the reviewer as well as further references to a broader literature to expand the scope of our manuscript.