Author Response

Reviewer #1 (Public Review):

This study used intersectional genetic approaches to stimulate a specific brainstem region while recording swallow/laryngeal motor responses. These results, coupled with histology, demonstrate that the PiCo region of the IRt mediates swallow/laryngeal behaviors, and their coordination with breathing. The data were gathered using solid methods and difficult electrophysiological techniques. This study and its findings are interesting and relevant. The analysis (and/or the presentation of the analysis) is incomplete, as there are analyses that need to be added to the manuscript. The interpretation of the data is mostly valid, but there are claims that are too speculative and are not well-supported by the results. The introduction and discussion would benefit from more citations and a deeper exploration of how this study relates to other work - especially a thorough accounting of and comparison to other studies concerning putative swallow gates.

General/major concerns:

The field of respiratory control is far from unified regarding the role of PiCo in breathing or any other laryngeal behaviors. If anything, the current consensus does not support the triple-oscillator hypothesis (in which PiCo is one of 3 essential respiratory oscillators). The name "PiCo", short for "post-inspiratory complex", suggests a function that has not been well-supported by data - it is a putative post-inspiratory complex, at best. I suggest putting this area in context with other discussions i.e. IRt (such as in Toor et al., 2019) or Dhingra et al. 2020 showed broad activation of many brainstem sites at the post-I period (including pons, BotC, NTS)

The reviewer’s comment refers to our previous publication and not the present one. With all due respect to the reviewer, the submitted study investigates PiCo’s involvement in swallow and laryngeal activation and its coordination with breathing.

We did not feel that it is appropriate for us to critique the Dhingra paper in the present study. However, since this seems to be important to this reviewer, we would like to clarify: Because of filter characteristics, and the low temporal and spatial resolution of these field recordings, the approach used by Dhingra is inappropriate for providing insights into the presence or absence of PiCo. We therefore developed an alternative approach, which provides more detailed insights into population activity, the Neuropixel approach. This Neuropixel recording from PiCo (black trace) exemplifies how field recordings (yellow) fail to pick up post-I activity. We could provide many more examples, but as stated above, addressing the study by Dhingra is tangential to the present study.

We would also emphasize that the study by Dhingra was never designed to provide negative evidence, and Dhingra et al. never claimed that their study demonstrates the absence of PiCo. Unfortunately, the data by Dhingra were misinterpreted by Swen Hülsmann in his Journal of Physiology editorial which created considerable confusion, but also sensation in the field. Objectively, Toor et al reproduced the Anderson study in rats as we will elaborate below. Unfortunately, Toor et al added to the confusion, by renaming the PiCo area into IRt. The field of respiration would have also been confused if the first study reproducing the Smith et al. 1991 study in a different rodent species would have refused to call this area preBötC and instead would have called it e.g. ventrolateral reticular field.

Did you perform control experiments in which the opto stimulations were done on animals without the genetic channels (for example, WT or uncrossed ChAT-ires-cre, etc.), or in mice with the genetic channels that weren't crossed (uncrossed Ai32 mice)? If so, please include. If not, why?

Yes, we performed many control experiments. Aside of many recordings in which viral injections were targeted outside PiCo, we also performed optogenetic stimulations in mice lacking channelrhodopsin. We have now added the following statements and supplemental figure.

Optogenetic stimulation in mice lacking channelrhodopsin

Stimulation of PiCo, across all stimulation durations, in 3 Ai32+/+ mice and 4 ChATcre:Vglut2FlpO:ChR2 mice where the ChR2 did not transfect ChATcre:Vglut2FlpO, as confirmed by a post-hoc histological analysis, resulted in no response (Fig. S3).

How do you know that your opto activations simulate physiological activation? First, the intensive optical activation at the stim site does not occur in those neurons naturally.

This seems like a generic critique of the optogenetic approach. In none of the 10,000+ published optogenetic studies is it known to what extent optogenetic activation stimulates exactly the same neurons and the same degree of activity as during a natural behavior. What we know is that PiCo neurons are activated during postinspiration (Anderson et al. 2016) and that optogenetic activation stimulates these neurons and that this activation evokes the same muscles in the same temporal sequence as a water-evoked swallow. We assume that the reviewer’s comment does not intend to imply that “swallows” evoked by nonspecifically stimulating the SLN is more physiological than the optogenetically-evoked swallows of a specific neuron population? From the reviewer’s other comments, it is obvious that the reviewer has no problems with the results of the Toor study that used exclusively SLN stimulations, an approach which is known to be very non-specific.

Doing a natural (water) stim for comparison is good, but it cannot necessarily be directly compared to the opto stim. The water stim would activate many other brainstem regions in addition to PiCo.

Can the reviewer provide any hard evidence that “many other brainstem regions” are activated by water stimulation in comparison to optogenetic stimulation?

A caveat is that opto PiCo stim =/= water stim (in terms of underlying mechanisms) should be included. Second, in looking at the differences between water vs opto swallows in Table S2: it appears that the ChAT animals (S2A) have something weaker than a swallow with opto stim. For the Vglut2 and ChAT/Vglut2 (S2B&C), the opto swallows also aren't as "strong" as the water swallows (the X and EMG amplitudes are smaller). The interpretation/discussion attributes this to the lack of sensory input during opto stim, but does not mention the strong possibility that there is a difference in central mechanisms occurring. It also seems to be dismissed with the characterization of the swallow as "all-or-none" (see note on Fig 3 results).

With all due respect, we are somewhat surprised that the reviewer dismisses the entire paragraph in the discussion that specifically addresses the comparison between water-swallows and PiCo-stimulated swallows. We discussed the possibility that PiCo stimulated swallows may not activate the full pathway/mechanism as does the water swallow. We carefully compared and confirmed that PiCo-stimulated swallows have the same temporal motor sequence of the same muscles as those activated in water swallows. As already stated, it is surprising that the reviewer has no problem with accepting the validity of previously published methods like electrical non-specific stimulations of the cNTS or SLN, a frequently used and accepted model to produce and study swallow.

The writing needs extensive copy editing to improve clarity and precision, and to fix errors.

Thank you for this comment, we have revised and reviewed the writing.

Results/Fig 1: What proportion had no/other motor response (non-swallow, non-laryngeal) to the opto stim? I can extrapolate by subtraction, but it would be nice to see the "no/other response" on the plot.

With all due respect to this reviewer, but it is not possible to address this question. Specifically, it is not possible to know if a “No response” (meaning “no behavioral output” occurred in response to PiCo stimulation), would have resulted in a swallow or laryngeal activation. However, figure 2 contains responses other than swallows, i.e. “non swallows”, which includes both laryngeal activation as well as “no responses” meaning “no behavioral response” in response to PiCo stimulation. This was determined to assess how the respiratory rhythm is affected when a swallow is not produced by PiCo stimulation.

The explanation of genetics is too spread out and confusing. There needs to be more detail about all the genetic tools used, using the standard language for such tools, in one spot. Please also provide a clear explanation of what those tools accomplish. Include a figure if necessary.

We apologize for creating confusion. We added more explanations to the text.

Pick a conventional designator/abbreviation for the different strains, define them in the methods and in the first paragraph of the results section, and use those abbreviations throughout. I think that using ChAT as an abbreviation for your ChAT-ires-cre x Ai32 mice is confusing because it makes it sound like you're talking about the enzyme rather than the specific strain/neurons. Saying "ChAT stimulated swallows... swallows evoked by water or ChAT" makes it sound like the enzyme choline acetyltransferase itself is stimulating swallow. As is convention, pick a more precise abbreviation like ChAT-cre/Ai32 or ChAT:Ai32 or ChAT-ChR2 or ChAT/EYFP. This goes for the other strains as well.

Thank you for pointing this out. To avoid confusion the strains/neurons are now referred to as: ChATcre:Ai32, Vglut2cre:Ai32, and ChATcre:Vglut2FlpO:ChR2

For Fig S2C&D, why does it say mCherry? Isn't it tdTomato? Is it just an anti-ChAT antibody and then the tdTomato Ai65 is only labeling Vglut2? I don't see this in the methods section.

Thank you for pointing this out. We apologize for our mistake, and we have corrected the manuscript to say tdTomato.

I also don't see methods for all the staining in Fig S3. The photomicrograph says Vglut2-cre Ai6, but there's no mention of Ai6 anywhere else. Which mice are these? Did you cross Vglut2-cre with an Ai6 reporter mouse? How can you image an Ai6 mouse (which I assume expresses ZsGreen? and that you excited at 488?) and a 488 anti-goat in the same section (that's the only secondary listed in the methods that would work with your goat anti-ChAT)? Is there an error in listing the fluorophores in the methods? Please give more details on the microscopy including which filters were used for the triple staining.

We have decided to remove the CTb data from the manuscript.

Regarding the staining: I would expect the staining/maps in for the 2 different ChAT/Vglut2 intersectional strains to be similar (Fig 5A/B and S2C/D). The photomicrographs look very different to me, while the heat maps (this goes for all the heat maps in the paper) have barely distinguishable differences. In Fig 5, the staining looks much stronger than in Fig S2C. Why does it look like there are so many more transfected neurons in Fig 5A2 than there are red neurons in the corresponding panel Fig S2C2? And for Fig 5A4 and Fig S2C44? The plot and results text for Fig 5 says the avg number of neurons was 123+¬11. The plot for Fig S2D says 112+¬15, but the results text says 242+¬12 (not sure which is the correct number).

Thank you for your comments. Previously the heat maps had different scale bars if you compare Fig 5A/B and S2C/D (now figure S4C/D). We changed the heat maps keeping the same scale for all of them. Discussing the representative photomicrography, even figure Fig 5A/B and S4C/D represents the same cluster of cells (PiCo Chat/Vglut+). Figure S4D states 242 ± 12 neurons (also stated in the results section).

However, we want to point out that there are several technical differences between both, 1) figure 5A represents the transfection promoted by the virus injection, impacting the number of cells stained/transfected (133 ± 16 neurons), 2) figure S4C/D represents a intersectional mouse ChATcre: Vglut2FlpO: Ai65; (242 ± 12 neurons). In this case, we have more tdTomato positive cells because this genetic approach is able to detect most of the Chat and Vglut2 cells. The difference between figures is considered normal for anatomical studies, in some studies the same bregma can show different number of cells. Thus, the differences are due to the differences in the type of approaches (viral expressions vs. intersectional approach).

We have also added additional experiments to figure 5 (now N=7) which has been reflected in the text and figures.

The results text for Fig S2C also says the staining is "similar to the previous ChAT staining...", which I assume refers to S2A/B. The plot and results text for Fig S2B reports 403+¬39 neurons, while S2D is either 112 or 242 (not sure?). The plots have different Y scales, which should be changed to be the same. But why do the photomicrographs and the heat maps look so similar? I would expect far fewer neurons to be stained in the intersectional mice (Fig 5 and Fig S2C/D) than in the ChAT staining (Fig S2A/B). I am having trouble reconciling the different presentations/quantifications and making sense of the data in these histology figures.

We removed “similar to the previous ChAT staining” and we have reviewed the heat maps. Since the original submission, we performed more experiments and now added more animals to the analysis (now N=7), each heat map represents the correct number of neurons in PiCo, respectively to each experiment.

The Y scales has been adjust to better demonstrate the Chat staining vs. the intersectional mice triple conditioned.

How can you distinguish PiCo from non-PiCo in the histology, especially in the ChAT-only staining? It seems that you have arbitrarily defined the PiCo region, and only counted neurons within that very constrained area.

Even in ChAT-only staining, the N.ambiguus is very distinct from the cholinergic neurons located more medial to the N.ambiguus. This can be unambiguously be confirmed by combining ChAT with glutamatergic in situ staining as done in the Anderson et al. study, or unambiguously be demonstrated with the viral approach as done in the present study. Thus, we don’t see why it is arbitrary to define the distribution of PiCo neurons. What is arbitrary is the definition of the preBötC, yet the field of respiration seems to have no problem with this. We assume that the reviewer knows that Dbx1 neurons are spread along the entire ventral respiratory column and dorsal portion of the PreBötzinger Complex up to the level of the XII nucleus. Yet it is commonly accepted for authors to refer to the PreBötzinger Complex by counting dbx1 neurons within a constrained area of what is believed to be the PreBötzinger Complex, even though the borders are arbitrary. It is e.g. known that some of the ventrally located preBötC neurons are presumed rhythmogenic while the more dorsally located Dbx1 neurons are premotor. The transition from rhythmogenic to premotor is gradual. Similarly, NK1 staining, or SST staining is not restricted to the preBötC and it is arbitrary to define where preBötC begins and what to include. Indeed, our PNAS paper indicates that inspiratory bursts can be generated by optogenetically stimulating Dbx1 neurons along the entire VRC column – so it is not clear where the rhythmogenic portion of the preBötC begins rostrocaudally and dorsoventrally and where the rhythmogenic portion and preBötC itself ends. Thus, we want to re-iterate and emphasize, that for the present study, we developed a method using the cre/FlpO approach to unambiguously define the PiCo region. It is surprising that this reviewer does not acknowledge this technical advance that added significantly more specificity to the anatomical and physiological characterization of PiCo, than the Toor et al. study, and also the Anderson et al. study.

I can see stained neurons in the area immediately outside of PiCo, and I'd like to see lower-magnification images that show the staining distribution in a broader region surrounding PiCo as well, especially in the rest of the reticular formation.

We characterized the PiCo area based on the histological phenotype and in vitro and in vivo experiments performed by Anderson et al., 2016. PiCo is an area located close to the NAmb, presenting the same ChATcre phenotype. As stated above, the distribution and agglomeration of the NAmb is clearly very compact, and different then the observed ChATcre: Vglut2FlpO: Ai65 neurons located outside of NAmb. It is also important to emphasize, that like is the case for the preBötC, other transmitter phenotypes of neurons are also present in the PiCo region (i.e. GABA or Dbx1). However, the study performed by Anderson et al, 2016 paper, described only the functions of cholinergic neurons located in PiCo, and we always planned to publish a paper of the other neurons within PiCo – this area e.g. contains pacemaker neurons etc. But, I hope that the reviewer acknowledges that many investigators have studied the preBötC for the past 30 years. Hence, much more information has been accumulated on this region (which btw was at least as controversial at the beginning), and it will likely take at least another 30 years to fully identify and characterize PiCo.

Similarly, how can you be sure you're stereotaxically targeting PiCo precisely (600um in diameter?) with your opto fiber (200um in diameter). Wouldn't small variations in anatomy put the fiber outside the tiny PiCo area?

We assume the reviewer means “stereotactically”. And yes, the reviewer is correct, it is necessary to position the laser at a consistent anatomical location. Placement of the optical fibers outside of this area does not result in activation of PiCo. We have added an additional supplemental figure (Figure S6) to address this.

Please put N's and stats results in Table S1 for both swallow and laryngeal activity. I took what I assume to be the Ns (10, 11, and 4) and did some stats like the ones you presented for the laryngeal duration. The differences between vagus duration for 40 and 200 ms pulse durations are all significant for each strain, by my calculations. Also, I think there must be an error in the orange swallow plot in Fig 3A. The orange dots don't correspond to the table values. I plotted all the Table S1 values for each strain. Each line looks similar to the blue laryngeal activation plot in Fig 3A. The slopes of the Vglut2 were less than the other strains, and the slopes for the swallow behavior were less than the laryngeal behavior for all strains. Otherwise, they all look similar. Please double-check your values/stats to address these discrepancies. If it is indeed true that the stim pulse duration affects swallow duration, revise the interpretations and manuscript accordingly.

We thank the reviewer for the diligence in reviewing our manuscript. But, with all due respect, the reviewer is incorrect and misunderstood the data. To clarify: Table S1 is only presenting data for laryngeal activation, swallow data is presented in Table S2. The orange data points in Fig 3A are not detailed in Table S1 or S2. Table S2 is the average of all swallows across all laser pulse durations since the laser pulse duration does not affect swallow behavior duration. All data will be publically available after publication of the manuscript.

Figure 3A is only representing the ChATcre:Vglut2FlpO:ChR2 column of Table S1

The N’s have been added to table S1

Please add more details on stats in general, including the specific tests that were performed, F values and degrees of freedom, etc.

Thank you, this has been added throughout the results section. Please refer to the results section for this addition. However below we have provided an example.

An example: A two-way ANOVA revealed a significant interaction between time and behavior (p<0.0001, df= 4, F= 23.31) in ChATcre:Vglut2FlpO:ChR2 mice (N=7).

How do you know that you're not just activating motoneurons in the NA when you stimulate your ChAT animals, especially given the results in Fig 1B? In this case, the phase-specific results could be explained by inhibitory inputs (during inspiration) to motoneurons in the region of the opto stim.

As stated in this paper as well as the Anderson et al 2016 paper (and for that matter also the Toor et al study) this is a caveat. This major caveat motivated the development and use of the ChATcre:Vglut2FlpO:ChR2 (specifically targeting the PiCo neurons that co-express ChAT and Vglut2, not laryngeal motor neurons) experiments that have mostly the same response as the ChATcre:Ai32 mice. We cannot say this is due to inhibitory inputs to laryngeal motoneurons, since the cre/FlpO specific experiments are not directly activating laryngeal motoneurons. But we do not want to entirely exclude that some premotor mechanisms may also occur in PiCo. The reviewer may know that there is overlap of rhythmogenic and premotor functions for the Dbx1 neurons in the PreBötC, But, addressing this issue is beyond the scope of this study. In fact, we are working on a separate connectivity study using novel, still unpublished antegrade and retrograde vectors that do not reveal any direct connections to laryngeal motoneurons. Hence, we expect that the connectivity from PiCo to laryngeal motoneurons is more complex and addressing this question cannot be done as a simple add-on to an already complex study. Again, we would refer to the PreBötzinger complex, where nobody expects that one study can resolve all the physiological and anatomical characterizations that have been accumulated over 30 years in one study. We would argue that in some ways, our cre/FlpO approach is more specific than the Dbx1 stimulations which activates not only rhythmogenetic PreBötzinger complex neurons, but also pre motoneurons as well as glia cells, and many neurons rostral and caudal to the PreBötzinger complex. We are aware of these caveats, and we have discussed this in the original submission, and also in the revision.

While the study from Toor et al is cited, there needs to be a much more thorough discussion of how their findings relate to the current study.

Many thanks for asking for a more thorough discussion of Toor et al., which we are happy to provide here. Perhaps we were too polite in our original manuscript to emphasize all the problems in that study.

They demonstrated that PiCo isn't necessary for the apneic portion of swallow. Inhibiting this region also didn't affect TI.

Please note – the fact that Toor et al did not find an effect on TI confirms Anderson et al. 2016: In Figure 3G,3F of the Nature paper, the reviewer will find that injections of DAMGO and SST into PiCo inhibited post-I activity without affect inspiratory duration. This figure also shows that the inspiratory burst can terminate in the absence of postinspiratory activity.

The reviewer states: “They demonstrated that PiCo isn't necessary for the apneic portion of swallow”. With all due respect to this reviewer, this is NOT correct. Toor et al showed that inhibiting PiCo did block SLN-evoked fictive-swallows but not the apnea caused by SLN stimulation. This is not the apnea caused by swallows (which was never studied by Toor), but by the SLN stimulation. The apnea evoked by SLN stimulation has most likely nothing to do with the apnea caused by swallows. Unfortunately, the Toor et al. makes the same misleading claim as the reviewer.

PiCo cannot be the sole source of post-I timing, and the evidence overwhelmingly favors the major involvement of other regions such as the pons.

This comment seems to be unrelated to the main thrust of this paper that studies PiCo’s involvement in swallow and laryngeal activation in coordination with breathing. However, since this comment seems to discredit the Ramirez lab in general, we would like to clarify that inhibiting PiCo with DAMGO and SST inhibits post-I activity (Anderson et al 2016, Fig.3G,3F). Thus, we don’t understand the rationale or actual data for the reviewer’s conclusion that PiCo cannot be the sole source of post-I timing? We also don’t understand the basis for the reviewer’s conclusion that “the evidence overwhelmingly favors the major involvement of other regions such as the pons”. We also want to add, that no-where in the Anderson et al. study did we state that the pons plays NO role. Indeed, we specifically stated: “In this context it will be interesting to resolve the role of the PiCo in specific postinspiratory behaviors and to identify how the PiCo interacts with other neural networks such as the Kolliker-Fuse nucleus, a pontine structure that has been hypothesized to gate postinspiratory activity and the periaqueductal grey a structure involved in vocalization and the control of postinspiration”.

They also showed that inhibition of all neurons (not just ChAT/Vglut) in the PiCo region suppresses post-I activity in eupnea. This suppression was overcome by the increased respiratory drive during hypoxia.

Before comparisons are made with Toor et al. it is important to note the species and methodological differences between Toor et al. rat anesthetized, vagotomized, paralyzed and artificially ventilated model which evaluated fictive swallows (deafferented and paralyzed). By contrast this study uses a mouse anesthetized, vagal intact, freely breathing model and evaluates natural physiologic swallow via water and central stimulation. It seems that the reviewers does not acknowledge one of the main innovations of this study. For this study we introduced a genetic approach to specifically target and activate ChATcre/Vglut2FlpO PiCo neurons. This has never been done before, and developing this approach took more than 4 years of breeding and crossing and testing different options.

As for Toor et al., these authors pharmacologically, bilaterally inhibited neurons in the area of PiCo with isoguvacine, a specific GABA-A agonist. Even though this pharmacological intervention does not specifically inhibit cholinergic/glutamatergic neurons in PiCo, these authors essentially confirm the study by Anderson et al. We do not find this finding controversial. Perhaps the reviewer finds the definition of PiCo “controversial”, because Toor et al called the identical area IRt instead of PiCo, even though they exactly reproduce the finding by Anderson. Toor et al. not only arrive at the same conclusion as Anderson but they added more details – none of which is contradicting the results by Anderson et al.: Here are excerpts from the Toor study “We therefore conclude that the ongoing activity of neurons in the IRt contributes to eupneic respiratory and sympathetic post-I activities without exerting significant control on other respiratory or cardiovascular parameters” “IRt significantly inhibited the post-I components of VNA” “IRt inhibition was also associated with a reduction in PNA” “increase in respiratory cycle frequency” “due to a reduction in TE“ “with no effect on TI observed”. “Bilateral microinjection of isoguvacine selectively reduced the magnitudes of post-I VNA and rSNA, but not PNA responses to acute hypoxemia”.

In this statement the reviewer probably refers to one particular aspect, i.e. the fact that Toor et al. did not significantly block some of the post-I activity – they state: “had no significant effect on the AUC of post-I rSNA (305+/- 24 vs 230+/- 28,p=0.16,n=6)”. Please note that there is a tendency, a reduction from 305-230. Perhaps the Toor study was not sufficiently powered to fully block the effect, perhaps the drug did not inhibit the entire PiCo. These are all open questions that a critical reader should know. The reviewer will agree that it is as difficult if not more difficult to demonstrate the absence of an effect. To arrive at a negative conclusion experiments should be done with the same scrutiny than to demonstrate a positive result. We also assume that the reviewer is familiar with animal experiments and will understand that pharmacological injections are often difficult to interpret, in particular in case of local in vivo injections. It is possible that Toor et al is inhibiting e.g. parts of the Bötzinger complex.

We have added to the manuscript the following statement: It is important to note that SLN stimulation does not only trigger swallows, but also changes in the overall stiffness and tension of the vocal cords (Chhetri et al., 2013) as well as prolonged hypoglossal activation independent of swallowing (Jiang, Mitchell, & Lipski, 1991). It has been hypothesized that inhibition of the IRt blocks fictive swallow but not swallow-related apnea. Yet this apnea was generated by SLN stimulation and not by a natural swallow stimulation (Ain Summan Toor et al., 2019). It is known that SLN stimulation causes endogenous release of adenosine that activates 2A receptors on GABAergic neurons resulting in the release of GABA on inspiratory neurons and subsequent inspiratory inhibition (Abu-Shaweesh, 2007), suggesting that the SLN evoked apnea may not be the same as a swallow related apnea. Moreover, microinjections of isoguvacine into the Bötzinger complex attenuated the apneic response but not the ELM burst activity (Sun, Bautista, Berkowitz, Zhao, & Pilowsky, 2011), suggesting the Bötzinger complex, not PiCo, could be involved in modulating apnea.

We would also like to add that our current study characterized swallow-related specific muscles and nerves in both water-triggered and PiCo-triggered swallows to better characterize the physiological properties of this swallow behavior. By contrast, Toor et al. only characterized nerve activities that are involved in multiple upper airway activities and breathing. It is somewhat surprising that the reviewer did not consider the fact that Toor et al. characterized putative swallows that were triggered by SLN stimulation and that Toor et al. were content with nerve-recordings and failed to confirm that the behavior that they evoked is actually a physiological swallow. Which, according to the comments from this reviewer (see above), indicates the possibility of differences in central mechanisms occurring between fictive swallow and physiological swallows.

While we have cited Toor et al and their truly excellent work in the broad iRt we did not feel it is appropriate to critique them for the fact that they are confusing the field by using a different anatomical term for the area that was clearly defined by us as an area containing cholinergic-glutamatergic neurons. We also did not feel it is appropriate to discuss results that are similar to comparing Apples and Oranges. Toor et al. never specifically manipulated glutamatergic-cholinergic neurons, thus their entire results rest on indirect stimulation affecting this general area – which will unavoidably also include laryngeal motoneurons. We don’t want to criticize this approach, since PiCo is heterogenous, which is another misunderstanding that we find in the reviewers’ critique. We used cholinergic-glutamatergic neurons to define this area. However, like the preBötC, PiCo is also heterogenous. This region contains inhibitory neurons, it also contains glutamatergic neurons that are not cholinergic, and cholinergic neurons that are not glutamatergic. Because of this heterogeneity we compared the effects of stimulating glutamatergic neurons and cholinergic neurons as well as cholinergic-glutamatergic neurons. This is an approach that is generally accepted in the field. As already stated, there is not a single marker that uniquely characterizes the PreBötC. Thus, when stimulating Dbx1 neurons, glutamatergic neurons, or Somatostatin neurons it only captures subpopulations of this region. The recently published study by Menuet et al. in eLife, used even more indirect methods to inhibit preBötC. They used a pan-neuronal CBA promotor that targets neurons irrespective of phenotype. It is not our intention to discredit this very elegant study, but we object the statement that we “have arbitrarily defined the PiCo region”.

This study has not demonstrated some of the things that are depicted in Fig 7 and included in the discussion. While swallow can inhibit inspiration, there are many mechanisms by which this can happen other than a direct inhibitory connection from the DGS to PreBotC. You cite Sun et al., 2011 findings of "a group of neurons that inhibits inspiration" during SLN stim, but don't mention that it is the BotC and that the paper shows that swallow apnea is dependent on BotC. That is also supported by the Toor study. I don't understand how post-I (aka E) can be discussed without discussion of the BotC - this is a glaring omission.

We have removed figure 7, which was only meant as a hypothetical schematic.

Why is it necessary for PiCo to innervate the cNTS?

This was a hypothesis based on CTb data that we have now removed.

That is true if the conjecture that PiCo gates swallowing is true, as the cNTS is the only known region for central swallow gating. However, PiCo could influence afferent input to the NTS less directly, and therefore not function as a gating hub per se. The experimental evidence does not warrant the claim that PiCo gates swallowing. The definition of a swallow gate(s) is a topic of much debate and no conclusive experimental evidence has emerged for swallow gating regions to exist anywhere except in the NTS. The current study's evidence also does not meet the criteria necessary to conclusively call PiCo a swallow gate. The authors should soften this claim and language throughout the manuscript.

Although we do not know of any studies that has optogenetically gated swallow in the cNTS, it seems the reviewer objects our use of the word “gate”. We have revised the manuscript and removed any wording stating PiCo is a swallow “gate”. It would be interesting to know whether the reviewer has the same objections of the use of the word “relay” as done by Toor et al.?

It is also unclear that PiCo acts directly on the swallow pattern generator to gate swallowing. It is not just "conceivable that the gating mechanism involves" the pons, but nearly certain. Swallow gating by respiratory activity may not be able to be ascribed to one particular location. At a minimum, it likely involves the NTS/DSG, pons, and possibly IRt (inclusive of PiCo). The authors are correct that "further studies are necessary to understand the interaction between PiCo and the pontine respiratory group on the gating swallow and other airway protective behaviors." This is why it shouldn't be stated that "this small brainstem microcircuits acts as a central gating mechanism for airway protective behaviors."

We have removed all language stating PiCo is a swallow gate.

PiCo is likely part of the VSG (and thus the swallow pattern generator). PiCo, as part of the IRt/VSG could indeed be surveilling afferent information and providing output that affects swallow or other laryngeal activation and the coordination of these behaviors with breathing. However, this is not the responsibility of PiCo alone. This role is likely shared by other parts of the SPG, and may require distributed SPG network participation to be functional. If one were to stim other regions of the distributed SPG, similar results might be expected. When Toor et al silenced the PiCo area (and locations that lie at least lightly beyond the borders of what the present study defines as PiCo), stim-evoked fictive swallows were greatly suppressed. However, swallow-related apnea was unaffected. This supports the role of PiCo as a premotor relay for swallow motor activation, but not as the site that terminates inspiration. Therefore, it cannot be called a gate.

We already addressed the issue that Toor never demonstrated that the “swallow-related” apnea was unaffected. Toor et al only demonstrated that the SLN-evoked apneas were unaffected, and their conclusions were only based on nerve recordings under fictive conditions (deafferented and paralyzed). Also, to the best of our knowledge, many aspects of the putative swallow pattern generator that this reviewer mentions are purely hypothetical. However, to avoid further arguments, we have removed the word gate and Figure 7 from this manuscript.

Similarly, Fig 7 does not accurately depict things that are already well-supported by evidence. PiCo should be included as part of the swallow pattern generator (VSG), not as a separate entity acting on it. The BotC and pons are glaring omissions. This study has not demonstrated the labeled inhibitory connection from DSG to PreBotC. The legend states speculations as fact and needs to be dialed way back to either include statements with solid experimental evidence or to clearly mark things as putative/speculative.

We have removed figure 7.

The discussion of expiratory laryngeal motoneurons needs to be expanded and integrated better into the discussion of swallow, post-I, and other laryngeal motor activation. Why can't PiCo just be premotor to ELMs?

If PiCo would “only” or “just” be premotor to ELM then it would not be expected that it could trigger an all-or-none swallow response with a temporal activity pattern similar to the one of a water-evoked swallow. We would also not expect that the activation of the activity pattern is independent of the laser stimulation duration as demonstrated in Figure 3. This was our reasoning why we originally called PiCo a “gate” because at the correct phase it will gate/trigger a complex swallow sequence. But, as stated above, we avoid the word gate in the revised manuscript.

Concerning the discussion of "PiCo's influence as a gate for airway protective behaviors is blurred...": The incomplete swallow motor sequence didn't seem super different in timing or duration compared to the fully transfected animals (comparing plots from Fig 6 to Fig S1, and Table S2 to Table S3. The values for swallow durations (XII and X) for each group for water and opto seem within similar ranges, as do the differences between water & opto-evoked swallows between strains. While the motor pattern is distinctive from the normal swallow, with laryngeal activity rather than submental activity leading, one might not even be able to call that a swallow. Is it evidence against a classic all-or-nothing swallow behavior any more than the graded swallow results from (fully transfected) Table S1?

We fully agree that it is possible that this unidentified behavior may not be a swallow. We have changed the name of this behavior to “upper airway motor activity.” However we also cannot rule out the possibility of this being some portion of a graded swallow which would argue that a graded swallow response is exact evidence against the classic all or nothing swallow behavior.

Please expand on this point and put it into context with others' results: "This brings into question whether this is the first evidence against the classic dogma of swallow as an "all or nothing" behavior, and/or whether this is an indication that activating the cholinergic/glutamatergic neurons in PiCo is not only gating the SPG, but is actually involved in assembling the swallow motor pattern itself."

This has been expanded and included citation of other studies. The following paragraph can be found in the discussion

Swallow has been thought of as an “all or nothing” response as early as 1883 (Meltzer, 1883). Whether modulating spinal or vagal feedback (Huff A, 2020b), central drive for swallow/breathing (Huff, Karlen-Amarante, Pitts, & Ramirez, 2022) or lesions in swallow related areas of the brainstem (Car, 1979; Robert W Doty, Richmond, & Storey, 1967; Wang & Bieger, 1991) swallow either occurred or did not. Swallows are thought to be a fixed action pattern, with duration of stimulation having no effect on behavior duration (Fig. 3) (Dick, Oku, Romaniuk, & Cherniack, 1993). Thus, it was particularly interesting that in instances when few PiCo neurons were transfected, either unilateral or bilateral, an unknown activation of upper airway activity occurred. Motor activity no longer outlasted laser stimulation rather was contained within, and the timing of the motor sequence was reversed in comparison to a water or PiCo evoked swallow (Fig. 6). Thus, if insufficient numbers of neurons are activated, PiCo’s influence to specifically activate swallow or laryngeal activation is blurred, resulting in the uncoordinated activation of muscles involved in both behaviors. This brings possible evidence against the classic dogma of swallow as an “all or nothing” behavior, or the presence of an entirely different behavior. We are not the first to bring possible evidence against the classic dogma, “small swallows” were described but failed to be discovered if this was in-fact a partial or incomplete swallow (Miller & Sherrington, 1915). The SPG is thought to consist of bilateral circuits (hemi-CPGs) that govern ipsilateral motor activities, but receive crossing inputs from contralateral swallow interneurons in the reticular formation, thought to coordinate synchrony of swallow movements (Kinoshita et al., 2021; Sugimoto, Umezaki, Takagi, Narikawa, & Shin, 1998; Sugiyama et al., 2011). Incomplete activation of PiCo activates the muscular components of a swallow, without establishing the coordinated timing and sequence of the pattern. It is possible that PiCo is involved in assembling the swallow motor pattern itself and unilateral activation of PiCo could either desynchronize swallow interneurons or activates only one side of the SPG. Since we did not record bilateral swallow related muscles and nerves this question needs to be further examined.

Reviewer #3 (Public Review):

Huff et.al further characterise the anatomy and function of a population of excitatory medullary neurons, the Post-inspiratory Complex (PiCo), which they first described in 2016 as the origin of the laryngeal adduction that occurs in the post-inspiratory phase of quiet breathing. They propose an additional role for the glutamatergic and cholinergic PiCo interneurons in coordinating swallowing and protective airway reflexes with breathing, a critical function of the central respiratory apparatus, the neural mechanics of which have remained enigmatic. Using single allelic and intersectional allelic recombinase transgenic approaches, Huff et al. selectively excited choline acetyltransferase (ChAT) and vesicular glutamate transporter-2 (VGluT2) expressing neurons in the intermediate reticular nucleus of anesthetised mice using an optogenetic approach, evoking a stereotyped swallowing motor pattern (indistinguishable from a water-induced swallow) during the early phase of the breathing cycle (within the first 10% of the cycle) or tonic laryngeal adduction (which tracked tetanically with stimulus length) during the later phase of the breathing cycle (after 70% of the cycle).

They further refine the anatomical demarcation of the PiCo using a combination of ChAT immunohistochemistry and an intersectional transgenic strategy by which fluorescent reporter expression (tdTomato) is regulated by a combinatorial flippase and cre recombinase-dependent cassette in triple allelic mice (Vglut2-ires2-FLPO; ChAT-ires-cre; Ai65).

Lastly, they demonstrate that the PiCo is anatomically positioned to influence the induction of swallowing through a series of neuroanatomical experiments in which the retrograde tracer Cholera Toxin B (CTB) was transported from the proposed location of the putative swallowing pattern generator within the caudal nucleus of the solitary tract (NTS) to glutamatergic ChAT neurons located within the PiCo. We would like to thank the reviewer for acknowledging the technical advances of the present study and for the positive statements in general.

Methods and Results

The experimental approach is appropriate and at the cutting edge for the field: advanced neuroscience techniques for neuronal stimulation (virally driven opsin expression within a genetically intersecting subset of neurons) applied within a sophisticated in vivo preparation in the anaesthetized mouse with electrophysiological recordings from functionally discrete respiratory and swallowing muscles. This approach permits selective stimulation of target cell types and simultaneous assessment of gain-of-function on multiple respiratory and swallowing outputs. This intersectional method ensures PiCo activation occurs in isolation from surrounding glutamatergic IRt interneurons, which serve a diverse range of homeostatic and locomotor functions, and immediately adjacent cholinergic laryngeal motor neurons within the nucleus ambiguous (seen by some as a limitation of the original study that first described the PiCo and its roll in post-I rhythm generation Anderson et al., 2016 Nature 536, 76-80). These experiments are technically demanding and have been expertly performed.

Again, we would like to thank the reviewer for these positive comments acknowledging the advances of the present study.

The supplemental tracing experiments are of a lower standard. CTB is a robust retrograde tracer with some inherent limitations, paramount of which is the inadvertent labelling of neurons whose axons pass through the site of tracer deposition, commonly leading to false positives. In the context of labelling promiscuity by CTB, the small number of PiCo neurons labelled from the NTS (maybe 5 or 6 at most in an optical plane that features 20 or more PiCo neurons) is a concern. Even assuming that only a small subset of PiCo neurons makes this connection with the presumed swallowing CPG within the cNTS, interpretation is not helped by the low contrast of the tracer labelling (relative to the background) and the poor quality of the image itself. The connection the authors are trying to demonstrate between PiCo and the cNTS could be solidified using anterograde tracing data the authors should already have at hand (i.e. EYFP labelling driven by the con-fon AAV vectors from PiCo neurons (shown in Fig5), which should robustly label any projections to the cNTS).

We fully agree with the reviewer that the CTB staining is of a lower standard and have removed this approach.

The retrograde labelling from laryngeal muscles seems unnecessary: the laryngeal motor pool is well established (within the nAmb and ventral medulla), and it would be unprecedented for a population of glutamatergic neurons to form direct connections with muscles (beyond the sensory pool).

The authors support their claim that PiCo neurons gate laryngeal activity with breathing through the demonstration that selective activation of glutamatergic and cholinergic PiCo neurons is sufficient to drive oral/pharyngeal/laryngeal motor responses under anaesthesia and that such responses are qualitatively shaped by the phase of the respiratory cycle within which stimulation occurs. Optical stimulation within the first 10% of the respiratory cycle was sufficient to evoke a complete, stereotyped swallow that reset the breathing cycle, while stimuli within the later 70% of the cycle, evoked discharge of the laryngeal muscles in a stimulus length-dependent manner. Induced swallows were qualitatively indistinguishable from naturalistic swallow induced by the introduction of water into the oral cavity. The authors note that a detailed interpretation of induced laryngeal activity is probably beyond the technical limits of their recordings, but they speculate that this activity may represent the laryngeal adductor reflex. This seems like a reasonable conclusion.

We thank the reviewer for this comment. Unfortunately, we felt compelled to remove the word “gating” based on the statements by reviewer 1.

The authors propose a model whereby the PiCo impinges upon the swallowing CPG (itself a poorly resolved structure) to explain their physiological data. The anatomical data presented in this study (retrograde transport of CTB from cNTS to PiCo) are insufficient to support this claim. As suggested above, complementary, high-quality, anterograde tracing data demonstrating connectivity between these structures as well as other brain regions would help to support this claim and broaden the impact of the study.

We fully agree with this reviewer. We have been working on a thorough anatomical characterization for more than 3 years using cutting edge anterograde and retrograde viruses in collaboration with vector experts at the University of Irvine. But these are partly novel, unpublished techniques that are in development, and require many careful controls and characterization. We feel that this is a separate study as it doesn’t relate to swallowing coordination and also includes partly different authors. We hope to submit this as a separate study later this year.

This study proposes that the PiCo in addition to serving as the site of generation of the post-I rhythm also gates swallowing and respiration. The scope of the study is small, and limited to the subfields of swallowing and respiratory neuroscience, however, this is an important basic biological question within these fields. The basic biological mechanisms that link these two behaviors, breathing and swallowing, are elusive and are critical in understanding how the brain achieves robust regulation of motor patterning of the aerodigestive tract, a mechanism that prevents aspiration of food and drink during ingestion. This study pushes the PiCo as a key candidate and supports this claim with solid functional data. A more comprehensive study demonstrating the necessity of the PiCo for swallow/breathing coordination through loss of function experiments (inhibitory optogenetics applied in the same transgenic context) along with robust connectivity data would solidify this claim.

Thanks again for the positive assessment of our study.

Author Response

Reviewer #1 (Public Review):

This paper explores the potential regulatory role of a previously unstudied phosphorylation site in the Src kinase SH3 domain. The data presented conclusively demonstrate that a phosphomimetic mutation of this site, src90E, causes an elevation in Src kinase activity, changes the structure of the Src catalytic domain as determined with a FRET sensor, disrupts certain SH3 domain interactions, causes changes in kinase intracellular dynamicity, and promotes cell invasiveness. Based on the behavior of the phosphomimetic mutant, the idea that constitutive phosphorylation of Y90 could have all of these effects is well-supported by the data. However, in wild-type cells or cells transformed by activated forms of Src, there is no constitutive phosphorylation of this site. Therefore, the question remains whether Y90 phosphorylation occurs to any significant extent in cells, and the data suggesting that it could do so is limited. It also remains to be conclusively established whether Y90 phosphorylation occurs via autophosphorylation.

Major comments:

1) Y90 was identified as a site of phosphorylation in Luo et al. It would be helpful if more information were provided about its significance relative to other sites identified in that study. Was it detected in non-transformed cells? Was it a major site? How does it relate to Y416 in abundance? The reference to the identification of the site in a different study from the White lab is made in the discussion but not in the introduction (this should be corrected). How abundant was it that study? A fuller description of its detection would strengthen the rationale for this study. Any additional phosphoproteomics studies that identified it should also be included.

As indicated in the manuscript (Figure 3C and new 3D), the amount of Y90 phosphorylation increases with the level of Src activation. Standard proteomic/phosphoproteomic data cannot be quantified in absolute values for technical reasons, only relative quantification is possible to some extent. To overcome this issue and address the question of the amount of Y90 phosphorylation, we newly prepared the corresponding stable isotope-labeled phosphopeptides and used them as internal standards. To the best of our knowledge, this allowed us to quantify for the first time the amount of specific tyrosine phosphorylation of Src kinase in cells. We found that in case of WT Src, the major phosphorylation site localized in the activation loop of the kinase domain, Y416, is phosphorylated in 22 % of molecules. In activated Src, this pool of Y416-phosphorylated molecules increases 2,5 times to 57 %. Y90 is phosphorylated in approximately 1 % of WT Src molecules but becomes 5 times more abundant in case of the activated kinase (5,3 % of phosphorylated molecules). This newly added data of absolute Src tyrosine phosphorylation (Figure 3D) is consistent with values we obtained from relative MS quantification of Src variants differing in catalytic activity (Figure 3C). Although the enrichment of Y90 phosphorylation in the catalytically active kinase is lower compared to Y416 phosphorylation in terms of percentage of phosphorylated molecules, it’s increment with respect to the basal state is significantly higher. We believe that this broader dynamic range of Y90 phosphorylation is in agreement with the demonstrated regulatory function of Y90 phosphorylation. We incorporated these new results and methodological approach into the revised manuscript. We also extended the original description of the MS protocol to include a description of relative quantification, which was included in the original manuscript.

Phosphorylation of Y90 was only detected in Luo et al. and Johnson et al. phosphoproteomic screens. However, phosphorylation of tyrosines homologous to Src Y90 was described in a vast number of proteins. Some of them are mentioned in the discussion e.g., Btk, Abl, p130Cas or Src family kinases Yes and Fyn. The presence of phosphorylation on homologous tyrosines and the evolutionary conserved nature of Y90 in SH3 domains supports relevance of Src Y90 phosphorylation despite the small number of studies that were able to identify it. In our opinion, this can be attributed to its low abundance in the basal state and the technical difficulties of its detection, as discussed below in response to point 2.

We emphasize the Luo et al. study in the introduction because it was the only study reporting Y90 phosphorylation at the time of the project’s initiation and led us to study Y90 further. Both studies are then mentioned in the discussion, which we believe is appropriate and sufficient.

2) Related to point 1, is there evidence from the literature indicating a significant site of phosphorylation in Src has been overlooked? Or, was this site only identified because of the recent advances in MS technology and increased sensitivity of this methodology? An introduction to these points would also enhance the rationale for the study.

In the manuscript discussion, we mention an early study (Erpel et al., 1995) which mapped conserved residues within the binding surface of the Src SH3 domain. It showed that mutation of Y90 to alanine led to partially deregulated Src and reduced affinity of the SH3 domain. Although they acknowledged the importance of Y90 for SH3 domain binding ability, they did not probe or discuss the effect of Y90 phosphorylation status. Furthermore, the level of Src Y90 phosphorylation in untransformed cells is relatively low (20-fold lower than Y416 phosphorylation). It is therefore not surprising that it has not been identified in most general phosphoproteomic studies performed on untransformed cells. In fact, in many of these studies, Y416 phosphorylation was not detected either. The detection of Y90 phosphorylation by Luo et al. likely reflects the fact that it was performed in Src527F-transformed cells, similarly Johnson et al. used HGF-activated cells. Last, we also cannot exclude that the tryptic peptides with Y90/pY90 are less detectable in MS depending on the experimental conditions. In fact, the "heavy" Y90 peptide was consistently much less (10-80 times less) detectable in our hands than the Y416 peptide. This could be because of its worse ionizability, stability in vacuum or some other technical reasons.

In our approach, we used immunoprecipitated Src molecules to maximize the amount of Src in the sample and targeted MS, which allowed us to specifically detect even low abundant ions/peptides. This represented the critical technical approach that allowed us to consistently detect Y90 phosphorylation in untransformed cells.

3) The explanation of the MS experiment designed to show that Y90 phosphorylation happens in cells is insufficient in the text. It is not clear why the SYF cells were not used and not clear why the FRET sensor constructs were used. It is also not clear whether or how the proteins were purified before MS analysis. Also, rather than showing the MS data as a relative level, it would be preferable to provide the number of spectra obtained for each peptide/phosphopeptide and compare this also to Y416. A fuller comparison between the phosphorylation of Y90 to that of Y416 is necessary in order to place the potential Y90-mediated phosphoregulation in context.

We are sorry for the confusing description. With the new quantification data, we have rewritten this section and hopefully made it clearer. We kept the original relative quantification data as they nicely show that abundance of Y90 phosphorylation increases with enhanced activity of Src. However, we added new MS analysis of Src tyrosine phosphorylation performed with labeled peptides as internal standards that provides absolute numbers of Y416 and Y90 phosphorylation in cells. The new <br /> measurements confirm the original data showing increased Y90 phosphorylation in activated Src variants and suggest that Y90 phosphorylation is not a rare event but represents an important regulatory element in Src activation. Our approach of MS quantification of phosphorylation events using labeled peptides as standards, allowed us, to the best of our knowledge, for the first time, to measure absolute quantities of Y416 and Y90 phosphorylation and therefore also the amount of activated Src molecules in cells.

For technical reasons, the SrcFRET biosensor was used in all these experiments. We attempted to analyze endogenous Src in several cell lines to assess its Y90 phosphorylation. However, in our hands, the amount of Src efficiently precipitated was never sufficient to detect the "very elusive" phosphopeptide containing Y90. We believe this was not caused by low amounts of Src in the cells, <br /> but rather because the anti-Src antibody performed much worse than the anti-GFP antibody used for SrcFRET biosensor (two high affinity epitopes) immunoprecipitation. We have previously shown that the SrcFRET biosensor functions in the same way as endogenous Src (Koudelková et al., 2019), and therefore we presume that it is phosphorylated in a similar manner and rate as endogenous Src.

4) I would like to see conclusive evidence that Y90 phosphorylation is due to autophosphorylation. This would involve relatively simple experiments. As one possibility, an IP kinase assay followed by immunoblotting with a site-specific antibody or MS or other types of phosphopeptide visualization/identification.

We further addressed the question of Y90 autophosphorylation using a kinase dead version of Src527F bearing K295M substitution. To quantify the amount of phosphorylated Src we applied the identical approach with labeled standards and measured phosphorylation levels of Y416 and Y90. Compared to Src WT and Src527F, phosphorylation of both tyrosines in the kinase dead variant was negligible despite the presence of endogenous Src and other SFKs in the U2OS cells we used for the experiments. These results suggest that phosphorylation of Y90 does indeed depend on the intrinsic kinase activity of Src and is therefore very likely autophosphorylation.

We have tried to address the question of Src autophosphorylation on Y90 by analyzing the level of Y90 phosphorylation in cells expressing a kinase-inactive SrcFRET construct with open conformation (527FKD) by quantitative MS. Despite the open conformation, SrcFRET527F-KD did not display any significant phosphorylation of neither Y90 nor Y416, even though we used U2OS cells which express endogenous Src and other SFKs. These results suggest that phosphorylation of Y90 depends on catalytic activity of the kinase rather than on compactness of its conformation and is therefore very likely autophosphorylation.

5) A few other mutations would be interesting to examine in both kinase and transformation assays for comparison to the mutants that were: Y527F Y416F; Y527F Y416F Y90E. The first is a low activity control and the second is for understanding whether Y90E could overcome the lack of Y416 phosphorylation.

Due to lack of time, we did not perform these experiments. However, we analyzed our new kinasedead 527F mutant for FRET and found that despite its inactive kinase domain and lack of Y416 phosphorylation, it still retains an open conformation. We believe that this is a strong indication that the Y90E kinase-dead mutant would behave the same way, maintaining an open conformation albeit the kinase domain is inactive.

6) I recommend that the results are discussed in a more circumspect manner. The results presented in Figure 7 on the double mutant, Y527F Y90F, suggest that phosphorylation of Y90 is not a very significant component of Src kinase regulation, at least in these biological contexts. That Y90 phosphorylation isn't a major regulatory factor does not diminish the value of the work describing Y90 phosphorylation. However, it does alter the interpretations. I encourage a more conservative discussion of its importance and a broader discussion of why it isn't a major site of Src phosphorylation, particularly if it is due to autophosphorylation.

We believe that given our new quantifications showing that Y90 phosphorylation is indeed considerably present and utilized in cells, the original discussion is consistent with the new data and does not need to be changed.

Reviewer #2 (Public Review):

The manuscript "Phosphorylation of tyrosine 90 in SH3 domain is a new regulatory switch controlling Src kinase" describes efforts to understand how phosphorylation of tyrosine (Y90) in the SH3 domain of Src affects the activity and function of this multi-domain kinase. The authors find that an Src variant containing a phospho-mimetic mutation (Glu) at position 90 demonstrates elevated activation levels in lysates and cells (Figure 1) and adopts a less compact autoinhibited conformation within the context of a SrcFRET biosensor in lysates (Figures 3A, 3B). A series of pulldown experiments with an isolated SH3 domain (Figure 2A, 2B) or full-length Src (Figure 2C, 2D) that contain the phospho-mimetic Y90E mutation demonstrates that phosphorylation of Tyr90 would likely disrupt the interaction of Src's SH3 domain with intermolecular binding partners and the linker that couples SH2 domain/C-tail binding to autoinhibition, which provides a mechanistic basis for the observed elevated kinase activity of Src Y90E. By performing a series of imaging experiments with a SrcFRET biosensor, the authors show that the Y90E mutation does not show enhanced localization at focal adhesions like a hyperactivated Src mutant (Y527F) that contains a non-phosphorylatable C-tail (Figure 4A). However, using ImFCS combined with TIRF microscopy (Figure 4B), the authors demonstrate that Src Y90E shows similarly reduced mobility (relative to the WT SrcFRET biosensor) at the plasma membrane (especially at focal adhesions) as Src Y527F. Consistent with the elevated kinase activity of Src Y90E, the authors go on to demonstrate that the Src Y90E variant shows an ability to transform fibroblasts-at levels that are intermediate between wild-type Src and the hyperactive Src mutant Y527F (Figure 5). Similarly, Src Y90E confers an intermediate level (between wild-type Src and Src Y527F) of invasiveness and ability to form spheroids. Together, these comprehensive experiments with a Y90 phospho-mimetic strongly support a model where phosphorylation of Src's SH3 domain at Tyr90 would lead to a more intramolecularly disengaged SH3 regulatory domain and enhanced kinase activity in cells.

Most of the conclusions in this paper are well supported by solid data, but confidence in several assays would be higher if additional technical detail or controls were provided and the biological significance of these findings would be higher if the role that Y90 phosphorylation plays in Src regulation and function were better delineated.

1) The kinase activity assays in Figures 1C,1D, and 7A need to be scaled to the Src variant levels present in the lysate (quantification of relative Src levels is not provided).

For kinase activity measurements, we used lysates of equal protein concentrations prepared from cell lines stably expressing Src variants. These cell lines were sorted and repeatedly tested for equal expression of Src constructs using immunodetection of Src on Western blots. We corrected the <br /> methods section and added this information to the description of kinase assays experimental setup.

2) More details are required for the experiments quantifying Y90 phosphorylation levels in Figure 3C. The experimental states that equal amounts of IP'd proteins were used for these analyses but there are no details on how this was confirmed. In addition, the experimental states that normalized intensities were used for your quantifying the Y90 phospho-peptide but no details are provided on how normalization was performed (the legend states that a base peptide was used but it is unclear what this means).

The paragraph on mass spectrometry analysis in the Materials and Methods section has been updated with the required information.

3) A key question is whether Y90 phosphorylation serves a regulatory role in Src's cellular activity and, if so, what is the regulatory network that mediates this phospho-event. Using a mass spectrometry readout with three Src variants (wild type vs. Y527F vs. E381G) that possess differing kinase activities, the authors demonstrate that Y90 phosphorylation levels correlate to Src's kinase activity (Figure 3C), which they suggest is an indication that this residue is an autophosphorylation site (or phosphorylated by another Src family kinase). However, as Src's kinase activity correlates with SH3 domain disengagement (which leads to a more accessible Y90), it is also entirely possible that another tyrosine kinase is responsible for this phosphorylation event. More importantly, it is unclear under which signaling regime Y90 phosphorylation would play a significant regulatory role. This phospho-event was observed in a previous phospho-proteomic study but it is unclear whether the phosphorylation levels of this site occur high enough stoichiometry to modulate the intracellular function of Src and whether there is a regulatory signaling network that influences Y90 phosphorylation levels.

We have tried to address the question of Src autophosphorylation on Y90 by analyzing the level of Y90 phosphorylation in cells expressing a kinase-inactive SrcFRET construct with open conformation (527F-KD) by quantitative MS. Despite the open conformation, SrcFRET527F-KD did not display any significant phosphorylation of neither Y90 nor Y416, even though we used U2OS cells which express endogenous Src and other SFKs. These results suggest that phosphorylation of Y90 depends on catalytic activity of the kinase rather than on compactness of its conformation and is therefore very likely autophosphorylation.

To further support our data on relevance of Y90 phosphorylation in cells, we performed a new MS analysis of Y90 and Y416 phosphorylation in WT and activated Src. This time we used corresponding stable isotope-labeled peptides and phosphopeptides as internal standards for MS quantification. This allowed us to measure absolute amounts of phosphorylated molecules and changes in their numbers, which is information that cannot be acquired by standard biochemical or proteomic approaches and is usually lacking for the majority of known phosphorylation sites. We found that in case of WT Src, the major phosphorylation site localized in the activation loop of the kinase domain, Y416, is phosphorylated in 22,6 % of molecules. In activated Src, this pool of Y416-phosphorylated molecules increases 2,5 times to 57 %. Y90 is phosphorylated in approximately 1 % of WT Src molecules but becomes 5,1 times more abundant in case of the activated kinase (5,3 % of phosphorylated molecules). This newly added data of absolute Src tyrosine phosphorylation (Figure 3D) is consistent with values we obtained from relative MS quantification of Src variants differing in catalytic activity (Figure 3C). Although the enrichment of Y90 phosphorylation in the catalytically active kinase is lower compared to Y416 phosphorylation in terms of percentage of phosphorylated molecules, it’s increment with respect to the basal state is significantly higher. We believe that this broader dynamic range of Y90 phosphorylation is consistent with and reflects the demonstrated regulatory function of Y90 phosphorylation. We incorporated these new results and methodological approach into the revised manuscript.

Author Response

Reviewer #1 (Public Review):

This manuscript focuses on a set of neurons from the border between the central and medial amygdala (AMGc/m-PAG ) that project to neurons in the periaqueductal gray (PAG) that gate ultrasonic vocalizations (USVs). These neurons suppress vocal production and are active in contexts where vocalizations would be inappropriate (e.g. in the presence of predator cues, or aggressive encounters with conspecifics). They then further characterized these neurons, demonstrating that like in males, these neurons are GABAergic in females and in both sexes, half of these neurons express estrogen receptor alpha (Esr1). To examine the inputs into these neurons, the authors performed monosynaptically-restricted transsynaptic rabies tracing and identified numerous cortical and subcortical projections. Of particular interest, neurons from the preoptic area of the hypothalamus (POA) in addition to terminating on PAG-USV neurons also project to AMGc/m-PAG neurons. Imaging the terminals of these neurons revealed elevated activity during vocalization-promoting contexts and optogenetically stimulating them resulted in evoking USVs. Together, these experiments further identify and quantify a circuit incorporating external factors (e.g. predatory factors, social interactions) in the drive to produce vocalizations.

The authors are commended for use of male and female mice, demonstrating that even though they produce USVs in different social contexts, AMGc/m-PAG neurons share a function in suppressing USV production in both sexes. They do this convincingly with a variety of methodologies while incorporating appropriate controls (e.g. light-only and GFP-control in optogenetic experiments). The experiments are performed in a logical order and the data generated is elaborate.

We appreciate the reviewer’s commendations and for their appreciation of the convincing insights provided by our study. We provide detailed responses to their recommendations in the following section. We hope the reviewer finds these revisions satisfactorily address their concerns.

Reviewer #2 (Public Review):

The existence of PAG-USV-producing neurons has been recently established, alongside two independent pathways, POA->PAG, and AMG->PAG, that promote and inhibit the production of ultrasound vocalizations in female and male mice, respectively. Because vocalizations can be modulated in a variety of contexts, such as in the presence of a predator, the authors first show that the AMG->PAG pathway is activated in situations where mice stop vocalizing, such as in the presence of a predator or aggressive conspecifics, and can inhibit natural vocalizations in contexts where females vocalize (extending to their previous findings in male mice). Interestingly, AMG->PAG neurons also receive input from POA neurons that are known to promote vocalizations via their connection to PAG interneurons that inhibit PAG-USV-producing neurons. This POA->AMG and PAG pathway is inhibitory and therefore its capacity to promote vocalizations via these two parallel pathways might be achieved by its inhibition of AMG and PAG neurons that inhibit the PAG-USV producing neurons. While these results hint at possible mechanisms that could underlie the hierarchical control of vocalization, and how different external signals impinge on existing pathways to produce behavior flexibility, the study is missing important elements to draw such conclusions. Overall, the study is also missing important information on how experiments were performed.

We appreciate the reviewer’s efforts to evaluate our manuscript and provide constructive feedback. In the following section, we have responded to all the reviewer’s comments and concerns and provide all but one of the previously missing elements and information. We also maintain that the results and additional analysis we provide in this manuscript go beyond merely hinting at possible mechanisms, and instead provide explicit synaptic mechanisms by which vocal-promoting and vocal suppressing signals interact in the mouse’s brain.

Author Response

Reviewer #1 (Public Review):

The authors worked towards a better understanding of the functional diversification of flavodoxins among diatoms, and this represents a quantum contribution building on the initial findings of Whitney, Lins, Hughes, Wells, Chappelle, and Jenkins (2011), with the inclusion of metatranscriptomic and other data from field collections and on-deck incubation experiments, relatively new genomic and transcriptomic datasets, and the adoption of reverse genetics tools that are not yet widely used in T. pseudonana. They hypothesize that clade I flavodoxins play a role in mitigating oxidative stress, while additional clade II flavodoxins would respond according to canon, in response to low iron availability.

The authors embarked on several field campaigns across environmental gradients where iron-responsive and oxidative stress-responsive flavodoxins were expected to show differential expression. The use of metatranscriptomics allowed taxa-specific assignment of relative transcript expression levels, and the results of both measurements across the environmental gradient and manipulative incubation experiments show the widespread taxonomic distribution of iron-responsive clade II flavodoxin. The fieldwork was well thought out, and biogeochemical trends comported to expectations. It's worth noting that the concomitant inclusion of geochemical data such as dissolved iron further strengthened the work. The authors also found clade I flavodoxins were not iron-responsive (as expected), but rather exhibited diel patterns in transcript abundance that suggest responses to photo-oxidative stress. Taken together, these field data are stunning.

We thank the reviewer for this kind assessment.

Lab experiments with five diatom species grown under varied iron and induced oxidative (H2O2) stress and transcript abundances for flavodoxin genes are reported. One reservation concerns the untoward and unknown effects of inducing outright iron starvation with the strong chelator, DFB (as opposed to achieving steady-state growth rate limitation from low iron by use of weak chelators such as EDTA). With DFB it is also difficult to predict sample timing (when cells have hit that "correct" and reproducible iron-limited space) when independent replicates are collected on different dates. Similarly, the use of DFB also makes it difficult to sample low and high iron cells at the same density or to maintain densities among replicate samples collected on different dates. pH and CO2 availability change with density unless special measures are taken.

We agree with the reviewer that DFB is a strong iron chelator that may affect diatom physiology in inadvertent ways. We designed the DFB experiments to allow us to screen multiple diatoms for whether they transcribed clade I and II flavodoxins in a short-term response to iron limitation.

We added the logic behind this experimental design (L177-179):

“In order to screen multiple diatoms for whether they transcribed clade I and II flavodoxins in response to iron limitation, we used the strong iron-chelator desferrioxamine B (DFB) and enhanced short-term iron limitation.”

Additionally, we now discuss the possible effect of DFB in our discussion (L395-410):

“Notably, we used the strong iron chelator DFB to enhance iron limitation in a variety of diatoms, as previously described (Andrew et al., 2019; Kranzler et al., 2021; Lampe et al., 2018; Timmermans et al., 2001; Wells, 1999), while recognizing that undesirable effects of DFB, that are not related to iron-limitation per se cannot be ruled out. Here, DFB was used in experiments designed to test whether transcription of the two flavodoxin clades differentially responded to iron limitation. The results from T. oceanica, and T. pseudonana agree with the literature, in which DFB was not added. In T. oceanica only the expression of one clade II flavodoxin was induced (Figure 2B-C, as in Lommer et al., 2012). The minor induction in mRNA of T. pseudonana clade I flavodoxin in response to iron limitation was detected in both long- and short-term adaptation to low iron, without added DFB (Goldman et al., 2019; Thamatrakoln et al., 2012). This flavodoxin seems to have diel regulation, and the observed induction might be specific to the circadian time and the setting of the diel cycle (Goldman et al., 2019).”

Based on the reviewer comments, we realized that our transcriptome sampling protocol was not clear. Because the diatom species have different growth rates, as well as different rates of growth-inhibition by iron limitation, we adjusted the sampling day for each species based on cell counts and photosynthetic efficiency. Importantly, the 9 samples (triplicates of 3 conditions) of each species were sampled together, at the same date and time. We also ensured that the biological replicates of each species and treatment had similar cell density at the time of harvest.

We clarified these concerns in the Results section (L188-206):

“For each diatom 6 replicates were grown in iron-replete conditions and 3 replicates in iron-limiting conditions until the low-iron cultures displayed a decrease in maximum photochemical yield of photosystem II (Fv/Fm), 3-6 days (depending on species, Figure 2 -figure supplement 1A-C, Figure 2A, supplementary file 1c), indicative of iron limitation, at which point transcriptome samples were collected for both the iron-limited and iron-replete conditions. Three of the iron-replete replicates were exposed to oxidative stress, mimicked by a lethal dose of H2O2, and transcriptome samples were collected about 1.5 h after exposure, when the cell phenotype (Fv/Fm or cell abundance) was unaltered from control.”

In the Materials and Methods section (L542-545)

"Cells were harvested by filtration onto 0.22 µm filters. Full details of the number of cells harvested per treatments, per species, and samples that failed library preparation are indicated in supplementary file 1c. The 9 samples of each diatom species were sampled together, at the same date and time. Filters were snap-frozen…”

A second set of lab experiments involved the (non-trivial) establishment and use of "knock out" clones of the clade I flavodoxin gene in the model diatom T. pseudonana to test the oxidative stress hypothesis. This is an exciting idea and the data suggest this flavodoxin may confer resistance to oxidative stress. The conclusion would be greatly strengthened if different phenotypes could be observed between WT and KO clones in response to environmentally relevant oxidative stress (such as supra-optimal irradiance), rather than exogenous H2O2 addition.

Based on the reviewer suggestion, we conducted a preliminary experiment with irradiation of up to 500 µE. As with the light level originally tested, there were no differences in growth rate or Fv/Fm between the WT and KO lines. We agree that future study of these knock-lines a series of much higher irradiation levels, photosynthetic-inhibitors, and other environmental stresses is interesting, but it is out of the scope of the current study.

We now also mention this in the revised manuscript (L417-419):

“Future studies in which the oxidative stress is driven by other environmental conditions as supra-optimal irradiation, UV radiation or biotic interactions are needed to further support the role of clade I flavodoxins in oxidative stress.”

We clarify that our use of exogenous H2O2 additions was based on previous studies with Phaeodactylum and T. pseudonana that indicate that exogenous addition of micromolar range of H2O2 is representative for other oxidative stress-responses (Graff van Creveld, 2015, Volpert 2018, Mizrachi 2019) (L185-188):

“Oxidative stress was induced by the lowest lethal dose of H2O2 (200-250 µM), as similar treatment was shown to be representative to other environmentally-relevant oxidative stressors in T. pseudonana and Phaeodactylum (Graff van Creveld et al., 2015; Mizrachi et al., 2019; Volpert et al., 2018).”

The relationship between the experimental conditions and results in Figure 3C and Supplemental Figure 3H was not clear.

Figure 3C summarize parts of Figure S3H information, Figure S3D-I present the individual clones, while Figure 3 only shows WT vs Flav-KO.

According to the reviewer comments, we modified Figure S3H (it is now Figure S3I), and specify this relationship in the legend:

“H-I. Percentage of Sytox Green-positive (dead) cells, measured by flow cytometry 24 h after treatment with H2O2 treatment. Orange and gray box plots represent a Flav-KO and WT respectively, single measurements are marked, color-coded by the individual colonies. H. Results of a single dose-response experiment. I. Results from additional experiments, experiments marked with an asterisk are summarized in main Figure 3C.”

In the introduction, the authors suggest that Fe-S-containing proteins are particularly sensitive to damage via oxygen and ROS and that reliance on ferredoxin (Fd) for electron shuttling carries an enhanced sensitivity to the ROS generated during photosynthesis. References would be helpful here. Fe-S cluster-containing proteins are not monolithic regarding their behavior or susceptibility towards ROS. My limited understanding is that (i) several 4Fe-4S cluster proteins (such as aconitase, isopropylmalate isomerase) are particularly sensitive but that (ii) this is less so for canonical 2Fe-2S cluster ferredoxins; (iii) in some phototrophs Fd catalyzes the reduction of molecular oxygen to superoxide, as part of a mechanism that keeps the electron transport chain less reduced under extremely high light. Thus, ferredoxins may not necessarily be susceptible to in vivo ROS-mediated damage.

Thank you for these comments.

We modified our original sentence (L37-39):

“Moreover, iron-sulfur-containing proteins are particularly sensitive to damage via oxygen and reactive oxygen species (ROS).”

Corrected sentence:

“Moreover, iron containing proteins are sensitive to damage via oxygen and reactive oxygen species (ROS), and Fd is down-regulated in response to oxidative stress (Singh et al., 2010, 2004).”

Reviewer #2 (Public Review):

In their manuscript, Van Creveld et al. set out to demonstrate divergent functions for two clades of flavodoxin in diatoms. To achieve their goals, the authors combined metatranscriptomic results originating from three separate research cruises in the North Pacific Ocean with laboratory experiments with a clade I flavodoxin knock-out mutant in the diatom Thalassiosira pseudonana. Overall, their field study confirmed that Clade II flavodoxin is mostly up-regulated under iron limitation in most diatoms that were represented in their metatranscriptomic data (Figure 5 A-F). Their field study also demonstrated that clade I flavodoxin is expressed at levels that are several orders of magnitude lower than clade II flavodoxin (figure 5H). The lower expression of clade I flavodoxin was also observed in laboratory culture experiments (Figure 2). The laboratory experiments also demonstrated that the clade I flavodoxins were responsive to iron limitation in some of the species studied (Their Figure 2C), such that the assignment of function based solely on the clade I and clade II flavodoxin classification may not always be straight forward, and that exceptions will likely be found as more diatom species are studied.

In their quest to determine whether Clade I flavodoxin plays a role in adaptation to oxidative stress, the authors created several knock-out mutants where the clade I flavodoxin is not functional. These mutant strains responded to iron limitation in the same way as the WT strains. However, the mutant strains defective in the clade I flavodoxin were more slightly more sensitive to oxidative stress (created by exposure to lethal doses of hydrogen peroxide) than the wild-type strains. The results of the oxidative stress challenges would have been stronger if a broader concentration range of hydrogen peroxide had been used in the experiments leading to a dose-response curve for both the mutant and wild-type strains.

Thank you for this suggestion. We now tested a broader range of H2O2 concentrations on the WT and KO strains and added a new Figure S3H, which includes responses to 0, 25, 50, 75, 100, 150, 200, 250 µM H2O2.

The supplemental information provided in the main manuscript holds a lot of important information. Take for example Figure S4 showing the placement of reads for Clade I and Clade II in a Maximum-likelihood tree for flavodoxin in the North Pacific Ocean. The results show that clade II flavodoxin is much more commonly found in the transcripts than clade I flavodoxin.

Perhaps different results would have been obtained by conducting a similar sampling of metatranscriptome in the Atlantic Ocean that is less subject to iron limitation.

We agree completely and would love to analyze metatranscriptomes from the Atlantic Ocean in the future.

Overall, the authors have provided results that support a role for Clade I flavodoxin in alleviating oxidative stress in Thalassiosira pseudonana, however, whether or not this role is universal for clade I flavodoxin in other diatom species will require further studies.

We agree with this assessment that additional experiments with additional diatoms is a fruitful research area into the future.

Author Response

Reviewer #1 (Public Review):

In their study Mas Sandoval and colleagues estimate, from human genomic data, two important parameters that measure how intermarriages have been affected by social stratification in the Americas: sex-biased admixture (SB), which refers to sex differences in the chances to intermarry with another ethnic group, and ancestry-based assortative mating (AM), which refers to the higher probability of partners to intermarry when they carry similar genetic ancestries. To do so, the authors train a deep neural network (DNN) with simulations of admixture with non-random mating and use ancestry tract length distributions to infer the two parameters. They show that their approach estimates SB and AM parameters with a relatively good accuracy in a number of scenarios. When applying the DNN to empirical data, they find solid evidence that social stratification has constrained the admixture processes in the Americas for the last centuries.

In contrast with the vast majority of population genetic studies, which assume random mating, this study assesses if mating has been random or not in American populations. Furthermore, the study is very valuable because it leverages, for the first time, a deep learning approach and local ancestry inference to co-estimate the extent of SB and AM from genomic data.

One limitation of the study, however, is that it assumes that (i) the admixture date in the simulations is known and equals 19 generations and (ii) admixture started at the same time in all admixed American populations. The authors also implicitly assume that the variance of the difference between male and female ancestry proportions only depends on AM, and not admixture timing. This may be problematic, as it has been shown that linkage disequilibrium between local ancestry tracts depends both on AM and admixture timing (Zaitlen et al., Genetics 2017).

To clarify the assumption of fixed admixture date, we have added the following sentence in the results section (line 170) where the model is firstly described: “In both models we assume a continuous admixture process that starts 19 generations ago, knowing that the populations analysed trace the first contact of Native American and European populations in the first half of 16th century and assuming a generation time of 26 .9 years (Wang et al., 2023). In contrast with the approaches that aim to find an admixture date assuming random mating, we assume that the admixture process starts with the contact, and it is continuous and modulated with the mating parameters.”

We thank the reviewer for such an important reference we had not included in our manuscript, whose findings support the basis of our approach. It is now included on line 70 to justify the analysis of the length of the ancestry tracts: Herein, we argue that the tract length information can measure the non-randomness of mating associated with genetic ancestry and, therefore, it can also monitor the permeability of socioeconomic and cultural barriers between subpopulations with different genetic ancestries (Zaitlen et al., 2017)

This is also suggested by the authors' results, showing that AM estimates are much lower in admixed Americans under the two-pulse model, relative to the one-pulse model, i.e., when admixture extends over time. Estimates of AM in admixed Americans may thus be biased, if admixture actually started less (or more) than 19 generations ago.

We evaluated the resemblance of the footprints left by either assortative mating or gene flow, by testing how a neural network trained on models with gene flow due to a second migration pulse predicts migration size on data generated by models without a second migration pulse but assortative mating only . We then tested how neural networks trained on models with assortative mating detect assortative mating from data with no assortative mating but only migration. Results are summarised in Figures 4 – supplement 1 – supplement 2 and show a strong correlation of the predicted size of the second migration pulse and the simulated level of assortative mating. Parallelly, there is also a strong correlation between the predicted assortative mating level and the size of the second migration pulse. Below, we respond to the reviewers in more detail regarding this question.

Another potential limitation concerns local ancestry inference. The authors assume that RFMix makes no errors when inferring ancestry tracts. This can be a concern, as recent studies have shown that RFMix has reduced accuracy compared to other methods (Hilmarsson et al., bioRxiv 2022).

In response to this comment, we performed a local ancestry analysis with Gnomix and generated the tract length profile according to the results obtained. One possible issue shared by Gnomix and RFMix is that they may infer a higher fraction of short tracts (at the expense of breaking longer ones). This issue was reported by Gravel et al. (2012). In this study, authors decided to filter out the short tracts because these tracts showed a high rate of false positives and false negatives. Therefore, we conducted an experiment to test if filtering out the shortest tract length window (i) improve the accuracy of the predictions of the simulated test data through the Mean Squared Error (MSE), ii) decrease the uncertainty of the estimations, and (iii) increase the correlation between Gnomix and RFMix-based estimates through the generalised variance.

We also tested a modification of the tract lengths profile by dividing (or not) the tract lengths profile by the total amount of tracts in either the Autosomes or the X chromosome. Our goal was to force the neural network to focus on the profile shape rather than on the absolute value of tracts at each window to mitigate the possible bias in the tract length profile. Our experimental set-up consisted of three combinations of modifications of the tract length profile, in addition to the non-modified one.

In Figures 4 supplement 3 – supplement 7, we show the predicted mating parameters using the modifications of the tract length profile outcoming from the local ancestry inference. Each point represents a prediction using RFMix and Gnomix tract length profiles (x and y axis, respectively) as input for each of the 1000 trained neural networks with the same architecture. We evaluated the uncertainty of the estimations for both Gnomix and RFMix and the correlation between them through the Generalised Variance. The Generalised Variance is the determinant of the covariance matrix, which increases with low values of covariance of the bivariate distribution and high values of the respective variances.The estimations of the parameters based on the tract length profile normalised by dividing by the total amount of fragments in Autosomes or X chromosome had both low values of Generalised Variances in the Gnomix-RFmix bivariate distribution of predicted parameters and low values of MSE in the prediction of simulated test data. These results indicate that by normalising the tract length profile by the total amount of fragments, the distribution is still informative and less sensitive to possible biases introduced by errors in the local ancestry analysis .

Therefore, we present the results obtained from this RFMix profile in the main figures and tables, while showing the other predictions in the supplementary figures.

In addition, the authors do not report a measure of uncertainty for the estimation of SB and AM, which is another important weakness. Interpretation of parameter estimates is limited if no measures of uncertainty are provided.

We now provide the 95% CI for each parameter obtained from the distribution of predicted parameters from the 1000 trained neural networks, for both RFMix and Gnomix for the tract length profile.

Finally, the authors compare the likelihood of two competing models, assuming a single or two admixture pulses, but do not determine the accuracy of their model choice procedure.

We now include the confidence intervals of the composite likelihood by replicating the test for each of the 1000 bootstrapped tract length profiles for each population. None of the 95% confidence intervals includes both negative and positive results and all of them support either the one pulse or the two pulses model, except for the sub-Saharan ancestry in the Columbian (CLM) population.

Overall, besides these methodological limitations, I expect that the study by Mas Sandoval and colleagues could be of great and broad interest for the scientific community studying population genetics, anthropology, sociology and history.

Reviewer #2 (Public Review):

This paper introduces a method to quantify how genetic ancestry drives non-random mating in admixed populations. Admixed American populations are structured by racial, gender, and class hierarchies. This has the potential to cause both ancestry-related assortative mating, in which the ancestry of mates tends to be correlated, and ancestry-related sex bias, in which individuals have a preference for mates with a particular ancestry composition. By applying their method to several African American and Latin American populations, Sandoval et al. further our understanding of ancestry-based population structure in this region more broadly.

Strengths

As many others have recently done, Sandoval et al. leverage the ability of a neural network to predict demographic parameters from high-dimensional population genomic data. Sandoval et al. first develop a clever probabilistic model of mating by defining the probability of a male and female mating as a function of the difference in ancestry between the individuals. They use this model to simulate population genomic data under various demographic scenarios, and then train a neural network on these simulated data. Finally, they apply the neural network to empirical data and learn the parameters of the underlying probability distribution, which can be related back to assortative mating and sex bias.

One clear strength of this paper is their ability to jointly assess assortative mating and sex bias, as well as their ability to apply their model to multiple contemporary admixed populations.

Importantly, the authors couch their results in an intersectional understanding of populations and consistently refer to research from historians and other social scientists throughout their paper, which reflects a very thoughtful awareness of the interdisciplinary nature of this research.

Weaknesses

The definition of assortative mating is conceptually confusing - in the text, assortative mating is introduced as genetic similarity between mates, i.e. positive assortative mating. However, based on the definition of assortative mating in their model, a population can have high assortative mating for a particular ancestry component even when there is non-zero sex bias for that component (e.g. males with low Native American ancestry are more likely to mate with females with high Native American ancestry). Fundamentally, this scenario cannot reflect positive assortative mating; rather, it reflects negative assortative mating (i.e. there is structured genetic dissimilarity between mates). However, the authors do not discuss the fact that the interpretation of the assortative mating parameter changes with the value of the sex bias parameter.

We acknowledge that our definition of assortative mating requires more clarity. We now define it on line 155 as: The AM parameter measures the non-randomness of mating associated to a genetic ancestry. This includes both positive assortative mating -genetic similarity between mates- (when SB is zero) and negative assortative mating -genetic dissimilarity between mates- (when SB is not zero). This approach allows accounting for the male-female way of negative assortative mating through SB parameter.

In addition, the results of the inference in ASW are difficult to interpret. They find that males of high African ancestry are more likely to mate with females of low African ancestry. This result seems counterintuitive given the body of literature that suggests sex-biased admixture in African Americans has greater male European and female African contributions. The authors do not suggest potential explanations for this observation.

We agree that results regarding the ASW population can be confusing. Our hypothesis to explain such results is that the sex bias parameter captures both sex-biased migrations and sex-biased admixture. Therefore, it is difficult to accommodate the complex genetic history of ASW. We have extended the discussion on this aspect as follows on line 380:

In addition, African American populations might have a complex genetic history involving on one hand male-biased sub-Saharan migration and on the other hand an admixture femalebiased in the sub-Saharan ancestry. However, our current model can only accommodate this demographic scenario with a single sex-bias parameter, and the results regarding this population should be interpreted with caution.

Lastly, the authors have not done any simulations to assess how accurate parameter estimates are if the demographic model is misspecified, which weakens the interpretability of the results.

We have performed a new analysis where we vary AM to generate tract length profiles to predict GFR, and viceversa. The results of this analysis are shown in the new figure 4Supplement 1. Results show how the footprint in the genome of the admixing populations of assortative mating and multiple pulse migration is similar. In the discussion we argue that both One Pulse and Two pulse models must be considered because they are supported by results obtained using X chromosome and Autosomes, respectively. We discuss how accounting for migration reduces AM values and how the resulting admixture dynamics resemble in both cases.

Author Response

Reviewer 1 (Public Review):

1) In Figure 2, electron microscopy images represent n=1 cell, making it hard to know how generalizable the mitochondrial phenotypes are. It would be useful to see a quantitative summary of a larger dataset indicating how frequently the mitochondrial defects are seen.

As requested, we performed quantitative analysis of mitochondrial ultrastructure in a larger dataset (n=163 analyzed in WT and n=206 in the KO) confirming that this finding is very consistent. This additional quantitative analysis that we included in the revised manuscript confirms a very significant and diffuse alteration of mitochondrial ultrastructure in Parl-/- vs WT spermatocytes (p=0.0002).

2) In Figure 3, representative images are shown for a single field from n=1 animal. It is hard to decisively conclude that the phenotype of Pink1-/-;Pgam5-/- and Ttc19-/- testes is completely normal based on this limited data. There may be other tubules outside the field of view that are abnormal, or more subtle changes in cell ratios. This conclusion would be significantly strengthened by cell counting (e.g. # round spermatids per Sertoli cell per tubule and # spermatocytes per Sertoli cell per tubule) or other quantitation. Likewise, the similarities in phenotype between Parl-/-, Parl-/-;Pink2-/-, and Parl-/-;Pgam5-/- should be more thoroughly documented. At least some additional images should be shown.

The goal of figure 3 is to indicate that WT, Pink1-/-;Pgam5-/- and Ttc19-/- have no gross morphological abnormality and have preserved sperm production in sharp contrast with Parl/-, Parl-/-;Pink1-/-, and Parl-/-;Pgam5-/- and the TKO that show total lack of sperm in the tubular lumen, indicating that the loss of Parl alone or in combination drives this phenotype. To strengthen these conclusions we performed additional work. We stained testis sections from all strains with an antibody for AIF-1, a marker of post-mitotic spermatids/spermatozoa included in Fig3-figure supplement 1. This additional experiment clearly confirms that production of differentiated germ cells occurs only in WT, Pink1-/-;Pgam5-/- and Ttc19-/-, but not in Parl-/, Parl-/-;Pink1-/-, and Parl-/-;Pgam5-/-. These results are consistent with the reproductive capacity of these mouse lines (the first group is fertile, the second is infertile). We acknowledge we cannot rule out minimal subclinical differences in reproductive fitness between the fertile mouse groups, but this is beyond the goal of our study.

3) In Figure 4, it looks like there is a significant decrease in CIV-driven respiration in Parl knockouts, but the text describes this as "did not significantly enhance" - that is, the absence of an increase. This result is difficult to interpret without further explanation.

We recognize this might be confusing but it is specified in the text that CIV driven (TMPD+ascorbate) respiration- relying on endogenous cytochrome c- is diminished (line 195) in Parl-/- testis mitochondria. This test reflects cytochrome c oxidase respiratory capacity/activity. We performed then an additional experiment just after the previous where we add exogenous cytochrome c in the cuvette to test the integrity of the outer mitochondrial membrane and checked if CIV-driven respiration increases or not after,compared to before, the addition of cytc. Exogenous cytochrome c does not cross intact mitochondrial outer membranes, so the test is performed to verify the good quality of mitochondrial preparations and/or pathological changes by looking if of the outer membrane integrity, not the function of CIV. CIV driven respiration increases only modestly after compared to before the addition of cytc and to a similar extent in both WT and Parl-/- indicating a good quality of the mitochondrial preparations and that the outer mitochondrial membrane of these mitochondria is overall well preserved in both WT and KO.

4) In Figure 5B, there is some variation in band intensity between replicates. Quantifying the band intensity relative to the loading control would help to increase confidence in the conclusion that coQ levels are reduced.

We performed this quantification, as suggested by the reviewer, and added the quantification in figure 5B. Quantification of the band intensity relative to the loading control confirms a significant difference between WT and KO. Moreover, we performed quantitative immunofluorescence of COQ4 in SCP-1 positive cells included now in Fig 5-figure supplement 1, which confirms a significantly decreased expression of COQ4 in Parl-/- primary spermatocytes.

5) GPX4 is not a Parl substrate, and no explanation is provided for why it might be reduced in Parl-/- testes. This makes the result and model difficult to interpret.

We thank the reviewer for pointing this out. We acknowledged this limitation in the discussion. We mentioned in the discussion that decreased GPX4 levels have been observed in other conditions (chemical inhibition, pathological conditions, etc.) and no mechanism has so far been demonstrated to our knowledge, but some evidence raises a possible link with CoQ deficiency that we discussed. Potential mechanisms including protein degradation are likely although unproven. This remains an important and intriguing issue to address in future studies.

6) Since Parl knockout induces necrosis in the brain, necrosis could be a contributing factor to cell death in spermatocytes alongside ferroptosis. No data is presented that can exclude this possibility.

Ferroptosis is actually considered, by some authors, a form of regulated necrosis (Seibt TM FRBM 2019). Therefore, we can affirm that PARL deletion leads to regulated necrosis in testis via ferroptosis through specific ferroptosis pathways that do not appear to be activated in the brain, or at least not overtly. Importantly, there is no recognized marker or specific molecular pathway for generic «accidental» necrosis that can be tested to differentiate between the 2 different cell death modalities.

7) The severe spermatogenesis phenotype implies that Parl knockout males should be infertile, but the fertility status is not described in the manuscript. It may be difficult to test fertility in these animals due to the neurodegeneration phenotype; if so, this can be clarified. If it is feasible to test fertility, demonstration of a fertility phenotype would significantly strengthen the conclusion that loss of Parl leads to spermatogenic arrest.

We specify in the text that Parl-/- mice are sterile due to total lack of sperm production caused by arrested spermatogenesis, as evidenced by detailed histological analysis and AIF1 staining. This is not due to the neurodegeneration since Parl-Ncre knockout have normal production of sperm as presented in the paper. Fertility in Parl-/- cannot be tested in vitro since these mice have no sperm due to the complete block of spermatogenesis, nor in vivo since they die young due to neurodegeneration. With these limitation Parl-/- males and WT females are kept together and in no single exception since the beginning of the colonies a pregnancy has ever been observed. Parl-/- mice are sterile.

Author Response

Reviewer #1 (Public Review):

The authors tried to measure the accuracy of the decision-making of honey bees by carrying out behavioural experiments in which they trained the bees to forage on artificial flowers of 5 different colours that offered different levels of reward. Subsequently, the bees' decision-making behaviour was tested with flowers of the same or different colours, with no reward present. The authors found that bees tend to approach a flower only when they are highly certain of a reward, and these decisions are made quickly. The majority of flowers were rejected by the bees. Based on the results of the tests, the authors created a model to identify what circuit elements or connections would be necessary to mimic the bees' decisions. This model could be potentially used for robotics.

The study is well supported by the signal detection theory and the experiments are well designed which is a major strength. However, the methods are not completely clear, so would be better to make a clearer description. Another weakness is the lack of clear explanations of the importance and relevance of the model.

Given the experimental design was optimal, the authors could potentially achieve the aims of this study.

Thank you for expressing your interest and providing constructive inputs. Based on your suggestions, we have thoroughly revised our manuscript to offer a more comprehensive explanation of the rationale behind our approach, as well as its comparison to existing knowledge and methods in the field. We believe that these revisions will significantly enhance the comprehensibility of our study and facilitate a better understanding of our findings.

Reviewer #2 (Public Review):

By elegantly designing experiments, MaBouDi et al. elucidated honeybee's behavioral strategy to quantitatively associate sensory cues with valences. The description is simple and concise enough to understand the logic. Particularly, the authors clearly demonstrated how sensory evidence and reward likelihood quantitatively affect the decision-making process and animals' response time. Their behavioral characterization approach and proposed model could also be helpful for studies using higher animal species. I have a few doubts regarding the definition of rejection behavior and the structure of the model that is critical to lead their main conclusions.

Thank you for your interest and valuable feedback. We greatly appreciate your input, and as a result, we have thoroughly reviewed your comments and implemented significant revisions to our manuscript. We have taken care to provide more comprehensive explanations of our methods, results, and the proposed model in order to enhance the overall comprehensibility of our study. Our intention is to ensure that readers can better understand our findings through these revisions.

Author Response

Reviewer #1 (Public Review)

Using in vitro assays that take advantage of thymic slices, with or without the ability to present pMHC antigens, the authors define an early period in which CCR4 expression is induced, which induces their migration to the medulla and likely encounter with cDC2 and other APCs. Notably, the timing for CCR4 expression precedes that of CCR7 and illustrates the potential role for this early expression to initiate the movement of post-positive selection thymocytes to the medulla. The evidence for supporting a role for CCR4, as well as CCR7, in sequential tolerance induction is provided using multiple approaches, and although the observed changes amount to small percent changes, the significance is clear and likely biologically relevant over the lifespan of a developing T cell repertoire. Overall, the model provides a holistic view of how tolerance to self-antigens is likely induced during T cell development, which makes this work highly topical and influential to the field.

We thank the reviewer for their comments and for highlighting the significance of identifying distinct roles for CCR4 and CCR7 in promoting medullary localization and inducing self-tolerance of thymocytes at different stages of T-cell development.

Reviewer #2: (Public Review )

This manuscript describes that CCR4 and CCR7 differentially regulate thymocyte localization with distinct outcomes for central tolerance. Overall, the data are presented clearly. The distinct roles of CCR4 and CCR7 at different phases of thymocyte deletion (shown in Figure 6C) are novel and important. However, the conclusion that expression profiles of CCR4 and CCR7 are different during DP to SP thymocyte development was documented previously. More importantly, the data presented in this manuscript do not support the conclusion that CCR7 is uncoupled from medullary entry. Moreover, it is unclear how the short-term thymus slice culture experiments reflect thymocyte migration from the cortex to the medulla.

We thank the reviewer for pointing out the significance of our finding that CCR4 and CCR7 regulate different phases of thymocyte deletion. We agree that prior reports, including our own (Cowan et al. 2014, Hu et al., 2015) have shown that CCR4 and CCR7 are expressed by different post-positive selection thymocytes. However, the expression data we present here provides a higher resolution perspective on the specific thymocyte subsets that express these two receptors, as well as the different timing with which the receptors are expressed after positive selection. These data, coupled with chemotaxis assays of the granular thymocyte subsets responding to CCR4 versus CCR7 ligands, and 2-photon imaging data showing that CCR4 and CCR7 are required for medullary accumulation of distinct thymocyte subsets, are critical for delineating the unexpectedly distinct roles of these two chemokine receptors in promoting medullary entry and central tolerance.

The reviewer raises an important question about our conclusion that CCR7 is “uncoupled” from medullary entry. We think there was likely a misunderstanding of our intended meaning, as we did not mean to imply that CCR7 does not promote medullary entry of thymocyte subsets; we have modified the wording of the abstract to replace “uncoupled” to clarify. As we detail in the Introduction, the role of CCR7 in directing chemotaxis of single-positive thymocytes towards the medulla and inducing their medullary accumulation is well established (Ehrlich et al., 2009; Kurobe et al., 2006; Kwan & Killeen, 2004; Nitta et al., 2009; Ueno et al., 2004). Instead, our data demonstrate that 1) the most immediate post-positive selection thymocyte subset (DP CD3loCD69+) does not require CCR7 for medullary entry, and 2) the next stage of post-positive selection thymocytes (CD4SP SM) express CCR7, but CCR7 recruits these cells only modestly into medulla. In contrast, CCR7 promotes robust medullary accumulation of more mature thymocyte subsets (CD4SP M1+M2), in keeping with the well-known role of CCR7 in promoting thymocyte medullary localization. We think these findings are highly significant for the field because currently, there is a widely held assumption that post-positive selection thymocytes that do not express CCR7 are located in the cortex, while those that express CCR7 are located in the medulla. Our data show that neither of these assumptions is true: CCR4 drives medullary accumulation of cells that do not yet express CCR7, and the earliest post-positive selection cells that express CCR7 continue to migrate in both the cortex and medulla. These findings form the basis of our statement that CCR7 expression is “not synonymous with” medullary localization. The finding that thymocytes do not robustly accumulate in the medulla in a CCR7-dependent manner until more the mature SP stages has important implications for central tolerance, as localization of thymocytes in the cortex versus medulla will impact which APCs and self-antigens they encounter when testing their TCRs for self-reactivity.

The reviewer also raised concerns about whether short-term thymus slice cultures reflect physiological thymocyte migration. Short-term live thymic slice cultures have been widely used to investigate the development, localization, migration, and positive and negative selection of thymocytes, as they have been shown to faithfully reflect these in vivo processes, including confirming the role of CCR7 in inducing chemotaxis of mature thymocytes from the cortex into the medulla (Au-Yeung et al., 2014; Dzhagalov et al., 2013; Ehrlich et al., 2009; Lancaster et al., 2019; Melichar et al., 2013; Ross et al., 2014). However, we acknowledge that thymic slices are not equivalent to intact thymuses and have now discussed limitations of this system in our revised Discussion.

Comment 1: Differential profiles in the expression of chemokine receptors, including CCR4, CCR7, and CXCR4, during DP to SP thymocyte development were well documented. Previous papers reported an early and transient expression of CCR4, a subsequent and persistent expression of CCR7, and an inverse reduction of CXCR4 (Campbell, et al., 1999, Cowan, et al., 2014, and Kadakia, et al. 2019). The data shown in Figures 1, 2, and 3 are repetitive to previously published data.

The expression profile of CCR4, CCR7 and CXCR4 on thymocytes has been documented previously in the studies cited above and in our prior publication (Hu et al., 2015). Campbell et al. (Campbell, Haraldsen, et al., 1999) investigated chemotactic effects of chemokines, but did not directly address expression of chemokine receptors by thymocyte subsets. Cowan et al. (Cowan et al., 2014) examined the expression of CCR4 versus CCR7 on DP and CD4SP thymocytes. However, our data provide a more detailed analysis of expression of these distinct chemokine receptors by subsets of DP, CD4SP, and CD8SP thymocyte subsets along the trajectory of differentiation after positive selection, using a gating scheme inspired by a study published after the above-cited papers (Breed et al., 2019). Our more nuanced evaluation of CCR4 versus CCR7 expression sets the stage for finding that they play distinct roles in promoting medullary entry and central tolerance of early- versus late-stage post-positive selection thymocytes. Without examining CCR4 and CCR7 expression patterns by distinct thymocyte subsets in detail, we would not have made the unexpected observation that although CCR7 is expressed at high levels by many CD4SP SM thymocytes, it does not induce strong chemotaxis or medullary accumulation of this subset, relative to its role in more mature SP thymocyte subsets. This finding has important implications for which APCs thymocytes encounter as they are tested for self-reactivity to enforce central tolerance. As we were working on these studies, Kadakia et al. reported that extinguishing CXCR4 expression was important for enabling medullary entry (Kadakia et al., 2019). Thus, we thought it was important to place CXCR4 in the context of CCR4 and CCR7 expression on thymocyte subsets in our study, and in doing so found another example of asynchronous chemokine receptor expression and function, further indicating that expression of a chemokine receptor alone is not a reliable marker of functional activity or thymocyte localization, as cells migrate dynamically between the cortex and medulla.

Through more extensive gating and simultaneous investigation of chemokine receptor expression and function, our data have provided new insights into how thymocytes respond to chemokine cues at different time points during their post-positive selection development. Moreover, our refined gating scheme (Figure 1) can be used to distinguish thymocyte subsets at different development stages without relying on chemokine receptor expression, thus providing an unbiased way of investigating chemokine receptor expression at different developmental stages.

Comment 2: The manuscript describes the lack of CCR7 at early stages during DP to SP thymocyte development (Figure 1-3). However, CCR7 expression is detected insensitively in this study. Unlike CCR4 detection with a wide fluorescence range between 0 and 2x104 on the horizontal axis, CCR7 detection has a narrow range between 0 and 2x103 on the vertical axis (Figure 1C, 1D, 4B, 4C, 6B, S2, S3), so that flow cytometric CCR7 detection in this study is 10-times less sensitive than CCR4 detection. It is therefore likely that the "CCR7-negative" cells described in this manuscript actually include "CCR7-low/intermediate" thymocytes described previously (for example, Figure S5A in Van Laethem, et al. Cell 2013 and Figure 6 in Kadakia, et al. J Exp Med 2019).

We provide new data to address the possibility that we were failing to detect low levels of CCR7 expression on early post-positive selection DPs (CD3loCD69+). We agree that CCR7 immunostaining of mouse cells is known to be more challenging than immunostaining of other chemokine receptors, including CCR4 and CXCR4. CCR7 immunostaining needs to be carried out at 37°C, which we did throughout our studies. We provide new data comparing CCR7 expression by Ccr7+/+ versus Ccr7-/- thymocyte subsets (Figure 1—figure supplement 2A-B), which confirm that CCR7 is not expressed at detectable levels by CD3loCD69+ DP cells above the background seen in CCR7-deficient cells. As thymocytes transition to theCD4SP SM stage, low/intermediate to high expression of CCR7 can be detected (Figure 1—figure supplement 2A). To further test whether we were failing to detect low levels of CCR7 by post-positive selection DPs, we incubated thymocytes at 37°C for up to 2 hours prior to immunostaining for CCR4 and CCR7, as a prior study indicated in vitro culture would enable increased cell surface expression of CCR7 by alleviating ligand-mediated CCR7 internalization (Britschgi et al., 2008). However, we did not observe increased CCR7 (or CCR4) expression by any thymocyte subset incubated at 37°C (Figure 1—figure supplement 2C-D). Lack of expression of CCR7 by CD3loCD69+ DP cells is consistent with their failure to undergo chemotaxis to CCR7 ligands in vitro, and initial expression of CCR7 by CD4SP SM is consistent with their chemotaxis towards CCR7 ligands in vitro (now show in greater detail in Figure 2—figure supplement 1), albeit at a much lower migration index than subsequent thymocyte subsets.

Comment 3: Low levels of CCR7 expression could be functionally evaluated by the chemotactic assay as shown in Figure 2. However, the data in Figure 2 are unequally interpreted for CCR4 and CCR7; CCR4 assays are sensitive where a migration index at less than 1.5 is described as positive (Figure 2A and 2B), whereas CCR7 assays are dismissal to such a small migration index and are only judged positive when the migration index exceeds 10 or 20 (Figure 2C and 2D). CCR7 chemotaxis assays should be carried out more sensitively, to equivalently evaluate the chemotactic function of CCR4 and CCR7 during thymocyte development.

We thank the reviewer for his insight about the possibility that we could have overlooked CCR7-mediated chemotaxis at lower migration indexes. When data from the chemotaxis assays were evaluated separately for each thymocyte subset, CCR7-mediated chemotaxis of CD4SP SM and subsequent DP CD3+CD69+ co-receptor reversing thymocytes could be detected. However, DP CD3loCD69+ thymocytes still did not undergo CCR7-meidated chemotaxis, but were responsive to the CCR4 ligand CCL22 (Figure 2—figure supplement 1).

We did not detect CCR7-mediated chemotaxis of CD4SP SM and DP CD3+CD69+ subsets in our previous analysis because their lower-level chemotactic index relative to mature thymocytes did not reach statistical significance when chemotaxis of all subsets were compared simultaneously (Figure 2D). We note that the magnitude of difference in the responsiveness of CD4SP SM cells compared to mature CD4SP and CD8SP M1 & M2 thymocytes (Figure 2D) is likely physiologically important as CCR7 deficiency results in severely reduced medullary accumulation of CD4SP M1+M2 cells, but only a very mild reduction in medullary accumulation of CD4SP SM cells, which is only detected with our new paired analyses in Figure 5C. We feel these new analyses provide important new insights and thank the reviewer for this suggestion.

Comment 4: Together, this manuscript suffers from the poor sensitivity for CCR7 detection both in flow cytometric analysis and chemotactic functional analysis. Conclusions that CCR7 is absent at early stages of DP to SP thymocyte development and that CCR7 is uncoupled from medullary entry are the overinterpretation of those results with the poor sensitivity for CCR7. The oversimplified scheme in Figure 3D is misleading.

We agree that the scheme in Figure 3D, as previously constructed, did not ideally display the difference in scale between thymocyte responses to CCR7 ligands versus CCR4 and CXCR4 ligands (as detected in vitro). Thus, we have now modified the schematic to include the mild response to CCR7 ligands that we observed in CD4SP SM thymocytes (comment 3) and to emphasize the higher chemotactic response of mature thymocytes to CCR7 ligands than of DPs and CD4SP SM to CCR4 ligands. Likewise, we have modified the manuscript to clarify the importance of CCR7 expression in the medullary entry and accumulation of mature thymocyte subsets.

We respectfully disagree that the sensitivity of CCR7 detection was poor in our flow cytometry and chemotactic analyses. Our CCR7 stains identified a range of CCR7 expression levels, from no expression by pre- and post-positive positive selection DP cells to high expression by CD4SP M1 cells, and we now provide new data confirming our ability to detect CCR7 expression (Figure 1—figure supplement 2), as described in response to Comment 3. Our chemotaxis assays detected CCR7 responses over a range of migration indexes from ~ 2 up to 100, showing our sensitive ability to detect CCR7-mediated chemotaxis in vitro (Figure 2 and Figure 2—figure supplement 1). In live thymic slices, we were also able to capture a range of biologic activities of CCR7, from mediating modest medullary accumulation of CD4SP SM cells to robust medullary accumulation of CD4SP M1+M2 cells (Figure 5A-C). Importantly, our results demonstrate that CCR7 is not the only chemokine receptor responsible for medullary entry and accumulation of thymocytes. Complex spatiotemporal regulation of thymocytes at distinct stages of development is achieved through tight orchestration of expression and signaling through multiple chemokine receptors, including CCR4, as shown by our data. However, our study does not negate an important role for CCR7 in mediating medullary entry of thymocytes, which we have clarified in the text.

Comment 5: The short-term thymus slice culture experiments should be described more carefully in terms of selection events during DP to SP thymocyte development, which takes at least 2 days for CD4 lineage T cells and approximately 4 days for CD8 lineage T cells (Saini, et al. Sci Signal 2010 and Kimura, et al. Nat Immunol 2016). The slice culture experiments in this manuscript examined cellular localization within 12 hours and chemokine receptor expression within 24 hours (Figures 4, 5) even for the development of CD8 lineage T cells (Figure S2), which are too short to examine entire events during DP to SP thymocyte development and are designed to only detect early phase events of thymocyte selection.

Experiments in Figures 4 and 5 were indeed designed to capture behaviors of thymocytes relatively early after introduction onto thymic slices. Figure 4 (and Figure 4—figure supplement 1) shows that the timing of CCR4 versus CCR7 expression after positive selection is dramatically different: CCR4 is expressed within hours of positive selection, concomitant with medullary entry, while CCR7 expression takes several days in the slices (sufficient time for CD8SP development, Figure 4—figure supplement 1). Figure 5 shows that medullary accumulation of CD4SP M1+M2 cells occurs robustly in the medulla of thymic slices within a couple of hours after introduction into the slices, and this localization is CCR7 dependent, while CCR4 induces more mild medullary accumulation of post-positive selection DPs. As indicated by the reviewer, it has been shown that it takes days for DP thymocytes to develop into mature CD4SP and CD8SP cells (Kimura et al., 2016; Lutes et al., 2021; Saini et al., 2010), as recapitulated in the thymus slice system (Figure 4—figure supplement 1) (Lutes et al., 2021). The relatively short time frame of our time-course experiments (up to 12 hours after addition of pre-positive selection thymocytes to positively selecting thymic slices) allowed us to detect expression of CCR4 within a few hours after positive selection and to determine that this timing correlated with medullary entry. Thus, the 12-hour time-course was important for temporal resolution of chemokine receptor expression and medullary localization after initial stages of positive selection.

Comment 6: It is unclear what the medullary density alteration measured in the thymus slice culture experiments represents. Although the manuscript describes that the increase in the medullary density reflects the entry of cortical thymocytes to the medulla (Figure 4E and S2E), this medullary density can be affected by other mechanisms, including different survival of the cells seeded on the top of different thymus microenvironments. Thymocytes seeded on the medulla may be more resistant to cell death than thymocytes seeded in the cortex, for example, because of the rich supply of cytokines by the medullary cells. So, the detected alterations in the medullary density may be affected by the differential survival of thymocytes seeded in the cortex and the medulla. Also, the medullary density is measured only within a short period of up to 12 hours. The use of MHC-II-negative slices and CCR4- or CCR7-deficient thymocytes in the thymus slice cultures may verify whether the detected alteration in the medullary density is dependent on TCR-initiated and chemokine-dependent cortex-to-medulla migration.

We thank the reviewer for pointing out these possibilities. The purpose of the positive selection timing experiment (Figure 4) was to establish the early correlation between receiving a positive selection signal, upregulating CCR4, and migrating into the medulla. In this system, cells only enter only the cortex in the first hour after migration in the slice, consistent with prior studies of localization of pre-positive selection thymocytes to the cortex (Ehrlich et al., 2009; Ross et al., 2014); subsequently, they move into the medulla. Because CCR7 is widely accepted to be essential for medullary entry, we feel it is important to demonstrate the disconnect between the timing of medullary entry and CCR7 expression in multiple ways. The timing experiment design utilized MHCII-/- and β2m-/- slices to show that positive selection was necessary for expression of CCR4. To test whether CCR4 or CCR7 were required for medullary entry of early post-positive selection DPs, we evaluated medullary accumulation of this subset from WT, Ccr4-/-, Ccr7-/-, and Ccr4-/-Cc7-/- mice. This experiment provided a more robust means of determining the extent to which CCR4 deficiency impacted medullary localization of a large cohort of cells that had passed positive selection (Figure 5), and again showed that the post-positive selection thymocytes, which express CCR4 but not CCR7, accumulate in the medulla in a CCR4-dependent manner. We note that in Figure 5, we show that all Ccr4-/-Ccr7-/- thymocyte subsets imaged have medullary:cortical density ratios of ~1, indicating an even distribution across cortex and medulla, which is highly consistent with an essential role for these two chemokine receptors in cooperating to mediate medullary accumulation of different stages of developing T cells.

The reviewer makes an interesting point that survival cues could differ in the cortex versus medulla. However, if thymocytes lacking one or both chemokine receptors had impaired survival because they didn’t enter a region of the thymus efficiently to receive survival cues, we would expect to detect increased apoptosis in Ccr4-/-, Ccr7-/-and Ccr4-/-Cc7-/- thymocytes. However, we found that chemokine receptor deficiencies resulted in diminished apoptosis of different thymocyte subsets (Figure 6). This finding is more consistent with reduced negative selection of these subsets due to reduced clonal deletion. We nonetheless discuss this possibility in our revised manuscript, as it important to consider that chemokine-mediated migration of thymocytes into different microenvironments could alter their access cytokines and other pro-survival cues.

Reviewer #3 (Public Review)

In this manuscript, Li et al. examine how the expression of the chemokine receptor CCR4 impacts the movement of thymocytes within the thymus. It is currently known that the chemokine receptor CCR7 is important for developing thymocytes to migrate from the cortical region into the medullary region and CCR7 expression is therefore often used to define medullary localization. This is important because key developmental outcomes, like enforcing tolerance to self-antigens amongst others, occur in the medullary environment. The authors demonstrate that the chemokine receptor CCR4 is induced on thymocytes prior to expression of CCR7 and thymocytes exhibit responsiveness to CCR4 ligands earlier in development. Using elegant live confocal microscopy experiments, the authors demonstrate that CCR4 expression is important for the entry and accumulation of specific thymocyte subsets while CCR7 expression is needed for the accumulation of more mature thymocyte subsets. The use of cells deficient in both CCR4 and CCR7 and competitive migration/accumulation experiments provide strong support for this conclusion. The elimination of CCR4 expression results in decreases in apoptosis of thymocyte subsets that have been signalled through their antigen receptor and are responsive to CCR4 ligands. As expected, more mature thymocyte subsets show decreased apoptosis when CCR7 is absent. Distinct antigen-presenting cells in the thymus express CCR4 ligands supporting a model where CCR4 expressing thymocytes can interact with thymic antigen-presenting cells for induction of apoptosis. The absence of CCR4 results in an increase in peripheral T cells that can respond to self-antigens presented by LPS-activated antigen-presenting cells providing further support for the model. Collectively, the manuscript convincingly demonstrates a previously unappreciated role for CCR4 in directing a subset of thymocytes to the medulla.

We thank the reviewer for appreciating the novelty of the finding that CCR4 directs distinct subsets of thymocytes into the medulla relative to CCR7, as supported by multiple lines of evidence.

Author Response

Reviewer #1 (Public Review):

The sustainability of vaccination programs is subject to multiple threats, from a pandemic like COVID-19 to political changes. The present study assesses different strategies, including gender-neutral vaccination, to better respond to threats in HPV national immunization programs. The authors showed that vaccinating boys against HPV (compared to vaccinating girls alone), would not only prevent more cases of cervical cancer but also limit the impact of disruptions in the program. Moreover, it would help attain the goal set by the World Health Organization of eliminating cervical cancer as a public health problem sooner, even in the case of disruptions.

Strengths and weaknesses: I found the manuscript well-written and easy to read. Decision-makers may find the results helpful in policy development and other researchers may use the study as an example to investigate similar scenarios in their local contexts. Nevertheless, there are some limitations. First, it should be considered that the present study is only applicable to India and other countries with a similar HPV context. Second, because it is a study based on a mathematical model, errors might arise from the assumptions considered for its construction. It also relies on the quality of the data used to construct and calibrate the model.

Models are important tools for decision-making, they allow us to assess different scenarios when obtaining real-world data is not feasible. They also allow to carried-out multiple sensitivity analyses to test the strengths of the results. The study carries out a necessary assessment of different vaccination strategies to minimize the impact on cervical cancer prevention due to disruptions in the HPV immunization program. By using a mathematical model, the authors are able to assess different scenarios regarding vaccination coverage rates, disruption time, and cervical cancer incidence. Therefore, decision-makers can consider the scenario which best represents their current situation.

The present study is not only valuable for decision-making, but also from a methodological point of view as future research can be conducted exploring more in deep the impact of vaccination disruptions and prevention measures.

The conclusions of this paper are mostly well supported by data, but some aspects of the methodology need clarification; furthermore, some aspects of the calculations can be improved. It would be more informative, and better for comparisons between the four scenarios, to have relative measures instead of the absolute numbers of cases prevented.

We thank the reviewer for the kind acknowledgement of the merits of the paper. We have tried to address the suggestions and questions as much as possible in the revised manuscript.

We agree to the points of weaknesses raised by the reviewer regarding the applicability of our study results is limited to other countries and the possible errors arising from a using a mathematical model. We have added more elaborate discussion of these points in the manuscript, as follows: - Page 15 lines 310-312: “Extrapolation of the results of this study to other populations will be limited to those sharing similar patterns of demography, social norms, and cervical cancer epidemiology as India.” - Page 17 lines 361-363: “…, within the limitations of our model, the modelbased estimates show that shifting from GO to GN vaccination may improve the resilience of the Indian HPV vaccination programme while also enhancing progress towards the elimination of cervical cancer.”

Furthermore, we have tried to clarify the rationale, advantages, and limitations of the measure of resilience we have adopted.

Reviewer #2 (Public Review):

This study evaluated the effect of population-based HPV vaccination programs in India which is suffering from the disease burden of cervical cancer. The authors used model simulations for estimating the outcomes by adopting the latest available data in the literature. The findings provide evidence-based support for policymakers to devise efficient strategies to reduce the impacts of cervical cancer in the country.

Strengths.

The study investigated the potential impact of cervical cancer elimination when HPV vaccination was disrupted (e.g., during the COVID-19 pandemic) and for meeting the WHO's initiatives. The authors considered several settings from the low to high effects of vaccination disruption when concluding the findings. The natural history was calibrated to local-specific epidemiological data which helps highlight the validity of the estimation.

Weaknesses.

Despite the importance and strengths, the current study may likely be improved in several directions. First, the study considered the scenario of using a recently developed domestic HPV vaccine but assuming vaccine efficacy based on another foreign HPV vaccine that has been developed and used (overseas) for more than 10 years. More information should be provided to support this important setting.

Second, the authors are advised to discuss the vaccine acceptability and particularly the feasibility to achieve high coverage scenarios in relatively conservative countries where HPV vaccines aim to prevent sexually transmitted infection. Third, as the authors highlighted, the health economics of gender-neutral strategies, which is currently missing in the manuscript, would be a substantial consideration for policymakers to implement a national, population-based vaccination program.

We thank the reviewer for the kind acknowledgement of the merits and strengths of the paper.

We have tried to address the reviewer’s three points of weaknesses as comprehensively as possible in the revised manuscript.

Regarding the first two points of weaknesses, we have provided more background information about the current situation of HPV introduction and screening in India (see the more specific replies below for where changes have been made), and some data of observed coverage in India in the states where HPV vaccination has been introduced.

Regarding the reviewer’s third point about the health economics of genderneutral strategies, we agree fully that it is an important aspect to consider for the local policymakers. However, a health economic assessment is out of the scope of the present paper. In the present paper, we are interested in highlighting the potential health benefits on GN HPV vaccination. Given the current context of HPV vaccination in India we think it is too early to provide a realistic assessment of the health-economic balance of GN vaccination. Please note that one manuscript (de Carvalho et al., MedRxiv, doi: https://doi.org/10.1101/2023.04.14.23288563) based on the same modelling exercise and reporting a health economic assessment of girls-only (routine and catch-up) HPV vaccination in India is currently submitted for peer-review.

Reviewer #3 (Public Review):

The authors put together a rigorous study to model the impact of HPV vaccine programme disruptions on cervical cancer incidence and meeting WHO elimination goals in a low-income country - using India as an example. The study explores possible scenarios by varying HPV vaccination strategies for 10-year-old children between a) increasing vaccine coverage in a girls-only vaccination programme and b) vaccinating boys in addition to girls (i.e a gender-neutral vaccination programme).

The main strength of this study is the strength of the modelling methodology in helping to make predictions and in contingency planning. The study methodology is rigorous and uses models that have been validated in other settings. The study employs a high level of detail in calibrating and adapting the model to the Indian context despite poor data availability. The detailed methodology allows future studies to employ the model and techniques with locally-contextualised parameters to study the potential impact of HPV vaccine programme disruptions in other countries.

The work in this field can begin to help lower-income countries explore varying HPV vaccination strategies to reduce cervical cancer incidence, keeping in mind the potential for future supply chains or other related disruptions. However, the scenarios could be better sculpted to model potentially realistic scenarios to guide policymakers to make decisions in situations with limited vaccine supplies - in other words comparing scenario alternatives based on a fixed number of vaccines being available. Using comparative alternatives will help policymakers grapple with the decisions that need to be made regarding planning national HPV vaccination programmes. The results could afford to provide readers with a clearer measure of vaccine strategy 'resilience'.

In all, the authors are able to successfully explore the potential impact of varying HPV vaccination strategies on cervical cancer cases prevented in the context of vaccine disruptions, and make valid conclusions. The results produced are rich in information and are worthy of deeper discussion.

We thank the reviewer for the kind acknowledgement of the merits and strengths of the paper.

Author Response

Reviewer #3 (Public Review):

The strongest aspects of this study are the structural analysis of the 90 residue KER domain. This is an important advance, discovering a founding member of a novel class of DNA binding motifs, termed a SAH-DBD (single alpha helix-DNA binding domain). Interestingly, they define a subregion of KER (termed "middle-A", residues 155-204 of Cac1) that has nearly the same DNA binding affinity and confers similar in vivo phenotypes as the full KER domain.

This study also shows that the biological role of KER partially overlaps compensatory factors in vivo, both within the same Cac1 protein subunit (e.g. the WHD domain) and also with other proteins acting in parallel (e.g. Rtt106). That is, the presence of either WHD or Rtt106 renders the drug-resistance and silencing assays employed here insensitive to loss of the KER domain.

However, the drug resistance and gene silencing phenotypes are inherently indirect measures of the most important claim of this work, that KER is a molecular ruler for DNA for the purpose of ensuring sufficiently large templates deposition of histone H3/H4 cargoes. Therefore, this study would be of greater impact if the authors more directly tested this measurement idea in assays that directly assess histone deposition. There are multiple options. Since the authors have in hand recombinant wild-type and mutant CAF-1 complexes, one could examine the number and/or spacing of nucleosomes formed during in vitro deposition reactions. Complementary in vivo experiments using the authors' existing mutant strains could be based on the finding that CAF-1 is particularly important for histone deposition onto nascent Okazaki fragments during DNA replication (Smith and Whitehouse, 2012; pmid: 22419157), and that the spacing pattern of nucleosomes on this DNA is greatly perturbed in cac1-delete cells.

Thank you for the suggestion of approaches to obtain data that more directly addresses changes in nucleosome assembly due to CAF-1 KER mutants. We considered using an in vitro nucleosome assembly assay, such as the reconstitution of nucleosomes onto gapped DNA using purified components developed by Kadyrova et al., 2013 (doi: 10.4161/cc.26310). However, they found defects only in the amount of nucleosome assembly and not changes in nucleosome spacing without CAF-1. In addition, we didn’t have the system set up and knew that it would be unlikely to produce data in the time needed for a revision of the manuscript, or even show spacing changes in nucleosomes at all. Therefore, we chose an assay system in yeast that already has been used to assess the impact of CAF-1 DNA binding mutants on nucleosome assembly (Smith and Whitehouse, 2012; pmid: 22419157 and Mattiroli et al., 2017 doi: 10.7554/eLife.22799). This approach, developed by Smith and Whitehouse, uses a degradable Ligase I system in yeast, which reveals Okazaki fragment lengths, and shows a defect when CAF-1 activity is knocked out (Smith and Whitehouse, 2012). This assay also showed that mutations or deletions in the Cac1 WHD DNA binding domain, led to increased lengths of Okazaki fragments (Mattiroli et al., 2017). As the WHD DBD impacts Okazaki fragment lengths, we reasoned that mutations in the KER DBD might also.

We generated numerous new yeast strains that included the degradable Ligase I system and collaborated with Dr. Duncan Smith of (Smith and Whitehouse, 2012; pmid: 22419157) to detect nascent Okazaki fragments in various CAC1 mutants in strains that were RTT106 or rtt106∆. We found that the Okazaki fragment lengths from cac1∆ yeast were larger and less discrete than from CAC1 yeast (as Dr Smith published previously) and that the Okazaki fragments from the cac1∆ rtt106∆ strain were barely detectable, presumably because they were too long to be resolved on the gel. However, the assay did not have sufficient resolution to detect changes between the Okazaki fragment length distribution between wild type CAC1 or the ∆KER, ∆middle-A and 2xKER mutants of CAC1, in either the RTT106 or rtt106∆ background. Therefore, we were unable to detect direct effects of the KER mutants on Okazaki-fragment lengths. We considered using the combination of KER mutants with the WHD mutants, but as this would not directly assess the effects of the KER mutants and CAF-1 proteins lacking the KER and the WHD don’t bind to DNA (Figure 3 in Mattiroli et al., 2017), we didn’t pursue it. As the complete deletion of the KER, shortening of the KER and lengthening of the KER did not give detectable changes in this assay, we also did not pursue the other mutants tested in the manuscript. Although, we are disappointed the experiment did not reveal effects that we had hoped for, this experiment provides support for the redundant functions of CAF-1 and Rtt106 in nucleosome assembly, which has not been shown using this assay. As such, we have added Figure 1-figure supplement 1g and text to the results section, methods section and strain table. We have included Prof. Duncan Smith and his student Anne Seck as authors.

Added text lines 195 to 207: “Finally, to assess the impact of deleting the KER more directly on nucleosome assembly in vivo, we examined histone deposition onto nascent Okazaki fragments during DNA replication as we have shown previously that the length of Okazaki fragment lengths are determined by histone deposition into nucleosomes and is disrupted upon deletion of CAC1 (Smith and Whitehouse, 2012). We compared CAF-1 mutants in the WT yeast background and in yeast lacking Rtt106. We found that the Okazaki fragment length distributions of the ∆KER mutant was indistinguishable from that of WT while that of cac1∆ was disrupted (Figure 1-figure supplement Figure 1-figure supplement 3g). That we did not detect effects on Okazaki-fragment lengths for the yCAF-1 mutants lacking the intact KER is consistent with the results of the viability and silencing assays for KER mutants, which also retained the WHD. Strikingly, the Okazaki fragments from rtt106∆ cac1∆ yeast were highly disrupted (Figure 1-figure supplement Figure 1-figure supplement 3g) further highlighting the redundancy between Rtt106 and Cac1 for assembling histones onto newly replicated DNA. Therefore, t”

Author Response

Reviewer #3 (Public Review):

The authors investigated the mechanism of transport of the GLUT5 sugar porter using enhanced sampling molecular dynamics simulations and biochemical analysis. The results suggest a possible general mechanism by which binding to a transported substrate stabilizes an occluded intermediate conformation between outward and inward-facing states of the alternating access conformational change of the protein, thereby enabling transport.

The authors also identified key elements of this transition, associated with residues involved in sugar binding, and through elegant biochemical experiments demonstrated how mutations of the latter affect the protein function, including mutations of gating residues that can recover the function of inactive mutants.

The general computational methodology used by authors is appropriate for addressing these questions and compared to other techniques has the advantage of bringing forth an unbiased molecular description of the transport process. The results are overall qualitatively in line with the proposed conclusions.

A major weakness of this work is that, in contrast to previous studies with the same type of methodology, the authors do not report error analysis or careful statistical assessment of the computational results. Therefore, it is not clear whether the latter is solid or if they support the proposed conclusions. The computational data might generally benefit from an improved methodological design, such as including more degrees of freedom (or collective variables) in the description of the minimum free energy pathway, e.g. the salt-bridges.

This has now been addressed in the essential revisions above.

Another weakness is that some of the details of the computational analysis are not reported, therefore other investigators would not know how to reproduce the results.

We have extended the methods section to include much more detail about the MSM construction and other computational analysis. Data files needed for reproduction are now found in a public repository with links provided in the Methods section.

Author Response

Reviewer #1 (Public Review):

This manuscript presents an inference technique for estimating causal dependence between pairs of neurons when the population is driven by optogenetic stimulation. The key issue is how to mitigate spurious correlations between unconnected neurons that can arise due to polysynaptic and other network-level effects during stimulation. The authors propose to leverage each neuron's refractory period (which begins at approximately random times, assuming Poisson-distributed spikes and conditional on network state) as an instrumental variable, allowing the authors to tease apart causal dependence by considering how the postsynaptic neuron fires when the presynaptic neuron must be muted (i.e., is in its refractory period). The idea is interesting and novel, and the authors show that their modified instrumental variable method outperforms similar approaches.

We wish to thank the reviewer for this positive assessment.

However, the scope of the technique is limited. The authors' results suggest that the proposed technique may not be practical because it requires considerable amounts of data (more than 10^6 trials for just 200 neurons, resulting in stimulation of more than 5000 times per neuron). Even with such data sizes, the method does not appear to converge to the true solution in simulations. The method is also not tested on any experimental data, making it difficult to judge how well the assumptions of the technique would be met in real use-cases. While the manuscript offers a unique solution to inferring causal dependence, its applicability for experimental data has not yet been convincingly demonstrated, and would, therefore primarily be of interest to those looking to build on these theoretical results for further method development.

We thank the reviewer for this assessment and agree that the requirement for this many trials makes the estimators practically unsuitable for identifying causal interactions in large systems. However, in the revised manuscript, we can observe that the IV estimator can be beneficial after even a few thousand trials when introducing a newly improved error measurement (which we discovered thanks to these reviews). Moreover, we agree that this work will be of interest to the more theoretically oriented community for methodological improvements; we believe that the methods and causal inference framework will be interesting and useful for the wider neuroscience community. For example, considering the first (new) example in the introduction, even under two-photon single-neuron stimulation, the IV framework should be used to avoid bias amplification.

Reviewer #2 (Public Review):

Lepperød et al. consider the problem of inferring the causal effect of a single neuron's activity on its downstream population. While modern methods can perturb neuronal activity, the authors focus on the issue of confounding that arises when attempting to infer the causal influence of a single neuron while stimulating many neurons together. The authors adapt two basic methods from econometrics that were developed to address causal inference in purely observational data: instrumental variables and difference-in-differences, both of which help correct for unobserved correlations that confound causal inference. The authors propose an experimental procedure where neurons have spike times measured with millisecond precision and a subset of neurons are optogenetically activated. As an instrumental variable, the authors propose using the refractoriness of a stimulated neuron, resulting in absent or delayed spiking which can be used to infer its causal effect in otherwise matched conditions.

Based on this, they develop a collection of estimators to measure the pairwise causal relationship of one neuron on another. By simulating a variety of small networks, the authors show that, provided enough data is present, the proposed causal methods provide estimates that better match underlying connectivity than methods based on ordinary least squares or naive cross-correlograms (CCHs). However, the methods proposed require extensive data and highly targeted stimulation to converge.

Strengths:

The value of the paper comes from its attempt to find neuroscience applications for methods from fields where causal analysis of observational data is required. Moreover, as the field develops improved methods of measuring anatomical neuronal connectivity using molecular, physiological, and structural approaches, the question of the causal influence of one neuron's spiking on another remains vital. The authors thoughtfully lay out the necessary conditions - and difficulties - required to establish this type of causal functional influence and suggest one potential approach. The collection of models tested highlighted both the strengths and difficulties of the suggested approaches.

We wish to thank the reviewer for the positive feedback, we are delighted to share your view that obtaining methodology for estimating causal influence is vital.

Weaknesses:

1) I found the paper's introduction to its analysis techniques to be very confusingly written, particularly as it is designed to bridge fields. It is vital that the ideas are communicated more clearly. Some topics are explained multiple times, even after being used previously, other ideas and notations are introduced and immediately dropped (e.g. the "do operator", the ratio of covariances in the introduction to instrumental variables), and still others are introduced with no clear explanation (e.g. the weight term w, the "|Y->Y-Y*" notation, and the notation in the methods with "Y(Z=0)").

We thank the reviewer to point out this lack of clarity and we extensively rewrote the paper to make it more accessible. The do operator is used in the methods to define Y(Z=0), but is now removed from the introduction to reduce the number of concepts introduced early in the text. The w term is now defined from the generative model. The difference in differences notation is written out fully to be clear and a sketch of the method intuition is added to Figure 1.

1) Of particular importance, the introduction of the Z,X, and Y variables in the first full paragraph on page five, it could be made much more clear that this method is pairwise: Z and X reference the spiking of one specific stimulated neuron at two time points and Y references one specific downstream neuron. 2) In the third paragraph of the same page, the authors refer to the "refractoriness of X" and "spiking of X onto Y", but this language confuses the neurons with variables in a way that took considerable time to unpack. 3) This was not helped by Figure 1b, which suggested that Z_i, X_i, and Y_i applied to all neurons and merely reflected time points around stimulation. 4) Similarly, the introduction of the Y* variable in the difference of differences method, which the authors view as one of the main contributions, is given little clear explanation or intuition. I assume "shifted on window-size left" means measuring the presence of spiking at the same time step as X, but I see no clear definition of this. 5) The confusion about variables remains when, in Figure 1d, a "transmission probability" goes below 0 and above 1.

1) Thank you for pointing out this lack of clarity, the suggested explanation of the variables XYZ is adopted.

2) The language is clarified such that variables and neurons are separated.

3) Figure is fixed such that variables refer to the neurons they represent.

4) We have now improved the explanation of DiD with a figure for intuition.

5) We have now redefined the “transmission probability” to effective connectivity to reduce confusion.

I also found the network models studied after the first section and the relevant variables difficult to understand with the detail necessary to interpret the results. For example, the cartoon in Figure 2a does not seem to match the text description. I see no explanation for the external "excitatory confounder" and "inhibitory confounder" terms, nor what is done to control the (undefined) \sigma_max/\sigma_min term. I don't see anything in the methods about distinct inhibitory and excitatory neurons either. Further, the violin plots (e.g. Fig 2d) seem quite noisy (e.g. is Br, DiD really bimodal?), and it is not clear what distribution is being covered by them. If this is computational simulations, I would imagine more samples could be generated. The same vagueness issues hold for the networks in section 2.4 and 2.7.

We have now clarified the implementation of the excitatory and inhibitory confounder and how we distinguish between excitatory and inhibitory neurons and defined the condition number. The violin plots were removed in Fig 2 since the large variance represented changes across external drive which produced largely incomparable statistics. To illustrate variance, we now show the standard deviation of the absolute error in line plots 2e and 2g.

2) Broadly speaking, the causal estimates appear better in the sense of having smaller errors, but it's not clear to me if they are actually good or not. What does an error of 0.4 mean in terms of your ability to estimate the causal structure, and what exactly does the Error(w{greater than or equal to}0) notion refer to? It would be useful to see actual reconstructions of ground truth versus causally inferred connectivity to better understand the method's strengths and weaknesses.

To improve clarity, we have added a paragraph in the text before figure 2 explaining a new error measure. Since the estimators give the transmission probability and not the inferred connection strength directly, we previously computed a regressed error as in Das & Fiete 2020. This error measure is equivalent to the sine of the angle between $W$ and $\hat{W}$. This error measure is not ideal and gives an indirect population measure with deviations scaled during the error regression. Upon further reflection, we realized that we could define the error directly using our definition of effective connectivity on the generative model to obtain a much cleaner and more interpretable measure. This further led us to remove one of the proposed methods (brew) as it did not perform well under this new error measure. All error measurements are updated in all figures. Error(w{greater than or equal to}0) means that we only look at positive weights; now clarified in the text

3) I found the section on optogenetic modeling to be unsatisfying in its realism. The general result that 1 photon excitation hits a wide collection of neurons is undisputed, but the simulation does not account for a number of key factors - optogenetic receptor expression is distributed across the axons and dendrites of a cell, not only soma, scattering in tissue greatly affects transmission, etc. Moreover, experiments that attempt to do highly targeted activation have other methods for exactly this reason, such as multiphoton activation or electrophysiology. The message of decreasing performance as a function of stimulus size is important, but I struggle with the idea of the model being "realistic".

We thank the reviewer for pointing out this unsatisfactory comparison with realistic scenarios. To mitigate we have changed the wording, but kept the simulation as is. As the reviewer pointed out optogenetic receptor expression is distributed, and here we have assumed an expression that only affects soma (experimentally plausible according to Grødem et al 2023 (10.1038/s41467-023-36324-3)), scattering in tissue is included according to the Kubelka-Munk model.

4) The authors spend a great deal of analysis of stimulation, but little time on measurement. It seems like this approach demands a highly precise measure of spike time to know if a neuron is firing or not at a given millisecond due specifically being in a refractory state. A stimulated but refractory neuron will still likely spike as soon as it can after the momentary delay, and given the noise in the network this difference might not be easily detectable in the delay-to-spike of the downstream neuron, even assuming one spike in the presynaptic neuron is likely to cause a spike in the downstream. It would be useful to see this aspect considered with the same detail as the rest of the study.

We thank the reviewer for pointing out this. We have now added a paragraph discussing this: “As outlined in \citep{ozturk2000ill}, ill-conditioning can affect statistical analysis in three ways and therefore similarly in inverse connectivity estimates from measured activity. First, measurement errors such as a temporal shift in spike time estimate e.g. due to low sampling frequency, inaccurate spike sorting, or general noisy measurement due to animal movement etc. In the presence of ill-conditioning the outputs will be sensitive (unstable) to small input changes. If errors are included in some variables, the inference procedures will require information about the distributional properties of these errors. Second, optimized inference can give misleading results in the presence of ill-conditioning, caused by bad design or sampling.

There will always exist a natural variability in the observations which necessitates the assessment of ill-conditioning before performing statistical analysis. Third, rounding errors can lead to small changes in input under ill-conditioning. This numerical problem is often not considered in neuroscience but will become evermore relevant when large-scale recordings require large-scale inferences.”

Author Response:

We would like to thank the reviewers for their thorough evaluation of the presented manuscript and herewith would like to address their comments and suggestions.

This study was funded by a NSF-grant awarded to Prof. Celio. The animal experimentation license (including animal husbandry, breeding and experiments) that is required by law to perform animal experiments was also issued to Prof. Celio. Therefore, with the retirement of Prof. Celio, the funding for the project was discontinued and the animal license was terminated. We are thus unable to answer the reviewers’ open questions with follow-up experiments. We would however like to discuss some of the reviewers’ open questions or concerns and hope this might be insightful to the interested reader.

Reviewer #1 (Public Review):

“First, they reported that chemogenetic activation of Foxb1 hypothalamic cell groups led to tachypnea. The authors tend to attribute this effect to the activation of hM3Dq expressed in the parvofox Foxb1 but did not rule out the participation of the PMd Foxb1 cell group which may as well have expressed hM3Dq, particularly considering the large volume (200 nl) of the viral construct injected. It is also noteworthy that the activation of the Foxb1hypothalamic cell groups in this experiment did not alter the gross locomotor activity, such as time spent immobile state.”

Because an AAV2 serotype was used for expression of the chemogenetic tools, the spread of viral infection was much more restricted to the injection site in chemogenetic animals than was observed the AAV5-based expression of optogenetic tools. The more restricted spread of viral infection with AAV2 serotypes has previously been shown by a range of other groups (e.g. see https://doi.org/10.3389/fnana.2019.00093). This limited spread of the AAV2 serotype in our chemogenetic animals, together with the absence of the very strong locomotor phenotype observed during optogenetic stimulation experiments makes us hypothesize, that the respiratory phenotype is largely attributable to the ParvafoxFoxb1 neurons.

“In the second experiment, the authors applied optogenetic ChR2-mediated excitation of the Foxb1+ cell bodies' axonal endings in the dlPAG leading to freezing […]. Here it is important to consider that optogenetic ChR2-mediated excitation of the axonal endings is likely to have activated the cell bodies originating these fibers, and one cannot ascertain whether the behavioral effects are related to the activation of the terminals in the PAGdl or the cell bodies originating the projection.”

We did not consider the possibility of backpropagation induced by optogenetic axon terminal stimulation at the time of experiments. We acknowledge that this is the major limitation of our optogenetic experiments that would have to be investigated with further animal experiments.

Reviewer #2 (Public Review):

“3) Fig. 5, a great effort has been made to illustrate the point that CCK and Foxb1 are differentially expressed. Why not just perform a double in situ experiment to directly illustrate the point?”

We came across the publication in which the Cck-expressing PMd neurons’ control escape behaviors, only when we were drafting the manuscript. Because this was already after the retirement of Prof. Celio and we were not able to conduct further experiments involving animals, we leveraged on in silico methods and the publicly available high-quality dataset on the gene expression of the posterior hypothalamic area. The applied in silico method of dimensionality reduction and cluster assignment is well established and widely accepted. We believe in the quality of the dataset and the reliability of these in silico results but we agree with the reviewer that an alternative would have been to illustrate the expression patterns of Cck and Foxb1 by in-situ hybridisation.

“4) Fig. 7 data on optogenetic stimulation on immobility and breathing, since not all mice showed the same phenotype, what is the criterion for allocating these mice to hit or no hit groups?"

We defined the group allocation criteria in the section titled “Optogenetic modulation of Foxb1 terminal in the dlPAG induces immobility” as follows:

“OnTarget_antPAG animals had the tip of the optic fiber implant located above the dlPAG at an anterior-posterior level AP-4.04mm (from bregma) or proxymal. The OffTarget group contains animals with fiber tips located below (i.e. ventral to) the dlPAG and/or located more distal than AP -4.04mm.”

Author Response

Reviewer #1 (Public Review):

The manuscript of Parab et al. reports a beautiful phenotype analysis of the vascular brain/meningeal anatomy in a variety of reporter lines and mutants for Wnt/β-catenin signaling and angiogenic cues (Vegfaa, Vegfab Vegfc, Vegfd) during zebrafish development.<br /> The present study extends the previous work of the same Parab, Quick, and Matsuoka, that focused on fenestrated vessel formation in the zebrafish myelencephalic choroid plexus (mCP). Vegfs were shown to regulate fenestrated vessel formation in combination, but not individually, and with only little effect on neighboring non-fenestrated brain vessel development. The fenestrated endothelium is thus known to have specific angiogenic requirements.

The scale of investigation has now changed, and fenestrated vessel formation has been examined throughout the brain, in both circumventricular organs (organum vasculosum of lamina terminalis) and other choroid plexuses (CPs) including the diencephalic CP and its interface with the pineal gland, the eye choroid (choriocapillaris), and the hypophysis vasculature. The original finding is that a regionspecific code of angiogenic cues controls fenestrated vessel formation. The authors show that fenestrated vessels form independently of Wnt/β-catenin signaling and BBB vascular development but require different combinations of Vegfa and Vegfc/d-dependent angiogenesis within and across brain regions. A previously unappreciated function of autocrine and paracrine Vegfc signaling is demonstrated in this brain region-specific regulation of fenestrated capillary development.

Twenty-one different fish lines accurately genotyped and characterized and including a new Reck mutant, have been instrumental to conduct vascular pattern analysis, using confocal and stereomicroscopy imaging combined with transmission EM. High-quality illustration and robust quantification methods, previously validated, have been used. The study is well organized and reflects the high expertise and strong methodology of the investigators. Data are presented in nine dense figures and the contribution of angiogenic ligands to fenestrated vessel formation can hardly be studied more indepth.

However, and this will be my only main concern, no information is provided on the regional diversity of angiogenic receptor expression that may correlate with the regional angiogenic factor code. Without asking for a spatial transcriptomic study, the combination of Vegfr-reporter lines or in situ hybridization with a combination of receptor probes would allow for generating a comprehensive set of ligand/receptor data relative to the regional angiogenic signaling pattern involved in fenestrated vessel formation.

We appreciate this reviewer’s positive and encouraging comments highlighting both the quality and significance of our study. As we commented in response to the Essential Revisions point #1, we anticipate that a detailed expression analysis of all four Vegf receptors at different developmental stages during CP and CVO vascularization will be best addressed with new technologies combined with optimizations of existing tools/protocols. Thus, we have provided a paragraph of discussion on our perspectives for potential Vegf receptors involved in CP and CVO vascularization in the current study.

We address each of the points raised by the reviewer below.

Reviewer #2 (Public Review):

Building on their previous studies, Parab et al used a larger collection of genetically modified zebrafish lines to map the precise expression domains of different VEGF isoforms in the brain and demonstrated that different combinations of VEGF isoforms differentially control the formation of fenestrated vessels at different locations in the brain.

The authors used three Wnt signaling mutants to convincingly show wnt signaling is essential for parenchymal angiogenesis, but not required for fenestrated vessel development, such as those in choroid plexus, suggesting fenestrated vessel and barrier vessel are differentially regulated. The previous work from this group has established that VEGF isoforms are critical for myelencephalic choroid plexus development. In this study, they carefully documented the developmental vessel patterning in the diencephalic choroid plexus/pineal gland interface. They also documented the local expression pattern of VEGF isoforms with a set of BAC transgenic fish, together with the phenotype of a series of VEGF mutant fish, the data well support that different combinations of VEGF isoforms regulate fenestrated vessel development at different brain locations.

Given a larger temporal and spatial domain, VEGFs are critical for all forms of vessel development, there are potential redundancy mechanisms to maintain hemostasis of VEGF signaling, in this study, no data is provided to address whether LOF of one form of VEGF affects the expression of other isoforms.

This work provided detailed evidence of different isoform combinations of VEGF regulate formation/patterning of the fenestrated vessel at CP, OVLT, and NH in zebrafish. It will be interesting to follow in the mammalian system, how well these findings are conserved, for example, which isoform of VEGF is critical for vascular patterning during the developmental stages of the pineal gland? How VEGF isoforms participate in choroid plexus development at different ventricle regions and subsequence secretory function maintenance. However, these tasks are challenging without a good genetic tool to locally manipulate VEGF isoform expression during mammalian brain vessel development.

We appreciate this reviewer’s favorable and encouraging comments highlighting both the quality and impact of our study. We also acknowledge the great importance of the points raised by the reviewer, including the Vegf redundancy mechanisms and also our results’ conservation in mammals.

Reviewer #3 (Public Review):

Parab et al. investigate the requirement of specific Vegf ligands during the embryonic development of new blood vessels in different brain regions. The authors implement their previously published experimental paradigm (Parab et al 2021 eLife) combined with new transgenic and mutant zebrafish lines to show that vegf ligands (vegfaa, vegfab, vegfc, and vegfd) are required in various combinations to drive angiogenesis in distinct brain regions. Specifically, they show that individual loss of different vegf ligands causes either undetectable or partial effects in angiogenesis, while combined loss of vegf ligands results in severe defects in brain region-specific angiogenesis. As different blood vessel types (i.e. arteries, veins, lymphatics) require specific angiogenic cues, this study provides interesting new data on how the combination of these signals drives brain region-specific vascular development.

While the conclusions of the paper are generally well supported by the data, the authors overstate some of their findings, particularly with respect to the development of fenestrated capillaries. In this study, the authors use the zebrafish transgenic reporter line, plvap:EGFP, as an indicator of fenestrations. However, the authors do not provide any evidence of fenestrations of the blood vessels of the choroid plexuses or the cranial vessels used for quantification (Figures 1, 3, and 4). While expression of Plvap protein is often used as a marker for non-blood brain barrier endothelial cells, as Plvap is the major component of the diaphragms of fenestrated capillaries, plvap:EGFP expression alone does not indicate fenestrations. This is an important point because previous work has demonstrated that targeted deletion of Plvap does not cause a loss of fenestrations, but instead a loss of the diaphragms associated with fenestrations (Stan et al 2012 Dev Cell; Gordon et al 2019 Development). Similarly, Plvap expression alone does not necessarily indicate fenestrations as an expression of Plvap is not sufficient for fenestration formation. In fact, Plvap has initially been expressed in brain endothelial cells during initial angiogenesis to the brain without evidence of fenestrations, and subsequently, Plvap expression disappears during the maturation of the BBB. Thus, to conclude that specific vegf ligands are required for the development of fenestrated capillaries, transmission electron microscopy (TEM) should be used on the capillaries examined in this study or the language describing the results should be modified accordingly. Conversely, the authors did show TEM for the choriocapillaris (Figure 5A-C) but did not show plvap:EGFP expression in these vessels.

Additionally, the authors' usage of the phrase "development of fenestrated vessels" suggests that the study was examining signals that regulate the formation of fenestrations and not angiogenesis of vessels that may become fenestrated as demonstrated here. Therefore, as Plvap expression does not necessarily equate fenestrations (and vice-versa), the title and some of the major claims of the study are somewhat overstated.

We appreciate this reviewer’s constructive comments and suggestions to improve this study. We agree with the reviewer that the descriptions of our findings in the original manuscript were not strictly accurate in some aspects. We have now addressed the concern of the Tg(plvap:EGFP) reporter specificity by conducting additional molecular and functional characterizations of Tg(plvap:EGFP)+ vs Tg(glut1b:mCherry)+ brain vasculature, as we have commented in response to the Essential Revisions point #2. In addition, we have made substantial revisions in describing our findings, including 1) the change of the phrase "development of fenestrated vessels" into a more appropriate phrase and 2) the clarification of the primary focus of this manuscript on “angiogenesis/vascularization”. We believe that our revised manuscript now more clearly conveys the finding of signals involved in angiogenesis/vascularization of CP and CVO vascular beds.

Author Response:

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

We are very glad that the editor and reviewers found our paper of broad interest to the community of population, evolutionary, and ecological genetics. We thank them for their positive feedback and insightful comments and suggestions. We have revised our manuscript to address some of the issues raised by the review. The main change we made was providing a detailed discussion of limitations of simulated genomes, focusing on considerations one needs to make when selecting a demographic model. This can be found in a new section “Limitations of simulated genomes” (pages 9-10). We made a few additional adjustments in other parts of the text based on the reviewers’ suggestions. They are all listed in the detailed point-by-point response to reviewers comments and questions below.

Editor:

1) It was noted that demographic models (or genomic parameters) that are inferred based on certain aspects of the genomic data (eg., site frequency spectrum, haplotype structure) may not recapitulate other aspects of the data. In other words, any inferred demographic models are expected to reliably reproduce only some aspects of the genetic variation data but not necessarily all. It would be helpful to emphasize this limitation in the manuscript and to include a table summarizing the types of variation that the demographic models for the catalogued species were based on.

This is a very important point, which we addressed in the revision by adding a section entitled “Limitations of simulated genomes”. This section discusses the considerations that one should make when selecting an inferred demographic model to implement in simulation. This includes the samples used in analysis, the method used for inference, as well as various filters. In this section we also point to the documentation page of the stdpopsim catalog, which provides information about each demographic model that can help users decide whether it is appropriate for their needs. We decided not to summarize this information in a succinct table in the manuscript because it is not straightforward to summarize the strengths and potential limitations of each model in a table. Instead, we will expand the summary provided for each demographic model in the documentation page to provide additional information. See response to the second reviewer’s comment on this topic for more details.

2) It will make stdpopsim more user-friendly to include an automated module that can visualize a demographic model given the corresponding parameters (or simulation scripts).

As mentioned in the response to the first reviewer’s comment on this subject, the documentation page of the stdpopsim catalog provides a brief summary for each demographic model, including a graphical representation. See response below for more details.

Reviewer #1:

In the introduction, the authors cite numerous efforts to generate high-quality reference genomes. That's not an issue in itself, but leading with this might send the message to some readers that it is these reference genome efforts that are driving the need for population genomics analysis and simulation tools, which is not really the case - why not instead give some citation attention to actual population genomics projects aiming to address the types of evolutionary questions this paper is concerned with? The reference genome citations would fit better in the section dealing with reference genomes, where they already appear.

Indeed, the desire to answer complex evolutionary questions is the main motivation for sequencing these genomes and also for generating realistic genome simulations. The reason we chose to lead with the genome-sequencing efforts is that high quality genome data is an important prerequisite for obtaining parameters for chromosome-scale simulations. So, with that perspective, these efforts which we cite are the driving force behind expansion of stdpopsim in the near future. Thus, we decided to leave these citations in the introduction. To balance things out, we now start the introduction with a statement about board questions in population genetics. Moreover, after we list the genome sequencing efforts, we added a list of specific types of questions that can be addressed by these newly emerging genomes, with relevant citations. The beginning of the introduction now reads:

“Population genetics allows us to answer questions across scales from deep evolutionary time to ongoing ecological dynamics, and dramatic reductions in sequencing costs enable the generation of unprecedented amounts of genomic data that can be used to address these questions (Ellegren, 2014). Ongoing efforts to systematically sequence life on Earth by initiatives such as the Earth Biogenome (Lewin et al., 2022) and its affiliated project networks, such as Vertebrate Genomes (Rhie et al., 2021), 10,000 Plants (Cheng et al., 2018) and others (Darwin Tree of Life Project Consortium, 2022), are providing the backbone for enormous increases in the amount of population-level genomic data available for model and non-model species. These data are being used, among other things, in inference of population history and demographic parameters (Beichman et al., 2018), studying adaptive introgression (Gower et al., 2021), distinguishing adaptation from drift (e.g. Hsieh et al., 2021), and understanding the implications of deleterious variation in populations of conservation concern (e.g. Robinson et al., 2023).”

Something that would be useful for the stdpopsim resource in general, though not necessarily something for the paper, would be some kind of more human-friendly representation of the demographic models implemented in the curated library. Perhaps I'm not looking in the right place, but as far as I can tell, if I want to study the curated demographic models, I need to go into the Python scripts on the stdpopsim GitHub page (e.g.

https://github.com/popsim-consortium/stdpopsim/tree/main/stdpopsim/catalog/BosTau). Here the various parameters and demographic events are hard-coded into the scripts. To understand the model being implemented, one thus needs to go dig into these scripts - something which is not necessarily very accessible to all researchers. Visual representations, such as the one for Anopheles gambiae in Fig 2. in the paper, are more widely accessible. I wonder if such figures could be produced for all the curated models and included in the GitHub folders alongside the scripts, perhaps aided by an existing model visualization software such as POPdemog. Again, I would not suggest that this is necessary for the paper, but if practically feasible I think it would be a useful addition to the resource in the longer term.

This is a very good point. The stdpopsim catalog actually has a documentation page that provides a brief summary for each demographic model, including a graphical representation. This graphical representation is generated using demesdraw applied to the demographic model object implemented in the code. Thus, potential users do not have to dig through the Python code to figure out the details of the demographic model. We used a similar approach to generate the image of the demographic history of A. gambiae for Fig. 2 of the paper. The documentation page is an important part of the stdpopsim catalog, and we now added a link to it in section “Data availability”, and we mention it in key places in the manuscript, such as the caption of Fig 2.

Reviewer #2:

An important update to the stdpopsim software is the capacity for researchers to annotate coding regions of the genome, permitting distributions of fitness effects and linked selection to be modeled. However, though this novel feature expands the breadth of processes that can be evaluated as well as is applicable to all species within the stdpopsim framework, the authors do not provide significant detail regarding this feature, stating that they will provide more details about it in a forthcoming publication. Compared to this feature, the additions of extra species, finite-site substitution models, and non-crossover recombination are more specialized updates to the software.

It would be helpful to provide additional information regarding the coding annotation (and associated distribution of fitness effects and linked selection) that is implemented in the current version of stdpopsim, but will be detailed in a forthcoming paper. This is not to take away from the forthcoming paper, but I believe this is the most important update to the software, and the current manuscript only brushes over it.

We agree that implementation of selection in simulations is a significant addition to stdpopsim. However, our intention in this manuscript is to focus on the separate effort we made in the last two years to expand the utility of stdpopsim to a more diverse set of species. We think the manuscript stands firmly even without discussing in detail the new features that allow modeling selection. The main reason we briefly mention these features in sections “Additions to stdpopsim” and “Basic setup for chromosome-level simulations” is because the released version of stdpopsim contains implemented DFEs for a few species, and we did not want to completely ignore this. We thus added a brief comment at the end of the “Basic setup” section (page 8) mentioning the three model species for which the stdpopsim catalog currently has annotations and implemented DFE models. We think that a more detailed description of how these features and how they should be used is best left to the manuscript that the PopSim community is currently writing (preprint expected later this year).

When it comes to simulating realistic genomic data, the authors clearly lay out that parameters obtained from the literature must be compatible, such as the same recombination and mutation rates used to infer a demographic history should also be used within stdpopsim if employing that demographic history for simulation. This is a highly important point, which is often overlooked. However, it is also important that readers understand that depending on the method used to estimate the demographic history, different demographic models within stdpopsim may not reproduce certain patterns of genetic variation well. The authors do touch on this a bit, providing the example that a constant size demographic history will be unable to capture variation expected from recent size changes (e.g., excess of low-frequency alleles). However, depending on the data used to estimate a demographic history, certain types of variation may be unreliably modeled (Biechman et al. 2017; G3, 7:3605-3620). For example, if a site frequency spectrum method was used to estimate a demographic history, then the simulations under this model from y stdpopsim may not recapitulate the haplotype structure well in the observed species. Similarly, if a method such as PSMC applied to a single diploid genome was used to estimate a demographic history, then the simulations under this model from stdpopsim may not recapitulate the site frequency spectrum well in the observed species. Though the authors indicate that citations are given to each demographic model and model parameter for each species, this may not be sufficient for a novice researcher in this field to understand what forms of genomic variation the models may be capable of reliably producing. A potential worry is that the inclusion of a species within stdpopsim may serve as an endorsement to users regarding the available simulation models (though I understand this is not the case by the authors), and it would be helpful if users and readers were guided on the type of variation the models should be able to reliably reproduce for each species and demographic history available for each species. It would be helpful to include a table with types of observed variation that the current set of 21 species (and associated demographic histories) are likely and unlikely to recapitulate well.

This is a very important point, which we now address in the section “Limitations of simulated genomes”, which we added to the manuscript. In this section, we expand on this topic and discuss various things that will affect the way simulated genomes reflect true sequence variation. This includes the choice of demographic inference method, but also the analyzed samples, and various filters. The main message of this section is that one should consider various things when deciding to implement a demographic model in simulation (or selecting a model among those implemented in stdpopsim). We also cite studies (including Beichman, et al. 2017), which compared different approaches to demography inference. However, we note that the conclusions of these comparisons are not as straightforward as the reviewer suggests. In particular, methods that make use of the site frequency spectrum (such as dadi) should be able to capture some aspects of haplotype structure, because this information is encoded in the demographic history. Furthermore, a demographic history inferred from a single genome (e.g., using PSMC) should do a reasonable job approximating some aspects of the site frequency spectrum. In other words, the aspects of genetic variation not modeled well by a given demographic inference method are not always predicted in a straightforward way. This is why we avoid summarizing this information in a table in the manuscript. The 2nd paragraph of the “Limitations of simulated genomes” section addresses some of these subtle considerations. In particular, we suggest that considering a demographic model for simulation requires some familiarity with the inference method and the way it was applied to data. Regarding the demographic models currently implemented in stdpopsim, we provide some information about each model in the documentation page of the catalog. When selecting a demographic model from the catalog, users should make use of this documentation to guide their decision. This is mentioned in the 3rd paragraph of the “Limitations of simulated genomes” section. Following-up on this issue, we intend to review the documentation and make sure it provides sufficient information for each demographic model. See this GitHub issue.

Reviewer #3:

- p5, 2nd paragraph: I think many Biologists, myself included, will think of horizontal gene transfer mostly as plasmids being transferred among bacteria and adding extra genetic material, not as homologous bacterial recombination. This made me confused about modelling horizontal gene transfer in the same way as gene conversion. It may be helpful for some readers if you specify that you are modelling this particular type of horizontal gene transfer. Some explanation along the lines of what is in Cury et al (2022) would be enough.

This is a good point. We modified the text in that sentence in the 2nd paragraph on page 5 to clarify that we are modeling non-crossover homologous recombination, and not incorporation of exogenous DNA (e.g., via plasmid transfer). The relevant part of the text now says:

“In bacteria and archaea, genetic material can be exchanged through horizontal gene transfer, which can add new genetic material (e.g., via the transfer of plasmids) or replace homologous sequences through homologous recombination (Thomas and Nielsen, 2005; Didelot and Maiden, 2010; Gophna and Altman-Price, 2022). However, the initial version of stdpopsim used crossover recombination to stand in for these processes. Although we cannot currently simulate varying gene content (as would be required to simulate the addition of new genetic material by horizontal gene transfer), the msprime and SLiM simulation engines now allow gene conversion, which has the same effect as non-crossover homologous recombination.

Following (Cury et al., 2022), we use this to include non-crossover homologous recombination in bacterial and archaeal species.”

- p5, 3rd paragraph: When you say gene conversion is turned off by default, you could refer to table 1 and briefly mention the consequence of ignoring gene conversion.

We agree that it is important to note that avoiding to model gene conversion may lead to faulty lengths of shared haplotypes across individuals. This is implied by the statement we make in the beginning of the 3rd paragraph on page 5, where we lay out the motivation for modeling gene conversion in simulation. Following the reviewer’s suggestion, we now added a statement about this in the end of that paragraph:

“Note that ignoring gene conversion may result in a slightly skewed distribution of shared haplotypes between individuals (see Table 1)”

-  p7, item 1 and p9, 1st paragraph: I am not sure what you mean by genetic map here, can you define this term? I am not sure if it is synonymous with gene annotations, a recombination map, or something else. The linkage map doesn't seem to make sense to me here.

The term ‘genetic map’ referred to the recombination map whenever it was used in the manuscript. To avoid any confusion, we now removed all mentions of ‘genetic map’, and use ‘recombination map’ instead. The recombination map is relevant in item 1 of page 7 because in species with poor assemblies you will not be able to reliably estimate recombination maps, making chromosome-scale simulations less effective. In the 1st paragraph of page 9, we discuss the issue of lifting over coordinates from one assembly to another, and if you have a recombination map estimated in one assembly, you might need to lift it over to another assembly to apply it in your simulation.

-  Table 1, last row, middle column: when you say "simulated population", I think it is a bit ambiguous. You mean "the true population that we are trying to simulate", but could be read as "the population data that was generated by simulation". I would delete the word simulated here.

What we mean here is that the selected effective population size should reflect the observed genetic diversity in real genomic data. We realize that the previous wording was confusing, and changed this to the following:

“Set the effective population size (Ne) to a value that reflects the observed genetic diversity”

-  Figure 2, and other places when you refer to mutation and recombination rate (eg p11, last paragraph), can you include the units (e.g. per base pair, per generation)?

Throughout the manuscript, rates are always specified per base per generation. In Figure 2, this is specified in the caption (3rd line). We added units in other places in section “Examples of added species” on pages 12-13, where they were indeed missing.

-  p11, "default effective population size": can you use a more descriptive word instead of the default? Maybe the historical average? Also, what is this value used for in the simulations when there is a demographic model specified (as in the case of Anopheles)?

We think that “default effective population size” is the most appropriate term to use here, since we are referring to the parameter in the species model in stdpopsim. It is correct that the value of this parameter should reflect the historical average size in some sense, but it is really unclear what this should be in the case of a species like Bos taurus, which experienced a very dramatic bottleneck in the recent past. We address this subtle, yet important, issue in the sentence preceding this one. If a demographic model is specified in simulation, it overrides the default effective population size, and its value is ignored (which is why we refer to it as ‘default’). We added a short sentence clarifying this in the 2nd paragraph of the “Bos Taurus” section (now page 12).

“Note that the default Ne is only used in simulation if a demographic model is not specified.”

-  p8, when you say "Such simulations are useful for a number of purposes, but they cannot be used to model the influence of natural selection on patterns of genetic variation.": You may want to bring up the discussion that many of these neutral parameters taken from the literature could have been estimated assuming genome-wide neutrality, and thus ignoring the effect of background selection. Therefore the parameter values might reflect some effect of background selection that was unaccounted for during their estimation.

This is an important subtle point, which we now address in the section “Limitations of simulated genomes”, which we added to the revised manuscript. In that section, we discuss various limitations of simulations, focusing on inferred demographic models. We address the potential influence of the segments selected for analysis toward the end of 2nd paragraph in that section (page 9):

“... all methods assume that the input sequences are neutrally evolving. This implies that technical choices, such as the specific genomic segments analyzed and various filters, may also influence the inferred model and its ability to model observed genetic variation.”

Interestingly, background selection in itself typically does not have a strong effect on the inferred model. This is something that is examined in the forthcoming publication that presents simulations with natural selection in stdpopsim.

-  Why are some concepts written in bold (eg effective population size, demographic model)? Were you planning to make a vocabulary box? I think this is a good idea given that you are aiming for a public that can include people who are not very familiar with some population genetics concepts.

In the “Examples of added species” section, we use boldface fonts to highlight the model parameters that were determined for each species. We added a statement clarifying this in the beginning of this section (page 11), and made sure that all the relevant parameters were consistently highlighted throughout this section. In other sections, we use boldface fonts only for titles. A few cases that did not conform to this rule were removed in the current version. We did not intend on adding a vocabulary box, but considered this when revising the manuscript, due to the reviewer’s suggestion. However, we found it difficult to converge on a small (yet comprehensive) set of terms with accurate and succinct definitions. We think that the important terms are adequately defined within the text of the manuscript, providing sufficient information also for readers who are not expert population geneticists.

- p4, 2nd paragraph: Are these automated scripts that are used to compare models publicly available? If you are suggesting that people use this approach generally when coming up with a simulation model (p8, penultimate paragraph), it would be helpful to have access to these automated scripts.

The scripts are part of the public stdpopsim repository on GitHub, and may be used by anyone. Some components of these scripts are more easy to apply in general, such as comparing a demographic model with one implemented separately by the reviewer. This step, for example, is achieved by application of the Demography.is_equivalent method in msprime. Other parts of the comparison depend on the specific structure of python objects used by stdpopsim, so they are not likely to be useful when implementing simulations outside the framework of stdpopsim.

-  p9, 1st paragraph, and p.12 2nd paragraph: instead of adjusting the mutation rate to fit the demographic model (and using an old estimate of the mutation rate), would it be ok to adjust the demographic model to fit the new mutation rate? E.g. with a new mutation rate that is the double of a previous estimate, would it be ok to just divide Ne by 2 such that Ne*mu is constant (in a constant population size model)? I imagine this could get complicated with population size changes.

In principle, this could be done if you were simulating neutrally evolving sequences (without modeling natural selection). Since the coalescence is scale-free, then you can scale down all population sizes and divergence times by a multiplicative factor, and scale up migration rates and the mutation rate by the same factor, and you get the exact same distribution over the output sequences. However, making sure you get the scaling right is tricky and is quite error-prone. Especially considering the fact that you have to do this every time the mutation rate of a species is updated. Moreover, once you start modeling natural selection, this scale-free property no longer holds. Thus, the simple solution we came up with in stdpopsim is to attach to each demographic model the mutation rate used in its inference.

Author Response:

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

We sincerely thank all the editors and reviewers for taking the time to evaluate this study. Here is our point-by-point response to the reviewers’ comments and concerns.

Reviewer #1 (Public Review):The study by Oikawa and colleagues demonstrates for the first time that a descending inhibitory pathway for nociception exists in non-mammalian organisms, such as Drosophila. This descending inhibitory pathway is mediated by a Drosophila neuropeptide called Drosulfakinin (DSK), which is homologous to mammalian cholecystokinin (CCK). The study creates and uses several Drosophila mutants to convincingly show that DSK negatively regulates nociception. They then use several sophisticated transgenic manipulations to demonstrate that a descending inhibitory pathway for nociception exists in Drosophila.

[…]

Weaknesses:

A minor weakness in the study is that it is unclear how DSK negatively regulates nociception. An earlier study at the Drosophila nmj shows that loss of DSK signaling impairs neurotransmission and synaptic growth. In the current study, loss of CCKLR-17D1 in Goro neurons seems to increase intracellular calcium levels in the presence of noxious heat. An interesting future study would be the examination of the underlying mechanisms for this increase in intracellular calcium.

We thank the reviewer for the kind and very positive evaluation of our manuscript. We agree that this study has not elucidated the intracellular molecular pathway(s) downstream of CCKLR-17D1 that are involved in the regulation of the activity of Goro neurons, and we think that it would definitely be an interesting topic for future research.       

Reviewer #1 (Recommendations For The Authors):

The response latencies for the control yw larvae seem large, with many larvae appearing to be insensitive to the thermal stimulus. Is this just an effect of the yw genetic background? A brief discussion of this might be helpful.

We thank the reviewer for pointing this out. We have also noticed that the yw control larvae tend to show longer response latencies than the other control strains, and in the revised manuscript, we have added the following sentence in the Result section (Lines 91–94):

“We have noticed that the yw control strain, which was used by us to generate the dsk and receptor deletion mutants, showed relatively longer response latencies to the 42 °C probe compared to the other control strains used in this study. This may be attributed to the effect of the genetic background, although, presently, the cause for this difference is unknown.”

Reviewer #2 (Public Review):_

This is an exceptional study that provides conclusive evidence for the existence of a descending pathway from the brain that inhibits nociceptive behavioral outputs in larvae of Drosophila melanogaster. […] The study raises many interesting questions for future study such as what behavioral contexts might depend on this pathway. Using the CAMPARI approach, the authors do not find that the DSK neurons are activated in response to nociceptive input but instead suggest that these cells may be tonically active in gating nociception. Future studies may find contexts in which the output of the DSK neurons is inhibited to facilitate nociception, or contexts in which the cells are more active to inhibit nociception._

Reviewer #2 (Recommendations For The Authors):I have no recommendations for the authors as this is a very complete and thoroughly executed study. The writing is crystal clear.

We thank the reviewer for the kind and very positive evaluation of our manuscript. We are happy to know that our current manuscript was deemed to be clear and convincing by the reviewer.

Reviewer #3 (Public Review):[…] Overall the authors use clean logic to establish a role for DSK and its receptor in regulating nociception. I have made a few suggestions that I believe would strengthen the manuscript as this is an important discovery.

Major comments:

1. It's not completely clear why the authors are staining animals with an FLRFa antibody. Can the authors stain WT and DSK KO animals with a DSK antibody? Also, can the authors show in supplemental what antigen the FLRFa antibody was raised against, and what part of that peptide sequence is retained in the DSK sequence? This overall seems like a weakness in the study that could be improved on in some way by using DSK-specific tools.

We thank the reviewer for this query. We would like to clarify that we first tried the FLRFa antibody to visualize an RFamide-type neuropeptide other than DSK in Drosophila and found that the staining pattern is quite similar to that of anti-DSK, as shown by Nichols et al. [1]. According to the original paper describing the anti-FLRFa antisera [2] (already cited in the reviewed manuscript), the antigen used to raise it was the Phe-Met-Arg-Phe-NH2 peptide conjugated with succinylated thyroglobulin, and the study experimentally shows that the antibody well binds to peptides containing Met-Arg-Phe-NH2 or Leu-Arg-Phe-NH2 sequence and has 100% cross-reactivity to FLRFa. As DSK contains Met-Arg-Phe-NH2 sequence [3], the cross-reaction of this antibody to DSK is consistent with the description of the original study.    

Although we were unable to use an antibody specific to DSK, our staining data with dsk deletion mutants and the expression pattern of DSK-2A-GAL4 corroborate each other (Figure 2 and Figure 2-figure supplement 1), which we believe provides compelling evidence for the specific expression of DSK in MP1 and Sv neurons, and for that DSK-2A-GAL4 is a reasonably effective tool to specifically manipulate DSK-expressing neurons.

2. What is the phenotype of DSK-Gal4 x UAS-TET animals? They should be hyper-reactive. If it's lethal maybe try an inducible approach.

We thank the reviewer for this question. Unfortunately, we have not attempted this experiment, although we agree that this would be a nice addition to further strengthen the study if TET worked well in the DSKergic neurons.

3. Figure 9. This was not totally clear, but I think the authors were evaluating spontaneous (i.e. TRPA1-driven) rolling at 35C. The critical question is "does activating DSK-expressing neurons suppress acute heat nociception" and this hasn't really been addressed. The inclusion of PPK Gal4 + DSK Gal4 in the same animal kind of clouds the overall conclusions the reader can draw. The essential experiment is to express UAS-dTRPA1 in DSK-Gal4 or GORO-Gal4 cells, heat the animals to ~29C, and then test latency to a thermal heat probe (over a range of sub and noxious temperatures). Basically prove the model in Figure 10 showing ectopic activation or inhibition for each major step, then test heat probe responses.

We thank the reviewer for suggesting ideas for alternative experiments to potentially strengthen our conclusion. Regarding experiments using heat probes, previous studies have demonstrated that (i) Blocking ppk1.9-GAL4-positive C4da neurons almost completely abolishes the larval nociceptive response to local heat stimulations [4]; (ii) Local heat stimuli above 39 °C readily activate C4da neurons and larval nociceptive rolling [5-9]; and (iii) Thermogenetically or optogenetically activating these neurons is sufficient to trigger Goro neurons and larval rolling [4, 10-12]. Thus, it has now been made clear that heat probes induce larval nociceptive rolling via excitation of the C4da pathway, and we believe that our experiments using thermogenetic activation of C4da neurons can be safely interpreted as an alternative to experiments using heat probes. Using heat probes demands a more complicated experimental set-up to be combined with CaMPARI imaging experiments, and this is another reason why we preferred to take the thermogenetic approach.

We have also considered the experiment using Goro-GAL4 instead of ppk-GAL4. However, if dTRPA1 artificially activates Goro neurons far downstream of the neuronal mechanism by which MP1 activation suppresses Goro neuron activity, the effect of MP1 activation may be bypassed and masked. As we currently do not know the epistasis between dTRPA function and the effect of MP1 activation in modulating the activity of Goro neurons, we rather chose to activate C4da neurons by using ppk-GAL4, which likely resulted in more natural activation of Goro neurons than dTRPA1-triggered direct activations.

4. It would also then be interesting to see how strong the descending inhibition circuit is in the context of UV burn. If this is a real descending circuit, it should presumably be able to override sensitization after injury.

Indeed, this is an interesting avenue to explore in future studies to understand the type of situation in which the DSKergic descending system functions to control nociception.

Reviewer #3 (Recommendations For The Authors):Overall this is a good story and the claims are generally supported with experimental evidence. The way to really improve this study would be to use more precise and definitive tools, like specific antibodies, specifically targeted genes, and better temporal control of the descending circuit to prove this is inducible sufficient to suppress acute thermal nociception and this occurs only via a descending pathway, etc. However this would be exponentially more work, and so the authors I guess need to weigh the cost-benefit of definitive proof vs. strong evidence for their claims. Overall I think this study will be the beginning of a new line of inquiry in the field that has the potential to guide our understanding also of mammalian descending pathways, and as such, this study is of value to the community.

We appreciate the reviewer’s multiple interesting ideas for experiments that could have been performed to further reinforce our findings. We agree that some experiments that the reviewer suggested would potentially strengthen this work if supplemented. However, as aforementioned, in our humble opinion, we think that the experiments that the reviewer suggested are either outside the scope of this paper or have no significant benefits over the experiments that were already conducted, and hence are not essential to the present study.

References

1. Nichols, R. and I.A. Lim, Spatial and temporal immunocytochemical analysis of drosulfakinin (Dsk) gene products in the Drosophila melanogaster central nervous system. Cell Tissue Res, 1996. 283(1): p. 107-16.

2. Marder, E., et al., Distribution and partial characterization of FMRFamide-like peptides in the stomatogastric nervous systems of the rock crab, Cancer borealis, and the spiny lobster, Panulirus interruptus. J Comp Neurol, 1987. 259(1): p. 150-63.

3. Nassel, D.R. and M.J. Williams, Cholecystokinin-like peptide (DSK) in Drosophila, not only for satiety signaling. Front Endocrinol, 2014. 5.

4. Hwang, R.Y., et al., Nociceptive neurons protect Drosophila larvae from parasitoid wasps. Curr Biol, 2007. 17(24): p. 2105-2116.

5. Tracey, W.D., Jr., et al., painless, a Drosophila gene essential for nociception. Cell, 2003. 113(2): p. 261-73.

6. Xiang, Y., et al., Light-avoidance-mediating photoreceptors tile the Drosophila larval body wall. Nature, 2010. 468(7326): p. 921-6.

7. Burgos, A., et al., Nociceptive interneurons control modular motor pathways to promote escape behavior in Drosophila. eLife, 2018. 7.

8. Honjo, K. and W.D. Tracey, Jr., BMP signaling downstream of the Highwire E3 ligase sensitizes nociceptors. PLoS Genet, 2018. 14(7): p. e1007464.

9. Im, S.H., et al., Tachykinin acts upstream of autocrine Hedgehog signaling during nociceptive sensitization in Drosophila. eLife, 2015. 4: p. e10735.

10. Ohyama, T., et al., A multilevel multimodal circuit enhances action selection in Drosophila. Nature, 2015. 520(7549): p. 633-9.

11. Honjo, K., R.Y. Hwang, and W.D. Tracey, Jr., Optogenetic manipulation of neural circuits and behavior in Drosophila larvae. Nat Protoc, 2012. 7(8): p. 1470-8.

12. Zhong, L., et al., Thermosensory and non-thermosensory isoforms of Drosophila melanogaster TRPA1 reveal heat sensor domains of a thermoTRP channel. Cell Rep, 2012. 1(1): p. 43-55.

Author Response:

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

We’d like to thank the three reviewers for reviewing in depth our work and providing insightful comments and suggestions.

Reviewer #1 (Recommendations For The Authors):

1) The evidence that MS023 is actually working in vivo in their last experiment (Fig 6) needs to be strengthened. This could be due to the timing of the experiment. Tail tips were collected 48 h after the final injection and analyzed by Western for ADMA and SDMA levels. They do see subtle changes, in the right directions, of SDMA and ADMA (but these changes are really not very obvious). Perhaps the inhibitor has already been largely metabolized two days after injection. Have they looked at MMA levels?

We have quantified the ADMA and SDMA levels of Fig. S6. We have not measured MMA levels. The text has been edited as follows:

“The average ADMA relative expression was 0.95 for vehicle treated mice and 0.83 for MS023 treated mice (p < 0.00041). The average SDMA relative expression was 0.92 for vehicle treated mice and 0.94 for MS023 treated mice (p < 0.17). These whole-body measurements as measured by tail biopsies show MS023 promotes the decrease of proteins with ADMA and a slight increase in proteins with SDMA. It is known that inhibition of type I PRMTs or PRMT1 deletion diminishes ADMA and increases SDMA due to substrate scavenging (Dhar et al, 2013).”

2) The authors need to explain why they would expect an increase in SDMA levels in these mice after MS023-treatment. 

We have edited the text as follows:

“It is known that inhibition of type I PRMTs or PRMT1 deletion diminishes ADMA and increases SDMA due to substrate scavenging (Dhar et al, 2013).”

3) In the discussion, it would be valuable to address the types of CRISPR-screens that could be performed in these MS023-expanded MSCs. They mention this as a benefit in the introduction, but to expand on this idea in the discussion.

The idea here was not necessarily to perform a CRISPR screen on the MS023-treated cells (although it is an interesting idea), but rather to correct the genetic mutation by CRISPR-Cas9 to enhance the success of genetically corrected autologous cell transplantation. The addition of MS023 to MuSC in vitro would allow to expand the cells while maintaining their self-renewal potential, thereby providing the opportunity to correct the mutation on the dystrophin gene using technologies such as CRISPR prime editing (Mbakam et al., 2022 Mol Ther Nucleic Acids 30:272-285). Our results demonstrating that MS023 enhances cell engraftment suggest that this method could be used to improve autologous cell transplantation efficiency. We have edited the text in the discussion as follows:

“Our findings suggest that type I PRMT inhibitors may have therapeutic potential for treating certain skeletal muscle diseases. For instance, to improve the efficacy of autologous cell therapy, the dystrophin-deficient MuSCs collected from DMD patient and corrected by CRISPR prime editing (Happi Mbakam et al, 2022) could be treated with MS023 to maintain their stemness and enhance their cell engraftment capacity.”

4) Also, could they address the potential value of MSC culture and expansion using a combination of SETD7 inhibition and PRMT1 inhibition?

Agreed. We have edited the text as follows:

“These findings suggest that inhibiting methyltransferases can affect MuSC fate and perhaps a combination of Setd7 and MS023 inhibitors would provide a more favorable combination to promote the expansion of MuSCs while maintaining their stem cell-like properties.”

Reviewer #2 (Recommendations For The Authors): 

In figure 2 the authors show that upon removal of MS023, the cells differentiate more efficiently. In figure 5E-F they show that the mice that received MS023-treated cells had more GFP mature muscle fibers. However, in figure 5C-D these cells have the same capacities to terminally differentiate. This reviewer was wondering if these cells would differentiate faster? Have the authors look into this?

The reviewer raises an interesting point. Our in vitro experiments shown in Supplemental Figure S1 indicate that MS023-treated cells are actively more cycling (more ki67+ cells) and are less committed to differentiation (less Pax7-MyoD+ cells), which would suggest that they would need to exit the cell cycle and differentiate faster to reach the same fusion capacity after 3 days of differentiation without MS023. Future experiments with a time course including earlier time points will be needed to confirm if these cells differentiate faster.

Reviewer #3 (Recommendations For The Authors): 

1) MS023 is a non-selective inhibitor of type I PRMTs. It has comparable IC50 values for PRMT1 and PRMT4 (CARM1), and lower IC50 values for PRMT6 and PRMT8. The authors argue that the cellular phenotype caused by MS023 is solely mediated via PRMT1, since the specific PRMT4-inhibitor TP-064 has no effects on MuSC expansion. TP-064 treatment was not used as a control for the transplantation and muscle strength measurement experiments. Are PRMT6 and PRMT8 expressed in MuSC and are thy inhibited by the applied concentrations of MS023? Kawabe et al reported that CARM1 methylates Pax7, thereby inducing Myf5 transcription during the asymmetric division of MuSC (PMID: 22863532). Is the expression of Myf5 reduced upon MS023 treatment? scRNAseq of MuSC 4-day after culture is too late to address this question, since the majority of the cells are already committed to differentiation. Staining for Myf5 using ex vivo cultured fibers or regenerating muscles in vivo should be used. 

Indeed, we mention throughout the text that MS023 is a type I PRMT inhibitor. We have edited the text as follows suggesting the effect are most likely mediated by inhibition of PRMT1 in vivo.

“Treatment of MuSCs with MS023 resulted in metabolic reprogramming of MuSCs, supporting a role for type I PRMTs as metabolic regulators. In vitro, MS023 has been shown to inhibit several type I enzymes at nM concentrations (Eram et al., 2016). It is well-documented that PRMT1 is the major cellular type I enzyme (Pawlak et al, 2000) and this is why PRMT1, but not the other type I PRMTs are embryonic lethal in mice (Guccione & Richard, 2019). The numerous published data in cellulo with MS023 are thus far only reproduced by PRMT1-deficiency by siRNA or knockout, suggesting that MS023 actions in vivo are predominantly mediated by inhibiting PRMT1 (Gao et al, 2019; Plotnikov et al, 2020; Wu et al, 2022; Zhu et al, 2019). Thus, the effects of MS023 on MuSCs are most likely mediated by inhibition of PRMT1.”

Moreover, we investigated the expression of other type I PRMTs as suggested by the reviewer. We investigated their expression from publicly available single cell RNAseq dataset (Oprescu SN et al, iScience 2020, 23:100993), which performed analysis on skeletal muscle at different time points post-cardiotoxin injury (uninjured, and 0.5, 2, 3.5, 5, 10, 21 days post-injury). The findings show that Prmt1 is by far the most expressed type I PRMT in MuSCs at every time point tested. Carm1 (Prmt4) is expressed at high level in a small/moderate subset of cells, especially during regeneration. Prmt6 is expressed at low level in a small proportion of cells, while Prmt8 expression was not detected. These findings are coherent with our observation that Prmt1 is the predominant type I Prmt in MuSCs, which further support our hypothesis that it is the main target of MS023. These findings were added in Suppl. Fig 1B.

The expression of Myf5 during asymmetric division is indeed well characterized on muscle fiber-associated MuSCs (Dumont et al., 2015 Nat Med 21:1455; Kawabe et al., 2012 Cell Stem Cell 11:333). As the reviewer states, the 4-day time point is too late to investigate Myf5 expression. Additionally, these cells were cultured ex-vivo and were not fiber-associated. Therefore, scRNAseq is not an ideal method to address the question of whether MS023 treatment modulates Myf5 expression, and further experiments would be required to examine Myf5 in an appropriate context (i.e. on ex-vivo cultured myofibers).

2) Figure 2 is not very informative, while the second paragraph of the result parts is excessive and too complicated. The extensive description of differential gene expression in each potential subpopulation is neither very informative nor helpful to convince the reader that the M3/M5 population has acquired more stemness-like features due to the MS023 treatment. From my point of view, the data just reflect the increased proliferative capacity of MS023-treated cells with elevated cell cycle markers, ribosomal protein, and metabolic state. Do the M1-M5 populations show any different distribution along the trajectory? The authors need to show cell trajectories for each sample and cluster in Figure S3A. It is also imperative to present the distribution of signature genes for each individual cluster. Essentially, M1-M5 all located together in one cloud. What justifies segregation into different subclusters? The color code for the different clusters (including the trajectories) to allow better distinction. 

MS023 treated MuSCs contain a subpopulation with higher Pax7 expression (Supplementary Figure S2F, S2G), which is consistent with the IF results in Figure 1 and emphasized in the abstract. Why are these data in the supplements and not in a main figure (e.g. in figure 2)?

We appreciate the thoughtful and detailed comments on our single-cell data. Please see below for a response to each point:

To address the concern that the results section is excessive, our intention was to simply provide the reader with a descriptive overview of the identity of each subcluster that the software identified. In fact, to ensure clarity and conciseness, we elected to provide only the names of a select few cluster markers rather than list all of the significant cluster markers that were generated. We kindly refer the reviewer to Supplementary Table S1 for a more extensive list of markers.

In response to the reviewer’s comment: “The color code for the different clusters (including the trajectories) to allow better distinction,” we agree that colour-coding is helpful, please refer to Figure 2A for a colour-coded map of the clusters.

To address the reviewer’s question regarding what justifies segregation into different subclusters for M1-M5, refer to Supplementary Table S1 for a list of uniquely enriched markers for each cluster. This list was filtered to include marker genes that were present only in a given cluster, thus contributing to its uniqueness and explains why that cluster was identified as being distinct from another given cluster.

Lastly, since the elevated Pax7 levels in MS023-treated MuSCs was already presented and discussed thoroughly in Figure 1, we elected to avoid repetition in the main Figures and presented the ridge plots showing elevated Pax7 in the Supplementary Material for Figure 2

3) The same group has reported previously that PRMT1-deficient MSCs show reduced expression of MyoD due to disruption of Eya1/Six1 recruitment to the MyoD promoter (PMID: 27849571). However, the scRNAseq result does reflect this finding. MyoD levels are not significantly changed in d4 MS023 compared with d4 (Supplementary figure S2G). The authors need to provide an explanation. Furthermore, the authors previously described that "the majority of PRMT1-deficient MSCs repressed Pax7 expression at day 3 while being Ki67 positive (Fig. 5B). How does that fit to the current observations, which indicate an increase of Pax7+ cells after MS023 treatment? This discrepancy needs to be resolved. 

While the scRNAseq does not show a reduction in overall MyoD expression in MS023-treated MUSCs, there is indeed a reduction in the proportion of MyoD+ myofiber-associated MuSCs (Figure 1C, 1D). Supplemental Figure S2G further shows a subpopulation in the d4MS023 group with lower MyoD expression that was not present in the d4 group, reflective of the findings in Figures 1C and 1D. Therefore, although the average expression was not shifted significantly with MS023, there was indeed a subpopulation of MuSCs with lower MyoD expression.

The reviewer additionally points out that Fig. 5B from a previous study (Blanc et al., 2017 MCB 37:e00457) performed by our group, shows that Pax7 expression was repressed at day 3 of culture in PRMT1-null MuSCs. However, this quantification was based on immunofluorescence staining where cells are marked positive or negative for Pax7 expression and does not look at the intensity of Pax7 expression levels. In our current study, we examine the expression levels of Pax7 in discrete subpopulations of MuSCs and found that there is a subpopulation of MuSCs that emerges with MS023 treatment that has higher Pax7 expression than untreated counterparts. Therefore, the results of the two experiments are not directly comparable. 

4) I do have a major problem with the interpretation of the metabolic changes in MS023-treated MuSC. In the abstract, the authors wrote, "These findings suggest that type I PRMT inhibition metabolically reprograms MuSCs resulting in improved self-renewal and muscle regeneration fitness." There is simply no causal evidence to support this claim, which is solely based on a correlation. If the authors want to maintain this claim they either need to stimulate OXPHOS and glycolysis by other means to see whether such a manipulation recapitulates the effects of MS023 or attenuate OXPHOS and glycolysis to see whether this abrogates the effects of MS023. To prove whether increased OXPHIS is a cause for improved self-renewal, the authors might simply co-treat MuSC with MS023 and an OXPHIS inhibitor and analyze consequences for the Pax7+/MyoD- population. 

We thank the reviewer for the excellent suggestions of experiments that would solidify a causal relationship between increased metabolism and increased self-renewal. We will certainly consider them for future studies. We agree that the relationship in the present study is correlative, and the text has been modified in the abstract as follows:

“Single cell RNA sequencing (scRNAseq) of ex vivo cultured MuSCs revealed the emergence of subpopulations in MS023-treated cells which are defined by elevated Pax7 expression and markers of MuSC quiescence, both features of enhanced self-renewal. Furthermore, the scRNAseq identified MS023-specific subpopulations to be metabolically altered with upregulated glycolysis and oxidative phosphorylation (OxPhos). Transplantation of MuSCs treated with MS023 had a better ability to repopulate the MuSC niche and contributed efficiently to muscle regeneration following injury. Interestingly, the preclinical mouse model of Duchenne muscular dystrophy had increased bilateral grip strength 10 days after a single intraperitoneal dose of MS023. Our  findings show that inhibition of type I PRMTs increased the proliferation capabilities of MuSCs with altered cellular metabolism, while maintaining their stem-like properties such as self-renewal and engraftment potential.”

5) Ryall et al reported that MuSCs undergo a metabolic switch from fatty acid oxidation to glycolysis with reduced intracellular NAD+ levels and reduced activity of SIRT1, leading to elevated H4K16 acetylation. Here, both OXPHOS and glycolysis are increased after treatment of MuSC with MS023. Are the NAD+ and H4K16ac levels changed in MS023-treated MuSC? 

This is another excellent study that would help to support a causal relationship between MS023 treatment and increase OXPHOS and glycolysis and could certainly be addressed in future studies.

6) In Ryall et al.'s results, there was no difference in the basal mitochondrial OCR between freshly isolated MuSCs and cultured MuSCs. Importantly, stimulation of OXPHOS will increase ROS concentration, resulting in premature differentiation of MuSC (PMID: 30106373). Furthermore, increased ROS levels will most likely enhance DNA damage rather than improve self-renewal. The authors have to address these issues and also monitor ROS and DNA damage levels. 

The lack of cell death upon treatment with MS023 in the present study would indicate that there is no major ROS-induced DNA damage occurring. Additionally, the propensity of MS023-treated MuSCs to retain their stemness while in long-term culture (Supplemental figure S1E) would indicate that in this context, premature differentiation is not a concern.

7) The authors used FACS-analysis of MuSCs three weeks after transplantation to demonstrate that MS023 treatment enables better engraftment into the MuSC niche. The six-fold increase of transplanted cells in the MuSC niche is difficult to understand, Why shall transplanted cells compete so efficiently with endogenous MuSC for repopulation of the niche? Is it possible that some of the transplanted MuSC are still lingering within the interstitium and erroneously counted as bona fide MuSC? The authors have to determine the localization of transplanted MuSC. Are all transplanted cells indeed situated in the proper niche or are they also present outside the basal lamina of muscle fibers? 

The hindlimbs which received the engraftment were irradiated 24h prior to engraftment, therefore the ability of endogenous MuSCs to compete is compromised. Additionally, Figure 5E shows that the regenerated muscle indeed has GFP negative fibers that would have been generated from endogenous MuSCs, indicating that MS023-treated MuSCs did not fully outcompete endogenous MuSCs.

8) The authors reported that an only 3-day treatment with MS023 is sufficient to dramatically improve muscle function in mdx mice even 30 days later, which is hard to swallow. What is the evidence that such strong effects are primarily mediated by stimulation of MuSC expansion? Are there other pathways or cells that respond to MS023 treatment and stimulate muscle strength? To support the claim of a 'better' stem cell function as the major cause for MS023-dependent stimulation of muscle strength in mdx mice, the authors need to determine the total number of Pax7+ cells, Pax7+/Ki67+, Pax7+/MyoD+, Pax7+/MyoD-, Pax7-/MyoD+ and myonuclei. It is also absolutely mandatory to include wildtype controls in the muscle strength measurements. Does MS023 treatment also increase muscle strength in wild-type controls? 

Agreed. We cannot exclude if the effect is mediated by an expansion of the MuSC pool or by an effect on other cell types, such as a direct impact on the myofibers. The manuscript has been modified to include the following text:

“Furthermore, our findings show that injection of MS023 in the dystrophic mouse model mdx led to enhanced muscle strength with effects lasting up to 30 days.  We cannot exclude if the effect of MS023 was mediated by an expansion of the MuSC pool or by an effect on other cell types, such as a direct impact on the myofibers. The goal of this experiment was to provide a therapeutic perspective for the possible use of type I PRMT inhibitor for the treatment of DMD.”

The goal of this figure was to provide a therapeutic perspective for the use of type I PRMT inhibitor for the treatment of DMD. Muscle wasting/weakness in DMD is a complex and multifactorial process (e.g., myofiber fragility, MuSC defects, chronic inflammation, fibrofatty accumulation). If MS023 can target multiple aspects of the physiopathology of the disease it would increase its therapeutic applicability. Further studies will be needed to determine the exact mechanism by which MS023 mediate its beneficial effect. These future studies could include the use of wild type control, as the reviewer suggests, to investigate the role of MS023 in a non-muscle degenerative context.

9) Ideally, a genetic inactivation-reactivation of PRMT1 should be done to validate the results with MS023 and to make sure that indeed the transient inhibition of PRMT1 is responsible for the beneficial effects of MS023. Of course, this would be a major effort when done in genetically manipulated mice and therefore is not adequate to ask for. However, it should be possible to use PRMT1-deficient MuSC, which the authors have in hand, and re-express PRMT1 in these cells with an AAV or a lentivirus. 

We agree that genetic ablation of PRMT1 is a key experiment to validate MS023 results. Please refer to previous work from our group, which shows that PRMT1-KO MuSCs have an enhanced self-renewal phenotype (Blanc et al., 2017 MCB 37:e00457), similar to what was observed in the present study with MS023 treatment.

10) Some claims are overstated and/or to aggressive. E.g.: "Therefore, through repression of type I PRMTs with MS023, we have reprogramed MuSCs to acquire a unique and previously uncharacterized identity." I do not see clear evidence that MS023 treatment 'reprograms' MuSC to a 'unique identity'. The observed changes are in large parts compatible with a simple stimulation of proliferation. 

The unique finding in our data is that treatment with MS023 resulted in a shift in identity as compared to the DMSO-treated proliferating MuSCs (M1, M2 and M4), creating transcriptionally distinct M3 and M5 clusters. M3 and M5 had elevated markers for metabolism (E.g. Eno1, Atp5k, etc) and early activation (E.g. Fos, Jun), while the untreated MuSCs in clusters M1, M2 and M4 did not. Furthermore, M3 and M5 had higher baseline levels of Pax7 expression when compared to untreated cells. Together, these findings describe a transitional subpopulation of MuSCs unique to MS023 treatment which not only harbour stem like/early activation markers Pax7, Fos and Jun, but also elevated proliferative markers related to cell cycle and energy metabolism. This particular combination of characteristics is unique to the MS023-treated MuSCs, thus identifying a unique subtype of MuSC identity. In accordance with our scRNAseq data, we validated experimentally that MS023-treated cells have higher energy metabolism and increased self-renewal potential, thereby confirming that the unique transcriptomic signature of these cells also lead to a different cell fate decision.

Author Response:

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

1) l. 80: "evolved from a fourth domain of cellular life": I am worried a little bit about putting together what I believe are too distinct hypothesis: (i) NCLDV deriving from a complex (ancestral) cellular life form (possibly proto-eukaryotic) by reductive evolution, and (ii) NCLDV forming or deriving from a fourth domain of cellular life. To clarify for non-expert reader, I would suggest rephrasing as "evolved reductive evolution, possibly from a fourth domain of cellular life...".

Following the reviewer’s recommendation, we have clarified the sentence by writing: “These observations are at odds with the suggestion that NCLDVs originated by reductive evolution, possibly from a fourth domain of cellular life (Colson et al., 2018; Legendre et al., 2012; Patil and Kondabagil, 2021).”.

2) l. 187-198: Please provide more information on which tool (with version number and parameter) was used to search genomes for MCPs. When I downloaded the HMM model and the faa file for the MCP from the figshare repository and tried to match the two, only a small number (4) of the MCP sequences actually matched the MCP HMM model with significant e-value, but I am not sure why? (for reference, I was using hmmsearch 3.3.2, default parameters)

We used HMMER version 3.3.2 using the default parameters (hmmbuild and hmmsearch algorithms). We now include this information in the relevant section of the Methods: “Next, we constructed a set of Hidden Markov Models (HMMER version 3.3.2, hmmbuild/hmmsearch using the default parameters) for each of the 4 core proteins involved in virion morphogenesis”.

We were able to reproduce the reviewer’s observation that the Major capsid curated HMM model returns 4 significant hits when used on the Major capsid multiple alignment file provided in FigShare (significant matches: 1. maverick2_NW_021681489.1_105940131438, 2. ncbincldv_NC_011335.1, 3. ncbincldv_NC_038553.1, 4. yutin_PLVACE1). This curated HMM model was one of the models used for searching homologous protein sequences and was built from a preliminary multiple sequence alignment comprising a different set of taxa (N. taxa = 48). In contrast, the multiple sequence alignment provided in Figshare is the final multiple sequence alignment of major capsid proteins that was used in phylogenetic analyses (N. taxa = 54). Therefore, we should not expect an exact match between the two files.

We have updated the Figshare repository with a compressed file containing all the HMMs used for searching protein homologues (n = 38), which can be validated on hmmsearch on the European Bioinformatic Institute’s website (https://www.ebi.ac.uk/Tools/hmmer/search/hmmsearch).) A separate compressed file contains the final multiple sequence alignments that were used in phylogenetic inference and hypothesis testing.

3) Figure 4: The acronyms should be explained in the legend (pPOLB, MCP, mCP, pro, atp, int, TIRs, etc)

We now provide an explanation of the acronyms used for the traits matrix on Figure 4: “Acronyms refer to genes and genomic features present in the viral genomes: pPOLB (protein-primed DNA polymerase B), MCP (major capsid protein), mCP (minor capsid protein), int (rve-type integrase), pro (adenoviral-like protease), atp (FtsK/HerA DNA packaging ATPase), TIRs (terminal inverted repeats).”

4) Figure 4: I believe that "TIRs" should be "Present in some members" for the virophages, based on https://doi.org/10.1186/s13062-015-0054-9? Interestingly, this group is typically the one that branches the deepest within virophages, which would be consistent with TIRs being an ancestral trait of the Maveriviricetes class (formerly Lavidaviridae family).

As suggested, we updated the terminal inverted repeats (TIRs) trait for virophages to “Present in some members” to account for the Rumen virophages described by Yutin, Kapitonov and Koonin (2015, doi: 10.1186/s13062-015-0054-9).

Additional changes:

1) Figure 1 has been updated and now shows a polytomy between Mavericks 1/2 and PLVs. This reflects more closely the conceptual framework for our analyses since the specific branching of these groups was not specified in the phylogenetic models.

2) We have added an Acknowledgements section to the end of the manuscript:

Acknowledgements

We wish to thank Peter Simmonds and Alexander Suh for their critical reading and comments on the manuscript, which served to improve this work. We also thank the reviewers for their recommendations and feedback. This work was supported by a doctoral scholarship (Dr. Jose Gregorio Hernandez Award) to JGNB made by the National Academy of Medicine of Venezuela and Pembroke College, Oxford.

Author Response

“It is unclear whether the Ter sites integrated by a single copy plasmid have any effect on the replication of this region but the data show that the observed effects are dependent on expression of the Tus protein.”

-The lack of perturbation of the TerB sequence on fork progression has extensively been studied previously in both Willis et al, 2014 and Larsen et. al, 2014. Furthermore, as the detection of the SMARD signal at the TerB sites is dependent on the 7.5kb probe that spans the TerB sites (orange probe, Fig 2B & 2D), it would be impossible to study the effect on replication in this region, with and without the integration of the single copy plasmid.

“The SMARD data do not reveal what proportion of forks are arrested at Tus/Ter, or how long the fork delay is imposed.”

-The percentage of fork stalling at the TerB sites, with and without Tus expression, has been quantified in Figure 2E & 2F. Essentially, 36% forks stall at the TerB block, i.e. 18% of the forks stall in both the 5’ to 3’ (orange) and 3’ to 5’ (blue) direction when the Tus-TerB block is active.

“It is not shown whether the replication inhibitor HU leads to the same widely spread gamma H2AX response.”

-While we have not shown gH2AX accumulation via ChIP after HU treatment, Supplementary Figure 5A & 5B clearly show increased gH2AX foci when the cells are treated with HU, suggesting a global replication stress response that is in stark contrast to the response to Tus-TerB.

Author Response

First, we would like to thank eLife and the reviewers for the positive assessment of our manuscript and for providing the medium to expand the scientific dialogue on the present study. We greatly appreciate the reviewers´ assessment and will revise the manuscript considering their input. Please see our provisional response below.

Reviewer #1’s main concern is centered on the evidential strength of the study’s conclusion that age-specific effects of birth weight on brain structure are more localized and less consistent across cohorts than age-uniform, stable effects. More specifically, the reviewer points out the evidence (or lack of such) for age-specific effects. The reviewer specifies four methodological concerns: that #1 no direct statistical comparisons are conducted between samples (beyond the spin-tests) and that #2 the differential composition of samples in terms of age distribution leads to the possibility that lack of results is explained by methodological differences. Further, he/she adds that #3 some datasets have a narrow age range precluding detection of age-related effects and #4 that the modeling strategy does not allow for non-linear interaction between age and BW suggesting the use of spline models instead in a _mega-_analytical fashion. Reviewer #1 also asks for greater clarity regarding the statistical models and the provision of effect-size maps.

As a response to the reviewer’s comments, we will submit a revised version of the manuscript with a revised description of the statistical models and submit effect-size maps in an accompanying repository. We will tackle the first two concerns by performing additional statistical analyses. The planned analyses will include across-sample reliability and within-sample reliability for the time*birth weight analyses. These analyses will address the concerns that the lack of age-specific effects is due to sample differences rather than a lack of biological effects (#2) and provide further explicit statistical comparisons across samples (#1).

We do not believe concern #3 is critically problematic since time*birth weight refers to a within-subject contrast, e.g. longitudinal-only based contrast. Birth weight, even when self-reported is a highly reliable measure and the sample sizes are relatively large (n = 635, 1759, and 3324 unique individuals). Note that the smaller dataset does have longer follow-up times and more observations per participant increasing the reliability of estimations in individual change. Structural MRI measures have very high reliability (often ICC > .95). Clearly, longitudinal brain change is less reliable, yet the present sample size and the high reliability of birth weight should provide enough statistical power to capture even small time-varying effects of birth weight on brain structure.

We agree that some - if not most - brain structures follow non-linear trajectories throughout life (#4). In the present study, age regressors are used only for accounting for variance in the data rather than capturing any effect of interest. Rather, it is the time*birth weight regressor that captures age-varying changes in brain structure. Time reflects within-subject follow-up time. In any case, the inclusion of non-linear regressors is superfluous in ABCD given the small age range and will not  regress out additional variance. Likewise, published data on UKB has shown that most age trajectories of cortical data follow grossly linear fits (age range in the UKB: 45 – 85 years). In LCBC, however, we recognize that non-linear modeling may be able to capture additional variance since it is a lifespan dataset. This may provide better estimates for birth weight and time*birth weight though it is unlikely that it has a big effect on the slope. In any case, #4 is a valid concern and we will repeat the main analysis using «spline models to fit age We will not follow the reviewer's suggestion for a mega-analytical approach. It is a valid approach, but it is best suited to other configurations of data than those used here where we used longitudinal data only. The use of cross-sectional data for deriving differential age trajectories is also not without problems.

We see Reviewer #2´s  concerns mainly being: #1 The translation of birth-weight effects on brain volume to years of aging is inadequate and not fully supported by the data. #2 Lack of data regarding the functional relevance of brain weight effects on brain structure; suggesting the inclusion of cognitive data and (birth weight – brain structure – cognition) mediation analyses. #3 The degree to which the association between birth weight and cortical area/volume is explained by overall somatic growth. #4 The lack of non-linear regressors for fitting age.

1 We will modify the section on translation of birth weight effects into years of aging. While it is often used - see for example the literature on brain age - and is relatable, it can be easily misinterpreted and does not reflect any objective measure of effect size. #2 The degree to which birth weight relates to cognition throughout the lifespan through individual variations in brain structure is a key question in the field for which currently we only have indirect evidence. Unfortunately, the study of this triple relationship is out of the scope of the current study. This, we agree with the reviewer, warrants further research. We disagree, however, in that mediation analyses are able to provide a satisfactory answer to this question. Also, we believe that the conclusions of the present study are not affected by the omission of cognitive data beyond what has been commented on in #1. The relationship between birth weight, structure, and cognition will be further discussed in the revised version of the manuscript.

3 Indeed, part of birth weight effects on brain structure can be explained by overall somatic growth. This study does not provide - and is not designed to provide - a specific mechanism for which birth weight is associated with brain structure. Yet, research has also shown that differences in birth weight partially reflect factors associated with variations in neurodevelopmental processes in gestation. Mechanisms are thought to be partially genetic (including geneticly influenced potential for somatic growth) and partially environmental prenatal influences. The amount of variance explained by such mechanisms is unknown and may differ across samples (e.g. population samples vs. monozygotic twins). In our view, the association is no less meaningful if mostly attributed to somatic growth if it is associated with partly modifiable variance in brain growth. The revised manuscript will include further considerations on the possible mechanisms explaining the association between birth weight and cortical structure throughout the lifespan.

Concern #4 is equivalent to Reviewer #1, concern #4. In short, the age regressor is used only to account for variance. Linear modeling in the ABCD and UKB datasets has been shown to fit relatively well the different cortical regions due to the samples’ narrow age range. In LCBC, however, we recognize that non-linear modeling may be able to capture additional variance since it is a lifespan dataset. This may provide better estimates for birth weight and time*birth weight though it is unlikely that has a big effect on the slope of the effects. We will repeat the main analysis using «spline models to fit age.

Author Response:

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

We thank the reviewers for their positive remarks, which we have addressed in detail below and which we have considered in our revised manuscript.

Reviewer #1 (Recommendations For The Authors):

The authors claim several times to have documented electrogenic chloride/oxalate exchange mediated by human SLC26A6. However, they fail to detect whole cell currents in SLC26A6-expressing HEK293 cells in oxalate bath, despite robust, saturable Cl- efflux from proteoliposomes into extracellular oxalate solution, as detected by AMCA fluorescence decay.

We interpret the low, and essentially non-detectable currents for Cl-/oxalate exchange as a consequence of the slow kinetics of transport. This lack of sensitivity is not unusual for electrogenic secondary-active transport processes recorded by patch-clamp electrophysiology in mammalian cells, which renders the recording in large X. laevis oocytes by two-electrode voltage clamp the preferred method for such investigations. In contrast to the non-detectable activity in electrophysiology, the pronounced signal in the ACMA assay reflects the influx of H+ as a consequence of the negative membrane potential established by the influx of the divalent anion oxalate, which we assume to occur in exchange with the monovalent Cl-.

Instances in the manuscript include:

Abstract Line 17 overstates the paper's findings as "we have characterized SLC26A6 as a strictly coupled exchanger of chloride with either bicarbonate or oxalate". To the extent that "strictly coupled" implies 1:1 stoichiometry, the authors conclude Cl-/bicarbonate exchange is electroneutral based on its lack of exchange current. In contrast, the lack of Cl/oxalate exchange current does not lead the authors to the same conclusion of electroneutrality for Cl-/oxalate exchange. The data presented do not measure the stoichiometry of Cl-/oxalate exchange.

We agree that our ACMA experiments do not strictly discriminate between coupled and uncoupled oxalate transport. However, it should be emphasized that, assuming that transport proceeds by an alternate access mechanism, uncoupled oxalate transport would require the change of the unloaded transporter between inward- and outward-facing conformations, which was shown to be unfavorable in Figure 1D.

We have reworded the sentence in the abstract to:

“Here we have determined the structure of the closely related human transporter SLC26A6 and characterized it as a coupled exchanger of chloride with bicarbonate and presumably also oxalate.”

Line 264 claims that "the paper's functional data has defined SLC26A6 as a coupled transporter that exchanges Cl- with either HCO3- or oxalate at equimolar stoichiometry."

We have changed the sentence to:

“Whereas our functional data has defined SLC26A6 as a coupled antiporter that exchanges Cl- with HCO3- and presumably also oxalate with equimolar stoichiometry…”

In lines 299-302, the authors claim to have "detected strict equimolar exchange of anions"...leading to the reasonable conclusion of electroneutral Cl-/HCO3- exchange and the reasonable but unsupported conclusion of coupled Cl/oxalate exchange.

We have reworded the sentence to:

“In the case of Cl-/HCO3- transport, we detect a strict equimolar exchange of anions binding to a conserved site in the mobile core domain of the transmembrane transport unit (Figure 4B, H). Although not shown unambiguously, we assume an analogous mechanism also for Cl-/oxalate exchange.”

Lines 505-508 in Methods claim that the AMCA proteoliposome assay "measured electrogenic oxalate transport." However, the assay documented extracellular oxalate- dependent anion transport that was most simply interpreted as coupled exchange.

The assay has detected H+ uptake into proteoliposomes as a consequence of electrogenic anion influx. In these experiments, oxalate is the only anion on the outside of vesicles and it requires to be transported to be able to observe any fluorescent change. The claim of electrogenic oxalate transport is thus justified. As described above, the assay does under the applied conditions not discriminate between uncoupled and coupled oxalate transport, however uncoupled oxalate transport would require the conformational change of an unloaded transporter, which was shown to be kinetically disfavored.

In contrast, other parts of the manuscript acknowledge that the evidence presented falls short of documenting stoichiometric chloride/oxalate exchange.

Results Line 151 sets out to "investigate a potentially electrogenic Cl-/oxalate exchanger. Similarly, results line 160 conservatively claims that Cl-/oxalate exchange occurs "presumably" with a 1:1 stoichiometry. This more accurate language needs to be used throughout the paper, replacing the more absolute but unjustified descriptions summarized earlier above.

We have now introduced the requested clarifications throughout.

I have otherwise only Minor points to suggest. Abstract:

"Among the eleven paralogs in humans.... ". This should be "at least 10," as the original

status of human SLC26A10 as a transcribed pseudogene vs. a truncated protein-expressing gene remains unresolved. The authors recognize this in the introduction, where on p. 3 they acknowledge "ten functional SLC26 paralogs in humans."

We have changed to ‘ten functional paralogs’

Introduction:

p. 4 line 45: membrane-inserted

We have introduced the correction.

Methods:

Construct Generation:

p. 25 lines 380-2: Add a sentence describing any C-terminal sequence extension added after C3 cleavage product, and whether/how it modified the PDZ-binding domain sequence. Has the modification been tested for PDZ-binding activity?

We have introduced the following sentence:

“As a consequence of FX cloning, expressed constructs include an additional serine at the N- terminus and an alanine at the C-terminus. Following C3-cleavage, SLC26A6 carries a seven residues long C-terminal extension (of sequence ALEVLFQ).”

We have not tested PDZ-domain binding but expect that the added residues interfere with interaction with the C-terminal binding motif.

Liposome Reconstitution:

p.28: lines 453-4: Please clarify the meaning of: "absorbance at 540 nm was used to detect liposome destabilization," followed immediately by "After the formation of stabilized liposomes".... Does destabilization mean liposomal leak of Eu.L1+ chromophore, with decline of absorbance? What is practically meant in terms of the number of 10 mL additions of 10% TTX-100 routinely added to generate stabilized liposomes without generating destabilized liposomes? Did this number vary from trial to trial? How did you know when to stop adding aliquots of TTX-100?

We have added the following sentence:

“For protein incorporation, 10 µl aliquots of 10% Triton X-100 were added in order to destabilize liposomes and permit protein incorporation. After reaching a plateau of the light scattering measured at 540 nm, 4 additional aliquots of Triton X-100 were added. The number of additions required for destabilization did not vary between reconstitutions.”

p.28 line 463: "dissolved" should be "suspended."

We have introduced the correction.

Bicarbonate Transport Assay

p. 29 line 480-1. How many repetitions represented by the phrase "sequential ultracentrifugation steps"- please provide a number or a range, as applicable.

We have defined the number of ultracentrifugation steps (two).

Pp 29-30, lines 485-7: define "cycles" - are these fluorometric excitation-emission cycles?

We have defined cycles as fluorometric excitation/emission cycles.

p. 30 line 489: delete "by"

We have deleted ‘by’

Name the fluorimeter used.

We have named the fluorimeter used as Tecan Infinite M1000 Pro microplate reader.

AMCA assay

Pp 30-31, lines 505-8: Add composition of extraliposomal oxalate-containing buffer. In Fig 1 Suppl Fig. 1 panels H and I, and Methods lines 505-508, with 150 mM oxalate substituting for 150 mM Cl- how was osmotic balance maintained in the external chloride solution?

We have added the composition of the oxalate-containing buffer. The osmolarity of the extracellular solution was not balanced.

Electrophysiology

p.32 line 532: What fold-increase of SLC26 protein levels was produced by inclusion of 3 mM valproic acid?

We consistently see an increase of expression upon addition of valproic for different membrane proteins acid but did not quantify it in this case.

Results:

Functional characterization of SLC26A6

Line 91: "comparably" to what? Otherwise, perhaps, "comparatively" was intended here?

We have changed to ‘comparatively’

Fig 1E legend: line 763 "time- and concentration-dependent". Same for line 791, line 799

We have introduced the correction.

Fig. 1G: Change Y axis legend to "Normalized [Eu.L1+] emission." Add bath ion composition for "neg" condition (black trace).

We have corrected the label on the Y-axis and added ion composition for neg.

Fig. 1H legend sentence 2 "in a concentration-dependent manner for liposomes (Mock) in

75 mM oxalate (n=5) and for SLC26A6 proteoliposomes in extracellular oxalate concentrations of 9.4 mM (n=3) etc

We have reworded the sentence:

“Traces show mean quenching of ACMA fluorescence in a time- and concentration-dependent manner for SLC26A6 proteoliposomes with outside oxalate concentrations of 9.4 mM (n = 3), 37.5 mM (n = 5), 75 mM (n = 6), 150 mM (n = 8, all from two independent reconstitutions). Neg. refers to liposomes not containing SLC26A6 assayed upon addition of 75 mM oxalate as defined in Figure1-figure supplement 1G.”

Fig. 1 Fig Suppl. 1. p.45 line 790: change "chemical formulas" to "2-D chemical structures"

We have introduced the change.

lines 799: Time-

We have introduced the change.

Fig 1 Fig Suppl. 1. p 46 lines 809-810: dashed lines indicating 0 pA are indeed red in panels A and B, but black in panels H and I.

We explicitly refer to recordings, where dashed lines at 0 pA are consistently in red.

Fig 2 Fig Suppl.1 p. 50, line 832: Two additional multi-class.

We have introduced the change.

Fig 2 Suppl Fig 2B, p. 51: Please label the residue numbers of the side chains coordinating the chloride binding site. Can those residues be indicated in Fig 2 Suppl Fig 2A in the appropriate helices? These residues might also be asterisked in the primary sequence alignment of Fig 2 Suppl Fig 3A.

We have labeled the residues in Fig. 2-figure supplement 2B but not 2A where the focus is on the general quality of the density in different parts of the protein. We have also labeled the same residues in Fig. 2-figure supplement 3A.

Fig.4 legend p. 59 line 882 – Deviating residues in SLC26A9 (typo A6) are highlighted in violet.

We have introduced the correction.

p. 60 lines 888-9: Please clarify the individual meaning of green and purple asterisks on defining the substrate cavity diameter; How do purple and green asterisks relate to the yellow and green lines in the graph? Should the asterisks be two green and two yellow asterisks, or should they be black? What is the meaning of the purple and green asterisks at the two upper corners of panel G with respect to the substrate cavity radii?

Please specify if y axis label "radius" refers to substrate cavity radius, and whether X axis label "distance" refers to axial distance along helix alpha10, alpha1, or of the helical pair. Is value "0 A" on the X axis anchored at the top of the helices as depicted in panels D-F? Is X-axis value 10.5A sited at the bottom of the helices? Please indicate on the panel G curves the x-y value range depicted in the inset images- or clarify that the inset images present the entire curves of panel G.

We have clarified these remarks in a revised legend:

“The radius of the substrate cavity of either protein is mapped along a trajectory connecting a start position at the entrance of each cavity (distance 0 Å) and an end position located outside of the cavity in the protein region (distance 10 Å). Both points are defined by asterisks in insets showing the substrate cavities for either transporter and they are indicated in the graph (green, cavity entrance towards the aqueous vestibule; violet, protein region).”

Fig. 4 p . 61-2 panel H and lines 907-8: Addition to the panel of the A5 "buried Cl- binding site" would be helpful, if possible to do without obscuring the A6 and A9 Cl-s.

Panels H and I show the cavity harboring the ion binding site in two orientations, including the surrounding residues. We prefer to show all surrounding residues for both orientations, even if this somewhat obscures the view on the ion in the left panels. An unobscured view of the ion in its cavity is provided in panels D and E.

p. 12. Results line 234: Please specify that "both proteins" here refers to A6 and A5 vs A6 and A9

We have specified this.

p. 13 Results lines 268-72: R404 is "ubiquitous in other mammalian paralogs..." should be changed to "shared by most but not all mammalian paralogs".

We have changed the text accordingly.

Fig 4C should have a red or purple asterisk placed under the yellow column corresponding to R404 of SLC26A6, so that the discussion can refer to it. It would also be helpful to remind the reader that R404 corresponds to conserved position 6 in Fig 4 Fig Suppl. 1 panels   D-G.

Here the authors might note that sulfate -transporting SLC26A1 and -A2 have the shorter side chain K residue.

We have marked the position in Figure 4C with an asterisk and added the following sentence to its legend:

Asterisk marks position that harbors a basic residue in all family members except for SLC26A9 where the residue is replaced by a valine. Whereas most paralogs, including the ones operating as bicarbonate exchangers, have an arginine at this site, the sulfate transporters SLC26A1 and 2 contain a smaller lysine.

We have added the following statement to the legend of Figure4-figure supplement 1:

“‘6’ indicates the position which contains a basic residue in all family members except for SLC26A9.”

Fig 5B legend p. 63 line 916. Please specify if the 14 independent experiments include both the symmetric Cl- conditions and the asymmetric Cl-/HCO3- conditions or only one condition.

The 14 independent experiments were only recorded in symmetric chloride conditions. We have changed the legend accordingly.

Fig 5C. It would be useful for readers to add the I-V trace of WT SLC26A6 taken from Fig 1 Suppl 1B (perhaps in gray), to document the specificity of the very low magnitude R404V whole cell current. Alternatively, please note (if the case) that WT SLC26A6 currents (Fig 1 Supple 1B) are indistinguishable from the blacked dashed zero current density line.

We have now displayed the I-V trace of WT SLC26A6 as grey dashed line for comparison and added a new panel that show the differences between the currents of R404V and WT recorded at 100 mV (Fig. 5D). Although the currents for R404V were consistently lager than for WT, the difference is not statistically significant. We have explicitly mentioned this in the text and the figure legend.

Fig 5E depiction of decline in ACMA fluorescence is missing from the legend. Legend references to panels E and F seem to correspond to Fig 5 panels F and G (lucigenin fluorescence), as noted in Results p 14 lines 280-3.

We have added the legend.

Chernova et al (2005) reported electroneutral human and mouse A6-mediated Cl/HCO3- exchange in Xenopus oocytes. They also observed electrogenic Cl-/oxalate exchange by mouse SLC26A6, but detected no current generated by human SLC26A6-mediated Cl-/Oxalate exchange. That paper (already cited) might be referred to more explicitly in connection to the authors' current findings of electroneutral Cl-/HCO3- exchange by human SLC26A6 as well as their inability to detect human SLC26A6-mediated Cl-/oxalate exchange current in HEK-293 whole cell recordings.

We now have included the reference in the discussion:

“Consequently, transport would be electroneutral in case of the monovalent HCO - and electrogenic in case of the divalent oxalate (Figure 1E-H), which was already proposed in a previous study (Chernova et al., 2005).   We also want to re-emphasize that the inability to measure discernable currents does not necessarily imply that the transport might not be electrogenic as, due to their slow kinetics, transport-mediated currents might be below the detection limit of patch-clamp electrophysiology.”

Reviewer #2 (Recommendations For The Authors):

-  It would be helpful if the authors briefly clarify the depiction/scheme of the hypothetical SLC26A6 outward-facing conformation. Is this gleaned from a prior structure of a related SLC family or distant homolog? Functional data? Biochemical/biophysical data? As well, I would also recommend labeling this within the figure (Figure 3-figure supplement 1D labeling, for instance - inward, hypothetical outward).  

As mentioned in the legend of Figure 3-figure supplement 1D, the outward conformation is hypothetical. We have also now mentioned this in the title. The displayed outward- conformation was constructed by manually moving the area depicting the core domain relative to the fixed gate domain.  

-  Have the authors attempted to block SLC26A6-mediated transport with the addition of a known inhibitor, such as niflumic acid? I understand that this may be technically challenging, but it would strengthen the transport assay data, especially in Figure 5D with the ACMA assay testing the SLC26A6 R404V mutant.  

We have not attempted to block the currents by addition of niflumic acid.  

-  It could be helpful to the reader to move the schematics in Figure 1-figure supplement 1C into Figure 1.  

We have now displayed the schematics illustrating the principle of the respective transport assays next to the data in Figure 1, but kept Figure 1-figure supplement 1C for a more detailed description of the assays.  

-  Figure 3-figure supplement 1D legend, should be "hypothetical" instead of "hypothetic."  

We have introduced the correction.

-  I might consider coloring the Cl- ion something that is distinct from the model colors that are used in the figures (see Figure 4 and Figure 4-figure supplement 1). This would help to clarify Figure4-figure supplement 1H, where I believed that the Cl- ions at first were from the SLC26A6 model at first glance.

We have used the green color for chloride throughout the manuscript and would prefer to keep it that way for consistency.

-  Labeling in Figure 5 legend (E, F) do not match the Figure (F,G). The description of the ACMA assay is absent from the figure legend (the real Figure 5E).

This has been corrected.

Reviewer #3 (Recommendations For The Authors):

None.  This  is  a  well-done  manuscript  and  I  have  no  further  suggestions.

Author Response

Reviewer #3 (Public Review):

In this manuscript, Man et al. describe a new signaling pathway for regulation of the voltage-gated calcium channel Cav1.2 and show that it can modulate synaptic plasticity in the hippocampus. Studies with specific inhibitors, phosphopecific antibodies, and gene knockdown show that activation of alpha-1 adrenergic receptors induces downstream activation of the serine/threonine protein kinase PKC and the tyrosine protein kinases Pyk2 and Src, which bind to the Cav1.2 channel through its large intracellular segment connecting domains II and III. This signaling complex leads to tyrosine phosphorylation of Cav1.2 and increased channel activity. Block of this novel signaling pathway in hippocampal slices with specific inhibitors of Pyk2 and Src reduced a specific component of long-term potentiation whose induction requires Cav1.2 channel activity.

This work is an important advance, as it presents a novel signaling pathway through which the ubiquitous neurotransmitter norepinephine and the neurohormone epinephrine can regulate synaptic plasticity, attention, learning, and memory. The experiments are comprehensive, carefully done, and clearly presented. The authors should consider revisions and responses to the points below.

1) Figure 2B, D. Inhibitors reduce Ica below control. Is there endogenous stimulation of this regulatory pathway under control conditions?

We now explicitly state in the Discussion: “Inhibitors of PKC, Pyk2, and Src reduce under nearly all conditions Cav1.2 baseline activity and also tyrosine phosphorylation of Cav1.2, Pyk2, and Src even when activators for alpha1 AR and PKC were present. Especially notable is the strong reduction of channel activity way below the control conditions by the Src inhibitor PP2 as well as the PKC inhibitor chelerythrine in Figure 2C. This effect is consistent with PP2 strongly reducing down below control conditions tyrosine phosphorylation of Src (Figure 8J), Pyk2 (Figure 8L), and Cav1.2 (Figure 9E) even with the PKC activator PMA present. These findings suggest that Pyk2 and Src experience significant although clearly by far not full activation under basal conditions as reflected by their own phosphorylation status, which translates into tyrosine phosphorylation of Cav1.2 under such basal conditions.” Because there are multiple ways Pyk2 and Src can be activated including Ca influx and cell-matrix interactions, defining the cause of this baseline activity has to remain beyond the scope of the current work.

2) As noted by the authors, it would be interesting to know if peptides from the linker between domains II and III block this signaling pathway. This would be an important result because, without this information, it is not clear if this is the correct functional site of interaction for this regulatory complex.

Briefly, we were not able to identify shorter loopII/III-derived peptides that would constitute the Pyk2 binding site and thus cannot displace Pyk2 from loop II/III either acutely with peptides or through mutagenesis of the binding site.

3) Figure 4B. The Brain IP for Src has a weak signal. The authors should replace this panel with a more convincing immunoblot.

We provide now the uncropped version in the raw dataset, which clearly illustrates clean, monospecific detection of the Src band over the full length of the blot. Also, please note that earlier work already reported that showed that Src binds to the C-terminus of Cav1.2 (Bence-Hanulec et al., 2000).

4) Scatter plots are provided for the electrophysiological results but not immunoblots. For immunoblots that are quanitified, it would be valuable to add a scatter plot of the replicates.

We now also provide scatter plots for the biochemical analysis.

Author Response

Reviewer #1 (Public Review):

The paper addresses why and how odor discrimination ability achieved after learning occurs in select contexts. The finding is that two related odors trigger near identical Kenyon cell responses when tested in isolation, but trigger different responses to the second odor if these are experienced in sequence within a small temporal window. The authors argue that this template comparison requires some activity downstream of Kenyon cells, that is recruited by MBONs. Overall, the experiments provide very nice physiological evidence for a neural mechanism that underlies a contextual basis for the precision of memory recall.

The experiments were well designed and done. The findings are interesting, but the pitch (e.g. the last paragraph of the discussion and the title of the paper) seems to both ignore the main finding of the paper and overstate the novelty of the idea that memory recall can be flexibly regulated by context. There should be more space dedicated to clearly articulated statements/descriptions of hypotheses and candidate mechanisms to explain the interesting phenomenon described here. For instance, explaining "enhanced template mismatch detection" by potential " real-time and delay line summation" of MBON activity is not super useful for the reader as seems to use one abstraction to explain another. The authors cite Lin et al, 2014 from Miesenbock's lab which shows a key role for GABAergic APL neurons in discrimination. Is there increased activation of APL neurons when similar odourants are being compared and discrimination is required? This seems like a simple physically embodied mechanism that could/ should be examined.

Overall, I think the idea that memories are recalled with high precision (less generalisation) only when increased precision is demanded, is a fact that sure is well appreciated by behavioral biologists even beyond the two papers cited here (Campbell et al., J Neurosci 2013; Xu and Südhof, Science 2013). The new findings fill in a physiological gap in this phenomenology. I think the paper would be greatly improved if the authors highlighted what and focused on the physiological correlate uncovered, and tried to communicate (or test) possible mechanistic origins for this in more physically accessible terms.

We thank Reviewer #1 for their appreciation of our findings. We are grateful for this extremely constructive feedback on re-focussing the pitch of the paper and have extensively revised the manuscript along these lines, particularly changing the Title, Introduction and Discussion. As suggested, we now highlight how similar stimuli can be categorized together, or apart, depending on the stimulus choices animals are presented during recall.

Reviewer #2 (Public Review):

One of the key questions in circuit neuroscience is how learned information guides behavior. Modi et al. investigated this question in Drosophila's mushroom bodies (MBs), where olfactory memory traces are formed during pavlovian olfactory conditioning. They have used optogenetics to restrict the formation of memory traces in selective output compartments of the Kenyon cell (KC) axon terminals, the principal intrinsic neurons of the MB, and tested how flies use these 'minimal memories' during learned olfactory discrimination. They found that memory traces formed in some compartments support discrimination between similar odor pairs, whereas others do not. They then investigated the neural basis of this difference by comparing the responses of relevant output neurons (MBONs) to similar and dissimilar odor pairs. They discovered that MBONs' responses could predict behavioral outcomes if odor presentation profiles during calcium imaging mimic olfactory experience during behavior. This paper and previous works support the idea that flies use olfactory memory templates flexibly to suit their behavioral needs. However, one key difference between this paper and the previous works is the site of discrimination. While previous studies using intensity discrimination have pointed towards spike-latency and on and off responses of the KCs as the main mechanism behind discrimination, Modi et al. have not detected any response difference for similar odor pairs among the KCs. Therefore, they concluded that a hitherto unknown mechanism creates these context-specific responses at the MBONs. The findings will advance our understanding of how memories are recalled during behavior. However, the authors need to bolster their data by including some critical controls that are currently missing.

We thank Reviewer #2 for highlighting how our work contributes to the literature and for pointing out the gaps in our discussion of previous work, as well as the missing controls.

Reviewer #3 (Public Review):

This manuscript by Modi et al represents a novel and significant advance in the neurobiology of memory retrieval. The authors employ a novel behavioral paradigm in order to investigate memory generalization and discrimination. They investigate the role of two different populations of dopamine neurons (DANs) targeting different compartments involved in aversion learning, i.e. α3 (MB630B) and γ2α'1 (MB296B).

The behavioral platform is clear and convincing but lacks natural reinforcement comparisons. The entire paper uses optogenetic reinforcement of said DAN populations.

The authors identify that the gamma DANs can enable both easy and hard odor discrimination, while the alphas DANs can only do easy.

The odors can be separated by calcium imaging analysis of Kenyon cells. Subsequent calcium imaging of the gamma DANs themselves showed that a single training event was insufficient to enable easy odor discrimination at the gamma DAN level, but strangely not for the hard discrimination that gamma DANs can mediate. Seemingly, this is due to the lack of the temporal contiguity of odors (present in behavioral experiments but not in the initial imaging experiments. However, in gamma DANs, Odour transitions enabled discrimination of odors in hard discrimination, based on the depression of calcium activity in DANs after training that was odor-specific. The same was not true for alpha DANs, though the authors used natural electric shock pairings instead of optogenetic stimulation of DANs for the alpha experiment. However, statistical comparisons are done within group and need also be provided for between the groups for both pre and post-training. The authors persuasively show that hard discrimination can only happen in transitions. They also argue that the same engram can be read in two different ways. This is convincing overall, but they claim it is happening downstream of the Kenyon cells just because they do not see it in the Kenyon cells, and I cannot comment on the modeling in Figure 5 (expertise).

Experimental methods used are appropriate, as are data analysis strategies.

The manuscript itself is well written in parts, though at times paragraphs are quite patchy, especially in the discussion. There are also a visible number of typos. The figures are well constructed, and generally well organized. The overall document is concise and has sufficient detail.

We appreciate the reviewer’s comments on the novelty and significance of our study.

Author Response

Reviewer #2 (Public Review):

In Rey et al., the authors goal was to characterize the development of a myelin-like (lacunar) expansion of glial membrane in Drosophila. Although myelin is largely considered a vertebrate innovation, there are a handful of invertebrate models that have been described with glial-derived "myelin," though these systems are not amenable to the same genetic control as Drosophila. To that end, the authors first newly-developed genetics and antibodies to characterize the presence of an axon initial segment (AIS) for adult Drosophila motor neurons that is present at the border between the central and peripheral nervous systems. They show that both sodium (Para) and potassium (Shal) channels, which are typically enriched at the AIS in mammalian neurons, are enriched at this border specifically on motor neurons. They then used multiple types of transmission electron microscopy to visualize this region and found that along with clustering of channels, there is an expansion of membranes from wrapping glia that is reminiscent of myelin. At times, this expansion spirally wraps around larger axons. Finally, they show that genetic ablation of wrapping glia results in an upregulation and redistribution of Para.

Major strengths of this manuscript include the creation of new genetic tools for visualization of subcellular features (e.g. channels) by both light microscopy and electron microscopy.

While this manuscript provides an interesting set of data, but suffers from a lack of quantification and annotation to allow the reader to judge whether this is a robust phenomenon. To increase the reader's confidence in these studies, substantially more quantification of the data is required.

Furthermore, to improve the accessibility of this manuscript, I have the following suggestions:

1) Please label the panels throughout the figures with an abbreviated genotype and what the fluorophores signify. Similarly, the presence of scale bars in uneven across the figures.

This was all corrected.

2) For panels where only one channel is shown, please show these in black and white, which is easier for the visually-impaired.

We have not done this, since the color adds another layer of information (e.g. paramCherry is in magenta, whereas anti-Para staining is in green) which in our view helps to make the complex figures easier to understand.

Author Response

Reviewer #1 (Public Review):

Inhibition of translation has been found as a conserved intervention to extend lifespan across a number of species. In this work, the authors systematically investigate the similarities and differences between pharmacological inhibition of protein synthesis at the initiation or elongation steps on longevity and stress resistance. They find that translation elongation inhibition is beneficial during times when proteostasis collapse is the primary phenotype such as proteasome dysfunction, hsf-1 mutants, and heat shock, but this intervention does not extend the lifespan of wt worms. While translation initiation inhibition extends the lifespan of wt worms and heat shock, but in an HSF-1 dependent manner. This work shows that a simple explanation of just inhibiting total protein synthesis and reduced folding load cannot explain all of the phenotypes seen from protein synthesis inhibition, as initiation and elongation inhibition repress overall translation similarly, but have different effects depending on the experiment tested. Using multiple interventions that target both initiation and elongation lends further support to their findings. These experiments are important for conceptualizing how translation inhibition actually extends lifespan and promotes proteostasis.

Major Comment:

The authors acknowledge that lifespan extension must not necessarily arise just from reducing protein synthesis, as elongation inhibition reduced protein synthesis but did not extend lifespan. Yet for the converse effects from elongation inhibition they seem to suggest that it arises from reducing protein synthesis. For example, regarding how elongation inhibition extends lifespan in an hsf-1 mutant, the authors suggest that "inhibition of elongation lowers the production of newly synthesized proteins and thus reduces the folding load on the proteostasis machinery", even though initiation inhibitors do not extend lifespan in an hsf-1 background (while presumably lowering the production of newly synthesized proteins).

Thank you for this excellent comment. It led us to conduct a crucial experiment with a new finding that is now Figure 6. As suggested, we asked if initiation inhibitors lower the concentration of newly synthesized protein in the hsf-1(sy441) background. The surprising answer is that initiation inhibitors lower the concentration of newly synthesized proteins in N2 but dramatically increase it in hsf-1(sy441). The failure to lower the concentration of newly synthesized proteins was true for the pharmacological inhibitors as well as RNAi against ifg-1. Therefore, inhibition of initiation requires HSF1 to lower the protein concentration. These new findings enable us to make a much more precise statement now added to the discussion:

Lines 372: “The inability of translation-initiation inhibitors to reduce the concentration of newly synthesized proteins in hsf-1(sy441) mutants and the inability to extend their lifespan shows that lowering the concentration of newly synthesized proteins is necessary for the beneficial effects. On the other hand, the finding that elongation inhibitors protect from proteotoxic stress but does not extend lifespan shows that lowering the concentration of newly synthesized proteins is sufficient to protect from proteotoxic stress but is not sufficient to extend lifespan in wild-type, which appears to require selective translation.”

Reviewer #2 (Public Review):

In this manuscript, Clay et al. investigate the underlying effects of reduced mRNA translation beneficial on protein aggregation and aging. They aim to test two pre-existing hypotheses: The selective translation model proposes that downregulation of overall translation increases the capacity of ribosomes to translate selected factors that in turn increase stress resistance against toxicity. The reduced folding load model suggests that during high mRNA translation rates, newly synthesized peptides and proteins can overwhelm the protein folding capacity of the cell and therefore cause protein toxicity. By generally lowering mRNA translation, lower loads of newly synthesized proteins should cause less protein folding stress and hence protein toxicity.

To understand how reduced mRNA translation mediates its beneficial effects in the context of the proposed models, the authors use different drugs established previously in other in vitro and in vivo systems to inhibit selected steps of translation. The systemic effects of translation initiation versus elongation inhibition in C. elegans are compared during heat shock, specific protein aggregation stresses and aging. These phenotypes are further tested for dependence on hsf-1, as contradictory data on the effect of translation inhibition during thermal stress in the context of hsf-1 dependency exist.

The data show that inhibition of translation initiation protects from heat stress and age-associated protein aggregation but on the contrary further sensitizes animals to protein toxicity induced by a misfunctioning proteasome. Further, inhibition of translation initiation increases lifespan in WT animals. The survival phenotypes observed during heat shock and regular lifespan assays are dependent of HSF-1, supporting the selective translation model. As stated in the manuscript, these findings themselves are not new, given that similar observations were made before using genetic models. Interestingly, the inhibition of translation elongation protects from heat stress, but, unlike initiation inhibition, also proteasome-misfunction-induced protein toxicity. Both phenotypes were observed to be independent of hsf-1. The authors further find that inhibiting elongation does not reduce protein aggregation in aged worms and does not prolong lifespan in wild-type animals. It does increase lifespan in short-lived hsf-1 mutants, where protein homeostasis is compromised. To a degree, these findings support the reduced folding load model. Overall, from these observations the authors summarize that the systemic consequences of lowering translation depend on the step in which translation is inhibited as well as the environmental context. The authors conclude that different ways to inhibit translation can protect from different insults by independent mechanisms.

Impact, strengths and weaknesses:

mRNA translation and its regulation is one of the most studied mechanisms connected to lifespan extension. However, gaps behind the protective effects of translation inhibition are so far unresolved, as stated by the authors. Therefore, testing existing hypotheses explaining the beneficial effects of translation inhibition is of great interest, not only for C. elegans researchers but a broad community working on the effects of misregulated translation during aging and disease. Overall, the conclusions made by the authors are generally supported by the data shown in this manuscript. However, some major gaps remain and need to be clarified and extended.

Thank you for your generous comments and thorough review.

Reviewer #3 (Public Review):

Clay and colleagues investigate the proteostasis and longevity benefits derived from translation inhibition in C. elegans by examining the impacts of chemical translation initiation inhibitors (IIs) and translation elongation inhibitors (EIs) on thermotolerance, protein folding stress, aggregation and longevity. They observe somewhat distinct impacts by the two chemical groups. IIs increased longevity in wild-type animals in an hsf-1 dependent manner, whereas, EIs only extended hsf-1 mutants' lifespan. Only EIs protected against proteasome dysfunction. Both protected against heat stress but with differing hsf-1 dependence. The authors utilize these observations to derive conclusions regarding two dominant points of view on the mechanism by which translation inhibition improves lifespan and proteostasis.

The study is based on interesting observations and several promising avenues of further investigation can be identified. However, the manuscript appears somewhat preliminary in nature, with many of the observations, while interesting, only explored superficially for mechanistic insights. The rationale behind some of the interpretations was also difficult to interpret. For example, the authors make conclusions about 'selective translation' being adopted upon IIs treatment without directly testing this. Protein aggregation, while possibly predictive, is not a reliable readout for selective translation of some mRNAs. Similarly, the evidence for a reduction in 'newly-synthesized protein load' by EIs is thin based on one reporter. Previous studies from the Blackwell lab have identified differential impacts of SKN-1 on select cytoprotective genes' expression and proteasomal gene expression based on inhibition of translation initiation or elongation. So there is precedence for both the differential impact of initiation vs. elongation inhibition as well as genetic background. There are several other such studies that reduce the impact of the observations presented here. With limited novelty and mechanistic insight, the impact of the study on the field is likely to be moderate.

We thank the reviewer for the thorough analysis and candid summary. Some of the criticisms rang true, and we have made considerable efforts to address them, both increasing the thoroughness of our study by establishing that these inhibitors inhibit initiation and elongation (new Figure 1) and by providing a novel mechanism showing that the ability of the initiation machinery requires HSF1 to lower the concentration of newly synthesized proteins.

Before we go into the specific criticisms, we would like to note that of the 30-40 eukaryotic translation inhibitors used in cell culture and yeast, very few have been validated in C. elegans. The go-to inhibitor was cycloheximide which, in our hands, is reliable in cell culture but unreliable in C. elegans, most likely due to its poor pharmacokinetics (data now added to the supplementary figures). To our knowledge, no C. elegans study investigating translation has made sure to equalize the concentration of newly synthesized proteins or could have because of a lack of validation of the chemical tools used in other organisms. Thus, the comment of reviewer #3 that we did not go far enough with the validation struck home, and in the revised version of Figure 1 we added more validation.

We are of the opinion that it is essential to ensure that both mechanisms reduce the concentration of newly synthesized proteins to the same degree to study mechanistic differences. Otherwise, one cannot deconvolute if any phenotypic difference is caused by the mechanistic difference or the degree of translation inhibition. The importance of monitoring the level of inhibition became evident in our new Figure 6, which shows that inhibition of the translation initiation machinery no longer reduces the concentration of newly synthesized proteins but increases it in the absence of HSF1.

Author Response

Reviewer #1 (Public Review):

The role of HCO3 (or possibly CO2) in regulating sACs is well established yet its physiological context is less clear. The heart is indeed an excellent choice of organ to study this. Isolated mitochondria offer a tractable model for studying the model, although are not without limitations. The quality of recordings is very high, as judged by the consistency of results (i.e. lack of clustering between biological repeats). My primary concern is about distinguishing the effect of pH and HCO3. A rise in HCO3 will also raise pH unless this had been compensated by CO2. It is unclear, from the legend or results, if the bicarbonate effect is due to HCO3 or pH. Was pH controlled by matching the rise in HCO3 with an appropriate level of CO2? The swings in pH are likely to be very large and, potentially, a confounding factor. Certainly, there will be an effect on the proton motive force. A more informative test would compare the effect of 0 CO2/0HCO3 at a pH set to say 7.2, 2.5% CO2/7.5 mM HCO3, and then 5% CO2/15 mM HCO3, etc. Control experiments would then repeat these observations over a range of pH (at zero CO2/HCO3) and over a range of CO2 (at constant HCO3). Data for zero bicarbonate are not informative, as this will never be a physiological setting (results claim 0-15 mM bicarb to represent physiology). Importantly, there seems to be no significant difference in 2A between 10 v 15 mM bicarb, i.e. the physiological range.

Thank you for your clear discussion and suggestions. We agree that pH must be controlled in these experiments to avoid the confounding situation you describe. In fact, pH was carefully controlled but this was not described adequately. To make this clearer the methods section was modified.

We agree that bicarbonate concentration is above 0 in living tissue. We used that value only as a reference in Figure 2A to examine which type of adenylyl cyclase (AC) is inside mitochondria, i.e., bicarbonate-activated soluble AC as opposed to transmembrane AC which is not bicarbonate activated. The wording has been corrected to better describe this. The question of what the physiological consequences are require an assay with higher signal-to-noise ratio. In effect, this is achieved in the experiments of Fig. 2B-C which show that physiologically relevant changes in bicarbonate have a large and significant influence on mitochondrial ATP production.

There is also a question on the validity of the model. A rise in respiratory rate will produce more CO2 in the matrix. This may raise matrix HCO3, and stimulate sACs therein, but the authors claim sACs are in the IMS, rather than the matrix. Since HCO3 is impermeable, it is unclear how sACs would detect HCO3 beyond the IMM. CO2 escaping the matrix will enter the continuum of the cytoplasmic space, which has finely controlled pH. Since membranes (including IMM) are highly permeable to CO2, the gradient between matrix and cytoplasm will be small (i.e. you only need a small gradient to drive a big flux, if the permeability is massive). Since CO2 can dissipate over a large volume, it is unlikely to accumulate to any degree. CO2 will be in equilibrium with HCO3 and pH (because there are carbonic anhydrases available). Since the cytoplasm has near-constant pH, [HCO3] must also be close to constancy. It is therefore difficult to imagine how HCO3 could change dramatically to meaningfully affect sACs and hence cAMP. Evidence for major changes in IMS pH in intact cells during swings of respiratory activity would be required to make this point. Indeed, for that reason, it would be more sensible to anchor sACS in the matrix, as there, HCO3 could rise to high levels, as it is impermeable, i.e. could be confined within the mitochondrion. I am therefore not convinced the numbers are favorable to the proposed mechanism to be meaningful physiologically.

The question of how CO2/bicarbonate signaling can work in the intermembrane space (IMS) is explicitly addressed in the revisions in the results and discussion sections. CO2, a membrane permeable gas, can easily cross the IMM (permeability coefficient of 0.01 to 0.33 cm/s). Once in the IMS, CO2 can combine with water to produce bicarbonate, a fast reaction in a physiological context (i.e. mM/s in physiological saline) that can be accelerated to nearly diffusion limits by carbonic anhydrase if present. The assumption that all the “CO2 escaping the matrix will enter the continuum of the cytoplasmic space” is not supported by the structure of cardiomyocyte mitochondria. As illustrated in new Fig 2H, only a small fraction (9-15%) of the IMS occurs along the mitochondrial periphery adjacent to the outer membrane and cytosol. Most of the IMS is contained inside the cristae (the intracristal space or ICS), which in interfibrillar mitochondrion are composed of closely packed extended flat membranes that create scores of alternating layers of matrix and ICS ideal for rapid gas exchange between compartments. The cristae are connected to the peripheral IMS through narrow “crista junction” openings that restrict solute diffusion between the peripheral IMS and ICS. Thus, rather than being dominated by the ionic equilibria of the cytosol, the crista compartments are functionally distinct from the peripheral IMS region and cytosol. We cite recent publications using super-resolution light microscopy in which gradients of 0.3-0.4 pH units (dependent on metabolic state) have been detected along the cristae of respiring cells and between the peripheral IMS and crista interiors. These diffusion effects would likely be even more pronounced for cardiac muscle mitochondria, which have larger, more densely packed lamellar cristae than other cell types. Thus, the microenvironment of the cristae provide confined spaces in close communication with the matrix CO2 production, and ideal for operation of a sAC signaling system.

Reviewer #2 (Public Review):

The authors explore the role of bicarbonate-regulated soluble adenylate cyclase in modulating cardiac mitochondrial energy supply. In isolated rat mitochondria, they show that cyclic AMP (but not the permeable cAMP analog 8-Br-cAMP) increases ATP production via a Ca-independent mechanism at a location in the intermembrane space of the mitochondria, rather than in the matrix, as previously reported. Moreover, they show that inhibition of EPAC, but not PKA, inhibits the response. The effect required supplementing the mitochondria with GTP and GDP to facilitate activation of the EPAC effector GTPase Rap1. The study provides interesting new information about how the heart might adapt to changes in energy supply and demand through complementary regulatory processes involving both Ca and cyclic AMP.

The authors nicely demonstrate that soluble adenylate cyclase is localized to mitochondria. They argue, based on the effects of cyclic AMP, which is accessible to the mitochondrial intermembrane space (IMS) but not the matrix, that the signalling pathway is located in the IMS. They also find that EPAC/Rap1 is the likely downstream effector of cyclic AMP, through yet unknown targets regulating oxidative phosphorylation.

A weakness is that the components of signaling (sAC, EPAC, and rap1) are not definitively localized to a specific mitochondrial compartment using the superresolution imaging methods employed.

Thank you for the concise summary of key findings. While the super-resolution data indicates sAC and CA are localized to the interior of the mitochondria (possibly co-localized in the same subspace), identification of the particular microcompartment is not possible from the imaging data alone. We explain clearly in the manuscript that the functional experiments are critical to the conclusion that the sAC signaling pathway most likely operates in the IMS.

Author Response

Reviewer #1 (Public Review):

This manuscript is interesting because of the exploration of a novel model organisms utilizing next-generation sequencing approaches, such as single-cell-RNA-seq. Despite the authors' efforts the manuscript lacks a cohesive narrative and suffers from being extremely preliminary in nature. For example, most of the figures are cut and pasted directly from the computational programs with very little formatting or thought to creating new knowledge from the data generated. Essentially the manuscript consists of 2-3 experiments where the authors performed single-cell-RNA-seq on different anatomical locations in the pig and also on a couple of different pig types (The Chenghua and Large White). The authors used standard computational pipelines consisting of Seurat, Monocle, Cell Chat, and others to characterize differences in their data.

There is potential in this manuscript but the authors should improve upon the manuscript by mining the data better and generating a better understanding of anatomical positions of pig skin by evaluating the Hox genes.

(1) Thanks for the reviewer's positive evaluation for our article and providing valuable feedback to improve the quality of our manuscript. To provide a more cohesive narrative, we have edited throughout the manuscript.

(2) Meanwhile, we also modified and formatted some figures including Figures 2-6, Figure 4—figure supplement 1 and 2, Figure 5—figure supplement 1 and 2, and Figure 6—figure supplement 1.

(3) We have analyzed these data of regional- or species-based differences more extensively, and the added content are in Result Section of “Heterogeneity of skin FBs in different anatomic sites” and “Heterogeneity of skin cells in different pig populations”.

(4) However, in our study, we did not identify any Hox gene among these differentially expressed genes in skin fibroblasts from both different anatomical sites and different pig populations. The differences of Hox code expression patterns might come from the heterogeneity of different species.

Reviewer #2 (Public Review):

The authors aimed to analyze different dermal compositions of various skin regions, focusing on fibroblast, endothelium and smooth muscle cells. They collect skin samples from six different skin regions of adult pig skin including the head, ear, shoulder, back, abdomen, and leg skins. After dissociating the tissues into single cells, they perform single-cell RNA analyses. A total of 215 thousand cells were analyzed. The authors identified distinct cell clusters, enriched molecules within each cell cluster, and the dynamic of cell cluster transition and interactions. Based on their findings, they conclude that tenascin N, collagen 11A1, and inhibin A are candidate genes for facilitating extracellular matrix accumulation.

Strength:

The methodology they used to prepare scRNA data is appropriate. Bioinformatic analyses are solid. The authors emphasize the heterogeneous phenotypes and composition ratios of smooth muscle cells, endothelial cells and fibroblasts in each skin region. They identify potential cell communication pathways among cell clusters. Expression of selective molecules on tissue sections were done.

Weakness:

While tenascin, collagen and inhibin are highlighted as genes important for ECM accumulation, there is no functional evaluation data. The discussion section is a compilation of comparisons, and is somewhat fragmentary. More significance from this dataset could have been extracted.

(1) We appreciate the reviewer's suggestions for evaluating the functional significance further. In our next research, we will perform some experiments in vitro and in vivo to explore the functions of these identified key genes.

(2) The discussion section have been greatly modified and it shall be more logical and readable.

Author Response

Reviewer #1 (Public Review):

Wu et al. provide a powerful cross-species approach to better understand brain cell-type specific responses to mutant tau and aging. Therefore, they use scRNAseq of established Drosophila models that they had previously used for bulk RNAseq (Mangleburg et al., 2020) at 1, 10 and 20 days of age, which thus allows them to study the contribution of pathogenic tau (R406W-mutant) in isolation in an experimentally highly controllable manner. They find a large overlap between tau-induced and aging-induced deregulated genes, however different cell-types were primarily affected, suggesting that expression of tau does not simply induce accelerated aging. When assessing cell number abundance in response to tau expression the authors noted that certain excitatory neurons were preferentially lost. They then examined innate immune pathways downstream of NFkB, which they had already uncovered in their previous bulk studies to be associated with tau expression. Also at the scRNAseq level, they find these pathways to be deregulated after expression of tau. In addition, in control cell types that are lost when tau is expressed, they find an inverse correlation of the expression of these pathways and cellular loss, suggesting they might be predictors of neurodegeneration severity. Finally, they use this finding uncovered in Drosophila and reexamined human Alzheimer's disease snRNAseq datasets, were they also find the NFkB pathway to be deregulated.

This study has several strengths. It demonstrates the power of studying taueffects in a tractable model and then using the obtained knowledge to pin-point relevant pathways in cross-sectional studies of human tauopathy, which are otherwise not easy to interpret given the overlayed effects of other disease triggers. By examining the single-cell level they uncover cell type specific effects, which would otherwise be hidden. This study also represents a valuable resource. Given that the authors have included multiple time points the dataset provides an opportunity to understand the evolution of cell-type specific tau effects over time. The authors have also included a replication dataset, which confirms the results of the primary analysis of neuronal loss. I also appreciate the efforts to understand the apparent increase in glia cell number after expression of tau. By combining computational and experimental methods the authors reach the well supported conclusion that in fact glial cell numbers remain constant but only appear increased due to the proportional nature of the scRNAseq data and profound loss of some neurons. Overall, it is interesting that the authors nominate the innate immunity and NFkB pathways in tauopathy, based on deregulated genes and also based on vulnerable neurons. Nevertheless, this is a correlative finding and as such does not proof that it is causal.

As noted [R3], above, we agree that our findings of NFkB dysregulation are correlative. We have performed new experiments to directly test the hypothesis that neuronal immune pathways are causally linked to tau-mediated neurodegeneration; however, the results were negative. These data are included in the revision and we also carefully discuss published work from other fly models of aging and neurodegeneration as well as mouse tauopathy that strongly suggest NFkB can directly modulate neurodegeneration.

The authors correctly point out the importance of aging as a risk factor for Alzheimer's disease. However, it is unclear whether their models actually capture age-dependent neurodegeneration. Alternatively, they might represent neurodevelopmental tau toxicity. In Figure 1B it can be seen that all vulnerable cell types are already lost at day 1, most notably a'/b'-KC, a/b-KC and G-KC with a >4-fold decrease. This raises the question whether the lost cells might developmentally have not correctly formed, as suggested by a study that the authors cite (Kosmidis et al., 2010). This distinction is important in order to strengthen the translational value of the study to human tauopathies.

The elav>tauR406W model manifests both developmental toxicity and age-dependent neurodegenerative changes. Our revision includes new data highlighting specific examples and includes a more balanced discussion of these issues.

The analysis of tau expression levels relative to its impact across cell types in Figure S8 is interesting, however has caveats. The profound neuronal loss makes the interpretation of the correlation analysis of tau levels vs. neuronal vulnerability difficult - since it might be that the individual surviving a'/b'KC, a/b-KC and G-KC cells are the ones that expressed little amounts of tau, while those that are missing used to express high tau. In addition, it is unclear from the methods whether the 3' UTR from the transformation vector to generate the models was included in the counting. The majority of reads would be expected to be there.

As suggested, we have repeated the alignment and analysis of MAPT expression including the short SV40 3’UTR (135bp) from the transformation vector. The result appears very similar to that from the previous analysis, and we have updated Figure 3–figure supplement 4 with these data. Based on the feedback from Reviewer 2, we also include a new plot highlighting the non-relation between MAPT expression and cell abundance changes (Figure 3–figure supplement 4B). We acknowledge the caveat that “missing” / dead cells may have previously expressed high levels of tau, leaving behind survivors with low tau levels. We have added mention of this possible caveat in the Results (lines 197-198). However, while this scenario might impact our interpretation of cell abundance changes, it is a less likely to confound our analyses of differential expression, in which the number of differentially expressed genes show very poor correlation, if any, with MAPT transgene expression (Figure 3–figure supplement 4C).

It would be relevant to know whether the animals were in the same genetic background. I.e. is UAS-TauR406W in the same background of the fly that was crossed to elav-Gal4 to serve as the control. This is not mentioned in the paper and also not in Mangleburg et al., 2020 which the authors refer to. There is a lot of tau-induced DEGs (~1/3 of the detected genes) and it would be relevant to know whether some of them might be due to genetic background.

Our experimental design mitigates the possibility of a substantial impact from genetic background; however, we have added text in the revision noting that this is an important consideration and possible confounder.

The finding of the authors that NFkB pathways are higher in cell types that degenerate more is interesting. However, in Figure 4D it is also apparent that multiple cell types that do not degenerate have comparably high expression. Therefore, it is not a sufficient factor to explain why some neurons are vulnerable vs. others are not, but rather predicts amongst the vulnerable neurons how much they will be lost. It would be helpful to make this distinction clear in the text.

We agree with the reviewer’s interpretation, and we have tried to make this more clear in both the results and discussion (lines 286-288; 387-390). Indeed, the NFkB expression level seems to be a marker for the severity of tau-triggered cell loss among the vulnerable cells.

Reviewer 2 (Public Review):

Wu et al. conducted longitudinal single-nucleus RNA sequencing in a Drosophila transgenic line expressing pathogenic tau (Arg406 ->Trp) and control to study presenile degenerative dementia with bitemporal atrophy. Their data is consistent with previous findings on Tau neurotoxicity, which significantly affects excitatory neurons in human brain samples and transgenic mice. Authors identify aging-like signatures, and an innate immune glial response, including the NFKB pathway, in the transgenic animals.

Strength: This is a great resource for the dissection of dynamic, age-dependent gene expression changes at cellular resolution for the fly community. The article's conclusions are largely supported by the data.

We thank the reviewer for recognizing the value of this work as a resource for the field.

Weakness: No additional orthogonal validation is done on the identified pathways using immunohistochemistry. Also, the authors hypothesized that innate immune signatures might serve as predictors of neuronal subtype vulnerability in tauopathies. Although their data support stronger immune responses in the mutant lines, these findings are not validated. Moreover, the Authors need to use appropriate control animals to compare the mutant Tau animals.

Our original manuscript included experimental validation demonstrating that (1) the apparent increases in glial cell abundance is likely due to changes in cell proportions, and we also (2) confirmed the expression of Relish in both neurons and glia of the adult fly brain. For our revision, we were guided by the requested Essential Revision #3 [R3]. We have therefore performed additional experiments directly testing whether manipulation of Relish/NFkB in neurons alters tau-induced neurodegeneration. While the results of these experiments were negative, we have incorporated them into the results and discussion, along with discussion of related published studies that support a causal role for NFkB immune pathways in tauopathy. Lastly, in our response to Essential Revision #4 [R4], we address concerns about the conservation of neurotoxic mechanisms between mutant and wildtype froms of MAPT and the use of mutant MAPT transgenic models for investigations of AD, along with the caveats. New analyses have been performed and textual revisions have been made in response this feedback

Reviewer3 (Public Review):

Understanding the changes in the brain during the progression of neurodegenerative diseases may provide a critical entry point towards medical treatments. Many genes have been directly or indirectly implemented in an array of neurodegenerative diseases, including the microtubule associated protein tau (MAPT). Various studies have shown that misexpression of tau can cause behavioral, genetic as well as molecular phenotypes that display properties of human neurodegenerative diseases connected to tauopathies. Here the authors use the fruit fly as model to assess phenotypic defects at single-cell resolution. Pan-neuronal misexpression of a mutant form of tau (R406W) and single-cell RNAseq at different time points provides the basis for the investigation.

The authors assess which cell-types are affected (by comparing it with previously described brain cell atlas identities) and find that certain cell types are missing (or less abundant) while other appear unaffected. They do this comparison in relative abundance; both neurons and glia cells are affected.

As next step they compare this with the cell-cluster changes during aging and compare both types of analysis; the investigation here includes the analysis of differentially expressed genes in defined cell clusters. One particularly affected pathway in response to tau is the NFκB signaling pathway. The authors investigate the gene expression changes of the NFκB signaling pathway in the current dataset in more detail. In the last section the authors compare singlecell transcriptomic analyses between fly and human postmortem tissue, showing that the NFκB signaling pathway might be a conserved aspect of neurodegeneration.

The manuscript is overall an elegant example of how single-cell RNAseq can be employed as tool to study the impact of genetic modulators of neurodegeneration (in this case tau) and that it allows direct comparison with human tissues. The results are clean, logically presented and accordingly discussed. It shows that such approaches are indeed powerful for genetic dissection of mechanisms at a descriptive level and opening doors for functional studies.

We thank the reviewer for this positive summary of our manuscript.

Author Response

Reviewer #1 (Public Review):

The goal of the authors was to understand how the kinase, hpk-1, could regulate and interrogate different aspects of cellular stress resilience. To this end, the authors uncovered that hpk-1 is coexpressed with several transcription factors known to regulate different stress responses and this coregulation only appears to occur in the nervous system. Taking a deeper dive, they convincingly find that hpk-1 overexpression in either serotonergic of GABAergic neurons can protect animals from heat stress or toxic protein aggregates. Interesting, it appears that hpk-1 functions in serotonergic neurons differently from GABAergic neurons in the induction of the heat shock response and autophagy.

Overall, the experiments and results are solid and the conclusions drawn reflect the result. The model suggests that the receiving cell deciphers that either heat shock response or autophagy can be induced in the same cell, but the data suggest otherwise. Perhaps the model should be reworked to reflect this point.

We thank the reviewer for their kind assessment and suggestion to refine our model. Indeed, we did not intend to imply that that the receiving cell/tissues were the same after each stimulus, but were attempting to simplify the diagram and condense space. In the revised manuscript we have altered the model (Figure 9B) to reflect that the recipient tissues are distinct.

Reviewer #2 (Public Review):

Lazaro-Pena et al. investigated how a conserved kinase called homeodomain interacting protein kinase (HPK-1), helps to preserve neuronal function, motlity and stress resilience during aging in the metazoan, C. elegans. HPK-1 is a member of the HIPK kinases that, in mammalian systems, regulate the activity of transcription factors (TFs), chromatin modifiers, signaling molecules and scaffolding proteins in response to cellular stress. The group finds that in C. elegans, HPK-1 depletion causes a premature shortening of lifespan and decreases motility and stress resilience in the whole animal. Conversely, increasing active, but not enzymatically dead, HPK-1 levels in the nervous system alone is sufficient to extend lifespan and mitigate the accumulation of aging-associated protein aggregates. The authors then identify a subset of neurons and cell stress response pathways that could be responsible for the contribution of HPK-1 to lifespan and neuronal health. This leads the authors to propose a hypothesis whereby HPK-1 activity in specific neurons preserves protein homeostasis and neuronal integrity, and thus limits the aging-induced decline in organismal function.

Overall, the authors test several functional readouts for neuronal activity to support their claim that HPK-1 activity limits functional decline during aging. These experiments are solid, and the use of a kinase dead HPK-1 in these experiments adds strong support to their claim that HPK-1 activity preserves organismal health. However, weaknesses in the experimental layout and rigor, and the statistical analyses of the publicly available data, limit the inferences that can be made, and further experimental evidence would be required to confirm the working model proposed by the authors.

We thank the reviewer for their thoughtful and balanced assessment of our study.

Author Response

The authors would like to thank the reviewers and editors for this thorough and constructive assessment of our paper. We look forward to addressing their suggestions for improvement of our work in a revised manuscript. In particular: (i) Reviewer 1 raises interesting questions regarding the potential impact of intrinsic cortical and mesh morphology on interpolation, smoothing and the resultant patterns of gene expression. We will test these ideas by developing a null model framework. (ii) Reviewer 2 suggests re-creating dense expression maps using an alternative Gaussian Processes for interpolation. We will implement this suggestion and compare the resulting maps with those generated by the current interpolation method. Notwithstanding these helpful lines of further enquiry, we believe our study provides a meaningful step forwards in multiscale analysis of the human brain by generating. validating, describing and annotating dense gene expression maps which can accelerate translation between neuroimaging and genomic analysis of the human cortical sheet.

Author Response

Reviewer #1 (Public Review):

First, we thank the reviewer for his instructive remarks. In the following we address the queries of Reviewer 1.

1.1) At several points, the authors make claims that I believe extend beyond the data presented here. For instance, in the Abstract (line 27), the authors state "the development of adult songs requires restructuring the entire HVC, including most HVC cell types, rather than altering only neuronal subpopulations or cellular components." The gene ontology analyses performed do suggest that there is a progression from cellular transcriptional changes to organ-level changes, however caution should be taken in claiming that "most HVC cell types" exhibit transcriptional changes. In fact, according to Fig. 3D most of the transcriptional changes appear restricted to neurons. As the authors themselves note elsewhere, claims at this resolution are difficult without support from single-cell approaches. I do not suggest that the authors need to perform single-cell RNA-seq for this work, but strong claims like this should be avoided.

We have revised our claim to more accurately reflect our findings. Our intended message is that testosterone treatment leads to extensive transcriptional changes in the HVC, likely affecting a majority of neuronal subpopulations rather than solely targeting specific cellular components. The revised text in lines 29-32 now reads: "Thus, the development of adult songs stimulated by testosterone results in widespread transcriptional changes in the HVC, potentially affecting a majority of neuronal subpopulations, rather than altering only specific cellular components."

1.2) Similarly the Abstract states that parallel regulation "directly" by androgen and estrogen receptors, as well as the transcription factor SP8, "lead" to the transcriptional and neural changes observed after testosterone treatment of females. However, experiments that demonstrate such a causal role have not been performed. The authors do perform a set of bioinformatic analyses that point in this direction - enrichment of androgen and estrogen receptor binding sites in the promoters of differentially expressed genes, high coexpression of SP8 with other genes, and the enrichment of predicted SP8 binding sites in coexpressed genes. However, further support for direct regulation, at the level that the authors claim, would require some form of transcription factor binding assay, e.g. ChIP-seq or CUT&RUN. I am fully aware that these assays are enormously challenging to perform in this system (and again I don’t suggest that these experiments need to be done for this work); however, statements of direct regulation should be tempered. This is especially true for the role of SP8. This does appear to be a compelling target, but without some manipulation of the activity of SP8 (e.g. through knockdowns) and subsequent analysis of gene expression, it is too much to claim that this transcription factor is a regulatory link in the testosterone-driven responses. SP8 does appear to be a highly connected hub gene in correlation network analysis, but this alone does not indicate that it acts as a hub transcription factor in a gene regulatory network.

We appreciate the reviewer's comment and have revised the statement concerning the role of SP8. Indeed, we document the coexpression of ESR2 and SP8, and our bioinformatics analysis suggests that SP8 might play an important role in transcriptomics. We have rephrased the statement in line 29-32 as follows: "Parallel gene regulation directly by androgen and estrogen receptors, potentially amplified by coexpressed transcription factors that are themselves steroid receptor regulated, leads to substantial transcriptomic and neural changes in specific behavior-controlling brain areas, resulting in the gradual seasonal occurrence of singing behavior." In addition, we have included discussions regarding limitations of promoter sequence analyses (lines 414 to 427).

1.3. Along these lines, the in-situ hybridizations of ESR2 and SP8 presented in Figure 5 need significant improvement. The signals in the red and green channels, SP8 and ESR2, look suspiciously similar, showing almost identical subcellular colocalization. This signal pattern usually suggests bleed-through during image acquisition, as it’s highly unlikely that the mRNA of both genes would show this degree of overlap. I would suggest that control ISHs be run with one probe left out, either SP8 or ESR2, and compare these ISHs with the dual label ISHs to determine if signal intensity and cellular distribution look similar. Furthermore, on lines 354-356 the authors write, "The fact that the two genes were expressed nearby in the same cell may indicate physical interactions between the gene pair and warrant further investigation into the nature of their relationship.". Yet, even if the overlap between ESR2 and SP8 shown in Figure 5 is confirmed, close localization of transcripts does not imply that the protein products physically interact. The STRING bioinformatic analysis is more convincing that there is a putative regulatory interaction between ESR2 and the SP8 locus, and this suggestion of protein-protein interaction is weak and should be omitted. In addition, the authors note that ESR2 has not been detected in the songbird HVC in a previous study. To further demonstrate the expression of ESR2 (and SP8) in HVC, it would be useful to plot their expression from the microarray data across the different testosterone conditions.

We repeated the coexpression study using confocal microscopy and fluorescent RNAScope in situ hybridization, which is now reflected in the revised Figure 5 and a new Figure 5 - Supplement Figure 1. We have also moderated our statement regarding the sparse co-expression of ESR2 and SP8 in HVC neurons. While the presence of co-expressing neurons may provide some anatomical basis for the bioinformatic findings, we have been cautious in our interpretation and have stated that "SP8 and ESR2 mRNAs exhibited low expression levels in HVC, co-localizing in a subset of cells, predominantly GABAergic cells" (lines 369-370). We have removed the speculation about potential protein interaction based on mRNA distribution. Additionally, we have highlighted that SP8 and ESR2 were differentially upregulated at T14d (lines 362-363).

1.4) My final concern lies in the interpretation of these results as generalizable to other sex hormone-modualated behaviors. On lines 452-455, the authors write, "This suggests that the testosterone (or estrogen)-triggered induction of adult behaviors, such as parental behavior and courtship, requires a much more extensive reorganization of the transcriptome and the associated biological functions of the brain areas involved than previously thought.". The experiments and argument likely apply to other neural systems to undergo large seasonal fluctuations in sex hormones and similar morphological changes. However, the authors argue that the large number of transcriptional changes seen here may generalize broadly to sex hormone modulated adult behaviors. I think there are a couple of problems with this argument. First, as described here and in past work, testosterone drives major morphological changes the song system of adult canaries; such dramatic changes are not seen for instance in sex hormone-receptive areas underlying mating behavior in adult mammals. Similarly, the study introduced testosterone into female birds which drives a greater morphological change in HVC relative to similar manipulations in males, which again may account for the large number of differentially expressed genes. I would temper the generality of these results and note how the experimental and biological differences between this system and other sex hormone-responsive systems and behaviors may contribute to the observed transcriptional differences.

We modified this statement in lines 473-478: “The testosterone-driven changes in female HVC morphology and function represent some of the most notable modifications known in the vertebrate brain. However, how this extensive, testosterone-induced gene regulation in the HVC applies to other seasonally testosterone-sensitive brain areas remains to be seen. Endpoint analysis of testosterone-induced singing in male canaries during the non-reproductive season also indicates considerable regulation of HVC transcriptomes (Frankl-Vilches et al., 2015; Ko et al., 2021)”.

Reviewer #2 (Public Review):

First, we would like to express our gratitude to Reviewer #2 for the constructive feedback. We have addressed the concerns in detail below:

2.1). The bulk of the manuscript details WGCNA, GO terms, and promoter ARE/ERE motif abundance, using the initial pairwise comparisons for each timepoint as input lists. However, there are no p/adjp values provided for these pair-wise comparisons that form the basis of all subsequent analyses. Nor are there supplementary tables to indicate how consistent the replicates are within each group or how abundantly the genes-of-interest are expressed. With the statistical tests used here, and the lack of relevant information in the supplementary tables, I cannot determine if the data support the authors’ conclusions. These omissions mar what is otherwise a conceptually intriguing line of investigation.

We appreciate the reviewer’s concerns. Please refer to our response addressing this point and the subsequent one (2.2) together in the section below.

Reviewer #3 (Public Review):

We appreciate the positive feedback from the reviewer and below addressed the issues pointed out by the reviewer.

3.1) My biggest concern is the sample size. Most of the time points only have 5 or 6 individuals represented, and I question whether these numbers provide sufficient statistical power to uncover the effects the authors are trying to explore. This is a particular problem when it comes to evaluating the supposed "transient" of testosterone on gene expression. There is currently little basis for distinguishing such effects from noise that accrues because of low power. This can be a major problem with studies of gene expression in non-model species, like canaries, where among-individual variability in transcript abundance is quite high. Thus, it is possible that one or two outliers at a given time point cause the effect testosterone at this time point to become indistinguishable from the controls; if so, then a gene may get put into the transient category, when in fact its regulation was not likely transient.

We acknowledge that our sample sizes may appear moderate. To address the concern regarding temporal regulation analysis, we followed Reviewer 3's suggestion and conducted a probe-level power analysis (point 2 of recommendations for the authors; labelled as point 3.9 below). We then excluded differentially expressed genes with a power less than 0.8 prior to conducting temporal classification. Consequently, 93% of our differentially expressed genes demonstrated a power ≥ 0.8 (9025/9710). Following further classification by temporal regulation pattern, we identified 29 constantly upregulated, 41 constantly downregulated, 39 dynamically regulated, and 8916 transiently regulated genes. If we apply a stricter constraint by requiring each differentially expressed gene to have at least two probe-sets with a power ≥ 0.8, 83% of differentially expressed genes (8033/9710) still have sufficient power.

We recognize that our sample size may not be sufficient to detect weakly differentially expressed genes. However, we have intentionally excluded these genes from the beginning (those with |log2(fold change)| ≤ 0.5 were excluded).

The scenario outlined by the reviewer, where outliers might cause the effect of testosterone to blend with controls, leading to misclassification, is indeed plausible. This could occur either because the genes are weakly regulated, or because the power to detect differential expression is insufficient, thus preventing these genes from surpassing the threshold to be deemed significantly differentially expressed. However, this also illustrates that the effect of testosterone does not regulate every gene in the same way.

We have appended a column indicating high power genes (≥ 0.8) in the DiffExpression.tsv file, available in the Dryad repository. The power analysis has been incorporated to the method section at lines 801-808 and result section at lines 188-192.

3.2) More on the transient categorization. Would a gene whose expression is not immediately upregulated (within 1 hour), but is upregulated later on (say in the 14d group) be considered transient? If so, this seems problematic. Aren’t the authors setting the null expectation of "non-transient" as a gene that does not increase immediately after 1 hour of treatment? The authors even recognize that it is quite surprising that gene expression changes after an hour. It may be that some genes whose regulation is classified as transient are simply slower to upregulate; but, really, would we say their expression in transient per se? Maybe I’m misunderstanding the categorizations?

We appreciate the reviewer's insightful discussion regarding the transient categorization. We understand that it is indeed more challenging for a gene to be classified as constantly regulated than transiently regulated, due to smaller effects by testosterone or being undetectable owing to low power. To address this concern, we further dissected the transiently regulated category by reporting the number of time points at which a gene is differentially expressed in Figure 2 - Figure supplement 1. Approximately half of the transiently regulated genes were only regulated at one time point, further illustrating that the effect of testosterone on gene expression was not constant during the time window we examined (see lines 184 - 187).

3.3) The authors don’t fully explain the logic for using females in this study to measure a "male-typical" behavior (singing). My understanding is that females have underlying circuitry to sign, and T administration triggers it; thus, this situation that creates a natural experiment in which we can explore T’s on brain and behavior, unlike in males which have fluctuating T. First, it might be good to clarify this logic for readers, unless perhaps I’m misunderstanding something. Second, I found myself questioning this logic a little. Our understanding of basic sex differences and the role that steroid hormones play in generating them has changed over the last few decades. There are, for example, a variety of genetic factors that underlie the development of sex differences in the brain (I’m especially thinking about the incredible work from Art Arnold and many others that harness the experimental power of the four core genotype mice). Might some of these factors influence female development, such that T’s effects on the female brain and subsequent ability to increase HVC size and sing is not the same as males.

Indeed, sex-chromosome dosage compensation is absent in birds leading to higher Z-chromosomal gene expression in males. We demonstrated substantial sex differences in gene expression in our earlier work [Ko, M.-C., Frankl-Vilches, C., Bakker, A., Gahr, M., 2021. The Gene Expression Profile of the Song Control Nucleus HVC Shows Sex Specificity, Hormone Responsiveness, and Species Specificity Among Songbirds. Frontiers in Neuroscience 15].

We have revised the introduction (lines 96-98) to clarify our rationale for using female canaries as a model for adult behavioral development, not as a model for male canaries. After testosterone treatment, these females start to sing, with song structure developing over time, similar to male seasonal progression. This approach eliminates the confounding effect of fluctuating testosterone levels seen in males, supported by distinct HVC transcriptomes in testosterone-implanted singing female canaries compared to males (Ko et al., 2021).

The revised paragraph reads as below: Female canaries (Serinus canaria) are typically non-singers, with their spontaneous songs displaying less complexity than their male counterparts (Hartley et al., 1997; Herrick and Harris, 1957; Ko et al., 2020; Pesch and Güttinger, 1985). Despite their infrequent singing, these females possess the necessary underlying circuitry that can be activated by testosterone. Following testosterone treatment, these females start to produce simple songs, which gradually evolve in structure over weeks—paralleling the seasonal progression of male singing (Hartog et al., 2009; Ko et al., 2020; Shoemaker, 1939; Vallet et al., 1996; Vellema et al., 2019). Moreover, testosterone induces the differentiation of song control-related brain nuclei in adult female canaries, a critical step for song development (Fusani et al., 2003; Madison et al., 2015; Nottebohm, 1980). In this study, we focus on these testosterone-treated female canaries as a model for adult behavioral development rather than a model for male canaries. This unique model allows us to examine transcriptional cascades in parallel with the differentiation of the song control system and the progression of song development, without the confounding impact of fluctuating testosterone levels seen in males, which often results in considerable individual differences in the non-reproductive season baseline singing behavior. This approach is backed by the observation that the HVC transcriptomes of testosterone-implanted singing female canaries are distinct from those of singing males (Ko et al., 2021).

3.4) I was surprised by the authors assertion that testosterone would only influence several tens or hundreds of genes. My read of the literature says that this is low, and I would have expected 100s, if not 1,000s, of genes to be influenced. I think that the total number of genes influenced by T is therefore quite consistent with the literature.

We apologize for any confusion caused by our statement. We did not mean to imply that testosterone only influences several tens or hundreds of genes, but rather that we did not expect such an extensive transcriptional regulation in the HVC by testosterone. We have clarified this in our revised manuscript, specifically in lines 450-451. Thank you for helping us to clarify this point.

3.5) I found the GO analyses presented herein uncompelling. As the authors likely know, not all GO terms are created equally. Some GO terms are enriched by hundreds of genes and thus reflect broad functional categories, whereas other GO terms are much more specific and thus are enriched by only a few genes. The authors report broad GO terms that don’t tell us much about what is happening in the HVC functionally. This is particularly the case when a good 50% of the genome is being differentially regulated.

We appreciate the reviewer's comment. We have added KEGG pathway enrichment analysis in Figure 3 - Figure supplement 1 as an alternative. However, we believe that the GO term enrichment results still provide valuable insights, and therefore we have retained them in Fig. 3.

3.6) The Genomatix analyses are similarly uncompelling. This approach to finding putative response elements can uncover many false positives, and these should always be validated thoroughly. Don’t get me wrong-I appreciate that these validations are not trivial, and I value the authors response element analysis.

We appreciate the reviewer's comment on the presence of AR or ER motifs in promoters and acknowledge that in mammals, AR and ER predominantly bind at distal enhancers rather than promoters. Our analysis focused on promoter regions due to the limitations of available tools and resources for our study species. We understand that this approach may not capture the full complexity of AR and ER regulation. We have revised our manuscript to note the limitations of our approach and clarify that the presence of AREs and EREs alone is not indicative of active receptor binding or direct regulation (lines 416-427).

3.7) I’m sceptical about the section of the paper that speculates about modification of steroid sensitivity in the HVC. These conclusions are based on analyses of mRNA expression of AKR1D1, SRD5A2, and the like. However, this does not reflect a different in the capacity to metabolize steroids, or at least there is little evidence to suggest this. Note that many of these transcripts have different isoforms, which could also influence steroidal metabolism.

We agree that mRNA expression levels of AKR1D1, SRD5A2, and other transcripts involved in steroid metabolism do not necessarily reflect changes in steroid metabolizing capacity. However, we believe that these changes in mRNA expression are indicative of potential changes in steroid sensitivity in the HVC, which could affect the neural response to steroids. We acknowledge that isoform differences of these transcripts may influence steroid metabolism and further studies are necessary to confirm our findings and elucidate the mechanisms underlying the observed changes in gene expression. In response to this comment, we have amended the text in lines 245-249 to reflect this consideration.

Author Response

Reviewer #1 (Public Review):

The authors have compiled and analysed a unique dataset of patients with treatment-resistant aggressive behaviours who received deep brain stimulation (DBS) of the posterior hypothalamic region. They used established analysis pipelines to identify local predictors of clinical outcomes and performed normative structural and functional connectivity analyses to derive networks associated with treatment response. Finally, Gouveia et al. perform spatial transcriptomics to determine the molecular substrates subserving the identified circuits. The inclusion of data from multiple centres is a notable strength of this retrospective study, but there are current limitations in the methodology and interpretation of findings that need to be addressed.

1) The validation of findings is heterogeneous and inconsistent across analysis pipelines. While the authors performed non-parametric permutation testing during sweet-spot mapping, structural and functional connectivity were validated using a 'four-fold consistency analysis'. The latter consists of a visual representation of streamlines and peak intensities after randomly dividing data into four groups, the findings were not validated quantitatively. If possible, the authors should apply permutation analysis in alignment with sweet-spot mapping and demonstrate the predictive ability of their identified networks in a LOO or k-fold cross-validation paradigm as carried out by similar studies. Given that the data has been derived from multiple centers, the prediction of left-out cohorts based on models generated by the remaining cohorts could be another means of validation. If validation is not possible, the authors should clearly state the limitations of their approach.

We appreciate the comment. We have now improved the validation of our connectomics analyses and removed the four-fold consistency analysis. For the functional connectivity analysis, we performed a 1000 permutation test (p<0.05). Similar brain areas were detected in the corrected and uncorrected maps. For the structural connectivity analysis, we used False Discovery Rate (FDR) correction at a significant level of p<0.001, as it is not feasible to perform a 1000 permutation test with this data. The structural connectome is composed of 12 million fibres, and every single permutation takes approximately 4 hours to be completed using our most powerful computational system. To perform 1000 permutations, it would take at least 4000 hours (i.e. 167 days or 5.5 months) of uninterrupted analysis to complete the test. However, it is important to highlight that an FDR correction at the level of p<0.001 is an extremely stringent method. This means that of the 23,000 fibres detected as being touched by the VATs, only 23 would be incorrect, while the remaining 22,977 are correct. Here again, we observed many similarities between the uncorrected and corrected maps, with the main anatomical structures being detected in both. The Methods section and Figures 4 and 5 were revised to reflect these changes.

2) In addition to a 'four-fold consistency analysis', functional connectivity was evaluated using LOOCV in a priori identified ROIs. Their network analysis, however, revealed a far more extensive network encompassing cortical, subcortical, and cerebellar structures. To avoid selection bias the authors should incorporate identified structures into their analysis and apply appropriate means of validation.

We thank the reviewer for this valuable suggestion. We originally did not explore the various significant areas but performed a more focused analysis intended to demonstrate that regions of the known ‘aggression network’ are indeed implicated in our findings. We performed a new analysis exploring the correlation between symptom improvement and the functional connectivity of all the areas described in Figure 5 (i.e., functional connectivity map). To this aim, we extracted individual connectivity values from the peak within each significant region and performed the same additive linear model, incorporating the functional connectivity of each area as well as the age of the patients to estimate individual symptomatic improvement. In addition, we performed a complete exploratory analysis considering the connectivity of any 2 brain structures and age. The resulting matrix shows to what extent functional connectivity to any two areas can be used to estimate clinical outcomes. Interestingly, this new analysis revealed the Periaqueductal Grey matter (PAG) to be the most important functionally connected area when investigated alone or in combination with brain structures critically involved in the regulation of emotional responses, namely the amygdala, anterior cingulate cortex, bed nucleus of the stria terminalis, nucleus accumbens, orbitofrontal cortex and fusiform gyrus. Also, the significance of the PAG connectivity was retained during leave-one-out cross-validation (LOOCV). The Methods, Results, Discussion and Figure 6 were revised. In addition, we added a new Table 2 and Supplementary File 1 to describe the new analysis and results.

3) Functional connectivity mapping: how were R-maps generated? The authors mention that patient-specific R-maps were p-thresholded and corrected for multiple comparisons, but it is not clear how group-level maps were generated. How did the authors perform regression on these maps? Were voxels that did not survive thresholding excluded?

This is a multiple-step analysis. First, it is necessary to localize the electrodes in each patient’s brain and estimate the volume of activated tissue (VAT) observed when stimulation parameters associated with symptomatic improvement are used. The VATs are then used as seeds for the next steps, during which we investigate how much functional influence the VTAs have on the other areas of the brain (i.e., individual r-map). This is done by correlating the BOLD time course of the VAT’s seed with the BOLD time course of all other voxels in the brain. The individual r-maps are then corrected for multiple comparisons to exclude voxels with potentially spurious correlations, resulting in an individual r-map that only included voxels surviving Bonferroni correction at the level of p<0.05. Finally, to create group-level maps, a voxel-wise linear regression analysis was performed to investigate whether each voxel of the map exerts more or less influence (corrected individual r-map with the functional connectivity of the patient’s VAT) or is more or less related to the clinical outcome (i.e. individual improvement). The last step is a permutation correction resulting in a significant group-level functional connectivity map (ppermute<0.05). We modified the Methods section and added a new Figure 1-figure supplement 1 illustrating this analysis.

4) The authors determined that age was a significant prédictor of the outcome, but it is unclear whether certain age groups presented with distinct etiologies underlying their aggressiveness. For example, aggression in epilepsy may show a better response to DBS as opposed to schizophrenia. How does patient outcome change when stratifying according to etiology? How does model performance change when controlling for etiology? The authors should include the etiology of aggressiveness in Table 1.

This is an interesting point. We observed a similar distribution between the pediatric and adult populations in relation to the most common etiologies reported. Epilepsy was the most frequent diagnosis in both populations (pediatric: 50%, adult: 62%), followed by autism spectrum disorder (pediatric: 34%, adult: 24%). The remaining etiologies were largely composed of single cases. A similar proportion of intellectual disability was also observed in pediatric and adult populations. Severe cases were observed in 75% of pediatric and 85% of adult patients. Moderate disability was present in 25% of pediatric and 15% of adult patients. Since several diagnoses were unique to some patients, the addition of this information to Table 1 could result in the identification of the patient. Thus, to preserve anonymity, the diagnoses were added to the end of Table 1 from more to less frequent. We have also revised the Results and Discussion sections to address this concern.

5) Stimulation parameters. The authors report average pulse widths of 219 µs and 142µs respectively, which is up to 4-fold higher as compared to DBS settings used conventionally in movement disorders and will significantly alter the volume of activated tissue. Did the authors account for the drastic increases in pulse width during VAT modeling?

We thank the reviewer for raising this point about the volume of activated tissue (VAT) modelled and the unusual pulse width observed in some patients in this cohort. These patients presented stimulation-induced sympathetic side effects when DBS was set with higher frequencies (e.g. increased heart rate and blood pressure). The chosen final parameters were the ones associated with a clinical benefit without generating side effects. There are a multitude of ways to estimate the VATs, from advanced axon cable models – the gold standard, which simulate axon membrane dynamics and require patient-specific diffusion-weighted imaging and tremendous computing power 1 - to simple heuristics-based models that estimate the rough extent of a VAT based on stimulation parameters without constructing an actual spatial model 2–4. The model employed in our study (and in a number of previous publications by our group 5–10) was the FieldTripSimBio ‘E-field norm’ finite element method (FEM) model. This model, which was first described by Horn et al. 11 and is freely available in Lead-DBS (https://www.lead-dbs.org/), strikes a balance between the sophisticated axon cable models and the simpler heuristic models. In particular, it constructs an electric field (E-field, by applying an electric field strength threshold, or activation threshold) and calculates the VAT associated with specific voltage settings and contact configurations, taking into account the conductivity of surrounding brain tissue and electrode components. Notably, studies comparing VAT modelling techniques 12 showed that ‘E-field norm’ FEM models closely approximate (<0.1 mm difference) the gold standard axon cable models in terms of the size of VATs constructed for monopolar stimulation settings. However, it should be acknowledged that the FieldTripSimBio model in Lead-DBS does not allow the user to specifically enter values for pulse width. Instead, it employs a standard activation/electric field strength threshold (0.2 V/mm) that reflects a combination of commonly modelled axon diameters (roughly 3.5 μm) and pulse width values (i.e., 60-90 μs). This threshold is based on work by researchers such as Astrom et al. 13 and reflects a ‘middle ground’ value that takes into account the fact that any VAT model will necessarily be an imperfect approximation of how electrical stimulation interfaces with brain tissue, depending heavily on aspects such as the diameter of local axons. Nonetheless, it is certainly understood that increased pulse width does meaningfully increase the effective range of stimulation (thus translating to a larger VAT) by lowering the activation threshold of nearby axons 12.

Given that our patient cohort included a small number of patients who were stimulated with higher pulse widths than the values assumed by our model (90 μs), it is reasonable to wonder whether we underestimated the size of these patients’ VATs. To address this aspect, we modelled these patients’ VATs using a simpler heuristic model 2 that does allow specific pulse width values to be selected by the user. More specifically, we computed a range of VATs for these patients using varied pulse width values (ranging from 90 μs up to their actual values). Not surprisingly, this endeavour did yield larger VATs when higher pulse widths were used. On average, the absolute difference in VAT diameter between 90 μs and 450 μs (the largest pulse width observed in this cohort) versions of these patients’ VATs was 2 mm. To check whether or not this difference could have potentially impacted our results, we repeated our probabilistic mapping analysis using altered VATs (specifically, VATs that were enlarged by 2 mm in diameter) for the patients with higher pulse widths. This new repeat analysis yielded a very similar average map to the original analysis: the overall map pattern and location/values of the peak corresponding to the most efficacious area for maximal symptom alleviation remaining unaltered, and only a few voxels on the periphery of the map changing in value by a couple of percentage points. This new supplementary analysis indicates that our results were not meaningfully altered by the unusual pulse width observed in these patients. We modified the Methods section to address some of these aspects and added a new Figure 3-figure supplement 2 illustrating both voxel efficacy maps.

6) Imaging transcriptomics. The methods described lack detail: How did the authors account for differences in expression across donors, samples, and regions during preprocessing of the Allen Human Brain Atlas? How was expression data collapsed into regions of interest? Did the authors apply any normalization? Recent publications have introduced reproducible workflows for processing and preparing the AHBA expression data for analysis that is publicly available.

7) 'genes with similar patterns of spatial distribution to the TFCE map were compiled in an extensive list'. It is unclear why authors used TFCE maps for spatial transcriptomics as opposed to the functional connectivity map featured in Figure 5. How was similarity measured between the TFCE map and the AHBA? How were candidate genes identified? Please provide a more comprehensive description of the analysis pipeline.

We apologize for the short description of this analysis. We performed a gene set analysis using the abagen toolbox (https://abagen.readthedocs.io/en/stable/index.html) to investigate genes with a spatial pattern distribution similar to one of clinically relevant functional connectivity. For this analysis, we used the Allen Human Brain Atlas (https://alleninstitute.org/) microarray data describing the cortical, subcortical, brainstem and cerebellar localization of over 20,000 genes in the human brain (3702 anatomical locations from 6 neurotypical adult brains) 14–17, along with a cell-specific aggregate gene set 18. These data are provided preprocessed, with gene expression values normalized across all brains, and registered to standard MNI space, allowing for a direct comparison between the spatial pattern of gene expression and the functional connectivity map (https://human.brain-map.org/microarray/search) 15. The TFCE maps were used to create clusters of clinically relevant functional connectivity with a spatial extent that overlaps with the anatomical locations from which microarray data was obtained. We parcellated both datasets (results of functional connectivity analysis and Allen Gene Atlas) according to the Harvard-Oxford brain atlas and correlated the spatial distribution of gene expression with the spatial distribution of the results of the functional connectivity mapping. The resultant list of candidate genes was used as input in gene ontology tools to investigate the associated biological processes and cell types. It is important to highlight that this process involves 2 corrections for multiple comparisons using FDR at q<0.005; one correction occurs at the level of the gene list to include only the most significant genes in the gene ontology analysis; a second correction occurs at the level of the gene ontology analysis to consider only the most significant biological processes. We have included some of these details in the revised Methods section.

8) What do the bar plots in Figure 7 (left) represent? P-values? The authors should label the axes to make this clear to the reader.

9) Interprétation of imaging transcriptomics: The authors identify a therapeutic circuit associated with deep brain stimulation of the posterior hypothalamic area, however, it is unclear how to reconcile genes associated with hormones, inflammation, and plasticity in this context. The authors mention and discuss genes implicated in hormonal processing, specifically oxytocin. The results provided in Figure 7, however, do not support this finding and it is unclear how the authors identified genes linked to oxytocin. In addition, the authors identified reductions in the number of microglia and astrocytes, while oligodendrocytes were overexpressed relative to the expected distribution of genes per cell type. These findings were attributed to DBS effects, however, both connectomic and transcriptomic data are acquired from healthy subjects, which suggests a physiological deficit/enrichment in a therapeutic circuit. How do the authors interpret findings given that no electrode implantation and stimulation were performed?

The analysis of normative datasets (functional and structural connectomics and spatial transcriptomics) is based on the idea of better understanding mechanisms of treatment considering our current knowledge of the average human brain. Unlike patient-specific studies in which imaging is acquired from a single patient or genetic profiles are extracted from tissue samples, these normative analyses rely on high-quality “atlases” derived from healthy subjects. In the case of functional and structural connectivity, these atlases are calculated from very large cohorts of subjects (around 1000 brain scans). Thus, imaging connectomics investigates the pattern of brain activity and structural connectivity related to a specific area of the brain (in this case, the volume of tissue activated (VATs) with DBS) and correlate these data with clinical outcomes to shed light on potential mechanisms of action. Similarly, the spatial transcriptomic analysis identifies spatial correlations between patterns of gene expression and brain characteristics detected by MRI 19 (in this case, the spatial pattern of functional connectivity) to investigate possible genetic underlying mechanisms. It is important to highlight that previous studies have shown that normative analyses yield results that are similar to the ones observed using patient-specific data 20–22. In the specific case of imaging connectomics, It has been shown that normative datasets can be used to create probabilistic models of optimal connectivity associated with patients’ outcomes that are meaningful to predict outcomes in patient-specific connectivity data 21. Thus, these exploratory data-driven approaches strive to simulate the presumed fingerprint that a particular patient’s individualized DBS intervention might modulate. They also allow the investigation of possible mechanisms of action in a large, previously inaccessible cohort of patients whose individual data are available. We apologize for the inaccuracy in Figure 7. Along with improving the Discussion section of the manuscript, we included the label for the bar plots in the left panel to improve the clarity of the graph and added the missing result from the KEGG 2021 Human Library that shows the oxytocin signalling pathway.

10) Data availability. Code used for data processing should be made openly available or shared as source data along with the Figures that were generated using the code. Sweet-spot, structural, and functional connectivity maps should be shared openly.

All tools and codes necessary for localizing the electrodes, estimating the volume of activated tissues, and analyzing imaging connectomics are freely available in Lead-DBS (https://www.lead-dbs.org/), a toolbox designed for DBS electrode reconstructions and computer simulations based on postoperative imaging. All codes for spatial transcriptomics are freely available in abagen (https://abagen.readthedocs.io/en/stable/), a toolbox designed to analyze the Allen Brain Atlas genetics data. Along with the codes, the websites for these tools provide manuals describing the step-by-step procedure for successful analysis. The datasets were made freely available at Zenodo (doi: 10.5281/zenodo.7344268). We improved our Data Availability Statement to address this concern.

Reviewer #2 (Public Review):

Deep brain stimulation (DBS) is an important, relatively new approach for treating refractory psychiatric illnesses including depression, addiction, and obsessive-compulsive disorder. This study examines the structural and functional connections associated with symptom improvement following DBS in the posterior hypothalamus (pHyp-DBS) for severe and refractory aggressive behavior. Behavioral assessments, outcome data, electrode placements, and structural and functional (resting-state) imaging data were collected from 33 patients from 5 sites. The results show structural connections of the effective electrodes (91% of patients responded positively) were with sensorimotor regions, emotional regulation areas, and monoamine pathways. Functional connectivity between the target, periaqueductal gray, and amygdala was highly predictive of treatment outcome.

Strengths.

This dataset is interesting and potentially valuable.

Weaknesses.

The figures seem to indicate that electrodes and symptom improvement is located lateral to the hypothalamus, perhaps in the subthalamic nucleus (STN). This is might explain why the streamlines from the tractography are strongest in motor regions. The inclusion of the monoaminergic based on the tractography is not warranted, as the resolution is not sufficient to demonstrate the distinction between the MFB (a relatively small bundle) and others flowing through this region to the brainstem.

This is an interesting point. The sweet spot identified in this work is indeed located in the posterior-inferior-lateral region of the posterior hypothalamic area, reaching the most superior part of the red nucleus, without including the STN. It is important to highlight that the voxel-efficacy mapping only shows voxels associated with a minimum of 50% symptomatic improvement following treatment. Thus, the areas not touching the red nucleus are also associated with excellent symptom alleviation. Although the structural connectivity mapping revealed tracts involved in motor and sensory information, it also showed tracts known to be involved in the regulation of emotions, such as the MFB, the Amygdalofugal Pathway and the ALIC. It is worth noting that this analysis is excellent for segregating the fibre tracts as relevant or not associated with a clinical improvement, but it is not capable of tearing apart the system to determine which of those are necessary for symptom alleviation. As a result, it is not possible to determine whether the motor projections are stronger or more relevant than others. However, the structural connectivity analysis presented here contributes to the body of knowledge on the network of aggressive behaviour and provides clinically relevant data that can be useful to improve future patient outcomes.

We agree with the reviewer that the engagement of the motor system is indeed highly relevant for the reduction of aggressive behaviours, as we have previously shown that aggressive behaviour is highly correlated with motor agitation 23,24. Additionally, in the context of ASD, self-injury behaviour is defined as a type of repetitive/stereotypic behaviour that results in physical injury to the patient’s own body. In relation to the involvement of the monoaminergic system, we would like to apologize for not being clear in the discussion of our findings. Although the functional and structural connectivity maps are related, they provide different means of exploring distinct aspects of the connectivity profile of each VAT. While the structural connectivity map may elucidate symptom improvement via direct fibre modulation (i.e. fibres that touch vs fibres that do not touch the VAT), the functional connectivity map investigates the functional dynamics of the network via BOLD signals (functional MRI). In this manuscript, we showed the functional connectivity (not fibre tracts) of the VATs with areas known to regulate monoamine production, such as the Raphe nuclei and the Substantia Nigra. Both serotonin and dopamine are critically involved in the control of aggressive behaviours, being the target of the main medications used to treat these symptoms in several patient populations. To address all the raised concerns, we incorporated a few sentences in the discussion, highlighting the relevance of the motor system and some limitations of our analysis. We also added a new Figure 3-figure supplement 1 and a discussion on the position of the sweet spot in relation to the red nucleus and subthalamic nucleus.

REFERENCES: 1. Gunalan, K., Howell, B. & McIntyre, C. C. Quantifying axonal responses in patient-specific models of subthalamic deep brain stimulation. Neuroimage 172, 263–277 (2018). 2. Dembek, T. A. et al. Probabilistic mapping of deep brain stimulation effects in essential tremor. Neuroimage Clin 13, 164–173 (2017). 3. Kuncel, A. M., Cooper, S. E. & Grill, W. M. A method to estimate the spatial extent of activation in thalamic deep brain stimulation. Clin. Neurophysiol. 119, 2148–2158 (2008). 4. Mädler, B. & Coenen, V. A. Explaining clinical effects of deep brain stimulation through simplified target-specific modeling of the volume of activated tissue. AJNR Am. J. Neuroradiol. 33, 1072–1080 (2012). 5. Elias, G. J. B. et al. Probabilistic Mapping of Deep Brain Stimulation: Insights from 15 Years of Therapy. Ann. Neurol. 89, 426–443 (2021). 6. Germann, J. et al. Brain structures and networks responsible for stimulation-induced memory flashbacks during forniceal deep brain stimulation for Alzheimer’s disease. Alzheimers. Dement. 17, 777–787 (2021). 7. Elias, G. J. B. et al. Mapping the network underpinnings of central poststroke pain and analgesic neuromodulation. Pain 161, 2805–2819 (2020). 8. Gouveia, F. V. et al. Case report: 5 Years follow-up on posterior hypothalamus deep brain stimulation for intractable aggressive behaviour associated with drug-resistant epilepsy. Brain Stimul. 14, 1201–1204 (2021). 9. Coblentz, A. et al. Mapping efficacious deep brain stimulation for pediatric dystonia. J. Neurosurg. Pediatr. 27, 346–356 (2021). 10. M Oliveira, L. et al. Probabilistic characterisation of deep brain stimulation in patients with tardive syndromes. J. Neurol. Neurosurg. Psychiatry (2021) doi:10.1136/jnnp-2020-324270. 11. Horn, A. et al. Connectivity Predicts deep brain stimulation outcome in Parkinson disease. Ann. Neurol. 82, 67–78 (2017). 12. Duffley, G., Anderson, D. N., Vorwerk, J., Dorval, A. D. & Butson, C. R. Evaluation of methodologies for computing the deep brain stimulation volume of tissue activated. J. Neural Eng. 16, 066024 (2019). 13. Astrom, M., Diczfalusy, E., Martens, H. & Wardell, K. Relationship between neural activation and electric field distribution during deep brain stimulation. IEEE Trans. Biomed. Eng. 62, 664–672 (2015). 14. Hawrylycz, M. J. et al. An anatomically comprehensive atlas of the adult human brain transcriptome. Nature 489, 391–399 (2012). 15. Arnatkeviciute, A., Fulcher, B. D. & Fornito, A. A practical guide to linking brain-wide gene expression and neuroimaging data. Neuroimage 189, 353–367 (2019). 16. Arnatkeviciute, A., Markello, R. D., Fulcher, B. D., Misic, B. & Fornito, A. Toward Best Practices for Imaging Transcriptomics of the Human Brain. Biol. Psychiatry 93, 391–404 (2023). 17. Markello, R. D. et al. Standardizing workflows in imaging transcriptomics with the abagen toolbox. Elife 10, (2021). 18. Seidlitz, J. et al. Transcriptomic and cellular decoding of regional brain vulnerability to neurogenetic disorders. Nat. Commun. 11, 3358 (2020). 19. Fornito, A., Arnatkevičiūtė, A. & Fulcher, B. D. Bridging the Gap between Connectome and Transcriptome. Trends Cogn. Sci. 23, 34–50 (2019). 20. Arnatkeviciute, A., Fulcher, B. D., Bellgrove, M. A. & Fornito, A. Imaging Transcriptomics of Brain Disorders. Biol Psychiatry Glob Open Sci 2, 319–331 (2022). 21. Wang, Q. et al. Normative vs. patient-specific brain connectivity in deep brain stimulation. Neuroimage 224, 117307 (2021). 22. Elias, G. J. B. et al. Normative connectomes and their use in DBS. in Connectomic Deep Brain Stimulation 245–274 (Elsevier, 2022). 23. Gouveia, F. V. et al. Bilateral Amygdala Radio-Frequency Ablation for Refractory Aggressive Behavior Alters Local Cortical Thickness to a Pattern Found in Non-refractory Patients. Front. Hum. Neurosci. 15, 653631 (2021). 24. Venetucci Gouveia, F. et al. Case report: 5 Years follow-up on posterior hypothalamus deep brain stimulation for intractable aggressive behaviour associated with drug-resistant epilepsy. Brain Stimul. (2021) doi:10.1016/j.brs.2021.07.062. 25. Gouveia, F. V. et al. Amygdala and Hypothalamus: Historical Overview With Focus on Aggression. Neurosurgery 85, 11–30 (2019). 26. Gouveia, F. V. et al. Longitudinal Changes After Amygdala Surgery for Intractable Aggressive Behavior: Clinical, Imaging Genetics, and Deformation-Based Morphometry Study-A Case Series. Neurosurgery 88, E158–E169 (2021). 27. Yan, H. et al. Deep brain stimulation for extreme behaviors associated with autism spectrum disorder converges on a common pathway: a systematic review and connectomic analysis. J. Neurosurg. 1–10 (2022). 28. Knotkova, H. et al. Neuromodulation for chronic pain. Lancet 397, 2111–2124 (2021). 29. Gray, A. M. et al. Deep brain stimulation as a treatment for neuropathic pain: a longitudinal study addressing neuropsychological outcomes. J. Pain 15, 283–292 (2014). 30. Miczek, K. A., Fish, E. W., De Bold, J. F. & De Almeida, R. M. M. Social and neural determinants of aggressive behavior: pharmacotherapeutic targets at serotonin, dopamine and gamma-aminobutyric acid systems. Psychopharmacology 163, 434–458 (2002). 31. Gouveia, F. V. et al. Reduction of aggressive behaviour following hypothalamic deep brain stimulation: involvement of 5-HT1Aand testosterone. bioRxiv (2023) doi:10.1101/2023.03.20.533520.

Author Response

Reviewer #1 (Public Review):

The strength of the manuscript is highlighted by the application of fractal formalism, which is commonly used in colloidal systems, in conjunction with MD simulation to study the phase separation of an IDP. The weakness lies in the fact that this study does not provide any discussion on how our understanding of the network structure and dynamical behavior of biomolecular condensates and their biological significance improves through this study. The experimental part remains weak, without any measurements of the dynamics of the condensates. Whether and how the formalism can distinguish between phase-separated condensates (WT) and classical protein aggregates (Y to A variant) remains unclear.

We thank the Reviewer for their careful reading of the manuscript and their appreciation of the link between IDP phase separation and colloid chemistry. Establishment of a quantitative framework behind this link, as given by the fractal formalism, and a multiscale model of the spatial organization of a biomolecular condensate, derived from MD simulations in combination with fractal scaling, are indeed two of our main contributions. In particular, to the best of our knowledge, ours is the first atomistically resolved model of the spatial organization of a biomolecular condensate at an arbitrary scale. The key features of the proposed model, as elaborated in the Discussion of the revised manuscript (p. 18, 20-21), are the coexistence of differently sized clusters inside a condensate, and a quantitative prediction of a particular scaling of mass with cluster size (Figure 5A), as further discussed below. Moreover, our results also point to the possible formation of pre-percolation clusters with sizes below the resolution limit of typical microscopy experiments, in agreement with recent observations (https://doi.org/10.1073/pnas.2202222119).

We agree that the full understanding of biomolecular condensates also requires a detailed treatment of the dynamical aspects. Following the Reviewer’s comments, we have provided significant new results in this regard and included an experimental characterization of fusion behavior (Videos 1, 2) and condensate dynamics by FRAP (Figure 1D, E and Figure 1—figure supplement 2) as well as a detailed analysis of diffusion and viscosity in the simulated systems (Figure 4C and Figure 4—figure supplement 1D-F). The newly performed FRAP experiments provide a direct measure of the condensate dynamics. Importantly, the measured recovery half-times for WT and R>K condensates resemble those of other well-characterized in vitro condensates. We have occasionally observed elongated, amorphous Y>A precipitates, albeit in low number and only at 50-fold higher concentration than the wild-type (45 mM and above, Figure 1C). While this may be consistent with the predictions of the fractal model and hint at the differences in mesoscopic organization between the WT and R>K condensates and the Y>A precipitates, the latter are rare and we are reluctant to draw major conclusions.

Furthermore, we could show that the WT diffusion coefficient is lower than for either mutant (Figure 4C, and Supplementary File 2). Clearly, this difference is not due to the effect of protein size or a higher solvent viscosity, but primarily indicates protein slow-down due to the more extensive interactions with partners (reflected also in higher average valency, Figure 2D, or probability of interactions Figure 2—figure supplement 1D). The fact that the WT diffusion coefficient drops by about 20% over the last 0.3 µs of the MD trajectory also correlates with the formation of a single percolating cluster in the system (Figure 2C). This is an expected effect on protein diffusion upon crossing the percolation threshold (https://doi.org/10.1038/ncomms11817, https://doi.org/10.1021/acs.jpcb.7b08785). Moreover, the difference in the recovery dynamics observed for WT and R>K mutant can be interpreted using the proposed model. Namely, accurate fitting of FRAP data was only possible if using at least two components (Figure 1—figure supplement 2). According to https://doi.org/10.1016/j.tcb.2004.12.001, these components indicate the contribution of particle diffusion and interaction (binding). Thus, recovery of the centrally bleached condensates is faster for WT than for the R>K mutant, which can be related to the higher compactness of the WT particles across scales as compared to R>K. On the other hand, the FRAP results for the condensates bleached in the peripheral area highlight the contribution of the binding component. Indeed, the recovery is about 3-fold faster for the R>K mutant, which could potentially be related to the lower valency of the interactions and the ease of the replacement of inactivated fluorescent species and/or exchange with proteins in the bulk. A further connection of the developed model and condensate dynamics concerns the multimodal description of diffusion in biomolecular condensates, together with multimodal fitting of FCS and FRAP data as used recently for interpreting single particle tracking results (https://doi.org/10.1016/j.bpj.2021.01.001). Namely, the polydisperse nature of the protein phase as suggested by the model translates to multimodal diffusion, reflecting the dynamics of protein clusters of different size. For instance, regularization fits used for DLS autocorrelation curves assume a multimodal character of the diffusion and are interpreted to reflect a multimodal distribution of cluster sizes in condensates (https://doi.org/10.1073/pnas.2202222119).

Finally, a way of testing the model prediction, which would merit a study in its own right, would involve static light scattering (SLS): a linearly decreasing scattering intensity as a function of the scattering vector in a log-log representation, as frequently seen for different colloidal systems, is expected by the fractal model. In fact, fractal dimension dF could directly be estimated from SLS experiments (https://doi.org/10.1038/339360a0) from the limiting value of scattering intensity for high values of the product of the scattering vector q and the average cluster size <Rg>. As a direct test of the predictions of the model, the experimental value of dF could then be compared with the predicted one. Moreover, techniques such as DLS and MALS could be used to measure independently masses and sizes of biomolecular condensates in vitro at different scales in order to test the validity of the particular scaling predicted by the fractal model. Such experiments are not trivial and are out of scope of the present study.

Reviewer #2 (Public Review):

A key aspect of the work is to use the simulations to explain differences between (i) dilute and dense phases and (ii) wild-type and mutant variants. Here, it would be important with a clearer analysis of convergence and errors to quantify which differences are significant.

Following the Reviewer’s suggestion, we now provide an analysis of convergence and statistical significance. Specifically, in Supplementary File 1 “Technical summary” we now report the average value, standard deviation and a block-average measure of convergence for all the key observables analyzed, including radius of gyration (Rg), valency (n), and compactness (), for all modeled systems. Furthermore, in the revised manuscript, we now also include the analysis of protein translational diffusion constants and solution viscosity for all modeled systems to assess the ability of the simulations to capture protein dynamics realistically (Figure 4C, Figure 4—figure supplement 1D-F, Supplementary File 2, see also above). Moreover, we include in the revised version a new figure depicting time evolution of average compactness in the 24-copy systems (Figure 4—figure supplement 1C). Thus, it can be seen that the two key model parameters derived from MD simulations of the 24-copy system – protein valency and compactness – reach a stable plateau over the last 0.3 µs (Figure 2D and Figure 4—figure supplement 1C), which were used for final analyses, with block-averaged deviations of less than 10% throughout (see below for details). All the differences in these parameters between single-copy and 24-copy simulations, as well as those between WT and mutation simulations, were found to be significant with p-values < 2.2 10-16 according to the Wilcoxon rank sum test with continuity correction (details in Supplementary File 1). Finally, considering the sampling limitations implicit in most MD studies, we clearly recognize the possibility that with longer simulation times or more protein copies per simulation box, the simulated systems may show a qualitatively different behavior. However, we emphasize that our derivation of the formalism that links the features of simulated ensembles on the scale of 10s of nanometers with their behavior on the scale of 100s of nanometers and beyond is independent of such limitations. Once longer, larger and more accurate simulations become available, one will be able to apply the formalism without alteration and obtain a model of the spatial organization of the condensate on an arbitrary scale, starting just from the local features of individual proteins. We now discuss these details on pp. 10, 11, 13 of the revised manuscript.

It would also be useful with a clearer description of how the analytical model is predictive, of which properties, and how they have been/can be validated. Which measurable quantities does the model predict?

As pointed above, the model predicts the existence and provides a quantitative description of pre-percolation finite-size clusters (https://doi.org/10.1016/j.molcel.2022.05.018, https://doi.org/10.1073/pnas.2202222119). More generally, the model provides the fractal dimension (dF) of protein clusters and enables evaluation of different scale-dependent properties of clusters of arbitrary size, including protein density as a function of cluster size (Figure 5—figure supplement 1C, Figure 5C). Importantly, the fractal dimension can be used in combination with local MD simulations and cluster–cluster aggregation algorithms to derive a detailed model of the 3D organization of fractal clusters of a chosen size at atomistic resolution (Figure 5A, B, and Videos 4, 5, and 6). Such detailed structural understanding of the interior organization of a condensate can, for example, be used to evaluate cavity sizes and interpret partitioning experiments. Since the differences in the morphology of WT and mutant protein clusters propagate across length scales, they can even be qualitatively characterized by the analysis of microscopic images (e.g. circularity, Figure 1—figure supplement 1C, see also discussion above). Finally, static light scattering (SLS) experiments give the possibility to test the model directly, which will be the subject of our future work. Namely, the fractal formalism predicts linear behavior in the log-log representation of the SLS intensity vs. scattering vector curves, while dF, which can directly be evaluated from such experiments, providing a quantitative point of comparison between theoretical predictions and experiment (see above).

In addition to these overall questions, a number of more specific suggestions follow below.

Major:

p. 7, line 120 (Fig. S1B) The proteins do not appear particularly pure based on the presented SDS PAGE analysis. How pure is the protein estimated to be, and is the presence of the other bands expected to affect e.g. the data presented in Fig. 1?

We have quantified the purity of the constructs by densitometry of the Coomassie stained gels and included it in Figure 1—figure supplement 1A: in the case of WT and R>K, we achieve purity higher than 91%. Importantly, the observed LLPS behavior of the constructs is consistent with the simulation and in agreement with other studies on R>K substitutions (https://doi.org/10.1073/pnas.2000223117; https://doi.org/10.1016/j.molcel.2020.01.025; https://doi.org/10.1073/pnas.2200559119; https://doi.org/10.1016/j.jmb.2019.08.008). In the case of Y>A, we have obtained the least pure protein (~65%), and must note that the precipitates observed in the experiments of Figure 1C are only present at the protein concentrations that are 50-fold higher as compared to WT (45 mM and above). Therefore, at such high total protein concentration, we cannot exclude the possibility that there might be some contamination affecting the behavior of this construct.

p. 7 & 8, lines 138-159: Has the method and energy function used to calculate the interact potential been validated by comparison to experiments, including studying the effect of varying the solvent? I see the computed error bars are very small, but am more interested in the average error when comparing to experiments. The numbers in water appear different from those e.g. reported by Krainer et al (https://doi.org/10.1038/s41467-021-21181-9), though the latter are also not immediately compared to experiments. Thus, it would be useful to know how much to trust these numbers.

We thank the Reviewer for raising this important point. To the best of our knowledge, the absolute binding free energies between Y-Y, Y-R or Y-K sidechain analogs or complete amino acids have never been determined experimentally, preventing a direct validation of the computed values and/or an evaluation of the average error when comparing to experiments. On the other hand, we did compare our data against the PMF curves presented by Krainer et al. (https://doi.org/10.1038/s41467-021-21181-9) for R-Y and Y-Y and the general trends are largely similar. In particular, in both analyses the R-Y interaction is stronger than the Y-Y interaction across different conditions, except at zero salt in Krainer et al. where the two are similar. When it comes to exact quantitative differences between the studies, it should first be pointed out that Krainer et al. studied capped amino-acids, while we used amino-acid side-chain analogs. The difference in the observed binding strengths is in part certainly related to the contribution of the capped backbone. Second, the values in Krainer et al. refer to the depth of the free energy minimum in the obtained PMFs and not to the resulting G values, as in our method. The latter includes integration over the PMF and an assumption of a standard-state concentration, which could also lead to significant differences. Finally, the differences could also be due to the intrinsic properties of the interaction potentials used. In particular, the prominent free-energy minima for the R-Y pair in the Krainer et al. study could only be obtained after refitting of the original AMBERff03ws charges on the Y bound to R via semi-empirical quantum-chemical calculations. On the other hand, the interaction potential used in our study was not adjusted to the system at hand, but rather comes from a published, widely used force field, the OPLS-AA (https://doi.org/10.1021/ja9621760), that was independently tested and validated experimentally in multiple studies. For example, OPLS-AA exhibits the low average error in absolute hydration free energy of ~0.5 kcal/mol, errors of only ~2% for heats of vaporization and densities (https://doi.org/10.1021/ja9621760), and a close agreement with osmotic coefficients (https://doi.org/10.1021/acs.jcim.9b00552) or a large range of organic compounds. This raises our confidence in the accuracy of the derived binding free energies, which directly or indirectly depend on these fundamental thermodynamic properties.

Regarding the method to evaluate PMF profiles, we have used a classical all-atom Monte Carlo approach originally developed by Jorgensen and coworkers (see, e.g., https://doi.org/10.1021/ar00161a004 and https://doi.org/10.1021/ja00168a022), as implemented in the widely used BOSS program (v. 4.8) (https://doi.org/10.1002/jcc.20297). This approach has been extensively tested against experimental data on ΔΔG values of various compounds in environments of different polarity (e.g., 2). Moreover, we have previously successfully applied this methodology in studies of the free energy of association of amino acid residues (https://doi.org/10.1021/jp803640e) and other biologically important groups (https://doi.org/10.1021/acs.jcim.9b00193). The results obtained have been compared with the available experimental data and demonstrated a good agreement. As for the small error bars in the plots, the fairly good convergence achieved in our PMF calculations is a result of extensive sampling combined with small system size, although obviously this is not always the case – see, for example, PMFs in our recent work (https://doi.org/10.1021/acs.jcim.9b00193).

The above points have been discussed on pp 7-8 of the revised manuscript.

p. 8, lines 149-154: Following up on the above, the authors also write "Importantly, only in the latter case are the R-Y interactions slightly more favorable than the K-Y ones (Figure S1C). While this can potentially contribute to increasing of Csat for the R>K mutant as compared to WT, the estimated thermodynamic effect is not too strong, especially if one considers that these interactions take place in an environment with largely water-like polarity. Therefore, the effect of R>K substitution on LLPS should be further explored in the context of protein-protein interactions." In the absence of estimates of the accuracy of the predictions, these sentences are somewhat unclear. Also, it is unclear what the authors mean by that the effect of R>K should be studied; there are already several examples of this (https://doi.org/10.1016/j.cell.2018.06.006 [already cited], https://doi.org/10.1038/s41557-021-00840-w & https://doi.org/10.1073/pnas.2000223117 come to mind, but there are likely more).

As pointed above, the free-energy values were obtained using well-established computational techniques and are expected to reflect realistic trends. However, considering that there exist no equivalent experimental results to assess the accuracy of the predicted free energies, they indeed must clearly be understood as predictions. This is now stated on pp. 7-8 of the revised manuscript. Furthermore, it seems that the vague phrasing on our part in the above paragraph resulted in a misunderstanding. Namely, when we talk about “further exploration”, we only meant it in relation to our study, i.e. a connection with the MD part, and not in relation to a wider literature on the topic. In other words, we simply wanted to refer to the fact that our binding free energies for individual residues do not provide sufficient information about interactions between Lge11-80 protein chains. Following the Reviewer’s comment, we have rephrased this part and included additional references on the known role of R and K residues on phase separation.

p. 8, lines 161-162: The authors perform MD simulations of Lge1 and variants using 24 copies and a box that gives them protein concentrations "in the mM concentration range". I realize that there's a concern about what is computationally feasible, but it would be important with an argument for this choice. Why is 24 expected to be enough to represent a condensate (I expect that there could be substantial finite-size effects)? What is the exact protein concentration in the simulations of the 24 chains [and of the 1-chain simulations]? How does this protein concentration compare to that in the condensates? The authors performed simulations in the NPT ensemble; how stable were the box dimensions?

The effective protein concentration for different 24-copy systems is 6-7 mM, depending on the system (Figure 2—figure supplement 1A). This concentration range was selected in order to get a reasonable system size for microsecond all-atom MD simulations, while still being approximately one order of magnitude lower than the semi-dilute regime of the protein at hand. As a testament to the internal consistency of our framework, the fractal model predicts the concentration inside WT condensates of the size observed in the experiment to indeed be in the mM range. Moreover, as seen in many other systems, the concentration inside the observed droplets is expected to be significantly higher than Csat (https://doi.org/10.1101/2020.10.25.352823). Here, we should again emphasize that we did not aim to model the process of phase separation in our all-atom MD. We rather use multicopy simulations for the analyses of the organization of the protein crowded phase and specifically, the mode of intermolecular interactions, and then use the fractal scaling to derive a model of the internal organization of condensates at arbitrary scales.

Regarding the experimental determination of the protein concentration in the condensates, we have used different approaches to estimate Csat and CD values: spin-down analyses (https://doi.org/10.1126/science.aaw8653), volumetry analysis (https://doi.org/10.1038/nchem.2803), estimation of concentration by fluorescent intensity of the condensates (https://doi.org/10.1016/j.molcel.2018.12.007; https://doi.org/10.1016/j.cell.2019.08.008), FCS (https://doi.org/10.1038/nchem.2803; https://doi.org/10.1016/j.cell.2019.10.011; https://doi.org/10.1126/science.aaw8653). However, different approaches yield values that vary by several orders of magnitude. That is the reason why we did not report definitive numbers. In general, there are uncertainties in the field about how to reliably measure protein concentrations in a condensate, necessitating the development of new approaches (https://doi.org/10.1101/2020.10.25.352823).

With regard to the convergence and potential finite-size effects, we agree that this is an important issue and have addressed it in the revised version. In general, the convergence of our observables such as valency or compactness (Figure 2C, D and Figure 4—figure supplement 1C) gives confidence that the simulations are at least in a local equilibrium, especially when it comes to short-range properties such as contact preferences as further elaborated in our reply to the Reviewer’s specific comment about convergence below (please, see also above for our response to Editor’s comment #5). Importantly, in all 24-copy systems, the average separation between protein images lies in the 12-15 nm range, and no instances of self-interaction between images due to PBC were observed (Supplementary File 1). Finally, analysis of fluctuations in box dimensions shows that they are all in the range of picometers and largely negligible when it comes to the analysis at hand (Supplementary File 1).

In order to highlight the realistic behavior of the simulated systems in the revised version, we now also report a detailed analysis of protein translational diffusion in MD simulations (Figure 4C and Figure 4—figure supplement 1D-F and Supplementary File 2). According to this analysis, single-molecule translational diffusion coefficients of Lge11-80 variants obtained from fitting of MSD curves with applied finite-size PBC correction and rescaling by the solvent viscosity (see Methods for details) are typically in the range of 100 µm2/s (Figure 4C and Supplementary File 2), which corresponds to experimentally measured values for different proteins of similar size. Importantly, the requisite finite-size corrections applied in the case of 24-copy systems are relatively small and amount to about 35-60%, while this is almost an order of magnitude higher (450-530%) for the single-copy simulations (Supplementary File 2). Please, see also the reply above to the Editor’s statements above for more details.

Also, did the authors include the Strep- and His-tags in the simulations? If not, why not?

We did not simulate the constant part of the constructs in order to: 1. expedite computation and 2. more directly expose the effect of different mutations. Since our comparison between simulation and experiment concerned largely qualitative observables, we have primarily focused on the relative differences between the three Lge11-80 variants. Importantly, the effect of mutations on the full-length protein and its different variants was analyzed in vivo in a previous publication (https://doi.org/10.1038/s41586-020-2097-z).

Throughout: One of my major concerns about this work is the general lack of analysis of convergence of the simulations. The authors must present some solid analysis of which results are robust given the relatively short simulations and potential for bias from the chosen starting structures.

First, we would like to emphasize that we did not attempt to capture the process of phase separation or characterize two coexisting phases, for which much larger ensembles and/or simulation times would be needed. Rather, our aim was to study the conformational behavior of individual protein chains in the context of a crowded protein mixture, taken as a model for the dense phase, and then use fractal scaling to provide a model of spatial organization of a condensate at an arbitrary length scale. Having said this, it is absolutely important to address how converged the key observables are, given the finite size of the all-atom simulation setup and the limited sampling used. In the revised manuscript, we have included an additional analysis of convergence of our simulations and could show that both key MD-derived parameters required by the fractal model, protein compactness and valency, display convergent behavior over the last third of 0.3 µs MD in the 24-copy systems (Figure 4—figure supplement 1C) and all analyses were performed over this region. In particular, the block averages of compactness and valency exhibit a standard deviation of only 2-4% and 4-8%, respectively, over the last 0.3 µs of MD simulations. Moreover, since we are interested in single-chain features in the context of a crowded mixture, our sampling corresponds effectively to 24 x 0.3 µs = 7.2 µs. Finally, a detailed analysis of convergence in conformational sampling was performed for single-copy simulations using calculations of configurational entropy as evaluated by the MIST formalism (Figure 4—figure supplement 1B). For instance, in the case of the weakly self-interacting Y>A, we do observe a close convergence in terms of the configurational entropy between two independent replicas on 1 µs MD trajectory (Figure 4—figure supplement 1B). However, we still recognize the possibility that with longer simulation times and/or more protein copies per simulation, the simulated systems may show a qualitatively different behavior, as discussed on pp. 10, 11, and 13 of the revised manuscript. Finally, we would like to reiterate the point that our derivation of the formalism that links the features of simulated ensembles on the scale of 10s of nanometers with their behavior on the scale of 100 s of nanometers and beyond is independent of such limitations. Once longer, larger and more accurate simulations become available, one will be able to apply the formalism without alteration and obtain a model of the spatial organization of the condensate on an arbitrary scale, starting just from the local features of individual proteins. We now discuss these details on pp. 10, 11, and 13 of the revised manuscript.

As an example, on p. 8 the authors discuss a potential asymmetry between the interactions found in the dilute (single-copy) and dense (24-mer) phases. These observations are somewhat in contrast to other observations in the field, namely that it is the same interactions that drive compaction of monomers as those that drive condensate formation.

Obviously, both the results in the literature and those presented here could be true. But in order to substantiate the statements made here, the authors should show some substantial statistical analyses to make it clear which differences are robust. The above holds for all parts of the computational/simulation work (e.g. other aspects of Fig. 2)

Note: this comment by the Reviewer echoes in several respects the comment 7 by the Editor. Because of this, our reply in some parts is identical to that given above to the Editor. We have decided to include it here for the ease of reading and completeness.

An expectation of the symmetry between intra- and intermolecular modes of interaction emerged from the background of polymer theory, which was primarily aimed to describe the behavior of homopolymers. In the case of heteropolymers such as proteins, the asymmetry in the aforementioned modes is rather intuitive. For instance, if there is only a single Y in a protein, then Y-Y contacts will not be possible in the intramolecular context, but could occur in multichain interactions. However, we agree with the Reviewer that this is an important issue and have deepened the analysis of this phenomenon in the revised manuscript.

First, our analysis shows that the observed asymmetry between intra- and intermolecular contexts is statistically significant and is likely not a consequence of limited sampling (pp. 10-11, Figure 3—figure supplement 1B-C). Moreover, the observed symmetry breaking is in line with the recent studies by Bremer et al. (https://doi.org/10.1038/s41557-021-00840-w) and Martin et al. (https://doi.org/10.1126/science.aaw8653), which have delineated the key requirements for the symmetry between single-chain and collective phase behavior to hold. Specifically, we have compared in detail the sequence composition of Lge11-80 with that of A1-LCD variants studied by Bremer et al. When it comes to aromatic composition, Lge1 is most similar to the -12F+12Y mutant of A1-LCD, and by this token, i.e. the high frequency of stickers tyrosines, should exhibit a strong coupling between single-chain and phase behavior. However, the net charge per residue (NCPR) in Lge11-80 of 0.075 is greater than that of A1-LCD (0.059) and this could contribute to the extent of decoupling, as suggested by Bremer et al. Moreover, Lge1 is extremely abundant in Arg (13.5 % as compared to 7.4 % in A1-LCD), and is in this sense most similar to the +7R A1-LCD mutant, which showed the greatest degree of decoupling between single-chain and phase behavior in Bremer et al., in agreement with what we see here. While these authors have demonstrated that NCPR is the primary determinant of decoupling in the case of A1-LCD mutants, their analysis showed that the nature of positive and negative residues involved also makes a significant difference. In particular, the significant excess of Arg residues, as context-dependent auxiliary stickers, could create the asymmetry between interactions that determine single-chain dimensions vs. collective phase behavior.

Furthermore, Martin et al. (https://doi.org/10.1126/science.aaw8653) have shown that an approximately uniform distribution of stickers along the sequence is required for the correspondence between the driving forces behind coil-to-globule transitions and phase separation to hold. We have analyzed the patterning of Tyr residues along the Lge11-80 sequence using Waro parameter used by Martin et al. (note that Tyr is the only aromatic in the Lge11-80 sequence). Interestingly, Lge11-80 exhibits a highly non-uniform patterning of Tyr residues, with Waro of the native Lge1 sequence (0.47) falling in the middle of the distribution for its shuffled variants (p=0.57). This is in contrast to the highly patterned sequences such as that of A1-LCD with p>0.99. Taken together, in addition to the relatively high NCPR, symmetry breaking in the case of Lge11-80 could be a consequence of its complex sequence composition, including both the non-uniform patterning of tyrosines and a high abundance of arginines. Provided that our simulations are long enough to provide an equilibrium picture and are on the length-scale of a single protein not strongly influenced by finite-size effects (these potential artifacts cannot be discounted), they actually can be seen as a demonstration of such symmetry breaking in a heteropolymer.

Furthermore, analysis of pairwise contacts suggests that intra- and intermolecular interactions rely on a similar pool of contacts by amino-acid type, but differ significantly if one analyzes specific sequence location of the interacting residues involved (Figure 2—figure supplement 1B and C). For example, one observes a high correlation between the frequencies of different contacts by amino-acid type when comparing intramolecular contacts in single-copy simulations and intermolecular contacts in 24-copy simulations (Figure 3—figure supplement 1B). This correlation is completely lost (Figure 3—figure supplement 1C) if one analyzes position-resolved statistics (2D pairwise contacts maps) or statistically defined interaction modes (Figure 3A, and Figure 3—figure supplement 1A). For example, although Tyr-Tyr interactions dominate in both cases, in single-copy simulations of WT Lge11-80 the C-terminal Tyr80 barely participates in any intramolecular interactions with other residues (Figure 3—figure supplement 1A), while in 24-copy simulations it is one of the most intermolecularly interactive residues (Figure 3). In other words, while the symmetry between intra- and intermolecular interactions can be observed at the level of pairwise contact types (similar type contact used for both), the distribution of these contacts along the peptide sequence is clearly different in the two cases. Finally, it should be mentioned that the parallel between single-copy and phase behavior in both homopolymers and heteropolymers is observed primarily at the level of thermodynamic variables such as LLPS critical temperature (Tc), coil-to-globule transition temperature (Tq) or the Boyle temperature (TB). It is possible that the noted correspondence extends primarily to such and similar thermodynamic variables, while and more structural, topological features of the globule in the single-molecule case and the network in the collective phase case remain uncoupled.

Interestingly, the core of intramolecular interactions observed for a single molecule at infinite dilution and in the crowded context remain approximately the same as reflected in the high correlation between intramolecular modes obtained in single and multichain simulations. Namely, proteins keep core self-contacts and establish new ones with neighbors, but do not donate everything to the intermolecular network losing “self-identity”, as in homopolymer melts. Similar effects have also been observed elsewhere: https://doi.org/10.1073/pnas.2000223117, https://doi.org/10.1073/pnas.1804177115.

Similarly, how were the errors of the radius of gyration for WT, R>K and Y>A mutants calculated? Is the Rg for WT significantly smaller than the values for the two mutants? And are the differences in Rg between single-copy and multi-copy simulations statistically significant? I am asking since converging the Rg of IDPs of this length in all-atom MD is not easy.

The errors for Rg values correspond to the standard deviations of the underlying distributions and are reported in Figure 4A and B, together with the corresponding means and an assessment of statistical significance of the difference. In particular, the character of the distributions (especially, for 24-copy systems) also suggests significant differences. In order to deepen this part in the revised version, we have added a new supplementary table (Supplementary File 1 “Technical summary) where we have included the average values of Rg together with the standard deviations for all modeled systems. Due to distributions being non-Gaussian, we have estimated the significance of the differences in Rgs between single-copy and multicopy simulations, as well as WT and mutants, using Wilcoxon rank sum test with continuity correction, with the resulting p-values < 2.2 10-16 for all cases.

p. 12, line 251: Has the MIST formalism been validated for IDPs; if so please provide a reference.

In the present work, we have evaluated the configurational entropy using a mutual information expansion approach with maximum-spanning-tree (MIST) approximation in internal-coordinate (bond-angle-torsion) representation. The latter is particularly well-suited for the analysis of IDPs as it allows one to avoid a number of artifacts (e.g., due to fitting of disordered ensembles to the average structure) associated with the more widely-used Cartesian-coordinate-based quasi-harmonic approaches. In particular, the MIST approach was used previously for the analysis of disordered protein ensembles (https://doi.org/10.1021/acs.jctc.8b00100). Here it should also be noted that, since intramolecular couplings are in general lower in IDPs, this makes them even better suited for MIST as compared to globular proteins. We have highlighted these points on p. 13 of the revision.

p. 5, line 105, p. 16 line 334 and p. 18 line 283: It is not completely clear what the predictions are and what/which experiments they are compared to. On p. 16, exactly what does the analytical model predict? As far as I understand, the results from the MD simulations are input to the model, but I am probably missing something. Which concrete and testable predictions does the model enable?

A key contribution of the present work is the development of a quantitative model that treats the spatial organization of a biomolecular condensate across scales using two key properties of individual polymer chains in the condensate - their average valency and compactness. The main predictions of the model concern the presence of a particular scaling of condensate mass with its radius, M(R), as captured by the fractal dimension, and the consequences this has on condensate morphology across scales. In the present manuscript, we have taken the first steps in testing these predictions in four different contexts. First, we could show that the MD simulations indeed match the predictions of fractal scaling for the three smallest clusters, which relates to the discussion on p. 16 that the Reviewer refers to. Here, it is important to understand that MD simulations in the first instance just give the average valency and compactness of individual chains in the dense phase. These values are then input into the fractal scaling formalism, which is conceptually fully independent from MD simulations, to obtain the dependence of condensate mass on its radius, M(R), at any desired length scale. The analysis presented in Figure 5—figure supplement 1B and discussed on p. 16 shows that the predictions of fractal scaling for the first three smallest clusters indeed correspond to what is seen in MD. This is a non-trivial correspondence and can be taken as direct evidence that fractal organization is present even at the shortest scale, i.e. at the level of MD simulation boxes.

Second, the model was used to reconstruct the spatial organization of clusters of arbitrary size at the atomistic level (Figure 5A and B, Videos 4, 5, and 6), enabling a structural understanding of the organization of condensate interior. One direct practical application of such understanding concerns the nature of cavity sizes and interpretation of dextran partitioning experiments (p. 20). Moreover, as pointed above, differences in morphology of protein clusters propagate across scales, and can be qualitatively characterized by the analysis of microscopic images (see also discussion above). In particular, the model correctly predicts the difference in the behavior of WT and R>K as opposed to Y>A variants, solely based on the predicted fractal dimension they exhibit. Ultimately, however, static light scattering experiments would give the best possibility to test the model directly and will be the topic of our future work. In particular, the fractal formalism predicts significant regions of linear behavior in such curves in log-log representation, while the fractal dimension df, provides a quantitative point of comparison between theoretical predictions and experimental measurements (Figure 5C). These points have been further discussed on p. 21 of the revised manuscript.

p. 19, lines 408-411: The authors find that when building clusters of Y>A from the simulations they find filamentous structures that they suggest explain the aggregation of the Y>A variant at high concentrations. While that sounds like an intriguing suggestion, it would be useful with a bit more detail about the robustness of this observation. For example, the simulations of Y>A appear similar to that of R>K; are the differences in topology really significantly different?

Fractal dimension, dF, is the key parameter that defines self-similar organization of differently sized protein clusters according to the fractal model. Consequently, the difference in morphology between R>K and Y>A mutants is reflected in different values of dF for the two. In particular, with a dF of 1.63, the Y>A mutant is predicted to form low-dimensional clusters, straddling the range between a linear (1-dimensional) and a planar (2-dimensional object), unlike WT and R>K variants, which both exhibit dF values greater than 2. The qualitative behavior of the three variants, whereby WT and R>K result in spherical condensates and Y>A does not, is consistent with this. Notably, we have observed sporadic precipitates at high protein concentration in the Y>A mutant, which may be consistent with the predictions of the fractal model. However, the material properties and possible influence of sample impurities in the Y>A case at high concentrations remain unclear. Moreover, the sporadic nature of Y>A precipitates prevents an adequate statistical analysis. Hence, in the revised manuscript we refrain from commenting on these infrequently observed precipitates.

Regarding MD simulations, the morphological differences between Y>A and R>K proteins can already be seen at the level of individual proteins in multicopy simulations, highlighted by the significantly different distribution of Rg (Figure 4B). This distribution in the case of Y>A has a prominently long tail, which indicates the possibility of adopting significantly more elongated configurations. Due to the self-similarity principle, such differences in morphology may propagate across length scales. Importantly, a recent publication included the experimental study of the possibility of IDRs to form low dimensional fractal systems upon disruption of the LLPS tendency by polyalanine insertion in synthetic elastin-like polypeptides (Roberts et al., Nature Materials, 2018).

Finally, I would suggest that the authors make their code and data available in electronic format.

All sharable data has been made available as part of the article package. Due to the heterogeneous character of our analysis, we do not have a single master code to be shared, but rather a collection of different scripts in combination with different software packages as indicated in the Methods section of the manuscript (GROMACS, MATLAB, R, FracVAL).

Author Response

Reviewer #1 (Public Review):

This manuscript confirms previous studies suggesting a great deal of heterogeneity of gene expression at the neural plate border in early vertebrate embryos, as neural, placodal, neural crest, and epidermal lineages gradually segregate. Using scRNA-seq, the study expands previous studies by using far larger numbers of genes as evidence of this heterogeneity. The evidence for this heterogeneity and the change in heterogeneity over time is compelling.

Many studies have suggested that there is considerable heterogeneity of gene expression in the developing neural plate border as the neural, neural crest, placodal and epidermal lineages segregate. Although the evidence for such heterogeneity was strong, until the advent of scRNA-seq, the extent of this heterogeneity was not appreciated. By using scRNA-seq at different stages of chick development, the authors sought to characterize how this heterogeneity develops and resolves over time.

The work is technically sound, and the level of analysis of gene expression, clustering, synexpression groups, and dynamic changes in gene modules over time is state-of-the-art. A weakness of the results as they stand now is that the conclusions of the analysis are not tested by the authors and thus, are over-interpreted. Such tests could be performed in future studies either by gain- and loss-of-function experiments or by using lineage tracing to demonstrate that the cell states the authors observe - especially the "unstable progenitors" they characterize - are biologically meaningful. The data will nevertheless be a useful resource to investigators interested in understanding the development of different cell lineages at the neural plate border.

We thank the reviewer for the positive assessment of our work. We agree that our models will need to be tested experimentally in the future, however, this will require a substantial amount of work. We therefore opted to share our data as a resource to be used by the community.

Reviewer #2 (Public Review):

The study of Thiery et al. aims to elucidate how cells undergo fate decisions between neural crest and (pan-) placodal cells at the neural plate border (NPB). While several previous single-cell RNA-Seq studies in vertebrates have included neural plate border cells (e.g. Briggs et al., 2018; Wagner et al., 2018; Williams et al., 2022), these previous studies did not provide conclusive insights on cell fate decisions between neural crest and placodes, due to either the limited number of genes recovered, the limited number of cells sampled or the limited numbers of stages included. The present study overcomes these limitations by analyzing almost 18,000 cells at six stages of development ranging from gastrulation until after neural tube closure (8 somite-stage), with an average depth of almost 4000 genes/cell. Using this extensive and high-quality data set, the study first describes the timing of segregation of neural crest and placodal lineages at the NPB suggesting that at late neural fold stages (somite stage 4) most cells have decided between placodal and neural crest fates. It then identifies gene modules specific for neural crest and placodal lineages and characterizes their temporal and spatial expression. Focusing on an NPB-specific subset of cells, the study then shows that initially most of these cells co-express neural crest and placodal gene modules suggesting that these are undecided cells, which they term "border-located unstable progenitors" (BLUPs). The proportion of BLUPs decreases over time, while cells classified as placodal or neural crest cells increases, with few BLUPs remaining at late neural fold stages (and a few scattered BLUPs even at somite stage 8). Based on these findings, the authors propose a new model of cell fate decisions at the NPB (termed the "gradient border model"), according to which the NPB is not defined by a specific transcriptional state but is rather a region of undecided cells, which diminishes in size between gastrulation and neural fold stages due to more and more cells committing to a placodal or neural crest fate based on their mediolateral position (with medial cells becoming specified as neural crest and lateral cells as placodal cells).

The study of Thiery et al. provides an unprecedentedly detailed, methodologically careful, and well-argued analysis of cell fate decisions at the NPB. It provides novel insights into this process by clearly demonstrating that the NPB is an area of indecision, in which cells initially co-express gene modules for ectodermal fates (neural crest and placodes), which subsequently become segregated into mutually exclusive cell populations. The paper is very well written and largely succeeds in presenting the very complex strategy of data analysis in a clear way. By addressing the earliest cell fate decisions in the ectoderm and one of the earliest cell fate decisions in the developing vertebrate embryo, this study will have a significant impact and be of interest to a wide audience of developmental biologists. There are, two conceptual issues raised in the paper that require further discussion.

We thank the reviewer for the positive comments on our work and its significance; we have addressed the conceptual issues below and in the revised version of the manuscript.

First, the authors suggest that their data resolve a conflict between two previously proposed models, the "binary competence model" and the "neural plate border model". The authors correctly describe, that the binary competence model proposed by Ahrens and Schlosser (2005) and Schlosser (2006) suggests that the ectoderm is first divided into two territories (neural and non-neural), which differ in competence, with the neural territory subsequently giving rise to the neural plate and neural crest and the non-neural territory giving rise to placodes and epidermis (sequence of cell-fate decisions: ([neural or neural crest]-[epidermal or placodal]). This model was proposed as an alternative to a "neural plate border state model", which instead suggests that initially the NPB is induced as a territory characterized by a specific transcriptional state, from which then neural crest and placodes are induced by different signals (sequence of cell fate decisions: neural-[placodal or neural crest]-epidermal) (see Schlosser, 2006, 2014). Instead in this paper, the authors contrast the binary competence model with a model they call the "neural plate border" model according to which the NPB can give rise to all four ectodermal fates with equal probability. However, I think this misses the main point of contention since all previously proposed models are in agreement that initially the neural plate border region is unspecified and can give rise to all four fates and that lineage restrictions only appear over time. "Binary competence" and "Neural plate border state" model, differ, however, in their predictions about the sequence, in which these fate restrictions occur.

We appreciate the reviewer's thoughtful feedback, but respectfully disagree with their comment regarding the sequence of events predicted by the neural plate border (NPB) model. While the NPB model does suggest that the NPB is a transcriptionally distinct state, it does not make specific predictions about the sequence of fate decisions. Although several papers cited in the Schlosser 2006 and 2014 reviews suggest that the NPB gives rise to all four ectodermal fates, none of them (and, to the best of our knowledge, no other primary paper referring to the NPB model) specifically defines the sequence of fate specification from the NPB.

The key points of the NPB model are that the NPB is defined by overlapping expression of early neural/non-neural markers (which is also observed in Xenopus – see Pieper et al., 2012 supplementary material), contains progenitors for all four ectodermal fates, and that this "state" exists prior to the emergence of definitive neural crest and placodal cells.

To investigate the heterogeneity in the order of cell fate decisions at the NPB, we carried out additional pairwise co-expression analyses of forebrain, mid-hindbrain, neural crest, and placodal gene modules, which reveals multiple different hierarchies of cell fate choice depending on a cell's axial positioning, as shown in Figure 6-figure supplement 1.

Considering these findings, we have expanded our discussion of the previously proposed binary competence and neural plate border models to highlight how neither of these models is sufficient to fully characterize the heterogeneity in cell fate decisions observed in our study. We hope this clarification will help address any concerns the reviewer may have had about the NPB model and its implications for our results.

Second, the authors should be more careful when relating their data to the specification or commitment of cells. Questions of specification and commitment can only be tested by experimental manipulation and cannot be inferred from a transcriptome analysis of normal development. So the conclusion that the activation of placodal, neural and neural crest-specific modules in that sequence suggests a sequence of specification in the same temporal order (lines 706-709) is not justified. Studies from the authors' own lab previously showed that epiblast cells from pre-gastrula stages are specified to express a large number of NPB border markers including neural crest and panplacodal markers, when cultured in vitro (Trevers et al., 2018; see also Basch et al., 2006 for early specification of the neural crest), which is not easily reconciled with this interpretation. I am not aware of any experimental evidence that shows that a panplacodal regulatory state is specified prior to neural crest in the chick (although I may have missed this). In Xenopus, experimental studies have shown instead that neural crest is specified and committed during late gastrulation, while the panplacodal states are specified much later, at neural fold stages (Mancilla and Mayor, 2006; Ahrens and Schlosser, 2005). It may well be the case that the relative timing of neural crest and panplacodal specification is different between species (and such easy dissociability may even be expected from the perspective of the binary competence model).

We very much agree with the reviewer that the definitions and correct terminology is important and apologise for lack of clarity. We have reworded the text carefully.

The reviewer is correct: specification of neural crest, placodes and neural plate is observed very early in chick, prior to gastrulation. However, in specification experiments tissue is removed from its normal environment to reveal what it does autonomously in the absence of additional signals. In the current study, we assess the activation of gene modules in normal development. We have therefore reworded the text to avoid ‘specification’ in this context.

Reviewer #3 (Public Review):

The goal of this work was to better understand how cell fate decisions at the neural plate border (NPB) occur. There are two prevailing models in the field for how neural, neural crest and placode fates emerge: (i) binary competence which suggests initial segregation of ectoderm into neural/neural crest versus placode/epidermis; (ii) neural plate border, where cells have mixed identity and retain the ability to generate all the ectodermal derivatives until after neurulation begins.

The authors use single-cell sequencing to define the development of the NPB at a transcriptional level and suggest that their cell classification identified increased ectodermal cell diversity over time and that as cells age their fate probabilities become transcriptionally similar to their terminal state. The observation of a placode module emerging before the neural and neural crest modules is somewhat consistent with the binary competence model but the observation of cells with potentially mixed identity at earlier stages is consistent with the neural plate border model.

Differences in the timing of analyses and techniques used can account for the generation of these two original models, and in essence, the authors have found some evidence for both models, possibly due to the period over which they performed their studies. However, the authors propose recognizing the neural plate border as an anatomical structure, containing transcriptionally unstable progenitors and that a gradient border model defines cell fate choice in concert with spatiotemporal positioning.

The idea that the neural plate border is an anatomical structure is not new to most embryologists as this has been well-recognized in lineage tracing and transplantation assays in many different species over many decades.

We appreciate the reviewers comment and agree that the neural plate border has previously been characterised anatomically. However, many studies have applied the term literally in reference to a transcriptional state which is specified through the expression of ‘neural plate border specifiers’, prior to segregation of the placodes and neural crest. Here we highlight that treating the neural plate border as a definitive transcriptional state which can be identified through the expression of ‘neural plate border specifiers’ is false. Instead, we find these ‘specifiers’ are upregulated within either neural crest, placodal or neural cell lineages over time. Cells at the neural plate border co-express these alternate lineage markers and therefore predicted to be undecided.

The authors don't provide molecular evidence for transcriptional instability in any cells. It's a molecular term and phenomenon inaccurately applied to these cells that are simply bipotential progenitors.

We thank the reviewer for pointing this out; we have therefore refrained from using the term unstable and instead refer to the cells as ‘undecided’ as suggested by reviewer 2.

Lastly, there's no evidence of a gradient that fits the proper biochemical or molecular definition. Graded or sequential are more appropriate terms that reflect the lineage determination or segregation events the authors characterize, but there's no data provided to support a true role for a gradient such as that achieved by a concentration or time-dependent morphogen.

We agree with the reviewer that ‘gradient’ was misleading. We have now replaced ‘gradient’ with ‘graded’ and expanded figure 6 to highlight the graded co-expression of gene modules associated with alternate fates. We have changed the title to reflect this.

A limitation of the study is that much of it reads like a proof-of-principle because validation comes primarily from known genes, their expression patterns in vivo, and their subsequent in vivo functions. Thus, the authors need to qualify their interpretations and conclusions and provide caveats throughout the manuscript to reflect the fact that no functional testing was performed on any novel genes in the emerging modules classified as placode versus neural or neural crest.

We agree with the reviewer that we do not provide any functional data to validate our predictions; it is for this reason that we submitted the manuscript as a ‘resource’ to make our data available to the community.

Lastly, a limitation of gene expression studies is that it provides snapshots of cells in time, and while implying they have broad potential or are lineage fated, do not actually test and confirm their ultimate fate. Therefore, in parallel with their studies, the authors really need to consider, the wealth of lineage tracing data, especially single-cell lineage tracing, which has been performed using the embryos of the same stage as that sequenced in this study, and which has revealed critical data about the potential cells through when and where lineage segregation and cell fate determination occurs.

The reviewer rightly points out the significance of the classical experiments in the context of the neural plate border. However, only one of the mentioned studies (Bronner-Fraser and Fraser, 1989), analyses cells at a single-cell level and does not assess placodes, while the remaining studies use tissue transplantation or cell population labelling. Although these studies provide valuable insights, they do not examine the fate or potential of single cells, nor do they reveal the transcriptional signature of these progenitors.

Our findings emphasize the transcriptional heterogeneity at the neural plate border, suggesting that distinct subsets of neural plate border progenitors undergo varying sequences of fate restrictions. The upcoming challenge will be to conduct clonal analysis alongside scRNAseq to determine if neural plate border progenitors with similar transcriptional signatures experience the same fate restrictions or if external factors, such as cell-cell signalling, dictate cell fate choices.

We have amended the manuscript to clarify that predictions of fate decisions require future validation through lineage tracing. Additionally, we have acknowledged in the introduction that previous studies have demonstrated the intermingling of neural, neural crest, and placodal progenitors at the neural plate border.

Author Respone

Reviewer #1 (Public Review):

This article describes the application of a computational model, previously published in 2021 in Neuron, to an empirical dataset from monkeys, previously published in 2018 in eLife. The 2021 modeling paper argued that the model can be used to determine whether a particular task depends on the perirhinal cortex as opposed to being soluble using ventral visual stream structures alone. The 2018 empirical paper used a series of visual discrimination tasks in monkeys that were designed to contain high levels of 'feature ambiguity' (in which the stimuli that must be discriminated share a large proportion of overlapping features), and yet animals with rhinal cortex lesions were unimpaired, leading the authors to conclude that perirhinal cortex is not involved in the visual perception of objects. The present article revisits and revises that conclusion: when the 2018 tasks are run through the 2021 computational model, the model suggests that they should not depend on perirhinal cortex function after all, because the model of VVS function achieves the same levels of performance as both controls and PRC-lesioned animals from the 2018 paper. This leads the authors of the present study to conclude that the 2018 data are simply "non-diagnostic" in terms of the involvement of the perirhinal cortex in object perception.

We appreciate the Reviewer’s careful reading and synthesis of the background and general findings of this manuscript.

The authors have successfully applied the computational tool from 2021 to empirical data, in exactly the way the tool was designed to be used. To the extent that the model can be accepted as a veridical proxy for primate VVS function, its conclusions can be trusted and this study provides a useful piece of information in the interpretation of often contradictory literature. However, I found the contribution to be rather modest. The results of this computational study pertain to only a single empirical study from the literature on perirhinal function (Eldridge et al, 2018). Thus, it cannot be argued that by reinterpreting this study, the current contribution resolves all controversy or even most of the controversy in the foregoing literature. The Bonnen et al. 2021 paper provided a potentially useful computational tool for evaluating the empirical literature, but using that tool to evaluate (and ultimately rule out as non-diagnostic) a single study does not seem to warrant an entire manuscript: I would expect to see a reevaluation of a much larger sample of data in order to make a significant contribution to the literature, above and beyond the paper already published in 2021. In addition, the manuscript in its current form leaves the motivations for some analyses under-specified and the methods occasionally obscure.

We believe that our comments outline our rationale for focusing our current analysis on data from Eldridge et al. In brief, these data provide compelling evidence against PRC involvement in perception, and are the only such data with PRC-lesioned/-intact macaques that we were able to secure the stimuli for. As such, data from Eldridge et al. provide a singular opportunity to address discrepancies between human and macaque lesion data. For this reason, we propose the current work as a Research Advance Article type, building off of a manuscript that was previously published in eLife.

Reviewer #2 (Public Review):

The goal of this paper is to use a model-based approach, developed by one of the authors and colleagues in 2021, to critically re-evaluate the claims made in a prior paper from 2018, written by the other author of this paper (and colleagues), concerning the role of perirhinal cortex in visual perception. The prior paper compared monkeys with and without lesions to the perirhinal cortex and found that their performance was indistinguishable on a difficult perceptual task (categorizing dog-cat morphs as dogs or cats). Because the performance was the same, the conclusion was that the perirhinal cortex is not needed for this task, and probably not needed for perception in general, since this task was chosen specifically to be a task that the perirhinal cortex might be important for. Well, the current work argues that in fact the task and stimuli were poorly chosen since the task can be accomplished by a model of the ventral visual cortex. More generally, the authors start with the logic that the perirhinal cortex gets input from the ventral visual processing stream and that if a task can be performed by the ventral visual processing stream alone, then the perirhinal cortex will add no benefit to that task. Hence to determine whether the perirhinal cortex plays a role in perception, one needs a task (and stimulus set) that cannot be done by the ventral visual cortex alone (or cannot be done at the level of monkeys or humans).

There are two important questions the authors then address. First, can their model of the ventral visual cortex perform as well as macaques (with no lesion) on this task? The answer is yes, based on the analysis of this paper. The second question is, are there any tasks that humans or monkeys can perform better than their ventral visual model? If not, then maybe the ventral visual model (and biological ventral visual processing stream) is sufficient for all recognition. The answer here too is yes, there are some tasks humans can perform better than the model. These then would be good tasks to test with a lesion approach to the perirhinal cortex. It is worth noting, though, that none of the analyses showing that humans can outperform the ventral visual model are included in this paper - the papers which showed this are cited but not discussed in detail.

Major strength:

The computational and conceptual frameworks are very valuable. The authors make a compelling case that when patients (or animals) with perirhinal lesions perform equally to those without lesions, the interpretation is ambiguous: it could be that the perirhinal cortex doesn't matter for perception in general, or it could be that it doesn't matter for this stimulus set. They now have a way to distinguish these two possibilities, at least insofar as one trusts their ventral visual model (a standard convolutional neural network). While of course, the model cannot be perfectly accurate, it is nonetheless helpful to have a concrete tool to make a first-pass reasonable guess at how to disambiguate results. Here, the authors offer a potential way forward by trying to identify the kinds of stimuli that will vs won't rely on processing beyond the ventral visual stream. The re-interpretation of the 2018 paper is pretty compelling.

We thank the Reviewer for the careful reading of our manuscript and for providing a fantistics synthesis of the current work.

Major weakness:

It is not clear that an off-the-shelf convolution neural network really is a great model of the ventral visual stream. Among other things, it lacks eccentricity-dependent scaling. It also lacks recurrence (as far as I could tell).

We agree with the Reviewer completely on this point: there is little reason to expect that off-the-shelf convolutional neural networks should predict neural responses from the ventral visual stream, for the reasons outlined above (no eccentricity-dependent scaling, no recurrence) as well as others (weight sharing is biologically implausible, as well as the data distributions and objective functions use to optimize these models). Perhaps surprisingly, these models do provide quantitatively accurate accounts of information processing throughout the VVS; while this is well established within the literature, we were careless to simply assert this as a given without providing an account of these data. We appreciate the Reviewer for making this clear and we have changed the manuscript in several critical ways in order to avoid making unsubstantiated claims in the current version. We hope that these changes also make it easier for the casual reader to appreciate the logic in our analyses. First, in the introduction, we outline some of the prior experimental work that demonstrates how deep learning models are effective proxies for neural responses throughout the VVS. We also demonstrate this model-neural fit in the current paper using electrophysiological recordings, but also including comments about the limitation of these models raised by the Reviewer.

In the introduction we also more clearly demarcate prior contributions from our recent computational work, and highlight how models approximate the performance supported by a linear readout of the VVS, but fail to reach human-level performance.

Results from these analyses were essential to understanding the logic of the paper but previously (as noted by the Reviewer) this critical evidence was cited but not directly presented. We include a description to these we describe these data in the introduction more thoroughly, and substantial change Figure 1, in order to visualize these data (b).

Moreover, we include a over of the methods and data used to generate these plots in the results and methods sections.

While there is little reason to expect that off-the-shelf convolutional neural networks should predict neural responses from the ventral visual stream, we believe that these modifications to the manuscript (to the introduction and figure one, as well as the results and methods sections) make clear that these models are nonetheless useful methods for predicting VVS responses and the behaviors that depend on the VVS.

To the authors' credit, they show detailed analysis on an image-by-image basis showing that in fine detail the model is not a good approximation of monkey choice behavior. This imposes limits on how much trust one should put in model performance as a predictor of whether the ventral visual cortex is sufficient to do a task or not. For example, suppose the authors had found that their model did more poorly than the monkeys (lesioned or not lesioned). According to their own logic, they would have, it seems, been led to the interpretation that some area outside of the ventral visual cortex (but not the perirhinal cortex) contributes to perception, when in fact it could have simply been that their model missed important aspects of ventral visual processing. That didn't happen in this paper, but it is a possible limitation of the method if one wanted to generalize it. There is work suggesting that recurrence in neural networks is essential for capturing the pattern of human behavior on some difficult perceptual judgments (e.g., Kietzmann et al 2019, PNAS). In other words, if the ventral model does not match human (or macaque) performance on some recognition task, it does not imply that an area outside the ventral stream is needed - it could just be that a better ventral model (eg with recurrence, or some other property not included in the model) is needed. This weakness pertains to the generalizability of the approach, not to the specific claims made in this paper, which appear sound.

We could not agree more with the Reviewer on these points. It could have been the case that these models' lack of correspondence with known biological properties (e.g. recurrence) led them to lack something important about VVS-supported performance, and that this would derail the entire modeling effort here. Surprisingly, this has not been the case, as is evident in the clear correspondence between model performance and monkey data in Eldridge et al. 2018. Nonetheless, we would expect that other experimental paradigms should be able to reveal these model failings. And future work evaluating PRC involvement in perception must contend with this very problem in order to move forward with this modeling framework. That is, it is of critical importance that these VVS models and the VVS itself exhibit similar failure modes, otherwise it is not possible to use these models to isolate behaviors that may depend on PRC.

A second issue is that the title of the paper, "Inconsistencies between human and macaque lesion data can be resolved with a stimulus-computable model of the ventral visual stream" does not seem to be supported by the paper. The paper challenges a conclusion about macaque lesion data. What inconsistency is reconciled, and how?

It appears that this point was lost in the original manuscript; we have tried to clarify this idea in both the abstract and the introduction. In summary, the cumulative evidence from the human lesion data suggest that PRC is involved in visual object perception, while there are still studies in the monkey literature that suggest otherwise (e.g. Eldridge et al. 2018). In this manuscript, we suggest that this apparent inconsistency is, in fact, simply a consequence of reliance on information interpretations of the monkey lesion data.

We have made substantive changes to the abstract so this is an obvious, central claim.

We have also made substantive changes to the introduction to make resolving this cross-species discrepancy a more central aim of the current manuscript.

Author Response

Reviewer #1 (Public Review):

This manuscript describes experiments that lead to a potentially impactful result and most of the data seem very nice. The authors conducted a mutant screen to find the gene BbCrpa from a fungus resistant to cyclosporine A (CsA). Microscopy indicates that the mode of action is likely sequestration of the toxin in vacuoles, mediated through the P4-ATPase pathway. They also show that expression of BbCrpa in Verticillium renders that fungus resistant to CsA. The paper then takes a very large jump across kingdoms and toxins and asks if BbCrpa, expressed in plants, will confer resistance to a different toxin (cinnamon acetate) that is produced by Verticillium. They conduct disease assays on Arabidopsis and cotton and show promising results, but these assays are less thoroughly completed. They provide microscopic evidence that the transgenics accumulate CIA in vacuoles, which is consistent with the mode of action of the other systems. Overall, my assessment of the paper is that the authors may have a nice story, but the transition to plants needs to be better described and potentially supported by additional experiments. For example, the authors seem to conclude that this resistance mechanism will be a very broad spectrum. Is there a second toxin-producing pathogen that could be used to assess whether this is true?

Thanks for your advice! To answer the question that "Is there a second toxin-producing pathogen that could be used to assess whether this is true?", we added the data of another t toxin-producing pathogens, Fusarium oxysporum and another V. dahliae race, L2-1, to support the conclusion that BbCrpa can confer resistance of plants against pathogens. As expected, the expression of BbCRPA in Arabidopsis and cotton could also significantly increase the resistance to the pathogens we tested. New data were shown in Figure 5-figure supplement 1G-J.

Reviewer #2 (Public Review):

The fungus B. bassiana is one of few fungal species resistant to cyclosporine A and tacrolimus, naturally occurring microbial compounds with antifungal and immunosuppressive properties. The authors studied the mechanism of this resistance and found a novel vesicle-mediated transport pathway that directs the compounds to vacuoles for degradation. This hitherto unknown mode of detoxification is initiated by the activity of a phospholipid flippase of the P4-ATPase type. Interestingly, transgenically expressing the fungal flippase in plant model systems induces a similar detoxification pathway and makes the plants resistant to certain fungal toxins of secondary metabolism.

Strengths

The genetic screening, isolation of cyclosporine A (CsA) resistant mutants, and characterization of the causative gene BbCrpa are very solid with two independent alleles, a synthetic knockout strain, and rescue of the mutant phenotype.

BbCrpa protein function in detoxification is demonstrated convincingly by expression in another CsA-sensitive strain, as is its reliance on sites conveying ATPase activity for proper function. It is also functionally different from a relatively closely related P4-ATPase from yeast.

Using fluorescently labeled CsA and tacrolimus (FK506), it is nicely demonstrated how the compounds are going through the anterograde pathway all the way to the vacuole.

The authors demonstrate that vacuolar targeting is key for the detoxifying function of BbCrpa and identify the targeting motif that contains a ubiquitination site.

A trans-species approach (actually, trans-kingdom) confers that BbCrpa can also enhance vacuolar targeting of small toxic compounds to vacuoles in plants, which is quite astounding, given that plant endomembrane transport has quite a number of differences from that of fungi.

Weaknesses

It is not clear at which temporal scale CsA is going through the different endosomal compartments.

Thanks for your comments! We agree your idea that it is better to provide indication of the temporal scale of CsA entering into different endosomal compartments. Actually, we had tried to trace the distribution of CsA in cells many times. Unfortunately, the fluorescence of 5-FAM is weak and decreases fast compared with eGFP or mRFP protein, which made us difficult to capture the transient localization and moving trace of CsA in the cells. Nevertheless, the trail of BbCrpa, which carried CsA from the vesicles to early/late endosome, and vacuoles, can reflect the pathway of the cargo (Figure 3K, Supplementary file 1).

Can it be ruled out that the fluorescently labeled CsA and the GFP-tagged BbCrpa are stripped off their label and we are seeing the free label only?

Thanks! We accept your comments. In order to rule out the interference from the cleaved fluorescent proteins (i.e., eGFP and mRFP) or chemical compound (i.e. 5-FAM), we took eGFP/mRFP and 5FAM as control. New data about the localization of eGFP/mRFP and 5-FAM in B. bassiana hypha were provided in the revised manuscript. Our observation indicated that the distribution of fluorescent materials alone is different with the labeled ones, confirming the bona fide localization of the fusion proteins or compound. Please see Figure 2-figure supplement 2.

Reviewer #3 (Public Review):

In this manuscript, the authors have attempted to determine the molecular mechanisms underlying the resistance of an insect fungal pathogen Bauveria barbicans to cyclosporine A (CsA) and tacrolimus (FK506), known antifungal secondary metabolites that are also used extensively as immunosuppressing agents in medicine. By screening the random insertion mutant library of this pathogen, they identified the gene responsible for conferring resistance to CsA and FK506. The amino acid sequence of the gene identified it to be P4-ATPase, designated BbCrpa, which was hypothesized to be involved in vesicle-mediated transport. The identity of this gene as a CsA resistance gene was confirmed by demonstrating that disruption of this gene in B. barbicans confers susceptibility to CsA and FK506 and the expression of the wild-type gene in the BbCRPA knockout strain restores resistance to these compounds. In addition, expression of this gene in a plant pathogen Verticillium dahliae confers resistance to CsA and FK506.

The authors hypothesized that CsA/FK506 detoxification in the resistant B. barbiana strain used is through the BbCRPA-mediated vesicle transport process transporting these toxic metabolites to vacuoles through trans-Golgi (TGN)-early endosome (EE)-late endosomes (LE) pathway. To test this hypothesis, they employed a dual labeling system using 5-carboxyfluorescein fluorescently labeled CsA and FK506 and fusions of red fluorescent proteins (RFP) with BbRab5 GTPase (a marker for early endosomes), BbRab7 GTPase (a marker for late endosome) and pleckstrin homology domain of human oxysterol binding protein (PHOSBP) (a marker for trans-Golgi). By looking at the distribution of fluorescein-labeled CsA and FK506 in the wild-type and ΔBbCRPA cells using confocal microscopy, the authors have provided compelling evidence that these metabolites are transported to the vacuole. The co-localization of CsA with endocytic marker proteins also appears to be convincing for the most part. The co-localization of CsA with mRFP:: PHOSBP as shown in Fig. 2D seems less compelling. Also, in the confocal micrographs presented in Fig. 2, the distinction between early and late endosomes seems less convincing. It seems that there is significant heterogeneity in the early endosome and late endosome populations in the fungal cells.

1) The co-localization of CsA with mRFP::PHOSBP as shown in Fig. 2D seems less compelling.

Thanks a lot! According to your suggestion, we repeated the observation and replaced the original figure with new one (Figure 2D). The new figures clear indicates the co-localization of CsA with mRFP::PHOSBP.

2) Also, in the confocal micrographs presented in Fig. 2, the distinction between early and late endosomes seems less convincing. It seems that there is significant heterogeneity in the early endosome and late endosome populations in the fungal cells.

Thanks! We agree with your comments! Rab5 is widely used as a marker for early endosomes, while Rab7 is used as a marker for late endosomes. Nevertheless, early endosome and late endosome are hardly to be distinguished strictly. According to your suggestion, we repeated the observation, and replaced the Figure 2F and 2L, and Figure 3E with new ones. Our results indicated that mRFP::BbRab5 appeared largely in the lumen of vacuoles and some punctaes (early endosomes), and mRFP::BbRab7 locates to vacuolar membrane and late endosomal compartments. These can be seen in our observations in Figure 3D and 3E, which are consistent with the observations in Fusarium graminearum described by Zheng et al. (Zheng et al., 2018, New Phytologist, 219: 654671, DOI: 10.1111/nph.15178).

The authors addressed the question of whether BbCrpa acts as a component involved in vesicle trafficking through the trans-Golgi-endosomes to vacuoles. Ten different eGFP-BbCrpa fusion proteins were constructed and shown to provide detoxification of CsA and FK506. The BbCrpa is localized to the apical plasma membrane and spitzenkorper region of the germ tube. The evidence for localization of BbCrpa in trans-Golgi and vacuole is clear. However, the experimental data shown in Fig. 3D-F claiming localization of BbCrpa in EEs and LEs are somewhat difficult for this reviewer to interpret. It is also not clear to this reviewer why the two FM4-64 staining patterns in Fig. 3C and Fig. 3F are strikingly different. The evidence for co-localization of the fluorescein-labeled CsA or FK506 with RFP-labeled BbCrpa in vacuoles (Fig.3 H and J) is convincing. Figs. 3L-M depicting dynamic trafficking of BbCrpa from TGN to vacuoles using timelapse microscopy is interesting. In Fig. 3M, eGFP should be labeled eGFP::Drs2p. The authors have identified the N-terminal vacuole targeting motif in BbCrpa and shown that the C-terminal sequence from aa1326 to aa1359 is important for detoxification of CsA and FK506 in B. barbiana. In particular, the importance of three Tyr residues located in the C-terminal domain of the enzyme for CsA resistance is interesting.

1) The experimental data shown in Fig. 3D-F claiming localization of BbCrpa in EEs and LEs are somewhat difficult for this reviewer to interpret.

Thanks for your comments! P4-ATPases are implicated in the initiation of vesicle biogenesis and moves along with the vesicle (Panatala et al., 2015, Journal of Cell Science, 128: 2021-2032, DOI: 10.1242/jcs.102715; van der Mark et a., 2013, International Journal of Molecular Sciences, 14, 7897-7922, DOI: 10.3390/ijms14047897). In this study, the crucial issue to be addressed was the journey of CsA to the vacuole, which might be through BbCrpa-mediated TGN-EE-LE vesicle transport pathway. Therefore, we observed the localization of BbCrpa in TGN, EEs and LEs. It has been reported that small GTPase Rab5 is localized to the early endosomes (Bucci et al., 1992, Cell, 70: 715-728, DOI: 10.1016/0092-8674(92)90306-w), while Rab7 is to the late endosomal compartment (Vitelli et al., 1997, The Journal of Biological Chemistry, 272: 4391-4397, DOI: 10.1074/jbc.272.7.4391). Thus, Rab5 and Rab7 are used as marker proteins to indicate EEs and LEs, respectively. Nevertheless, Rab5 and Rab7 could also be observed in MVB (multivesicular bodies) or vacuoles (Toshima J.Y.,et al., 2014, Bifurcation of the endocytic pathway into Rab5-dependent and -independent transport to the vacuole, Nature Communication, 5:3498, DOI: 10.1038/ncomms4498; Zheng et al., 2018, New Phytologist, 219: 654-671, DOI: 10.1111/nph.15178). According to your suggestion, we repeated our observation and replaced Figure 3E with new one. In Figure3D, we can see mRFP::BbRab5 appeases largely in the lumen of vacuoles and some punctaes (early endosomes), and in Figure 3E mRFP::BbRab7 locates to late endosomal compartments and vacuolar membrane, which are consistent with the observations in Fusarium graminearum described by Zheng et al.(Zheng et al., 2018, New Phytologist, 219: 654671, DOI: 10.1111/nph.15178).

2) It is also not clear to this reviewer why the two FM4-64 staining patterns in Fig. 3C and Fig. 3F are strikingly different.

Thanks for your comments! FM4-64 is a styryl dye that can bind to the outer lipid leaflet of the plasma membrane and enter into cells through endocytosis (Scheuring et al., 2015, Methods in Molecular Biology, 1242:83-92, DOI: 10.1007/978-1-4939-1902-4_8). When the dye is internalized, it can be observed firstly in the membrane of vesicles and endosomal compartments, and then appears in the vacuolar membrane (Jelníková et al., 2010, Plant Journal, 61(5): 883-892, DOI: 10.1111/j.1365-313X.2009.04102.x; Löfke et al., 2013, Journal of Integrative Plant Biology, 55(9): 864-875, DOI: 10.1111/jipb.12097). In Figure 3C, we tried to show the evidence that eGFP::BbCrpa appears in vesicles that are stained by FM4-64, while in Figure 3F, we aimed to indicate eGFP::BbCrpa accumulates in mature vacuoles. Hence, FM4-64 staining patterns in Figure 3C and Figure 3F are somehow different.

3) In Fig. 3M, eGFP should be labeled eGFP::Drs2p.

Thanks for your reminder! We have modified it according to the suggestion. Please see Figure 3M.

Finally, the authors overexpressed BbCrpa gene in transgenic Arabidopsis and cotton plants to show that transgenic plants expressing this enzyme are protected from the toxic effects of the toxin cinnamyl acetate (CA) produced by the fungal pathogen Verticillium dahlia which causes vascular wilt disease in these plants. The data reported in Fig. 5A show that the transgenic Arabidopsis seed is able to germinate in presence of CA, whereas the nontransgenic control seed is not able to germinate. Evidence is presented that CA accumulates in the vacuole in transgenic Arabidopsis. However, the seedlings emerging from transgenic seeds are only partially protected from CA (Fig. 5A). It is also clear from the data presented in Figs. 5B-G that expression of the BbCrpa gene in transgenic Arabidopsis and cotton affords protection from infection by V. dahlia although no evidence for the expression of this gene at the protein level is presented. However, it seems likely that the transgenic lines only show delayed disease symptoms and are not truly resistant to this pathogen. The authors did not state clearly if Verticillium wilt disease resistance assays were performed on homozygous transgenic plants and their corresponding null segregants as negative controls. They also fail to provide evidence that the transgenic Arabidopsis and cotton challenged with the pathogen are able to grow to maturity and set viable seeds.

1）However, the seedlings emerging from transgenic seeds are only partially protected from CA (Fig. 5A).

Thanks for your comments! In this study, at the concentration of 50 μg/ml, the germination of wildtype seeds of Arabidopsis was severely inhibited while the transgenic seeds were still able to germinate. However, the growth of transgenic seedling was suppressed obviously compared with that of the untreated seedlings (Figure 5A). According to your suggestion, in the revised manuscript, we used “tolerance”, rather than “resistance” to weaken the statement.

2) However, it seems likely that the transgenic lines only show delayed disease symptoms and are not truly resistant to this pathogen. The authors did not state clearly if Verticillium wilt disease resistance assays were performed on homozygous transgenic plants and their corresponding null segregants as negative controls. They also fail to provide evidence that the transgenic Arabidopsis and cotton challenged with the pathogen are able to grow to maturity and set viable seeds.

Thanks for your comments! In our routine procedure for generating transgenic plant lines, we identified non-transgenic plants in the segregative generation (usually in T1 generation) of transformats, and then used these non-transgenic plants (null lines) as control to rule out the somatic variation from tissue culture. In the meantime, the homologous transgenic plants were identified in the segregative generation and propagated by selfing. Relevant descriptions were added in the section of Methods & Materials (Lines 783-795).

We agree with your opinion. The resistance displayed in seedlings does not always match that in maturity stage, because the resistance of host to Verticilium pathogen can be affected by environmental conditions, for example, temperature and nutrition. Nevertheless, in the case of our transgenic plants, the resistance to Verticilium disease is endowed from the detoxic function of BbCrpa. Theoretically, such transgenic traits will not be significantly affected by the environmental conditions if the expression of transgenes is stable. We had detected the expression level of BbCRPA during plants growth and in deferent generation (to T5 generation). The expression of BbCRPA gene was stable and the resistance to the diseases is descendible.

Author Response

Reviewer #1 (Public Review):

This is a well-conceived and well-executed investigation of how activation loop autophosphorylation and IN-box autophosphorylation synergistically activate AURKB/INCENP. An elegant chemical ligation strategy allowed construction of the intermediate phospho-forms so that the contributions of each phosphorylation event to structure, dynamics, and activity could be dissected. Autophosphorylation at both sites serves to rigidify both AURKB and the IN-box, and to coordinate opening, twisting, and activation loop movements. Consistent with previous findings, both sites are necessary for enzymatic activity; further, this work finds that activation loop autophosphorylation occurs slowly in cis while INbox autophosphorylation occurs quickly in trans.

Due to abundant previous work in the field, many of the conclusions of this paper were expected. However, that does not diminish the quality of the work, and the addition of how kinase dynamics contribute to activation is important for AURKB and many other kinases. The experimental results are clear and interpreted appropriately, with good controls. The computational work is also clearly explained and directly tied to the function of the enzyme, making it highly complementary to the experimental findings and to previously published structures.

We thank the reviewer for positive words about our work.

Some minor limitations of the study:

1) Of note when interpreting the HDX data, there is no coverage of the peptide containing the activation loop autophosphorylation site T248 (Fig S2A), and as mentioned in the Discussion, the time scale of HDX is not able to capture differences in exchange in very flexible regions like the activation loop.

The peptides spanning the region containing the phosphorylated Aurora BThr248 are not shown in our coverage map because they do not meet the stringent quality criteria for peptides that we used for HDX analysis (see Material and Methods). However, we have compared these peptides in phosphorylated and unphosphorylated enzyme complex manually and added a paragraph 4 on page 4.

“The peptides spanning the region containing the phosphorylated Aurora BThr248 are not shown in our coverage map (Figure 1-figure supplement 2A) because they did not pass the stringent peptide quality filter based on intensity, and redundancy of the peptide. However, upon manual analysis, we did not detect any changes in deuterium uptake between the phosphorylated and unphosphorylated forms in this region. Deuterium exchange in this part of the protein (which is also observed in the peptides immediately upstream of Aurora BThr248, see Supplementary file 5) is very rapid, independently of enzyme phosphorylation, so that complete exchange occurs even at the earliest time points. This is in contrast to the second part of the activation loop, which includes the Aurora BaEF helix (labeled region 3 in Figure 1A; peptide 254-260 in Supplementary file 5), where we clearly see HDX protection upon phosphorylation. It is possible that phosphorylation causes dynamic changes on a very fast scale (seconds or faster) in the part of the protein encompassing Aurora BThr248, but we could not detect them due to the limitation of our approach, which operates on the scale of minutes.“

Also, we analyzed the peptides covering this region to confirm the extent of phosphorylation in the loop (Figure 3-figure supplement 2A).

2) Some data lack robust statistical analysis, which would make the findings more compelling.

We have now included statistical analysis throughout the paper where it was possible.

3) One point that might be clarified is how the occupancy of T248 was confirmed to be either fully phosphorylated in the [AURKB/IN-box]IN-deltaC or fully dephosphorylated in the IN-box K846N/R827Q mutant. Especially because T248 autophosphorylation is found to occur in cis, it is unclear how incubating the [AURKB/IN-box]IN-deltaC with traces of wild-type [AURKB/IN-box]all-P would ensure that T248 is phosphorylated.

We confirmed phosphorylation occupancy of Aurora BThr248 by mass spectrometry in [Aurora B/IN-box]allP and [Aurora B/IN-box]loop-P but not for [AURKB/IN-box]no-P (Figure 3-figure supplement 2A). To achieve complete phosphorylation in the activation loop of [Aurora B/IN-box]IN-DC, we incubated this construct with fully active [Aurora B/IN-box]all-P. This is because, according to previous kinetic analysis, cisphosphorylation of Aurora BThr248 is only obligatory when the entire enzyme population is in the nonphosphorylated state. Once a fraction of the Aurora B population has been partially or fully activated, phosphorylation of Aurora BThr248 can also occur in trans, by the already activated enzyme. In other words, our model proposes an obligatory initial intramolecular step followed by propagation of activation in trans as reported by (Zaytsev, Segura-Peña at al, eLife.2016). We have now clarified this on page 11, paragraph 2 where we explain the results of the autoactivation kinetics.

“It is noteworthy that phosphorylation of the activation loop in cis is the first necessary step in the autoactivation process, assuming a completely unphosphorylated enzyme pool. However, a partially active or fully active enzyme can phosphorylate the activation loop in trans. This type of activation mechanism with an initial intramolecular activation step followed by an intermolecular step of activation have been previously reported for PAK2 (J. Wang et al., 2011) and for Aurora B (Zaytsev et al., 2016).”

Reviewer #2 (Public Review):

This study presents a dynamic, multi-step model for the activation of Aurora-B kinase through the interaction with INCENP and autophosphorylation. This interaction is critical to the proper execution of chromosome segregation, and key details of the mechanism are not resolved. The study is an advance on previous studies on Aurora-B and the related kinase Aurora-C, primarily because it clarifies the roles of the different phosphorylation sites. However, major differences in the details of the molecular interactions are presented that are not clearly backed up by the evidence due to limitations in the approach, when compared to previous work based on crystal structures.

Strengths. The experimental approach to the analysis of the Aurora-B/INCENP interaction is sound and novel and it is striking example of preparation of proteins in specific phosphorylation states, and of using HDX to characterise localised changes in the structural dynamics of a protein complex. The authors have generated two intermediate phosphorylation states of the complex, enabling them to dissect their contributions to the regulation of structural dynamics and activity of the complex.

Weaknesses. The major weakness of the study is the molecular dynamics simulation. The resulting model of the complex differs from the crystal structure of the Aurora-C/IN-box structure in key details, and these are neither described clearly nor explained. The challenges/limitations of simulation of phosphorylated proteins should be described.

We thank the reviewer for the positive words about our work and the criticism that helped us to improve our manuscript. We have now extended the MD studies that confirm our original observations regarding the entropic nature of the IN-box and the effect of phosphorylation on the structure and dynamics of [Aurora B/IN-box]. We clarify that the conformation of [Aurora B/IN-box]all-P observed in the simulations is not the final folded state, but a productive intermediate in the activation pathway of [Aurora B/IN-box]. For this reason, the differences from the [Aurora C/IN-box] crystal structure do not indicate flaws in the simulations. On the contrary, we believe that the data from the MD simulations provide crucial insights into the dynamical properties of the system that could not otherwise be assessed.

Author Response

Reviewer #2 (Public Review):

1) It has been reported that PHD fingers can bind to DNA in addition to lysine-methylated histone H3. Can the authors address whether or not the enhanced selectivity of PHD-nucleosome interactions over PHD-peptide interactions is due to PHD-DNA binding?

We apologize for not making this clearer in our initial manuscript. We did test the ability of our PHD readers to bind nucleosomal DNA of various lengths and observed no significant engagement (Figure 1 - figure supplement 1B). This is emphasized in the revised text.

2) What's the binding affinities of PHD-nucleosome interactions and PHD-peptide interactions, respectively?

The relative EC50 (EC50rel) for these interactions (Figure 1 - figure supplement 1C-D and Figure 2 - figure supplement 1H) are consistent with others using Alpha/dCypher technologies (e.g. doi.org/10.1101/ 2022.02.21.481373v1; which also contains a detailed description of EC50rel calculation and the difference between this value and an equilibrium Kd).

3) Histone H4K5acK8ac is a well-known site-specific histone acetylation mark for gene transcriptional activation, much more so than histone H3 acetylation. Does H4K5K8 acetylation enhance PHD-H4K3me3 binding in nucleosome?

We appreciate the reviewer for asking this question. In our studies, we tested H4K5ac and H4K8ac binding individually but do not have a nucleosome with the dual H4K5ac8ac nucleosome. Given the limited amount of time and resources we had for making more nucleosomes, we felt our efforts were better spent on developing heterotypic nucleosomes to answer the more striking cis vs. trans question posed by both reviewers.

4) The authors provided the data showing cis histone H3 tail lysine acetylation effects on PHDH4K3me3 binding. What about trans histone H3 lysine acetylation effects?

Thank you for this suggestion. To address this, we expended considerable resources to create new fully PTM-defined heterotypic (to accompany our homotypic) nucleosomes (note nomenclature to minimize confusion with asymmetric/symmetric DNA methylation) to directly test whether MLL1’s activity enhancement in the context of H3 tail acetylation occurs in cis or in trans. As shown in Figure 2D, enhancement of H3K4 methylation only occurs with heterotypic nucleosomes that have an available H3K4 residue with tail acetylation in cis (H3K4me3 • H3K9acK14acK18ac (hereafter H3triac)) and is not seen in H3K4 methylatable nucleosomes with tail acetylation in trans (H3 -> H3K4acK9acK14acK18ac (hereafter H3tetraac)). These exciting new findings greatly strengthen our study and provide more definitive mechanistic details of H3ac → H3K4me regulation.

Author Response

Reviewer #2 (Public Review):

This manuscript reports an experiment involving learning and imaging of neural activity in rats. The goal was to test if scopolamine, which is an antagonist of acetylcholine receptors, could cause memory loss (amnesia). Two types of learning were tested: first, rats learned to prefer a short path, compared to a detour path, between two rewarded locations in a linear maze; second, in a subset of the experimental sessions, a shock zone was activated in the middle of the short path, and rats had to learn to avoid it. As a control, some sessions had a clear plexiglass barrier placed in the middle of the short path, which should not have aversive properties. The order of sessions was different for different groups of rats, but shock learning was always followed by a number of 'extinction' sessions without shock. In some groups, shock learning was accompanied by a systemic (intraperitoneal) scopolamine injection, 30 min before the start of the session. This manipulation was performed on most rats once but at a different slot in the sequence of sessions (sometimes before the drug-free shock learning, sometimes after it, sometimes in the absence of preceding barrier sessions, sometimes after them). In what follows, I might use the terms 'control group' and 'test group' to refer to sessions without scopolamine and sessions with it, respectively.

The main behavioural results are that rats increase their visits to the short path with learning, then visit less the short path once the shock zone is active or when the barrier is there. When re-tested in later sessions, rats trained in the absence of the scopolamine injection still avoid the short path, while most of the rats that were given the scopolamine injection do not avoid it, suggesting a deficit in encoding or recall of the shock zone memory.

In addition to these behavioural manipulations, the authors image the activity of dorsal CA1 hippocampal neurons using calcium imaging. They detect the existence of place cells, which increase their firing on specific portions of the short path of the maze (the long path data is not analysed). When comparing data before the shock training to during shock training, control place cells were more stable (i.e. had increased between-session correlations) and had more recurrent place fields (i.e. spatially active in one session then still active in another session) with respect to test data (rats injected with scopolamine). When comparing the pre-shock session and the first extinction session, place cell activity was less similar in the control rats (no scopolamine) compared to the scopolamine rats; but note that for scopolamine rats, extinction occurred earlier, so instead of using the first extinction session, the 4th session post-shock training was used to match the control data. Place cell activity has been shown to allow decoding of the animal's position; here, position decoding accuracy was lower around the shock zone in the control group compared to the scopolamine group.

From these analyses, the manuscript proposes the following findings: 1) scopolamine injections impair avoidance learning, and 2) scopolamine affects the long-term response of place cells to the aversive experience (less "remapping"). These findings are interpreted to support the idea that 3) place cell remapping is involved in avoidance/aversive learning and 4) that scopolamine, as an antagonist of the muscarinic acetylcholine receptors in the hippocampus (or elsewhere?), produces amnesia.

These findings, if properly supported, would be very interesting to a wide range of researchers interested in the neural bases of memory and learning, specifically aversive memory and spatial learning. The manuscript is well-written, has the advantage of using both male and female rats (which have consistent results), is one of the rare studies to date to perform calcium imaging of the hippocampus in rats, and records along learning of two simple tasks which seems relatively ecological and produce robust learning effects, and uses an experimental manipulation (injection of scopolamine) instead of purely correlational measures. The authors make some analytical effort in equalizing the number of trials across different sessions (even though I do not believe this fixes the existing confounds). I particularly appreciated the nice '3D' trajectory plots that show the unfolding of behaviour along a given session and that each individual's data are generally shown in the figures.

However, the experiment and its analysis seem to have some major flaws both in the experimental design (which may be difficult to fix) as well as in the analysis (which might be easier to fix), which prevent proper interpretation of the results. Specifically:

• To demonstrate finding 1 (that scopolamine specifically impairs avoidance learning), a control would be needed to show that the injection procedures do not impair general behaviour (e.g. motivation, attention, level of stress) as well as other forms of learning. Indeed, the control rats - as far as I understood - are not injected with saline, which would have been an appropriate control. One example of non-specific confounding effects of scopolamine is that it could, for example, reduce sensitivity to pain, thus to some extent decreasing the relevance of the shock to the rats, but many other interpretations are possible; most revolve around the idea that instead of scopolamine impairing learning, scopolamine might impair behaviour, which might, in turn, impair learning. Indeed, the scopolamine injection is shown to decrease running speed and the number of trials run, even before exposure to the shock. Related to this, the "short path preference" does not seem to be quantified properly: instead of simply using the number of visits to the short path, a better measure would be to compute a relative preference index quantifying visits to the short path with respect to visits to the long path [e.g. (num short - num long) / num total], to focus on the preference regardless of global changes in behavioural activity levels. In summary: the proposition that scopolamine specifically impairs avoidance learning has not been convincingly demonstrated; the possibility that it even impairs any form of learning is not currently demonstrated either.

We have run an additional saline-only control group (along with additional scopolamine groups) to demonstrate that scopolamine does impair avoidance learning. As shown in the Supplement to Fig. 1, rats receiving saline alone show significantly greater post-training avoidance of the short path than mice receiving scopolamine.

• For similar reasons, finding 2 (that the place cell response to aversive learning is affected by the scopolamine injection) is subject to the same lack of controls and existence of possible confounds noted above. Specifically, running more slowly and running fewer laps would have affected the overall amount of excitation of place cells during the session, which might affect plasticity, as well as the amount of reactivations/replay at the reward sites, which is likely to have effects in terms of memory consolidation. One way to potentially control for this would be to have the control group run the same amount of trials as the test group (but then session durations would be different); it is unclear how to prevent the difference in running speed. To be able to claim that the effects of scopolamine are specific to aversive learning, a control with either no learning (perhaps the long path data would be useful for this?) or appetitive learning (e.g. of a reward location, which also involves place field reorganization in some cases) would be useful.

We recognize and understand the referee’s valid concerns about these potential confounds. The revised paper makes more clear that the scopolamine condition is designed as a control for whether an aversive stimulus is remembered or forgotten, and the barrier condition is a control for whether a novel path-blocking stimulus is aversive or neutral. As the referee points out, there are other confounding variables that may covary with these manipulated factors. The revised discussion (ll. 544-559) acknowledges potential confounds arising from learning-induced behavior changes, and argues that even though it is not possible to perfectly control for such changes, our current study does so more effectively than most prior studies because the rat’s behavior during isolated beeline trials is highly stereotyped and thus more similar across experimental conditions than in any prior study that we know of. The revised discussion also acknowledges the difficulty of dissociating acquisition versus extinction effects upon remapping (ll. 633-644). The significance of the barrier manipulation is now acknowledged more clearly in the abstract, introduction, and discussion.

• Statement 3 implies a causal link between the two first statements, suggesting that place cell remapping would be necessary for the memory of an aversive experience or aversive location. Given the weakness of the arguments supporting the first 2 statements, this is also non convincingly demonstrated. In any case, the current paradigm would not be sufficient to make a causal link, but it might be sufficient to show a correlational link by showing a correlation between the amount of remapping and memory performance, such as presented in supplementary figure 5 (which would still be informative even if results from the scopolamine sessions were removed?).

We agree with the referee’s point that our study’s evidence is correlational in nature. The revised manuscript prominently acknowledges this in the concluding sentence of the discussion’s opening paragraph (ll. 500-503), which now reads: “While these results do not definitively prove that place cell remapping is causally necessary for storing memories of aversive encounters, they provide correlational evidence that remapping occurs selectively under conditions where a motivationally significant (rather than neutral) stimulus occurs and is subsequently remembered rather than forgotten.”

• Statement 4 - that scopolamine causes amnesia - is both not fully defined (what form of amnesia?) and not supported by the findings for the reasons mentioned above.

It is unclear what remedy the referee is recommending for this concern. We do not present the idea that scopolamine is an amnesic drug as a novel conclusion of our study; rather, this is a widely accepted view in the literature that motivated our experimental design decision to use scopolamine as a tool for dissociating whether an aversive event was remembered or forgotten. We recognize that scopolamine’s effects on memory may vary with experimental conditions. In the revised discussion, we extensively compare and contrast our experimental findings with acute and chronic results from numerous prior studies using scopolamine and other cholinergic drugs (ll. 561-678). We hope this is sufficient to address the referees concerns.

• In addition to these concerns, a study cited here (Sun et al 2021) mentions a few references in the discussion regarding how muscarinic receptor agonists might affect the link between spikes and calcium signals. Scopolamine is a muscarinic receptor antagonist and might thus have related/reversed effects. Thus, the technique used (calcium imaging) does not seem the best to address questions related to scopolamine. The current manuscript also mentions some findings that were not replicated (e.g. lack of over-representation of the shock zone) which are probably due to the fact that the finding relied on extra-field isolated spikes, which are less likely to be detected via calcium imaging.

This is an excellent point which is now addressed in the revised discussion; it is acknowledged (ll. 536-539) that mAChRs can regulate calcium signals and thus that scopolamine may have different effects on spikes detected from calcium imaging versus electrophysiology, which in turn could account for discrepancies between our finding that place fields do not migrate to aversively reinforced locations and Milad et al.’s (2019) findings that they do. The Sun et al. (2021) paper used calcium imaging methods similar to ours, so this factor is less likely to account for the discrepancy between their prior finding that place fields were acutely disrupted by scopolamine and our current finding that they were not.

• If anything, perhaps targeted injections in a specific brain region (e.g. dorsal CA1), instead of systemic injection, might give a more precise picture of the effects of scopolamine on place cells and spatial memory, but I do not know if this is technically possible.

Unfortunately the necessary placement of the GRIN lens above the recording location prevented the direct application of scopolamine there via cannulae. To date only one series of experiments has demonstrated single unit place cell recordings with direct microdialysis (Brazhnik et al. 2003, 2004). To study the effects of aversive learning across many days, we needed to utilize a recording method capable of tracking many cells across long time periods. However, systemic scopolamine has been widely used to study both learning (Anagnostaras et al. 1999; Huang et al. 2011; Svoboda et al. 2017) and its effects on place cells (Douchamps et al. 2013; Newman et al. 2017; Sun et al. 2021), thus by utilizing this method we can directly compare our findings with previous work. We have added a paragraph to the discussion (ll. 665-678) in which it is explained why the main conclusions of our study (namely, that place cell cell remapping is related to storage of memories for aversive events) do not depend upon whether or not scopolamine’s pharmacological actions were localized to the hippocampus (almost certainly they were not).

My conclusion would be that the experiment either needs to be redesigned to address the original question (effect of scopolamine on place cell firing and aversive learning) or that some of the data could be still used to address different questions which have not been addressed with calcium imaging before, e.g. learning of the short path, activity on the short path vs long path, effects on behaviour and place cell activity of learning & extinction of the barrier and shock zone avoidance; perhaps without focusing on the scopolamine manipulations, which seem to introduce many confounds.

Author Response

Reviewer #2 (Public Review):

Kim et al. examined the properties of neuronal connections responsible for inhibitory cell activation to show that the characteristics examined were similar in humans and rodents. This is important, as it suggests that the many rodent studies carried out over the past decades are physiologically relevant to humans.

Strengths

1) Human brain tissues are difficult to obtain, hence the study provides valuable insights

2) An impressive multipronged approach was used for cell classifications

3) Despite the lack of novel findings, the revelation of the similarities between human and rodent synapses is important and has far-reaching implications. This important finding suggests the knowledge generated from rodent research is, at least partly, physiologically relevant to and transferrable to humans.

Weaknesses

1) The study is descriptive by design, and hence provides limited conceptual advances, especially with the retrospect that synaptic properties are similar between humans and rodents (although see strength #3). For example, very similar findings and techniques have already recently been reported by a number of the same authors in the Campagnola et al., Science 2022 paper.

We agreed that stimulus protocols of connectivity assays with multiple patch-clamp recordings in this study had been adapted from the recent publication (Campagnola et al., Science 2022). In this previous study, especially for human synaptic connectivity data, the main cell type categorization was at the level of excitatory and inhibitory neurons which identified based on morphological features and observed PSP characteristics (e.g., direction of membrane potential changes) when it connected each other. However, we went further to identify interneuron subclasses in the connectivity assays using virally labeled slice cultures and post-hoc HCR staining in addition to intrinsic classifier, which is not investigated from the recent publication (Campagnola et al., 2022). Therefore, following scientific findings and their implications are not the same shown in the previous study and we think this study provides a significant advance of our understanding in human cortical circuits organization.

2) Despite the fact that normal physiology was reported, the use of pathological human brain tissue could affect the results.

We agreed that the use of pathological human brain tissue to investigate normal physiology is not ideal, however, as mentioned in the METHODS below (section of “Acute slice preparation”), our surgically resected neocortical tissues show minimal pathology, and we believe these tissue preparations can be used to address normal physiological properties of human neurons. Importantly, we saw no effect of disease state (epilepsy vs. tumor) on the intrinsic or synaptic properties that we measured. Our METHODS state that “Surgically resected neocortical tissue was distal to the pathological core (i.e., tumor tissue or mesial temporal structures). Detailed histological assessment and using a curated panel of cellular marker antibodies indicated a lack of overt pathology in surgically resected cortical slices (Berg et al., 2021).”. We also state in the RESULTS that “These tissues were distal to the epileptic focus or tumor, and have shown minimal pathology when examined (Berg et al., 2021). Brain pathology was evaluated using six histological markers that were independently scored by three pathologists. Surgically resected tissues have been used extensively to characterize human cortical physiology and anatomy (Berg et al., 2021).”. Lastly, this is the best possible human tissue available for us to conduct physiological experiments. It is an unavoidable caveat of this work that our healthy brain tissue was derived from a donor brain exhibiting a serious disease.

3) The manuscript may not be easy to understand for the uninvited, because many concepts and abbreviations were not properly introduced.

Thank you for pointing this oversight out. We updated our manuscript and made sure that we fully describe all abbreviations. We now changed the abbreviation of MPC back to multiple patch-clamp recording, and some other abbreviations such as LAMP5, SLC17A7, DLX are now better explained. We have also changed the order of multiple figures (i.e., Figure 5 – Figure supplements to Figure 3 – Figure supplements) and removed some complicated figures (e.g., Figure 1 – Figure supplement 1) to present the data in a fashion that can be understood by a more general reader.

4) The statistical treatment is not ideal, so some conclusions may not be valid.

We performed additional statistical analyses as suggested and implemented in the text of the RESULTS.

Furthermore, we also made additional Figure supplements (Figure 4 – Figure supplement 3, Figure 4 – Figure supplement 4, Figure 6 – Figure supplement 2, and Figure 6 – Figure supplement 3) to support our conclusions.

5) The mixed usage of acute and cultured slices is not ideal and likely affects the outcome.

We agree that the mixed usage of acute and cultured slices is not ideal, and it could affect the interpretation of outcome. Therefore, we performed additional analyses to see if there is any correlated change of synaptic property (i.e., paired pulse ratio) along the days after slice culture (now implemented in Figure 4 – Figure supplement 4 and Figure 6 – Figure supplement 3) and we didn’t find any significant correlation. However, we noticed the short-term synaptic dynamics are rather differentiated between acute and slice culture condition shown in Figure 4 – Figure supplement 1d. We think this is due to sampling bias rather than tissue preparation difference and these points are now more carefully described in the DISCUSSION as “This difference we observed in this study, i.e., more facilitating synapses were detected in slice cultures than in acute slices, could either reflect an acute vs. slice culture difference. However, we believe it is more likely to reflect a selection bias for PVALB neurons when patching in unlabeled acute slices, and that the AAV-based strategy with a pan-GABAergic enhancer allows a more unbiased sampling of interneuron subclasses whose properties are preserved in culture. In support of this, PPR analysis as a function of days after slice culture shows no relationship to acute versus slice culture preparation (Figure 4 – Figure supplement 4, Figure 6 – Figure supplement 3). Furthermore, we have observed that viral targeting of GABAergic interneurons greatly facilitates sampling of the SST subclass in the human cortex compared to unbiased patch-seq experiments (Lee et al., 2022), and this selection bias likely explains synapse type sampling differences in cultured slices compared to acute preparations.”.

Author Response

We would like to extend our thanks to the reviewers who took the time to carefully read our paper and provide thoughtful insights and suggestions on how to strengthen our conclusions. All reviewers agreed that our study presented strong data supporting a role for triglyceride lipase brummer (bmm) in regulating testis lipid droplets and spermatogenesis in Drosophila, and that our findings advance our understanding of lipid biology during sperm development. Reviewers also made several helpful suggestions on how to strengthen our manuscript even further. Below, we provide a brief outline of our plans to revise this manuscript in response to reviewer comments.

The majority of reviewer comments will be addressed by text changes, rearranging figures to add images, and making a model to visually represent our findings. Together, these changes will ensure we clearly communicate our data and conclusions with readers, and properly contextualize our findings. See below for details on our planned revisions.

Reviewer #1 (Public Review):

In this study, the authors investigate the role of triglycerides in spermatogenesis. This work is based on their previous study (PMID: 31961851) on triglyceride sex differences in which they showed that somatic testicular cells play a role in whole body triglyceride homeostasis. In the current study, they show that lipid droplets (LDs) are significantly higher in the stem and progenitor cell (pre-meiotic) zone of the adult testis than in the meiotic spermatocyte stages. The distribution of LDs anti-correlates with the expression of the triglyceride lipase Brummer (Bmm), which has higher expression in spermatocytes than early germline stages. Analysis of a bmm mutant (bmm[1]) - a P-element insertion that is likely a hypomorphic - and its revertant (bmm[rev]) as a control shows that bmm acts autonomously in the germline to regulate LDs. In particular, the number of LDs is significantly higher in spermatocytes from bmm[1] mutants than from bmm[rev] controls. Testes from males with global loss of bmm (bmm[1]) are shorter than controls and have fewer differentiated spermatids. The zone of bam expression, typically close to the niche/hub in WT, is now many cell diameters away from the hub in bmm[1] mutants. There is an increase in the number of GSCs in bmm[1] homozygotes, but this phenotype is probably due to the enlarged hub. However, clonal analyses of GSCs lacking bmm indicate that a greater percentage of the GSC pool is composed of bmm[1]-mutant clones than of bmm[rev]-clones. This suggests that loss of bmm could impart a competitive advantage to GSCs, but this is not explored in greater detail. Despite the increase in number of GSCs that are bmm[1]-mutant clones, there is a significant reduction in the number of bmm[1]-mutant spermatocyte and post-meiotic clones. This suggests that fewer bmm[1]-mutant germ cells differentiate than controls. To gain insights into triglyceride homeostasis in the absence of bmm, they perform mass spec-based lipidomic profiling. Analyses of these data support their model that triglycerides are the class of lipid most affected by loss of bmm, supporting their model that excess triglycerides are the cause of spermatogenetic defects in bmm[1]. Consistent with their model, a double mutant of bmm[1] and a diacylglycerol O-acyltransferase 1 called midway (mdy) reverts the bmm-mutant germline phenotypes.

There are numerous strengths of this paper. First, the authors report rigorous measurements and statistical analyses throughout the study. Second, the authors utilize robust genetic analyses with loss-of-function mutants and lineage-specific knockdown. Third, they demonstrate the appropriate use of controls and markers. Fourth, they show rigorous lipidomic profiling. Lastly, their conclusions are appropriate for the results. In other words, they don't overstate the results.

We thank the Reviewer for their positive assessment of our paper.

There are a few weaknesses. Although the results support the germline autonomous role of bmm in spermatogenesis, one potential caveat that the mdy rescue was global, i.e., in both somatic and germline lineages. The authors did not recover somatic bmm clones, suggesting that bmm may be required for somatic stem self-renewal and/or niche residency. While this is beyond the scope of this paper, it is possible that somatic bmm does impact germline differentiation in a global bmm mutant.

In the revised manuscript, we will more clearly delineate when we used global versus germline-only loss of mdy to rescue bmm mutant phenotypes in the testis. We will also acknowledge the possibility that somatic bmm may play a role in germline differentiation in a global bmm mutant.

Regarding data presentation, I have a minor point about Fig. 3L: why aren't all data shown as box plots (only Day 14 bmm[rev] does). Finally, the authors provide a detailed pseudotime analysis of snRNA-seq of the testis in Fig. S2A-D, but this analysis is not sufficiently discussed in the text.

We will make text and presentation changes in the revised manuscript to describe our data more clearly, and will add text to describe our pseudotime analysis of single-cell RNA seq data in more detail.

Overall, the many strengths of this paper outweigh the relatively minor weaknesses. The rigorously quantified results support the major aim that appropriate regulation of triglycerides are needed in a germline cell-autonomous manner for spermatogenesis.

This paper should have a positive impact on the field. First and foremost, there is limited knowledge about the role of lipid metabolism in spermatogenesis. The lipidomic data will be useful to researchers in the field who study various lipid species. Going forward, it will be very interesting to determine what triglycerides regulate in germline biology. In other words, what functions/pathways/processes in germ cells are negatively impacted by elevated triglycerides. And as the authors point out in the discussion, it will be important to determine what regulates bmm expression such that bmm is higher in later stages of germline differentiation.

We agree with the reviewer about the many interesting future directions for this project. We will therefore add a model figure in the revised manuscript to visualize our findings and highlight remaining questions about how bmm and triglycerides support normal spermatogenesis in Drosophila.

Reviewer #2 (Public Review):

Summary:

Here, the authors show that neutral lipids play a role in spermatogenesis. Neutral lipids are components of lipid droplets, which are known to maintain lipid homeostasis, and to be involved in non-gonadal differentiation, survival, and energy. Lipid droplets are present in the testis in mice and Drosophila, but not much is known about the role of lipid droplets during spermatogenesis. The authors show that lipid droplets are present in early differentiating germ cells, and absent in spermatocytes. They further show a cell autonomous role for the lipase brummer in regulating lipid droplets and, in turn, spermatogenesis in the Drosophila testis. The data presented show that a relationship between lipid metabolism and spermatogenesis is congruous in mammals and flies, supporting Drosophila spermatogenesis as an effective model to uncover the role lipid droplets play in the testis.

We thank the Reviewer for their positive assessment of our paper.

Strengths and weaknesses:

The authors do a commendably thorough characterization of where lipid droplets are detected in normal testes: located in young somatic cells, and early differentiating germ cells. They use multiple control backgrounds in their analysis, including w[1118], Canton S, and Oregon R, which adds rigor to their interpretations. The authors employ markers that identify which lipid droplets are in somatic cells, and which are in germ cells. The authors use these markers to present measured distances of somatic and germ cell-derived lipid droplets from the hub. Because they can also measure the distance of somatic and germ cells with age-specific markers from the hub, these results allow the authors to correlate position of lipid droplets with the age of cells in which they are present. This analysis is clearly shown and well quantified.

The quantification of lipid droplet distance from the hub is applied well in comparing brummer mutant testes to wild type controls. The authors measure the number of lipid droplets of specific diameters, and the spatial distribution of lipid droplets as a function of distance from the hub. These measurements quantitatively support their findings that lipid droplets are present in an expanded population of cells further from the hub in brummer mutants. The authors further quantify lipid droplets in germline clones of specified ages; the quantitative analysis here is displayed clearly, and supports a cell autonomous role for brummer in regulating lipid droplets in spermatocytes.

Data examining testis size and number of spermatids in brummer mutants clearly indicates the importance of regulating lipid droplets to spermatogenesis. The authors show beautiful images supported by rigorous quantification supporting their findings that brummer mutants have both smaller testes with fewer spermatids at both 29 and 25C. There is also significant data supporting defects in testis size for 14-day-old brummer mutant animals compared to controls. The comparison of number of spermatids at this age is not significant, which does not detract from the the story but does not support sperm development defects specifically caused by brummer loss at 14 days. Their analysis clearly shows an expanded region beyond the testis apex that includes younger germ cells, supporting a role for lipid droplets influencing germ cell differentiation during spermatogenesis.

We thank the reviewer for pointing out this inaccuracy in our manuscript. In the revised manuscript we will choose more precise language to describe defects in sperm development in 14-day-old bmm mutants.

The authors present a series of data exploring a cell autonomous role for brummer in the germline, including clonal analysis and tissue specific manipulations. The clonal data indicating increased lipid droplets in spermatocyte clones, and a higher proportion of brummer mutant GSCs at the hub are convincing and supported by quantitation. The authors also show a tissue specific rescue of the brummer testis size phenotype by knocking down mdy specifically in germ cells, which is also supported by statistically significant quantitation. The authors present data examining the number of spermatocyte and post-meiotic clones 14 days after clonal induction. While data they present is significant with a 95% confidence interval and a p value of 0.0496, its significance is not as robust as other values reported in the study, and it is unclear how much information can be gained from that specific result.

We thank the reviewer for raising this point. In the revised manuscript we will display the p-value clearly to ensure our statistical output is clear for readers to evaluate our conclusions regarding bmm mutant clones 14 days after clone induction.

The authors do a beautiful job of validating where they detect brummer-GFP by presenting their own pseudotime analysis of publicly available single cell RNA sequencing data. Their data is presented very clearly, and supports expression of brummer in older somatic and germline cells of the age when lipid droplets are normally not detected. The authors also present a thorough lipidomic analysis of animals lacking brummer to identify triglycerides as an important lipid droplet component regulating spermatogenesis.

Impact:

The authors present data supporting the broad significance of their findings across phyla. This data represents a key strength of this manuscript. The authors show that loss of a conserved triglyceride lipase impacts testis development and spermatogenesis, and that these impacts can be rescued by supplementing diet with medium-chain triglycerides. The authors point out that these findings represent a biological similarity between Drosophila and mice, supporting the relevance of the Drosophila testis as a model for understanding the role of lipid droplets in spermatogenesis. The connection buttresses the relevance of these findings and this model to a broad scientific community.

Reviewer #3 (Public Review):

In this manuscript, Chao et al seek to understand the role of brummer, a triglyceride lipase, in the Drosophila testis. They show that Brummer regulates lipid droplet degradation during differentiation of germ and somatic cells, and that this process is essential for normal development to progress. These findings are interesting and novel, and contribute to a growing realisation that lipid biology is important for differentiation.

We thank the Reviewer for their positive assessment of our manuscript.

Major comments:

1) The data in Figs 1 and 2, while helpful in setting the scene, do not add much to what was previously shown by the same group, namely that lipid droplets are present in both early germ cells and early somatic cells in the testis, and that Bmm regulates their degradation (PMID: 31961851). Measuring the distance of lipid droplets from the hub, while helpful in quantifying what is apparent, that only stem and early differentiated stages have lipid droplets, is not as informative as the way data are presented later (Fig. 2I), where droplets in specific stages are measured. Much of this could be condensed without much overall loss to the manuscript.

We thank the reviewer for this comment and will condense the first part of the paper in our revised manuscript.

2) It would be important to show images of the clones from which the data in Fig. 2I are generated. The main argument is that Bmm regulates lipid droplets in a cell autonomous manner; these data are the strongest argument in support of this and should be emphasised at the expense of full animal mutants (which could be moved to supplementary data).

We thank the reviewer for this comment, and will add an image in our revised manuscript showing lipid droplets in bmm mutant spermatocyte clones.

Similarly, the title of Fig. S2 ("brummer regulates lipid droplets in a cell autonomous manner") should be changed as the figure has no experiments with cell (or cell-type)-specific knockdowns/mutants. This figure does show changes in lipid droplets in both lineages in bmm mutants, so an appropriate title could be "brummer regulates lipid droplets in both germ and soma".

We thank the reviewer for this comment, and will adjust the S2 figure legend title in the revised manuscript.

3) Interestingly, the clonal data show that bmm is dispensable in germ cells until spermatocyte stages, as no increase in lipid droplet number is seen until then. This should be more clearly stated, as it indicates that the important function of Bmm is to degrade lipid droplets at the transition from spermatogonial to spermatocyte stages. This is consistent with the phenotypes observed in which late stage germ cells are reduced or missing. However, the effect on niche retention of the mutant GSCs at the expense of neighbouring wildtype GSCs is hard to explain. Are lipid droplets in mutant GSCs larger than in control? Is there any discernible effect of bmm mutation on lipids in GSCs? Additionally, bam expression is delayed, suggesting that bmm may have roles on cell fate in earlier stages than its roles that can be detected on lipid droplets.

We thank the reviewer for this comment. We will include more text in the revised manuscript to clarify the key role bmm plays in regulating lipid droplets at the spermatogonia-spermatocyte transition. We will also add more detail and potentially data to our description of how bmm affects lipid droplets in cells at the earliest stages of germline development.

4) The bmm loss-of-function phenotype could be better described. Some of the data is glossed over with little description in the text (see for example the reference to Fig. 3A-C). For instance, in the discussion, the text states "loss of bmm delays germline differentiation leading to an accumulation of early-stage germ cells" (p13, l.259-60). However, this accumulation has not been clearly shown, or at least described in the manuscript. Most of the data show a reduction (or almost complete absence) of differentiated cell types. This could indeed be due to delayed differentiation, or alternatively to a block in differentiation or to death of the differentiated cells. The clonal data presented show a decrease in the number of cells recovered, but do not allow inferences as to the timing of differentiation, making it hard to distinguish between the various possibilities for the lack of differentiated spermatids. Apart from data showing that GSCs are more likely to remain at the niche, no further data are shown to support the fact that mutant germ cells accumulate in early stages. While additional experiments could help resolve some of these issues, much of this could also be resolved by tempering the conclusions drawn in the text.

We thank the reviewer for these comments. In the revised manuscript we will temper our conclusions regarding bmm’s precise role in spermatogenesis by discussing different mechanisms (e.g. differentiation or death) that could lead to the phenotypes we observe.

5) In the discussion (p.14, l-273 onwards), the authors suggest that products of triglyceride breakdown are important for spermatogenesis. However, an alternative interpretation of the results presented here (especially those using the midway mutant) could be that triglycerides impede normal differentiation directly. Indeed, preventing the cells' ability to produce triglycerides in the first place can rescue many of the defects observed. A better discussion of these results with a model for the function of triglycerides and their by-products would be a great improvement to this manuscript.

We thank the reviewer for this comment. To ensure our data is clearly communicated with readers, we will add a model to the paper suggesting how triglyceride and its by-products influence spermatogenesis.

Together, these changes will strengthen our overall finding that bmm-mediated regulation of testis triglyceride is important for normal sperm development. Because our findings in flies align with and extend data from rodent models, the developmental mechanisms we uncovered about how triglyceride lipase bmm regulates testis lipid droplets and sperm development will likely operate in other species.

Author Response

Reviewer #1 (Public Review): 

“I recommend that the authors revisit their calculation methods to provide a more convincing conclusion on the presence of positive epistasis for fitness in their dataset.” 

The reviewer is right that the present description of the fitness calculation can be found insufficient. Below we provide relevant derivation. It will be included in our planned revision, probably as a supplementary text: 

Expected fitness effect of multiple mutations

Fitness is the number of offspring divided by the number of progenitors, w\=No/Np. This can be the number of cells left by one cell (including itself in the case of budding cells) over a unit of time. 

Assume that an organism carries multiple mutations—α, β, … ω—which are in heterozygous loci, their wild-type counterparts are marked universally with +. The fitness effect of a single mutation is wα/+, and so on. Fitness can be converted to relative fitness, i.e., expressed as a quotient of the wild-type fitness, wα/+/w+/+, and so on. Under the multiplicative model of mutation accumulation, an expected joint effect of multiple mutations on relative fitness is a product of individual quotients: 

wexp/w+/+ = (wα/+/w+/+) (wβ/+/w+/+) … (wω/+/w+/+). 

When a population is continuously growing, log-transformation of fitness is typically applied as it equates the rate of growth. In particular, it could be the number of doublings completed over a unit of time: 

log2(wexp/w+/+) = log2[(wα/+/w+/+) (wβ/+/w+/+) … (wω/+/w+/+)]. 

After replacing the above log multiplicative formula with its log additive equivalent, all its terms can be normalized by dividing by log2 fitness of the wild type which turns them into relative doubling rates, for example, log2(wα/+)/log2(w+/+)=rDRα/+. The joint effect of multiple mutations is then equal to 

rDRexp = 1 + (rDRα/+  ̶ 1) + (rDRβ/+  ̶ 1) + … + (rDRω/+  ̶ 1) 

or 

rDRexp = 1 + ∑d  

where d\=rDR ̶ 1 (see Fig. 2B in the main text). 

Reviewer #2 (Public Review):

“The initiation and interpretation of the results were apparently performed in a vacuum of a century of work on genomic balance.” 

Indeed, we neither introduce nor discuss results obtained with organisms other than the budding yeast. We accept that researchers working with other beasts may expect to see such considerations. We will introduce them into a revised submission, to the extent allowable for non-review articles. The suggestions provided by the reviewer will be followed. 

“If there is an increase in the general transcriptome size, then there might not be much reduction of the proteosome subunits as claimed and the increases might be somewhat less than indicated.” 

- together with – 

“A second experiment that would clarify the results would be to perform estimates of the general transcriptome size. If the general transcriptome size is actually increased, the claims of reduced expression of the proteosome might need to be revised.” 

Multicellular organisms may be well heterogenic across their bodies, in terms of the cell number and (transcriptome) composition. We believe that any proper sampling of their mRNA requires careful absolute, and not only relative, quantification. We worked with clones prepared to be mostly homogeneous (about two-three divisions under conditions promoting strong growth). We maintain that we are allowed to rely on chromosomal averages expressed as proportions of the total mRNA. Our monosomic counts were very strictly around 50% of those predicted for euploids, we do not see any danger of erroneous sampling or calculation in our wonderfully simple case. Regarding the problem of a possible general increase in the transcriptome size which would help to compensate for the decrease in the proteasomic mRNAs, we adopted a strictly linear interpretation. That is, in a doubled transcriptome not only mRNAs for the proteasome but also all other proteins (prey species for the proteasome) would be doubled and thus no relief in proteolysis would happen. We follow here previous findings that in yeast the fractions of individual mRNAs are reflected in the fractions of ribosome-bound mRNA fragments and (at least roughly) mature proteins (e.g., the cited by us Larrimore et al. 2020). 

“The claim of Torres et al that there are no global modulations in trans is counter to the knowledge that transcription factors are typically dosage sensitive and have multiple targets across the genome … Taken as a whole it would seem to suggest that there are many inverse relationships of global gene expression with chromosomal dosage in both yeast disomies and monosomies.” 

Well, the debate about mRNA compensation in relation to aneuploidy in yeast has been intense and sometimes heated. Disparate claims can be found but we are left with an overwhelming impression that the relation between the total amount of mRNA and the number of chromosome copies is pretty much (perhaps not ideally) linear. We will consider our wording again. We will admit that such a rigid relation is somewhat unusual compared to other eukaryotes. But again, we see strict halves of mRNA for the monosomic chromosomes.  

“To clarify the claims of this study, it would be informative to produce distributions of the various ratios of individual gene expression in monosomy versus diploid as performed by Hou et al. 2018.” 

Such distributions are already prepared and will be likely presented in a revised msc. 

“The authors claim there are no genes that are compensated on the varied chromosome but considering how many genes are upregulated across the genome, it would seem that a subset are probably upregulated on the cis chromosome as well and approach the diploid level, i.e. are dosage compensated.” 

Perhaps we misstated our conclusions somewhere but it was obvious to us that some genes were upregulated on the cis chromosome (monosomic), some other were downregulated, the net result was the average 50% (Fig. 3). 

Reviewer #3 (Public Review): 

“1) In Figure 3b (and line 179) …  What the data really show is that the level of overexpression is not correlated with the fitness effect of the deletion (since all the p values are not significant). The authors need to correct their conclusions.” 

That’s right! We already mended it as we are preparing for revision. 

“2) Why are some monosomic strains removed from the transcriptomics analysis, especially when the chromosome IV and XV strains show very strong positive epistasis? The authors need to provide an explanation here.” 

We run out of money. In more scientific terms, we believe our sample of eight strains is unbiased and sufficient. It was truly randomly chosen. We were glad to see that it covers both slightly and most strongly affected (with profound epistasis) monosomics. All of them displayed parallel shifts in the transcriptome (RP up, proteasome down). We judged we could stop here, the additional five strains would be unlikely to change our main conclusions. 

“3) The authors stated that diploidy observed in chromosome VII and XIII strains were due to endoreplication after losing the marked chromosomes (lines 97 and 117). Isn't chromosome missegregation an equally possible explanation? Since monosomic cells are generated by chromosome missegregation during mitosis, another chromosome missegregation event may occur to rescue the fitness (or viability) of monosomic cells in these strains.” 

We believe that it happened in this way as the reviewer suggests, at least in most cases. By “endoreduplication”, we understand any event making two chromosomes of one, not necessarily additional DNA replication. We will check our text to make it clear in this respect.

Author Response

Reviewer #2 (Public Review):

The authors dissected the effects of mycolacton on endothelial cell biology and vessel integrity. The study follows up on previous work by the same group, which highlighted alterations in vascular permeability and coagulation in patients with Buruli ulcer. It provides a mechanistic explanation for these clinical observations, and suggests that blockade of Sec61 in endothelial cells contributes to tissue necrosis and slow wound healing. Overall, the generated data support their conclusions and I only have two major criticisms:

• Replicating the effects of mycolactone on endothelial parameters with Ipomoeassin F (or its derivative ZIF-80) does not demonstrate that these effects are due to Sec61 blockade. This would require genetic proof, using for example endothelial cells expressing Sec61A mutants that confer resistance to mycolactone blockade. The authors claimed in the Discussion that they could not express such mutants in primary endothelial cells, but did they try expressing mutants in HUVEC cell lines? Without such genetic evidence all statements claiming a causative link between the observed effects on endothelial parameters and Sec61 blockade should be removed or rephrased. The same applies to speculations on the role of Sec61 in epithelial migration defects in discussion. Data corresponding to Ipomoeassin F and ZIF-80 do not add important information, and may be removed or shown as supplemental information.

• While statistical analysis is done and P values are provided, no information is given on the statistical tests used, neither in methods nor results. This must be corrected, to evaluate the repeatability and reproducibility of their data.

We respectfully but fundamentally disagree with the comments regarding the Sec61 dependence of the effects that we observed. We showed that loss of glycocalyx and basement membrane components underpinned the phenotypic changes in endothelial cells (morphological changes, loss of adhesion, increased permeability, and reduced ability to repair scratch wounds). We demonstrated that we could phenocopy permeability increases and elongation phenotype by knocking down the type II membrane protein B3Galt6, and reverse the adhesion defect by exogenous provision of the secreted laminin-511 heterotrimer.

Our conclusion that mycolactone mediates these effects via Sec61 inhibition is not based solely on the use of alternative inhibitors but is built on several pillars of evidence:

First, the proteomics data conforms entirely to predictions based on the topology of affected vs. non-effected proteins, and agrees with independently published proteomic datasets from T lymphocytes, dendritic cells and sensory neurons (ref.12), as well as biochemical studies performed using in vitro translocation assays (ref.11,34). Furthermore, the pattern of membrane protein down regulation observed in our experiments fits perfectly with established models of protein translocation mechanisms, particularly with respect to the lack of effect on specific topologies of multipass membrane proteins, tail anchored- and type III membrane proteins (ref.34-36).

Second, since Sec61 very highly conserved amongst mammals and is found in all nucleated cells, it is hard to conceptualise a framework in which mycolactone targets Sec61 in some cells and not others, as this reviewer suggests might be the case for epithelial cells [noting that the work being referred to (ref.29) predates our 2014 work showing that mycolactone is a Sec61 inhibitor (ref.7)]. Indeed, mycolactone has been shown to target Sec61 in multiple independent approaches including forward genetic screens involving random mutagenesis and CRISPR/Cas9 (ref.10, PMID: 35939511). Genetic evidence has previously been provided for the Sec61 dependence of mycolactone effects in epithelial cells (ref.10,17). We have unpublished genetic evidence that the rounding and detachment of epithelial cells due to mycolactone is reduced when resistance mutations are over expressed, and will consider including this in the next version of the manuscript.

Third, given this weight of evidence, one would be hard-pressed to provide an alternative explanation for the specific down-regulation of glycosaminoglycan-synthesising enzymes and adhesion/basement membrane molecules while most cytosolic and non-Sec61 dependent membrane proteins are unchanged or upregulated. However, seeking to be as rigorous as possible we have here shown that a completely independent Sec61 inhibitor produces the same phenotype at the gross and molecular level. Ipomoeassin F (Ipom-F) is a glycolipid, not a polyketide lactone, yet they both compete for binding with cotransin in Sec61α (ref.6). There is significant overlap in the cellular responses to mycolactone and Ipom-F, including the induction of the integrated stress response (ref.17, PMID: 34079010), which we observed again in the current data, providing further evidence that this approach is useful when genetic approaches are technically unattainable.

Therefore, we are confident the effects seen on endothelial cells are Sec61-dependent. We are happy to provide more detail on our lengthy attempts at over-expressing mycolactone resistant SEC61A1 genes in HUVECs; primary endothelial cells derived from the umbilical vein. We are highly experienced in this area, and have previously stably expressed these proteins in epithelial cell lines, reproducing the resistance profile (ref.10,17). Notably though, these cells do not have normal ‘fitness’ in the absence of challenge. Since endothelial cells (and endothelial cell lines; PMID: 12560236) are extremely hard to transfect with plasmids, with efficiency routinely 5-10% (including in our hands), we developed a lentivirus system. We were eventually (after multiple attempts using different protocols) able to transduce primary HUVECs with constructs expressing GFP (at an efficiency of about 10-20%) and select/expand these under puromycin selection. Never-the-less, we never recovered any cells that expressed the flag-tagged SEC61A1 wild type or SEC61A1 carrying the resistance mutant D60G. We also attempted to select D60G-transduced cells with mycolactone epimers, an approach that can help the cells compete against non-transduced cells in culture flasks (ref.10). We concluded that primary endothelial cells are unable to tolerate the expression of additional Sec61α, and this was incompatible with survival.

It’s also important to note that most endothelial cell specialists would agree that endothelial cell lines are not good models of endothelial behaviour. We tested the HMEC-1 cell line, but found it did not express prototypical endothelial marker vWF in the expected way. Therefore we focussed our efforts on primary endothelial cells. Should we be able to overcome the dual challenge of the necessity to work in primary cells, and the difficulty of over-expressing Sec61, we will update this paper at a later date with this data, and will also expand the above arguments.

We apologise for the embarrassing oversight of not including information about the statistical analyses we used, which of course we will correct in full in the revised version. However, we would like to provide this information to readers of the current version of the manuscript. All data were analysed using GraphPad Prism Version 9.4.1:

Figure 1: one-way ANOVA with Dunnett’s (panel A) or Tukey’s (panel B) correction for multiple comparisons

Figure 2 supplement: one-way ANOVA with Tukey’s correction for multiple comparisons (analysed panel)

Figure 3: one-way ANOVA with Tukey’s (panel B) or Dunnett’s (panel E&F) correction for multiple comparisons

Figure 4: one-way ANOVA with Dunnett’s correction for multiple comparisons (all analysed panels)

Figure 5 and supplement: one-way ANOVA with Dunnett’s correction for multiple comparisons (all analysed panels)

Figure 6: one-way ANOVA with Dunnett’s correction for multiple comparisons (analysed panel)

Figure 6 supplement: one-way ANOVA with Dunnett’s correction for multiple comparisons (all analysed panels)

Figure 7: two-way ANOVA with Tukey’s correction for multiple comparisons (all analysed panels; panels B&C also included the Geisser Greenhouse correction for sphericity)

Figure 7 supplement: Panels A&D used a repeated measures one-way ANOVA with Dunnett’s correction for multiple comparisons (panel D also included the Geisser Greenhouse correction for sphericity). Panels B,C&E used a two-way ANOVA with Tukey’s correction for multiple comparisons (panels B&C also included the Geisser Greenhouse correction for sphericity)

Author Response:

We would like to thank the reviewers and editor for their insightful comments and suggestions. We will update the manuscript accordingly. We are particularly glad to read that our software package constitutes a set of “well-written analysis routines” which have “the potential to become very valuable and foundational tools for the analysis of neurophysiological data”. Both reviewers have identified a number of weaknesses in the manuscript, and we would like to take this opportunity to provide a response to some of the remarks and clarify the objectives of our work. We would like to stress that this kind of toolkit is in continual development, and the manuscript offered a snapshot of the package at one point during this process. Since the initial submission several months ago, several improvements have been implemented and further improvements are in development by our group and a growing community of contributors. The manuscript will be updated to reflect these more recent changes, some which will directly address the reviewers’ remarks.

It was first suggested that the manuscript should better showcase the value of the analysis pipeline. As noted by the first reviewer, the online repository (i.e. GitHub page) conveys a better sense of how the toolbox can be used than the present manuscript. Our original intention was to illustrate some examples of data analysis in Figure 4 by adding the corresponding Pynapple command above each processing step. Each step takes a single line of code, meaning that, for example, one only needs to write three lines of code to decode a feature from population activity using a Bayesian decoder (Fig. 4a), or to compute a cross-correlograms of two neurons during specific stimulus presentation (Fig. 4b), or to compute the average firing rate of two neurons around a specific time of the experimental task (Fig. 4c). In our revision, we will include code snippets which will clearly show the required steps for each of these analyses. In addition, we will more clearly point the reader to the online tools (e.g. Jupyter notebooks), which offer an easier and clearer way to demonstrate the use of the toolbox.

Another remark concerns our claim that the package does not have dependencies. We agree that this claim was not well-worded. Our intention was to say that the package exclude dependencies such as scikit-learn, tensorflow or pytorch, which are often used in signal processing and which can be tedious to install. Pynapple still depends on a few packages including the most common ones: Numpy, Scipy, and Pandas. We will rephrase this statement in the manuscript and emphasize the importance of minimal dependencies for long-term backwards-compatibility in scientific computing.

We will complete the bibliography to make sure we properly reference all the packages designed for similar purpose. To note, some are not citable per se (i.e. no associated paper) but will be discussed.

It was suggested that the manuscript should better describe the integration of Pynapple into a full experimental data pipeline. This is an interesting point, which was briefly mentioned in the third paragraph of the discussion. Pynapple was not originally designed to pre-process data. However, it can load any type of data stream after the necessary pre-processing steps. Overall, this modularity is a key aspect of the Pynapple framework, and this is also the case for the integration with data pre-processing pipelines, for example spike sorting in electrophysiology and detection of region of interest in calcium imaging. We do not think there should be an integrated solution to the problem but, instead, to make it possible that any piece of code can be used for data irrespective of how the dataset was acquired. This is why we focused on making data loading straightforward and easy to adapt to any situation. This feature enables any user with any data modality and any long-established (often in-house) pre-processing scripts/software to utilize Pynapple in the analysis phase of their pipeline. Overall, not imposing a certain format compatibility from data acquisition phase is a strength for any analysis package.  

Finally, the reviews raised the issue of data and intermediate result storage. We agree that this is a critical issue. In the long term, we do not believe that the current implementation of NWB is the right answer for data involved in active analysis, as it is not possible to overwrite a NWB file. This would require the creation of a new NWB file each time an intermediate result is saved, which will be computationally intensive and time consuming, further increasing the odds of writing error. Theoretically, users who need to store intermediate results in a flexible way could use any methods they prefer, writing their own data files and wrappers to reload these data into Pynapple object. However, it is desirable for the Pynapple ecosystem to have a standardized format for storing data. We are currently improving this feature by developing save and loads methods for each Pynapple core object. We aim to provide an output format that is very simple to read in future Pynapple releases. This feature will be available in the coming weeks and will be described in the revised manuscript.

Author Response:

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

We would like to thank all Reviewers for their careful evaluation of our work. Below please find our responses and comments.

Reviewer #1 (Recommendations For The Authors):

1) The detection of cell-released GLP-1 is addressed in an indirect, averaged way in Fig. 2 - Supplement 1. This question seems like a good opportunity for an antagonist experiment (Exendin-9), which presumably would require much lower concentrations than those used to antagonize a saturating dose of GLP-1. It would also be much more convincing if GLPLight1 could be used to detect stimulated release of GLP-1 from the GLUTag cells.

We tried multiple times to acutely stimulate GLUTag cells using Forskolin and IBMX, but unfortunately we did not observe any robust fluorescence increase of GLPLight1. The only observation that was consistent was the higher baseline fluorescence of GLPLight1, and the reduced maximal response to saturating GLP-1 when GLPLight1 expressing HEK cells were cultured overnight with GLUTag cells. We considered this assay to be at best qualitative and — despite the aforementioned attempts — could not determine quantitative values.

2) The excitation-ratiometric response of the sensor, shown in Fig. 1D, is usually accompanied by strong pH-dependence of sensor function. It would be valuable to characterize this pH-dependence, using permeabilized cells in which the pH is changed; the ability of small (0.2-0.5 unit) pH changes to produce changes in fluorescence, as well as to affect the dynamic range of the sensor, should be characterized. This will prevent the misidentification of agents that affect cellular pH as having (for instance) an inhibitory effect on the binding of GLP-1 to GLPLight.

The pH sensitivity of cpGFP-based sensors is a valid concern. However, considering that the cpGFP module from GLPLight1 is intracellular (and thus largely protected from potential extracellular pH changes) we assume that GLPLight1 signal should be robust in most in-vivo or cell-based assays. In fact we have previously characterized this for a similarly-built neuropeptide sensor (PMID: 35145320) and believe that this will be the case also for GLPLight1.

3) The reported Kd for Exendin-9 is in the low nM range. Please explain the partial response at 1000x the concentration (including a discussion of the Kd of GLP-1 itself, as well as its off kinetics, and a comparison of this assay to the assays used previously).

The partial response is due to the presence of 1 uM GLP-1 in the imaging buffer, which is in constant competition with Exendin-9 for the binding to GLPLight1. Because GLP-1 has similar affinity as Exendin9 (see for example PMIDs: 34351033 and 21210113) and both are present at saturating concentration, we did expect to observe a partial response from GLPLight1. In this study, we did not exactly determine the on and off kinetics of both GLP-1 and Exendin9 on the GLPLight1 sensor due to technical challenges: to perform these experiments, we would need to set up a perfusion system where we could remove the unbound ligand and either wash off the bound ligand with buffer or compete it out with an antagonist. Unfortunately, we currently do not have access to such a set up.

4) Are the turn-on kinetics in Fig. 2C limited by drug application or by association? Are the on-rates much slower for the lower concentrations used for Fig. 2C? This is important for knowing how fast responses are likely to be at the lower concentrations likely to be achieved by endogenous release.

If we consider Fig 2B and 2C, we assumed the on-kinetics to be mostly driven by association since the ligand is expected to be homogeneously distributed.

The on-rate kinetics are indeed slower when lower concentrations of GLP-1 are used as shown in (Figure 2b) where we observe a TauOn of 4.7s with 10 uM GLP-1 and much slower kinetics when GLP-1 is applied a 1 uM for example (Figure 3d). As a result, we chose to incubate the ligand with GLPLight1 expressing cells for at least 30 minutes before the measurement of the dose-response to be close to equilibrium.

5) The parameters for the fitted dose-response curves in Fig 2C should be listed. The ~4x discrepancy between the dose-response in HEK-293 cells and neurons should be discussed. Are there known auxiliary subunits, dimerization, or lipid dependence that might account for this? It seems important to understand this if the sensors are to be used in an assay that may compare different systems.

We added the EC50 values to Fig 2C as requested. We did not consider a 4x discrepancy to be significant, because the measurement error in the EC50 region is relatively high and this difference seemed to be within the error range. In fact, the 95% confidence interval ranges are 7.8 to 11.1 nM in Neurons and 23.8 to 32.1 nM for HEK cells, if we consider the upper and lower boundaries of each, the difference drops to around 1-fold. We also performed a statistical test to compare the two fits (Extra sum of squares F-test) that confirmed the two fits were not significantly different (P value = 0.3736). Of course, the interaction partners and membrane composition are different in HEK cells and neurons and probably have an influence on the EC50 of GLPLight1, but their exact influence is unclear.

6) It seems surprising that removal of the endogenous N-terminal secretory sequence is actually helpful for membrane expression. Do the authors have any suggested explanation for this?

GLPLight1 contains an N-terminal hemagglutinin (HA) secretory motif. The hmGLP1R sequence that we chose also contained an endogenous secretory sequence that most likely interfered with the membrane transport mechanism and resulted in a lower sensor expression with both secretory sequences. We thus decided to keep the HA instead of endogenous to remain consistent with other sensors created in-house.

7) In Fig. 1, supplement 3, are the transient responses real? Do they occur with the control construct?

While we have not measured the G-protein recruitment on GLPLight-ctr, we have often observed this phenomenon for various receptors and ligands. The transient responses are thus most likely an artifact after manual addition of the ligand possibly due to:

-       Temperature difference

-       Exposure of the plate to ambient light before resuming measurement (phosphorescence)

-       Re-suspension of the cells affecting the proximity to the detector

-       Other unknown variables

If these responses were real, we would also expect them to be more sustained over time.

8) Please include a sentence or two explaining the luminescence complementation assay, and a reference.

We updated the results section of the manuscript with a section describing the luminescence complementation assay along with a reference:

“Next, we compared the coupling of GLPLight1 and its parent receptor (WT GLP1R) to downstream signaling. We first measured the agonist-induced membrane recruitment of cytosolic mini-G proteins and β-arrestin-2 using a split nanoluciferase complementation assay (Dixon et al., 2016). In this assay both the sensor/receptor and the mini-G proteins contains part of a functional luciferase (smBit on the sensor/receptor and LgBit for Mini-G proteins) that becomes active only when these two partners are in close proximity (Wan et al., 2018).”

Bravo to the authors for already making the sensor plasmids available at addgene.com. It would be helpful to include the plasmid IDs and/or a URL in the manuscript.

We would like to thank Reviewer #1 for noticing this. We have updated the data availability section of the manuscript and added the AddGene plasmid numbers of the constructs generated in this study.

Reviewer #2 (Recommendations For The Authors):

1) There are some parts of the introduction that need clarification. For example, GLP1 is quoted as an anorexigenic peptide, however, that is probably only true for centrally- derived GLP1. There is no evidence that enteroendocrine-derived GLP1 (the major pool) is anorexigenic- it is likely to be substantially degraded by DPPIV before reaching the brain. In any case, the discovery of GLP1 was always one of glucose-dependent insulin secretion, with the brain system being described decades later. Overall, the intro needs to be slightly reframed. While the tools presented here are more useful for assessment of central GLP1-releasing circuitry, they are ultimately based upon GLP1R signaling that is much better validated in the periphery.

We have slightly reframed the introduction accordingly.

2) "The human GLP1R (hmGLP1R) is a prime target for drug screening and drug development efforts, since GLP-1 receptor agonists (GLP1RAs) are among the most effective and widely-used weight-loss drugs available to date (Shah and Vella, 2014)." GLP1R was for two decades the breakthrough drug for treatment of type 2 diabetes mellitus and correction of glucose tolerance as assessed through HbA1c. It is only through reporting on millions of patients receiving GLP1RA that the weight loss effects were noted, leading to Phase1-3 trials and eventual approval for obesity indication. Again, some slight reframing of the introduction is required here.

Also for this point, we have slightly reframed the introduction accordingly.

3) GLP1 was applied at a maximal dose of 10 uM, which is 10-fold higher than maximal. Can the authors confirm absence of cytotoxic effects of exposing to peptide at such concentration? Ex4 (9-39) at such concentrations is usually cytotoxic at least in primary tissue.

We did not observe any obvious cytotoxic effect of GLP-1 at this concentration in HEK293T cells or Neurons.

4) "As expected, GLPLight1 responded to both GLP1RAs with almost maximal activation, on par with GLP1 (Figure 2a)." Such a claim is difficult to interpret without concentration-response curves, since the maximal concentration of liraglutide and semaglutide might not have been achieved in these experiments.

We agree with this statement is difficult to interpret without further clarification. We know from the literature that GLP-1, liraglutide and semaglutide all have very high affinity to the hmGLP1R (PMID: 31031702). We also proved that GLPLight signal saturates at concentrations above 1 uM of GLP-1 (figure 2C), we thus applied a 10x excess of all ligands and considered this signal as maximal.

5) "These results indicate that GLPLight1 can serve as a direct readout of pharmacological drug action on the hmGLP1R with higher temporal resolution than previously available approaches, such as downstream signaling assays (Zhang et al., 2020)." Many investigators use cAMP imaging to investigate GLP1R signaling, which is arguably of similar spatiotemporal resolution, also with the advantage of FRET quantification in some cases (e.g. EpacVV). Direct GLP1R signaling can also be inferred using cell lines heterologously-expressing GLP1R. Thus, the advantage of the current probes is that they can be used to readout direct GLP1R activation in native cells/tissues where promiscuous class B binding might limit signaling measures or where endogenous GLP1 release needs to be investigated.

We have edited the manuscript text accordingly.

6) "State-of-the-art techniques for detecting endogenous GLP-1 or glucagon release in vitro from cultured cells or tissues consist of costly and time-consuming antibody- based assays (Kuhre et al., 2016) or analytical chemistry procedures (Amao et al., 2015)." Agreed, but non-specificity/cross-reactivity of such assays is more prohibitive/problematic (e.g. against glicentin).

We have edited the introduction accordingly.

7) The studies using co-culture of GLUTag and GLP1Light1-HEK293 cells, whilst interesting, are not entirely convincing in their current form. Firstly, co-culture could influence GLP1Light expression levels (can the authors label FLAG?). Secondly, specificity of the response is not tested e.g. by adding Ex4 (9-39). Thirdly, titration with GLUTag conditioned media is not performed.

We partially addressed this issue in the answer to comment #1 from Reviewer #1. We previously performed a FLAG staining of GLPLight1 in the presence or absence of GLUTag cells and we did not notice any obvious difference. This goes in line with the fact that GLPLight1 is signaling inert, and the presence of GLP1 should not interfere with the surface expression of the sensor. We also checked that HEK293T cells did not express high levels of GLP1R according to the BioGPSCell line Gene Expression profile (https://maayanlab.cloud/Harmonizome/gene_set/HEK293/BioGPS+Cell+Line+Gene+Expression+Profiles).

We also tried to add GLUTag media after stimulation in bolus to GLPLight1 expressing cells and observed no response. This indicated that the “sniffer” cells must be present in close proximity to GLUTag cells for an extended period of time to observe any substantial difference in response, justifying our choice of experimental setup.

8) "Given that our photocage was placed at the very N-terminus of photo-GLP1, our results show that this caging approach prevents the peptide's ability to activate GLP1R but, at the same time, preserves its ability to interact with the ECD." An alternative hypothesis is that PhotoGLP1 does activate GLP1R, but this is undetectable with the sensitivity of GLP1Light. PhotoGLP1 cAMP concentration-response assays are needed (uncaged versus cage) to properly characterize and validate the compound (as would be standard for any newly-described GLP1R peptide ligand).

While we agree that there is a chance that Photo-GLP1 could activate GLP1R at high concentrations, we think that the characterization of Photo-GLP1 has to be determined by the end user directly with the technique of choice (GLPLight1 in our case) in order to get a reliable comparison of potency and efficacy. We modified the text accordingly to more accurately reflect the direct conclusions from our data, as follows:

“our results show that this caging approach prevents the peptide's ability to activate GLPLight1”.

9) "Surprisingly, GLPLight1 shows a fluorescent response in all three uncaged areas, while its fluorescence remained unaltered throughout the rest of the FOV, indicating high spatial localization of the response to GLP-1 (Figure 3f)." Why is this surprising?

We agree that this result is, indeed, not surprising and would like to thank Reviewer #2 for spotting this mistake, which has now been corrected in the manuscript.

10) The localized PhotoGLP1 experiments are interesting and show the utility of the ligand. There is however activation outside of the region of uncaging, which would argue against a pre-bound ECD mode of action. Possibly some PhotoGLP1 is pre- bound to the ECD, and some is freely diffusing? Alternatively, the scan area might be below the diffraction limit/accuracy of the microscope?

We would like to thank Reviewer #2 for this comment and agree with their observation. There could be some free Photo-GLP1 that gets photo-activated and binds regions around the uncaging area (similar to what has been observed for Photo-OXB:,PMID: 36481097). The activation around the uncaging area could also be due to lateral diffusion of the activated receptor on the membrane. There is also most likely some light diffraction at the uncaging area that could account for this phenomenon. To increase the spatial resolution, future studies could involve uncaging during sensor imaging via two-photon microscopy.

11) What was the rationale for caging native GLP1, which is then susceptible to DPPIV-mediated degradation? Would the N-terminal cage and first 2 amino acids also not be cleaved by DPPIV, thus rendering the tool of limited in vivo application? Conversely, PhotoGLP1 provides a template for similar light-activated (stabilized) GLP1R agonists such as Ex4 or liraglutide.

Thank you for making us aware of this (in vivo) limitation. We designed photoGLP1 as a tool for neurobiological experiments in the brain, where DPPIV expression would be low compared to peripheral organs (https://www.proteinatlas.org/ENSG00000197635-DPP4/tissue). We also envisage that the presence of the photocage would be enough to hinder the binding to DPP4 that cuts the first 2 AA. This hypothesis, however, was never tested experimentally, and we, therefore, acknowledge the limitation in the manuscript. We would furthermore like to thank the reviewers for his comment on additional photo-caged GLP1 agonists, which could be developed future studies.

12) It wasn't clear how GLP1Light could be used as a HTS screen for drug discovery? Surely, conventional systems (e.g. GLP1R + BAR/Ca2+/cAMP reporting) allow signal bias, an important component of GLP1RA action, to be assessed. Or could GLP1Light1 be used as a pre-screen to exclude any ligands that do not orthosterically bind GLP1R?

We would like to thank Reviewer #2 for this comment and would like to offer some clarification. We indeed thought that GLPLight1 could be used as a first line of screening to exclude ligands that do not bind in the orthosteric pocket. It is also a rather flexible method as the fluorescence increase of those sensors can be monitored using various techniques/devices that are available in most labs (e.g. microscopy, plate reader, flow cytometry).

13) Limitations of GLP1Light1 and PhotoGLP1 are not acknowledged in the discussion.

We would like to thank Reviewer #2 for pointing out the lack of description of the limitations of these tools, which have now been added to the Discussion.

14) Full characterization of PhotoGLP1 is missing, to include UV/Vis, Tr and HRMS.

PhotoGLP1 was fully characterized by UV/Vis and HRMS, and all experimental and analytical data was uploaded as supplementary data when the manuscript was initially submitted for publication in eLife.

Reviewer #3 (Recommendations For The Authors):

1) The ~1000 fold lower EC50 for GLP1 of GLPLight1 compared with native GLP1R needs to be openly acknowledged as a major limitation of the sensor, as this will substantially reduce the types of experiment for which it will be useful. Because it needs 1000 times higher GLP1 levels than wild type GLP1R to be activated, it is unlikely, for example, to be useful for monitoring the dynamics of activation of native GLP1R in vivo. The claim that the sensor could be used for in vivo imaging for fibre photometry is therefore an exaggeration.

We would like to first thank Reviewer #3 for this comment and to further provide some clarification. We recognized that the data presented in this manuscript might have been confusing when comparing the affinity of GLP1R (using cAMP) and GLPLight1 (using the fluorescence increase because there is no coupling to cAMP). We believe that the low EC50 measured in the cAMP assay cannot accurately be compared to GLPLight1 response because it is an enzymatically amplified process. In order to support this claim, we included another set of experiments where we titrated agonist- induced recruitment of miniGs protein to the GLP1R receptor and found an EC50 of 3.8 nM for native GLP-1 using this assay (added as panel l in Figure1 Supplement 3). We thus confirmed that the nature of the assay itself has a drastic influence on the EC50 measured and it is not unusual to observe 100x fold difference of EC50 for the same receptor-ligand pair.

We believe that the miniGs protein recruitment is a better comparison to GLPLight1 because it is not enzymatically amplified. This assay reveals that GLPLight1 has around 8-fold lower affinity to GLP1 compared to its parent receptor, which is in line with the EC50 loss observed previously for other GPCR-based sensors of this class. We are thus confident that GLPLight1 has to potential to be used in vivo under specific circumstances, specifically in brain tissue. We elaborated on this point in the Discussion part of the manuscript.

2) Fig2 suppl 1 is described as demonstrating a reduced response of GLPLight1 to GLP-1 when HEK cells with were cultured with GLUTag cells. However, it is speculation to conclude that this is because GLP1Light1 was partially pre-activated by endogenous GLP-1, without demonstrating the response of GLPLight1 before and after GLUTag cell stimulation. Unless additional data are generated, the presented data do not convincingly demonstrate that GLP1Light1 can detect GLP1 released from GLUTag cells.

We would like to thank Reviewer #3 for this comment which has been addressed already in the replies to Comment#1 from Reviewer #1 and Reviewer #2.

3) The authors should openly acknowledge that photo-uncaging the GLP1 probe might not be very helpful for monitoring the temporal dynamics of the GLP1-GLP1R interaction, because unless all the photocaged glp1 is released by the light stimulus, the activation of photo-released GLP1 will be slowed by the remaining caged GLP1, and the dynamics will be slower than for native GLP1. This makes it unsuitable for many temporal questions, although it might be useful to deliver GLP1 in a spatial restricted manner.

We do agree that the biggest advantage of Photo-GLP1 is its ability to be activated in a very localized manner. We also agree that the presence of caged Photo-GLP1 will influence the binding of the uncaged GLP-1. Nevertheless, there is still an advantage of using Photo-GLP1 in some assays such as pharmacological activation on brain slices. In fact, we have shown for our Photo-OXB molecule that the perfusion of OXB was much slower at eliciting neuronal depolarization compared to uncaging of Photo- OXB (see PMID: 36481097). We think that this was mainly due to the slow diffusion kinetics of the peptide into the brain tissue. We also think that uncaging can provide a more controlled activation with varying laser power and uncaging duration.

4) To claim (as currently in the discussion) that GLPLight1 has potential to be used for investigating the dynamics of endogenous GLP1, the authors would need to compare the dynamics of the GLP1Light sensor with wild type GLP1R. We do not know that its activation dynamics will reproduce native glp1r.

We would like to thank Reviewer #3 for this comment and would like to offer some clarification. Since GLPLight1 does not couple to intracellular signaling, it was impossible to compare its activation kinetics to GLP1R WT using the same assay. However, we can offer a relative comparison since we know that GLPLight1 takes around 50 seconds to be activated using 1 µM GLP-1 (figure 2B) and that it takes a similar time for GLP1R to be activated in the miniG protein recruitment assay (Fig 1 Supplement 3) using 100 nM GLP-1. Considering that GLPLight1 has a lower affinity than the GLP1R (8-10x lower), we think that the activation kinetics of both the sensor and GLP1R are comparable.

Additional comments:

1) In fig 2A,B, it is not clear whether the trace shows a partial reversal of GLP1- triggered activation by Ex9, or Ex9-independent receptor desensitization. A control trace is required to show the kinetics of GLP1-triggered activation without the addition of Ex9.

We would like to thank Reviewer #3 for this comment. We can exclude the possibility of Ex9-independent desensitization because GLPLight1 has been shown to be signaling inert to all G-proteins, Beta arrestin-2 and cAMP. Moreover, we have observed that the fluorescence signal was stable for more than 30 minutes for the GLP-1 titrations, even at high concentrations of ligand.

2) It would be helpful if the pEC50 for WT GLP1 were also shown in table 1, for comparison with the GLP1 mutants.

We would like to thank Reviewer #3 for this comment, and we have now added the respective pEC50 for WT GLP1 to Table 1.

3) Fig2 suppl 1. The methods and analysis for this figure are inadequately explained. To show that the HEK-GLPLight1 cells are responding to GLP1 released from GLUTag cells, the GLPLight1 response needs to be shown before and after GLUTag cell stimulation with an agent that should trigger GLP-1 release.

We would like to thank Reviewer #3 for this comment which has been partially addressed already in the replies to Comment#1 from Reviewer #1 and Reviewer #2.

Since we did not observe any response to acute stimulation of GLUTag cells we considered the high glucose concentration present in the culture media being a stimulation agent for GLUTag cells, which has been previously reported (PMID: 17643200).

4) Fig 3g and others: The end of the photo activation period needs to be represented correctly on the timeline. In 3g, the bar that should indicate when photoactivation was applied does not end at the zero time point (which is labelled as the time relative to photoactivation).

We would like to thank Reviewer #3 for pointing this out. The shaded area representing the photo-activation has been matched accordingly.

5) Discussion para 1: the authors claim their data show that ligand induced activation of human GLP1R occurs more slowly than others similar GPCR sensors - they should give actual data to substantiate this claim, since the time course of glp1r activation has not been analysed and compared with other sensors in the manuscript.

We added data to support this claim to the discussion: “As a reference, other previously-characterized class-A GPCR-based neuropeptide biosensors showed sub- second activation kinetics (Duffet et al., 2022a; Ino et al., 2022).”

6) Methods: what wavelength was used for recording emission from GLP1Light1? The excitation wavelength is given, but I can't see the emission wavelength(s). In fig 1d, the excitation and emission spectra should be depicted in different colours/line properties, otherwise this figure is very confusing.

We updated figure1d and changed the colors to improve data visualization. Regarding the missing wavelength, we would like to clarify that both wavelengths were already described in the methods section as: “The excitation and emission spectra were measured at λem =560nm and λex\= 470nm, respectively, on a TECAN M200 Pro plate reader at 37 °C. “. We would be happy to rewrite this paragraph, if necessary, shall it remain unclear to the reader.

Author Response:

Reviewer #1 (Public Review):

This manuscript features a key technical advance in single-molecular force spectroscopy. The critical advance is to employ a click chemistry (DBCO-cycloaddition) for making a stable covalent connection between a target biomacromolecule and solid support in place of conventional antigen-antibody binding. This tweak dramatically improves the mechanical stability of the pulling system such that the pulling/relaxation can be repeated up to a thousand times (the previous limit was a few hundred cycles at best). This improvement is broadly applicable to various molecular interactions and other types of single-molecule force spectroscopy allowing for more statistically reliable force measurements. Another strength of this method is that all conjugation steps are chemically orthogonal (except for Spy-catcher conjugation to the termini of a target molecule) such that the probability of side reactions could be reduced.

The reliability of kinetic and thermodynamic parameters obtained from single-molecule force spectroscopy depends on statistics, that is, the number of pulling measurements and their distribution. By extending the number of measurements, this robust method enables fundamental/critical statistical assessment of those parameters. That is, it is an important and interesting lesson from this study that ~200 repeats can yield statistically reasonable parameters.

The authors carried out carefully designed optimization steps and inform readers of the critical aspects of each. The merit, quality, and rigor as a method-oriented manuscript are impressive. Overall, this is an excellent study.

We appreciate for the positive evaluation for our work. Additionally, the minor suggestions were helpful to improve our manuscript. Thank you!

Reviewer #2 (Public Review):

In this study, the authors have developed methods that allow for repeatedly unfolding and refolding a membrane protein using a magnetic tweezers setup. The goal is to extend the lifespan of the single-molecule construct and gather more data from the same tether under force. This is achieved through the use of a metal-free DBCO-azide click reaction that covalently attaches a DNA handle to a superparamagnetic bead, a traptavdin-dual biotin linkage that provides a strong connection between another DNA handle and the coverslip surface, and SpyTag-SpyCatcher association for covalent connection of the membrane protein to the two DNA handles.

The method may offer a long lifetime for single-molecule linkage; however, it does not represent a significant technological advancement. These reactions are commonly used in the field of single-molecule manipulation studies. The use of multiple tags including biotin and digoxygenin to enhance the connection's mechanical stability has already been explored in previous DNA mechanics studies by multiple research labs. Additionally, conducting single-molecule manipulation experiments on a single DNA or protein tether for an extended period of time (hours or even days) has been documented by several research groups.

One of the unique features of our work is the development of a robust single-molecule tweezer method that is applicable to membrane proteins, rather than simply making another stable system. As re-written in Introduction, it is not straightforward as we have to consider the membrane reconstitution. We believe that our work is expected to overcome the bottleneck in membrane protein studies that arises when using single-molecule tweezer methods.

To improve the delivery of the contextual information, we revised Introduction, Results, and Discussion. The first four paragraphs in the Introduction briefly review previous tweezer methods with an improved stability and delineate where our work is placed. In the first paragraph of the Results, we also briefly discussed how and why our DBCO tethering strategy differs from previous DBCO methods. In the first paragraph of the Discussion, we compared the previous methods regarding the stability improvement.

Additionally, the revised manuscript now includes new findings – the full dissection of structural transitions of a helical membrane protein, the observation of hidden helix-coil transitions at a constant force, and the estimation of kinetic pre-exponential factors. We believe that the new findings provide important insights into membrane protein folding, in addition to the usefulness of our method itself for membrane protein studies. We extensively edited the main text and Methods accordingly. Relevant figures are Figures 6 and 7, Figure 6–figure supplements 1–3, and Figure 7–source data 1.

Reviewer #3 (Public Review):

The authors describe a method to tether proteins via DNA linkers in magnetic tweezers and apply it to a model membrane protein. The main novelty appears to be the use of DBCO click chemistry to covalently couple to the magnetic bead, which creates stable tethers for which the authors report up to >1000 force-extension cycles. Novel and stable attachment strategies are indeed important for force spectroscopy measurements, in particular for membrane proteins that are harder and therefore less studied in this regard than soluble proteins, and recording >1000 stretch and release cycles is an impressive achievement. Unfortunately, I feel that the current work falls short in some regards to exploring the full potential of the method, or at least does not provide sufficient information to fully assess the performance of the new method. Specific questions and points of attention are included below.

We appreciate for the positive evaluation. We were able to largely improve our manuscript while preparing our responses to the comments. Thank you!

- The main improvement appears to be the more stable and robust tethering approach, compared to previous methods. However, the stability is hard to evaluate from the data provided. The much more common way to test stability in the tweezers is to report lifetimes at constant force(s). Also, there are actually previous methods that report on covalent attachment, even working using DBCO. These papers should be compared.

As shown in Figure 4E, we evaluated the robustness of our method in a way suggested by you – the lifetime measurement at a constant force. Specifically, ~12 hours at 50 pN. Definitely, our tweezer approach established here is the most robust method for membrane protein studies. Please refer to the section “Assessing robustness of our single-molecule tweezers” in page 7 and line 31.

We discussed the previous covalent methods for which quantitative data are presented in light of the system stability. Please refer to the first paragraph of Discussion. We also briefly discussed how and why our DBCO tethering strategy differs from previous DBCO methods, in the first paragraph of Results.

- The authors use the attachment to the surface via two biotin-traptavidin linkages. How does the stability of this (double) bond compare to using a single biotin? Engineered streptavidin versions have been studied previously in the magnetic tweezers, again reporting lifetimes under constant force, which appears to be a relevant point of comparison.

The papers in this comment showed that the tethering lifetimes of biotin-streptavidin variants were affected by the asymmetric bead anchoring point. However, the situation does not apply to our work as we do not anchor traptavidin to beads. Besides, the stability comparison between the single- and double-biotin systems is not the main point of our work, so we do not have the answer to the question. However, we cited the reference in the first paragraph of Discussion where we discuss the system stability.

- Very long measurements of protein unfolding and refolding have been reported previously. Here, too, a comparison would be relevant.

We briefly discussed the relevant previous works in the first paragraph of Discussion.

In light of this previous work, the statement in the abstract "However, the weak molecular tethers used in the tweezers limit a long time, repetitive mechanical manipulation because of their force-induced bond breakage" seems a little dubious. I do not doubt that there is a need for new and better attachment chemistries, but I think it is important to be clear about what has been done already.

The sentence is in Abstract, so we also had to consider the conciseness. By simply adding the phrase “used for the membrane protein studies”, we can place our work into a more proper context.

In page 2 and line 3, “…However, the weak molecular tethers used for the membrane protein studies have limited long-time, repetitive molecular transitions due to force-induced bond breakage…”

- Page 5, line 99: If the PEG layer prevents any sticking of beads, how do the authors attach reference beads, which are typically used in magnetic tweezers to subtract drift?

The PEG layer consists of biotin-PEG and methyl-PEG at a 1:27.5 molar ratio. As the reference beads are coated with streptavidin, they are attached to the PEG layer by the regular biotin-streptavidin interaction. In page 19 and line 7, you can refer to “…The polystyrene beads are attached to the PEG surface via biotin-streptavidin interaction. The beads are used as reference beads for the correction of microscope stage drifts…”

- Figure 3 left me somewhat puzzled. It appears to suggest that the "no detergent/lipid" condition actually works best, since it provides functional "single-molecule conjugation" for two different DBCO concentrations and two different DNA handles, unlike any other condition. But how can you have a membrane protein without any detergent or lipid? This seems hard to believe.

We explained the raised point in page 6 and line 18,

“…Indeed, the best condition was in the absence of any detergents or lipids (Figure 3; no detergents/lipids only during the conjugation step). This situation is possible because membrane proteins are sparsely tethered to the chamber surface, which kept them from aggregating. However, not using detergents or lipids means that the membrane proteins are definitely deformed from their native folds. Therefore, we sought an optimal solubilization condition for membrane proteins during the DBCO-azide conjugation step...”

Figure 3 also seems to imply that the bicelle conditions never work. The schematic in Figure 1 is then fairly misleading since it implies that bicelles also work.

The buffer conditions shown in Figure 3 are those ONLY during the DBCO-azide conjugation step. In this step, the bicelle conditions did not work. Therefore, after the conjugation in 0.5% DDM, the buffer was exchanged with a bicelle solution. This process is shown in Figure 2 and the finally assembled system is depicted in Figure 1.

To clarify this point, we put a note “Buffer conditions only during the DBCO-azide conjugation step” just above the buffer conditions in Figure 3. You can also find for the relevant exchange step in page 6 and line 31, “…Following a 1 h incubation of the beads in the single-molecule chamber at 25°C, unconjugated beads were washed, and the detergent micelles were exchanged with bicelles to reconstitute the lipid bilayer environment for membrane proteins…”

- When it comes to investigating the unfolding and refolding of scTMHC2, it would be nice to see some traces also at a constant force. As the authors state themselves: magnetic tweezers have the advantage that they "enable constant low-force measurements" (page 8, line 189). Why not use this advantage?<br /> In particular, I would be curious to see constant force traces in the "helix coil transition zone". Can steps in the unfolding landscape be identified? Are there intermediates?

Yes, please refer to Figure 6. We were able to dissect three distinct transitions from the fully unstructured state to the native state, including the helix-coil transitions. We also reconstructed the folding energy landscape using a deconvolution method.

Please refer to the pertinent sections in the main text, which are titled “Structural transitions and folding energy landscape over extended time scales” and “Mechanistic dissection of folding transitions”.

- Speaking of loading rates and forces: How were the forces calibrated? This seems to not be discussed.

We wrote an additional section in Methods titled “Instrumentation of single-molecule magnetic tweezers”, where we discuss the force calibration. For the actual force calibration data, please see Figure 4–figure supplement 1A.

In page 20 and line 10, “…The mechanical force applied to a bead-tethered molecule was calibrated as a function of the magnet position using the formula F = k_B_T∙L/δx_2 derived from the inverted pendulum model96, where _F is the applied force, k_B is the Boltzmann constant, _T is the absolute temperature, L is the extension, and _δx_2 is the magnitude of lateral fluctuations…”

And how were constant loading rates achieved? In Figure 4 it is stated that experiments are performed at "different pulling speeds". How is this possible? In AFM (and OT) one controls position and measures force. In MT, however, you set the force and the bead position is not directly controlled, so how is a given pulling speed ensured?<br /> It appears to me that the numbers indicated in Figures 4A and B are actually the speeds at which the magnets are moved. This is not "pulling speed" as it is usually defined in the AFM and OT literature. Even more confusing, moving the magnets at a constant speed, would NOT correspond to a constant loading rate (which seems to be suggested in Figure 4A), given that the relationship between magnet positions and force is non-linear (in fact, it is approximately exponential in the configuration shown schematically in Figure 1).

You are correct, so we simply modified the “pulling speed” to “magnet speed” in the figure caption. The loading rates provided in the figure (with the notation <>) were average loading rates in 1–50 pN to provide rough estimates. We actually specified it in the caption as “average force-loading rate”. However, this can be misleading at a glance, so we just deleted all the loading-rate values in the figure and caption.

- Finally, when it comes to the analysis of errors, I am again puzzled. For the M270 beads used in this work, the bead-to-bead variation in force is about 10%. However, it will be constant for a given bead throughout the experiment. I would expect the apparent unfolding force to exhibit fluctuations from cycle to cycle for a given bead (due to its intrinsically stochastic nature), but also some systematic trends in a bead-to-bead comparison since the actual force will be different (by 10% standard deviation) for different beads. Unfortunately, the authors average this effect away, by averaging over beads for each cycle (Figure 4). To me, it makes much more sense to average over the 1000 cycles for each bead and then compare. Not surprisingly, they find a larger error "with bead size error" than without it (Figure 5A). However, this information could likely be used (and the error corrected), if they would only first analyze the beads separately.

We might be wrong, but there seems to be a misunderstanding. First, we added Figure 5–figure supplement 1 where you can see individual traces. As expected, the levels of unfolding forces/sizes appear consistent during the progress of pulling cycles. Second, the advantage of averaging for different beads is that you can effectively remove the bead size effect. This “averaging-out” is the key strategy in our kinetic analysis. Based on the error estimation, if you average the values of kinetic parameters obtained from different beads, you can then estimate them with reasonably small errors despite the bead size variations. This becomes more evident after initial hundreds of pulling cycles. The errors for 200 and 1000 cycles are of only ~1% difference, indicating that you do not need to blindly run the pulling cycles. These results are based on the “averaging-out” strategy, which is the merit of our analysis. For more details, please see the section in the main text titled “Assessing statistical reliability of pulling-cycle experiments”, where relevant figures, figure supplements, and Method sections are referred.

What is the physical explanation of the first fast and then slow decay of the error (Figure 5B)? I would have expected the error for a given bead after N pulling cycles to decrease as 1/sqrt(N) since each cycle gives an independent measurement. Has this been tested?

If the sampling was from one population (here, unfolding probability profile), the error would follow a 1/√n decay as expected for the standard error. In our analysis, however, we estimated the expected “mean” errors, regardless of detailed shapes of the unfolding probability profiles. To this end, we sampled the data from different possible profiles (shown in Figure 5–figure supplement 5). We then averaged all the error plots to obtain the plot of the mean errors during progress of pulling cycles (black curve in Figure 5D). In this case, the plot does not have to follow the standard error curve represented by the factor 1/√n.

We tested this by fitting with the model function of y = A/√n, for various lower limit of N = 10, 30, 50, 100, 300, and 500 in the regression analysis (Figure 5–figure supplement 6). The results of the reduced chi-square (χ2) used for a goodness-of-fit test (χ2 = 1 for the best fit) indicates that the two-term exponential model (χ2 = 1.60) shows a better fit than the reciprocal square root model (χ2 = 2.30–6.01). The regression model adopted in our analysis is a phenomenological model that more properly describes the error decay curve. The trend of the first fast and then slow decay is not unusual because it is also expected for the reciprocal square root model – the plot 1/√n decays fast and then slowly, too (Figure 5–figure supplement 6).

Author Response:

eLife assessment

The authors present an exciting idea about how to integrate morphogens into a gene regulatory network with the dynamics of morphogenesis and cell movement. It represents a novel methodology, but in its current form the hypotheses, data and relationships described do not provide a sufficiently compelling model to disentangle cause and effect or elucidate the impact of cell movements on differentiation dynamics the zebrafish mesoderm.

Our aim in this work was not to disentangle causal relationships between signalling, cell movements and gene-regulatory interactions. As discussed in the specific responses below, and in the discussion of the pre-print, this would require precise experimental manipulations within the context of a modelling framework that enables multi-scalar integration of each of these three dynamic components. What we do present here is a) computational methodology to reverse-engineer GRNs in the context of tissue morphogenesis (Spiess et al.,) and b) experiments to narrow down a candidate GRN capable of recapitulating gene expression dynamics in vitro and in vivo (Fulton et al.,). We see this as the first step in tackling the causal relationships of cell movements, signalling and cell fate decision making and propose a working model for future studies to build on.

Reviewer #1 (Public Review):

In the manuscript " Cell Rearrangement Generates Pattern Emergence as a Function of Temporal Morphogen Exposure" by Fulton et al., the authors set out to link cell dynamics and single-cell gene expression states, in order to understand the dynamics of cell differentiation. This important challenge is tackled by studying somitogenesis in the zebrafish embryo and combining reverse-engineering gene regulatory networks (GRNs) with cell tracking data. The differentiation of the presomitic cells is evaluated by the differential tbx marker expression through in situ HCR and antibody staining, and live imaging of reporters. Through mathematical modelling taking into consideration the HCR tbx data, live reporter data of the morphogen activity, and the 3D tracking data at different stages, the authors find a candidate model of a gene regulatory network that recapitulates both in vivo and in vitro patterns of the dynamics of cell differentiation. Using this live-modelling approach, the authors move on to question the impact of cell movement on gene expression and conclude that pattern emerges as a function of cell rearrangements tuning the temporal exposure of the cells to the morphogen gradients.

The major strength of the manuscript is the development of a unique method for addressing cell differentiation dynamics by combining static gene expression data with live cell dynamics. Bridging spatiotemporal information is key to understanding tissue and embryo development and this work provides a great basis for it. A potential weakness is how one selects which of the GRNs predicted from the live-modelling is physiologically relevant to the system of interest, since it requires fitting techniques.

The major goal of the paper is mostly achieved. This is evident by the proposed model predicting well the dynamics of differentiation both in vivo and in vitro. To fully support the conclusion that cell rearrangements are necessary for patterning, the addition of functional experiments targeted in this direction might be beneficial.

We agree with the reviewer that functional evidence for a role of cell rearrangement in pattern formation is lacking from the pre-print. We will adjust our title and conclusions to reflect this in a revised version.

Reviewer #2 (Public Review):

Fulton et al. seek to understand the interplay between "morphogen exposure, intrinsic timers of differentiation, and cell rearrangement" that together regulate the differentiation process within the presomitic mesoderm tissue (PSM) in developing Zebrafish embryos. A combination of live-cell microscopy to measure cell movements, static measurements of gene expression, and computational and mathematical methods was used to develop a model that captures the observed differentiation profile in the PSM as a function of cell rearrangements and morphogen signaling.

The authors motivate their investigation into the link between cell rearrangements and differentiation by first comparing differentiation timing in vitro and in vivo. The authors report that a subset of cells differentiating in vitro do so synchronously while cells differentiating in vivo do so with a wide range of differentiation trajectories. By following a small group of photo-labeled cells, it is suggested that the variation of differentiation timing in vivo is related to variation in cell movements in the tissue. To explain these observations in terms of gene expression within single cells, a novel method to combine cell tracks with fixed measurements of gene expression is first used to estimate gene expression dynamics (AGET) in live cells within a tissue. A final ODE-based gene regulatory network (GRN) model is selected based on a combination of data fitting to AGETs and tissue level measurements, further in vitro experiments, and literature criteria. Importantly this model incorporates information from diverse experimental sources to generate a single unified model that can be potentially used in other contexts such as predicting how differentiation is perturbed by genetic mutations affecting cell rearrangement. The authors then use this GRN model to explain how cells starting from the same position in the PSM can have different fates due to differential movement along the A-P axis. Lastly, the model predicts and, the authors experimentally validate, that the expression of differentiation markers can be heterogeneously expressed between neighboring PSM cells.

The presented research addresses the important topic of patterning regulation accounting for individual cell motion. contributes to larger tissue patterns, this work may directly contribute to our understanding of how regulation across biological scales. Additionally, the methodology to estimate AGET is especially intriguing because of its potential applicability to a wide variety of developmental processes.

However several issues weigh down the strengths of this paper. First, some conclusions and interpretations in the paper do not obviously follow the data and require further clarification. Second, the authors should consider alternative explanations and models and include some discussion about instances where the final GRN model may not fit as well. Finally, the current manuscript lacks clarity in its presentation and this makes it difficult to follow and understand.

Major concerns:

1. A key conclusion made in this paper is that differentiation times show a high variability even when neighboring PSM cells are compared. This is based on the photoconversion experiment shown in Figure 2A-C, where a group of cells is labeled and over time, a trail of labeled cells is visible. It is crucial to understand which compartment is labeled, i.e. progenitor vs. maturation zone vs. PSM. If cells in the progenitor/marginal zone are labeled, the underlying reason for the trailing effect is not a difference in differentiation time, but rather, a difference in the timing of when cells exit the progenitor zone. This needs to be distinguished in my view. In other words, while the timing of progenitor zone exit varies (needs to), once cells are within the PSM, do they still show a difference in differentiation timing? From previous experimental evidence I would expect that in fact, PSM cells differ only very little in differentiation timing. My statement is based on previously published labeling experiments done in posterior PSM cells, not tail bud cells (in chick embryos), which showed that labeled neighboring PSM cells were incorporated into the same adjacent somites, without evidence of a 'trail' (see figure 4H in Dubrulle et al. 2001). In the case of single cell labeling, it was found that these are actually incorporated into the same somite (or adjacent one), even if labeled in the posterior PSM (Stern et al. 1988). The situation in zebrafish appears similar (see Griffin & Kimelman 2002 and Müller et al. 1996). Additionally, the scheme in Figure 2K suggests that the trailing effect reflects a sequential exit from the progenitor zone that is controlled and timed.

We place the labels in a region of the taibud containing tbxta and tbx16 positive mesodermal progenitors and not in the PSM. Therefore, we are examining the timing of exit, and show this is correlated with the onset of tbx6 expression. Taken together with previous work (Thomson et al., 2021; 10.1016/j.cdev.2021.203748), it demonstrates that in zebrafish embryos, non-directional cell movements generate a progressive exit of cells from the progenitor region in the tailbud towards the PSM. We will make these points clear in a revised version of the manuscript.

2. The data on cell movement needs to be presented more clearly. Currently, this data is mainly presented in Figure 3D, which does not provide a good description of the cell movements. Visualization of the single cell tracks and the different patterns that are in the tissue along with the characterization of the movement/timescales is needed to better communicate the data and to tie it to the main conclusions.

A thorough analysis of the tracking data and cell movements in the tailbud are presented in a previous paper (Thomson et al., 2021; 10.1016/j.cdev.2021.203748), and is cited in the pre-print.

3. The conclusion "As a result of their different patterns of movement, and therefore different Wnt and FGF dynamics, the simulated T-box gene expression dynamics differ in both cells." (Line 249) is not convincing: what part of the data shows that it is not the other way around, i.e. the signaling activities control the movement? The way I understand the rationale of this analysis: the authors take the cell movement tracks as a given input into the problem, and then ask, what signaling environment is the cell exposed to? The challenge with this view is two-fold: first, the authors seem to assume that a cell moves into a new environment and is hence exposed to a different level of signal, while in reality, these signaling gradients act short-range and maybe even at a cellular scale and hence a moving cell would carry Wnt-ligands with it, essentially contributing to the signaling environment. This aspect of 'niche construction' seems to be missing. Second, it has been shown (in chick embryos) that cell movement is, in turn, controlled by signaling levels, how would this factor into this model?

See response to reviewer 1, we have revised our conclusions to make it clear that we are not demonstrating a causal role of cell movements in this process. We instead provide a modelling framework to interrogate these complex multi-scale interactions.

4. On the comparison with the in vitro model:<br /> A. The interpretation of cells differentiating synchronously or coherently in vitro seems inconsistent with the data presented in figure 1. To me figure 1F/G does not seem compatible with the previous figure 1D/E since 1F seems to describe cells that upregulate tbx6 over a range of times, in a manner analogous to what is reported in vivo, i.e. figure 2.

We agree that once initiated, tbx6 expression is variable between individual cells as shown in Figure 1. Our conclusion is that, whatever the rate of increase in expression, cells initiate their increase at the same time (200 mins). We will make this clear in a revised version.

B. The authors conclude that in vitro, single PSM cells differentiate 'synchronously' and hence differently to what is seen in vivo, where the authors conclude that there is a "range of time scales". As noted above, the situation in vivo can be explained by a timed exit from the progenitor zone, while PSM differentiation is proceeding similarly in all PSM cells. In this view, what is seen in vitro is that all those cells that undergo PSM differentiation, initiate this process in culture more synchronously but it is the exit from the progenitor state, not the dynamics of differentiation, that might be regulated differently in vivo vs. in vitro.

We agree with this statement- the process we are examining is the timing of tbx6 onset, a proxy for the timing of switching from a progenitor to a PSM cell state. However, we don’t see how this is different from the ‘dynamics of differentiation’ as these processes are directly related.

C. Another important point to clarify is that the overall timing of differentiation is entirely different in the in vitro experiment: as has been shown previously (Rohde et al. 2021, Figure S12) both the period of the clock and the overall time it takes to differentiate is very substantially increased, in fact, more than doubled. This aspect needs to be taken into account and hence the conclusion: "Our analysis revealed that cells undergo a range of temporal trajectories in gene expression, with the fastest cells transiting through to a newly formed somite in 3 hours; half the time taken for cells to fully upregulate tbx6 in vitro (Figure 2K-L).)" (line 142) appears misleading, as it seems to emphasize how fast some cells in vivo differentiate. However, given the overall slowing down seen in vitro, which more than doubles the time it takes for differentiation (see Rohde et al. 2021, Figure S12), this statement needs to be refined.

This is indeed an interesting observation and will be discussed in a revised version.

5. The GRN proposed in this work includes inhibition of ntl/brachyury by Fgf (Figure 3f). However, it has been shown that Fgf signaling activates, not inhibits, ntl (see for instance dnFgfr1 experiments in Griffin et al., 1995). This does not seem compatible with the presented GRN, can the authors clarify?

Experiments in which signalling and/or transcription function are disrupted in vivo are very different interpret from analysing the impact of gene expression alone. As discussed, and highlighted by the reviewers, there exists a complex interplay where signals can impact cell movements and vice versa. What we propose in this work is a working model of this process through which this interplay can be explored.

6. The authors use static mRNA in situ hybridization and antibody stainings to characterize Wnt and Fgf signaling activities. First, it should be clarified in Figure 3A that this is not based on any dynamic measurement (it now states Tcf::GFP, as if GFP is the readout, so the label should be GFP mRNA). Second, and more importantly, it is not clear how this quantification has been done. Figure 3C shows a single line, while the legend says n=6 and "all data plotted"..can this be clarified? Without seeing the data it is not possible to judge if the profiles shown (the mean) are convincing. As this experimental result is used to inform the model and the remainder of the paper, it is of critical importance to provide convincing evidence, in this case, based on static snapshots.

This will be clarified in a revised version of the paper.

7. Although the AGET analysis and this specific GRN model development are of interest and warrant the explanation the authors have provided, I would be careful not to overstate the findings. In particular, I believe the word "predicted" is used too loosely throughout the manuscript to describe the agreement between model and experiments. For example, my understanding of Figure 4, and what is described in the supplemental diagram, is that the in vitro experiments are used to further refine the model selection process. Therefore, it should not be stated as a prediction of the selected model. This is not to say the final model is not predictive, but it's difficult to assess the predictive power of this model since it hasn't been tested in independent experimental conditions (e.g. by perturbing cell movement and using the model to predict the expected differentiation boundary).

We will take care with the use of the term ‘predicted’ in a revised version of the paper. The reviewer is correct that this result was used to select from an existing set of GRNs.

Reviewer #3 (Public Review):

Fulton et al. look to apply approaches for tackling the readout of gene regulatory networks (GRNs) to a system where cell position itself is continually changing. The objective is highly laudable. GRN analysis has proven to be a powerful approach for understanding how cell fates are determined by morphogenetic inputs, but it has thus far been applied in a limited number of systems. Here, the authors look to substantially extend the application of GRNs to more dynamic systems. The theoretical and experimental approaches are integrated to achieve the analysis of the GRN. In principle, this has wide potential impact and applicability to other systems.

Unfortunately, in its current form, the manuscript does not do justice to the central aims of the authors. The manuscript is unclear in nearly all sections, and figures and analysis can be substantially improved. The quantifications are not shown in a fitting manner. The modelling itself stands as the strongest part of the manuscript, but improvements are needed. Currently, the main claims of the authors cannot be evaluated based on the quality of the presented data.

This reviewer has provided a list of minor corrections that will greatly improve a revised version of the manuscript for our next submission.

Author Response

Reviewer #1 (Public Review):

In this manuscript, Soto-Feliciano et al. investigate the tumor suppressive role of MLL3 in hepatocellular carcinoma (HCC). The authors used a variety of techniques including hydrodynamic tail vain injection (HTVI), CRISPR deletion, and shRNA to disrupt MLL3 expression in mouse models. They clearly show that MLL3 acts as a tumor suppressor in the context of MYC-induced HCC. They show that MLL3 acts by activating the Cdkn2a locus. Genomic analysis showed that MLL3 binds to enhancers and promoters, and specifically interacts with the Cdkn2a promoter. When MLL3 was downregulated, Cdkn2a levels fell and this corresponded to changes in relevant histone marks targeted by MLL3. The authors were also able to show that reintroduced MLL3 expression in a dox inducible system could rescue CDKN2A locus expression, which in turn reduced colony formation and induced apoptosis. Human genomic correlation showed that MLL3 and Cdkn2a mutations are generally mutually exclusive. Overall, the conclusions of the manuscript are well supported by a logical series of experiments with good controls and orthogonal approaches. While it would be useful to examine another HCC model such a CTNNB1-driven model, the current paper is convincing in its conclusions.

We thank the reviewer for their positive and constructive comments and suggestions. Our study primarily used MYC as the driving oncogene for two reasons: first, in an initial in vivo screen of 12 candidate tumor suppressors, MLL3 was the strongest hit that its loss cooperated with the Myc oncogene to drive HCC (Figure 1—figure supplement 1); second, in human HCCs, KMT2C (gene encoding MLL3) mutations and deletions co-occur with MYC gains and amplification.

Based on the reviewer’s suggestion, we examined MLL3 loss in conjunction with CTNNB1 activation, using HTVI of a transposon containing the constitutively active Ctnnb1. However, we did not observe oncogenic cooperation between Ctnnb1 activation and Kmt2c loss; no mice developed liver tumors by the experimental endpoint (5 months post HTVI, Figure 1—figure supplement 3). Additionally, analysis of genomic data from human HCCs showed no significant co-occurrence between CTNNB1 and KMT2C alterations (Figure 1A). These results suggest that, similar to other epigenetic regulators, the tumor suppressive function of MLL3 is likely oncogene-specific. Our in vivo screen results that nominated MLL3 as a tumor suppressor also reinforce this functional interaction with MYC oncogene. We have updated the text to reflect the context specificity of MLL3 as a tumor suppressor in our study.

Reviewer #2 (Public Review):

Soto-Feliciano et al. have characterized the function of MLL3 in hepatocellular carcinoma (HCC) suppression. MLL3 is recurrently mutated in human HCC. The authors show that Mll3 mutations cooperate with Myc overexpression to drive HCC cancer in mice. They identify Cdkn2a as a critical direct target of MLL3. Overall, the manuscript makes a compelling case that MLL3 is a bona fide HCC tumor suppressor, that it directly binds and activates the Cdkn2a locus, and that Cdkn2a acts downstream of MLL3 to suppress HCC initiation.

The strengths of the paper include mouse modeling techniques that clearly demonstrate a role for MLL3 in suppressing Myc-driven HCC, a detailed characterization of MLL3 binding sites and target gene expression, and the combined weight of several functional studies showing that MLL3 induces apoptosis in hepatocytes/HCC by inducing p16 and ARF. The major conclusions appear well-supported by the data.

The paper does have some weaknesses. Some of the genomic data require clarification. Furthermore, the authors draw broad conclusions about an epistatic relationship between MLL3 and CDKN2A based on mutually exclusive mutation patterns in human cancers. Those conclusions are not as well-supported as the mechanistic conclusions. The incidences of MLL3 and CDKN2A mutations in HCC are both relatively low (1% and 5% respectively), so it seems difficult to draw any conclusions from mutually exclusive profiles.

One additional criticism is that the paper is a bit reductive. The link to CDKN2A offers a satisfying explanation for how MLL3 suppresses HCC, but the model may oversimplify the functions of MLL3.

We thank the reviewer for their constructive comments and suggestions, which we addressed as follows with point-by-point response provided below. We agree with the concerns regarding the mutational analyses of KMT2C and CDKN2A in human cancers and the working model of the manuscript. We have removed the majority of the mutational analyses from the Results section. Importantly, our latest integrative analyses of RNA-seq and MLL3 ChIP-seq revealed other potential downstream effectors of MLL3 tumor suppressive functions (Figure 3A and Figure 3—figure supplement 1B). We have modified the Results and the Discussion to reflect this more nuanced view of MLL3 function in cancer. Nonetheless, we believe that other data continue to support our conclusion that CDKN2A is a dominant effector of MLL3 tumor suppressive functions in our model.

Reviewer #3 (Public Review):

The enhancer chromatin-modifying enzyme MLL3 functions as a tumor suppressor in multiple human cancers, however, the mechanisms underlying its tumor suppressive function remain unclear. The manuscript of SotoFeliciano et al. focused on Myc-driven liver cancer and aimed to address and fill the gap. The authors used an elegant genetic design and approach to manipulate the overexpression of the Myc oncogene and knockout of the Mll3 tumor suppressor gene in mouse liver cancer models. Their genetic mouse models showed that loss of Mll3 constrains Myc-driven liver tumorigenesis, with tumors having a slightly later onset compared to mice with Myc overexpression in conjunction with p53 inactivation. Because MLL3 is a major histone-modifying enzyme for enhancer-associated H3K4 monomethylation and is responsible for enhancer activation and the following target gene transcription, they performed ChIP-seq analysis to study the roles of Mll3 in Myc-driven mouse liver cancer. Interestingly, their ChIP-seq studies revealed that loss of Mll3 preferentially limits Mll3 enrichments at promoters and thereby attenuates promoter-associated H3K4 trimethylation and target gene transcription, whereas the unchanged Mll3 genomic binding between the two genotypes (Myc;sgTrp53 and Myc;sgKmt2c) is largely located within enhancer (intergenic) regions. They further demonstrated that the cdkn2a locus is a genomic and transcriptional target of Mll3 in Myc-driven mouse liver cancer. Supporting their findings, genomic inactivations of MLL3 and CDKN2A displays mutual exclusivity in human liver cancer and many other cancer types. Furthermore, they described a possible mechanism for MLL3's role in MYC-driven liver cancer that MLL3 mediates MYC-induced apoptosis in a CDKN2A-dependent manner by manipulating Myc overexpression, Mll3 function, and Cdkn2a regulation in their genetic mice models. This manuscript describes a potential function of MLL3 in the control of tumor suppressor gene expression via modulating their promoter chromatin landscapes. More importantly, loss of normal function of MLL3 or the downstream effector CDKN2A may impair MYC-induced apoptosis, and in turn, lead to MYC-induced tumorigenesis.

Overall, the manuscript is well written, organized, and focused on an interesting topic, and with data presented supports the authors' claims.

We thank the reviewer for their positive and constructive comments and suggestions.

Author Response

Reviewer #1 (Public Review):

This manuscript represents a substantial and well-executed body of work that contributes new data on 32 hymenopteran genomes, systematically identifies viral endogenization and domestication events, and tests whether this phenomenon is more common in hymenopteran species with specific lifestyles, eg. endoparasitism. The authors developed a pipeline to identify endogenization that improves upon previously described pipelines and is more comprehensive for the identification of endogenization events from a variety of virus types. Significant findings include the identification of previously undocumented cases of viral endogenization in several hymenopteran species and also moderate statistical support for a higher rate of dsDNA virus endogenization and domestication in endoparasitoids.

1) The authors have tested whether the lifestyle of hymenopteran species (endoparasitism, ectoparasitism, or free-living) is related to the incidence of virus endogenization and domestication. Addressing this kind of question has only become possible with the availability of genome sequences from many taxa so that any results can be statistically supported by appropriate sample sizes. It appears that the authors have not included new genomic data from hymenopteran genomes that have been published since 2019, which are of similar or better quality than the data used in this manuscript. A number of taxa with endogenous viruses (and also without) have become available since then. The best solution would be for the authors to use their pipeline to incorporate the new data, which may have an impact on their findings and could even strengthen their conclusions about virus domestication being more common in endoparasitoids. If this is not possible, the authors should at least justify their decision not to include the most recent data and discuss how it could affect their results.

The first step of our pipeline is to extract all candidate loci from each genome. Then all these loci are clustered and further analyzed to infer endogenization and domestication (sequence alignments, phylogeny, dN/dS, genomic context, mapping…). Thus, adding new genomes requires re-run the whole analysis from the very beginning which represents a huge amount of work and computational resources together with their associated carbon costs. Additionally, this work was part of Benjamin Guinet’s Phd project which was defended on 21 Marsh 2023. In conclusion, we will not be able run again the whole pipeline in all the genomes published since 2019.

2) Please summarize in the main manuscript (results or discussion) what the limitations of the pipeline to detect EVEs and dEVEs are - what are important factors to consider, including the availability of closely related "free-living" viruses, and of closely related wasp species for dN/dS analyses.

We added a paragraph in the discussion section to discuss the limitations of our pipeline as follow:

“Because the identification of EVEs necessitates the availability of related viruses in the database, we should see these numbers as an underestimation of the real number of EVEs. In addition, our pipeline necessitates the availability of either related species sharing the same EVEs (or at least the presence of paralogs within a single species) or the availability of RNAseq data to infer domestication. Because these last conditions were only met for 701 out 1261 EVEs, the results we obtained here regarding domestication should be seen as an underestimation of the prevalence of the phenomenon.”

3) In this manuscript, a description of the methods that precede the results would make it much easier to appreciate the results shown. It appears that this is allowed in cases where it makes sense, according to the author's instructions.

The first paragraph of the result section was intended to give this overview on the methods used. However, we tried to give more details on the methods in this paragraph. We hope that this will increase readability of the paper.

4) The sensitivity and specificity of methods analysis are commendable, as is the availability of substantial supplementary data and scripts on GitHub. However, more effort could be made to align numbers reported in the text and in figures so that readers can verify support for the conclusions described.

To align numbers reported in the text and the figures, we added a new excel sheet within the supplementary file 6 named “Figure_data” in which we report the data used to build the figures 2A, 3A and 3B.

Reviewer #2 (Public Review):

Guinet et al address the question of whether the divergent lifestyles in hymenopteran insects determine the rates of acquisition and domestication of viral genetic elements. As endoparasitoids are intimately associated with their hosts and often develop as broods herein, they predicted that the acquisition rate is higher compared to free-living and ectoparasitoid hymenopterans. Following viral domestication in the new recipient wasp genome, these viral elements have been shown to contribute to endoparasitism by promoting the delivery of secreted compounds in insect hosts (where immature wasps develop). Because of this functional importance, the authors predicted that the rate of domestication is also higher in endoparasitoid wasps. I was impressed with the solid and rigorous approach that was followed to test these two hypotheses. The authors carefully ruled out confounding factors, including contamination of genome assemblies. Previously characterized hymenopteran genomes were included as positive controls to assess the developed pipelines. There was also great merit in using a Bayesian model to study endogenization within the phylogenetic framework. To summarize, this multi-pronged strategy to mine animal genomes for viral genetic elements has the potential of becoming a new benchmark for future studies.

Although the authors do partially achieve their aim of coupling endogenization with an endoparasitoid lifestyle, I am afraid some of the assumptions and generalizations hinder a more solid conclusion. I feel that categorizing hymenopterans either as free-living, endoparasitoids, or ectoparasitoids is an oversimplification. Many of the authors' arguments to associate endogenization with endoparasitoids also apply to free-living eusocial hymenopterans. Both endoparasitoid and eusocial insects can be relatively more exposed to viruses because of intimate conspecific interactions within confined spaces. As endoparasitoids intimately interact with their host, so do eusocial insects with their social guests (melittophiles, myrmecophiles, and termitophiles). Perhaps, you could even argue that some gregarious insects also fit the bill. I would be interested to see whether the conclusions hold when "free-living" is further subdivided and "eusocial" is a separate category.

To answer this question, we reran the study by separating the free-living category into "eusocial" and "free-living" subcategories. All of the new eusocial assignations and their accompanying bibliographies have been added to the Supplemental file 1 under the columns "lifestyle2" and "ref-lifestyle2". All of the new GLM results have been added to the Supplemental file 6. We also made a new violin plot figure names “Figure 4-figure supplement 3” which contains the GLM coefficients distribution of the model run on A only and A to D scaffolds.

We also added a few lines in the M&M to explain this analysis “The same analysis was carried out by splitting the free-living category into two sub-categories, namely eusocial and free-living. A new glm model was then built (GLM(Number EVEs ~ free-living + eusocial + endoparasitoid + ectoparasitoid * Branch_length, family = zero inflated neg binomial). (Lines 754-756).

Overall, the new models that included free-living eusocial hymenopterans revealed the exact same patterns as found in the main analysis. We added a new section entitled “Conclusions hold when eusociality is taken into account” to report the results.

In conclusion, when "free-living" is further subdivided, the mains findings still hold.

Second, I wonder why the authors did not include Wolbachia infection as an explanatory variable to explain the endogenization rate. Wolbachia bacteria infect the insect germline and are often associated with phages. These phages could thus be a major source of viral genetic elements. Having said that, I do not see any Symbioviridae, the phylogenetic clade in which these phages reside (https://doi.org/10.1371/journal.pgen.1010227), in Figure 2B - so perhaps this is a minor point.

In this study we chose to concentrate our attention on eukaryotic viruses, since we reasoned that they have better opportunities to integrate into the insect genomes du to their intimate relationship. This is the reason why we eliminated from our database all phage proteins (as specified in line 511).

Finally, in addition to the dsDNA virus - endoparasitoids relationship, the authors also detect a link between ssRNA viruses and free-living hymenopterans. (Maybe eusociality is biasing these results?)

Thanks to reviewer’s comment, we realize that the sentence referring to this point was misleading. In our initial analysis, ectoparasitoid species showed in fact less domestication events involving ssRNA viruses compared to all other lifestyles (Figure 4-figure supplement 1-L). We clarified this sentence in the main text as follow: “except for a lower rate of domestication of ssRNA viruses in ectoparasitoids compared to other lifestyles (Figure 4-figure supplement 1-L)” (lines 195-196).

The same effect was observed when including eusociality (see Figure 4-figure supplement 3-L).

Author Response

Reviewer #1 (Public Review):

Estimating the effects of mutations on the thermal stability of proteins is fundamentally important and also has practical importance, e.g, for engineering of stable proteins. Changes can be measured using calorimetric methods and values are reported as differences in free energy (dG) of the mutant compared to wt proteins, i.e., ddG. Values typically range between -1 kcal/mol through +7 kcal/mol. However, measurements are highly demanding. The manuscript introduces a novel deep learning approach to this end, which is similar in accuracy to ROSETTA-based estimates, but much faster, enabling proteomewide studies. To demonstrate this the authors apply it to over 1000 human proteins.

The main strength here is the novelty of the approach and the high speed of the computation. The main weakness is that the results are not compared to existing machine learning alternatives.

We thank Prof. Ben-Tal for taking the time to assess our work, and for his comments and suggestions below.

Reviewer 2 (Public Review):

Summary:

This work presents a new machine-learning method, RaSP, to predict changes in protein stability due to point mutations, measured by the change in folding free energy ΔΔG.<br /> The model consists of two coupled neural networks, a 3D selfsupervised convolutional neural network that produces a reduceddimensionality representation of the structural environment of a given residue, and a downstream supervised fully-connected neural network that, using the former network's structural representation as input, predicts the ΔΔG of any given amino-acid mutation. The first network is trained on a large dataset of protein structures, and the second network is trained using a dataset of the ΔΔG values of all mutants of 35 proteins, predicted by the biophysics-based method Rosetta.

The paper shows that RaSP gives good approximations of Rosetta ΔΔG predictions while being several orders of magnitude faster. As compared to experimental data, judging by a comparison made for a few proteins, RaSP and Rosetta predictions perform similarly. In addition, it is shown that both RaSP and Rosetta are robust to variations of input structure, so good predictions are obtained using either structures predicted by homology or structures predicted using AlphaFold2.<br /> Finally, the usefulness of a rapid approach such as RaSP is clearly demonstrated by applying it to calculate ΔΔG values for all mutations of a large dataset of human proteins, for which this method is shown to reproduce previous findings of the overall ΔΔG distribution and the relationship between ΔΔG and the pathological consequences of mutations. The RaSP tool and the dataset of mutations of human proteins are shared.

Strengths:

The single main strength of this work is that the model developed, RaSP, is much faster than Rosetta (5 to 6 dex), and still produces ΔΔG predictions of comparable accuracy (as compared with Rosetta, and with the experiment). The usefulness of such a rapid approach is convincingly demonstrated by its application to predicting the ΔΔG of all single-point mutations of a large dataset of human proteins, for which using this new method they reproduce previous findings on the relationship between stability and disease. Such a large-scale calculation would be prohibitive with Rosetta. Importantly, other researchers will be able to take advantage of the method because the code and data are shared, and a google colab site where RaSP can be easily run has been set up. An additional bonus is that the dataset of human proteins and their RaSP ΔΔG predictions, annotated as beneficial/pathological (according to the ClinVar database) and/or by their allele frequency (from the gnomAD database) are also made available, which may be very useful for further studies.

Weaknesses:

The paper presents a solid case in support of the speed, accuracy, and usefulness of RaSP. However, it does suffer from a few weaknesses.

The main weakness is, in my opinion, that it is not clear where RaSP is positioned in the accuracy-vs-speed landscape of current ΔΔGprediction methods. The paper does show that RaSP is much faster than Rosetta, and provides evidence that supports that its accuracy is comparable with that of Rosetta, but RaSP is not compared to any other method. For instance, FoldX has been used in large-scale studies of similar size to the one used here to exemplify RaSP. How does RaSP compare with FoldX? Is it more accurate? Is it faster? Also, as the paper mentions in the introduction, several ML methods have been developed recently; how does RaSP compare with them regarding accuracy and CPU time? How RaSP fares in comparison with other fast approaches such as FoldX and/or ML methods will strongly affect the potential usefulness and impact of the present work.

Second, this work being about presenting a new model, a notable weakness is that the model is not sufficiently described. I had to read a previous paper of 2017 on which this work builds to understand the self-supervised CNN used to model the structure, and even so, I still don't know which of 3 different 3D grids used in that original paper is used in the present work.

A third weakness is, I think, that a stronger case needs to be made for fitting RaSP to Rosetta ΔΔG predictions rather than experimental ΔΔGs. The justification put forward by the authors is that the dataset of Rosetta predictions is large and unbiased while the dataset of experimental data is smaller and biased, which may result in overfitting. While I understand that this may be a problem and that, in general, it is better to have a large unbiased dataset in place of a small biassed one, it is not so obvious to me from reading the paper how much of a problem this is, and whether trying to fix it by fitting the model to the predictions of another model rather than to empirical data does not introduce other issues.

Finally, the method is claimed to be "accurate", but it is not clear to me what this means. Accuracy is quantified by the correlation coefficient between Rosetta and RaSP predictions, R = 0.82, and by the Mean Absolute Error, MAE = 0.73 kcal/mol. Also, both RaSP and Rosetta have R ~ 0.7 with experiment for the few cases where they were tested on experimental data. This seems to be a rather modest accuracy; I wouldn't claim that a method that produces this sort of fit is "accurate". I suppose the case is that this may be as accurate as one can hope it to be, given the limitations of current experimental data, Rosetta, RaSP, and other current methods, but if this is the case, it is not clearly discussed in the paper.

We thank the reviewer for their detailed comments and suggestions.

As discussed in our general comments above and also below, we have now added additional benchmarking, making it easier to compare the accuracy of RaSP with other methods. Regarding the model description, we have now added a more detailed description of also the 3D CNN.

Regarding whether to fit the model to experiments or computational data, we agree that it is not clear cut that the former would also not work. Indeed, a main problem is that in both cases it is hard to answer which approach is better because of the scarcity of experimental data. One major problem with the larger sets of experimental data is, as we mention, the bias and variability; another is the provenance. While some databases exist, they are rarely exactly raw data, and for example may contain ∆∆G values estimated from ∆Tm values. In the revised manuscript we now explain better why we chose to target Rosetta, but also acknowledge that one might also have used experiments.

As to the question of accuracy, we agree completely that the methods could be better. One problem, however, is that it is very difficult to answer how much better because of problems with experiments. As mentioned also by reviewer 1, variation across different experiments suggest that even a “perfect” predictor would only achieve Pearson correlation coefficients in the range 0.7–0.8 (https://doi.org/10.1093/bioinformatics/bty880). Clearly, this is an issue with imperfect data curation (it is possible to measure ∆∆G quite accurately), but in the absence of larger and better curated experiments, one will not expect much better accuracy than what we report here. This is now discussed in the revised manuscript.

Reviewer 3 (Public Review):

The authors present a machine learning method for predicting the effects of mutations on the free energy of protein stability. The method performs similarly to existing methods, but has the advantage that it is faster to run. Overall this is reasonable and a faster method will likely have some potential uses. However, not improving performance beyond the reasonable but not great performance of existing methods of course makes this a less useful advance. The authors provide predictions for a set of human proteins, but the impact of their method would be much greater if they provided predictions for all substitutions in all human proteins, for example. In places the text somewhat overstates the performance of computational methods for predicting free energy changes and is potentially misleading about when ddGs are predicted vs. experimentally measured. In addition, the comparison to existing methods is rather slim and there isn't a formal evaluation of how well RASP discriminates pathological from benign variants.

We thank the reviewer for taking time to read our work and for their various suggestions.

Author Response

Reviewer #1 (Public Review):

Alignment between high dimensional data which express their dynamics in a subspace is a challenge which has recently been addressed both with analytic-based solutions like the Procrustes transformation, and, most interestingly, via deep learning approaches based on adversarial networks. The authors have previously proposed an adversarial network approach for alignment which relied on first dimensionally-reducing the binned neural spikes using an autoencoder. Here, they use an alternative approach to align data without use of an initial dimensional-reduction step.

The results are fairly clear - the Cycle-GAN approach works better than their previous ADAN approach and one based on dimensionality reduction followed by the Procrustes transform. In general, a criticism of this entire field is to understand what alignment teaches us about the brain or how it specifically will be used in a BCI context.

There are a few issues with the paper.

1.) To increase the impact of their work, the investigators have now used it to align data in multiple types of tasks. There was an unanswered question about this related to neuroscience - does alignment in one task predict alignment for another?

This is a great question! We anticipate that it will be challenging for an alignment learned on one task to be used on another task, because we know that M1 decoders trained on data from one behavior often do not generalize when tested using a different behavior (Naufel et al., 2019)*. The same nonlinearities that prevent zero-shot decoding across tasks are also likely to impair the ability of an aligner trained on data from one task to successfully align data from another task. Furthermore, the results of Naufel et al. indicate that even if neural alignment is successful, we would need a decoder already trained on the new task to produce reliable predictions-- in which case the data needed to train that decoder could simply be used for alignment. A systematic study of the relation between the ability to align and decode from data is well warranted, but beyond the scope of our current work.

*Naufel, S., Glaser, J. I., Kording, K. P., Perreault, E. J., & Miller, L. E. (2019). A muscle-activity-dependent gain between motor cortex and EMG. Journal of neurophysiology, 121(1), 61-73.

Action in the text: none.

2) Investigators use decoding as a way of comparing alignment performance. The description of the cycle GAN was not super detailed, and it wasn't clear whether there was any dynamic information stored in the network that might create questions of causality in actual use. It seems that input is simply the neural activity at a current time point rather than neural activity across the trial, which would alleviate this concern. However, they mention temporal alignment but never describe in detail whether all periods of spikes are properly modeled by the system or if only subsets of data (specific portions of task or non-task time) will work. Perhaps this is more a question of the Wiener filter, for which precise details are missing.

As intuited by the reviewer, we did only use the neural activity at a current time point as the inputs for Cycle-GAN training, so the system is causal and can be used in real time. We have modified the text to clarify this.

We apologize for any confusion caused by our use of the term "temporal alignment", which was for the sake of consistency with earlier-published, CCA-based alignment methods (e.g., in Gallego et al., 2020), but is indeed confusing. In the revised manuscript, we have switched to the term ‘trial alignment’ which we believe better reflects this pre-processing step, and we have included additional explanations in the introduction.

Importantly, while CCA-style trial alignment is not required by our methods, we do still preprocess our data to exclude behaviors not related to the investigated task. Since monkeys were resting or performing task-irrelevant movements during inter-trial period, we chose to use data only from trial start to trial end, but without any explicit trial matching or alignment (see Appendix 1 - Behavior tasks). In the revised manuscript, we now show that our methods still works well when applied even to the continuous recordings, with Cycle-GAN significantly outperforming both ADAN and PAF.

Action in the text (page 2, lines 72-74): clarifying CCA description and replacing “temporal alignment” with “trial alignment”.

Action in the text (page 5, lines 191-192): stating that ADAN and Cycle-GAN have no knowledge of dynamics.

Action in the text (page 6, lines 258-272): documenting performance on full-day recordings without trial matching.

Action in the text (page 13, lines 647-649): again, stating that Cycle-GAN has no knowledge of dynamics.

3) In general, precise details of the algorithms should have been provided.

We appreciate the reviewer noting this-- in the submitted manuscript, the full descriptions of Cycle-GAN and ADAN were included as supplementary methods in Appendix 4, but we did not extensively reference this and it may have been missed. In the revised manuscript, we added more references to Appendix 4 and in the Methods section of the main text. We provided further details on the choice of hyperparameters for each method (including PAF) in Appendix 4 itself.

Action in the text (page 13, lines 643-644): added “For a full description of the ADAN architecture and its training strategy, please refer to “ADAN based aligner” in Appendix 4 and (Farshchian et al., 2018).”

Action in the text (page 14, lines 669): added “Further details about the Cycle-GAN based aligner are provided in “Cycle-GAN based aligner”, Appendix 4.” Action in the text (Appendix 4 Tables 1-2): We have added a summary table of hyperparameters for each method in Appendix 4 (ADAN: Appendix 4 Table 1; CycleGAN: Appendix 4 Table 2).

4) Cross validation for day-0 alignment is not explained.

As mentioned above, the training and validation details of day-0 models were included in Appendix 4, which was not extensively referenced in the manuscript and may have been missed. We have now added more references to the Appendix in the revised manuscript.

Action in the text (page 13, lines 627-629): added “(Note that this LSTM based decoder is only used for latent space discovery, not the later decoding stage that is used for performance evaluation (see “ADAN day-0 training” in Appendix 4 for full details)).”

5) Details of statistical tests is not provided.

We apologize for this omission. In the revised manuscript, we have added a section in the methods summarizing all the statistical tests. In addition, we added the sample sizes for each stat reported in the results section.

Action in the text (page 15, lines 754-768): new Methods section added.

6) (minor) The idea that for neurons that have disappeared that the CycleGAN can "infer their response properties", seems an incorrect description. A proper description should be that it "hallucinates" their response properties?

We prefer to avoid the term “hallucinate”, due to its recent increased (appropriate) use in the context of large language models describing content generation that is “nonsensical or unfaithful to the provided source content” (as per the Wikipedia article on hallucination in AI). The synthetized “responses” of vanished neurons are not nonsensical, but are indeed, inferred: they are the model’s best estimate of how these neurons would have responded, had they been observed. While not explored further here, this prediction could be of potential scientific use: a strong discrepancy between predicted and observed activity might be a clue to look for further evidence of learning or remodeling of neural representations of behavior.

Action in the text: none.

Reviewer #2 (Public Review):

In this manuscript, the authors use generative adversarial networks (GANs) to manipulate neural data recorded from intracortical arrays in the context of intracortical BCIs so that these decoders are robust. Specifically, the authors deal with the hard problem where signals from an intracortical array change over time and decoders that are trained on day 0 do not work on day K. Either the decoder or the neural data needs to be updated to achieve the same performance as initially. GANs try to alter the neural data from day K to make it indistinguishable to day 0 and thus in principle the decoder should perform better. The authors compare their GAN approach to an older GAN approach (by an overlapping group of authors) and suggest that this new GAN approach is somewhat better. Major Strengths are multiple datasets from behaving monkeys performing various tasks that involve motor function. Comparison between two different GAN approaches and a classical approach that uses factor analysis. The weakness is insufficient comparison to another state-of-the-art approach that has been applied on the same dataset (NoMAD, Karpowicz et al. BioRxiv.)

The results are very reasonable and they show their approach, Cycle GANs, does slightly better than the traditional GAN approach. However, the Cycle GANs have many more modules and also as I understand it performs a forward backward mapping of the day - 0 and day - k and thus theoretically better. But, it seems quite slow.

We are concerned that the reviewer may have mistaken the Cycle-GAN training time (the time it takes to find an alignment, Figure 4B) with its inference time (the time it takes to transform data once an alignment has been found). Whereas inference time is critical for practical deployment of a model, we argue that Cycle-GAN's somewhat longer training time is not a substantial barrier to use: it is still reasonably fast (a few minutes) and training will only need to be performed on the order of once per day. We have modified the y-axis label of Figure 4B to make this distinction clearer.

We have also now added information on the inference speed of trained models to the paper: we find that both Cycle-GAN and ADAN perform the inference step in under 1 ms per 50 ms sample of data – this is because the forward map in both models consists of a fully connected network with only two hidden layers. We also note that while forward-backward mapping between days does occur during Cycle-GAN training, only the forward mapping is performed during inference.

Action in the text (page 7, lines 303-306): added inference time for Cycle-GAN and ADAN.

I think the results are interesting but as such, I am not sure this is such a fundamental advance compared to the Farashcian et al. paper, which introduced GANs to improve decoding in the face of changing neural data. There are other approaches that also use GANs and I think they all need to be compared against each other. Finally, these are all offline results and what happens online is anyone's real guess. Of course, this is not just a weakness of this study but many such studies of its ilk.

Author Response

Reviewer #1 (Public Review):

The study by Yang et al. reports a new mechanistic role of vinculin in inhibiting the Mef2c nuclear translation and sclerostin expression in osteocytes and promoting bone formation. The authors showed the reduction of vinculin in aged bone human bone samples. A 10kb DMP-1-Cre mouse model was generated that deleted vinculin in osteocytes. They found that vinculin deletion caused bone loss and decreased bone formation associated with increased sclerostin expression. This increase does not affect the protein level of transcription factor Met2c but interestingly enhances nuclear translocation. Vinculin is interested in Mef2c and appears to retain Mef2c in the cytosol. As expected, as a component of the mechanosensory focal adhesion complex, bone formation via tibial loading was decreased in vinculin deletion. Intriguingly, the bone loss associated with estrogen deficiency through ovariectomy was attenuated. Overall, the study unveiled an important role concerning a key player of focal adhesion and the study was well designed and executed. The paper would be strengthened by including a more thorough discussion including variables such as male vs. female, and cortical vs. trabecular bone as the vinculin deletion appeared to primarily affect trabecular bone while mechanical loading exerts anabolic effects on both bone types. The effect of estrogen deficiency effect is interesting and is worth some discussion.

Strengths:

The paper shows a novel mechanism that vinculin retains Mef2c in the cytosol via protein interaction to prevent it from migrating to the nucleus and increases transcription of sclerostin, an inhibitory factor for Wnt/β-catenin signaling, a critical pathway for osteoblast activity and bone formation. They employed various in vivo and in vitro models as well as human tissue samples including generating conditional knockout of vinculin in osteocytes in vivo and vinculin gene knockdown in MLO-Y4 cells. They also used physiological/pathological relevant models, tibial loading, and ovariectomy to study the role of vinculin under mechanical loading and estrogen deficiency. The adopted standard techniques to study bone properties include microCT, bone formation, bone histomorphometry, histochemistry as well as biochemical assays such as immunoprecipitation, ChIP assays, etc.

The study is comprehensive and thorough and the noticeable uniqueness is that after observing the phenotypes from in vivo data, they further explored the underlying mechanisms using cell models. The experiments in general are well-designed and presented with adequate repeats and statistical analysis. The paper is also logically written and the figures were clearly labeled.

We highly thank the reviewer for his/her positive comments and helpful suggestions.

Minor weaknesses:

More discussion is necessary concerning the potential difference in responses between male and female. Most of the studies were conducted in male mice except ovariectomy mice.

During the revision, we have added new results from µCT analysis. Our new results showed that vinculin loss significantly reduced bone mass in 6-month-old female mice (Figure 2-figure supplement 1. a-d).

It is interesting that the cKO of vinculin in osteocytes primarily affects trabecular bones with limited effect on cortical bones. However, sclerostin is increased in cortical bones. The promotion of bone formation by mechanical loading appears to affect both cortical and trabecular bones. If focal adhesion is a key mechanosensory complex, how to reconcile the different responses in the cKO model?

We thank the reviewer for raising this good point. In fact, we do not know why there was no marked cortical bone loss in cKO mice. During the revision, we performed three-point bending analysis to determine whether vinculin loss impaired in the mechanical properties of the long bone and found that the ultimate force and total energy absorption before fracture were decreased in the femur of 3-month-old male cKO compared to those in control mice (Figure 3-figure supplement 1. e, f). Furthermore, our new results from the calcein double labeling experiments showed that both MAR and BFR of femur cortical bones were slightly but significantly reduced in cKO mice relative to those in control mice (Figure 3-figure supplement 1. a-c).

The OVX response is interesting and it is worthwhile to elaborate more regarding the potential underlying mechanism and what's the relationship between estrogen and mechanical loading and if the action of estrogen on vinculin shares any similar mechanisms with mechanical loading, etc.

We feel that the relationship between estrogen and mechanical loading could be quite complex, which deserves further investigation in the future. Thank you for this good point.

Reviewer #3 (Public Review):

This study by Wang et al. investigates the role of the focal adhesion protein vinculin in osteocytes and its effect on bone mass. First, they showed decreased levels of vinculin in osteocytes in trabecular bone from aged individuals compared to young, suggesting a potential role for vinculin in regulating bone mass with aging. Next, they deleted vinculin in late osteoblasts and osteocytes in young and older mice and found decreased bone mineral density and trabecular bone mass. This was due to impaired bone formation, which the authors attributed to increased sclerostin levels. Further in vitro experiments showed that vinculin regulates sclerostin via the transcription factor Mefc2. Conditional knockout of vinculin in late osteoblasts and osteocytes had no effect on the bone of mice lacking Sost, further implicating an essential role for sclerostin in mediating the effects of vinculin in osteocytes. Interestingly, the vinculin conditional knockout mice had an impaired response to mechanical loading, suggesting an important role for vinculin in the osteocyte mechanoresponse. Finally, the authors showed that while ovariectomy increased osteoclast formation and bone resorption in control mice, it had no effect on the bone of the vinculin conditional knockout mice.

Overall, the authors show convincing data for the important role of vinculin in osteocytes in regulating the anabolic effects of bone formation under physiological conditions. They also show that osteocyte vinculin may be a regulator of bone resorption under conditions mimicking postmenopausal osteoporosis. However, not all of the conclusions are fully supported by the data.

Strengths:

The use of both in vivo and in vitro approaches to determine the role of vinculin in osteocytes provides compelling evidence for its importance under basal conditions and in regulating the anabolic effects of mechanical loading. The in vitro assays nicely demonstrate a potential mechanism through Mef2c/ECR5.

The creation of the vinculin and Sost double conditional knockout mouse model provides further convincing evidence for the causative role of sclerostin in the effects of vinculin knockout in osteocytes.

The use of both young and older male mice links nicely with the human samples where vinculin expression appears to be reduced in osteocytes with aging. The authors need to be careful in describing 14-month-old mice as aged though, as these mice would not be typically thought of as old.

Weaknesses:

The methods section is lacking in basic details (e.g., there is no information on the CRISPR deletion of Vcl in the MLO-Y4 cells). While referencing their previous papers is fine, a brief description of the methods should be included in this paper.

During the revision, we have added the method of CRISPR deletion of Vcl in the MLO-Y4 cells (page 22, line 16-20).

While much of the data linking vinculin to sclerostin is convincing, it is surprising that the authors show decreased trabecular bone volume in the vinculin cKO mice, yet show increased sclerostin levels in the cortical bone. If increased sclerostin is responsible for impaired bone formation in the vinculin cKO mice, why is there no cortical bone phenotype?

Please see our response to above Point 3.

It would be important for the authors to also show the sclerostin immunostaining in the trabecular bone of these animals.

During the revision, we utilized IHC to demonstrate that cKO mice also displayed increased level of the sclerostin protein in cortical bone compared to the control group (Figure 4-figure supplement 1. a, b).

The authors do not provide any potential explanation for the effects of vinculin cKO in the ovariectomized mice. Under physiological conditions, osteocyte vinculin has no effect on osteoclast number or bone resorption. How is osteocyte vinculin affecting osteoclasts after ovariectomy? Are there differences in the osteocyte expression of Rankl or Opg in response to the loss of estrogen in the vinculin cKO and control mice?

During the revision, we used IHC staining to measure the expression of Rankl and Opg in the cortical bone of control and cKO mice treated with or without OVX surgery. The results showed that in cKO mice, the increase in Rankl induced by OVX was lower than that in the control group, while the increase in Opg induced by OVX was higher than that in the control group. The Rankl/Opg ratio was decreased in cKO mice induced by OVX (Figure 7-figure supplement 1. a-d).

From their in vitro experiments, the authors deduce that loss of vinculin affects osteocyte attachment. However, their images would suggest that it is the formation of dendrites that is strongly inhibited in the cells lacking vinculin. It is surprising that no investigation of osteocyte dendrite number or connectivity was performed in the vinculin cKO mice. This is particularly important as a decrease in osteocyte dendrites and connectivity has been observed in the bones of aged mice (see Tiede-Lewis et al., Aging. 2017) and osteocyte dendrites are important for mechanosensation.

Please see our response to above Point 4.

Author Response

Reviewer #1 (Public Review):

This manuscript investigates the mechanisms of 'summiting disease' using a previously characterised Drosophila model. The authors also show that E. muscae infiltrates the brain likey through a defective blood-brain barrier and populates regions of the brain in the medial protocerebrum. It likely releases metabolites into the haemolymph of summiting flies that has the ability to induce summiting in uninfected flies. They also show that a burst of locomotor activity precedes death. To understand the circuit basis of this, they perform a screen of more than a hundred neuronal lines and genes to identify an active DPN1>pars intercerebralis neurons> corpora allata>JH axis as being invovled in the summiting behaviour while not affecting death.

Thank you for your succinct summary of our paper.

Reviewer #2 (Public Review):

In this study, the authors aim to uncover the neuroanatomical and metabolite underpinnings of an intriguing phenomenon observed in some insects due to the infection of fungal pathogens. They very cleverly develop a high-throughput assay to examine and quantify this behaviour in a tractable model organism - Drosophila melanogaster which the authors have previously shown to also exhibit this phenomenon. They characterize the details of this behaviour and clearly show the temporal gating of this summiting-followed-by-death behavior to occur shortly before the dusk transition. They go on to examine using a candidate (over 200) screen approach potential neuronal circuits and genes based on the hypothesis that they may be related to 'arousal and gravitaxis'. They narrow down to a line that is restricted to the PI based on the fact that it has a significant effect on the summiting behaviour and that it is known to affect locomotion. They can demonstrate that flies when a subset of PI neurons (R19G10) are transiently activated, they will show summiting even without exposure to the pathogen. Based on Syt-eGFP staining they conclude that PI communicates with the carpora cardiaca (CA). They also show that CA itself is necessary for this behavior, but cannot demonstrate the role of Juvenile hormones using their pharmacological methods.

The authors then describe an automated classifier to identify an upcoming summiting behaviour. Further, they use this real-time classifier to stage different steps of the summiting and match it to the extent of pathology observed by microscopy. They also ask whether the constituents of the hemolymph differ between the summiting and not-yet summiting flies for which they conduct metabolome analysis of the hemolymphs. They are also able to show that cross-injection of uninfected or infected but not summiting flies can be induced to show summiting-like behaviour upon injection with the hemolymph. Finally, they propose the sequence by which the fungal pathogen may modulate the behaviours of the host fly so as to execute this highly gated act of increased locomotion prior to death.

This is a good summary of our findings.

Strengths

• The detailed characterization of the behaviour in D melanogaster and development of the high-throughput behavioural arena.

• Development of the automated classifier which appears to accurately predict this behaviour.

• Narrowing down to a small group of PI neurons having a strong impact on this behaviour although sufficiency is not clearly demonstrated.

Thank you for highlighting these areas of our paper. With respect to demonstrating sufficiency of the PI neurons, we believe this actually an area of comparative strength for the manuscript. With thermogenetic and complementary optogenetic experiments, we demonstrated that activating the PI-CA neurons induces a burst of locomotion consistent with that seen during summiting. A similar burst of locomotion is seen when thermogenetically activating DN1ps. These experiments demonstrate that activity in these neurons is sufficient to induce a pattern of activity like that seen during summiting. In future studies, once the molecular effectors are identified, we may be able to show that fungal alteration of the physiology of these neurons alone is sufficient to induce a burst of locomotion, but that experiment is beyond our current capabilities and beyond the scope of this study.

Weaknesses

• The evidence of temporal (circadian) gating is weak despite the proposed DN1p - PI - CA connections.

• The eventual modification of the behavior to enable enhanced locomotion and negative geotaxis to occur appears to be mediated by yet unknown factors

• The metabolite analysis did not help to narrow down to candidates that can be speculated to cause this behaviour.

With respect to evidence for temporal gating, in this study we did not aim to address the underpinnings of the timing of summiting behavior in this study and did not mean to suggest that the timing of summiting behavior is explained by DN1ps being fly clock neurons. As previously stated in response to high level comments from the editor, we interpret the data presented here as evidence that host neurons (which just happen to be clock neurons) are manipulated by the fungus to inducethe characteristic burst of pre-death locomotor activity that we believe is the key feature of summiting. We have added the following paragraph in the discussion (see Host circadian and pars intercerebralis neurons mediate summiting) to clarify this point:

“Our data indicate that the host circadian network is involved in mediating the increased locomotor activity that we now understand to define summiting. However, our data do not speak to how the timing of this behavior is determined in the zombie-fly system. That is, we have yet to address the mechanisms underlying the temporal gating of summiting and death. Our observation that E. muscae-infected fruit flies continue to die at specific times of day in the absence of proximal lighting cues (Fig 1-S1) suggests that the timing of death is under circadian control and aligns with previous work in E. muscae-infected house flies (Krasnoff et al., 1995). Given that molecular clocks are prevalent across the tree of life, it is likely that two clocks (one in the fly, one in E. muscae) are present in this system. Additional work is needed to determine if the host clock is required for the timing of death under free-running conditions and to assess if E. muscae can keep time.”

We agree that there are many unknown factors at play in this behavior. These include molecular effectors produced by the fungus that alter the physiology of host neurons, and the specific mechanisms by which JH release from the CA alters locomotion. We have endeavored to transparently present what we do and don’t know at this time and hope to be able to address these additional elements in subsequent studies.

It is true that we were unable to determine the identity of compounds driving summiting behavior. However, our analysis did serve to inform which compounds may play a role in summiting by virtue of their overabundance. While we do not yet know the structure of these compounds, their consistent detection in our samples and our new knowledge of their molecular weight with very high accuracy means that these are prime candidates to isolate, purify and functionally test moving forward.

Reviewer #3 (Public Review):

The fungus Entomophthora muscae infects flies and in turn manipulates the flies to produce a summiting behavior that is believed to enhance spore dispersal that happens upon the eventual death of the fly. In this study, the authors undertake a Herculean effort to identify the neural pathways that are manipulated by the fungus to cause summiting. In a major advance, the authors develop techniques that allow them to track behaviors of infected flies over the course of several days. This allows them to investigate summiting behaviors that occur just prior to death with unprecedented detail. In their analysis, the authors find that summiting flies show a burst of increased locomotion just prior to death. Importantly, they show that this burst of locomotion is not seen in flies that are dying from other causes (starvation or desiccation). The burst of locomotion is also found to coincide with an increase in elevation that occurs with summiting, but other results indicate that a change in elevation may be an indirect consequence of increased locomotion. With this new knowledge in hand, the authors screen for genes and neuronal pathways that either disrupt or enhance the burst of locomotion that is characteristic of summiting. These experiments clearly indicate that neurons and genes controlling circadian rhythms play a major role in summiting behaviors. The authors focus their attention on a particular subset of clock neurons (DN1p) as potentially mediating summiting behavior. It is worth noting that DN1p neurons have been implicated in a variety (and in some cases contradictory) of circadian processes and that the interpretation of manipulations of these neurons may be an oversimplification. In particular, prior studies have implicated these cells in temperature entrainment/compensation so interpreting thermogenetic manipulations of these cells might be complicated. The authors also zoom in on a specific region of the brain containing neurons of the pars intercebralis, since they find infiltration by the fungus in this region and the effects of drivers targeting the PI. Converging and convincing lines of evidence to suggest that the PI neurons output to the corpora allata and effects of summing may be mediated by the CA. The already impressive series of experiments are further clinched by the development of a machine vision-based classifier that allows the authors to automatically identify summiting flies so that they may be collected for metabolomic analyses. The authors are automatically emailed and seemingly roused themselves in the middle of the night in order to obtain the precious flies they needed. They find a bunch of compounds that appear in summiting flies and even inject hemolymph from the infected animals into naive flies to find that circulating compounds can affect behaviors. Overall, this paper is a tour de force that addresses a system of long-standing interest and brings it into the modern age. Many new questions are now raised for the future by this fascinating study.

Thank you for your gracious summary of our work and for recognizing the multifaceted approach we have taken to begin to understand the mechanistic basis of summiting. We agree that there are many new questions raised by this work and hope to address them in future publications.

Author Response

Reviewer #1 (Public Review):

The manuscript by Gochman and colleagues reports the discovery of a very strong sensitization of TRPV2 channels by the herbal compound cannabidiol (CBD) to activation by the synthetic agonist 2aminoethoxydiphenyl borate (2-APB). Using patch-clamp electrophysiology the authors show that the ~100-fold enhancement by micromolar CBD of TRPV2 current responses to low concentrations of 2-APB reflects a robust increase in apparent affinity for the latter agonist. Cryo-EM structures of TRPV2 in lipid nanodiscs in the presence of both drugs report two-channel conformations. One conformation resembles previously solved structures whereas the second conformation reveals two distinct CBD binding sites per subunit, as well as changes in the conformation of the S4-S5 linker. Interestingly, although TRPV1 and TRPV3 are highly homologous to TRPV2 and both CBD binding sites are relatively conserved, the CBD-induced sensitization towards 2-APB is observable only for TRPV3 but not for TRPV1. Moreover, the simultaneous substitution of non-conserved residues in the CBD binding sites and the pore region of TRPV1 with the amino acids present in TRPV2 fails to confirm strong CBD-induced sensitization. The authors conclude that CBD-dependent sensitization of TRPV2 channels depends on structural features of the channel that are not restricted to the CBD binding site but involve multiple channel regions.

These are important findings that promote our understanding of the molecular mechanisms of TRPV family channels, and the data provide convincing evidence for the conclusions.

We appreciate the supportive evaluation of the reviewer.

Reviewer #2 (Public Review):

In this manuscript, Gochman et al. studied the molecular mechanism by which cannabidiol (CBD) sensitizes the TRPV2 channel to activation by 2-APB. While CBD itself can activate TRPV2 with low efficacy, it can sensitize TRPV2 current activated by 2-APB by two orders of magnitude. The authors showed, via single-channel recording, that the CBD-dependent sensitization arises from an increase in Po when the channel binds to both CBD and 2-APB. The authors then used cryo-EM to investigate how CBD binds to TRPV2 and identified two CBD binding sites in each subunit, with one site being previously reported and the other being newly discovered.

TRPV1 and TRPV2 are two channels closely related to TRPV2. All three channels can be activated by CBD and 2-APB, but only TRPV2 and 3 are strongly sensitized by CBD. To understand the molecular basis of the different sensitivity to CBD, the authors compared the residues within the CBD binding sites and generated mutants by swapping non-conserved residues between TRPV1 and TRPV2. They then performed patch-clamp recordings on these mutants and found that mutations on non-conserved residues indeed influenced the CBD-dependent sensitization, thereby supporting the observed CBD binding sites.

Unexpectedly, the authors did not identify the binding site of 2-APB, despite its robust effect in electrophysiology recordings, especially when combined with CBD. Although previous structural studies of TRPV2 have reported 2-APB binding sites, the associated densities in these studies were not wellresolved. Therefore, the authors called on the field to re-examine published structural data with regard to the 2-APB binding sites.

Overall, this is an important study with well-designed and well-conducted experiments.

We appreciated the supportive comments of the reviewer.

Reviewer #3 (Public Review):

In this paper, Gochman et al examine TRPV1-3 channel sensitization by CBD, specifically in the context of 2-APB activation. The authors primarily used classic electrophysiological techniques to address their questions about channel behavior but have also used structural biology in the form of cryo-EM to examine drug binding to TRPV2. The authors have carefully observed and quantified sensitization of the rat TRPV2 channel to 2-APB by CBD. While this sensitization has been reported previously (Pumroy et al, Nat Commun 2022), the authors have gone into much more detail here and carefully examined this process from several angles, including a comparison to some other known methods of sensitizing TRPV2. Additionally, the authors have also revealed that CBD sensitizes rat TRPV1 and mouse TRPV3 to 2-APB, which has not been reported previously. Up to this point, the work is well thought through and cohesive.

The major weakness of this paper is that the authors' efforts to track down the structural and molecular basis for CBD sensitization neither give insight into how sensitization occurs nor provide a solid footing for future work on the topic. The structural work presented in this paper lacks proper controls to interpret the observed states and the authors do nothing to follow up on a potentially interesting second binding site for CBD. Overall, the structural work feels detached from the rest of the paper. The mutations chosen to examine sensitization are based on setting up TRPV1 in opposition to TRPV2 and TRPV3, which makes little sense as all three channels show sensitization by CBD, even if to different extents. The authors chose their mutations based on the assumption that response to CBD is the key difference between the channels for sensitization, yet the overall state of each channel or the different modes of activation by 2-APB seem to be more likely candidates. As a result, it is not particularly surprising that none of the mutations the authors make reduce CBD sensitization in TRPV2 or increase CBD sensitization in TRPV1.

A difficulty in examining TRPV1-3 as a group is that while they are highly conserved in sequence and structure, there are key differences in drug responses. While it does seem likely that CBD would bind to the same location in TRPV1-3, there is extensive evidence that 2-APB binds at different sites in each channel, as the authors discuss in the paper. Without more basic information about where 2-APB binds to each channel and confirmation that CBD does indeed bind TRPV1-3 at the same site, it may not be possible to untangle this particular mode of channel sensitization.

We appreciate this reviewer’s perspective and we too were disappointed that our approach did not yield more definitive answers to why some TRPV channels are more sensitive to CBD. We have revised the results and discussion sections to more clearly articulate what we think our results reveal. We have also added a section to the discussion to present the idea that the differential sensitivity of TRPV channels to CBD may have more to do with where 2-APB binds and how it activates the channel than CBD. These challenging points are all excellent and they have helped us to present our message more clearly.

Author Response

Reviewer #1 (Public Review):

In this study, authors examine immune signatures from patients that experienced mild, moderate, or severe COVID-19 symptoms and followed them for months to evaluate whether there was a correlation between their immune activation phenotypes, disease severity, and long COVID. Authors observed higher T cell activation/proliferation marker expression in blood samples of patients with severe disease whereas other cell types were more or less unchanged. The authors also examined the cytokine profile of the patient's serum samples to determine the potential drivers of T cell activation phenotypes. Authors then perform T-cell responses to viral peptides to determine the differences in activation phenotypes with disease severity.

The major strengths of the paper appear in the evaluation of the appropriate cohort of human samples and following them over a period of months. Additionally, the authors perform detailed T-cell analysis in an unbiased way to determine any possible activation correlations with disease severity. The authors also perform antigen-specific T-cell analysis via peptide stimulation which adds to the overall findings. However, there are a number of drawbacks that need to be mentioned. Firstly, the phenotypes of T cells prior to the 3-month time-point are not known. Hence, there is no information on baseline or during the early phase of infection. Secondly, the response is largely obtained from blood. How much information about T cells in blood correlate with lung disease is a matter of concern. Analysis of lungs, where actual disease manifestation is ideal, however close to impossible in the human cohort. Alternatively, analysis of local lymph node aspirate or nasal swabs could be useful. Thirdly, the claim that bystander T cell activation plays a role seems loose, specifically the IL-15 in vitro data. Moreover, the analysis of T cells seems very focused on activation/proliferation phenotypes. Alternative T cell phenotypes such as regulatory, IL-10 producing, or FoxP3 expression are not extensively analyzed.

Major points

1) In Figure 1, the CD4 T cell activation phenotypes do not seem consistent across the groups. Why does moderate vs. severe show increases in CXCR3 expression but not mild vs. severe? The same goes for other markers. Performing T cell stimulation with class II peptides specific for CoV-2 and looking at IFN etc. to determine antigen-specific T cells and then gating on these activation/proliferation markers may be a better way to observe differences.

Figure 1 shows activation phenotypes of total CD4+ T-cells. We performed similar analysis on SARS-C0V-2 spike-specific CD4+ T cells as suggested by Reviewer 1 (using 15-mer peptides overlapping by 10 amino acids which are able to stimulate both CD4+ and CD8+ T cells- see Figure 5), but we did not observe differences between the groups (data not shown). Importantly, as reported in the discussion (page 18 from “Our data does not support the persistence of SARS-CoV-2 antigens at 3 months….”) we did not observe significant activation of spike-specific CD4+ or CD8+ T cells which suggests that T cell activation in these patients at 3 months is not driven by persistence of SARS-CoV-2 spike antigens.

2) One major drawback is the control patients. It would have helped to include a batch of samples from uninfected patients. Or to have the plasma/blood from patients before COVID-19 symptoms. This way there is a baseline for each group that could be compared. It is difficult to draw broad conclusions across the group at 3 months if we do not know their baseline phenotypes.

We did not have access to blood samples from these patients prior to COVID-19 infection. However, we have now added an analysis of matched samples from the same patients at 12 months post infection (N=33, see Figure 2- figure supplement 1 and also response to Reviewer 2). These data show a significant decrease in T cell activation at 12 months compared to 3 months. T cell activation has decreased to largely undetectable “baseline” levels at 12 months, that are similar between patients who had experienced mild, moderate or severe COVID-19. This lack of T cell activation at 12 months likely reflects the T cell profiles that patients will have had prior to COVID-19 infection.

3) Although the authors focused on activating/proliferating markers to correlate with disease severity, this analysis does not consider alternate T cell phenotypes such as the ones with regulatory or anti-inflammatory phenotypes. Did authors detect differences in T cells with regulatory profiles such as expression of IL-10, FoxP3, etc. in their unsupervised UMAP analysis or otherwise flow experiments?

Due to limited blood volumes we were unable to analyse regulatory/anti-inflammatory T cells phenotypes. Our serum cytokine data does not suggest statistically significant differences in serum IL-10 levels in patients with mild, moderate or severe disease. However, it is possible that we may have missed differences in FoxP3+ regulatory or IL-10- producing T cells.

Reviewer #2 (Public Review):

The manuscript is well written, the data are based on well-performed experiments, and the conclusions are supported by the data. The authors study thoroughly the global phenotype of T and NK cells and also analyze antigen-specific T cell frequencies. The data confirm that individuals who had severe COVID-19 disease (required ventilation and/or ITU admission) have slightly more activated CD4 and CD8 T cells at 3 months post-infection and report more frequently long COVID symptoms, yet the novelty of this manuscript is to show that these two are not linked to each other. Moreover, the manuscript confirms that patients across all disease severities mount and maintain memory T cell and antibody responses to SARS-CoV-2.

The authors find that patients who recovered from severe COVID-19 3 months ago have more activated CD4+ and CD8+ T cells than patients who recovered from the mild disease. Although the difference is significant, the frequency of CD4+ T cells with an activated phenotype is increased only by about 2-fold (~2% vs ~1%), while the frequency of activated CD8+ T cells is about 6% vs 4%, which should be added to the results to better describe the extent of the activation.

As the authors mention in the discussion, it cannot be excluded that the more activated T cell phenotype in patients who recovered from severe COVID-19 is not rather a consequence of the increased comorbidities associated with this group. However, their Luminex analysis of the serum shows that the levels of cytokines TNF-a, IL-4, IL-12, IL-15, and IL-17A decline by 8 and 12 months, suggesting that the immune activation by 3 months is most likely a consequence of the previous severe viral infection.

To strengthen this point, PBMC is probably not available at a later time point, to see if the increased T cell activation decreases in line with the serum cytokines. Yet, the authors should at least try to repeat the experiments of coculturing CD3+ T cells from healthy volunteers with the serum of mild/severe patients at 8-12 months post-recovery (Fig. 3 D-E).

Thank you for these suggestions. We had access to PBMCs from N=33 matched patients at 12 months post admission and have now performed analyses of these samples. Our results show that CD4+ and CD8+ T cell activation at 12 months is significantly decreased compared to that observed at 3 months (Figure 2- figure supplement 1). We show that the frequencies of Ki67+ CD38+ CD4+ and CD8+ T cells are significantly decreased at 12 compared to the 3-month time point. Similarly, the frequencies of CXCR3+ CD4+ and CD8+ T cells are strongly decreased at 12 compared to 3 months post admission. Activated HLA-DR+ CD38+ and granzyme B+ CD8+ T cells are also significantly decreased from 3 to 12 months post admission. Unsupervised UMAP analyses shows that the cell distribution and density of CD4+ and CD8+ T cell populations was similar across all severity groups at 12 months post infection, while major differences are observed at 3 months between the patient groups (Figure 2- figure supplement 1 K). We added this information in the manuscript at page 9 (see tracked changes) and in Figure 2- figure supplement 1.

Thank you for suggesting we repeat the co-culture experiments of healthy donor PBMCs with serum of mild and severe patients at 12 months post admission. We co-cultured healthy PBMCS from the same donors (only 3 out of the 4 donors used for the 3 months experiment were still accessible) with serum from mild and severe patients at 12 months. Notably, we observed that IL-15R upregulation did not occur upon co-culture of healthy donor PBMCs with plasma from severe patients at 12 months. This suggests that factors inducing IL-15R upregulation present in the 3 months plasma may be absent in the 12 month plasma. We have added this new data to the manuscript (Figure 3E)

The authors tried to find if the activated T cell phenotype or increased serum cytokines at 3 months post-infection is linked with increased long COVID symptoms. The study does not find any direct association when the data are adjusted for age, sex, and severity. This is the only novelty of this study, yet it is an important piece of information in the attempt to broaden our understanding of the underlying causes of long COVID symptoms.

Overall, it would be important to understand if increased frequencies of T cell activation (~2-fold) and increased levels of serum cytokines at 3 months following severe COVID-19 that resulted in ventilation and/or ITU admission is specific to severe SARS-CoV-2 infection, or if similar consequences are resulting also from other severe acute viral infections. Addressing this question is beyond the scope of the manuscript, yet it should be discussed.

We agree this is an important question. In H7N9 influenza infection persistent T cell activation was associated with fatal disease while T cell activation early during infection associated with positive clinical outcomes (Wang et al. Nature Comms 2018). Aging is also known to alter T cell function and the persisting low-grade inflammation present in elderly individuals may also facilitate the persistence of bystander activated T cells (Yunis et al Trends in Microbiology 2023). We have added these considerations in the discussion (page 18-19).

Reviewer #3 (Public Review):

In this paper, the authors used a cohort study to link immune signatures in blood 30 days after COVID-19 infection as possible predictors of prolonged symptomatology. The paper partially achieves its aims. While the selected analyses are comprehensive, the cohort design is appropriate and the mechanistic ex vivo work is clever and convincing, the strength of conclusions is somewhat limited by the selection of imprecise clinical endpoints, and the lack of analyses examining T regulatory signatures.

Strengths of the paper are:

• The paper includes a comprehensive and structured immune analysis.

• The paper is extremely clearly written.

• The use of manual gating and unsupervised analysis in Fig 1 is complementary and helpful.

• Bystander T cell experiments with IL-15 are useful and attempt to explore mechanisms from human samples which are traditionally very challenging.

• The experiments shown in Figure 4 documenting equal Cov2 T cell responses in all 3 cohorts are an extremely important result.

Major concerns are:

• The significance of the study is somewhat limited by the small sample size.

• The symptomatic outcome scale for PASC is blunt and poorly captures severity. More state-of-the-start scales of symptomatic severity and heterogeneity exist for PASC. I suggest this and other papers as an example: https://pubmed.ncbi.nlm.nih.gov/36454631/

Thank you for this suggestion. Additional outcome measures have now been included in our analysis, refer to the results section and the updated Figure 6- source data 1 A-B.

• The omission of analyses examining T regulatory functions is a missed opportunity and these may be impaired in this population.

We have acknowledged this as a limitation of the study.

• This is a challenging question that can be applied to many exploratory studies of this nature: how can we rule out the possibility that statistically significant differences in Figs 1, 2 & 3 are statistically significant but biologically meaningless? All cellular and cytokine measures of immune responses shown in these figures are not routinely measured in the clinic. Are there studies that can be cited to show that these differences are sufficient to have a causal impact on prolonged symptoms and tissue damage rather than just correlations with these outcomes?

This is a challenging question, and we were unable to find studies correlating these measurements with tissue damage and prolonged symptoms. In our study we however suggest that prolonged T cell activation is not related to ongoing long-COVID symptoms.

Author Response

Reviewer #1 (Public Review):

This study presents an important finding on human m6A methyltransferase complex (including METTL3, METTL14 and WTAP). The evidence supporting the claims of the authors is convincing, although the model and assays need to be further modified. The work will be of interest to biologists working on RNA epigenetics and cancer biology.

In mammals, a large methyltransferase complex (including METTL3, METTL14 and WTAP) deposits m6A across the transcriptome, and METTL3 serves as its catalytic core component. In this manuscript, the authors identified two cleaved forms of METTL3 and described the function of METTL3a (residues 239-580) in breast tumorigenesis. METTL3a mediates the assembly of METTL3-METTL14-WTAP complex, the global m6A deposition and breast cancer progression. Furthermore, the METTL3a-mTOR axis was uncovered to mediate the METTL3 cleavage, providing potential therapeutic target for breast cancer. This study is properly performed and the findings are very interesting; however, some problems with the model and assays need to be modified. It is widely known that METTL3 and METTL14 form a stable heterodimer with the stoichiometric ratio of 1:1 (Wang X et al. Nature 534, 575-578 (2016), Su S et al. Cell Res 32(11), 982-994 (2022), Yan X et al. Cell Res 32(12), 1124-1127 (2022)), the numbers of METTL3 and METTL14 in the model of Fig 7P are not equivalent and need to be modified.

We thank for reviewer’s good suggestion. We will modify the model in Fig. 7P.

Reviewer #2 (Public Review):

In this study, Yan et al. report that a cleaved form of METTL3 (termed METTL3a) plays an essential role in regulating the assembly of the METTL3-METTL14-WTAP complex. Depletion of METTL3a leads to reduced m6A level on TMEM127, an mTOR repressor, and subsequently decreased breast cancer cell proliferation. Mechanistically, METTL3a is generated via 26S proteasome in an mTOR-dependent manner.

The manuscript follows a smooth, logical flow from one result to the next, and most of the results are clearly presented. Specifically, the molecular interaction assays are well-designed. If true, this model represents a significant addition to the current understanding of m6A-methyltransferase complex formation.

A few minor issues detailed below should be addressed to make the paper even more robust. The specific comments are contained below.

1) The existence of METTL3a and METTL3b.

In this study, the author found the cleaved form of METTL3 in breast cancer patient tissues and breast cancer cell lines. Is it a specific event that only occurs in breast cancer? The author may examine the METTL3a in other cell lines if it is a common rule.

We thank reviewer for point this out. We discovered the cleaved form of METTL3 in breast cancer, and we also observed this cleaved METTL3 in other cell lines such as lung cancer cell lines, renal cancer cell lines, HCT116 and MEF, suggesting that it is a common rule. We will add these results in the revised manuscript.

2) Generation of METTL3a and METTL3b.

1) Figure 1 shows that METTL3a and METTL3b were generated from the C-terminal of full-length METTL3. Because the sequence of METTL3a is involved in the sequences of METTL3b, can METTL3b be further cleaved to produce METTL3a?

Although the sequence of METTL3a is involved in the sequences of METTL3b, overexpression of METTL3b in T47D, MDA-MB-231 and 293T cells did not show METTL3a expression (please see Figures 3A, 3C, 3G), suggesting that METTL3b can not be further cleaved to produce METTL3a, and the METTL3 cleavage may require its N-terminal region. We will add this in the discussion.

2) Based on current data, the generation of METTL3a and METTL3b are separated. Are there any factors that affect the cleavage ratio between METTL3a and METTL3b?

We thank for reviewer’s excellent question. In this study, we show that both METTL3a and METTLb are produced through proteasomal cleavage, and both of them are positively regulated by the mTOR pathway. On the other hand, we indeed observed the differential cleavage ratios between METTL3a and METTL3b across different cell lines. For example, METTL3a/METTLb ratio was greater than 1 in MDA-MB-231 cells (see Figure 7C), less than 1 in T47D and 293T cell lines (see Figure 7A and 7B), and equal to 1 in MEF cells (see Figure 7O). Based on these results, we speculate that there may be some factors that control the cleavage ratio between METTL3a and METTL3b, which warrants further investigation. We will add this in the discussion.

3) In Figure 2G, the author shows the result that incubation of the Δ198+Δ238 METTL3 protein with T47D cell lysates cannot produce the METTL3a and METTL3b variants. The author may also show the results that Δ198 METTL3 protein or Δ238 METTL3 protein incubates with T47D cell lysates, respectively.

Following the reviewer’s suggestion, we will perform in vitro cleavage assays by incubation of METTL3-Δ238 or METTL3-Δ198 with T47D cell lysates, and will incorporate this result in the revised manuscript.

4) As well as many results published in previous studies, the in vitro methylation assay shows that WT METTL3 is capable of methylating RNA probe (figure 2H). The main point of this study is that METTL3a is required for the METTL3-METTL14 assembly. However, the absence of METTL3a in the in vitro system did not inhibit METTL3-METTL14 methylation activity. Moreover, the presence of METTL3a even resulted in a weak m6A level.

The main point of this study is that METTL3a is required for the METTL3-WTAP interaction, but dispensable for the METTL3-METTL14 assembly (see Figure 4A-4B). In this in vitro methylation assays, METTL3 and METTL14 is capable of methylating RNA probe in the absent of WTAP. In this condition, we found that METTL3 WT as well as its different variants (METTL3-Δ238, METTL3-Δ198, METTL3b and METTL3a) except the catalytically dead mutant METTL3 APPA showed methylation activity in vitro.

5) In Figure 4A, the author suggests that WTAP cannot be immunoprecipitated with METTL3a and 3b because WTAP interacted with the N-terminal of METTL3. If this assay is performed in WT cells, the endogenous full-length METTL3 may help to form the complex. In this case, WTAP is supposed to be co-immunoprecipitated.

We thank reviewer for point this out. METTL3 interacts with WTAP through its N-terminal (1-33aa) (1). Consistently, we find that the two cleaved forms METTL3a and METTL3b which lack the N-terminal region are not able to bind with WTAP. In Figure 4A, we overexpressed METTL3 WT as well as its different variants METTL3-Δ238, METTL3-Δ198, METTL3b and METTL3a respectively in WT cells, and compared their binding abilities with WTAP or METTL14 among these overexpressed METTL3 variants. We acknowledge that the exogenous METTL3a and METTL3b interact with endogenous full-length METTL3, and the endogenous full-length METTL3 may help them to form the complex with WTAP. But it is also noteworthy that the exogenous expression levels of METTL3a and METTL3b are much higher than that of endogenous full-length METTL3 (see Figure 3A and 3C). In this case, METTL3a or METTL3b predominantly interacts with itself, METTL3, METTL14 or other potential interacting proteins through its C-terminal region, this may greatly dilute the condition for the interaction between WTAP and endogenous full-length METTL3. Moreover, in Figure 4A, the comparison is among overexpressed METTL3 variants, this week indirect interaction through much lower expression levels of endogenous protein is not comparable to the direct interaction between the overexpressed METTL3 variant and WTAP.

Reference:

1. Schöller, E., Weichmann, F., Treiber, T., Ringle, S., Treiber, N., Flatley, A., Feederle, R., Bruckmann, A., and Meister, G. (2018). Interactions, localization, and phosphorylation of the m6A generating METTL3–METTL14–WTAP complex. RNA 24, 499-512.

Author Response

Reviewer #1 (Public Review):

The authors present a study of visuo-motor coupling primarily using wide-field calcium imaging to measure activity across the dorsal visual cortex. They used different mouse lines or systemically injected viral vectors to allow imaging of calcium activity from specific cell-types with a particular focus on a mouse-line that expresses GCaMP in layer 5 IT (intratelencephalic) neurons. They examined the question of how the neural response to predictable visual input, as a consequence of self-motion, differed from responses to unpredictable input. They identify layer 5 IT cells as having a different response pattern to other cell-types/layers in that they show differences in their response to closed-loop (i.e. predictable) vs open-loop (i.e. unpredictable) stimulation whereas other cell-types showed similar activity patterns between these two conditions. They analyze the latencies of responses to visuomotor prediction errors obtained by briefly pausing the display while the mouse is running, causing a negative prediction error, or by presenting an unpredicted visual input causing a positive prediction error. They suggest that neural responses related to these prediction errors originate in V1, however, I would caution against over-interpretation of this finding as judging the latency of slow calcium responses in wide-field signals is very challenging and this result was not statistically compared between areas.

Surprisingly, they find that presentation of a visual grating actually decreases the responses of L5 IT cells in V1. They interpret their results within a predictive coding framework that the last author has previously proposed. The response pattern of the L5 IT cells leads them to propose that these cells may act as 'internal representation' neurons that carry a representation of the brain's model of its environment. Though this is rather speculative. They subsequently examine the responses of these cells to anti-psychotic drugs (e.g. clozapine) with the reasoning that a leading theory of schizophrenia is a disturbance of the brain's internal model and/or a failure to correctly predict the sensory consequences of self-movement. They find that anti-psychotic drugs strongly enhance responses of L5 IT cells to locomotion while having little effect on other cell-types. Finally, they suggest that anti-psychotics reduce long-range correlations between (predominantly) L5 cells and reduce the propagation of prediction errors to higher visual areas and suggest this may be a mechanism by which these drugs reduce hallucinations/psychosis.

This is a large study containing a screening of many mouse-lines/expression profiles using wide-field calcium imaging. Wide-field imaging has its caveats, including a broad point-spread function of the signal and susceptibility to hemodynamic artifacts, which can make interpretation of results difficult. The authors acknowledge these problems and directly address the hemodynamic occlusion problem. It was reassuring to see supplementary 2-photon imaging of soma to complement this data-set, even though this is rather briefly described in the paper.

We will expand on the discussion of caveats as suggested.

Overall the paper's strengths are its identification of a very different response profile in the L5 IT cells compared other layers/cell-types which suggests an important role for these cells in handling integration of self-motion generated sensory predictions with sensory input. The interpretation of the responses to anti-psychotic drugs is more speculative but the result appears robust and provides an interesting basis for further studies of this effect with more specific recording techniques and possibly behavioral measures.

Reviewer #2 (Public Review):

Summary:

This work investigates the effects of various antipsychotic drugs on cortical responses during visuomotor integration. Using wide-field calcium imaging in a virtual reality setup, the researchers compare neuronal responses to self-generated movement during locomotion-congruent (closed loop) or locomotion-incongruent (open loop) visual stimulation. Moreover, they probe responses to unexpected visual events (halt of visual flow, sudden-onset drifting grating). The researchers find that, in contrast to a variety of excitatory and inhibitory cell types, genetically defined layer 5 excitatory neurons distinguish between the closed and the open loop condition and exhibit activity patterns in visual cortex in response to unexpected events, consistent with unsigned prediction error coding. Motivated by the idea that prediction error coding is aberrant in psychosis, the authors then inject the antipsychotic drug clozapine, and observe that this intervention specifically affects closed loop responses of layer 5 excitatory neurons, blunting the distinction between the open and closed loop conditions. Clozapine also leads to a decrease in long-range correlations between L5 activity in different brain regions, and similar effects are observed for two other antipsychotics, aripripazole and haloperidol, but not for the stimulant amphetamine. The authors suggest that altered prediction error coding in layer 5 excitatory neurons due to reduced long-range correlations in L5 neurons might be a major effect of antipsychotic drugs and speculate that this might serve as a new biomarker for drug development.

Strengths:

• Relevant and interesting research question:

The distinction between expected and unexpected stimuli is blunted in psychosis but the neural mechanisms remain unclear. Therefore, it is critical to understand whether and how antipsychotic drugs used to treat psychosis affect cortical responses to expected and unexpected stimuli. This study provides important insights into this question by identifying a specific cortical cell type and long-range interactions as potential targets. The authors identify layer 5 excitatory neurons as a site where functional effects of antipsychotic drugs manifest. This is particularly interesting as these deep layer neurons have been proposed to play a crucial role in computing the integration of predictions, which is thought to be disrupted in psychosis. This work therefore has the potential to guide future investigations on psychosis and predictive coding towards these layer 5 neurons, and ultimately improve our understanding of the neural basis of psychotic symptoms.

• Broad investigation of different cell types and cortical regions:

One of the major strengths of this study is quasi-systematic approach towards cell types and cortical regions. By analysing a wide range of genetically defined excitatory and inhibitory cell types, the authors were able to identify layer 5 excitatory neurons as exhibiting the strongest responses to unexpected vs. expected stimuli and being the most affected by antipsychotic drugs. Hence, this quasi-systematic approach provides valuable insights into the functional effects of antipsychotic drugs on the brain, and can guide future investigations towards the mechanisms by which these medications affect cortical neurons.

• Bridging theory with experiments

Another strength of this study is its theoretical framework, which is grounded in the predictive coding theory. The authors use this theory as a guiding principle to motivate their experimental approach connecting visual responses in different layers with psychosis and antipsychotic drugs. This integration of theory and experimentation is a powerful approach to tie together the various findings the authors present and to contribute to the development of a coherent model of how the brain processes visual information both in health and in disease.

Weaknesses:

• Unclear relevance for psychosis research

From the study, it remains unclear whether the findings might indeed be able to normalise altered predictive coding in psychosis. Psychosis is characterised by a blunted distinction between predicted and unpredicted stimuli. The results of this study indicate that antipsychotic drugs further blunt the distinction between predicted and unpredicted stimuli, which would suggest that antipsychotic drugs would deteriorate rather than ameliorate the predictive coding deficit found in psychosis. However, these findings were based on observations in wild-type mice at baseline. Given that antipsychotics are thought to have little effects in health but potent antipsychotic effects in psychosis, it seems possible that the presented results might be different in a condition modelling a psychotic state, for example after a dopamine-agonistic or a NMDA-antagonistic challenge. Therefore, future work in models of psychotic states is needed to further investigate the translational relevance of these findings.

We fully agree that it is unclear how the effects of antipsychotics in mice relate to the drug effects that would be observed in schizophrenic patients. It is also correct that the reduction of the difference between closed and open loop locomotion onset response in L5 IT neurons (Figure 4) is not what we would have expected to find under the assumption that psychosis is characterized by a blunted distinction between predicted and unpredicted stimuli. We are not sure how to interpret this finding. However, it is probably important to note that the difference is only reduced when using a normalized comparison. Looking just at the subtraction of the two curves, the difference between closed and open loop locomotion onset responses remains unchanged before and after antipsychotic drug injection. The finding of a decorrelation of layer 5 activity, however, is easier to interpret under the assumption that layer 5 functions as an internal representation. If speech hallucinations, for example, are the consequence of a spurious activation of internal representations in speech processing areas of cortex, then antipsychotics might reduce the probability of these spurious activation events by reducing the lateral influence between layer 5 neurons in different cortical areas.

We do indeed plan to address the question of how antipsychotics influence cortical processing in mouse models of schizophrenia in the future.

• Incomplete testing of predictive coding interpretation

While the investigation of neuronal responses to different visual flow stimuli Is interesting, it remains open whether these responses indeed reflect internal representations in the framework of predictive coding. While the responses are consistent with internal representation as defined by the researchers, i.e., unsigned prediction error signals, an alternative interpretation might be that responses simply reflect sensory bottom-up signals that are more related to some low-level stimulus characteristics than to prediction errors.

This is correct – we will expand on the discussion of this point in the manuscript.

Moreover, This interpretational uncertainty is compounded by the fact that the used experimental paradigms were not suited to test whether behaviour is impacted as a function of the visual stimulation which makes it difficult to assess what the internal representation of the animal actual was. For these reasons, the observed effects might reflect simple bottom-up sensory processing alterations and not necessarily have any functional consequences. While this potential alternative explanation does not detract from the value of the study, future work would be needed to explain the effect of antipsychotic drugs on responses to visual flow. For example, experimental designs that systematically vary the predictive strength of coupled events or that include a behavioural readout might be more suited to draw from conclusions about whether antipsychotic drugs indeed alter internal representations.

We agree that much additional work will be necessary to identify internal representation neurons. However, it is difficult to envision how behavioral output could be used to make inferences about internal representations in sensory areas of cortex. In humans, for example, there is evidence that internal representations in visual cortex and behavioral output are not always directly related: binocular rivalry activates representations of both stimuli shown in visual cortex, while the conscious experience that drives behavioral output is only of one of the two stimuli. Hence, we would assume that the internal representation in visual cortex does not necessarily relate to behavioral output.

• Methodological constraints of experimental design

While the study findings provide valuable insights into the potential effects of antipsychotic drugs, it is important to acknowledge that there may be some methodological constraints that could impact the interpretation of the results. More specifically, the experimental design does not include a negative control condition or different doses. These conditions would help to ensure that the observed effects are not due to unspecific effects related to injection-induced stress or time, and not confined to a narrow dose range that might or might not reflect therapeutic doses used in humans. Hence, future work is needed to confirm that the observed effects indeed represent specific drug effects that are relevant to antipsychotic action.

We agree that both dosages and a broader spectrum of non-antipsychotic compounds will need to be investigated. We are in the process of building a screening pipeline to perform exactly these types of experiments. We would however argue that the paper already includes a control condition in the form of the amphetamine data (Figure 7). While it is possible that amphetamine might have an effect that exactly cancels out potential i.p. injection- or stress-induced changes, we would argue it is more probable that these changes had no measurable effect on Tlx3 positive L5 IT neuron calcium activity per se. We will provide additional evidence that time or injection stress alone do not result in the observed effects.

Conclusion:

Overall, the results support the idea that antipsychotic drugs affect neural responses to predicted and unpredicted stimuli in deep layers of cortex. Although some future work is required to establish whether this observation can indeed be explained by a drug-specific effect on predictive coding, the study provides important insights into the neural underpinnings of visual processing and antipsychotic drugs, which is expected to guide future investigations on the predictive coding hypothesis of psychosis. This will be of broad interest to neuroscientists working on predictive coding in health and in disease.

Reviewer #3 (Public Review):

The study examines how different cell types in various regions of the mouse dorsal cortex respond to visuomotor integration and how antipsychotic drugs impacts these responses. Specifically, in contrast to most cell types, the authors found that activity in Layer 5 intratelencephalic neurons (Tlx3+) and Layer 6 neurons (Ntsr1+) differentiated between open loop and closed loop visuomotor conditions. Focussing on Layer 5 neurons, they found that the activity of these neurons also differentiated between negative and positive prediction errors during visuomotor integration. The authors further demonstrated that the antipsychotic drugs reduced the correlation of Layer 5 neuronal activity across regions of the cortex, and impaired the propagation of visuomotor mismatch responses (specifically, negative prediction errors) across Layer 5 neurons of the cortex, suggesting a decoupling of long-range cortical interactions.

The data when taken as a whole demonstrate that visuomotor integration in deeper cortical layers is different than in superficial layers and is more susceptible to disruption by antipsychotics. Whilst it is already known that deep layers integrate information differently from superficial layers, this study provides more specific insight into these differences. Moreover, this study provides a first step into understanding the potential mechanism by which antipsychotics may exert their effect.

Whilst the paper has several strengths, the robustness of its conclusions is limited by its questionable statistical analyses. A summary of the paper's strengths and weaknesses follow.

Strengths:

The authors perform an extensive investigation of how different cortical cell types (including Layer 2/3, 4 , 5, and 6 excitatory neurons, as well as PV, VIP, and SST inhibitory interneurons) in different cortical areas (including primary and secondary visual areas as well as motor and premotor areas), respond to visuomotor integration. This investigation provides strong support to the idea that deep layer neurons are indeed unique in their computational properties. This large data set will be of considerable interest to neuroscientists interested in cortical processing.

The authors also provide several lines of evidence that visuomotor information is differentially integrated in deep vs. superficial layers. They show that this is true across experimental paradigms of visuomotor processing (open loop, closed loop, mismatch, drifting grating conditions) and experimental manipulations, with the demonstration that Layer 5 visuomotor integration is more sensitive to disruption by the antipsychotic drug clozapine, compared with cortex as a whole.

The study further uses multiple drugs (clozapine, aripiprazole and haloperidol) to bolster its conclusion that antipsychotic drugs disrupt correlated cortical activity in Layer 5 neurons, and further demonstrates that this disruption is specific to antipsychotics, as the psychostimulant amphetamine shows no such effect.

In widefield calcium imaging experiments, the authors effectively control for the impact of hemodynamic occlusions in their results, and try to minimize this impact using a crystal skull preparation, which performs better than traditional glass windows. Moreover, they examine key findings in widefield calcium imaging experiments with two-photon imaging.

Weaknesses:

A critical weakness of the paper is its statistical analysis. The study does not use mice as its independent unit for statistical comparisons but rather relies on other definitions, without appropriate justification, which results in an inflation of sample sizes.

We will expand on both analyses and justifications throughout.

For example, in Figure 1, independent samples are defined as locomotion onsets, leading to sample sizes of approx. 400-2000 despite only using 6 mice for the experiment. This is only justified if the data from locomotion onsets within a mouse is actually statistically independent, which the authors do not test for, and which seems unlikely. With such inflated sample sizes, it becomes more likely to find spurious differences between groups as significant. It also remains unclear how many locomotion onsets come from each mouse; the results could be dominated by a small subset of mice with the most locomotion onsets. The more disciplined approach to statistical analysis of the dataset is to average the data associated with locomotion onsets within a mouse, and then use the mouse as an independent unit for statistical comparison. A second example, for instance, is in Figure 2L, where the independent statistical unit is defined as cortical regions instead of mice, with the left and right hemispheres counting as independent samples; again this is not justified. Is the activity of cortical regions within a mouse and across cortical hemispheres really statistically independent? The problem is apparent throughout the manuscript and for each data set collected.

This may partially be a misunderstanding. Figures 1F-1K indeed use locomotion onsets as a unit, but there were no statistical comparisons. In these Figures we were addressing the question of whether locomotion onsets in closed loop differ from those in open loop. Thus, we quantify variability as a unit of locomotion onsets. The question of mouse-to-mouse variability of this analysis is a slightly different one. We did include the same analysis (for visual cortex) with the variability calculated across mice as Figure S2. We will expand this supplementary figure with the equivalent data of Figure 3 to further address this concern.

For Figure 1L (we assume the reviewer means Figure 1L, not Figure 2L), the unit we used for analysis was cortical area. We will update and improve the analysis. This was indeed not optimal, and we will replace the statistical testing with hierarchical bootstrap (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7906290/) to account for nested data.

An additional statistical issue is that it is unclear if the authors are correcting for the use of multiple statistical tests (as in for example Figure 1L and Figure 2B,D). In general, the use of statistics by the authors is not justified in the text.

We will update and improve the analysis shown in Figure 1L.

In Figures 2B and 2D, we think adding family-wise error correction would be slightly misleading. We could add a correction – our conclusions would remain unchanged almost independent of the choice of correction (most of the significant p values are infinitesimally small, see Table S1). However, our interpretation is not focusing on one particular comparison (of many possible comparisons) that is significant - all comparisons between closed and open loop data points were significant in the L5 IT recordings and none of them were significant in the recordings in C57BL/6 mice that expressed GCaMP brain-wide.

Finally, it is important to note that whilst the study demonstrates that antipsychotics may selectively impact visuomotor integration in L5 neurons, it does not show that this effect is necessary or sufficient for the action of antipsychotics; though this is likely beyond the scope of the study it is something for readers to keep in mind.

We fully agree, it is still unclear how the effects we observe in our work relate to the treatment relevant effects in patients. We will expand on this point in the discussion.

Author Response:

Reviewer #1 (Public Review):

[…] The manuscript contains a large amount of data that make a major inroad on a new type of link between telomere replication and regulation of the telomerase. Nevertheless, the detailed choreography of the events as well as the role of PCNASUMO remain elusive and the data do not fully explain the role of the Stn1/Elg1 interaction. The data presented do not sufficiently support the claim that SUMOPCNA is a positive signal for telomerase activation.

We thank the reviewer for her/his review efforts and opinion. We will resubmit a new version of the manuscript in which we will clarify some of the criticisms presented.

Reviewer #2 (Public Review):

[…] The conclusions are largely supported by experiments examining protein-protein interactions at low resolution and ambiguous regarding directness of interactions like co-IP and yeast two-hybrid (Y2H) combined with genetics. However, some results appear contradictory and there's a lack of rigor in the experimental data needed to support claims. There is significant room for improvement and this work could certainly attain the quality needed to support the claims. The current version needs substantial revision and lacks the necessary experimental detail. Stronger support for the claims would add detail to help distinguish competing models.

We thank the reviewer for her/his positive opinion. We will resubmit a new version of the manuscript in which we will clarify some of the criticisms presented by the referees, and add all the missing experimental details.

Reviewer #3 (Public Review):

This paper reveals interesting physical connections between Elg1 and CST proteins that suggest a model where Elg1-mediated PCNA unloading is linked to regulation of telomere length extension via Stn1, Cdc13, and presumably Ten1 proteins. Some of these interactions appear to be modulated by sumolyation and connected with Elg1's PCNA unloading activity. The strength of the paper is in the observations of new interactions between CST, Elg1, and PCNA. These interactions should be of interest to a broad audience interested in telomeres and DNA replication.

We thank the reviewer for her/his positive opinion. We will resubmit a new version of the manuscript in which we will clarify some of the criticisms presented.

What is not well demonstrated from the paper is the functional significance of the interactions described. The model presented by the authors is one interpretation of the data shown, and proposes that the role of sumolyation is temporally regulate the Elg1, PCNA and CST interactions at telomeres. This model makes some assumptions that are not demonstrated by this work (such as Stn1 sumolyation, as noted) and are left for future testing. Alternative models that envision sumolyation as a key in promoting spatial localization could also be proposed based on the data here (as mentioned in the discussion), in addition to or instead of a role for sumolyation in enforcing a series of switches governing a tightly sequenced series of interactions and events at telomeres. Critically, the telomere length data from the paper indicates that the proposed model depicts interactions that are not necessary for telomerase activation or inhibition, as telomeres in pol30-RR strains are normal length and telomeres in elg1∆ strains are not nearly as elongated as in stn1 strains. One possibility mentioned in the paper is the PCNAS and Elg1 interactions are contributing to the negative regulation of telomerase under certain conditions that are not defined in this work. Could it also be possible that the role of these interactions is not primarily directed toward modulating telomerase activity? It will be of interest to learn more about how these interactions and regulation by Sumo function intersect with regulation of telomere extension.

We present compelling evidence for a role of SUMOylated PCNA in telomere length regulation. Figure 1 shows that this modification is both necessary and sufficient to elongate the telomeres, indicating that PCNA SUMOylation plays a positive role in telomere elongation. The model we present is consistent with all our results. There are, of course, possible alternative models, but they usually fail to explain some of the results. We agree that the fact that pol30-RR presents normal-sized telomeres implies that SUMO-PCNA is not required for telomerase to solve the "end replication problem", but rather is needed for "sustained" activity of telomerase. Since elongated telomeres (by absence of Elg1 or by over-expression of SUMO-PCNA) was the phenotype monitored, this may require sustained telomerase activity. Similar results were seen in the past for Rnr1 (Maicher et al., 2017), and this mode depends on Mec1, rather than Tel1 (Harari and Kupiec, 2018). Telomere length regulation is complex, and we may not yet understand the whole picture. It appears that for normal “end replication problem” solution, very little telomerase activity may be needed, and spontaneous interactions at a low level may suffice. Future work may find the conditions at which telomerase switches from "end replication problem" to "sustained" activity. We will add further explanations on this subject to the Discussion section.

We suspect, but could not prove, a role for Stn1 SUMOylation in the interactions. SUMOylation is usually transient, and notoriously hard to detect, and despite the fact that many telomeric proteins are SUMOylated, Stn1 SUMOylation could not be shown directly by us and others (Hang et al, 2011).

Author Response

eLife assessment

This study provides valuable information on the biogenesis of eccDNAs during spermatogenesis, i.e., eccDNAs in spermatogenic cells are not derived from miotic recombination hotspots but represent oligonucleosomal DNA fragments from apoptotic male germ cells, whose ends are ligated through microhomology-mediated end-joining. The study is currently incomplete because the method of bioinformatics needs more details and data interpretation should take the amplification bias into consideration.

We highly appreciate the positive assessment.

The negative assessment of our bioinformatics method is probably based on Reviewer #2’ comemnts. While Reviewer #1 considered that “Results from sequencing data analysis were presented elegantly”, Reviewer #2 overlooked some details and raised several critiques regarding our bioinformatics method. We respectfully disagree with many of his or her critiques: (I) Reviewer #2 considered that our method was not fully described. However, we have illustrated the principle and steps of our eccDNA detection method by Figure 4C and Figure 4-figure supplement 2, and submited our source codes to GitHub. (II) Reviewer #2 had concerns on the reliability of our method. However, we have revealed that it has comparible sensitivity and specificity with established bioinformatics tools (Figure 4—figure supplement 2C), and even higher accuracy on the assignment of eccDNA boundaries (Figure 4—figure supplement 2A). (III) Reviewer #2 also believed that “the similarity between the eccDNA profiles of human and mouse sperm remains uncertain”. However, we believe that our Fig. 5 have clearly shown that human sperm eccDNAs have exactly the same characteritics with mouse sperm eccDNAs. Nevertheless, in revised manuscript, we will add more description to help readers to better understand our method, and perform additional analyses to further back up our claims.

The amplification bias is indeed a problem of Circle-seq. Following editors’ and Reviewer #1’s insightful suggestions, we will analyze other datasets generated either by rolling circle amplification or not to see how our findings are affected. Additionally, we will consider to add one section to remind readers of the limitations of rolling-circle amplification-based Circle-seq and our data interpretation.

Reviewer #1 (Public Review):

This study aims to address the mechanism of eccDNA generation during spermatogenesis in mice. Previous efforts for cataloging eccDNA in mammalian germ cells have provided inconclusive results, particularly in the correlation between meiotic recombination and the generation of eccDNA. The authors employed an established approach (Circle-seq) to enrich and amplify eccDNA for sequencing analyses and reported that sperm eccDNA is not associated with miotic recombination hotspots. Rather, the authors reported that eccDNAs are widespread, and oligonucleosomal DNA fragments from sperm undergoing apoptosis, with the ligation of DNA ends by microhomology-mediated end-joining, would be a major source of eccDNA.

The strength of the study includes evaluating the eccDNA contents not only in sperm but also from earlier stages of cells in spermatogenesis. The differences in eccDNA size peaks between sperm and other progenitors, in particular, the unique peak in sperm around 360 bp, are intriguing. Results from sequencing data analysis were presented elegantly.

We are grateful to Reviewer #1 for his or her recognition of the strength of this study.

I also have critiques. First, the lack of eccDNA quality control step is a concern. Previous studies employed electron microscopy to ensure that DNA species are mostly circular before rolling-circle amplification. Phi29 polymerase is widely used for DNA amplification, including whole genome amplification of linear chromosomal DNA. Phi29 polymerase has a high processivity and strand displacement activity. When those activities occur within a molecule, it creates circular DNA from linear DNA in vitro. In vitro-created eccDNA from linear DNA would be randomly distributed in the genome, which may explain the low incidence of common eccDNA between replicates. Therefore, it will be crucial to show that DNA prior to amplification is dominantly circular. Electron microscopy would be challenging for the study because the relatively small number of cells were processed to enrich eccDNA. An alternative method for quality controls includes spiking samples with linear and circular exogenous DNA and measuring the ratios of circular/linear control DNA before and after column purification/exonuclease digestion. eccDNA isolation procedures can be validated by a very high circular/linear control DNA ratio.

We highly appreciate Reviewer #1’s insightful suggestions. We would like to perform eccDNA quality control by introducing circular exogenous DNA into our samples and measuring its ratio to endogenous linear DNA before and after eccDNA isolation procedures.

Another critique is regarding the limitation of the study. It is important to remind the readers of the limitations of the study. As the authors mentioned, rolling circle amplification preferentially increases the copy numbers of smaller eccDNA. Therefore, the native composition of eccDNA is skewed. In addition, the candidate eccDNAs are identified by split reads or discordant read pairs. The details of the mapping process are unclear from the methods, but such a method would require reads with high mapping quality; the identification of eccDNA is expected to require sequencing reads that are mapped to genomic locations uniquely with high confidence, and reads mapped to more than one genomic location, such as highly similar repeat sequences or duplications, are eliminated. Such identification criteria would favor eccDNA formed by little or no homology at the junction sequences, and eliminate eccDNA formed by long homologies at the ends, such as eccDNA formed exclusively by satellite DNA. Therefore, it is not surprising that the authors found the dominance of microhomology-mediated eccDNA. It remains to be determined whether small eccDNA with microhomologies are the dominant species of eccDNA in the native composition. In this regard, it is noted that similar procedures of eccDNA enrichment (column purification, exonuclease digestion, and rolling circle amplification ) revealed variable sizes and characteristics of eccDNA in sperm (human from Henriksen et al. or mice from this study), dependent on the methods of sequencing (long-read or short-read sequencing). Considering these limitations, the last sentence of the introduction, "We conclude that germline eccDNAs are formed largely by microhomology mediated ligation of nucleosome protected fragments, and barely contribute to de novo genomic deletions at meiotic recombination hotspots" needs to be revised.

We thank Reviewer #1 for pointing out limitations of the study. We will take into account and integrate the perspectives of Reviewer #1 in our revised manuscript. We will also try to analyze eccDNA datasets generated by long-read sequencing to see how our conclusions might be affected. However, we envision that it might be challenging to examine the contribution of microhomology-mediated ligation to eccDNA biogenesis using long-read sequencing data as the sequencing error rate of nanopore long-read sequencing data is very high.

Small eccDNA (microDNA) data from various mouse tissues are available from the study by Dillion et al., (Cell Reports 2015). Authors are encouraged to examine whether the notable findings in this study (oligonucleosomal-sized eccDNA peaks and the association with apoptotic cell death) are unique to sperm or common in the eccDNA from other tissues.

We are thankful to Reviewer #1 for this suggestion. We would like to analyze additional eccDNA sequencing datasets to see whether our findings are unique to sperm or common for other tissues.

Reviewer #2 (Public Review):

This study presents a useful investigation of eccDNAs in spermatogenesis of mouse. It provides evidence about the biogenesis of eccDNAs and suggests that eccDNAs are derived from oligonucleosmal DNA fragmentation during apoptosis by MMEJ and may not be the direct products of germline deletions. However, the method of data analyses were not fully described and data analysis is incomplete. It provides additional observations about the eccDNA biogenesis and can be used as a starting point for functional studies of eccDNA in sperms. However, many aspects about data analyses and data interpretations need to be improved.

We thank Reviewer #2 for his or her critical reading. However, we respectfully disagree with some critiques on our data analyses (see below). Anyway, we will provide more method details in addition to Fig. 4C and Figure 4-figure supplement 2 that have illustrated the principle and steps of our method, as well as the performance in comparison with established methods. We will also perform additional analyses and make some clarifications in revised manuscript (see below).

• Most of the conclusions made by the work are only based on the bioinformatics analyses, the validation of these foundlings using other method (biochemistry/molecular biology method) are missing. For example, no QC results presented for the eccDNA purification, which may show whether contaminates such as linear DNA or mitochondria DNA have been fully removed. Additionally, it is also helpful to use simple PCR to test the existence of identified eccDNAs in sperm or other samples to validate the specificity of the Circle-seq method.

Following both this Reviewer’s and Reviewer #1’s suggestions, we will introduce circular exogenous DNA into our samples and measure its ratio to endogenous linear DNA and to mitochondria DNA before and after eccDNA isolation procedures. We will also try to perform PCR to test the existence of identified eccDNAs.

• The reliability of the data analysis methods is uncertain, as the authors constructed and utilized their own pipeline to identify eccDNAs, despite the availability of established bioinformatics tools such as ECCsplorer, eccFinder, and Amplicon Architect. Moreover, the lack of validation of the pipeline using either ground truth datasets or simulation data raises concerns about its accuracy. Additionally, the methodology employed for identifying eccDNA that encompasses multiple gene loci remains unclear.

In fact, we have compared the performance between our method and established methods for identification of eccDNA regions, such as Circle_finder, Circle_Map and ecc_finder. Our method has comparable sensitivity and specificity with existing methods, especially Circle_finder and Circle_Map (Figure 4—figure supplement 2C). We also used one specific genomic region to show that existing methods identified the same eccDNA regions but misassigned the eccDNA boundaries (Figure 4—figure supplement 2A). These results have been shown in Figure 4—figure supplement 2. We will highlight the information to make it more clear in our revised manuscript. We will further detect eccDNAs by ECCsplorer for comparison. Since Amplicon Architect is more specifically designed for detection of ecDNAs, it will not be included in our comparison. We will also try to perform PCR to validate the identified eccDNAs.

As pointed out by Reviewer #2, similar to ECCsplorer, Circle_finder, Circle_Map and ecc_finder, our method fails to identity ecDNAs that encompass multiple gene loci. We will remind readers of this limitation in our revised manuscript.

• Although the author stated that previous studies utilizing short-read sequencing technologies may have incorrectly annotated eccDNA breakpoints, this claim requires careful scrutiny and supporting evidence, which was not provided in the manuscript.

As abovementioned, we used one specific genomic region to show that existing methods all misassigned the eccDNA boundaries (Figure 4—figure supplement 2A). In revised manuscript, we will provide necessary statistics to support this claim.

• The similarity between the eccDNA profiles of human and mouse sperm remains uncertain, and therefore, analyses of human eccDNA data and comparisons between the two are necessary if the authors claim that their findings of widespread eccDNA formation in mouse spermatogenesis extend to human sperms.

We believe that our Fig. 5 have clearly shown that human sperm eccDNAs are also originated from oligonucleosomal fragmentation (Fig. 5A-C), not associated with meiotic recombination hotspots (Fig. 5D and E) but formed by microhomology directed ligation (Fig. 5F and G). These findings are consistent with what we observed in mouse sperm eccDNAs. Nevertheless, we will analyze additional public datasets to further back up our claim in revised manuscript.

Author Response

Reviewer #1 (Public Review):

This is an interesting manuscript that proposes a new approach to for accounting for viral diversity within hosts in phylogenetic analyses of pathogens. Concretely, the authors consider sites for which a minor allele exist as an additional base in the substitution model. For example, if at a particular site 60% of reads have an C and 40% have a G, then this site is assigned Cg, as opposed to an C which is typical of analysing consensus sequences. Because we typically model sequence evolution as a Markovian process, as is the case here, the data become naturally more informative, given that there are more states in the Markov chain when adding these bases. As a result, phylogenetic trees estimated using these data are better resolved than those from consensus sequences. The branches of the trees are probably also longer, which is why temporal signal becomes more apparent.

I commend the authors on their rigorous simulation study and careful empirical data analyses. However, I strongly suggest they consider whether treating minor alleles as an additional base is biologically realistic and whether this may have implication for other analyses, particularly when there is very high within-host diversity and the number of states in becomes very large.

We thank the reviewer for the helpful and thorough review. We have included a paragraph in the Discussion regarding the biological interpretation of the 16-state model (Line 344-351), as well as the consequences when there’s high within-host diversity (Line 398).

Reviewer #2 (Public Review):

I agree that minor genetic variation could potentially be used to more accurately infer who-infected- whom in an outbreak scenario. Indeed, the use of minor genetic variation has proven very useful in reconstructing transmission chains for chronic infections such as HIV (e.g., see applications using Phyloscanner). To me, it seems that considering the full spectrum of viral genetic diversity within infected hosts would necessarily do the same if not better than considering only consensus-level viral sequence data. This is because there is a necessarily a loss of data and potentially a loss of information when going from considering the genetic composition of viral populations within a host to only considering the consensus sequences of those viral populations. As such, Ortiz et al.'s hypothesis stated on lines 66-70 is a reasonable one, and I was looking forward to seeing this hypothesis evaluated in detail in this manuscript.

R2.1 There are several parts of this manuscript I really like. In particular, encoding within-sample diversity as character states and using that alternative representation of sequence data for phylogenetic inference (as shown in Figure 3) is a very interesting idea, I think. There are some limitations that are not explicitly mentioned, however. For example, when using this 16-character state representation for phylogenetic inference, they assume independence between nucleotide sites. This is a major assumption that can be violated when considering longitudinal intrahost data and transmission dynamics in an outbreak setting, given genetic linkage between sites.

We have generated another set of simulations where the starting tree was a coalescent tree rather than a random phylogeny. This is described in the Results section, Line 228, and Figure 4—figure supplement 2. By using a coalescent tree, we increase the genetic linkage between sites. For all metrics used, the 16-state model performed better than the consensus sequence model. It is also important to note, as the reviewer points out, that longitudinal isolates should be removed from transmission inference, as we do in Figure 7 and Figure 7—figure supplement 2.. This point is now reflected in the Results (Line 286) and Methods (Line 534).

I have several major concerns about the work as it stands, particularly in the context of the SARS-CoV-2 application.

Concerns not related to the SARS-CoV-2 application:

R2.2 Figure 4 shows that a model using within-sample diversity can more accurately reconstruct evolutionary histories than a model that uses only consensus-level genetic data. This is really interesting. The Materials and Methods section (particularly lines 351-354) indicates that the sequence data were generated using certain specified substitution rates. The rates specified seem to be chosen in such a way to facilitate finding an improvement when using within-sample diversity. I don't know whether the relative rates of these 'substitutions' at all mirror "real-life". It would be very useful to have a broader set of analyses here to examine the effect of these 'substitution' rates on the utility of incorporating within-sample diversity into phylogenetic inference. (Also, 1, 100, 200 (line 353) inconsistent with 1, 20, 200 in Supp Table 3)

We have now corrected Supp Table 3 to reflect the rates described in the Methods section.

We defined our model with three rates: rate of minor variant acquisition, rate of minor-major variant switch, and rate of minor variant loss. We chose the rates for the simulations (1, 100, 200) to reflect a low rate of minor variant acquisition (1) and high rates of minor-major variant switch (200) and minor variant loss (100). These rates will result in pure bases (A,C,G and T) 100 times more likely to be present than low frequency variants, as seen in the base frequencies in Supp Table 1 and 3, which would in turn minimize the effect of including minor variations. We chose these rates to reflect the high turnover of minor variation often observed in real data and the frequencies of minor alleles in the SARS-CoV-2 dataset, but we agree with the reviewer that this may not always be the case. We also agree with the reviewer that changing the parameters in the simulations also affects the effect of including low frequency variation in the model. As such, we have now included simulations using different sets of rates (Figure 4—figure supplement ):

1) With a high rate of variant switch and loss compared to acquisition (1, 10, 100), reducing the frequency of minor variation.

2) With a lower rate of switch and loss (1, 10, 10), promoting a stable landscape of low frequency variation.

3) With no low frequency variation (Jukes-cantor model)

R2.3 Figure 5 is very interesting, particularly the results at bottleneck sizes of 1-10. What are the 'substitution' rates that are inferred here from using this simulated dataset? The Material and Methods section also does not mention the within-host viral generation time anywhere, as far as I can see (~line 384 states the mutation rate per base per generation cycle but not the length of the generation cycle anywhere).

Fastsimcoal2 is a coalescent simulator of population histories over several generations, given a population size and a mutation rate. For our purposes, transmissions are simulated as bottlenecks of constant size, and a generation is represented by each time step in the outbreak simulation, which corresponds to 1 day. This is further clarified in the Methods section (Line 475).

Concerns related to the SARS-CoV-2 application:

R2.4 I am very concerned about the testing of this hypothesis on the SARS-CoV-2 data presented. First, 1% is a very low variant calling threshold. Second, analysis of the 17 samples that were resequenced (out of 454) indicated that on average, 39% of iSNVS (intrahost single nucleotide variants) called between duplicate runs were only observed in one of the two runs (line 117). Their analysis in Figure 1 indicates that these discrepant (and seemingly spurious) variants occur at higher levels in high Ct samples (which makes sense; Figure 1b). They therefore decide to limit their analyses to samples with Ct values <= 30. This results in 249 samples. However, if we look at Figure 1b, only ~10% of iSNVs called across duplicate runs with Ct = 30 are shared! That means that 90% of iSNVs in the set appear to be spurious. If we assume that each duplicate run of a sample has approximately the same number of spurious iSNVs, then approximately 82% of iSNVs called in a sample with a Ct of 30 would be spurious. This fraction decreases with samples that have lower Ct values, but even at a Ct of 27, only ~60% of iSNVs called across duplicate runs are shared. All the downstream SARS-CoV-2 analyses based on within-host sample diversity therefore are based on samples where the large majority of considered sample diversity is not real. This leads to me necessarily discounting all of those downstream SARS-CoV-2 results.

We agree with the reviewer that, as the results show, datasets that incorporate within-sample low frequency variation are expected to have considerably more noise than using exclusively consensus sequences, and perhaps this wasn’t properly discussed in the manuscript. We have incorporated some notes about this in the Discussion section (Line 408-413).

The 1% variant frequency threshold was used to generate the analysis of Fig. 1 and Supp. Fig. 1-4. Looking at these results, we decided to establish the Ct cutt-off of 30 as mentioned by the reviewer, as well as a variant frequency threshold of 2% (as shown in the x-axis of Fig. 2). We overlooked this second variant frequency threshold in the manuscript, which has been added. As shown in Supp. Fig 4, this variant frequency threshold will increase the concordance between technical replicates, although some level of noise persists.

R2.5 Lines 153-167: I can't figure out how to square the quantitative results given in this paragraph with what is shown in Figure 2. To me, Figure 2 shows only that Technical Replicates have higher probabilities of sharing a variant than with 'No' relationship. What would also be helpful here so that the reader can get a better feel for the data would be to see the iSNV frequencies plotted over time for the longitudinal replicate samples in the supplement and, for the 'epidemiological' samples to show 'TV plots' in the supplement (as in Fig 3c in McCrone et al. eLife)

Figure 2 shows that technical replicates, longitudinal replicates, epidemiological samples and, in some instances, from the same department have a higher probability of sharing low frequency variants than those with no relationship (also shown in Supp Figure 5). However, also shown in Figure 2 is that the 95% CI is very wide, and therefore in many instances low frequency variants won’t be shared between epidemiological samples or samples from the same department.

We have also added Figure 2—figure supplement showing the low frequency variants plotted over time for longitudinal replicates. Unlike McCrone et al, we don’t have proven transmission between pairs of samples, although we believe our analysis also shows a pattern of shared low frequency variants among potential epidemiological links.

R2.6 Figure 6 and associated text: (a) root-to-tip distance: what units is this distance in? (b) That the authors find a temporal signal in these transmission clusters (where all consensus sequences within a cluster are the same) is interesting but also a bit baffling to me. Given the inference of very small transmission bottlenecks in previous studies (e.g., Martin & Koelle - reanalysis of Popa et al.; Lythgoe et al.; Braun et al.), I don't understand where the temporal signal comes in. Do the samples become more genetically diverse over the outbreak (this seems to be indicated in lines 260-262 but never shown and unlikely given bottleneck sizes)? Additional analyses to help the reader understand WHY within-sample diversity allows for the identification of temporal signal is important. This could involve plotting genetic diversity of the samples by collection date or some other, similar analyses.

a) The units of the y-axis (root-to-tip distance) are measured in substitutions per genome. This is now reflected in the legend of the figure.

b) As shown in Figure 5, even at small bottleneck sizes we are able to pick some of the diversity that evolves during the course of an outbreak. As hinted by the reviewer, the smaller the bottleneck the less diversity we can leverage for phylogenetic inference, and in fact for some epidemiological samples all the diversity will be lost during transmission, which is why many of the within-sample variants are not shared between the epidemiologically related samples. Figure 6 is indeed showing that the genetic distance (measured as number of substitutions per genome) increases per collection date. We have also added a Figure 6—figure supplement showing the increase in low frequency variants within outbreaks as the outbreaks progress in time (explained in Line 261 of the Results section), which explain in part the increasing temporal signal in clusters.

R2.7 Paragraph consisting of lines 229-238 and Figure 7: This analysis stops abruptly. What are the conclusions here? Figure 7a (right) seems inconsistent to me with Figure 7b and 7C results. Also, the main hypothesis put forward in this paper is that within-sample sequence data can better resolve who-infected-whom in an outbreak setting. Figure 7b and 7c however are never compared against analogous panels that use just consensus sequences. (Even though the consensus sequences are the same, according to Figure 7a, the inferences shown in Figures 7b and 7c could use additional data such as collection times, etc. that would provide information even when using exclusively consensus-level data). Also, do the analyses in Figures 7b and 7c use the 16-character state model at all? I think Supp Figure 9 is relevant here but not sure how?)

We have extended this section of the results to make it more coherent and clear (Line 284-293) and in the Discussion (Line 385-395). As added into the Discussion, we agree with the reviewer that even with equal sequences some inferences about transmission can be made with epidemiological data, specially collection dates. However, such data can’t be used to infer the genetic structure of the cluster, which complicates any analysis that can use a phylogenetic as input.

Additional concerns:

R2.8 Some of the stated conclusions, particularly in the Discussion section and in the Abstract, do not seem to be supported by the presented results. For example, line 27: 'within-sample diversity is stable among repeated serial samples from the same host': Figure 2 does not show this conclusively. Line 28: 'within-sample diversity... is transmitted between those cases with known epidemiological links': Figure 2 also does not show this conclusively. Line 29: 'within-sample diversity... improves phylogenetic inference and our understanding of who infected whom': Figure 7b/c results using within-sample diversity is never compared against results that use only consensus, so improvement not demonstrated. Line 272-273: 'samples with shorter distance in the consensus phylogeny were more likely to share low frequency variants'. Line 287: 'We demonstrated that phylogenies... were heavily biased'.

Line 27 and Line 28: We agree with the reviewer that the genomic analysis of SARS-CoV-2 sequences show only partial congruence within technical replicates and epidemiological links. We have appropriately addressed this in the Abstract.

Line 29 and Fig 7: Transmission inference using the consensus sequence in Figure 7b/c couldn’t be performed because the lack of any genetic difference between the consensus sequence meant that all sequences had the same transmission likelihood. This is now better explained in the Discussion section, lines 385-395.

Line 272-273: We have removed this section as we did not perform this analysis, as pointed out by the reviewer.

Line 287: The conclusion expressed in line 287 (now line 340) has been changed.

R2.9 The manuscript at times does not cite previous work that is highly relevant and thus overstates the novelty of the current work. For example: lines 21-23: '..conventional whole-genome sequencing phylogenetic approaches to reconstruct outbreaks exclusively use consensus sequences...' Phyloscanner uses within-sample diversity, for example, as does SCOTTI. These are finally cited in the discussion section (~line 310), but because this previous work is not acknowledged earlier in the manuscript, the novelty of the work presented here is somewhat overstated.

We have included background information in the introduction regarding the use of within-sample diversity for transmission inference (Line 69-73), as well as emphasizing that the novelty of our work lies more in the use of within-sample diversity in phylogenetic inference rather than exclusively transmission inference (Line 74, and other instance along the manuscript).

In sum, I think that the 16 character-state model is a very interesting model. More analyses on simulated data would be helpful to expand on when below-the-consensus level genetic data would truly be informative of phylogenetic relationships and who-infected-whom in outbreak settings. The SARS-CoV-2 analyses are very worrisome to me, given the inclusion of samples where the majority of considered within-sample genetic diversity is very likely not real. Some of the stated conclusions appear to either be at odds with the results presented or not directly evaluated.

Author Response

Reviewer #2 (Public Review):

Targeted genetic engineering with programmable nucleases and other targetable enzymes (aka "genome editing") has emerged as a technology with curative potential in hemoglobinopathies, sickle cell disease, and beta-thalassemia. Multiple ongoing clinical trials are evaluating such editing using distinct approaches: elevation of fetal hemoglobin (HbF), direct repair of the mutation causing SCD, and engineering of a Hb variant. The present work explores a different strategy: the targeted engineering of the promoter of a paralog of adult beta-globin known as HBD. This is a timely effort because there has emerged, over the past decade, a clear and charted path for advancing any such approach to human clinical trials. The study identifies three transcription factor binding sites as divergent in the HBD promoter vs the HBB one. A homology-directed repair (HDR)-based scheme using oligonucleotide repair templates in combination with a CRISPR-Cas9-induced double-strand break (DSB) is designed and used to generate pools of human immortalized cells bearing one, two, or all three such de novo introduced TF binding sites at the HBD promoter. Only the latter scheme is shown to measurably increase HBD (following erythroid differentiation) in pools of cells and single-cell-derived clones as gauged by qPCR and HPLC. A similar analysis is performed on pools of erythroid-like cells generated from genome-edited human hematopoietic stem and progenitor cells (HSPCs), as well as genetically clonal erythroid colonies bearing the edits of interest; trends in these data support the observations made on the immortalized cells. Overall the data support the notion that HBD promoter genome editing has the potential as a strategy to normalize hemoglobin synthesis in hemoglobinopathies. Further, the data support an advance of this approach down a well-established path of preclinical development in such cases: increasing the efficiency of genome editing in HSPCs to what would be deemed therapeutically useful, assessing the genotoxic burden from the editing, evaluating the potential negative impact on stemness, and determining whether this approach would normalize hemoglobin synthesis in the erythroid progeny of patient HSPCs.

We thank reviewer 2 for their input on our manuscript, especially sharing their insight from a clinical path perspective.

The genome editing scheme for the "KDT" strategy in Fig 1B involves the introduction of three binding sites for transcription factors at progressively increasing distances from the site of the DSB induced by Cas9. It would be of interest to determine from the next-generation-sequencing data whether partial gene conversion tracks are observed at the edited locus (Elliott and Jasin MCB 18: 93), and if yes, whether these affect in some way the pool-level measurement by qPCR on HBD mRNA levels (Fig 1D).

For the analysis of our NGS reads, we utilized the CRISPResso2 analysis pipeline. After CRISPResso2 aligns the reads and makes allelic calls of either unmodified, NHEJ, or HDR. It is important to note that the KDT of our HBD knock-in construct, is not identical to the HBB promoter. Through simple searching of the CRISPResso aligned-reads, we did not find any HBB promoter sequence present. In this regard, our CRISPResso analysis does not seem to find any gene-conversions between HBB and HBD. However, we cannot rule out the possibility of gene conversions altogether – it can be that since our primers for NGS anneal specifically to HBD, we are unable to amplify, and therefore unable to see, the alleles in which these gene conversion events occurred.

The data in Fig 2A show an analysis of transcription factor and RNA pol II occupancy following genome editing at HBD. The figure legend refers to these data as having been obtained on single-cell-derived clones bearing the edits in homozygous or heterozygous form, but it is unclear from fig 2A, which clones were used for which analysis.

We have now clarified this point in the figure legend.

The data in Fig 3C present an analysis of HBD levels in erythroid colonies derived from genome-edited HSPCs. It would be helpful to clarify whether an individual dot represents a single such colony (this would seem to be the case from the cognate figure legend). If so, what number of such colonies would one need to obtain to gain a clearer sense of the effect on HBD levels from the various genome editing strategies used?

Indeed, each dot represents a singular colony. We have now expanded this dataset from colonies derived from n=2 HSPC donors to n=4 HSPC donors. Figure 3C and Supp Fig 3 have been updated accordingly.

It would be helpful to comment, in the Discussion, on potential genome editing strategies to obtain high-efficiency pool-level uniform long-track gene conversion that is necessary to obtain high HBD levels in the progeny of edited CD34 cells. Would this be a good application of the AAV6 strategy developed by the Sangamo and Porteus groups? Would prime editing as developed by Liu be an option here?

Prime editing can introduce small insertions, but still has limitations of low-editing efficiency (https://doi.org/10.1016/j.tibtech.2023.03.004). Additionally, our KDT construct would require a larger insertion that prime editing would not be able to facilitate easily. In light of the adverse effects using AAV6 for biotech company Graphite Bio, we will not suggest this in the discussion.

It would be equally helpful, in the Discussion, to place the level of HbA2 obtained via the strategy shown in the manuscript in the context of other genome-editing-based approaches for normalizing Hb synthesis in the hemoglobinopathies (ie HbF elevation by editing the BCL11A enhancer, or the gammaglobin promoter; or direct repair of the SCD mutation; or engineering of Hb Makassar).

We have now added a new section in the discussion summarizing some of the recent genome editing approaches for hemoglobinopathies. Specifically, we mention CRISPR Therapeutics’ clinical trial on the BCL11A enhancer, David Liu’s most recent paper on base-editing to correct the SCD mutation, and Annarita Miccio’s recent paper on disrupting a repressor binding site on the gamma-globin promoter.

Reviewer #3 (Public Review):

This is a well-written and referenced paper from the laboratory of an outstanding senior investigator. Dr. Corn and colleagues demonstrate convincingly that correction of three transcription factor binding sites in the delta-globin gene promoter results in high levels of delta-globin expression in HUDEP-2 clonal cell populations (Fig. 2B and C) and in CD34+ HSPC (hematopoietic stem and progenitor cells) clonal cell expansions (Fig. 3C). Although correction of the mutant KLF1 binding site has previously been shown to upregulate delta-globin gene transgenes, this new data demonstrate that correction of multiple factor binding sites is required to achieve high-level expression of the delta-globin gene in the endogenous beta-globin gene locus. The results are important because high delta-globin protein levels inhibit the formation of sickle hemoglobin (HbS) polymers that cause sickle cell disease.

We thank reviewer 3 for their feedback on our manuscript.

Unfortunately, high levels of delta-globin gene expression were not observed after editing of pooled (non-clonal) populations of HUDEP-2 cells (Fig. 1D) or CD34+ HSPC pooled cell populations (Fig. 3B). This result suggests that correction of all 3 promoter elements on individual alleles in CD34+ HSPC populations is far below the level required to be clinically relevant.

We have added to the discussion on ways to improve HDR efficiency. Additionally, we show new data where we utilize an HDR enhancer drug and show that we can increase HDR and overall HBD in edited pooled populations of HSPCs (Fig 3 C and D)

Also, NHEJ is high in CD34+ HSPC (Fig. 3A); therefore, promoter deletions will inactivate many alleles, and total hemoglobin levels in erythrocytes derived from populations of edited CD34+ HSPC will be much less than normal (29 pg/cell). These cells would be extremely beta-thalassemic.

We were not completely sure about the origin of this point, since our edits are aimed at HBD, which makes up less than 5% of total hemoglobins under normal conditions. NHEJ occurring in HBB (e.g. when doing HDR for direct correction) would potentially yield thalassemic cells. But indels in the HBD promoter might at most cause a 5% decrease in total globin levels (if delta expression was completely destroyed). We have performed a new experiment to explicitly address this point. We edited n=4 CD34+ HSPCs donors and compared unedited populations to populations edited with Cas9+HBD gRNA but no repair template. This represents a “worst case” scenario, in which there can be no HDR-based promoter engineering and only NHEJ. These data are included this in Supplementary Figure 3. We observed high editing efficiency of 61 – 78% in the HBD promoter. We performed qRT-PCR of the beta-like globins in edited pools and normalized to HBA, reasoning that HBA is a neutral control for absolute levels of each globin in the beta locus because HBA is located in a different locus. By qRT-PCR, HBD transcripts were decreased by half compared to mock treated cells, while HBB and HBG1/2 were non-significantly affected. But as mentioned above, HBD expression makes up less than 5% of total hemoglobins, and therefore a half reduction in HBD represents a total reduction of 2.5% of globins. We do acknowledge that this experiment does not specifically quantify the rates of large deletions that might span from delta to beta, and further studies would be needed to address this point. But if such large deletions do exist, they do not greatly affect beta expression. We have included this in the results and the discussion section.

Author Response

Reviewer #1 (Public Review):

In this interesting manuscript, Nasser et al explore long-term patterns of behavior and individuality in C. elegans following early-life nutritional stress. Using a rigorous, highly quantitative, high-throughput approach, they track patterns of motor behavior in many individual nematodes from L1 to young adulthood. Interestingly, they find that early-life food deprivation leads to decreased activity in young larvae and adults, but that activity between these times, during L2-L4, is largely unaffected. Further, they show that this "buffering" of stress requires dopamine signaling, as L2-L4 activity is significantly reduced by early-life starvation in cat-2 mutants. The paper also provides evidence that serotonin signaling has a role in modulating sensitivity to stress in L1 larvae and adults, but the size of these effects is modest. To evaluate patterns of individuality, the authors use principal components analysis to find that three temporal patterns of activity account for much of the variation in the data. While the paper refers to these as "individuality types," it may be more reasonable to think of these as "dimensions of individuality." Further, they provide evidence that stress may alter the strength and/or features of these dimensions. Though the circuit mechanisms underlying individuality and stress-induced changes in behavior remain unknown, this paper lays an important foundation for evaluating these questions. As the authors note, the behaviors studied here represent only a small fraction of the behavioral repertoire of this system. As such, the findings here are an interesting and very promising entry point for a deeper understanding of behavioral individuality, particularly because of the cellular/synaptic-level analysis that is possible in this system. This paper should be of interest to those studying C. elegans behavior and also more generally to those interested in behavioral plasticity and individuality.

We thank the reviewer for finding our results interesting.

Reviewer #2 (Public Review):

This paper set out to understand the impact of early life stress on the behavior and individuality of animals, and how that impact might be amplified or masked by neuromodulation. To do so, the authors built on a previously established assay (Stern et al 2017) to measure the roaming fraction and speed of individuals. This technique allowed the authors to assess the effects of early life starvation on behavior across the entire developmental trajectory of the individual. By combining this with strains with mutant neuromodulatory systems, this enabled the authors to produce a rich dataset ripe for analysis to analyze the complicated interactions between behavior, starvation intensity, developmental time, individuality, and neuromodulatory systems.

The richness of this dataset - 2 behavioral measures continuous across 5 developmental stages, 3 different neuromodulatory conditions (with the dopamine system subject to decomposition by receptor types) and 4 different levels of starvation, with ~50-500 individuals in each condition-underlies the strength of this paper. This dataset enabled the authors to convincingly demonstrate that starvation triggers a behavioral effect in L1 and adult animals that is largely masked in intermediate stages, and that this effect becomes larger with increased severity of starvation. Furthermore, they convincingly show that the masking of the effect of starvation in L2-L4 animals depends on dopaminergic systems. The richness of the dataset also allowed a careful analysis of individuality, though only neuromodulatory mutants convincingly manipulated individuality, recapitulating earlier research. Nonetheless, a few caveats exist on some of their findings and conclusions:

We thank the reviewer for the constructive comments. In the revised manuscript we include additional analyses and textual changes as detailed below, to address the points raised.

1) Lack of quantitative analysis for effects within developmental stages. In making the argument for buffered effects of starvation on behavior during periods of larval development, the authors make claims regarding the temporal structure of behavior within specific stages. However, no formal analysis is performed and and the traces are provided without confidence intervals, making it difficult to judge the significance of potential deviations between starvation conditions.

In the revised manuscript, we include additional analyses of roaming fraction effects across shorter developmental-windows, showing within-stage differences in behavioral patterns following starvation (Figure 1 - figure supplement 1E; Figure 3 - figure supplement 1C). In addition, we further temper and rewrite our conclusions to clearly describe these effects (now- “…while 1 day of early starvation modified within-stage temporal behavioral structures by shifting roaming activity peaks to later time-windows during the L2 and L3 stages…” in p. 4 and “Interestingly, during the L2 intermediate stage the effects on roaming activity patterns were more pronounced during earlier time-windows of the stage…” in p. 8).

2) Incorrect inferences from differences in significance demonstrating significant differences. The authors claim that there is an increase in PC1 inter-individual variation in tph-1 individuals, however the difference in significance is not evidence of a significant difference between conditions (see Nieuwenhuis et al. 2011). This undermines claims about an interaction of starvation, neuromodulators, and individuality.

In the revised manuscript we provide now a direct comparison of PCs inter-individual variances between starved and unstarved populations, demonstrating significant differences in inter-individual variation in specific PC individuality dimensions following stress (Figure 6 and Figure 6 - figure supplement 1). These results include the increase in PC1 inter-individual variation in tph-1 mutants following 3 and 4 days of starvation (Figure 6A,E).

3) Sensitivity of analysis to baseline effects and assumptions of additive/proportional effects. The neuromodulatory and stress conditions in this paper have a mixture of effects on baseline activity and differences from baseline. The authors normalize to the roaming fraction without starvation, making the reasonable assumption that the effect due to starvation is proportional to baseline, rather than an additive effect. This confound is most visible in the adult subpanel of figure 5d, where an ~2-3 fold difference in relative roaming due to starvation is clearly noted, however, this is from a baseline roaming fraction in tph-1 animals that are ~2 fold higher, suggesting that the effect could plausibly be comparable in absolute terms.

Unavoidably, any such assumptions on the expected interaction between multiple effects will be a gross simplification in complicated nonlinear systems, and the data are largely shown with sufficient clarity to allow the reader to make their own conclusions. However, some of the interpretations in the paper lean heavily on an assumption that the data support a direct interpretation (e.g. "neuronal mechanisms actively buffer behavioral alterations at specific development times") rather than an indirect interpretation (e.g. that serotonin reduces baseline roaming fraction which makes a fixed sized effect more noticeable). Parsing the differences requires either more detailed mechanistic study or careful characterization of the effect of different baselines on the sensitivity of behavior to perturbation-barring that it's worth noting that many of these interactions may be due to differences in biological and experimental sensitivity to change under different conditions, rather than a direct interaction of stress and neuromodulatory processes or evidence of differing neuromodulatory activity at different stages of development.

In the revised manuscript we added a discussion of the potential complicated interactions between neuromodulation and stress, altering baseline levels and deviations from baseline. We also discuss the interpretation of the results in the context of non-linear systems in which sensitivity of the behavioral response to underlying variations may be modified by specific neuromodulatory and environmental perturbations, without assuming direct differences in neuromodulatory states over development or across individuals (p. 16).

Reviewer #3 (Public Review):

In this study, Nasser et al. aim to understand how early-life experience affects 1) developmental behavior trajectory and 2) individuality. They use early life starvation and longitudinal recording of C. elegans locomotion across development as a model to address these questions. They focus on one specific behavioral response (roaming vs. dwelling) and demonstrate that early life (right after embryo hatching) starvation reduces roaming in the first larval (L1) and adult stages. However, roaming/dwelling behavior during mid-larval stages (L2 through L4) is buffered from early life starvation. Using dopamine and serotonin biosynthesis null mutant animals, they demonstrated that dopamine is important for the buffering/protection of behavioral responses to starvation in mid-larval stages, while in contrast, serotonin contributes to early-life starvation's effects on reduced roaming in the L1 and adult stages. While the technique and analysis approaches used are mostly solid and support many of the conclusions made in the manuscript for part 1), there are some technical limitations (e.g., whether the method has sufficient resolution to analyze the behaviors of younger animals) and confounding factors (e.g., size of the animal) that the authors do not yet sufficient address, and can affect interpretation of the results. Additionally, much of the study is descriptive and lacks deep mechanistic insight. Furthermore, the focus on a single behavioral parameter (dwelling vs. roaming) limits the broad applicability of the study's conclusions. Lastly, the manuscript does not provide clear presentation or analysis to address part 2), the question of how early life experience affect individuality.

We thank the reviewer for these important comments. As described below, in the revised manuscript we include new analyses (following extraction of size data), showing behavioral modifications across different conditions/genotypes also in size-matched individuals (within the same size range) (Figure 1 - figure supplement 1F; Figure 3 - figure supplement 1D,E; Figure 5 - figure supplement 1B,D). We also made edits to the text to describe these results (Methods p. 21 and Results section). In addition, while we can detect behavioral changes using our imaging method even in young L1 worms across conditions and genotypes (described in Stern et al. 2017 and this manuscript), as the reviewer correctly pointed out, we may miss some milder behavioral effects due to lower spatial imaging resolution in younger worms. We are now referring to this spatial resolution limitation in the revised manuscript (discussion part). Lastly, in the revised manuscript we added clearer and more direct analyses of changes in inter-individual variation in multiple PC dimensions following early stress, by directly comparing variation between starved and unstarved individuals within the mutant and wild-type populations (Figure 6; Figure 6 - figure supplement 1). These analyses show significant changes in inter-individual variation within specific PC individuality dimensions following early stress. Also, we made textual changes along the manuscript to increase the clarity of presentation of these results.

Author Response

Reviewer #2 (Public Review):

Using an approach that combines synthetic genetic array (SGA) analysis with high-throughput microscopic analysis of the GFP-tagged yeast ORF collection in the budding yeast, Saccharomyces cerevisiae, this study has examined the contribution of the critical checkpoint kinases Mec1 and Rad53 to the subcellular relocalization of 322 candidate proteins in response to HU- and MMS-induced replication stress. Previous studies have established that Mec1 is required for Rad53 activation during replication stress and that Mec1 also serves checkpoint functions independent of Rad53. Unexpectedly, this study identifies groups of proteins whose stress-induced relocalization is dependent on Rad53 but not Mec1. This data indicates that Rad53 mediates some replication stress responses in a non-canonical manner that is independent of Mec1.

The authors confirm their initial observations from the screening approach by focusing on the Rad53-dependent and Mec1-independent focus formation of GFP-Rad54. Moreover, using mass-spec analysis the authors demonstrate that some Rad53 phosphorylation sites known to be critical for Rad53 activation, including a consensus Mec1 phosphorylation site, are phosphorylated after replication stress even in the absence of Mec1. Motivated by this finding the authors screen for potential kinase and phosphatase pathways that may regulate Rad53 function during MMS-induced replication stress. Top hits identified include members of the retrograde signaling pathway, which is confirmed by conventional genetic assays while mass spec analysis supports the involvement of Rtg3 in mediating Rad53 phosphorylation during replication stress in the absence of Mec1.

Overall this is a solid study reporting unexpected new findings that significantly advance our view of the global replication checkpoint response. The data are generally of high quality, well presented and quantified, and overall support the authors' claims. The mass spec approach used here to identify Rad53 phosphorylation sites offers an unbiased alternative to the simpler and more widely employed gel-shift method to monitor Rad53 activation. The hits identified in the various screens presented here provide a platform for potential follow-up studies by the community. The main drawback is that it remains unclear how Rtg3 promotes Rad53 activation. However, this could be considered to be beyond the scope of this study.

We thank the reviewer for their positive assessment of our experimental data. We have made the changes requested by the reviewer to increase the clarity of Figure 5, and performed a second replicate to show that the FACS data are reproducible.

Reviewer #3 (Public Review):

The work by Ho et al describes the identification of Mec1/Tel1 independent activation of Rad53 after MMS treatment, which could lead to changes of GFP fusion signals for several dozens of proteins and this was partly dependent on Rtg3. Starting from an unbiased, targeted screen, the authors identified proteins whose GFP fusion signals changed intensity in rad53∆ but not in mec1∆ cells using live cell imaging, including Rad54. Using Rad54 as a readout for the subsequent experiments, a second screen amongst kinases/phosphatases and their regulators found that rtg2-3 mutants reduced Rad54-GFP intensity. Mass spectrometry data identified Rad53 phosphorylate sites in mec1∆ tel1∆ cells, consistent with the cell biological data described above. Overall, the work was well done and supported the main conclusions. The concept of Mec1/Tel1-independent and Rtg3-dependent Rad53 activation connects checkpoint signaling with the retrograde pathway.

We thank the reviewer for their positive assessment of our experimental work and appreciate the suggestions that ultimately led to interesting new data. As outlined below, we attempted to perform most of the experiments suggested by the reviewer. Unfortunately, experiments with a MEC1-AID degron allele were inconclusive, with details summarized below. However, we have identified additional proteins whose re-localization is affected by Rtg3, providing stronger support for its role in the replication stress response. We revised the manuscript to add these new details.

Author Response

Reviewer 1 (Public Review):

In this paper, Reato, Steinfeld et al. investigate a question that has long puzzled neuroscientists: what features of ongoing brain activity predict trial-to-trial variability in responding to the same sensory stimuli? They record spiking activity in the auditory cortex of head-fixed mice as the animals performed a tone frequency discrimination task. They then measure both overall activity and the synchronization between neurons, and link this ’baseline state’ (after removing slow drifts) of cortex to decision accuracy. They find that cortical state fluctuations only affect subsequent evoked responses and choice behavior after errors. This indicates that it’s important to take into account the behavioral context when examining the effects of neural state on behavior.

Strengths of this work are the clear and beautiful presentation of the figures, and the careful consideration of the temporal properties of behavioral and neural signals. Indeed, slowly drifting signals are tricky as many authors have recently addressed (e.g. Ashwood, Gupta, Harris). The authors are well aware of the difficulties in correlating different signals with temporal and cross-correlation (such as in their ’epoch hypothesis’). To disentangle such slow trends from more short-lived state fluctuations, they remove the impact of the past 10 trials and continue their analyses with so-called ’innovations’ (a term that is unusual, and may more simply be replaced with ’residuals’).

The terms ‘innovations’ and ‘residuals’ are sometimes used interchangeably. We used innovations because that’s how they were introduced in the signal processing literature (i.e., Kailath, T (1968). ”An innovations approach to least-squares estimation–Part I: Linear filtering in additive white noise.” IEEE transactions on automatic control). We try to be explicit in the text about the formal definition of this quantity, to avoid problems with terminology.

I do wonder if this throws out the baby with the bathwater. If the concern is statistical confound, the ’session permutation’ method (Harris) may be better suited. If the concern is that short-term state fluctuations are more behaviorally relevant (and obscured by slow drifts), then why are the results with raw signals in the supplement (Suppfig 8) so similar?

The concern was statistical confound, although this concern is ameliorated when using a mixed model approach and focusing on fixed effects. However, our approach allowed us to assess the relative importance of slow versus single-trial timescales in the predictive relationship between cortical state (and arousal) and behavior, revealing that, in the conditions of our experiment, only the fast timescales are relevant. Because of this, we think that the baby wasn’t thrown out with the bathwater as, qualitatively, no new phenomenology was revealed when the slow components of the signals were included. In hindsight, it is true that the results we obtained suggest that maybe the effort we made to isolate the fast component of the signals was unjustified. However, this can only be known after both options have been tried, as we did. Moreover, we started using innovations based on the results in Figure 2 where, as we show, the use of innovations does make a difference, even at the level of fixed effects in a mixed model. We agree that we could have used the ‘session permutation’ method, but given the depth at which we have explored this issue in the manuscript already, and the clarity of the results, we think that adding a third method would only make reading the manuscript more difficult without adding any substantially new content.

While the authors are correct that go-nogo tasks have drawbacks in dissociating sensitivity from response bias, they only cursorily review the literature on 2AFC tasks and cortical state. In particular, it would be good to discuss how the specific method - spikes, EEG (Waschke), widefield (Jacobs) and algorithm for quantifying synchronization may affect outcomes. How do these population-based measures of cortical state relate to those described extensively with slightly different signals, notably LFP or EEG in humans (e.g. work by Saskia Haegens, Niko Busch, reviewed in https://doi.org/10.1016/j.tics.2020.05.004)? This review also points out the importance of moving beyond simple measures of accuracy and using SDT, which would be an interesting improvement for this paper too.

We thank the reviewer for pointing us towards the oscillation-based brain-state literature in humans. We have expanded the paragraph in the discussion where we compare our results with previous work in order to (i) elaborate on the literature on 2AFC tasks, (ii) specifically address the literature linking alpha power in the pre-stimulus baseline and psychophysical performance, and (iii) mention different methods for assessing desynchronization. Our view is that absence of lowfrequency power is a robust measure which can be assessed using different types of signals (spikes, imaging, LFP, EEG). That said, the relationship between desynchronization and behavior appears subtle and variable, specially within discrimination paradigms. These issues are discussed in the paragraph starting in line 527 in the text.

Regarding the use of SDT, we had already established that our main finding could be expressed as a significant interaction between FR/Synch and the stimulus-strength regressor, when predicting choice after errors (Supplementary Fig. 4A in original manuscript), which is equivalent to a cortical state-dependent increase in d′ after the mice made a mistake. In order to consider a possible effect of cortical state on the ‘criterion’ (i.e., an effect on the bias of the mice towards either response spout), we re-run this GLMM but adding the cortical state regressors as main effects. The results show that the FR-Synch predictor is only significantly greater than zero as an interaction after errors (p = 0.0025). As a main effect, it’s not significantly different from zero neither after errors (p = 0.28), nor after correct trials (p = 0.97). We have included this analysis as Figure 3-figure supplement 1B (replacing the previous Supplementary Fig. 4A) and commented on them in the text (lines 222-225).

Reviewer 2 (Public Review):

The relationship between measures of brain state, behavioral state, and performance has long been speculated to be relatively simple - with arousal and engagement reflecting EEG desynchronization and improved performance associated with increases in engagement and attention. The present study demonstrates that the outcome of the previous trial, specifically a miss, allows these associations to be seen - while a correct response appears less likely to do so. This is an interesting advance in our understanding of the relationship between brain state, behavioral state, and performance.

This is probably just a typo, but we would like to clarify that the relevant outcome in the previous trial is not a miss, but an incorrect choice in an otherwise valid trial (i.e., a trial with a response within the allowed response window).

While the study is well done, the results are likely to be specific to their trial structure and states exhibited by the mice. To examine the full range of arousal states, it needs to be demonstrated that animals are varying between near-sleep (e.g. drowsiness) and high-alertness such as in rapid running. The fact that the trials occurred rapidly means that the physiological and neural variables associated with each trial will overlap with upcoming trials - it takes a mouse more than a few seconds to relax from a previous miss or hit, for example. Spreading the rapidity of the trials out would allow for a broader range of states to be examined, and perhaps less cross-talk between adjacent trials. The interpretation of the results, therefore, must be taken in light of the trial structure and the states exhibited by the mice.

We thank the reviewer for the positive assessment of our work and also for raising this point in particular. This motivated us to look more carefully at this issue, with results that, we believe, strengthen our study.

Author Response

Reviewer #1 (Public Review):

In this work, Roche et al. study a 13-year long time series of microbiome samples from wild baboons from Kenya. The data used in this work challenge a previous finding from the same authors that temporal dynamics in microbiome changes are largely individualized. Using a multinomial logistic-normal modeling approach, the authors detect that co-variance in temporal dynamics in microbial pair-wise associations among individuals occurs more frequently between relatives. Furthermore, the authors identify that microbial phylogenetic proximity is associated with consistent co-abundance changes over time and that their metric of universal microbial relationships is robust across hosts and is detected even in human longitudinal data. The authors conduct a thorough statistical revision of publicly available results, highlighting this time (e.g. compared to Björk et al, doi: 10.1038/s41559-022-01773-4) the consistently shared microbial properties between individuals, rather that the individual microbial signatures highlighted in their previous work.

Thank you for this summary. We would like to briefly clarify that we do not see the current work as inconsistent with our prior finding in Björk et al. that microbiome taxonomic compositions are idiosyncratic and asynchronized. However, this new analysis, which focuses on abundance correlations between pairs of taxa, indicates that the personalized compositions and dynamics we observed in Björk et al. are probably not attributable to personalized microbiome ecologies. In other words, Björk et al. showed that microbial taxa found in the guts of different baboons can be quite distinct (and remain so over time, giving rise to semi-stable individual signatures). The current study shows that, despite this taxonomic individuality, the correlations between pairs of microbes in the baboon gut are often quite consistent. To give a basic example, hot weather and ice cream, when observed, are often observed together (positively correlated), but while some places have a lot of both, some have little of either. This idea is discussed in more detail below (see response R6) and in the revised Discussion section (lines 572 to 586).

Strengths:

This work is foundational in its compelling effort to generate a rigorous method to evaluate coabundance dynamics in longitudinal microbiome data. The approach taken will likely inspire developments that will sharpen the capacity to extract co-varying microbial features, taking into account seasonality, diet, age, relatedness, and more. To the best of my understanding, their hierarchical model integrated into the Gaussian process to analyze microbial dynamics is reasonably robust and they clearly explain the implementation. Furthermore, this work introduces and defines the concept of a universality score for microbial taxon pairs. Overall, the work presented is clear and convincing and provides tools for the community to benefit from both methods and results. Furthermore, conceptually, this work stresses the value of consistent and shared microbial dynamics in groups, which enriches our understanding of host-associated microbial ecology, otherwise understood to be largely dependent on external fluctuations.

Weakness:

It is not entirely clear the extent to which the presented results revise, refute, or support the previously published analysis performed by the authors on the same dataset (doi: 10.1038/s41559-022-01773-4), which was more focused on individuality.

We agree the relationship between Björk et al. and the current manuscript was unclear in our original submission. We now elucidate the relationship between these papers in the Discussion (lines 572 to 586). Briefly, Björk et al. found that microbiome taxonomic compositions are idiosyncratic and asynchronized. The current analysis finds that pairwise bacterial abundance correlations are predominantly shared and not highly personalized. We think the most likely explanation is that, as mentioned by Reviewer 2 below, the current analyses do not account for the role that environmental gradients play in the gut. If these environments differ asynchronously across hosts, it could lead to shared abundance correlations, but individualized microbiome compositions and individualized single-taxon dynamics. We discuss this possibility and other potential explanations in the revised Discussion (lines 572 to 586).

Reviewer #2 (Public Review):

The authors of this paper identify a knowledge gap in our understanding of the generalizability of ecological associations of gut bacteria across hosts. Theoretically, it is possible that ecological associations between bacteria are consistent within a host organism but differ between hosts, or that they are universal across hosts and their environmental gradients. The authors utilize longitudinal data with a unique temporal resolution, on Amboseli baboons, 56 individuals who were sampled for gut microbiome hundreds of times over a decade. This data allows disentangling ecological dynamics within and across individuals in a way that as far as I know has never been done before. The authors show that ecological relationships among baboon gut bacteria, measure through a correlation based on covariation, are largely universal (similar within and across host individuals) and that the most universally covarying taxa are almost always positively associated with each other. They also compare these results with two sets of human data, finding similar patterns in one human data set but not in the other.

The main aim of this paper is to establish whether gut microbial ecologies are universal across hosts, and this the authors generally show to be true in a thorough and convincing way. However, some re-assessment or re-assurance on the solidity of their chosen method of estimating co-variation would be needed to fully assess the robustness of subsequent results. Specifically, the authors measure the correlation between microbial taxa from data on their abundance co-variation across samples. While necessary steps have been taken to validate the estimates across spurious correlations due to the compositional nature and autocorrelation structures present in the data, I worry that the sparsity of the data might influence the estimation of positive and negative correlations in a slightly different manner. There exist more microbial taxa than samples in the data and some taxa are present in as few as 20% of the samples, meaning that the covariation data will have a large amount of 0-0 pairs. I worry that the abundance of 0-0 pairs in the data might inflate the measures of positive co-variation, making taxa seem highly positively correlated in abundance when they in fact are missing from many samples. Of course, mutual absence is also a form of biologically meaningful covariation but taking the larger number of taxa than samples and the inability of sequencing technology to detect all low-abundance taxa in a sample, I am currently not convinced that all of the 0-0 pairs are modeled as a realistic and balanced way as a continuum of the other non-zero co-variation between taxa in the data. This may become problematic when positive and negative relationships are compared: The authors state that even though most associations between taxa were negative, the most universally correlated taxa pairs (taxa pairs with strongest correlations in abundance both within and between hosts) were enriched in positive associations. It may be possible that this is influenced by the fact that zero inflation in the data lends more weight to positive links than negative links. Whether these universal positive correlations are driven by positive non-zero abundance covariation or just 0-0 links in the data is currently unclear.

Thank you for pointing out this weakness in our original analyses. As described in response R1 above, your hunch was correct: zero inflation biased our correlation patterns such that taxa pairs with a high frequency of joint zero observations (i.e., where both members of the pair had very low or zero abundances) tended to be positively correlated (Fig. R1). Consequently, as you suggested, zero inflation in the data lent more weight to positive links than negative links in our data set. To address this problem in the revised manuscript, we now restrict our analyses to taxon pairs whose joint zero-abundance observations were less than 5% of all samples across hosts (pairs to the left of the dashed vertical line in Fig. R1 above). We also restricted our analyses to taxa observed in at least 50% of all samples. The first of these criteria was the most restrictive. As described above, our new filtering procedure retained 1,878 of the original 7,750 ASV-ASV pairs; 57 of the original 66 phylum-phylum pairs; and 473 of the original 666 class/order/family-level pairs.

Another additional result that would benefit from a more clear context is the result that taxa correlation patterns were more similar between phylogenetically close taxa and between genetically close host individuals. The former notion is to be expected if taxa abundances are driven by environmental (or host physiology-related) selective forces that favor bacteria with similar phenotypes. This yields more support to the idea that covariation is environmentally driven rather than driven by the ecological network of the bacteria themselves, and this could be more clearly emphasized. The latter notion of covariation being more similar in genetically related hosts is currently impossible to disentangle from the notion that covariation patterns were more similar with individuals harboring a more similar baseline microbiome composition since microbiome composition and genetic relatedness were apparently correlated. To understand if something about relatedness was actually influential over correlation pattern similarity, one would need to model that effect on top of the baseline similarity effect. Currently, it is not clear if this was done or not.

We agree that shared responses to environmental gradients within hosts—especially immune profiles and pH—could explain both of these findings. These ideas are now described in the Discussion in lines 559 to 562.

We also now report partial Mantel tests to control for baseline similarity in microbiome composition when testing for shared microbial correlation patterns among genetic relatives. Controlling for baseline similarity had little effect on the results, and we now report the statistics for this partial Mantel (Fig. 5B; Table S7; r2=0.009; partial Mantel p-value=0.002). See lines 391-392.

The authors also slightly overemphasize the generalizability of their results to humans, taking that only one of the human data sets they compare their results to, shows similar patterns. While they mention that the other human data set (that was not similar in patterns to theirs) was different in some key aspects (sampling frequency was much higher), the other human data set was also dissimilar to the other two (it only contained infants, not adults). Furthermore, to back up the statement that higher sampling frequency would be the reason this data set had dissimilar covariation between taxa, one would need to show that the temporal variation in this data set was different from the baboon one and show that these covariation patterns were sensitive to timescale by subsampling either data to create mock data sets with different sampling frequency and see how this would change the inference of ecological associations.

We have revised the text to tone down the generalizability of our results to humans. For instance, the abstract (line 58) now states that “universality in baboons was similar to that in human infants, and stronger than one data set from human adults” but does not state that our results are generalizable to humans.

We also considered sub-sampling the data set from Johnson et al., from daily to monthly scales, but unfortunately that data set is only 17 days long, so doing so is impossible. This is now stated in the Discussion in line 619, which states, “However, without the ability to subsample Johnson et al. [7] to monthly scales (this data set is only 17 days long), it is impossible to test this prediction.”

To the extent that the results are robust, particularly regarding to the main result of the universality of gut microbial ecological associations, the impact of this paper is not small. This question has never been so thoroughly and convincingly addressed, and the results as they stand have the power to strongly influence the expectations of gut microbial ecology across many different systems. Moreover, as the authors point out, evidence for universal gut microbial ecology is important for the future development of probiotics. An important point here, underemphasized by the authors, is that universal gut microbe ecologies will allow specific interventions that use gut microbe ecology to manipulate emergent community properties of microbiomes to be more beneficial for the host, rather than just designing compositional cocktails that should fit all. In addition to the main finding of this study, the unique data set and the methods developed as part of this study (e.g. the universality score, the enrichment measures, the model of log-ratio dynamics, the assessment of covariation from time-ordered abundance trajectories) will doubtlessly be translatable to many other studies in the future.

Thank you for these suggestions. We now mention these implications in the introduction (line 82-84) and in the discussion in lines 537-539 and line 630.

Reviewer #3 (Public Review):

This is a well-executed study, offering thorough analysis and insightful interpretations. It is wellwritten, and I find the conclusions interesting, important, and well-supported.

Thank you for your supportive comments.

References

1. Silverman JD, Roche K, Holmes ZC, David LA, Mukherjee S. Bayesian Multinomial Logistic Normal Models through Marginally Latent Matrix-T Processes. Journal of Machine Learning Research. 2022;23:1-42.
2. Quinn TP, Richarrson MF, Lovell D, Crowley TM. propr: An R-package for Identifying Proportionally Abundant Features Using Compositional Data Analysis Scientific Reports. 2017;7:16252.
3. Cao Y, Lin W, Li H. Large covariance estimation for compositional data via compositionadjusted thresholding. . J Am Stat Assoc. 2019:759-72.
4. Friedman J, Alm EJ. Inferring correlation networks from genomic survey data. PLoS Comput Biol. 2012;8(9):e1002687. doi: 10.1371/journal.pcbi.1002687. PubMed PMID: 23028285; PubMed Central PMCID: PMCPMC3447976.
5. Risely A, Schmid DW, Muller-Klein N, Wilhelm K, Clutton-Brock TH, Manser MB, et al. Gut microbiota individuality is contingent on temporal scale and age in wild meerkats. Proc Biol Sci. 2022;289(1981):20220609. Epub 20220817. doi: 10.1098/rspb.2022.0609. PubMed PMID: 35975437; PubMed Central PMCID: PMCPMC9382201.
6. Wilmanski T, Diener C, Rappaport N, Patwardhan S, Wiedrick J, Lapidus J, et al. Gut microbiome pattern reflects healthy ageing and predicts survival in humans. Nat Metab. 2021;3(2):274-86. Epub 20210218. doi: 10.1038/s42255-021-00348-0. PubMed PMID: 33619379; PubMed Central PMCID: PMCPMC8169080.
7. Johnson AJ, Vangay P, Al-Ghalith GA, Hillmann BM, Ward TL, Shields-Cutler RR, et al. Daily Sampling Reveals Personalized Diet-Microbiome Associations in Humans. Cell Host & Microbe. 2019;25(6):789-802. Epub 2019/06/14. doi: 10.1016/j.chom.2019.05.005. PubMed PMID: 31194939.
8. Franzosa EA, Huang K, Meadow JF, Gevers D, Lemon KP, Bohannan BJM, et al. Identifying personal microbiomes using metagenomic codes. Proceedings of the National Academy of Sciences. 2015;112(22):E2930-E8. doi: 10.1073/pnas.1423854112. PubMed PMID: WOS:000355832200014.
9. Faith JJ, Guruge JL, Charbonneau M, Subramanian S, Seedorf H, Goodman AL, et al. The long-term stability of the human gut microbiota. Science. 2013;341(6141):1237439. Epub 2013/07/06. doi: 10.1126/science.1237439. PubMed PMID: 23828941; PubMed Central PMCID: PMC3791589.
10. Bik EM, Costello EK, Switzer AD, Callahan BJ, Holmes SP, Wells RS, et al. Marine mammals harbor unique microbiotas shaped by and yet distinct from the sea. Nat Commun. 2016;7:10516. Epub 20160203. doi: 10.1038/ncomms10516. PubMed PMID: 26839246; PubMed Central PMCID: PMCPMC4742810.
11. Caporaso JG, Lauber CL, Costello EK, Berg-Lyons D, Gonzalez A, Stombaugh J, et al. Moving pictures of the human microbiome. Genome Biology. 2011;12(5):R50. doi: Artn R50 Doi 10.1186/Gb-2011-12-5-R50. PubMed PMID: ISI:000295732700014.
12. Costello EK, Lauber CL, Hamady M, Fierer N, Gordon JI, Knight R. Bacterial community variation in human body habitats across space and time. Science. 2009;326(5960):1694-7. doi: Doi 10.1126/Science.1177486. PubMed PMID: ISI:000272839000053.
13. Dolinsek J, Goldschmidt F, Johnson DR. Synthetic microbial ecology and the dynamic interplay between microbial genotypes. Fems Microbiology Reviews. 2016;40(6):961-79. doi: 10.1093/femsre/fuw024. PubMed PMID: WOS:000387995000010.
14. Louca S, Polz MF, Mazel F, Albright MBN, Huber JA, O'Connor MI, et al. Function and functional redundancy in microbial systems. Nat Ecol Evol. 2018;2(6):936-43. Epub 2018/04/18. doi: 10.1038/s41559-018-0519-1. PubMed PMID: 29662222.
15. Rainey PB, Quistad SD. Toward a dynamical understanding of microbial communities. Philos Trans R Soc Lond B Biol Sci. 2020;375(1798):20190248. Epub 2020/03/24. doi: 10.1098/rstb.2019.0248. PubMed PMID: 32200735; PubMed Central PMCID: PMCPMC7133524. 16. Martiny JB, Jones SE, Lennon JT, Martiny AC. Microbiomes in light of traits: A phylogenetic perspective. Science. 2015;350(6261):aac9323. doi: 10.1126/science.aac9323. PubMed PMID: 26542581.
16. Debray R, Herbert RA, Jaffe AL, Crits-Christoph A, Power ME, Koskella B. Priority effects in microbiome assembly. Nat Rev Microbiol. 2022;20(2):109-21. Epub 20210827. doi: 10.1038/s41579-021-00604-w. PubMed PMID: 34453137.
17. Gloor GB, Reid G. Compositional analysis: a valid approach to analyze microbiome highthroughput sequencing data. Can J Microbiol. 2016;62(8):692-703. Epub 2016/06/18. doi: 10.1139/cjm-2015-0821. PubMed PMID: 27314511.
18. Joseph TA, Pasarkar AP, Pe'er I. Efficient and Accurate Inference of Mixed Microbial Population Trajectories from Longitudinal Count Data. Cell Syst. 2020;10(6):463-9 e6. Epub 20200624. doi: 10.1016/j.cels.2020.05.006. PubMed PMID: 32684275.
19. Aijo T, Muller CL, Bonneau R. Temporal probabilistic modeling of bacterial compositions derived from 16S rRNA sequencing. Bioinformatics. 2018;34(3):372-80. doi: 10.1093/bioinformatics/btx549. PubMed PMID: 28968799; PubMed Central PMCID: PMCPMC5860357.
20. Coyte KZ, Rao C, Rakoff-Nahoum S, Foster KR. Ecological rules for the assembly of microbiome communities. PLoS Biol. 2021;19(2):e3001116. Epub 20210219. doi: 10.1371/journal.pbio.3001116. PubMed PMID: 33606675; PubMed Central PMCID: PMCPMC7946185.
21. Coyte KZ, Schluter J, Foster KR. The ecology of the microbiome: Networks, competition, and stability. Science. 2015;350(6261):663-6. doi: 10.1126/science.aad2602. PubMed PMID:
23. Palmer JD, Foster KR. Bacterial species rarely work together. Science. 2022;376(6593):581-2. Epub 20220505. doi: 10.1126/science.abn5093. PubMed PMID:
25. Reese AT, Pereira FC, Schintlmeister A, Berry D, Wagner M, Hale LP, et al. Microbial nitrogen limitation in the mammalian large intestine. Nat Microbiol. 2018. Epub 2018/10/31. doi: 10.1038/s41564-018-0267-7. PubMed PMID: 30374168.
26. Firrman J, Liu L, Mahalak K, Tanes C, Bittinger K, Tu V, et al. The impact of environmental pH on the gut microbiota community structure and short chain fatty acid production. FEMS Microbiol Ecol. 2022;98(5). doi: 10.1093/femsec/fiac038. PubMed PMID: 35383853.
27. de Vos WM, Tilg H, Van Hul M, Cani PD. Gut microbiome and health: mechanistic insights. Gut. 2022;71(5):1020-32. Epub 20220201. doi: 10.1136/gutjnl-2021-326789. PubMed PMID: 35105664; PubMed Central PMCID: PMCPMC8995832.
28. Tamames J, Sanchez PD, Nikel PI, Pedros-Alio C. Quantifying the Relative Importance of Phylogeny and Environmental Preferences As Drivers of Gene Content in Prokaryotic Microorganisms. Front Microbiol. 2016;7:433. Epub 20160331. doi: 10.3389/fmicb.2016.00433. PubMed PMID: 27065987; PubMed Central PMCID: PMCPMC4814473.
29. Gloor GB, Macklaim JM, Pawlowsky-Glahn V, Egozcue JJ. Microbiome datasets are compositional: and this is not optional. Front Microbiol. 2017;8:2224. Epub 2017/12/01. doi: 10.3389/fmicb.2017.02224. PubMed PMID: 29187837; PubMed Central PMCID: PMCPMC5695134.
30. Caporaso JG, Lauber CL, Walters WA, Berg-Lyons D, Lozupone CA, Turnbaugh PJ, et al. Global patterns of 16S rRNA diversity at a depth of millions of sequences per sample. Proceedings of the National Academy of Sciences. 2011;108:4516-22. doi: Doi 10.1073/Pnas.1000080107. PubMed PMID: ISI:000288451300002.

Author Response

Reviewer #1 (Public Review):

This study was designed to examine the bypass of Ras/Erk signaling defects that enable limited regeneration in a mouse model of hepatic regeneration. The authors show that this hepatocyte proliferation is marked by expression of CD133 by groups of cells. The CD133 appears to be located on intracellular vesicles associated with microtubules. These vesicles are loaded with mRNA. The authors conclude that the CD133 vesicles mediate an intercellular signaling pathway that supports cell proliferation. These are new observations that have broad significance to the fields of regeneration and cancer.

The primary observation is that the limited regeneration observed in livers with Ras/Erk signaling defects is associated with CD133 expression by groups of cells. The functional significance of CD133 was tested using Prom1 KO mice - the data presented are convincing.

The major weakness of the study is that some molecular mechanistic details are unclear - this is, in part, due to the extensive new biology that is described. Nevertheless, the data used to support some key points in this study are unclear:

We fully agree that some details of the molecular mechanisms are yet to be elucidated for the CD133+ vesicles (intercellsomes, as we named). This is the first report of a new direct cell-cell communication mechanism provoked in stress response to proliferative signal deficit.

Despite a huge body of literature, many questions remain open for the molecular mechanisms of exosomes/EVs.

a) What is the evidence that the observed CD133 groups of cells are not due to clonal growth. Is this conclusion based on the time course (the groups appear more rapidly than proliferation) or is this based on the GFP clonal analysis?

This is indeed a very critical point for this study. Our initial thought and efforts were indeed on finding evidence that supports clonal expansion of progenitor cells. However, the experiments showed that the CD133+ cells were negative for all other stem/progenitor cell markers and that they are mature hepatocytes. CD133 expression was upregulated dramatically in regenerating livers and disappeared upon completion of liver regeneration. Furthermore, suppression of Ras-Erk signaling by Shp2 and Mek inhibitors robustly induced CD133 expression in a variety of cancer cell lines in culture in vitro.

At 2 days after PHx, we already observed big colonies, which were unlikely derived from a single initiating cell (Figure 1). The GFP clonal analysis unambiguously demonstrated the heterogenous origin of the clustered cells (Figure 3). We detected mixed GFP-positive and -negative cells within each colony, without a single colony consisting entirely of GFP-positive cells. The original colony sizes were estimated to be 10 cells or more (Figures 3G and Figure 3–figure supplement 1B). Thus, both the sizes and compositions in the GFP clonal analyses support the assertion that CD133+ cell clusters originated from multiple mature hepatocytes.

b) What is the evidence that the CD133 vesicles mediate intercellular communication. This is an exciting hypothesis, but what is the evidence that this happens? Is this inferred from IEG mRNA diversity? or some other data. Is there direct evidence of transfer - for example, the does the GFP clonal analysis show transfer of GFP that is not mediated by clonal proliferation? Moreover, since the hepatocytes are isogenic, what distinguishes the donor and recipient cells?

Increased clarity concerning what is hypothesis and what is directly supported by data - would improve the presentation of this study.

Per the reviewer’s advice, we have clarified these points in the revised version. Our proposal that CD133 vesicles mediate intercellular communication was supported by these experimental results.

A). Data in Fig. 5 suggest direct trafficking of the vesicles, as CD133 existed on the filaments that bridge the tightly contacting cells. This was confirmed by two different CD133 antibodies in mouse and human. We are now conducting correlative light and electron microscopy to characterize these bridges and the exchange event at the cell-cell border. Of note, CD133+ vesicles are negative for CD9, CD63 or CD81, markers for exosomes/EVs. We could only isolate CD133+ vesicles from cell lysates in vitro and mouse tissue lysates, but not from cell supernatants from which exosomes/EVs are isolated.

B). More direct evidence of the transfer was presented in Fig. 6H, showing Myc-tagged CD133 molecules transferred from one cell to another. We are engineering a knock-in system to track the endogenous CD133.

C). Further experimental evidence was provided in the single and double gene KO experiments in Fig. 8E-G, suggesting the functional significance of CD133 in intercellular communication.

D). In addition to the data above, the IEG mRNA diversity analyses based on scRNA-seq support the mRNA exchange model. The isogenic CD133+ SKO hepatocytes were found to lack different IEG transcripts randomly. This is why we propose a mutually sharing model, rather than a donor and recipient model. Importantly, the mRNA diversity (entropy) model also illustrates the association of CD133 and “stemness", as described in the discussion.

In sum, we believe that a most reasonable interpretation of the current data set is a model of direct cell-cell communication via CD133+ vesicles. We take the reviewer’s point and have made changes to the text to better distinguish conclusion and hypothesis, which will be validated in future studies.

Reviewer #2 (Public Review):

The manuscript by Kaneko set out to understand the mechanisms underlying cell proliferation in hepatocytes lacking Shp2 signals. To do this, the authors focused on CD133 as the proliferating clusters of cells in the Shp2 knockout (SKO) livers are CD133 expressing. After excluding the contribution of progenitors that are CD133 to this cell population, the authors focused on the intrinsic regulation of CD133 by Met/Shp2 regulated Ras/Erk parthway and showed upregulation of CD133 to be a compensatory signal to overcome loss of Ras/Erk signal and suggested Wnt10a in the regulation of CD133 signal. The study then focused on the observed filament localization of CD133 in the CD133+ cluster of cells. The study went on to identify the CD133+ vesicles that contain primarily mRNA vs. microRNA like other EVs. Specifically, the authors identified several mRNA species that encode IEGs, indicating a potential role for these CD133+ vesicles in cell proliferation signal transmission to neighboring cells via delivery of the IEG mRNAs as cargos. Finally, they showed that the induction of CD133 (and by derivative, the CD133+ vesicles) are necessary for maintaining cell proliferation in the cell cluster with high proliferation capacities in the SKO livers; and in intestinal crypt organoids treated with Met inhibitors to block Ras/ERk signal.

1) The identification of CD133+ vesicles is largely based on staining and costainings. Though the experiments are very well done with many controls and approaches, the authors may want to perform one or two key experiments with EM to definitively demonstrate the colocalization. For example, the mCherry experiment in Fig6H and the colocalization experiments for CD133 and HuR in Fig 7.

Many thanks for the suggestion. We are now establishing a correlative light and electron microscopy system, as the classic immunogold staining method was not sufficient for this purpose. Further characterization of CD133+ vesicles is now a major focus of research in our lab, to establish, substantiate or modify the model and hypothesis presented in this article. The long-term goal is to elucidate how cells strive to proliferate under insufficient proliferative signal, which is likely relevant to drug resistance and tumor relapse.

2) Since CD133+ marks the 50nM intracellsome defined by the authors, it is unclear what the CD133- vesicles used as controls are. Are they regular EVs that are larger in size? This needs better clarification as they are used as a control for many experiments such as Fig 7A.

Per the advice, we added more explanation to the revised text. We used regular EVs as the control, since they are the well-studied intercellular communication vesicles. Since the EVs are highly heterogenous, we did not choose to select a specific subpopulation of EVs. We used the well-established polymer-based precipitation method to isolate the EV fraction from cell culture supernatant for RNA-seq analysis. We did detect the enrichment of micro-RNAs in the isolated EVs, consistent with reports in the literature. Strikingly, the CD133 vesicles isolated from cell lysates showed a completely distinct RNA profile, relative to the EVs.

Author Response

Reviewer #1 (Public Review):

This paper raises an interesting question about learning signals. The most intriguing property of this system is the one-to-one convergence, plasticity, and apparently linear input/output function of the SFL-to-SAM relay. These properties suggest that, unlike structures like the insect mushroom body or mammalian cerebellum, in which the intermediate layer is thought to increase the dimensionality of the representation, the SAMs should be thought of more like the weights of a linear readout of the SFL inputs by the LNs.

What learning signal guarantees appropriate weight changes? In a few places (the section on "associativity" and the section on AFs), it is suggested that SAMs can themselves, through coordinated local activity, cause LTP, which the authors call "self LTP-induction." But what is the purpose of such plasticity? It doesn't seem like it would permit, for example, LTP which associates a pattern of SFL activity with the appropriate LNs for the correct vs. the incorrect action. Presumably, appropriately routed information from the NMs and AFs sends the appropriate learning signals to the right places. Does the pattern of innervation of NMs and AFs reveal how these signals are distributed across association modules? Does this lead to a prediction for the logic of the organization of the association modules?

We extended the discussion section to clarify some of these points. One paragraph describes our idea of “self-LTP induction” (L712-744). In addition, we address the potential role of the neuromodulatory fibers (NMs) and ascending fibers (AFs) in a paragraph titled "Perspective on the involvement of the ascending fibers (AF) and the neuromodulatory fibers (NM) in the supervision of learning" (L786). Answering how these signals manifest across different association modules requires a larger reconstruction.

One challenge for a reader who is not an expert on the VL is that the manuscript in its present form lacks discussion about the impact (or hypothesized impact) of the VL on behavior. There is a reference to a role for LNs suppressing attack behavior, but a more comprehensive picture of what the readout layer of this system is likely controlling would be helpful.

To contextualize how the VL circuitry can allow for the coincident detection of visual stimuli and environmental cues (punishing or rewarding) to control the stereotypic attack behavior of the octopus, we added two discussion sections: "Perspective on the VL involvement in octopus associative learning" (L774) and "Perspective on the involvement of ascending fibers (AF) and neuromodulatory fibers (NM) in the supervision of learning" (L786).

The authors do a thorough job of characterizing the "fan-out" architecture from SFL axons to SAMs and CAMs. A few key numbers remain to characterize the "fan-in" architecture of LNs. There appears to be a 400:1 convergence from AMs to LNs. Is it possible to estimate the approximate number of presynaptic inputs per LN? The text around Figure 7 states a median of 162 sites per 100μm dendrite length. One could combine this with an estimate of the total dendritic length for one of these cells from previously available data to estimate the number of inputs per LN. This would help determine the degree of overlap of different association modules in Figure 11, which would be interesting from a computational perspective.

Due to the limitations associated with a small EM volume, our study focused on the fanout of the VL network. We agree that a better understanding of the fan-in part of the VL network is crucial. To the best of our knowledge, previous data have not provided estimates on the dendritic length of the LNs, due to low-resolution images or lack of 3D imaging (Hochner/Shomrat/Young experiment). We intentionally avoided making a largely inaccurate estimate of the fan-in part of the network based on our data. We believe that future research can aim to combine neuron labeling with EM, or other super-resolution techniques, to allow for detailed assessment of the large neuron arborization.

Reviewer #2 (Public Review):

Octopuses are known for their abilities in solving complex tasks and numerous apparently complex cognitive behaviours such as astonishment at octopuses learning how to open jars by watching others and the mind-boggling camouflage. They are very clever molluscs. The octopus shows the famously advanced brain plan but it is one that has little research progress due to its large size and structural complexity. This was originally recognised by the work of BB Boycott, JZ Young, EG Gray, and others in mid last century. Since then, however, little progress has been achieved towards a modernday description of the octopus neural network particularly in the higher-order brain lobe, despite intense interest and indeed research progress concerning their complex behavioural and cognitive abilities.

This study applied a combination of EM-based imaging, neural tracing, and analyses to start revealing a further detailed view of a part of the lateral gyrus of the vertical lobe (learning and memory centre) of the common European octopus. It is a long overdue contribution and starts to bring octopus neuroscience a step close to the details of some vertebrates achieved. The new findings of neurons and the associated network provide new insights into this very complex but unfamiliar brain, allowing to propose a functional network that may link to the octopus memory formation. Also, this work could be of potential interest to a broad audience of neuroscientists and marine biologists as well as those in bio-imaging and deep-learning fields.

Strengths:

Current knowledge of the neuroanatomy and the associating network of the octopus vertical lobe (learning and memory centre) remains largely based on the pioneering neuroanatomical studies in the '70s, this work indeed provides a rich and new dataset using modern-day imaging technology and reveals numerous previously-unknown neuron types and the resulting further complex network than we thought before. This new dataset reveals hundreds of cell processes from seven types of neurons located in one gyrus of the vertical lobe and can be useful for planning further approaches for advanced microscopy and other approaches including electrophysiological and molecular studies.

Another strength of this study is to apply the current fashion of the deep learning technique to accelerate the imaging process on this octopus complex neural network. This could trigger some inventions to develop new algorithms for further applications on those non-model animals.

Weakness/limitations:

In an effort to match the key claims of the first connectome of the octopus vertical lobe, mapping up an entire vertical lobe is essential. However, also understandably, given challenges in imaging a large-sized brain region, this study managed to image a very small proportion of the anterior part of the lateral gyrus. Along with the current limited dataset, a partially reconstructed neural network of one gyrus, it is unclear whether the wiring pattern found in this study would appear as a similar arrangement throughout an entire lateral gyrus. Furthermore, it is also unknown if another 4 gyri might keep a similar pattern of neural network as it found in the lateral gyrus. Considering some recent immunochemistry evidence that showed distinct different signals in different gyri in terms of heterogeneity of neuron types amongst gryi, to assume this newly discovered network can represent the wiring pattern across an entire 5-gyrus vertical lobe is inadequate.

We revised the introduction (L106-113) to address this important point, and added discussion section titled "How well does a partial connectome of a small portion of one VL lateral lobule represent the connectivity patterns across all five VL lobuli?” (L894). We clarify what we believe is likely conserved across VL gyri and what is more likely to differ.

As this study is the first big step to reveal the complex network in the octopus vertical lobe system, the title may be changed to "Toward the connectome of the Octopus vulgaris vertical lobe - new insights into a memory acquisition network".

We appreciate the reviewer's suggestion for a new title for our manuscript. We feel, however, that the current title reflects the scope of our work and the significant step it makes toward understanding the neuronal network of the octopus vertical lobe. We do not claim to provide the octopus' VL connectome but how its connectomics unravels its workings and underlying principles. After deliberations, we decided to leave the title unchanged.

Author Response

Reviewer #1 (Public Review):

In this study the authors sought to address the issue of whether the Steller's sea cow -- a massive extinct sirenian ("sea cow") species that differs from its living relatives (manatees and dugongs) not only in body mass but also in having inhabited cold climates in the northern Pacific -- had hemoglobin adaptations that enhanced the species' thermoregulatory capacities relative to those of the extant species, which are restricted to relatively warm waters. To do so, the authors synthesized recombinant hemoglobin proteins of all the major sea cow lineages and used these data to assess differences in O2 binding, Hb solubility, responses to allosteric effectors, and thermal sensitivity. The work presented is very innovative and in my opinion convincingly demonstrates that the Steller's sea cow had remarkable hemoglobin adaptations that allowed for an extreme range extension into cool waters despite several physiological constraints that are inherent to the sirenian (and paenungulate, afrotherian, etc.) clade. I did not detect any obvious weaknesses of the paper, whereas the use of ancient DNA to resurrect 'extinct' hemoglobins, and the various analyses of these extinct hemoglobins alongside those of extant relatives is very exciting and are major strengths of the paper that make this study a very important advance for our understanding of Steller's sea cow's paleophysiology, as well as our understanding of the potential for extreme hemoglobin phenotypes that have not been documented among living species. Moving forward, these methods can be used to study aspects of the paleophysiology of other recently extinct mammals. I applaud the authors on an excellent and innovative study that significantly augments our understanding of the Steller's sea cow.

We sincerely appreciate the constructive comments of this reviewer.

Reviewer #2 (Public Review):

This manuscript is an impressive "resurrection" of physiology regarding an enigmatic though unfortunately extinct species, and their potential adaptation to cold-water environments. I am largely convinced of their findings, which I feel are very straightforward and thorough.

One place where the authors perhaps fell a bit short was regarding some conclusions associated with maternal/fetal oxygen delivery. The sirenian versions of fetal & embryonic hemoglobin genes have been identified and assessed to some degree in previously published work the same research group. I feel the manuscript would have benefited from actual analysis of the fetal & embryonic hemoglobin (epsilon, gamma, zeta) to strengthen their assertions.

Again, we appreciate the kind words and valid concern of this reviewer regarding a potential shortcoming of the maternal/fetal gas exchange discussion. As noted above, we previously collected physiological data from two pre-natal Steller’s sea cow Hb isoforms that were initially intended to form a stand-alone publication on this topic. However, we have elected to include this data here to better support our claims and provide a more complete picture of maternal/fetal oxygen delivery in this extinct species.

Reviewer #3 (Public Review):

Signore et al. synthesized and functionally characterized the recombinant adult hemoglobin (Hb) proteins of extant, extinct, and ancestral sirenians to explore the putative role of Hb in helping Steller's sea cows adapt to life in extremely cold waters. The functional comparisons show that the Hb of the subarctic Steller's sea cows differs in multiple biochemical properties relative to the Hbs of the two extant sirenians in the study, the Florida manatee, and the dugong and also from the Hb inferred for the common ancestor of Steller's sea cow and dugong. Specifically, the Steller's sea cow shows reduced oxygen binding affinity, reduced sensitivity to the allosteric cofactors DPG, Cl-, and H+, increased solubility, and reduced thermal sensitivity. DPG plays an important role in regulating Hb oxygen affinity in mammals, and the lack of sensitivity to it is unique to the Hb of Steller's sea cow. Sequence comparisons show that the Hb of the Steller's sea cow differs at 11 amino acids from that of its sister group, the dugong, one of which is intriguing because it occurs in a position that is invariable among mammals at a site that is critical for DPG binding, a change from Lys to Asn in position 82 of the mature β/δ globin chain. To test the significance of this change, the authors use site directed mutagenesis to insert back a Lys in the Steller's sea cow Hb background (β/δ82Asn→Lys) and test its biochemical properties. The functional assays with the β/δ82Asn→Lys mutant indicate that reverting this position to its ancestral state drastically altered the biochemical properties of the Steller's sea cow Hb, making it functionally similar to the Hbs of manatee, dugong, and the Hb inferred for the common ancestor of Steller's sea cow and dugong.

The study's strength lies in comparing the different recombinant Hbs in an explicit evolutionary framework. The conclusions are supported by the analyses, and the results are relevant in the fields of evolutionary biology, physiology, and biochemistry because they suggest that a single amino acid substitution in a protein can have profound biochemical consequences that impact whole organism physiology.

We concur with the excellent synopsis of this reviewer. The finding that most of the functional differences between Steller’s sea cow and other sirenian Hbs can be attributed to a single amino acid replacement mirrors earlier sentiments of hemoglobin adaptation by pioneers in the field (e.g. Max Perutz). By contrast, more recent studies highlight the importance of multiple causative replacements of smaller effect and the significance of genetic background in hemoglobin evolution/adaptation (which is also evident for Steller’s sea cow Hb). We hope that the present work helps to bridge these two important evolutionary forces.

Author Response

Reviewer #3 (Public Review):

The authors described the one family showing autoinflammatory phenotypes with L236P variant of TNFAIP3 gene. The variant has not been reported on and they evaluated the function of this variant using in vitro and in silico methods. I think this is well-written manuscript and I agree with their interpretation about the pathogenicity of this variant, but the new finding is poor. The variant information was only a new finding.

I recommend the revision of the following points.

In Table 2, T647P seemed to be pathogenic which was evaluated with in vitro assay by Kadowaki.

The Kadowaki study indeed showed reduced NFκB activity for the Thr647Pro variant. This information has now been added to Table 2. However, the variant is highly frequent in the control population. According the ACMG guidelines, the variant does not fulfil all the conditions to be considered as pathogenic as its allele frequency is not compatible with the disease frequency. Therefore, as we cannot conclude on the pathogenic effect of the variation, we have described it as a variant of unknown significance.

Two other missense variants, V377I (Niwano, Rheumatology 2022) and T602S (Jiang W, Cellular Immunol 2022) were recently reported. These should be included in the discussion.

We have analyzed all additional missense variations, including the V377M (we found the report of a variation involving V377, but it was V377M and not V377I) and T602S variations and have added them to Table 2.

Author Response

Reviewer #2 (Public Review):

Granell et al. investigated genetic factors underlying wheezing from birth to young adulthood using a robust data-driven approach with the aim of understanding the genetic architecture of different wheezing phenotypes. The association of 8.1 million single nucleotide polymorphisms (SNPs) with wheeze phenotypes derived from birth to 18 years of age was evaluated in 9,568 subjects from five independent cohorts from the United Kingdom. This meta-genome-wide association study (GWAS) revealed the suggestive association of 134 independent SNPs with at least one wheezing subtype. Among these, 85 genetic variants were found to be potentially causative. Indeed, some of these were located nearby well-known asthma loci (e.g., the 17q21 chromosome band), although ANXA1 was revealed for the first time to play an important role in early-onset persistent wheezing. This was strongly supported by functional evidence. One of the top ANXA1 SNPs associated with wheezing was found to be potentially involved in the regulation of the transcription of this gene due to its location at the promoter region. This polymorphism (rs75260654) had been previously evidenced to regulate the ANXA1 expression in immune cells, as well as in pulmonary cells through its association as an eQTL. Protein-protein network analyses revealed the interaction of ANXA1 with proteins involved in asthma pathophysiology and regulation of the inflammatory response. Additionally, the authors conducted a murine model, finding increased anxa1 levels after a challenge with house dust mite allergens. Mice deficient in anxa1 showed decreased lung function, increased eosinophilia, and Th2 cell levels after allergen stimulation. These results suggest the dysregulation of the immune response in the lungs, eosinophilia, and Th2-driven exacerbations in response to allergens as a result of decreased levels of anxa1. This coincides with evidence of lower plasmatic ANXA1 levels in patients with uncontrolled asthma, suggesting this locus is a very promising candidate as a target of novel therapeutic strategies.

Limitations of this piece of work that need to be acknowledged:

(1) the manual and visual inspection of Locus Zoom plots for the refinement of association signals and identification of functional elements does not seem to be objective enough;

This is an important observation and we have now added the following text in the Discussion which can be found on lines 400-2 Revised Main Manuscript:

“Finally, the manual and visual inspection of Locus Zoom plots for the refinement of association signals and identification of functional elements was an objective approach which might have undermined the findings.“

(2) the sample size is limited, although the statistical power was improved by the assessment of very accurate disease sub-phenotype;

This point was already mentioned as a limitation and it can now be found in lines 349-365 Revised Main Manuscript:

“By GWAS standards, our study is comparatively small and may be considered to be underpowered. The sample size may be an issue when using an aggregated definition (such as “doctor-diagnosed asthma”) but is less likely to be an issue when primary outcome is determined by deep phenotyping. This is indirectly confirmed in our analyses. Our primary outcome was derived through careful phenotyping over a period of more than two decades in five independent birth cohorts, and although comparatively smaller than some asthma GWASs, our study proved to be powered enough to detect previously identified key associations (e.g. chr17q21 locus). Precise phenotyping has the potential to identify new risk loci. For example, a comparatively small GWAS (1,173 cases and 2,522 controls) which used a specific subtype of early-onset childhood asthma with recurrent severe exacerbations as an outcome, identified a functional variant in a novel susceptibility gene CDHR3 (SNP rs6967330) as an associate of this disease subtype, but not of doctor-diagnosed asthma(51). This important discovery was made with a considerably smaller sample size but using a more precise asthma subtype. In contrast, the largest asthma GWAS to date had a ~40-fold higher sample size(7), but reported no significant association between CDHR3 and aggregated asthma diagnosis. Therefore, with careful phenotyping, smaller sample sizes may be adequately powered to identify larger effect sizes than those in large GWASs with broader outcome definitions(52).”

(3) association signals with moderate significance levels but with strong functional evidence were found;

We do not think of this as a limitation but as a strength. We were able to support our genetic results with evidence from experimental mouse models.

(4) no direct replication of the findings in independent populations including diverse ancestry groups was described.

This point was already mentioned as a limitation and it can now be found in lines 375-391 and 392-399 Revised Main Manuscript.

“We are cognisant that there may be a perception of the lack of replication of our GWAS findings. We would argue that direct replication is almost certainly not possible in other cohorts, as phenotypes for replication studies should be homogenous(56). However, there is a considerable heterogeneity in LCA-derived wheeze phenotypes between studies, and although phenotypes in different studies are usually designated with the same names, they differ between studies in temporal trajectories, distributions within a population, and associated risk factors(57). This heterogeneity is in part consequent on the number and the non-uniformity of the timepoints used, and is likely one of the factors responsible for the lack of consistent associations of discovered phenotypes with risk factors reported in previous studies(58). This will also adversely impact the ability to identify phenotype-specific genetic associates. For example, we have previously shown that less distinct wheeze phenotypes in PIAMA were identified compared to those derived in ALSPAC(59). Thus, phenotypes that are homogeneous to those in our study almost certainly cannot readily be derived in available populations. This is exemplified in our attempted replication of ANXA1 findings in PIAMA cohort (see OLS, Table E12). In this analysis, the number of individuals assigned to persistent wheezing in PIAMA was small (40), associates of this phenotype differed to those in STELAR cohorts, and the SNPs’ imputation scores were low (<0.60), which meant the conditions for replication were not met.”

“Our study population is of European descent, and we cannot generalize the results to different ethnicities or environments. It is important to highlight the under-representation of ethnically diverse populations in most GWASs(9). To mitigate against this, large consortia have been formed, which combine the results of multiple ethnically diverse GWASs to increase the overall power to identify asthma-susceptibility loci. Examples include the GABRIEL(6), EVE(60) and TAGC(7) consortia, and the value of diverse, multi-ethnic participants in large-scale genomic studies has recently been shown(61). However, such consortia do not have the depth of longitudinal data to allow the type of analyses which we carried out to derive a multivariable primary outcome.”

Nonetheless, the robustness and consistency of the findings supported by different analytical and experimental layers is the major strength of this study.

The authors successfully achieved the aims of the study, strongly supported by the results presented. This study not only provides an exciting novel locus for wheezing with potential implications in the development of alternative therapeutic strategies but also opens the path for better-powered research of asthma genetics, focused on accurate disease phenotypes derived by innovative data-driven approaches that might speed up the process to disentangle the missing heritability of asthma, making use of still useful GWAS approaches.

Author Response

Reviewer #1 (Public Review):

This study aimed to estimate contact parameters associated with the transmission of SARS-CoV-2 in unvaccinated South African households over one year. The authors found no correlation between the frequency or duration of contacts and infection risk. Similar parameters (e.g., sharing a room with the index patient) also failed to yield an association. Reassuringly, a robust association was found with the Ct of the index case; female sex and individuals aged 13-17 years were also associated with increased risk. In a more general analysis, obesity, age >5 and <60 y, and non-smoking status were associated with increased risk.

Strengths of the study are its relatively large size (131 households involving 497 people) with detailed proximity data; frequent testing to enable high ascertainment of infections; and ability to exclude individuals seropositive at baseline. Additionally, several outcomes were evaluated in the models, partly to accommodate uncertainty in the index case. Different model structures were evaluated to gauge robustness.

Limitations of the study include the fact that many index cases were likely enrolled after their infectious period, and it is possible that apparent secondary cases in the household arose from a shared exposure with the index case but had a longer latent period. Each of these factors could weaken the perceived effect of close contacts. Statistically, there is the vexing question of what age (gender, smoking, etc.) really represents mechanistically, and whether the models may be conditioning on a collider. Another statistical consideration is that many household contacts were excluded from the study because they were seropositive at baseline. In effect, their households may already have been "challenged" with the virus, and there may be heterogeneities in household susceptibility that are not fully considered by the simple exclusion of individuals with evidence of prior infection. Separating these household types in the analysis might have yielded different results.

Although conditioning on a collider in the multivariable analysis is theoretically possible, it would be important to consider these co-variates as they had been found to be associated with transmission in both this and previous studies. The fact that there is also no signal for the contact parameters on univariate analysis support these findings.

We added another sensitivity analysis where any households where individuals were excluded due to seropositivity was excluded from the analysis. The results showed that none of the contact patterns were significantly associated with SARS-CoV-2 transmission in the household.

All that said, it is telling that in these households, infection is not clearly linked to typically defined close contacts. This is an important result that complements other strong evidence that aerosols are the dominant route of transmission for SARS-CoV-2. This information is critical for the design of effective intervention strategies. Additionally, the authors outline how future studies can be designed to improve on this work.

Reviewer #3 (Public Review):

The manuscript by Kleynhans et al analyzes data from household contacts of SARS-CoV-2 cases at two sites in South Africa. Proximity sensors were distributed to household members following diagnosis of the "index case" and measured the frequency and duration of close contacts (defined as being face-to-face within 1.5 meters for at least 20 seconds). The authors then examined the association between the duration, frequency, and average duration of contacts and the risk of a diagnosis of SARS-CoV-2 among household members in the subsequent two weeks, for both contact with the index case and all cases within the household. The risk of infection among household members was high (~60%), but was not significantly associated with the contact metrics examined. The findings may indicate that aerosols may be the predominant mode of SARS-CoV-2 transmission within households; however, there are also a number of limitations associated with the design and analysis of the study, which the authors acknowledge and which may limit the interpretability of the conclusions of this study.

One important study limitation has to do with the design of the study: Sensors were not distributed to household members until a day or two after the diagnosis of the index case. Since individuals are most infectious with SARS-CoV-2 just prior to symptom onset, contact patterns were measured only after most transmission from the index case likely occurred. Furthermore, household members may have limited their contact with the index case, particularly if the index case attempted to isolate following their diagnosis, so the contact patterns measured are unlikely to be representative of typical mixing within the household.

Another important limitation has to do with the analytical approach: The logistic regression model assumes that the first person in the household to test positive for SARS-CoV-2 (i.e. the index case) infected all subsequent cases within the household. However, this approach does not account for chains of transmission within the household or transmission from outside the household (possibly from the same source that infected the index case). While this concern is partially addressed by also assessing the association between the risk of infection and contact with all infected household members, more sophisticated methods could be used to infer the most likely infector of each case. The possibility of multiple introductions of the virus from outside the household is also only partially addressed by excluding households in which more than one variant was detected. While these limitations (and others) are appropriately acknowledged by the authors in the Discussion, nevertheless they limit the conclusions that can be drawn from the study results.

We will be considering an analytic framework to distinguish the chains of transmission in future analyses.

It is also worth noting that the contact metrics as defined and analyzed in the model may not be the measures that are most relevant to transmission. The authors examined three different contact measures: the median daily duration of contact, the median daily frequency of contact, and the median daily average duration of contact (i.e. the ratio of the two previous measures). They chose to examine the median daily values because contact duration was heavily skewed and the number of days of follow-up varied after data cleaning, but it may be that longer-duration contacts important to transmission are not appropriately captured by these metrics. Indeed, the median daily duration of the contact is quite short (only ~18 minutes on average). It would be useful to also evaluate a measure such as the total cumulative duration of contact and frequency of contacts divided by the number of days of follow-up, which differs from the measures they calculate and would take into account more prolonged and frequent contacts.

Additional contact parameters added as described in response to reviewer 2.

Lastly, the measures of association reported in the manuscript are the odds ratios (ORs) associated with one additional second of contact per day. This is not a very biologically meaningful unit of measure, and when rounded to two significant digits, the ORs are not surprisingly 1.0 with 95% confidence intervals that also round to 1.0. It would be more interpretable to report the ORs associated with a 1-minute (rather than 1-second) increase in the duration of contact, and the biological interpretation of the ORs should be described in the text.

All time-based contact parameters now expressed in minutes.

Author Response

Reviewer #1 (Public Review):

Rapan et al. analyzed the cytoarchitectonic of the prefrontal cortex based on observer-independent analysis, confirming previous parcellations based on cyto-, myelo-, and immunoarchitectonic approaches, but also defining novel subdivisions of areas 10, 9, 8B, and 46 and identified the receptor density "fingerprint" of each area and subdivision. Furthermore, they analyzed the functional connectivity of the prefrontal cortex with caudal frontal, cingulate, parietal, and occipital areas to identify specific features for the various prefrontal subdivisions. Altogether, this study corroborates previous parcellations of the prefrontal cortex, adds new cortical subdivisions, and provides a neurochemical description of the prefrontal areas useful for comparative considerations and for guiding functional and clinical studies.

Strengths:

• This study provides a detailed cytoarchitectonic map of the prefrontal cortex enriched with receptor density and functional connectivity data.

• The authors shared the data via repositories and applied their map to a macaque MRI atlas to further facilitate data sharing.

Weaknesses:

• The temporal cortex should be included in the functional connectivity analysis as it is known from anatomical studies that most prefrontal areas display rich connectivity with temporal areas. The aim of creating a comprehensive view of the frontal cortex makes the manuscript data-rich but cursory in discussing the relevant anatomical and functional literature.

One of the main concerns pointed out by reviewers was that the functional connectivity analysis is incomplete without temporal lobe areas. Although our initial decision was to use only our parcellation scheme, we fully agree with reviewers. Thus, we have extended our functional connectivity analysis, and combined our frontal, parietal, cingulate and occipital parcellations with temporal areas as defined in the atlas of Kennedy and colleagues (Markov et al., 2014). In the revised version of the manuscript, old Figures 13-17, and related Supplementary Figures 12 and 13, have been replaced with new Figures 12-15 and Figures 12-15 – Figure supplements 12, 13 and 14, and the results are described in the updated Results, chapter 3.4 Functional connectivity analysis. The Discussion has also been adjusted regarding the updated results.

Reviewer #2 (Public Review):

Rapan and colleagues did perform an impressive multi-modal parcellation of the macaque frontal cortex. In addition to qualitative cytoarchitectonic and resting-state functional fMRI data analyses, the authors based their parcellation on quantitative receptor density analysis of 14 receptors. Compared with the classic Walker map of the macaque frontal cortex, the authors produced a more refined map. Those results should be discussed in light of previous work on the same topic (Petrides et al. 2012 Cortex; Reveley et al. 2017 Cerebral Cortex; Saleem and Logothetis 2012).

In the Discussion, under chapter “4.1 Comparison with previous architectonic maps of macaque prefrontal region” (pages 44-52), we compared our parcellation to previously published maps, including the work of Petrides and colleagues (i.e., Petrides and Pandya 1984, 1994, 1999,2002, 2006, 2009; Petrides 2000, 2005; Petrides et al. 2012). With the exception of Caminiti et al. (2017), which integrates work by Belmalih et al. (2009); Borra et al. (2011, 2019) and Gerbella et al. (2010, 2013), we had restricted our citations to original mapping studies because we find it is important to discuss their reliability and objectivity, since they have been widely used in tracer-tract and neuroimaging studies, as well as the parcellation maps depicted in 3D atlases. Indeed, Saleem and Logothetis (Saleem & Logothetis, 2012) use the maps of Carmichael and Price (1994), Petrides and Pandya (1999, 2002) Petrides (2005) and Preuss and Goldman-Rakic (1991) for the parcellation of the prefrontal cortex in their atlas, and Reveley et al. (Reveley et al., 2017) use the map of Saleem and Logothetis (2012) in their 3D atlas. We now provide this information in the Introduction (lines 126-132):

“In recent years, several digital macaque atlases have been created (Bezgin et al., 2012; Frey et al., 2011; McLaren et al., 2009; Moirano et al., 2019; Reveley et al., 2017; Van Essen et al., 2012) based on the previous parcellations. Indeed, maps of Carmichael and Price (1994), Petrides and Pandya (1999, 2002), Petrides (2005) and Preuss and Goldman-Rakic (1991), used in atlas of Saleem and Logothetis (2012), have been brought into stereotaxic space by Reveley et al. (2017).”

Author Response

Reviewer #1 (Public Review):

The authors have approached the study of the mechanism of maturation of retroviruses lattice, where Gag polyprotein is the main component. The Gag polyprotein is common to all retroviruses and makes up most of the observed lattice underlying the virion membrane. Within the lattice, 95% of the monomers are Gag, and 5% are Gag-Pol, which has the 6 domains of Gag followed by protease, reverse transcriptase and integrase domains (coming from Pol) embedded within the same polyprotein. For the maturation and infectivity of HIV retrovirus, the Gag proteins within the immature lattice must be cleaved by the protease formed from a dimer of Gag-Pol. Importantly, the lattice covers only 1/3 to 2/3 of the available space on the membrane. The incompleteness of the lattice results in a periphery of Gag monomers with unfulfilled intermolecular contacts. Recently, the structure of the immature lattice has been partially resolved using sub-tomogram averaging cryotomography (cryET) and it has been shown that the incompleteness of the lattice provides more accessible targets for the protease (Tan A. et al. 2021). Based on these, the authors have wondered: does the incompleteness of the lattice allow for dynamic rearrangements that ensure that protease domains embedded within the lattice can find one another to dimerize and activate? To answer this, they started from experimental cryoET data and used reaction-diffusion simulations of assembled Gag lattices with varying energies and kinetic rates to test how lattice structure and stability can support the dimerization of the Gag-Pols. They found that although they represent only 5% of the monomers that assemble into the lattice, the stochastic assembly ensure that at least a pair of them are adjacent within the lattice. They next showed that if the molecules are distant from one another, they would need to detach, diffuse, and reattach stochastically at the site of another Gag-Pol molecule.

I consider the work very interesting, which could contribute to a very important aspect of retroviruses maturation such as their infectivity. However, the observations made by the authors do not necessarily answer their initial question which seemed to be focused on studying the possible role of the incompleteness of the lattice on the protease activation rather than the mechanism of Pol activation itself. Maybe this is only a nuance to be polished in the writing.

The weakness of the work comes from both the fact their entire study has been done by computational methods and the exclusion in their computational approaches of well-known cellular components with a role in retrovirus maturation, which might obey to the fact of keeping their models into the simplest possible since handling atomistic models is already a heavy task. Maybe complementary molecular or structural studies would strengthen their results.

We appreciate Reviewer #1’s comments and interest in our work. In the revised manuscript, we have clarified our writing to emphasize that our primary goal was to quantitatively interrogate the mechanism by which dimerization of two protease domains can occur, as it is essential for activation of the protease and the subsequent maturation process. We do not address the steps that follow the essential dimerization process. The molecular details concerning how this (dimerized) protease enzyme initiates cleavage and ultimately cleaves the entire lattice off the membrane would be the subject of a separate study (see page 4).

Regarding the concern about the computational nature of our work, we agree that the model is a (necessarily) simplified representation of the true biological system, and as we elaborate further in the Discussion section, including more components in the model would strengthen connection to experiment, which we plan in future work (see page 26). However, we have not relied solely on our computational results, but made several direct and quantitative comparisons to experimental structural (now!), biochemical, and imaging data. We have also validated the parameters of our model against theory.

We agree that it would be interesting and worthwhile to make more direct comparisons as well to more molecular models of the immature lattice, particularly in future work with the inclusion of more specific co-factors like IP6, which we discuss in the Discussion section. We show that already in its current form, our model provides new quantitative evidence on the mechanisms that would allow two protease domains to find one another to initiate maturation, in a way that is consistent with structural and biochemical data. (See revisions on page 26 and 28 of the manuscript)

Reviewer #2 (Public Review):

Immature lattice assembly remains an arcane topic, and these simulations provide high resolution data such as assembly kinetics and large-scale lattice rearrangement. Further, the authors extend their model to compare directly with experiments, e.g. SNAP-HALO dimerization, which provides a basis to interpret their conclusions. The manuscript is difficult to read, as it is a technical manuscript that overuses jargon; overall, it seems written for a specialized audience. Additionally, there are several aspects of the model design that remain opaque, such as the implicit lipid method and the suppression of multi-site nucleation. Further, analyses such as time auto-correlation and mean first passage time are not given much context by the authors. Altogether, it is the opinion of this reviewer that several revisions to the manuscript should be incorporated to improve clarity and strengthen the significance of the authors' efforts.

We appreciate the constructive comments from Reviewer #2. We've revised the text in multiple places to minimize the use of technical jargon and provide clearer explanations for specialized terms or concepts. Specifically, we've provided more detailed descriptions of the implicit lipid method and the rationale behind the suppression of multi-site nucleation to help readers better understand our model design. Additionally, we've added more context for the analyses such as time auto-correlation and mean first passage time, describing their significance and relevance to our study.

Reviewer #3 (Public Review):

The manuscript concerns the cleavage of the Gag polyprotein lattice from the HIV virion membrane, a key stage in HIV lifecycle, and one that is required for HIV to become infectious. Since cleavage requires homodimerization between the small fraction (5%) of such Gag polyproteins that carry a protease domain, referred to as Gag-Pol, this raises questions regarding how such homodimerization can take place, and whether it can happen on the required timescales, given that Gag-Pol is typically embedded in a lattice that is observed to form one large connected component.

The authors address these questions in silico, using particle-based reaction diffusion simulations. Such simulations are rigid-body and "structure-resolved" meaning that they rigidly incorporate the geometry of the polyproteins, and their various binding interfaces, based on existing structural data. Other aspects of the simulations are also in-line with available data, including copy numbers, lattice curvature, and dissociation rates. This focused approach is a strength of the work and allows the authors to make credible claims that their simulations have relevance to HIV (as does their commitment to comparison with HALO-SNAP-based measurements of dimerization kinetics as well as iPALM experiments that characterize lattice dynamics).

A central part of the model is that it allows for the "possibility of imperfect alignment of molecules in the lattice", presumably due to the incompatibility of regular hexagonal tiling and surfaces with non-zero Gaussian curvature, such as a sphere. This is implemented via the ad-hoc imposition of a free-energy penalty when complete hexamers are formed, implying that hexamers are less stable than six ideal bonds. By varying this strain penalty, the authors can change the stability of the lattice independently of individual binding affinities, allowing its use as an effective fitting parameter when comparing to HALO-SNAP data. In the latter case, agreement between simulation and experiment can only be found at moderate levels of lattice stability.

However, such energetic penalties are present whenever the polyprotein structure must undergo deformations which, on surfaces with nonzero Gaussian curvature, should be the case for partial tilings as well as complete ones (where all six interfaces form bonds). This, therefore, appears to be a weakness of the work. An elastic implementation of polyprotein structure, for example, would permit strain to accumulate (and therefore stresses to propagate) throughout the lattice naturally, irrespective of whether complete hexamers were formed, and might reasonably be expected to impact the likelihood of different lattice structures. Whilst it is not clear how or whether this would lead to qualitatively or quantitatively different results, it is nevertheless worth remarking upon since the authors high-level claim is that lattice structure is an important determinant of the mean-firstpassage times to dimerization.

Overall, I find this to be a valuable study, carried out in a solid and comprehensive manner. The primary impact of the work appears to be twofold: the unification of different experimental measurements under a single model, and the further identification of the salient parts of that model that most impact biological function. The results advance the understanding of one of the steps of the HIV lifecycle, via a better description of the mechanisms underpinning Gag-Pol dimerization. Notably, the authors stop short of drawing parallels to many related concepts and models in statistical physics, such as those concerning percolation and diffusion limited aggregation as well as the notions of dislocations and defects in crystalline matter on curved surfaces. These might reasonably have provided a basis for better understanding and quantification of the authors' simulations, as well as improving the scope for extensions and conceptual clarity.

We appreciate the constructive and positive comments from Reviewer #3. We have revised the text and expanded the discussion given the points the reviewer has brought up.

We have quantified the number and organization of incomplete hexamers or defects in the simulated lattices. This allows us to compare with experimental structures and also quantify how the assembly parameters would impact this organization. As we now remark on pages 10-11, with reversible binding during assembly, we see that fewer defects are present in the lattice, indicating that the hexameric lattice can improve its organization and stability when unbinding reactions can correct for weak contacts in the lattice. We thus speculate that because our lattices are statistically in good agreement with experiment even when binding is irreversible, that the assembly process does not rely on a significant amount of annealing. Otherwise the lattice structures would be more ideal.

We further discuss on pages 27 that a model that also incorporates forces to control interactions would be important to measure the mechanical stability of the lattice, particularly as it couples to membrane bending. However, models that naturally incorporate forces (via interaction potentials) can be difficult to tune with respect to their kinetics and free energies, which are nontrivial to calculate, unlike our modeling approach here.

In these sections we have now drawn parallels, as suggested by the reviewer, to the literature on defects in crystalline matter on curved surfaces, and their possible consequences for the mechanics of the lattice (Ref 41 Negri et al, Deformation and failure of curved colloidal crystal shells. Proc Natl Acad Sci U S A, 2015. 112(47)) (Ref 59 Zandi, R. and D. Reguera, Mechanical properties of viral capsids. Phys Rev E Stat Nonlin Soft Matter Phys, 2005. 72(2 Pt 1): p. 021917.)

We did not include a connection to Diffusion-limited aggregation, as this process seems to result in fractal-like structures that lack the specific hexagonal order of the Gag protein lattice. The proteins impose a significant orientational order on the assembly process that makes growth significantly more compact, at least under the assembly conditions we used. Even for our fastest rates, the process is still only moderately diffusion influenced (ka <<109M-1s-1), with typically multiple collisions needed before succesful binding occurs, consistent with most protein-protein interactions.

Author Response

Reviewer #2 (Public Review):

The manuscript by Mohebi et al. examines a critical open question regarding the interaction of cholinergic interneurons of the striatum and transmitter release from dopaminergic axons in behaving animals. Activation of cholinergic interneurons in the striatum can evoke dopamine release in brain slices and in vivo as measured with voltammetry. However, it remains an open question in what context and to what extent this acetylcholine-mediated dopamine occurs in behaving animals. Here, the authors argue that CIN activity triggers dopamine release in the nucleus accumbens which encodes the motivation to obtain a reward through increasing "ramps" of dopamine release. Their data suggest that the ramps are not reflected in the firing of dopaminergic neurons. Rather, they provide compelling evidence that the ramps of dopamine release correlate with ramps in cholinergic interneuron activity as measured with GCaMP6. What's more, the authors show that ACh-mediated dopamine release has no paired-pulse depression, a striking result that differs from all prior ex vivo brain slice data. The manuscript is extremely well written and the data are of very high quality. Overall, this study represents an important step forward in our understanding of how ACh-mediated dopamine release regulates behavior, and more broadly how axons can generate behaviors independently from somatic activity.

Major comments

1) The complete absence of any short-term plasticity in CIN-mediated dopamine release is a striking result that is important for the field. The authors should strengthen this result with additional quantitative analysis demonstrating the lack of STP. They have analyzed paired-pulse ratios, but they should analyze this for stimuli at the higher frequencies (4 Hz, etc) that are more physiologically relevant. For example, Fig 1e shows a CIN-evoked DA release at many optically-stimulated frequencies. The authors should quantify short-term plasticity by generating fits of the single stimulus signal and comparing the mathematical sum predicted from 4 stim DA signals at different frequencies to the recorded data. A similar analysis has been done with Ca signals (Koester and Sakmann, 2000).

Thank you for this very helpful suggestion. We have performed this analysis as recommended, and now confirm the lack of STP even at the higher frequencies (see new Supplementary Figure 1).

2) The authors show that optical activation of CINs results in DA release as measured by dLight. To clearly establish that these signals are generated by DA release driven by nicotinic receptors (and not a partial effect of some unknown artifact), it would be useful to show that the optical CIN-evoked dLight signals shown in Fig. 1 are inhibited by nicotinic receptor antagonists such as DHbE. This control experiment would significantly strengthen the result shown here.

We agree that combining drug manipulations with photometry would be useful, but as noted above this is not a methodology in our current technical repertoire.

3) Similarly, the authors show clear correlations between CIN activity and DA release during behavior. The authors should consider determining whether CINs play a causal role in triggering DA release during behavior. For example, does infusion of DHbE in the NAc prevent the light-mediated DA release during behavior? As an alternative hypothesis, some groups have been suggesting that CIN activity has almost no direct influence over DA. Therefore, testing whether a causal relationship exists between CINs and DA release would be an important experiment in addressing these two opposing viewpoints.

As noted above we are not currently able to combine drug manipulations with photometry in behaving animals.

4) The ramps that are described in this manuscript are an order of magnitude faster (increasing over 100s of milliseconds) than ramps described in other studies that occur over seconds. In fact, the two signals may be completely different functionally. Discussion of this topic would be helpful.

Dopamine ramps have indeed been reported over multiple different time scales, and as discussed in Berke 2018, this seems to reflect the duration of the approach behavior. We think further discussion of this topic is better saved for another paper, especially as we are now actively studying ramping over longer time scales (Krausz et al. 2023).

Reviewer #3 (Public Review):

This report by Mohebi et al. provides new answers to old questions by showing that the activity of striatal cholinergic interneurons (CINs) escalates progressively during specific reward-related behaviors and that this correlates with previously observed ramps in dopamine (DA) release in the nucleus accumbens core. The report is strong and provides evidence for the authors' hypothesis that DA ramps are independent of DA neuron activity, but are instead the result of CIN activity and corresponding acetylcholine (ACh) release. The authors further demonstrate that the fidelity of CIN activation and consequent driving of DA release is even more robust in vivo than observed ex vivo slice preparations, which is fundamental for understanding the role of ACh-DA interactions in behavior. The findings complement the authors' previous evidence ventral tegmental area (VTA) DA neuron firing patterns do not show a ramping pattern; the previously reported VTA data are appropriately included here (in Fig. 3) to illustrate the absence of VTA firing during the time-locked increases in CIN activity and DA release. The present studies stop short of showing a direct link between CIN activity and DA release, however, which would require examining DA release during behavior in the presence of an antagonist of nicotinic ACh receptors. The authors also extend the understanding of the regulation of DA release by acetylcholine (ACh) by showing that optical activation of CINs in vivo promotes DA release responses that do not attenuate with repetitive stimulation. This contrasts with previous results in ex vivo striatal slices in which ACh-evoked DA release has been found to decline progressively from rundown and/or receptor desensitization. The authors propose that in vivo, AChE may be more effective in curtailing local ACh levels than in slices because of the slightly lower temperature typically used for slice studies, as well as the use of superfusion that might facilitate some AChE washout (AChE inhibitors are still effective in slices, of course). Overall, the report not only provides evidence for the cellular substrate for DA ramps but also shows the robustness of ACh-driven DA release in vivo. A few points to strengthen the report are listed below.

1) The authors give a few details about how CINs were activated at the beginning of the results, but say only that DA dynamics were monitored using fiber photometry. Given that the methods are at the end, a brief summary should be given here to indicate whether this means direct monitoring of DA or indirect via GCaMP, for example. It would be helpful to note the sensor used in the abstract, as well. In this light, as it were, RdLight1 should be described upon the first mention.

We have now clarified in both abstract and text that we are using the direct DA sensor RdLight1.

2) The authors show that infusion of DHbE in the NAc likelihood of decisions to approach the center port, as did antagonism of DA receptors. This supports the authors' argument that ramping of CIN activity and consequent ACh release underlies observed ramps in DA release. However, to show a causal interaction requires testing whether the observed DA ramps are absent after DHbE infusion in the NAc, under the same conditions that attenuated behavior.

As noted above we are not currently able to combine drug manipulations with photometry in behaving animals.

3) In Fig. 3, the y-axis title for the upper panels should specify VTA, not simply "rate". This is stated in the legend, but should also be specified in the figure panel.

We have updated the y-axis titles in this figure.

4) A recent preprint in BioRxiv by AC Krok, NX Tritsch et al. shows a related correlation between ACh and DA release in vivo in a reward task, as well as differences in other conditions. This report shows also that cortical input to CINs indeed plays a role, as suggested in the concluding sections of the present report. Consideration of the data in the preprint in the context of the present results could be valuable for the field.

We have also noted those pre-prints with interest, even though they investigated different brain regions using different approaches. There are established differences between CIN-DA interactions in dorsal vs. ventral striatum that we suspect are relevant here. But given the rapid pace of developments in this subfield, we prefer not to speculate too much at this point and instead review the overall body of work once it is published.

Author Response

Reviewer #1 (Public Review):

“The abstract does not adequately summarize the content of the paper. There is no mention of stimulation, or bilateral connectivity, which is a large part of the paper. The names of all five species should appear in the abstract, not just X. laevis.”

In the revised manuscript, we have included all the names of the species and types of stimuli used to elicit fictive vocalizations in the abstract. In regard to bilateral connectivity, we believe that the reviewer was referring to the rostral-caudal connections between the parabrachial nucleus and nucleus ambiguus, which are critical for fast, but not for slow trill production. We have added this piece of information in the abstract. Furthermore, we have clarified the bilateral nature of the two central pattern generators (CPGs) in male X. laevis. In our previous study (Yamaguchi et al., 2017), we demonstrated that transections of the two commissures (one at the parabrachial nuclei level and the other at the nucleus ambiguus level) did not eliminate fictive advertisement calls in male X. laevis brains, indicating the presence of fast and slow trill CPGs in left and right hemi-brains of male X. laevis. This information was originally included in the results section (“Unilateral transection desynchronizes the fast clicks, but not the slow clicks across species”). We have now added this information to the introduction section to provide a clearer description of the anatomical organization of the two CPGs (p5, ln10 – 14).

“The conclusion that the "fast and slow CPGs identified in male X. laevis are conserved across species." is contradicted by the last paragraph, which states, "Fast trill-like CPGs are likely present only in fast clickers..." This inherent contradiction needs to be resolved.”

To resolve this contradiction, we have revised sentences in the abstract to clarify our findings. Specifically, we now state that “We found that even though the courtship calls of different Xenopus species vary in their click rates and duration, the CPGs used to generate clicks are conserved across species. The fast CPGs found in male X. laevis, which critically rely on reciprocal connections between the parabrachial nucleus and the nucleus ambiguus, are conserved among species that produce fast clicks. Similarly, the slow CPGs found in the caudal brainstem of male X. laevis are shared among species that produce slow clicks” (p2, ln 10 – 15) By making this change, we hope to provide greater clarity regarding our findings and help to resolve any contradictions.

“The testosterone results are over-emphasized.” “The conclusion that there is differential expression of testosterone receptors in the brain of each species is completely speculative and not supported by the data presented here.”

We have extensively revised our manuscript to ensure a more accurate interpretation of the results regarding testosterone experiments. Revised conclusions are outlined below:

Abstract: “In addition, our results suggest that testosterone plays a role in organizing fast CPGs in fast-click species, but it does not appear to have the same effect in slow-click species.” (p2, ln 15 – 17)

Introduction: “Additionally, we found that fast trill-like CPGs are present only in species that produce fast clicks and their presence appears to be regulated by testosterone in these species. “ (p6, ln 2 – 4)

Discussion: “However, this effect of testosterone appears to be limited to the fast clicker species. Male X. tropicalis, a slow clicker species, has been shown to have comparable plasma levels of testosterone to male X. laevis (mean plasma levels of testosterone of male X. laevis: 13 to 22ng/ml (Hecker et al., 2005; Hayes et al., 2010), male X. tropicalis: ~20ng/ml, (Olmstead et al., 2009)), yet the synapses between the PBN and laryngeal motoneurons in male X. tropicalis remained weak, and PBN showed little activity during fictive advertisement calls. These results suggest that testosterone acts differently on the central vocal pathways of fast and slow clickers, promoting the emergence of fast trill-like CPGs in X. laevis but not in X. tropicalis. Although further experiments with controlled testosterone levels are necessary to validate these results, we hypothesize that changes in the androgen receptors (e.g., expression patterns, ligand affinity) may have contributed to the divergence of fast and slow clickers.“ (p26, ln 13 – 24)

”The use of the word "development" implies embryology. Here, adults were treated and looked at 13 weeks later. There is no data presented about development. ”

In our revised manuscript, we have replaced the term “development” with “presence” or “acquisition” of neural circuitry to enhance the clarity and help to prevent any potential misunderstandings.

Author Response

Reviewer #1 (Public Review):

By the in vitro DNA damage response (DDR) assay with a defined DNA substrate using Xenopus extracts and in vitro binding assays with purified proteins, the authors nicely showed the role of APE1 (APEX1) in ATRIP recruitment for DDR activation, particularly a non-enzymatic (structural) role of APE1 in the binding to both ssDNAs and ATRIP. The results described in the paper are very convincing to support the authors' claim. However, these studies lack the quantification with proper statistics (and/or mentioning the reproducibility of the results). And, given the important discovery of APE1 in the DDR activation in vitro, it would be nice to demonstrate the role of APE1(APEX1) in ATR activation in vivo using siRNA-mediated knockdown of mammalian cells or yeast cells.

Thanks for the suggestion. As shown in our response to the #2 Essential Revisions, we have addressed this question by additional experiment and added extra description in our revised manuscript showing that APE1 is important for the ATR DDR following oxidative stress in culture human cell U2OS cells (Figure 1-figure supplement 1B). In addition, we have performed at least three independent experiments and statistical analysis to support our claims.