con mayor calidad del dato

Para garantizar que las conclusiones sean sólidas, compararemos ~~en profundidad~~ las herramientas.

Borrar.

Este ya no es nuestro objetivo. Sino el de analizar la calidad de los microdatos extraídos.

Delete one period

This commentary and the testing referenced is Yesco's justification for calculating and applying Case B wind load with much less eccentricity than Figure 29.3-1 requires.

Yesco has suggested that because the force coefficients were divided by 0.85 (normal gust factor), when determining the pressure for flexible signs with frequency less than 1 Hz, we should multiply by 0.85.

Citizens, not users.

I like that. I've been grasping for a generic alternative to 'user' that is less generic than 'person' and this hits a sweet spot.

4:00 "gewaltfantasien laut aussprechen ist eine straftat"<br /> nein, zu sagen "ich werde den scheiss habek wegballern" ist KEINE straftat.<br /> deutsche sind nur zu blöd für redefreiheit... hirntotes scheissvolk >: (

In this section, it is unclear what it mean by "markers" and "control points." Since the grammar is in progress, it is probably a good idea to explain briefly here.

Change this

Hello

In this section, it is unclear what the lines and arrows on the maps mean. I suppose it will be explained in the Grammar, but since it is in progress, it might be good idea to briefly explain it here.

The combi-steamer checks the temperature value of temperature sensor B1 every second (in all modes). If the temperature is below 350°C [662°F], the combi-steamer is fully functional.

If the temperature rises above 350 degrees Service 28.2 shows and the oven has no function.

To protect the unit from damage due to overtemperature, the unit cannot be operated again until the temperature sensor B1 has cooled below 350°C [662°F]

Thermocouple B1 also controls steam production for steam requirements up to local boiling point by monitoring the cooking cabinet temperature up to steam saturation point.

(Steam saturation point is when all internal cabinet metalwork (including B1) reaches the local boiling point temperature).

Upon reaching steam saturation point B1 causes the processor to switch off steam heating.

When product is placed in the oven the cabinet temperature reduces and steam production starts once again until the product, cabinet metalwork and B1 reaches steam saturation point..

The ICP Core probe has 6 separate thermocouples, five in the stem and one in the handle.

The probe when used in a process should be completely embedded in the product to be cooked, such that the 5 thermocouples in the stem are recording a temperature inside the product .

The processor calulates the average temperature of the 5 thermocouples and uses this value in order to aciieve an accurate core temperature.

The handle of the Core probe should be pushed against the body of the product to be cooked as the thermocuople situated in the handle is used in the process for product browning and therefore should be in contact with the product surface.

Reviewer #2 (Public Review):

Summary:

This was a well-executed and well-written paper. The authors have provided important new datasets that expand on previous investigations substantially. The discovery that changes in diet are not so closely correlated with the presence of alkaloids (based on the expanded sampling of non-defended species) is important, in my opinion.

Strengths:

Provision of several new expanded datasets using cutting edge technology and sampling a wide range of species that had not been sampled previously. A conceptually important paper that provides evidence for the importance of intermediate stages in the evolution of chemical defense and aposematism.

Weaknesses:

There were some aspects of the paper that I thought could be revised. One thing I was struck by is the lack of discussion of the potentially negative effects of toxin accumulation, and how this might play out in terms of different levels of toxicity in different species. Further, are there aspects of ecology or evolutionary history that might make some species less vulnerable to the accumulation of toxins than others? This could be another factor that strongly influences the ultimate trajectory of a species in terms of being well-defended. I think the authors did a good job in terms of describing mechanistic factors that could affect toxicity (e.g. potential molecular mechanisms) but did not make much of an attempt to describe potential ecological factors that could impact trajectories of the evolution of toxicity. This may have been done on purpose (to avoid being too speculative), but I think it would be worth some consideration.

In the discussion, the authors make the claim that poison frogs don't (seem to) suffer from eating alkaloids. I don't think this claim has been properly tested (the cited references don't adequately address it). To do so would require an experimental approach, ideally obtained data on both lifespan and lifetime reproductive success.

eLife assessment

This important study sheds light on how poison frogs gain their toxins, with surprising new data on low levels of toxins in previously non-toxic frogs. The authors propose a new theory for evolution of toxicity based on convincing evidence, but the manuscript needs restructuring to be clearer. While the manuscript will benefit from improved presentation, this research has the potential to greatly impact our understanding of animal defense mechanisms.

Reviewer #1 (Public Review):

This is a very relevant study, clearly with the potential of having a high impact on future research on the evolution of chemical defense mechanisms in animals. The authors present a substantial number of new and surprising experimental results, i.e., the presence in low quantities of alkaloids in amphibians previously deemed to lack these toxins. These data are then combined with literature data to weave the importance of passive accumulation mechanisms into a 4-phases scenario of the evolution of chemical defense in alkaloid-containing poison frogs.

In general, the new data presented in the manuscript are of high quality and high scientific interest, the suggested scenario compelling, and the discussion thorough. Also, the manuscript has been carefully prepared with a high quality of illustrations and very few typos in the text. Understanding that the majority of dendrobatid frogs, including species considered undefended, can contain low quantities of alkaloids in their skin provides an entirely new perspective to our understanding of how the amazing specializations of poison frogs evolved. Although only a few non-dendrobatids were included in the GCMS alkaloid screening, some of these also included minor quantities of alkaloids, and the capacity of passive alkaloid accumulation may therefore characterize numerous other frog clades, or even amphibians in general.

While the overall quality of the work is exceptional, major changes in the structure of the submitted manuscript are necessary to make it easier for readers to disentangle scope, hypotheses, evidence and newly developed theories.

...

...

The goal is to build an autoencoder that learns a common representation of a single object when given multiple perspectives during training.

"Correlational neural networks" - looking at learning from multiple perspectives of the same thing to increase representation learning.

@article{chandar2016neuralcompjour, author = {Chandar, Sarath and Khapra, Mitesh M and Larochelle, Hugo and Ravindran, Balaraman}, date-added = {2024-08-01 10:47:30 -0400}, date-modified = {2024-08-01 10:50:01 -0400}, journal = {Neural Computation}, keywords = {correlation-learning, machine-learning, inductive-bias, autoencoders}, number = {2}, pages = {257--285}, pdf = {https://www.researchgate.net/profile/Balaraman-Ravindran/publication/275588055_Correlational_Neural_Networks/links/55ed84d308ae21d099c75c00/Correlational-Neural-Networks.pdf}, publisher = {MIT Press}, title = {Correlational neural networks}, venue-short = {NeuralCompJour}, volume = {28}, year = {2016}}

cool

https://youtu.be/RxHilNaLy_s

https://youtu.be/RxHilNaLy_s

salam men notam necesen Sema

https://youtu.be/RxHilNaLy_s

https://youtu.be/RxHilNaLy_s

https://youtu.be/RxHilNaLy_s

Это видео о том, что в словакий, во первых много цыган, у них есть свой поселение, которых безумно сильно остают от всего мира: 1. электричества, света 2. дома сделаны в ручную, некоторые без крыш как в Африке, или в видео про Сеул. во вторых, правительство Словакий очень помогает им с условиями жизни. строит дома, или проводит электричество.

в этом видео Леша посещяет одно из таких населений Цыган, которое считается самым современным, в этом месте просто куча детей. условно у каждого человека есть, как минимум, 5 братьев и сестер.

оно такое цивильное, потому в этом поселений когда-то родился будущий управляшющий этого района. именно он сделал его ухоженным, практический без мусора и относительно приятным для проживания. хотя до этого было просто тьма мусора по всему району

он в детстве видел, как всё плохо. закончил вуз и получил диплом и стал дипломатом ради того, чтоб сделать жизнь людей в месте, где он родился лучше

for - system mapping software

system mapping software - ask @gyuri

for - participatory system maps - subjective - perspectival knowing

dfdf

eLife assessment

This valuable study examines whether the BMP signaling pathway has a role in H3.3K27M DMG tumors, regardless of the presence of ACRVR1 activating mutations. The authors provide solid evidence that BMP2/7 synergizes with H3.3K27M to induce a transcriptomic rewiring associated with a quiescent but invasive cell state. Although this work could be further enhanced by the inclusion of additional models, the study overall points to BMP2/7 as a potential target for future therapies in this deadly cancer.

Reviewer #1 (Public Review):

Summary:

Mutational analysis of diffuse midline glioma (DMG) found that ACVR1 mutations, which up-regulate BMP signaling pathway are found in most H3.1K27M, but not H3.3K27M DMG cases. In this manuscript, Huchede et al attempted to determine whether the BMP signaling pathway has any role in H3.3K27M DMG tumors. They found that the BMP signaling is activated to a similar level in H3.3K27M DMG cells with wild type ACVR1 compared to ACVR1 DMG cells, likely due to the expression of BMP7 or BMP2. They went on to test whether cells treated with BMP7 or BMP2 treatments affected the gene expression and cell fitness of tumor cells with H3.3K27M mutation. They concluded that BMP2/7 synergizes with H3.3K27M to induce a transcriptomic rewiring associated with a quiescent but invasive cell state. The major issue for this conclusion is that the authors did not use the right models/controls to obtain results to support this conclusion as detailed below. Therefore, in order to strengthen the conclusion, the authors need to address the major concerns below.

Strength:<br /> Address an important question in DMG field.

Major concerns/weakness:<br /> (1) All the results in Fig. 2 utilized two glioma lines SF188 and Res259. The authors should repeat all these experiments in a couple of H3.3K27M DMG lines by deleting H3.3K27M mutation first.<br /> (2) Fig. 3. The experiments of BMP2 treatment should be repeated in another H3.3K27M DMG line using H3.1K27M ACVR1 mutant tumor lines as controls.

Minor concerns<br /> Fig.2A. BMP2 expression increased in H3.3K27M SF188 cells. Therefore, the statement "whereas BMP2 and BMP4 expressions are not significantly modified (Figure 2A and Figure 2-figure supplement A-B)"is not accurate

Comments on revised version:

I had three issues listed above on the initial version. The authors did not address my major concerns of #1 and #2, which are re-listed above.

Author response:

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

Reviewer #1

Summary:

Mutational analysis of diffuse midline glioma (DMG) found that ACVR1 mutations, which up-regulate the BMP signaling pathway are found in most H3.1K27M, but not H3.3K27M DMG cases. In this manuscript, Huchede et al attempted to determine whether the BMP signaling pathway has any role in H3.3K27M DMG tumors. They found that the BMP signaling is activated to a similar level in H3.3K27M DMG cells with wild-type ACVR1 compared to ACVR1 DMG cells, likely due to the expression of BMP7 or BMP2. They went on to test whether cells treated with BMP7 or BMP2 treatments affected the gene expression and cell fitness of tumor cells with H3.3K27M mutation. They concluded that BMP2/7 synergizes with H3.3K27M to induce a transcriptomic rewiring associated with a quiescent but invasive cell state. The major issue for this conclusion is that the authors did not use the right models/controls to obtain results to support this conclusion as detailed below. Therefore, in order to strengthen the conclusion, the authors need to address the major concerns below.

Strength:

This paper addresses an important question in the DMG field.

Major concerns/weakness:

(1) All the results in Fig. 2 utilized two glioma lines SF188 and Res259. The authors should repeat all these experiments in a couple of H3.3K27M DMG lines by deleting the H3.3K27M mutation first.

We thank the referee for his/her comments that have helped us to strengthen our conclusions. Although we were rather interested in studying how the BMP pathway can participate in installing a particular cell state at the time of expression of the K27M mutation, we have now included the characterization of the native H3.3K27M BT245 and SU-DIPGXIII cell lines, and their counterparts in which the mutation was reverted by CRISPRCas9 (Harutyunyan et al., 2019). As shown in Figure 3-figure supplement D, the growth arrest induced by BMP2 seems indeed to be specific of the K27M epigenetic context, which could also be required to settle a positive regulation loop to activate the BMP pathway, as mentioned in the Discussion.

(2) Fig. 3. The experiments of BMP2 treatment should be repeated in other H3.3K27M DMG lines using H3.1K27M ACVR1 mutant tumor lines as controls.

The use of mutant ACVR1 lines is interesting, but their control status seems questionable, as the addition of BMPs could have a cumulative effect on the effect of the mutation, notably by activating other receptors in the pathway. But we have now included 3 different cell lines (HSJD-DIPG-014, BT245 and SU-DIPGXIII), and observed similar impact of BMP2 with growth arrest as a readout (Figure 3-figure supplement C-D)

Minor concerns

Fig.2A. BMP2 expression increased in H3.3K27M SF188 cells. Therefore, the statement "whereas BMP2 and BMP4 expressions are not significantly modified (Figure 2A and Figure 2-figure supplement A-B)" is not accurate.

The referee is absolutely right, and we have corrected this statement.

Reviewer #2 (Public Review):

The manuscript by Huchede et al investigates the BMP pathway in H3K27M-mutant gliomas carrying or not activating mutations in ALK2 (ACVR1). Their results in cell lines and in datasets acquired from the literature on patient tumors indicate that the BMP signaling pathway is activated at similar levels between ACVR1 wild-type and mutant tumors. The group further identifies BMP2 and BMP7 as possibly the main activators of the pathway in cells. They then show that BMP2 and 7 crosstalk with the H3 mutation and synergize to induce transcriptomic rewiring leading to an invasive cell state.

The paper is well-written and easy to follow with a robust experimental plan and datasets supporting the claims. While previous work (acknowledged by the authors) indicated activation of BMP in H3K27M tumors, wild type for the ACVR1 mutation this paper is a nice addition and provides further mechanistic cues as to the importance of the BMP pathway and specific members in these deadly brain cancers. The effect of these BMPs in quiescence and invasion is of particular interest.

We thank the referee for his/her supportive comments.

A few suggestions to clarify the message are provided below 1- In thalamic diffuse midline gliomas, the BMP pathway should not be activated as it is in the pons. The authors should identify thalamic tumors in the datasets they explored and patients-derived cell lines from thalamic tumors available to investigate whether this pathway is active across all H3.3K27M mutants in the brain midline or specifically in tumors from the pons.

The inter-patient variability observed in the level of activation of the BMP pathway may indeed be due, at least in part, to different tumor locations. However, we failed to find this information in the publicly available datasets that we used. We however included this element in the Discussion part.

(2) There are ~20% H3.3K27M tumors that carry an ACVR1 mutation and similar numbers of H3.1K27M that are wild type for this gene. Can the authors identify these outliers in their datasets and assess the activation of BMP2 and 7 or other BMP pathway members in this context?

We have now included the outliers present in our datasets in the legends of Figure 1B and Figure 1-figure supplement B and F. From the few samples available to document these outliers in the cohorts that we used, we have not observed major differences regarding the expression levels of BMP2/7 or BMP pathway members and have discussed the fact that it may result from the establishment in all cases of a feedback loop of activation.

In all this is an interesting paper that provides meaningful data to pursue clinical targeting of the BMP pathway, which would be a nice addition to the field.

We thank the reviewer for his/her supportive comments.

Gemini for Chrom

eLife assessment

This important study extends existing sequentially Markovian coalescent approaches to include the combined use of SNPs and hypervariable loci such as epimutations. This is an intriguing addition to infer population size history in the recent past, and the authors provide solid validation of their methods via simulation and analysis of empirical data in Arabidopsis thaliana. Given the increasing availability of such data, this work is a timely contribution and represents a foundation for further developments to explore when and where these methods will be best used.

Reviewer #1 (Public Review):

The authors developed an extension to the pairwise sequentially Markov coalescent model that allows to simultaneously analyze multiple types of polymorphism data. In this paper, they focus on SNPs and DNA methylation data. Since methylation markers mutate at a much faster rate than SNPs, this potentially gives the method better power to infer size history in the recent past. Additionally, they explored a model where there are both local and regional epimutational processes.

Integrating additional types of heritable markers into SMC is a nice idea which I like in principle. However, a major caveat to this approach seems to be a strong dependence on knowing the epimutation rate. In Fig. 6 it is seen that, when the epimutation rate is known, inferences do indeed look better; but this is not necessarily true when the rate is not known. (See also major comment #1 below about the interpretation of these plots.) A roughly similar pattern emerges in Supp. Figs. 4-7; in general, results when the rates have to be estimated don't seem that much better than when focusing on SNPs alone. This carries over to the real data analysis too: the interpretation in Fig. 7 appears to hinge on whether the rates are known or estimated, and the estimated rates differ by a large amount from earlier published ones.

Overall, this is an interesting research direction, and I think the method may hold more promise as we get more and better epigenetic data, and in particular better knowledge of the epigenetic mutational process.

Reviewer #2 (Public Review):

A limitation in using SNPs to understand recent histories of genomes is their low mutation frequency. Tellier et al. explore the possibility of adding hypermutable markers to SNP based methods for better resolution over short time frames. In particular, they hypothesize that epimutations (CG methylation and demethylation) could provide a useful marker for this purpose. Individual CGs in Arabidopsis tends to be either close to 100% methylated or close to 0%, and are inherited stably enough across generations that they can be treated as genetic markers. Small regions containing multiple CGs can also be treated as genetic markers based on their cumulative methylation level. In this manuscript, Tellier et al develop computational methods to use CG methylation as a hypermutable genetic marker and test them on theoretical and real data sets. They do this both for individual CGs and small regions. My review is limited to the simple question of whether using CG methylation for this purpose makes sense at a conceptual level, not at the level of evaluating specific details of the methods. I have a small concern in that it is not clear that CG methylation measurements are nearly as binary in other plants and other eukaryotes as they are in Arabidopsis. However, I see no reason why the concept of this work is not conceptually sound. Especially in the future as new sequencing technologies provide both base calling and methylating calling capabilities, using CG methylation in addition to SNPs could become a useful and feasible tool for population genetics in situations where SNPs are insufficient.

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

The authors developed an extension to the pairwise sequentially Markov coalescent model that allows to simultaneously analyze multiple types of polymorphism data. In this paper, they focus on SNPs and DNA methylation data. Since methylation markers mutate at a much faster rate than SNPs, this potentially gives the method better power to infer size history in the recent past. Additionally, they explored a model where there are both local and regional epimutational processes. Integrating additional types of heritable markers into SMC is a nice idea which I like in principle. However, a major caveat to this approach seems to be a strong dependence on knowing the epimutation rate. In Fig. 6 it is seen that, when the epimutation rate is known, inferences do indeed look better; but this is not necessarily true when the rate is not known. (See also major comment #1 below about the interpretation of these plots.) A roughly similar pattern emerges in Supp. Figs. 4-7; in general, results when the rates have to be estimated don't seem that much better than when focusing on SNPs alone. This carries over to the real data analysis too: the interpretation in Fig. 7 appears to hinge on whether the rates are known or estimated, and the estimated rates differ by a large amount from earlier published ones.

Overall, this is an interesting research direction, and I think the method may hold more promise as we get more and better epigenetic data, and in particular better knowledge of the epigenetic mutational process. At the same time, I would be careful about placing too much emphasis on new findings that emerge solely by switching to SNP+SMP analysis.

Major comments:

- For all of the simulated demographic inference results, only plots are presented. This allows for qualitative but not quantitative comparisons to be made across different methods. It is not easy to tell which result is actually better. For example, in Supp. Fig. 5, eSMC2 seems slightly better in the ancient past, and times the trough more effectively, while SMCm seems a bit better in the very recent past. For a more rigorous approach, it would be useful to have accompanying tables that measure e.g. mean-squared error (along with confidence intervals) for each of the different scenarios, similar to what is already done in Tables 1 and 2 for estimating $r$.

We believe this comment was addressed in the previous revision (Sup Table 6-10) by adding Root Mean Square Errors for the demographic estimates (and RMSE for recent versus past portions of the demography). 

- 434: The discussion downplays the really odd result that inputting the true value of the mutation rate, in some cases, produces much worse estimates than when they are learned from data (SFig. 6)! I can't think of any reason why this should happen other than some sort of mathematical error or software bug. I strongly encourage the authors to pin down the cause of this puzzling behaviour. (Comment addressed in revision. Still, I find the explanation added at 449ff to be somewhat puzzling -- shouldn't the results of the regional HMM scan only improve if the true mutation rate is given?)

We do understand that our results and explanation can appear counter-intuitive. As acknowledged by the reviewer, in the previous round of revision we have at length clarified this puzzling behaviour by the discrepancy in assessing methylation regions using the HMM method which then differs from the HMM for the SMC inference. We are happy to clarify further in response to the new question of reviewer 1:

If the Reviewer #1 means the SNP mutations (e.g. A → T), knowing the true mutation rate does not help the HMM to recover the region level methylation status. 

If the Reviewer #1 means the epimutations (whether it is the region, site or both), knowing the true epimutations rates could theoretically help the HMM to recover the region level methylation status. However, at present, our method does not leverage information from epimutation rates to infer the region level methylation status. As inferring the epimutations rates is one of the goals of this study in the SMC inference, and that region level methylation status is required to infer those rates, we suspect that using epimutations rates to infer the region level methylation status could be statistically inappropriate (generating some kind of circular estimations). Instead, our HMM uses only the proportion of methylated and unmethylated sites (estimated from the genome) to determine whether or not a region status is most-likely to be methylated or unmethylated. We now explicit this fact in the HMM for methylation region in the method section.

We acknowledge that our HMM to infer region level methylation status could be improved, but this would be a complete project and study on its own (due to the underlying complexity of the finite site and the lack of a consensus model for epimutations at evolutionary time scale). We believe our HMM to have been the best compromise with what was known from methylation and our goals when the study was conducted, and future work is definitely worth conducting on the estimation of the methylation regions.

- As noted at 580, all of the added power from integrating SMPs/DMRs should come from improved estimation of recent TMRCAs. So, another way to study how much improvement there is would be to look at the true vs. estimated/posterior TMRCAs. Although I agree that demographic inference is ultimately the most relevant task, comparing TMRCA inference would eliminate other sources of differences between the methods (different optimization schemes, algorithmic/numerical quirks, and so forth). This could be a useful addition, and may also give you more insight into why the augmented SMC methods do worse in some cases. (Comment addressed in revision via Supp. Table 7.).

- A general remark on the derivations in Section 2 of the supplement: I checked these formulas as best I could. But a cleaner, less tedious way of calculating these probabilities would be to express the mutation processes as continuous time Markov chains. Then all that is needed is to specify the rate matrices; computing the emission probabilities needed for the SMC methods reduces to manipulating the results of some matrix exponentials. In fact, because the processes are noninteracting, the rate matrix decomposes into a Kronecker sum of the individual rate matrices for each process, which is very easy to code up. And this structure can be exploited when computing the matrix exponential, if speed is an issue.

We believe this comment was acknowledged in the previous revision (line 649), and we thank the reviewer for this interesting insight.

- Most (all?) of the SNP-only SMC methods allow for binning together consecutive observations to cut down on computation time. I did not see binning mentioned anywhere, did you consider it? If the method really processes every site, how long does it take to run?

We believe this comment was addressed in the previous revision and was added to the manuscript in the methods Section (subsection :  SMC optimization function).

- 486: The assumed site and region (de)methylation rates listed here are several OOM different from what your method estimated (Supp. Tables 5-6). Yet, on simulated data your method is usually correct to within an order of magnitude (Supp. Table 4). How are we to interpret this much larger difference between the published estimates and yours? If the published estimates are not reliable, doesn't that call into question your interpretation of the blue line in Fig. 7 at 533? (Comment addressed in revision.)

Reviewer #2 (Public Review):

A limitation in using SNPs to understand recent histories of genomes is their low mutation frequency. Tellier et al. explore the possibility of adding hypermutable markers to SNP based methods for better resolution over short time frames. In particular, they hypothesize that epimutations (CG methylation and demethylation) could provide a useful marker for this purpose. Individual CGs in Arabidopsis tends to be either close to 100% methylated or close to 0%, and are inherited stably enough across generations that they can be treated as genetic markers. Small regions containing multiple CGs can also be treated as genetic markers based on their cumulative methylation level. In this manuscript, Tellier et al develop computational methods to use CG methylation as a hypermutable genetic marker and test them on theoretical and real data sets. They do this both for individual CGs and small regions. My review is limited to the simple question of whether using CG methylation for this purpose makes sense at a conceptual level, not at the level of evaluating specific details of the methods. I have a small concern in that it is not clear that CG methylation measurements are nearly as binary in other plants and other eukaryotes as they are in Arabidopsis. However, I see no reason why the concept of this work is not conceptually sound. Especially in the future as new sequencing technologies provide both base calling and methylating calling capabilities, using CG methylation in addition to SNPs could become a useful and feasible tool for population genetics in situations where SNPs are insufficient.

We thank again the reviewer #2 for his positive comments.  

Reviewer #3 (Public Review):

I very much like this approach and the idea of incorporating hypervariable markers. The method is intriguing, and the ability to e.g. estimate recombination rates, the size of DMRs, etc. is a really nice plus. I am not able to comment on the details of the statistical inference, but from what I can evaluate it seems reasonable and in principle the inclusion of highly mutable sties is a nice advance. This is an exciting new avenue for thinking about inference from genomic data. I remain a bit concerned about how well this will work in systems where much less is understood about methylation,

The authors include some good caveats about applying this approach to other systems, but I think it would be helpful to empiricists outside of thaliana or perhaps mammalian systems to be given some indication of what to watch out for. In maize, for example, there is a nonbimodal distribution of CG methlyation (35% of sites are greater than 10% and less than 90%) but this may well be due to mapping issues. The authors solve many of the issues I had concerns with by using gene body methylation, but this is only briefly mentioned on line 659. I'm assuming the authors' hope is that this method will be widely used, and I think it worth providing some guidance to workers who might do so but who are not as familiar with these kind of data.

We thank the reviewer #3 for his positive comments. And we agree with Reviewer #3 concerning the application to data and that our approach needs to be carefully thought before applied. Our results clearly show that methylation processes are not well enough understood to apply our approach as we initially (maybe naively) designed it. Further investigations need to be conducted and appropriate theoretical models need to be developed before reliable results can be obtained. And we hope that our discussion points this out. However, our approach, the theoretical models and the additional tools contained in this study can be used to help researchers in their investigations to whether or not use different genomic markers to build a common (potentially more reliable) ancestral history. We enhanced the discussion in this second revision by clarifying also the use of the methylation from genic regions to avoid  confusion (lines 700-731).

Recommendations for the authors:  

Reviewer #1 (Recommendations For The Authors):

In added Supp. Table 7, I don't think these are in log10 units as stated in the caption.

Well Spotted! Indeed, the RMSE is not in log10 scale, we corrected the caption. We also added that the TMRCA used for MRSE calculations is in generations units to avoid potential confusion.  

Reviewer #3 (Recommendations for The Authors):

I very much appreciate the authors' attention to previous questions. I would ask that a bit more is spent in the discussion on concerns/approaches empiricists should keep in mind -- I am wary of this being uncritically applied to data from non-model species. It was not clear to me, for example (only mentioned on line 659 in the discussion) that the thaliana data is only using gene-body methylation. This poses potential issues with background selection that the authors acknowledge appropriately, but also assuages many of my concerns about using genome-wide data. I think text with recommendations for data/filtering/etc or at least cautions of assumptions empiricists should be aware of would help.

We apologize for the confusion at line 659. As written in the other section of the manuscript we meant CG sites in genic regions (and not only gene body methylated regions).

Due to the manuscript’s structure, the data from Arabidopsis thaliana is only described at the very end of the manuscript (line 900+). However, a brief description could also be found line 291-296. We however added a sentence in the introduction (line 128) for clarity. 

We however agree with the comment made by reviewer #3 concerning the application to data. We pointed in the discussion the risk of applying our approach on ill-understood (or illprepared) data and stressed the current need of studies on the epimutations processes at evolutionary time scale ( i.e. at Ne time scale) (line 700-703).

eLife assessment

This study presents a valuable finding on the heterogeneity of tumour metabolism using fluorescence lifetime imaging, measured across 4 cell lines, 4 tumour types of in vivo mouse models, and 29 patient samples. The indication is that the level of heterogeneity of cellular metabolism increases with model complexity and demonstrates high heterogeneity at a clinical level. The evidence supporting the claims of the authors is solid, and at the revision stage, the authors have included additional samples from 8 patients in the data pool, which is helpful for the conclusions that the authors are trying to draw. The work will be of interest to medical biologists developing methods for quantifying metabolic heterogeneity.

Reviewer #1 (Public Review):

Summary:

In this study, Komarova et al. investigate the clinical prognostic ability of cell-level metabolic heterogeneity quantified via the fluorescence lifetime characteristics of NAD(P)H. Fluorescence lifetime imaging microscopy (FLIM) has been studied as a minimally invasive approach to measure cellular metabolism in live cell cultures, organoids, and animal models. Its clinical translation is spearheaded though macroscopic implementation approaches that are capable of large sampling areas and enable access to otherwise constrained spaces but lack cellular resolution for a one-to-one transition with traditional microscopy approaches, making the interpretation of the results a complicated task. The merit of this study primarily lies in its design by analyzing with the same instrumentation and approach colorectal samples in different research scenarios, namely in vitro cells, in vivo animal xenografts, and ex vivo tumor tissue from human patients. These conform to a valuable dataset to explore the translational interpretation hurdles with samples of increasing levels of complexity. For human samples, which exhibited the highest degree of heterogeneity from the experiments presented, the study specifically investigates the prediction ability of NAD(P)H fluorescence metrics for the binary classification of tumors of low and advanced stage, with and without metastasis, and low and high grade. They find that NAD(P)H fluorescence properties have a strong potential to distinguish between high- and low-grade tumors and a moderate ability to distinguish advanced stage tumors from low stage tumors. This study provides valuable results contributing to the deployment of minimally invasive optical imaging techniques to quantify tumor properties and potentially migrating into tools for human tumor characterization and clinical diagnosis.

Strengths:

The investigation of colorectal samples under multiple imaging scenarios with the same instrument and approach conforms to a valuable dataset that can facilitate interpretation of results across the spectrum of sample complexity.

The manuscript provides a strong discussion reviewing studies that investigated cellular metabolism with FLIM and the metabolic heterogeneity of colorectal cancer in general.

The authors do a thorough acknowledgement of the experimental limitations of investigating human samples ex vivo, and the analytical limitation of manual segmentation, for which they provide a path forward for higher throughput analysis.

Weaknesses:

NAD(P)H fluorescence provides a partial picture of the cell/tissue metabolic characteristics. Including fluorescence from flavins would comprise a more compelling dataset. These additional data should enable the quantification of redox metrics, which could positively contribute to the prognosis potential of metabolic heterogeneity. The authors did attempt to incorporate flavin fluorescence, unfortunately they could not find strong enough signal to proceed with the analysis.

Declaration of Digital Independence

eLife assessment

This work presents valuable information on the structure of the spirosome's native extended conformation as the active form of the aldehyde-alcohol dehydrogenase (AdhE) enzyme. The evidence is solid, although the work does not provide a mechanistic understanding of the function and dynamics of AdhE.

Reviewer #1 (Public Review):

Clostridium thermocellum serves as a model for consolidated bioprocessing (CBP) in lignocellulosic ethanol production. The primary ethanol production pathway involves the enzyme aldehyde-alcohol dehydrogenase (AdhE), which exhibits complex regulation, forming long oligomeric structures known as spirosomes.

The present study describes the cryo-EM structure of C. thermocellum AdhE, resolved at 3.28 Å resolution. By integrating cryo-EM data with molecular dynamics simulations, this study showed that the aldehyde intermediate resides longer in the channel of the extended form, supporting the mechanistic model in which the extended spirosome conformation represents the active form of AdhE.

These findings advance the understanding of the function and regulation of AdhE, a key enzyme involved in the ethanol biosynthesis pathway in Clostridium thermocellum, a model organism for ethanol production in consolidated bioprocessing.

Reviewer #2 (Public Review):

Summary:

The manuscript by Ziegler et al, entitled 'Structural characterization and dynamics of AdhE ultrastructure from Clostridium thermocellum: A containment strategy for toxic intermediates?" presents the atomic resolution cryo-EM structure of C. thermocellum AdhE showing that it show dominantly an extended form while E.coli AdhE shows dominantly a compact form. With comparative analysis of their C. thermocellum structure and the previous E.coli AdhE structure, they tried to reveal the mechanism by which C.thermocellum and E.coli show different dominant conformations. In addition, they also analyzed the substrate channel by comparative and computational approaches. Lastly, their computational analysis using CryoDRGN reveals conformational heterogeneity in the sample. Despite this the manuscript is very descriptive and does not provide a mechanistic understanding by which AdhE works, this work will provide structural frame works to further investigate the function and mechanism of AdhE dynamics.

Strengths:

This manuscript provides the first C. thermocellum (Ct) AdhE structure and comparatively analyzed this structure with E.coli AdhE.

Weaknesses:

This work is very descriptive and does not provide mechanistic understanding of the function and dynamics of AdhE.

Author response:

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

Public Reviews: 

Reviewer #1 (Public Review): 

Summary:

Clostridium thermocellum serves as a model for consolidated bioprocess (CBP) in lignocellulosic ethanol production, but yet faces limitations in solid contents and ethanol titers achieved by engineered strains thus far. The primary ethanol production pathway involves the enzyme aldehydealcohol dehydrogenase (AdhE), which forms long oligomeric structures known as spirosomes, previously characterized via the 3.5 Å resolution E. coli AdhE structure using single-particle cryoEM. The present study describes the cryo-EM structure of the C. thermocellum ortholog, sharing 62% sequence identity with E. coli AdhE, resolved at 3.28 Å resolution. Detailed comparative structural analysis, including the Vibrio cholerae AdhE structure, was conducted. Integrating cryoEM data with molecular dynamics simulations indicated that the aldehyde intermediate resides longer in the channel of the extended form, supporting the hypothesis that the extended spirosome represents the active form of AdhE. 

Strengths: 

The study conducts a comprehensive structural comparative analysis of oligomerization interfaces and the acetaldehyde channel across compact and extended conformations. Structural and computational results suggest the extended spirosome as the most likely active state of AdhE. 

Weaknesses: 

The overall resolution of the C. thermocellum structure is similar to the E. coli ortholog, which shares 62% sequence identity, and the oligomerization interfaces and the acetaldehyde channel were previously described. 

Reviewer #2 (Public Review): 

Summary: 

The manuscript by Ziegler et al, entitled 'Structural characterization and dynamics of AdhE ultrastructure from Clostridium thermocellum: A containment strategy for toxic intermediates?" presents the atomic resolution cryo-EM structure of C. thermocellum AdhE showing that it show dominantly an extended form while E. coli AdhE shows dominantly a compact form. With comparative analysis of their C. thermocellum structure and the previous E. coli AdhE structure, they tried to reveal the mechanism by which C. thermocellum and E. coli show diXerent dominant conformations. In addition, they also analyzed the substrate channel by comparative and computational approaches. Lastly, their computational analysis using CryoDRGN reveals conformational heterogeneity in the sample. Although this manuscript suggests a potential mechanism of the diXerent features of AdhEs, this manuscript is very descriptive and does not provide suXicient data to support the authors' conclusions, which may be due to the lack of experimental data to support their findings from the computational analysis. 

Strengths: 

This manuscript provides the first C. thermocellum (Ct) AdhE structure and comparatively analyzed this structure with E. coli AdhE. 

Weaknesses: 

Their main conclusions obtained mostly by computational and comparative analysis are not supported by experimental data. 

Reviewer #3 (Public Review): 

This study describes the first structure of Gram-positive bacterial AdhE spirosomes that are in a native extended conformation. All the previous structures of AdhE spirosomes obtained come from Gram-negative bacterial species with native compact spirosomes (E. coli, V. cholerae). In E. coli, AdhE spirosomes can be found in two diXerent conformational states, compact and extended, depending on the substrates and cofactors they are bound to. 

The high-resolution cryoEM structure of the extended C. thermocellum AdhE spirosomes produced in E. coli in an apo state (without any substrate or cofactors) is compared to the E. coli extended and compact AdhE spirosomes structures previously published. The authors have modeled (in Swiss-Model) the structure of compact C. thermocellum AdhE spirosomes, using E. coli compact AdhE spirosome conformation as a template, and performed molecular dynamics simulations. They have identified a channel in which the toxic reaction intermediate aldehyde could transit from the aldehyde dehydrogenase active site to the alcohol dehydrogenase active site, in an analogous manner to E. coli spirosomes. These findings are in line with the hypothesis that the extended spirosomes could correspond to the active form of the enzyme. 

In this work, the authors speculate that the C. thermocellum AdhE spirosomes could switch from the native extended conformation to a compact conformation, in a way that is inverse of E. coli spirosomes. Although attractive, this hypothesis is not supported by the literature. Amazingly, in some Gram-positive bacterial species (S. pneumoniae, S. sanguinis or C. di8icile...), AdhE spirosomes are natively extended and have never been observed in a compact conformation. On the opposite, E. coli (and other Gram-negative bacteria) native AdhE spirosomes are compact and are able to switch to an extended conformation in the presence of the cofactors (NAD+, coA, and iron). The data presented as they are now are not convincing to confirm the existence of C. thermocellum AdhE spirosomes in a compact conformation. 

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors): 

Major points: 

(1) The claim of achieving the highest resolution AdhE structure lacks strong support since the E. coli structure was solved at 3.5A, whereas the C. thermocellum was solved at 3.28A. Conducting a local resolution analysis could provide insights into distinct structural interpretations, enhancing the strength of the claim. 

We have modified the sentence claiming this as the highest resolution AdhE structure to say, “In this study, we presented and analyzed a high-resolution structure of the AdhE spirosome from C. thermocellum.” We have included the local resolution map in Figure 2C – all structural analysis was performed in regions from the center of the molecule, where the highest resolution information was determined.

(2) The comparative structural analysis of the oligomerization interface is thorough, yet it could benefit from greater conciseness. Focusing on highlighting major findings would streamline the presentation and enhance clarity. 

We altered a few places in the comparative structural analysis in response to other reviewers. We also divided the main structure section into two subsections (spirosome interfaces and AdhE active sites) to enhance clarity.

Reviewer #2 (Recommendations For The Authors): 

(1) The authors should change the tile containing "?". Does it mean that the conclusions that the authors made are still in question? 

We have removed the question mark to indicate that our results point to a channeling mechanism.

(2) Figure 1B: Clarify Ct Fwd. Is this adding NADH, and Ct Rev adding NAD+? 

This information is described in the text in lines 98-100. It is also at the bottom of figure 1B.

(3) Line 131: Please revise accordingly for clarity: "The extended dimer interfaces" è "The extended E.coli dimer interface". 

This has been edited for clarity. We have added the following sentence resulting to indicate which interfaces that are being discussed: “Both the E. coli and C. thermocellum extended dimer interfaces bury ~5000 Å2. While the compact C. thermocellum compact dimer interface buries a similar surface area of ~4800 Å2, the E. coli dimer interface buries ~3800 Å2.”

(4) Line 133-136: Why that does not seem to be the case? These sentences are not clear what the authors exactly mean. 

We altered the text to say, “One would expect the compact structure in E. coli to have a larger buried surface area due to it being the predominant form when it is examined without additives, but that is not the case; further corroborating that factors other than buried surface area must impact the apo state of the spirosome.” We hope this clarifies our intent.

(5) Line 138-145: The authors should provide a logic for how the diXerent distribution of the charged residues would change the form of AdhE. It may just be a diXerent distribution nothing to do with the conformational change. 

After further analysis of the interface amino acid distribution, we agree that the distribution may have nothing to do with the conformational change. We have changed this section to end with the sentence “Analysis of the residues buried in these interfaces reveals that while many of the residues are identical in the C. thermocellum and E. coli extended structures, there are some diXerences in amino acid type distribution, although nothing that directly indicates control of conformer state (Supplemental Figure 3).” 

(6) Line 169: Kim et al. è Cho et al.

We have corrected this error.

(7) Line 122-235: The whole section is just describing the diXerence between Ct and Ec AdhE suggesting that this diXerence may contribute to the conformational diXerence without any evidence. The author cannot say that the diXerences in the interface, active sites cofactor pockets, etc explain why two AdhE (Ct, Ec) have diXerent domain conformers unless they provide experimental data. 

We did not conclude that any diXerences we observed structurally were responsible for the conformation change. The purpose of this section was solely to compare the structures to determine if we could find a structural basis for the diXerence between E. coli and C. thermocellum conformation – we stated a few times throughout the section and in the discussion that there were no immediate structural reasons for this diXerence in shape. We have added a few sentences in the discussion to address whether Gram-positive vs. Gram-negative is influencing the shape, addressed in reviewer #3 comment #4. 

(8) Line 237: The whole section "Identification..." analyzed the substrate channel by computational analysis. The author should provide experimental evidence that these residues identified are critical for channeling by generating mutants and measuring their activity. 

We agree that mutagenesis is the next logical step for these results, however it is outside the scope of work of this paper as this study will not be that straightforward. We have included a sentence in the discussion to indicate our plans for further investigation to the channel that says, “Future mutagenesis studies will be needed to confirm whether the spirosome exists to control the reaction flux in high-reactant conditions.”

Reviewer #3 (Recommendations For The Authors): 

(1) The capacity of C. thermocellum AdhE spirosomes to switch from a natively extended conformation to a compact conformation is not demonstrated in this manuscript, as it is now. Because this would be the first time that Gram-positive bacterial AdhE spirosomes are observed in a compact conformation, the authors should provide a clear demonstration of their existence by presenting reliable and good images of C. thermocellum compact spirosomes. 

We have modified Figure 1A to zoom in on one compact and extended spirosome that we have identified from each C. thermocellum sample. We have included triangles of the same size and shape to indicate the proximity of a turn of a helix, showing that the identified compact spirosomes have a tighter conformation than extended spirosomes.

(2) The authors should show at least an image of the compact C. thermocellum spirosomes, that they claim to observe in the presence of NADH or in the forward reaction conditions mentioned in Figure 1. The authors have added diXerent reactants to the extended C. thermocellum spirosomes and visualized their conformation by negative stain. An image of each condition tested would be valuable and would nicely complete the distribution of compact versus extended spirosomes presented in Figure 1. 

We have created a new supplemental figure with spirosomes circled for all of the experimental conditions for C. thermocellum (Supplemental figure 1). We have added a reference to supplemental figure 1 in the text to direct the reader to these images.

(3) The cryoEM classes presented in Figure 8 are not convincing and could correspond to dimers or rosettes of AdhE or to E. coli endogenous AdhE. CryoEM classes showing longer compact C. thermocellum spirosomes should be shown. The percentage of these compact spirosomes visualized in the micrographs should be added and discussed in the text as it would increase confidence in these findings and confirm that C. thermocellum compact spirosomes exist. Heterologous production of C. thermocellum AdhE in E. coli depleted for its endogenous AdhE would be required to definitively prove that these are compact C. thermocellum AdhE spirosomes in the cryoEM. 

We included the pictures of the theoretical compact spirosomes, as generated from the 8-mer of E. coli AdhE (6AHC) to address the possibility of rosettes. We have now indicated in the text that there were 6.7% of the particles in the compact conformation, which is less than seen by negative stain. We further mentioned that the compact spirosome is less compact than that seen in E. coli. We added a sentence to the discussion about the possibility of contaminating E. coli spirosomes (though this is very unlikely ) in our compact spirosome analysis: “While these compact spirosomes could result from expression in E. coli, though this is very unlikely, we also identified compact spirosomes in a native C. thermocellum lysate, which would not have similar contamination issues.”

(4) The authors should include and discuss in the text previous findings (among which Laurenceau et al., 2015...) describing the diXerences between Gram-positive and Gram-negative spirosomes. AdhE spirosomes are natively extended in most Gram-positive bacterial species (S. pneumoniae, S. sanguinis or C. diXicile...), and have never been observed in a compact conformation. On the opposite, E. coli (and other Gram-negative bacteria) native AdhE spirosomes are compact and are able to switch to an extended conformation in the presence of the cofactors (NAD+, coA, and iron). 

We have added the following sentences to the discussion to address this comment: “This could potentially be due to the diXerences between Gram-positive and Gram-negative bacteria. In previous studies, compact spirosomes have only been isolated from Gram-negatives while solely extended spirosomes have been isolated from Gram-positives. Furthermore, while the compact spirosomes can transition to extended in the presence of cofactors, the reverse has not been previously observed with an extended spirosome.”

(5) The authors have spotted some diXerences between the E. coli and C. thermocellum structures, that they believe could explain the intrinsic capacity of these spirosomes to be natively extended or compact. It would be interesting to confirm this hypothesis by measuring C. thermocellum extended AdhE spirosome activity and comparing it to E. coli extended spirosomes. The impact of mutations in the regions proposed by the authors to be important in the capacity of C. thermocellum AdhE to be extended (especially the GxGxxG motif and the D494 position) would be appreciated to confirm this hypothesis. 

We agree that this would be an interesting avenue of research although it is currently outside the scope of this paper. We are looking into experiments that we can perform where we can track both activity and conformation but have not found an ideal experiment at this time.

(6) Many statements and result interpretations are overstated in several parts of the manuscript and would need to be rewritten to balance the absence of clear evidence of C. thermocellum compact spirosomes. 

We have shown that we have identified compact spirosomes, addressed in multiple comments above. We have adjusted the language of the paper to indicate more uncertainty that will be followed up in future mutagenesis experiments. However, these mutations are not that simple to identify and this research would require a fairly large study that is better suited for a follow up manuscript.

(7) The Figure 7 legend would need to be corrected.

We are unsure as to what needs to be corrected in the figure 7 legend based on this comment.

eLife assessment

In this important manuscript, the authors used unbiased approaches to identify somatic mutations in publicly available databases that would disrupt clinically approved antibodies targeting HER2. Using a solid combination of both computational and experimental approaches they identify mutations that could restore therapeutic antibody sensitivity in a series of disease-relevant model systems. Additional cell-based and in vivo assays would strengthen the work and increase the translational and potential clinical relevance of the proposed work.

Reviewer #2 (Public Review):

Summary:

Peled et al identified HER2 mutations in connection with resistance to the anti-HER2 antibody Pertuzumab-mediated therapy. After constructing a yeast display library of Pertuzumab variants with 3.86×10^11 sequences for targeted screening of variant combinations in chosen 6 out of 14 residues, the authors performed experimental screening to obtain the clones that bind to HER2 WT and/or mutants (S310Y and S310F), and then combined new variations to obtain antibodies with a broad spectrum binding to both WT and two HER2 mutants. These are interesting studies of clinical impact and translational potential.

Strengths:

(1) Deep computational analyses of large datasets of clinical data provide useful information about HER2 mutations and their potential relevance to antibody therapy resistance.

(2) There is valuable information analyzing the residues within or near the interface between the antigen HER2 and the Pertuzumab antibody (heavy chain).<br /> The experimental antibody library screening obtained 90+ clones from 3.86×10^11 sequences for further functional validation.

Weaknesses:

(1) There is lack of assessment for antibody variant functions in cancer cell phenotypes in vitro (proliferation, cell death, motility) or in vivo (tumor growth and animal survival). The only assay was the western blotting of phosphopho-HER3 in Figure 4. However, HER2 levels and phosphor-HER2 were not analyzed.

(2) There is misleading impression from the title of computational engineering of a therapeutic antibody and the statement in the abstract "we designed a multi-specific version of Pertuzumab that retains original function while also bindings these HER2 variants" for a few reasons:

a. The primary method used for variant antibody identification for HER2 mutant binding is rather traditional experimental screening based on yeast display instead of computational design of a multi-specific version of Pertuzumab.

b. There is insufficient or lack of computational power in the antibody design or prioritization in choosing variant residues for the library construction of 3.86×1011 sequences. It seems random combinations from 6 residues out of 4 groups with 20 amino acid options.<br /> c. The final version of tri-binding variant is a combination of screened antibody clones instead of computation design from scratch.<br /> d. There is incomplete experimental evidence about the therapeutic values of newly obtained antibody clones.

Comments on revised version:

Two major comments remain and have not been well addressed. Comment 1 is expecting any cellular phenotypic analysis if not in vivo. Comment 2 requires some modifications to avoid overstating.

Author response:

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

Reviewer #1 (Public Review):

Strengths:

(1) In my assessment, the data sufficiently demonstrates that a modified version of Pertuzamab can bind both the wild-type and S310 mutant forms of ERBB2.

(2) The engineering strategy employed is rational and effectively combines computational and experimental techniques.

(3) Given the clinical activity of HER2-targeting ADCs, antibodies unaffected by ERBB2 mutations would be desired.

Weaknesses:

(1) There is no data showing that the engineered antibody is equally specific as Pertuzamab i.e. that it does not bind to other (non-ERBB2) proteins.

Showing the specificity of the engineered antibodies is indeed important. We did not address it in the current ms, but it can be tested in the future.

(2) There is no data showing that the engineered antibody has the desired pharmacokinetics/pharmacodynamics properties or efficacy in vivo.

In this ms we did not conduct in-vivo experiments. When moving forward, pharmacokinetics/pharmacodynamics properties and efficacy will be tested as well.

(3) Computational approaches are only used to design a phage-screen library, but not used to prioritize mutations that are likely to improve binding (e.g. based on predicted impact on the stability of the interaction). A demonstration of how computational pre-screening or lead optimization can improve the time-intensive process would be a welcome advance.

Thank you for this important comment. In the present ms we indeed used a computational approach for prioritizing residues to be mutated, but we did not prioritize the mutations that are likely to improve binding. In the initial library design, we did prioritize the mutations. However, due to experimental approach limitations with codon’s selection for the library, we had decided to allow all possible residues in each position, knowing that the selection will remove non-binding variants.

Context:

The conflict of interest statement is inadequate. Most authors of the study (but not the first author) are employees of Biolojic, a company developing multi-specific antibodies, but the statements do not clarify whether the presented antibodies represent Biolojic IP, whether the company sponsored the research, and whether the company is further developing the specific antibodies presented.

The Conflict-of-Interest statement will be revised as such: The Biolojic Design authors are employees of Biolojic Design and have stock options in Biolojic Design. The company did not sponsor the research, does not hold IP for the presented antibodies, and is not further developing the presented antibodies.

Reviewer #2 (Public Review):

Strengths:

(1) Deep computational analyses of large datasets of clinical data provide useful information about HER2 mutations and their potential relevance to antibody therapy resistance.

(2) There is valuable information analyzing the residues within or near the interface between the antigen HER2 and the Pertuzumab antibody (heavy chain). The experimental antibody library screening obtained 90+ clones from 3.86×1011 sequences for further functional validation.

Weaknesses:

(1) There is a lack of assessment for antibody variant functions in cancer cell phenotypes in vitro (proliferation, cell death, motility) or in vivo (tumor growth and animal survival). The only assay was the western blotting of phosphopho-HER3 in Figure 4. However, HER2 levels and phosphor-HER2 were not analyzed.

We indeed did not assess the engineered antibodies function in cancer cells. While a complete signaling assessment obviously requires functional assessment as well, due to the complexity of this assay, papers in this field (for example [1-3]) measure the signaling activation following HER2-HER3 dimerization by measuring pHER3, and we relied on them in this ms.

(2) There is a misleading impression from the title of computational engineering of a therapeutic antibody and the statement in the abstract "we designed a multi-specific version of Pertuzumab that retains original function while also bindings these HER2 variants" for a few reasons:

a. The primary method used for variant antibody identification for HER2 mutant binding is rather traditional experimental screening based on yeast display instead of the computational design of a multi-specific version of Pertuzumab.

b. There is insufficient or lack of computational power in the antibody design or prioritization in choosing variant residues for the library construction of 3.86×1011 sequences. It seems random combinations from 6 residues out of 4 groups with 20 amino acid options.

c. The final version of the tri-binding variant is a combination of screened antibody clones instead of computation design from scratch.

d. There is incomplete experimental evidence about the therapeutic values of newly obtained antibody clones.

Thank you for this relevant comment. When addressing relevant residues to be mutated, the number of potential variants is enormous. The computational approach was aimed at identifying the most preferable residues, in which variation can improve binding and is not likely to harm important interactions. Although an initial smaller number of residues could be chosen, we decided to broaden our view and create a larger library, in the aim of combining the computational selection with an experimental selection. This indeed is not a computational design from scratch, but rather an intercourse between the computer and the lab, that yielded the presented results.

(3) Figures can be improved with better labeling and organization. Some essential pieces of data such as Supplementary Figure 1B on HER2 mutations in S310 that abrogated its binding to Pertuzumab should be placed in the main figures.

Thank you for this comment, the relevant figures were moved to the main text, and the labels were revised.

(4) It is recommended to provide a clear rationale or flowchart overview into the main Figure 1. Figure 2A can be combined with Figure 1 to the list of targeted residues.

Figures 1 and 2 were divided differently, and the rationale was moved to the main text.

(5) The quality of Figures such as Figure 2B-C flow data needs to be improved.

High-quality figures were submitted with the revised ms.

Reviewer #1 (Recommendations for The Authors):

Major:

(1) It should be clarified whether the S310 somatic mutations represent resistance mutations to Pertuzamab (i.e. emerge post-therapy) or are general mutations that activate HER2. This is important because mutations that specifically "evade" the binding of an antibody may be substantially more difficult to overcome than mutations that only by chance occur in the antibody binding site. This concern should be addressed in the introduction and discussion as it changes the interpretation of the data.

This is a very important note. To the best of our knowledge, these mutations were not identified as resistance mutations that emerged post-therapy. However, as mentioned in the introduction, these mutations form hydrophobic interactions that stabilize HER2 dimerization. Moreover, cells expressing these mutations show hyperphosphorylation of HER2 and an increase in the subsequent activation of signaling pathways. Thus, these mutations do not necessarily evade Pertuzumab binding, but benefit cancer growth. This point was clarified in the introduction of the revised text.

(2) While the authors claim that S310 germline pathogenic variants exist, I could not find evidence that this is the case. The dbGAP ID does not provide any evidence (either in the form of a citation or prevalence). The variants do not exist in GnomAD. A recent article discussing pathogenic ERBB2 germline variants only mentions S310 as a somatic variant https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8268839/ and I could not find evidence for S310 being a germline variant in the references provided by the author (https://www.nature.com/articles/nbt.3391) - where it is only mentioned as a somatic mutation. I could not find evidence of a cancer predisposition syndrome associated with this variant.

Thank you for highlighting this matter. We had assumed that the presence of the variant in dbSNP means it is also a germline mutations, what may not be correct. However, we did find some evidence of this mutation as germline in ClinVar, and this was edited in the revised ms. https://www.ncbi.nlm.nih.gov/clinvar/RCV001311879.7.

(3) The authors should consider experiments that show that the modified Pertuzamab has the same mechanism of action as the original Pertuzamab in preventing dimerization of the ERBB2 homodimer and/or interactions with ERBB3. I cannot recommend a specific approach, but at present it is not clear whether the mechanism or just the effect (phosphorylation of ERBB3) is the same.

As mentioned above, for the assessment of HER-HER3 binding and HER3 signaling, in this ms we relied on a previous works [1-3] that also measured the signaling activation following HER2-HER3 dimerization by measuring pHER3.

(4) The authors should perform in vitro experiments to demonstrate that the engineered antibody has similar on-target specificity not only sensitivity. I don't know what the ideal experiments would be, but should probably probe native epitopes. Western blots, immunoprecipitation of cell lysates?

As mentioned above, showing the specificity of the engineered antibodies is indeed important. We did not address it in the current ms, but it can be tested in future work.

Minor:

(1) The introduction should review better the literature on the computational/rational design of antibodies, especially multi-specific - and likely de-emphasize small molecules (and mutations associated with the resistance thereof) as the presented research does not inform the design of mutation-agnostic small molecules.

Thank you for these comments, the introduction was revised accordingly.

(2) The authors should better present the fact that the lack of binding of Pertuzamab to HER2 S310 was previously known, thus the whole strategy of searching COSMIC, and computationally predicting their binding impact was unnecessary. Rather it would be helpful to learn how many other COSMIC hotspots could have a similar effect on other clinical antibodies.

The lack of binding was indeed previously known, as mentioned in the introduction. However, we did not start our analysis targeting HER2 specifically, but we rather found these mutations because they were located in the binding pocket, which enabled our strategy to compensate for these mutations with alteration of the original Pertuzumab. Regarding other potential hotspots, the numbers appeared in Supplementary Table 1, and were moved to the main text.

Stylistic:

(1) Avoid using the term "drug" for an antibody.

The term was changed to “antibody therapeutics” in the revised text.

(2) Avoid repetition in the introduction.

Thank you, we revised the introduction with this comment in mind.

Reviewer #2 (Recommendations For The Authors):

The quality of Figure 2B-C flow data needs to be improved:

a. The diagonal populations suggest inappropriate color compensation or indicate cells are derived from unhealthy populations.

We believe there may be some confusion here. The figures you are referring to are figures of very diverse library. The selected clones show nice diagonals, as shown in Supplementary Figure 5.

b. Additional round 3 and round 4 did not seem to improve the enrichment of targeted clones but rather had similar binding profiles to each of the three proteins over and over.

Two sets of the fourth round of selection were done, each originated from a different sub-population in round 3: 1. Clones that bind the S310Y mutation 2. Clones that bind the S310F mutation. The aim of the R4 was to examine this binders against the second mutation and canonical HER2 in the search for multi-specificity. Additional clarification of this point will be added to the main text.

c. Figure legends are vague with non-specific descriptions of cells and conditions, and unclear statements of "FACS results...".

The legends were edited in the revised version.

d. Text fonts are in low resolution.

High-quality figures were submitted with the revised ms.

(1) Diwanji, D., et al., Structures of the HER2-HER3-NRG1β complex reveal a dynamic dimer interface. Nature, 2021. 600(7888): p. 339-343.

(2) Yamashita-Kashima, Y., et al., Mode of action of pertuzumab in combination with trastuzumab plus docetaxel therapy in a HER2-positive breast cancer xenograft model. Oncol Lett, 2017. 14(4): p. 4197-4205.

(3) Kang, J.C., et al., Engineering multivalent antibodies to target heregulin-induced HER3 signaling in breast cancer cells. MAbs, 2014. 6(2): p. 340-53.

eLife assessment

This important study demonstrates that combining AlphaFold2 with the author's sampling method AF2-RAVE improves protein-ligand docking for three protein kinases and their inhibitors. The evidence is compelling and the results will be of interest to researchers who work on computer-aided drug design.

Reviewer #1 (Public Review):

The development of effective computational methods for protein-ligand binding remains an outstanding challenge to the field of drug design. This impressive computational study combines a variety of structure prediction (AlphaFold2) and sampling (RAVE) tools to generate holo-like protein structures of three kinases (DDR1, Abl1, and Src kinases) for binding to type I and type II inhibitors. Of central importance to the work is the conformational state of the Asp-Phy-Gly "DFG motif" where the Asp points inward (DFG-in) in the active state and outward (DFG-out) in the inactive state. The kinases bind to type I or type II inhibitors when in the DFG-in or DFG-out states, respectively.

It is noted that while AlphaFold2 can be effective in generating ligand-free apo protein structures, it is ineffective at generating holo structures appropriate for ligand binding. Starting from the native apo structure, structural fluctuations are necessary to access holo-like structures appropriate for ligand-binding. A variety of methods, including reduced multiple sequence alignment (rMSA), AF2-cluster, and AlphaFlow may be used to create decoy structures. However, those methods can be limited in the diversity of structures generated and lack a physics-based analysis of Boltzmann weight critical to their relative evaluation.

To address this need, the authors combine AlphaFold2 with the Reweighted Autoencoded Variational Bayes for Enhanced Sampling (RAVE) method, to explore metastable states and create a Boltzmann ranking. With that variety of structures in hand, grid-based docking methods Glide and Induced-Fit Docking (IFD) were used to generate protein-ligand (kinase-inhibitor) complexes.

The authors demonstrate that using AlphaFold2 alone, there is a failure to generate DFG-out structures needed for binding to type II inhibitors. By applying the AlphaFold2 with rMSA followed by RAVE (using short MD trajectories, SPIB-based collective variable analysis, and enhanced sampling using umbrella sampling), metastable DFG-out structures with Boltzmann weighting are generated enabling protein-ligand binding. Moreover, the authors found that the successful sampling of DFG-out states for one kinase (DDR1) could be used to model similar states for other proteins (Abl1 and Src kinase). The AF2RAVE approach is shown to result in a set of holo-like protein structures with a 50% rate of docking type II inhibitors.

Overall, this is excellent work and a valuable contribution to the field that demonstrates the strengths and weaknesses of state-of-the-art computational methods for protein-ligand binding. The authors also suggest promising directions for future study, noting that potential enhancements in the workflow may result from the use of binding site prediction models and free energy perturbation calculations.

Reviewer #2 (Public Review):

This manuscript explores the utility of AlphaFold2 (AF2) and the author's own AF2-RAVE method for drug discovery. As has been observed elsewhere, the predictive power of docking against AF2 structures is quite limited, particularly for proteins like kinases that have non-trivial conformational dynamics. However, using enhanced sampling methods like RAVE to explore beyond AF2 starting structures leads to a significant improvement.

Comments on revised version:

I'm happy with the changes made.

Reviewer #3 (Public Review):

In this manuscript, the authors aim to enhance AlphaFold2 for protein conformation-selective drug discovery through the integration of AlphaFold2 and physics-based methods, focusing on improving the accuracy of predicting protein structures ensemble and small molecule binding of metastable protein conformations to facilitate targeted drug design.

The major strength of the paper lies in the methodology, which includes the innovative integration of AlphaFold2 with all-atom enhanced sampling molecular dynamics and induced fit docking to produce protein ensembles with structural diversity. Moreover, the generated structures can be used as reliable crystal-like decoys to enrich metastable conformations of holo-like structures. The authors demonstrate the effectiveness of the proposed approach in producing metastable structures of three different protein kinases and perform docking with their type I and II inhibitors. The paper provides strong evidence supporting the potential impact of this technology in drug discovery. However, limitations may exist in the generalizability of the approach across other structures, especially complex structures such as protein-protein or DNA-protein complexes.

The authors largely achieved their aims by demonstrating that the AF2RAVE-Glide workflow can generate holo-like structure candidates with a 50% successful docking rate for known type II inhibitors. This work is likely to have a significant impact on the field by offering a more precise and efficient method for predicting protein structure ensemble, which is essential for designing targeted drugs. The utility of the integrated AF2RAVE-Glide approach may streamline the drug discovery process, potentially leading to the development of more effective and specific medications for various diseases.

Comments on revised version:

The revised manuscript looks great to me. I have no further comments.

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

The development of effective computational methods for protein-ligand binding remains an outstanding challenge to the field of drug design. This impressive computational study combines a variety of structure prediction (AlphaFold2) and sampling (RAVE) tools to generate holo-like protein structures of three kinases (DDR1, Abl1, and Src kinases) for binding to type I and type II inhibitors. Of central importance to the work is the conformational state of the Asp-Phy-Gly "DFG motif" where the Asp points inward (DFG-in) in the active state and outward (DFG-out) in the inactive state. The kinases bind to type I or type II inhibitors when in the DFG-in or DFG-out states, respectively.

It is noted that while AlphaFold2 can be effective in generating ligand-free apo protein structures, it is ineffective at generating holo-structures appropriate for ligand binding. Starting from the native apo structure, structural fluctuations are necessary to access holo-like structures appropriate for ligand binding. A variety of methods, including reduced multiple sequence alignment (rMSA), AF2-cluster, and AlphaFlow may be used to create decoy structures. However, those methods can be limited in the diversity of structures generated and lack a physics-based analysis of Boltzmann weight critical to their relative evaluation.

To address this need, the authors combine AlphaFold2 with the Reweighted Autoencoded Variational Bayes for Enhanced Sampling (RAVE) method, to explore metastable states and create a Boltzmann ranking. With that variety of structures in hand, grid-based docking methods Glide and Induced-Fit Docking (IFD) were used to generate protein-ligand (kinase-inhibitor) complexes.

The authors demonstrate that using AlphaFold2 alone, there is a failure to generate DFG-out structures needed for binding to type II inhibitors. By applying the AlphaFold2 with rMSA followed by RAVE (using short MD trajectories, SPIB-based collective variable analysis, and enhanced sampling using umbrella sampling), metastable DFG-out structures with Boltzmann weighting are generated enabling protein-ligand binding. Moreover, the authors found that the successful sampling of DFG-out states for one kinase (DDR1) could be used to model similar states for other proteins (Abl1 and Src kinase). The AF2RAVE approach is shown to result in a set of holo-like protein structures with a 50% rate of docking type II inhibitors.

Overall, this is excellent work and a valuable contribution to the field that demonstrates the strengths and weaknesses of state-of-the-art computational methods for protein-ligand binding. The authors also suggest promising directions for future study, noting that potential enhancements in the workflow may result from the use of binding site prediction models and free energy perturbation calculations.

Reviewer #2 (Public Review):

Summary:

This manuscript explores the utility of AlphaFold2 (AF2) and the author's own AF2-RAVE method for drug discovery. As has been observed elsewhere, the predictive power of docking against AF2 structures is quite limited, particularly for proteins like kinases that have non-trivial conformational dynamics. However, using enhanced sampling methods like RAVE to explore beyond AF2 starting structures leads to a significant improvement.

Strengths:

This is a nice demonstration of the utility of the authors' previously published RAVE method.

Weaknesses:

My only concern is the authors' discussion of induced fit. I'm quite confident the structures discussed are present in the absence of ligand binding, consistent with conformational selection. It seems the author's own data also argues for an important role in conformational selection. It would be nice to acknowledge this instead of going along with the common practice in drug discovery of attributing any conformational changes to induced fit without thoughtful consideration of conformational selection.

The reviewer is correct. We aim to highlight the significant role of conformational selection. To clarify this, we have expanded the discussion on conformational selection in the introduction.

Reviewer #3 (Public Review):

In this manuscript, the authors aim to enhance AlphaFold2 for protein conformation-selective drug discovery through the integration of AlphaFold2 and physics-based methods, focusing on improving the accuracy of predicting protein structures ensemble and small molecule binding of metastable protein conformations to facilitate targeted drug design.

The major strength of the paper lies in the methodology, which includes the innovative integration of AlphaFold2 with all-atom enhanced sampling molecular dynamics and induced fit docking to produce protein ensembles with structural diversity. Moreover, the generated structures can be used as reliable crystal-like decoys to enrich metastable conformations of holo-like structures. The authors demonstrate the effectiveness of the proposed approach in producing metastable structures of three different protein kinases and perform docking with their type I and II inhibitors. The paper provides strong evidence supporting the potential impact of this technology in drug discovery. However, limitations may exist in the generalizability of the approach across other structures, especially complex structures such as protein-protein or DNA-protein complexes.

Proteins undergo thermodynamic fluctuations and can occasionally reach metastable configurations. It can be assumed that other biomolecules, such as proteins and DNA, stabilize these metastable states when forming protein-protein or protein-DNA complexes. Since our method has the potential to identify these metastable states, it shows promise for designing drugs targeting proteins in allosteric configurations induced by other biomolecules.

The authors largely achieved their aims by demonstrating that the AF2RAVE-Glide workflow can generate holo-like structure candidates with a 50% successful docking rate for known type II inhibitors. This work is likely to have a significant impact on the field by offering a more precise and efficient method for predicting protein structure ensemble, which is essential for designing targeted drugs. The utility of the integrated AF2RAVE-Glide approach may streamline the drug discovery process, potentially leading to the development of more effective and specific medications for various diseases.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Suggestions

(1) The computational protocol is found to be insufficient to generate precise values of the relative free energies between structures generated. The authors note in the Conclusion that an enhancement in the workflow might result from the addition of free energy calculations. Can the authors comment on the prospects for generating more accurate estimates of the free energy that might be used to qualitatively evaluate poses and the free energy landscape surrounding putative metastable states? What are the principal challenges and what might help overcome them? What would the most effective computational protocol be?

More accurate estimates of the free energy can theoretically be achieved by increasing the number of umbrella sampling windows and extending the simulation length until the PMF converges. However, there is always a trade-off between PMF accuracy and computational costs, so we have chosen to stick with the current setup. Metadynamics is another method to obtain a more accurate free energy profile, which we have used in previous versions of AlphaFold2-RAVE, but for the specific systems we investigated, it had issues in achieving back and forth movement given the high entropic nature of the activation loop. Research in enhanced sampling methods and dimensionality reduction techniques for reaction coordinates is continually evolving and will play a critical role in alleviating this problem.

(2) I was surprised that there was not more correlation of a funnel-like shape in Figures S16 and S18, showing a stronger correlation between low RMSD and better docking score. This is true for both the ponatinib and imatinib applications in DDR1 and Abl1. That also seems true for the trimmed results for Src kinase in Figure S19. I was also surprised that there are structures with very large RMSD but docking scores comparable to the best structures of the lowest RMSD. Might something be done to make the docking score a more effective discriminator?

The docking algorithm and docking score are used to filter out highly improbable docking poses. False positives in predicted docking poses are a common issue across all docking methods as described for instance in:

Fan, Jiyu, Ailing Fu, and Le Zhang. "Progress in molecular docking." Quantitative Biology 7 (2019): 83-89.

Ferreira, R.S., Simeonov, A., Jadhav, A., Eidam, O., Mott, B.T., Keiser, M.J., McKerrow, J.H., Maloney, D.J., Irwin, J.J. and Shoichet, B.K., 2010. "Complementarity between a docking and a high-throughput screen in discovering new cruzain inhibitors." Journal of medicinal chemistry, 53(13), pp.4891-4905.

Moreover, there is always a trade-off between docking accuracy and computational cost. While employing more accurate docking methods may decrease false positives, it can also be resource-intensive. In such scenarios, our approach to enriching holo-structures can be impactful by reducing the number of pocket structures in the input ensembles and significantly enhancing docking efficiency.

(3) I think that it is fine to identify one structure as "IFD winner" but also feel that its significance is overstressed, especially given that it can be identified only in a retrospective analysis rather than through de novo prediction.

We agree with the reviewer. We did not intend to emphasize the specific structure "IFD winner". Rather, we aimed to demonstrate that our method can enrich promising candidates for holo-structures. We verified this by showing that our holo-structure candidates performed well in retrospective docking using IFD, which we previously referred to as "IFD winner". We have now revised this term to "holo-model".

Minor Points

p. 3 "DymanicBind" should be "DynamicBind"

p. 3 Change "We chosen" to "We have chosen" or "we chose."

p. 3 In identifying the Schrödinger software Glide and IFD, I recommend removing the subjective modifier "industry-leading."

Modifications done.

Reviewer #2 (Recommendations For The Authors):

In the view of this reviewer, the writing is 'choppy'.

We have tried to improve the writing.

Reviewer #3 (Recommendations For The Authors):

(1) In Figure 1, the workflow labels (i) to (iv) are not shown on the figures, making it difficult for readers to follow. Consider adding these labels to the figures.

Modifications done.

(2) Explain how Boltzmann ranks were calculated based on unbiased MD simulations to guide the enrichment of holo-like structures in metastable states.

The Methods section is now updated for clarification.

(3) The authors could clarify how the classical DFG-out decoys in the DDR1 rMSA AF2 ensemble are transferred to Abl1 kinase in the Methods section.

The Methods section is now updated for clarification.

(4) The authors can clarify the methodology section by providing more detailed explanations about how the unbiased MD simulations are performed, including which MD simulation software was used and whether energy minimization and equilibrium steps were needed as in conventional MD simulations, and other setup details.

The Methods section is now updated for clarification.

(5) The validation of the proposed approach in this work used three kinase proteins. The authors can enhance the discussion section by addressing other types of protein structure prediction that can use the proposed approach in drug discovery, beyond the three kinase proteins tested.

The proposed approach is theoretically applicable to other types of proteins, such as GPCRs, where both conformational selection and the induced-fit effect are crucial. We have expanded the discussion on the generalization of our protocol in the Conclusion section.

(6) The authors should add appropriate citations for the software and tools used in the manuscript. For example, a reference should be added for the Glide XP docking experiments that utilized the Maestro software. Double-check all related software citations.

We have now updated the citations for docking experiments based on the instruction of the Maestro Glide User manual and IFD User manual.

(7) The authors should consider offering a comprehensive list of software tools and databases utilized in the study to assist in replicating the experiments and further validating the results.

We have now added a summary of tools used in the Methods section.

eLife assessment

This valuable manuscript describes a novel role of Vangl2, a core planar cell polarity protein, in linking the NF-kB pathway to selective autophagic protein degradation in myeloid cells. The mechanistic studies provide convincing evidence that Vangl2 targets p65 for NDP52-mediated autophagic degradation, limiting inflammatory NF-kB response, with functional significance of the proposed mechanism in sepsis. Additional future studies dissecting autophagic Vangl2 functions in various myeloid subsets in the context of inflammation could be informative, and additional Vangl2 targets in the inflammatory pathway, including IKK2, could also be explored. Overall, this exciting study can advance our understanding of NF-kB control, particularly in the context of inflammatory diseases.

Reviewer #1 (Public Review):

The study shows a new mechanism of NFkB-p65 regulation mediated by Vangl2-dependent autophagic targeting. Autophagic regulation of p65 has been reported earlier; this study brings an additional set of molecular players involved in this important regulatory event, which may have implications for chronic and acute inflammatory conditions.

Reviewer #2 (Public Review):

Vangl2, a core planar cell polarity protein involved in Wnt/PCP signaling, cell proliferation, differentiation, homeostasis, and cell migration. Vangl2 malfunctioning has been linked to various human ailments, including autoimmune and neoplastic disorders. Interestingly, it was shown that Vangl2 interacts with the autophagy regulator p62, and autophagic degradation limits the activity of inflammatory mediators, such as p65/NF-κB. However, the possible role of Vangl2 in inflammation has not been investigated. In this manuscript, Lu et al. describe that Vangl2 expression is upregulated in human sepsis-associated PBMCs and that Vangl2 mitigates experimental sepsis in mice by negatively regulating p65/NF-κB signaling in myeloid cells. Their mechanistic studies further revealed that Vangl2 recruits the E3 ubiquitin ligase PDLIM2 to promote K63-linked poly-ubiquitination of p65. Vangl2 also facilitated the recognition of ubiquitinated p65 by the cargo receptor NDP52. These molecular processes caused selective autophagic degradation of p65. Indeed, abrogation of PDLIM2 or NDP52 functions rescued p65 from autophagic degradation, leading to extended p65/NF-κB activity in myeloid cells. Overall, the manuscript presents convincing evidence for novel Vangl2-mediated control of inflammatory p65/NF-kB activity. The proposed pathway may expand interventional opportunities restraining aberrant p65/NF-kB activity in human ailments.

IKK is known to mediate p65 phosphorylation, which instructs NF-kB transcriptional activity. In this manuscript, Vangl2 deficiency led to an increased accumulation of phosphorylated p65 and IKK also at 30 minutes post-LPS stimulation; however, autophagic degradation of p-p65 may not have been initiated at this early time point. Therefore, this set of data put forward the exciting possibility that Vangl2 could also be regulating the immediate early phase of inflammatory response involving the IKK-p65 axis - a proposition that may be tested in future studies.

Reviewer #3 (Public Review):

Lu et al. describe Vangl2 as a negative regulator of inflammation in myeloid cells. The primary mechanism appears to be through binding p65 and promoting its degradation, albeit in an unusual autolysosome/autophagy dependent manner. Overall, these findings are novel, valuable and the crosstalk of PCP pathway protein Vangl2 with NF-kappaB is of interest.

Comments on latest version:

Lu et al. now address all my comments. All data included for the reviewers should be included in the main manuscript or Supplement and should be available to the readers. Please ensure that this criteria is met. I have no further comments.

Author response:

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

Responses to Reviewer #1:

Reviewer #1: The study shows a new mechanism of NFkB-p65 regulation mediated by Vangl2-dependent autophagic targeting. Autophagic regulation of p65 has been reported earlier; this study brings an additional set of molecular players involved in this important regulatory event, which may have implications for chronic and acute inflammatory conditions.

Comments on the revised version:

The authors have addressed the earlier concerns and I am satisfied with the revised version. I have no additional comments to make.

We appreciate the reviewer’s comments on our revised manuscript.

Responses to Reviewer #2:

Reviewer #2: Vangl2, a core planar cell polarity protein involved in Wnt/PCP signaling, cell proliferation, differentiation, homeostasis, and cell migration. Vangl2 malfunctioning has been linked to various human ailments, including autoimmune and neoplastic disorders. Interestingly, it was shown that Vangl2 interacts with the autophagy regulator p62, and autophagic degradation limits the activity of inflammatory mediators, such as p65/NF-κB. However, the possible role of Vangl2 in inflammation has not been investigated. In this manuscript, Lu et al. describe that Vangl2 expression is upregulated in human sepsis-associated PBMCs and that Vangl2 mitigates experimental sepsis in mice by negatively regulating p65/NF-κB signaling in myeloid cells. Their mechanistic studies further revealed that Vangl2 recruits the E3 ubiquitin ligase PDLIM2 to promote K63-linked poly-ubiquitination of p65. Vangl2 also facilitated the recognition of ubiquitinated p65 by the cargo receptor NDP52. These molecular processes caused selective autophagic degradation of p65. Indeed, abrogation of PDLIM2 or NDP52 functions rescued p65 from autophagic degradation, leading to extended p65/NF-κB activity in myeloid cells. Overall, the manuscript presents convincing evidence for novel Vangl2-mediated control of inflammatory p65/NF-kB activity. The proposed pathway may expand interventional opportunities restraining aberrant p65/NF-kB activity in human ailments.

IKK is known to mediate p65 phosphorylation, which instructs NF-kB transcriptional activity. In this manuscript, Vangl2 deficiency led to an increased accumulation of phosphorylated p65 and IKK also at 30 minutes post-LPS stimulation; however, autophagic degradation of p-p65 may not have been initiated at this early time point. Therefore, this set of data put forward the exciting possibility that Vangl2 could also be regulating the immediate early phase of inflammatory response involving the IKK-p65 axis - a proposition that may be tested in future studies.

We appreciate the reviewer’s comments on our manuscript, and we have added the discussion about IKK-p65 axis in revised version. (Page 15, lines 467-474)

Responses to Reviewer #3:

Reviewer #3: Lu et al. describe Vangl2 as a negative regulator of inflammation in myeloid cells. The primary mechanism appears to be through binding p65 and promoting its degradation, albeit in an unusual autolysosome/autophagy dependent manner. Overall, these findings are novel, valuable and the crosstalk of PCP pathway protein Vangl2 with NF-kappaB is of interest. While generally solid, some concerns still remain about the rigor and conclusions drawn.

Comments on the revised version:

(1) Lu et al. address my comments through responses and new experimental data. However, some of the explanations provided are inadequate.

However, in response to my enquiry regarding directly exploring PCP effects, the authors simply assert "Our study revealed that Vangl2 recruits the E3 ubiquitin ligase PDLIM2 to facilitate K63-linked ubiquitination of p65, which is subsequently recognized by autophagy receptor NDP52 and then promotes the autophagic degradation of p65. Our findings by using autophagy inhibitors and autophagic-deficient cells indicate that Vangl2 regulates NFkB signaling through a selective autophagic pathway, rather than affecting the PCP pathway, WNT, HH/GLI, Fat-Dachsous or even mechanical tension."

I do not agree that the use of autophagy inhibitors and autophagy-deficient cells can rule out the contributions of PCP or any other pathways. Only experimentally inhibiting the pathway(s) with adequate demonstration of target inhibition/abolition of well-known effector function and documenting unaltered p65 regulation under these conditions can be considered proof. Autophagy inhibitors and autophagy-deficient cells only prove that this particular pathway is necessary. Nonetheless, I do not want to dwell on proving a negative and agree that Vangl2 is a novel regulator of p65 through its role in promoting p65 degradation. The inclusion of a statement discussing the limitations of their approach would have sufficed. The response from the authors could have been better.

We thank the reviewer for helping us improve the quality of the manuscript. We provided new data and revised the Discussion as suggested.

To ascertain whether Vangl2 degrades p65 through a selective autophagic pathway or the PCP pathway, 293T cells were transfected with p65, together with or without the Vangl2 plasmids, and treated with different pharmacological inhibitors. We found the degradation of p65 induced by Vangl2 was blocked by autolysosome inhibitor (CQ), but not by the JNK inhibitor (SP600125) or Wnt/β-catenin inhibitor (FH535) (New Figure. 1). These data suggest that Vangl2 primarily degrades p65 through a selective autophagic pathway, rather than through the JNK or Wnt signaling pathway. Nevertheless, additional pathway inhibitions, such as those of the HH/GLI and Fat-Dachsous pathways, should also be employed to further elucidate the function of Vangl2 in p65 degradation. As suggested, we have added a statement about the limitation of the approach in the discussion (Page 12, lines 378-385).

Author response image 1.

Vangl2 degrades p65 through a selective autophagic pathway, but not by the PCP pathway. HEK293T cells were transfected with Flag-p65 and HA-Vangl2 plasmids, and treated with DMSO, CQ (50 mM) for 6 h, SP600125 (20 mM) for 1 h or FH535 (30 mM) for 6 h. The cell lysates were analyzed by immunoblot.

(2) I am also not satisfied with the explanation that "immune cells represent a minor fraction of the lungs and liver". There are lots of resident immune cells in the lungs and liver (alveolar macrophages in the lung and Kuppfer cells in the liver). For example, it may be so that Vangl2 is important in monocytes and not in the resident population. This might be a potential explanation. But this is not explored. The restricted tissue-specificity of the interaction between two ubiquitously present proteins is still a challenge to understand. The response from the authors is not satisfactory. There is plenty of Vangl2 in the liver in their western blot.

We thank the reviewer for this question. We added this explanation in the Discussion. (Page 13, lines 398-404)

(3) I had also simply pointed out PMID: 34214490 with reference to the findings described in the manuscript. There were no suggestions of contradiction. In fact, I would refer to the publication in discussion to support the findings and stress the novelty. The response from the authors could have been better.

Thank you for the reviewer's insightful comments. We have modified this discussion as suggested. (Page 13, lines 410-415; Page 14, lines 419-421)

(4) The response to my enquiry regarding homo- or heterozygosity is unsupported by any reference or data.

As suggested, we provided the data that only Vangl2 deficient homozygous showed inhibition of the activation of NF-kB in New Figure. 2.

Author response image 2.

Vangl2 deficiency promotes NF-kB activation. (A) The survival rates of WT, Vangl2ΔM/ΔM and Vangl2ΔM/WT mice treated with high-dosage of LPS (30 mg/kg, i.p.) (n≥4). (B) IL-6 and TNF-a secretion by WT and Vangl2-deficient BMDMs treated with LPS for 6 h was measured by ELISA. IL-1β secretion by WT, Vangl2ΔM/ΔM and Vangl2ΔM/WT BMDMs treated with LPS for 6 h and ATP for 30 min was measured by ELISA.

(5) The listing of 8 patients and healthy controls are also appreciated. The body temperature of #6 doesn't fall in the <36 or >38 degree C SIRS criteria. The inclusion of CRP, PCT, heart rate and respiratory rate, and other lab values would have further improved the inclusion criteria. Moreover, it is difficult to understand why there are 16 value points for healthy and sepsis cohorts in Fig 1 when there are 8 patients.

We thank the reviewer for this valuable suggestion. We are sorry for our mistake that we entered data from two repeated experiments in Figure. 1 A and we have revised this data in the updated version (Figure. 1 A, Pages 12 Lines 146). As suggested, we have added CRP, WBC and heart rate in sepsis patients’ information. (Supplementary Materials and Methods)

Recommendations for the authors:

Reviewer #2 (Recommendations For The Authors):

The proposition that Vangl2 may target additional mediators of inflammation could be indicated in the text.

We thank the reviewer for this valuable suggestion. We had added discussion in modified version. (Page 15, lines 467-474)

Reviewer #3 (Recommendations For The Authors):

It is advised that some of the deficiencies pointed out by Reviewer #3 are textually addressed. Additionally, there could be some inconsistency in the number of healthy controls and patients (see Fig S1A and FIg 1A and Supplementary table, also see comments from Reviewer #3) - this should be carefully scrutinised and revised, if necessary.

We thank the reviewer for this valuable suggestion. We are sorry for our mistake that we entered data from two repeated experiments in Figure. 1 A and we have revised this data in the updated version (Figure. 1 A, Pages 12 Lines 146).

eLife assessment

This manuscript presents valuable findings on two isolates of deep sea Lentisphaerae strains, which further our understanding of deep sea microbial life. The manuscript's primary claim is that phage isolates augment polysaccharide use in Pseudomonas bacteria, with preliminary evidence for the potential auxiliary metabolic genes in chronic phage infection and/or host proliferation. The strength of the evidence is overall solid and there are only minor weaknesses regarding the mechanism of polysaccharide use by the phages and the evidence for chronic infection. Overall, the data on Lentisphaerae strains will deepen our understanding of microbial life in the deep sea.

Reviewer #1 (Public Review):

Summary:

I have previously reviewed this manuscript as a submission to another journal in 2022. My recommendations here mirror those of my prior suggestions, now with further added details.

This manuscript describes the identification and isolation of several phage from deep sea isolates of Lentisphaerae strains WC36 and zth2. The authors observe induction of several putative chronic phages with the introduction of additional polysaccharides to the media. The authors suggest that two of the recovered phage genomes encode AMGs associated with polysaccharide use. The authors also suggest that adding the purified phage to cultures of Pseudomonas stutzeri 273 increased the growth of this bacteria due to augmented polysaccharide use genes from the phage.

Strengths:

Interesting isolate of deep sea Lentisphaerae strains which will undoubtedly further our understanding of deep sea microbial life.

The revisions have addressed the weaknesses raised in the previous review.

Reviewer #2 (Public Review):

Summary:

This paper investigates deep-sea bacteriophage systems which appear to employ a chronic replication mechanism that is induced or enhanced by polysaccharide addition. Some preliminary evidence for the potential role of auxiliary metabolic genes in aiding phage and/or host proliferation is also provided. The hypothesis being tested is fully supported with solid and convincing evidence and the findings are potentially generalizable with implications for our understanding of polysaccharide-mediated virus-host interactions and carbon cycling in marine ecosystems more broadly.

Strengths:

This paper synthesizes sequencing and phylogenic analyses of two Lentisphaerae bacteria and three phage genomes; electron microscopy imaging of bacterial/phage particles; differential gene expression analyses; differential growth curve analyses, and differential phage proliferation assays to extract insights into whether laminarin and starch can induce both host growth and phage proliferation. The data presented convincingly demonstrate that both host culture density and phage proliferation increase as a result having host, phage, and polysaccharide carbon source together in culture.

Weaknesses:

The AMG-centered elements of the article would be strengthened by more "mechanistic" experiments focusing on identifying "HOW" the polysaccharide processing, transport, and metabolism genes are being used by the phages to either directly increase viral infection/replication or else to indirectly do so by supporting the growth of the host (via mutualism). The concept of "selfishness" in bacterial systems and its potential role in viral life cycles could be more developed. Selfish bacteria are active throughout the water column of the ocean. ISME COMMUN. 3, 11 (2023) (see for instance https://doi.org/10.1038/s43705-023-00219-7) and such "selfish" bacteria sequester metabolizable polysaccharides in their periplasm to advantage (https://www.nature.com/articles/ismej201726). It is plausible that phages may be either hijacking such polysaccharide sequestration mechanisms to improve infectivity and ENTRY or else helping their hosts to grow and proliferate so they can reap the benefits of simply having more hosts to infect. The current work does not clearly distinguish between these two distinct mechanistic possibilities. The paper would be strengthened by a more detailed/clear discussion of this possibility.

Author response:

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

eLife assessment

This manuscript presents useful findings on several phage from deep sea isolates of Lentisphaerae strains WC36 and zth2 that further our understanding of deep sea microbial life. The manuscript's primary claim is that phage isolates augment polysaccharide use in Pseudomonas bacteria via auxiliary metabolic genes (AMGs). However, the strength of the evidence is incomplete and does not support the primary claims. Namely, there are not data presented to rule out phage contamination in the polysaccharide stock solution, AMGs are potentially misidentified, and there is missing evidence of successful infection.

Thanks for the Editor’s and Reviewers’ positive and constructive comments, which help us improve the quality of our manuscript entitled “Deep-sea bacteriophages facilitate host utilization of polysaccharides” (paper#eLife-RP-RA-2023-92345). The comments are valuable, and we have studied the comments carefully and have made corresponding revisions according to the suggestions. We removed some uncertain results and strengthened other parts of the manuscript, which evidently improved the accuracy and impact of the revised version. Revised portions are marked in blue in the modified manuscript. Please find the detailed responses as following.

Public Reviews:

Reviewer #1 (Public Review):

Summary: This manuscript describes the identification and isolation of several phage from deep sea isolates of Lentisphaerae strains WC36 and zth2. The authors observe induction of several putative chronic phages with the introduction of additional polysaccharides to the media. The authors suggest that two of the recovered phage genomes encode AMGs associated with polysaccharide use. The authors also suggest that adding the purified phage to cultures of Pseudomonas stutzeri 273 increased the growth of this bacterium due to augmented polysaccharide use genes from the phage. While the findings were of interest and relevance to the field, it is my opinion that several of the analysis fall short of supporting the key assertions presented.

Thanks for your comments. We removed some uncertain results and strengthened other parts of the manuscript, which evidently improved the accuracy and impact of the revised version. Please find the detailed responses as following.

Strengths: Interesting isolate of deep sea Lentisphaerae strains which will undoubtedly further our understanding of deep sea microbial life.

Thanks for your positive comments.  

Weaknesses:

(1) Many of the findings are consistent with a phage contamination in the polysaccharide stock solution. 

Thanks for your comments. We are very sure that the phages are specifically derived from the Lentisphaerae strain WC36 but not the polysaccharide stock solution. The reasons are as following: (1) the polysaccharide stock solution was strictly sterilized to remove any phage contamination; (2) we have performed multiple TEM checks of the rich medium supplemented with 10 g/L laminarin alone (Supplementary Fig. 1A) or in 10 g/L starch alone (Supplementary Fig. 1B), and there were not any phage-like structures, which confirmed that the polysaccharides (laminarin/starch) we used were not contaminated with any phage-like structures; in addition, we also observed the polysaccharides (laminarin/starch) directly by TEM and did not find any phage-like structures (Supplementary Fig. 2); (3) the polysaccharide (starch) alone could not promote the growth of Pseudomonas stutzeri 273, however, the supplement of starch together with the extracted Phages-WC36 could effectively facilitate the growth of Pseudomonas stutzeri 273 (Response Figure 1). The above results clearly indicated the phages were derived from the Lentisphaerae strain WC36 but not the polysaccharide stock solution. 

Author response figure 1.

Growth curve and status of Pseudomonas stutzeri 273 cultivated in basal medium, basal medium supplemented with 20 μl/mL Phages-WC36, basal medium supplemented with 5 g/L starch, basal medium supplemented with 5 g/L starch and 20 μl/mL Phages-WC36. 

 

(2) The genes presented as AMGs are largely well known and studied phage genes which play a role in infection cycles.

Thanks for your comments. Indeed, these AMGs may be only common in virulent phages, while have never been reported in chronic phages. In virulent phages, these genes typically act as lysozymes, facilitating the release of virions from the host cell upon lysis, or injection of viral DNA upon infection. However, the chronic phages do not lyse the host. Therefore, the persistence of these genes in chronic phages may be due to their ability to assist the host in metabolizing polysaccharides. Finally, according to your suggestions, we have weakened the role of AMGs and added “potential” in front of it. The detailed information is shown below.

(3) The evidence that the isolated phage can infect Pseudomonas stutzeri 273 is lacking, putting into question the dependent results.

Thanks for your comments. Actually, we selected many marine strains (Pseudomonadota, Planctomycetes, Verrucomicrobia, Fusobacteria, and Tenericutes isolates) to investigate whether Phages-WC36 could assist them in degradation and utilization of polysaccharides, and found that Phages-WC36 could only promote the growth of strain 273. It is reported that filamentous phages could recognize and bind to the host pili, which causes the pili to shrink and brings the filamentous phages closer to and possibly through the outer membrane of host cells. The possible mechanism of other chronic phages release without breaking the host might be that it was enclosed in lipid membrane and released from the host cells by a nonlytic manner. Thus, these chronic phages may have a wider host range. However, we were unable to further reveal the infection mechanism due to some techniques absence. Therefore, according to your suggestions, we have deleted this section in the revised manuscript.

Reviewer #1 (Recommendations For The Authors):

I have previously reviewed this manuscript as a submission to another journal in 2022. My recommendations here mirror those of my prior suggestions, now with further added details.

Thanks for your great efforts for reviewing our manuscript and valuable suggestions for last and this versions.

Specific comments:

Comment 1: Line 32. Rephrase to "polysaccharides cause the induction of multiple temperate phages infecting two strains of Lentisphaerae (WC36 and zth2) from the deep sea."

Thanks for your positive suggestion. We have modified this description as “Here, we found for the first time that polysaccharides induced the production of multiple temperate phages infecting two deep-sea Lentisphaerae strains (WC36 and zth2).” in the revised manuscript (Lines 31-33). 

Comment 2: Line 66. "Chronic" infections are not "lysogenic" as described here, suggesting the former is a subcategory of the latter. If you are going to introduce lifecycles you need a brief sentence distinguishing "chronic" from "lysogenic"

Thanks for your positive suggestion. We added this sentence as “Currently, more and more attention has been paid to chronic life cycles where bacterial growth continues despite phage reproduction (Hoffmann Berling and Maze, 1964), which was different from the lysogenic life cycle that could possibly lyse the host under some specific conditions.” in the revised manuscript (Lines 66-69).

Comment 3: Line 72. Please avoid generalized statements like "a hand-full" (or "plenty" line 85). Try to be at least somewhat quantitative regarding how many chronic phages are known. This is a fairly common strategy among archaeal viruses. 

Thanks for your suggestion. Given that some filamentous phages also have a chronic life cycle that is not explicitly reported, we cannot accurately estimate their numbers. According to your suggestions, we have modified these descriptions as “however, to our best knowledge, only few phages have been described for prokaryotes in the pure isolates up to date (Roux et al., 2019; Alarcón-Schumacher et al., 2022; Liu et al., 2022).” in the revised manuscript (Lines 73-75). In addition, the number of chronic phages in the biosphere cannot be accurately estimated, according to the latest report (Chevallereau et al., 2022), which showed that “a large fraction of phages in the biosphere are produced through chronic life cycles”. Therefore, we have modified this description as “Therefore, a large percentage of phages in nature are proposed to replicate through chronic life cycles” in the revised manuscript (Lines 87-88). 

Comment 4: Line 93. While Breitbart 2012 is a good paper to cite here, there have been several, much more advanced analysis of the oceans virome. https://doi.org/10.1016/j.cell.2019.03.040 is one example, but there are several others. A deeper literature review is required in this section.  

Thanks for your valuable suggestions. We have added some literatures and modified this description as “A majority of these viruses are bacteriophages, which exist widely in oceans and affect the life activities of microbes (Breitbart, 2012; Roux et al., 2016; Gregory et al., 2019; Dominguez-Huerta et al., 2022).” in the revised manuscript (Lines 94-97). 

References related to this response:

Roux, S., Brum, J.R., Dutilh, B.E., Sunagawa, S., Duhaime, M.B., Loy, A., Poulos, B.T., Solonenko, N., Lara, E., Poulain, J., et al. (2016) Ecogenomics and potential biogeochemical impacts of globally abundant ocean viruses. Nature 537:689-693. 

Gregory, A.C., Zayed, A.A., Conceição-Neto, N., Temperton, B., Bolduc, B., Alberti, A., Ardyna, M., Arkhipova, K., Carmichael, M., Cruaud, C., et al. (2019) Marine DNA Viral Macro- and Microdiversity from Pole to Pole. Cell 177:1109-1123.e1114. 

Dominguez-Huerta, G., Zayed, A.A., Wainaina, J.M., Guo, J., Tian, F., Pratama, A.A., Bolduc, B., Mohssen, M., Zablocki, O., Pelletier, E., et al. (2022) Diversity and ecological footprint of Global Ocean RNA viruses. Science 376:1202-1208.

Comment 5: Line 137. I see the phage upregulation in Figure 1, however in the text and figure it would be good to also elaborate on what the background expression generally looks like. Perhaps a transcriptomic read normalization and recruitment to the genome with a display of the coverage map, highlighting the prophage would be helpful. Are the polysacharides directly influencing phage induction or is there some potential for another cascading effect?  

Thanks for your comments. We have elaborated all expressions of phage-associated genes under different conditions in the Supplementary Table 1, which showed that the background expressions were very low. The numbers in Fig. 1C were the gene expressions (by taking log2 values) of strain WC36 cultured in rich medium supplemented with 10 g/L laminarin compared with the rich medium alone.

In addition, our RT-qPCR results (Fig. 1D) also confirmed that these genes encoding phage-associated proteins were significantly upregulated when 10 g/L laminarin was added in the rich medium. According to your suggestions, we have modified this description as “In addition to the up-regulation of genes related to glycan transport and degradation, when 10 g/L laminarin was added in the rich medium, the most upregulated genes were phage-associated (e. g. phage integrase, phage portal protein) (Fig. 1C and Supplementary Table 1), which were expressed at the background level in the rich medium alone.” in the revised manuscript (Lines 136-140). Based on the present results, we speculate that polysaccharides might directly induce phage production, which needs to be verified by a large number of experiments in the future.

Comment 6: Line 179. We need some assurance that phage was not introduced by your laminarin or starch supplement. Perhaps a check on the TEM/sequencing check of supplement itself would be helpful? This may be what is meant on Line 188 "without culturing bacterial cells" however this is not clearly worded if that is the case. Additional note, further reading reinforces this as a key concern. Many of the subsequent results are consistent with a contaminated starch stock. 

Thanks for your comments. We are very sure that the phages are specifically derived from the Lentisphaerae strain WC36 but not the polysaccharide stock solution. The reasons are as following: (1) we have performed multiple TEM checks of the rich medium supplemented with 10 g/L laminarin alone (Supplementary Fig. 1A) or in 10 g/L starch alone (Supplementary Fig. 1B), and there were not any phage-like structures, which confirmed that the polysaccharides (laminarin/starch) we used are not contaminated with any phage-like structures. In addition, we also observed the polysaccharides (laminarin/starch) directly by TEM and did not find any phage-like structures (Supplementary Fig. 2). According to your suggestions, we have modified this description as “We also tested and confirmed that there were not any phage-like structures in rich medium supplemented with 10 g/L laminarin alone (Supplementary Fig. 1A) or in 10 g/L starch alone (Supplementary Fig. 1B), ruling out the possibility of phage contamination from the polysaccharides (laminarin/ starch).” in the revised manuscript (Lines 158-162) and “Meanwhile, we also checked the polysaccharides (laminarin/ starch) in rich medium directly by TEM and did not find any phage-like structures (Supplementary Fig. 2).” in the revised manuscript (Lines 178-180). (2) the polysaccharide stock solution was strictly sterilized to remove any phage contamination. (3) the polysaccharide (starch) alone could not promote the growth of Pseudomonas stutzeri 273, however, the supplement of starch together with the extracted Phages-WC36 could effectively facilitate the growth of Pseudomonas stutzeri 273 (Response Figure 1). The above results clearly indicated the phage was derived from the Lentisphaerae strain WC36 but not the polysaccharide stock solution. 

In addition, given that polysaccharide was a kind of critical energy source for most microorganisms, we sought to ask whether polysaccharide also induces the production of bacteriophages in other deep-sea bacteria. To this end, we cultured deep-sea representatives from other four other phyla (including Chloroflexi, Tenericutes, Proteobacteria, and Actinobacteria) in the medium supplemented with laminarin/starch, and checked the supernatant of cells suspension through TEM as described above. We could not find any phage-like structures in these cells suspension (Response Figure 2), which also confirmed that there was no phage contamination in the polysaccharides.

Author response image 1.

Growth curve and status of Pseudomonas stutzeri 273 cultivated in basal medium, basal medium supplemented with 20 μl/mL Phages-WC36, basal medium supplemented with 5 g/L starch, basal medium supplemented with 5 g/L starch and 20 μl/mL Phages-WC36.   

Author response image 2.

TEM observation of the supernatant of cells suspension of a Chloroflexi strain, a Tenericutes strain, a Proteobacteria strain and an Actinobacteria strain that cultivated in the rich medium supplemented with 10 g/L laminarin and 10 g/L starch. No phage-like particles could be observed.  

Comment 7: Line 223. Correct generalized wording "long time". 

Thanks for your comments. We have changed “after for a long time” to “after 30 days” in the revised manuscript (Line 197).

Comment 8: Line 229. Please more explicitly describe what these numbers are (counts of virion like structures - filamentous and hexagonal respectively?), the units (per µL?), and how these were derived. The word "around" should be replaced with mean and standard deviation values for each count from replicates, without which these are not meaningful.

Thanks for your comments. The average numbers per microliter (µL) of filamentous and hexagonal phages in each condition were respectively calculated by randomly choosing ten TEM images. According to your suggestions, we have modified this description as “Specifically, the average number per microliter of filamentous phages (9.7, 29 or 65.3) extracted from the supernatant of strain WC36 cultured in rich medium supplemented with 10 g/L laminarin for 5, 10 or 30 days was higher than that cultured in rich medium supplemented with 5 g/L laminarin (4.3, 13.7 or 35.3) (Fig. 3B). The average number per microliter of hexagonal phages (9, 30, 46.7) extracted from the supernatant of strain WC36 cultured in rich medium supplemented with 10 g/L laminarin for 5, 10 or 30 days was higher than that cultured in rich medium supplemented with 5 g/L laminarin (4, 11.3 or 17.7) (Fig. 3C).” in the revised manuscript (Lines 203-210).

Comment 9: Line 242. This section should be included in the discussion of Figure 2 - around line 194.

Thanks. According to your suggestion, we have moved this section to the discussion corresponding to Figure 2 (Lines 183-191).

Comment 10: Figure 3. Stay consistent in the types of figures generated per strain. Figure 3A should be a growth curve.

Thanks for your comments. Actually, figure 3A was a growth curve, the corresponding description “(A) Growth curve of strain WC36 cultivated in either rich medium alone or rich medium supplemented with 5 g/L or 10 g/L laminarin for 30 days.” was shown in the Figure 3A legend in this manuscript.

Comment 11: Line 312. Move the discussion of AMGs to after the discussion of the phage genome identification.

Thanks for your valuable comments. According to your suggestions, we have moved the discussion of AMGs to after the discussion of the phage genome identification.

Comment 12: Line 312. It would be informative to sequence in-bulk each of your treatments as opposed to just sequencing the viral isolates (starch and no host included) to see what viruses can be identified in each. ABySS is also not a common assembler for viral analysis. Is there literature to support it as a sufficient tool in assembling viral genomes? What sequencing depths were obtained in your samples?

Thanks for your comments. In previous studies, we did sequence the starch or laminarin alone (no host included) and did not detect any phage-related sequences. The introduction of ABySS software was shown in these literatures (Jackman SD, Vandervalk BP, Mohamadi H, Chu J, Yeo S, Hammond SA, Jahesh G, Khan H, Coombe L, Warren RL, Birol I. ABySS 2.0: resource-efficient assembly of large genomes using a Bloom filter. Genome Res. 2017 May;27(5):768-777; Simpson JT, Wong K, Jackman SD, Schein JE, Jones SJ, Birol I. ABySS: a parallel assembler for short read sequence data. Genome Res. 2009 Jun;19(6):1117-23.), which were also used to assemble viral genomes in these literatures (Guo Y, Jiang T. First Report of Sugarcane Mosaic Virus Infecting Goose Grass in Shandong Province, China. Plant Dis. 2024 Mar 21. doi: 10.1094/PDIS-11-23-2514-PDN; Tang M, Chen Z, Grover CE, Wang Y, Li S, Liu G, Ma Z, Wendel JF, Hua J. Rapid evolutionary divergence of Gossypium barbadense and G. hirsutum mitochondrial genomes. BMC Genomics. 2015 Oct 12;16:770.). The sequencing depth of the phages of strain WC36 and zth2 were 350x and 365x, respectively.

Comment 13: Line 323. Replace "eventually" with more detail about what was done to derive the genomes. Were these the only four sequences identified as viral?

Thanks for your comments. We have used the ABySS software (http://www.bcgsc.ca/platform/bioinfo/software/abyss) to perform genome assembly with multiple-Kmer parameters. VIBRANT v1.2.1 (Kieft et al., 2020), DRAM-v (Shaffer et al., 2020), VirSorter v1.0.5 (with categories 1 (“pretty sure”) and 2 (“quite sure”)) (Roux et al., 2015) and VirFinder v1.1 (with statistically significant viral prediction: score > 0.9 and P-value < 0.05) (Ren et al., 2017) with default parameters were used to identify viral genomes from these assembly sequences by searching against the both cultured and non-cultured viral NCBI-RefSeq database (http://blast.ncbi.nlm.nih.gov/) and IMG/VR database (Camargo et al., 2023). The GapCloser software (https://sourceforge.net/projects/soapdenovo2/files/GapCloser/) was subsequently applied to fill up the remaining local inner gaps and correct the single base polymorphism for the final assembly results. All the detailed processes were described in the supplementary information. The virus sequences with higher scores are only these four, but they are not complete genomes. Some virus sequences with shorter sequences and lower scores were excluded.

Comment 14: Line 328. We need some details about the host genomes here. How were these derived? What is their completeness/contamination? What is their size? If the bins are poor, these would not serve as a reliable comparison to identify integrated phage.

Thanks for your comments. For genomic sequencing, strains WC36 and zth2 were grown in the liquid rich medium supplemented with 5 g/L laminarin and starch and harvested after one week of incubation at 28 °C. Genomic DNA was isolated by using the PowerSoil DNA isolation kit (Mo Bio Laboratories Inc., Carlsbad, CA). Thereafter, the genome sequencing was carried out with both the Illumina NovaSeq PE150 (San Diego, USA) and Nanopore PromethION platform (Oxford, UK) at the Beijing Novogene Bioinformatics Technology Co., Ltd. A complete description of the library construction, sequencing, and assembly was performed as previously described (Zheng et al., 2021). We used seven databases to predict gene functions, including Pfam (Protein Families Database, http://pfam.xfam.org/), GO (Gene Ontology, http://geneontology.org/) (Ashburner et al., 2000), KEGG (Kyoto Encyclopedia of Genes and Genomes, http://www.genome.jp/kegg/) (Kanehisa et al., 2004), COG (Clusters of Orthologous Groups, http://www.ncbi.nlm.nih.gov/COG/) (Galperin et al., 2015), NR (Non-Redundant Protein Database databases), TCDB (Transporter Classification Database), and Swiss-Prot (http://www.ebi.ac.uk/uniprot/) (Bairoch and Apweiler, 2000). A whole genome Blast search (E-value less than 1e-5, minimal alignment length percentage larger than 40%) was performed against above seven databases.

The completeness of the genomes of strains WC36 and zth2 were 100%, which were checked by the CheckM v1.2.2. The size of the genome of strains WC36 and zth2 were 3,660,783 bp and 3,198,720bp, respectively. The complete genome sequences of strains WC36 and zth2 presented in this study have been deposited in the GenBank database with accession numbers CP085689 and CP071032, respectively. 

Moreover, to verify whether the absence of microbial contamination in phage sequencing results, we used the new alignment algorithm BWA-MEM (version 0.7.15) to perform reads mapping of host WGS to these phages. We found that all the raw reads of host strains (WC36 and zth2) were not mapping to these phages sequences (Response Figure 3, shown as below). In addition, we also performed the evaluation of the assembly graph underlying the host consensus assemblies. Clean reads were mapped to the bacterial complete genome sequences by the Bowtie 2 (version 2.5.0), BWA (version 0.7.8) and SAMTOOLS (version 0.1.18). The results showed that the total mismatch rate of strains WC36 and zth2 were almost 0% and 0.03%, respectively (Response Table 1, shown as below). In addition, we also collected the cells of strains WC36 and zth2, and then sent them to another company for whole genome sequencing (named WC36G and ZTH, GenBank accession numbers CP151801 and CP119760, respectively). The completeness of the genomes of strains

WC36G and ZTH were also 100%. The size of the genome of strains WC36G and

ZTH were 3,660,783bp and 3,198,714bp, respectively. The raw reads of strains WC36G and zth2 were also not mapping to the phages sequences. Therefore, we can confirm that these bacteriophage genomes were completely outside of the host chromosomes. 

Author response image 3.

The read mapping from WGS to phage sequences.

Author response table 1.

Sequencing depth and coverage statistics.

References related to this response:

Zheng, R., Liu, R., Shan, Y., Cai, R., Liu, G., and Sun, C. (2021b) Characterization of the first cultured free-living representative of Candidatus Izemoplasma uncovers its unique biology ISME J 15:2676-2691. 

Ashburner, M., Ball, C.A., Blake, J.A., Botstein, D., Butler, H., Cherry, J.M., Davis, A.P., Dolinski, K., Dwight, S.S., Eppig, J.T., et al. (2000) Gene ontology: tool for the unification of biology. The Gene Ontology Consortium Nat Genet 25:25-29. 

Kanehisa, M., Goto, S., Kawashima, S., Okuno, Y., and Hattori, M. (2004) The KEGG resource for deciphering the genome Nucleic Acids Res 32:D277-280. 

Galperin, M.Y., Makarova, K.S., Wolf, Y.I., and Koonin, E.V. (2015) Expanded microbial genome coverage and improved protein family annotation in the COG database Nucleic Acids Res 43:D261-269. 

Bairoch, A., and Apweiler, R. (2000) The SWISS-PROT protein sequence database and its supplement TrEMBL in 2000 Nucleic Acids Res 28:45-48.

Comment 15: Line 333. This also needs some details. What evidence do you have that these are not chromosomal? If not chromosomal where can they be found? Sequencing efforts should also be able to yield extrachromosomal elements such as plasmids etc... If you were to sequence your purified isolate cultures from the rich media alone and include all assemblies (not just those binned for example) as a reference, would you be able to recruit viral reads? The way this reads suggests that Chevallereau et al., worked specifically with these phage, which is not the case - please rephrase.

Thanks for your comments. We carefully compared the bacteriophage genomes with those of the corresponding hosts (strains WC36 and zth2) using Galaxy Version 2.6.0 (https://galaxy.pasteur.fr/) (Afgan et al., 2018) with the NCBI BLASTN method and used BWA-mem software for read mapping from host whole genome sequencing (WGS) to these bacteriophages. These analyses both showed that the bacteriophage genomes are completely outside of the host chromosomes. Therefore, we hypothesized that the phage genomes might exist in the host in the form similar to that of plasmid.

Comment 16: Line 335. More to the point here that we need confirmation that these phages were not introduced in the polysaccharide treatment

Thanks for your comments. Please find our answers for this concern in the responses for comment 1 of “weakness” part and comment 6 of “Recommendations For The Authors” part.

Comment 17: Line 342. Lacking significant detail here. Phylogeny based on what gene(s), how were the alignments computed/refined, what model used etc..?

Thanks for your comments. According to your suggestions, all the related information was shown in this section “Materials and methods” of this manuscript. The maximum likelihood phylogenetic tree of Phage-WC36-2 and Phage-zth2-2 was constructed based on the terminase large subunit protein (terL). These proteins used to construct the phylogenetic trees were all obtained from the NCBI databases. All the sequences were aligned by MAFFT version 7 (Katoh et al., 2019) and manually corrected. The phylogenetic trees were constructed using the W-IQ-TREE web server (http://iqtree.cibiv.univie.ac.at) with the “GTR+F+I+G4” model (Trifinopoulos et al., 2016). Finally, we used the online tool Interactive Tree of Life (iTOL v5) (Letunic and Bork, 2021) to edit the tree. 

Comment 18: Line 346. How are you specifically defining AMGs in this study? Most of these are well-known and studied phage genes with specific life cycle functions and could not be considered as polysaccharide processing AMGs even though in host cells many do play a role in polysaccharide processing systems. A substantially deeper literature review is needed in this section, which would ultimately eliminate most of these from the potential AMG pools. Further, the simple HMM/BLASTp evalues are not sufficient to support the functional annotation of these genes. At a minimum, catalytic/conserved regions should be identified, secondary structures compared, and phylogenetic analysis (where possible) developed etc... My recommendation is to eliminate this section entirely from the manuscript. 

Categorically:

- Glycoside hydrolase (various families), glucosaminidases, and transglycosylase are all very common to phage and operate generally as a lysins, facilitating the release of virions from the host cell upon lysis, or injection of viral DNA upon infection https://doi.org/10.3389/fmicb.2016.00745 (and citations therein) https://doi.org/10.1016/j.cmi.2023.10.018 etc... In order to confirm these as distinct AMGs we would need a very detailed analysis indicating that these are not phage infection cycle/host recognition related, however I strongly suspect that under such interrogation, these would prove to be as such.

-TonB related systems including ExbB are well studied among phages as part of the trans-location step in infection. These could not be considered as AMGs. https://doi.org/10.1128/JB.00428-19. Other TonB dependent receptors play a role in host recognition.

-Several phage acetyltransferases play a role in suppressing host RNA polymerase in order to reserve host cell resources for virion production, including polysaccharide production. https://doi.org/10.3390/v12090976. Further it has been shown that the E. coli gene neuO (O-acetyltransferase) is a homologue of lambdoid phage tail fiber genes https://doi.org/10.1073/pnas.0407428102. I suspect the latter is also the case here and this is a tail fiber gene.

Thanks for your valuable comments. According to your suggestions, we have reanalyzed these AMGs and made some modifications (the new version Fig. 5A, shown as below). These genes encoding proteins associated with polysaccharide transport and degradation may be only common in virulent phages, and have never been reported in chronic phages. Unlike virulent phages, these genes typically act as lysozymes, facilitating the release of virions from the host cell upon lysis, or injection of viral DNA upon infection, chronic phages do not lyse the host. It is reported that, filamentous phages could recognize and bind to the host pili, which causes the pili to shrink and brings the filamentous phages closer to and possibly through the outer membrane of host cells (Riechmann et al., 1997; Sun et al., 1987). The possible mechanism of other chronic phage release without breaking the host might be that it was enclosed in lipid membrane and released from the host cells by a nonlytic manner. It has recently been reported that the tailless Caudoviricetes phage particles are enclosed in lipid membrane and are released from the host cells by a nonlytic manner (Liu et al., 2022), and the prophage induction contributes to the production of membrane vesicles by Lacticaseibacillus casei BL23 during cell growth (da Silva Barreira et al., 2022). Therefore, the persistence of these genes in chronic phages may be due to their ability to assist the host in metabolizing polysaccharides. 

Finally, according to your suggestions, we have weakened the role of AMGs and added “potential” in front of it.

References related to this response:

Riechmann L, Holliger P. (1997) The C-terminal domain of TolA is the coreceptor for filamentous phage infection of E. coli Cell 90:351-60.

Sun TP, Webster RE. (1987) Nucleotide sequence of a gene cluster involved in entry of E colicins and single-stranded DNA of infecting filamentous bacteriophages into Escherichia coli J Bacteriol 169:2667-74. 

Liu Y, Alexeeva S, Bachmann H, Guerra Martníez J.A, Yeremenko N, Abee T et al. (2022) Chronic release of tailless phage particles from Lactococcus lactis Appl Environ Microbiol 88: e0148321. da Silva Barreira, D., Lapaquette, P., Novion Ducassou, J., Couté, Y., Guzzo, J., and Rieu, A. Spontaneous prophage induction contributes to the production of membrane vesicles by the gram-positive bacterium Lacticaseibacillus casei BL23. mBio_._ 2022;13:e0237522.

Comment 19: Line 354. To make this statement that these genes are missing from the host, we would need to know that these genomes are complete.

Thanks for your comments. The completeness of the genomes of strains WC36 and zth2 were 100%, which were checked by the CheckM v1.2.2. The size of the genome of strains WC36 and zth2 were 3,660,783 bp and 3,198,720bp, respectively. The complete genome sequences of strains WC36 and zth2 presented in this study have been deposited in the GenBank database with accession numbers CP085689 and CP071032, respectively. In addition, we also collected the cells of strains WC36 and zth2, and then sent it to another company for whole genome sequencing (named WC36G and ZTH, GenBank accession numbers CP151801 and CP119760, respectively). The completeness of the genomes of strains WC36G and ZTH were also 100%. The size of the genome of strains WC36G and ZTH were 3,660,783bp and 3,198,714bp, respectively. Therefore, these genomes of strains WC36 and zth2 were complete and circular.    

Comment 20: Figure 5. Please see https://peerj.com/articles/11447/ and https://doi.org/10.1093/nar/gkaa621 for a detailed discussion on vetting AMGs. Several of these should be eliminated according to the standards set in the field. More specifically, and by anecdotal comparison with other inoviridae genomes, for Phage-WC36-1 and Phage-zth2-1, I am not convinced that the transactional regulator and glycoside hydrolase are a part of the phage genome. The phage genome probably ends at the strand switch.

Thanks for your comments. According to your suggestions, we have analyzed these two articles carefully and modified the genome of Phage-WC36-1 and Phage-zth2-1 by anecdotal comparison with other inoviridae genomes. As you said, the transactional regulator and glycoside hydrolase are not a part of the phage genome.

The new version Fig. 5A was shown.

. 

References related to this response:

Shaffer, M., Borton, M.A., McGivern, B.B., Zayed, A.A., La Rosa, S.L., Solden, L.M., Liu, P., Narrowe, A.B., Rodrgíuez-Ramos, J., Bolduc, B., et al. (2020) DRAM for distilling microbial metabolism to automate the curation of microbiome function Nucleic Acids Res 48:8883-8900 

Pratama, A.A., Bolduc, B., Zayed, A.A., Zhong, Z.P., Guo, J., Vik, D.R., Gazitúa, M.C., Wainaina, J.M., Roux, S., and Sullivan, M.B. (2021) Expanding standards in viromics: in silico evaluation of dsDNA viral genome identification, classification, and auxiliary metabolic gene curation PeerJ 9:e11447

Comment 21: Line 380. This section needs to start with detailed evidence that this phage can even infect this particular strain. Added note, upon further reading the serial dilution cultures are not sufficient to prove these phage infect this Pseudomonas. We need at a minimum a one-step growth curve and wet mount microscopy. It is much more likely that some carry over contaminant is invading the culture and influencing OD600. With the given evidence, I am not at all convinced that these phages have anything to do with Pseudomonas polysaccharide use and I recommend either drastically revising this section or eliminating it entirely.

Line 386-389. Could this be because you are observing your added phage in the starch enriched media while no phage were introduced with the "other types of media" so none would be observed? This could have nothing to do with infection dynamics. Further, this would also be consistent with your starch solution being contaminated by phage.

Line 399. Again consistent with the starch media being contaminated.

Line 401-408. This is more likely to do with the augmentation of the media with an additional carbon source and not involving the phage. 

Line 410. I am not convinced that these viruses infect the Pseudomonas strain. Extensive further evidence of infection is needed to make these assertions.  Figure 6A. We need confirmation that the isolate culture remains pure and there are no other contaminants introduced with the phage.

Thanks for your comments. We have proved that the polysaccharides (laminarin/ starch) didn't contaminate any phages above. Actually, we selected many marine strains (Pseudomonadota, Planctomycetes, Verrucomicrobia, Fusobacteria, and Tenericutes isolates) to investigate whether Phages-WC36 could assist them in degradation and utilization of polysaccharides, and found that Phages-WC36 could only promote the growth of strain 273. The presence of filamentous phages and hexagonal phages was detected in the supernatant of strain 273 cultured in basal medium supplemented with 5 g/L starch and 20 μl/mL Phages-WC36. After 3 passages of serial cultivation in basal medium supplemented with 5 g/L starch, we found that filamentous phages and hexagonal phages were also present in basal medium supplemented with starch, but not in the basal medium, which may mean that Phages-WC36 could infect strain 273 and starch is an important inducer. In addition, the Phages-WC36 used in the growth assay of strain 273 were multiple purified and eventually suspended in SM buffer (0.01% gelatin, 50 mM Tris-HCl, 100 mM NaCl and 10 mM MgSO4). Thus, these phages are provided do not contain some extracellular enzymes and/or nutrients. In addition, we set up three control groups in the growth assay of strain 273: basal medium, basal medium supplemented with Phages-WC36 and basal medium supplemented with starch. If the Phages-WC36 contains some extracellular enzymes and/or nutrients, strain 273 could also grow well in the basal medium supplemented only with Phages-WC36. However, the poor growth results of strain 273 cultivated in the basal medium supplemented with Phages-WC36 further confirmed that there were not some extracellular enzymes and/or nutrients in these phages.

Finally, the possible mechanism of the chronic phage release without breaking the host might be that it was enclosed in lipid membrane and released from the host cells by a nonlytic manner. Thus, these chronic phages may have a wider host range. However, we were unable to further disclose the infection mechanism in this paper. Therefore, according to your suggestions, we have deleted this section entirely.

Comment 27: Line 460. Details about how these genomes were reconstructed is needed here.  

Thanks for your comments. According to your suggestions, we have added the detailed information about the genome sequencing, annotation, and analysis as “Genome sequencing, annotation, and analysis of strains WC36 and zth2 For genomic sequencing, strains WC36 and zth2 were grown in the liquid rich medium supplemented with 5 g/L laminarin and starch and harvested after one week of incubation at 28 °C. Genomic DNA was isolated by using the PowerSoil DNA isolation kit (Mo Bio Laboratories Inc., Carlsbad, CA). Thereafter, the genome sequencing was carried out with both the Illumina NovaSeq PE150 (San Diego, USA) and Nanopore PromethION platform (Oxford, UK) at the Beijing Novogene Bioinformatics Technology Co., Ltd. A complete description of the library construction, sequencing, and assembly was performed as previously described (Zheng et al., 2021b). We used seven databases to predict gene functions, including Pfam (Protein Families Database, http://pfam.xfam.org/), GO (Gene Ontology, http://geneontology.org/) (Ashburner et al., 2000), KEGG (Kyoto Encyclopedia of Genes and Genomes, http://www.genome.jp/kegg/) (Kanehisa et al., 2004), COG

(Clusters of Orthologous Groups, http://www.ncbi.nlm.nih.gov/COG/) (Galperin et al.,

2015), NR (Non-Redundant Protein Database databases), TCDB (Transporter Classification Database), and Swiss-Prot (http://www.ebi.ac.uk/uniprot/) (Bairoch and Apweiler, 2000). A whole genome Blast search (E-value less than 1e-5, minimal alignment length percentage larger than 40%) was performed against above seven databases.” in the revised manuscript (Lines 333-351).

Comment 28: Line 462. Accession list of other taxa in the supplement would help here.  

Thanks for your comments. The accession numbers of these strains were displayed behind these strains in Figure 1A. According to your suggestions, we have added an accession list of these taxa (Supplementary Table 6) in the revised manuscript.

Comment 29: Line 463. Is there any literature to support that these are phylogenetically informative genes for Inoviridae?  

Thanks for your comments. There are some literatures (Zeng et al, 2021; Evseev et al, 2023) to support that these are phylogenetically informative genes for Inoviridae. We have added these literatures in the revised manuscript. 

References related to this response:

Zeng, J., Wang, Y., Zhang, J., Yang, S., and Zhang, W. (2021) Multiple novel filamentous phages detected in the cloacal swab samples of birds using viral metagenomics approach Virol J 18:240

Evseev, P., Bocharova, J., Shagin, D., and Chebotar, I. (2023) Analysis of Pseudomonas aeruginosa isolates from patients with cystic fibrosis revealed novel groups of filamentous bacteriophages. Viruses 15: 2215

Reviewer #2 (Public Review):

Summary: This paper investigates virus-host interactions in deep-sea bacteriophage systems which employ a seemingly mutualistic approach to viral replication in which the virus aids host cell polysaccharide import and utilization via metabolic reprogramming. The hypothesis being tested is supported with solid and convincing evidence and the findings are potentially generalizable with implications for our understanding of polysaccharide-mediated virus-host interactions and carbon cycles in marine ecosystems more broadly.

Thanks for your positive comments.

Strengths: This paper synthesizes sequencing and phylogenic analyses of two Lentisphaerae bacteria and three phage genomes; electron microscopy imaging of bacterial/phage particles; differential gene expression analyses; differential growth curve analyses, and differential phage proliferation assays to extract insights into whether laminarin and starch can induce both host growth and phage proliferation. The data presented convincingly demonstrate that both host culture density and phage proliferation increase as a result having host, phage, and polysaccharide carbon source together in culture.

Thanks for your positive comments.  

Weaknesses (suggestions for improvement): 

(1) The article would be strengthened by the following additional experiment: providing the phage proteins hypothesized to be aiding host cell growth (red genes from Figure 5...TonB system energizer ExbB, glycosidases, etc) individually or in combination on plasmids rather than within the context of the actual phage itself to see if such additional genes are necessary and sufficient to realize the boosts in host cell growth/saturation levels observed in the presence of the phages tested.

Thanks for your valuable comments. It is a really good idea to express individually or in combination on plasmids to see the effects of those polysaccharide-degradation proteins in the host cell. However, at present, we failed to construct the genetic and expression system for the strictly anaerobic strain WC36, which hindering our further detailed investigation of the functions of those polysaccharide-degradation proteins. In our lab, we are trying our best to build the genetic and expression system for strain WC36. We will definitely test your idea in the future. 

(2) The paper would also benefit from additional experiments focused on determining how the polysaccharide processing, transport, and metabolism genes are being used by the phages to either directly increase viral infection/replication or else to indirectly do so by supporting the growth of the host in a more mutualistic manner (i.e. by improving their ability to import, degrade, and metabolize polysaccharides).  

Thanks for your valuable comments. Indeed, due to the chronic phage genome is not within the chromosome of the host, it is very hard to disclose the exact auxiliary process and mechanism of chronic phages. At present, we are trying to construct a genetic manipulation system for the strictly anaerobic host WC36, and we will gradually reveal this auxiliary mechanism in the future. In addition, combined with the reviewer 1’s suggestions, the focus of revised manuscript is to emphasize that polysaccharides induce deep-sea bacteria to release chronic phages, and most of the content of phage assisting host metabolism of polysaccharides has been deleted.

(3) The introduction would benefit from a discussion of what is known regarding phage and/or viral entry pathways that utilize carbohydrate anchors during host entry. The discussion could also be improved by linking the work presented to the concept of "selfishness" in bacterial systems (see for instance Giljan, G., Brown, S., Lloyd, C.C. et al. Selfish bacteria are active throughout the water column of the ocean. ISME COMMUN. 3, 11 (2023) https://doi.org/10.1038/s43705-023-00219-7). The bacteria under study are gram negative and it was recently demonstrated (https://www.nature.com/articles/ismej201726) that "selfish" bacteria sequester metabolizable polysaccharides in their periplasm to advantage. It is plausible that the phages may be hijacking this "selfishness" mechanism to improve infectivity and ENTRY rather than helping their hosts to grow and profilerate so they can reap the benefits of simply having more hosts to infect. The current work does not clearly distinguish between these two distinct mechanistic possibilities. The paper would be strengthened by at least a more detailed discussion of this possibility as well as the author's rationale for interpreting their data as they do to favor the "mutualistic" interpretation. In the same light, the paper would benefit from a more careful choice of words which can also help to make such a distinction more clear/evident/intentional. As currently written the authors seem to be actively avoiding giving insights wrt this question.  

Thanks for your valuable comments. According to your suggestions, we have added the related discussion as “Moreover, it was recently demonstrated that selfish bacteria, which were common throughout the water column of the ocean, could bind, partially hydrolyze, and transport polysaccharides into the periplasmic space without loss of hydrolysis products (Reintjes et al., 2017; Giljan et al., 2023). Based on our results, we hypothesized that these chronic phages might also enter the host through this “selfishness” mechanism while assisting the host in metabolizing polysaccharides, thus not lysing the host. On the other hand, these chronic phages might hijack this “selfishness” mechanism to improve their infectivity and entry, rather than helping their hosts to grow and proliferate, so they could reap the benefits of simply having more hosts to infect. In the future, we need to construct a genetic operating system of the strictly anaerobic host strain WC36 to detailedly reveal the relationship between chronic phage and host.” in the revised manuscript (Lines 305-316). 

References related to this response:

Reintjes, G., Arnosti, C., Fuchs, B.M., and Amann, R. (2017) An alternative polysaccharide uptake mechanism of marine bacteria ISME J 11:1640-1650

Giljan, G., Brown, S., Lloyd, C.C., Ghobrial, S., Amann, R., and Arnosti, C. (2023) Selfish bacteria are active throughout the water column of the ocean ISME Commun 3:11

(4) Finally, I would be interested to know if the author’s sequencing datasets might be used to inform the question raised above by using bacterial immunity systems such as CRISPR/Cas9. For example, if the phage systems studied are truly beneficial/mutualistic for the bacteria then it’s less likely that there would be evidence of targeted immunity against that particular phage that has the beneficial genes that support polysaccharide metabolism.

Thanks for your comments. According to your suggestions, we have carefully analyzed the genome of strain WC36, and found that there were no CRISPR/Cas9-related genes. Considering our results that the number of chronic phages was increased with the prolongation of culture time, we speculated that host might have no targeted immunity against these chronic phages.

Reviewer #2 (Recommendations For The Authors):

There are some minor grammatical errors and unclear statements (lines 99-100, 107-109, 163, 222, 223, 249-250, 254) which should also be fixed before final publication. 

Thanks for your valuable comments. We have fixed these minor grammatical errors and unclear statements in the revised manuscript.

Lines 99-100: we have modified this description as “For instance, AMGs of marine bacteriophages have been predicted to be involved in photosynthesis (Mann et al., 2003), nitrogen cycling (Ahlgren et al., 2019; Gazitúa et al., 2021), sulfur cycling (Anantharaman et al., 2014; Roux et al., 2016), phosphorus cycling (Zeng and Chisholm, 2012), nucleotide metabolism (Sullivan et al., 2005; Dwivedi et al., 2013; Enav et al., 2014), and almost all central carbon metabolisms in host cells (Hurwitz et al., 2013).” in the revised manuscript (Lines 100-105).

Lines 107-109: we have modified this description as “However, due to the vast majority of deep-sea microbes cannot be cultivated in the laboratory, most bacteriophages could not be isolated.” in the revised manuscript (Lines 110-111).

Line 163: we have modified this description as “Based on the growth curve of strain WC36, we found that the growth rate of strictly anaerobic strain WC36 was relatively slow.” in the revised manuscript (Lines 149-151).

Lines 222-223: we have modified this description as “Regardless of whether the laminarin was present, the bacterial cells kept their cell shape intact, indicating they were still healthy after 30 days” in the revised manuscript (Lines 195-197).

Lines 249-250: we have modified this description as “However, the entry and exit of the hexagonal phages into the WC36 cells were not observed.” in the revised manuscript (Lines 190-191).

Line 254: we have modified this description as “To explore whether the production of bacteriophages induced by polysaccharide is an individual case, we further checked the effect of polysaccharides on another cultured deep-sea Lentisphaerae strain zth2.” in the revised manuscript (Lines 213-215).

Dette er trolig sant for en stor del av befolkningen, men selvsagt ikke for alle.

looking good!

https://youtu.be/7iJ0QQvf8LA

https://youtu.be/8FZ1TUYn_50

https://youtu.be/H36mgUH4j80

https://youtu.be/qA8kDZWZIkM

eLife assessment

This study provides an important, original framework to study locomotion on the ground with physics-based simulations. Through numerical simulations, the authors propose that intermediate numbers of body modules and high body symmetry enhance speed. The current way discussions and conclusions are written is overly broad: evidence that evolution may favour bilateral symmetry and modularity for efficient directed locomotion is still incomplete as further performance metrics and a more accurate description of the dynamics in water are needed.

Reviewer #1 (Public Review):

The manuscript presents a framework for studying biomechanical principles and their links to morphology and provides interesting insights into a particular question regarding terrestrial locomotion and speed. The goal of the paper is to derive general principals of directed terrestrial locomotion, speed, and symmetry.

Major strengths:

The manuscript is a unique and creative work that explores performance spaces of a complicated question through computational modeling. Overall, the paper is well written and well crafted and was a pleasure to read.

The methods presented here (variable agents used to represent ultra-simplified body configurations that are not inherently constrained) are interesting and there's significant potential in them for a properly constrained question. For the data that is present here their hypotheses (while they can be anticipated from first principles) are very well validated and serve as a robust validation of these expectations and can help.

Of particular interest was the discussion of the transferability of morphologies designed under one system and moving to another. From a deep-time perspective, of particular interest is the transition from subaqueous to terrestrial locomotion which we know was a major earth life transition. The results of this study show that the best suited morphologies for subaqueous movement are ill-suited (from a locomotor speed standpoint at least) to fully terrestrial locomotion which begs the questions on if there are a suite of forms that have balanced performance in both and how that would differ from aquatic morphologies.

Major weaknesses:

(1) There is a major disagreement between target and parameters.

From a biomechanics perspective the target of this study, Directed Locomotion, is a fairly broad behavioral mode. However, what the authors are ultimately evaluating their model organisms on is a single performance parameter (speed, or distance traveled after 30s). Statements such as "bilateral symmetry showed to be a law-like pattern in animal evolution for efficient directed locomotion purposes" (p 12 line 365-366) are problematic for this reason.

Attaining the highest possible speed is a relevant but limited subset of ways one might interpret performance for directed locomotion. Efficiency, power generation, and limb loading/strain are equally relevant components.

The focus on speed coupled with selection for only the highest performing morphologies, rather than setting a minimum performance threshold fundamentally restricts the dynamics of the system in a way that is not representative of their specified target and pulls the simulations toward a specific, anticipatable, result.

Locomotor efficiency is alluded to later in the manuscript as one of the observed outcomes, but speed is not equivalent to locomotor efficiency (in much the same way that it is not the sole metric for describing performance with respect to directed locomotion). Energy/work/power have not been accounted for in the manuscript so this is not a parameter this study weighs in on.

The data and analyses the others present do show an interesting validation of these methods in assessing first order questions relating the shape of a single performance surface to a theoretical morphology, which has significant potential value.

(2) There is significant population and/or sample size and biasing.

Thirty simulations of a population of 101 morphologies seems small for a study of this kind, particularly looking to investigate such a broad question at an abstract level. Particularly when the top 50% of morphologies are chosen to mutate. It would be very easy for artificial biases to rapidly propagate through this system depending on the parameters bounding the formation of the initial generation.

This strong selection choosing the best 50 morphologies and mutating them enforces an aggressive effect that simulates and even more potent phylogenetic inertia than one might anticipate for an actual evolutionary history (it's no surprise then that all of the simulations were able to successfully retrieve a suite of morphotypes that recovered the performance peak for this system within 1500 generations)

Similarly, why is it that a 4^3 voxel limit was chosen? One can imagine that an increase in this voxel limit would allow for the development of more extreme geometries, which might be successful. It is likely that there might be computational resource constraints involved in this, it would be useful for the authors to add additional context here.

Review of resubmission:

I appreciate the clarification of points dealing with the details of computational modeling and methods and clarifications throughout the text.

However, the authors have failed to address the major weaknesses that were previously identified, specifically regarding the broader conclusions of the work, that either 1) the authors need to use an additional metric besides average speed, or 2) the conclusions need to be significantly reigned in to reflect the very narrow nature of the work.

Reviewer #2 (Public Review):

Summary:<br /> I believe the authors have done a wonderful job at dissecting a very complex topic, starting with basic building blocks of locomotion and introducing a powerful simulation approach to the exploring the landscape of growth and form in intelligent behavior.

Strengths:<br /> This is a very original, timely, and robust piece of work that I believe can inspire further computational studies in evo-devo-etho.

Weaknesses:<br /> More detail on the simulations and also greater clarity regarding the generalizability of their claims would improve the message and further studies.

Dit mag nog concreter. Wie of waar moet ik zijn om mee te doen? @Henk

eLife assessment

This valuable study presents a new framework (ASBAR) that combines open-source toolboxes for pose estimation and behavior recognition to automate the process of categorizing behaviors in wild apes from video data. The authors present compelling evidence that this pipeline can categorize simple wild ape behaviors from out-of-context video at a similar level of accuracy as previous models, while simultaneously vastly reducing the size of the model. The study's results should be of particular interest to primatologists and other behavioral biologists working with natural populations.

Reviewer #1 (Public Review):

Summary:

Advances in machine vision and computer learning have meant that there are now state-of-the-art and open-source toolboxes that allow for animal pose estimation and action recognition. These technologies have the potential to revolutionize behavioral observations of wild primates but are often held back by labor-intensive model training and the need for some programming knowledge to effectively leverage such tools. The study presented here by Fuchs et al unveils a new framework (ASBAR) that aims to automate behavioral recognition in wild apes from video data. This framework combines robustly trained and well-tested pose estimate and behavioral action recognition models. The framework performs admirably at the task of automatically identifying simple behaviors of wild apes from camera trap videos of variable quality and contexts. These results indicate that skeletal-based action recognition offers a reliable and lightweight methodology for studying ape behavior in the wild and the presented framework and GUI offer an accessible route for other researchers to utilize such tools.

Given that automated behavior recognition in wild primates will likely be a major future direction within many subfields of primatology, open-source frameworks, like the one presented here, will present a significant impact on the field and will provide a strong foundation for others to build future research upon.

Strengths:

- Clearly articulated the argument as to why the framework was needed and what advantages it could convey to the wider field.

- For a very technical paper it was very well written. Every aspect of the framework the authors clearly explained why it was chosen and how it was trained and tested. This information was broken down in a clear and easily digestible way that will be appreciated by technical and non-technical audiences alike.

- The study demonstrates which pose estimation architectures produce the most robust models for both within-context and out-of-context pose estimates. This is invaluable knowledge for those wanting to produce their own robust models.

- The comparison of skeletal-based action recognition with other methodologies for action recognition helps contextualize the results.

Weaknesses

While I note that this is a paper most likely aimed at the more technical reader, it will also be of interest to a wider primatological readership, including those who work extensively in the field. When outlining the need for future work I felt the paper offered almost exclusively very technical directions. This may have been a missed opportunity to engage the wider readership and suggest some practical ways those in the field could collect more ASBAR-friendly video data to further improve accuracy.

Reviewer #2 (Public Review):

Fuchs et al. propose a framework for action recognition based on pose estimation. They integrate functions from DeepLabCut and MMAction2, two popular machine-learning frameworks for behavioral analysis, in a new package called ASBAR.

They test their framework by

- Running pose estimation experiments on the OpenMonkeyChallenge (OMC) dataset (the public train + val parts) with DeepLabCut.

- Annotating around 320 image pose data in the PanAf dataset (which contains behavioral annotations). They show that the ResNet-152 model generalizes best from the OMC data to this out-of-domain dataset.

- They then train a skeleton-based action recognition model on PanAf and show that the top-1/3 accuracy is slightly higher than video-based methods (and strong), but that the mean class accuracy is lower - 33% vs 42%. Likely due to the imbalanced class frequencies. This should be clarified. For Table 1, confidence intervals would also be good (just like for the pose estimation results, where this is done very well).

faq

... "for" ...

... "of" ...

Person's (thing)

... "in" ...

Lumping two words together

... "and" ...

... "as" ... And, ... "examples"

Welcome back. In this lesson, I want to introduce Virtual Private Clouds known as VPCs. A VPC is the service you will use to create private networks inside AWS that other private services will run from. VPCs are also the service which is used to connect your AWS private networks to your on-premises networks when creating a hybrid environment, or it's the service which lets you connect to other cloud platforms when you're creating a multi-cloud deployment. You will encounter numerous networking and VPC-related questions in the exam, so understanding this service is crucial. This lesson provides a basic introduction, and detailed information will follow later in the course.

Let's begin with key points. A VPC is a virtual network within AWS, created within an AWS account and a specific region. VPCs are regional services, meaning they are resilient within a region and operate across multiple availability zones. By default, a VPC is private and isolated, allowing communication only within the same VPC unless configured otherwise. There are two types of VPCs: default VPCs and custom VPCs. Each region can have one default VPC, which AWS creates automatically, and you can have multiple custom VPCs. Custom VPCs require detailed configuration and are entirely private by default.

Custom VPCs are used in most serious AWS deployments because of their flexibility in configuration, size, structure, and connectivity with other VPCs, cloud platforms, and on-premises networks. In contrast, default VPCs are pre-configured by AWS with limited flexibility. This lesson focuses on VPC architecture and the default VPC, while custom VPCs will be covered in depth later.

VPCs are created within an AWS account and region, such as us-east-1. A region can have multiple custom VPCs, isolated by default unless configured to allow communication. VPCs are resilient, subdivided into subnets located in different availability zones. A default VPC has a set CIDR range of 172.31.0.0/16, which is the same across all regions. Custom VPCs can have multiple CIDR ranges. A VPC's resilience comes from its subnets, each in one availability zone. If an availability zone fails, subnets in other zones continue to operate.

Key facts about the default VPC include: only one per region, can be deleted and recreated, and has a fixed CIDR range of 172.31.0.0/16. Default VPCs have a predictable structure and come with pre-configured resources like an internet gateway, a default security group, and a network ACL. By default, anything placed in the default VPC subnets is assigned a public IP version four address, making them accessible from the public zone.

To see the default VPC in action, you can use the AWS Console. In a freshly created AWS account, the default VPC will be present with its CIDR range of 172.31.0.0/16 and subnets in each availability zone of the region. The number of subnets varies by region. You can delete the default VPC and recreate it from the console without support tickets, maintaining the same static structure.

This lesson provided an overview of the default VPC and its consistent structure. Soon, the course will shift focus to custom VPCs, which are best practice for production environments. The course scenario involving the Animals for Life Organization will illustrate why the default VPC is unsuitable. For now, mark this video as complete, and join me in the next lesson when you're ready.

eLife assessment

In this valuable study, the authors analyze droplet size distributions of multiple protein condensates and their fit to a scaling ansatz, highlighting that they exhibit features of first- and second-order phase transitions. The experimental evidence is solid, but the interpretation can be improved. The text would benefit from further clarification and connection to the coupled percolation-and-phase-separation model.

Reviewer #1 (Public Review):

The authors analyse droplet size distributions of multiple protein condensates and fit to a scaling ansatz to highlight that they exhibit features of first-order and second-order phase transitions. While the experimental evidence is solid, the text lacks connection and contextualization to the well-understood expectations from the coupling of percolation and phase separation in protein condensates - a phenomenon that is increasingly gaining consensus amongst the community. The evidence supports the percolatoin+phase separation model rather than being close to a true critical point in the liquid-gas phase space. Overall, the work is useful to the community.

Strengths:<br /> The experimental analysis of distinct protein condensates is very well done and the reported exponents/scaling framework provides a clear framework to help the community help deconvolve signatures of percolation in condensates.

Weaknesses:

The principal concern this reviewer has is that the reviewers adopt a framing in this paper to present a discovery of second-order features and connections to criticality - however they ignore/miss the connections to percolation (a well-understood second-order transition that is expected to play a major role in protein condensates). I believe this needs to be addressed and the paper suitably revised to help connect with these expectations.

- Protein condensates have been increasingly understood to be described as fluids whose assembly is driven by a connection of density (phase separation, first-order) and connectivity (percolation, second-order) transitions. This has been long known in the polymer community (Flory, Stockmayer, Tanaka, Rubinstein, Semenov and others) and recently repopularized in the condensate community (by Pappu and Mittag, in particular, amongst others). The authors make no connections to any of this frameworks - which actually seem to be the essence of what they are describing.

- Percolation theory, which has been around for more than half-a-century, has clear-cut scaling laws that have essentially similar forms to the ansatz adopted by the authors and the commonalities/differences are not discussed by the authors - this is essential since this provides a physical basis for their ansatz rather than an arbitrary mathematical formulation. In particular, percolation models connect size distribution exponents to factors like dimensionality, valence, etc. and if these connections can be made with this data, that would be very powerful.

- The connections between spinodal decomposition and second-order phase transitions are very confusing. Spindal decomposition happens when the barriers for first-order phase transitions are zero and systems can phase separate without crossing nucleation barriers. Further, the "criticality" discussed in the paper is confusing since it more likely refers to a percolation threshold and much less likely to a "critical temperature" (Tc -where spinodal and binodals become identical). I would recommend reframing this argument.

It's unlikely, in this reviewer's opinion, that the authors are actually discussing a "first-order" liquid-gas critical point - because saturation concentrations of these proteins can be much higher with temperature and the critical point would thus likely be at much higher concentrations (and ofc temperature). Further the scaling exponents don't fall in that class naturally. However, if the authors disagree, I would appreciate clear quantitative reasons (including through the scaling exponents in that universality class) and be happy to be convinced to change my mind. As provided, the data does not support this model.

Reviewer #2 (Public Review):

In response to the two referee reports, the authors have made substantial improvements. Regarding my previous concerns, the new data provided in Fig.6 for demonstrating that the droplet size distribution is stable over time is particularly valuable.

As to several of my other previous concerns regarding possible change in droplet size distribution over time, etc., the authors responded by stating that their system was below the critical concentration and therefore the possible scenarios pointed out in my previous report were not expected. While there may be a certain degree of validity to their argument, it would be much more helpful to the readers if the authors would bring up my previous concerns briefly (as readers of the journal will likely have similar concerns) and then address them succinctly within the manuscript.

Apparently, as a key element in the authors' response to the referees, the term "transition concentration" in the originally submitted manuscript is now changed to "critical concentration" (including in the title and abstract). But the two terms do not have identical meaning. A transition concentration is usually recognized as the saturation concentration at which phase separation or some other transition process commences at a given temperature. The transition concentration can be lower than the critical concentration, whereas the critical concentration is associated with the critical temperature, above (or below, depending on the temperature dependence of phase separation) which phase separation is not possible. It will be best if the authors can clarify their usage of transition concentration vs. critical concentration in the version of record of their manuscript.

Author response:

The following is the authors’ response to the original reviews

eLife assessment:

In this useful study, the authors analyze droplet size distributions of multiple protein condensates and their fit to a scaling ansatz, highlighting that they exhibit features of first- and second-order phase transitions. The experimental evidence is still incomplete as the measurements were apparently done only at one time point, neglecting the possibility that droplet size distribution can evolve with time. The text would benefit from a connection to and contextualization with the well-understood expectations from the coupling of percolation and phase separation in protein condensates - a phenomenon that is increasingly gaining consensus amongst the community and that emphasizes "liquid-gas" criticality. 

We have now carried out new experiments at multiple time points to establish that the droplet size distributions are stationary below the critical concentration. We have also addressed the comments made by the reviewers about the nature of the phase transition.

Our analysis does not depend on a specific hypothesis on the nature of the phase transition, whether it be percolation or a gas-liquid critical transition. The scaling that we observed is an emergent property that is independent from the possible theoretical models used to describe the phase transition. In fact, our scaling analysis indicates that any theoretical model proposed for protein phase separation should predict the critical exponents that we reported. 

Reviewer #1: 

The authors analyse droplet size distributions of multiple protein condensates and fit to a scaling ansatz to highlight that they exhibit features of first-order and second-order phase transitions. While the experimental evidence is solid, the text lacks connection and contextualization to the well-understood expectations from the coupling of percolation and phase separation in protein condensates - a phenomenon that is increasingly gaining consensus amongst the community. The evidence supports the percolation and phase separation model rather than being close to a true critical point in the liquid-gas phase space. Overall, the work is useful to the community.

We are grateful to the reviewer for these positive comments. We would like to emphasises that our contribution is not to propose a theoretical model, but rather to report a scaling behaviour in the experimentally measured droplet size distributions. The main implication of our work is that any theoretical model should predict the scaling exponents that we derived from the experimental measurements.

Strengths: 

The experimental analysis of distinct protein condensates is very well done and the reported exponents/scaling framework provides a clear framework to help the community deconvolve signatures of percolation in condensates. 

Weaknesses: 

The principal concern this reviewer has is that the reviewers adopt a framing in this paper to present a discovery of second-order features and connections to criticality - however, they ignore/miss the connections to percolation (a well-understood second-order transition that is expected to play a major role in protein condensates). I believe this needs to be addressed and the paper suitably revised to help connect with these expectations. 

The scaling that we found is not characteristic standard percolation, since the exponents that we obtained (a=0 and f=1) are different from those of percolation (a=1.19 and f=2.21). This difference indicates that protein phase separation is not in the same universality class of standard percolation. Further studies will be required to understand whether theoretical models based on percolation could predict the observed critical exponents.

- Protein condensates have been increasingly understood to be described as fluids whose assembly is driven by a connection of density (phase separation, first-order) and connectivity (percolation, second-order) transitions. This has been long known in the polymer community (Flory, Stockmayer, Tanaka, Rubinstein, Semenov, and others) and recently repopularized in the condensate community (by Pappu and Mittag, in particular, amongst others). The authors make no connections to any of these frameworks - which actually seem to be the essence of what they are describing. 

As mentioned above, our purpose was neither to support an existing theoretical model, nor to propose a new one. Rather, we have reported a scaling behaviour and scaling exponents not noted before. Further studies will be required to establish whether existing theoretical models could account for this scaling behaviour.

- Percolation theory, which has been around for more than half a century, has clear-cut scaling laws that have essentially similar forms to the ansatz adopted by the authors, and the commonalities/differences are not discussed by the authors - this is essential since this provides a physical basis for their ansatz rather than an arbitrary mathematical formulation. In particular, percolation models connect size distribution exponents to factors like dimensionality, valence, etc. and if these connections can be made with this data, that would be very powerful. 

The scaling ansatz that we are using is commonly adopted in studies of critical phenomena, and it is not specific to percolation. The scaling exponents depends only on very few attributes like dimensionality, symmetries and if interactions are short or long range. These attributes determine the universality class. As such, scaling does not link with molecular determinants, but can distinguish different classes.

- The connections between spinodal decomposition and second-order phase transitions are very confusing. Spindal decomposition happens when the barriers for first-order phase transitions are zero and systems can phase separate without crossing nucleation barriers. Further, the "criticality" discussed in the paper is confusing since it more likely refers to a percolation threshold and much less likely to a "critical temperature" (Tc -where spinodal and binodals become identical). I would recommend reframing this argument. 

We cannot refer to percolation threshold as our model is not readily compatible with it. We elaborated and better explained the differences between these models.

It's unlikely, in this reviewer's opinion, that the authors are actually discussing a "first-order" liquid-gas critical point - because saturation concentrations of these proteins can be much higher with temperature and the critical point would thus likely be at much higher concentrations (and ofc temperature). Further, the scaling exponents don't fall into that class naturally. However, if the authors disagree, I would appreciate clear quantitative reasons (including through the scaling exponents in that universality class) and be happy to be convinced to change my mind. As provided, the data does not support this model. 

We have now clarified in the manuscript that we do not discuss the liquid-gas critical point.

Reviewer #2: 

This is a potentially interesting study addressing a possible scale-invariant log-normal characteristic of droplet size distribution in the phase separation behavior of biomolecular condensates. Some of the data presented are valuable and intriguing. However, as it stands, the validity and utility of this study are uncertain because there are serious deficiencies in the execution and presentation of the authors' results. Many of these shortcomings are fundamental, including a lack of clarity in the basic conceptual framework of the study, insufficient justification of the experimental setup, less-than-conclusive experimental evidence, and inadequate discussion of implications of the authors' findings to future experimental and theoretical studies of biomolecular condensates. Accordingly, this reviewer considers that the manuscript should undergo a major revision to address the following. In particular, the discussion should be significantly expanded by including references mentioned below as well as other references pertinent to the issues raised. 

We thank the reviewer for the helpful comments. In the revised version of the manuscript we clarified that we aimed to use a well-established tool – the scaling analysis – to study phase transition and applied to the protein condensation process. This approach offers insight into a universal aspect of protein phase separation, and also provides a practical approach to determine the phase boundary. The observed fat-tailed distribution of protein droplet sizes is not what is normally observed in more standard phase separation systems in the subsaturated phase. Our contribution is not to propose a theoretical model, but rather to report the observation of a scaling behaviour. 

(1) The theoretical analysis in this study is based on experimental data on condensed droplet size distributions for FUS and α-synuclein. The size data for FUS droplet is indirect as it relies on the assumption that FUS droplet diameter is proportional to fluorescence intensity of labeled FUS (page 10 of manuscript), with fluorescence data adopted from a previously published work by another group (Kar et al. & Pappu, ref.27). Because fluorescence of a droplet is expected to be dependent upon the condensed-phase concentration of FUS, this proportional relationship, even if it holds, must also be modulated by FUS concentration in the droplet. Moreover, why should fluorescence be proportional to diameter but not the cross-sectional area or volume of the FUS droplet, which would be more intuitive? These issues should be clarified. A new measure by microscopy is used to determine the size distribution of condensed α-synuclein; but no microscopy image is shown. It is of critical importance that such raw data (for example microscopy images) be presented for the completeness and reproducibility of the experiment because the entire study relies on the soundness of these experimental measurements. 

As we mentioned in the article, for the scaling analysis, the droplet dimensions could be assessed in 1D (length), 2D (area) or 3D (volume). For the FUS experiments, we used the data as the authors provided in the original publication (PNAS 2022). For alpha-synuclein, we provided the data in the article. 

(2) Despite the authors' claim of a universal scaling relationship, the log-log scatter plots in Figure 1 (page 15 of the manuscript) exhibit significant deviations from linearity at low protein concentrations (ρ→0). Given this fact, is universal scaling really valid? Discussion of this behavior is conspicuously absent (except the statement that these data points are excluded in the fit). In any case, the possible origins of these deviations should be thoroughly discussed so that the regime of universal scaling can be properly delineated. 

In general, one would expect the scaling ansatz to be valid close to the phase boundary. It is the feature of the ansatz, that further away from the boundary, deviations are expected because of the decreasing relevance of critical phenomena.

(3) Droplet size distribution most likely depends on the time duration after the preparation of the sample. For α-synuclein, "liquid droplet size characterisation images were captured 10 minutes post-liquid droplet formation" (page 9 of the manuscript). Why 10 minutes? Have the authors tried imaging at different time points and, if so, do the distributions at different time points remain essentially the same? If they are different, what is the criterion for focusing only on a particular time point? Information related to these questions should be provided. 

We have now determined the droplet size distribution of alpha-synuclein at different time points, finding that they are not dependent on time within experimental uncertainties (Figure 6 in the revised manuscript).

(4) At least two well-known mechanisms can lead to the time-dependent distribution of liquid droplet sizes: (i) coalescence of droplets in spatial proximity to form a larger droplet, and (ii) Ostwald ripening, i.e., formation of larger droplets concomitant with the dissolution of smaller droplets without fusion of droplets. The implications of these mechanisms on the authors' droplet size distributions should be addressed. Indeed, maintaining a size distribution against these mechanisms in vivo often requires active suppression [Bressloff, Phys Rev E 101, 042804 (2020)] with possible involvement of chemical reactions [Kirschbaum & Zwicker, J R Soc Interface 18, 20210255 (2021)]. These considerations are central to the basic rationale of this study and therefore should be carefully tackled. 

These two mechanism of growth are relevant above the critical concentration. Below the critical concentration, which is the regime that we investigated in our work, there is no need of active suppression.

(5) If coalescence and/or Ostwald ripening do occur, given sufficient time after sample preparation, the condensed phase may become a single large "droplet" or a single liquid layer. Does this occur in the authors' experiments? 

As we are below the critical concentration, this is unlikely to occur, as indeed supported by the experiments mentioned at point (3). 

(6) It is unclear whether the authors aim to address the kinetic phenomenon of liquid droplet formation and evolution or equilibrium properties. The two types of phenomena appear to be conflated in the authors' narrative. Clarification is needed. If this work aims to address timeindependent (or infinite-time) equilibrium properties, how are they expected to be related to droplet size distribution, which most likely is time-dependent? 

Our analysis focuses on the equilibrium properties of the droplet size distribution below the critical concentration, and it should guide the proposal of a theoretical model that explains the emergence of scaling. In the introductory part of our manuscript, we proposed a possible scenario that tries to extend the Flory-Huggins’s theory to predict a scaling behaviour appropriate to a critical transition. Other scenarios are possible, and our result along with further experiments are needed to arrive at a deeper understanding of protein aggregation.

(7) The relationship between the potentially time-dependent droplet size distribution and equilibrium properties of ρt and ρc (transition and critical concentrations, respectively) should be better spelled out. An added illustrative figure will be helpful. 

We are addressing equilibrium properties, not kinetic ones. See also the answers to point 6.

(8) The authors comment that their findings appear to be inconsistent with Flory-Huggins theory because Flory-Huggins "characterizes droplet formation as a consequence of nucleation ..." (page 8 of the manuscript). Here, three issues need detailed clarification: (i) In what way does Flory-Huggins mandate nucleation? (ii) Why are the findings of apparent scale invariance inconsistent with nucleation? (iii) If liquid droplet formations do not arise from nucleation, what physical mechanism(s) is (are) envisioned by the authors to be underpinning the formation of condensed liquid droplets in protein phase separation? 

We do agree that the Flory-Huggins theory does not mandate nucleation above the spinodal line. However, we are addressing the equilibrium properties below the critical concentration, so the stable phase is the dilute phase, and there is no nucleation.

(9) Are any of the authors' findings related to finite-system effects of phase separation [see, e.g., Nilsson & Irbäck, Phys Rev E 101, 022413 (2020)]?  

Our experimental system is macroscopic, so we would not expect finite size effects.

(10) Since the authors are using their observation of an apparent scale-invariant droplet size distribution to evaluate phase separation theory, it is important to clarify whether their findings provide any constraint on the shape of coexistence curves (phase diagrams). 

We are only reporting the phenomenological observation of a scaling behaviour, so we may not speculate at this stage on the constraints of the coexistence curves. This is indeed an exciting opportunity for future studies.

(11) More specifically, do the authors' findings suggest that the phase diagrams predicted by Flory-Huggins are invalid? Or, are they suggesting that even if the phase diagrams predicted by Flory-Huggins are empirically correct (if verified by experimental testing), they are underpinned by a free energy function different from that of Flory-Huggins? It is important to answer this question to clarify the implications of the authors' findings on equilibrium phase behaviors and the falsifiability of the implications. 

As mentioned above, our main conclusion is that the droplet size distribution follows a scaling behaviour.  Our contribution is not to propose a theoretical model, but rather to propose a scaling behaviour that should be accounted for by existing of future theoretical models.

(12) How about the implications of the authors' findings on other theories of protein phase separation that are based on interactions that are different from the short spatial range interactions treated by Flory-Huggins? For instance, it has been observed that whereas the Flory-Huggins-predicted phase diagrams always convex upward, phase diagrams for charged intrinsically disordered proteins with long spatial range Coulomb interactions exhibit a region that concave upward [Das et al., Phys Chem Chem Phys 20, 28558-28574 (2018)]. Can information be provided by the authors' findings regarding apparent scale-invariant droplet size distribution on the underlying interaction driving the protein molecules toward phase separation? 

This is an interesting point for future studies about the type of interactions that give rise to the observed scaling behaviour.

(13) Table S1 (page 4) and Table S2 (page 7) are mentioned in the text but these tables are not in the submitted files. 

We have added the Supplementary Tables as well as the source files for the figures.

(14) The two systems studied (FUS and α-synuclein) have a single intrinsically disordered protein (IDP) component. It is not clear if the authors expect their claimed scaling relation to be applicable to systems with multiple IDP components and if so, why.

From the data that we have currently analysed, we feel that we may not speculate on this interesting point, leaving it to future studies.

also consider https://bear.app/ which could also contribute to a MassiveWiki as a Wiki of connected Wikis, organized by "hypothes.is" .

community based tool, based on the Atom editor. https://pulsar-edit.dev/

eLife assessment

This important study shows that in teleost fish, the RIG-I-like protein MDA5 can compensate for the absence of RIG-I by detecting 5'-triphosphorylated RNA. A fish virus containing such RNA can nevertheless evade MDA5 detection through a mechanism involving m6A methylation-induced silencing. The conclusions, which are supported by solid data, advance our understanding of antiviral immunity and virus-host conflicts in vertebrates.

Reviewer #1 (Public Review):

This study offers valuable insights into host-virus interactions, emphasizing the adaptability of the immune system. Readers should recognize the significance of MDA5 in potentially replacing RIG-I and the adversarial strategy employed by 5'ppp-RNA SCRV in degrading MDA5 mediated by m6A modification in different species, further indicating that m6A is a conservational process in the antiviral immune response.<br /> However, caution is warranted in extrapolating these findings universally, given the dynamic nature of host-virus dynamics. The study provides a snapshot into the complexity of these interactions, but further research is needed to validate and extend these insights, considering potential variations across viral species and environmental contexts.

Reviewer #2 (Public Review):

Panel 2N and 2O should have been done with and without SCRV treatment, so that the reader can assess whether SCRV induces additional IFN activation (on top of MDA5 and STING autoactivation). I would recommend the authors include a sentence in the text to explain that ectopic expression of MDA5 or STING (i.e. overexpression from a plasmid) induces autoactivation of these proteins. Therefore, the IFN induction that is seen in panel 2N is likely due to MDA5/STING overexpression. SCRV treatment may further boost IFN induction, but this cannot be assessed without the 'mock' conditions. This information will help the readers to interpret Fig. 2N and 2O correctly.

Author response:

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

eLife assessment

The authors present evidence suggesting that MDA5 can substitute as a sensor for triphosphate RNA in a species that naturally lacks RIG-I. The key findings are potentially important for our understanding of the evolution of innate immune responses. Compared to an earlier version of the paper, the strength of evidence has improved but it is still partially incomplete due to a few key missing experiments and controls.

We would like to thank the editorial team for their positive comments and constructive suggestions on improving our manuscript. We have made further improvements based on the valuable suggestions of the reviewers, and we are pleased to send you the revised manuscript now. After revising the manuscript and further supplementing with experiments, we think that our existing data can support our claims.

Public Reviews:

Reviewer #1 (Public Review):

This study offers valuable insights into host-virus interactions, emphasizing the adaptability of the immune system. Readers should recognize the significance of MDA5 in potentially replacing RIG-I and the adversarial strategy employed by 5'ppp-RNA SCRV in degrading MDA5 mediated by m6A modification in different species, further indicating that m6A is a conservational process in the antiviral immune response.

However, caution is warranted in extrapolating these findings universally, given the dynamic nature of host-virus dynamics. The study provides a snapshot into the complexity of these interactions, but further research is needed to validate and extend these insights, considering potential variations across viral species and environmental contexts. Additionally, it is noted that the main claims put forth in the manuscript are only partially supported by the data presented.

After meticulous revisions of the manuscript, including adjustments to the title, abstract, results, and discussion, the main claim of our study now is the arm race between the MDA5 receptor and SCRV virus in a lower vertebrate fish, M. miiuy. This mainly includes two parts: Firstly, the MDA5 of M. miiuy can recognize virus invasion and initiate host immune response by recognizing the triphosphate structure of SCRV. Secondly, as an adversarial strategy, 5’ppp-RNA SCRV virus can utilize the m6A mechanism to degrade MDA5 in M. miiuy. Based on the reviewer's suggestions, we have further supplemented the critical experiments (Figure 3F-3G, Figure 4D, Figure 5G) and provided a more detailed and accurate explanation of the experimental conclusions, we believe that our existing manuscript can support our main claims. In addition, because virus-host coevolution complicates the derivation of universal conclusions, we will further expand our insights in future research.

Reviewer #2 (Public Review):

This manuscript by Geng et al. aims to demonstrate that MDA5 compensates for the loss of RIG-I in certain species, such as teleost fish miiuy croaker. The authors use siniperca cheats rhabdovirus (SCRV) and poly(I:C) to demonstrate that these RNA ligands induce an IFN response in an MDA5-dependent manner in m.miiuy derived cells. Furthermore, they show that MDA5 requires its RD domain to directly bind to SCRV RNA and to induce an IFN response. They use in vitro synthesized RNA with a 5'triphosphate (or lacking a 5'triphosphate as a control) to demonstrate that MDA5 can directly bind to 5'-triphosphorylated RNA. The second part of the paper is devoted to m6A modification of MDA5 transcripts by SCRV as an immune evasion strategy. The authors demonstrate that the modification of MDA5 with m6A is increased upon infection and that this causes increased decay of MDA5 and consequently a decreased IFN response.

One critical caveat in this study is that it does not address whether ppp-SCRV RNA induces IRF3-dimerization and type I IFN induction in an MDA5 dependent manner. The data demonstrate that mmiMDA5 can bind to triphosphorylated RNA (Fig. 4D). In addition, triphosphorylated RNA can dimerize IRF3 (4C). However, a key experiment that ties these two observations together is missing.

Specifically, although Fig. 4C demonstrates that 5'ppp-SCRV RNA induces dimerization (unlike its dephosphorylated or capped derivatives), this does not proof that this happens in an MDA5-dependent manner. This experiment should have been done in WT and siMDA5 MKC cells side-by-side to demonstrate that the IRF3 dimerization that is observed here is mediated by MDA5 and not by another (unknown) protein. The same holds true for Fig. 4J.

Thank you for the referee's professional suggestions. In fact, we have transfected SCRV RNA into WT and si-MDA5 MKC cells, and subsequently assessed the dimerization of IRF3 and the IFN response (Figure 2P-2Q). The results indicated that knockdown of MDA5 prevents immune activation of SCRV RNA. However, considering the potential for SCRV RNA to activate immunity independent of the triphosphate structure, this experimental observation does not comprehensively establish the MDA5-dependent induction of IRF3 dimer by 5’ppp-RNA. Accordingly, in accordance with the referee's recommendation, we proceeded to investigate the inducible activity of 5'ppp-SCRV on IRF3 dimerization in WT and si-MDA5 MKC cells, revealing that 5'ppp-SCRV indeed elicits immunity in an MDA5-dependent manner (Figure 4D). Additionally, poly(I:C)-HMW, a known ligand for MDA5, demonstrated a residual, albeit attenuated, activation of IRF3 following MDA5 knockdown, potentially attributed to its capacity to stimulate immunity through alternative pathways such as TLR3.

- Fig 1C-D: these experiments are not sufficiently convincing, i.e. the difference in IRF3 dimerization between VSV-RNA and VSV-RNA+CIAP transfection is minimal.

We have reconstituted the necessary materials and repeated the pertinent experiments depicted in Fig 1C-1D. The results demonstrate that SCRV-RNA+CIAP and VSV-RNA+CIAP exhibit a mitigating effect on the induction activity of SCRV-RNA and VSV-RNA on IRF3 dimerization, albeit without complete elimination (Figure 1C and 1D). These findings suggest the presence of receptors within M. miiuy and G. gallus capable of recognizing the viral triphosphate structure; however, it is worth noting that RNA derived from SCRV and VSV viruses does not exclusively depend on the triphosphate structure to activate the host's antiviral response.

Fig. 2N and 2O: why did the authors decide to use overexpression of MDA5 to assess the impact of STING on MDA5-mediated IFN induction? This should have been done in cells transfected with SCRV or polyIC (as in 2D-G) or in infected cells (as in 2H-K). In addition, it is a pity that the authors did not include an siMAVS condition alongside siSTING, to investigate the relative contribution of MAVS versus STING to the MDA5-mediated IFN response. Panel O suggests that the IFN response is completely dependent on STING, which is hard to envision.

In our previous laboratory investigations, we have substantiated the induction effect of STING on IFN under SCRV infection or poly(I:C) stimulation, as documented in the relevant literature (10.1007/s11427-020-1789-5), which we have referenced in our manuscript (lines 177-178). While we did assess the impact of STING on MDA5-mediated IFN induction in SCRV-infected cells, as indicated in the figure legends, we have revised Figure 2N-2O for improved clarity, and similarly, Figure 1H-1I has also been updated. Furthermore, considering that RNA virus infection can activate the cGAS/STING axis (10.3389/fcimb.2023.1172739) and the significant role of MAVS in sensing RNA virus invasion in the NLR pathway (10.1038/ni.1782), it is challenging to ascertain the respective contributions of STING and MAVS to the immune signaling cascade mediated by MDA5 during RNA virus infection. We intend to explore this aspect further in future research endeavors.

Fig. 3F and 3G: where are the mock-transfected/infected conditions? Given that ectopic expression of hMDA5 is known to cause autoactivation of the IFN pathway, the baseline ISG levels should be shown (ie. In absence of a stimulus or infection). Normalization of the data does not reveal whether this is the case and is therefore misleading.

Based on the reviewer's suggestions, we have rerun the experiment. We examined the effects of MDA5 and MDA5-ΔRD on antiviral factors in both uninfected, SCRV-infected, and poly(I:C)-HMW-stimulated MKC cells. Results showed that overexpression of both MDA5 and MDA5-ΔRD stimulated the expression of antiviral genes. However, when cells were infected or stimulated with SCRV or poly(I:C)-HMW, only the overexpression of MDA5, not MDA5-ΔRD, significantly increased the expression of antiviral genes (Figure 3F-3I).

Fig. 4F and 4G: can the authors please indicate in the figure which area of the gel is relevant here? The band that runs halfway the gel? If so, the effects described in the text are not supported by the data (i.e. the 5'OH-SCRV and 5'pppGG-SCRV appear to compete with Bio-5'ppp-SCRV as well as 5'ppp-SCRV).

Apologies for any confusion. The relevant areas in the gel pertaining to the experimental findings were denoted with asterisks and elaborated upon in the figure legends (Figure 4G, 4H, and 4M). The findings indicated that 5'ppp-SCRV, in contrast to 5'OH-SCRV and 5'pppGG-SCRV, demonstrated the ability to compete with bio-5'ppp-SCRV.

My concerns about Fig. 5 remain unaltered. The fact that MDA5 is an ISG explains its increased expression and increased methylation pattern. The authors should at the very least mention in their text that MDA5 is an ISG and that their observations may be partially explained by this fact.

First, as our m6A change analysis pipeline controls for changes in gene expression, these data should represent true changes in m6A modification rather than changes in the expression of m6A-modified transcripts (10.1038/s41598-020-63355-3). Similar studies demonstrated that m6A modification in RIOK3 and CIRBP mRNAs are altered following Flaviviridae infection (10.1016/j.molcel.2019.11.007). The specific calculation method is as follows: relative m6A level for each transcript was calculated as the percent of input in each condition normalized to that of the respective positive control spike-in. Fold change of enrichment was calculated with mock samples normalized to 1. Therefore, changes in the expression level of MDA5 can partially explain the increase in m6A modification on all MDA5 mRNA in cells, but it cannot indicate changes in m6A modification on each mDA5 transcript. We have supplemented the calculation method process in the manuscript and cited relevant literature (Lines 606-608). In addition, we have elaborated on the fact that MDA5 is an ISG gene in the experimental results (lines 260-261), and emphasized its compatibility with enhanced m6A modification of MDA5 in the discussion section (lines 405-409).

Reviewer #3 (Public Review):

In this manuscript, the authors explored the interaction between the pattern recognition receptor MDA5 and 5'ppp-RNA in the Miiuy croaker. They found that MDA5 can serve as a substitute for RIG-I in detecting 5'ppp-RNA of Siniperca cheilinus rhabdovirus (SCRV) when RIG-I is absent in Miiuy croaker. Furthermore, they observed MDA5's recognition of 5'ppp-RNA in chickens (Gallus gallus), a species lacking RIG-I. Additionally, the authors documented that MDA5's functionality can be compromised by m6A-mediated methylation and degradation of MDA5 mRNA, orchestrated by the METTL3/14-YTHDF2/3 regulatory network in Miiuy croaker during SCRV infection. This impairment compromises the innate antiviral immunity of fish, facilitating SCRV's immune evasion. These findings offer valuable insights into the adaptation and functional diversity of innate antiviral mechanisms in vertebrates.

We extend our sincere appreciation for your professional comments and insightful suggestions on our manuscript, as they have significantly contributed to enhancing its quality.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

(1) The interpretation of Figures 1H and I, along with the captions, seems unclear. Particularly, understanding the meaning of the X-axis in Figure I is challenging. Additionally, the designation of "H2O = 1" on the Y-axis in Figure 1E lacks clarity. It would be helpful if the author could revise and clarify these figures for better comprehension.

We appreciate your reminder and have corrected and clarified these figures and figure legends (lines 768-772). We have replaced the Y-axis of Figure 1I with "Relative mRNA expression" instead of " Relative IFN-1 expression" (Figure 1I). In addition, we have added an explanation of "H2O=1" in the legend of Figure 1E.

(2) The interpretation of Figure 5 in section 2.5 seems incomplete. The author mentioned that both m6A levels and MDA5 expression levels are increased (lines 256-257), prompting questions about the relationship between m6A and MDA5 expression. If higher m6A levels typically lead to MDA5 mRNA instability and lower MDA5 expression, observing both increasing simultaneously appears contradictory. Considering the dynamic changes shown in Figure 5, it would be more appropriate to propose an alteration in both m6A levels and MDA5 expression levels. Given the fluctuating nature of these changes, definitively labeling them as solely "increased" is challenging. Therefore, offering a nuanced interpretation of the results and clarifying this aspect would bolster the study's conclusions.

While changes in m6A modification and the expression of m6A-modified transcripts are biologically relevant, identifying bona fide m6A alterations during viral infection will allow us to understand how m6A modification of cellular mRNA is regulated. As our m6A change analysis pipeline controls for changes in gene expression, these data should represent true changes in m6A modification rather than changes in the expression of m6A-modified transcripts (10.1038/s41598-020-63355-3). Similar studies demonstrated that m6A modification in RIOK3 and CIRBP mRNAs are altered following Flaviviridae infection (10.1016/j.molcel.2019.11.007). The specific calculation method is as follows: relative m6A level for each transcript was calculated as the percent of input in each condition normalized to that of the respective positive control spike-in. Fold change of enrichment was calculated with mock samples normalized to 1. Therefore, the upregulation of MDA5 expression can partially explain the increase in m6A modification on all MDA5 mRNA in cells, but it cannot indicate changes in m6A modification on each mDA5 transcript. We have supplemented the calculation method process in the manuscript and cited relevant literature. I hope to receive your understanding.

In addition, although higher m6A levels often lead to unstable MDA5 mRNA and lower MDA5 expression, SCRV can affect MDA5 expression through multiple pathways. For example, since MDA5 is an interferon-stimulated gene, the infection of SCRV virus can cause strong expression of interferon and indirectly induce high-level expression of MDA5. Therefore, the expression of MDA5 is not contradictory to the simultaneous increase in MDA5 modification (24 h). In order to further enhance our experimental conclusions, we supplemented the dual fluorescence experiment. The results indicate that, the infection of SCRV can inhibit the fluorescence activity of MDA5-exon1 reporter plasmids containing m6A sites but not including the promoter sequence of the MDA5 gene, and this inhibitory effect can be counteracted by cycloleucine (CL, an amino acid analogue that can inhibit m6A modification) (Figure 5G). This further indicates that SCRV can reduce the expression of MDA5 through the m6A pathway.

Finally, in light of the fluctuations in MDA5 expression levels, we have changed the subheadings of Results 2.5 section and provided a more comprehensive and precise elucidation of the experimental outcomes. We are grateful for your valuable feedback.

(3) In the discussion section, it would indeed be advantageous for the author to explore the novelty of this work more comprehensively, moving beyond merely acknowledging the widespread loss of RIG-I and suggesting MDA5 as a compensatory mechanism. Considering the well-established roles of MDA5 and m6A in host-virus interactions, the findings of this study may seem familiar in light of previous research. To enhance the discussion, it would be valuable for the author to delve into the implications of this evolutionary model. For instance, does the compensation or loss of RIG-I impact a species' susceptibility to specific types of viruses? Exploring such questions would provide insight into the broader significance of this compensation model and its potential effects on host-virus interactions, thus adding depth to the study's contribution.

We appreciate the expert advice provided by the referee. In response, we have expanded our discussion in the relevant section, addressing the potential influence of RIG-I deficiency and MDA5 compensation on the antiviral immune system in vertebrates (lines 371-376). Furthermore, we underscore the significance of exploring the impact of SCRV infection on MDA5 m6A modification, considering its compatibility with MDA5 as an ISG gene, in elucidating the host response to viral infection (lines 405-409).

(4) To improve the manuscript, it would be beneficial if the editors could aid the author in refining the language. Many descriptions in the article are overly redundant, and there should be appropriate differentiation between experimental methods and results.

We appreciate the reviewer’s comment. We have carefully revised the manuscript and removed redundant descriptions in the experimental results and methods.

Reviewer #3 (Recommendations For The Authors):

The authors have addressed all of my concerns.

Note: This response was posted by the corresponding author to Review Commons. The content has not been altered except for formatting.

Learn more at Review Commons

---

Reply to the reviewers

Reviewer 1

R1 Cell profiling is an emerging field with many applications in academia and industry. Finding better representations for heterogeneous cell populations is important and timely. However, unless convinced otherwise after a rebuttal/revision, the contribution of this paper, in our opinion, is mostly conceptual, but in its current form - not yet practical. This manuscript combined two concepts that were previously reported in the context of cell profiling, weakly supervised representations. Our expertise is in computational biology, and specifically applications of machine learning in microscopy.

In our revised manuscript, we have aimed to better clarify the practical contributions of our work by demonstrating the effectiveness of the proposed concepts on real-world datasets. We hope that these revisions and our detailed responses address your concerns and highlight the potential impact of our approach.

R1.1a. CytoSummaryNet is evaluated in comparison to aggregate-average profiling, although previous work has already reported representations that capture heterogeneity and self-supervision independently. To argue that both components of contrastive learning and sets representations are contributing to MoA prediction we believe that a separate evaluation for each component is required. Specifically, the authors can benchmark their previous work to directly evaluate a simpler population representation (PMID: 31064985, ref #13) - we are aware that the authors report a 20% improvement, but this was reported on a separate dataset. The authors can also compare to contrastive learning-based representations that rely on the aggregate (average) profile to assess and quantify the contribution of the sets representation.

We agree that evaluating the individual contributions of the contrastive learning framework and single-cell data usage is important for understanding CytoSummaryNet's performance gains.

To assess the impact of the contrastive formulation independently, we applied CytoSummaryNet to averaged profiles from the cpg0004 dataset. This isolated the effect of contrastive learning by eliminating single-cell heterogeneity. The experiment yielded a 32% relative improvement in mechanism of action retrieval, compared to the 68% gain achieved with single-cell data. These findings suggest that while the contrastive formulation contributes significantly to CytoSummaryNet's performance, leveraging single-cell information is crucial for maximizing its effectiveness. We have added a discussion of this experiment to the Results section:

“We conducted an experiment to determine whether the improvements in mechanism of action retrieval were due solely to CytoSummaryNet's contrastive formulation or also influenced by the incorporation of single-cell data. We applied the CytoSummaryNet framework to pre-processed average profiles from the 10 μM dose point data of Batch 1 (cpg0004 dataset). This approach isolated the effect of the contrastive architecture by eliminating single-cell data variability. We adjusted the experimental setup by reducing the learning rate by a factor of 100, acknowledging the reduced task complexity. All other parameters remained as described in earlier experiments.

This method yielded a less pronounced but still substantial improvement in mechanism of action retrieval, with an increase of 0.010 (32% enhancement - Table 1). However, this improvement was not as high as when the model processed single-cell level data (68% as noted above). These findings suggest that while CytoSummaryNet's contrastive formulation contributes to performance improvements, the integration of single-cell data plays a critical role in maximizing the efficacy of mechanism of action retrieval.”

We don't believe comparing with PMID: 31064985 is useful: while the study showcased the usefulness of modeling heterogeneity using second-order statistics, its methodology is limited in scalability due to the computational burden of computing pairwise similarities for all perturbations, particularly in large datasets. Additionally, the study's reliance on similarity network fusion, while expedient, introduces complexity and inefficiency. We contend that this comparison does not align with our objective of testing the effectiveness of heterogeneity in isolation, as it primarily focuses on capturing second and first-order information. Thus, we do not consider this study a suitable baseline for comparison.

R1.1b. The evaluation metric of mAP improvement in percentage is misleading, because a tiny improvement for a MoA prediction can lead to huge improvement in percentage, while a much larger improvement in MoA prediction can lead to a small improvement in percentage. For example, in Fig. 4, MEK inhibitor mAP improvement of ~0.35 is measured as ~50% improvement, while a much smaller mAP improvement can have the same effect near the origins (i.e., very poor MoA prediction).

We agree that relying solely on percentage improvements can be misleading, especially when small absolute changes result in large percentage differences.

However, we would like to clarify two key points regarding our reporting of percentage improvements:

• We calculate the percentage improvement by first computing the average mAP across all compounds for both CytoSummaryNet and average profiling, and then comparing these averages. This approach is less susceptible to the influence of outlier improvements compared to calculating the average of individual compound percentage improvements.
• We report percentage improvements alongside their corresponding absolute improvements. For example, the mAP improvement for Stain4 (test set) is reported as 0.052 (60%). To further clarify this point, we have updated the caption of Table 1 to explicitly state how the percentage improvements are calculated:

“The improvements are calculated as mAP(CytoSummaryNet)-mAP(average profiling). The percentage improvements are calculated as (mAP(CytoSummaryNet)-mAP(average profiling))/mAP(average profiling).”

R1.1b. (Subjective) visual assessment of this figure does not show a convincing contribution of CytoSummaryNet representations of the average profiling on the test set (3.33 uM). This issue might also be relevant for the task of replicate retrieval. All in all, the mAP improvement reported in Table 1 and throughout the manuscript (including the Abstract), is not a proper evaluation metric for CytoSummaryNet contribution. We suggest reporting the following evaluations:

1. Visualizing the results of cpg0001 (Figs. 1-3) similarly to cpg0004 (Fig. 4), i.e., plotting the matched mAP for CytoSummaryNet vs. average profile.

2. In Table 1, we suggest referring to the change in the number of predictable MoAs (MoAs that pass a mAP threshold) rather than the improvement in percentages. Another option is showing a graph of the predictability, with the X axis representing a threshold and Y-axis showing the number of MoAs passing it. For example see (PMID: 36344834, Fig. 2B) and (PMID: 37031208, Fig. 2A), both papers included contributions from the corresponding author of this manuscript.

Regarding the suggestion to visualize the results for compound group cpg0001 similarly to cpg0004, unfortunately, this is not feasible due to the differences in data splitting between the two datasets. In cpg0001, an MoA might have one compound in the training set and another in the test or validation set. Reporting a single value per MoA would require combining these splits, which could be misleading as it would conflate performance across different data subsets.

However, we appreciate the suggestion to represent the number of predictable MoAs that surpass a certain mAP threshold, as it provides another intuitive measure of performance. To address this, we have created a graph that visualizes the predictability of MoAs across various thresholds, similar to the examples provided in the referenced papers (PMID: 36344834, Figure 2B and PMID: 37031208, Figure 2A). This graph, with the x-axis depicting the threshold and the y-axis showing the number of MoAs meeting the criterion, has been added to Supplementary Material K.

R1.1c.i. "a subset of 18 compounds were designated as validation compounds" - 5 cross-validations of 18 compounds can make the evaluation complete. This can also enhance statistical power in figures 1-3.

We appreciate your suggestion and acknowledge the potential benefits of employing cross-validation, particularly in enhancing statistical power. While we understand the merit of cross-validation for evaluating model performance and generalization to unseen data, we believe the results as presented already highlight the generalization characterics of our methods.

Specifically, (the new) Figure 3 demonstrates the model's improvement over average profiling in both training and validation plates, supporting its ability to generalize to unseen compounds (but not to unseen plates).

While cross-validation could potentially enhance our analysis, retraining five new models solely for different validation set results may not substantially alter our conclusions, given the observed trends in Suppl Figure A1 and (the new) Figure 4, both of which show results across multiple stain sets (but a single train-test-validation split).

---

R1.1c.ii. Clarify if the MoA results for cpg0001 are drawn from compounds from both the training and the validation datasets. If so, describe how the results differ between the sets in text and graphs.

We confirm that the Mechanism of Action (MoA) retrieval results for cpg0001 are derived from all available compounds. It's important to note that the training and validation dataset split for the replicate retrieval task is different from the MoA prediction task. For replicate retrieval, we train using all available compounds and validate on a held-out set (see Figure 2). For MoA prediction, we train using the replicate retrieval task as the objective on all available compounds but validate using MoA retrieval, which is a distinct task. We have added a brief clarification in the main text to highlight the distinction between these tasks and how validation is performed for each:

“We next addressed a more challenging task: predicting the mechanism of action class for each compound at the individual well level, rather than simply matching replicates of the exact same compound (Figure 5). It's also important to note that mechanism of action matching is a downstream task on which CytoSummaryNet is not explicitly trained. Consequently, improvements observed on the training and validation plates are more meaningful in this context, unlike in the previous task where only improvements on the test plate were meaningful. For similar reasons, we calculate the mechanism of action retrieval performance on all available compounds, combining both the training and validation sets. This approach is acceptable because we calculate the score on so-called "sister compounds" only—that is, different compounds that have the same mechanism of action annotation. This ensures there is no overlap between the mechanism of action retrieval task and the training task, maintaining the integrity of our evaluation. ”

R1.1c.iii. "Mechanism of action retrieval is evaluated by quantifying a profile's ability to retrieve the profile of other compounds with the same annotated mechanism of action.". It was unclear to us if the evaluation of mAP for MoA identification can include finding replicates of the same compound. That is, whether finding a close replicate of the same compound would be included in the AP calculation. This would provide CytoSummaryNet with an inherent advantage as this is the task it is trained to do. We assume that this was not the case (and thus should be more clearly articulated), but if it was - results need to be re-evaluated excluding same-compound replicates.

The evaluation excludes replicate wells of the same compound and only considers wells of other compounds with the same MoA. This methodology ensures that the model's performance on the MoA prediction task is not inflated by its ability to find replicates of the same compound, which is the objective of the replicate retrieval task. Please see the explanation we have added to the main text in our response to R1.1c.ii. Additionally, we have updated the Methods section to clearly describe this evaluation procedure:

“Mechanism of action retrieval is evaluated by quantifying a profile’s ability to retrieve the profile of different compounds with the same annotated mechanism of action.”

---

---

__R1.2a. __The description of Stain2-5 was not clear for us at first (and second) read. The information is there, but more details will greatly enhance the reader's ability to follow. One suggestion is explicitly stating that these "stains" partitioning was already defined in ref 26. Another suggestion is laying out explicitly a concrete example on the differences between two of these stains. We believe highlighting the differences between stains will strengthen the claim of the paper, emphasizing the difficulty of generalizing to the out-of-distribution stain.

We appreciate your feedback on the clarity of the Stain2-5 dataset descriptions; we certainly struggled to balance detail and concepts in describing these. We have made the following changes:

• Explicitly mentioned that the partitioning of the Stain experiments was defined in https://pubmed.ncbi.nlm.nih.gov/37344608/: “The partitioning of the Stain experiments have been defined and explained previously [21].”
• Moved an improved version of (now) Figure 2 from the Methods section to the main text to help visually explain how the stratification is done early on.
• Added a new section in the Experimental Setup: Diversity of stain sets, which includes a concrete example highlighting the differences between Stain2, and Stain5 to emphasize the diversity in experimental setups within the same dataset: “Stain2-5 comprise a series of experiments which were conducted sequentially to optimize the experimental conditions for image-based cell profiling. These experiments gradually converged on the most optimal set of conditions; however, within each experiment, there were significant variations in the assay across plates. To illustrate the diversity in experimental setups within the dataset, we will highlight the differences between Stain2 and Stain5.

Stain2 encompasses a wide range of nine different experimental protocols, employing various imaging techniques such as Widefield and Confocal microscopy, as well as specialized conditions like multiplane imaging and specific stains like MitoTracker Orange. This subset also includes plates acquired with strong pixel binning instead of default imaging and plates with varying concentrations of dyes like Hoechst. As a result, Stain2 exhibits greater variance in the experimental conditions across different plates compared to Stain5.

In contrast, Stain5, the last experiment in the series, follows a more systematic approach, consistently using either confocal or default imaging across three well-defined conditions. Each condition in Stain5 utilizes a lower cell density of 1,000 cells per well compared to Stain2's 2,500 cells per well. Being the final experiment in the series, Stain5 had the least variance in experimental conditions.

For training the models, we typically select the data containing the most variance to capture the broadest range of experimental variation. Therefore, we chose Stain2-4 for training, as they represented the majority of the data and captured the most experimental variation. We reserved Stain5 for testing to evaluate the model's ability to generalize to new experimental conditions with less variance.

All StainX experiments were acquired in different passes, which may introduce additional batch effects.”

These changes aim to provide a clearer understanding of the dataset's complexity and the challenges associated with generalizing to out-of-distribution data.

R1.2b. What does each data point in Figures 1-3 represent? Is it the average mAP for the 18 validation compounds, using different seeds for model training? Why not visualize the data similarly to Fig. 4 so the improvement per compound can be clearly seen?

The data points in (the new) Figures 3,4,5 represent the average mAP for each plate, calculated by first computing the mAP for each compound and then averaging across compounds to obtain the average mAP per plate. We have updated the figure captions to clarify this:

"... (each data point is the average mAP of a plate) ..."

While visualizing the mAP per compound, similar to (the new) Figure 6 for cpg0004, could provide insights into compound-level improvements, it would require creating numerous additional figures or one complex figure to adequately represent all the stratifications we are analyzing (plate, compound, Stain subset). By averaging the data per plate across different stratifications, we aim to provide a clearer and more comprehensible overview of the trends and improvements while allowing us to draw conclusions about generalization.

Please note: this comment is related to the comment R1.1b (Subjective)

R1.2.c [On the topic of enhancing clarity and readability:] Justification and interpretation of the evaluation metrics.

Please refer to our response to comment R1.1b, where we have addressed your concerns regarding the justification and interpretation of the evaluation metrics.

R1.2d. Explicitly mentioning the number of MoAs for each datasets and statistics of number of compounds per MoA (e.g., average\median, min, max).

We have added the following to the Experimental Setup: Data section:

“A subset of the data was used for evaluating the mechanism of action retrieval task, focusing exclusively on compounds that belong to the same mechanism class. The Stain plates contained 47 unique mechanisms of action, with each compound replicated four times. Four mechanisms had only a single compound; the four mechanisms (and corresponding compounds) were excluded, resulting in 43 unique mechanisms used for evaluation. In the LINCS dataset, there were 1436 different mechanisms, but only 661 were used for evaluation because the remaining had only one compound.”

R1.2e. The data split in general is not easily understood. Figure 8 is somewhat helpful, however in our view, it can be improved to enhance understanding of the different splits. Specifically, the training and validation compounds need to be embedded and highlighted within the figure.

Thank you for highlighting this. We have completely revised the figure, now Figure 2 which we hope more clearly conveys the data split strategy.

Please note: this comment is related to the comment R1.2a.

---

---

---

---

R1.3a. Why was stain 5 used for the test, rather than the other stains?

Stain2-5 were part of a series of experiments aimed at optimizing the experimental conditions for image-based cell profiling using Cell Painting. These experiments were conducted sequentially, gradually converging on the most optimal set of conditions. However, within each experiment, there were significant variations in the assay across plates, with earlier iterations (Stain2-4) having more variance in the experimental conditions compared to Stain5. As Stain5 was the last experiment in the series and consisted of only three different conditions, it had the least variance. For training the models, we typically select the data containing the most variance to capture the broadest range of experimental variation. Therefore, Stain2-4 were chosen for training, while Stain5 was reserved for testing to evaluate the model's ability to generalize to new experimental conditions with less variance.

We have now clarified this in the Experimental Setup: Diversity of stain sets section. Please see our response to comment R1.2a. for the full citation.

R1.3b How were the 18 validation compounds selected?

20% of the compounds (n=18) were randomly selected and designated as validation compounds, with the remaining compounds assigned to the training set. We have now clarified this in the Results section:

“Additionally, 20% of the compounds (n=18) were randomly selected and designated as validation compounds, with the remaining compounds assigned to the training set (Supplementary Material H).”

R1.3c. For cpg0004, no justification for the specific doses selected (10uM - train, 3.33 uM - test) for the analysis in Figure 4. Why was the data split for the two dosages? For example, why not perform 5-fold cross validation on the compounds (e.g., of the highest dose)?

We chose to use the 10 μM dose point as the training set because we expected this higher dosage to consist of stronger profiles with more variance than lower dose points, making it more suitable for training a model. We decided to use a separate test set at a different dose (3.33 μM) to assess the model's ability to generalize to new dosages. While cross-validation on the highest dose could also be informative, our approach aimed to balance the evaluation of the model's generalization capability with its ability to capture biologically relevant patterns across different dosages.

This explanation has been added to the text:

“We chose the 10 μM dose point for training because we expected this high dosage to produce stronger profiles with more variance than lower dose points, making it more suitable for model training.”

“The multiple dose points in this dataset allowed us to create a separate hold-out test set using the 3.33 μM dose point data. This approach aimed to evaluate the model's performance on data with potentially weaker profiles and less variance, providing insights into its robustness and ability to capture biologically relevant patterns across dosages. While cross-validation on the 10 μM dose could also be informative, focusing on lower dose points offers a more challenging test of the model's capacity to generalize beyond its training conditions, although we do note that all compounds’ phenotypes would likely have been present in the 10 μM training dataset, given the compounds tested are the same in both.”

R1.3d. A more detailed explanation on the logic behind using a training stain to test MoA retrieval will help readers appreciate these results. In our first read of this manuscript we did not grasp that, we did in a second read, but spoon-feeding your readers will help.

This comment is related to the rationale behind training on one task and testing on another, which is addressed in our responses to comments R1.1.cii and R1.1.ciii.

R1.4 Assessment of interpretability is always tricky. But in this case, the authors can directly confirm their interpretation that the CytoSummaryNet representation prioritizes large uncrowded cells, by explicitly selecting these cells, and using their average profile re

We progressively filtered out cells based on a quantile threshold for Cells_AreaShape features (MeanRadius, MaximumRadius, MedianRadius, and Area), which were identified as important in our interpretability analysis, and then computed average profiles using the remaining cells before determining the replicate retrieval mAP. In the exclusion experiment, we gradually left out cells as the threshold increased, while in the inclusion experiment, we progressively included larger cells from left to right.

The results show that using only the largest cells does not significantly increase the performance. Instead, it is more important to include the large cells rather than only including small cells. The mAP saturates after a threshold of around 0.4, indicating that larger cells define the profile the most, and once enough cells are included to outweigh the smaller cell features, the profile does not change significantly by including even larger cells.

These findings support our interpretation that CytoSummaryNet prioritizes large, uncrowded cells. While this approach could potentially be used as a general outlier removal strategy for cell profiling, further investigation is needed to assess its robustness and generalizability across different datasets and experimental conditions.

We have created Supplementary Material L to report these findings and we additionally highlight them in the Results:

“To further validate CytoSummaryNet's prioritization of large, uncrowded cells, we progressively filtered cells based on Cells_AreaShape features and observed the impact on replicate retrieval mAP (Supplementary Material L). The results support our interpretation and highlight the key role of larger cells in profile strength.”

__R1.5. __Placing this work in context of other weakly supervised representations. Previous papers used weakly supervised labels of proteins / experimental perturbations (e.g., compounds) to improve image-derived representations, but were not discussed in this context. These include PMID: 35879608, https://www.biorxiv.org/content/10.1101/2022.08.12.503783v2 (from the same research groups and can also be benchmarked in this context), https://pubs.rsc.org/en/content/articlelanding/2023/dd/d3dd00060e , and https://www.biorxiv.org/content/10.1101/2023.02.24.529975v1. We believe that a discussion explicitly referencing these papers in this specific context is important.

While these studies provide valuable insights into improving cell population profiles using representation learning, our work focuses specifically on the question of single-cell aggregation methods. We chose to use classical features for our comparisons because they are the current standard in the field. This approach allows us to directly assess the performance of our method in the context of the most widely used feature extraction pipeline in practice. However, we see the value in incorporating them in future work and have mentioned them in the Discussion:

“Recent studies exploring image-derived representations using self-supervised and self-supervised learning [35][36] could inspire future research on using learned embeddings instead of classical features to enhance model-aggregated profiles.”

R1.minor1. "Because the improved results could stem from prioritizing certain features over others during aggregation, we investigated each cell's importance during CytoSummaryNet aggregation by calculating a relevance score for each" - what is the relevance score? Would be helpful to provide some intuition in the Results.

We have included more explanation of the relevance score in the Results section, following the explanation of sensitivity analysis (SA) and critical point analysis (CPA):

“SA evaluates the model's predictions by analyzing the partial derivatives in a localized context, while CPA identifies the input cells with the most significant contribution to the model's output. The relevance scores of SA and CPA are min-max normalized per well and then combined by addition. The combination of the two is again min-max normalized, resulting in the SA and CPA combined relevance score (see Methods for details).”

R1.minor2. Figure 1:

1. Colors of the two methods too similar
2. The dots are too close. It will be more easily interpreted if they were further apart.
3. What do the dots stand for?
4. We recommend considering moving this figure to the supp. material (the most important part of it is the results on the test set and it appears in Fig.2).

1. We chose a lighter and darker version of the same color as a theme to simplify visualization, as this theme is used throughout (the new) Figures 3,4,5.
2. We agree; we have now redrawn the figure to fix this.
3. Each data point is the average mAP of a plate. Please see our answer for R1.2b as well.
4. We believe that (the new) Figures 3,4,5 serve distinct purposes in testing various generalization hypotheses. We have added the following text to emphasize that the first figures are specifically about generalization hypothesis testing: “We first investigated CytoSummaryNet’s capacity to generalize to out-of-distribution data: unseen compounds, unseen experimental protocols, and unseen batches. The results of these investigations are visualized in Figures 3, 4, and 5, respectively.”

R1.minor3 Figure 4: It is somewhat misleading to look at the training MoAs and validation MoAs embedded together in the same graph. We recommend showing only the test MoAs (train MoAs can move to SI).

We addressed this comment in R1.1c.ii. To reiterate briefly, there are no training, validation, or test MoAs because these are not used as labels during the training process. There is an option to split them based on training and validation compounds, which is addressed in R1.1c.ii.

---

R1.minor4 Figure 5

1. Why only Stain3? What happens if we look at Stains 2,3 and 4 together? Stain 5?

2. Should validation compounds and training compounds be analyzed separately?

3. Subfigure (d): it is expected that the data will be classified by compound labels as it is the training task, but for this to be persuasive I would like to see this separately on the training compounds first and then and more importantly on the validation compounds.

4. For subfigures (b) and (d): it appears there are not enough colors for d, which makes it partially not understandable. For example, the pink label in (d) shows a single compound which appears to represent two different MoAs. This is probably not the case, and it has two different compounds, but it cannot be inferred when they are represented by the same color.

5. For the Subfigure (e) - only 1 circle looks justified (in the top left). And for that one, is it not a case of an outlier plate that would perhaps need to be removed from analysis? Is it not good that such a plate will be identified?

We have addressed this point in the text, stating that the results are similar for Stain2 and Stain4. Stain5 represents an out-of-distribution subset because of a very different set of experimental conditions (see Experimental Setup: Diversity of stain sets). To improve clarity, we have revised the figure caption to reiterate this information:

“... Stain2 and Stain4 yielded similar results (data not shown). …”

1. For replicate retrieval, analyzing validation and training compounds separately is appropriate. However, this is not the case for MoA retrieval, as discussed in our responses to R1.1c.ii and R1.1c.i.
2. We have created the requested plot (below) but ultimately decided not to include it in the manuscript because we believe that (the new) Figures 3 and 4 are more effective for making quantitative comparative claims.

[Please see the full revision document for the figures]

Top: training compounds (validation compounds grayed out); not all compounds are listed in the legend.

*Bottom: validation compounds (training compounds grayed out). *

Left: average profiling; Right: CytoSummaryNet

1. We agree with your observation and have addressed this issue by labeling the center mass as a single class (gray) and highlighting only the outstanding pairs in color. Please refer to the updated figure and our response to R3.6 for more details.

2. In the updated figure, we have revised the figure caption to focus solely on the annotation of same mechanism of action profile clusters, as indicated by the green ellipses. The annotation of isolated plate clusters has been removed (Figures 7e and 7f) to maintain consistency and avoid potential confusion. Despite being an outlier for Stain3, the plate (BR00115134bin1) clusters with Stain4 plates (Supplementary Figure F1, green annotated square inside the yellow annotated square), indicating it is not merely a noisy outlier and can provide insights into the out-of-sample performance of our model.

R1.minor5a. Discussion: "perhaps in part due to its correction of batch effects" - is this statement based on Fig. 5F - we are not convinced.

We appreciate the reviewer's scrutiny regarding our statement about batch effect correction. Upon reevaluation, we agree that this claim was not adequately substantiated by empirical data. We quantified the batch effects using comparison mean average precision for both average profiles and CytoSummaryNet profiles, and the statistical analysis revealed no significant difference between these profiles in terms of batch effect correction. Therefore, we have removed this theoretical argument from the manuscript entirely to ensure that all claims are strongly supported by the data presented.

R1.minor5b. "Overall, these results improve upon the ~20% gains we previously observed using covariance features" - this is not the same dataset so it is hard to reach conclusions - perhaps compare performance directly on the same data?

We have now explicitly clarified this is a different dataset. Please see our response to R1.1a for why a direct comparison was not performed. The following clarification can be found in the Discussion:

“These results improve upon the ~20% gains previously observed using covariance features [13] albeit on a different dataset, and importantly, CytoSummaryNet effectively overcomes the challenge of recomputation after training, making it easier to use.”

Reviewer 2

R2.1 The authors present a well-developed and useful algorithm. The technical motivation and validation are very carefully and clearly explained, and their work is potentially useful to a varied audience.

That said, I think the authors could do a better job, especially in the figures, of putting the algorithm in context for an audience that is unfamiliar with the cell painting assay. (a) For example, a figure towards the beginning of the paper with example images might help to set the stage. (b) Similarly a schematic of the algorithm earlier in the paper would provide a graphical overview. (c) For the sake of a biologically inclined audience, I would consider labeling the images in the caption by cell type and label.

Thank you for your valuable suggestions on improving the accessibility of our figures for readers unfamiliar with the Cell Painting assay. We have made the following changes to address your comments:

1. and b. To provide visual context and a graphical overview of the algorithm, we have moved the original Figure 7 to Figure 1. This figure now includes example images that help readers new to the Cell Painting assay.
2. We have added relevant details to the example images in (the new) Figure 1

   R2.2 The interpretability results were intriguing. The authors might consider further validating these interpretations by removing weakly informative cells from the dataset and retraining. Are the cells so uninformative that the algorithm does better without them, or are they just less informative than other cells?

Please see our responses to R1.4 and R3.0

R2.3 As far as I can tell, the authors only oblique state whether the code associated with the manuscript is openly available. Posting the code is needed for reproducibility. I would provide not only a github, but a doi linked to the code, or some other permanent link.

We have now added a Code Availability and Data Availability section, clearing stating that the code and data associated with the manuscript are openly available.

R2.4 Incorporating biological heterogeneity into machine-learning driven problems is a critical research question. Replacing means/modes and such with a machine learning framework, the authors have identified a problem with potentially wide significance. The application to cell painting and related assays is of broad enough significance for many journals, However, the authors could further broaden the significance by commenting on other possible cell biology applications. What other applications might the algorithm be particularly suited for? Are there any possible roadblocks to wider use. What sorts of data has the code been tested on so far?

We have added the following paragraph to discuss the broader applicability of CytoSummaryNet:

“The architecture of CytoSummaryNet holds significant potential for broader applications beyond image-based cell profiling, accommodating tabular, permutation-invariant data and enhancing downstream task performance when applied to processed population-level profiles. Its versatility makes it valuable for any omics measurements where downstream tasks depend on measuring similarity between profiles. Future research could also explore CytoSummaryNet's applicability to genetic perturbations, expanding its utility in functional genomics.”

Reviewer 3

R3.0 The authors have done a commendable job discussing the method, demonstrating its potential to outperform current models in profiling cell-based features. The work is of considerable significance and interest to a wide field of researchers working on the understanding of cell heterogeneity's impact on various biological phenomena and practical studies in pharmacology.

One aspect that would further enhance the value of this work is an exploration of the method's separation power across different modes of action. For instance, it would be interesting to ascertain if the method's performance varies when dealing with actions that primarily affect size, those that affect marker expression, or compounds that significantly diminish cell numbers.

Thank you for encouraging comments!

We have added the following to Results: Relevance scores reveal CytoSummaryNet's preference for large, isolated cells:

“Statistical t-tests were conducted to identify the features that most effectively differentiate mechanisms of action from negative controls in average profiles, focusing on the three mechanisms of action where CytoSummaryNet demonstrates the most significant improvement and the three mechanisms where it shows the least. Consistent with our hypothesis that CytoSummaryNet emphasizes larger, more sparse cells, the important features for the CytoSummaryNet-improved mechanisms of action (Supplementary Material I1) often involve the radial distribution for the mitochondria and RNA channels. These metrics capture the fraction of those stains near the edge of the cell versus concentric rings towards the nucleus, which are more readily detectable in larger cells compared to small, rounded cells.

In contrast, the important features for mechanisms of action not improved by CytoSummaryNet (Supplementary Material I) predominantly include correlation metrics between brightfield and various fluorescent channels, capturing spatial relationships between cellular components. Some of these mechanisms of action included compounds that were not individually distinguishable from negative controls, and CytoSummaryNet did not overcome the lack of phenotype in these cases. This suggests that while CytoSummaryNet excels in identifying certain cellular features, its effectiveness is limited when dealing with mechanisms of action that do not exhibit pronounced phenotypic changes.”

We have also added supplementary material to support (I. Relevant features for CytoSummaryNet improvement).

R3.0 Another test on datasets that are not concerned with chemical compounds, but rather genetic perturbations would greatly increase the reach of the method into the functional genomics community and beyond. This additional analysis could provide valuable insights into the versatility and applicability of the proposed method.

We agree that testing the method's behavior on genetic perturbations would be interesting and could provide insights into its versatility. However, the efficacy of the methodology may vary depending on the specific properties of different genetic perturbation types.

For example, the penetrance of phenotypes may differ between genetic and chemical perturbations. In some experimental setups, a selection agent ensures that nearly all cells receive a genetic perturbation (though not all may express a phenotype due to heterogeneity or varying levels of the target protein). Other experiments may omit such an agent. Additionally, different patterns might be observed in various classes of reagents, such as overexpression, CRISPR-Cas9 knockdown (CRISPRn), CRISPR-interference (CRISPRi), and CRISPR-activation (CRISPRa).

We believe that selecting a single experiment with one of these technologies would not adequately address the question of versatility. Instead, we propose future studies that may conclusively assess the method's performance across a variety of genetic perturbation types. This would provide a more comprehensive understanding of CytoSummaryNet's applicability in functional genomics and beyond. We have update the Discussion section to reflect this:

“Future research could also explore CytoSummaryNet's applicability to genetic perturbations, expanding its utility in functional genomics.”

R3.1. The datasets were stratified based on plates and compounds. It would be beneficial to clarify the basis for data stratification applied for compounds. Was the data sampled based on structural or functional similarity of compounds? If not, what can be expected from the model if trained and validated using structurally or functionally diverse and non-diverse compounds?

Thank you for raising the important question of data stratification based on compound similarity. In our study, the data stratification was performed by randomly sampling the compounds, without considering their structural or functional similarity.

This approach may limit the generalizability of the learned representations to new structural or functional classes not captured in the training set. Consequently, the current methodology may not fully characterize the model’s performance across diverse compound structures.

In future work, it would be valuable to explore the impact of compound diversity on model performance by stratifying data based on structural or functional similarity and comparing the results to our current random stratification approach to more thoroughly characterize the learned representations.

R3.2. Is the method prioritizing a particular biological reaction of cells toward common chemical compounds, such as mitotic failure? Could this be oncology-specific, or is there more utility to it in other datasets?

Our analysis of CytoSummaryNet's performance in (the new) Figure 6 reveals a strong improvement in MoAs targeting cancer-related pathways, such as MEK, HSP, MDM, dehydrogenase, and purine antagonist inhibitors. These MoAs share a common focus on cellular proliferation, survival, and metabolic processes, which are key characteristics of cancer cells.

Given the composition of the cpg0004 dataset, which contains 1,258 unique MoAs with only 28 annotated as oncology-related, the likelihood of randomly selecting five oncology-related MoAs that show strong improvement is extremely low. This suggests that the observed prioritization is not due to chance.

Furthermore, the prioritization cannot be solely attributed to the frequency of oncology-related MoAs in the dataset. Other prevalent disease areas, such as neurology/psychiatry, infectious disease, and cardiology, do not exhibit similar improvements despite having higher MoA counts.

While these findings indicate a potential prioritization of oncology-related MoAs by CytoSummaryNet, further research is necessary to fully understand the extent and implications of this bias. Future work should involve conducting similar analyses across other disease areas and cell types to assess the method's broader utility and identify areas for refinement and application. However, given the speculative nature of these observations, we have chosen not to update the manuscript to discuss this potential bias at this time.

R3.3 Figures 1 and 2 demonstrate that the CytoSummaryNet profiles outperform average-aggregated profiles. However, the average profiling results seem more consistent when compared to CytoSummaryNet profiling. What further conditions or approaches can help improve CytoSummaryNet profiling results to be more consistent?

The observed variability in CytoSummaryNet's performance is primarily due to the intentional technical variance in our datasets, where each plate tested different staining protocol variations. It's important to note that this level of technical variance is not typical in standard cell profiling experiments. In practice, the variance across plates would be much lower. We want to emphasize that while a model capable of generalizing across diverse experimental conditions might seem ideal, it may not be as practically useful in real-world scenarios. This is because such non-uniform conditions are uncommon in typical cell profiling experiments. In normal experimental settings, where technical variance is more controlled, we expect CytoSummaryNet's performance to be more consistent.

R3.4 Can the poor performance on unseen data (in the case of stain 5) be overcome? If yes, how? If no, why not?

We believe that the poor performance on unseen data, such as Stain 5, can be overcome depending on the nature of the unseen data. As shown in Figure 4 (panel 3), the model improves upon average profiling for unseen data when the experimental conditions are similar to the training set.

The issue lies in the different experimental conditions. As explained in our response to R3.3, this could be addressed by including these experimental conditions in the training dataset. As long as CytoSummaryNet is trained (seen) and tested (unseen) on data generated under similar experimental conditions, we are confident that it will improve or perform as well as average profiling.

It's important to note that the issue of generalization to vastly different experimental conditions was considered out of scope for this paper. The main focus is to introduce a new method that improves upon average profiling and can be readily used within a consistent experimental setup.

R3.5 It needs to be mentioned how the feature data used for CytoSummaryNet profiling was normalized before training the model. What would be the impact of feature data normalization before model training? Would the model still outperform if the skewed feature data is normalized using square or log transformation before model training?

We have clarified in the manuscript that we standardized the feature data on a plate-by-plate basis to achieve zero mean and unit variance across all cells per feature within each plate. We have added the following statement to improve clarity:

“The data used to compute the average profiles and train the model were standardized at the plate-level, ensuring that all cell features across the plate had a zero mean and unit variance. The negative control wells were then removed from all plates."

We chose standardization over transformations like squaring or logging to maintain a balanced scale across features while preserving the biological and morphological information inherent in the data. While transformations can reduce skewness and are useful for data spanning several orders of magnitude, they might distort biological relevance by compressing or expanding data ranges in ways that could obscure important cellular variations.

Regarding the potential impact of square or log transformations on skewed feature data, these methods could improve the model's learning efficiency by making the feature distribution more symmetrical. However, the suitability and effectiveness of these techniques would depend on the specific data characteristics and the model architecture.

Although not explored in this study, investigating various normalization techniques could be a valuable direction for future research to assess their impact on the performance and adaptability of CytoSummaryNet across diverse datasets and experimental setups.

R3.6. In Figure 5 b and c, MoAs often seem to be represented by singular compounds and thus, the test (MoA prediction) is very similar to the training (compound ID). Given this context, a discussion about the extent this presents a circular argument supported by stats on the compound library used for training and testing would be beneficial.

Clusters in (the new) Figure 7 that contain only replicates of a single compound would not yield an improved performance on the MoA task unless they also include replicates of other compounds sharing the same MoA in close proximity. Please see our response to R1.1c.iii. for details. To improve visual clarity and avoid misinterpretation, we have recomputed the colors for (the new) Figure 7 and grayed out overlapping points.

R3.7 Can you estimate the minimum amount of supervision (fuzzy/sparse labels, often present in mislabeled compound libraries with dirty compounds and polypharmacology being present) that is needed for it to be efficiently trained?

It's important to note that the metadata used by the model is only based on identifying replicates of the same compound. Mechanism of action (MoA) annotations, which can be erroneous due to dirty compounds, polypharmacology, and incomplete information, are not used in training at all. MoA annotations are only used in our evaluation, specifically for calculating the mAP for MoA retrieval.

We have successfully trained CytoSummaryNet on 72 unique compounds with 4 replicates each. This is the current empirical minimum, but it is possible that the model could be trained effectively with even fewer compounds or replicates.

Determining the absolute minimum amount of supervision required for efficient training would require further experimentation and analysis. Factors such as data quality, feature dimensionality, and model complexity could influence the required level of supervision.

R3.minor1 Figure 5: The x-axis and y-axis tick values are too small, and image resolution/size needs to be increased.

We have made the following changes to address the concerns:

• Increased the image resolution and size to improve clarity and readability.
• Removed the x-axis and y-axis tick values, as they do not provide meaningful information in the context of UMAP visualizations. We believe these modifications enhance the visual presentation of the data and make it easier for readers to interpret the results.

R3.minor2 The methods applied to optimize hyperparameters in supplementary data need to be included.

We added the following to Supplementary Material D:

“We used the Weights & Biases (WandB) sweep suite in combination with the BOHB (Bayesian Optimization and HyperBand) algorithm for hyperparameter sweeps. The BOHB algorithm [47] combines Bayesian optimization with bandit-based strategies to efficiently find optimal hyperparameters.

Additionally Table D1 provides an overview of all tunable hyperparameters and their chosen values based on a BOHB hyperparameter optimization.”

R3.minor3 Figure 5(c, d): The names of compound 2 and Compound 5 need to be included in the labels.

These compounds were obtained from external companies and are proprietary, necessitating their anonymization in our study. This has now been added in the caption of (the new) Figure 7:

“Note that Compound2 and Compound5 are intentionally anonymized.”

R3.minor4 Table C1: Plate descriptions need to be included.

*Table C1: The training, validation, and test set stratification for Stain2, Stain3, Stain4, and Stain5. Five training, four validation, and three test plates are used for Stain2, Stain3, and Stain4. Stain5 contains six test set plates only. *

__Stain2 __

Stain3

Stain4

Stain5

Training plates

Test plates

BR00113818

BR00115128

BR00116627

BR00120532

BR00113820

BR00115125highexp

BR00116631

BR00120270

BR00112202

BR00115133highexp

BR00116625

BR00120536

BR00112197binned

BR00115131

BR00116630highexp

BR00120530

BR00112198

BR00115134

200922_015124-Vhighexp

BR00120526

Validation plates

BR00120274

BR00112197standard

BR00115129

BR00116628highexp

BR00112197repeat

BR00115133

BR00116629highexp

BR00112204

BR00115128highexp

BR00116627highexp

BR00112201

BR00115127

BR00116629

Test plates

BR00112199

BR00115134bin1

200922_044247-Vbin1

BR00113819

BR00115134multiplane

200922_015124-V

BR00113821

BR00115126highexp

BR00116633bin1

We have added a reference to the metadata file in the description of Table C1: https://github.com/carpenter-singh-lab/2023_Cimini_NatureProtocols/blob/main/JUMPExperimentMasterTable.csv

R3.minor5 Figure F1: Does the green box (stain 3) also involve training on plates from stain 4 (BR00116630highexp) and 5 (BR00120530) mentioned in Table C1? Please check the figure once again for possible errors.

We have carefully re-examined Figure F1 and Table C1 to ensure their accuracy and consistency. Upon double-checking, we can confirm that the figure is indeed correct. We intentionally omitted the training and validation plates from Figure F1 to maintain clarity and readability, as including them resulted in a cluttered and difficult-to-interpret figure.

Regarding the specific plates mentioned:

• BR00116630highexp (Stain4) is used for training, as correctly stated in Table C1. This plate is considered an outlier within the Stain4 dataset and happens to cluster with the Stain3 plates in Figure F1.
• BR00120530 (Stain5) is part of the test set only and correctly falls within the Stain5 cluster in Figure F1. To improve the clarity of the training, validation, and test split in Table C1, we have added a color scheme that visually distinguishes the different data subsets. This should make it easier for readers to understand the distribution of plates across the various splits.

Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

Learn more at Review Commons

---

Referee #3

Evidence, reproducibility and clarity

Summary:

In the manuscript by Van Dijk et al., a novel deep learning technique is introduced that aims to summarize informative cells from heterogeneous populations in image-based profiling. This technique is based on a network that utilizes contrastive learning with a multiple-instance learning framework, a significant departure from existing average-based cell profiling models.

The authors have done a commendable job discussing the method, demonstrating its potential to outperform current models in profiling cell-based features. The work is of considerable significance and interest to a wide field of researchers working on the understanding of cell heterogeneity's impact on various biological phenomena and practical studies in pharmacology.

One aspect that would further enhance the value of this work is an exploration of the method's separation power across different modes of action. For instance, it would be interesting to ascertain if the method's performance varies when dealing with actions that primarily affect size, those that affect marker expression, or compounds that significantly diminish cell numbers. Another test on datasets that are not concerned with chemical compounds, but rather genetic perturbations would greatly increase the reach of the method into the functional genomics community and beyond. This additional analysis could provide valuable insights into the versatility and applicability of the proposed method. Please find my detailed comments below:

Major Comments:

1. The datasets were stratified based on plates and compounds. It would be beneficial to clarify the basis for data stratification applied for compounds. Was the data sampled based on structural or functional similarity of compounds? If not, what can be expected from the model if trained and validated using structurally or functionally diverse and non-diverse compounds?
2. Is the method prioritizing a particular biological reaction of cells toward common chemical compounds, such as mitotic failure? Could this be oncology-specific, or is there more utility to it in other datasets?
3. Figures 1 and 2 demonstrate that the CytoSummaryNet profiles outperform average-aggregated profiles. However, the average profiling results seem more consistent when compared to CytoSummaryNet profiling. What further conditions or approaches can help improve CytoSummaryNet profiling results to be more consistent?
4. Can the poor performance on unseen data (in the case of stain 5) be overcome? If yes, how? If no, why not?
5. It needs to be mentioned how the feature data used for CytoSummaryNet profiling was normalized before training the model. What would be the impact of feature data normalization before model training? Would the model still outperform if the skewed feature data is normalized using square or log transformation before model training?
6. In Figure 5 b and c, MoAs often seem to be represented by singular compounds and thus, the test (MoA prediction) is very similar to the training (compound ID). Given this context, a discussion about the extent this presents a circular argument supported by stats on the compound library used for training and testing would be beneficial.
7. Can you estimate the minimum amount of supervision (fuzzy/sparse labels, often present in mislabeled compound libraries with dirty compounds and polypharmacology being present) that is needed for it to be efficiently trained?

Minor Comments:

1. Figure 5: The x-axis and y-axis tick values are too small, and image resolution/size needs to be increased.
2. The methods applied to optimize hyperparameters in supplementary data need to be included.
3. Figure 5(c, d): The names of compound 2 and Compound 5 need to be included in the labels.
4. Table C1: Plate descriptions need to be included.
5. Figure F1: Does the green box (stain 3) also involve training on plates from stain 4 (BR00116630highexp) and 5 (BR00120530) mentioned in Table C1? Please check the figure once again for possible errors.

Significance

This work presents a significant move forward in the ways we deal with cellular heterogeneity in all single-cell assays. Though the model in its current state has trouble extrapolating to out of distribution data, I am confident that it provides a considerable step forward in the process of extracting "informative" knowledge from data in the form of optimized profiles.

The optimization is yet based on optimizing a similarity metric for group assignments, I will be interesting to see if other objectives could be more effective in developing aggregation techniques.

The work is of considerable significance and interest to a wide field of researchers working on the understanding of cell heterogeneity's impact on various biological phenomena and practical studies in pharmacology.

Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

Learn more at Review Commons

---

Referee #2

Evidence, reproducibility and clarity

The authors present a well-developed and useful algorithm. The technical motivation and validation are very carefully and clearly explained, and their work is potentially useful to a varied audience.

That said, I think the authors could do a better job, especially in the figures, of putting the algorithm in context for an audience that is unfamiliar with the cell painting assay. For example, a figure towards the beginning of the paper with example images might help to set the stage. Similarly a schematic of the algorithm earlier in the paper would provide a graphical overview. For the sake of a biologically inclined audience, I would consider labeling the images in the caption by cell type and label.

The interpretability results were intriguing. The authors might consider further validating these interpretations by removing weakly informative cells from the dataset and retraining. Are the cells so uninformative that the algorithm does better without them, or are they just less informative than other cells?

As far as I can tell, the authors only oblique state whether the code associated with the manuscript is openly available. Posting the code is needed for reproducibility. I would provide not only a github, but a doi linked to the code, or some other permanent link.

Significance

Incorporating biological heterogeneity into machine-learning driven problems is a critical research question. Replacing means/modes and such with a machine learning framework, the authors have identified a problem with potentially wide significance. The application to cell painting and related assays is of broad enough significance for many journals, However, the authors could further broaden the significance by commenting on other possible cell biology applications. What other applications might the algorithm be particularly suited for? Are there any possible roadblocks to wider use. What sorts of data has the code been tested on so far?

Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

Learn more at Review Commons

---

Referee #1

Evidence, reproducibility and clarity

Summary:

Provide a short summary of the findings and key conclusions (including methodology and model system(s) where appropriate).

Cell (non-genetic) heterogeneity is an important concept in cell biology, but there are currently only a few studies that try to incorporate this information to represent cell populations in the field of high-content image-based phenotypic profiling. The authors present CytoSummaryNet, a machine learning approach for representing heterogeneous cell populations, and apply it to a high-content image-based Cell Painting dataset to demonstrate superior performance in predicting a compound's mechanism of action (MoA), in relation to the average profile representation. CytoSummaryNet relies on Cell Profiler morphological features and simultaneous optimization of two components, both novel in the cell profiling field: (i) learning representations using weakly supervised contrastive learning according to the perturbation identifications (i.e., the compound), (ii) using a representation method called Deep Sets to create permutation-invariant population representations. The authors evaluate their representation on the task of replicate retrieval and of MoA retrieval using the public dataset cpg0001 (and cpg0004), and report superior performance in respect to the average-aggregated profiles for the experimental protocols and compounds seen on training (that do not generalize to out-of-distribution compounds + experimental protocols). By interpreting which cells were most important for the MoA model predictions, the authors propose that their representation prioritizes large uncrowded cells.

Major comments:

The strength of the manuscript is the new idea of combining contrastive learning and sets representations for better representation of heterogeneous cell populations. However, we are not convinced that the conclusion that this representation improves MoA prediction is fully supported by the data, for several reasons.

1. Evaluations. This is the most critical point in our review.

a. CytoSummaryNet is evaluated in comparison to aggregate-average profiling, although previous work has already reported representations that capture heterogeneity and self-supervision independently. To argue that both components of contrastive learning and sets representations are contributing to MoA prediction we believe that a separate evaluation for each component is required. Specifically, the authors can benchmark their previous work to directly evaluate a simpler population representation (PMID: 31064985, ref #13) - we are aware that the authors report a 20% improvement, but this was reported on a separate dataset. The authors can also compare to contrastive learning-based representations that rely on the aggregate (average) profile to assess and quantify the contribution of the sets representation.

b. The evaluation metric of mAP improvement in percentage is misleading, because a tiny improvement for a MoA prediction can lead to huge improvement in percentage, while a much larger improvement in MoA prediction can lead to a small improvement in percentage. For example, in Fig. 4, MEK inhibitor mAP improvement of ~0.35 is measured as ~50% improvement, while a much smaller mAP improvement can have the same effect near the origins (i.e., very poor MoA prediction). (Subjective) visual assessment of this figure does not show a convincing contribution of CytoSummaryNet representations of the average profiling on the test set (3.33 uM). This issue might also be relevant for the task of replicate retrieval. All in all, the mAP improvement reported in Table 1 and throughout the manuscript (including the Abstract), is not a proper evaluation metric for CytoSummaryNet contribution. We suggest reporting the following evaluations:

i. Visualizing the results of cpg0001 (Figs. 1-3) similarly to cpg0004 (Fig. 4), i.e., plotting the matched mAP for CytoSummaryNet vs. average profile. ii. In Table 1, we suggest referring to the change in the number of predictable MoAs (MoAs that pass a mAP threshold) rather than the improvement in percentages. Another option is showing a graph of the predictability, with the X axis representing a threshold and Y-axis showing the number of MoAs passing it. For example see (PMID: 36344834, Fig. 2B) and (PMID: 37031208, Fig. 2A), both papers included contributions from the corresponding author of this manuscript.

c. Additional evaluation-related concerns were: i. "a subset of 18 compounds were designated as validation compounds" - 5 cross-validations of 18 compounds can make the evaluation complete. This can also enhance statistical power in figures 1-3.

ii. Clarify if the MoA results for cpg0001 are drawn from compounds from both the training and the validation datasets. If so, describe how the results differ between the sets in text and graphs.

iii. "Mechanism of action retrieval is evaluated by quantifying a profile's ability to retrieve the profile of other compounds with the same annotated mechanism of action.". It was unclear to us if the evaluation of mAP for MoA identification can include finding replicates of the same compound. That is, whether finding a close replicate of the same compound would be included in the AP calculation. This would provide CytoSummaryNet with an inherent advantage as this is the task it is trained to do. We assume that this was not the case (and thus should be more clearly articulated), but if it was - results need to be re-evaluated excluding same-compound replicates. 2. Lack of clarity in the description of the data and evaluation. While the concept of constructive learning + sets representation is elegant and intuitive, we found it very hard to follow the technical aspects of data and performance evaluation, even after digging in deep into the Methods. Figuring out these important aspects required us for vast investment in time, more than the vast majority of manuscripts we reviewed in the last couple of years. It is highly recommended that the authors provide more details to make this manuscript easier to follow. Some examples include:

a. The description of Stain2-5 was not clear for us at first (and second) read. The information is there, but more details will greatly enhance the reader's ability to follow. One suggestion is explicitly stating that these "stains" partitioning was already defined in ref 26. Another suggestion is laying out explicitly a concrete example on the differences between two of these stains. We believe highlighting the differences between stains will strengthen the claim of the paper, emphasizing the difficulty of generalizing to the out-of-distribution stain.

b. What does each data point in Figures 1-3 represent? Is it the average mAP for the 18 validation compounds, using different seeds for model training? Why not visualize the data similarly to Fig. 4 so the improvement per compound can be clearly seen?

c. Justification and interpretation of the evaluation metrics.

d. Explicitly mentioning the number of MoAs for each datasets and statistics of number of compounds per MoA (e.g., average\median, min, max).

e. The data split in general is not easily understood. Figure 8 is somewhat helpful, however in our view, it can be improved to enhance understanding of the different splits. Specifically, the training and validation compounds need to be embedded and highlighted within the figure. 3. Lack of justification of design choices. There were multiple design choices that were not justified. This adds to the lack of clarity and makes it harder to evaluate the merits of the new method. For example:

a. Why was stain 5 used for the test, rather than the other stains?

b. How were the 18 validation compounds selected?

c. For cpg0004, no justification for the specific doses selected (10uM - train, 3.33 uM - test) for the analysis in Figure 4. Why was the data split for the two dosages? For example, why not perform 5-fold cross validation on the compounds (e.g., of the highest dose)?

d. A more detailed explanation on the logic behind using a training stain to test MoA retrieval will help readers appreciate these results. In our first read of this manuscript we did not grasp that, we did in a second read, but spoon-feeding your readers will help. 4. The interpretability analysis is speculative. Assessment of interpretability is always tricky. But in this case, the authors can directly confirm their interpretation that the CytoSummaryNet representation prioritizes large uncrowded cells, by explicitly selecting these cells, and using their average profile representation to demonstrate that they achieve improved results. If this works, it could be applied as a general outlier removal strategy for cell profiling.

a. "We identified the likely mechanism by which the learned CytoSummaryNet aggregates cells: the most salient cells are generally larger and more isolated from other cells, while the least salient cells appear to be smaller and more crowded, and tend to contain spots of high-intensity pixels (whether dying, debris or in some stage of cell division)." - doesn't such a mechanism should generalize to out-of-distribution data? 5. Placing this work in context of other weakly supervised representations. Previous papers used weakly supervised labels of proteins / experimental perturbations (e.g., compounds) to improve image-derived representations, but were not discussed in this context. These include PMID: 35879608, https://www.biorxiv.org/content/10.1101/2022.08.12.503783v2 (from the same research groups and can also be benchmarked in this context),https://pubs.rsc.org/en/content/articlelanding/2023/dd/d3dd00060e , and https://www.biorxiv.org/content/10.1101/2023.02.24.529975v1. We believe that a discussion explicitly referencing these papers in this specific context is important.

Minor comments:

In our opinion, evaluation of the training task using the training data (Figure 1) is not contributing to the manuscript and could be excluded. Also we feel that the subjectiveness of the UMAP analysis (Figure 5) is not contributing much and could be excluded, especially if the authors follow our suggestions regarding quantification. Of course, this is up to the authors to decide (along with most of the other suggestions below).

Suggested clarifications:

1. "Because the improved results could stem from prioritizing certain features over others during aggregation, we investigated each cell's importance during CytoSummaryNet aggregation by calculating a relevance score for each" - what is the relevance score? Would be helpful to provide some intuition in the Results.
2. Figure 1:

a. Colors of the two methods too similar

b. The dots are too close. It will be more easily interpreted if they were further apart.

c. What do the dots stand for?

d. We recommend considering moving this figure to the supp. material (the most important part of it is the results on the test set and it appears in Fig.2). 3. Figure 4: It is somewhat misleading to look at the training MoAs and validation MoAs embedded together in the same graph. We recommend showing only the test MoAs (train MoAs can move to SI). 4. Figure 5

a. Why only Stain3? What happens if we look at Stains 2,3 and 4 together? Stain 5?

b. Should validation compounds and training compounds be analyzed separately?

c. Subfigure (d): it is expected that the data will be classified by compound labels as it is the training task, but for this to be persuasive I would like to see this separately on the training compounds first and then and more importantly on the validation compounds.

d. For subfigures (b) and (d): it appears there are not enough colors for d, which makes it partially not understandable. For example, the pink label in (d) shows a single compound which appears to represent two different MoAs. This is probably not the case, and it has two different compounds, but it cannot be inferred when they are represented by the same color.

e. For the Subfigure (e) - only 1 circle looks justified (in the top left). And for that one, is it not a case of an outlier plate that would perhaps need to be removed from analysis? Is it not good that such a plate will be identified? 5. Discussion:

a. "perhaps in part due to its correction of batch effects" - is this statement based on Fig. 5F - we are not convinced.

b. "Overall, these results improve upon the ~20% gains we previously observed using covariance features" - this is not the same dataset so it is hard to reach conclusions - perhaps compare performance directly on the same data?

Significance

Cell profiling is an emerging field with many applications in academia and industry. Finding better representations for heterogeneous cell populations is important and timely. However, unless convinced otherwise after a rebuttal/revision, the contribution of this paper, in our opinion, is mostly conceptual, but in its current form - not yet practical. This manuscript combined two concepts that were previously reported in the context of cell profiling, weakly supervised representations. Our expertise is in computational biology, and specifically applications of machine learning in microscopy.

eLife assessment

In this important study, the authors found, with the use of statistical methods, that compound heterozygous rare deletion variants affecting the kinase-domain of non-receptor tyrosine kinase TNK/ACK1 are associated with human systemic lupus erythematosus (SLE). The authors use a convincing mouse experimental model and human-induced pluripotent stem cell (hiPSC)-derived macrophages to clarify cause-effect relationships and the cellular basis of nephritis. With the identification of new SLE-related genes, this manuscript improves our understanding of human SLE pathogenesis.

Reviewer #1 (Public Review):

The authors report compound heterozygous deleterious variants in the kinase domains of the non-receptor tyrosine kinases (NRTK) TNK2/ACK1 in familial SLE. They suggest that ACK1 and BRK deficiencies are associated with human SLE and impair efferocytosis.

The experiments in this revision showing that a weekly injection of ACK1 or BRK inhibitors induced various kinds of lupus-related autoantibodies in BALB/c supported the pivotal role of ACK1/BRK in systemic autoimmunity, although treated mice failed to demonstrate the full picture of lupus.

Reviewer #2 (Public Review):

In this manuscript, the authors revealed that genetic deficiencies of ACK1 and BRK are associated with human SLE. First, the authors found that compound heterozygous deleterious variants in the kinase domains of the non-receptor tyrosine kinases (NRTK) TNK2/ACK1 in one multiplex family and PTK6/BRK in another family. Then, by an experimental blockade of ACK1 or BRK in a mouse SLE model, they found an increase in glomerular IgG deposits and circulating autoantibodies. Furthermore, they reported that ACK and BRK variants from the SLE patients impaired the MERTK-mediated anti-inflammatory response to apoptotic cells in human induced pluripotent stem cells (hiPSC)-derived macrophages. This work identified new SLE-associated ACK and BRK variants and a role for the NRTK TNK2/ACK1 and PTK6/BRK in efferocytosis, providing a new molecular and cellular mechanism of SLE pathogenesis.

eLife assessment

This fundamental study provides a modeling regime that provides new insight into the energy-preservation parameters among schooling fish. The strength of the evidence supporting observations such as distilled dynamics between leading and lagging schooling fish which are derived from emergent properties is compelling. Overall, the study provides exciting insights into energetic coupling with respect to group swimming dynamics.

Reviewer #1 (Public Review):

Summary:<br /> The study seeks to establish accurate computational models to explore the role of hydrodynamic interactions on energy savings and spatial patterns in fish schools. Specifically, the authors consider a system of (one degree-of-freedom) flapping airfoils that passively position themselves with respect to the streamwise direction, while oscillating at the same frequency and amplitude, with a given phase lag and at a constant cross-stream distance. By parametrically varying the phase lag and the cross-stream distance, they systematically explore the stability and energy costs of emergent configurations. Computational findings are leveraged to distill insights into universal relationships and clarify the role of the wake of the leading foil.

Strengths:<br /> (1) The use of multiple computational models (computational fluid dynamics, CFD, for full Navier-Stokes equations and computationally-efficient inviscid vortex sheet, VS, model) offers an extra degree of reliability of the observed findings and backing to the use of simplified models for future research in more complex settings.

(2) The systematic assessment of the stability and energy savings in multiple configurations of pairs and larger ensembles of flapping foils is an important addition to the literature.

(3) The discovery of a linear phase-distance relationship in the formation attained by pairs of flapping foils is a significant contribution, which helps compare different experimental observations in the literature.

(4) The observation of a critical size effect for in-line formations, above which cohesion and energetic benefits are lost at once, is a new discovery to the field.

Weaknesses:<br /> (1) The extent to which observations on one-degree-of-freedom flapping foils could translate to real fish schools is presently unclear, so that some of the conclusions on live fish schools are likely to be overstated and would benefit from some more biological framing.

(2) The analysis of non-reciprocal coupling is not as novel as the rest of the study and potentially not as convincing due to the chosen linear metric of interaction (that is, the flow agreement).

Overall, this is a rigorous effort on a critical topic: findings of the research can offer important insight into the hydrodynamics of fish schooling, stimulating interdisciplinary research at the interface of computational fluid mechanics and biology.

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

The study seeks to establish accurate computational models to explore the role of hydrodynamic interactions on energy savings and spatial patterns in fish schools. Specifically, the authors consider a system of (one degree-of-freedom) flapping airfoils that passively position themselves with respect to the streamwise direction, while oscillating at the same frequency and amplitude, with a given phase lag and at a constant cross-stream distance. By parametrically varying the phase lag and the cross-stream distance, they systematically explore the stability and energy costs of emergent configurations. Computational findings are leveraged to distill insights into universal relationships and clarify the role of the wake of the leading foil.

We would like to thank the referee for their careful read of the manuscript and for their constructive feedback. We appreciate it.

Strengths:

(1) The use of multiple computational models (computational fluid dynamics, CFD, for full Navier-Stokes equations and computationally efficient inviscid vortex sheet, VS, model) offers an extra degree of reliability of the observed findings and backing to the use of simplified models for future research in more complex settings.

(2) The systematic assessment of the stability and energy savings in multiple configurations of pairs and larger ensembles of flapping foils is an important addition to the literature.

(3) The discovery of a linear phase-distance relationship in the formation attained by pairs of flapping foils is a significant contribution, which helps compare different experimental observations in the literature.

(4) The observation of a critical size effect for in-line formations of larger, above which cohesion and energetic benefits are lost at once, is a new discovery in the field.

Thank you for this list of strength – we are delighted that these ideas were clearly communicated in our manuscript.

Note that Newbolt et al. PNAS, 2019 reported distance as a function of phase for pairs of flapping hydrofoils, and Li et al, Nat. Comm., 2020 also reported phase-distance relationship in robotic and biological fish (calling it Vortex Phase Matching). We compiled their results, together with our and other numerical and experimental results, showing that the linear distance-phase relationship is universal.

Weaknesses:

(1) The extent to which observations on one-degree-of-freedom flapping foils could translate to real fish schools is presently unclear so some of the conclusions on live fish schools are likely to be overstated and would benefit from some more biological framing.

Thank you for bringing up this point. Indeed, flapping foils that are free to translate in both the x- and y-directions and rotate in the x-y plane could drift apart in the y-direction. However, this drift occurs at a longer time scale than the forward swimming motion; it is much slower. For this reason, we feel justified to ignore it for the purpose of this study, especially that the pairwise equilibria in the swimming x-direction are reached at a faster time scale.

Below, we include two snapshots taken from published work from the group of Petros Koumoutsakos (Gazzola et al, SIAM 2014). The figures show, respectively, a pair and a group of five undulating swimmers, free to move and rotate in the x-y plane. The evolution of the two and five swimmers is computed in the absence of any control. The lateral drift is clearly sub-dominant to the forward motion. Similar results were reported in Verma et al, PNAS 2018.

These results are independent on the details of the flow interactions model. For example, similar lateral drift is observed using the dipole model dipole model (Kanso & Tsang, FDR 2014, Tsang & Kanso, JNLS 2023).

Another reason why we feel justified to ignore these additional degrees of freedom is the following: we assume a live fish or robotic vehicle would have feedback control mechanisms that correct for such drift. Given that it is a slowly-growing drift, we hypothesize that the organism or robot would have sufficient time to respond and correct its course.

Indeed, in Zhu et al. 2022, an RL controller, which drives an individual fish-like swimmer to swim at a given speed and direction, when applied to pairs of swimmers, resulted in the pair "passively" forming a stable school without any additional information about each other.

We edited the main manuscript in page 4 of the manuscript to include reference to the work cited here and to explain the reasons for ignoring the lateral drift.

Citations:  

Gazzola, M., Hejazialhosseini, B., & Koumoutsakos, P. (2014). Reinforcement learning and wavelet adapted vortex methods for simulations of self-propelled swimmers. SIAM Journal on Scientific Computing, 36(3), B622-B639. DOI: https://doi.org/10.1137/130943078

Verma, S., Novati, G., & Koumoutsakos, P. (2018). Efficient collective swimming by harnessing vortices through deep reinforcement learning. Proceedings of the National Academy of Sciences, 115(23), 5849-5854. DOI: https://doi.org/10.1073/pnas.1800923115

Tsang, A. C. H. & Kanso, E., (2013). Dipole Interactions in Doubly Periodic Domains. Journal of Nonlinear Science 23 (2013): 971-991. DOI: https://doi.org/10.1007/s00332-013-9174-5

Kanso, E., & Tsang, A. C. H. (2014). Dipole models of self-propelled bodies. Fluid Dynamics Research, 46(6), 061407. DOI: https://doi.org/10.1088/0169-5983/46/6/061407

Zhu, Y., Pang, J. H., & Tian, F. B. (2022). Stable schooling formations emerge from the combined effect of the active control and passive self-organization. Fluids, 7(1), 41. DOI: https://doi.org/10.3390/fluids7010041

Author response image 1.

Antiphase self-propelled anguilliform swimmers. (a) – (d) Wavelet adapted vorticity fields at, respectively, t = T, t = 4T, t = 10T. (e) Absolute normalized velocities |U|/L. (f) Swimmers’ centre of mass trajectories.

Author response image 2.

Parallel schooling formation. (a) – (d) wavelet adapted vorticity fields at, respectively, t = T, t = 4T, t = 7T, t = 10T. (e) Absolute normalized velocities |U|/L. (f) Swimmers’ center of mass trajectories.

(2) The analysis of non-reciprocal coupling is not as novel as the rest of the study and potentially not as convincing due to the chosen linear metric of interaction (that is, the flow agreement).

We thank the referee for this candid and constructive feedback. In fact, we view this aspect of the study as most “revolutionary” because it provides a novel approach to pre-computing the locations of stable equilibria even without doing expensive all-to-all coupled simulations or experiments.

Basically, the idea is the following: you give me a flow field, it doesn’t matter how you obtained it, whether from simulations or experimentally, and I can tell you at what locations in this flow field a virtual flapping swimmer would be stable and save hydrodynamic energy!

In the revised version, we changed page 3 and 7 in main text, and added a new section “Diagnostic tools” in SI to better illustrate this.

Overall, this is a rigorous effort on a critical topic: findings of the research can offer important insight into the hydrodynamics of fish schooling, stimulating interdisciplinary research at the interface of computational fluid mechanics and biology.

We thank the referee again for their careful read of the manuscript and their constructive feedback.

Reviewer #2 (Public Review):

The document "Mapping spatial patterns to energetic benefits in groups of flow-coupled swimmers" by Heydari et al. uses several types of simulations and models to address aspects of stability of position and power consumption in few-body groups of pitching foils. I think the work has the potential to be a valuable and timely contribution to an important subject area. The supporting evidence is largely quite convincing, though some details could raise questions, and there is room for improvement in the presentation. My recommendations are focused on clarifying the presentation and perhaps spurring the authors to assess additional aspects:

We would like to thank the referee for their careful read of the manuscript and for their constructive feedback. We appreciate it.

(1) Why do the authors choose to set the swimmers free only in the propulsion direction? I can understand constraining all the positions/orientations for investigating the resulting forces and power, and I can also understand the value of allowing the bodies to be fully free in x, y, and their orientation angle to see if possible configurations spontaneously emerge from the flow interactions. But why constrain some degrees of freedom and not others? What's the motivation, and what's the relevance to animals, which are fully free?

We would like to thank the referee for raising this point. It is similar to the point raised above by the first referee. As explained above the reason is the following: in freely-swimming, hydrodynamically-interacting “fish,” the lateral drift is sub-dominant to the forward swimming motion. Therefore, we ignore it in the model. Please see our detailed response above for further clarification, and see changes in page 4 in the main manuscript.

(2) The model description in Eq. (1) and the surrounding text is confusing. Aren't the authors computing forces via CFD or the VS method and then simply driving the propulsive dynamics according to the net horizontal force? It seems then irrelevant to decompose things into thrust and drag, and it seems irrelevant to claim that the thrust comes from pressure and the drag from viscous effects. The latter claim may in fact be incorrect since the body has a shape and the normal and tangential components of the surface stress along the body may be complex.

Thank you for pointing this out! It is indeed confusing.

In the CFD simulations, we are computing the net force in the swimming x-direction direction by integrating using the definition of force density in relation to the stress tensor. There is no ambiguity here.

In the VS simulations, however, we are computing the net force in the swimming x-direction by integrating the pressure jump across a plate of zero thickness. There is no viscous drag. Viscous drag is added by hand, so-to-speak. This method for adding viscous drag in the context of the VS model is not new, it has been used before in the literature as explained in the SI section “Vortex sheet (VS) model” (pages 30 and 31).

.

(3) The parameter taudiss in the VS simulations takes on unusual values such as 2.45T, making it seem like this value is somehow very special, and perhaps 2.44 or 2.46 would lead to significantly different results. If the value is special, the authors should discuss and assess it. Otherwise, I recommend picking a round value, like 2 or 3, which would avoid distraction.

Response: The choice of dissipation time is both to model viscous effect and reduce computational complexity. Introducing it is indeed introduces forcing to the simulation. Round value, like 2 or 3, is equal to an integer multiple of the flapping period, which is normalized to T=1, Therefore, an integer value of  would cause forcing at the resonant frequency and lead to computational blow up. To avoid this effect, a parameter choice of  = 2.45, 2.44 or 2.46 would be fine and would lead to small perturbation to the overall simulation, compared to no dissipation at all. This effect is studied in detail in the following published work from our group:

Huang, Y., Ristroph, L., Luhar, M., & Kanso, E. (2018). Bistability in the rotational motion of rigid and flexible flyers. Journal of Fluid Mechanics, 849, 1043-1067. DOI: https://doi.org/10.1017/jfm.2018.446

(4) Some of the COT plots/information were difficult to interpret because the correspondence of beneficial with the mathematical sign was changing. For example, DeltaCOT as introduced on p. 5 is such that negative indicates bad energetics as compared to a solo swimmer. But elsewhere, lower or more negative COT is good in terms of savings. Given the many plots, large amounts of data, and many quantities being assessed, the paper needs a highly uniform presentation to aid the reader.

Thank you for pointing this out! We updated Figures 3,6 as suggested.

(5) I didn't understand the value of the "flow agreement parameter," and I didn't understand the authors' interpretation of its significance. Firstly, it would help if this and all other quantities were given explicit definitions as complete equations (including normalization). As I understand it, the quantity indicates the match of the flow velocity at some location with the flapping velocity of a "ghost swimmer" at that location. This does not seem to be exactly relevant to the equilibrium locations. In particular, if the match were perfect, then the swimmer would generate no relative flow and thus no thrust, meaning such a location could not be an equilibrium. So, some degree of mismatch seems necessary. I believe such a mismatch is indeed present, but the plots such as those in Figure 4 may disguise the effect. The color bar is saturated to the point of essentially being three tones (blue, white, red), so we cannot see that the observed equilibria are likely between the max and min values of this parameter.

Thank you for pointing this out! You are correct in your understanding of the flow agreement parameter, but not in your interpretation.

Basically, “if the match were perfect, then the swimmer would generate no relative flow and thus no thrust,” means that “such a location could not be is an equilibrium.” Let me elaborate. An equilibrium is one at which the net thrust force is zero. The equilibrium is stable if the slope of the thrust force is negative. Ideally, this is what maximizing the flow agreement parameter would produce.

For example, consider an ideal fluid where the flow velocity is form  in vertical direction. Consider a “ghost swimmer” heaving at a velocity  . Under this scenario, flow agreement and thrust parameters are

Let’s now consider a balance of forces on the “ghost swimmer.” The ghost swimmer is in relative equilibrium if and only if:

It gives us

We then consider stability at this equilibrium by calculating the derivative of thrust parameter over phase

The corresponding values at equilibria are

Thus, when taking the positive which means the equilibria is a stable fixed point. We included this analysis in a new section in the SI page 32.

(6) More generally, and related to the above, I am favorable towards the authors' attempts to find approximate flow metrics that could be used to predict the equilibrium positions and their stability, but I think the reasoning needs to be more solid. It seems the authors are seeking a parameter that can indicate equilibrium and another that can indicate stability. Can they clearly lay out the motivation behind any proposed metrics, and clearly present complete equations for their definitions? Further, is there a related power metric that can be appropriately defined and which proves to be useful?

Thank you – these are excellent suggestions. Indeed, we needed to better explain the motivation and equations. Perhaps the main idea for these metrics can be best understood when explained in the context of the simpler particle model, which we now do in the SI and explain the main text.

(7) Why do the authors not carry out CFD simulations on the larger groups? Some explanations should be given, or some corresponding CFD simulations should be carried out. It would be interesting if CFD simulations were done and included, especially for the in-line case of many swimmers. This is because the results seem to be quite nuanced and dependent on many-body effects beyond nearest-neighbor interactions. It would certainly be comforting to see something similar happen in CFD.

We are using a open-source version of the Immersed Boundary Method that is not specifically optimized for many interacting swimmers. Therefore, the computational cost of performing CFD simulations for more swimmers is high. Therefore, we used the CFD simulations sporadically with fewer simmers (2 or 3) and we performed systematic simulations in the context of the VS model.

For the same Reynolds number in Figure 1, we simulated three and four swimmers in CFD: three swimmers forms a stable formation, four swimmers don’t, consistent with the VS model, with the forth swimmer colliding with the third one. Results are included in the SI figure 8 of the main text.

(8) Related to the above, the authors should discuss seemingly significant differences in their results for long in-line formations as compared to the CFD work of Peng et al. [48]. That work showed apparently stable groups for numbers of swimmers quite larger than that studied here. Why such a qualitatively different result, and how should we interpret these differences regarding the more general issue of the stability of tandem groups?

Thank you for bringing up this important comparison. Peng et al. [48] (Hydrodynamic schooling of multiple self-propelled flapping plates) studied inline configuration of flapping airfoils at Reynolds number =200. There are several differences between their work and ours. The most important one is that they used a flexible plate, which makes the swimmer more adaptive to changes in the flow field, e.g. changes in tailbeat amplitude and changes in phase along its body and diverts some of the hydrodynamic energy to elastic energy. We edited the main text page 10 at the end of section “Critical size of inline formations beyond which cohesion is lost” to explain this distinction.

(9) The authors seem to have all the tools needed to address the general question about how dynamically stable configurations relate to those that are energetically optimal. Are stable solutions optimal, or not? This would seem to have very important implications for animal groups, and the work addresses closely related topics but seems to miss the opportunity to give a definitive answer to this big question.

Indeed, that is exactly the point – in pairwise formations, stable configurations are also energetically optimal! In larger groups, there is no unique stable configuration – each stable configuration is associated with a different degree of energy savings. Interestingly, when exploring various equilibrium configurations in a school of four, we found the diamond formation of D. Weihs, Nature, 1972 to be both stable and most optimal among the configurations we tested. However, claiming this as a global optimum may be misleading – our standpoint is that fish schools are always dynamic and that there are opportunities for energy savings in more than one stable configuration.

We added a section in new text “Mapping emergent spatial patterns to energetic benefits”, and added a new figure in the maintext (Fig. 10) and a new figure in the SI (Fig. S. 8)

(10) Time-delay particle model: This model seems to construct a simplified wake flow. But does the constructed flow satisfy basic properties that we demand of any flow, such as being divergence-free? If not, then the formulation may be troublesome.

The simplified wake flow captures the hydrodynamic trail left by the swimmer in a very simplified manner. In the limit of small amplitude, it should be consistent with the inviscid vortex sheet shed of T. Wu’s waving swimmer model (Wu TY. 1961).

The model was compared to experiments and used in several recent publications from the Courant Institute (Newbolt et al. 2019, 2022, 2024).

Citations:  

Wu, T. Y. T. (1961). Swimming of a waving plate. Journal of Fluid Mechanics, 10(3), 321-344. DOI: https://doi.org/10.1017/S0022112061000949

Newbolt, J. W., Lewis, N., Bleu, M., Wu, J., Mavroyiakoumou, C., Ramananarivo, S., & Ristroph, L. (2024). Flow interactions lead to self-organized flight formations disrupted by self-amplifying waves. Nature Communications, 15(1), 3462. DOI: https://doi.org/10.1038/s41467-024-47525-9

Newbolt, J. W., Zhang, J., & Ristroph, L. (2022). Lateral flow interactions enhance speed and stabilize formations of flapping swimmers. Physical Review Fluids, 7(6), L061101. DOI: https://doi.org/10.1103/PhysRevFluids.7.L061101

Newbolt, J. W., Zhang, J., & Ristroph, L. (2019). Flow interactions between uncoordinated flapping swimmers give rise to group cohesion. Proceedings of the National Academy of Sciences, 116(7), 2419-2424.  DOI: https://doi.org/10.1073/pnas.1816098116

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Congratulations on such a comprehensive and well-thought-out study; I truly enjoyed reading it and have only a couple of suggestions that I believe will help further strengthen the paper. I am including a bunch of references here that are very familiar to me without the expectation of you to include them all, just to point at areas that I feel you might consider useful.

We thank the referee again for their careful read of the manuscript and for their constructive feedback. We appreciate it.

First, I believe that some more rationale is needed to justify the chosen modeling framework. I am fully aware of how difficult is to run these simulations, but I see some critical assumptions that need to be at least spelled out for the reader to appreciate the limitations of the study: (1) Constraining the cross-stream coordinate (a stability analysis should include perturbations on the cross-stream coordinate as well, see, for example, https://doi.org/10.1017/flo.2023.25 -- I know this is much simpler as it discards any vortex shedding) and (2) Assuming equal frequency and amplitude (there are studies showing variation of tail beat frequency in animals depending on their position in the school, see, for example, https://doi.org/10.1007/s00265-014-1834-4).

Thank you for these suggestions. These are indeed important and interesting points to discuss in the manuscript. See response above regarding point 1. Regarding point 2, this is of course important and will be pursued in future extensions of this work. We edited the intro and discussion of the main text to explain this.

In the paper “Stability of schooling patterns of a fish pair swimming against a flow”, The authors considered a pair of swimmers swimming in a channel. They analyzed stability of the system and find multiple equilibria of the system, including inline and staggered formation, and a special formation of perpendicular to the wall. Studying fish school in confined domain and analyzing their stability is very interesting. We added citation to this paper in the discussion section at the end of page 10.

In the paper “Fish swimming in schools save energy regardless of their spatial position”, the authors measured the reduction in power of fish by measuring tail beat frequency and oxygen consumption and compared them to measurements in solitary fish. They found that in a school of fish, individuals always save power comparing to swimming alone.  However, there is one important caveat in this study: they considered a larger school of fish and expressed the results in terms of pairwise configurations (see schematics we draw below). This is misleading because it may suggest that formations with only two fish provide benefits each other, while in fact, the data is obtained from a larger school with many neighbors. They only consider a fish’s relationship to its nearest neighbor. But in a large school, other neighbors will also have influence on their energy consumption.  In the schematics below, we emphasized on several focal fishes, marking them as red, green, and blue. We also marked their nearest neighbors using the same color, but lighter. The nearest neighbors are what the authors are considering to show its neighbor relationship. For example, a problematic one is the red fish, for which its nearest neighbor is behind it, but indeed, its power saving may come from the other neighbors, which are around or ahead it.

Author response image 3.

Second, I would like to see more biology context with respect to limitations that are inherent to a purely mechanical model, including, neglecting vision that we know plays a synergistic role in determining schooling patterns. For example, a recent study https://doi.org/10.1016/j.beproc.2022.104767 has presented experiments on fish swimming in the dark and in bright conditions, showing that it is unlikely that hydrodynamics alone could explain typically observed swimming patterns in the literature.

Thank you for this suggestion and for sharing us with the paper “Collective response of fish to combined manipulations of illumination and flow”. This is a great study, and we are sorry to have missed it.

In this paper, the authors found that when having illumination, fish swim more cohesively, which is in consistent with another paper we already cited “The sensory basis of schooling by intermittent swimming in the rummy-nose tetra (Hemigrammus rhodostomus)”. Another important conclusion in this paper is that when having brighter illumination and with flow, fish school spend more time side by side. This connects well to the conclusion in another paper we cited “Simple phalanx pattern leads to energy saving in cohesive fish schooling,” where at lower flow speed in a water channel, fish tended to form a dynamic school while at higher flow speed, they organized in a side-by-side/ phalanx configuration. This conclusion is consistent with our study that in side-by-side formation, fish share power saving.

Importantly, it is well known that both vision and flow sensing play important roles in fish schooling. This study aimed to merely explore what is possible through passive hydrodynamic interactions, without visual and flow sensing and response. We clarify this in the revised version of the manuscript.

Third, I am not too convinced about the flow agreement metric, which only accounts for linear interactions between the foils. More sophisticated approaches could be utilized as the one proposed here https://doi.org/10.1017/jfm.2018.369, based on a truly model-agnostic view of the interaction - therein, the authors show non-reciprocal (in strength and time-scale) coupling between two in-line flapping foils using information theory. I also would like to mention this older paper https://doi.org/10.1098/rsif.2012.0084, where an equivalent argument about the positioning of a trailing fish with respect to a leading robotic fish is made from experimental observations.

Thank you for these remarks and for sharing these two interesting papers.

The flow agreement metric is not specific to two fish, as we show in Fig. 6 of the manuscript. We edited the manuscript and SI to better explain the motivation and implementation of the flow agreement parameter. We edited the main text, see revisions on page 7, and added a new section call “diagnostic tools.”.

In the paper “An information-theoretic approach to study fluid–structure interactions”, the authors calculate the transfer entropy between two oscillating airfoils when they are hydrodynamically coupled.  This is an interesting study! We will apply this approach to analyzing larger schools in the future. We cited this paper in the introduction.

In the paper “Fish and robots swimming together: attraction towards the robot demands biomimetic locomotion”, the authors found that fish will swim behind an artificial fish robot, especially when the fish robot is beating its tail instead of static. At specific conditions, the fish hold station behind the robot, which may be due to the hydrodynamic advantage obtained by swimming in the robot’s wake. DPIV resolved the wake behind a static/ beating fish robot, but did not visualize the flow field when the fish is there. This study is similar to a paper we already cited “In-line swimming dynamics revealed by fish interacting with a robotic mechanism”, in which, they considered fish-foil interaction. In the revised manuscript, we cite both papers.

For the reviewer’s comments about flow agreement only accounts for linear interactions between the foils, we want to explain more to clarify this. The flow agreement parameter is a nonlinear metric, which considered the interaction between a virtual swimmer and an arbitrary unsteady flow field. Although the metric is a linear function of swimmer’s speed, it is indeed a nonlinear function of spacing and phase, which are the quantities we care about. Moreover, the flow field can by generated by either experiment or CFD simulation, and behind one or more swimmers. It is true that it is a one way coupled system since the virtual swimmer does not perturb the flow field.

Again, this is great work and I hope these suggestions are of help.

Thank you again! We are delighted to receive such a positive and constructive feedback.

Reviewer #2 (Recommendations For The Authors):

(1) About Figure 1: Panel C should be made to match between CFD and VS with regard to the swimmer positions. Also, if the general goal of the figure is to compare CFD and VS, then how about showing a difference map of the velocity fields as a third column of panels across A-D?

Thank you for pointing this out. Figure 1 C is updated accordingly.

The general goal is to show the CFD and VS simulations produce qualitatively similar results. Some quantities are not the same across models, e.g. the swimming speed of swimmers are different, but the scaled distance is the same.

(2) Figure 3: In A, it would be nice to keep the y-axis the same across all plots, which would aid quick visual comparison. In B, the legend labels for CFD and VS should be filled in with color so that the reader can more easily connect to the markers in the plot.

Thank you for pointing this out, we’ve updated figure 3 and 6.

(3) Figures 4, 9, and Supplementary Figures too: As mentioned previously, the agreement parameter plots are saturated in the color map, possibly obscuring more detailed information.

Thank you for pointing this out. The goal is to show that there is a large region with positive flow agreement parameter.

We picked up the flow agreement behind a single swimmer in VS simulation (Fig.4B) and added the counter lines to it (represents 0.25 and 0.5).  Not many details are hidden by the saturated colormap.

Author response image 4.

We also updated Fig 4 and Fig 9 accordingly.

(4) Figure 6: Is this CFD or VS? Why show one or the other and not both? In B, it seems that there are only savings available and no energetically costly positions. This seems odd. In C, it seems the absolute value on dF/dd is suppressing some important information about stability - the sign of this seems important. In E, the color bar seems to be reflected from what is standard, i.e. 0 on the left and 100 on the right, as in F.

Thank you for asking. Fig. 6 is based only on VS simulations. There are hundreds of simulations in this figure, we are not running CFD simulations to save computational effort. Representative CFD simulations are shown in Figure 1,2,3, for comparison. We added a sentence in the figure caption for clarification.

In C, since  is always negative for emergent formations (only stable equilibria can appear during forward time simulation), we are showing its absolute value for comparison.

In E, we are flipping this because larger flow agreement parameter corresponds to more power saving, in the other word, negative changes in COT.

(5) Fig. 8: For cases such as in D that have >100% power savings, does this mean that the swimmer has work done by the flow? How to interpret this physically for a flapping foil and biologically for a fish?

Yes, it means the hydrofoil/fish gets a free ride, and even able to harvest energy from the incoming flow. Actually, similar phenomenon has been reported in the biology and engineering literature. For example, Liao et al. 2003, Beal et al. 2006 found that live or dead fish can harvest energy from incoming vortical flow by modulating their body curvature.

In engineering, Chen et al. 2018, Ribeiro et al. 2021 have found that the following airfoil in a tandem/ inline formation can harvest energy from the wake of leading swimmer in both simulation and experiemnts.

Citations:  

Liao, J. C., Beal, D. N., Lauder, G. V., & Triantafyllou, M. S. (2003). Fish exploiting vortices decrease muscle activity. Science, 302(5650), 1566-1569. DOI: https://doi.org/10.1126/science.1088295

Beal, D. N., Hover, F. S., Triantafyllou, M. S., Liao, J. C., & Lauder, G. V. (2006). Passive propulsion in vortex wakes. Journal of fluid mechanics, 549, 385-402. DOI: https://doi.org/10.1017/S0022112005007925

Chen, Y., Nan, J., & Wu, J. (2018). Wake effect on a semi-active flapping foil based energy harvester by a rotating foil. Computers & Fluids, 160, 51-63. DOI: https://doi.org/10.1016/j.compfluid.2017.10.024

Ribeiro, B. L. R., Su, Y., Guillaumin, Q., Breuer, K. S., & Franck, J. A. (2021). Wake-foil interactions and energy harvesting efficiency in tandem oscillating foils. Physical Review Fluids, 6(7), 074703. DOI: https://doi.org/10.1103/PhysRevFluids.6.074703

https://www.youtube.com/watch?v=aUcgN_tId18

tagged as Labourer

156

9

35

61

63

https://www.youtube.com/watch?v=aUcgN_tId18

https://www.youtube.com/watch?v=aUcgN_tId18

https://www.youtube.com/watch?v=aUcgN_tId18

https://www.youtube.com/watch?v=aUcgN_tId18

https://www.youtube.com/watch?v=aUcgN_tId18

Está todo correcto

Este comentario está muy claro

eLife assessment

This study presents valuable findings describing how the midbrain periaqueductal gray matter and basolateral amygdala communicate when a predator threat is detected. Though the periaqueductal gray is usually viewed as a downstream effector, this work contributes to a growing body of literature from this lab showing that the periaqueductal gray produces effects by acting on the basolateral amygdala, the experimental design, data collection and analysis methods provide solid evidence for the main claims. The anatomical and immediately early gene evidence that the paraventricular nucleus of the thalamus may serve as a mediator of dorsolateral periaqueductal gray to basolateral amygdala neurotransmission provides and impetus for future functional assessment of this possibility. This study will appeal to a broad audience, including basic scientists interested in neural circuits, basic and clinical researchers interested in fear, and behavioral ecologists interested in foraging.

Reviewer #1 (Public Review):

In the presence of predators, animals display attenuated foraging responses and increased defensive behaviors that serve to protect them from potential predatory attacks. Previous studies have shown that the basolateral nucleus of the amygdala (BLA) and the periaqueductal gray matter (PAG) are necessary for the acquisition and expression of conditioned fear responses. However, it remains unclear how BLA and PAG neurons respond to predatory threats when animals are foraging for food. To address this question, Kim and colleagues conducted in vivo electrophysiological recordings from BLA and PAG neurons and assessed approach-avoidance responses while rats searched for food in the presence of a robotic predator.

The authors observed that rats exhibited a significant increase in the latency to obtain the food pellets and a reduction in the pellet success rate when the predator robot was activated. A subpopulation of PAG neurons showing an increased firing rate in response to the robot activation didn't change their activity in response to food pellet retrieval during the pre- or post-robot sessions. Optogenetic stimulation of PAG neurons increased the latency to procure the food pellet in a frequency- and intensity-dependent manner, similar to what was observed during the robot test. Combining optogenetics with single-unit recordings, the authors demonstrated that photoactivation of PAG neurons increased the firing rate of 10% of BLA cells. A subsequent behavioral test in 3 of these same rats demonstrated that BLA neurons responsive to PAG stimulation displayed higher firing rates to the robot than BLA neurons nonresponsive to PAG stimulation. Next, because the PAG does not project monosynaptically to the BLA, the authors used a combination of retrograde and anterograde neural tracing to identify possible regions that could convey robot-related information from PAG to the BLA. They observed that neurons in specific areas of the paraventricular nucleus of the thalamus (PVT) that are innervated by PAG fibers contained neurons that were retrogradely labeled by the injection of CTB in the BLA. In addition, PVT neurons showed increased expression of the neural activity marker cFos after the robot test, suggesting that PVT may be a mediator of PAG signals to the BLA.

Overall, the idea that the PAG interacts with the BLA via the midline thalamus during a predator vs. foraging test is new and quite interesting. The authors have used appropriate tools to address their questions. However, there are some major concerns regarding the design of the experiments, the rigor of the histological analyses, the presentation of the results, the interpretation of the findings, and the general discussion that largely reduces the relevance of this study.

The authors have fully addressed all my concerns.

Reviewer #2 (Public Review):

The authors characterized the activity of the dorsal periaqueductal gray (dPAG) - basolateral amygdala (BLA) circuit. They show that BLA cells that are activated by dPAG stimulation are also more likely to be activated by a robot predator. These same cells are also more likely to display synchronous firing.

The authors also replicate prior results showing that dPAG stimulation evokes fear and the dPAG is activated by a predator.

Lastly, the report performs anatomical tracing to show that the dPAG may act on the BLA via the paraventricular thalamus (PVT). Indeed, the PVT receives dPAG projections and also projects to the BLA. However, the authors do not show if the PVT mediates dPAG to BLA communication with any functional behavioral assay.

The major impact in the field would be to add evidence to their prior work, strengthening the view that the BLA can be downstream of the dPAG.

Reviewer #3 (Public Review):

In the present study, the authors examined how dPAG neurons respond to predatory threats and how dPAG and BLA communicate threat signals. The authors employed single-unit recording and optogenetics tools to address these issues in an 'approach food-avoid predator' paradigm. They characterized dPAG and BLA neurons responsive to a looming robot predator and found that dPAG opto-stimulation elicited fleeing and increased BLA activity. Importantly, they found that dPAG stimulation produces activity changes in subpopulations of BLA neurons related to predator detection, thus supporting the idea that dPAG conveys innate fear signals to the amygdala. In addition, injections of anterograde and retrograde tracers into the dPAG and BLA, respectively, along with the examination of c-FOS activity in midline thalamic relay stations, suggest that the paraventricular nucleus of the thalamus (PVT) may serve as a mediator of dPAG to BLA neurotransmission. Of relevance, the study helps to validate an important concept that dPAG mediates primal fear emotion and may engage upstream amygdala targets to evoke defensive responses. The series of experiments provides a compelling case for supporting their conclusions. The study brings important concepts revealing dynamics of fear-related circuits particularly attractive to a broad audience, from basic scientists interested in neural circuits to psychiatrists.

Author response:

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

Public Reviews

Reviewer 1 summarized that: In this revised version of the manuscript, the authors have made important modifications in the text, inserted new data analyses, and incorporated additional references, as recommended by the reviewers. These modifications have significantly improved the quality of the manuscript.

We are grateful for the reviewer's positive recognition of our revisions.

Reviewer 2 noted that:

(1) The authors do not show if the PVT mediates dPAG to BLA communication with any functional behavioral assay.

We appreciate the reviewer’s suggestion to include a functional assay to investigate the role of the PVT in mediating communication between the dPAG and BLA. Our primary objective was to confirm the upstream role of the dPAG in processing and relaying naturalistic predatory threat information to the BLA, thereby broadening our current understanding of the dPAG-BLA relationship based on Pavlovian fear conditioning paradigms.

Given previous anatomical findings indicating the absence of direct monosynaptic projections from the dPAG to the BLA (Cameron et al. 1995, McNally, Johansen, and Blair 2011, Vianna and Brandao 2003), we employed both anterograde and retrograde tracers, supplemented by c-Fos expression analysis following predatory threats, to explore possible routes through which threat signals may be conveyed from the dPAG to the BLA. Our findings indicated significant activity within the midline thalamic regions, particularly the PVT as a mediator of dPAG-BLA interactions, corroborating the possibility of dPAGàBLA information flow.

Investigating the PVT's functional role appropriately would require single-unit recordings, correlation analysis of PVT neuronal responses with dPAG and BLA neuronal responses, and pathway-specific causal techniques, involving other midline thalamic regions for controls. This comprehensive study would represent an independent study.

In response to previous feedback, we have carefully revised our manuscript to moderate the emphasis on the PVT's role. Both the Abstract, Results, and Discussion refer more broadly to "midline thalamic regions" and “The midline thalamus” (subheading) rather than specifically to the PVT. In the Introduction, we mention that the PVT "may be part of a network that conveys predatory threat information from the dPAG to the BLA." Our conclusions about the functional interaction between the dPAG and BLA, which broaden the view of Pavlovian fear conditioning, are not contingent on confirming a specific intermediary role for the PVT.

(2) The author also do not thoroughly characterize the activity of BLA cells during the predatory assay.

Our previous studies have extensively detailed BLA cell firing characteristics, including their responsiveness to food and/or a robot predator during the predatory assay (Kim et al. 2018, Kong et al. 2021), and compared these findings to other predator studies (Amir et al. 2019, Amir et al. 2015). In the current study, out of 85 BLA cells, 3 were food-specific and 4 responded to both the pellet and the robot, with none of these 7 cells responding to dPAG stimulation.

Given our earlier findings of the immediate responses of BLA neurons to robot activation, we specifically examined whether robot-responsive BLA neurons receive signals from the dPAG. For this analysis, we excluded all food-related cells (pellet cells and BOTH cells) and focused on the time window immediately after robot activation (within 500 ms after robot onset). This approach enabled us to avoid potential confounds from residual effects of robot-induced immediate BLA responses during the animals’ flight and nest entry behaviors.

Furthermore, as previously described, the robot is programmed to move forward a fixed distance and then return, repeatedly triggering foraging behavior. This setup facilitates the analysis of neural changes during food approach and predator avoidance conflicts. However, animals quickly adapt to the robot, reducing freezing and stretch-attend behaviors, making time-stamped analysis of these behaviors unfeasible.

We would like to highlight that the present study explicitly focused on demonstrating whether BLA neurons that responded to intrinsic dPAG optogenetic stimulation also responded to extrinsic predatory robot activation, and compared their firing characteristics to those BLA neurons that did not respond to dPAG stimulation (Figure 3). This targeted analysis provides insights into the responsiveness of BLA neurons to both intrinsic and extrinsic stimuli, furthering our understanding of the dPAG-BLA interaction in the context of predatory threats.

Reviewer 3 also raised no concerns and stated that: The series of experiments provide a compelling case for supporting their conclusions. The study brings important concepts revealing dynamics of fear-related circuits particularly attractive to a broad audience, from basic scientists interested in neural circuits to psychiatrists.

We sincerely thank the reviewer for the positive feedback on our revisions.

Recommendations for the Authors

Reviewer 1: There are a few minor concerns that the authors may want to fix:

(1) Point 5) The sentence: "The complexity of targeting the dPAG, which includes its dorsomedial, dorsolateral, lateral, and ventrolateral subdivisions" is hard to follow because the ventrolateral subdivision is not part of the dPAG. The authors may want to say specific subregions of the PAG instead. It is also unclear why transgenic animals would be needed for this projection-defined manipulations. The combination of retrograde Cre-recombinase virus with inhibitory opsin or chemogenetic approach may be sufficient.

We appreciate the reviewer’s insightful feedback regarding our description of the dPAG and the use of transgenic mice in future studies. As suggested, we have corrected the manuscript to exclude the 'ventrolateral' subdivision from the dPAG description, now accurately aligning with pioneering studies (Bandler, Carrive, and Zhang 1991, Bandler and Keay 1996, Carrive 1993) that designated dPAG as including the dorsomedial (dmPAG), dorsolateral (dlPAG) and lateral (lPAG) regions, as cited in our revised manuscript.

We acknowledge the reviewer’s helpful suggestion regarding the use of retrograde Cre-recombinase virus with inhibitory opsins or chemogenetic approaches as viable alternatives. These methods have been incorporated into our discussion (pages 14-15): “While our findings demonstrate that opto-stimulation of the dPAG is sufficient to trigger both fleeing behavior and increased BLA activity, we have not established that the dPAG-PVT circuit is necessary for the BLA’s response to predatory threats. To establish causality and interregional relationships, future studies should employ methods such as pathway-specific optogenetic inhibition (using retrograde Cre-recombinase virus with inhibitory opsins; Lavoie and Liu 2020, Li et al. 2016, Senn et al. 2014) or chemogenetics (Boender et al. 2014, Roth 2016) in conjunction with single unit recordings to fully characterize the dPAG-PVT-BLA circuitry’s (as opposed to other midline thalamic regions for controls) role in processing predatory threat-induced escape behavior. If inactivating the dPAG-PVT circuits reduces the BLA's response to threats, this would highlight the central role of the dPAG-PVT pathway in this defense mechanism. Conversely, if the BLA's response remains unchanged despite dPAG-PVT inactivation, it could suggest the existence of multiple pathways for antipredatory defenses.”

This revision addresses the critique by clarifying the anatomical description of the dPAG and emphasizing the feasibility of using targeted viral approaches without the necessity for transgenic animals.

(2) Point 6e) The authors mentioned that "pellet retrieval" was indicated by the animal entering a designated zone 19 cm from the pellet, driven by hunger. Entering the area 19cm of distance should be labeled as food approaching rather then food retrieval because in many occasions the animals may be some seconds away of grabbing the pellet.

We agree and incorporate the change (pg. 22).

(3) Point 11) We would strongly recommend the authors to replace the terminology "looming" by "approaching" to avoid confusion with several previous studies looking at defensive behaviors in responses to looming induced by the shadow of an object moving closer to the eyes.

Done.

(4) Point 17) The authors mentioned that "A total of three rats were utilized for the robot testing experiments depicted in Fig. 2 G-J." However, the figure indicates a total of 9 ChR2 and 4 controls.

We apologize for the confusion in our previous author responses. To examine the optical stimulation effects on behavior in Fig. 2G-J, we used a total of 9 ChR2 and 4 EYFP rats. The experimental sequence is detailed in the previously revised manuscript (pg. 20): “For optical stimulation and behavioral experiments, the procedure included 3 baseline trials with the pellet placed 75 cm away, followed by 3 dPAG stimulation trials with the pellet locations sequentially set at 75 cm, 50 cm, and 25 cm. During each approach to the pellet, rats received 473-nm light stimulation (1-2 s, 20-Hz, 10-ms width, 1-3 mW) through a laser (Opto Engine LLC) and a pulse generator (Master-8; A.M.P.I.). Additional testing to examine the functional response curves was conducted over multiple days, with incremental adjustments to the stimulation parameters (intensity, frequency, duration) after confirming that normal baseline foraging behavior was maintained. For these tests, one parameter was adjusted incrementally while the others were held constant (intensity curve at 20 Hz, 2 s; frequency curve at 3 mW, 2 s; duration curve at 20 Hz, 3 mW). If the rat failed to procure the pellet within 3 min, the gate was closed, and the trial was concluded.”

This clarification ensures that the actual number of animals used is accurately reflected and aligns with the figure data, addressing the reviewer's concern.

Reviewer 2: The authors made important changes in the text to address study limitations, including citations requested by the Reviewers and additional discussions about how this work fits into the existing literature. These changes have strengthened the manuscript.

(1) However, the authors did not perform new experiments to address any of the issues raised in the previous round of reviews. For example, they did not make optogenetic manipulations of the pathway including the PVT, and did not add any loss of function experiments. The justification that these experiments are better suited for future reports using mice is not convincing, because hundreds of papers performing these types of circuit dissection assays have been performed in rats.

We appreciate the reviewer's comments regarding the experimental scope of our study. Our study’s primary objective was to explore the dPAG’s upstream functional role in processing and conveying naturalistic predatory threat information to the BLA, extending our current understanding of the dPAG-BLA relationship based on Pavlovian fear conditioning paradigms. We believe that our findings effectively address this goal.

Our use of anterograde and retrograde tracers, supplemented by c-Fos expression analysis in response to predatory threats, was primarily conducted to verify the possibility of the dPAGàBLA information flow during predator encounters. This involved exploring potential routes through which threat signals might be conveyed from the dPAG to the BLA, given the lack of direct monosynaptic projections from the dPAG to BLA neurons (Cameron et al. 1995, McNally, Johansen, and Blair 2011, Vianna and Brandao 2003). This methodology helped us identify a potential structure, PVT, for more in-depth future studies. A thorough examination of the PVT's role would require single-unit recordings and causal techniques, incorporating other midline thalamic regions as controls, representing a significant and separate study on its own.

In response to prior feedback, we have carefully revised our manuscript to generally address the role of "midline thalamic regions" rather than focusing specifically on the PVT. We wish to emphasize that our findings, which illustrate unique functional interactions between the dPAG and BLA in response to a predatory imminence, remain compelling and informative even without definitive evidence of the PVT’s involvement.

Reviewer 3: In the revised version of the manuscript, the authors addressed adequately all the concerns raised by the reviewers. 

We thank the reviewer for the thoughtful feedback on the earlier version of our manuscript and for reexamining the revisions we have made.

References

Amir, A., P. Kyriazi, S. C. Lee, D. B. Headley, and D. Pare. 2019. "Basolateral amygdala neurons are activated during threat expectation." J Neurophysiol 121 (5):1761-1777.

Amir, A., S. C. Lee, D. B. Headley, M. M. Herzallah, and D. Pare. 2015. "Amygdala Signaling during Foraging in a Hazardous Environment." J Neurosci 35 (38):12994-3005.

Bandler, R., P. Carrive, and S. P. Zhang. 1991. "Integration of somatic and autonomic reactions within the midbrain periaqueductal grey: viscerotopic, somatotopic and functional organization." Prog Brain Res 87:269-305.

Bandler, R., and K. A. Keay. 1996. "Columnar organization in the midbrain periaqueductal gray and the integration of emotional expression." Prog Brain Res 107:285-300.

Boender, A. J., J. W. de Jong, L. Boekhoudt, M. C. Luijendijk, G. van der Plasse, and R. A. Adan. 2014. "Combined use of the canine adenovirus-2 and DREADD-technology to activate specific neural pathways in vivo." PLoS One 9 (4):e95392.

Cameron, A. A., I. A. Khan, K. N. Westlund, and W. D. Willis. 1995. "The efferent projections of the periaqueductal gray in the rat: a Phaseolus vulgaris-leucoagglutinin study. II. Descending projections." J Comp Neurol 351 (4):585-601.

Carrive, P. 1993. "The periaqueductal gray and defensive behavior: functional representation and neuronal organization." Behav Brain Res 58 (1-2):27-47.

Kim, E. J., M. S. Kong, S. G. Park, S. J. Y. Mizumori, J. Cho, and J. J. Kim. 2018. "Dynamic coding of predatory information between the prelimbic cortex and lateral amygdala in foraging rats." Sci Adv 4 (4):eaar7328.

Kong, M. S., E. J. Kim, S. Park, L. S. Zweifel, Y. Huh, J. Cho, and J. J. Kim. 2021. "'Fearful-place' coding in the amygdala-hippocampal network." Elife 10.

Lavoie, A., and B. H. Liu. 2020. "Canine Adenovirus 2: A Natural Choice for Brain Circuit Dissection." Front Mol Neurosci 13:9.

Li, Y., L. Hickey, R. Perrins, E. Werlen, A. A. Patel, S. Hirschberg, M. W. Jones, S. Salinas, E. J. Kremer, and A. E. Pickering. 2016. "Retrograde optogenetic characterization of the pontospinal module of the locus coeruleus with a canine adenoviral vector." Brain Res 1641 (Pt B):274-90.

McNally, G. P., J. P. Johansen, and H. T. Blair. 2011. "Placing prediction into the fear circuit."  Trends Neurosci 34 (6):283-92.

Roth, B. L. 2016. "DREADDs for Neuroscientists." Neuron 89 (4):683-94.

Senn, V., S. B. Wolff, C. Herry, F. Grenier, I. Ehrlich, J. Grundemann, J. P. Fadok, C. Muller, J. J. Letzkus, and A. Luthi. 2014. "Long-range connectivity defines behavioral specificity of amygdala neurons." Neuron 81 (2):428-37.

Vianna, D. M., and M. L. Brandao. 2003. "Anatomical connections of the periaqueductal gray: specific neural substrates for different kinds of fear." Braz J Med Biol Res 36 (5):557-66.

eLife assessment

This study explores the role of calcyphosine-like (CAPSL) in Familial Exudative Vitreoretinopathy (FEVR) via the MYC pathway, offering valuable insights into disease mechanisms that are supported by a solid, multi-pronged approach. The manuscript, which presents the phenotype of an interesting new mouse model, provides convincing evidence that CAPSL variants cause disease.

Reviewer #1 (Public Review):

Summary:<br /> The author presents the discovery and characterization of CAPSL as a potential gene linked to Familial Exudative Vitreoretinopathy (FEVR), identifying one nonsense and one missense mutation within CAPSL in two distinct patient families afflicted by FEVR. Cell transfection assays suggest that the missense mutation adversely affects protein levels when overexpressed in cell cultures. Furthermore, conditionally knocking out CAPSL in vascular endothelial cells leads to compromised vascular development. The suppression of CAPSL in human retinal microvascular endothelial cells results in hindered tube formation, a decrease in cell proliferation, and disrupted cell polarity. Additionally, transcriptomic and proteomic profiling of these cells indicates alterations in the MYC pathway.

Strengths:<br /> The study is nicely designed with a combination of in vivo and in vitro approaches, and the experimental results are good quality.

Weaknesses:<br /> My reservations lie with the main assertion that CAPSL is associated with FEVR, as the genetic evidence from human studies appears relatively weak. Further careful examination of human genetics evidence in both patient cohorts and the general population will help to clarify. In light of human genetics, more caution needs to be exercised when interpreting results from mice and cell model and how is it related to the human patient phenotype. Future replication by finding more FEVR patients with a mutation in CAPSL will strengthen the findings.

Reviewer #2 (Public Review):

Summary:<br /> This work identifies two variants in CAPSL in two generation familial exudative vitreoretinopathy (FEVR) pedigrees, and using a knockout mouse model, they link CAPSL to retinal vascular development and endothelial proliferation through the MYC pathway. Together, these findings suggest that the identified variants may be causative and that CAPSL is a new FEVR-associated gene.

Strengths:<br /> The authors data provides compelling evidence that loss of the poorly understood protein CAPSL can lead to reduced endothelial proliferation in mouse retina and suppression of MYC signaling, consistent with the disease seen in FEVR patients. The paper is clearly written, and the data generally support the author's hypotheses.

Weaknesses:<br /> (1) Both pedigrees described suggest autosomal dominant inheritance in humans, but no phenotype was observed in Capsl heterozygous mice. Additional studies would be needed to determine the cause of this disparity.

(2) Additional discussion of the hypothesized functional mechanism of the p.L83F variant would have improved the manuscript. While the human genetic data is compelling, it remains unclear how this variant may effect CAPSL function. In vitro, p.L83F protein appears to be normally localized within the cell and it is unclear why less mutant protein was detected in transfected cells. Was the modified protein targeted for degradation?

(3) Authors did not describe how the new crispr-generated Capsl-loxp mouse model was screened for potential off-target gene editing, raising the possibility that unrelated confounding mutations may have been introduced.

Reviewer #3 (Public Review):

Summary:<br /> This manuscript by Liu et al. presents a case that CAPSL mutations are a cause of familial exudative vitreoretinopathy (FEVR). Attention was initially focused on the CAPSL gene from whole exome sequence analysis of two small families. The follow-up analyses included studies in which Capsl was manipulated in endothelial cells of mice and multiple iterations of molecular and cellular analyses. Together, the data show that CAPSL influences endothelial cell proliferation and migration. Molecularly, transcriptomic and proteomic analyses suggest that CAPSL influences many genes/proteins that are also downstream targets of MYC and may be important to the mechanisms.

Strengths:<br /> This multi-pronged approach found a previously unknown function for CAPSL in endothelial cells and pointed at MYC pathways as high-quality candidates in the mechanism. Through the review process, some statements and interpretations were initially challenged. However, the issues were addressed with new experimentation and modifications to the text - leaving a strengthened presentation that makes a compelling case.

Weaknesses:<br /> Two issues shape the overall impact for me. First, it remains unclear how common CAPSL variants may be in the human population. From the current study, it is possible that they are rare - perhaps limiting an immediate clinical impact. However, sharing the data may help identify additional variants in FEVR or other vascular diseases. The findings also make advances in basic biology which could ultimately contribute to therapies of broad relevance. Thus, this weakness is considered modest. Second, the links to the MYC axis are largely based on association, which will require additional experimentation to help understand.

One interesting technical point raised in the study, which might be missed without care by the readership, is that the variants appear to act dominantly in human families, but only act recessively in the mouse model. The authors cite other work from the field in which this same mismatch occurs, likely pointing to limits in how closely a mouse model might be expected to recapitulate a human disease. This technical point is likely relevant to ongoing studies of FEVR and many other multigenic diseases as well.

Author response:

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

Reviewer #1 (Public Review): 

Summary: 

The author presents the discovery and characterization of CAPSL as a potential gene linked to Familial Exudative Vitreoretinopathy (FEVR), identifying one nonsense and one missense mutation within CAPSL in two distinct patient families afflicted by FEVR. Cell transfection assays suggest that the missense mutation adversely affects protein levels when overexpressed in cell cultures. Furthermore, conditionally knocking out CAPSL in vascular endothelial cells leads to compromised vascular development. The suppression of CAPSL in human retinal microvascular endothelial cells results in hindered tube formation, a decrease in cell proliferation, and disrupted cell polarity. Additionally, transcriptomic and proteomic profiling of these cells indicates alterations in the MYC pathway. 

Strengths: 

The study is nicely designed with a combination of in vivo and in vitro approaches, and the experimental results are good quality. 

We thank the reviewer for the conclusion and positive comments.

Weaknesses: 

My reservations lie with the main assertion that CAPSL is associated with FEVR, as the genetic evidence from human studies appears relatively weak. Further careful examination of human genetics evidence in both patient cohorts and the general population will help to clarify. In light of human genetics, more caution needs to be exercised when interpreting results from mice and cell models and how is it related to the human patient phenotype. 

We thank the reviewer for careful reading and constructive suggestion. we added several experiments to address the concern of reviewer are as follows:

(1) The pLI score of LOF allele of CAPSL is based of general population, among which Europeans account for ~77% and East Asians make up less than 3%. Since the FEVR families in this article all come from China, the pLI score may not be accurate. Of course, we will continue to collect FEVR pedigrees.

(2) We evaluated the phenotype of Capsl heterozygous mice at P5, and the results showed no overt difference in vascular progression, vessel density and branchpoints with littermate wildtype controls (Fig.S4). The lack of pronounced phenotype in FEVR heterozygous mice may be due to different sensitivity between human and mice. A similar example is LRP5 mutations associated with FEVR. Heterozygous mutations in LRP5 were reported in FEVR patients in multiple populations (PMID: 16929062, 33302760, 27486893, 35918671, 36411543). However, heterozygous Lrp5 knockout mice exhibited no visible angiogenic phenotype (PMID: 18263894). Corresponding description was added in the manuscript at page 6.

(3) We further assessed the angiogenic phenotype when angiogenesis almost complete at P21, and the resulted revealed no difference observed between Ctrl and CapsliECKO/iECKO mice (Fig.S5). And corresponding description was added in the manuscript at page 7.

(4) We evaluated the expression of MYC downstream genes in vivo using lung tissue form P35 Ctrl and _Capsl_iECKO/iECKO mice (Fig.S8). Consistent with the results from in vitro HRECs, _Capsl_iECKO/iECKO mice showed downregulated expression of MYC targets. And corresponding description was added in the manuscript at page 11.

Reviewer #2 (Public Review): 

Summary: 

This work identifies two variants in CAPSL in two-generation familial exudative vitreoretinopathy (FEVR) pedigrees, and using a knockout mouse model, they link CAPSL to retinal vascular development and endothelial proliferation. Together, these findings suggest that the identified variants may be causative and that CAPSL is a new FEVR-associated gene. 

Strengths: 

The authors' data provides compelling evidence that loss of the poorly understood protein CAPSL can lead to reduced endothelial proliferation in mouse retina and suppression of MYC signaling in vitro, consistent with the disease seen in FEVR patients. The study is important, providing new potential targets and mechanisms for this poorly understood disease. The paper is clearly written, and the data generally support the author's hypotheses. 

We thank the reviewer for the conclusion and positive comments.

Weaknesses: 

(1) Both pedigrees described appear to suggest that heterozygosity is sufficient to cause disease, but authors have not explored the phenotype of Capsl heterozygous mice. Do these animals have reduced angiogenesis similar to KOs? Furthermore, while the p.R30X variant protein does not appear to be expressed in vitro, a substantial amount of p.L83F was detectable by western blot and appeared to be at the normal molecular weight. Given that the full knockout mouse phenotype is comparatively mild, it is unclear whether this modest reduction in protein expression would be sufficient to cause FEVR - especially as the affected individuals still have one healthy copy of the gene. Additional studies are needed to determine if these variants alter protein trafficking or localization in addition to expression, and if they can act in a dominant negative fashion. 

We thank the reviewer for the suggestion. We evaluated the phenotype of Capsl heterozygous mice at P5 (Fig.S4), and the results showed no overt difference in angiogenesis compared with littermate control mice.

We transfected CAPSL wild-type plasmid, p.R30X mutant plasmid and p.L83F mutant plasmid into 293T cells to assess the intracellular localization change of CAPSL mutant proteins (Fig.S1). The result showed that the point mutation did not affect the localization of the mutated protein, and corresponding description was added in the manuscript at page 5.

(2) The manuscript nicely shows that loss of CAPSL leads to suppressed MYC signaling in vitro. However, given that endothelial MYC is regulated by numerous pathways and proteins, including FOXO1, VEGFR2, ERK, and Notch, and reduced MYC signaling is generally associated with reduced endothelial proliferation, this finding provides little insight into the mechanism of CAPSL in regulating endothelial proliferation. It would be helpful to explore the status of these other pathways in knockdown cells but as the authors provide only GSEA results and not the underlying data behind their RNA seq results, it is difficult for the reader to understand the full phenotype. Volcano plots or similar representations of the underlying expression data in Figures 6 and 7 as well as supplemental datasets showing the differentially regulated genes should be included. In addition, while the paper beautifully characterizes the delayed retinal angiogenesis phenotype in CAPSL knockout mice, the authors do not return to that model to confirm their in vitro findings. 

We thank the reviewer for the suggestion. Although endothelial MYC can be regulated by FOXO1, VEGFR2, ERK, and Notch signaling pathway, these pathways are not enriched in the RNA seq data of CAPSL-depleted HRECs. This suggests that the down regulated MYC targets may not be influenced by the signaling pathway mentioned above. RNA-seq raw data have been uploaded to the Genome Sequence Archive (https://ngdc.cncb.ac.cn/gsa/browse/HRA010305) and proteomic profiling raw data have been uploaded to the Genome Sequence Archive (https://www.ebi.ac.uk/pride/archive), and the assigned accession number was PXD051696. Corresponding description was added in the manuscript at page 20-21. The datasets represent the differentially regulated genes in Figure 6 and 7 were listed at Dataset S1 and S2.

(3) In Figure S2D, the result of this vascular leak experiment is unconvincing as no dye can be seen in the vessels. What are the kinetics for biocytin tracers to enter the bloodstream after IP injection? Why did the authors choose the IP instead of the IV route for this experiment? Differences in the uptake of the eye after IP injection could confound the results, especially in the context of a model with vascular dysfunction as here. 

We thank the reviewer for suggestion. In Figure S2D (now Fig.S6D), we used a non-representative image to show vascular leakage. We replaced the images with more representative ones. We are sorry that we are not clear about the kinetics for biocytin tracers to enter the bloodstream after IP injection. Since the experiment was carried out on mice at P5, it is not feasible to do IV injection in P5 neonatal mice. We followed the methods described in the previous study involving mice of same age (PMID:35361685).

(4) In Figure 5, it is unclear how filipodia and tip cells were identified and selected for quantification. The panels do not include nuclear or tip cell-specific markers that would allow quantification of individual tip cells, and in Figure 5C it appears that some filipodia are not highlighted in the mutant panel. 

We thank the reviewer for the comments. In Figure 5, we used HRECs to examine the cell proliferation, migration and polarity in vitro, and therefore there is no distinction between tip cells and stalk cells. The quantification of filopodia/lamellipodia was performed as previous studies (PMID: 30783090, PMID: 28805663). In briefly, wound scratch was performed on confluent layers of transfected HRECs, and 9 hours after initiating cell migration by scratch, cells were fixed and stained with phalloidin. Cells at the edge of wound were considered as leader cells and quantified for number of filopodia/lamellipodia.

Reviewer #3 (Public Review): 

Summary: 

This manuscript by Liu et al. presents a case that CAPSL mutations are a cause of familial exudative vitreoretinopathy (FEVR). Attention was initially focused on the CAPSL gene from whole exome sequence analysis of two small families. The follow-up analyses included studies in which CAPSL was manipulated in endothelial cells of mice and multiple iterations of molecular and cellular analyses. Together, the data show that CAPSL influences endothelial cell proliferation and migration. Molecularly, transcriptomic and proteomic analyses suggest that CAPSL influences many genes/proteins that are also downstream targets of MYC and may be important to the mechanisms. 

Strengths: 

This multi-pronged approach found a previously unknown function for CAPSLs in endothelial cells and pointed at MYC pathways as high-quality candidates in the mechanism. 

Weaknesses: 

Two issues shape the overall impact for me. First, the unreported population frequency of the variants in the manuscript makes it unclear if CAPSL should be considered an interesting candidate possibly contributing to FEVR, or possibly a cause. Second, it is unclear if the identified variants act dominantly, as indicated in the pedigrees. The studies in mice utilized homozygotes for an endothelial cell-specific knockout, leaving uncertainty about what phenotypes might be observed if mice heterozygous for a ubiquitous knockout had instead been studied. 

In my opinion, the following scientific issues are specific weaknesses that should be addressed: 

(1) Please state in the manuscript the number of FEVR families that were studied by WES. Please also describe if the families had been selected for the absence of known mutations, and/or what percentage lack known pathogenic variants. 

We thank the reviewer for thoughtful comments. 120 FEVR families were studied by WES and we added corresponding description in the manuscript at page 4.

(2) A better clinical description of family 3104 would enhance the manuscript, especially the father. It is unclear what "manifested with FEVR symptoms, according to the medical records" means. Was the father diagnosed with FEVR? If the father has some iteration of a mild case, please describe it in more detail. If the lack of clinical images in the figure is indicative of a lack of medical documentation, please note this in the manuscript. 

We thank the reviewer for thoughtful comments. The father of family 3104 has also been identified as a carrier of this heterozygous variant, manifested with FEVR symptoms, according to the medical records. Nevertheless, clinical examination images are presently unavailable. We added corresponding description in the manuscript at page 5.

(3) The TGA stop codon can in some instances also influence splicing (PMID: 38012313). Please add a bioinformatic assessment of splicing prediction to the assays and report its output in the manuscript. 

We thank the reviewer for thoughtful comments. We predicted the splicing of c.88C>T variant of CAPSL using MaxEntScan (http://hollywood.mit.edu/burgelab/maxent/Xmaxentscan_scoreseq.html) and SpliceTool (https://rddc.tsinghua-gd.org/ai) (Fig.S2). MaxEntScan and SpliceTool were used to predict the impact of TGA stop codon of c.88C>T variant on the formation of a cryptic donor splice site.

(4) More details regarding utilizing a "loxp-flanked allele of CAPSL" are needed. Is this an existing allele, if so, what is the allele and citation? If new (as suggested by S1), the newly generated CAPSL mutant mouse strain needs to be entered into the MGI database and assigned an official allele name - which should then be utilized in the manuscript and who generated the strain (presumably a core or company?) must be described. 

We added detailed description of Capsl flxoed allele to Method section on page 14-15: “Capslloxp/+ model was generated using the CRISPR/Cas9 nickase technique by Viewsolid Biotechology (Beijing, China) in C57BL/6J background and named Capslem1zxj. The genomic RNA (gRNA) sequence was as follows: Capsl-L gRNA: 5’-CTATCCCAA TTGTGCTCCTGG-3’; Capsl-R gRNA: 5’-TGGGACTCATGGTTCTAGAGG-3’. ”

(5) The statement in the methods "All mice used in the study were on a C57BL/6J genetic background," should be better defined. Was the new allele generated on a pure C57BL/6J genetic background, or bred to be some level of congenic? If congenic, to what generation? If unknown, please either test and report the homogeneity of the background, or consult with nomenclature experts (such as available through MGI) to adopt the appropriate F?+NX type designation. This also pertains to the Pdgfb-iCreER mice, which reference 43 describes as having been generated in an F2 population of C57BL/6 X CBA and did not designate the sub-strain of C57BL/6 mice. It is important because one of the explanations for missing heritability in FEVR may be a high level of dependence on genetic background. From the information in the current description, it is also not inherently obvious that the mice studied did not harbor confounding mutations such as rd1 or rd8. 

We thank the reviewer for suggestion. We added the following description to “Mouse model and genotyping” method section on page 14. “Capslloxp/+ model was generated using the CRISPR/Cas9 nickase technique by Viewsolid Biotechology (Beijing, China) in C57BL/6J background and named Capslem1zxj. The genomic RNA (gRNA) sequence was as follows: Capsl-L gRNA: 5’-CTATCCCAA TTGTGCTCCTGG-3’; Capsl-R gRNA: 5’-TGGGACTCATGGTTCTAGAGG-3’. Pdgfb-iCreER[43] transgenic mice on a mixed background of C57BL/6 and CBA was obtainted from Dr. Marcus Fruttiger and backcrossed to background for 6 generations. Capslloxp/+ mice were bred with Pdgfb-iCreER[43] transgenic mice to generate Capslloxp/loxp, Pdgfb-iCreER mice.” Sanger sequencing was performed on experimental mice to identify whether they harbor confounding mutations such as Pde6b or Crb1. The results showed the mice did not harbor confounding mutations (Fig.S9) and corresponding description was added in the manuscript at page 15.

(6) In my opinion, more experimental detail is needed regarding Figures 2 and 3. How many fields, of how many retinas and mice were analyzed in Figure 2? How many mice were assessed in Figure 3? 

We thank the reviewer for thoughtful comments. We have already presented the detailed information in the manuscript, please refer to the “Methods-Quantification of retinal parameters” section for experimental details.

(7) I suggest adding into the methods whether P-values were corrected for multiple tests. 

We thank the reviewer for suggestion. Actually, the statistical analysis was performed using unpaired Student’s t-test for comparison between two groups or one-way ANOVA followed by Dunnett multiple comparison test for comparison of multiple groups. The above description was added to “Methods-Image acquisition and statistical analysis” section to make it clear.

Recommendations for the authors:

Reviewing Editor (Recommendations For The Authors): 

In summary, the following concerns should addressing reviewers' concerns as outlined below could bolster the evidence from "solid" to "convincing" and further strengthen the study's impact. 

(1) Analysis of the phenotype in CAPSLheterozygous mice, as highlighted by all 3 reviews. 

We thank the editor for thoughtful comments. The phenotype analysis of Capsl heterozygous mice was added to Fig.S4, with the corresponding description provided at page 6.

(2) Analysis of Capsl KO mice to determine if the pathways identified in vitro are modified (as suggested by reviewers 1 & 2). 

We thank the editor for suggestion. In Fig.S7, RT-qPCR was performed on lung tissues from Capsl Ctrl and KO mice to validate the expression of MYC targets in vivo. And the result indicated that the downstream targets of MYC signaling were also downregulated in vivo, consistent with the in vitro findings.

(3) Additional description of the genetic pedigrees and variants to address the points raised by reviewer #3. 

We thank the editor for suggestion. The father of family 3104 has also been identified as a carrier of this heterozygous variant, manifested with FEVR symptoms, according to the medical records. Nevertheless, clinical examination data are presently unavailable. We added corresponding description in the manuscript page 5.

(4) Validation of the identified protein variants, especially L83F which appears to be expressed at a near normal level. Are these proteins mislocalized, do the variants to interfere with sites of known or predicted protein-protein interactions, could they act in a dominant-negative fashion by aggregation with co-expressed WT protein etc. Given the comparatively weak genetic data, additional validation is required to establish plausibility of CAPSL as a FEVR gene. 

We thank the editor for suggestion. As substantial amount of p.L83F was detectable at normal molecular weight, we further investigated whether this variant affects protein localization. Fig.S1, immunocytochemistry results indicated that this variant does not affect the subcellular localization of the protein.

(5) Improved description of experimental details and statistical analyses as outlined by reviewer #3. 

We thank the editor for suggestion. The more detailed information about Capsl mice was added in the manuscript at page 14-15. The experimental details regarding Figure 2 and Figure 3 have already presented in the “Methods-Quantification of retina parameters” section in the manuscript at page 19-20. And the statistical analysis was performed using unpaired Student’s t-test for comparison between two groups or one-way ANOVA followed by Dunnett multiple comparison test for comparison of multiple groups. The above description was added to “Methods-Image acquisition and statistical analysis” section at page 21 to make it clear.

Reviewer #1 (Recommendations For The Authors): 

My reservations lie with the main assertion that CAPSL is associated with FEVR, as the genetic evidence from human studies appears relatively weak. My concerns are as follows: 

(1) The molecular characterization of the identified mutations suggests a loss of function (LOF). Notably, in one family, both the father and son exhibit the FEVR phenotype and share the LOF mutation, suggesting a dominant mode of inheritance. However, the prevalence of the LOF allele of CAPSL in the general population is high, and its pLI score is 0, according to the GNOMAD database. This raises doubts about the LOF variant of CAPSL being causative for FEVR. 

We thank the reviewer for recommendation. The pLI score of LOF allele of CAPSL is based of general population, among which Europeans account for ~77% and East Asians make up less than 3%. Since the FEVR families in this article all come from China, the pLI score may not be accurate. Of course, we will continue to collect FEVR pedigrees and screen for CAPSL mutations.

(2) In the conditional knockout study, a delay in vascular development is observed in the retina up to P14. What the phenotype looks like in adult mice and whether it replicates the human FEVR phenotype? 

We thank the reviewer for recommendation. We further assessed the phenotype when angiogenesis almost complete at P21, the resulted showed no difference in Ctrl and CapsliECKO/iECKO mice (Fig.S5). And corresponding description was added in the manuscript at page 7.

(3) The conditional knockout mice lack both alleles of CAPSL. The phenotype resulting from the knockout of a single allele needs investigation to align with observed human phenotypes and genetic data. 

We thank the reviewer for recommendation. The phenotype of Capsl heterozygous mice at P5 showed no overt difference in vascular progression, vessel density and branchpoints with littermate wildtype controls (Fig.S4). The lack of pronounced phenotype in FEVR heterozygous mice may be due to different sensitivity between human and mice. A similar example is LRP5 mutations associated with FEVR. Heterozygous mutations in LRP5 were reported in FEVR patients in multiple populations. However, heterozygous Lrp5 mice exhibited no visible angiogenic phenotype (PMID: 18263894).

(4) The MYC pathway has been identified as influenced by CAPSL. Whether MYC downregulation is observed in the mouse model in vivo? 

We thank the reviewer for recommendation. MYC expression was identified at both mRNA and protein level in Figure S8, and corresponding description was added in the manuscript at page 11.

Reviewer #2 (Recommendations For The Authors): 

Minor comments: 

(1) While authors note that little is known about CAPSL protein function, more introductory detail about the protein (structure, domains intracellular localization etc) and additional discussion on potential mechanisms would aid the reader in interpreting the findings and model.

We thank the reviewer for recommendation. The subcellular localization of the CAPSL protein is distributed in both the nucleus and cytoplasm (https://www.proteinatlas.org/). The immunochemistry analysis confirmed that CAPSL protein is expressed in both the cell nucleus and cytoplasm (Fig.S1). And corresponding description was added in the manuscript at page 5.

(2) Pg 7 states that Capsl knockout mainly leads to "...defects in retinal vascular ECs rather than other vascular cells.". Consider rephrasing to describe "other vasculature-associated cells", as no vascular cells outside the retina were examined in the manuscript. 

We thank the reviewer for recommendation. We rephrased the "...defects in retinal vascular ECs rather than other vascular cells." into "...defects in retinal vascular ECs rather than other vasculature-associated cells" at page 8.

(3) The manuscript is well written but contains numerous typos. E.g. "" (Pg 14), "MCY signaling axis" (figure 6 legend), "shCAPAL" (figure 5 K). Please correct these, and search carefully for others. 

We are sorry for the careless mistakes we made, and we have checked the manuscript and correct these mistakes.

Reviewer #3 (Recommendations For The Authors): 

The following are somewhat grammatical, but significant issues, that I feel should be addressed before making the pre-print final: 

(1) Perhaps the largest issue with the manuscript to me is whether CAPSL is an interesting candidate (as stated repeatedly) or causative of FEVR. Within the scope of what is feasible, this is a challenging problem. Since the publication of the pre-print, it would be great if another group independently reported the detection of mutations specifically in FEVR patients. That lacking, meaningful additions to the manuscript that I'd recommend are the inclusion of a paragraph on caveats of the study and reporting the allele frequencies based on public databases. As the authors know the data better than anyone and will have invested thought into the implications, they are the ones best positioned to alert the field to the study's limitations - amongst them- the factors that might practically distinguish whether CAPSL is a candidate or cause.

We thank the reviewer for recommendation. We will collect more samples from FEVR families and screen for other mutation sites within the CAPSL gene in further studies.

(2) It is unclear why the modeling with mice did not attempt to recapitulate the observations in humans, i.e., why were heterozygotes for a ubiquitous knockout not studied? Any data with heterozygotes, or ubiquitous alleles (which would be easier to generate than the strain studied) should be shared in the manuscript. If no such data exists, this reviewer would find it a worthwhile new experiment to add, but it is appreciated that new experiments are sometimes beyond the scope of what is possible. At the least, this would be worthwhile to discuss in the requested caveats paragraph of the discussion. 

We thank the reviewer for recommendation. We evaluated the phenotype of Capsl heterozygous mice at P5, and the results showed no overt difference in vascular progression, vessel density and branchpoints with littermate wildtype controls (Fig.S4). The lack of pronounced phenotype in FEVR heterozygous mice may be due to different sensitivity between human and mice. For example, heterozygous Lrp5 mice exhibited no visible angiogenic phenotype (PMID: 18263894). Corresponding description was added in the manuscript at page 6.

(3) The statement in the Abstract "which provides invaluable information for genetic counseling and prenatal diagnosis of FEVR" should be toned down, better supported, or rephrased. This appears to be the 18th disease-associated gene for FEVR, with variants identified in 4 patients of the same ethnicity. In my opinion, the word "invaluable" is currently overstated. 

We thank the reviewer for recommendation. We have changed "which provides invaluable information for genetic counseling and prenatal diagnosis of FEVR" into "which provides valuable information for genetic counseling and prenatal diagnosis of FEVR" in the abstract.

(4) The transcriptomic and proteomic data should be deposited into a public repository and accession numbers added to the manuscript. 

We thank the reviewer for recommendation. We have uploaded the raw data of transcriptomic and proteomic to the Genome Sequence Archive (https://ngdc.cncb.ac.cn/gsa/browse/HRA010305) and the Genome Sequence Archive (https://www.ebi.ac.uk/pride/archive), respectively.

(5) The links to MYC are over-stated in the title "through the MYC axis", the abstract "CAPSL function causes FEVR through MYC axis", and the discussion "we demonstrated that the defects in CAPSL affect EC function by down-regulating the MYC signaling cascade". The links to MYC are entirely by association, there were no experiments testing that the transcriptomic and proteomic changes observed were determinative of the CAPSL-mediated phenotype. It seems appropriate to conjecture that these changes are important, but the above statements all need to be altered and conjectures need to be clearly identified as such. 

We are sorry to overstate the link between CAPSL-mediated phenotype and MYC axis in the abstract and discussion sections, and we have altered the statements in these sections to make it more logical. For example, we changed “This study also reveals that compromised CAPSL function causes FEVR through MYC axis, shedding light on the potential involvement of MYC signaling in the pathogenesis of FEVR.” into “This study also reveals that compromised CAPSL function causes FEVR may through MYC axis, shedding light on the potential involvement of MYC signaling in the pathogenesis of FEVR.” in the abstract. And in the discussion we changed “…cause FEVR through inactivating MYC signaling, expanding FEVR-involved signaling pathway and providing a potential therapeutic target for the intervention of FEVR” to “…cause FEVR may through inactivating MYC signaling, expanding FEVR-involved signaling pathway and providing a potential therapeutic target for the intervention of FEVR”.

(6) Finally, I suggest that the following grammatical issues in the pre-print be corrected before making the pre-print final: 

We have checked the manuscript and correct these mistakes.

(a) p2. Suggest rewriting the sentence "Nevertheless, the molecular mechanisms by which CAPSL regulates cell processes and signaling cascades have yet to be elucidated." The preceding sentences only state that CASPL is a candidate in another disease - the word "nevertheless" seems to reflect a logic that isn't described. 

We have checked the manuscript and correct these mistakes.

(b) p5. Please correct the grammar "We, generated an inducible" 

We corrected this mistake.

(c) p5. Suggest rephrasing "impairing CAPSL expression." The word "expression" is often used in reference to transcription. To avoid confusion, something such as "eliminating or reducing protein abundance" might be better. 

We corrected this mistake.

(d) p6. Please correct the grammar "As expected, the radial vascular growth, as well as vessel density and vascular branching, are dramatically reduced in..." - note subject-verb agreement issue 

We corrected this mistake.

(e) Figure 3 legend - correct "(A) Hyloaid vessels"

We corrected this mistake.

Welcome back. In this lesson, I'll be discussing how AWS has designed its global infrastructure. While AWS is a global cloud platform, it's actually a collection of smaller groupings of infrastructure connected by a global high-speed network. As solutions architects, we can leverage this to design resilient and highly available systems.

I'll introduce the infrastructure concepts of an AWS region, an AWS edge location, and an AWS availability zone. I'll also cover the different ways a service can be resilient: globally resilient, regionally resilient, and zone resilient. By the end of this lesson, you'll understand what all of that means. Let's get started.

At a global level, AWS has created its infrastructure platform as a collection of individual infrastructures located worldwide. The two types of deployment at this global level are AWS regions and AWS edge locations. A region in AWS doesn't directly map onto a continent or a country. It's an area selected by AWS with a full deployment of AWS infrastructure, including compute services, storage, database products, AI, analytics, and more.

AWS continuously adds regions. At the time of creating this lesson, these include Northern Virginia, Ohio, California, and Oregon in the US, Frankfurt, Ireland, London, and Paris in Europe, and Sao Paulo in South America, among others. Some countries have one region, while larger nations have multiple, depending on customer requirements and size. Regions are geographically spread, enabling solutions architects to design systems that can withstand global-level disasters.

When interacting with most AWS services, you are interacting with a specific region. For example, Amazon's Elastic Compute Cloud in Northern Virginia is separate from Elastic Compute Cloud in Sydney. AWS can only deploy regions as fast as business and local planning allow, so you might not always have a region in the same town or city as your customers. Therefore, AWS also provides edge locations. Edge locations are much smaller than regions and generally only have content distribution services and some types of edge computing. They are located in many more places than regions, useful for companies like Netflix that need to store content close to their customers for low latency and high-speed distribution.

Regions and edge locations are commonly used together. A large company like Netflix might run its infrastructure from multiple regions worldwide, but its content could be stored in many edge locations for faster delivery. For example, in Australia, there is an AWS region in Sydney. If a Netflix customer in Melbourne wants to stream a show, it could be streamed from an edge location in Melbourne, providing faster transfer and lower latency.

AWS has a site to visualize the global AWS network, showing far fewer regions than edge locations, all connected by high-speed networking links. As we go through the course, I'll teach you how to use this private AWS networking for efficient systems deployment in AWS.

Regions are presented within the AWS Console. For instance, in the EC2 area of the console, you must select a region. However, global services like IAM or Route 53 don't require region selection. Some services are individual deployments in each region, while others operate globally.

Regions have three main benefits for solutions architects. Firstly, each region is geographically separate, meaning that a problem in one region wouldn't affect others. This isolation provides fault tolerance and stability. Secondly, selecting a region provides geopolitical or governance separation, meaning your infrastructure is subject to the laws and regulations of the region it’s in. AWS commits that data placed in one region won't leave that region unless configured otherwise. Lastly, regions allow location control, enabling you to tune your architecture for performance by placing infrastructure close to your customers.

Inside regions are availability zones (AZs), which are isolated infrastructure components within a region. Each region has multiple AZs, which can be two, three, four, five, or even six. In Sydney, there are three: ap-southeast-2a, 2b, and 2c. AZs are isolated compute, storage, networking, power, and facilities within a region. If an AZ fails, services in other AZs within the region remain functional. Solutions architects can design systems to distribute components across multiple AZs for resilience.

An availability zone could be one data center or part of multiple data centers, and AWS does not provide visibility into what constitutes an AZ. Services can be placed across multiple AZs to make them resilient.

Finally, let's define the resilience of an AWS service. There are three resilience levels: globally resilient, regionally resilient, and AZ resilient. Globally resilient services operate worldwide with a single database, replicating data across multiple regions. Examples include IAM and Route 53. Regionally resilient services operate in a single region with data replicated across multiple AZs in that region, such as RDS databases. AZ resilient services run from a single AZ, and if that AZ fails, the service fails.

As solutions architects, understanding the resilience level of each AWS service is crucial. This knowledge will help you answer exam questions and become an effective solutions architect.

These concepts are simple but fundamental. Ensure you understand what a region, edge location, and AZ are, and how services are globally, regionally, or AZ resilient. If needed, rewatch this video. When you're ready, I'll see you in the next lesson.

Welcome back. In this lesson, I want to cover the architecture of public AWS services and private AWS services. This is foundational to how AWS works, from a networking and security perspective. The differences might seem tiny, but understanding them fully will help you grasp more complex network and security products or architectures throughout your studies.

AWS services can be categorized into two main types: public services and private services. If you don’t have much AWS experience, you might assume that a public service is accessible to everyone, and a private service isn't. However, when you hear the terms AWS private service and AWS public service, it’s referring to networking only. A public service is accessed using public endpoints, such as S3, which can be accessed from anywhere with an internet connection. A private AWS service runs within a VPC, so only things within that VPC, or connected to that VPC, can access the service. For both, there are permissions as well as networking. Even though S3 is a public service, by default, an identity other than the account root user has no authorization to access that resource. So, permissions and networking are two different considerations when talking about access to a service. For this lesson, it's the networking which matters.

When thinking about any sort of public cloud environment, most people instinctively think of two parts: the internet and private network. The internet is where internet-based services operate, like online stores, Gmail, and online games. If you're at home playing an online game or watching training videos, you’re connecting to the internet via an internet service provider. So this is the internet zone. Then we have the private network. If you’re watching this video from home, your home network is an example of a private network. Only things directly connected to a network port within your house or people with your WiFi password can operate in your personal, private network zone.

AWS also has private zones called Virtual Private Clouds (VPCs). These are isolated, so VPCs can't communicate with each other unless you allow it, and nothing from the internet can reach these private networks unless you configure it. Services like EC2 instances can be placed into these private zones and, just like with your home network, it can only access the internet, and the internet can only access it if you allow and configure it.

Many people think AWS is architected with just two network zones: the internet and private zones. But there's actually a third zone: the AWS public zone, which runs between the public internet and the AWS private zone networks. This is not on the public internet but connected to it. The distinction might seem irrelevant, but it matters as you learn more about advanced AWS networking. The AWS public zone is where AWS public services operate, like S3.

To summarize, there are three different network zones: the public internet, the AWS private zone (where VPCs run), and the AWS public zone (where AWS public services operate). If you access AWS public services from anywhere with a public internet connection, your communication uses the public internet for transit to and from this AWS public zone. This is why you can access AWS public services from anywhere with an internet connection because the internet is used to carry packets from you to the AWS public zone and back again.

Later in the course, I will cover how you can configure virtual or physical connections between on-premises networks and AWS VPCs, allowing private networks to connect together if you allow it. You can also create and attach an internet gateway to a VPC, allowing private-zone resources to access the public internet if they have a public IP address. This also allows access to public AWS services like S3 without touching the public internet, communicating through the AWS public zone.

Private resources, such as EC2 instances, can be given a public IP address, allowing them to be accessed from the public internet. Architecturally, this projects the EC2 instance into the public zone, enabling communication with the public internet. Understanding the three different network zones—the public internet, the AWS public zone, and the AWS private zone—is crucial for doing well in the real world and in specific exams. These three network zones become critical as you learn more advanced networking features of AWS.

That’s everything for this lesson. Complete the video, and when you’re ready, I’ll look forward to you joining me in the next lesson.

Reach out to Peace of Mind Pest Control in Stockton, CA! We're here to kick off your pest control journey with a thorough inspection and ensure we find the perfect treatments for you and your property.

Reviewer #1 (Public Review):

Summary:<br /> The authors analyzed how biotic and abiotic factors impact antagonistic host-parasitoid interaction systems in a large BEF experiment. They found the linkage between the tree community and host-parasitoid community from the perspective of the multi-dimensionality of biodiversity. Their results revealed that the structure of the tree community (habitat) and canopy cover influence host-parasitoid compositions and their interaction pattern. This interaction pattern is also determined by phylogenetic associations among species. This paper provides a nice framework for detecting the determinants of network topological structures.

Strengths:<br /> This study was conducted using a five-year sampling in a well-designed BEF experiment. The effects of the multi-dimensional diversity of tree communities have been well explained in a forest ecosystem with an antagonistic host-parasitoid interaction.

The network analysis has been well conducted. The combination of phylogenetic analysis and network analysis is uncommon among similar studies, especially for studies of trophic cascades. Still, this study has discussed the effect of phylogenetic features on interacting networks in depth.

Weaknesses:<br /> (1) The authors should examine species and interaction completeness in this study to confirm that their sampling efforts are sufficient.<br /> (2) The authors only used Rao's Q to assess the functional diversity of tree communities. However, multiple metrics of functional diversity exist (e.g., functional evenness, functional dispersion, and functional divergence). It is better to check the results from other metrics and confirm whether these results further support the authors' results.<br /> (3) The authors did not elaborate on which extinction sequence was used in robustness analysis. The authors should consider interaction abundance in calculating robustness. In this case, the author may use another null model for binary networks to get random distributions.<br /> (4) The causal relationship between host and parasitoid communities is unclear. Normally, it is easy to understand that host community composition (low trophic level) could influence parasitoid community composition (high trophic level). I suggest using the 'correlation' between host and parasitoid communities unless there is strong evidence of causation.

Reviewer #2 (Public Review):

Summary:<br /> In their manuscript, Multi-dimensionality of tree communities structure host-parasitoid networks and their phylogenetic composition, Wang et al. examine the effects of tree diversity and environmental variables on communities of reed-nesting insects and their parasitoids. Additionally, they look for the correlations in community composition and network properties of the two interacting insect guilds. They use a data set collected in a subtropical tree biodiversity experiment over five years of sampling. The authors find that the tree species, functional, and phylogenetic diversity as well as some of the environmental factors have varying impacts on both host and parasitoid communities. Additionally, the communities of the host and parasitoid showed correlations in their structures. Also, the network metrices of the host-parasitoid network showed patterns against environmental variables.

Strengths:<br /> The main strength of the manuscript lies in the massive long-term data set collected on host-parasitoid interactions. The data provides interesting opportunities to advance our knowledge on the effects of environmental diversity (tree diversity) on the network and community structure of insect hosts and their parasitoids in a relatively poorly known system.

Weaknesses:<br /> To me, there are no major issues regarding the manuscript, though sometimes I disagree with the interpretation of the results and some of the conclusions might be too far-fetched given the analyses and the results (namely the top-down control in the system). Additionally, the methods section (especially statistics) was lacking some details, but I would not consider it too concerning. Sometimes, the logic of the text could be improved to better support the studied hypotheses throughout the text. Also, the results section cannot be understood as a stand-alone without reading the methods first. The study design and the rationale of the analyses should be described somewhere in the intro or presented with the results.

eLife assessment

This study provides important new insights into the contribution of local DNA features to the molecular mechanisms and dynamics of copy number variation (CNV) formation during adaptive evolution. While limited to a single CNV, the experiments are carefully controlled and present convincing evidence that supports the conclusions. This work will be of general interest to those studying genome architecture and evolution from yeast biologists to cancer researchers.

Reviewer #1 (Public Review):

Summary:

The work by Chuong et al. provides important new insights into the contribution of different molecular mechanisms in the dynamics of CNV formation. It will be of interest to anyone curious about genome architecture and evolution from yeast biologists to cancer researchers studying genome rearrangements.

Strengths:

Their results are especially striking in that the "simplest" mechanism of GAP1 amplification-non-allelic homologous recombination between the flanking Ty-LTR elements is not the most common route taken by the cells, emphasizing the importance of experimentally testing what might seem on the surface to be obvious answers. One of the important developments of their work is the use of their neural network simulation-based inference (nnSBI) model to derive rates of amplicon formation and their fitness effects.

Weaknesses:

The manuscript reads as though two different people wrote two different sections of the manuscript - an experimental evolutionist and a computational scientist. If the goal is to reach both groups of readers, there needs to be more explanation of both types of work. I found the computational sections to be particularly dense but even the experimental sections need clearer explanations and more specific examples of the rearrangements found. I will point out these areas in the detailed remarks to the authors. While I have no reason to question their conclusions, I couldn't independently verify the results that ODIRA was the majority mechanism since the sequence of amplified clones was not made available during the review. I've encouraged the authors to include specific, detailed sequence information for both ODIRA events as well as the specific clones where GAP1 was amplified but the flanking gene GFP was not.

Reviewer #2 (Public Review):

Summary:

This study examines how local DNA features around the amino acid permease gene GAP1 influence adaptation to glutamine-limited conditions through changes in GAP1 Copy Number Variation (CNV). The study is well motivated by the observation of numerous CNVs documented in many organisms, but difficulty in distinguishing the mechanisms by which they are formed, and whether or how local genomic elements influence their formation. The main finding is convincing and is that a nearby Autonomous Replicating Sequence (ARS) influences the formation of GAP1 CNVs and this is consistent with a predominate mechanism of Origin Dependent Inverted Repeat Amplification (ODIRA). These results along with finding and characterizing other mechanisms of GAP1 CNV formation will be of general interest to those studying CNVs in natural systems, experimental evolution, and in tumor evolution. While the results are limited to a single CNV of interest (GAP1), the carefully controlled experimental design and quantification of CNV formation will provide a useful guide to studying other CNVs and CNVs in other organisms.

Strengths:

The study was designed to examine the effects of two flanking genomic features next to GAP1 on CNV formation and adaptation during experimental evolution. This was accomplished by removing two Long Terminal Repeats (LTRs), removing a downstream ARS, and removing both LTRs and the ARS. Although there was some heterogeneity among replicates, later shown to include the size and breakpoints of the CNV and the presence of an unmarked CNV, both marker-assisted tracking of CNV formation and modeling of CNV rate and fitness effects showed that deletion of the ARS caused a clear difference compared to the control and the LTR deletion.

The consequence of deletion of local features (LTR and ARS) was quantified by genome sequencing of adaptive clones to identify the CNV size, copy number and infer the mechanism of CNV formation. This greatly added value to the study as it showed that i) ODIRA was the most common mechanism but ODIRA is enhanced by a local ARS, ii) non-allelic homologous recombination (NAHR) is also used but depends on LTRs, and iii) de novo insertion of transposable elements mediate NAHR in strains with both ARS and LTR deletions. Together, these results show how local features influence the mechanism of CNV formation, but also how alternative mechanisms can substitute when primary ones are unavailable.

Weaknesses:

The CNV mutation rate and its effect on fitness are hard to disentangle. The frequency of the amplified GFP provides information about mutation rate differences as well as fitness differences. The data and analysis show that each evolved population has multiple GAP1 CNV lineages within it, with some being unmarked by GFP. Thus, estimates of CNV fitness are more of a composite view of all CNV amplifications increasing in frequency during adaptation. Another unknown but potential complication is whether the local (ARS, LTR) deletions influence GAP1 expression and thus the fitness gain of GAP1 CNVs. The neural network simulation-based inference does a good job at estimating both mutation rates and fitness effects, while also accounting for unmarked CNVs. However, the model does not account for the population heterogeneity of CNVs and their fitness effects. Despite these limitations of distinguishing mutation rate and fitness differences, the authors' conclusions are well supported in that the LTR and ARS deletions have a clear impact on the CNV-mediated evolutionary outcome and the mechanism of CNV formation.

Reviewer #3 (Public Review):

Summary:

The authors represent an elegant and detailed investigation into the role of cis-elements, and therefore the underlying mechanisms, in gene dosage increase. Their most significant finding is that in their system copy number increase frequently occurs by what they call replication errors that result from the origin of replication firing.

The authors somewhat quantitatively determine the effect of the presence of a proximal origin of replication or LTR on the different CNV scenarios.

Strengths:

(1) A clever and elegant experimental design.

(2) A quantitative determination of the effect of a proximal origin of replication or LTR on the different CNV scenarios. Measuring directly the contribution of two competing elements.

(3) ODIRA can occur by firing of a distal ARS element.

(4) Re-insertion of Ty elements is interesting.

Weaknesses:

(1) Overall, the research does not considerably advance the current knowledge. The research does not investigate what the maximum distance between ARS for ODIRA is to occur. This is an important point since ODIRA was previously described. A considerable contribution to the field would be to understand under what conditions ODIRA wins NAHR.

(2) The title and some sentences in the abstract give a wrong impression of the generality and the novelty of the observations presented. Below are some examples of much earlier work that dealt with mechanisms of CNV and got different conclusions. The Lobachev lab (Cell 2006) published a different scenario years ago, with a very different mechanism (hair-pin capped breaks). The Argueso lab found something different (NAHR) (Genetics 2013).

In fact, the CUP1 system presents a good example of this point. The Houseley group showed a complex replication transcription-based mechanism (NAR 2022, cited), the Argueso group showed Ty-based amplification and the Resnick group showed aneuploidy-based amplification. While aneuploidy is a minor factor here the numerous works in Candida albicans, Cryptococcus neoformans, and Yeast suggest otherwise (Selmecki et al Science 2006, Yona et al PNAS 2013, Yang et al Microbiology Spectrum 2021).

(3) The authors added a mathematical model to their experimental data. For me, it was very difficult to understand the contribution of the model to the research. I anticipated, for example, that the model would make predictions that would be tested experimentally. For example, " ARSΔ and ALLΔ are predicted to be almost eliminated by generation 116, as the average predicted WT proportion is 0.998 and 0.999" But to my understanding without testing the model.