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

In this manuscript, the authors interrogated single cells of yeast as they developed into quiescence after the natural depletion of glucose from the culture medium. To do so, they constructed a microfluidic platform to track individual cells in a batch culture of cells transitioning from a growing phase into stationary phase. They then used a number of assays to monitor the metabolic changes that accompany this transition. They observed that internal pH dropped during development of quiescence, with some cells showing a rapid drop and others showing a delayed and heterogeneous drop. At diauxie, cells transition from fermentation to respiration, and the minority of cells that showed a rapid drop in pH were respiration-deficient (R-) and unable to resume growth, while cells with a delayed pH drop were respiration proficient (R+) and able to resume growth. Using established markers to follow the previously described structural changes that accompany the development of quiescence, they found that the pH changes were temporally related to these structural changes. They suggest that the dynamic changes in intracellular pH promote waves of structural remodeling that eventually leads to a transition of cells to a gel-like state. The early drop in pH in R- cells was proposed to lead to a precocious transition of the cytoplasm, contributing to the inability of these cells to resume growth.

This study is well done, well-written, and the results are clearly presented and generally convincing. Data accumulation and analysis were well documented and appropriate statistics were used. While the authors provided details of the materials used to construct the microfluidic device, it would be appropriate for them to provide a detailed blueprint (and a video, for example) to other investigators who would like to employ this device in their studies of quiescence once this manuscript has been published.

Significance

The cell fate program that is set in motion as yeast cells transition from fermentation to respiration is still not well understood. The development of the microfluidic platform described in this manuscript could make a significant contribution to our understanding of the succession of metabolic and structural changes occurring during this transition. The Sagot lab has made a series of important contributions in this area, and the application of single cell tracking to monitor the temporal program of these changes represents a major technical advance that will be of general interest to researchers interested in defining the developmental programs that contribute to cellular quiescence and longevity.

Expertise: yeast chromatin biology

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

These investigators have teamed up to solve a technical problem that has thwarted efforts to get a clear picture of the chronology of events in yeast cultures as they naturally exhaust their nutrient supply. This is a challenge because the time course is long and the density of the culture makes single cell analysis problematic. Previous studies involving abrupt starvation have shown that there is a pH drop when nutrients are eliminated, but abrupt starvation also leads to rapid loss of viability compared to what is observed with cells as they respond and adapt to changes in their environment. This microfluidic devise and an intracellular pH detector allowed them to follow pH change as cells transition from fermentation to respiration and stationary phase. About 15% of the population responds completely differently than the other 85%, making this single cell analysis crucial. It also provides a negative control of sorts, to further substantiate the correlations they draw. This 15% fails to enter the respiratory phase and dies rapidly. The pH also drops rapidly and is correlated with loss of mitochondrial function and aggregation of proteins. The 85% of cells that succeed in shifting to respiration suffer the same pH drop, but it is much slower and is correlated with slower protein aggregation, P Body, actin body, and proteosome storage Granule assembly. They also followed the cytoplasmic transition to a glassy state, based on the mobility of protein foci and lipid droplets. This transition occurs at the same pH in both populations but with completely different timing. This recapitulates the transition observed after abrupt starvation. It shows that the same transition occurs in viable, quiescent cells and provide further evidence that it is correlated with pH changes.

The only concern I have is that they used only one strain, which reduces the universality of their findings. Moreover, it is ambiguous which strain was used. The strain table says they used S288c which is known to carry a hap1 mutation that compromises respiration and isn't the best choice for studying respiring cells. The text mentions that they are working in the BY background, (where most of their GFP studies have been carried out. The BY strains have the same hap1 mutation and several other unknown polymorphisms that prevent ethanol utilization and biomass increase after the diauxic shift.

Significance

This kind of single cell analysis is clearly the way forward and will have many further applications to understanding how cells adapt to their environment. The paper is well written and the figures are well laid out and easy to understand. It is a significant advance for the field and will set the bar for future experiments. However, this work was done with a single strain that is known to be defective in respiration. It would be extremely valuable to know if their results with this strain are generalizable to other lab and wild strains.

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

Learn more at Review Commons

---

Reply to the reviewers

Response/revision plan

(Point-by-point response)

---

Reviewer #1 (Evidence, reproducibility and clarity (Required)):

The manuscript by Pennauer et al is the first to systematically investigate the role of class I&II Arfs using a knockout approach. It builds on earlier work by the Kahn lab who used an RNAi approach (Volpicelli-Daley et al. 2005) and is complementary to the overexpression approach used by the Hauri lab (Ben-Tekaya et al, 2010). The work is elegant and the data are strong. I am strongly in favor of publishing this work and my comments are technical in nature (2-5) and a request for some text changes (1). have the following comments for improvements:

1- When it comes to evaluating the role of depletions of Arfs on cell fitness, it would be better to use a non-transformed cell line. I am not asking the authors to go through the painstaking process of generating knockout cell lines in RPE1 cells for instance. Rather, I suggest that the authors make the reader aware that conclusions about cell survival have to be taken with care due to the use of a transformed cell line.

We will add this valid point to the Discussion.

2- Why do Arf1 and Arf4 ko cells grow more slowly. Is it a higher rate of cell death? Is it a block in a certain phase of the cell cycle. Given the link of the Golgi to G2-M entry, I think that an analysis of the cell cycle distribution would add more depth to these data. If the cell cycle distribution is unaffected, then I would conclude that that the difference in doubling time are due to reduced cell survival. If there is an effect on the cell cycle distribution, then the conclusion of the authors is safe that no single Arf is required for survival

We plan to analyze cell cycle distribution.

3- It is not clear to me how many cells were quantified in Figure 2D-F. I suppose that each dot represents a cell. In this case, the number of cells quantified is a bit low. Such a quantification of fluorescence intensities in two channels in the same region is a simple task and I think it should be no problem obtaining at least 100 cells per condition.

We will add the number of cells analyzed to the figure legends: At least 40 Golgis were quantified in each experiment. thus >100 in total.

4- Is the drop in the ratio of beta-COP/GM130 in Arf1 depleted cells reflecting reduced recruitment to the Golgi? Because the Golgi is bigger, it might be reflecting a reduced density in the number of coatomer molecules per surface area. If it is due to reduced recruitment, then the ratio of membrane/cytosolic betaCOP should be altered. This of course requires to show that the knockout does not affect total levels of coatomer. I think that such fractionation experiments would be a valuable addition to the manuscript and increase the depth of the data.

We are currently performing immunoblot analysis to determine bCOP levels.

In the Figure below, we have plotted the total intensity of GM130 or bCOP per Golgi from our immunoflurescence data. Total intensity of GM130 significantly increased in the cell lines lacking Arf1, consistent with the increase in Golgi volume. The amount of bCOP at the Golgi remained constant, resulting in reduced bCOP/GM130 ratio. Deletion of Arf1 thus results in reduced rate of coat recruitment that is compensated by an increase in Golgi mass. In the simplest model, reduced formation of Golgi-exit carriers causes Golgi growth until exit carrier formation allows for the required flux.

We propose to include this data in the revised manuscript.

FIGURE

5- The finding that Arf4-ko cells exhibit a defect on retrieval of ER-resident proteins is exciting, and in my opinion, it is the most significant finding in this manuscript. How can this be reconciled with the lack of an ARf4 ko effect on coatomer recruitment to the Golgi. Looking carefully at the data, I see that in 2 out of 3 experiments, Arf4 ko reduced the betaCOP/GM130 ratio. This is why I think it is crucial to perform more experiments and add more cells to increase the confidence in the data. Reduced retrieval of ER chaperones is frequently found in tumors and we still don't understand the reason behind this. Therefore, this finding is of significance beyond the community of cell biologists.

We plan to repeat quantitation with COPI for better statistical validity.

6- I find Figure 6A confusing. Why do Arf1 overexpressing parental HeLa cells exhibit less Arf1 than control cells?

In order not to overload the immunoblot of Arf overexpressing lysates, a smaller aliquot (1/20) was loaded. We will indicate this directly below the blots to make this more obvious in the revised figure.

7- Why was the following condition not tested: Arf4ko cells with Arf1 overexpression. Given the importance of Arf1 in retrograde (Golgi-to-ER) trafficking, I would expect a partial rescue of the retrieval of ER chaperones.

We will to do this experiment.

Reviewer #1 (Significance (Required)):

**Significance of the work:**

The paper is important because it is the first to examine the role of Arfs using a knockout approach. Another very important finding is that Arf4 depleted cells exhibit problems with retrieval of ER chaperones. This is a very novel finding and to the best of my knowledge

**Audience:**

The primary audience is of course the community working on membrane trafficking, organelle biology and proteostasis. However, I think that the data on the role of Arf4 in retrieval of ER chaperones might be of relevance for cancer biologists. Secretion of ER chaperones is frequently found in many tumors and we still do not understand why this is happening and what the significance thereof is.

**My own expertise:**

Export from the endoplasmic reticulum Golgi fragmentation in cancer cell migration Rho GTPases Kinase signaling Pseudoenzymes Cell migration of breast cancer cells Proteostasis in multiple myeloma

**Referee Cross-commenting**

Just a follow-up comment from my side:

I agree that it has not been unequivocally established that Arf1 is the main/sole of retrograde transport. However, even less established is the role of Arf4 in this process. The authors show that it is mainly Arf1 depletion that reduces the amount of COPI at the Golgi (ratio of COPI/GM130). Thus, I remain very surprised that it is actually the Arf4 depletion that results in reduced retrieval.

What is the significance of having less COPI at the Golgi in Arf1-ko cells? Certainly, the Golgi is not more "leaky". Does the level of COPI at the Golgi not reflect the strength of retrograde trafficking? Maybe there is no less COPI at the Golgi, and it only appears to be less, because the Golgi is bigger. This is why a simple fractionation experiment would be good. Something like making a cytosol and a microsome fraction and looking at the ratio of COPI (Cyt/Mem).

If both reviewers think it is too much, or unlikely to work, then I am happy to drop this point.

Below are my comments to the evaluations by the other two reviewers:

1- I agree with most comments that the two other reviewers made. Some of them are actually overlapping with mine (e.g. the use of a cell line other than HeLa).

2- I am not sure whether the impact of the paper would improve by adding data on Arf6.

3- To the comment on Golgi polarity. Maybe we could be more specific here and say that it would be sufficient to show that a trans-Golgi protein and a cis-Golgi protein can be separated by fluorescence microscopy, or whether we alternatively want them to actually do it by immunogold labeling for EM (which is more difficult).

4- I agree with reviewer 2 that the work proposed needs 1-3 months. I think reviewer 3 is a bit too optimistic with 1 month, because her/his comment on using a cell line other than HeLa cannot be addressed in just a month.

---

Reviewer #2 (Evidence, reproducibility and clarity (Required)):

Pennauer uses HeLa cells and CRISPR/Cas9 to delete the 5 members of the class I and class II ARF family of small regulatory GTPases either individually or in combinations. The characterization of the KO cells is excellent and convincingly demonstrates that true KOs were generated. The quality of the data presented is high. Using the KO cells she documents minor alterations in Golgi architecture and the recruitment of vesicular coats in cells deleted of all ARFs except ARF4. In contrast, there is a significant lack of retention/recycling to the ER of KDEL-containing ER proteins in ARF4 KO cells, with numerous ER chaperones now released into the medium (the ARF4 KO secretome). This is a well-done study that showcases the ability of ARF4 alone to sustain cellular life (quite a surprise to this reviewer). Yet, the characterization of the phenotypes is somewhat minimal and the conclusions would be more robustly supported by additional experiments. Specifically:

1. The authors completely ignore class III ARF6 and this paper would be much more comprehensive and informative if analysis of that ARF was also included (ARF6 has been seen at the Golgi and also mediates endosomal trafficking that intersects with the TGN).

In agreement with the reviewers' consensus in cross-commenting, we consider Arf6ko to be beyond the scope of this study.

Although the overall Golgi architecture seems to be largely conserved, it remains essential to test whether Golgi polarity is similarly maintained, and such data would significantly expand the significance of the reported findings

We have performed super-resolution microscopy of wild-type and Arf1ko Golgis for GM130 and TGN46 as cis- and trans-Golgi markers, respectively, showing that polarity is still intact for Arf1ko, the morphologically most affected knockout cell line. We plan to include the following Figure in the revised manuscript.

FIGURE

Golgi complexes were imaged by superresolution microscopy for GM130 (green) and TGN46 (red), and displayed as maximum intensity projections, or tomographic 2D slices. Scale bar, 3 μm.

Since there is a defect in retrieval of KDEL-proteins, it would be important to show the intracellular localization of the KDEL-R in the cells (especially in the ARF4 KO cells that don't retrieve KDEL-GFP) - is the receptor degraded, stuck in some specific place - knowing that would increase the impact of this study and provide a mechanistic explanation for the observed phenotype

We plan to perform immunoblot analysis for KDELR to test for changes in levels in Arf deletion cells, and immunofluorescence microscopy to analyze changes in KDELR localization.

The rescue experiments in Figure 6 are good as far as they go, but this experiment would be much more informative if in addition to the same class rescue, the other class ARFs (at least one!) were also characterized.

We will to do this experiment.

This is maybe a little too much to ask, but since the authors propose a mechanistic explanation for the ARF4 KO KDEL phenotype as being due to different effectors recruited by this ARF (in this case different COPI isotypes - this study would increase in impact by actually testing this mechanisms by assessing whether ARF4Q71L mutant preferentially bound any particular isotype of COPI or even try to do mass spec to identify relevant effectors for this extremely interesting ARF.

We also think that this additional analysis is beyond the scope of this study.

The Discussion is a very limited and would be more impactful by adding some discussion of organismal effects of ARF deletions (many are embryonic lethal while cells seems to live quite happily) or mutations (links to cancer come to mind here), as well as some mention of data from yeast ARF (what is and isn't essential in those cells). As is, the authors miss an opportunity to highlight the importance of their findings as they relate to current knowledge of ARFology.

We agree to add a discussion of information on embryonic lethality and disease.

Reviewer #2 (Significance (Required)):

This is an important paper for the ARF field and people interested in ARF signaling will be glad to read about the findings and perhaps also use the developed KO cell lines - this is a significant advancement. The impact would be even higher if some of the experiments suggested above were incorporated into the manuscript.

---

Reviewer #3 (Evidence, reproducibility and clarity (Required)):

This paper describes the application of CRISPR/Cas9 to systematically delete from HeLa cells all four Arf genes, either singly or in combination. The authors find it is possible to generate a number of double deletions (notably one lacking both Class I Arfs), and a triple deletion lacking all but Arf4. The authors characterise the structure of the Golgi in these mutants as well as retention of ER residents. The work is a comprehensive study of an exceptionally high technical standard. There is excellent validation of the deletions, and then the application of a wide range of methods including immunofluorescence, electron microscopy and mass-spectrometry, all with careful and extensive quantitation. The finding that cells can survive without Class I Arfs is interesting and unexpected, as is the fact that Arf4 alone is sufficient. This work will provide an excellent platform for future studies on Arf protein function in human cells. There are of course many questions that arise from these findings, but given the scope and quality of the work they would seem better left for future publications. There is one experiment that could be added, and some additions needed to the text for clarity and minor adjustments to the figures (all listed below), but if these are addressed, this would be a high quality paper of wide audience to a cell biological audience.

**Specific comments:**

1) Have the authors tested the levels and/or localisation of the KDEL receptor in the various lines? This is not essential, but if it were easily done, it would add to the work on ER resident secretion.

We plan to perform immunoblot analysis for KDELR to test for changes in levels in Arf deletion cells, and immunofluorescence microscopy to analyze changes in KDELR localization.

2) The work is entirely done in HeLa cells. The authors should note that the situation might be different in other cells types and cell lines. For instance, the DepMap CRISPR database suggests that quite a lot of cell lines are strongly affected by loss of Arf1.

We agree to add a discussion on known effects in other tissues.

3) Figure 2. Please show single channels as grey scale, and only merge as RGB. This is easier to see, especially for the colour blind. Likewise, Figure 3D would be clearer in greyscale rather than green, and 6B better in grey than in red.

We will make these changes.

4) Figure 5C. A brief comment is needed as to why it might be that BiP and calreticulin are not so efficiently secreted when Arf5 is knocked out in addition to Arf4.

This was a mistake in labeling that lane and will be corrected. It should read "Arf3+5ko" not "Arf4+5ko. Thank you for pointing this out.

5) Discussion:

a) The authors should relate these studies to work in other species. For instance, in yeast reduction of Arf levels causes the Golgi to enlarge (PubMed ID 9487133).

We can discuss this.

1. b) Some more discussion is needed of the fact that Arfs may not all act in the same part of the Golgi, which could explain some of the differences observed between the various deletions.

We can add this point in the discussion.

Reviewer #3 (Significance (Required)):

The Arf GTPases have been studied extensively for over 30 years as major regulators of Golgi function. They are essential for the recruitment to Golgi membranes of both COPI and clathrin/AP-1 coats, as well as various other proteins that regulate Golgi function. In addition, they have been reported to have roles in viral replication, and even other cellular processes such as lipid droplet formation and mitochondrial division. In humans there are four Arfs, Arf1 and Arf3 (Class I Arfs), and Arf4 and Arf5 (Class II Arfs). All are present on the Golgi, but their precise individual roles have remained unclear. Attempts have been made to deplete individual Arfs using RNAi, but incomplete knockdowns have made the results hard to interpret.

**Referee Cross-commenting**

There is probably no need for a prolonged debate about this, but I agree that the importance of Arf4 is striking, but it reflects the nature of this work that CRISPR has finally allowed these sorts of questions to be addressed unequivocally. COPI is also involved in recycling of Golgi resident enzymes, and it may be that Arf1 acts in this role.

If the authors check levels of COPI by blotting, and measure the intensity over the Golgi by quantitative IMF, that will reveal whether stability or membrane association if affected without fractionation which is probably less reliable.

If they want to do some extra experiments, then it would be quite easy to check the levels of some Golgi enzymes, or look at lectin binding as a proxy for glycosylation enzyme levels.

Overall, I agree with the positive comments of Reviewers 1 and 2, and it good that we all recognise the quality and importance of the work. However, I feel that one or two of their requests go beyond the scope of a single publication, or would add rather little for a lot of additional work. It is of course easy to propose experiments that someone else has to do!

**[On] Reviewer 1:**

Point 4. I agree that it would be useful to perform a blot to determine if the levels of coatomer are effected in the various KO lines. I am not sure if Reviewer 1 is also proposing fractionation to determine cytosol vs membrane ratio, but if so, then this would be less useful as peripheral membrane proteins tend to fall off membranes during fractionation and so such analysis is generally questionable. A blot, and clarification of the way the COPI/GM130 ratio is determined, would answer the key points in a relatively straightforward way.

Point 5. I agree that the defect in retrieval of ER residents in Arf4-KO is striking, but it a clear effect even if the reviewer does not understand it themselves! It does not seem so surprising to me, given that Arf4 is likely to act on the early Golgi were such retrieval occurs from. However, the experiment suggested by myself and Reviewer 32 of checking the levels and localisation of the KDEL-receptor would seem to me a good first step to addressing possible mechanism, and certainly sufficient for an initial publication.

Point 7. I am not sure that it has been unequivocally established that Arf1 is important in retrograde traffic. The reality is that many labs have taken Arf1 as being representative of all others and so concentrated biochemical and in vivo studies on this protein. This paper is really important as it highlights the need to investigate both Class I and Class II Arfs, and to bear in mind that their roles in vivo may well be more distinct than their in vitro properties would lead one to suspect. Perhaps, the simplest explanation for this is that the GEFs that activate them have a strong preference for one or the other.

Follow up Comment 1. I was not suggesting that the authors repeat all this in a cell line other than HeLa cells, as this is clearly impractical. HeLa cells are widely used, and so the findings are useful, and whilst it seems certain that some other cell lines would give different data (and indeed the DepMap data show this), then testing one other line would not change the conclusions much. All I wanted the authors to do is to clearly state in the text that what they see in HeLa cells may well be different in other cell lines. This does not detract from the fact that their HeLa cells will provide an excellent platform for focused studies on the role of individual Arfs.

Follow up Comment 2. I agree that Arf6 is not relevant to this paper (as discussed in detail below).

Follow up Comment 3. agree that a simple IMF experiment would suffice to check polarity and immuno-EM is technically very demanding and would add little in this context. The authors have already shown that the Golgi forms stacks in the KO cell lines, and I cannot see how this could occur without the stack being polarised - it has to form at one end and then mature to the other. In addition, after decades of working on the Golgi I have yet to see a credible report of a change to cells causing a loss of Golgi polarity, but maintaining a stacked structure. If the Golgi is not polarised it could not form a stack.

Follow up Comment 4. I agree that one month is perhaps too short to look at KDEL-R, COPI levels and checking polarity by IMF. As noted above, I am NOT suggesting that they repeat all this in a different cell line.

**[On] Reviewer 2.**

Point 1. I agree with Reviewer 1 that the authors are correct to ignore Arf6. It is a completely different GTPase with a distinct function in a different part of the cell. The family of Arf1-Arf5 arose in metazoans from a single Arf, but Arf6 had already split away from the Arf1-5 family in the last eukaryotic common ancestor, as Arf6 is present in plants and yeasts. There is overwhelming evidence that Arf1-Arf5 are partially redundant and this has hampered their study. Arf6 does not share these roles. The fact that it is acts on endosomes and has been reported to be on the Golgi (which is not widely agreed), is also true of many other GTPases. Indeed, other distant relatives of Arf1-5 are actually on the Golgi (Arl1, Arl5 etc), but these are also not relevant as like Arf6 they do not bind coat proteins and other major effectors of Arf1-5.

Point 2. As noted above, it is hard to see how polarity could be affected given that a Golgi stack is formed, but, at most, a simple application of IMF would seem sufficient to confirm this.

Point 3. Agreed.

Point 5. I agree with the reviewer that this is (much!) too much to ask for an initial publication. Various labs have already reported analysis of the effectors of Class II Arfs and they tend to overlap with Class I. Moreover, it is quite possible that the difference of role in vivo reflects differing interactions with regulators.

Point 6. Agreed.

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

This paper describes the application of CRISPR/Cas9 to systematically delete from HeLa cells all four Arf genes, either singly or in combination. The authors find it is possible to generate a number of double deletions (notably one lacking both Class I Arfs), and a triple deletion lacking all but Arf4. The authors characterise the structure of the Golgi in these mutants as well as retention of ER residents. The work is a comprehensive study of an exceptionally high technical standard. There is excellent validation of the deletions, and then the application of a wide range of methods including immunofluorescence, electron microscopy and mass-spectrometry, all with careful and extensive quantitation. The finding that cells can survive without Class I Arfs is interesting and unexpected, as is the fact that Arf4 alone is sufficient. This work will provide an excellent platform for future studies on Arf protein function in human cells. There are of course many questions that arise from these findings, but given the scope and quality of the work they would seem better left for future publications. There is one experiment that could be added, and some additions needed to the text for clarity and minor adjustments to the figures (all listed below), but if these are addressed, this would be a high quality paper of wide audience to a cell biological audience.

Specific comments:

1) Have the authors tested the levels and/or localisation of the KDEL receptor in the various lines? This is not essential, but if it were easily done, it would add to the work on ER resident secretion.

2) The work is entirely done in HeLa cells. The authors should note that the situation might be different in other cells types and cell lines. For instance, the DepMap CRISPR database suggests that quite a lot of cell lines are strongly affected by loss of Arf1.

3) Figure 2. Please show single channels as grey scale, and only merge as RGB. This is easier to see, especially for the colour blind. Likewise, Figure 3D would be clearer in greyscale rather than green, and 6B better in grey than in red.

4) Figure 5C. A brief comment is needed as to why it might be that BiP and calreticulin are not so efficiently secreted when Arf5 is knocked out in addition to Arf4.

5) Discussion:

a) The authors should relate these studies to work in other species. For instance, in yeast reduction of Arf levels causes the Golgi to enlarge (PubMed ID 9487133).

b) Some more discussion is needed of the fact that Arfs may not all act in the same part of the Golgi, which could explain some of the differences observed between the various deletions.

Significance

The Arf GTPases have been studied extensively for over 30 years as major regulators of Golgi function. They are essential for the recruitment to Golgi membranes of both COPI and clathrin/AP-1 coats, as well as various other proteins that regulate Golgi function. In addition, they have been reported to have roles in viral replication, and even other cellular processes such as lipid droplet formation and mitochondrial division. In humans there are four Arfs, Arf1 and Arf3 (Class I Arfs), and Arf4 and Arf5 (Class II Arfs). All are present on the Golgi, but their precise individual roles have remained unclear. Attempts have been made to deplete individual Arfs using RNAi, but incomplete knockdowns have made the results hard to interpret.

Referee Cross-commenting

There is probably no need for a prolonged debate about this, but I agree that the importance of Arf4 is striking, but it reflects the nature of this work that CRISPR has finally allowed these sorts of questions to be addressed unequivocally. COPI is also involved in recycling of Golgi resident enzymes, and it may be that Arf1 acts in this role.

If the authors check levels of COPI by blotting, and measure the intensity over the Golgi by quantitative IMF, that will reveal whether stability or membrane association if affected without fractionation which is probably less reliable.

If they want to do some extra experiments, then it would be quite easy to check the levels of some Golgi enzymes, or look at lectin binding as a proxy for glycosylation enzyme levels.

Overall, I agree with the positive comments of Reviewers 1 and 2, and it good that we all recognise the quality and importance of the work. However, I feel that one or two of their requests go beyond the scope of a single publication, or would add rather little for a lot of additional work. It is of course easy to propose experiments that someone else has to do!

Reviewer 1:

Point 4. I agree that it would be useful to perform a blot to determine if the levels of coatomer are effected in the various KO lines. I am not sure if Reviewer 1 is also proposing fractionation to determine cytosol vs membrane ratio, but if so, then this would be less useful as peripheral membrane proteins tend to fall off membranes during fractionation and so such analysis is generally questionable. A blot, and clarification of the way the COPI/GM130 ratio is determined, would answer the key points in a relatively straightforward way.

Point 5. I agree that the defect in retrieval of ER residents in Arf4-KO is striking, but it a clear effect even if the reviewer does not understand it themselves! It does not seem so surprising to me, given that Arf4 is likely to act on the early Golgi were such retrieval occurs from. However, the experiment suggested by myself and Reviewer 3 of checking the levels and localisation of the KDEL-receptor would seem to me a good first step to addressing possible mechanism, and certainly sufficient for an initial publication.

Point 7. I am not sure that it has been unequivocally established that Arf1 is important in retrograde traffic. The reality is that many labs have taken Arf1 as being representative of all others and so concentrated biochemical and in vivo studies on this protein. This paper is really important as it highlights the need to investigate both Class I and Class II Arfs, and to bear in mind that their roles in vivo may well be more distinct than their in vitro properties would lead one to suspect. Perhaps, the simplest explanation for this is that the GEFs that activate them have a strong preference for one or the other.

Follow up Comment 1. I was not suggesting that the authors repeat all this in a cell line other than HeLa cells, as this is clearly impractical. HeLa cells are widely used, and so the findings are useful, and whilst it seems certain that some other cell lines would give different data (and indeed the DepMap data show this), then testing one other line would not change the conclusions much. All I wanted the authors to do is to clearly state in the text that what they see in HeLa cells may well be different in other cell lines. This does not detract from the fact that their HeLa cells will provide an excellent platform for focused studies on the role of individual Arfs.

Follow up Comment 2. I agree that Arf6 is not relevant to this paper (as discussed in detail below).

Follow up Comment 3. agree that a simple IMF experiment would suffice to check polarity and immuno-EM is technically very demanding and would add little in this context. The authors have already shown that the Golgi forms stacks in the KO cell lines, and I cannot see how this could occur without the stack being polarised - it has to form at one end and then mature to the other. In addition, after decades of working on the Golgi I have yet to see a credible report of a change to cells causing a loss of Golgi polarity, but maintaining a stacked structure. If the Golgi is not polarised it could not form a stack.

Follow up Comment 4. I agree that one month is perhaps too short to look at KDEL-R, COPI levels and checking polarity by IMF. As noted above, I am NOT suggesting that they repeat all this in a different cell line.

Reviewer 2.

Point 1. I agree with Reviewer 1 that the authors are correct to ignore Arf6. It is a completely different GTPase with a distinct function in a different part of the cell. The family of Arf1-Arf5 arose in metazoans from a single Arf, but Arf6 had already split away from the Arf1-5 family in the last eukaryotic common ancestor, as Arf6 is present in plants and yeasts. There is overwhelming evidence that Arf1-Arf5 are partially redundant and this has hampered their study. Arf6 does not share these roles. The fact that it is acts on endosomes and has been reported to be on the Golgi (which is not widely agreed), is also true of many other GTPases. Indeed, other distant relatives of Arf1-5 are actually on the Golgi (Arl1, Arl5 etc), but these are also not relevant as like Arf6 they do not bind coat proteins and other major effectors of Arf1-5.

Point 2. As noted above, it is hard to see how polarity could be affected given that a Golgi stack is formed, but, at most, a simple application of IMF would seem sufficient to confirm this.

Point 3. Agreed.

Point 5. I agree with the reviewer that this is (much!) too much to ask for an initial publication. Various labs have already reported analysis of the effectors of Class II Arfs and they tend to overlap with Class I. Moreover, it is quite possible that the difference of role in vivo reflects differing interactions with regulators.

Point 6. Agreed.

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

Pennauer uses HeLa cells and CRISPR/Cas9 to delete the 5 members of the class I and class II ARF family of small regulatory GTPases either individually or in combinations. The characterization of the KO cells is excellent and convincingly demonstrates that true KOs were generated. The quality of the data presented is high. Using the KO cells she documents minor alterations in Golgi architecture and the recruitment of vesicular coats in cells deleted of all ARFs except ARF4. In contrast, there is a significant lack of retention/recycling to the ER of KDEL-containing ER proteins in ARF4 KO cells, with numerous ER chaperones now released into the medium (the ARF4 KO secretome). This is a well-done study that showcases the ability of ARF4 alone to sustain cellular life (quite a surprise to this reviewer). Yet, the characterization of the phenotypes is somewhat minimal and the conclusions would be more robustly supported by additional experiments. Specifically:

1. The authors completely ignore class III ARF6 and this paper would be much more comprehensive and informative if analysis of that ARF was also included (ARF6 has been seen at the Golgi and also mediates endosomal trafficking that intersects with the TGN).
2. Although the overall Golgi architecture seems to be largely conserved, it remains essential to test whether Golgi polarity is similarly maintained, and such data would significantly expand the significance of the reported findings
3. Since there is a defect in retrieval of KDEL-proteins, it would be important to show the intracellular localization of the KDEL-R in the cells (especially in the ARF4 KO cells that don't retrieve KDEL-GFP) - is the receptor degraded, stuck in some specific place - knowing that would increase the impact of this study and provide a mechanistic explanation for the observed phenotype
4. The rescue experiments in Figure 6 are good as far as they go, but this experiment would be much more informative if in addition to the same class rescue, the other class ARFs (at least one!) were also characterized.
5. This is maybe a little too much to ask, but since the authors propose a mechanistic explanation for the ARF4 KO KDEL phenotype as being due to different effectors recruited by this ARF (in this case different COPI isotypes - this study would increase in impact by actually testing this mechanisms by assessing whether ARF4Q71L mutant preferentially bound any particular isotype of COPI or even try to do mass spec to identify relevant effectors for this extremely interesting ARF.
6. The Discussion is a very limited and would be more impactful by adding some discussion of organismal effects of ARF deletions (many are embryonic lethal while cells seems to live quite happily) or mutations (links to cancer come to mind here), as well as some mention of data from yeast ARF (what is and isn't essential in those cells). As is, the authors miss an opportunity to highlight the importance of their findings as they relate to current knowledge of ARFology.

Significance

This is an important paper for the ARF field and people interested in ARF signaling will be glad to read about the findings and perhaps also use the developed KO cell lines - this is a significant advancement. The impact would be even higher if some of the experiments suggested above were incorporated into the manuscript.

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

The manuscript by Pennauer et al is the first to systematically investigate the role of class I&II Arfs using a knockout approach. It builds on earlier work by the Kahn lab who used an RNAi approach (Volpicelli-Daley et al. 2005) and is complementary to the overexpression approach used by the Hauri lab (Ben-Tekaya et al, 2010). The work is elegant and the data are strong. I am strongly in favor of publishing this work and my comments are technical in nature (2-5) and a request for some text changes (1). have the following comments for improvements:

1- When it comes to evaluating the role of depletions of Arfs on cell fitness, it would be better to use a non-transformed cell line. I am not asking the authors to go through the painstaking process of generating knockout cell lines in RPE1 cells for instance. Rather, I suggest that the authors make the reader aware that conclusions about cell survival have to be taken with care due to the use of a transformed cell line.

2- Why do Arf1 and Arf4 ko cells grow more slowly. Is it a higher rate of cell death? Is it a block in a certain phase of the cell cycle. Given the link of the Golgi to G2-M entry, I think that an analysis of the cell cycle distribution would add more depth to these data. If the cell cycle distribution is unaffected, then I would conclude that that the difference in doubling time are due to reduced cell survival. If there is an effect on the cell cycle distribution, then the conclusion of the authors is safe that no single Arf is required for survival

3- It is not clear to me how many cells were quantified in Figure 2D-F. I suppose that each do represents a cell. In this case, the number of cells quantified is a bit low. Such a quantification of fluorescence intensities in two channels in the same region is a simple task and I think it should be no problem obtaining at least 100 cells per condition.

4- Is the drop in the ratio of beta-COP/GM130 in Arf1 depleted cells reflecting reduced recruitment to the Golgi? Because the Golgi is bigger, it might be reflecting a reduced density in the number of coatomer molecules per surface area. If it is due to reduced recruitment, then the ratio of membrane/cytosolic betaCOP should be altered. This of course requires to show that the knockout does not affect total levels of coatomer. I think that such fractionation experiments would be a valuable addition to the manuscript and increase the depth of the data.

5- The finding that Arf4-ko cells exhibit a defect on retrieval of ER-resident proteins is exciting, and in my opinion, it is the most significant finding in this manuscript. How can this be reconciled with the lack of an ARf4 ko effect on coatomer recruitment to the Golgi. Looking carefully at the data, I see that in 2 out of 3 experiments, Arf4 ko reduced the betaCOP/GM130 ratio. This is why I think it is crucial to perform more experiments and add more cells to increase the confidence in the data. Reduced retrieval of ER chaperones is frequently found in tumors and we still don't understand the reason behind this. Therefore, this finding is of significance beyond the community of cell biologists.

6- I find Figure 6A confusing. Why do Arf1 overexpressing parental HeLa cells exhibit less Arf1 than control cells?

7- Why was the following condition not tested: Arf4ko cells with Arf1 overexpression. Given the importance of Arf1 in retrograde (Golgi-to-ER) trafficking, I would expect a partial rescue of the retrieval of ER chaperones.

Significance

Significance of the work:

The paper is important because it is the first to examine the role of Arfs using a knockout approach. Another very important finding is that Arf4 depleted cells exhibit problems with retrieval of ER chaperones. This is a very novel finding and to the best of my knowledge

Audience:

The primary audience is of course the community working on membrane trafficking, organelle biology and proteostasis. However, I think that the data on the role of Arf4 in retrieval of ER chaperones might be of relevance for cancer biologists. Secretion of ER chaperones is frequently found in many tumors and we still do not understand why this is happening and what the significance thereof is.

My own expertise:

Export from the endoplasmic reticulum Golgi fragmentation in cancer cell migration Rho GTPases Kinase signaling Pseudoenzymes Cell migration of breast cancer cells Proteostasis in multiple myeloma

Referee Cross-commenting

Just a follow-up comment from my side:

I agree that it has not been unequivocally established that Arf1 is the main/sole of retrograde transport. However, even less established is the role of Arf4 in this process. The authors show that it is mainly Arf1 depletion that reduces the amount of COPI at the Golgi (ratio of COPI/GM130). Thus, I remain very surprised that it is actually the Arf4 depletion that results in reduced retrieval.

What is the significance of having less COPI at the Golgi in Arf1-ko cells? Certainly, the Golgi is not more "leaky". Does the level of COPI at the Golgi not reflect the strength of retrograde trafficking? Maybe there is no less COPI at the Golgi, and it only appears to be less, because the Golgi is bigger. This is why a simple fractionation experiment would be good. Something like making a cytosol and a microsome fraction and looking at the ratio of COPI (Cyt/Mem).

If both reviewers think it is too much, or unlikely to work, then I am happy to drop this point.

Below are my comments to the evaluations by the other two reviewers:

1- I agree with most comments that the two other reviewers made. Some of them are actually overlapping with mine (e.g. the use of a cell line other than HeLa).

2- I am not sure whether the impact of the paper would improve by adding data on Arf6.

3- To the comment on Golgi polarity. Maybe we could be more specific here and say that it would be sufficient to show that a trans-Golgi protein and a cis-Golgi protein can be separated by fluorescence microscopy, or whether we alternatively want them to actually do it by immunogold labeling for EM (which is more difficult).

4- I agree with reviewer 2 that the work proposed needs 1-3 months. I think reviewer 3 is a bit too optimistic with 1 month, because her/his comment on using a cell line other than HeLa cannot be addressed in just a month.

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

Learn more at Review Commons

---

Reply to the reviewers

Reviewer #1 (Evidence, reproducibility and clarity (Required)): **Summary:** In this study authors investigated the role of NAMPT, NAD+ and PARP1/parthanatos in skin inflammation using a zebrafish psoriasis model with an hypomorphic mutation of spint1a and human organotypic 3D skin models of psoriasis. Authors showed that genetic deletion and/or pharmacological inhibition of Nampt/PARP1/AIFM1/NADPH oxidases reduced oxidative stress, inflammation, keratinocyte DNA damage, hyperproliferation and cell death in zebrafish models of chronic skin inflammation. Authors also showed the expression of pathology-associated genes in human organotypic 3D skin models of psoriasis with pharmacological inhibition of Nampt/PARP1/AIFM1/NADPH oxidases. The key finding of this study is that PARP1 hyperactivation caused by ROS-induced DNA damage mediates skin inflammation through parthanatos. **Major comments:** This is a very comprehensive study to investigate the role of PARP1 in skin inflammation. The main conclusion was made based on the genetic inhibition and/or pharmacological inhibition of Nampt/PARP1/AIFM1/NADPH oxidases. Although the finding of this study that NAMPT-derived NAD+ fuels PARP1 to promote skin inflammation through parthanatos is interesting and important, there are lots of major concerns and questions, which have to be addressed to better support the main conclusion. In addition, the data and methods were not presented with sufficient detail.

1. This study is heavily relied on pharmacology inhibition. However, the specificity and selectivity of many inhibitors were not tested in this study.

At least 3 concentrations of each inhibitor were tested and the lowest one able to rescue the phenotype was then used for further testing (please, see Table S1). More importantly, the specificity of all compounds used were confirmed by genetic inhibition of their targets.

Fig. 1: it is quite confusing how NAD+ increases H2O2 levels? Is NAD+ cell permeable? It is not clear if NAD+ has been really up taken by cells in the larvae. If NAD+ fuels PARP1 to promote skin inflammation, why NAM treatment increased H2O2 levels but NMN precursor failed to increase skin oxidative stress? No reasonable explanation has been provided.

This is an interesting point. We have shown that exogenous NAD+ added in the water of larvae increased larval NAD+ (please, see Fig. 2K). It has been shown that neurons can take up NAD+ through CX43 (Fig. S7), so a similar mechanism may operate in larval skin. As regards, the effect of NAM and NMN, a recent study has demonstrated that NAM supplementation increased zebrafish larval NAD+; however, NA, NMN and NR failed to boost larval NAD+ level (PMID: 32197067). These results are consistent with our data.

Fig. 1E and 1G: it is not clear what is the green channel. Similarly, there is no clear description what is red or green in many other figures.

To help the interpretation of larval pictures, we have indicated in all figures what is analyzed in each fluorescent channel.

1. Fig. 1K and 1L: It is hard to understand why FK-866 reduced H2O2 release, but it increased neutrophils infiltration. How to interpret this conclusion?

Fig. 2C-D: Why low doses FK-866 reduced neutrophil infiltration whereas high dose FK-866 increased neutrophil infiltration?

Answer to 4&5: As it was explained in lines 145-156, FK-866 induces NF-kB activation in the muscle and neutrophil infiltration in this tissue when used at 100 uM. This result may be deleted if the reviewers think it is confusing, since a 10 uM dose was used in all subsequent experiments to study the impact of Nampt in skin inflammation. This dose has no effects in the muscle but robustly reduced skin H2O2 production and neutrophil skin infiltration.

Fig. 2I-J: it is not clear how NF-kB activity was measured. Is that based on green fluorescence shown in Fig. 2J? if so, the representative images were not consistent with the quantification data shown in I. Similarly, many other representative images were also not consistent with their quantification data throughout the manuscript. For example, Fig. 3C/D, 3E/F, 3G/H, 3L/M, Figure S2C/D, S2G/H, Fig. 4C/D, 4J/K.

The quantification of NFkB was measured in the skin, as it has already been reported previously (Candel et al., 2014). This is indicated in M&M section. The images show the whole larvae and NFkB is expressed at high levels in different tissues, such as neuromasts. To clarify this, we have included an additional figure to explain the ROI used for quantification of H2O2 and NfkB (Fig. S1G).

Figure S1C, Nampta/Namptb protein expression should be checked and shown after its KO using crispr/cas9 technique.

Unfortunately, we have used to different antibodies and both failed to crossreact with zebrafish Nampta/b. However, we have included the efficiency of CRISPR-Cas9 in Fig. S1F of the revised version. The efficiency is relatively low, probably indicating that is indispensable for zebrafish development, as occurs in mice (PMID 28333140).

Fig. 3I: protein expression of nox1, nox4 and nox 5 should be checked after genetic inhibition using CRISPR/Cas9 technique.

Unfortunately, we do not have antibodies able to recognize zebrafish Nox1, Nox4 and Nox5. However, we have provided the efficiency of the gRNA used for each gene (Fig. S3) and it is about 65%.

Fig. 4: If Olaparib treatment increased DNA damage, will it increase PARP1 activation and PAR formation?

As it has widely used in mammalian models, parthanatos is triggered by overactivation of PARP1 following DNA damage. Therefore, although inhibition of olaparib may further induces DNA damage, it blocks parthanatos. This is consistent with our results showing that olaparib reduces PARylation (Fig. S4H) and cell death (Figs. 4J, 4K).

Fig. 4M: it is not clear what staining has been done. No difference was observed among different groups.

As indicated in the figure legends, pγH2Ax+ (green) keratinocytes (red) are shown. We have indicated this in the figure and include arrows to show pγH2Ax+ cells. The quantitation of this experiment (Fig. 4L) show that FK-866 robustly reduced, while olaparib increases, keratinocyte DNA damage.

Authors used N-phenylmaleimide (NP) to block AIF nuclear translocation. How does this inhibitor work? what is its actual effect on AIF nuclear translocation? Experiments are required to show this inhibitor actually blocks AIF nuclear translocation.

NP has been shown to block AIFM1 nuclear translocation, since it inhibits cysteine proteases which are required for its cleavage which precedes nuclear translocation (PMID 8879205). Although we have shown that genetic inhibition of Aifm1 rescues skin inflammation, confirming the specificity of the inhibitor, we agree on this point. Therefore, we have performed additional experiments and showed nuclear Aifm1 in keratinocyte aggregates of Spint1-deficient larvae and that NP treatment blocked nuclear translocation (Fig. S6C). In addition, we have also shown increased nuclear translocation of AIFM1 in keratinocytes of lesional skin from psoriasis patients (Figs. 6C, 6D).

Figure S4: it is hard to understand why lane #2 with Olaparib has the highest PAR signal.

We are sorry for this mistake labeling the WB. The right legend is: 1 +/+, 2 -/- treated with DMSO, 3 -/- treated with FK-866 and 4 -/- treated with olaparib.

Does spint1a-/- zebrafish show parthanatos cell death? It is not clear how cell death was measured.

We have shown that skin keratinocytes from Spint1a-deficient fish show increased cell death, as assayed by TUNEL, that is fully reversed by olaparib (Figs. 4J, 4K). In addition, skin keratinocytes from the mutant fish also have increased PARylation that is reversed by either FK-866 or olaparib (Fig. S4G, S4H). Further, pharmacological and genetic inhibition of Aifm1 inhibition and forced expression of Parga also rescue skin inflammation. Finally, we have included new experiments showing Aifm1 nuclear translocation in both Spint1a-deficient larvae and psoriasis patient lesional skin. Therefore, all these results show that Spint1a-deficient fish show parthanatos cell death-induced inflammation.

NAD+ levels were regulated by 3 different pathways. Expression of many genes involved in these 3 pathways were altered in psoriasis. However, it is not clear if the other two pathways play a role in PARP1-mediated inflammation.

NAD+ salvage pathway has been shown to be the major pathway regulating NAD+ levels in most tissues. The inhibition of this pathway with FK-866 rescues all skin phenotypes observed in Spint1a-deficient larvae as well as in organotypic 3D skin models of psoriasis. These results were further validated using another inhibitor (GMX1778) and genetic inhibition. Therefore, our results support that the salvage pathway is the one involved in psoriasis and inhibition of this pathway would rescue inflammation. However, it will be worthy to investigate if other pathways play a role in psoriasis and specifically upon inhibition of the salvage pathway.

**Minor comments:**

1. Page 9: To test this hypothesis, we used N-phenylmaleimide (NP), a chemical inhibitor of Aifm1 translocation from the nucleus to the mitochondria (Susin et al., 1996). The statement is not correct.

We are sorry for this mistake. It has been amended to: “To test this hypothesis, we used N-phenylmaleimide (NP), a chemical inhibitor of Aifm1 translocation from the mitochondria to the nucleus (Susin et al., 1996).”

Page 12: To the best of our knowledge, this is the first study demonstrating the existence of parthanatos in vivo. This statement is not correct.

We have removed this statement.

Figure S3 and S6E: they should be presented in an easy understandable way for the general readers.

We have explained in the legends the graph output of TIDE analysis.

Figure legends should be presented in a clearer way.

We have tried our best writing the legends. All suggestions and request were made.

Reviewer #1 (Significance (Required)): Parthanatos is a new type of cell death distinct from apoptosis, necrosis, necroptosis and plays a pivotal role in ischemic stroke and neurodegenerative diseases (Wang Y et a., Science. 2016; Kam TI et al., Science 2018). The current study may provide new evidence of the importance of PARP1 and parthanatos in skin inflammation and potential targets for the treatment of skin inflammation. We thank the reviewer’s opinion on the significance of our study.

The reviewer has the expertise in oxidative stress, PARP1 and parthanatos research. Reviewer #2 (Evidence, reproducibility and clarity (Required)): **Summary:** The manuscript entitle "NAMPT-derived NAD+ fuels PARP1 to promote skin inflammation through parthanatos" is well written, divided and organized. This work demonstrated that models of psoriasis are characterized by ROS stress, inflammation and cell death. It was clear that NAMPT, a rate-limiting enzyme of NAD salvage pathway, and PARP1, a Poly-ADP-ribose polymerase, could be targeted to decrease ROS stress and inflammation that are contributing to cell death through parthanatos pathway. However, it was not clear that NAD+ are the responsible for fuel these processes in the psoriasis models analyzed. Nevertheless, the present work demonstrated that the cell death observed in the psoriasis model analyzed was correlated to an unidentified programmed cell death pathway, parthanatos that up to date has not been demonstrated.

We are pleased with the reviewer’s comments on our study.

**Major comments:** Most of the data showed confirmed that inhibition of NAMPT or PARP1 seems to be beneficial for the relief of some characteristics related to oxidative stress and inflammation in the skin. However, the author should show data about NAD+ levels only instead of the ratio NAD+/NADH to state that NAMPT-derived NAD+ is promoting oxidative stress (line 366-368) (fig2K).

The data shown in Fig 2K are NAD+ plus NADH. Considering that cytosolic and nuclear NAD+/NADH ratio typically ranges from 100 to 1000 (PMID: 21982715), these data mainly show intracellular NAD+ concentration in larvae.

Some data images are not convincing, or they don't really show an increase or decrease as the author showed in graph data. (Fig1D, 1E - 1F,1G).

The quantification of H2O2 and NFkB was measured in the skin, as it has already been reported previously (Candel et al., 2014). To clarify this, we have shown the ROI used for quantification of H2O2 and NfkB in Fig. S1G.

What is the relevance to analyze muscle and what is the relevance of the results obtained, since the effect of FK-866 in muscle increases the NFKB activity?

This is essentially a similar concern raised by reviewer 1. FK-866 induces NF-kB activation in the muscle and neutrophil infiltration in this tissue when used at 100 uM. This result may be deleted if the reviewers think it is confusing, since a 10 uM dose was used in all subsequent experiments to study the impact of Nampt in skin inflammation. This dose has no effects in the muscle but robustly reduced skin H2O2 production and neutrophil infiltration.

Figure S4H is not convincing with what the author wrote.

We are sorry for this mistake labeling the WB. The right legend is: 1 +/+, 2 -/- treated with DMSO, 3 -/- treated with FK-866 and 4 -/- treated with olaparib. Both FK-866 and olaparib rescue PARylation in the skin of Spint1a-deficient larvae.

The author should make the keratinocyte aggregation experiment with FK-866 treatment to better substantiate what they are proposing.

These results are shown in Figs. 2E and 2F.

**Minor comments:** Line 281: "NP, a chemical inhibitor of Aifm1 translocation from the nucleus to the mitochondria..." should be the opposite: NP, a chemical inhibitor of Aifm1 translocation from mitochondria to nucleus.

We are sorry for this mistake. It has been amended.

Line 299 "figure 6A" should be Figure 6B.

We have checked and it is correct.

How the author explains the relationship between all the results being related to NAMPT and supposedly to NAD+, but an important precursor to make NAD through salvage pathway (NMN) and a well NAD+ booster didn't show any effect?

This is an interesting point that was also raised by reviewer 1. A recent study has demonstrated that NAM supplementation increased zebrafish larval NAD+; however, NA, NMN and NR failed to boost larval NAD+ level (PMID: 32197067). This explains our results. We have discussed this point in the revised manuscript.

Line 178: should be NAMPT inhibitor stead of FK-866 inhibitor.

Thanks a lot. It has been amended.

Line 191-192: I suggest reformulating this sentence since the result showed was only the ratio NAD/NADH.

Please, see our response above. We are measuring NAD+ plus NADH. We have amended the text to clarify this fact.

Reviewer #2 (Significance (Required)): The present work greatly demonstrated the relevance of PARP1 and NAMPT in the field of inflammation and ROS in the skin that contribute to diseases like psoriasis. Although it is not a lethal disease, as the author mentioned, it affects the physical and mental health of the individual. Understanding the mechanism that underlie this condition would help to trigger new and more efficient treatments. It was clear that the result showed a promising strategy in targeting NAMPT and PARP1. Furthermore, inhibitor for them is already know and may be useful for future treatment of psoriasis disease. We thank this comments on the impact of our study.

Reviewer #3 (Evidence, reproducibility and clarity (Required)): This study shows NAMPT derived NAD facilitates PARP activation to promote skin inflammation via parthanatos. The authors used the zebrafish model and organoid models of psoriasis and observed that inhibition of NAMPT reduces inflammation in zebrafish and human skin organoid models. They also observed that NADPH oxidase-derived oxidative stress activates PARP, and PARP inhibition or over-expression of PARG or AIF mimics protection mediated by NAMPT inhibition. This is an interesting study, but there are several weaknesses to support the conclusions of this study. While pharmacological inhibition is a powerful tool, complementary methods (knock out of PARP-1) are critical for this paper's conclusions. PARP inhibitor used in this study may not specifically inhibit PARP1 but other PARPs too. Therefore, genetic knockout of PARP will make the make this conclusions/interpretation of this study strong.

We thank these comments on our manuscript. All pharmacological inhibitions used in this study were confirmed by genetic experiments, including Parp1. The genetic inhibition of Parp1 is shown in Figs. S4C-S4F.

Additional comments include: This study's primary focus is PARP activation and PAR-mediated parthanatos, but it is not shown how different inhibitors used in this study and supplementations of NAD alter PARP activation and PAR formation.

We have shown through the quantitation of PARylation that Spint1a-deficient skin shows increased PAR activity and that pharmacological inhibition of either Nampt or Parp was able to fully reverse it (Figs S4g & S4H). In addition, we have also shown a dramatically increased PAR activity in lesional skin biopsies from psoriasis patients (Fig. 6E).

NAMPT is not the only NAD biosynthesis pathway; how other NAD pathways respond when NAMPT is inhibited with FK-866.

NAD+ salvage pathway has been shown to be the major pathway regulating NAD+ levels in most tissues. The inhibition of this pathway with FK-866 rescues all skin phenotypes observed in Spint1a-deficient larvae as well as in organotypic 3D skin models of psoriasis. Therefore, our results support that the salvage pathway is the one involved in psoriasis and inhibition of this pathway would rescue inflammation. However, we agree that it will be worthy to investigate if other pathways play a role in psoriasis and specifically upon inhibition of the salvage pathway. However, this is out of the scope of this manuscript.

PARG is used in this study, but the protein levels of PARG are not shown, and it is not clear whether the PARG overexpression is sufficient to reduce PAR levels in the models used. AIF pharmacological and genetic manipulation of AIF is used, but it is not shown that AIF translocates to the nucleus in this model.

We agree on these points, so we have analyzed Aifm1 translocation in Spint1a-deficiet larvae and psoriasis patient lesional skin (please, see above our response to reviewer 1) and PARylation upon forced expression of Parga (Fig. 5M).

Does NAMPT inhibition reduce NAPD oxidase activity?

Our results indicate that Nampt inhibition reduce NAPDH oxidase activity, since a drastic reduction of H2O2 production was observed in the skin of Spint1a-deficient larvae treated with FK-866.

PAR plots provided in fig S4 need quantification, and the blots (Fig S4 G&H) should be run on the same gel to make sure the exposure levels are the same. It is not clear which group is represented in lane 4 of Fig S4 G.

We have provided the quantitation. The problem is that we mislabeled the legend of Fig. S4H. The right legend is: 1 +/+, 2 -/- treated with DMSO, 3 -/- treated with FK-866 and 4 -/- treated with olaparib. Therefore, either Nampt or Parp inhibition robustly reduces PARylation of Spint1a-deficient skin to the levels of their wild type counterparts.

Reviewer #3 (Significance (Required)): This study in interesting potentially showing the role of PARP-1 activation and Parthanatos in skin inflammation. It could be very significant if above identified weaknesses are addressed.

We are pleased with this reviewer’s assessment on the significance of our study.

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

This study shows NAMPT derived NAD facilitates PARP activation to promote skin inflammation via parthanatos. The authors used the zebrafish model and organoid models of psoriasis and observed that inhibition of NAMPT reduces inflammation in zebrafish and human skin organoid models. They also observed that NADPH oxidase-derived oxidative stress activates PARP, and PARP inhibition or over-expression of PARG or AIF mimics protection mediated by NAMPT inhibition. This is an interesting study, but there are several weaknesses to support the conclusions of this study. While pharmacological inhibition is a powerful tool, complementary methods (knock out of PARP-1) are critical for this paper's conclusions. PARP inhibitor used in this study may not specifically inhibit PARP1 but other PARPs too. Therefore, genetic knockout of PARP will make the make this conclusions/interpretation of this study strong.

Additional comments include:

This study's primary focus is PARP activation and PAR-mediated parthanatos, but it is not shown how different inhibitors used in this study and supplementations of NAD alter PARP activation and PAR formation. NAMPT is not the only NAD biosynthesis pathway; how other NAD pathways respond when NAMPT is inhibited with FK-866 PARG is used in this study, but the protein levels of PARG are not shown, and it is not clear whether the PARG overexpression is sufficient to reduce PAR levels in the models used. AIF pharmacological and genetic manipulation of AIF is used, but it is not shown that AIF translocates to the nucleus in this model. Does NAMPT inhibition reduce NAPD oxidase activity? PAR plots provided in fig S4 need quantification, and the blots (Fig S4 G&H) should be run on the same gel to make sure the exposure levels are the same. It is not clear which group is represented in lane 4 of Fig S4 G.

Significance

This study in interesting potentially showing the role of PARP-1 activation and Parthanatos in skin inflammation. It could be very significant if above identified weaknesses are addressed.

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

Summary:

The manuscript entitle "NAMPT-derived NAD+ fuels PARP1 to promote skin inflammation through parthanatos" is well written, divided and organized. This work demonstrated that models of psoriasis are characterized by ROS stress, inflammation and cell death. It was clear that NAMPT, a rate-limiting enzyme of NAD salvage pathway, and PARP1, a Poly-ADP-ribose polymerase, could be targeted to decrease ROS stress and inflammation that are contributing to cell death through parthanatos pathway. However, it was not clear that NAD+ are the responsible for fuel these processes in the psoriasis models analyzed. Nevertheless, the present work demonstrated that the cell death observed in the psoriasis model analyzed was correlated to an unidentified programmed cell death pathway, parthanatos that up to date has not been demonstrated.

Major comments:

Most of the data showed confirmed that inhibition of NAMPT or PARP1 seems to be beneficial for the relief of some characteristics related to oxidative stress and inflammation in the skin. However, the author should show data about NAD+ levels only instead of the ratio NAD+/NADH to state that NAMPT-derived NAD+ is promoting oxidative stress (line 366-368) (fig2K). Some data images are not convincing, or they don't really show an increase or decrease as the author showed in graph data. (Fig1D, 1E - 1F,1G). What is the relevance to analyze muscle and what is the relevance of the results obtained, since the effect of FK-866 in muscle increases the NFKB activity?<br> Figure S4H is not convincing with what the author wrote. The author should make the keratinocyte aggregation experiment with FK-866 treatment to better substantiate what they are proposing.

Minor comments:

Line 281: "NP, a chemical inhibitor of Aifm1 translocation from the nucleus to the mitochondria..." should be the opposite: NP, a chemical inhibitor of Aifm1 translocation from mitochondria to nucleus. Line 299 "figure 6A" should be Figure 6B. How the author explains the relationship between all the results being related to NAMPT and supposedly to NAD+, but an important precursor to make NAD through salvage pathway (NMN) and a well NAD+ booster didn't show any effect? Line 178: should be NAMPT inhibitor stead of FK-866 inhibitor. Line 191-192: I suggest reformulating this sentence since the result showed was only the ratio NAD/NADH.

Significance

The present work greatly demonstrated the relevance of PARP1 and NAMPT in the field of inflammation and ROS in the skin that contribute to diseases like psoriasis. Although it is not a lethal disease, as the author mentioned, it affects the physical and mental health of the individual. Understanding the mechanism that underlie this condition would help to trigger new and more efficient treatments. It was clear that the result showed a promising strategy in targeting NAMPT and PARP1. Furthermore, inhibitor for them is already know and may be useful for future treatment of psoriasis disease.

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:

In this study authors investigated the role of NAMPT, NAD+ and PARP1/parthanatos in skin inflammation using a zebrafish psoriasis model with an hypomorphic mutation of spint1a and human organotypic 3D skin models of psoriasis. Authors showed that genetic deletion and/or pharmacological inhibition of Nampt/PARP1/AIFM1/NADPH oxidases reduced oxidative stress, inflammation, keratinocyte DNA damage, hyperproliferation and cell death in zebrafish models of chronic skin inflammation. Authors also showed the expression of pathology-associated genes in human organotypic 3D skin models of psoriasis with pharmacological inhibition of Nampt/PARP1/AIFM1/NADPH oxidases. The key finding of this study is that PARP1 hyperactivation caused by ROS-induced DNA damage mediates skin inflammation through parthanatos.

Major comments:

This is a very comprehensive study to investigate the role of PARP1 in skin inflammation. The main conclusion was made based on the genetic inhibition and/or pharmacological inhibition of Nampt/PARP1/AIFM1/NADPH oxidases. Although the finding of this study that NAMPT-derived NAD+ fuels PARP1 to promote skin inflammation through parthanatos is interesting and important, there are lots of major concerns and questions, which have to be addressed to better support the main conclusion. In addition, the data and methods were not presented with sufficient detail.

1. This study is heavily relied on pharmacology inhibition. However, the specificity and selectivity of many inhibitors were not tested in this study.
2. Fig. 1: it is quite confusing how NAD+ increases H2O2 levels? Is NAD+ cell permeable? It is not clear if NAD+ has been really up taken by cells in the larvae. If NAD+ fuels PARP1 to promote skin inflammation, why NAM treatment increased H2O2 levels but NMN precursor failed to increase skin oxidative stress? No reasonable explanation has been provided.
3. Fig. 1E and 1G: it is not clear what is the green channel. Similarly, there is no clear description what is red or green in many other figures.
4. Fig. 1K and 1L: It is hard to understand why FK-866 reduced H2O2 release, but it increased neutrophils infiltration. How to interpret this conclusion?
5. Fig. 2C-D: Why low doses FK-866 reduced neutrophil infiltration whereas high dose FK-866 increased neutrophil infiltration?
6. Fig. 2I-J: it is not clear how NF-kB activity was measured. Is that based on green fluorescence shown in Fig. 2J? if so, the representative images were not consistent with the quantification data shown in I. Similarly, many other representative images were also not consistent with their quantification data throughout the manuscript. For example, Fig. 3C/D, 3E/F, 3G/H, 3L/M, Figure S2C/D, S2G/H, Fig. 4C/D, 4J/K.
7. Figure S1C, Nampta/Namptb protein expression should be checked and shown after its KO using crispr/cas9 technique.
8. Fig. 3I: protein expression of nox1, nox4 and nox 5 should be checked after genetic inhibition using CRISPR/Cas9 technique.
9. Fig. 4: If Olaparib treatment increased DNA damage, will it increase PARP1 activation and PAR formation?
10. Fig. 4M: it is not clear what staining has been done. No difference was observed among different groups.
11. Authors used N-phenylmaleimide (NP) to block AIF nuclear translocation. How does this inhibitor work? what is its actual effect on AIF nuclear translocation? Experiments are required to show this inhibitor actually blocks AIF nuclear translocation.
12. Figure S4: it is hard to understand why lane #2 with Olaparib has the highest PAR signal.
13. Does spint1a-/- zebrafish show parthanatos cell death? It is not clear how cell death was measured.
14. NAD+ levels were regulated by 3 different pathways. Expression of many genes involved in these 3 pathways were altered in psoriasis. However, it is not clear if the other two pathways play a role in PARP1-mediated inflammation.

Minor comments:

1. Page 9: To test this hypothesis, we used N-phenylmaleimide (NP), a chemical inhibitor of Aifm1 translocation from the nucleus to the mitochondria (Susin et al., 1996). The statement is not correct.
2. Page 12: To the best of our knowledge, this is the first study demonstrating the existence of parthanatos in vivo. This statement is not correct.
3. Figure S3 and S6E: they should be presented in an easy understandable way for the general readers.
4. Figure legends should be presented in a clearer way.

Significance

Parthanatos is a new type of cell death distinct from apoptosis, necrosis, necroptosis and plays a pivotal role in ischemic stroke and neurodegenerative diseases (Wang Y et a., Science. 2016; Kam TI et al., Science 2018). The current study may provide new evidence of the importance of PARP1 and parthanatos in skin inflammation and potential targets for the treatment of skin inflammation.

The reviewer has the expertise in oxidative stress, PARP1 and parthanatos research.

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

Learn more at Review Commons

---

Reply to the reviewers

We thank the reviewers for the positive assessment of our work and for the constructive comments that helped us to improve the quality of our manuscript. We have carefully considered each point and have addressed most by modifying the manuscript text to increase clarity of our work. Based on a suggestion by Reviewer 2 we have also included the results of a new experiment.

In addition to addressing all comments of the reviewers, we have expanded the part of the study analysing the functionality of Caulobacter’s DnaA Nt in the heterologous host E. coli. Furthermore, we have replaced our original set of fluorescence data by a new data set that has been acquired using optimized measurement parameters (bottom read and 100 for the detector gain - see Material and Methods for details), which have improved the signal-to-noise ratio and the overall quality of the fluorescence profiles. Importantly, these new data do not change, but rather strengthen, our conclusions.

Reviewer #1 (Evidence, reproducibility and clarity (Required)):

Felletti et al provide compelling new evidence that a CDS element in the dnaA mRNA is required for nutrient dependent translationol control. This provides a mechanisms by which dnaA translation is shut off during carbon starvation, and is supported by a rather rigorous analysis of the mRNA performed both in vitro and in vivo. Overall it was a pleasure to read and the data are generally very compelling. My specific comments are below:

**Major Comments:**

While the authors rule out differences in charging of different ala-tRNAs as controlling the nutrient dependent repression in translation, the authors assume that this must be due to the nascent sequence. However, could it also be possible that all ala-tRNA isoacceptors have lower charging after C-starvation?

We thank the reviewer for raising this important point. As Reviewer 1 pointed out, we cannot conclusively exclude that carbon starvation could lead to reduced charging levels of all isoacceptor Ala-tRNAs. However, based on the available literature, we consider it unlikely. In a first work by Elf et al 2003 (confirmed later by Dittmar et al 2005 and Subramaniam et al 2014) the authors argued that under amino acid-limiting conditions the charging levels of the different isoacceptor tRNAs depend directly on their codon usage during translation. Importantly, in our work we could show that Nt mediates the inhibition of translation independent of the synonymous codon choice, suggesting that aa-tRNA levels are not limiting in our experimental conditions. To address this comment of Reviewer 1, we discussed this matter in a greater detail in the revised version of the manuscript (line 374-379).

**Minor comments:**

It was observed many years ago that tmRNA is required for the proper timing of DNA replication initiation in Caulobacter (Cheng and Keiler J Bact 2009). Since the AAI motif is appearing to alter translation elongation, it might be interesting to discuss the AAI motif may be linked to ribosome arrest and rescue.

We appreciate this suggestion. Cheng and Keiler 2009 proposed an indirect involvement of the tmRNA in the transcriptional regulation of DnaA over Caulobacter’s cell cycle. In the revised version of the manuscript, we mention the tmRNA and ArfB protein as possible factors involved in ribosome rescue following Nt-induced ribosome stalling and we refer to Keiler et al 2000 and Feaga et al 2014.

Line 49 - add "initiation"

The word “initiation” was added to the text.

Line 61 - is "cleared" meant to be proteolyzed or simply meaning to have a lower protein level?

We apologize if we were not clear. We rephrased the text as follows: “[…] DnaA levels decrease at the onset of carbon starvation […]”.

Line 92-93 - is this 5' UTR based on a previously defined TSS determined in their previous study?

dnaA TSS has been first determined by primer extension (Zweiger and Shapiro 1994) and later by global 5’RACE (Schrader et al 2014 and Zhou et al 2015). In the new version of the manuscript, we include references to these previous studies (line 94).

Line 115-118 - this is interesting, might this conserved 5' UTR be added to rfam?

We thank the reviewer for this suggestion. We will submit our alignment to rfam after publication of the manuscript in a journal.

Line 126-127, 131,189 - Is the 3nt sequence the authors found here considered a Shine-Dalgarno site? I would imagine that this would be too small to consider this. Perhaps calling it SD-like sequence might be more appropriate.

We agree with this comment. In the new version of the manuscript, we refer to the identified 3-nucleotide sequence as a “SD-like sequence”.

Lines 136-140, 208-210 - Would the authors consider this upstream site with a potential CUG start codon a standby site? It appears to fit many of the criteria which could be used to define one.

According to our probing data, the mRNA region in proximity of the CUG start codon forms a very stable stem-loop structure. Based on our previous experience (especially the extensive work by the Wagner lab), typical ribosome standby sites only occur in largely unstructured regions. Furthermore, in Supplementary Fig. 4 we show that the deletion of stem P4 does not affect eGFP expression levels. For these reasons, we consider it unlikely that the putative CUG start codon is part of a ribosome standby site.

Lines 253-255 - this is a beautiful experiment, but very hard to understand from the text. Perhaps add a sentence or two to explain it in more detail.

We thank the reviewer for this comment. In the revised version of the manuscript, we provide a more detailed description of the dfsNt reporter mutant. We hope this will address the reviewer’s concerns.

Line 307 - add "synonomous"

The word “synonymous” was added in the revised version of the manuscript

When dnaA is depleted, it was observed that the chromsome can be erroneously segregated by the ParA/B/S system (mera et al PNAS). Does this occur in C-starvation when DnaA levels drop?

In a separate study we have also observed that in a fraction of DnaA depleted cells the origin of replication is erroneously translocated from the stalked to the swarmer cell pole. We have not studied this phenomenon under carbon starvation, as it lies outside the scope of this paper. However, if the ParA/B/S remains functional under carbon starvation, this might also happen in G1-arrested starved cells.

Reviewer #1 (Significance (Required)):

Appears to be quite significant to researchers studying regulation of bacterial cell cycle and translation. Since DnaA is conserved across bacteria, and this mechanism works in E. coli, it appears that the findings will likely be important in many bacterial systems.

Referee Cross-commenting

All the reviewer comments I read seem reasonable. Specifically, I found the point about E. coli 30S ribosomes is very important that the authors address. This could be done in writing, but should be better listed as a caveat to those experiments.

As suggested by the reviewers, we have partially rephrased some parts of the text describing the toeprint results. Moreover, we have inserted in the main text and in Fig. 1 legend explicit references to the use of purified E. coli 30S subunits and tRNA-fMet. We believe these changes will address the reviewers’ concerns.

Reviewer #2 (Evidence, reproducibility and clarity (Required)):

**Summary:** The Jonas lab provide good evidence that they have found a new mechanism to regulate the amount of the DnaA protein by a starvation signal. The DnaA protein is the key chromosome replication initiator probably for most bacteria and as such DnaA is the target of many regulatory inputs. The authors created an accurate reporter system that allows them to dissect the 5' mRNA translated and untranslated sequences of dnaA and they have convincingly demonstrated that the N-terminal DnaA peptide sequence and not the RNA mediate the response to starvation by glucose exhaustion. This is potentially a model example for global translational responses in bacteria.

**Major comments:**

The main conclusion, i.e. that the DnaA leader peptide "Nt" mediates this response is convincing. However, there were 2 major problems that should be easily addressed. These do not subtract from the main conclusion.

Problem 1

E. coli 30S subunits were used in the "Toeprint" assay of Fig. 1. Obviously Caulobacter 30S Ribosome subunits should have been used, or a justification should be given. One remedy would be to make this supplementary information.

We thank Reviewer 2 for this comment. We agree that it would be better to use Caulobacter 30S ribosome subunits in our toeprint experiments. However, because toeprint assays with E. coli 30S ribosome were already established in our lab (i.e. the Wagner lab, where the assays were performed) and because works by other groups have shown that E. coli 30S subunits can be used to study the translation of mRNAs from other bacteria, we decided to use this experimental set up. Based on our results, we also had no reason to doubt the suitability of the E. coli 30S subunits. The toeprint showed that translation starts at the in silico predicted translation start site, which was further confirmed by our in vivo mutagenesis experiments. For these reasons, we are confident that the toeprint assays indicate the true translational start site. However, we acknowledge that we could have been more explicit about the use of the purified E. coli 30S subunits and tRNA-fMet in toeprinting assay. To increase clarity and transparency, in this revised version of the manuscript, some parts of the main text were rephrased and references to the use of E. coli 30S and tRNA-fMet were introduced (including Fig. 1 legend). We hope that these changes will address the reviewer’s concerns.

Problem 2

The results in Fig. 6B could be due to the Nt simply making the hybrid protein more unstable in E. coli. This is the main impression given by the drop in signal. In this case, the conclusion would be wrong, and Nt is not transferring a starvation translation block from C. crescentus to E. coli. Nt is just making the protein unstable. These results should be treated as preliminary pending protein stability measurements. However, this defect does not subtract from the other main points and without the Fig. 6 E. coli experiments they still make a complete and interesting story. One remedy would be to make this also supplementary information.

It is indeed striking that a drop of normalised fluorescence is observed for the 5’UTRdnaA-Nt construct in E. coli but not in Caulobacter. In order to address if this behavior can be explained by reduced protein stability, we have performed a translation shut-off assay using the 5’UTRdnaA-Nt E. coli reporter construct. The results of this experiment (shown in Supplementary Fig. 9A and described in line 327-329) show that the normalised fluorescence remains stable over 10 hours after chloramphenicol addition to the culture, ruling out that the presence of Nt significantly affects eGFP protein stability in E. coli. Importantly, this experiment also showed that in contrast to the chloramphenicol treated culture, in which the OD600 decreased after reaching stationary phase, the OD600 of the non-treated cultures slightly increased between 2 and 10 hours (Supplementary Fig. 9A). Because this increase was not observed in carbon starved Caulobacter cultures, we consider the different growth dynamics between E. coli and Caulobacter to be the most likely explanation for differences in eGFP accumulation at later time points during the experiment.

To further strengthen our E. coli data, we have analysed additional relevant Nt mutants that we identified as most critical mutants in our Caulobacter experiments presented in Fig. 5, namely dfsNt, mutD1, mutD2, ΔAAI and AAI>DDK. Determination of Δt and Δf values for the E. coli strains carrying these different Nt constructs showed similar results as for the corresponding constructs in Caulobacter. Collectively, these new data further support the notion that Nt operates in E. coli through a conserved inhibitory mechanism of translation. These data are now included in a reorganized new version of Fig. 6 (panels A, B) as well as in Supplementary Fig. 9.

**Minor comments:**

There are also 6 minor issues that are easily addressed, most by small changes to the text, and these should improve this otherwise fine manuscript.

Issue 1

Line 88 Fig. 1A shows DnaA degradation upon entering stationary phase from a low glucose media and not a simple starvation response to one component like glucose. Did the authors consider trying simple washout experiments, i.e. resuspend the cells in glucose-free media? This would have the advantage of suddenly exposing the cells to starvation and thereby studying the sudden response rather than the slower lingering response which would be due to many factors and not just glucose removal.

In a previous work from our lab (Leslie et al 2015), we have conclusively shown that the downregulation of DnaA synthesis depends primarily on the nutrient content of the growth medium.

Besides being in continuity with our previous work, we think that the starvation protocol that we used in the present study, and that was also used by the Sean Crosson lab (Boutte et al. 2012), might better reproduce what happens in the natural environment when nutrient levels gradually decrease until becoming limiting for bacterial growth.

Issue 2

Reference 16 should be cited are the first publication to show that glucose and other starvations induce DnaA degradation in Caulobacter.

We have added Reference 16 to the first sentence of the results section, in which we state that DnaA levels decrease when cells are shifted from a glucose-supplemented minimal medium to a glucose-limiting medium.

Issue 3

Fig. 1D shows that the TOEprint is not changed by adding the ribosome, very surprising considering its size and SD docking & alignment. 2 Minor bands then appear when the tRNA-Met is further added. These are presumably the "toeprints". A control with just the added tRNA-Met would make this result much more significant.

In the existing literature, there is a common consensus to consider real toeprints (i.e., indicative of the presence of an assembled translation pre-initiation complex) as only those bands that appear faintly in the presence of the 30S ribosome subunit but that become clearly enhanced upon addition of the initiator tRNA-fMet. Some examples can be found in Hoekzema et al 2019, Romilly 2014, Romilly 2020. In cases when the translation start site is buried in a structural element, the intensity of the toeprint signal is further increased when the mRNA is rendered unfolded, as seen in our data.

tRNA-30S-independent bands always show up in toeprint experiments, but their intensities differ with the sequence of the mRNA and sometimes the choice of RT used for primer extension. Addition of initiator tRNA-fMet alone is commonly not done in toeprint experiments (see references mentioned above). Finally, we want to point out again (see also our answer on “Problem 1”) that the toeprint data are very much consistent with our in silico predictions and our in vivo mutagenesis data. Therefore, we are confident that the observed toeprint upstream of the AUG corresponds to the true ribosome binding site.

Issue 4

Why does the cell OD drop, e.g. in Fig. 2, is it cell death from starvation?

We don’t think that the slight reduction of OD600 observed in our experiments is due to cell death. Based on our knowledge, carbon starved cells remain viable up to 24 hours after the starvation onset. Instead, we have observed a cell volume reduction, which may at least partially explain the observed OD600 decrease.

Issue 5

Line 327 Discussion "This study reveals a new mechanism, by which some bacteria can regulate the synthesis of the replication initiator DnaA in response to nutrient availability by modulating the rate of translation." Rate of translation or rate of translation abortions (as implied in Fig. 6)?

The rate of translation is the result of multiple contributions such as initiation, elongation, abortion and termination. Our data indicate that Nt is a regulator of DnaA translation elongation responding specifically to the nutritional state of the cell. Translation abortion could be one of the possible outcomes (but not the only one) of ribosome stalling. For these reasons, in the new version of the manuscript, we added the word “elongation” at the end of the sentence mentioned by Reviewer 2 (line 354).

Issue 6

It seems that that for most experiments with the eGFP the translation and protein decay components of the signal could have been easily uncoupled by running a parallel +chloramphenicol control. For example, this would simplify the interpretation of Fig. 6 where Nt eGFP stabilities are an issue and it is important to establish that comparable protein stability with and without the Nt peptide.

To address the reviewer’s comment, we have now included a chloramphenicol control experiment (stability assay) performed with E. coli carrying the 5’UTRdnaA-Nt reporter construct (Supplementary Fig. 9A). Please, see the response above for more details. For the experiments with the Caulobacter 5’UTRdnaA-Nt reporter we show in Supplementary Fig. 7 that the Nt peptide has no destabilising effect on eGFP.

Reviewer #2 (Significance (Required)):

Caulobacter crescentus is a model bacterium that has provided many insights into bacterial physiology that are now exploited to understand many organisms. These present results may provide one such example. It is known that the first amino acids of translated peptides can influence increase or impede exit from the ribosome, so this is a potential translation-level regulatory point that might be used by many organisms. This manuscript gives a concrete and important example of such usage suggesting that it many be widespread. Therefore, this work should find a wide audience and it should stimulate research in many other systems.

My lab also studies Caulobacter crescentus and we studied the same dnaA gene and protein including starvation responses. We at present do not have projects on dnaA but we do study other regulators and regulatory mechanisms of chromosome replication in Caulobacter crescentus.

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

Summary:

The Jonas lab provide good evidence that they have found a new mechanism to regulate the amount of the DnaA protein by a starvation signal. The DnaA protein is the key chromosome replication initiator probably for most bacteria and as such DnaA is the target of many regulatory inputs. The authors created an accurate reporter system that allows them to dissect the 5' mRNA translated and untranslated sequences of dnaA and they have convincingly demonstrated that the N-terminal DnaA peptide sequence and not the RNA mediate the response to starvation by glucose exhaustion. This is potentially a model example for global translational responses in bacteria.

Major comments:

The main conclusion, i.e. that the DnaA leader peptide "Nt" mediates this response is convincing. However, there were 2 major problems that should be easily addressed. These do not subtract from the main conclusion.

Problem 1

E. coli 30S subunits were used in the "Toeprint" assay of Fig. 1. Obviously Caulobacter 30S Ribosome subunits should have been used, or a justification should be given. One remedy would be to make this supplementary information.

Problem 2

The results in Fig. 6B could be due to the Nt simply making the hybrid protein more unstable in E. coli. This is the main impression given by the drop in signal. In this case, the conclusion would be wrong, and Nt is not transferring a starvation translation block from C. crescentus to E. coli. Nt is just making the protein unstable. These results should be treated as preliminary pending protein stability measurements. However, this defect does not subtract from the other main points and without the Fig. 6 E. coli experiments they still make a complete and interesting story. One remedy would be to make this also supplementary information.

Minor comments:

There are also 6 minor issues that are easily addressed, most by small changes to the text, and these should improve this otherwise fine manuscript.

Issue 1

Line 88 Fig. 1A shows DnaA degradation upon entering stationary phase from a low glucose media and not a simple starvation response to one component like glucose. Did the authors consider trying simple washout experiments, i.e. resuspend the cells in glucose-free media? This would have the advantage of suddenly exposing the cells to starvation and thereby studying the sudden response rather than the slower lingering response which would be due to many factors and not just glucose removal.

Issue 2

Reference 16 should be cited are the first publication to show that glucose and other starvations induce DnaA degradation in Caulobacter.

Issue 3

Fig. 1D shows that the TOEprint is not changed by adding the ribosome, very surprising considering its size and SD docking & alignment. 2 Minor bands then appear when the tRNA-Met is further added. These are presumably the "toeprints". A control with just the added tRNA-Met would make this result much more significant.

Issue 4

Why does the cell OD drop, e.g. in Fig. 2, is it cell death from starvation?

Issue 5

Line 327 Discussion "This study reveals a new mechanism, by which some bacteria can regulate the synthesis of the replication initiator DnaA in response to nutrient availability by modulating the rate of translation." Rate of translation or rate of translation abortions (as implied in Fig. 6)?

Issue 6

It seems that that for most experiments with the eGFP the translation and protein decay components of the signal could have been easily uncoupled by running a parallel +chloramphenicol control. For example, this would simplify the interpretation of Fig. 6 where Nt eGFP stabilities are an issue and it is important to establish that comparable protein stability with and without the Nt peptide.

Significance

Caulobacter crescentus is a model bacterium that has provided many insights into bacterial physiology that are now exploited to understand many organisms. These present results may provide one such example. It is known that the first amino acids of translated peptides can influence increase or impede exit from the ribosome, so this is a potential translation-level regulatory point that might be used by many organisms. This manuscript gives a concrete and important example of such usage suggesting that it many be widespread. Therefore, this work should find a wide audience and it should stimulate research in many other systems.

My lab also studies Caulobacter crescentus and we studied the same dnaA gene and protein including starvation responses. We at present do not have projects on dnaA but we do study other regulators and regulatory mechanisms of chromosome replication in Caulobacter crescentus.

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

Felletti et al provide compelling new evidence that a CDS element in the dnaA mRNA is required for nutrient dependent translationol control. This provides a mechanisms by which dnaA translation is shut off during carbon starvation, and is supported by a rather rigorous analysis of the mRNA performed both in vitro and in vivo. Overall it was a pleasure to read and the data are generally very compelling. My specific comments are below:

Major Comments:

While the authors rule out differences in charging of different ala-tRNAs as controlling the nutrient dependent repression in translation, the authors assume that this must be due to the nascent sequence. However, could it also be possible that all ala-tRNA isoacceptors have lower charging after C-starvation?

Minor comments:

It was observed many years ago that tmRNA is required for the proper timing of DNA replication initiation in Caulobacter (Cheng and Keiler J Bact 2009). Since the AAI motif is appearing to alter translation elongation, it might be interesting to discuss the AAI motif may be linked to ribosome arrest and rescue.

Line 49 - add "initiation"

Line 61 - is "cleared" meant to be proteolyzed or simply meaning to have a lower protein level?

Line 92-93 - is this 5' UTR based on a previously defined TSS determined in their previous study?

Line 115-118 - this is interesting, might this conserved 5' UTR be added to rfam?

Line 126-127, 131,189 - Is the 3nt sequence the authors found here considered a Shine-Dalgarno site? I would imagine that this would be too small to consider this. Perhaps calling it SD-like sequence might be more appropriate.

Lines 136-140, 208-210 - Would the authors consider this upstream site with a potential CUG start codon a standby site? It appears to fit many of the criteria which could be used to define one.

Lines 253-255 - this is a beautiful experiment, but very hard to understand from the text. Perhaps add a sentence or two to explain it in more detail.

Line 307 - add "synonomous"

When dnaA is depleted, it was observed that the chromsome can be erroneously segregated by the ParA/B/S system (mera et al PNAS). Does this occur in C-starvation when DnaA levels drop?

Significance

Appears to be quite significant to researchers studying regulation of bacterial cell cycle and translation. Since DnaA is conserved across bacteria, and this mechanism works in E. coli, it appears that the findings will likely be important in many bacterial systems.

Referee Cross-commenting

All the reviewer comments I read seem reasonable. Specifically, I found the point about E. coli 30S ribosomes is very important that the authors address. This could be done in writing, but should be better listed as a caveat to those experiments.

Transparent Peer Review

---

Download the complete Review Process [PDF] including:

• reviews
• authors' reply
• editorial decisions

</br>

Transparent Peer Review

---

Download the complete Review Process [PDF] including:

• reviews
• authors' reply
• editorial decisions

</br>

Transparent Peer Review

---

Download the complete Review Process [PDF] including:

• reviews
• authors' reply
• editorial decisions

</br>

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

Learn more at Review Commons

---

Reply to the reviewers

The authors wish to thank all three Reviewers for their appreciative comments regarding our ECPT and for very useful suggestions. Response to all points raised are presented below, we hope that the responses and new experiments proposed in the following pages will fully address remaining concerns.

Reviewer’s comments to the BiorXiv paper by Chesnais et al, 2021

“High content Image Analysis to study phenotypic heterogeneity in endothelial cell monolayers”

https://www.biorxiv.org/content/10.1101/2020.11.17.362277v3

---

Reviewer #1 (Evidence, reproducibility and clarity (Required)):

The authors highlight the importance of endothelial heterogeneity using endothelial cells from different tissues. They examined aortic and pulmonary endothelium as well as HUVECs. They cultured the cells in identical conditions and also stimulated them with a physiological concentration of vascular endothelial growth factor as well high concentrations as would be found in cancers. They developed a profiling tool that allowed analysis of individual endothelial cells within a monolayer and quantification of inter-endothelial junctions, Notch activation, proliferation and other features.

**Major comments**

1. It would be useful to apply this technology one step beyond two-dimensional culture, to use vessels opened up longitudinally so that one can see the monolayer of endothelial cells and assess whether it is relevant in primary material in situ. I think this would be a major utility of the whole approach.

R: We thank this reviewer for the suggestion. In vivo analysis is not in the objectives of the paper. However, we propose to perform “En face” staining of murine blood vessels following the protocol in the reference below. We will perform stainings for murine CDH5 (VE-Cadherin), NOTCH1 intracellular domain, HES1 and DNA which parallel that used in vitro on human EC. We will then apply our revised ECPT workflow and present data in a new Figure.

En Face Preparation of Mouse Blood Vessels. Ko KA, Fujiwara K, Krishnan S, Abe JI. J Vis Exp. 2017 May 19;(123):55460. doi: 10.3791/55460. PMID: 28570508

2. There are some very nice images here but disappointed not see a field that could show staining and markers for several of the target proteins and thus show the heterogeneity and randomness or organisation of the endothelial cells.

R: We thank the reviewer for the appreciative comment. We propose to include representative microphotographs to illustrate the heterogeneity of different EC monolayers in the revised version of the manuscript. Furthermore, to further illustrate these aspects we will also include spatial correlation maps of cells and features measured with ECPT as explained below.

---

---

---

---

3.

• The Notch signalling is an important aspect of this work, particularly evidence of lateral inhibition would have been of value. For example, one might expect cells adjacent to each other to have alternating high and low NICD. R: We thank the reviewers for the suggestion. To address this, we are currently developing a new module to perform spatial autocorrelation analysis based on cell maps built using ECPT. In particular we have developed a new module to export cell maps as spatial objects in R which can be then analysed using the adespatial R package and provide correlation metrics such as the Moran’s autocorrelation index (see reference below). The index works with continuous data, removing the need to establish arbitrary thresholds and thus provides formal metrics to demonstrate heterogeneity in EC monolayers. We have derived this index as an example of such analysis for synthetic data and for one ECPT cell map as shown below.

Figure 1: Moran’s spatial autocorrelation analysis using R and adespatial package. Moran’s index has values between –1 and 1. If adjacent cells had a consistent tendency to acquire alternate high and low NICD values, the corresponding bivariate Moran’s index would have an I+ value ~ 0 and an I- value approaching -1. In the example cell map both I+ and I- have relatively small absolute values and large p values which suggest a random cell distribution. The analysis was performed on synthetic data and ECPT derived data (HUVEC at baseline).

• *

Community ecology in the age of multivariate multiscale spatial analysis

S Dray et al, Ecological Monographs, 2012. doi:10.1890/11-1183.1

• NICD staining alone does score the extent of the signalling because of many factors that can influence the transport of the cleaved NICD. Really a marker of Notch signalling downstream e.g. HES or HEY family, DLL4 fis needed to give more information about this critical aspect. R): We thank the reviewer for the suggestion. We are currently performing HES1 staining (with no Pha staining) along with a new NICD mAb (see below). Preliminary qualitative data (Fig 2) show that HES1 staining also reveals single cell heterogeneity of NOTCH activation in the same monolayer. We will include ECPT analysis of HES1 and correlation with NICD and other features as suggested. We will reformat the current Fig 5 to include HES1 analysis and improved metrics of NOTCH pathway activation including spatial analysis (point 3 above).

Figure 2: HES1 immunostaining on HUVEC (Image enhanced for visualisations). Cell nuclei labelled as 1, 2 and 3 have raw mean grey values of HES1 signal equal to 2271, 11210 and 48261 (C2/C1 and C3/C2 >4 folds).

---

---

---

I really do not think that in Figure 5 it is justified to have a red line drawn through the cloud of points. The correlation coefficient is so low that this is meaningless. The failure to distinguish a P value from biological relevant is worrying. Much better comparison would have been between NICD staining and a downstream gene regulated by notch.

R: We appreciate the reviewer’s concerns and are presenting our analyses of NOTCH activation using new immunostainings (HES1) and robust metrics for NOTCH activation as discussed above. We will therefore remove the mentioned corelation plots from the reviewed version of the manuscript.

It is important to know that the antibodies used for staining have be validated by the investigators. They would need to show a single band on Western blots or be able to block staining on immunohistochemistry. We all know the manufacturers can be unreliable and use high concentrations of proteins for Western blots. These should be added as a supplementary figure.

R: While the paper was under revision the AB8925 (NICD, Abcam) has been retracted from the market. To address this major concern, we have decided to acquire a different antibody targeting the intracellular portion of NOTCH receptor and validated its specificity by western blot. Fig 3 below, show western blots demonstrating a clean band at ~98 Kd as expected for cleaved NOTCH1 intracellular domain (NICD).

We are currently repeating the whole experiment presented in the current version of the manuscript and the ECPT analysis using the new antibody and including HES1 one of the canonical NOTCH target genes as also suggested by this and other Reviewers. We will provide WB analysis of all antibodies used in the paper in a supplementary figure in the revised manuscript.

Figure 1, WB analysis (NOTCH1 intracellular domain, AB52627, Abcam). of HUVEC (lanes 2,3), HAoEC (lanes 4,5) and HPMEC (lanes 6,7)

Reviewer #1 (Significance (Required)):

This represents a valuable and thorough methodology likely to be highly useful to many groups and show new insights into endothelial biology.

Wide audience, cancer, cardiology, vascular disease-covid.

My expertise >100 papers on angiogenis in cancer, basic mechanism, therapy models, bioinformatics IHC, patients, clinical trial. H score 190 Google Scholar

R: We thank Reviewer One for their very appreciative comments and we hope that the proposed revisions will fully address remaining concerns.

Reviewer #2 (Evidence, reproducibility and clarity (Required)):

The manuscript by Chesnais et al reports development of workflow for analysis of cultured endothelial cells , which they call Endothelial Cell Profiling tool (ECPT). Using ECPT they analyse several parameters in three different endothelial cell types (HuAEC, HUVEC and HPMEC), such as cell morphology, activation of cytoskeleton, VE-cadherin junctions, cell proliferation and Notch activation, under steady conditions and upon treatment with VEGF. The analysis allows to observe some predicted changes, such as increase in cell cycle and junctional activation in cells treated with VEGF-A, and such changes are highly heterogeneous. Overall, this is a potentially useful albeit not revolutionary tool for batch analysis of cultured endothelial cell phenotypes.

I have the following comments:

1. To make their case the authors should provide a comparison with other currently used approaches for EC phenotypic analysis in vitro - what is the advantage of using ECPT? The authors repeatedly use the term "single-cell level of analysis ", but this is in fact the case of any IF based analysis of cultured cells.

R: We thank the reviewer for the suggestions. Indeed, several tools for imaging based single cell phenotyping are available. However, ECPT represents an improvement under several aspects. First, it allows improved segmentation of difficult-to-segment and heterogeneous cells; second, ECPT allows multi-parametric analysis on large image datasets in a semi-automated and structured way facilitating downstream data analysis; third, ECPT is open source.

Furthermore, ECPT is a very flexible workflow including tools which facilitate and automate several tasks such as systematic images re-labelling and grouping. We will now draft a table including a complete list of features and improvements in comparison to other available tools and include it in revised manuscript in appendix1 and include analyses which are not implemented in any currently available software such as spatial autocorrelation.

I strongly recommend to stain HPMECs for PROX1, these cells are frequently 100% lymphatic endothelial cells. In this case the authors compare different lineages and not blood endothelial cells from different locations.

R: We thank the reviewer for the suggestion. We will address this with a new characterisation as supplementary figure in the revised manuscript. We are currently performing a qRT-PCR screening of several EC marker including arterious, venous and lymphatic markers (e.g., CXCR4, Tie2, CDH5, PROX1, LYVE1 as well as baseline NOTCH1 and Dll4 and downstream genes such as HES1 and HEY2.

Please provide evidence for specificity of NICD antibody.

R: We thank this reviewer for the suggestion. Please see response to Reviewer one point 5.

Figure 1: HPMEC picture appears out of focus

R: We thank this reviewer for noticing, we will now include a clearer picture in revised version of the manuscript.

Figure 3 A - it is not entirely clear what is the difference between activated and stressed phenotype, they look quite similar.

---

R: We will clarify the definitions of cell activation in revised version of the manuscript and present this analysis as supplementary material to demonstrate the flexibility of our ECPT rather than in main figures. We have removed Pha staining from the new experiments we are performing to allow HES1 staining and address NOTCH signalling in more details. The assessment of Pha and stress fibres in previous experiments will be moved to supplementary material. The classification is based on PhA staining using CPA classifier which was trained to distinguish among the two by the presence of stress fibres. The general rule to place cells in the stressed category during training of the CPA model was the observation of stress fibres crossing the nucleus while cells with peripheral bundles of actin were placed in the activated category.

---

Figure 5 - what is the difference in NICD localization between "high" and "On" conditions?

R:

Since it has been noted by this and other reviewers that this classification might be difficult to interpret and in fact, the established thresholds are somehow arbitrary, we will completely revise the way we present analysis of NOTCH activation data including downstream analysis and more formal metrics of spatial correlation and extent of activation eliminating the need to impose thresholds (also see response to Reviewer one point 3).

Since the authors make a correlation between Notch activity and junctional stabilization, it would be important to confirm this by other means, such as analysis of Notch target genes.

R: We thank this reviewer for the comment which resonate with this and other Reviewers’ comments. We will include HES1 analysis in the revised manuscript, please see Response to point 6 and reviewer’s one point 3 above.

• *

**Technical and minor**

1. Methods mentions HDMECs (human dermal microvascular endothelial cells) but the authors discuss HPMEC throughout the text 2. Please add scale bars on all microscopy pictures. 3. Please provide the information on what isoform of VEGF-A was used for stimulation and the rationale for selecting the concentration.

R: We thank this reviewer for flagging these imprecisions and we will fix them in revised version of the manuscript.

Reviewer #2 (Significance (Required)):

The authors provide a workflow for the phenotypic analysis of cultured cells. Such tool is potentially useful, although the examples the authors show do not reveal striking examples of why such analysis is better in comparison to existing approaches. My guess is that the analysis may be faster and less tedious, once the training sets are generated, but this is not specified. My speciality is endothelial cells biology.

R: We thank this reviewer for their very useful and appreciative comments. As mentioned above we will expand appendix 1 to fully explain potential and utility of our ECPT and review the main text to clearly highlight main advantages.** We hope that our plan for revision will fully address remaining concerns.

Reviewer #3 (Evidence, reproducibility and clarity (Required)):

**SUMMARY:**

The manuscript by Chesnais et al. presents a novel endothelial cell (EC) profiling tool (ECPT) which provides spatial and phenotypic information from individual ECs, and was tested with a variety of specialized EC subtypes (arterial, venous, microvascular). They present a high throughput immunostaining and imaging-based platform using culture of human ECs on 96-well plates and capture of fixed, stained samples on a Perkin Elmer Operetta CLS system. The authors report the use of this ECPT tool to investigate EC phenotypes from human umbilical vein ECs (HUVEC), human aortic ECs (HAoEC) and human pulmonary microvascular (HPMEC) in relation to 50 ng/mL VEGF stimulation for 48 hours, and the general parameters of proliferation, Notch activation and stress fiber rearrangement (F-actin), and present this as a prospective platform to examine differences in EC phenotypes and responses at a more individualized level.

**MAJOR COMMENTS:**

1. Fundamentally, the advantage of single cell technologies is the ability to segregate populations to make novel observations. One area that would be of interest to explore in this manuscript using this ECPT platform would be reporting the results from single cell analysis that is then subsequently pooled within a sub-population, rather than sub-stratifying populations to reflect the multiple phenotypes that may be present within a single "confluent" well. With analysis of EC heterogeneity, it would be of interest to differentiate heterogeneity within EC subtypes at the culture/treatment conditions presented, and heterogeneity between EC subtypes.

R: We thank this reviewer for the suggestions, we believe that the new approach to evaluate heterogeneity through spatial autocorrelation can provide a much better and clearer picture of this aspect (see responses to Reviewers One point 3 and Two points 6 and 7. Furthermore, we are currently restructuring the ECPT data structure to a more intuitive layout (list of lists rather than a single huge data frame) without affecting downstream data presentation. We will also update our Shiny App to enable the user to perform analyses on data subsets of interest without any R coding, we will present examples and walkthrough of this approach in appendix.

2.

The term "stable IEJ" is used and refers to 48h after seeding 40,000 cells on a 96-well plate, but it is unclear how the authors defined or demonstrated a "stable" junction. In previous reports, longer-term culturing of EC monolayers well beyond the point of confluence has been shown to result in junctional complex rearrangement (Andriopoulou P et al. Arterioscler Thromb Vasc Biol. 1999; reviewed in Bazzoni G & Dejana E. Physiol Rev. 2004). To this point, the fact that the different EC subtypes investigated had different percentages of "quiescent cells" suggests that the monolayers were not completely quiescent. The statement that the IEJ classification is "an immediate index of EC activation in contrast to quiescence" should be further supported by references or data. The definition of quiescent EC as simply non-proliferating, non-migrating is somewhat reductionist, and oversimplifies EC states. The authors state that HAoEC and HUVEC "...appeared more active...", but it is unclear what "active" means, and whether this may simply reflect that these cells had not yet reached confluence or quiescence in the 48h total culture time. As well, it is unclear how "migratory phenotypes" could occur in confluent monolayers. It would be helpful to see the data for these observations. If leaving ECs longer in culture, are the authors able to achieve a higher percentage of quiescent cells?

---

R: We thank this reviewer for the very insightful comments and for suggesting the references. Indeed, we considered these aspects carefully. Regarding cell culture density and confluency, we previously tested seeding densities of 30000-60000 cell/well of 96 well plates (0.32 cm2, ~95000-190000cells/cm2) and we selected 40000 as the maximum seeding density allowing adhesion of >99% of cells. For HUVEC, a seeding density of 40000 cells/well (125000 cells/cm2) produced a high-density culture immediately after seeding (close to what reported for long-confluent cultures in Andriopoulou P et al, ATVB 1999, 140000 cells/cm2). We allowed further 48h culture aiming to achieve junctional “stabilisation” and “maximal” cell density. For consistency, we also seeded 40000 HAoEC and HPMEC per well in all our experiments, however both cell types are significantly larger than HUVEC (Fig 4). For all cells cultures we used EGM2 medium which has few differences with that reported in Andriopoulou P et al, namely, absence of antibiotics and antimycotics and use of defined cocktail of recombinant growth factors instead of Endothelial Cells Growth Supplement. In the past we compared HUVEC cultured in EGM2 and supplemented M199 medium and in our experience EGM2 promotes higher proliferation rates in sub-confluent cultures but similar morphology upon confluency. Is notable that several other factors (including flow, matrix, perivascular cells) are absent in our culture conditions and therefore the homeostatic balance found in vivo might not be fully achievable under our experimental conditions. However, we argue that the described culture conditions should be sufficient to reach a bona fide relatively quiescent EC phenotype in culture.

Save these considerations, we agree with this reviewer that providing examples of longer-term cultures would help substantiating our findings and further validate the ECPT approach. We will perform a supplementary experiment to evaluate this aspect by comparing 48h cultures with longer culture times (72h and 96h). Furthermore, we will expand the methods section with the details discussed above and in relation to the suggested references.

• *

Regarding the definition of “stable IEJ” and “active EC”, we used this terminology referring exclusively to our measures of IEJ stability (STB index) and Pha based cell classification where we used the terms of “quiescent”, “active” or “stressed”. Therefore, all statements mentioning more or less “stable IEJ” or “active” EC are relative to the specific context of our experiment (not in absolute terms).

Overall, we appreciate that the terminology we employed is a source confusion and might suggest inappropriate over-interpretation of our results. We will correct the text in the manuscript to avoid this confusion and to clarify that our observations are valid within the context of our in vitro conditions. In particular, we will present the data regarding junctions as proportions of different junction per cell, and we will rename cell “activation” categories based on PhA immunostaining using more neutral terms (e.g., No Fibres, Peripheral Bundles, Stress fibres). Finally, we will also attempt to generalise our observations to more physiologic context by performing immunostaining on “en face” preparation of murine blood vessels (cfr response to R1 point 1).

Fig 4: Cell area density distribution for HUVEC, HAoEC and HPMEC in baseline conditions.

Could the authors comment on the baseline NICD immunoreactivity in the nuclei in HAoEC and HPMEC compared to HUVEC? Is this a reflection of active NOTCH signaling? Or rather, is it possible contact-inhibition (and downregulation of NOTCH) may not have occurred? Demonstration of EC quiescence would help to ensure similar cell cycle states. The definition of "Notch-positive" and "Notch-negative" cells is a bit misleading, as NICD levels and localization are a better indication of canonical Notch activation, and not the presence or absence of Notch protein(s). Further, NICD activation is also dependent on the levels of Notch ligands, which was not addressed. Are the authors able to confirm "OFF", "Low", "High", and "ON" classifications based on NICD intensity and localization with downstream Notch gene activation at a single-cell level? Or correlation between NICD status and the phase of cell cycle or proliferation status?

R: We thank this reviewer for the comment. Overall, NICD either nuclear or cytoplasmic can give a measure of how much a cell is relaying canonical notch signalling in a small timescale (minutes, which is also the timescale affected by lateral inhibition, Sjoqvist M and Andersson ER, Dev Biol, 2019). By evaluating single cells in the context of their population in multiple fields of view and samples we can get an indication of how frequently a particular cell type tends to actively transduce canonical NOTCH (under confluent conditions). As this and other reviewers have pointed out NOTCH signal transduction mediated by NICD can be affected by several factors limiting the potential to infer actual activation of the pathway (i.e., downstream gene transcription. As suggested by this and other reviewers we are including measures of downstream gene activation, in particular we have included HES1 staining in our workflow, and we will include these data in a new analysis (also see response to R1 point 3). We will also provide new metrics of spatial autocorrelation to evaluate the tendency to lateral inhibition (R1 point 3) and correlation between parameters using continuous mesures and therefore we will remove the previous classification based on thresholds. Finally, we are performing a qRT-PCR screening to assess baseline levels of DLL4, NOTCH1 and JAG1 which we will present as supplementary material.

Do as I say, Not(ch) as I do: Lateral control of cell fate

Sjoqvist M and Andersson ER, Dev Biol, 2019

PMID: 28969930

The existing workflow/platform is adapted for images obtained from the Operetta CLS system (Perkin Elmer) and Harmony software (proprietary), which may not be available for broader users in the EC field. It would be helpful to include ImageJ macros for the bulk automatic import of TIFF, renaming and upscaling of resolution/bit quality to match the formats that are compatible with the software.

R: We thank this reviewer for the comment. We have now included an ImageJ macro (available in the GitHub repository) which in principle can import and elaborate images from any source. We didn’t include a specific option in our current user interface because the relabelling operates by parsing original filenames into fields which are then renamed according to user input and each HT platform adopt different regular expression to encode filename. Any user with a basic literacy in ImageJ macro scripting can achieve relabelling and elaboration of their own file given that their filenames use regular expressions which can be parsed. Also, it is relatively easy (again by modifying the macro) to include user defined pre-processing steps including image scaling. An example of parsing method for Operetta CLS filenames is provided in appendix 1.

Could the authors comment on the manpower (hours from start to finish for experiments, staining, imaging, analysis, etc.) and cost of the ECPT pipeline relative to emerging single cell technologies such as single cell-RNA sequencing.

Further, one major advantage of imaging technologies is the ability to assess live cell dynamics, which are particularly relevant in response to stimuli and agonists. Have the authors utilized the ECPT platform for these approaches, in particular, to assess the differential EC subtype dynamics in proliferative conditions?

R: In terms of manpower the workflow is not very demanding. Our current dataset is based on images extracted form 4 independent experiments (18 wells each). The process is sequential, therefore a single user trained in cell biology, automated microscopy and in the use of the different ECPT components (ImageJ, CP, CPA and R) could perform the experiment alone. The timing of each experiment will depend on circumstantial factors, however once the ECPT is trained for specific user’s requirements (which can require some trials and errors depending on user’s experience) the whole process from cell fixation and staining, through image acquisition (~2 h acquisition for each experiment on an Operetta CLS system), to dataset build-up can take less than one week. For example, elaborating the current image database (~6000 images for four fluorescence channels) which data are presented throughout, had the following raw elaboration times on a Mac Book Pro 2017 equipped with an intel i7 processor and 16 Gb of RAM:

- Image pre-processing and relabelling ~1h

- Generation of probability maps for VEC and NICD ~3h

- CP pipeline run (Cell segmentation, objects measurements and classification) ~16h

- Data import (R studio) ~20m

• *

We will measure these parameters more precisely in the new experimental run and present timings for each step in a new table in appendix 1.

• *

After main dataset is created R studio can perform most statistical analyses and data plotting almost instantly.

• *

We fully appreciate the value of employing ECPT in live imaging setups and we believe it is one of the most promising future applications. We didn’t address live microscopy experiments in the context of ECPT development and validation presented in the current manuscript therefore we cannot present example data or proof of concept. However, we can confidently comment that time lapse experiment would not endow further layers of complexity in terms of image analysis workflow. Therefore, given appropriate set of live markers (e.g., transgenic fluorescently tagged CDH5 for EC segmentation and junctions analysis) we believe that the current implementation of ECPT is already fully equipped to facilitate elaboration and analysis of imaging data derived from time lapse experiments.

The authors should discuss the ability to amend or revise of the ECPT platform to incorporate analysis of additional markers that may be obtained through imaging, and discuss greater implications and utility to specifically tailor the workflow for other researchers in vascular biology, or to other monolayer culture systems. Further, they should better highlight the novel observations obtained with the ECPT compared to traditional methodology.

R: We thank this reviewer for the comments. We will provide evidence of ECPT flexibility within this manuscript by including, during the time of this review process, a new analysis for downstream NOCTH signalling (HES1). We will move analysis of cell “activation” (i.e., stress fibres analysis) to supplementary information and include a more through discussion of how automated single cell classification could improve content, speed, reliability and robustness of quantification tasks which are currently exposed to long and tedious processing times and conscious/unconscious observer biases.

**MINOR COMMENTS:**

We thank this reviewer for the very thorough revision of the manuscript. It is truly invaluable to us to improve it. Below responses to specific technical points, we will fix all stylistic, formatting and typographical issues in revised version of the manuscript.

1. There are minor typographical, capitalization and grammatical errors throughout.

R1: Thanks, we will fix these in updated version of the manuscript.

Why was fibronectin used to coat plates, and what was rationale for using this ECM substrate versus gelatin (most commonly used in EC cultures) or type I collagen?

R2: We used fibronectin for immunostaining experiments similar to what reported in our previous work (Veschini et al, 2007, 2011, Wiseman et al, 2019) and also in Andriopoulou P et al,1999. In general, in our experience FN gives better cell adhesion in comparison to gelatin when culturing EC on glass or other substrates different from cell culture plastic. FN is the cell culture substrate recommended by Promocell therefore, we also used FN for cell expansion to avoid any phenotypic change which might have been caused by switch in cell culture substrate.

3. Based on the various box plots present throughout the figures, it appears that some parameters have a large range of values. Is it possible or helpful to set minimum and maximum exclusionary criteria? Further, in the way that these data are presented, it is difficult to appreciate the effects of a treatment such as 48h of VEGF, as the magnitude of STB Index difference, for example, appears small, and it is difficult to understand whether these significant differences are biologically relevant, as assessed.

R3: We agree that in absence of exclusion criteria it is difficult to infer biologic meaning out of subtle differences (e.g., the tiny difference in STB index between HAoEC in presence or absence of VEGF). In the current version of the manuscript, we attempted to be agnostic in regards whether some observed small but significant mean differences could endow biologic meaning and discussed larger variation as biologically meaningful, for example the differences in STB index among cell types. We argue that tiny differences in the distribution of some selected parameter across experimental conditions could reflect underlying mechanisms masked by biologic noise, therefore catching a glimpse of these variations via ECPT could inspire novel experiments to specifically address their full biologic significance.

To the interest of better understanding of the current manuscript we will re elaborate our data to provide more immediate metrics and highlight outstanding features.

---

Use of arrows and further description in Figure 1 would help the reader understand what specific features are different in the various EC subtypes. As well, the representative micrographs for HPMEC appear blurry compared to other panels (Fig. 1).

In Figure 2, the panels in A, B and C do not correspond horizontally, and it may be cleared to demonstrate "Segmentation & features extraction" overlays from the same representative micrographs shown in panel A. Labeling of the individual panels and software used for panel B would help the readership understand what is being quantified and how. The second panel in "C" appears blurry.

In Figure 3, labelling the color code for quiescent, activated and stressed categories on graphs and in legend would be helpful to easily identify populations.

R4-6: Thanks, we will fix these in updated version of the manuscript.

For Figure 4, line separators or more obvious grouping to distinguish discontinuous, linear and stabilized junction types in panel A. What proportion of the different EC subtypes contains discontinuous, linear and stabilized junctions at confluence/quiescence? Is there a correlation between discontinuous junctions and proliferating cells?

---

R7: We will perform new analyses to address correlation between proliferation and junctions and proliferation vs HES1. We will restructure data presentation on junctions to display different proportion of junctions per cell or per cell type rather than a unified value (STB index).

---

It would be useful to distinguish the effects of published mediators on junctional integrity in intact EC monolayers (i.e. histamine; VEGF) from those shown in this automated quantitation. It appears that 50 ng/mL of VEGF treatment for 48h only slightly increases STB index based on panel C.

R7c__: __We agree that increase of STB index in HAoEC and HPMEC upon VEGF treatment might not be highly biologically meaningful, save consideration in point 3 above. However, difference in HUVEC (+- VEGF) is visually appreciable in images (i.e., VEGF treated HUVEC seem to have more linear junctions) therefore we believe that the ~16 units difference in STB index is biologically meaningful. As discussed in point 7 above, we will restructure data presentation to better clarify these aspects.

---

Figure 5 panel B should provide legend in graphs/figures or figure legends to highlight the color-coding matching the OFF, Low, High and ON groups. Further, it is unclear the difference between "High" and "ON" groups. The authors state that "thresholds were selected empirically", however, it is unclear whether this was derived through utilization of known Notch activators or inhibitors, and how this relates to the threshold of Notch activity necessary to enhance proliferation or maintain quiescence. In Supplementary Figure 4 (which I believe is mislabelled as Supplementary Figure 5), shows only a weak positive correlation between nuclear NICD intensity and mean STB index. It would be of interest to see the plot from Supplementary Figure 5 for each of the EC subtypes, in the presence and absence of VEGF. As well, for Figure 5, on C and D panels, it would improve clarity to revise "Low" and "High" descriptors with "Low NICD activity" and "High NICD activity".

R8: As discussed above we will revise our analyses to remove NOTCH categories and instead show spatial autocorrelation analyses which work on continuous data.

In Supplementary Table 1, "Widt/length" should be "Width/length"

---

R9: Thanks, we will fix this in updated version of the manuscript.

For Supplementary Figure 3, it would be of use to show DNA distribution intensities from proliferating, non-confluent EC subtypes to demonstrate the validity of this methodology to identify cells in G0/G1, S and G2/M phases, as highlighted in panel A. Could the authors comment on the discrepancy between the percentage of cells identified as quiescent by ECPT and the percentage of cells in G0/G1? The comment that "VEGF induced a small detectable increase in proliferation rate in all EC" is curious, as a dose of 50 ng/mL of VEGF should be a relatively strong stimulator of proliferation/migration in ECs.

R10: We will perform ECPT analysis on sub-confluent or sparse cells to further validate our analysis. Qualitative data on preliminary images seems to confirm that the proliferation rate in sparse cells is very high (>70%). To perform the evaluation we followed and improved a previously published method (Roukos et al, Nat Prot, 2015)

Regarding the relation between cell in G0/G1 and assessment of “quiescent” phenotype (which nomenclature will be revised as discussed above), it is important to highlight that we reported data on stress fibres analysis (i.e., classification into “quiescent”, “activated” and “stressed” cells) only on the cells in G0/G1 (i.e., we excluded proliferating cells from this analysis as we assumed that all proliferating cells would be “not quiescent” and bias our estimation).

For Supplementary Figure 5, "Nuclear NOTCH intensity" on the Y-axis should read "Nuclear NICD intensity", as it does not appear that Notch was stained. It would also be of benefit to overlay the ranges for "OFF, Low, High and "ON" to appreciate ranges of activation. Is there any correlation between NICD nuclear intensity and proliferative index?

R11: We will present correlation between NICD or HES1 and proliferation in revised version of the manuscript.

Definitions should be provided for many terms. i.e. vascular endothelial-cadherin (VE-CAD; CDH5); HUVEC (human umbilical vein endothelial cell); HAoEC (human aortic endothelial cell); HDMEC (human dermal microvascular endothelial cell); NICD (NOTCH intracellular domain); VEGF (vascular endothelial growth factor); etc. at first appearance.

---

R12: We will add this information in revised version of the manuscript.

For EC subtypes purchased from commercial vendor, it would be of interest to understand how many unique donors these cells/data were derived from, and whether there are any differences in basic donor information such as age, sex, etc. Further, Promocell catalogs proliferative rate for each of their lot numbers, and it would be of interest how this compares to the values determined using the ECPT software analysis package.

R13: We will add this information in revised version of the manuscript.

1 In the "Cell culture" section of the methods, HDMEC from Promocell are listed, however, the manuscript and figures show data from HPMEC. Both EC subtypes are available from Promocell, however, HDMEC are from dermal origin.

1 Vascular endothelial-cadherin should be abbreviated "VE-CAD" or "CDH5" and not "VEC", as this is not a standard or gene notation, and will likely be confused with the more common abbreviations for venous or vascular EC. It seems as though "CDH5" is used most commonly throughout manuscript, so this should be used throughout.

1 The authors refer to "activated NOTCH" when describing antibodies in the methods, however, it would be clearer to the reader to simply refer to the antibody target (NICD), and mention that this reflects canonical NOTCH downstream activation.

The sentence in the "Immunostaining" methods "CDH5 is a lineage marker..." should be moved to results/discussion as these details are out of place in methods.

How were the 3 areas captured per wells designated? Were these locations the automated, and the same for all wells?

"Appendix - Figure" notation should be revised to "Appendix Figure" for consistency and to avoid confusion.

R14-19: Thanks, we will fix these in updated version of the manuscript.

How were artifacts and mis-segmented cell objects excluded?

R20: We will add this information in the revised appendix. As general rules, cells containing NaNs values in any of the parameters, cells fragments or merged cells (evaluated using area measurements) and cells with no detectable junctions were all excluded (total cell excluded from analysis were ~ 2.5 % of the initial dataset).

• *

In "Statistical analysis" "Tuckey's" should be "Tukey's". "HSD" should be defined "honestly significant difference" or simply removed, as Tukey's is most common name.

In "Statistical analysis", "significative" should be "significant" or "statistically significant".

Scale bars should be added to micrographs.

---

R21-23: Thanks, we will fix these in updated version of the manuscript.

Could the authors comment on the necessity of µclear plates, which substantially increases the cost per plate/experiment.

R24: m**clear plates were used to allow image acquisition with a 40x water immersion objective in the Operetta CLS (impossible with standard 96 well plates). Cell grown on coverslips and mounted on microscopy slides could be used as well with significant increase in acquisition time (Wiseman et al, 2019).

• *

Were other seeding densities and times investigated?

R25: We will evaluate sparse cells in revised version of the manuscript as discussed above.

---

More description on potentially novel observations between these three primary EC subtypes would be informative for the readership to appreciate

The references do not appear in chronological order. Further, consistency of reference formatting should be reviewed, and appropriate journal name abbreviations should be used.

R26-27: Thanks, we will fix these in updated version of the manuscript.

Reviewer #3 (Significance (Required)):

• This manuscript presents a conceptual and technical advance, introducing a high throughput imaging platform to assess endothelial phenotypes
• Within the field of angiogenesis, several tools exist, either proprietary, or leveraging ImageJ software to assist in assessment of cells. The ECPT provides a more complex analysis platform to integrate analysis of multiple endpoints
• This work would be of interest to vascular biology laboratories to adopt a more comprehensive view of heterogeneous endothelial phenotypes in vitro
• As a vascular biology researcher, I have had extensive experience with in vitro culture of various endothelial cell subtypes from human and mouse. My field of expertise gives me the perspective of the nuances of the direct handling and phenotyping of ECs, and have worked specifically worked with HUVEC, HAoEC and HPMEC, and assessed the impact of key factors relevant in angiogenesis such as VEGF, Notch and other mediators.

R: We thank the reviewer for the very appreciative comments, and we hope that with the revised version of the manuscript we will be able to fully address remaining concerns.

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:

The manuscript by Chesnais et al. presents a novel endothelial cell (EC) profiling tool (ECPT) which provides spatial and phenotypic information from individual ECs, and was tested with a variety of specialized EC subtypes (arterial, venous, microvascular). They present a high throughput immunostaining and imaging-based platform using culture of human ECs on 96-well plates and capture of fixed, stained samples on a Perkin Elmer Operetta CLS system. The authors report the use of this ECPT tool to investigate EC phenotypes from human umbilical vein ECs (HUVEC), human aortic ECs (HAoEC) and human pulmonary microvascular (HPMEC) in relation to 50 ng/mL VEGF stimulation for 48 hours, and the general parameters of proliferation, Notch activation and stress fiber rearrangement (F-actin), and present this as a prospective platform to examine differences in EC phenotypes and responses at a more individualized level.

MAJOR COMMENTS:

1. Fundamentally, the advantage of single cell technologies is the ability to segregate populations to make novel observations. One area that would be of interest to explore in this manuscript using this ECPT platform would be reporting the results from single cell analysis that is then subsequently pooled within a sub-population, rather than sub-stratifying populations to reflect the multiple phenotypes that may be present within a single "confluent" well. With analysis of EC heterogeneity, it would be of interest to differentiate heterogeneity within EC subtypes at the culture/treatment conditions presented, and heterogeneity between EC subtypes.
2. The term "stable IEJ" is used and refers to 48h after seeding 40,000 cells on a 96-well plate, but it is unclear how the authors defined or demonstrated a "stable" junction. In previous reports, longer-term culturing of EC monolayers well beyond the point of confluence has been shown to result in junctional complex rearrangement (Andriopoulou P et al. Arterioscler Thromb Vasc Biol. 1999; reviewed in Bazzoni G & Dejana E. Physiol Rev. 2004). To this point, the fact that the different EC subtypes investigated had different percentages of "quiescent cells" suggests that the monolayers were not completely quiescent. The statement that the IEJ classification is "an immediate index of EC activation in contrast to quiescence" should be further supported by references or data. The definition of quiescent EC as simply non-proliferating, non-migrating is somewhat reductionist, and oversimplifies EC states. The authors state that HAoEC and HUVEC "...appeared more active...", but it is unclear what "active" means, and whether this may simply reflect that these cells had not yet reached confluence or quiescence in the 48h total culture time. As well, it is unclear how "migratory phenotypes" could occur in confluent monolayers. It would be helpful to see the data for these observations. If leaving ECs longer in culture, are the authors able to achieve a higher percentage of quiescent cells?
3. Could the authors comment on the baseline NICD immunoreactivity in the nuclei in HAoEC and HPMEC compared to HUVEC? Is this a reflection of active NOTCH signaling? Or rather, is it possible contact-inhibition (and downregulation of NOTCH) may not have occurred? Demonstration of EC quiescence would help to ensure similar cell cycle states. The definition of "Notch-positive" and "Notch-negative" cells is a bit misleading, as NICD levels and localization are a better indication of canonical Notch activation, and not the presence or absence of Notch protein(s). Further, NICD activation is also dependent on the levels of Notch ligands, which was not addressed. Are the authors able to confirm "OFF", "Low", "High", and "ON" classifications based on NICD intensity and localization with downstream Notch gene activation at a single-cell level? Or correlation between NICD status and the phase of cell cycle or proliferation status?
4. The existing workflow/platform is adapted for images obtained from the Operetta CLS system (Perkin Elmer) and Harmony software (proprietary), which may not be available for broader users in the EC field. It would be helpful to include ImageJ macros for the bulk automatic import of TIFF, renaming and upscaling of resolution/bit quality to match the formats that are compatible with the software.
5. Could the authors comment on the manpower (hours from start to finish for experiments, staining, imaging, analysis, etc.) and cost of the ECPT pipeline relative to emerging single cell technologies such as single cell-RNA sequencing. Further, one major advantage of imaging technologies is the ability to assess live cell dynamics, which are particularly relevant in response to stimuli and agonists. Have the authors utilized the ECPT platform for these approaches, in particular, to assess the differential EC subtype dynamics in proliferative conditions?
6. The authors should discuss the ability to amend or revise of the ECPT platform to incorporate analysis of additional markers that may be obtained through imaging, and discuss greater implications and utility to specifically tailor the workflow for other researchers in vascular biology, or to other monolayer culture systems. Further, they should better highlight the novel observations obtained with the ECPT compared to traditional methodology.

MINOR COMMENTS:

1. There are minor typographical, capitalization and grammatical errors throughout.
2. Why was fibronectin used to coat plates, and what was rationale for using this ECM substrate versus gelatin (most commonly used in EC cultures) or type I collagen?
3. Based on the various box plots present throughout the figures, it appears that some parameters have a large range of values. Is it possible or helpful to set minimum and maximum exclusionary criteria? Further, in the way that these data are presented, it is difficult to appreciate the effects of a treatment such as 48h of VEGF, as the magnitude of STB Index difference, for example, appears small, and it is difficult to understand whether these significant differences are biologically relevant, as assessed.
4. Use of arrows and further description in Figure 1 would help the reader understand what specific features are different in the various EC subtypes. As well, the representative micrographs for HPMEC appear blurry compared to other panels (Fig. 1).
5. In Figure 2, the panels in A, B and C do not correspond horizontally, and it may be cleared to demonstrate "Segmentation & features extraction" overlays from the same representative micrographs shown in panel A. Labeling of the individual panels and software used for panel B would help the readership understand what is being quantified and how. The second panel in "C" appears blurry.
6. In Figure 3, labelling the color code for quiescent, activated and stressed categories on graphs and in legend would be helpful to easily identify populations.
7. For Figure 4, line separators or more obvious grouping to distinguish discontinuous, linear and stabilized junction types in panel A. What proportion of the different EC subtypes contains discontinuous, linear and stabilized junctions at confluence/quiescence? Is there a correlation between discontinuous junctions and proliferating cells? It would be useful to distinguish the effects of published mediators on junctional integrity in intact EC monolayers (i.e. histamine; VEGF) from those shown in this automated quantitation. It appears that 50 ng/mL of VEGF treatment for 48h only slightly increases STB index based on panel C.
8. Figure 5 panel B should provide legend in graphs/figures or figure legends to highlight the color-coding matching the OFF, Low, High and ON groups. Further, it is unclear the difference between "High" and "ON" groups. The authors state that "thresholds were selected empirically", however, it is unclear whether this was derived through utilization of known Notch activators or inhibitors, and how this relates to the threshold of Notch activity necessary to enhance proliferation or maintain quiescence. In Supplementary Figure 4 (which I believe is mislabelled as Supplementary Figure 5), shows only a weak positive correlation between nuclear NICD intensity and mean STB index. It would be of interest to see the plot from Supplementary Figure 5 for each of the EC subtypes, in the presence and absence of VEGF. As well, for Figure 5, on C and D panels, it would improve clarity to revise "Low" and "High" descriptors with "Low NICD activity" and "High NICD activity".
9. In Supplementary Table 1, "Widt/length" should be "Width/length"
10. For Supplementary Figure 3, it would be of use to show DNA distribution intensities from proliferating, non-confluent EC subtypes to demonstrate the validity of this methodology to identify cells in G0/G1, S and G2/M phases, as highlighted in panel A. Could the authors comment on the discrepancy between the percentage of cells identified as quiescent by ECPT and the percentage of cells in G0/G1? The comment that "VEGF induced a small detectable increase in proliferation rate in all EC" is curious, as a dose of 50 ng/mL of VEGF should be a relatively strong stimulator of proliferation/migration in ECs.
11. For Supplementary Figure 5, "Nuclear NOTCH intensity" on the Y-axis should read "Nuclear NICD intensity", as it does not appear that Notch was stained. It would also be of benefit to overlay the ranges for "OFF, Low, High and "ON" to appreciate ranges of activation. Is there any correlation between NICD nuclear intensity and proliferative index?
12. Definitions should be provided for many terms. i.e. vascular endothelial-cadherin (VE-CAD; CDH5); HUVEC (human umbilical vein endothelial cell); HAoEC (human aortic endothelial cell); HDMEC (human dermal microvascular endothelial cell); NICD (NOTCH intracellular domain); VEGF (vascular endothelial growth factor); etc. at first appearance.
13. For EC subtypes purchased from commercial vendor, it would be of interest to understand how many unique donors these cells/data were derived from, and whether there are any differences in basic donor information such as age, sex, etc. Further, Promocell catalogs proliferative rate for each of their lot numbers, and it would be of interest how this compares to the values determined using the ECPT software analysis package.
14. In the "Cell culture" section of the methods, HDMEC from Promocell are listed, however, the manuscript and figures show data from HPMEC. Both EC subtypes are available from Promocell, however, HDMEC are from dermal origin.
15. Vascular endothelial-cadherin should be abbreviated "VE-CAD" or "CDH5" and not "VEC", as this is not a standard or gene notation, and will likely be confused with the more common abbreviations for venous or vascular EC. It seems as though "CDH5" is used most commonly throughout manuscript, so this should be used throughout.
16. The authors refer to "activated NOTCH" when describing antibodies in the methods, however, it would be clearer to the reader to simply refer to the antibody target (NICD), and mention that this reflects canonical NOTCH downstream activation.
17. The sentence in the "Immunostaining" methods "CDH5 is a lineage marker..." should be moved to results/discussion as these details are out of place in methods.
18. How were the 3 areas captured per wells designated? Were these locations the automated, and the same for all wells?
19. "Appendix - Figure" notation should be revised to "Appendix Figure" for consistency and to avoid confusion.
20. How were artifacts and mis-segmented cell objects excluded?
21. In "Statistical analysis" "Tuckey's" should be "Tukey's". "HSD" should be defined "honestly significant difference" or simply removed, as Tukey's is most common name.
22. In "Statistical analysis", "significative" should be "significant" or "statistically significant".
23. Scale bars should be added to micrographs.
24. Could the authors comment on the necessity of µclear plates, which substantially increases the cost per plate/experiment.
25. Were other seeding densities and times investigated?
26. More description on potentially novel observations between these three primary EC subtypes would be informative for the readership to appreciate
27. The references do not appear in chronological order. Further, consistency of reference formatting should be reviewed, and appropriate journal name abbreviations should be used.

Significance

• This manuscript presents a conceptual and technical advance, introducing a high throughput imaging platform to assess endothelial phenotypes
  • Within the field of angiogenesis, several tools exist, either proprietary, or leveraging ImageJ software to assist in assessment of cells. The ECPT provides a more complex analysis platform to integrate analysis of multiple endpoints
  • This work would be of interest to vascular biology laboratories to adopt a more comprehensive view of heterogeneous endothelial phenotypes in vitro
  • As a vascular biology researcher, I have had extensive experience with in vitro culture of various endothelial cell subtypes from human and mouse. My field of expertise gives me the perspective of the nuances of the direct handling and phenotyping of ECs, and have worked specifically worked with HUVEC, HAoEC and HPMEC, and assessed the impact of key factors relevant in angiogenesis such as VEGF, Notch and other mediators.

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 manuscript by Chesnais et al reports development of workflow for analysis of cultured endothelial cells , which they call Endothelial Cell Profiling tool (ECPT). Using ECPT they analyse several parameters in three different endothelial cell types (HuAEC, HUVEC and HPMEC), such as cell morphology, activation of cytoskeleton, VE-cadherin junctions, cell proliferation and Notch activation, under steady conditions and upon treatment with VEGF. The analysis allows to observe some predicted changes, such as increase in cell cycle and junctional activation in cells treated with VEGF-A, and such changes are highly heterogeneous. Overall, this is a potentially useful albeit not revolutionary tool for batch analysis of cultured endothelial cell phenotypes.

I have the following comments:

1. To make their case the authors should provide a comparison with other currently used approaches for EC phenotypic analysis in vitro - what is the advantage of using ECPT? The authors repeatedly use the term "single-cell level of analysis ", but this is in fact the case of any IF based analysis of cultured cells.
2. I strongly recommend to stain HPMECs for PROX1, these cells are frequently 100% lymphatic endothelial cells. In this case the authors compare different lineages and not blood endothelial cells from different locations.
3. Please provide evidence for specificity of NICD antibody.
4. Figure 1: HPMEC picture appears out of focus
5. Figure 3 A - it is not entirely clear what is the difference between activated and stressed phenotype, they look quite similar.
6. Figure 5 - what is the difference in NICD localization between "high" and "On" conditions?
7. Since the authors make a correlation between Notch activity and junctional stabilization, it would be important to confirm this by other means, such as analysis of Notch target genes.

Technical and minor

1. Methods mentions HDMECs (human dermal microvascular endothelial cells) but the authors discuss HPMEC throughout the text
2. Please add scale bars on all microscopy pictures.
3. Please provide the information on what isoform of VEGF-A was used for stimulation and the rationale for selecting the concentration.

Significance

The authors provide a workflow for the phenotypic analysis of cultured cells. Such tool is potentially useful, although the examples the authors show do not reveal striking examples of why such analysis is better in comparison to existing approaches. My guess is that the analysis may be faster and less tedious, once the training sets are generated, but this is not specified. My speciality is endothelial cells biology.

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

The authors highlight the importance of endothelial heterogeneity using endothelial cells from different tissues. They examined aortic and pulmonary endothelium as well as HUVECs. They cultured the cells in identical conditions and also stimulated them with a physiological concentration of vascular endothelial growth factor as well high concentrations as would be found in cancers. They developed a profiling tool that allowed analysis of individual endothelial cells within a monolayer and quantification of inter-endothelial junctions, Notch activation, proliferation and other features.

Major comments

1. It would be useful to apply this technology one step beyond two-dimensional culture, to use vessels opened up longitudinally so that one can see the monolayer of endothelial cells and assess whether it is relevant in primary material in situ. I think this would be a major utility of the whole approach.
2. There are some very nice images here but disappointed not see a field that could show staining and markers for several of the target proteins and thus show the heterogeneity and randomness or organisation of the endothelial cells. For example are any clusters of a subtype of endothelial cells around proliferating cells.
3. The Notch signalling is an important aspect of this work, particularly evidence of lateral inhibition would have been of value. For example, one might expect cells adjacent to each other to have alternating high and low NICD. NICD staining alone does score the extent of the signalling because of many factors that can influence the transport of the cleaved NICD. Really a marker of Notch signalling downstream e.g. HES or HEY family ,DLL4 fis needed to give more information about this critical aspect.
4. I really do not think that in Figure 5 it is justified to have a red line drawn through the cloud of points. The correlation coefficient is so low that this is meaningless. The failure to distinguish a P value from biological relevant is worrying. Much better comparison would have been between NICD staining and a downstream gene regulated by notch.
5. It is important to know that the antibodies used for staining have be validated by the investigators. They would need to show a single band on Western blots or be able to block staining on immunohistochemistry. We all know the manufacturers can be unreliable and use high concentrations of proteins for Western blots. These should be added as a supplementary figure.

Significance

This represents a valuable and thorough methodology likely to be highly useful to many groups and show new insights into endothelial biology.

Wide audience, cancer, cardiology, vascular disease-covid.

My expertise >100 papers on angiogenis in cancer, basic mechanism, therapy models, bioinformatics IHC, patients, clinical trial. H score 190 Google Scholar

Transparent Peer Review

---

Download the complete Review Process [PDF] including:

• reviews
• authors' reply
• editorial decisions

</br>

Transparent Peer Review

---

Download the complete Review Process [PDF] including:

• reviews
• authors' reply
• editorial decisions

</br>

Transparent Peer Review

---

Download the complete Review Process [PDF] including:

• reviews
• authors' reply
• editorial decisions

</br>

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

Learn more at Review Commons

---

Reply to the reviewers

Point-by-point response, comments in (blue), our response in (black)

Note: we included 6 Figures in our response, yet the ReviewCommons system does not appear to support including images as part of the response. These Figures are in the original "Initial Response" file available to ReviewCommons. We requested that Review Commons post our "Initial Response" file that contains these figures so that this information is available.

Reviewer #1

*In the paper by Gowthaman et al., the authors aim at better understanding the molecular mechanisms controlling divergent non-coding transcription (DNC). They describe a high-throughput yeast genetic screen using two strains in which two loci consisting of a coding and a divergent non-coding transcription unit (CGC1-SUT098 or ORC2-SUT014) were replaced by a bidirectional fluorescent reporter construct encoding mCherry in the coding direction and YFP in the non-coding direction. The two reporter strains were crossed with the yeast deletion library and mutants leading to increased or decreased YFP signal were selected as potential DNC repressors or activators. The two screens identified a number of common potential repressors and activators. Components of the Hda1C histone deacetylase complex were identified as DNC repressors in both screens. This phenomenon was confirmed genome-wide by performing NET-Seq in WT as well as hda1D and hda3D strains. This experiment allowed to identify 1517 DNC transcripts repressed by Hda1. Further analyses indicate that Hda1C represses DNC genome-wide independently of expression levels and that loss of Hda1 does not substantially affect coding transcription.

Live-cell imaging of transcription was then used to show that loss of Hda1 increases DNC transcription frequency rather than duration providing novel information on the link between DNC transcription initiation kinetics and chromatin regulation. Finally, using Chip-seq, the authors show that the level of acetylation over the divergent non-coding units is increased in the absence of Hda1 and some experiments suggest that H3K56 acetylation also contributes to DNC regulation, further strengthening the importance of elevated histone acetylation in efficient DNC.

Importantly, several components of the SWI/SNF chromatin remodeling complex were identified as activators confirming earlier observations (Marquardt et al., 2014). SAGA subunits were also among potential DNC activators, however these effects could not be confirmed through validation experiments. The authors conclude that DNC may be independent of specific activators and mainly due to transcriptional noise resulting from the adjacent NDR.

Overall this paper is very well structured, clearly written and the experiments are well controlled. The genetic screen identifies novel factors involved in the regulation of DNC. The study clearly demonstrates that the level of acetylation is a key regulator of divergent non-coding transcription and that histone deacetylation by Hda1 reduces the frequency of DNC initiation events. While this conclusion is strongly supported by the Net-Seq and Chip-seq metagene analyses, the fluorescence mCherry and YFP values or qRP-PCR analyses of specific genes do not always behave as expected when looking at absolute values rather than mCherry/YFP or GCG1/SUT098 ratios, which is sometimes disturbing when reading the paper. Therefore, the following points should be clarified.*

We are grateful for the kind appreciation of our manuscript and clarify the remaining questions in the revised manuscript.

**Major points**

#1.1: Figures 2 and S2A: Figures 2C and D show the mCherry/YFP fluorescence and GCG1/SUT098 RT-qPCR gene expression ratios respectively, which are consistent with a repressive effect of Hda1C on DNC transcription and a potential DNC activating effect of SAGA components. However, the absolute mCherry and YFP or GCG1 and SUT098 expression values presented in Figures S2A and S2B show the opposite: loss of Hda1C subunits rather leads to a decrease in mCherry with not much effect on YFP; moreover loss of Hda3 results in decreased SUT098, which is inconsistent with the whole model. The same comment is valid for the SAGA mutants. It would be good to provide some explanation for these a priori contradictory observations, especially for the Hda1c mutants, which are the major focus of the study. The Net-Seq analyses are certainly more reliable since less subject to protein or RNA stability effects, which may underlie some of the inconsistencies between protein and RNA absolute levels.

Thank you for this comment. We offer enhanced clarity in the revised manuscript.

In general, transcription in each direction shows a weak yet highly statistically relevant positive correlation (Spearman rho = 0.26, p-value = 4.94e-24). We are enclosing a plot based on NET-seq data that supports the correlation in each direction of a NDR as part of our response below (RFig.1). To unpick relative effects the ratio captures these effects well, in our experience better than the individual fluorescence measurements or RT-qPCR. Of course, we are ultimately interested in transcription and fluorescence measurements or RT-qPCR of steady-state RNA are only an approximation. Resources and time constraints limit how many mutations we can examine by techniques such as NET-seq, which are arguably most informative. The positive correlation between transcription in each direction has the effect that relative differences can manifest themselves through detectable effects of the other fluorophore. As this reviewer mentions, we can be most confident of results that we could further validate by NET-seq or live-cell imaging.

(INSERT Rfig1)

RFig1: Scatterplot of NET-seq data for DNC/host gene pairs. Each point corresponds to a bidirectional gene promoter overlapping with a nucleosome-depleted region (NDR). The values represent NET-seq FPKM values in protein-coding (x-axis) vs non-coding (y-axis) directions. These data support a statistically significant correlation (Spearman test: rho = 0.2554876, p-value = 4.939658e-24).

#1.2: Figure 3: this figure examines the effect of Hda1 and Hda3 on the 1517 DNC transcripts. Does loss of this HDAC also increase the expression of all the other 2219 non-coding transcripts identified by Net-Seq, which would make Hda1C a more general repressor of non-coding transcription?

We have performed the analysis for all other non-coding transcripts in Hda1C mutant NET-seq data and added it as part of this response RFig2. Quantification of CUTs, SUTs and other lncRNAs that are not resulting from DNC in Hda1C mutants results in a slight increase in the nascent transcription that is not statistically significant. These data do not offer strong support for the idea that Hda1C represents a more general repressor. We added this plot as novel supplementary figure S3D and adjusted the text of the revised manuscript (line 214).

(INSERT Rfig2)

RFig2: Metagene plot of NET-seq data for non-coding RNA that are not classified as DNC. Metagene plot shows genomic windows [TSS - 100 bp, TSS + 500 bp] relative to the annotated starts of ncRNA transcripts.

#1.3: Moreover, does loss of Hda1 or Hda3 reveal DNC transcripts that were not detected in wild-type? This may increase even more the number of genes with divergent transcription.

We are grateful for the opportunity to clarify this point. We noticed that the yeast genome shows evidence for much more non-coding transcription than annotated. In this paper, we used TranscriptomeReconstructoR for a data-driven annotation of yeast non-coding transcripts, with an emphasis on the boundaries. See also:__ ( DOI: 10.1186/s12859-021-04208-2 ). The set of non-coding transcripts was for example informed by the previously published NET-seq data on wild-type samples (Churchman et al., 2009; Marquardt et al., 2014; Harlen et al., 2016; Fischl et al., 2017). We have clarified relevant Methods sections to make this point more accessible (line 733). The combination of these NET-seq datasets gives a very good sequencing coverage. The Hda1C mutant NET-seq data does not have a better coverage than this combined reference set, so it would be very hard to find new transcripts without prior evidence in our exhaustive set of combined NET-seq data. However, our Supplementary table S3 contains the fold-change values for all DNC transcripts in mutant compared to wild type. Loci with a high fold-change could arguably be regarded as hda-specific. __

#1.4: Figures S3A, B, C: are the 3 groups of DNCs derepressed to the same extent by loss of Hda1 or Hda3? This is difficult to judge given the differences in y-axis scales. Figures S3D, E: the authors show the Net-Seq snapshots for the GCG1 and ORC2 loci. It would be good to add the quantifications as presented in Figure 3 for YPL172C and YDRr216C.

Thank you for the suggestion. We replaced S3A-C with plots that show the same range of the y-axis in the supplementary figure. Hda1C represses DNC in all three cohorts stratified by DNC expression strength. We also added a quantification boxplot for NET-seq signal in the GCG1 and ORC2 loci in revised S3F-I.

#1.5: Figures S4A, B, C and D are not well explained. What does the y axis frequency correspond to? Is it the % of cells showing a signal? Is the intensity of SUT098 higher because the transcription initiation frequency is higher and therefore the transcription site signal is more intense?

We improved the annotation for the supplementary figure S4. We clarified in the legend that the y-axis frequency represents the percentage of frames recorded for transcription initiation spots (TS). The bars represent transcription intensity in all the frames recorded, with active transcription ‘ON’ and without TS ‘OFF’. The intensity increases with higher initiation rates and thus the intensity of SUT098 transcription initiation is high.

#1.6: Figures S4 A-I should be more specifically cited in the text.

We have cited the figures in the text in the revised version.

#1.7: Figure 5A: it is really unexpected and unclear why the mCherry/YFP in the WTH3/hda1D and WTH3/hda1D/H3K56mut is increasing compared to WTH3, since DNC is supposed to increase. Similar comment for Figure S5C. This should be clarified in the text.

Thank you for pointing this out. We missed to address this in the text. The isogenic control “H3 wild type” carries only one copy of the two genes coding for H3, which has a general effect on transcription. We added data showing this as part of our response (RFig3.), and explained this part more clearly in the revised text (line 263). Essentially, the genetic background of the yeast synthetic histone mutant collection sensitizes for a decreased ratio of mCherry/YFP (RFig3.). This result is also included in table S2, where deletions of the histone genes HHT2 (H3) and HHF2 (H4) are listed as shared repressors in both screens. Hda1C mutations show the increased ratio in the sensitized “H3 wild type” background, but not in backgrounds we tested that contain a wild-type dosage of histone genes.

These data remain valid to support the genetic interaction of hda1D along with the substitution mutants of H3K56.

(INSERT Rfig3)

RFig3. Fluorescence signal values of H3WT and BY4741 strains with GCG1pr FPR. The H3WT affects general transcription of coding transcript and decreases the ratio of mCherry/YFP fluorescence.

#1.8: More generally, as already mentioned above, the fluorescence data are expressed either as mCherry/YFP ratio or as absolute values. It would be good to systematically show the ratios and the absolute values of mCherry and YFP signal; the same for coding and DNC RT-qPCR as well as Net-Seq values when available.

We ensured that the absolute data values for flow cytometry and qPCR have been represented in the supplementary figures S2 and S5. The FPKM values for NET-seq data for individual transcript units are provided in the supplementary table S3.

#1.9: Figures S5A and B are not referred to in the text. It should be mentioned and explained how normalization to H3 affects the levels of acetylated H3 over the NDR.

We now refer to the figures in the main text and explained the rationale for normalization.

#1.10:* p. 12 "Our data thus suggest to extend the transcriptional noise hypothesis with activities limiting DNC transcription to account for genome-wide variation in non-coding transcription".

If DNC is the result of "transcriptional noise", it is surprising that in the case of CGC1-SUT098, the transcription frequency is higher in the non-coding versus the coding direction. Is the SUT098 behaving like the coding unit in this case? The authors should comment on that. *

This is very interesting point. One interpretation of the “transcriptional noise” hypothesis is indeed that non-coding transcription is at low level. We selected loci with high DNC expression, so these loci are somewhat contradictory to this idea a priori. Nevertheless, identifying a biological function of non-coding RNAs is challenging, and it remains to be tested if SUT098 represents particularly “loud noise” or if the high transcription indicates that it carries a yet unknown cellular function. In theory, this screen is suitable to identify factors that may be required to induce DNC, perhaps even specifically. To identify such factors a locus with high DNC is needed to facilitate detection, since our previous screen using the PPT1/SUT129 system had lower SUT expression and failed to identify such mutants systematically. This is important, since a mutation lowering DNC needs to start from a sufficiently high fluorescence signal to distinguish it from background fluorescence. Since the results presented did not clearly uncover such factors, we favor the hypothesis of DNC arising due to the promoter architecture at NDRs, see also positive correlation plot in RFig1. The many repressive pathways are also acting on highly expressed DNCs, which is certainly an interesting information provided by this manuscript.

**Minor comments**

#1.11: p. 4 should one talk about Hda1C-linked histone acetylation facilitates... (should be deacetylation...??)

Done.

#1.12: The authors should explain why they chose two coding/non-coding pairs that are cac2D insensitive and whether other criteria, such as level of DNC transcription, were also considered, since GCG1-SUT098 represents one of the most highly expressed divergent non-coding transcripts.

The GCG1 and ORC2 loci were chosen based on i) high DNC levels, ii) a low fold-change of NET-seq data in the cac2∆ and iii) a DNC region free from other transcriptional units. However, this was based on the state-of-the-art annotation in 2015 when we started this project. Also, when we categorized genes as affected by cac2∆, we used a fold-change expression cut-off that suggested that about a third of DNCs are repressed by CAF-I. It appears that we still underestimated the effect of CAF-I, since our data show that the target regions of our new screens are also affected by CAF-I. DNC expression at these loci is high, which would result in a low fold-change in mutants that further increased DNC here.

#1.13: It is hard to understand why both the H3K56A and H3K56Q mutations lead to increased DNC, a result already presented in the Marquardt et al. 2014 paper. It would be helpful to provide a more extensive explanation or hypothesis.

The H3K56 substitution mutant Q is expected to mimic the acetylation state and A is devoid of post-translational modifications. We observe an increase of signal ratio in the mutants because the H3K56ac is both responsible for incorporation and eviction of -1 nucleosomes (Marquardt et al., 2014). Mutations affecting H3K56 can thus result in less -1 nucleosome density and more DNC through reducing incorporation or enhancing eviction. We have improved the revised text to highlight this. We have clarified this in the text (line 271).

#1.14: What defines the level of DNC repression? How does the level of repression correlate with the level of coding transcription?

We have added RFig.1 to address the question about correlation. There is a statistically significant positive correlation between transcription in each direction by NET-seq data in wild type samples genome-wide. However, the correlation is weak (rho = 0.26), which is consistent with locus-specific adjustments of transcriptional strength in each direction. For DNC, several chromatin-based pathways contribute to repression. The resulting level of DNC transcription thus reflects the combined action of several pathways. Here, we characterize Hda1C as a novel player with a genome-wide effect on this phenomenon. Elucidating the mechanistic interplay at specific target DNC loci will be an exciting future research question.

Reviewer #1 (Significance (Required)):

This is a very interesting and innovative study using cutting edge genetic approaches, genome-wide sequencing as well as single cell imaging to extend our understanding of non-coding transcription regulation and its potential impact on gene expression. It is a nice continuation and complement of an earlier study from the same author (Marquardt et al., 2014) and will certainly be of interest to a large chromatin biology audience.

We are grateful for the appreciation of our research on this topic.

Reviewer #2

Promotors are frequently transcribed in both directions. The divergent, \upstream' transcript is frequently unstable. Transcription initiation is regulated through the acetylation of promoter-proximal nucleosomes, where HDAC-dependent deacetylation of histones typically represses transcription initiation.*

*The current manuscript addresses the question whether initiation of coding and divergent, non-coding (DNC) transcription is regulated by the same factors. Previously Marquardt and others showed that H3K56ac-mediated histone exchange has a differential effect on coding and DNC transcription.

Using a clever reporter system, the authors screened for positive and negative regulators that preferentially affect DNC transcription. They discover the Hda1 deacetylase complex as a DNC-biased repressor and diverse HATs as DNC-biased activators. The role of activators could not be validated, presumably due to high variability of the system.*

Focusing on Hda1c the authors present data suggesting a larger effect of Hda1c on 'upstream' nucleosomes associated with DNC transcription than in coding transcription. Genome-wide NET-seq mapping was consistent with this differential regulation. Life cell imaging of one specific case argues that Hda1-mediated repression reduced the time between initiation events. The authors employ state of the art methods and in general the data are of very good quality. The effect size is very small, which raises the broader question whether the results, while statistically significant is biological relevant. I have a few comments that the authors may use to revise their manuscript.

Thank you for the appreciation of our very good data quality. We hope our revision plan will help to clarify some confusion about the scope and effect size.

#2.1) The differentially regulated coding and DNC transcription are defined by a directionality score. The screen was performed with two reporter loci that are strongly biased for DNC transcription (the idea to detect activators did not work out). Considering that coding and DNC transcription may not be totally independent because of the proximity of target nucleosomes, and sense and antisense transcription may compete for regulators, the question arises how levels of coding transcription affect DNC transcription in wildtype and mutants. The authors stratified their results according to levels of DNC transcription, but discussion and data analysis of the effect of coding transcription on the directionality score may be relevant.

We added the plot in RFig.1 above to address the question of correlation between transcription in each direction. NET-seq data supports a weak but highly statistically significant positive correlation between transcription in each direction genome-wide (rho = 0.26, p-value = 4.94e-24). We agree that it is relevant to discuss the effect of coding transcription on the directionality scores and revised the discussion accordingly (line 315). We have used both the coding and DNC signal values to create the comprehensive quadrant scatter plot in Fig. 1D-E. Analysis of mutants along the diagonal illustrates that many mutations affect coding transcription as well as DNC. The directionality score measures deviations from the axis of positive correlation, which requires us to use the information of both fluorophores.

#2.2) The study is strong where the findings can be generalized. The single-molecule live-cell imaging analysis, while done properly, has only limiting impact, because the corresponding coding transcript could not be detected. This si more an anecdotal finding.

There seems to be a misunderstanding, the live-cell imaging measurements of transcription for SUT098 are stand-alone data. SUT098 by itself is a transcription unit, so we measure DNC of this unit independently from GCG1 that has much lower expression. The measurements are specific to SUT098 transcription and the quantification provides new information about the mechanisms involved in the regulation of DNC. We clarified the text in this regard (line 233).

#2.3) The effect size is small (20%, on average) and the variability is high. The fact that the HATs that emerged as very robust activators of DNC transcription could not be validated and that the Hda2 subunit of the HDAC complex was not found statistically significant show the limitations of the study. To their credit, the authors discuss these limitations appropriately.

We have worked on the Methods in the revised manuscript to clarify this confusion (line 712). For the screen, the median signal values represent data from up to 50,000 individual cells. These experiments are remarkably accurate and highly reproducible, especially for molecular biology where n=3 is common. We have uploaded these data to the FlowCore public repository. We encourage any colleague to exploit the opportunity to analyze these data independently to experience the high data quality. With high number of observations, 20% average is a large effect and reflects a rather big shift of the population. As is standard for genetic screens, resource constraints are prohibitive to pursue all hits. In addition, it is expected that only some hits will be affecting transcription of DNC since the fluorescence reporter can be affected by many other cellular events. We focused on the effects on DNC in this manuscript.

There seems to be some misunderstanding, Hda2 is a statistically significant hit in the ORC2/SUT14pr screen; this information is in Fig. 1E. The Hda1C subunits are labeled in purple.

#2.4) Figure S3C suggests that the Hda effect is largest at genes that are poorly expressed, and smaller at more average expression levels. Are we looking at a phenomenon that mainly applies to repressed genes?

Thank you very much for this suggestion. We replaced S3A-C with revised panels where the data is shown with the same y-axis scale, please see also #1.4. We believe the revised presentation also helps to clarify that the mutations increase DNC for all cohorts stratified by DNC expression.

**Minor issues**

#2.5) The NET-seq study involves two replicates. How well did they correlate?

The WT and mutant NET-seq replicates have good correlation (Spearman’s correlation coefficient was above 0.6 for WT and above 0.8 for the mutants).

(INSERT Rfig4)

RFig4. Correlation scatter plot of individual NET-seq replicates of WT, hda1D and hda3D. Spearman correlation coefficients of WT, hda1D and hda3D are 0.677, 0.8 and 0.825, respectively.

#2.6) For the live-cell imaging replicates were not mentioned. Were replicate studies performed?

We have updated the text to make this important point more accessible (line 230). For live-cell imaging studies, transcription is recorded as movies of cells over time. We took multiple movies, and pooled the data from all the cells to improve statistical power. Data from each movie represent individual repeats. We monitored 130 cells on average for the WT and mutant strains over time.

#2.7) Fig 4E is not mentioned in the text (mislabeled as 4D)

Done.

#2.8) Fig S5 is not mentioned in the main text.

__Done.

__Reviewer #2 (Significance (Required)):

In summary, this is a high-quality study that presents the results of a genome-wide screen that will be of interest to colleagues in the narrower field. Due to the small effects the results may appeal less to a general readership.

We are grateful for appreciating our manuscript as a high-quality study. We hope our revisions help to clarify confusion concerning effect size.

Reviewer #3

In this manuscript, Gowthaman et al describe the results and follow up of their screen aimed at identifying regulators of divergent noncoding (DNC) transcription in S. cerevisiae. From this screen, they identify Hda1C as a repressor of DNC transcription, and perform follow experiments to support and detail this finding. In addition to RTqPCR to confirm the reporter and endogenous changes, the authors perform NET-seq to look at global DNC alteration upon Hda1C subunit deletion and identify a number of non-coding transcripts with altered expression levels. In addition, the authors perform live cell imaging to demonstrate that there is a modest restriction of initiation frequency when one of the subunits of Hda1C is deleted. Finally, the authors explore changes to pan-H3 acetylation and the genetic overlap between Hda1C and H3K56ac demonstrating independent genetic pathways, but overall increases in H3 acetylation over DNCs when Hda1C is deleted. Overall, the screen and results are of interest, but the authors overstate some of the conclusions (perhaps most importantly within the title!). I have the following suggestions to improve the manuscript:

Thank you for recognizing the interest in our results. We have revised the manuscript to state the conclusions more cautiously.

**Major comments**

#3.1. The title of the manuscript is based on the single molecule live cell imagining experiments presented in Figure 4. While there is a statistically significant decrease in initiation frequency from deletion of one Hda1C subunit, there is no statistical decrease in deletion of the other two. Furthermore, these experiments were performed at one locus. As a result, I find the title to be an overstatement of the findings of the paper and suggest the authors refocus on the more robust findings of the manuscript.

Live-cell imaging requires extensive engineering of the target loci. Perhaps this was lost in the Methods, but it is a 5-step process to integrate the stem-loops. We tried to engineer other loci, but this is far from trivial and this technique does not work for all loci tested. The hairpins are also unstable, and need to be carefully checked prior to experimentation, which challenges scaling this approach up to a higher-throughput. It appears that we undersold this point, but the fact that we now provide a locus and strains for the community that makes such studies possible for DNC represent a tremendous achievement. Since hda1D also decreases time between initiations, we generalized the finding to Hda1C.

However, we recognized that the reviewer makes a helpful suggestion to choose a more careful title since there is no statistically significant reduction of initiation frequency in some mutants. We have revised the title to “__Hda1C limits divergent non-coding transcription and restricts transcription initiation frequency__” in the revised manuscript to address this point.

#3.2. Relatedly, in Figure 4, the authors present the findings from the single molecule live cell imaging experiments. Within this experiment, the authors include a cac2 deletion (CAF-1 subunit) strain, and observe a modest effect, similar to hda1 deletion. This is surprising as the authors mentioned this location (GCG1/SUT098) was selected as CAF-1 was NOT shown to regulate the DNC previously (Marquardt et al 2014; as mentioned at the beginning of the Results section). The similar decrease in initiation frequency between cac2 deletion and hda1 deletion further concerns me regarding the use of these data as the headlining finding.

We believe there is a misunderstanding. We clarify that selection of the GCG1 locus was based on a cut-off value for cac2D effect, as is also shown in Fig S1C. The fold-change is small, but since DNC transcription of the chosen loci is high in wild type, an increase in a mutant would not necessarily give a high fold-change. Hence, we need to be cautious to conclude that CAF-I does not regulate DNC at this locus. The fold-change analysis suggested it, but it remained possible. CAF-I appears to affect even more loci than initially identified with the chosen cut-off. We see the same trend as in Hda1C mutants as in cac2, which offers support to the exciting idea that modulation of the initiation frequency may be a shared mechanism by chromatin-based regulators acting on DNC.

#3.3. It is unclear to me why the change in mRNA expression is included within the screen. Why not solely look at the expression change of the DNC? Importantly, the authors note in the discussion that perhaps the reason the SAGA complex was identified was due to regulating mRNA expression and not DNC expression and therefore was identified in the screen. Could the authors not just present the fold change in DNC expression using their YFP reporter, and not the YFP vs mCherry?

The regulation of initiation frequency in each direction is super-imposed on a general positive correlation __(rho = 0.26, p-value = 4.94e-24) between the coding and non-coding directions__, please see also RFig.1. For the purpose of this study about selective effects on the direction of transcription, it is vital to incorporate both sides of the reporter. Otherwise, we would select for factors that activate or repress the transcription from the target promoter NDR. This point is accessible in Fig.1D-E, where mutations that affect YFP usually also have an effect on mCherry. The aim of this study was to identify mutants that affect the relative expression, and therefore a focus on one fluorophore would not improve the analysis. We clarified this important point more accessibly in the revised manuscript (line 315).

Please also note that all the raw data are available, so colleagues are in the position to perform their independent analyses. We believe that it is very valuable for the community to have access to these data since they may be useful for other purposes and could be analyzed in many different ways. In fact, we have tried several methods and approaches over the years and present what we believe is most appropriate in this manuscript. For example, Hda1C comes out as a convincing hit with a range of different approaches to analyze the data, which is also a reason we feel confident about the characterization of Hda1C.

#3.4. This is absolutely beyond the scope of the paper, but limiting the screen to only nonessential proteins likely misses important regulators. In the future, perhaps the authors could pursue a SATAY screen to look for essential proteins as well? Again, the findings of this paper are appropriate, and the screen is a great undertaking, but I want to suggest this to the authors for potential future projects.

Thank you for this excellent suggestion. We agree that capturing the role of essential factors would be very informative, and the saturated transposition approach would be promising. However, as the reviewer points out, performing these analyses is beyond the scope of the current manuscript.

#3.5. The authors perform NETseq experiments in deletion strains and identify ~1500 DNC transcripts with altered expression. Later the authors look into the mechanism and demonstrate an increased H3ac in hda1 deletion strains. The authors could enhance the representation of these datasets by correlating the change in H3ac with the change in DNC transcription - do they correlate?

Thank you for bringing up this excellent point. We present the correlation data of change in H3ac and DNC transcription in the hda1D mutant (RFig5.). The ChIP-seq and NET-seq values of hda1D were divided by respective WT values in order to quantify the relative increase of H3 acetylation or nascent transcription in hda1D). The data showed a weak (Spearman rho= 0.23) but significant (pval=3.0e-20) positive correlation between the ratio values. The hda1D-dependent increase in H3 acetylation correlates with hda1D-dependent increase of RNAPII occupancy in DNC transcripts. We enhanced our representation of these data by including this plot as S5D in the revised manuscript as suggested.

(INSERT Rfig5)

RFig5__: Scatterplot of hda1D/WT NET-seq (y-axis) and ChIP-seq (x-axis) ratios. Each point corresponds to a bidirectional gene promoter overlapping with an NDR. The x-axis shows ChIP-seq ratios, and the y-axis shows the NET-seq ratios. These data support Spearman correlation test: rho = 0.234 and a statistically significant p-value = 3.0e-20.__

#3.6. In Figure 5, the authors argue that Hda1C works non-redundantly with K56ac, using point mutants to mutate K56 to A or Q. Did the screen identify anything else in the K56ac pathway? Rtt109 or Asf1, for example? Because Hda1C deacetylates H3, including but not limited to K56, it is a bit surprising the K56 point mutations result in a larger increase in SUT098-YFP levels. The authors discuss within the text that Hda1C has multiple targets; but coming back to my previous point that CAF-I was not supposed to impact this location, I am having a hard time understanding these results.

This is an excellent point. We improved the manuscript by highlighting other factors with links to H3K56ac in our scatter plots, for example Rtt109 in Fig 2A. Nevertheless, the reviewer may wish to satisfy his/her curiosity by exploring table S2 in more detail. Table S2 lists the top candidates from both screens.

We hope our answer to point #3.2 helped to clarify the aspect of this comment related to CAF-I.

**Minor comments**

#3.7. The authors follow up the screen using RTqPCR for GCG1/SUT098 in newly made deletion strains. I was surprised the authors choose this locus rather than the ORC2/SUT014 locus, as the screen showed a strong increase for this reporter. While I appreciate generating the deletion strains within the reporter is beyond necessary, assessing the endogenous locus within the deletion strains by RTqPCR seems reasonable.

We chose GCG1 locus since the fold change in directionality by genetic screen was high for the activator mutants. We will perform this experiment and add the missing validation experiment for the ORC2 locus in the revised manuscript.

#3.8. The authors tend to show their genomic data as metaplots; it would be nice to see heatmaps where more can be gleaned from the display of all the loci. This applies to the NET-seq data (Figure 3) and the ChIP-seq data (Figure 5).

We appreciate the suggestion and generated the requested heatmaps using the NET-seq tracks of WT and hda mutants (RFig6.). The heatmap represents the same genomic intervals as on the corresponding metagene plot (Figure 3A). We find that the differences between WT and hda samples are more clearly accessible at first glance on the metagene plot rather than on the heatmap. We believe that this could be because the heatmaps do not represent what transcripts have in common and rather underlines the differences. In contrast, the metagene plots reveal the common trends by taking the average of signal. We thus prefer showing metagene plots in the manuscript, as they allow for overlay of multiple tracks on the same plot, thus enhancing visual comparison for the readers.

(INSERT Rfig6)

RFig6. Heatmap representing NET-seq data in WT, hda1D and hda3D. Genomic intervals covering [TSS - 100 bp, TSS + 500 bp] of DNC transcripts (n=1517) are shown. The color indicates the log2-transformed NET-seq values.

#3.9. In Figure 5B, the authors present H3ac ChIP-seq data, presented as a ratio of H3ac/total H3. While this is a perfectly acceptable way to present the data, I was surprised to see a decrease in total H3 levels when examining the supplemental data. Has this decrease in H3 occupancy upon hda1 deletion been shown previously? This finding should be discussed within the manuscript.

We appreciate that the reviewer noticed this. We do not think this has been explicitly stated before, as the focus thus far had been on the effects towards the mRNA. However, the effect is not statistically significant between the WT and hda1D as observed in S5B. We thus prefer to remain cautious about this conclusion.

#3.10. In Supplemental Figure S3, the authors break down the NET-seq data by DNC FPKM, which is very nice. Very minor point that the font here is quite small.

Thanks, we improved the font size. Note that we also revised the y-axis scale in response to comment #1.4.

Reviewer #3 (Significance (Required)):

\*Significance:** *

The regulation of divergent non-coding RNAs is an understudied field. In this paper, the authors perform a screen for all non-essential yeast proteins in regulating the expression of these ncRNAs. The screen results and follow up defining the role of Hda1C in broadly repressing the expression of these ncRNAs is of interest to the field.

We are grateful to the reviewer for highlighting the interest of our work to the field.

\*Context:** *

This work follows from Marquardt's previous 2014 study that identify Caf1 as regulating DNCs in S. cerevisiae.

\*Audience:** *

Broadly, the chromatin and transcription field. Anyone interested in how chromatin regulates transcription, regulation of ncRNAs, and functions of histone modifying enzymes.

\*Expertise:** *

I am a member of the chromatin and transcription field, largely performing genomic experiments. We do not perform microscopy, although sufficiently understand the experiments and results presented here.

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 this manuscript, Gowthaman et al describe the results and follow up of their screen aimed at identifying regulators of divergent noncoding (DNC) transcription in S. cerevisiae. From this screen, they identify Hda1C as a repressor of DNC transcription, and perform follow experiments to support and detail this finding. In addition to RTqPCR to confirm the reporter and endogenous changes, the authors perform NET-seq to look at global DNC alteration upon Hda1C subunit deletion and identify a number of non-coding transcripts with altered expression levels. In addition, the authors perform live cell imaging to demonstrate that there is a modest restriction of initiation frequency when one of the subunits of Hda1C is deleted. Finally, the authors explore changes to pan-H3 acetylation and the genetic overlap between Hda1C and H3K56ac demonstrating independent genetic pathways, but overall increases in H3 acetylation over DNCs when Hda1C is deleted. Overall, the screen and results are of interest, but the authors overstate some of the conclusions (perhaps most importantly within the title!). I have the following suggestions to improve the manuscript:

Major comments:

1. The title of the manuscript is based on the single molecule live cell imagining experiments presented in Figure 4. While there is a statistically significant decrease in initiation frequency from deletion of one Hda1C subunit, there is no statistical decrease in deletion of the other two. Furthermore, these experiments were performed at one locus. As a result, I find the title to be an overstatement of the findings of the paper and suggest the authors refocus on the more robust findings of the manuscript.
2. Relatedly, in Figure 4, the authors present the findings from the single molecule live cell imaging experiments. Within this experiment, the authors include a cac2 deletion (CAF-1 subunit) strain, and observe a modest effect, similar to hda1 deletion. This is surprising as the authors mentioned this location (GCG1/SUT098) was selected as CAF-1 was NOT shown to regulate the DNC previously (Marquardt et al 2014; as mentioned at the beginning of the Results section). The similar decrease in initiation frequency between cac2 deletion and hda1 deletion further concerns me regarding the use of these data as the headlining finding.
3. It is unclear to me why the change in mRNA expression is included within the screen. Why not solely look at the expression change of the DNC? Importantly, the authors note in the discussion that perhaps the reason the SAGA complex was identified was due to regulating mRNA expression and not DNC expression and therefore was identified in the screen. Could the authors not just present the fold change in DNC expression using their YFP reporter, and not the YFP vs mCherry?
4. This is absolutely beyond the scope of the paper, but limiting the screen to only nonessential proteins likely misses important regulators. In the future, perhaps the authors could pursue a SATAY screen to look for essential proteins as well? Again, the findings of this paper are appropriate, and the screen is a great undertaking, but I want to suggest this to the authors for potential future projects.
5. The authors perform NETseq experiments in deletion strains and identify ~1500 DNC transcripts with altered expression. Later the authors look into the mechanism and demonstrate an increased H3ac in hda1 deletion strains. The authors could enhance the representation of these datasets by correlating the change in H3ac with the change in DNC transcription - do they correlate?
6. In Figure 5, the authors argue that Hda1C works non-redundantly with K56ac, using point mutants to mutate K56 to A or Q. Did the screen identify anything else in the K56ac pathway? Rtt109 or Asf1, for example? Because Hda1C deacetylates H3, including but not limited to K56, it is a bit surprising the K56 point mutations result in a larger increase in SUT098-YFP levels. The authors discuss within the text that Hda1C has multiple targets; but coming back to my previous point that CAF-I was not supposed to impact this location, I am having a hard time understanding these results.

Minor comments:

1. The authors follow up the screen using RTqPCR for GCG1/SUT098 in newly made deletion strains. I was surprised the authors choose this locus rather than the ORC2/SUT014 locus, as the screen showed a strong increase for this reporter. While I appreciate generating the deletion strains within the reporter is beyond necessary, assessing the endogenous locus within the deletion strains by RTqPCR seems reasonable.
2. The authors tend to show their genomic data as metaplots; it would be nice to see heatmaps where more can be gleaned from the display of all the loci. This applies to the NET-seq data (Figure 3) and the ChIP-seq data (Figure 5).
3. In Figure 5B, the authors present H3ac ChIP-seq data, presented as a ratio of H3ac/total H3. While this is a perfectly acceptable way to present the data, I was surprised to see a decrease in total H3 levels when examining the supplemental data. Has this decrease in H3 occupancy upon hda1 deletion been shown previously? This finding should be discussed within the manuscript.
4. In Supplemental Figure S3, the authors break down the NET-seq data by DNC FPKM, which is very nice. Very minor point that the font here is quite small.

Significance

Significance:

The regulation of divergent non-coding RNAs is an understudied field. In this paper, the authors perform a screen for all non-essential yeast proteins in regulating the expression of these ncRNAs. The screen results and follow up defining the role of Hda1C in broadly repressing the expression of these ncRNAs is of interest to the field.

Context:

This work follows from Marquardt's previous 2014 study that identify Caf1 as regulating DNCs in S. cerevisiae.

Audience:

Broadly, the chromatin and transcription field. Anyone interested in how chromatin regulates transcription, regulation of ncRNAs, and functions of histone modifying enzymes.

Expertise:

I am a member of the chromatin and transcription field, largely performing genomic experiments. We do not perform microscopy, although sufficiently understand the experiments and results presented here.

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

This reviewer considers himself a generalist with insight into chromatin-based gene regulation, but no first-hand experience with the yeast system or single-molecule imaging.

Promotors are frequently transcribed in both directions. The divergent, *upstream' transcript is frequently unstable. Transcription initiation is regulated through the acetylation of promoter-proximal nucleosomes, where HDAC-dependent deacetylation of histones typically represses transcription initiation. The current manuscript addresses the question whether initiation of coding and divergent, non-coding (DNC) transcription is regulated by the same factors. Previously Marquardt and others showed that H3K56ac-mediated histone exchange has a differential effect on coding and DNC transcription.

Using a clever reporter system, the authors screened for positive and negative regulators that preferentially affect DNC transcription. They discover the Hda1 deacetylase complex as a DNC-biased repressor and diverse HATs as DNC-biased activators. The role of activators could not be validated, presumably due to high variability of the system.

Focusing on Hda1c the authors present data suggesting a larger effect of Hda1c on 'upstream' nucleosomes associated with DNC transcription than in coding transcription. Genome-wide NET-seq mapping was consistent with this differential regulation. Life cell imaging of one specific case argues that Hda1-mediated repression reduced the time between initiation events. The authors employ state of the art methods and in general the data are of very good quality. The effect size is very small, which raises the broader question whether the results, while statistically significant is biological relevant. I have a few comments that the authors may use to revise their manuscript.

1) The differentially regulated coding and DNC transcription are defined by a directionality score. The screen was performed with two reporter loci that are strongly biased for DNC transcription (the idea to detect activators did not work out). Considering that coding and DNC transcription may not be totally independent because of the proximity of target nucleosomes, and sense and antisense transcription may compete for regulators, the question arises how levels of coding transcription affect DNC transcription in wildtype and mutants. The authors stratified their results according to levels of DNC transcription, but discussion and data analysis of the effect of coding transcription on the directionality score may be relevant.

2) The study is strong where the findings can be generalized. The single-molecule live-cell imaging analysis, while done properly, has only limiting impact, because the corresponding coding transcript could not be detected. This si more an anecdotal finding.

3) The effect size is small (20%, on average) and the variability is high. The fact that the HATs that emerged as very robust activators of DNC transcription could not be validated and that the Hda2 subunit of the HDAC complex was not found statistically significant show the limitations of the study. To their credit, the authors discuss these limitations appropriately.

4) Figure S3C suggests that the Hda effect is largest at genes that are poorly expressed, and smaller at more average expression levels. Are we looking at a phenomenon that mainly applies to repressed genes?

Minor issues

5) The NET-seq study involves two replicates. How well did they correlate?

6) For the live-cell imaging replicates were not mentioned. Were replicate studies performed?

7) Fig 4E is not mentioned in the text (mislabeled as 4D)

8) Fig S5 is not mentioned in the main text.

Significance

In summary, this is a high-quality study that presents the results of a genome-wide screen that will be of interest to colleagues in the narrower field. Due to the small effects the results may appeal less to a general readership.

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

In the paper by Gowthaman et al., the authors aim at better understanding the molecular mechanisms controlling divergent non-coding transcription (DNC). They describe a high-throughput yeast genetic screen using two strains in which two loci consisting of a coding and a divergent non-coding transcription unit (CGC1-SUT098 or ORC2-SUT014) were replaced by a bidirectional fluorescent reporter construct encoding mCherry in the coding direction and YFP in the non-coding direction. The two reporter strains were crossed with the yeast deletion library and mutants leading to increased or decreased YFP signal were selected as potential DNC repressors or activators. The two screens identified a number of common potential repressors and activators. Components of the Hda1C histone deacetylase complex were identified as DNC repressors in both screens. This phenomenon was confirmed genome-wide by performing NET-Seq in WT as well as hda1D and hda3D strains. This experiment allowed to identify 1517 DNC transcripts repressed by Hda1. Further analyses indicate that Hda1C represses DNC genome-wide independently of expression levels and that loss of Hda1 does not substantially affect coding transcription.

Live-cell imaging of transcription was then used to show that loss of Hda1 increases DNC transcription frequency rather than duration providing novel information on the link between DNC transcription initiation kinetics and chromatin regulation. Finally, using Chip-seq, the authors show that the level of acetylation over the divergent non-coding units is increased in the absence of Hda1 and some experiments suggest that H3K56 acetylation also contributes to DNC regulation, further strengthening the importance of elevated histone acetylation in efficient DNC.

Importantly, several components of the SWI/SNF chromatin remodeling complex were identified as activators confirming earlier observations (Marquardt et al., 2014). SAGA subunits were also among potential DNC activators, however these effects could not be confirmed through validation experiments. The authors conclude that DNC may be independent of specific activators and mainly due to transcriptional noise resulting from the adjacent NDR.

Overall this paper is very well structured, clearly written and the experiments are well controlled. The genetic screen identifies novel factors involved in the regulation of DNC. The study clearly demonstrates that the level of acetylation is a key regulator of divergent non-coding transcription and that histone deacetylation by Hda1 reduces the frequency of DNC initiation events. While this conclusion is strongly supported by the Net-Seq and Chip-seq metagene analyses, the fluorescence mCherry and YFP values or qRP-PCR analyses of specific genes do not always behave as expected when looking at absolute values rather than mCherry/YFP or GCG1/SUT098 ratios, which is sometimes disturbing when reading the paper. Therefore, the following points should be clarified.

Major points:

Figures 2 and S2A: Figures 2C and D show the mCherry/YFP fluorescence and GCG1/SUT098 RT-qPCR gene expression ratios respectively, which are consistent with a repressive effect of Hda1C on DNC transcription and a potential DNC activating effect of SAGA components. However, the absolute mCherry and YFP or GCG1 and SUT098 expression values presented in Figures S2A and S2B show the opposite: loss of Hda1C subunits rather leads to a decrease in mCherry with not much effect on YFP; moreover loss of Hda3 results in decreased SUT098, which is inconsistent with the whole model. The same comment is valid for the SAGA mutants. It would be good to provide some explanation for these a priori contradictory observations, especially for the Hda1c mutants, which are the major focus of the study. The Net-Seq analyses are certainly more reliable since less subject to protein or RNA stability effects, which may underlie some of the inconsistencies between protein and RNA absolute levels.

Figure 3: this figure examines the effect of Hda1 and Hda3 on the 1517 DNC transcripts. Does loss of this HDAC also increase the expression of all the other 2219 non-coding transcripts identified by Net-Seq, which would make Hda1C a more general repressor of non-coding transcription?

Moreover, does loss of Hda1 or Hda3 reveal DNC transcripts that were not detected in wild-type? This may increase even more the number of genes with divergent transcription.

Figures S3A, B, C: are the 3 groups of DNCs derepressed to the same extent by loss of Hda1 or Hda3? This is difficult to judge given the differences in y-axis scales. Figures S3D, E: the authors show the Net-Seq snapshots for the GCG1 and ORC2 loci. It would be good to add the quantifications as presented in Figure 3 for YPL172C and YDRr216C.

Figures S4A, B, C and D are not well explained. What does the y axis frequency correspond to? Is it the % of cells showing a signal? Is the intensity of SUT098 higher because the transcription initiation frequency is higher and therefore the transcription site signal is more intense?

Figures S4 A-I should be more specifically cited in the text.

Figure 5A: it is really unexpected and unclear why the mCherry/YFP in the WTH3/hda1D and WTH3/hda1D/H3K56mut is increasing compared to WTH3, since DNC is supposed to increase. Similar comment for Figure S5C. This should be clarified in the text.

More generally, as already mentioned above, the fluorescence data are expressed either as mCherry/YFP ratio or as absolute values. It would be good to systematically show the ratios and the absolute values of mCherry and YFP signal; the same for coding and DNC RT-qPCR as well as Net-Seq values when available.

Figures S5A and B are not referred to in the text. It should be mentioned and explained how normalization to H3 affects the levels of acetylated H3 over the NDR. p. 12 "Our data thus suggest to extend the transcriptional noise hypothesis with activities limiting DNC transcription to account for genome-wide variation in non-coding transcription".

If DNC is the result of "transcriptional noise", it is surprising that in the case of CGC1-SUT098, the transcription frequency is higher in the non-coding versus the coding direction. Is the SUT098 behaving like the coding unit in this case? The authors should comment on that.

Minor comments:

p. 4 should one talk about Hda1C-linked histone acetylation facilitates... (should be deacetylation...??) The authors should explain why they chose two coding/non-coding pairs that are cac2D insensitive and whether other criteria, such as level of DNC transcription, were also considered, since GCG1-SUT098 represents one of the most highly expressed divergent non-coding transcripts.

It is hard to understand why both the H3K56A and H3K56Q mutations lead to increased DNC, a result already presented in the Marquardt et al. 2014 paper. It would be helpful to provide a more extensive explanation or hypothesis.

What defines the level of DNC repression? How does the level of repression correlate with the level of coding transcription?

Significance

This is a very interesting and innovative study using cutting edge genetic approaches, genome-wide sequencing as well as single cell imaging to extend our understanding of non-coding transcription regulation and its potential impact on gene expression. It is a nice continuation and complement of an earlier study from the same author (Marquardt et al., 2014) and will certainly be of interest to a large chromatin biology audience.

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

Learn more at Review Commons

---

Reply to the reviewers

Point-by-point response:

*Reviewer #1 (Evidence, reproducibility and clarity (Required)):

**SUMMARY**

This MS tackles a largely unknown topic of vessel formation: how vessels anastomose and lumenise. The authors demonstrate that a matrix protein svep1 produced by neural tube during zebrafish embryogenesis plays a key role with blood flow to orchestrate anastomose formation. Actually in absence of this protein concomitantly with blood flow reduction results in significant decrease of lumenised DLAV segments.

In absence of svep1 they observed an expansion of apelin positive endothelial cells connected with a defect in tip/stalk cell specification. Interestingly the phenotype is amplified by blocking the kinase activity of VEGFR2

**MAJOR COMMENTS**

The most solid evidence on the role of blood flow in cooperating with svep1 relies on the use of tricaine, which reduces heart contractility. Interestingly the authors report some data by using embryo lacking cardiac troponin T2. In my opinion I suggest the author to better analyze the phenotype obtained by the deletion of svep1 together a dose-dependent reduction of tnnt2. This approach is more elegant and physiologic than the use of a chemical compound. Furthermore this approach will allow to better analyze the relations ship between blood flow and the expression of svep1 in neural tube. It should be relevant to establish a sort of flow threshold required to dampen lumenisation. *

Response: We appreciate the comment and have previously attempted to titrate the tnnt2 morpholino as published to have a graded reduction in blood flow. In our hands, this has not proved to be a robust approach, but we are willing to give it another try. In addition, we propose use alternative compounds to tricaine for blood flow reduction without affecting neural physiology. Alternatively we will use a-bungarotoxin mRNA injection to selectively affect neural activity to immobilize the embryos without effects on heart rate and blood flow (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4526548/)

To further improve the findings here reported I suggest to analyze the expression of klf2, which is a well known mechano-sensor of blood flow in several animal species including zebrafish.

Response: We will perform klf2 expression analysis

It's likely that apelin is relevant in the observed phenotype. Which is the phenotype of a double mutant lacking both apl and svep1? Is there a direct influence of blood flow on apl expression?

Response: We will investigate the double loss of function. However, double mutants would take some time, and a combination of morpholino and mutant would likely be the first and best option to answer this question in a reasonable time frame. The effect of flow on apl expression can be tested.

Is there any suggestion that this mechanism is oprative in mammalian?

Response: This is an interesting question and certainly relevant for follow up studies. At present, we can only speculate on a possible connection with flow, given that Svep1 mutations have recently been associated with artherosclerosis. However, whether the anastomosis defect we identify is conserved remains to be seen.

*Reviewer #1 (Significance (Required)):

The data here reported might represent a step forward in the field because a new mechanism is suggested.

The interest is sufficiently broad.

Reviewer #2 (Evidence, reproducibility and clarity (Required)):

**Summary:**

The authors demonstrated that loss of svep1 in zebrafish contributed to defective anastomosis of intersegmental vessels, in addition, such Svep1 acted synergistically with blood flow to modulate vascular network formation in the zebrafish trunk.

**Major comments:**

The expression of svep1 is localized in neurons of neural tube, dorsal epithelial cells (as indicated by transgenic zebrafish) and ventral somite boundary (as indicated by in situ) but is excluded from endothelial cells nor the vasculature. It remains puzzling and the authors have not addressed this very reason of how a gene that is expressed in non-vascular tissue play a crucial role in vessel anastomosis, ie DLAV, ISV lumenization, during angiogenesis. As the entire story of this svep1 is related to its function in angiogenic sprout and lumen formation of vascular tissues, it will be helpful for reader to be able to put the pieces together of how such gene may be functionally involved in such angiogenic process. Previous publication of this gene involved in lymphoangiogenesis, as in this manuscript the authors could provide more evidence of how such gene and its localized expression contribute to different tissue in the vascular system, ie DLAV, instead of the neural tube, dorsal epidermis or ventral somite boundary.*

Response: We appreciate the wish to understand exactly how non-endothelial expression of Svep1 causes an endothelial phenotype selectively under reduced flow conditions. The very nature of this new phenotype requires analysis in vivo, and can not easily be transferred to an ex vivo assay. Therefore, selective loss of function in different cell populations is not easily available. More importantly, the interpretation of such efforts, when mosaic, are marred with issues. At this point, we feel that full molecular characterization of how Svep1 affects endothelial cells during anastomosis will require entirely new approaches and lies beyond what can be achieved in this manuscript.

We will however attempt to clarify the findings and the potential mechanisms in the discussion.

Another puzzling point is that tricaine is the center of the subject in this study. As the authors claim that tricaine-dependent blood flow reduction synergistically augmented the effect of svep1 deficiency. However, tricaine is known acting on neural voltage-gated sodium channels, whether svep1 function was affected by tricaine in the neural tissues and possibly its expression, the authors could provide more explanation and argument in the discussion.

Response: As mentioned in our response to reviewer 1, we will perform additional experiments to try to clarify whether an effect of tricaine on neuronal sodium channels contributes to the phenotype.

It is unclear on p12 "These results suggest that while svep1 loss-of-function produces a cardiac defect that enhances the effect of tricaine on reducing blood flow, svep1 has an additive effect in modulating blood vessels anastomosis" that svep1 deficiency enhances the effect of tricaine leading to reduced blood flow, however, it is not accurate to state that svep1 loss-of-function produces a cardiac defect. It is not sure if the effect of svep1 was actually neural rather than cardiovascular tissue, for example, tricaine acts on neural voltage-gated sodium channel that slowing down heart beat. Whether the authors can explore the possibility that svep1 function in neural rather than cardiovascular tissues, may be discuss why the authors think svep1 enhances the blood flow defect (tnnt2a knockdown or tricaine) on angiogenesis such as DLAV phenotype.

Response: We will attempt to dissect potential contributions by neural effects from cardiac and flow related effects as stated above. Tnnt2 MO and alternative drugs to reduce heart function selectively will be used. We will also clarify the discussion.

On p13, the authors stated that svep1 expression was inhibited by reduced blood flow, however, is it really the effect of reduced blood flow or caused by the chemical tricaine? If tnnt2a knockdown showed a similar phenotype, then it may be more convincing.

Response: see above

\*Minor comments:**

The work on "svep1 loss-of-function and knockdown are rescued by flt1 knockdown" was beautifully done and it is very clear and convincing.

The last two sections, "Vegfa/Vegfr signalling is necessary for ISV lumenisation maintenance and DLAV formation" and "Vegfa/Vegfr signalling inhibition exacerbates svep1 loss-of-function DLAV phenotype in reduced flow conditions" are more related to the flt1 knockdown phenotype. These 3 different sections are actually related in the sense that the rescue phenotype should be explained in the vegf signaling pathway. They are better off to discuss more cohesively about this vegf pathway that will help readers to appreciate more their work in svep1. *

Answer: We agree and will do so.

*Reviewer #2 (Significance (Required)):

This manuscript of svep1 in zebrafish provides new insight in angiogenesis, particularly in development of vessel anastomosis in zebrafish embryo, is very significant in the field and readers who are interested in angiogenesis and zebrafish development, including myself.

Reviewer #3 (Evidence, reproducibility and clarity (Required)):

This manuscript reports that the secreted extra-cellular matrix protein Svep1 plays a role in vascular anastomosis during developmental angiogenesis in zebrafish. Further, the study demonstrates that flow and Svep1 modulate the vascular network in a synergistic fashion. This is a high quality manuscript presenting novel data which compellingly support the conclusions that are made. I have no suggestions for further experimentation but list minor points below.

1. The final paragraph of Discussion is underdeveloped in that it claims regulation of phenotypic robustness in angiogenesis and its failure promises crucial insights into the mechanisms causing breakdown of vascular homeostasis in human disease. However, this issue is not pursued in any substantial way in Discussion. For example, are there known mutations in humans which lead to anastomosis defects and, if so, do any of them relate to the molecules or signaling pathways which are the subject of this manuscript? *

Response: We agree with the wish to see more substantial discussion of the issue of phenotypic robustness and potential links to human disease. The question of anastomosis itself is something that has not been addressed in humans, as it is a rather detailed phenotype observable where predictive patterning occurs and can be dynamically studied. As such, there is a lack of literature and knowledge on signalling pathways that drive anastomosis in humans, and also not many that have been identified in experimental systems or animal models. Flt1 and Vegf signalling, junctional molecules and a few other pathways have been shown to be involved, but nothing is known so far about Svep1 and anastomosis in other system. We will attempt to complement the discussion to make this more clear.

• There are typographical errors in the text so a further proof-read is required. *

Response: thank you, these will be corrected

*Reviewer #3 (Significance (Required)):

This manuscript provides an incremental conceptual advance in our understanding of the molecular mechanisms responsible for vascular anastomosis during developmental angiogenesis. The manuscript will be of interest to developmental biologists and vascular biologists.

My field of expertise pertains to angiogenesis and lymphangiogenesis in the setting of cancer and other diseases. *I am not a developmental biologist.

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

This manuscript reports that the secreted extra-cellular matrix protein Svep1 plays a role in vascular anastomosis during developmental angiogenesis in zebrafish. Further, the study demonstrates that flow and Svep1 modulate the vascular network in a synergistic fashion. This is a high quality manuscript presenting novel data which compellingly support the conclusions that are made. I have no suggestions for further experimentation but list minor points below.

1. The final paragraph of Discussion is underdeveloped in that it claims regulation of phenotypic robustness in angiogenesis and its failure promises crucial insights into the mechanisms causing breakdown of vascular homeostasis in human disease. However, this issue is not pursued in any substantial way in Discussion. For example, are there known mutations in humans which lead to anastomosis defects and, if so, do any of them relate to the molecules or signaling pathways which are the subject of this manuscript?
2. There are typographical errors in the text so a further proof-read is required.

Significance

This manuscript provides an incremental conceptual advance in our understanding of the molecular mechanisms responsible for vascular anastomosis during developmental angiogenesis. The manuscript will be of interest to developmental biologists and vascular biologists.

My field of expertise pertains to angiogenesis and lymphangiogenesis in the setting of cancer and other diseases. I am not a developmental biologist.

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

Summary:

The authors demonstrated that loss of svep1 in zebrafish contributed to defective anastomosis of intersegmental vessels, in addition, such Svep1 acted synergistically with blood flow to modulate vascular network formation in the zebrafish trunk.

Major comments:

The expression of svep1 is localized in neurons of neural tube, dorsal epithelial cells (as indicated by transgenic zebrafish) and ventral somite boundary (as indicated by in situ) but is excluded from endothelial cells nor the vasculature. It remains puzzling and the authors have not addressed this very reason of how a gene that is expressed in non-vascular tissue play a crucial role in vessel anastomosis, ie DLAV, ISV lumenization, during angiogenesis. As the entire story of this svep1 is related to its function in angiogenic sprout and lumen formation of vascular tissues, it will be helpful for reader to be able to put the pieces together of how such gene may be functionally involved in such angiogenic process. Previous publication of this gene involved in lymphoangiogenesis, as in this manuscript the authors could provide more evidence of how such gene and its localized expression contribute to different tissue in the vascular system, ie DLAV, instead of the neural tube, dorsal epidermis or ventral somite boundary.

Another puzzling point is that tricaine is the center of the subject in this study. As the authors claim that tricaine-dependent blood flow reduction synergistically augmented the effect of svep1 deficiency. However, tricaine is known acting on neural voltage-gated sodium channels, whether svep1 function was affected by tricaine in the neural tissues and possibly its expression, the authors could provide more explanation and argument in the discussion.

It is unclear on p12 "These results suggest that while svep1 loss-of-function produces a cardiac defect that enhances the effect of tricaine on reducing blood flow, svep1 has an additive effect in modulating blood vessels anastomosis" that svep1 deficiency enhances the effect of tricaine leading to reduced blood flow, however, it is not accurate to state that svep1 loss-of-function produces a cardiac defect. It is not sure if the effect of svep1 was actually neural rather than cardiovascular tissue, for example, tricaine acts on neural voltage-gated sodium channel that slowing down heart beat. Whether the authors can explore the possibility that svep1 function in neural rather than cardiovascular tissues, may be discuss why the authors think svep1 enhances the blood flow defect (tnnt2a knockdown or tricaine) on angiogenesis such as DLAV phenotype.

On p13, the authors stated that svep1 expression was inhibited by reduced blood flow, however, is it really the effect of reduced blood flow or caused by the chemical tricaine? If tnnt2a knockdown showed a similar phenotype, then it may be more convincing.

Minor comments:

The work on "svep1 loss-of-function and knockdown are rescued by flt1 knockdown" was beautifully done and it is very clear and convincing.

The last two sections, "Vegfa/Vegfr signalling is necessary for ISV lumenisation maintenance and DLAV formation" and "Vegfa/Vegfr signalling inhibition exacerbates svep1 loss-of-function DLAV phenotype in reduced flow conditions" are more related to the flt1 knockdown phenotype. These 3 different sections are actually related in the sense that the rescue phenotype should be explained in the vegf signaling pathway. They are better off to discuss more cohesively about this vegf pathway that will help readers to appreciate more their work in svep1.

Significance

This manuscript of svep1 in zebrafish provides new insight in angiogenesis, particularly in development of vessel anastomosis in zebrafish embryo, is very significant in the field and readers who are interested in angiogenesis and zebrafish development, including myself.

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

This MS tackles a largely unknown topic of vessel formation: how vessels anastomose and lumenise. The authors demonstrate that a matrix protein svep1 produced by neural tube during zebrafish embryogenesis plays a key role with blood flow to orchestrate anastomose formation. Actually in absence of this protein concomitantly with blood flow reduction results in significant decrease of lumenised DLAV segments.

In absence of svep1 they observed an expansion of apelin positive endothelial cells connected with a defect in tip/stalk cell specification. Interestingly the phenotype is amplified by blocking the kinase activity of VEGFR2

MAJOR COMMENTS

The most solid evidence on the role of blood flow in cooperating with svep1 relies on the use of tricaine, which reduces heart contractility. Interestingly the authors report some data by using embryo lacking cardiac troponin T2. In my opinion I suggest the author to better analyze the phenotype obtained by the deletion of svep1 together a dose-dependent reduction of tnnt2. This approach is more elegant and physiologic than the use of a chemical compound. Furthermore this approach will allow to better analyze the relations ship between blood flow and the expression of svep1 in neural tube. It should be relevant to establish a sort of flow threshold required to dampen lumenisation.

To further improve the findings here reported I suggest to analyze the expression of klf2, which is a well known mechano-sensor of blood flow in several animal species including zebrafish.

It's likely that apelin is relevant in the observed phenotype. Which is the phenotype of a double mutant lacking both apl and svep1? Is there a direct influence of blood flow on apl expression?

Is there any suggestion that this mechanism is oprative in mammalian?

Significance

The data here reported might represent a step forward in the field because a new mechanism is suggested.

The interest is sufficiently broad.

Transparent Peer Review

---

Download the complete Review Process [PDF] including:

• reviews
• authors' reply
• editorial decisions

</br>

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

Learn more at Review Commons

---

Reply to the reviewers

We thank the reviewers for their comments, criticisms and suggestions that will help to improve the quality of our manusrcipt.

Please find enclosed in this initial response our answer to each point raised by the reviewers.

Please note that for several answers normally come along with an additional figure that could be added in the full revised version of the manuscript. However, these additional figures could not be added in the way we have to submit our answers but we are ready to send a pdf file including our answers with the additional figures upon request.

Reviewer #1 (Evidence, reproducibility and clarity (Required)):

The paper by Genest et al. describes the effect of flotillins and sphingosine kinase 2 to stabilize AXL as a mechanism to promote epithelial-mesenchymal transition in breast (cancer) cells. The potential role of vesicles trafficking EMT-promoting proteins is of high interest in the field, also for exploring new opportunities of pharmacological targeting. However, the paper fails to convincingly demonstrate that the proposed mechanism is of real importance to support or promote EMT for the following main reasons:

1-a) The role of flotillins is studied only by overexpression and in the context of non-cancerous MCF10A cells, while breast cancer cells of epithelial-like origin are not analyzed.

Regarding the first part of the point raised here, we are not sure to understand correctly the sentence “[…] while breast cancer cells of epithelial-like origin are not analyzed”. Indeed, we used the breast cancer cell line MDA-MB-231 and a derived cell line that we generated by knocking down flotillin expression (MDA-MB-231shFlot2) in the second part of this study (Figure 6C, F and H and S7A, E and F). This previously characterized cell line allowed us to demonstrate that abolishing flotillin overexpression was sufficient to significantly inhibit the invasive properties of MDA-MB-231 cells (Planchon et al, J Cell Science 2018, https://doi.org/10.1242/jcs.218925

Although flotillin upregulation induces some major mechanisms of the EMT process in MCF10A cells, flotillin downregulation was not sufficient to reverse the EMT phenotype in MDA-MB-231 cells. This could be explained by the fact that EMT is a multifactorial process and that MDA-MB-231 cells went through too many irreversible changes leading to this process. By contrast, when we analyzed EMT markers after SphK2 inhibition or knock down in MCF10AF1F2 and in MDA-MB-231 cells (Figure 6A-C), we could observe a significant decrease in ZEB1 expression.

1-b) This is contrast with the purpose of the paper (see abstract, introduction, patients' data) which is to study tumors and EMT. Effect of shRNAs is also not reported, making it difficult to estimate the importance on the EMT phenotype.

As we mentioned in our manuscript, previous studies by other groups who downregulated flotillin expression in different cancer cell lines using siRNA approaches or re-expression of miRNAs that inhibit flotillin expression, already showed flotillin participation in EMT (for review please see, Gauthier-Rouvière et al, Cancer Metastasis Review, 2020, **doi: 10.1007/s10555-020-09873-y).

In this context, the novelty and the first goal of our study was to investigate how strong is the contribution of flotillin upregulation to EMT induction. To achieve this goal, we chose on purpose to use non-tumoral epithelial cells that do not harbor the anomalies already favoring EMT, unlike the cancer cell lines used in previous studies. In these non-tumoral models (the human MCF10A and mouse NMuMG mammary epithelial cell lines), we ectopically overexpressed flotillins (MCF10AF1F2 and NMuMGF1F2) to levels similar to what observed in invasive breast cancer cells. Using this approach, we found that flotillin overexpression is enough to induce EMT.

1-c) Then, alteration of EMT should be concluded also from other non-genetic functional parameters, not just by markers. For instance: was morphology of the cells changed? Was cell migration affected with F1F2?

Our conclusion that flotillin upregulation is sufficient to induce EMT in MCF10AF1F2 and NMuMGF1F2 cells is not based only on genetic functional parameters or markers. For instance, Figure S1 (panels H and I) shows a strong modification of the cell morphology and of the actin cytoskeleton organization in NMuMG cells upon flotillin upregulation. NMuMGF1F2 cells became flat and lost their apical F-actin belt and exhibited an increase in stress fibers.

As shown below (Additional Figure 1), similar modifications of the cell morphology and of the F-actin cytoskeleton organization occur also when flotillins are upregulated in MCF10A cells (see below the comparison of MCF10A and MCF10AF1F2 cells) (these data could be added in the manuscript).

ADDITIONAL FIGURE 1 CAN NOT BE ADDED BUT IS AVAILABLE UPON REQUEST

Additional figure 1: Upregulation of flotillins in MCF10A cells leads to changes in the cell morphology and in F-actin cytoskeleton organization. Comparison of the morphology and of the actin cytoskeleton organization in MCF10AmCh and MCF10AF1F2 cells. Confluent cells were fixed and stained for F-actin (green) using Alexa488-conjugated-Phalloidin and for nuclei (blue) using Hoechst (in panel A flotillin2-mCherry signal is shown). (A) Upper panels show the maximum intensity projection images (MIP) of MCF10AmCh (control) and MCF10AF1F2 (flotillin overexpression) cells obtained from a stack of images acquired by confocal microscopy. Lower panels show magnified images from the boxed areas, including one single plane and the x-z and y-z projections along the indicated axes. (B) 3D reconstruction images obtained from the region in the boxed area from the MIP-images shown in A.

These data show that in MCF10AF1F2 cells the apical actin belt is lost and the height of the cellular monolayer is lower compared with control MCF10AmCh cells.

We also analyzed the migration capacity of these cells (shown in Figure 3G of the submitted manuscript). Briefly, using a Boyden chamber assay, we showed that flotillin upregulation significantly increased migration of MCF10A cells (Figure 3G). We previously demonstrated that flotillin upregulation also promotes cell invasion in 3D using a spheroid assay (Planchon et al, J Cell Science, 2018, https://doi.org/10.1242/jcs.218925**). As shown below (Additional Figure 2), using a wound healing assay, we also observed that cell velocity is higher in flotillin-overexpressing NMuMGF1F2 cells than in control NMuMG cells (this could be added to the manuscript).

ADDITIONAL FIGURE 2 CAN NOT BE ADDED BUT IS AVAILABLE UPON REQUEST

Additional figure 2: Upregulation of flotillins in NMuMG cells increases cell velocity in a 2D migration assay. (A) Representative images of NMuMGmCh (control) and NMuMGF1F2 cells during wound healing. The yellow dashed line indicates the leading edge of the migrating monolayer at the indicated times. The trajectory of 60 individual cells was tracked and the cell velocity and persistence of migration were extracted. The histogram shows the velocity quantification (mean ± SEM of 4 independent experiments). (B) Representative trajectories of individual cells.

2) AXL up-regulation is not very strong (2-fold). What is unclear is if the minimal AXL increase due to F1F2 really provides a significant contribution to the EMT phenotype (as the authors conclude). The siRNA experiment knocks down all AXL, not just the F1F2-induced levels, making it difficult to estimate the real effect of the mechanism proposed.

As shown in figure 3A and D, in MCF10AF1F2 cells compared with MCF10AmCh cells, we measured a significant 2.5 ± 0.7-fold increase in the AXL protein level. We do not think that this can be considered as a minimal increase.

Considering that flotillin upregulation may affect simultaneously different receptors (Figure S2I, Figure S6A-F), we did not expect that downregulating a single receptor would have a major impact on the level of EMT markers and on cell migration. Yet, after knocking down AXL in MCF10AF1F2 cells, we observed a decrease in ZEB1 and N-cadherin expression and the re-expression of E-cadherin (Figure 3D-F) and the inhibition of cell migration (Figure 3G). The fact that we observed such an effect by downregulating AXL, which according to Reviewer #1 is minimally increased, might be explained by its well-known ability to act not alone but through cross-talk with other signaling receptors (Graham et al, Nature Reviews Cancer 2014; Halmos and Haura, Science Signaling 2016; Colavito et al, Journal of Oncology 2020).

As suggested by Reviewer #1, ideally, it would be interesting to bring back AXL to its level in MCF10AmCh cells to better evaluate only the contribution of its increase. However, adjusting so precisely the efficacy of AXL downregulation by siRNA seems quite difficult to achieve.

3) Why didn’t the author focus on EphA4 (or to a lesser extent ALK), which showed better regulation ?

As we mentioned (page 18) “the available tools allowed us to validate this result only for AXL, but not for EphA4 and ALK”**.

Nevertheless, for EphA4, we showed in Figure S6 that it is located in flotillin-positive late endosomes (Figure S6 A and C, for MCF10AF1F2 and NMuMGF1F2 cells, respectively) in a phosphorylated form (using an antibody against P-Y588/Y596-EphA4 that works in NMuMG cells, Figure S6D). However, the signals obtained by western blotting using the same antibody were too low to validate any significant variation of EphA4 Y-phosphorylation status, as suggested by the results from the phospho-RTK array.

Regarding ALK, the increase in its phosphorylation, suggested by the phospho-RTK array, remains puzzling to us. By western blotting of cell lysates and in the presence of positive controls, we did not detect any positive signal for phosphorylated ALK and even for total ALK in MCF10A and MCF10AF1F2 cells. In addition, to our knowledge, ALK expression in MCF10A cells has never been reported in the literature. These observations did not encourage us to pursue our investigations on ALK.

Moreover, several points led us to focus on AXL. Indeed, AXL expression is associated with the acquisition of a mesenchymal cell phenotype, invasive properties, and resistance to treatments and AXL is an attractive therapeutic target against which several inhibitors are in preclinical and clinical development (Shen Y et al. Life Sciences 2018). Moreover, AXL expression in tumors is attributed to post-transcriptional regulation, but the mechanisms are totally unknown. Understanding how its stabilization and signaling can be triggered by flotillin-mediated endocytic pathways is new and of high significance for the cancer field and the trafficking community.

3) The conclusions of the manuscript are contradicted by the reported clinical data. In Figure S4 the authors clearly observe co-expression of Flotillin 1 and AXL prevalently in luminal breast cancers, which is the subtype known to not be driven by EMT. This evidence already indicates that this (otherwise interesting) mechanism is not relevant to EMT in breast cancer. So, the conclusions are not supported by the data, and the experimental setup and model chosen are not appropriate to generalize the findings to cancer.

We acknowledge that flotillin 1/AXL co-expression is highest in the luminal subtype. If this co-expression was observed only in this particular subtype, we would have agreed that it excluded that flotillins and AXL co-overexpression may participate in EMT in tumor cells. However, our results show that flotillin 1 and AXL are co-expressed also in other subtypes that have undergone EMT. Considering this observation and the influence of flotillin upregulation on AXL overexpression we reported here, we believe that the point raised by the Reviewer is not sufficient to exclude that the co-upregulation of flotillins and AXL can participate in EMT induction in breast cancer cells.

**Minor (here the most important):**

4) The point of the Figure 2 is not clear. Why this part should have such a central role in the story? The entire data presented are not followed up in the rest of the paper. Moreover, in some cases upregulations also questionably significant (like RAS and STAT3 are not even 2 fold).

Moreover, the error bars are so small that it seems unrealistic that the plots indicate three independent experiments.

Because the activation of oncogenic signaling pathways is crucial to promote EMT, we think that analyzing these pathways in the context of flotillin upregulation is coherent with the message of the paper.

To our knowledge, the amplitude of up- or down-regulation has nothing to do with its significance. The amplitude also depends strongly on the context (stimulation with an agonist, overexpression of GEF, etc). For instance, increases lower than 2-fold are frequently reported (Bodin and Welch, Mol Biol Cell, 2005; Miura SI et al, Arteriosclerosis, Thromb and Vasc Biology, 2003; Matsunaga-Udagawa R et al, J Bio Chem 2010)** when assessing the activity of Ras or small GTPases, but they represent real upregulations. Furthermore, Ras activation is supported by the downstream 4-fold activation of ERK that we measured (Figure 2C).

In Figure 2, panels B, C, E, F and J, considering the amplitude of the mean increases shown, the error bars corresponding to SEM do not seem disproportionately small.

As the Reviewer seems to insinuate that we have not performed independent experiments, we are presenting in the table below the detailed results all obtained from independent experiments.

Panel

Parameter measured

Number of independent experiments

Fold of increase value in MCF10AF1F2 cells compared with MCF10AmCh cells in each experiment

Mean

SEM

p-value

B

Ras-GTP

5

1.95 ; 1.96 ; 1.18 ; 1.67 ; 1.86

1.72

0.14

0.001

C

Phospho- ERK

5

1.24 ; 5.43 ; 3.22 ; 6.11 ; 3.52

3.71

0.73

0.0042

E

Phospho-AKT

4

2.29 ; 6.54 ; 3.76 ; 2.6

3.8

0.97

0.0276

F

Phospho-STAT3

4

1.63 ; 1.63 ; 2.42 ; 1.60

1.82

0.20

0.0066

J

Phospho-SMAD3

8

4.1 ; 5.12 ; 6.29 ; 1.82 ; 2.58 ; 6.66 ; 2.82 ; 5.40

4.35

0.64

0.0001

In the legend to figure 2 panels C, E, F, J, “The histograms show […] with control MCF10AmCh **cells calculated from 4 independent experiments” was corrected by “The histograms show […] with control MCF10AmCh cells calculated from at least 4 independent experiments” as data shown in panel J were actually calculated from 8 independent experiments.

5) More robust statistical analysis should be provided in the Figure 1 to support that EMT is suppressed with F1F2 overexpression. For instance a more standard GSEA on hallmark signatures.

To avoid confusion, we understand that Reviewer #1 meant “… that EMT is induced with F1F2 overexpression” and not “… suppressed …”.

As recommended by Reviewer #1, we performed a GSEA on the hallmark signature and the results are already included in the current revised version of our manuscript (figure 1C).

6) In Figure 3 E-Cadherin is rescued with siAXL in the IF but not in the western blot.

Using siRNA transfection, we can have a mosaic effect due to the fact that not all the cells of the sample are transfected and thus efficiently knocked down. This mosaicism was clear when we analyzed E-cadherin by immunocytochemistry. Indeed, in some cells, probably the ones that have been more efficiently transfected with the AXL siRNA, E-cadherin expression is clearly seen. By western blotting, which provides a global analysis in which transfected and non-transfected cells are mixed, this was not significantly higher than in MCF10AF1F2 cells transfected with a control siRNA, although there was a trend towards increased E-cadherin expression in MCF10AF1F2 transfected with the AXL siRNA.

For the revised version of our manuscript we will try to improve the efficacy of the AXL siRNA and test whether we can fully rescue E-cadherin expression. The corresponding panel could be modified according to the data we will obtain.

7) Some sentences require clarifications. The authors should be more clear on why ZEB2 antibody was not available or what they mean with "Unfortunately the available tools..".

Page 7: we wrote «no anti-Zeb2 antibody is available». We should have said: «none of the anti-Zeb2 antibodies tested worked in MCF10A cells». We decided to remove “no anti-Zeb2 antibody is available” from the sentence to avoid confusion in the revised version of our manuscript.

Page 19: we wrote «unfortunately the available tools» to refer **the available tools against EphA4 and ALK that did not allow us to validate the data obtained using the phospho-RTK array showing that the Y-phosphorylation of these two RTK is increased in cells with upregulated flotillins. (see also our answer to major point 2).

8) Western blot from the CHX experiment should be shown, at least in the supplements. Again, the standard deviation in this experiment is minimal, was this really an average of three independent experiments (and not three western on the same lysates)?

As asked, a representative western blot is now shown in Figure 3C in the current revised version of the manuscript.

As indicated in the legend to the figure already in the initial version of our manuscript: “**The results are the mean ± SEM of 6 to 8 independent experiments depending on the time point, and are expressed as the percentage of AXL level at T0”. We wish to reassure Reviewer#1 that the results are really based on western blots performed on different lysates obtained in independent experiments. We can show the Reviewer these data obtained from independent experiments if necessary.

9) All conclusions are derived from one single cells MCF10a. NMuMG cells are shown at the beginning but not used for the rest of the paper. Anyway, if this wants to be a cancer research paper, then cancer cells needs to be used.

It is true that we did not use a cancer cell line at the beginning of the paper because, as expected, flotillin knock-down did not allow to revert the mesenchymal phenotype of MDA-MB-231 cells toward an epithelial one. If this had been obtained, we would have used these cells from the beginning of the paper. The lack of reversion of the mesenchymal phenotype after flotillin knock-down was expected. Indeed, the EMT process is multifactorial and the decrease of flotillins alone is obviously not sufficient to reverse it in a tumor cell line bearing multiple oncogenic mutations. Moreover, because we wanted to assess whether flotillin upregulation is sufficient in normal cells to acquire the properties of tumor cells and particularly to induce EMT, we used human MCF10A and murine NMuMG cells, two non-tumoral epithelial cell lines. Until now, the studies carried out on the effects of flotillin overexpression have used tumor cells that already harbor pro-oncogenic perturbations, preventing to show that flotillin overexpression alone activates oncogenic processes leading to EMT, and to identify the downstream mechanisms.

Nevertheless, we have used the MDA-MB-231 cell line in several experiments to analyze: i) AXL distribution and internalization following the knock-down of flotillins (Figures 4 and S5), ii) SphK2 and flotillin 2 co-localization and co-endocytosis (Figures 5A and D and S7A), iii) the impact of SphK2 inhibition on AXL expression level distribution and endocytosis (Figure 6), iv) SphK2 expression level upon flotillin knock-down (Figure S7E) and AXL expression level upon SphK1 inhibition (Figure S7F). With these experiments performed in MDA-MB-231 cells, we showed that AXL and SphK2 colocalize in flotillin-positive late endosomes and are co-endocytosed from the plasma membrane containing flotillin-rich domains to flotillin-positive vesicles. We also demonstrated that flotillins and SphK2 control the rate of AXL endocytosis and its stabilization.

We recently obtained additional data with HS578T cells, another triple negative breast cancer cell line, on the co-trafficking of AXL and flotillins as well as the co-trafficking of SphK2 and flotillins (Additional Figure 3, this data could be added in the fully revised version of our manuscript).

In addition, we observed that inhibiting SphK2 also decreased the level of AXL in HS578T cells. This data could be added in the revised version of the manuscript (see data in our answer to Point #1 from Reviewer #3).

• ADDITIONAL FIGURE 3 CAN NOT BE ADDED BUT IS AVAILABLE UPON REQUEST*

Additional figure 3: Co-trafficking of SphK2 and AXL with flotillin 1 in intracellular vesicles in HS578T cells. HS578T cells co-expressing Flot1-mCherry with SphK2-GFP (A) or AXL-GFP (B) were monitored by time lapse spinning disk confocal video-microscopy. On the right of each panel are shown still images at different time points (min) of the boxed area. The colored arrows allow following three distinct vesicles that are positive for Flot1-mCherry and Sphk2-GFP, or AXL-GFP.

10) The methods section contains inconsistent data about patients' samples (9 are indicated, but the Figure S4 features 37). Then, where those other 527 come from?

We corrected the manuscript and added all characteristics regarding the 37 patients in the “Supplementary information” section.

The 527 patients are from another cohort and were used for the analysis of the correlation between the mRNA levels of FLOT1 and p63 in breast cancer biopsies from 527 patients (Figure 2I). This cohort was described in our previous study (Planchon et al. J Cell Science 2018, https://doi.org/10.1242/jcs.218925). In the revised version of our manuscript, we now refer to this previous article in the “Result” section and in the legend to figure 2I to explain the origin and characteristics of this cohort.

11) Some figures do not match with the legends or with the description in the text. It has not been easy to review this paper.

We apologize as we indeed made one mistake in figure 2 that was inserted into the manuscript and that was actually figure S2 (that appeared twice). However, the correct figure 2 was uploaded on the website of Review Commons and BioRxiv. Regarding the comments made in point 4, it seems that Reviewer #1 examined the correct figure 2 that was uploaded and that matches the legend indicated in the manuscript.

Besides this mistake, we do not see any other mismatch between figures and legends.

Reviewer #1 (Significance (Required)):

I am a cancer biologist working on EMT.

**Referee Cross-commenting** I have nothing to comment on other's reviews.

Reviewer #2 (Evidence, reproducibility and clarity (Required)): Genest and co-authors present in this paper new fascinating evidence on how intracellular trafficking can modulate oncogenic signalling.

First of all, they show how overexpression of Flotillin1 and 2 in non-cancerous breast lines can induce a strong reprogramming towards an EMT phenotype. They analyse mRNA and protein expression, intracellular distribution of activated proteins, cell phenotypes to demonstrate a strong activation of oncogenic signalling pathways. They then identify AXL as a key player in this process and show how this protein is stabilised upon Flotillin expression. The authors use an amazing variety of approaches to study the endocytosis and the trafficking of endogenous, GFP-tagged, Halo-tagged and Myc-tagged AXL in different cell lines and their data are strong and very convincing, the images are of very high quality and the analysis rigorous. Their data strongly support the hypothesis that high Flotillin levels triggers AXL endocytosis and accumulation in non-degradative late endosomes where signalling remains active. The authors then show how SphK2 has a key role in AXL stabilisation, it colocalises with Flotillin, AXL and CD63 and its activity (which they block by using inhibitors or siRNA) is necessary for flotillin-induced AXL stabilisation and EMT induction. The paper is extremely well written, the data flow logically and they are appropriately presented and analysed. I don't have any major comment and I believe the paper is suitable for publication.

We thank the Reviewer for the positive appreciation on our manuscript.

I have only some minor comments/questions: 1) did the authors try to colocalise AXL with endogenous Flotillin in MDA-MB-231 cells? They could use the antibodies used in Fig S1B. Of note, the authors have shown it in luminal tumours in Fig S4C.

We performed co-immunofuorescence experiments to detect endogenous AXL with endogenous Flotillin in MDA-MB-231 cells. As shown below (Additional Figure 4), we could find AXL and Flotillin being present in the same intracellular endosomes. Images could be added in the revised version of the manuscript.

ADDITIONAL FIGURE 4 CAN NOT BE ADDED BUT IS AVAILABLE UPON REQUEST

Additional figure 4: Endogenous AXL and flotillin 1 are found in the same in intracellular vesicles in MDA-MB-231 cells. MDA-MB-231 cells were fixed and labelled with relevant antibodies directed against Flotillin1 and AXL. Scale bar in the main image : 10 µm. Scale bars in the magnified images from the boxed area : 1 µm. Arrows indicate flotillin and AXL positives vesicles

2) In Fig6G, it appears that AXL-Flotillin colocalization is lost upon SphK2 inhibition. Is this the case? It could be that the correct lipids are necessary for the formation of Flotillin-positive internalisation domains and this could be very interesting and reinforce the model proposed in the paper.

In figure 6G, cells were not permeabilized. Thus, only AXL at the cell surface was labelled using an antibody against the extracellular domain of AXL. Because flotillin 2 is tagged with mCherry, this allowed its visualization revealing its localization both at the cell surface and intracellularly in the inset of the lower pane l of figure 6G.

After 6 hours of treatment using the opaganib inhibitor, we did not notice any major change in AXL-flotillin colocalization at the cell surface. Somehow, this is expected because blocking the generation of S1P is more likely to inhibit the invagination of flotillin-rich membrane microdomains rather than their formation.

3) I would remove the sentence on line 995-997 "to our knowledge this is the first report to describe ligand-independent AXL stabilization..." as the cells are not serum starved in all experiments and animal serum can contain variable amounts of the ligand GAS6.

We understand and agree with Reviewer #2, this sentence has been modified by “**To our knowledge this is the first report to describe AXL stabilization following its endocytosis”

Please note that the authors don't have to necessarily address comments 1-2, their paper is already very rich in convincing data.

Reviewer #2 (Significance (Required)):

AXL is a major oncogene that promotes EMT in a variety of tumour types. Understanding how its signalling can be triggered by endocytic pathways even in cells that are non-cancerous is very important and of high significance for the cancer field and the trafficking community.

---

Reviewer #3 (Evidence, reproducibility and clarity (Required)):

This is an interesting and well written paper describing that upregulated flotillin promotes an endocytic pathway called upregulated flotillins-induced trafficking (UFIT) that mediates AXL endocytosis and allows its stabilization. Consequently, stabilized AXL in these flotillin-positive late endosomes enhances activation of oncogenic signaling pathways that promotes EMT. The authors suggest that Flotillin upregulation-induced AXL stabilization requires the activity of SphK2. However, this latter point is not supported by the data and further studies are needed to support this important conclusion.

**Major concerns:**

1. Most of the conclusions are based on effects of high concentrations (50 uM) of an ill-defined SphK2 inhibitor. The experiment described in Figure 6C-H need to be confirmed by downregulation of SphK2.

We understand that Reviewer #3 is concerned that in our experimental conditions, the effects we observed could be really explained by a specific inhibition of SphK2.

From the literature, among all the inhibitors described for SphK2, opaganib (ABC294640) is the most specific inhibitor available. It was shown to have no inhibitory effect on SphK1 up to 100 µM (French et al, J Pharmacol Experimental Exp Ther 2010; Neubauer HA and Pitson SM, The FEBS Journal 2013). In agreement, we found that PF543, the most specific SphK1 inhibitor, had no effect on AXL level (Figure S7F), unlike incubation with opaganib (Figure 6A and C), and that was confirmed in MCF10AF1F2 cells by the knock down of SphK2 with a specific siRNA (Figure 6B).

In the literature, depending on the cell lines, opaganib is used in vitro in the 10 to 60 µM range. Opaganib IC50 on recombinant SphK2 was established at 60 µM (French et al, J Pharmacol Experimental Exp Ther 2010). In our experiments, opaganib was used at a concentration of 50 µM, below the IC50 value, as previously done by Nichols’ group (Riento and al, PloS ONE, 2018). In most of our experiments (Figure 6, A, D, E-I, Figure S7D), opaganib was added for a maximum of 10 hours, which is shorter compared to what done in other studies (24-48 hours). Furthermore, it was shown that an opaganib concentration of 50 µM does not have any inhibitory effect in vitro on 20 protein kinases tested, including PKA, PKB, PKC, CDK, MAP-K, PDK1 and Src (French et al, J Pharmacol Experimental Exp Ther 2010).

In addition to inhibit SphK2, acting in a sphingosine-competitive manner, opaganib also was shown to act as an antagonist of estrogen receptor (ER), and inhibits ER-positive breast cancer tumor formation in vivo (Antoon JW et al, Endocrinology 2010). If Reviewer #3 is concerned about the possibility that the opaganib downstream effects we observed in our study might be explained by ER inhibition, we remind that we used cellular models that do not express ER. Indeed, the MDA-MB-231 cell line is a triple negative breast cancer cell line. MCF10A cells also do not express ER (Lane MA et al, Oncolgy Report, 1999,)** and our transcriptomic analysis (Table S1) did not reveal any increase in the expression of ER genes in MCF10AF1F2 cells in which flotillins are upregulated, thus eliminating a possible non-specific effect of opaganib in these cells.

In conclusion, we hope that these arguments help to convince Reviewer #3 that our experiments were performed in conditions where we carefully limited the possibility of opaganib off-target effects, on the basis of the currently available opaganib-related data from the literature.

We totally agree with Reviewer #3 that complementary experiments by downregulating SphK2 must be used. In agreement, we already downregulated SphK2 by siRNA in MCF10AF1F2 cells. This led to a significant decrease in AXL and ZEB1 expression. In the current revised version of the manuscript we have added data obtained with similar siRNA experiments performed in MDA-MB-231 cells (now Figure 6C). In agreement, we observed AXL and ZEB1 downregulation.

As shown below (Additional Figure 5) we recently obtained similar data in HS578T cells, showing that inhibiting SphK2 also affects AXL protein level in this triple negative breast cancer cell line (these data could be added in the manuscript).

ADDITIONAL FIGURE 5 CAN NOT BE ADDED BUT IS AVAILABLE UPON REQUEST

Additional figure 5: SphK2 inhibition decreases AXL level in HS578T cells. HS578T cells were incubated with opaganib (50µM, 10 hours) (A) or with siRNA Ctrl or siRNA SphK2 for 72 hours (B). Cell lysates were blotted with relevant antibodies against AXL, SphK2 and actin. The histograms show AXL level (normalized to actin) expressed as fold-increase compared with the control condition, and data are the mean ± SEM of 3 (A) and 4 (B) independent experiments.

Reviewer #3 also asks to use the siRNA approach on experiments shown in previous panels D-H (now panels E-I) of figure 6.

In complement to Figure 6D (now Figure 6E), experiments using a siRNA against SphK2 to show that “**AXL decrease upon SphK2 inhibition is not due to protein synthesis inhibition” are on-going and the obtained data could be added in the full revised version of our manuscript.

However, we are unfavorable to use a siRNA against SphK2, in addition to opaganib, in the experiments done to measure the effect of SphK2 inhibition on the rate of AXL internalization (previously in Figure 6E and F, now Figure 6F and G) and the level of AXL at the cell surface (previously in Figure 6G and H, now Figure 6H and I). Indeed, we carefully chose a short (4 hours) incubation with opaganib at the end of which the total cellular level of AXL was not yet decreased, allowing to measure unambiguously a defect in AXL endocytosis or a change in the level of AXL at the cell surface. We believe that it would be very difficult to achieve similar experiments using a siRNA against SphK2. It would require to determine the exact time after siRNA transfection leading to a sufficient SphK2 level reduction but in conditions where AXL level is still maintained. We think that due to the mosaic transfection efficiency, being able to precisely synchronize the effect of a siRNA at its beginning is impossible.

1. Does overexpression of SphK2 reverse the effects of the SphK2 inhibitor? In a similar manner, does overexpression of SphK2 enhance stabilization of AXL?

To answer the first question, it is not clear for us how to test whether SphK2 overexpression can reverse the effects of the SphK2 inhibitor because the ectopically expressed SphK2 would also be sensitive to the inhibitor. This would require to overexpress a SphK2 mutant that is catalytically active but insensitive to the inhibitor, and to our knowledge, such a mutant does not exist.

Regarding the second question, we are currently generating a retroviral DNA construct allowing to overexpress SphK2 homogeneously in the cell population. Then we will test whether it further increases AXL level through its stabilization. This will be tested in cells upregulated for flotillin. As we showed in Figure 6 A and D (previously Figure 6 A and C) that AXL level depends on SphK2 activity only in cells that overexpress flotillins, we anticipate that there will be no impact in a cell line with a moderate level of flotillin. Results could be added in the fully revised manuscript.

1. Although the authors suggest recruitment of SphK2 and formation of S1P in UFIT, there are no measurements of S1P. Also, there is no indication that SphK2 is activated despite the fact that ERK and AKT are activated in UFIT and are known to phosphorylate and activate SphK2. Is SphK2 that is recruited to flotillin phosphorylated?

To answer the first point raised by Reviewer#3, we recently performed, in collaboration with a lipidomic platform, a comparative analysis by quantitative mass-spectrometry of S1P levels between MCF10AmCh and MCF10AF1F2 cells. As we anticipated, the results show a 3,5-fold increase in S1P in MCF10AF1F2 cells compared with MCF10AmCh (Additional Figure 6). This data agrees with the fact that we found that the SphK2 catalytic activity is required for the UFIT pathway mediated AXL stabilization. This result is also in agreement with the study from the Nichols’ group which detect a decrease in S1P in cells in which flotillins were knocked out (Riento et al, PloS ONE, 2018). The results regarding the analysis of S1P level along with the complete methodology used will be added in the fully revised version of our manuscript.

ADDITIONAL FIGURE 6 CAN NOT BE ADDED BUT IS AVAILABLE UPON REQUEST

Additional figure 6: Upregulation of flotillins in MCF10A cells promotes an increase in the level of Sphingosine-1-phosphate. The level of sphingosine-1-phosphate was compared by quantitative mass-spectrometry analysis from three independent samples of MCF10AmCh and MCF10AF1F2 cells. The results are expressed in pmol equiv / 1 . 106 cells. The graph shows the value for each sample and the bar horizontal bars indicate the mean value for each condition.

Regarding the second point, we would like to clarify that we do not think that SphK2 interacts directly or indirectly with flotillins because SphK2 did not co-immunoprecipitate with flotillins (not shown). Thus, investigating by western blotting SphK2 phosphorylation status in flotillin immunoprecipitates is pointless. In theory, we could investigate the activity-related phosphorylation status of SphK2 associated with flotillin rich-membranes and endosomes. But this seems difficult to achieve because unfortunately, the only two commercially available antibodies against phosphorylated SphK2 are not described to work for immunofluorescence staining. One is against the Thr578 residue (https://www.abcam.com/sphk2-phospho-t578-antibody-ab215750.html), identified as phosphorylated downstream of ERK by Sarah Spiegel’s group (Hait et al, J Biol Chem, 2007). The second is designed to recognize specifically the phospho-Thr614 residue (https://www.abcam.com/sphk2-phospho-t614-antibody-ab111948.html), but this site has not been rigorously demonstrated to be phosphorylated downstream of AKT or ERK or to stimulate SphK2 activity. Thus, considering the lack of appropriate tools and considering that we already showed, using opaganib, that the catalytic activity of SphK2 is required for the UFIT pathway, we believe that investigating the phosphorylation status of SphK2 reflecting its activity in flotillin-positive vesicles will be complicated to achieve in a reasonable amount of time and we think that it will not bring a higher value to our present study.

To answer more broadly to the question “Is SphK2 recruited to flotillin phosphorylated?”, we anticipate that it could be the case at least on the Ser419 and Ser420 residues because Nakamura’s group demonstrated that the phosphorylation of these sites favors the nuclear export of SphK2 (Ding G et al, J Biol Chem, 2007). This group developed an antibody against these phospho-sites, potentially working by immunofluorescence. However, as it is unknown whether phosphorylation of these residues influences SphK2 activation status, we do not plan to perform immunofluorescence experiments with this tool (not available commercially) because the results would not address the Reviewer’s question.

1. It should be determined whether the optogenetic system used to induce flotillin oligomerization also induces recruitment and activation of SphK2.

As we already have all the available tools, optogenetic experiments will be performed to answer this point and the results could be added to the fully revised version of our manuscript.

As suggested, we plan to perform experiments in which exogenous S1P will be added to cells with a moderate flotillin expression level to check whether it could recapitulate the effect of flotillin upregulation on AXL expression. Results could be added to the fully revised version of the manuscript.

However, our current results on the localization and the involvement of SphK2 suggest that the generation of S1P involved in the UFIT pathway occurs at the plasma membrane and in late endosomes. Because the exogenous S1P that will be added in the culture medium will not go through the plasma membrane, we anticipate that it could be insufficient to mimic all the mechanisms of the UFIT pathway. Its effect will be limited to the plasma membrane. In addition, these mechanisms are very likely based on a local concentration of S1P in some microdomains (at the plasma membrane and in intracellular membranes) scaffolded by flotillins. It will be very difficult to mimic such local concentration of S1P just by adding S1P to the cells.

We agree that identifying the S1P receptors involved would be of valuable interest for a better characterization of the UFIT pathway. However, we think that this is beyond the scope of our present study. Among the five known S1P receptors, we do not know if any could be involved in membrane remodeling at the plasma membrane to promote endocytosis. To our knowledge, involvement of S1P receptors in endocytosis has never been reported. However, based on the work by Nakamura’s group (Kajimoto et al, Nat Comm, 2013 and Kajimoto et al, J Biol Chem, 2018), the S1P1 and S1P3 receptors are involved in membrane remodeling and cargo sorting from the outer membrane of late endosomes (where flotillins accumulate in our cell models). We could hypothesize that these receptors are influenced by flotillins and are involved in the UFIT pathway. But we think that testing this hypothesis would be the subject of a distinct study.

At the plasma membrane, we totally agree that the effect of S1P could be mediated, as suggested by De Camilli’s group (Shen et al, Nat Cell Biol 2014), by the formation of tubular endocytic structure rich in sphingosine after acute cholesterol extraction. Reciprocally, in our cell models, upregulated flotillins, thanks to their ability to bind to sphingosine (demonstrated by Nichols’ group (Riento et al, PloS ONE, 2018)) and to oligomerize, could create sphingosine-rich membrane regions.

1. There is a commercial antibody for endogenous SphK2 that can be used to validate and substantiate the data with GFP-SphK2. (F1000Res . 2016 Dec 6;5:2825. doi: 10.12688/f1000research.10336.2. eCollection 2016. Validation of commercially available sphingosine kinase 2 antibodies for use in immunoblotting, immunoprecipitation and immunofluorescence)

We thank Reviewer #3 for this suggestion and advice. Being able to detect the localization of endogenous SphK2 in late endosome would be valuable for our study. We already tried with no success with antibodies from Sigma and Cell Signaling Technology (not described to work in immunofluorescence experiments).

We will follow the advice from Reviewer #3 and test the anti-SphK2 antibody from ECM-Biosciences mentioned in the article by Neubauer and Pitson F1000 research, 2016. If we obtain interesting results, they will be included in the revised version of our manuscript.

However, in experiments using SphK2-GFP, we noticed that in live cells, the signal in late endosomes was completely lost after fixation using paraformaldehyde. Similarly, we also observed in live cells that NBD-Sphingosine, added in the culture medium, quickly accumulated in flotillin-positive late endosomes (Additional Figure 7, this data could be added in the fully revised version of the manuscript), but this accumulation was no longer detectable after fixation. Based on these observations, we believe that SphK2 recruitment to flotillin-positive late endosomes is highly labile probably because it mainly involves its interaction with sphingosine molecules that are enriched in these intracellular compartments. This is supported by our observation that addition of opaganib, characterized as a sphingosine competitive inhibitor, displaces SphK2-GFP from flotillin-positive late endosomes in live cells (Figure S7D). In addition, we showed that SphK2-Halo is more recruited in CD63-positive late endosomes in cells overexpressing flotillins (Figure 5E). This could be due to a higher concentration of sphingosine promoted by flotillins (that bind to sphingosine) accumulating in these compartments.

Thus, we will try the immunofluorescence staining of endogenous SphK2 using the recommended antibody, but it might be difficult to detect its presence in flotillin-rich late endosomes in fixed cells. The data could be added in the fully revised version of the manuscript.

ADDITIONAL FIGURE 7 CAN NOT BE ADDED BUT IS AVAILABLE UPON REQUEST

Additional figure 7: Visualization of NBD-sphingosine in flotillin-positive late endosomes. Live HS578T, MDA-MB-231 and MCF10AF1F2 cells expressing Flot1-mCherry were monitored by time lapse spinning disk confocal video-microscopy, 5 min after addition of fluorescent NBD-Sphingosine in the culture medium. On the right are shown still images corresponding to the boxed areas to illustrate the accumulation of NBD-sphingosine in virtually all flotillin-positive endosomes.

Reviewer #3 (Significance (Required)): This is an interesting paper. If the authors confirm the involvement of Sphk2 and mechanism of action of S1P, this would be an important contribution to the field.

Modifications done in the initial revised-version of our manuscript (at the time of the initial response). A full revised version will be provided after all the additional experiments asked by all the Reviewers will be achieved.

Revisions are highlighted in grey in the initial revised-version of the manuscript

1) Figure 1 has been modified and now includes results from a GSEA analysis as recommended by Reviewer #1. The texts of the corresponding legend and of the “Results” and “Methods” sections have been modified accordingly.

1) The Figure 2 version that was inserted in the manuscript was wrong because it was a copy of Figure S2. However, the correct Figure 2 was uploaded to the Review Commons website and accessible for the Reviewers. The correct Figure 2 is now inserted in the manuscript.

2) In the legend to panels C, E, F, J of Figure 2, the sentence: “The histograms show […] with control MCF10AmCh cells calculated from 4 independent experiments” was corrected to “The histograms show […] with control MCF10AmCh cells calculated from at least 4 independent experiments” because data shown in panel J are actually calculated from 8 independent experiments.

3) Figure 6 has been modified with the addition of panel C showing the effect of SphK2 downregulation by siRNA on AXL and ZEB1 level in MDA-MB-231 cells. The text has been modified accordingly.

4) In Figure 3 C, representative western blots have been added as asked by Reviewer #1.

5) In the Supplementary information section, the full clinicopathological characteristics of only 9 patients were indicated, whereas Figure S4 mentioned 37 patients. We corrected this mistake and now provide the characteristics of all patients.

6) In the sentence “Conversely, it induced ZEB 1 and 2 mRNA expression (Figures 1H and S1K) and ZEB1 protein expression (Figures 1I and S1L) (no anti-ZEB2 antibody is available)”, we removed “no anti-ZEB2 antibody is available”.

7) The sentence previously on line 995-997 "to our knowledge this is the first report to describe ligand-independent AXL stabilization..." has been modified to “**To our knowledge this is the first report to describe AXL stabilization following its endocytosis”

8) We are now referring to reference 18 (Planchon et al. J Cell Science, 2018) for the description of the cohort of 527 patients with breast cancer because this was missing.

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

This is an interesting and well written paper describing that upregulated flotillin promotes an endocytic pathway called upregulated flotillins-induced trafficking (UFIT) that mediates AXL endocytosis and allows its stabilization. Consequently, stabilized AXL in these flotillin-positive late endosomes enhances activation of oncogenic signaling pathways that promotes EMT. The authors suggest that Flotillin upregulation-induced AXL stabilization requires the activity of SphK2. However, this latter point is not supported by the data and further studies are needed to support this important conclusion.

Major concerns:

1. Most of the conclusions are based on effects of high concentrations (50 uM) of an ill-defined SphK2 inhibitor. The experiment described in Figure 6C-H need to be confirmed by downregulation of SphK2.
2. Does overexpression of SphK2 reverse the effects of the SphK2 inhibitor? In a similar manner, does overexpression of SphK2 enhance stabilization of AXL?
3. Although the authors suggest recruitment of SphK2 and formation of S1P in UFIT, there are no measurements of S1P. Also, there is no indication that SphK2 is activated despite the fact that ERK and AKT are activated in UFIT and are known to phosphorylate and activate SphK2. Is SphK2 that is recruited to flotillin phosphorylated?
4. It should be determined whether the optogenetic system used to induce flotillin oligomerization also induces recruitment and activation of SphK2.
5. Most importantly, it has not been established that the effects are mediated by S1P. Does addition of S1P enhance stabilization of AXL? Are the effects of S1P mediated by a S1P receptor? If so, which S1P receptor? There are several specific agonists and antagonists of S1PRs that can be utilized to answer this question. It's also possible that the effects of S1P are mediated by intracellular actions as were suggested by the De Camilli group (Nat Cell Biol. 2014 Jul;16(7):652-62).
6. There is a commercial antibody for endogenous SphK2 that can be used to validate and substantiate the data with GFP-SphK2. (F1000Res . 2016 Dec 6;5:2825. doi: 10.12688/f1000research.10336.2. eCollection 2016. Validation of commercially available sphingosine kinase 2 antibodies for use in immunoblotting, immunoprecipitation and immunofluorescence)

Significance

This is an interesting paper. If the authors confirm the involvement of Sphk2 and mechanism of action of S1P, this would be an important contribution to the field.

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

Genest and co-authors present in this paper new fascinating evidence on how intracellular trafficking can modulate oncogenic sigalling.

First of all, they show how overexpression of Flotillin1 and 2 in non-cancerous breast lines can induce a strong reprogramming towards a EMT phenotype. They analyse mRNA and protein expression, intracellular distribution of activated proteins, cell phenotypes to demonstrate a strong activation of oncogenic signalling pathways. They then identify AXL as a key player in this process and show how this protein is stabilised upon Flotillin expression. The authors use an amazing variety of approaches to study the endocytosis and the trafficking of endogenous, GFP-tagged, Halo-tagged and Myc-tagged AXL in different cell lines and their data are strong and very convincing, the images are of very high quality and the analysis rigorous. Their data strongly support the hypothesis that high Flotillin levels triggers AXL endocytosis and accumulation in non-degradative late endosomes where signalling remains active. The authors then show how SphK2 has a key role in AXL stabilisation, it colocalises with Flotillin, AXL and CD63 and its activity (which they block by using inhibitors or siRNA) is necessary for flotillin-induced AXL stabilisation and EMT induction.

The paper is extremely well written, the data flow logically and they are appropriately presented and analysed.

I don't have any major comment and I believe the paper is suitable for publication.

I have only some minor comments/questions:

1) did the authors try to colocalise AXL with endogenous Flotillin in MDA-MB-231 cells? They could use the antibodies used in Fig S1B. Of note, the authors have shown it in luminal tumours in Fig S4C.

2) In Fig6G, it appears that AXL-Flotillin colocalization is lost upon SphK2 inhibition. Is this the case? It could be that the correct lipids are necessary for the formation of Flotillin-positive internalisation domains and this could be very interesting and reinforce the model proposed in the paper.

3) I would remove the sentence on line 995-997 "to our knowledge this is the first report to describe ligand-independent AXL stabilization..." as the cells are not serum starved in all experiments and animal serum can contain variable amounts of the ligand GAS6.

Please note that the authors don't have to necessarily address comments 1-2, their paper is already very rich in convincing data.

Significance

AXL is a major oncogene that promotes EMT in a variety of tumour types. Understanding how its signalling can be triggered by endocytic pathways even in cells that are non-cancerous is very important and of high significance for the cancer field and the trafficking community.

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

The paper by Genest et al. describes the effect of flotillins and sphingosine kinase 2 to stabilize AXL as a mechanism to promote epithelial-mesenchymal transition in breast (cancer) cells. The potential role of vesicles trafficking EMT-promoting proteins is of high interest in the field, also for exploring new opportunities of pharmacological targeting. However, the paper fails to convincingly demonstrate that the proposed mechanism is of real importance to support or promote EMT for the following main reasons:

1) The role of flotillins is studied only by overexpression and in the context of non-cancerous MCF10A cells, while breast cancer cells of epithelial-like origin are not analyzed. This is contrast with the purpose of the paper (see abstract, introduction, patients' data) which is to study tumors and EMT. Effect of shRNAs is also not reported, making it difficult to estimate the importance on the EMT phenotype. Then, alteration of EMT should be concluded also from other non-genetic functional parameters, not just by markers. For instance: was morphology of the cells changed? Was cell migration affected with F1F2?

2) AXL up-regulation is not very strong (2-fold). What is unclear is if the minimal AXL increase due to F1F2 really really provides a significant contribution to the EMT phenotype (as the authors conclude). The siRNA experiment knocks down all AXL, not just the F1F2-induced levels, making it difficult to estimate the real effect of the mechanism proposed. Why didn't the author focus on EphA4 (or to a lesser extent ALK), which showed better regulation?

3) The conclusions of the manuscript are contradicted by the reported clinical data. In Figure S4 the authors clearly observe co-expression of Flotillin 1 and AXL prevalently in luminal breast cancers, which is the subtype known to not be driven by EMT. This evidence already indicates that this (otherwise interesting) mechanism is not relevant to EMT in breast cancer. So, the conclusions are not supported by the data, and the experimental setup and model chosen are not appropriate to generalize the findings to cancer.

Minor (here the most important):

4) The point of the Figure 2 is not clear. Why this part should have such a central role in the story? The entire data presented are not followed up in the rest of the paper. Moreover, in some cases upregulations also questionably significant (like RAS and STAT3 are not even 2 fold). Moreover, the error bars are so small that it seems unrealistic that the plots indicate three independent experiments.

5) More robust statistical analysis should be provided in the Figure 1 to support that EMT is suppressed with F1F2 overexpression. For instance a more standard GSEA on hallmark signatures.

6) In Figure 3 E-Cadherin is rescued with siAXL in the IF but not in the western blot.

7) Some sentences require clarifications. The authors should be more clear on why ZEB2 antibody was not available or what they mean with "Unfortunately the available tools..".

8) Western blot from the CHX experiment should be shown, at least in the supplements. Again, the standard deviation in this experiment is minimal, was this really an average of three independent experiments (and not three western on the same lysates)?

9) All conclusions are derived from one single cells MCF10a. NMuMG cells are shown at the beginning but not used for the rest of the paper. Anyway, if this wants to be a cancer research paper, then cancer cells needs to be used.

10) The methods section contains inconsistent data about patients' samples (9 are indicated, but the Figure S4 features 37). Then, where those other 527 come from?

11) Some figures do not match with the legends or with the description in the text. It has not been easy to review this paper.

Significance

I am a cancer biologist working on EMT.

Referee Cross-commenting

I have nothing to comment on other's reviews.

Transparent Peer Review

---

Download the complete Review Process [PDF] including:

• reviews
• authors' reply
• editorial decisions

</br>

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

Learn more at Review Commons

---

Reply to the reviewers

The authors do not wish to provide a response at this time.

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

Recommendation: Publish after revisions

Duart and colleagues have put forth a manuscript detailing the under-appreciated effect of intrahelical salt bridge formation to the insertion transmembrane helices. Utilizing an in vitro and in vivo applicable construct of vehicle protein leader peptidase (Lep) from Escherichia coli, the authors were able to use glycosylation patterns as a way to quantify the apparent free energy of membrane insertion for multiple transmembrane helices. These results have demonstrated the importance of taking salt bridge formation into account when developing membrane protein prediction tools; however, prior to publication, further analyses would be beneficial for supporting their quantitative conclusions.

Major Comments:

• It would be helpful if the authors detailed their process in deciding which of the 136 potential salt bridge-containing helices were chosen for further investigations.
• Considering the data presented in Fig. 3c, it may be useful to also include charged pair mutations in the i, i+3 positions in the analyses of helix G and helix A, as these positions are the most likely to form salt bridges. This would act as a useful positive control, to see if the mutations would improve the insertion of the sequence.
• Page 14, Line 255: Authors state "The salt bridge contributes approximately ~0.5 kcal/mol to the apparent experimental free energy of membrane insertion." Can this change in apparent free energy be attributed completely to the mutation? Are there any potential interactions between the inserted helix and the natural H2 transmembrane sequence of Lep that could be changing with the various mutations?
• It is promising to see these results in the context of the Lep protein, but the authors should consider the effect these salt bridges may have in the context of the full protein. Creating mutants of Halorhodopsin or calcium ATPase would determine the impact of potential salt bridge disruption on protein folding, which would provide some context on the functional consequences of these mutations.
• Authors have presented a strong argument for the inclusion of potential salt bridge formation in the prediction of transmembrane helices; however, they have not detailed the necessary steps for developing a new system. It would be encouraging to see recommendations on the next steps towards better prediction software.

Minor comments:

• Page 8, Line 118: Authors state "Our results showed a tendency to better insertion when charge pairs were placed in positions (i, i+1; i, i+3; i, i+4) that are permissive with salt bridge formation (Fig. 1c), actually an effect not observed in the predictions (Fig. 1b)." It is important to clarify that this "better" insertion in Fig. 1c is compared to each respective predicted value in Fig. 1b. Currently, it reads as if the authors are suggesting the introduction of the charged pair residues is helping the insertion of the unaltered L4/A15 sequence.
• Figure 1a: Add cytoplasmic and lumen identifiers for clarity.
• Figure 3b: Slashes for "Opp charge" and "Same charge" in the legend appear to be reversed according to the values presented in Table 2.

Significance

This paper increases our understanding of salt bridges in membrane protein structure and function, as performed systematically by a lab with major expertise in this research area.

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

As the title states, Duart et al. have examined the energetic cost of the translocon-assisted insertion of TM helices containing salt bridges into membranes. The paper has three parts: (1) model studies using the methods of Hessa et al. (12,13), (2) statistical analysis of salt bridges in membrane proteins of known 3D structure, and (3) Hessa et al. measurements of selected i,i+4 salt-bridge containing TM helices from halorhodopsin (PDB ID = 3QBG) and calcium ATPase (1SU4). The major overall conclusion is that for i,i+4 salt bridges yield a more favorable insertion free energy of about 0.5 to 0.7 kcal/mol.

The subject of the paper is broadly interesting, but it suffers from several problems that must be addressed before it can be considered seriously for publication. The comments below are described in terms of the three parts.

(1) Figure 1b reports tabular values of predicted Hessa et al. DG values for sequences that contain K, D, or K & D substitutions in a parent L4A15 parent sequence, which has a favorable DG of about -0.5 kcal/mol. For all of the other sequences, DG is predicted to be unfavorable 1.4 kcal/mol to 3.5 kcal/mol. Figure 1c presents triplicate experimental measurements of DG for the sequences in Figure 1b shown as bar graphs. All of the sequences yield unfavorable DG values of about +0.2 kcal/mol except for the parent sequence that has a favorable value close to the predicted value.

There are several problems with these data and their presentation. Fig. 1b should also include the measured DGs with standard deviations in addition to the predicted values. In Fig. 1c, the positive values are plotted on different scale than the sole negative value. This causes the authors to insert a break in the bar representing the sole negative value. The bars are color coded in a mysterious way that is not clearly described in the figure legend. In any case, the measured DG values are all about the same.

A fundamental problem with the measurements is that the method of Hessa et al (12) should have been adhered to rigidly. As those authors noted "The quantification [of DG values]is maximally sensitive for H-segments with DGapp values close to zero (p < 0.5 in Fig. 1d); therefore, for each kind of residue we balanced the contribution from the central residue by varying the number of Leu residues until an H-segment with DG in the range -1.2 to 1.2 kcal/mol was found." The measurements reported Fig. 1b and 1c are far outside the maximum sensitivity range. The Western blots upon which the numbers are based should have been included (perhaps as an appendix).

(2) The statical analysis seems fine and is useful.

(3) Fig.4, halorhodopsin helix G measurements. The table of Fig. 4a should be expanded to include both in vitro (panel e) and in vivo data (panel f). It is not entirely clear where the values given in the table are from, but presumably from the in vitro data (panel e). Fig. 5, calcium ATPase helix A. Comments similar to those regarding Fig. 4 apply. The division of the long helix into a short greasy one and longer one carrying more charges is interesting, but it seems to add little to the main intent of the paper to assess the thermodynamic properties of helices containing salt-bridges.

Overall, the paper would be stronger if it focused mostly on the part (1) experiments to arrive at definitive answer to the energetics of salt-bridge insertion. As it stands, it is a smash up of ideas and experiments.

Significance

Surprisingly,there have been no reported measurements that I am aware that examine the energetic cost of inserting TM helices containing salt bridge into membranes. This paper is a start in that direction.

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:

In the manuscript „Intra-helical salt bridge contribution to membrane protein insertion" the authors investigate the effect of salt bridge formation between positively and negatively charged amino acids on the insertion behavior of α-helical protein segments into the membrane. Generally it is believed that polar or even charged residues prevent stable membrane insertion of α-helical protein segments, but some of these authors had already shown in a previous paper that such residues are more frequent than expected in transmembrane helices. In the current study, the authors investigate in detail the role of intra-helical salt bridge formation on stable membrane insertion. Using an in vitro membrane insertion assay based on the E. coli leader peptidase (Lep) protein, they found better membrane insertion for helical segments with opposite charge pairs placed at positions compatible with intra-helical salt bridge formation (positions i→i+1; i→i+3 and i→i+4). Furthermore, the authors performed a database screen which revealed that oppositely charged residues are overrepresented at these positions. Finally they picked two candidate membrane proteins from the database (Halorhodopsin and calcium ATPase 1) and proved the presence of an intra-helical salt bridge and determined the contribution of the salt bridge to the apparent free energy of membrane insertion (ΔGapp), which was in the range of 0,5-0,7 kcal/mol.

Major comments

1. It seems that the data in Fig. 3b has been mixed up, making it difficult to judge the conclusions. The bars with forward slash seem to represent the "same charge" data and the bars with backward slash seem to represent the "opposite charge" data (exactly contrary to the figure legend). In general the forward and backward slash representation is not easily distinguishable, and for the position i+4 both bars contain a forward slash (making it impossible to discriminate same and opposite charge). Please use filled and unfilled bars instead. Furthermore the bar diagram in Fig. 3a is stacked for opposite and same charge, whereas in Fig. 3b the respective bars are placed next to each other. Additionally the label of the y-axis in Fig. 3c is misleading, as it is not the "Frac. of opp. charged pairs" but the fraction of oppositely charged pairs that form salt bridges.
2. The authors don´t give details no how the log odds ratios and the respective p-values have been determined. Please include this in the Materials and Methods section. What does a p-value of 0.00e+00 mean (see Table 2, Spacing: +3, "All Log odds")?
3. What is the proof that for the isolated helix A from the calcium ATPase 1 the membrane embedded part is identical to the full-length protein? The authors investigated two different helix A peptides, the full-length helix ranging from L49-F78, and one short fragment ranging from L49-A69 containing the more hydrophilic N-terminal region, which is the membrane-embedded region in the full-length protein. The authors state: "In contrast, when only the membrane-embedded sequence was included, the Lep chimera was mainly doubly-glycosylated (Fig. 5c, lane 3), suggesting that helix A is properly inserted when the full helical sequence is present." In my opinion this conclusion cannot be drawn from the data presented. The authors used an isolated helical segment, so in my opinion it is much more likely that the isolated full-length helix inserted via its hydrophobic C-terminal part (L60-F78) into the membrane. The authors themselves state in their manuscript: "It has been previously shown that the position in the membrane of TM helices in protein folded structures does not always correspond to the thermodynamically favored positions in the membrane of the isolated helices." Also the i→i+5 mutant points into that direction, because the effect of disturbing the intra-helical salt bridge for the helix A is much less pronounced compared to the similar data in Fig. 4f for the Halorhodopsin protein. In my opinion this shows that most probably only one charged residue (R63) is embedded inside the membrane (with a membrane embedded part of L60-F78).

Minor comments:

1. line 151: ",see Figure 2)" Typo: Bracket missing.
2. line 172: "Other known structural features can also be hinted at, including aromatic ring stacking by His-Trp pair [20] at i→i+6." Please give some more examples of important structural features of membrane proteins, which can be seen in your analysis (e.g. I think that also the glycine zipper can be seen in the i→i+4 data set).
3. line 255: "The salt bridge contributes approximately ~0,5 kcal/mol to the apparent experimental free energy of membrane insertion." Please explain that this value was derived from the comparison of the ΔGexp between the wt and the i→i+5 mutant. Please comment also on the large difference between the predicted (ΔGpred) and the experimental values (ΔGexp), even if no salt-bridge is involved (e.g. for the DD mutant).
4. line 348: "Asp-Lys pairs at position i, i+4 and Glu-Lys pairs at position i→ i+3 are the most prevalent as seen previously in Figure 2. They are both among the most prevalent oppositely charged pairs and the charged pairs that form the highest number of salt bridges in membrane protein structures. This is in stark contrast to Glu-Arg pair at position i→ i+1 that although as frequent in pairs as Asp-Lys and Glu-Lys at positions i→i+4 and i→i+3 respectively, only form salt bridges in one-fourth of the cases." Fig. 2 shows that each charge pair has a different prevalence depending on the order (e.g. Asp-Lys and Lys-Asp pairs). I think for this statement the sum of both prevalences should be taken into account, and as the sum is not easy discernible from Fig. 2, it would help to include a table containing the sums. Furthermore, it would be good to refer also to Fig. 3, which also contains a part of the discussed data.
5. line 402: "c-myc tag (Glu-Gln-Lys-Leu-Ile-Ser-Glu-Glu-Asp-Leu, EQKLISEEDL) was added in Ct in hanging with de PCR primer before cloning." Please revise the sentence and I think the one letter code for the c-myc tag is sufficient (please correct this also in line 428).
6. line 420: "A region's total ΔG is the sum of these individual scores weighted on where in the region the residue, a residue in the middle of the helix has a higher weight than residues at the ends." Please revise the sentence, the meaning is unclear.
7. line 436: "Total protein was quantified and equal amounts of protein submitted to Endo H treatment or mock-treated, followed by SDS-PAGE analysis and transferred into a PVDF transfer membrane (ThermoFisher Scientific)." Please revise the sentence.
8. line 498: "Topological files with sequence and membrane topology are created with the help of the RCSB secondary structure file and only membranes annotated as pure α-helices were retained." I assume that the description contains a typo (membranes annotated as pure α-helices?)
9. line 507: typo "..., but we did not clustered the proteins" 14: line 560: "The individual value of each experiment in represented by a solid dot being represented as a green square the experimental ΔG value for the L4/A15 sequence from [2]." Please revise the sentence. 15: line 562: "The wt and simple mutants are shown in white bars." Typo: single mutants 16: line 563: "Charges at compatible distances with salt bridge formation (i→i+1; i→i+3; and i→i+4) are shown in yellow. Not compatible distances with salt bridge formation (i→i+2; and i→i+5) are shown in dark gray. Compatible distances but not compatible amino acid pair (i, i+4 DD pair) is shown in clear gray." The given colors don´t match with the figure (i→i+1 = brown; i→i+3 = orange; i→i+4 = yellow and i→i+4 DD pair = white) 17: line 597: "The different monomers are shown in transparent blue, purple and indigo." The different colors are hardly distinguishable in the figure. 18: Figure 4a: The table could be simplified. I think the column "charges" can be removed, as it contains not really charges and the names of the peptides already contain the same information. The column "Å" contains only a value for the wt (and not for the DK i,i+5 mutant) and as the distance for the wt is also given in Fig. 4g, this column can be also removed.
10. Fig. 4f: The marker lane is hardly visible (completely dark lane)
11. Fig. 5b: The column "Å" contains only values for the wt sequences (long and short). See also comment 18.
12. Fig 5d: Why is in the SDS gel a mass shift between the wt and the i→i+5 mutant visible, even though the peptide mass is equal.
13. There are several changes of font type or format changes (e.g. line 210-214). Please correct this.

Significance

As a structural biologist with a focus on membrane-proteins, I understand that the study is concerned on intra-helical salt bridges, but the implications of inter-helical salt bridges should also be discussed, at least in the introduction or outlook. The authors propose that their results are important for the improvement of membrane protein topology prediction methods, so for this aim it is also necessary to take any potential inter-helical salt bridges into account. In this context, it would be relevant to point point out that there even exist extended rows of salt bridges between transmembrane segments (charge-zippers), which serve an important structural element in several membrane proteins.

The article is well written and most of the conclusions drawn from the experimental results are convincing. I agree with the authors that their results are relevant for future improvement of membrane protein topology prediction software, which so far does not take the possibility of salt bridge formation into account. Therefore, I recommend publication after clarification/revision of the abovementioned points.

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

Learn more at Review Commons

---

Reply to the reviewers

We thank the three reviewers for their helpful and valuable comments. We plan to address their criticisms in a revised manuscript and hope that our manuscript will then be significantly improved.

Reviewer #1 (Evidence, reproducibility and clarity (Required)):

The authors have presented a very interesting and compelling set of data regarding the impact of conditional deletion of the only known pathway allowing the uptake of pyruvate into mitochondria. The paper comprises two interwoven stories that are both important. The first is the remarkable finding that the majority of excitatory neurons in the cortex (i.e. those under the influence of the CaMKII promoter) show remarkable metabolic flexibility as they tolerate elimination of pyruvate oxidation, considered the major supplier of ATP in neurons. The data on this seem clear although the authors did not delve into the potential mechanisms of metabolic compensation that likely occurs. Instead they examined whether there was some mal-adaptive compensation and they found clear evidence of this: in the absence of MPC activity the mice are much more prone to epileptic seizures, unveiled experimentally by relatively standard protocols (kindling). The authors present largely very convincing evidence that this mal-adaptive compensation in turn ends up decreasing the activity of KV7.2/7.3 channels whose job is normally to limit runaway repetitive firing by mediating an hyperpolarizing K+ efflux following an action potential. This channel, put on the map as it was one of the downstream targets modulated by cholinergic metabotropic activation, is also know known to be controlled by Calmodulin and therefore cytosolic Ca levels. Overall, I think at its core this manuscript is interesting and important. There however several weaknesses, I fear, will diminish the impact on the eventual readership. If these points can be addressed, it will strengthen the longevity of these findings:

1) It is puzzling why the authors resorted to using shRNA-mediated KD of MPC1 for some of the in vitro studies when they have gone to the trouble of making a floxed CRE-dependent mouse. Primary cells (e.g. Fig 1) or organotypic cultures (Fig. 6) from these mice would have made a more consistent set of starting conditions to compare data across the manuscript. As there viruses expressing the CRE recombinase are widely available this could have been used on mice simply harboring the floxed gene it they are worried about waiting for the expression of the CaMKII promoter for in-vitro conditions.

This is indeed a good point. Indeed initially, when we started these experiments, we tried to use viruses expressing the CRE recombinase in cultured neurons from mice harboring the floxed gene as proposed by the reviewer. However, for reasons that we do not fully understand, the use of AAVs or lentiviruses expressing the CRE was found to be deleterious for the cultured neurons. In view of this toxicity we tried using TAT-CRE recombinase, a recombinant cell-permeant fusion recombinase, which we added directly to the medium. However, this strategy proved to be poorly efficient. We finally used cultures of Cre-floxed neurons in which we tried to knockout MPC1 gene using 4-hydroxytamoxifen in the culture medium. However, we did not obtain satisfying results because, as previously reported, cortical neurons grow poorly in the presence of 4-hydroxytamoxifen (Nichols et al., Cell Death and Disease, 2018. https://doi.org/10.1038/s41419-018-0607-9). For these reasons we turned to the shRNA strategy and to the use of 3 small molecule inhibitors of the MPC each with different chemical structures. Both the RNA interference and the pharmacological approaches gave similar results, reinforcing our confidence in the specificity of the results, and the unlikelihood of off-target effects.

2) The data in Figure 5 gets a little less convincing as using extracellular glutamate to drive Ca elevations is so non-physiological that the results might really be distorted by the participation of something irrelevant to the story, even though it supports the overall interpretation for a role of Ca/CaM in the control of the channel. Similarly, the use of RU360 should be done with caution. The drug, although a useful antagonist of MCU in purified mitochondria, is famously finicky with respect to its ability to cross membranes and could well have off target impact. A much cleaner experiment would be to suppress the expression of MCU via KD. Presumably in the MPC-deficient neurons, this would have minimal impact on Ca signals. Given the frequent ambiguity associated with interpreting pharmacological results, coupled to the central importance of this finding in interpreting the entire paper, I think carrying out experiments with molecular genetic manipulation of MCU is warranted.

The main point of this figure is to study the capacity of MPC1 KO neurons to handle intracellular calcium increase and to regulate calcium homeostasis. To this end, we used strategies described to acutely increase cytosolic calcium, either through membrane depolarization with KCl (Rienecker et al., ASN Neuro. 2020. https://doi.org/10.1177/1759091420974807) or through activation of glutamate receptors using glutamate (For example see Wong, Vis Neurosci, 1995 : DOI: 10.1017/s0952523800009469). It is important to mention that the concentration of glutamate used in our experiments (10 microM for 2 min) is well below the concentration normally used to induce excitotoxicity (100-500 microM for 30min). The fact that both stimulations provided similar results and clearly indicated a defect in the clearance of cytosolic calcium in MPC-deficent neurons.

Regarding the concern with RU360, we are aware of the problems with plasma membrane permeability associated with this compound, and for this reason we included a membrane permeabilizer (0.02% pluronic acid) to facilitate its entry into the cell. This was indicated in the Material and Methods section (line 585) as well as in the figure legend (line 948). In order to clarify this methodology, we will add this information in the main text. It should be noted that this concern would not apply to the electrophysiogical experiments, since in this case the compound was injected directly into the cell. We would like to add that we chose to inhibit the MCU using a chemical inhibitor rather than a shRNA because of the well known difficulty in obtaining a complete loss of function of the MCU using RNA interference (Nichols et al., Cell Death and Disease, 2018. https://doi.org/10.1038/s41419-018-0607-9). Nevertheless, as recommended by the reviewer, we will attempt to downregulate the expression of MCU using shRNA.

3) The authors have not really made clear in this paper whether the ability to suppress the phenotype of the MPC deficiency with ketones is really related to a providing TCA cycle support or instead a pharmacological impact on non-TCA related targets (such as the Kv7.2/7.3 channels). Presumably the use of other ketones might circumvent this. The action of ketone bodies has been a topic of considerable interest in neuroscience, given the clinical relevance for childhood epilepsies. Previous studies for example have argued for direct inhibition of the vesicular glutamate transporter (Juge et al. Neuron 2010). The use of other ketones (acetoacetate) would narrow down the interpretations of the data.

Our results point to 2 two possible mechanisms of ketone bodies: i) providing acetyl-CoA to the Krebs cycle, thereby stimulating OXPHOS and ii) direct action of 3-beta hydroxybutyrate on the activity of Kv7/7.3 channels. The reviewer is asking whether, in addition to 3-beta hydroxybutyrate, other ketone bodies, acetone or acetoacetate, may display antiepileptic activity, which would probably indicate that providing substrates to the TCA cycle is sufficient to prevent neuron-intrinsic hyperactivity and seizures. We agree that this in an interesting question and we will now test the effect of acetoacetate on PTZ-induced seizures in MPC KO mice.

**other**

1) In vitro - scramble controls only serve to demonstrate there is no general effect of treating cells with shRNAs, but do not address if there is an off-target effect. The most convincing thing here would be to have an shRNA-insensitive variant that rescues.

We have used 2 different shRNAs and 3 chemically unrelated inhibitors of the MPC and in all cases we obtained similar results. Therefore, we think that it is unlikely that the effects we observe are due to an off-target activity. The experiment proposed by the reviewer is interesting but extremely difficult. The idea would be to reintroduce a shRNA-insensitive MPC1 into MPC1-deficient neurons treated with shRNA. This is difficult as it is known that the expression level of MPC1 needs to be matched to that of MPC2, otherwise it leads to depolarization of the mitochondria. Obtaining the right level of MPC1 would be extremely difficult to achieve in practice.

2) Does rescuing CaMK binding to KCNQ channels rescue the phenotypes?

The question raised by the Reviewer implies that CaM is not constitutively bound to KCNQ channels, which is a matter of debate. As we pointed out in the discussion, ‘Intracellular calcium decreases CaM-mediated KCNQ channel activity (32, 36) by detaching CaM from the channel or by inducing changes in configuration of the calmodulin-KCNQ channel complex (36).’ The CaM-KCNQ tethering is also described in a review by Alaimo and Villaroel, 2018 (doi:10.3390/biom80300579): ‘[…] CaM was first defined as an integral subunit constitutively tethered to the C-terminal region of Kv7.2/3 channels since Kv7.2 mutants that were deficient in CaM binding were unable to generate measurable currents [5,21]. However, this model has been questioned since Kv7.2 channels, carrying a hB mutation [40] or Kv7.4 hA mutated channels [41] that do not bind CaM, can still reach the plasma membrane and are functional.’

When considering to manipulate CaM binding to KCNQ, it should also be considered that previous studies on this matter have mainly worked with heterologous systems and through genetic manipulations of CaM (by expression of a dominant negative or by overexpression of CaM) or of the KCNQ binding motif.

Based on both theoretical and practical issues, we, thus, believe that it is not feasible to implement a straightforward approach that would be compatible with our mouse model.

An alternative, indirect approach, as indicated by Reviewer #3, would be to test the effect of Ca2+ chelators. Although this is likely to introduce confounding effects through the inhibition of other Ca2+-dependent channels, we propose to focus on trying this option and assess whether a XE991-sensitive component will be unmasked in MPC1 deficient cells.

3) As the authors imply that BHB activates KCNQ channels, showing this directly in their prep would provide some convincing data. If this is true, why doesn't BHB increase firing rate of WT neurons?

Activation of KCNQ channels is expected to reduce (not increase) neuronal firing. When exposed to BHB, we indeed found that WT cells also show a trend towards decreased excitability (p=0.08). We will report this trend in the revised figure 5F. Given that KCNQ channels are already available to be recruited upon repetitive firing in WT cells (to a larger extent as compared to KO, as indicated by our data with XE991) it is conceivable that a further potentiating effect of BHB at the concentration used for ex vivo recordings (2 mM) will be limited.

4) How does the anti-epileptic effects of ketones in this study relate to previous reports of regulation of KATP channels? One of main concerns is that ketones might have a parallel anti-epileptic effect in the MPC1 KO mice that is unrelated to the mechanism proposed here.

The ketogenic diet is likely to exert several effects including disruption of glutamatergic synaptic transmission, inhibition of glycolysis, and activation of ATP-sensitive potassium channels as pointed out by the reviewer. We do not exclude that inhibition of the MPC could also have an impact on the KATP channels and we are currently exploring this possibility. However, such work to dissect the potential implication of the KATP channels would go well beyond the scope of the present paper. Nevertheless, we will plan to certainly raise this important possibility in the discussion.

**Minor comments:**

1- What is the MPC1 KO efficiency in CaMK neurons? The western blot in 2c is from the whole cortex and therefore does not show that.

This is indeed a good comment, however, please note that the estimation of MPC1 KO efficiency has also been evaluated in synaptosomes isolated from MPC1 KO cortices. These structures are mainly isolated from neurons (Carlin et al., JCB, 1980. 10.1083/jcb.86.3.831). As shown in figure 2C, these synaptosomes are massively enriched for CamKII and contain less astrocytic marker GFAP in comparison to the whole cortex. The amount of MPC1 in the synaptosomes prepared from the KO animals is strongly decreased. Nevertheless, as recommended by the reviewer, we plan to quantify the efficiency of the KO by performing a double immunostaining for MPC1 and a specific marker for neurons.

2- Mitochondrial Ca2+ levels are not measured directly, for which there are many tools. This is needed to demonstrate definitively that there is a defect in Ca2+ handling."

The reviewer raised an important point and we plan to monitor the levels of mitochondrial calcium in MPC-deficient neurons using the mito-Aequorin, a luminescent quantitative probe targeted to mitochondria (Granatiero et al., Cold Spring Harb. Protoc. 2014. 10.1101/pdb.top066118)

Reviewer #1 (Significance (Required)):

see above.

**Referee Cross-commenting**

It seems we are in reasonable agreement about the pros & cons of the manuscript. I agree that alternative approaches to RU360 are warranted.

Reviewer #2 (Evidence, reproducibility and clarity (Required)):

De la Rossa and colleagues examined the consequences of conditionally knocking out MPC1,a subunit of the mitochondrial pyruvate carrier. They found that despite decreased levels of oxidative phosphorylation in excitatory neurons, phenotypically these conditional knockout mice were normal at rest. However, when challenged by inhibition of GABA neurotransmission, these animals developed severe seizure activity and expired. These authors then showed that neurons with an absence of MPC1 were hyperexcitable in part through abnormal calcium homeostasis, which was associated with a reduction in M-type inhibitory potassium channel activity. Intriguingly, the ketogenic diet and the major ketone body beta-hydroxybutyrate were able to reverse these changes.

This is a carefully conducted research study that reveals cell type-specific alterations of MPC1 deletion and functional consequences. The study design was logical and involved an exhaustive array of methodologies. The manuscript was generally well written and organized, and there are no major concerns. This study shows a direct causal relationship between impaired bioenergetics at the level of mitochondrial, and subsequent behavioral seizures, and is perhaps the most direct demonstration to date that an intrinsic disturbance of metabolic function can result in seizure activity (through changes in calcium regulation and impairment of ion channel activity). This will be an important contribution to the scientific literature.

**MINOR:**

1. Page 4, line 86: Would recommend changing "paroxystic" to "paroxysmal" (the latter which is a more recognized term). We will make the change.

Page 5, line 124: recommend including the concentration of beta-hydroxybutyrate used when first mentioned. In general, concentration and dose information were difficult to find, as well as route of administration (for kainate, page 7, line 175). This type of information was not conveniently presented.

We will follow the reviewer’s recommendation.

Page 5, line 128: "both overcomed" is awkward. Would recommend using "both reversed".

We fully agree and will make the change in the revised manuscript.

Page 8, line 193: the authors probably meant "astro-MPC1-WT mice", not "neuro-MPC1-WT mice".

Thank you for the acurate look. This will be changed.

Page 12, lines 280-282: the authors might want to mention that chronic exposure of BHB might reduce the hyperexcitability of neuro-MPC1-KO mice.

This point could indeed be discussed.

Please review entire manuscript and use consistent tense. For example, on page 13, line 309, to maintain the past tense, it should read "We first assessed whether..."

Thanks for the recommendation.

Page 13, line 318: the authors used 10 mM BHB when examining calcium levels, but they earlier used 2 mM. They need to explain why they used a different concentration; and 2 vs 10 mM are quite different.

The reviewer makes a valid point. When we performed the in vitro experiments, we used 10 mM BHB, which is slightly higher than the amount of ketone bodies measured in the blood of mice fed on a ketogenic diet for 2 days (Supplementary figure 4). This concentration of BHB has also been used by others (see for example: Izumi et al., JCI 1998, 101:1121-1132). Later on, when electrophysiology experiments were performed, the person in charge of these experiments followed a previously published protocol by Yellen and colleagues, in which the authors had used 2 mM BHB (Ma et al., J. Neurosci 2007,27: 3618-3625). This explains the differences between the concentrations used in vitro and in vivo.

Page 13, line 323: it is not necessary to say "...interesting study published during the preparation of this manuscript." This phrase should be deleted, and the relevant reference simply cited.

We will follow the reviewer’s recommendation.

The authors need to explain more clearly in the beginning what exactly is meant by "paradoxical" hyperactivity. They provide greater meaning later in the manuscript, but this should be clarified at the outset.

We will explain why we used this adjective in the beginning as recommended by the reviewer.

Reviewer #2 (Significance (Required)):

This is a very important study to show how primary defects in metabolism (i.e., disruption of the mitochondrial pyruvate carrier) can lead to epilepsy. Moreover, it details a primary mechanism that connects cellular bioenergetics to membrane excitability (through changes in calcium homeostasis and M-current function).

This is a well-conducted study that utilizes a multiplicity of experimental tools to link biochemistry with seizure activity. This type of study is not so readily done, and strengthens the notion that primary defects in metabolism can result in epileptic seizures.

This study is unique and attempts successfully to be more than just correlational. Hence it is a valuable contribution to the field.

The audience will likely consist of metabolic geneticists, neurologists/epileptologists, and neuroscientists. This is a beautiful study that runs the translational spectrum from biochemistry to behavior.

My expertise is in the field of translational epilepsy research, with a focus on mitochondria, metabolism, the ketogenic diet and ketone bodies. Thus, I am qualified to critically evaluate the entire manuscript.

**Referee Cross-commenting**

After reading comments and reviewing the manuscript again, would agree with Reviewer #1, and would change recommendation to MAJOR REVISION.

Reviewer #3 (Evidence, reproducibility and clarity (Required)):

This manuscript tests the genetic requirement of the mitochondrial pyruvate carrier (MPC) in regulation of neuronal excitability. The authors find that MPC deficiency in glutamatergic neurons is associated with aerobic glycolysis, inhibition of the M-type K channels, and neuronal hyperexcitability that manifests in increased sensitivity to chemical pro-convulsants without changes in resting conditions. Alterations in Ca homeostasis in MPC-deficient neurons is consistent with reduced mitochondrial membrane potential and attendant diminution of mitochondrial calcium buffering capacity. The authors further show that the effect of MPC deficiency can be phenocopied by treatment of wild type neurons with a chemical inhibitor of the mitochondrial Ca uniporter (MCU). Based on these data, it is proposed that reduced mitochondrial Ca uptake causes neuronal hyperexcitability in the absence of MPC. Overall, the manuscript presents detailed electrophysiology and in vivo seizure studies. However, there is significant disconnect between the actual data in Fig. 6 and the authors' conclusions/proposed mechanism. In particular, the evidence for the role of Ca in the hyperexcitability due to MPC deficiency is the weak link in the authors' argument.

1. The studies linking reduced mitochondrial Ca uptake to hyperexcitability in MPC-deficient neurons (Fig. 6) have several limitations that significantly weaken the paper: 1a. The Ca measurements in cortical neurons (Fig. 6A-F) are performed under conditions (glutamate/KCl) that are fundamentally different from those used in electrophysiology of CA1 pyramidal neurons (Fig. 6G-N). The electrophysiological excitation is much briefer and less extreme than the chemical stimulation, and it is not clear that the Ca dysregulation occurs at the earliest times (see Fig. 6A).

This point was also raised by reviewer 1. Please see our response to point 2.

1b. The conclusion that MCU is functionally responsible for MPC's effect on neuronal excitability is singularly based on the use of RU360 as a chemical inhibitor of MCU but the specificity of this reagent is questionable. Evidence for a cause and effect relationship that directly implicates altered MCU/mitochondrial Ca buffering has not been provided.

This accurate point was also raised by the reviewer 1. Please see our response to point 2 for a complete response. We will downregulate expression of MCU using shRNAs. We will also measure the mitochondrial calcium level in the hope of better understanding whether the phenotype of the MPC-deficient mice is due to impaired mitochondrial calcium uptake.

1c. There is a large variation in the effect of 10 uM RU360 on firing frequency, comparing Fig. 6H and N (blue traces), including the shape of the traces and values at ramp number 6. This calls into question the reliability of the comparisons in each separate figure.

Data presented in each single graph in the main Figures were obtained from groups of littermates through recordings conducted in consecutive days. Some caution is warranted when comparing data between different figures (i.e. between different experimental series), as several factors may contribute to inter-experiment variability, including variability between different batches of animals. However, the difference pointed out by the reviewer regarding the values of cell firing reported in Fig. 6H and N is only apparent. When applying depolarizations with ramps of 5s, a fair amount of WT cells infused with RU-360 show high instantaneous firing frequency, especially for the last ramps that steeply reach high current levels. This leads to accommodation/inactivation of the action potential towards the end of the ramps, as shown in the example trace in Fig 6G. As a result, the current-frequency plot deviates from linearity, as it is the case in Fig 6H (blue trace) and, even more evidently, in Fig 6N. We have now reanalyzed the same recordings from WT cells infused with 10 µM RU-360 and measured the firing frequency in response to a square depolarizing step (250 pA) of 0.5 or 1 second. No difference was found between the firing frequencies of the cells from Fig 6H and Fig. 6N (group 1 and group 2, respectively, in the figure below). Although the ramps may lead to some distortion for higher stimulation levels, we have decided to show results from ramps consistently throughout the main figures because this protocol with continuously increasing currents allows us to measure more precisely the rheobase and the firing threshold (as opposed to the stepwise increments of a square stimulation).

1d. The calcium > PIP2 > M-type K+ channel axis is well established but has not been fully explored in the context of MPC deficiency. The use of a calcium chelator will likely be informative in this context, and would be better evidence for a role of Ca in the MPC effects.

Although the use of a Ca2+ chelator such as BAPTA is likely to introduce confounding effects through the inhibition of other Ca2+-dependent channels, we will try this option and assess whether a XE991-sensitive component will be unmasked in MPC deficient cells.

1e. The ability of BHB to rescue various parameters in this and other figures in the paper is interesting but does not directly speak to the specific mechanism as to how MPC deficiency affects neuronal excitability. BHB's effect is consistent with the metabolic flexibility of neurons when the TCA cycle cannot be fueled by glucose/pyruvate (as in GLUT1 or MPC deficiency).

The mechanism we propose to explain the hyperexcitability of MPC-deficient neurons relies on the low mitochondrial membrane potential and their decreased capacity to buffer calcium. Based on our data, we propose that calcium accumulation in the cytosol disrupts the CaM-KCNQ interaction leading to hyperexcitability. Indeed, BHB could act in two possible (and parallel) ways. 1: directly on the M-type channels, 2. on mitochondria by providing acetylCoA to the TCA cycle. The use of an alternative ketone body will be informative in disentangling these two possibilities.

The manuscript (and the field) will benefit from a more scholarly discussion and integration of published literature:

2a. The published studies on the outcome of pharmacologic MPC inhibition in neurons (Ref 18, Divakaruni et al.) are not only consistent with the bioenergetic effect in Fig. 1, but more importantly, show that interference with MPC does not lead to broad deficiencies in energy metabolism but rather remodel fuel utilization patterns to alternative substrates that feed the TCA cycle (BHB, leucine, etc). For this reason, terms such as "mitochondrial dysfunction" and "OXPHOS deficiency" used throughout the manuscript to describe MPC deficiency are vague and imprecise. In addition, this metabolic flexibility may explain lack of defects under resting conditions. In light of these considerations, the argument as to whether aerobic glycolysis in MPC-deficient neurons explains the lack of phenotype in resting conditions (p 17) seems one-sided. Overall, the studies in ref 18 are relevant to the current manuscript and should be better integrated in the discussion.

We fully agree with the possibility that the rewiring of cell metabolism in MPC-deficient neurons in the presence of leucine, BHB and other metabolites could explain the lack of phenotype in resting conditions. We thank the reviewer for this highly relevant comment which we will include in the revised discussion.

2b. Several references are cited to describe the role of OXPHOS vis-à-vis aerobic glycolysis in neuronal function. At times, however, the authors' statements are not consistent with what these papers actually show (or do not show). For example, see the use of refs 6 and 44 on p17 of the discussion, where the authors state that aerobic glycolysis uncoupled from OXPHOS is sufficient to provide ATP for normal neurotransmission, but this does not mean OXPHOS is not needed.

We agree that these references are not appropriate here and they will be removed.

2c. Although the XE991 experiments support an important role for the M-type channels in the altered excitability with deficiency, it is not clear that the proposed mechanism can explain all of the electrophysiological differences, particularly those resting properties that are measured without a Ca challenge to the neurons. It would be good to discuss other possible mechanisms that could affect neuronal excitability.

Our results point to M-type channels as important players in the phenotype of the MPC-deficient mice. Previous reports indicate that inhibition of this channel by XE991 can modulate input resistance, membrane potential and firing threshold of pyramidal cells (e.g. Shah et al, 2018, doi/10.1073/pnas.0802805105; Hu et al. 2007, DOI:10.1523/JNEUROSCI.4463-06.2007; Petrovic et al., 2012, doi:10.1371/journal.pone.0030402). We also found that XE991 induced a shift towards more negative potentials in the firing threshold of WT cells, but not in MPC1 deficient cells (-3.3±0.6 vs. -0.4±1.0, n=9, 8, p=0.027). However, we agree with the reviewer that the phenotype is probably highly complex and that additional mechanisms may contribute to modulate the intrinsic excitability of MPC-deficient neurons. One such mechanism could be closure of KATP channels, which we are currently investigating. This will be discussed.

Reviewer #3 (Significance (Required)):

The significance of the advance: The studies provide genetic evidence for the role of MPC in neuronal excitability.

The work in the context of existing literature: Please see specific comments above under point 2 regarding the need for a scholarly discussion and integration of existing literature.

Audience that might be interested: mitochondrial bioenergetics and metabolism and metabolic control of neuronal excitation.

Keywords describing expertise: metabolism, mitochondria and electrophysiology.

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

This manuscript tests the genetic requirement of the mitochondrial pyruvate carrier (MPC) in regulation of neuronal excitability. The authors find that MPC deficiency in glutamatergic neurons is associated with aerobic glycolysis, inhibition of the M-type K channels, and neuronal hyperexcitability that manifests in increased sensitivity to chemical pro-convulsants without changes in resting conditions. Alterations in Ca homeostasis in MPC-deficient neurons is consistent with reduced mitochondrial membrane potential and attendant diminution of mitochondrial calcium buffering capacity. The authors further show that the effect of MPC deficiency can be phenocopied by treatment of wild type neurons with a chemical inhibitor of the mitochondrial Ca uniporter (MCU). Based on these data, it is proposed that reduced mitochondrial Ca uptake causes neuronal hyperexcitability in the absence of MPC. Overall, the manuscript presents detailed electrophysiology and in vivo seizure studies. However, there is significant disconnect between the actual data in Fig. 6 and the authors' conclusions/proposed mechanism. In particular, the evidence for the role of Ca in the hyperexcitability due to MPC deficiency is the weak link in the authors' argument.

1. The studies linking reduced mitochondrial Ca uptake to hyperexcitability in MPC-deficient neurons (Fig. 6) have several limitations that significantly weaken the paper:

1a. The Ca measurements in cortical neurons (Fig. 6A-F) are performed under conditions (glutamate/KCl) that are fundamentally different from those used in electrophysiology of CA1 pyramidal neurons (Fig. 6G-N). The electrophysiological excitation is much briefer and less extreme than the chemical stimulation, and it is not clear that the Ca dysregulation occurs at the earliest times (see Fig. 6A).

1b. The conclusion that MCU is functionally responsible for MPC's effect on neuronal excitability is singularly based on the use of RU360 as a chemical inhibitor of MCU but the specificity of this reagent is questionable. Evidence for a cause and effect relationship that directly implicates altered MCU/mitochondrial Ca buffering has not been provided.

1c. There is a large variation in the effect of 10 uM RU360 on firing frequency, comparing Fig. 6H and N (blue traces), including the shape of the traces and values at ramp number 6. This calls into question the reliability of the comparisons in each separate figure.

1d. The calcium > PIP2 > M-type K+ channel axis is well established but has not been fully explored in the context of MPC deficiency. The use of a calcium chelator will likely be informative in this context, and would be better evidence for a role of Ca in the MPC effects.

1e. The ability of BHB to rescue various parameters in this and other figures in the paper is interesting but does not directly speak to the specific mechanism as to how MPC deficiency affects neuronal excitability. BHB's effect is consistent with the metabolic flexibility of neurons when the TCA cycle cannot be fueled by glucose/pyruvate (as in GLUT1 or MPC deficiency).

1. The manuscript (and the field) will benefit from a more scholarly discussion and integration of published literature:

2a. The published studies on the outcome of pharmacologic MPC inhibition in neurons (Ref 18, Divakaruni et al.) are not only consistent with the bioenergetic effect in Fig. 1, but more importantly, show that interference with MPC does not lead to broad deficiencies in energy metabolism but rather remodel fuel utilization patterns to alternative substrates that feed the TCA cycle (BHB, leucine, etc). For this reason, terms such as "mitochondrial dysfunction" and "OXPHOS deficiency" used throughout the manuscript to describe MPC deficiency are vague and imprecise. In addition, this metabolic flexibility may explain lack of defects under resting conditions. In light of these considerations, the argument as to whether aerobic glycolysis in MPC-deficient neurons explains the lack of phenotype in resting conditions (p 17) seems one-sided. Overall, the studies in ref 18 are relevant to the current manuscript and should be better integrated in the discussion.

2b. Several references are cited to describe the role of OXPHOS vis-à-vis aerobic glycolysis in neuronal function. At times, however, the authors' statements are not consistent with what these papers actually show (or do not show). For example, see the use of refs 6 and 44 on p17 of the discussion, where the authors state that aerobic glycolysis uncoupled from OXPHOS is sufficient to provide ATP for normal neurotransmission, but this does not mean OXPHOS is not needed.

2c. Although the XE991 experiments support an important role for the M-type channels in the altered excitability with deficiency, it is not clear that the proposed mechanism can explain all of the electrophysiological differences, particularly those resting properties that are measured without a Ca challenge to the neurons. It would be good to discuss other possible mechanisms that could affect neuronal excitability.

Significance

The significance of the advance: The studies provide genetic evidence for the role of MPC in neuronal excitability.

The work in the context of existing literature: Please see specific comments above under point 2 regarding the need for a scholarly discussion and integration of existing literature. Audience that might be interested: mitochondrial bioenergetics and metabolism and metabolic control of neuronal excitation.

Keywords describing expertise: metabolism, mitochondria and electrophysiology.

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

De la Rossa and colleagues examined the consequences of conditionally knocking out MPC1,a subunit of the mitochondrial pyruvate carrier. They found that despite decreased levels of oxidative phosphorylation in excitatory neurons, phenotypically these conditional knockout mice were normal at rest. However, when challenged by inhibition of GABA neurotransmission, these animals developed severe seizure activity and expired. These authors then showed that neurons with an absence of MPC1 were hyperexcitable in part through abnormal calcium homeostasis, which was associated with a reduction in M-type inhibitory potassium channel activity. Intriguingly, the ketogenic diet and the major ketone body beta-hydroxybutyrate were able to reverse these changes.

This is a carefully conducted research study that reveals cell type-specific alterations of MPC1 deletion and functional consequences. The study design was logical and involved an exhaustive array of methodologies. The manuscript was generally well written and organized, and there are no major concerns. This study shows a direct causal relationship between impaired bioenergetics at the level of mitochondrial, and subsequent behavioral seizures, and is perhaps the most direct demonstration to date that an intrinsic disturbance of metabolic function can result in seizure activity (through changes in calcium regulation and impairment of ion channel activity). This will be an important contribution to the scientific literature.

MINOR:

1. Page 4, line 86: Would recommend changing "paroxystic" to "paroxysmal" (the latter which is a more recognized term).
2. Page 5, line 124: recommend including the concentration of beta-hydroxybutyrate used when first mentioned. In general, concentration and dose information were difficult to find, as well as route of administration (for kainate, page 7, line 175). This type of information was not conveniently presented.
3. Page 5, line 128: "both overcomed" is awkward. Would recommend using "both reversed".
4. Page 8, line 193: the authors probably meant "astro-MPC1-WT mice", not "neuro-MPC1-WT mice".
5. Page 12, lines 280-282: the authors might want to mention that chronic exposure of BHB might reduce the hyperexcitability of neuro-MPC1-KO mice.
6. Please review entire manuscript and use consistent tense. For example, on page 13, line 309, to maintain the past tense, it should read "We first assessed whether..."
7. Page 13, line 318: the authors used 10 mM BHB when examining calcium levels, but they earlier used 2 mM. They need to explain why they used a different concentration; and 2 vs 10 mM are quite different.
8. Page 13, line 323: it is not necessary to say "...interesting study published during the preparation of this manuscript." This phrase should be deleted, and the relevant reference simply cited.
9. The authors need to explain more clearly in the beginning what exactly is meant by "paradoxical" hyperactivity. They provide greater meaning later in the manuscript, but this should be clarified at the outset.

Significance

This is a very important study to show how primary defects in metabolism (i.e., disruption of the mitochondrial pyruvate carrier) can lead to epilepsy. Moreover, it details a primary mechanism that connects cellular bioenergetics to membrane excitability (through changes in calcium homeostasis and M-current function).

This is a well-conducted study that utilizes a multiplicity of experimental tools to link biochemistry with seizure activity. This type of study is not so readily done, and strengthens the notion that primary defects in metabolism can result in epileptic seizures.

This study is unique and attempts successfully to be more than just correlational. Hence it is a valuable contribution to the field.

The audience will likely consist of metabolic geneticists, neurologists/epileptologists, and neuroscientists. This is a beautiful study that runs the translational spectrum from biochemistry to behavior.

My expertise is in the field of translational epilepsy research, with a focus on mitochondria, metabolism, the ketogenic diet and ketone bodies. Thus, I am qualified to critically evaluate the entire manuscript.

Referee Cross-commenting

After reading comments and reviewing the manuscript again, would agree with Reviewer #1, and would change recommendation to MAJOR REVISION.

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

The authors have presented a very interesting and compelling set of data regarding the impact of conditional deletion of the only known pathway allowing the uptake of pyruvate into mitochondria. The paper comprises two interwoven stories that are both important. The first is the remarkable finding that the majority of excitatory neurons in the cortex (i.e. those under the influence of the CaMKII promoter)show remarkable metabolic flexibility as they tolerate elimination of pyruvate oxidation, considered the major supplier of ATP in neurons. The data on this seem clear although the authors did not delve into the potential mechanisms of metabolic compensation that likely occurs. Instead they examined whether there was some mal-adaptive compensation and they found clear evidence of this: in the absence of MPC activity the mice are much more prone to epileptic seizures, unveiled experimentally by relatively standard protocols (kindling). The authors present largely very convincing evidence that this mal-adaptive compensation in turn ends up decreasing the activity of KV7.2/7.3 channels whose job is normally to limit runaway repetitive firing by mediating an hyperpolarizing K+ efflux following an action potential. This channel, put on the map as it was one of the downstream targets modulated by cholinergic metabotropic activation, is also know known to be controlled by Calmodulin and therefore cytosolic Ca levels. Overall, I think at its core this manuscript is interesting and important. There however several weaknesses, I fear, will diminish the impact on the eventual readership. If these points can be addressed, it will strengthen the longevity of these findings:

1) It is puzzling why the authors resorted to using shRNA-mediated KD of MPC1 for some of the in vitro studies when they have gone to the trouble of making a floxed CRE-dependent mouse. Primary cells (e.g. Fig 1) or organotypic cultures (Fig. 6) from these mice would have made a more consistent set of starting conditions to compare data across the manuscript. As there viruses expressing the CRE recombinase are widely available this could have been used on mice simply harboring the floxed gene it they are worried about waiting for the expression of the CaMKII promoter for in-vitro conditions.

2) The data in Figure 5 gets a little less convincing as using extracellular glutamate to drive Ca elevations is so non-physiological that the results might really be distorted by the participation of something irrelevant to the story, even though it supports the overall interpretation for a role of Ca/CaM in the control of the channel. Similarly, the use of RU360 should be done with caution. The drug, although a useful antagonist of MCU in purified mitochondria, is famously finicky with respect to its ability to cross membranes and could well have off target impact. A much cleaner experiment would be to suppress the expression of MCU via KD. Presumably in the MPC-deficient neurons, this would have minimal impact on Ca signals. Given the frequent ambiguity associated with interpreting pharmacological results, coupled to the central importance of this finding in interpreting the entire paper, I think carrying out experiments with molecular genetic manipulation of MCU is warranted.

3) The authors have not really made clear in this paper whether the ability to suppress the phenotype of the MPC deficiency with ketones is really related to a providing TCA cycle support or instead a pharmacological impact on non-TCA related targets (such as the Kv7.2/7.3 channels). Presumably the use of other ketones might circumvent this. The action of ketone bodies has been a topic of considerable interest in neuroscience, given the clinical relevance for childhood epilepsies. Previous studies for example have argued for direct inhibition of the vesicular glutamate transporter (Juge et al. Neuron 2010). The use of other ketones (acetoacetate) would narrow down the interpretations of the data.

other

1) In vitro - scramble controls only serve to demonstrate there is no general effect of treating cells with shRNAs, but do not address if there is an off-target effect. The most convincing thing here would be to have an shRNA-insensitive variant that rescues.

2) Does rescuing CaMK binding to KCNQ channels rescue the phenotypes?

3) As the authors imply that BHB activates KCNQ channels, showing this directly in their prep would provide some convincing data. If this is true, why doesn't BHB increase firing rate of WT neurons?

4) How does the anti-epileptic effects of ketones in this study relate to previous reports of regulation of KATP channels? One of main concerns is that ketones might have a parallel anti-epileptic effect in the MPC1 KO mice that is unrelated to the mechanism proposed here.

Minor comments:

1- What is the MPC1 KO efficiency in CaMK neurons? The western blot in 2c is from the whole cortex and therefore does not show that. 2- Mitochondrial Ca2+ levels are not measured directly, for which there are many tools. This is needed to demonstrate definitively that there is a defect in Ca2+ handling."

Significance

see above.

Referee Cross-commenting

It seems we are in reasonable agreement about the pros & cons of the manuscript. I agree that alternative approaches to RU360 are warranted.

Transparent Peer Review

---

Download the complete Review Process [PDF] including:

• reviews
• authors' reply
• editorial decisions

</br>

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

This study seeks to define how human lysosomes selectively downregulate membrane proteins and identify the machinery involved in this process. To this end, the authors screened a set of 30 lysosome membrane proteins (LMPs) in a cycloheximide chase assay in a human cell line which led to the identification of RNF152 (an E3 ligase) as a particularly short lived LMP. Further experiments demonstrate that RNF152 degradation is ubiquitin, ESCRT and lysosome dependent. They also show that the E3 ubiquitin ligase activity of RNF152 is critical for its turnover. The overall technical quality of the experiments is high and conclusions about the degradation of RNF152 are mostly reasonable. My most significant concern is that while compelling data is provided for RNF152 turnover, the authors over-reach in their efforts to generalize their findings to other LMPs. Given that the E3 ligase activity of RNF152 is so important for its turnover, RNF152 might be a special case. Consistent with this, the authors did not characterize other LMPs with similarly high rates of turnover. Although it would be interesting if RNF152 regulates the stability of other LMPs, until such proteins are identified, the authors should be more cautious in their interpretation. Speculation on this matter is reasonable so long as it is labeled as such. Even with respect to RNF152 turnover mechanisms, the overall conclusions would be significantly strengthened by a demonstration that the endogenously expressed, untagged protein behaves in a similar manner to what was described for the GFP-tagged transgene. With respect to the question about how long it would take for the authors to address these concerns, I cannot give a precise answer as it would depend on whether they decide to much more narrowly interpret their findings and temper their major claims (less than a month) or to expand efforts to generalize results (time frame unknown and perhaps not feasible).

1. As a specific (but not the only) example of over-reaching in generalizing the findings, the abstract ends with the following statement: "Thus, our study uncovered a conserved mechanism to down-regulate lysosome membrane proteins." My concern is that although this mechanism might be generalizable, the authors have only presented data for RNF152.
2. There is a complete reliance on over-expressed, GFP-tagged RNF152. There is no demonstration that the endogenously expressed protein undergoes such high rates of turnover. It is thus possible that the data does not reflect the normal turnover pathway for this protein.
3. In Figure 2B, why is the loss of full length RNF152-GFP not accompanied by an increase in the signal for free GFP during these pulse-chase experiments?
4. Figure 2E: Were all of the pairs of Input and IP immunoblots subject to the same exposure and image adjustments?
5. Figure 3C-E: The RNF152 mutants have slowed but not eliminated degradation. Is this dependent on their association with or ubiquitination by the endogenouslyh expressed RNF152?
6. Methods section indicates that t-tests were performed for all statistics. However, many experiments contain multiple comparisons and are thus ideally suited to t-tests. The authors should either justify the use of t-tests or provide a more suitable statistical analysis.
7. Although the model in Figure 7 shows the E3 (RNF152) ubiquitinating other proteins and promoting their ESCRT-dependent sorting into ILVs, this study did not identifying any such clients of RNF152.

Minor

Page 3: "Without treatment, almost all types of LSD patients will develop severe neurodegeneration in the central nervous system." This statement is misleading as there are multiple forms of LSDs that do not result in neurodegeneration and it is only these LSDs which can be successfully treated via enzyme replacement therapies. Unfortunately, the neuropathic LSDs remain largely untreatable due largely to issues of blood brain barrier permeability.

Page 3: "As we age, the lysosome membrane gradually accumulates damaged proteins and loses its activity, which dampens the cell's ability to remove pathogenic protein aggregates and damaged organelles, eventually leading to cell death and inflammation (Carmona-Gutierrez et al., 2016; Cheon et al., 2019; Yambire et al., 2019)." The references provided do not provide sufficient direct support for this broad statement.

Page 10: STED imaging results (currently "data not shown") should be supported by showing the relevant data.

Camera and objective information should be provided for microscopy studies.

Significance

The identification of a generalizable mechanism for the turnover of mammalian LMPs would represent a significant advance and would raise many interesting questions about mechanisms, regulations and physiological impact. While this studies contributes some interesting new clues to this topic, it falls short of unambiguously establishing how most LMPs are turned over in human cells. The data with respect to RNF152 is intriguing as it supports the idea that a novel form of ESCRT-dependent protein clearance occurs at the limiting membrane of lysosomes. However, it remains very much unclear to what extent this can be generalized to other proteins.

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 mechanisms involved in lysosome membrane protein turnover are not well understood. Weichao et al. used a cycloheximide chase screen and overexpression 30+ lysosome membrane proteins in HEK293 cells to identify LMPs (lysosome membrane proteins) with fast turnover rates. They identified RNF152 as a suitable candidate for study given its high turnover rate and physiological relevance. They showed that RNF152's levels were regulated by ubiquitination by mutating cytosolic lysine residues and RNF152's ring domain and finding that these changes increased RNF152 stability. The researchers found that knocking down ESCRTIII and overexpressing a dominant-negative mutant of VPS4 increased RNF152 levels at steady-state and delayed RNF152 turnover. When expressed in yeast, RNF152 is localized on vacuole membrane and is also subject to regulation by the ESCRT pathway. Early ESCRT pathway members are essential for RNF152 degradation in yeast but not in mammalian cells. Taken together, these findings are important for furthering our understanding of how the levels of lysosome membrane proteins are regulated. A better understanding of ESCRT mediated LMP degradation is important not only for understanding mechanisms involved in controlling lysosomal activities but also for therapeutic development for many diseases involving dysregulation of LMP protein levels.

However, the following concerns should be addressed before the paper is published:

1. The authors have found that among 30 LMPs, three LMPs, LAPTM4A, RNF152, and OCA2, have half-lives less than 9 hours. RNF152 is a ubiquitin ligase and the authors showed that auto-ubiquitination is important for the recognition by the ESCRT machinery. Can the authors speculate how the ligase activity of RNF152 is regulated? Also, is similar mechanism involved in LAPTM4A and OCA2 turnover? Are these two proteins also ubiquitinated?
2. The authors should at least demonstrate that endogenous RNF152 levels and turnover are also regulated by ESCRT III and VPS4, using the stable cell lines the authors have already made. All of the mammalian cell experiments are performed using overexpression of RNF152, and an endogenous experiment would inspire confidence that the author's findings are not an artifact of over-expression.
3. While the authors showed that the K->R and C->S mutants of RNF152 have increased stability, it would be more compelling if they could perform an IP using HA-ubiquitin to prove this effect is due to a loss/reduction of RNF152 ubiquitination and not due to other changes in the protein. Another concern is whether mutating 8 lysine or 4 cysteine residues simultaneously would affect the folding of the protein, leading to abnormal aggregation in the cell.
4. For some of the data, statistical analysis is missing: a. All of the cycloheximide chase experiments. b. statistical significance for the puncta vs membrane GFP signal data shown in figure 6f c. The flow cytometry data
5. Fig. 4A and Fig. S2A, why MG132 treatment affects the levels of free GFP if it's inside of the lysosome?
6. Make sure that the figures are properly referenced in the text, there is one instance where the authors referenced figure 2d, when they clearly meant to reference figure 2e, and figure 2e where the authors meant to reference figure 2f etc.

Minor Comments:

1. In figure 1A, at CHX 3h, there's ~40% reduction of GFP-RNF152, however, in the rest of the figures, such as figure 2B，at CHX 2h, there's ~70-80% reduction of GFP-RNF152. How to explain the difference in the kinetics?
2. In figure 2F, it is hard to differentiate when the underline for input ends and the underline for IP begins unless the reader zooms in, please separate them a bit more.
3. Fig. 4F, it's very hard to see the red and green signals, maybe get rid of the DAPI channel increase the intensity for both green and red channels, and zoom in?
4. Scale bars are missing in the insert images in figure 1C, figure 4G and figure 6E.
5. In figure S1, the labels do not match with the blot for GFP-TMEM106B time points.

Significance

These findings are important for furthering our understanding of how the levels of lysosome membrane proteins are regulated. A better understanding of ESCRT mediated LMP degradation is important not only for understanding mechanisms involved in controlling lysosomal activities but also for therapeutic development for many diseases involving dysregulation of LMP protein levels.

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

Lysosomes play key roles in cellular homeostasis by functioning as a signaling hub for growth control and acting as a terminal catabolic station. Deregulation of lysosomes are now linked to multiple human diseases including cancer, neurodegeneration and etc. An emerging topic of interests in lysosomal biology is the regulation of lysosomal proteostasis and how it impacts the overall fitness and functionality of the lysosome per se. Zhang et al presents here a case study of quality control of lysosomal membrane proteins, with a focus on the turnover of a lysosomal anchor E3 ubiquitin ligase RNF152. They showed that RNF152 is rapidly degraded through an ESCRT-dependent fashion and that this mechanism is also conserved in yeast.

Major comments:

1. The writing of the manuscript including the abstract could be further polished. The manuscript in its present form appears to be a technical report that does not sufficiently convey the significance of this study.
2. Cyclohexamide is commonly used in studying the half-lives of proteins of interests. This is not a new method authors developed in the first place.
3. The data of protein turnover was presented by plotting the relative level of proteins as a function of time. But the use of degradation kinetics was all over the place in the manuscript, which is inappropriate scientifically. The authors should first generate fit to first order decay to acquire a degradation rate constant, k (min-1) and calculate half-life (T1/2) from there.
4. What are the functional consequences of RNF152 degradation? What are the biological impacts at both lysosomal and cellular levels in RNF152-depleted cells?
5. Given the rapid turnover of RNF152 at basal state, one can predict that this protein may become functionally important under specific circumstances, for example, certain stress. This aspect is worth exploring.
6. The authors chose RNF152 over OCA2, a melanosome-specific protein. However, OCA2 was shown to colocalize with LAMP2 much better than RNF152.

Minor comments:

1. Mislabeling and typo errors detected in the text: a. Page 7 "As expected, the full-length GFP-RNF152 and other lysosomal proteins such as LAMP2 and cathepsin D (CTSD) were enriched by Lyso-IP. In contrast, PDI (ER), Golgin160 (Golgi), EEA1 (endosomes), and GAPDH (cytosol) were not enriched (Figure 2D)." - should be Figure 2E instead. b. Page 7 "Our result confirmed that the lysosome population of GFP-RNF152 is quickly turned over, while LAMP2 is very stable on the lysosome (Figure 2E)." - should be Figure 2F instead. c. Page 14 "knocking down either TSG101 or both TSG101 and RNF152 only had a minor impact on the degradation kinetics of GFP-RNF152 (Figure S3A-B)." - should be ALIX instead of RNF152.
2. Stable cells expressing GFP-RNF152 or 3xFLAG-RNF152 were primarily used in this study. It will be useful to perform some experiments by examining the endogenous counterpart using antibodies against RNF152. For example, Figure 2D and 2E.
3. For all the flow cytometry analysis, the value of GFP intensity in respective graphs should be indicated.
4. Statistics analysis was not performed on Figure 5D.
5. In Figure 6D and J, what are the reasons for the appearance of multiple peaks, particularly, by the red line?
6. In Figure 3A, the question marks should be removed to avoid confusion. "Predicted" can be used instead if there is no direct evidence from mass spec analysis.
7. In Figure 3C, the authors identified two mutants including KR and CS that are refractory to degradation. It will be more insightful by showing the ubiquitination of these two mutants as in Figure 3B.

Significance

Multiple mechanisms including ESCRT complex have been reported to regulate the quality control of lysosomes. Understanding the roles of each mechanisms and selection of their substrates in maintenance of lysosomal integrity is of great interest in cell biology. Zhang and colleagues showed a case study of RNF152, a substrate of ESCRT-dependent degradation, but did not further pursue the biological functions of RNF152. This somewhat limits the conceptual advance of the study.

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

Learn more at Review Commons

---

Reply to the reviewers

The authors do not wish to provide a response at this time.

Transparent Peer Review

---

Download the complete Review Process [PDF] including:

• reviews
• authors' reply
• editorial decisions

</br>

Transparent Peer Review

---

Download the complete Review Process [PDF] including:

• reviews
• authors' reply
• editorial decisions

</br>

Transparent Peer Review

---

Download the complete Review Process [PDF] including:

• reviews
• authors' reply
• editorial decisions

</br>

Transparent Peer Review

---

Download the complete Review Process [PDF] including:

• reviews
• authors' reply
• editorial decisions

</br>

Transparent Peer Review

---

Download the complete Review Process [PDF] including:

• reviews
• authors' reply
• editorial decisions

</br>

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

Learn more at Review Commons

---

Reply to the reviewers

We would like to thank the editor for their consideration and the reviewers for their time and thoughtful comments. Below we have written a point-by-point response to their comments and concerns. The original comments are displayed in italic fonts, whereas our responses are in regular fonts for clarity.

Reviewer #1 (Evidence, reproducibility and clarity (Required)):

The submitted manuscript 'Identification of phenotype-specific networks from paired gene expression-cell shape imaging data' of Barker et al. uses a convolute of different bioinformatics tools (see Key Resource Table) to analyze reported RNA sequencing data and to correlate derived pathways with imaging features of breast cancer cell lines based on specific pathway constructions. The thin red line of the data presentation in the manuscript is not obvious.

\*Major concerns:***

1.1 The main biological 'finding' of the study RAP1 'as a potential mediator between the sensing of mechanical stimuli and regulation of NFkB activity' is reported and therefore the assumption 'how exactly extra-cellular mechanical cues are sensed by the cell and passed on to NFkB in breast cancer is not understood' is misleading. Please review: https://www.nature.com/articles/ncb2080** (human breast cancers with NF-κB hyperactivity show elevated levels of cytoplasmic Rap1. Similar to inhibiting NF-κB, knockdown of Rap1 sensitizes breast cancer cells to apoptosis) https://pubmed.ncbi.nlm.nih.gov/17510404**/ (RAP1 is a crucial element in organizing acinar structure and inducing lumen formation), and https://pubmed.ncbi.nlm.nih.gov/21429211/**.

R1.1 We thank the reviewer for pointing out these references. Teo et al. (which is cited in our manuscript) provides evidence that Rap1 regulates IKK and therefore NFkB in breast cancer, while Itoh et al. and McSherry et al. focus on Rap1’s ability to modulate migration and morphogenesis as do other similar papers cited in our manuscript. None of the papers show the significance of the Rap1-NFkB interaction in the explicit context of cell shape with Teo et al. only speculatively mentioning a potential relevance of Rap1/NFkB in migration (“Given that NF-κB is critical for [. . . ] stimulating invasion, our results document a clinical setting wherein Rap1-mediated regulation of NF-κB could be critical.”). We appreciate that the specific sentence the reviewer has drawn attention to is slightly misleading given Teo et al. and we will amend it in the revised manuscript. However, here, we use a novel methodology on a past dataset to link the concepts introduced by these 3 papers within the specific context of cellular morphology in breast cancer cells.

Specifically, in this manuscript, we identify a Rap1 expression module correlated with cell shape and find that it is at the network confluence of transcription factors activated by cell shape. This, along with our findings of modulation of NFkB co-activators, as well as previous work showing that it is a key mechano-transductive transcription factor leads us to hypothesize that Rap1 mediates the regulation and mechano-sensing of cell shape via its interaction with NFkB.

It is also important to note that, while we build on the findings of Teo et al., McSherry et al. and Itoh et al. relating NFkB, Rap1 and cell shape, we also use our method to focus on other proteins of interest. These are drawn attention to in the discussion, with the ARNT KO/TNFalpha module being the most highly correlated gene expression module with the morphological features. Also, the importance of transcriptional co-regulators of NFkB are illustrated in the network propagation, with both NR0B2 and PPARGC1A mentioned in the discussion. However, our analysis naturally concentrates on the node with the most apparent literature support, which as your reading suggests is Rap1. The significance of this manuscript is that it is an unbiased systems-based methodology used to link cell-shape with signaling, via transcription in a context-specific manner (i.e. in the context of breast cancer). This produces a phenotype-specific network that has allowed us to connect diverse mechanisms and hypotheses put forward by other authors and further our understanding of how signaling manages the sensing and regulation of cell shape in breast cancer. The methodology is also applicable to any paired transcriptomics/phenotype dataset.

1.2 Besides, Fig. 2 and 3 are unrelated to this main statements.

R1.2 Figure 2 shows the results of our morphological cluster identification and subsequent differential expression analysis. Since these were included as parts of the network, we included them to give the reader an idea of the components included by this step. Figure 3A shows the network that was generated by our pipeline which forms the basis of all subsequent biological exploration, and the discovery of Rap1 and nodes important for the regulation of cell shape. Figure 3B shows that no bias was introduced by using the specific algorithm for our network generation. As such Figures 2 and 3 are related to the generation of the cell shape-specific network that forms the basis of our study. We will amend the text and figure legends to clarify this point more carefully, and we will consider moving some of the panels to the supplementary materials to simplify the message.

1.3. The spotted RAP1 (by TFs JARD2 and RUNX2) finding is not obvious without Fig. 4 results, a network propagation of functional TFs in differentially activated processes (basal vs. luminal) in the cell shape regulatory network. Please show that RAP1 could be not identified without the network based on TF and DEG only.

R1.3 The Rap1 hypothesis is supported by both Figure S2E and Figure 4. Figure S2E shows that the Rap1 pathway-enriched gene expression module is the most differentially expressed module among those incorporated in our cell shape regulatory network. This suggests that this module is correlated to cell shape on a transcriptomic level, but does not necessarily mean anything within the context of intracellular signaling. Figure 4 shows that this gene expression module is at the confluence of activated transcription factors as specified by our constructed signaling network. This is an interesting finding as it implies (unlike Figure S2E) that the Rap1 gene expression module is relevant to intra-cellular signaling.

While the Rap1 module is indeed differentially expressed and could in theory have been found just by the DE analysis as being important, the network approach enables us to integrate these modules of co-expressed genes within known signaling networks. This allows us to go further than just making comments about expression and transcription factor activity, to discussing how signaling networks interact with our identified gene expression modules. This in turn allows us to construct more sophisticated hypotheses about cell shape regulation.

Particularly, we use this analysis to reinforce the association with Rap1 by illustrating that the Rap1 network node lies at the confluence of transcription factors activated in luminal-like and basal-like cell shapes. We also use the network to identify highly central nodes (such as PPARGC1A, CTNNB1 and ESR1) and other proteins identified in the network propagation (YAP1, IKBKB and ARNT). Furthermore, the network is used as a means of integrating gene expression modules in their signaling network environment. The method by which this embedding was done (in-going edges being transcription factors regulating the module and out-going edges being signaling proteins contained within the module) adds context specificity to a network that is otherwise generalised to many cell-types and contexts.

The lack of clarity on how we arrived at Rap1 as a key tenet of our discussion, as well as the added value to the methodology of the network analysis is something that we will certainly work on in our revision and we thank the reviewer for their valuable feedback. We will also move Figure S2E from the supplementary figures to the main Figure panels, as it is an important part of how we arrived upon Rap1 as a module of particular interest.

1.4 More complex fluorescence phenotypes are available and do not match the complexity of the RNASeq data, data input and pathway construction with only 10 simple cell shape features. Conversely, relative 'monoclonal' breast cancer cell lines may are the only application for this workflow.

R1.4 We thank the reviewer for their comment and respectfully disagree. These cell shape features were sufficient for the original authors Sero et al. to predict TF activities (PMID: 26148352) and Sailem et al. to identify clinically predictive metagenes (PMID: 27864353). Although these features seem simplistic, they concisely summarise a highly complex phenotype and are proven to encode metastatic potential (PMCID: PMC6976289) and to be prognostic markers for breast cancer progression (PMID: 28977854). Accordingly, these features are re-analysed using our workflow to better understand the signaling that drives them. Understanding other features that might match the complexity of our expression data is possible using the presented method, but is outside of the scope of the research within this manuscript as our research question focuses on the regulation of cell shape in breast cancer.

As evidence that this cell shape network is applicable beyond cell lines, we can perform an analysis on breast cancer patient data from the TCGA, demonstrating the relationship of our network’s components with metastasis, which is highly related to cell shape.

1.5 Image features Fig. 1 and 5 do not match

R1.5 It is true that the features in Fig.1 and Fig.5 do not exactly match. This is because these two datasets came from different studies. While they are slightly different features, the essential phenotypes that they quantify are the same. For example, in Fig 5., cytoplasm area, cytoplasm perimeter, nucleus area, nucleus length, nucleus width and nucleus perimeter, are clearly basic morphological features that are analogous to features in Fig.1, such as cell area, cell width to length, nucleus area, nucleus width to length. It is a solid assumption that the biological processes that drive these features are common, and our results illustrate this. Meanwhile, the existence of features measured in the dataset in Fig.5, that are not analogous to those used in the dataset used for model construction, provide convenient negative controls in order to determine that our network describes regulatory processes specific to the features used in network construction.

1.6 with Fig. 5 being a rather indirect 'proof' of usability.

R1.6 The purpose of Figure 5 is not to demonstrate usability but to prove that the network that we have identified is indeed representing cell signalling that controls cell shape. It shows that perturbations inside our network have a significantly stronger effect on phenotypes used in the creation of the network than perturbations outside our network. Moreover, it shows that this is not the case for phenotype features not used in the creation of our network, underlining that our network is both accurate and specific to the phenotypes used as input. We will add further clarifications in the text and figure legend to explain this better.

1.7 Fig. 1a has not achieved a visual descriptive state and asking a lot.

R1.7 We apologize for the lack of clarity here and will revise the figure to better present the method and reduce confusion.

Reviewer #2 (Evidence, reproducibility and clarity (Required)):

The authors combine single cell morphology and gene expression data to identify signaling activities implicated in the control of cellular morphogenesis. They describe a reasonable bioinformatics pipeline from gene expression shifts between two morphological phenotypes to pathways, then to common transcription factors to signaling. As far as I can assess the situation (I am not familiar with all the tools they use) the proposed pipeline works convincingly.

We would like to thank their reviewer for their thoughtful comments and to clarify that the analysis has not been done on single cell gene expression data, but rather on bulk RNAseq data. This is further explained in the point-by-point responses below.

2.1 However, I am concerned that the logic underlying this analysis is only partially valid. The link between signaling and morphology may be more direct than via TF-based gene expression regulation. Many signals (and many of the kinases the authors test for validation) are implicated in morphology control as direct upstream regulators of cytoskeleton dynamics and adhesion. This also applies to the GTPase Rap1, which the authors fish out as the most differentially expressed signal between two types of morphologies. In addition to the indirect effect of Rap1 on morphology via NFKB regulation suggested by the authors, Rap1 will affect morphology probably very directly through activation of Rac -> F-actin and RIAM -> nascent adhesions. At minimum, the authors should discuss this complexity as a caveat of their approach. And dependent on the impact the authors hope to have with this story, I believe they should experimentally resolve the ambiguity of direct vs indirect signaling for some of their key interpretations.

R2.1 We thank the reviewer for making this fair point about the limitations of our data, i.e that we are not able to directly observe and delineate exactly how upstream signaling has modulated or is modulated by cell shape. As a first major clarification, which we will make sure to include in the text, the Rap1 node doesn’t necessarily represent the Rap1 GTPase itself. It is a co-expression module that is enriched in activators and downstream effectors of Rap1 signalling. As such when we are talking about the Rap1 module we mean the subnetwork of Rap1 signalling rather than the specific small GTPase itself. Thus, we can’t assume that Rap1 itself is the key node in this subnetwork and it is therefore complicated to design specific experiments to test the direct or indirect modulation of cell shape by Rap1. We agree however that additional information regarding the role of this module in regulating cell shape would be interesting and valuable.

For this study, we have access to transcriptomic and cell shape data from 14 cell lines with transcriptomic and cell shape data. Using the expression data, we will quantify Rap1 expression module activity and its relationship to NFkB transcriptional activity across these cell lines. By comparing in these cell lines the effect on morphology when NFkB and the Rap1 module are combinatorially activated or deactivated, we can disseminate between competing hypotheses for direct and indirect activities of these two factors on cell shape. For example, if the overwhelming source of Rap1 module’s function was via direct interaction with F-actin, then Rap1 module activity would be predictive of cell shape, regardless of NFkB activation. A caveat to this is our limited access to only 14 cell lines. Additionally, if necessary, we have access to drugs that can induce cytoskeletal defects and perturb morphology directly. This can be used to disrupt the relationship between Rap1 and F-actin that the reviewer has identified and gauge the effect on cell shape. If such an intervention disrupts the relationship between Rap1 signaling module/NFkB transcriptional activity and cell shape then we can hypothesise that the activity of Rap1 signaling is greater than just its direct activity on F-actin. Finally, we can perform a knock-down of Rap1 (or selected components of its module) and NFkB itself and gauge the effect of such a perturbation on the Rap1 gene expression module and cell shape.

That being said, these signaling modulations (whether indirect or direct) are reflected accordingly in differentially activated transcription factors, and therefore can be observed and recorded from expression data. This is an interesting finding, as it implies that signaling processes not explicitly making use of transcription factors (such as those that directly affect adhesion complexes, regulating cytoskeletal proteins etc) can still have their activity gauged through their indirect downstream expression signatures. In any case, our findings illustrate that there is a cell shape-specific modulation of the Rap1 module in breast cancer, reflected in the expression data. Rap1 almost certainly has some direct contributions to cytoskeletal dynamics in breast cancer (PMID: 10805781, PMID: 30156466 and PMID: 22644079), but here we observe clearly how it also is modulating transcription factors, that we hypothesise may contribute to the development of a morphological and transcriptomic ‘niche’ in a more robust and long-term fashion. Nonetheless, the points discussed by the reviewer are valuable and in our revision we will discuss this as a potential caveat.

In defense to the presented premise, the authors start out by looking for correlation between gene expression and morphology, and they find some signal. Correlation analysis, especially in large data sets, tends to be pretty robust and specific, even on presence of strong confounders. Thus, even though the correlation expression-morphology, which points indirectly at morphology-regulating signaling modules, is likely to be super-imposed by direct morphology-regulating signaling pathways the proposed approach will not be able to detect, the presented analysis is valuable, in principle.

We thank the reviewer for their positive comments.

That said, I have a number of substantial concerns also with the implementation and presentation of the approach.

2.2 First, on the presentation side, for a paper that talks about cell morphology it is strange to have not a single figure panel showing an image of cells, or at least cell outlines. As a reader I would like to get visual impression of how different a high vs low Rap1 gene expresser is, for example.

R2.2 - We agree that it would greatly help the clarity and message of the manuscript. As we are not able to use the public and previously published data that have been used for our paper due to copyright laws associated with journal publications, we will generate relevant images representative of the respective cell shapes and include them in the manuscript.

2.3 Along the same lines, it is not quite clear to me when the authors collate entire cell lines into a single phenotype, do they switch then to population-based analysis? That is, for example the volcano plots in 2B,C are they representing an average gene expression shift?

R2.3 - We apologise for the lack of clarity. The morphological clustering and differential expression illustrated in Figure 2 is to find expression signatures responsible for distinct breast cancer cell shapes. We link our expression data and our imaging data via the breast cancer cell lines and so are limited to studying expression in bulk per cell lines. The volcano plots in 2B,C are of the differentially expressed genes (as calculated by DESEQ2) in morphologically distinct clusters of cell lines as specified by Figure 2A. This was collated so we could observe transcriptomic differences accounting for cellular morphology, rather than differences in cell lines (these being already well characterised and not in the scope of this manuscript).

We will add additional clarifications in our revised manuscript to further explain this.

2.4 How heterogeneous are the morphological signals?

R2.4 We provide values of standard deviation for the morphological features of the derived clusters (page 4 - “Clustering based on morphology reveals distinctive cell-line shapes”). The heterogeneity of the morphological clusters is minimised as per the elbow plot shown in Figure S2A. This plot illustrates the decreasing total within-cluster variation of the cell shape groups as the number of clusters (k) is increased. The point of inflection represents the optimum number of clusters (in our case k=3). Aside from this, we note that one of those 3 clusters is significantly more heterogeneous both morphologically speaking and biologically speaking (illustrated Figure 2A) and so we used the other two which showed more informative gene expression profiles and could be annotated roughly with breast cancer subtypes (basal and luminal - although is alignment was not perfect since the grouping was based only on the morphology). We took the more heterogeneous cluster (cluster A, Figure 2A) to be the least relevant cluster in terms of morphology and biologically significance, also because it contained the non-tumorigenic cell-line MCF10A.

2.5 Are the correlations between gene expression and morphology computed with single cell data as the basis?

R2.5 Due to the availability only of bulk RNA sequencing data for the cell lines for which we also had high content imaging data, all analyses are done using bulk RNA sequencing data at the cell line level. We will clarify this in the text and methods section.

2.6 Could the volcano plots be sharpened by accounting for the single cell variation in morphology instead of lumping the cells into two morphological classes?

R2.6 As per the limitations mentioned in R2.3 and R2.5, we cannot study gene expression at the single cell level. Furthermore, the utility of using morphological clusters was so that we could observe morphological transcriptomic traits rather than those specific to cell lines.

2.7 On the back end of the paper, when the authors apply kinase inhibitors to validate some of the claimed pathways, it would be nice for the reader to see the morphological effects of these inhibitors. And to relate the kinase induced shifts to the morphological heterogeneity that is the basis for the study driving, initial correlation analysis? At the end of the day, the proof is in the pudding.

R2.7 - We thank the reviewer for his very good point, and it would certainly improve the manuscript. We will attempt to source a visual illustration of the effect of kinase inhibitors on the breast cancer cells. However, the dataset we source this data from is publicly available, but unpublished and so our use is constrained by the terms of use of LINCs. We will contact the LINCS consortium to acquire permission and if they allow, we will certainly include them in our revised manuscript.

2.8 Finally, cell morphology regulation is a pretty foundational process of life. One therefore wonders whether the pathways the authors pulled out of their analysis work also in other cell types, beyond breast cancer cells? What if they pooled data from different cell types that cover the morphological state space more broadly?

R2.8 We thank the reviewer for this interesting point. We have observed that many if not most of the general processes identified, i.e. developmental pathways, extracellular matrix regulatory pathways and adhesion pathways, are already known to be associated with cell shape regulation and mechanotransduction in many different cell types. Thus, at the ‘big picture’ level, our findings hold across multiple cell types. However, the precise wiring of our network seems to be breast cancer specific. This is evidenced by the fact that when we try to use the LINCS data from other cell types to see if our network still holds in these contexts, we do not observe significant increase in changes in morphology when perturbing central nodes within our network compared to outside of our network. This is not unexpected: depending on the individual molecular background of each tissue and tumour type, signalling networks are known to be wired differently. Indeed, our method that uses the context- and cell feature- specific gene expression modules and the transcription factors that regulate them as a basis for extracting our cell shape signalling network allow for identification of exactly its specific wiring in the context used for training, i.e. the breast cancer cells. It would be very interesting to repeat our analysis on similar data on a different tissue type to identify parts of the network that seem to be identical versus those that differ. We have not been able yet to identify public systematic high content imaging data for a different tissue across multiple cell lines, but we will continue to look for such a dataset in the literature and through our network of collaborators. We will also explore the possibility of extracting such information from images from the TCGA to perform this analysis across patients of a specific cancer type, although admittedly we are not sure how feasible it would be to extract analogous features as the ones used for the breast cancer network from the images available. We will also add this point to the discussion.

Reviewer #2 (Significance):

The premise of this manuscript is very exciting and interesting: Is it possible to identify from a correlation of cell morphology and single cell gene expression the underlying cell signaling states that control morphology? Answers to this will begin to shed some light on the black box relation of morphology as an informant of cell states, which has been exploited by pathologists, physiologists, and cell biologists for more than a century, and which has seen a sharp revival recently thanks to deep learning, which is exceedingly good at finding correlations between data patterns.

This paper would have a broad audience in quantitative cell biology, systems biology, and perhaps also life data science and cancer (although the cancer aspect is marginal).

Reviewer #3 (Evidence, reproducibility and clarity (Required)):

\*Summary***

In this work Barker et al. used computational approaches to analyze several existing data sets (including morphology and expression) in a common context of signaling-regulatory network that correlates with cell morphological features. They identified several pathways and associated transcription factors that their expression levels correlate with specific cell morphological features. The work thus has two main contributions. First, it provides a network of signaling pathways and regulons that may affect the morphological features of breast cancer cells. Second, the computational procedure can be general to study other systems.

\*Main comments***

3.1 I assume that in all analyses using the packages listed in the manuscript, some parameters need to be selected. The authors need to provide these details, and discuss whether the results are robust against parameter choice (at least to certain degree).

R3.1 We thank the reviewer for this comment. Parameters need to be selected in WGCNA and PCSF. In WGCNA they are selected based on the guidance given by the authors of the package, little significant variation in the results was observed when these were changed (i.e. the make-up of the gene expression modules naturally changed, but the processes enriched in the modules correlated to cell shape were the same).

PCSF is a more sensitive step because the best solutions to the sub-network identification problem are observed when the network edge weights are permuted over a number of iterations. Following this, the union of the produced array of networks is taken to be the solution. Obviously, biology is an inherently noisy system and so this formulation of the PCSF algorithm can capture latent network architecture that the deterministic variation cannot. This introduces extra parameters based on the requirement to introduce random noise to the network, along with the standard PCSF parameters (seen here: https://rdrr.io/github/IOR-Bioinformatics/PCSF/man/PCSF_rand.html) that are used to take into account of user variations in network degree distribution, edge-weight distribution, etc. It is normal for some tuning of parameters to be required for users to tailor their PCSF to their supplied network. We used degree distribution to gauge whether our network appeared to be of a biological ‘scale-free’ distribution and selected parameters based on that. This provides an affirmation that our resulting network is consistent with how we understand the topology of biological networks, and as a result the parameters selected are not arbitrary.

Nonetheless, we also tested variations in these parameters and found that although levels of significance in our validation would vary, the trends apparent from our validation did not (i.e., that targeting kinases within our network produced a larger effect on cell shape than those outside). From this we were assured that our conclusions mentioned in the discussion were robust to parameter selection. All details of parameters used are currently in our gitlab page, but we will additionally include them in the methods section.

3.2 Cells show some degree of heterogeneity both in cell expression and morphological features, which can be affected by many factors. Wu et al. (Sci Adv 2020, 6 (4): eaaw6938) identified several subgroups of MD-MBA-231 cells with persistent (over generations) distinct morphology, expression profiles, and metastasis potential. Another possible main factor to cell morphology heterogeneity is cell cycle stage. I understand that the analyses in this work are limited by the types of data available, for example, the expression data are largely bulk. One exception might be the data shown in Fig 5. Besides giving the fold changes after kinase inhibitor treatment, the authors may also analyze the variance of cells before and after treatment to estimate the relative extent of cell-cell heterogeneity relative to the effect due to treatment.

R3.2 We thank the reviewer for the comment and suggestion. We will perform such an analysis and include this point in the discussion. However, to reassure the reviewer, we are not concerned about this affecting our analysis in a major way as, in data from high content microscopy experiments such as the ones we used, hundreds of cells are sampled and the resulting quantified phenotypes are represented by the average from single cells, after removing outliers. Similarly in the bulk RNAseq experiments the dominant cell phenotype/expression profile would be mainly represented in the data. We are therefore reasonably confident that both sets of data used from each cell line indeed represented the most common phenotype for that cell line.

3.3. As related to point 2, In Fig. 1B, I am surprised that cell cycle only correlates with cell area significantly, while one knows that cells undergo dramatic change during cell cycle. For example, cells would turn to be roundish for mitosis. How would the authors explain the results? Is it possible that there is sampling bias towards interphase cells?

R3.3 We thank the reviewer for this comment and apologise for the confusion. The y axis of Fig.1B relates to gene expression modules identified in the expression data. These were named based on any informative term that could be associated with the genes within the modules as implemented by gene set enrichment. The goal was to provide more informative names than the default module names that are based on colours. ‘Cell cycle’ as a term in Reactome is a particularly generalisable gene set and was applied to the gene expression module in question because it was the only informative term identified for it. This singular gene expression module does not represent all transcriptomic activity associated with the cell cycle process. Indeed, the term ‘Cell cycle’ was also enriched in the ‘Hedgehog off-state’ gene expression module (Supplementary table 5). As the enrichment is based only on the genes: HAUS8;MCM8;NCAPH2;MIS12;BIRC5;CENPM;SPDL1;FBXO5;TYMS;TUBB4A, which are not necessarily the major cell cycle-relevant genes, we agree that the name of the specific module is not ideal and can cause confusion so we will rename this module. We will also go over the naming of all the modules to ensure that the names are indeed representative of the module functions.

\*Minor comments***

3.4 In Fig 1B, I have trouble to understand the biological relevance of some module names, like "Green", "indianred4"?

R3.4 Our pipeline uses WGCNA which constructs gene expression modules completely from gene expression data. We named modules based on terms we could find associated with the genes within a module. Some modules did not have any informative terms associated with them and so we opted to keep the default name of those modules that WGCNA supplies (based on colours). We will attempt to make this clearer in our revised manuscript, by adding a better explanation, and renaming these modules to something that makes it clear that we could not assign a clear function such as non-annotated (NA) module 1,2,3 etc.

3.5 Fig 3B: I can't find a detailed explanation on how the combined score was calculated.

R3.5 Thank you for pointing this out. This is described in Chen et al. 2013 as part of the enrichR package for gene set enrichment analysis. We will add this detail in the methods section under “Quantification and Statistical Analysis“.

3.6 Some of the cell features in Fig 5A are not in Fig 1. Are they from the same analysis? Any explanation?

R3.6 As the two datasets were acquired in two completely different studies there isn’t a 1-1 correspondence of the phenotype features, however several of them essentially represent the same phenotype. For example, in Fig 5., cytoplasm area, cytoplasm perimeter, nucleus area, nucleus length, nucleus width and nucleus perimeter, are analogous to features in Fig.1, such as cell area, cell width to length, nucleus area, nucleus width to length. The intersection between the features in these two datasets is not exact however, and we use the features in Fig.5 not used in the network construction as a negative control. This allows us to show that our network is phenotype-specific to the morphology features it was trained on. We will clarify this in the manuscript.

3.7) It is interesting that the authors have identified a number of pathways known to be related to mechanosensing. Does the Hippo-YAP/TAZ pathway appear in their analysis?

R3.7 Yes, YAP1 is also significantly highly ranked in our network propagation of activated transcription factors in Fig.4 in both luminal- and basal- shaped cell lines. Furthermore, since submitting we have been experimenting with identifying subnetworks of our regulatory network using maximum-flow. Here we assess the interaction between the Rap1 module (given its centrality to our discussion) with NFkB and what is the most efficient ‘flow’ of information between these two nodes given our network. To our interest, we identify LATS2, DVL1 (genes within the Rap1 module), YAP1 and TAZ as key mediating factors between these nodes. This implicates the Hippo-YAP/TAZ pathway as being of particular importance in the interface of our identified gene expression module and our derived signalling network. As an illustration of this we include here one of the preliminary networks derived from this analysis.

Figure - Network describing maximum flow (top 20 edges) between Rap1 signaling module as the source node and NFKB1 as the target node. The super-node representing the gene expression module is coloured red (with the interface nodes used to embed it in the signaling network coloured blue) and NFKB1 coloured azure. Edge thickness indicates weight, which was used as the maximum capacity used in the max flow calculation.

We will go over the figures and move barplots and more technical information to the supplement to make space for more figures of this nature that better illustrate the processes involved in the regulation of cell shape as derived from our analysis.

Reviewer #3 (Significance):

Extensive studies link cell morphological features and cell expression states, with some important work from one of the authors (Chris Bakal). This topic gains further interest recently. For example, Wu et al. (Sci Adv 2020, 6 (4): eaaw6938) demonstrated that cell shape encodes metastasis potential. Wang et al. (Sci. Adv. 2020 6 (36): eaba9319) traced cell dynamics in cell feature space. Given the context, the work of Barker et al. is a timely study to establish a cell shape-signaling network from integrated analysis of several different types of data. While some previous studies have related the NFkB network and cell morphology, this work further provides unbiased analysis on the relation between cell morphological features and multiple pathways/transcription factors. It is interesting, for example, that the study identifies the correlation between the Rap1 pathway and cell morphology, which was not well studied previously. As the authors acknowledged, there are some limitations of their approach. For example, the relations identified are correlative instead of causal without further verification. The resultant network may have sampling bias. Despite these limitations, I suggest that this work will be a nice contribution to the field and can provide a basis for further studies.

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 this work Barker et al. used computational approaches to analyze several existing data sets (including morphology and expression) in a common context of signaling-regulatory network that correlates with cell morphological features. They identified several pathways and associated transcription factors that their expression levels correlate with specific cell morphological features. The work thus has two main contributions. First, it provides a network of signaling pathways and regulons that may affect the morphological features of breast cancer cells. Second, the computational procedure can be general to study other systems.

Main comments

1) I assume that in all analyses using the packages listed in the manuscript, some parameters need to be selected. The authors need to provide these details, and discuss whether the results are robust against parameter choice (at least to certain degree).

2) Cells show some degree of heterogeneity both in cell expression and morphological features, which can be affected by many factors. Wu et al. (Sci Adv 2020, 6 (4): eaaw6938) identified several subgroups of MD-MBA-231 cells with persistent (over generations) distinct morphology, expression profiles, and metastasis potential. Another possible main factor to cell morphology heterogeneity is cell cycle stage. I understand that the analyses in this work are limited by the types of data available, for example, the expression data are largely bulk. One exception might be the data shown in Fig 5. Besides giving the fold changes after kinase inhibitor treatment, the authors may also analyze the variance of cells before and after treatment to estimate the relative extent of cell-cell heterogeneity relative to the effect due to treatment.

3) As related to point 2, In Fig. 1B, I am surprised that cell cycle only correlates with cell area significantly, while one knows that cells undergo dramatic change during cell cycle. For example, cells would turn to be roundish for mitosis. How would the authors explain the results? Is it possible that there is sampling bias towards interphase cells?

Minor comments

4) In Fig 1B, I have trouble to understand the biological relevance of some module names, like "Green", "indianred4"?

5) Fig 3B: I can't find detailed explanation on how the combined score was calculated.

6) Some of the cell features in Fig 5A are not in Fig 1. Are they from the same analysis? Any explanation?

7) It is interesting that the authors have identified a number of pathways known to be related to mechanosensing. Does the Hippo-YAP/TAZ pathway appear in their analysis?

Significance

Extensive studies link cell morphological features and cell expression states, with some important work from one of the authors (Chris Bakal). This topic gains further interest recently. For example, Wu et al. (Sci Adv 2020, 6 (4): eaaw6938) demonstrated that cell shape encodes metastasis potential. Wang et al. (Sci. Adv. 2020 6 (36): eaba9319) traced cell dynamics in cell feature space. Given the context, the work of Barker et al. is a timely study to establish a cell shape-signaling network from integrated analysis of several different types of data. While some previous studies have related the NFkB network and cell morphology, this work further provides unbiased analysis on the relation between cell morphological features and multiple pathways/transcription factors. It is interesting, for example, that the study identifies the correlation between the Rap1 pathway and cell morphology, which was not well studied previously. As the authors acknowledged, there are some limitations of their approach. For example, the relations identified are correlative instead of causal without further verification. The resultant network may have sampling bias. Despite these limitations, I suggest that this work will be a nice contribution to the field and can provide a basis for further studies.

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 combine single cell morphology and gene expression data to identify signaling activities implicated in the control of cellular morphogenesis. They describe a reasonable bioinformatics pipeline from gene expression shifts between two morphological phenotypes to pathways, then to common transcription factors to signaling. As far as I can assess the situation (I am not familiar with all the tools they use) the proposed pipeline works convincingly. However, I am concerned that the logic underlying this analysis is only partially valid. The link between signaling and morphology may be more direct than via TF-based gene expression regulation. Many signals (and many of the kinases the authors test for validation) are implicated in morphology control as direct upstream regulators of cytoskeleton dynamics and adhesion. This also applies to the GTPase Rap1, which the authors fish out as the most differentially expressed signal between two types of morphologies. In addition to the indirect effect of Rap1 on morphology via NFKB regulation suggested by the authors, Rap1 will affect morphology probably very directly through activation of Rac -> F-actin and RIAM -> nascent adhesions. At minimum, the authors should discuss this complexity as a caveat of their approach. And dependent on the impact the authors hope to have with this story, I believe they should experimentally resolve the ambiguity of direct vs indirect signaling for some of their key interpretations.

In defense to the presented premise, the authors start out by looking for correlation between gene expression and morphology, and they find some signal. Correlation analysis, especially in large data sets, tends to be pretty robust and specific, even on presence of strong confounders. Thus, even though the correlation expression-morphology, which points indirectly at morphology-regulating signaling modules, is likely to be super-imposed by direct morphology-regulating signaling pathways the proposed approach will not be able to detect, the presented analysis is valuable, in principle.

That said, I have a number of substantial concerns also with the implementation and presentation of the approach. First, on the presentation side, for a paper that talks about cell morphology it is strange to have not a single figure panel showing an image of cells, or at least cell outlines. As a reader I would like to get visual impression of how different a high vs low Rap1 gene expresser is, for example. Along the same lines, it is not quite clear to me when the authors collate entire cell lines into a single phenotype, do they switch then to population-based analysis? That is, for example the volcano plots in 2B,C are they representing an average gene expression shift? How heterogeneous are the morphological signals? Are the correlations between gene expression and morphology computed with single cell data as the basis? Could the volcano plots be sharpened by accounting for the single cell variation in morphology instead of lumping the cells into two morphological classes? On the back end of the paper, when the authors apply kinase inhibitors to validate some of the claimed pathways, it would be nice for the reader to see the morphological effects of these inhibitors. And to relate the kinase induced shifts to the morphological heterogeneity that is the basis for the study driving, initial correlation analysis? At the end of the day, the proof is in the pudding.

Finally, cell morphology regulation is a pretty foundational process of life. One therefore wonders whether the pathways the authors pulled out of their analysis work also in other cell types, beyond breast cancer cells? What if they pooled data from different cell types that cover the morphological state space more broadly?

Significance

The premise of this manuscript is very exciting and interesting: Is it possible to identify from a correlation of cell morphology and single cell gene expression the underlying cell signaling states that control morphology? Answers to this will begin to shed some light on the black box relation of morphology as an informant of cell states, which has been exploited by pathologists, physiologists, and cell biologists for more than a century, and which has seen a sharp revival recently thanks to deep learning, which is exceedingly good at finding correlations between data patterns.

This paper would have a broad audience in quantitative cell biology, systems biology, and perhaps also life data science and cancer (although the cancer aspect is marginal).

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

The submitted manuscript 'Identification of phenotype-specific networks from paired gene expression-cell shape imaging data' of Barker et al. uses a convolute of different bioinformatics tools (see Key Resource Table) to analyze reported RNA sequencing data and to correlate derived pathways with imaging features of breast cancer cell lines based on specific pathway constructions. The thin red line of the data presentation in the manuscript is not obvious.

Major concerns:

1. The main biological 'finding' of the study RAP1 'as a potential mediator between the sensing of mechanical stimuli and regulation of NFkB activity' is reported and therefore the assumption 'how exactly extra-cellular mechanical cues are sensed by the cell and passed on to NFkB in breast cancer is not understood' is misleading. Please review: https://www.nature.com/articles/ncb2080 (human breast cancers with NF-κB hyperactivity show elevated levels of cytoplasmic Rap1. Similar to inhibiting NF-κB, knockdown of Rap1 sensitizes breast cancer cells to apoptosis) https://pubmed.ncbi.nlm.nih.gov/17510404/ (RAP1 is a crucial element in organizing acinar structure and inducing lumen formation), and https://pubmed.ncbi.nlm.nih.gov/21429211/. Besides, Fig. 2 and 3 are unrelated to this main statements.
2. The spotted RAP1 (by TFs JARD2 and RUNX2) finding is not obvious without Fig. 4 results, a network propagation of functional TFs in differentially activated processes (basal vs. luminal) in the cell shape regulatory network. Please show that RAP1 could be not identified without the network based on TF and DEG only.
3. More complex fluorescence phenotypes are available and do not match the complexity of the RNASeq data, data input and pathway construction with only 10 simple cell shape features. Conversely, relative 'monoclonal' breast cancer cell lines may are the only application for this workflow. Image features Fig. 1 and 5 do not match, with Fig. 5 being a rather indirect 'proof' of usability.
4. Fig. 1a has not achieved a visual descriptive state and asking a lot.

Significance

The 'Review Commons' efficiently facilitates the reviewing process for the corresponding journals due to the broader 'audience'. On the other hand, authors face less restrictions and pressure for the same reason. Although I really like the idea of pulling the reviews upstream into the preprint process, I would like to answer here also with a kind of pre-review to avoid entering partly immature manuscript to 'Review Commons'. Review Commons might install an automated 'sanity check' of manuscripts in the future to keep the quality of submissions higher?

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

Learn more at Review Commons

---

Reply to the reviewers

Review Commons Reviews for Refereed Preprint RC-2021-00693

Ferrari G. et al., DLL4 and PDGF-BB regulate migration of human iPSC-derived skeletal myogenic progenitors.

Reviewer #1 (Evidence, reproducibility and clarity): Summary:

The paper presented by Ferrari et al., aims to improve the migration capacity of hiPSC- derived myogenic progenitors. For this purpose, the authors used a previously published well characterized hiMPs model and focussed on the modulation of NOTCH and PDGF signaling pathways. The rational to target these pathways was based on muscle cells migrations molecular events observed during developmental described in the literature.

Major comments: Are the key conclusions convincing?

This is a very interesting paper. Few clarifications as suggested below need to be done before being fully convincing. Enrichment test and heat maps and the network analysis are not well explained in terms of which genes were selected and why, and in terms of which gene set were selected and why. In some cases, the information may be given in the paper, but it is not easy for the reader to find it. It should be stated more clearly. For example, in Fig2C why these eight were chosen for the heat maps and why not other genes known to be involved in myogenesis, cell migration etc. Similar comment for figure 3 A, D and G. Another example, in Fig 2E, on what basis are some gene sets chosen to be shown in this figure when there are many more significant in the supplementary table.

We thank the Reviewer for their positive feedback and for this comment. Although some answers to the queries could be found within the figure legends, we agree that figures could have been more self-explanatory, and we will amend them accordingly. We will also add additional information into the main text to clarify those specific points.

In response to the specific queries:

• All enrichment heat maps were generated from GO lists or KEGG pathways.
• 2C: these were chosen instead of other myogenic or cell migration markers for consistency with our previous study (Figure 2C in Gerli et al Stem Cell Reports 2019).
• 3A, D, G: details of the GO lists used to generate heat maps were available in the relative figure legend.
• 2E: enrichment pathways – we listed pathways shared between at least 2 of the three groups and with relevance to cellular migration.

  Figure 4F is impossible to interpret without a clear description of how the subnetwork is extracted, was a list of gene list submitted to string, if so which genes and why? Secondly, why are there many nodes with no edges? Is it all of the nodes that are in that GO-Term, if so it needs to be clarified? Was this the most strongly deregulated go-Term according to string analysis?

We thank the Reviewer for this comment. This specific GO list was selected for its highly relevant title/topic, i.e.: “positive regulation of cell migration”. Details on this point could also be found in the specific figure legend, where we specified how the network is extracted and constructed. There are several nodes with no edges as the edges represent predicted functional association and therefore, a lack of edges suggests a lack of interaction.

Figure 4 B, C, D and E: (1) The authors should clarify what figure 4B is? Is 1,2,3,4 different time point? Treated or untreated cells?

We apologise with the Reviewer for not having provided enough information on this point. 1,2,3 and 4 are four sequential time points of untreated cells. We will amend the figure to make this clearer.

(2) Figure C: Is the graph showing the cell distribution of both treated and untreated cells? If yes is it possible to give a different shape for the control cells and see if indeed more control green shape would be observed in this plot? (In the supplementary data there is the distribution showing the treated v untreated, but the clusters are not visible)

We thank the Reviewer for this helpful comment. We agree that this will increase the quality of the figure. We will distinguish treated and control cells within figure 4C by replacing dots with different shapes for treated and untreated samples.

(3) Would it be possible to take some of the parameters in Figure 4D and show the distribution in treated vs untreated and perform the statistical analysis? (eg is there a significant difference for the parameter total distance between control and treated?). Or, may be just show some of the results in figure S4C and E in the main text.

We thank the Reviewer for this comment. We agree that it will be better to move S4C into the main figure and we will action this point in the revised version of the manuscript.

(4) Why pooling the 3 independent experiment together? Looking at the data in Figure S4, it seems that one treated sample is very similar to the control, thus weakening the conclusion. The replicates in this figure are biological replicates. Yet the papers present 4/5 different cell lines, so why only 3 of them are used here? Is there some explanation regarding the outsider (cell line age, number of division etc). Might be worth adding data from the other cell lines (1 or 2 more).

We thank the Reviewer for this point. The experiment shown in figure S4E has been performed with one cell line (N5) and independent experimental replicates were assessed for the statistical analysis. We are not sure why there appears to be an outlier in some cases, and this is why it was important to replicate this experiment three times. However, we will also repeat this experiment with another cell line applying more stringent conditions to strengthen this point.

(5) Figure 4 H and I: What are the statistic actually comparing: treated v untreated for each cell lines or different cell lines against each other? If the former, then how is it possible to have a 139 fold change with such a weak p value of 0.042? If the latter, then why is a p-value given for each of the 3 cell lines? Also, the number and source of replicates is unclear - N=3 is stated, so was each cell line done in triplicate? If so, how many fields per replicate?

We are happy to clarify this point for the Reviewer. The statistical analysis compares treated vs. untreated samples within the same genotype. The high fold change observed is likely due to the large standard deviation of the dataset, which was also highlighted as raw data in the figure panel (bottom part of each picture in white colour font). For this reason, we have repeated this experiment multiple times and validated it across three independent cell lines.

Would additional experiments be essential to support the claims of the paper? Request additional experiments only where necessary for the paper as it is, and do not ask authors to open new lines of experimentation. It would important to also show the migratory capacity of these cells in vivo.

Are the suggested experiments realistic in terms of time and resources? It would help if you could add an estimated cost and time investment for substantial experiments. Human muscle cells engraftment and tracking in immunodeficient mice could be easily done. Engrafted muscle can be harvested 2-3 weeks after engraftment, and measurement of the distance from the engraftment point could be done (Site of injection could be labelled with tattoo die). This would be a month/month and half of work. Immunodeficient mice would cost around £1500 (n=6 mice per group => total of 12 mice) plus the cost of housing.

Are the data and the methods presented in such a way that they can be reproduced? Are the experiments adequately replicated and statistical analysis adequate? See comments in first paragraph. The authors should probably be able to answer easily to the different concerns raised above.

We thank the Reviewer for these comments. We agree that the suggested in vivo experiment might strengthen our work and we are currently sourcing all required materials to perform it. Additionally, we will perform a similar, quasi-vivo, experiment to study migration in a species-specific setting by delivering cells in 3D models in vitro (e.g. Maffioletti SM et al., Cell Reports 2018). This strategy will provide a solid alternative to the in vivo assay, in the eventuality that the xenogeneic setting will limit the resolution of the proposed transplantation experiment.

Minor comments: typo "Onthology" should be "Ontology" in figure 2E. Some of the data in Figure S4E should be moved to the main text.

Thank you for highlighting these minor comments. We will correct the typo and move data from figure S4 into the main figure 4.

Reviewer #2 (Evidence, reproducibility and clarity): In this manuscript, Ferrari and colleagues provide solid data indicating that the Notch ligand DLL4 and PDGF-BB regulate the migration of myogenic progenitors derived from human pluripotent stem cells (PSC). These studies built from recent work by the same group (Gerli et al, Stem Cell Reports, 12:461, 2019), in which the authors documented that Notch and PDGF-BB signaling enhances migration and expression of stem cell markers while inducing perivascular cell features in muscle satellite cells. Here the authors perform similar in vitro studies in PSC-derived myogenic progenitors and conclude that the same effect is observed in this population of cells. The results are clear and well presented.

Throughout the manuscript, the authors emphasize the importance of such findings for the future therapeutic application of a PSC-based therapy to treat patients with muscular dystrophy since multiples skeletal muscles need to be targeted in this group of diseases. Unfortunately, the authors do not provide transplantation data. The results would be highly meaningful if they show that observed in vitro changes (transcriptomes and chamber assay) result in meaningful migration in vivo using the systemic delivery, but as it is, the data do not support the claims and conclusions.

We thank Reviewer 2 for their comments. We were pleased to read that they found our study and data solid, clear and well presented. Although we agree with the Reviewer that in vivo evidence would strengthen our findings, we would like to highlight that our study did not aim to be a translational investigation of the therapeutic potential of treated hPSC derivatives for muscle cell therapy (we believe our manuscript’s title reflects this). We see this work more as a foundational study to establish the required evidence for future, follow up transplantation studies focused on the therapeutic potential of this approach (something requiring a dedicated project, funding and months/years of work).

Moreover, we believe that xenogeneic transplants are of limited use to investigate a complex species-specific phenomenon such as transendothelial cell migration. For this very reason we moved back to intraspecific transplantsin past studies (e.g.. Tedesco et al Sci Transl Med 2012). However, as a key aim of our study is to obtain data specific to human cells and given that we already performed mouse-in-mouse in vivo intra-arterial delivery experiments using DLL4 and PDGFBB treated primary cells in Gerli et al. Stem Cell Reports 2018, we are therefore proposing and planning to:

• Test transendothelial migration with another quasi-vivo microfluidic assay orthogonal to the reported transwell experiments. This will model intraspecific (i.e., human-in-human) transendothelial migration under flow conditions.
• Assess evidence of migration in human 3D muscles setting up a novel invasion assay in our in vitro 3D muscle models.
• Perform intramuscular delivery of treated vs. untreated cells as per Reviewer 1 request to assess migration in skeletal muscle in vivo. This approach will optimise in vivo experiments in a 3Rs compliant fashion, avoiding invasive procedures in animals to study intravascular delivery.

Reviewer #2 (Significance): Significance is limited if only in vitro data are provided. However if the authors are able to show enhanced engraftment upon systemic transplantation of human PSC-derived myogenic progenitors upon DLL4 and PDGF-BB treatment, the significance would be high.

Please see our reply to the previous point.

In terms of existing literature, there are publications reporting systemic delivery of murine PSC-derived myogenic progenitors as well as transcriptome and in vitro migration studies. It would probably be appropriate to cite them.

We apologies to the Reviewer for this oversight. We will add the following papers which include systemic delivery of murine PSC-derived myogenic progenitors as well as transcriptome and migration studies: Matthias N et al., Exp Cell Res 2015; Incitti T et al., PNAS 2019.

If systemic engraftment is observed, the manuscript would be of interest to the skeletal muscle and stem cell biology/regenerative medicine community.

Please see our reply to the initial point.

Reviewer #3 (Evidence, reproducibility and clarity):

In this manuscript, the authors exploited the signal-mediated activation of NOTCH and PDGF pathways, by one week-long delivery of DLL4 and PDGF-BB to cultures of hiPSC-derived myogenic progenitors in vitro, to improve their migration ability. They performed transcriptomic and functional analyses across human and mouse primary muscle stem cells and human hiPSC-derived myoblasts, including genetically corrected hiPSC derivatives, to show that DLL4 and PDGF-BB treatment modulates pathways involved in cell migration, including enhanced trans-endothelial migration in transwell assays.

The increased migratory ability, and in particular enhancing extravasation, is a fundamental property required for optimal performance of hiPSC myogenic derivatives, upon their intravascular delivery; hence, the finding reported here are of extremely high potential interest in term of solution of one of the major bottle-neck of cell therapy. However, there are important issues that need to be resolved by the authors with additional experimentation, that I recommend performing, in order to improve this manuscript.

We sincerely thank the Reviewer for acknowledging the extremely high relevance and potential of our paper for muscle gene and cell therapies and for providing constructive feedback to improve our manuscript.

1) The most critical issue here is that the authors fail to provide evidence that DLL4/PDGF-BB-treated cultures of hiPSC-derived myogenic progenitors do not lose their myogenic potential and are able to form myotubes, upon interruption of treatment. It would be also important to determine when (how many days after withdrawal of DLL4/PDGF-BB) the full myogenic properties of these cells are recovered. From the RNAseq datasets shown by the authors, it appears that DLL4/PDGF-BB-treated hiPSC-derived myoblasts do not express the key genes of myogenic identity (MyoD) and early differentiation (myogenin), while expressing genes of mesenchymal/vessel-derived lineages. It is imperative that the authors show that these changes are reversible, upon withdrawal of DLL4/PDGF-BB. This should be show by an unbiased transcriptomic analysis (RNAseq) of hiPSC-derived myoblasts after withdrawal of DLL4/PDGF-BB, that should be integrated with functional evidence showing that these cells can resume their ability to form differentiated myotubes, upon exposure to myogenic culture cues in vitro.

We thank the Reviewer for this comment. We agree that this is an important and feasible experiment which will add important information to our work. We performed similar work in our previous study and already observed phenotype reversion of treated cells upon release of the stimuli within a few passages in cultures. However, we agree that this requires systematic assessment and quantification. To this aim, we will assess the reversibility of the DLL4 & PDGF-BB effect by stopping treatment at day 7 and then assessing skeletal myogenic differentiation capacity of target cells at sequential passages and time points post-treatment. Analysis of the differentiation index at different time points will provide functional evidence on the myogenic potential of hiPSC-derived myogenic progenitors post-withdrawal of DLL4 & PDGF-BB. We believe that the Reviewer’s suggestion for transcriptomic analysis via RNA-seq might be overly costly for the purpose of identifying the myogenic potential of treated cells post-withdrawal of treatment, and that qPCR panels alongside immunofluorescence staining may be sufficient.

2) A parallel evidence in vivo should be also provided, showing that DLL4/PDGF-BB-treated hiPSC-derived myoblasts do not express MyoD and myogenin when delivered intravascularly, but regain their expression after they have crossed the vessel endothelium and have entered the skeletal muscles.

We thank the Reviewer for suggesting this experiment. We agree that this would be a very interesting point to address; however, it might be very challenging to address this question with the proposed in vivo experiment. Nonetheless, we believe that with a combination of in vitro and in vivo assays we will be able to satisfactorily answer the question: Do DLL4 and PDGF-BB-treated myogenic progenitors re-gain myogenic potential upon entering skeletal muscle tissue? To this aim, we aim to analyse muscles following intramuscular transplantation of treated and untreated cells. Moreover, to model intra-vascular delivery and have high resolution imaging, we aim to adapt a microfluidic platform to perform trans-endothelial assays and selectively differentiate cells that successfully cross the blood vessel layer. Although likely to be very challenging, we will attempt to capture or stain those very cells in order to assess the expression of myogenic markers as requested by the Reviewer.

If these experiments could firmly demonstrate that DLL4/PDGF-BB-treatment reversibly promotes migratory properties of hiPSC-derived myoblasts (as predicted, but not demonstrated in previous works from the same group, using mouse or human primary muscle stem cells - Cappellari et al. 2013; Gerli et al. 2019), then this work could be a great interest in term of basic and translational biology and clearly suitable for publication in a top journal.

We thank the Reviewer for this constructive feedback and for seeing the great potential of our work in terms of basic and translational biology. We assume there was a typo in the sentence in brackets with a missing “as” (“..not demonstrated as in previous work...”): we indeed demonstrated the effect of DLL4 and PDGFBB in vivo extensively in our previous work.

Other points:

• Fig. 2A. it looks like there are some outlier RNAseq sample replicates that might negatively impact at the statistical level on the subsequent analysis. This issue is likely due to the heterogeneity of the samples (both untreated and treated) and could be resolved by replacing outlier samples with new replicates.

We thank the Reviewer for this comment. Although we agree that replacing those samples with new replicates might improve our statistical analyses, this will be financially challenging at this stage and perhaps also not completely reflecting the real variability of the experimental setup.

• Along the same line as above, sample heterogeneity following treatment might be resolved by a better understanding of optimal doses of DLL4/PDGF-BB and time of exposure, which I recommend the authors to define by additional experiments.

We thank the Reviewer for this comment. This is a potentially interesting experiment, which we have not performed as we took advantage of previous knowledge and dose-response on primary mouse and human myoblasts. Overall, we believe that this experiment might not be strictly required at this stage, given that we have already solid evidence of response in hiMPs with a defined concentration and exposure time of DLL4 and PDGFBB.

Reviewer #3 (Significance):

If these experiments could firmly demonstrate that DLL4/PDGF-BB-treatment reversibly promotes migratory properties of hiPSC-derived myoblasts (as predicted, but not demonstrated in previous works from the same group, using mouse or human primary muscle stem cells - Cappellari et al. 2013; Gerli et al. 2019), then this work could be a great interest in term of basic and translational biology and clearly suitable for publication in a top journal and could be interesting for a wide audience in regenerative medicine.

We thank the Reviewer once again for this constructive feedback and for seeing the great potential of our work in terms of basic and translational biology, as well as for regenerative medicine.

Please note that the following statement will be added to the Acknowledgements section of our revised manuscript: "For the purpose of Open Access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission. This work was supported by the Francis Crick Institute which receives its core funding from Cancer Research UK, the UK Medical Research Council, and the Wellcome Trust (FC001002)".

Once again, we sincerely thank all Reviewers for their positive, constructive and insightful comments, which motivate us to further improve our work. We also thank the Review Commons editorial team for guidance and assistance.

Prof. Francesco Saverio Tedesco, University College London and The Francis Crick Institute, London, UK.

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

In this manuscript, the authors exploited the signal-mediated activation of NOTCH and PDGF pathways, by one week-long delivery of DLL4 and PDGF-BB to cultures of hiPSC-derived myogenic progenitors in vitro, to improve their migration ability. They performed transcriptomic and functional analyses across human and mouse primary muscle stem cells and human hiPSC-derived myoblasts, including genetically corrected hiPSC derivatives, to show that DLL4 and PDGF-BB treatment modulates pathways involved in cell migration, including enhanced trans-endothelial migration in transwell assays. The increased migratory ability, and in particular enhancing extravasation, is a fundamental property required for optimal performance of hiPSC myogenic derivatives, upon their intravascular delivery; hence, the finding reported here are of extremely high potential interest in term of solution of one of the major bottle-neck of cell therapy. However, there are important issues that need to be resolved by the authors with additional experimentation, that I recommend performimg, in order to improve this manuscript.

1) The most critical issue here is that the authors fail to provide evidence that DLL4/PDGF-BB-treated cultures of hiPSC-derived myogenic progenitors do not lose their myogenic potential and are able to form myotubes, upon interruption of treatment. It would be also important to determine when (how many days after withdrawal of DLL4/PDGF-BB) the full myogenic properties of these cells are recovered. From the RNAseq datasets shown by the authors, it appears that DLL4/PDGF-BB-treated hiPSC-derived myoblasts do not express the key genes of myogenic identity (MyoD) and early differentiation (myogenin), while expressing genes of mesenchymal/vessel-derived lineages. It is imperative that the authors show that these changes are reversible, upon withdrawal of DLL4/PDGF-BB. This should be show by an unbiased transcriptomic analysis (RNAseq) of hiPSC-derived myoblasts after withdrawal of DLL4/PDGF-BB, that should be integrated with functional evidence showing that these cells can resume their ability to form differentiated myotubes, upon exposure to myogenic culture cues in vitro.

2) A parallel evidence in vivo should be also provided, showing that DLL4/PDGF-BB-treated hiPSC-derived myoblasts do not express MyoD and myogenin when delivered intravascularly, but regain their expression after they have crossed the vessel endothelium and have entered the skeletal muscles. If these experiments could firmly demonstrate that DLL4/PDGF-BB-treatment reversibly promotes migratory properties of hiPSC-derived myoblasts (as predicted, but not demonstrated in previous works from the same group, using mouse or human primary muscle stem cells - Cappellari et al. 2013; Gerli et al. 2019), then this work could be a great interest in term of basic and translational biology and clearly suitable for publication in a top journal.

Other points:

• Fig. 2A. it looks like there are some outlier RNAseq sample replicates that might negatively impact at the statistical level on the subsequent analysis. This issue is likely due to the heterogeneity of the samples (both untreated and treated) and could be resolved by replacing outlier samples with new replicates.
• Along the same line as above, sample heterogeneity following treatment might be resolved by a better understanding of optimal doses of DLL4/PDGF-BB and time of exposure, which I recommend the authors to define by additional experiments.

Significance

If these experiments could firmly demonstrate that DLL4/PDGF-BB-treatment reversibly promotes migratory properties of hiPSC-derived myoblasts (as predicted, but not demonstrated in previous works from the same group, using mouse or human primary muscle stem cells - Cappellari et al. 2013; Gerli et al. 2019), then this work could be a great interest in term of basic and translational biology and clearly suitable for publication in a top journal and could be interesting for a wide audience in regenerative medicine.

Expertise of this reviewer:

Muscle regeneration; Muscular Dystrophies; Signaling and Epigenetics

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

In this manuscript, Ferrari and colleagues provide solid data indicating that the Notch ligand DLL4 and PDGF-BB regulate the migration of myogenic progenitors derived from human pluripotent stem cells (PSC). These studies built from recent work by the same group (Gerli et al, Stem Cell Reports, 12:461, 2019), in which the authors documented that Notch and PDGF-BB signaling enhances migration and expression of stem cell markers while inducing perivascular cell features in muscle satellite cells. Here the authors perform similar in vitro studies in PSC-derived myogenic progenitors and conclude that the same effect is observed in this population of cells. The results are clear and well presented.

Throughout the manuscript, the authors emphasize the importance of such findings for the future therapeutic application of a PSC-based therapy to treat patients with muscular dystrophy since multiples skeletal muscles need to be targeted in this group of diseases. Unfortunately, the authors do not provide transplantation data. The results would be highly meaningful if they show that observed in vitro changes (transcriptomes and chamber assay) result in meaningful migration in vivo using the systemic delivery, but as it is, the data do not support the claims and conclusions.

Significance

Significance is limited if only in vitro data are provided. However if the authors are able to show enhanced engraftment upon systemic transplantation of human PSC-derived myogenic progenitors upon DLL4 and PDGF-BB treatment, the significance would be high.

In terms of existing literature, there are publications reporting systemic delivery of murine PSC-derived myogenic progenitors as well as transcriptome and in vitro migration studies. It would probably be appropriate to cite them.

If systemic engraftment is observed, the manuscript would be of interest to the skeletal muscle and stem cell biology/regenerative medicine community.

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:

The paper presented by Ferrari et al., aims to improve the migration capacity of hiPSC- derived myogenic progenitors. For this purpose, the authors used a previously published well characterized hiMPs model and focussed on the modulation of NOTCH and PDGF signaling pathways. The rational to target these pathways was based on muscle cells migrations molecular events observed during developmental described in the literature.

Major comments:

• Are the key conclusions convincing? This is a very interesting paper. Few clarifications as suggested below need to be done before being fully convincing. Enrichment test and heat maps and the network analysis are not well explained in terms of which genes were selected and why, and in terms of which gene set were selected and why. In some cases, the information may be given in the paper, but it is not easy for the reader to find it. It should be stated more clearly. For example, in Fig2C why these eight were chosen for the heat maps and why not other genes known to be involved in myogenesis, cell migration etc. Similar comment for figure 3 A, D and G. Another example, in Fig 2E, on what basis are some gene sets chosen to be shown in this figure when there are many more significant in the supplementary table. Figure 4F is impossible to interpret without a clear description of how the subnetwork is extracted, was a list of gene list submitted to string, if so which genes and why? Secondly, why are there many nodes with no edges? Is it all of the nodes that are in that GO-Term, if so it needs to be clarified? Was this the most strongly deregulated go-Term according to string analysis? Figure 4 B, C, D and E:

(1) The authors should clarify what figure 4B is? Is 1,2,3,4 different time point? Treated or untreated cells?

(2) Figure C: Is the graph showing the cell distribution of both treated and untreated cells? If yes is it possible to give a different shape for the control cells and see if indeed more control green shape would be observed in this plot? (In the supplementary data there is the distribution showing the treated v untreated, but the clusters are not visible)

(3) Would it be possible to take some of the parameters in Figure 4D and show the distribution in treated vs untreated and perform the statistical analysis? (eg is there a significant difference for the parameter total distance between control and treated?). Or, may be just show some of the results in figure S4C and E in the main text.

(4) Why pooling the 3 independent experiment together? Looking at the data in Figure S4, it seems that one treated sample is very similar to the control, thus weakening the conclusion. The replicates in this figure are biological replicates. Yet the papers present 4/5 different cell lines, so why only 3 of them are used here? Is there some explanation regarding the outsider (cell line age, number of division etc). Might be worth adding data from the other cell lines (1 or 2 more).

(5) Figure 4 H and I: What are the statistic actually comparing: treated v untreated for each cell lines or different cell lines against each other? If the former, then how is it possible to have a 139 fold change with such a weak p value of 0.042? If the latter, then why is a p-value given for each of the 3 cell lines? Also, the number and source of replicates is unclear - N=3 is stated, so was each cell line done in triplicate? If so, how many fields per replicate?

• Would additional experiments be essential to support the claims of the paper? Request additional experiments only where necessary for the paper as it is, and do not ask authors to open new lines of experimentation. It would important to also show the migratory capacity of these cells in vivo. -Are the suggested experiments realistic in terms of time and resources? It would help if you could add an estimated cost and time investment for substantial experiments. Human muscle cells engraftment and tracking in immunodeficient mice could be easily done. Engrafted muscle can be harvested 2-3 weeks after engraftment, and measurement of the distance from the engraftment point could be done (Site of injection could be labelled with tattoo die). This would be a month/month and half of work. Immunodeficient mice would cost around £1500 (n=6 mice per group => total of 12 mice) plus the cost of housing.
• Are the data and the methods presented in such a way that they can be reproduced? Are the experiments adequately replicated and statistical analysis adequate?

See comments in first paragraph. The authors should probably be able to answer easily to the different concerns raised above.

Minor comments:

typo "Onthology" should be "Ontology" in figure 2E. Some of the data in Figure S4E should be moved to the main text.

Significance

Describe the nature and significance of the advance, existing literature, audience: Generating iPSC cell lines with an improved capacity to migrate will be of high interest for the neuromuscular field, and could be a potential therapeutic strategy applicable for many neuromuscular disorders.

Muscle cell engraftment is quite challenging as the capacity of these cells to populate different muscles is very poor. Improving the cell migration, survival and proliferation may thus help to improve the muscle cell engraftment strategy.

Expertise:

I have an expertise in neuromuscular disorders, muscle stem cells (human and murine, in vitro and in vivo), as well as an expertise in omics analysis.

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

Learn more at Review Commons

---

Reply to the reviewers

Point-by-point response to reviewer comments

---

Reviewer #1 (Evidence, reproducibility and clarity (Required)):

In the current manuscript, Millarte et al reports a novel role of Rabaptin5 in selectively clearing damaged endosomes via canonical autophagy. They have identified FIP200 as a novel interactor of Rabaptin5 under basal conditions using yeast-two hybrid screening and further confirmed the interaction of Rabaptin5 with FIP200 with immunoprecipitation. They next used Chloroquine and monitored colocalization of the Rabaptin5 with WIPI2, ATG16L1 and LC3B to demonstrate the potential interaction of Rabaptin5 with the autophagic machinery. They have primarily used Gal-3 as a marker of membrane damage after 30 minutes of Chloroquine treatment. In order to further elucidate the role of Rabaptin5 in autophagic induction mediated by Chloroquine, they have silenced Rabaptin5, FIP200, ULK1 and ATG13 and observed a decrease in the number of LC3 or WIPI2 autophagosome formation. Based on these observations they tested if Rabaptin5 interacts with ATG16L1 upon Chloroquine treatment and confirmed their interaction with potential interaction sites of both Rabaptin5 with ATG16L1 with IP. The authors confirmed the interaction of Rabaptin5 with ATG16L1 by complementing the KO line with the mutant form of Rabaptin5 containing alanine residues in its consensus motif. Finally, they have used Salmonella and SCV as a model to study the role of Rabaptin5 in endomembrane damage and monitored a 50% decrease in the removal of Salmonella in Rabaptin5 KO or KD cells.

Major concerns One of the major concerns is the membrane damage reported by chloroquine which is known to induce lysosomal swelling and further targeting of the swollen compartments to degradation by direct conjugation of LC3 onto single membrane as a form of non-canonical autophagy. The evidence regarding membrane damage by Gal3 colocalization on the Rabaptin5 vesicles is preliminary. As suggested by the authors the canonical autophagy pathway recognizing damaged membranes recruits also ALIX to the damaged membrane which was not observed in Supplementary Figure 2. The link to membrane damage by chloroquine and monensin with Rabaptin5 is not convincing as there is not sufficient evidence of membrane damage. In relation to this issue authors should consider using other damage markers as Gal8, p62 or NDP52 to provide additional claim with respect to membrane damage induced by chloroquine.

To expand on the question of CQ treatment damaging early endosomes, we also tested for Gal8 on Rabaptin5-positive enlarged endosomes and quantified the fraction of Rabaptin5-positive rings positive for Gal3 and Gal8 after 30 min of CQ treatment. We propose to include this data in Figure 2:

• *

*

• *

We have tested the importance of Gal3 and p62 by siRNA-mediated knockdown where we found a robust inhibition of induction of WIPI2 puncta with CQ, but not with Torin1. Formation of LC3 puncta was less reduced, similar to knockdowns of FIP200, ATG13, or Rabaptin5.

We propose to add these knockdown experiments as a supplementary figure:

• *

• *

*

One of the main claims here is that Rabaptin5 regulates the targeting of damaged endosomes to autophagy. Clearly, these are early endosomes as stated in the abstract. However, the evidence presented here showing these are early endosomes is not convincing. Analysing Gal3 and Gal8 positive vesicles that are Rabaptin5 positive and an early endosomal marker will be important in this context. For example, there need to be additional evidence showing that early endosomes are targeted to autophagy. Is the degradation of TfR affected by this targeting? Did the authors look at the effect of Bafilomycin A1? If this process affects exclusively early endosomes, it should be BafA1 independent. This will direct more into the cellular function of this process.

Rabaptin5 is a bona fide marker of Rab5-positive early sorting endosomes. As a control, we confirmed colocalization of Rabaptin5 with transferrin receptor, another endosomal marker, on CQ-induced rings (Fig. 2B). We now also analyzed swollen endosomes with triple-staining for Rabaptin5, transferrin receptor, and Gal3 as shown in this gallery (30 min CQ, as in Fig. 2). All Rabaptin5-positive swollen endosomes (rings) were positive for transferrin receptor and ~80% for mCherry-Gal3.

• *

*

• *

We further tested transferrin receptor levels with and without CQ. Since CQ inhibits autophagic flux, this assay may not be very sensitive. Nevertheless, we found a significant reduction of ~15% and ~30% after overnight incubation with CQ in parental HEK293A cells and in Rbpt5-KO cells re-expressing wild-type Rabaptin5, resp., but no reduction in Rbpt5-KO cells expressing the Rabaptin5-AAA mutant defective in binding to ATG16L1:

• *

*

• *

As to the effect of BafA1, see our general response on top. The osmotic effect of CQ or Mon on endosomes that leads to membrane breakage requires an acidic pH. Preincubation with BafA1 neutralizes the pH, prevents osmotic swelling by CQ/Mon, and was shown to block LC3 lipidation (Florey et al., 2015, Jacquin et al., 2017). When BafA1 was added simultaneously, CQ was found to induce LC3 despite the presence of BafA1 (Mauthe et al., 2018), and Mon was shown to still be able to break endosomal membranes and recruit LC3 to EEA1-positive endosomes (Fraser et al., 2019). However, CQ-induced LC3 recruitment to latex bead-containing phagosomes or entotic vacuoles, i.e. LAP-like autophagy, was blocked (Florey et al., 2015). Consistent with this literature, we found increased LC3B lipidation already within 30 min of CQ treatment independently of BafA1 (no preincubation).

• *

*

• *

Upon longer incubations, LC3B lipidation is very strong already with BafA1 alone so that the effect of CQ cannot be assessed anymore, since both drugs inhibit autophagic flux.

Furthermore, we found a CQ-dependent increase in WIPI2- and LC3B-positive puncta to be insensitive to BafA1 (panel A below). Colocalization of Rabaptin5 to LC3B and LC3B to Rabaptin5 significantly increased upon CQ treatment independently of the presence of BafA1 (no pretreatment), indicating that at least a large part of CQ-induced LC3B puncta is not due to LAP-like autophagy.

• *

*

Minor concerns Both for Figure 2 and Supplementary Figure 7 it will be clearer to have the images in colour rather than black and white for better interpretation.

We thought the grayscale images were clearer, but are happy to provide color images.

The interaction of FIP200 and ATG16L1 with Rabaptin5 is well characterized with immunoprecipitation and imaging but the interaction of Rabaptin5 in presence of chloroquine with FIP200 and ATG16L1 DWD are missing and it will be important to include if in the presence of chloroquine these interactions will increase or not.

We can do co-IP experiments also upon CQ treatment.

In order to further support the role of Rabaptin5 for LC3 lipidation upon chloroquine induced membrane damage, western blots of WT, +Rabaptin5, Rabaptin5 KO, Rabaption5 KO +WT or +AAA cell lines were analysed. However, the lysates were collected upon 30 minutes of chloroquine treatment which does not correlate with the imaging performed in Figure 2 as the number of LC3 vesicles did not show an increase upon 30 minutes of chloroquine treatment. The authors should include the 150 minutes time point for the LC3 lipidation in these conditions.

Because CQ inhibits autophagic flux, LC3-II accumulates after longer times in all cell lines. The differences can only be seen early.

The experiments with Salmonella are of great quality. The relationship of Rabaptin5 with SCV and the endomembrane damage induced by Salmonella could be further elucidated with Rabaptin5 positive vesicles at early infection stages. It is not very clear from the text how authors link the endosomal network previously described for chloroquine with infection. It would be important here to show that Salmonella mutants unable to damage endosomal membranes do not have an effect. In Figure 7 panel C, the time points on graphs are in hours but it should be in minutes. corrected.

Since Salmonella require T3SS for infection of HEK cells and T3SS causes the membrane damage, the proposed experiment is difficult.

The events of targeting the damaged membranes for degradation was well characterized by the recognition of these membranes by Gal3, Gal8 and recruitment of autophagic receptors to the site of damage (Chauhan et al. 2016; Jia et al. 2019; Thurston et al. 2012; Maejima et al. 2013; Kreibich et al. 2015). This manuscript introduces a new potential platform for the formation of autophagic machinery on endosomes with the interaction of Rabaptin5 with FIP200 and ATG16L1, however more evidence is required to link this to the clearance of damaged membranes. Previously it was shown that endolysosomal compartments that were neutralized and swollen by monensin and chloroquine had been directed to degradation by direct conjugation of LC3 to single membranes via noncanonical autophagy, but here authors propose another mechanism for this event via canonical autophagy.

As discussed in the general response above, the literature reports CQ and Mon to initiate both canonical autophagy and LAP-like autophagy, the latter particularly on phagosomes containing latex beads or entotic vacuoles. Our results – including the additional data above –concern the effects of CQ and Mon damaging early endosomes and causing recruitment of galectins and ubiquitination, triggering autophagy dependent on the ULK complex and WIPI2 as hallmarks of canonical autophagy, and Rabaptin5. The reviewer comments highlighted the possibility of LAP-like autophagy occurring in parallel, perhaps on endosomes that are not broken, which might explain the relative insensitivity of LC3 puncta induced by CQ and Mon – compared to the strong and robust reduction of WIPI2 puncta – on the knockdown of FIP200, ATG13, or Rabaptin5. In an alternative explanation, inhibition of autophagic flux causes remaining canonical autophagy to accumulate, while WIPI2 puncta are strongly inhibited. In support of the latter interpretation, ULK inhibition by MRT68921 (Fig. 4C and D) or FIP200 knockout (Fig. 6B and C) abolished CQ-induced LC3 structures, suggesting that – unlike on phagosomes or entotic vacuoles – there is little LAP-like autophagy. We propose to revise the manuscript to discuss these considerations more clearly.

Reviewer #1 (Significance (Required)):

Overall this work is very novel and shows some evidence of early endosomal autophagy. It could be relevant for some for of receptor-mediated signalling (although it is not discussed by the authors) My experience is in intracellular trafficking of pathogens and membrane damage.

**Referee Cross-commenting**

In my opinion, the only way you can distinguish between double or single membrane is by EM. For me, the important part is to show this is targeting of early endosomes to autophagy, either using other early endosomal markers, analysing by WB some early endosome receptors such as TfR or other inhibitors. If the authors are able to address some these comments, I agree the paper will be in a better position for publication.

---

Reviewer #2 (Evidence, reproducibility and clarity (Required)):

Millarte et al study the role of Radaptin-5 (Rbpt5) during early endosome damage recognition by autophagy. The authors focus on using chloroquine (CQ) as an agent to induce endosomal swelling/damage and suggest that Rbpt5 is required for the recruitment of the autophagy machinery to perturbed endosomes. They further use salmonella infection as a model to confirm the role of Rbpt5 in this process. The authors initially show that Rbpt5 binds to FIP200 and subsequently focus on its interaction with ATG16L1 and identify a mutant that is unable to bind ATG16L1 or mediate the recognition of early endosomes by autophagy. Overall, this is an interesting study which provides molecular insights of how early endosomes can be targeted by the autophagy machinery where Rbpt5 may act as an autophagy receptor. Some specific comments are as follows:

Fig.3A: siRbpt5 seems to induce the localization of LC3 to ring-like structures during CQ treatment. Are these LAP-like structures (e.g. sensitive to BafA1 treatment)? And were they included in the quantification in Fig.3C?

Ring-like LC3 structures were also counted.

As discussed in the general remarks above, it is a possibility that knockdown-resistent LC3 recruitment (particularly rings) is due to a CQ-induced LAP-like process. The alternative explanation is that there is residual canonical autophagy upon knockdown of Rabaptin5, ATG13, or FIP200: while WIPI2 puncta are strongly reduced, LC3-positive structures accumulate due to inhibition of autophagic flux. In support of the latter interpretation, ULK inhibition by MRT68921 (Fig. 4C and D) or FIP200 knockout (Fig. 6B and C) abolished CQ-induced LC3 puncta or rings.

We can also test BafA1 treatment. Certainly, we will revise the text to discuss this point in more detail.

• *

Fig.4A&B: Since Rbpt5 KD has a weak effect on LC3 puncta formation (Fig.3) and to distinguish the effects of CQ in inducing LAP, the effects of ATG13 and ULK1 KD should be assessed by localising Rbpt5 with WIPI2 or ATG16L1.

We can do that.

Fig.4: It is not clear why ULK1 KD would affect Torin1-induced autophagy but not LC3/WIPI2 localisation during CQ induced early endosome-damage. As the ULK inhibitors can target other pathways, the authors should confirm this finding in ULK1/2 double KO or KD cells.

We have used **MRT68921, because it is frequently used in the literature for this purpose with high specificity. It was used for example by Lystad et al. (2019) together with VPS34IN1 to block all canonical autophagy to analyze exclusively noncanonical effects of monensin treatment. We could perform ULK1/2 double knockdowns, but since ULK2 cannot be detected on immunoblots in HEK293 cells, the result would be interpretable only when there is an effect.

Fig.5: The contribution of FIP200 in the interaction between Rbpt5 and ATG16L1 is unclear. Is binding between Rbpt5 and ATG16L1 mediated by ATG16L1's interaction with FIP200? The plasmid details describing the delta-WD40 deletion plasmid used in this study are missing and could be important to confirm that the detla-WD40 still retains binding to FIP200.

We will of course include the details on the deletion plasmid, which were missing by mistake. Our WD deletion construct of ATG16L1 consists of residues 1–319, precisely deleting just the WD40 repeats, but retaining the FIP200 interaction sequence and the second membrane binding segment (b).

We did a co-immunoprecipitation experiment and found both wild-type ATG16L1 and the ∆WD mutant to co-immunoprecipitate with FIP200:

• *

*

Fig.5E: the authors should test Rbpt5 AAA mutant binding to FIP200. Since the mutant appears to express less, its binding to ATG16L1 should be quantified or repeated with more comparable expression levels.

We will quantify the immunoblots and perhaps attempt getting more equal expression levels.

Fig.6: CQ treatment can induce various endosomal damage (in addition to early endosomes) and LC3 lipidation processes (e.g. LAP-like). The authors show that Rbpt5 is specifically involved in damaged early endosome autophagy. In this figure, it would be important to distinguish CQ-induced LC3 puncta as a result of early endosome damage or other lipidation processes (e.g. canonical or non-canonical autophagy). The use of FIP200 KO cells shows that LC3 puncta is inhibited. However, here a specific readout to look at early endosome recognition by autophagy is important. The authors can localize early endosome markers (EEA1) with autophagy players (e.g. WIPI2 and LC3). This is also relevant to other figures (e.g. supplementary figure 7E).

Rabaptin5 is a bona fide marker of Rab5-positive early sorting endosomes. As a control, we confirmed colocalization of Rabaptin5 with transferrin receptor, another endosomal marker, on CQ-induced rings (Fig. 2B). We also analyzed swollen endosomes with triple-staining for Rabaptin5/ transferrin receptor/ Gal3 as shown in this gallery (30 min CQ, as in Fig. 2). All Rabaptin5-positive swollen endosomes (rings) were positive for transferrin receptor and ~80% for mCherry-Gal3.

• *

*

• *

Our results are in agreement with Fraser et al. (2019) where they use EEA1 as an endosomal marker upon monensin treatment.

We also performed a colocalization analysis for Rabaptin5 and LC3B, showing enhanced colocalization after CQ treatment for 150 min: ~20% of LC3B is (still) pos for Rabaptin5 after 150 min of CQ treatment:

*

Fig.6F&G: the authors should show representative images of these localization images quantified here. These can be added in the supplementary figures.

We are happy to do this.

**Minor comments:**

Fig.2E: FIP200 seems to be highly overexpressed in this image. Commercial antibodies that recognise endogenous FIP200 are widely used and should be tested to confirm the colocalisation between FIP200 and Rbpt5.

We plan to do this.

Fig.7C image: the different setting denoted by +/-, +/+ ..etc are not clearly defined.

We will improve this.

Reviewer #2 (Significance (Required)):

This is a interesting study and provides important mechanistic insights underlying the recognition of perturbed early endosomes by the autophagy machinery. Researchers interested in endosomal trafficking or autophagic substrate recognition are likely to benefit from this study.

**Referee Cross-commenting**

In my opinion, the authors have attempted to distinguish single membrane from double membrane LC3 lipidation by looking at the ULK complex requirement. As other reviewers suggested, this can be further confirmed by using ATG16L1 mutants. It is important however that these experiments are supplemented by co-localising autophagy proteins with alternative early endosome markers when Rbpt5 is inhibited.

I think if the authors are able to address the suggested experiments, this would help improve the manuscript and make it suitable for publication.

---

Reviewer #3 (Evidence, reproducibility and clarity (Required)):

Millarte and colleagues find that Rabaptin5, a critical regulator of Rab5 activity, and a protein localized to early endosomes, interacts with FIP200 and ATG16L1. This interaction is confirmed and validated by a number of approaches (yeast 2 H, co-immunoprecipitation) and the binding sites of Rabaptin5 are mapped on FIP200 and ATG16L1. More precisely the binding site for ATG16L1 is nicely mapped on Rabaptin 5 by analogy with other ATG16L1 binders. The authors investigate the significance of this binding of Rabaptin5 to the autophagy proteins by proposing this interaction is required for targeting "autophagy to damaged endosomes". Endosomes are damaged with short treatments of chloroquine, a well studied compound previously shown to inhibit autophagy by disrupting fusion of autophagosomes with lysosomes. They propose the recruitment of autophagy (proteins) to the damaged endosomes may allow them to be eliminated. They use another model (phagocytosis of salmonella) to probe the role for rabaptin5 and its partners FIP200 and ATG16L1 in the well-documented role of autophagy on the elimination of salmonella in SCVs (Salmonella containing vacuole) formed from endosomes. Using short infection time points, and the Rabaptin5 mutants which prevent ATG16L1 binding they suggest Rabaptin5 binding contributes to elimination and killing of Salmonella by recruitment of ATG16L1.

**Major comments:**

1. The authors make an unfortunate and confusing choice of wording in the title and the text of "autophagy being recruited" to damaged early endosomes. A protein can recruit another protein but it can not recruit a process or pathway to a membrane.

In the title we use the term "target". It is OK for us to avoid the expression "recruiting autophagy".

The authors conclude that Rabaptin5 may have a role in autophagy directed to damaged early endosomes. The conclusion that Rabaptin5 binds FIP200 and ATG16L1 are convincing. The main issue is however in identifying what sort of process they are following. They have shown WIPI2 and LC3 can be recruited to early endosomes after 30 min chloroquine treatment but there is no data to explain the consequences of the binding of these proteins. They do not provide proof that canonical autophagosomes are formed which engulf and remove the damaged endosomes, nor do they show that the recruitment of WIPI2 is to a single membrane (presumably damaged early endosomes) which would be a non-canonical pathway. They often use the terminology "chloroquine-induced autophagy" (see Figure 4) but have virtually no proof they have induced either canonical or non-canonical pathways in their experiments. The only evidence they provide that there is some alteration in a membrane-mediated event is increase in lipidation of LC3 in Figure 6. The authors must follow either an early endosome protein or cargo to demonstrate lysosome-mediated degradation indicative of autophagy, or demonstrate the process is a variation on non-canonical autophagy.

We analyzed transferrin receptor levels with and without CQ to test degradation of an early endosomal marker protein. Since CQ inhibits autophagic flux, this assay may not be very sensitive. Nevertheless, we found a significant reduction of ~15% and ~30% after overnight incubation with CQ in parental HEK293 cells and in Rbpt5-KO cells re-expressing wild-type Rabaptin5, resp., but no reduction in Rbpt5-KO cells expressing the Rabaptin5-AAA mutant defective in binding to ATG16L1:

• *

*

There are concerns about the replicates done for many experiments in particular the co-immunoprecipitations which are not quantified (Figure 1 and 5).

We will quantify these blots.

The rescue experiments, even if done with stable cells lines made in the parental HEK293 cell line should be viewed with caution because of the very different amounts of Rabaptin5 (see Figure 6A). The overexpression of Rabaptin5 has not been well studied and comparisons with the mutants are therefore preliminary (Figure 6F and G).

Fig 6A shows that Rabaptin5 levels are similar except for +Rbpt, where they are higher, and R-KO, which has none. Additional Rabaptin5 seems not to significantly enhance early WIPI and ATG16L1 colocalization.

Conclusions about the role of the ULK complex, or ULK1 versus ULK2, should be expanded by studying the activity of the complex (phosphorylation of ATG13 for example) in order to make the conclusions more significant.

We consider this to be beyond the scope of this study. Rabaptin5-dependent autophagy depends on the components of the ULK complex.

**Minor comments:**

1. Much of the labelling in the immunofluorescence images is not visible even on the screen version.

We were careful to have the signals within the dynamic range of the image, but we can enhance the signals for better visibility.

The LC3-lipidation experiment (Figure 6D) should be re-analysed by normalization to the loading control. The result may be significantly different and is open to re-interpretation. The quality of this western blot is also very poor.

Quantitation was based on the ratio between LC3B-I and -II or the **percentage of II of the total, always within the same lane and therefore largely independent of loading.

Reviewer #3 (Significance (Required)):

This manuscript topic fits into the field of study of canonical versus non-canonical autophagy. This literature is best described as "LAP" first discovered by Doug Green, (Sanjuan in 2009) but more recently as a phenomena induced by monesin, and viral infection amongst others. Most relevant to this study are the references (in the text) by Florey (Autophagy 2015), Fletcher (EMBO J, 2018) and others. However, this manuscript fails to cite and consider the critical findings in a key study published by Lystad et al., Nature Cell Biology 2019, which examines the role of ATG16 in both canonical and non-canonical autophagy. The current study if placed into the context of the Lystad study would have significantly more value, and potentially make the findings more significant.

We did not refer to Lystad et al. (2019), because they analyzed different ATG16L1 mutants on their contribution to monensin-induced processes on LC3 lipidation after completely blocking canonical autophagy with the ULK inhibitor MRT68921 and the VPS34 inhibitor VPS34IN1. The Rabaptin5-dependent CQ-induced processes are blocked by MRT68921 (Fig. 4C). We plan to refer to this study in the revision.

Furthermore, the short chloroquine treatments used here could be of interest to the field if using the cited study of Mauthe et al., (which very clearly defines the effect of chloroquine after long (5 hrs treatment)) the authors would to revisit and repeat some of the key experiments in order to demonstrate the effects of 30 minute treatment. Does such short treatment block fusion? Does it affect the pH of the acidic compartments? Does it inactivate the endocytitic pathway? As the manuscript stands the lack of this understanding of the effect of chloroquine at short times, makes the observations difficult to be place into any biological context.

This reviewer has expertise in autophagy, autophagosome formation and is familiar with the areas of endocytosis and infection.

**Referee Cross-commenting**

I think a major concern about the manuscript which is present in all reviews is the lack of clarity about what type of membrane LC3 is added to- is this the damaged endosome or a forming autophagosome? This leads to the question of what type of process is being observed here? non-canonical versus canonical autophagy.

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

Millarte and colleagues find that Rabaptin5, a critical regulator of Rab5 activity, and a protein localized to early endosomes, interacts with FIP200 and ATG16L1. This interaction is confirmed and validated by a number of approaches (yeast 2 H, co-immunoprecipitation) and the binding sites of Rabaptin5 are mapped on FIP200 and ATG16L1. More precisely the binding site for ATG16L1 is nicely mapped on Rabaptin 5 by analogy with other ATG16L1 binders. The authors investigate the significance of this binding of Rabaptin5 to the autophagy proteins by proposing this interaction is required for targeting "autophagy to damaged endosomes". Endosomes are damaged with short treatments of chloroquine, a well studied compound previously shown to inhibit autophagy by disrupting fusion of autophagosomes with lysosomes. They propose the recruitment of autophagy (proteins) to the damaged endosomes may allow them to be eliminated. They use another model (phagocytosis of salmonella) to probe the role for rabaptin5 and its partners FIP200 and ATG16L1 in the well-documented role of autophagy on the elimination of salmonella in SCVs (Salmonella containing vacuole) formed from endosomes. Using short infection time points, and the Rabaptin5 mutants which prevent ATG16L1 binding they suggest Rabaptin5 binding contributes to elimination and killing of Salmonella by recruitment of ATG16L1.

Major comments:

1. The authors make an unfortunate and confusing choice of wording in the title and the text of "autophagy being recruited" to damaged early endosomes. A protein can recruit another protein but it can not recruit a process or pathway to a membrane.
2. The authors conclude that Rabaptin5 may have a role in autophagy directed to damaged early endosomes. The conclusion that Rabaptin5 binds FIP200 and ATG16L1 are convincing. The main issue is however in identifying what sort of process they are following. They have shown WIPI2 and LC3 can be recruited to early endosomes after 30 min chloroquine treatment but there is no data to explain the consequences of the binding of these proteins. They do not provide proof that canonical autophagosomes are formed which engulf and remove the damaged endosomes, nor do they show that the recruitment of WIPI2 is to a single membrane (presumably damaged early endosomes) which would be a non-canonical pathway. They often use the terminology "chloroquine-induced autophagy" (see Figure 4) but have virtually no proof they have induced either canonical or non-canonical pathways in their experiments. The only evidence they provide that there is some alteration in a membrane-mediated event is increase in lipidation of LC3 in Figure 6. The authors must follow either an early endosome protein or cargo to demonstrate lysosome-mediated degradation indicative of autophagy, or demonstrate the process is a variation on non-canonical autophagy.
3. There are concerns about the replicates done for many experiments in particular the co-immunoprecipitations which are not quantified (Figure 1 and 5).
4. The rescue experiments, even if done with stable cells lines made in the parental HEK293 cell line should be viewed with caution because of the very different amounts of Rabaptin5 (see Figure 6A). The overexpression of Rabaptin5 has not been well studied and comparisons with the mutants are therefore preliminary (Figure 6F and G).
5. Conclusions about the role of the ULK complex, or ULK1 versus ULK2, should be expanded by studying the activity of the complex (phosphorylation of ATG13 for example) in order to make the conclusions more significant.

Minor comments:

1. Much of the labelling in the immunofluorescence images is not visible even on the screen version.
2. The LC3-lipidation experiment (Figure 6D) should be re-analysed by normalization to the loading control. The result may be significantly different and is open to re-interpretation. The quality of this western blot is also very poor.

Significance

This manuscript topic fits into the field of study of canonical versus non-canonical autophagy. This literature is best described as "LAP" first discovered by Doug Green, (Sanjuan in 2009) but more recently as a phenomena induced by monesin, and viral infection amongst others. Most relevant to this study are the references (in the text) by Florey (Autophagy 2015), Fletcher (EMBO J, 2018) and others. However, this manuscript fails to cite and consider the critical findings in a key study published by Lystad et al., Nature Cell Biology 2019, which examines the role of ATG16 in both canonical and non-canonical autophagy. The current study if placed into the context of the Lystad study would have significantly more value, and potentially make the findings more significant.

Furthermore, the short chloroquine treatments used here could be of interest to the field if using the cited study of Mauthe et al., (which very clearly defines the effect of chloroquine after long (5 hrs treatment)) the authors would to revisit and repeat some of the key experiments in order to demonstrate the effects of 30 minute treatment. Does such short treatment block fusion? Does it affect the pH of the acidic compartments? Does it inactivate the endocytitic pathway? As the manuscript stands the lack of this understanding of the effect of chloroquine at short times, makes the observations difficult to be place into any biological context.

This reviewer has expertise in autophagy, autophagosome formation and is familiar with the areas of endocytosis and infection.

Referee Cross-commenting

I think a major concern about the manuscript which is present in all reviews is the lack of clarity about what type of membrane LC3 is added to- is this the damaged endosome or a forming autophagosome? This leads to the question of what type of process is being observed here? non-canonical versus canonical autophagy.

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

Millarte et al study the role of Radaptin-5 (Rbpt5) during early endosome damage recognition by autophagy. The authors focus on using chloroquine (CQ) as an agent to induce endosomal swelling/damage and suggest that Rbpt5 is required for the recruitment of the autophagy machinery to perturbed endosomes. They further use salmonella infection as a model to confirm the role of Rbpt5 in this process. The authors initially show that Rbpt5 binds to FIP200 and subsequently focus on its interaction with ATG16L1 and identify a mutant that is unable to bind ATG16L1 or mediate the recognition of early endosomes by autophagy. Overall, this is an interesting study which provides molecular insights of how early endosomes can be targeted by the autophagy machinery where Rbpt5 may act as an autophagy receptor. Some specific comments are as follows:

Fig.3A: siRbpt5 seems to induce the localization of LC3 to ring-like structures during CQ treatment. Are these LAP-like structures (e.g. sensitive to BafA1 treatment)? And were they included in the quantification in Fig.3C?

Fig.4A&B: Since Rbpt5 KD has a weak effect on LC3 puncta formation (Fig.3) and to distinguish the effects of CQ in inducing LAP, the effects of ATG13 and ULK1 KD should be assessed by localising Rbpt5 with WIPI2 or ATG16L1.

Fig.4: It is not clear why ULK1 KD would affect Torin1-induced autophagy but not LC3/WIPI2 localisation during CQ induced early endosome-damage. As the ULK inhibitors can target other pathways, the authors should confirm this finding in ULK1/2 double KO or KD cells.

Fig.5: The contribution of FIP200 in the interaction between Rbpt5 and ATG16L1 is unclear. Is binding between Rbpt5 and ATG16L1 mediated by ATG16L1's interaction with FIP200? The plasmid details describing the delta-WD40 deletion plasmid used in this study are missing and could be important to confirm that the detla-WD40 still retains binding to FIP200.

Fig.5E: the authors should test Rbpt5 AAA mutant binding to FIP200. Since the mutant appears to express less, its binding to ATG16L1 should be quantified or repeated with more comparable expression levels.

Fig.6: CQ treatment can induce various endosomal damage (in addition to early endosomes) and LC3 lipidation processes (e.g. LAP-like). The authors show that Rbpt5 is specifically involved in damaged early endosome autophagy. In this figure, it would be important to distinguish CQ-induced LC3 puncta as a result of early endosome damage or other lipidation processes (e.g. canonical or non-canonical autophagy). The use of FIP200 KO cells shows that LC3 puncta is inhibited. However, here a specific readout to look at early endosome recognition by autophagy is important. The authors can localize early endosome markers (EEA1) with autophagy players (e.g. WIPI2 and LC3). This is also relevant to other figures (e.g. supplementary figure 7E).

Fig.6F&G: the authors should show representative images of these localization images quantified here. These can be added in the supplementary figures.

Minor comments:

Fig.2E: FIP200 seems to be highly overexpressed in this image. Commercial antibodies that recognise endogenous FIP200 are widely used and should be tested to confirm the colocalisation between FIP200 and Rbpt5.

Fig.7C image: the different setting denoted by +/-, +/+ ..etc are not clearly defined.

Significance

This is a interesting study and provides important mechanistic insights underlying the recognition of perturbed early endosomes by the autophagy machinery. Researchers interested in endosomal trafficking or autophagic substrate recognition are likely to benefit from this study.

Referee Cross-commenting

In my opinion, the authors have attempted to distinguish single membrane from double membrane LC3 lipidation by looking at the ULK complex requirement. As other reviewers suggested, this can be further confirmed by using ATG16L1 mutants. It is important however that these experiments are supplemented by co-localising autophagy proteins with alternative early endosome markers when Rbpt5 is inhibited.

I think if the authors are able to address the suggested experiments, this would help improve the manuscript and make it suitable for publication.

Referee Cross-commenting

In my opinion, the authors have attempted to distinguish single membrane from double membrane LC3 lipidation by looking at the ULK complex requirement. As other reviewers suggested, this can be further confirmed by using ATG16L1 mutants. It is important however that these experiments are supplemented by co-localising autophagy proteins with alternative early endosome markers when Rbpt5 is inhibited.

I think if the authors are able to address the suggested experiments, this would help improve the manuscript and make it suitable for publication.

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

In the current manuscript, Millarte et al reports a novel role of Rabaptin5 in selectively clearing damaged endosomes via canonical autophagy. They have identified FIP200 as a novel interactor of Rabaptin5 under basal conditions using yeast-two hybrid screening and further confirmed the interaction of Rabaptin5 with FIP200 with immunoprecipitation. They next used Chloroquine and monitored colocalization of the Rabaptin5 with WIPI2, ATG16L1 and LC3B to demonstrate the potential interaction of Rabaptin5 with the autophagic machinery. They have primarily used Gal-3 as a marker of membrane damage after 30 minutes of Chloroquine treatment. In order to further elucidate the role of Rabaptin5 in autophagic induction mediated by Chloroquine, they have silenced Rabaptin5, FIP200, ULK1 and ATG13 and observed a decrease in the number of LC3 or WIPI2 autophagosome formation. Based on these observations they tested if Rabaptin5 interacts with ATG16L1 upon Chloroquine treatment and confirmed their interaction with potential interaction sites of both Rabaptin5 with ATG16L1 with IP. The authors confirmed the interaction of Rabaptin5 with ATG16L1 by complementing the KO line with the mutant form of Rabaptin5 containing alanine residues in its consensus motif. Finally, they have used Salmonella and SCV as a model to study the role of Rabaptin5 in endomembrane damage and monitored a 50% decrease in the removal of Salmonella in Rabaptin5 KO or KD cells.

Major concerns One of the major concerns is the membrane damage reported by chloroquine which is known to induce lysosomal swelling and further targeting of the swollen compartments to degradation by direct conjugation of LC3 onto single membrane as a form of non-canonical autophagy. The evidence regarding membrane damage by Gal3 colocalization on the Rabaptin5 vesicles is preliminary. As suggested by the authors the canonical autophagy pathway recognizing damaged membranes recruits also ALIX to the damaged membrane which was not observed in Supplementary Figure 2. The link to membrane damage by chloroquine and monensin with Rabaptin5 is not convincing as there is not sufficient evidence of membrane damage. In relation to this issue authors should consider using other damage markers as Gal8, p62 or NDP52 to provide additional claim with respect to membrane damage induced by chloroquine.

One of the main claims here is that Rabaptin5 regulates the targeting of damaged endosomes to autophagy. Clearly, these are early endosomes as stated in the abstract. However, the evidence presented here showing these are early endosomes is not convincing. Analysing Gal3 and Gal8 positive vesicles that are Rabaptin5 positive and an early endosomal marker will be important in this context. For example, there need to be additional evidence showing that early endosomes are targeted to autophagy. Is the degradation of TfR affected by this targeting? Did the authors look at the effect of Bafilomycin A1? If this process affects exclusively early endosomes, it should be BafA1 independent. This will direct more into the cellular function of this process.

Minor concerns Both for Figure 2 and Supplementary Figure 7 it will be clearer to have the images in colour rather than black and white for better interpretation.

The interaction of FIP200 and ATG16L1 with Rabaptin5 is well characterized with immunoprecipitation and imaging but the interaction of Rabaptin5 in presence of chloroquine with FIP200 and ATG16L1 WD are missing and it will be important to include if in the presence of chloroquine these interactions will increase or not.

In order to further support the role of Rabaptin5 for LC3 lipidation upon chloroquine induced membrane damage, western blots of WT, +Rabaptin5, Rabaptin5 KO, Rabaption5 KO +WT or +AAA cell lines were analysed. However, the lysates were collected upon 30 minutes of chloroquine treatment which does not correlate with the imaging performed in Figure 2 as the number of LC3 vesicles did not show an increase upon 30 minutes of chloroquine treatment. The authors should include the 150 minutes time point for the LC3 lipidation in these conditions.

The experiments with Salmonella are of great quality. The relationship of Rabaptin5 with SCV and the endomembrane damage induced by Salmonella could be further elucidated with Rabaptin5 positive vesicles at early infection stages. It is not very clear from the text how authors link the endosomal network previously described for chloroquine with infection. It would be important here to show that Salmonella mutants unable to damage endosomal membranes do not have an effect. In Figure 7 panel C, the time points on graphs are in hours but it should be in minutes.

The events of targeting the damaged membranes for degradation was well characterized by the recognition of these membranes by Gal3, Gal8 and recruitment of autophagic receptors to the site of damage (Chauhan et al. 2016; Jia et al. 2019; Thurston et al. 2012; Maejima et al. 2013; Kreibich et al. 2015). This manuscript introduces a new potential platform for the formation of autophagic machinery on endosomes with the interaction of Rabaptin5 with FIP200 and ATG16L1, however more evidence is required to link this to the clearance of damaged membranes. Previously it was shown that endolysosomal compartments that were neutralized and swollen by monensin and chloroquine had been directed to degradation by direct conjugation of LC3 to single membranes via noncanonical autophagy, but here authors propose another mechanism for this event via canonical autophagy.

Significance

Overall this work is very novel and shows some evidence of early endosomal autophagy. It could be relevant for some for of receptor-mediated signalling (although it is not discussed by the authors) My experience is in intracellular trafficking of pathogens and membrane damage.

Referee Cross-commenting

In my opinion, the only way you can distinguish between double or single membrane is by EM. For me, the important part is to show this is targeting of early endosomes to autophagy, either using other early endosomal markers, analysing by WB some early endosome receptors such as TfR or other inhibitors. If the authors are able to address some these comments, I agree the paper will be in a better position for publication.

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

Learn more at Review Commons

---

Reply to the reviewers

We would like to thank our reviewers for their critical reading and constructive comments. We have addressed all of their points and have included below a detailed reference to the changes we made accordingly. We have also added an additional supplemental figure.

Reviewer #1 :

**Major comments:**

1. The authors highlight in their conclusion that the new Python library has the potential to accelerate and expand microscopy development. I agree with this statement since classes and methods do not need to be written in Python from scratch anymore. However, I would recommend that the authors include in their conclusion the value of the library for reproducibility if the final python acquisition code is shared along with publications. Nowadays, scientists frequently write in their publications that LabView or a specific commercial scope's acquisition software was used without any acquisition code. Python-Microscope would have the potential to change this trend, and the authors need to stress this aspect and its value for reproducibility in science accordingly. This is a good point. We have added the following to the discussion section.

“A further advantage of the approach provided by Microscope is in increasing reproducibility in science. Scientists frequently write in their publications that LabView or a specific commercial scope's acquisition software was used without any specific acquisition settings, code or macros to assist with reproduction. This is especially critical in complex experimental setups where specifics of acquisition are particularly important. Microscope has the potential to change this trend, allowing authors to freely publish simple code demonstrating exactly how their control and acquisition operates. Additionally, the defined device interfaces allow such code to be ported to other specific hardware with minimal changes.”

The authors need to provide a more comprehensive overview of the currently used data acquisition strategies in their introduction. Currently, they highlight the acquisition software provided by vendors for data acquisition (mainly used by life scientists and not necessary scope builders/developers), Micro-Manager (mainly used by life scientists; currently also restricted to wide-field systems), and LabView (for advanced microscope systems; used by advanced developers).

However, most advanced microscope builders use MatLab (Chmyrov et al. Nature Methods (2013) - https://doi.org/10.1038/nmeth.2556, Ta et al. Nature Communications (2015) - https://doi.org/10.1038/ncomms8977 , etc.), Python (York et al. Nature Methods (2013) - https://doi.org/10.1038/nmeth.2687, etc.), and LabView to write their acquisition software. Since the manuscript focused on advanced microscopes, the authors need to position their library with respect to Matlab and Python's current use as well.

We thank the reviewer for pointing out the omission of Matlab control solutions and extending the references to other Python based approaches. We have also added a reference to the Pycro-Manager framework for Micro-Manager which has been published since our original submission.

We have added Matlab to the LabVIEW generalised control software section which now reads:

“custom control software often in LabVIEW or Matlab, both proprietary software. LabVIEW offers a visual programming environment that is commonly used for building instruments in the physical sciences, whereas Matlab is a programming platform with a focus on numeric computing.”

And extended the description sections in the introduction with the following paragraphs and references:

“Matlab is a numerical focused programming environment that scientists are often familiar with for data processing. It has frequently been used for microscopy, leveraging a number of available Matlab sub packages to provide GUI’s and easy access to complex data processing steps. The use of Matlab for microscope control is common in the field but the actual code is rarely shared and often custom to a single microscope setup and associated to image reconstruction (Chmyrov et al., 2013, Ta et al., 2015). Exceptions are ScanImage for the control of laser scanning microscopes (Pologruto et al., 2003), and Matlab Instrument Control (MIC) for the control of individual microscope components (Pallikkuth et al., 2018). Matlab provides a textual programming language simplifying code sharing and version control, however, Matlab is proprietary closed source software and the general requirement of many extensions significantly adds to the cost of implementing many systems.”

“There is currently an increasing number of software options for microscope control in Python, many of which are in the form of custom scripts specific to a microscope (Alvelid and Testa, 2019, York et al., 2013) but some provide a fully integrated microscope control environments, namely PYME https://www.python-microscopy.org/ for SMLM and ACQ4 (Campagnola et al., 2014) for electrophysiology. While this code is freely available and can be modified, their design around a specific setup, technique, or environment reduces its potential for code reuse in other projects.”

The authors need to give (a) software provided by vendors, (b) LabView, and (c) Micro-Manager, more credit.

(a) Several microscope vendors (e.g., Abberior Instruments - https://imspectordocs.readthedocs.io/en/latest/specpy.html ) allow their scopes can be externally controlled to enable the execution of customer-driven acquisition strategies which the vendor's acquisition software itself might not have implemented with. The authors might want to include that scope vendors aim for more customer modifiable acquisition software.

The reviewer makes a good point, especially in the fact that a number of microscope vendors provide Python interfaces for their systems. We have added the following text:

Several microscope vendors, such as Abberior Instruments and Zeiss, provide Python interfaces to enable instrument control from Python. These are all very useful additions to proprietary systems, however they have a fundamental draw back that each manufacture produces their own abstractions meaning code from one system is not compatible with another. Although these interfaces leverage the substantial Python infrastructure they are not generalisable and hence fail to enhance portability or reproducibility.

The fact that these companies are providing Python interfaces to their instruments indicates the general interest of the community in Python as a programming language to extend hardware capabilities. This demonstrates the potential benefit of an entirely Python based interface to a wide range of hardware.

(b) The authors criticize that LabView code can be hard to understand, reproduce and maintain. However, similar to writing good code in general, there are best practice strategies for writing good LabView code to ensure scalability, readability, and maintainability available as well (https://learn.ni.com/learning-paths/labview-core-3-2016-english ). The primary problem might lie more on the side of lousy coding practice than on LabView's side to perform appropriately.

This is a fair point and we have revised the manuscript as indicated below. However, it remains true that it is much harder for a non-expert to write high quality code in LabView than in Python. This is particularly evident in complex systems.

We have changed the section about LabView to read:

“The visual nature of the programming environment makes simple projects easy but systems with a large number of hardware components or complicated control architecture can become hard to understand, reproduce, and maintain. Although this complication can be reduced with good programming practices, it is not uncommon to outsource such work to a commercial company \citep{chhetri2020software} because good code writing in LabView is significantly more challenging than in popular general purpose languages such as Python. Additionally, the LabView work flow does not integrate well into modern distributed source control infrastructure such as mercurial or git, a necessity for modern open source development.”

(c) The authors should include the current effort by Pinkard et al. (Pinkard et al. Nature Methods (2021) - https://doi.org/10.1038/s41592-021-01087-6 ) in their discussion.

A pre-print version of this paper was available on arXiv and cited in our original submission. Now this paper is published we have included the published reference and the following text has been added to our discussion section.

“As mentioned in the introduction, micromanager has a recently introduced Python interface, Pycro-Manager (Pinkard et al. 2021). This simplifies connections between micromanager based hardware interfaces and Python based analysis and control. Although this reduces the effort in using Python for control and online analysis compared to other approaches it does not provide direct access to the hardware via Python. This interface keeps the existing micromanager infrastructure. Particularly new hardware interfaces still need code in both C/C++ and Java before they are accessible via the Python interface.”

The authors might want to explain how they plan to facilitate the library's adoption and the long-term maintenance within the microscopy community. Do they plan to create a new category on Image.sc, which would allow the community to interact with the developers? etc. Furthermore, who will keep writing wrappers to the libraries provided by the vendors? etc

This is a critical point, as the reviewer states, community involvement is essential to continuation of the project and provide a useful tool going forward. We have already published several systems utilising this software platform and are working hard to expand its user base. We have asked for people to post question on the image.sc forums (https://image.sc/) and we also interact with developers and users on the github issue pages (https://github.com/python-microscope/microscope/issues). We have recently implemented a fully automated microscope on a simple motorized stand from Zaber. This provides a fully automated microscopy solution for a very low cost.

We have edited the end of the discussion to read

Microscope is a free and open source project currently being used in several labs with an open development approach. Our aim is that the microscope development community will find it a useful tool and engage in this development to increase its general usefulness. With that aim in mind, we perform our development conversations and user support in the open as github issues and the project is an image.sc community partner. In particular, expanding the number of devices supported by Microscope would be extremely beneficial. However, adding support for a device requires physical access to the device and the current list of supported devices echoes the devices we and our collaborators have access to. This is a chicken and egg problem. Python-Microscope needs broad device support to be widely adopted by the community but it needs contributions from the community to support those devices. We believe that, Microscope currently provides enough devices and infrastructure to support adoption by more developers. There are contribution guidelines within the ``Get Involved' section of the documentation, available online at https://www.python-microscope.org/doc/get-involved.

The authors stress using their library for complex scopes but do not provide an example of complex implementation (they only provide paper references). Only a code for a simple time-series is provided. It would be very beneficial to provide the code for implementing a complex microscope and its GUI with the author's library as separate figures or in the paper's supplement. This would also support point 1 in the review.

The GUI elements provided by Python-Microscope are deliberately minimal implementations to allow basic connectivity and functionality of specific hardware to be tested. Python-Microscope is specifically designed to provide a hardware interface layer separate from the user interface. We provide a very simple examples to demonstrate how easily devices can be controlled. For more complete examples we have developed two associated packages providing GUIs, both are referenced in the text, BeamDelta is an optical alignment tool, while Microscope-Cockpit provides a full user interface to complex microscope systems. We have added a supplemental figure demonstrating the full GUI provided by Cockpit.

**Minor comments:**

It would help the paper if several phrases would be changed: Title: 'Python-Microscope: High-performance control of arbitrarily complex and scalable bespoke microscopes." To: e.g., Python-Microscope: A new open-sources Python library for the control of microscopes

Why? The authors use the word "high-performance" to address their Python library's trigger feature within the text. Unfortunately, that is not how most people would use the term for. Therefore, it should be avoided not only in the title but throughout the text. Furthermore, the word "complex" combined with microscopes should be avoided. A complex microscope is, for most microscope builders, a microscope that needs precise times and synchronization, includes several feedback active feedback loops, incorporates several devices, is very stable, etc. The context in which the term "complex microscopes" is used here is when the authors talk about the library's features to connect devices to servers either locally or remotely. I agree that the library can connect devices over arbitrary complex networks, but using the term "arbitrary complex microscopes" would be misleading considering the library's current speed limitations, the limited number of currently integrated devices, etc.

We have changed the title to:

Python-Microscope: A new open-source library for the control of microscopes

1. Various section titles: "Library features" would be more suitable than "Use Cases" since the individual new features at the new library are described in this section. Also, the description of the individual features should be mentioned more accurately. The following list might be a better, more accurate fit: (1) "Device modularity" instead of "Device independence."

Also, the current title "Write once, run with any device" is inaccurate since the wrapper for multiple devices has not been implemented. (2) "Experiment- and scope-specific layout" instead of "Experiments as programs." (3) "Complex network integration" instead of "Easy implementation of complex systems and scalability" (see reasoning under point a). (4) "Hardware and software trigger integration" instead of "High performance, " (5) "Developer-friendly programming features" instead of "Simple development tool."

We have renamed the specified sections and subsections title and expanded the description in the list of use cases to be more accurate.

1. The authors should avoid using the term "Microscope" when talking about "Python-Microscope." It facilitates the manuscript's readability since it is occasionally not evident in the paper if they refer to the library or a microscope. We have changed “Microscope” to “Python-Microscope” in multiple places of the manuscript where it was unclear whether we were referring to the software or to a physical microscope.

2. The authors should avoid the phrase "pythonic software platform" in the abstract since Python-Microscope is a library / Python package and not a software platform. Furthermore, the term "pythonic" describes the desired way to write Python code. It means code that does not just get the syntax right but follows the Python community conventions and uses the language in the way it is intended to be used. Instead, it might be advisable to write, "Python-Microscope offers elegant Python-based tools to control microscopes...". We have changed the abstract as suggested.

Figure 1 should be supported by comments, e.g., #Load packages, #Parameter Initialization, #Create Devices, # Set camera parameters, etc.

Comments have been added the sample code.

The paragraph under the section "Experiments as programs" about the advantages of using Python (starting from "We have developed the software in Python, ...") should be moved into the Introduction section.

We have moved this segment to the end of the introduction.

Reviewer #2:

1)The introduction does a good job describing the current situation (using multiple software from multiple vendors simultaneously, Micro-Manager, Labview), although it could be highlighted a bit more that several groups have created custom Python code for microscope control (such as https://github.com/ZhuangLab/storm-control, https://github.com/Ulm-IQO/qudi, https://github.com/fedebarabas/tormenta, https://github.com/AndrewGYork/tools), some with at least the hope that their code will be generally usable. It also could be noted that the Micro-Manager device abstraction layer has been accessible from Python for more than a decade (currently the Python 3 interface is at https://github.com/micro-manager/pymmcore).

We have significantly expanded the references to previous Python code and made other changes to the relevant sections as detailed in the response to reviewer #1 and quoted below. We have made reference to the recently published Pycro-Manager package (the previous version referenced the arXiv preprint of this paper. It should be noted that although the Python bindings for mmcore have been available for more than a decade, they have been rarely used, the only published paper referencing them appears to be the whitepaper from a workshop on microscope control software published on arXiv in 2020 (https://arxiv.org/abs/2005.00082).

“There is currently an increasing number of software options for microscope control in Python, many of which are in the form of custom scripts specific to a microscope (Alvelid and Testa, 2019, York et al., 2013) but some provide a fully integrated microscope control environments, namely PYME https://www.python-microscopy.org/ for SMLM and ACQ4 (Campagnola et al., 2014) for electrophysiology. While this code is freely available and can be modified, their design around a specific setup, technique, or environment reduces its potential for code reuse in other projects”

2) Manuscripts describing software tools have to balance the goal to "announce" and advertise the software package with the goal to objectively explain the design principles and choices made. In my opinion, this manuscript finds a nice balance, and leaves the reader with a decent understanding of the capabilities, advantages, limitations and high level architecture of the Python-Microscope package. Possible exceptions are the use of the word "elegant" in the abstract, and extensive use of the word "bespoke" that I mainly know from real estate agent language and that likely is confusing to many readers for whom English is a second language.

We have reworded the abstract to say

“Python-Microscope offers simple to use Python-based tools to control microscopes…”

We use the term “bespoke” to refer to the construction of novel optical microscopes, as opposed to controlling existing integrated systems from commercial vendors. We have reworded paper to refer to custom built microscopes and optical systems to clarify this point.

As far as I am aware, "Microscope" is the most developed microscope abstraction layer written in pure Python. Remarkably, its design (device classes that inherit from a device-base class and have their own function calls, supplemented with "Settings" that can be declared by each device), is extremely similar to that of the Micro-Manager device abstraction layer (where "Settings" are called "Properties"), with the main difference being that one is written in Python and the other in C++ with C bindings. Writing these device classes in Python hopefully brings the advantage that more people can write them, however, the Micro-Manager C interface has the advantage that it can be used from any programming language on any platform, hence is more future proof than pure Python code. The downside of having multiple microscope device abstraction layers is that resources will be diluted and confuse partners in industry (which toolkit should they support with their limited resources?). The number of devices supported is currently much, much greater in the Micro-Manager platform than in Microscope, and a translation layer to make Micro-Manager device adapters in Microscope does not seem out of the question, and could possibly benefit many.

We are aware of the similarity between our approach and that in micromanager. There is therefore significant overlap and possible duplication of effort, however when we started this project we reviewed the Python bindings of micromanager core and felt that using this approach would add significantly, not only to our development effort, but also to end user effort as they would also have to install Micro Manager and its Python bindings. In addition, we believe that there is significant value in having a pure Python implementation. As the reviewer suggests "Python is at the moment probably the most widely used computer programming language by scientists". Having Python-Microscope in a language that the end user can code, invites them to look into the “box” and eases the process for these, possible casual, Python users to contribute with fixes and support for new devices.

Reviewer #3:

• I miss more information regarding the latency of the device-server and software triggering, how fast can it be? How much delay would you have between computers/devices? For example, could we have the devices sincronized at the microsecond range? I think this is super important so that the reader knows if it's worth using a software triggering approach with Python-Microscope or they should buy a DAQ instead. We generally expect high performance hardware to require hardware triggering, software triggers are very unlikely to be performant, or reliable enough to achieve ms, yet alone, µs timing accuracy and reproducibility. Software triggering is implemented as a basic approach to allow simple low speed hardware control, such as basic image snapping. Our systems all utilise external timing devices to provide digital triggering and, in some cases, analogue voltage control. This is becoming increasingly easy with high performance microprocessors such as the ardiuno or higher spec solutions such as National Instruments DAQ boards. We are currently investigating the recently released Raspberry Pi Pico boards, which provide very high performance digital triggering at very low cost (~£4). We are passionately promoting open source, low cost solutions, so requiring a NI DAQ board and LabView licenses goes against the spirit of this project.

1b) It's good though that they don't want to limit themselves to software triggering but also mention hardware triggering, but it's important to better explain where are the limitations.

This is a significant issue but we feel it is beyond the scope of this paper. We utilise microscope as a low level interface to hardware for our systems. The hardware control software has no internal knowledge of device connectivity eg which filter wheel might be in front of what camera, so any integrated control, such as synchronising light sources and cameras is beyond the scope of this package. We use the cockpit package as a GUI and to provide this higher level control integration. We then utilise hardware timing devices interfaced to cockpit to run experiments. We feel that this is a relatively cheap and approachable solution while allowing high performance from even complex systems.

1c) Needs info adding to the text, but in general python-microscope doesn’t concern itself with this, just allows setting of trigger types and you are then responsible for triggering.

As suggested by the reviewer, Python-Microscope does not generally concern itself with triggering. It allows setting of trigger types in a consistent manner, and on relevant devices can initiate a software trigger event. The end of the section “Fast and furious” now reads:

“The microscope interface was designed with the concept of triggers that activate the individual devices and software triggers are handled as simply another trigger type. This approach provides an interface that supports software triggers but is easily upgraded to hardware triggers. The source of such hardware triggers can be other devices --- typically a camera --- or a dedicated triggering device. The recommended procedure is to prepare an experiment template that is then loaded on a dedicated timing device which triggers all other devices, as described in Carlton 2010.

The existence of fast and cheap microprocessors and single board computers mean providing a dedicated hardware timing to sequence and synchronise a number of devices is relatively easy and extremely cost effective. We would recommend systems are designed around using an external device to provide hardware triggers to devices. This provides reliable timing and much more flexible sequencing than directly connecting outputs from one device to trigger inputs for another.”

1d) I also miss information about the triggering, do the software offer a platform that can synchronize devices, or that's more left to the developer to do? They say they can generalize to arbitrarily complex devices so therefore I think it needs to be specified how. Same with the server feature, how fast is that link?

The software triggering depends very much upon the individual devices and delays such as context switching within the OS. We offer no solution to synchronise devices. Our claim to generalise to arbitrarily complex systems is based on the fact that you can trivially run devices on different computers to allow horizontal scaling. If you wish to have 25 cameras, simply run them on different computers, then none will be speed limited by computational resources. Synchronisation can be achieved by an external hardware timing device as described above.

The server link is passed over standard ethernet, likely now 1GB/s, however data packets must be serialised before transmission and deserialised on receipt by Python, as well as standard network overhead and latency. We have only seen network limitations on image transfer from cameras to remote server computers. This has not been a significant issue as the cameras drivers typically have memory buffers, which can be enlarged to cope with backlogs, as well as the Python-Microscope image transmission processes acting on a FIFO memory queue. Possibly long experiments utilising fast, high pixel count cameras could saturate these buffers, but such a specialised application could use specialised solutions such as multi-path networking or a computer with a very large amount of RAM for temporary buffering.

2a) Some critical comments are that, first of all there are not so many drivers yet available (for example Hamamatsu camera).

The reviewer is correct, device support is critical. There are two components to this, a) the resources to implement new devices, and b) the physical hardware to enable testing and debugging of these devices. We have focused on the hardware that we own and use but hardware support is expanding. As described in our reply to reviewer #1, we hope that a community of experienced hardware and software developers will evolve and help support new devices. We have instructions on how to support new hardware devices and are happy to help interested parties. We also plan to apply for continuing funding to enable us to further develop Python-Microscope, especially to expand its range of supported hardware,

The well defined interface with the abstract base class in Python enforces what is required for a minimal implementation of a specific device type. Most devices are relatively easily supported by reference to existing devices of the same type. For instance, a stage is likely to be communicated to by serial over USB, taking simple text commands and returning easy to interpret responses. Adding a new device simply involves defining what commands to send and how to deal with the replies from the hardware. With a suitable manual this can typically be done with a few hours of programming and testing.

2b) I guess this paper is also to show proof of concept and then upon interest they will include more devices, but in that case it should be more documented how one can contribute to the project and generate new drivers. For example, if we want to try it tomorrow in our setups, and we have a specific device such as an Hamamatsu camera, What should we do? Should we contact the authors, write an issue in the github page or write the driver ourself?

We have added the following paragraph on contributing to the project at the end of discussion section of the paper:

Microscope is a free and open source project currently being used in several labs with an open development approach. Our aim is that the microscope development community will find it a useful tool and engage in this development to increase its general usefulness. With that aim in mind, we perform our development conversations and user support in the open as github issues and the project is an image.sc community partner. In particular, expanding the number of devices supported by Microscope would be extremely beneficial. However, adding support for a device requires physical access to the device and the current list of supported devices echoes the devices we and our collaborators have access to. This is a chicken and egg problem. Python-Microscope needs broad device support to be widely adopted by the community but it needs contributions from the community to support those devices. We believe that, Microscope currently provides enough devices and infrastructure to support adoption by more developers. There are contribution guidelines within the ``Get Involved' section of the documentation, available online at https://www.python-microscope.org/doc/get-involved.

• Second, the graphical interface is maybe good enough for developers and builders but in order to have a solid microscope that biologists are going to use it needs a bit more work in that direction. The GUI in microscope is extremely basic and designed for quick testing. For a microscope system aimed at biological users we would recommend using Microscope-Cockpit, our paper is now referenced and a supplemental figure shows an example of its interface, or implementing an alternative more specialised GUI. We have released Python-Microscope as a separate package to separate low level hardware control from a GUI front end, enable relatively easy automated control of microscope systems directly from Python, or allow others to create GUI base interfaces without having to deal with interfacing to specific hardware.

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

Pinto et al present a new python based software to control microscopes. Overall the work is very interesting and will help microscopists to accelerate their development by providing new tool to integrate the different hardwares.

A few aspects commented below need to be clarified to help potential future users to integrate the software for the correct microscopes/hardware.

In general the software is mostly targeted to developers that want to build microscopes, as they mention in the discussion. Some positive features are (1) the ability to have experiments as scripts, (2) the software triggering, (3) the device-server structure, and (4) the ability to have virtual devices to try out the code and the testing I see in the github page. I think it's robust especially and mostly for the device-layer of the software. It's also positive that one can install it in python and import it in your programs, so it can be incorporated into other software fairly easy.

I miss more information regarding the latency of the device-server and software triggering, how fast can it be? How much delay would you have between computers/devices? For example, could we have the devices sincronized at the microsecond range? I think this is super important so that the reader knows if it's worth using a software triggering approach with Python-Microscope or they should buy a DAQ instead. It's good though that they don't want to limit themselves to software triggering but also mention hardware triggering, but it's important to better explain where are the limitations.

I also miss information about the triggering, do the software offer a platform that can synchronize devices, or that's more left to the developer to do? They say they can generalize to arbitrarily complex devices so therefore I think it needs to be specified how. Same with the server feature, how fast is that link?

Some critical comments are that, first of all there are not so many drivers yet available (for example Hamamatsu camera). I guess this paper is also to show proof of concept and then upon interest they will include more devices, but in that case it should be more documented how one can contribute to the project and generate new drivers. For example, if we want to try it tomorrow in our setups, and we have a specific device such as an Hamamatsu camera, What should we do? Should we contact the authors, write an issue in the github page or write the driver ourself?

Second, the graphical interface is maybe good enough for developers and builders but in order to have a solid microscope that biologists are going to use it needs a bit more work in that direction.

Significance

Microscope control software, especially is open source, can help the rapid integration of new hardware and accelerate overall microscopy development.

I see this paper as an important starting point platform for future more user friendly Python-microscope controlling software.

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

This manuscript describes Python-Microscope, a library/framework written in Python to control custom-built microscopes. Modern light microscopes consist of many computer controllable components and data sensors, and software has become an integral component of such systems. Microscopy is such a fast moving and diverse technology that a significant (>25%?) fraction of microscope systems can not be cookie-cutter, standardized systems, but are custom-built, assembled using commercial microscope stands and/or hardware from vendors such as Thorlabs. For many, creating the software to control such custom-built systems is more laborious and difficult than building the actual optical setup, and software toolkits to make this easier (such as the one presented in this manuscript) are of great interest to everyone working in this area. Python is at the moment probably the most widely used computer programming language by scientists, and a well-thought-out environment for microscope control from the Python language is a welcome addition.

The introduction does a good job describing the current situation (using multiple software from multiple vendors simultaneously, Micro-Manager, Labview), although it could be highlighted a bit more that several groups have created custom Python code for microscope control (such as https://github.com/ZhuangLab/storm-control, https://github.com/Ulm-IQO/qudi, https://github.com/fedebarabas/tormenta, https://github.com/AndrewGYork/tools), some with at least the hope that their code will be generally usable. It also could be noted that the Micro-Manager device abstraction layer has been accessible from Python for more than a decade (currently the Python 3 interface is at https://github.com/micro-manager/pymmcore).

Manuscripts describing software tools have to balance the goal to "announce" and advertise the software package with the goal to objectively explain the design principles and choices made. In my opinion, this manuscript finds a nice balance, and leaves the reader with a decent understanding of the capabilities, advantages, limitations and high level architecture of the Python-Microscope package. Possible exceptions are the use of the word "elegant" in the abstract, and extensive use of the word "bespoke" that I mainly know from real estate agent language and that likely is confusing to many readers for whom English is a second language.

As far as I am aware, "Microscope" is the most developed microscope abstraction layer written in pure Python. Remarkably, its design (device classes that inherit from a device-base class and have their own function calls, supplemented with "Settings" that can be declared by each device), is extremely similar to that of the Micro-Manager device abstraction layer (where "Settings" are called "Properties"), with the main difference being that one is written in Python and the other in C++ with C bindings. Writing these device classes in Python hopefully brings the advantage that more people can write them, however, the Micro-Manager C interface has the advantage that it can be used from any programming language on any platform, hence is more future proof than pure Python code. The downside of having multiple microscope device abstraction layers is that resources will be diluted and confuse partners in industry (which toolkit should they support with their limited resources?). The number of devices supported is currently much, much greater in the Micro-Manager platform than in Microscope, and a translation layer to make Micro-Manager device adapters in Microscope does not seem out of the question, and could possibly benefit many.

Expected audience:

This manuscript will be of interest to those scientists who build/assemble their own microscope systems and write software code to control their operation.

Field of expertise:

I think a lot about microscope control software and how it can help scientists do their experiments.

Significance

see above.

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:

In this manuscript, Pinto et al. report Python-Microscope, a new open-source Python library for microscopy control. The new library lets microscope builders implement individual microscope devices as Python Classes with devices specific parameters and methods. Furthermore, the new Python library supports remote procedural calls and turns individual devices into a resource accessible over a network. Moreover, it has been designed to support hardware as well as software triggers. Finally, it provides several developer-friendly features; it is equipped with simple GUI programs for different device types, and it can simulate devices without the need for physical access to the hardware.

Major comments:

1. The authors highlight in their conclusion that the new Python library has the potential to accelerate and expand microscopy development. I agree with this statement since classes and methods do not need to be written in Python from scratch anymore. However, I would recommend that the authors include in their conclusion the value of the library for reproducibility if the final python acquisition code is shared along with publications. Nowadays, scientists frequently write in their publications that LabView or a specific commercial scope's acquisition software was used without any acquisition code. Python-Microscope would have the potential to change this trend, and the authors need to stress this aspect and its value for reproducibility in science accordingly.
2. The authors need to provide a more comprehensive overview of the currently used data acquisition strategies in their introduction. Currently, they highlight the acquisition software provided by vendors for data acquisition (mainly used by life scientists and not necessary scope builders/developers), Micro-Manager (mainly used by life scientists; currently also restricted to wide-field systems), and LabView (for advanced microscope systems; used by advanced developers). However, most advanced microscope builders use MatLab (Chmyrov et al. Nature Methods (2013) - https://doi.org/10.1038/nmeth.2556, Ta et al. Nature Communications (2015) - https://doi.org/10.1038/ncomms8977 , etc.), Python (York et al. Nature Methods (2013) - https://doi.org/10.1038/nmeth.2687, etc.), and LabView to write their acquisition software. Since the manuscript focused on advanced microscopes, the authors need to position their library with respect to Matlab and Python's current use as well.
3. The authors need to give (1) software provided by vendors, (2) LabView, and (2) Micro-Manager, more credit. (1) Several microscope vendors (e.g., Abberior Instruments - https://imspectordocs.readthedocs.io/en/latest/specpy.html ) allow their scopes can be externally controlled to enable the execution of customer-driven acquisition strategies which the vendor's acquisition software itself might not have implemented with. The authors might want to include that scope vendors aim for more customer modifiable acquisition software. (2) The authors criticize that LabView code can be hard to understand, reproduce and maintain. However, similar to writing good code in general, there are best practice strategies for writing good LabView code to ensure scalability, readability, and maintainability available as well (https://learn.ni.com/learning-paths/labview-core-3-2016-english ). The primary problem might lie more on the side of lousy coding practice than on LabView's side to perform appropriately. (3) The authors should include the current effort by Pinkard et al. (Pinkard et al. Nature Methods (2021) - https://doi.org/10.1038/s41592-021-01087-6 ) in their discussion.
4. The authors might want to explain how they plan to facilitate the library's adoption and the long-term maintenance within the microscopy community. Do they plan to create a new category on Image.sc, which would allow the community to interact with the developers? etc. Furthermore, who will keep writing wrappers to the libraries provided by the vendors? etc Several useful software packages have been written in the past, but their existence was often not for long (after 2-3 years, most packages simply can not be used anymore). The concept of software maintenance is frequently not addressed/considered. Therefore, could the authors expand this aspect in an additional section of their paper?
5. The authors stress using their library for complex scopes but do not provide an example of complex implementation (they only provide paper references). Only a code for a simple time-series is provided. It would be very beneficial to provide the code for implementing a complex microscope and its GUI with the author's library as separate figures or in the paper's supplement. This would also support point 1 in the review.

Minor comments:

1. It would help the paper if several phrases would be changed: a. Title: 'Python-Microscope: High-performance control of arbitrarily complex and scalable bespoke microscopes." To: e.g., Python-Microscope: A new open-sources Python library for the control of microscopes Why? The authors use the word "high-performance" to address their Python library's trigger feature within the text. Unfortunately, that is not how most people would use the term for. Therefore, it should be avoided not only in the title but throughout the text. Furthermore, the word "complex" combined with microscopes should be avoided. A complex microscope is, for most microscope builders, a microscope that needs precise times and synchronization, includes several feedback active feedback loops, incorporates several devices, is very stable, etc. The context in which the term "complex microscopes" is used here is when the authors talk about the library's features to connect devices to servers either locally or remotely. I agree that the library can connect devices over arbitrary complex networks, but using the term "arbitrary complex microscopes" would be misleading considering the library's current speed limitations, the limited number of currently integrated devices, etc. b. Various section titles: "Libraray features" would be more suitable than "Use Cases" since the individual new features at the new library are described in this section. Also, the description of the individual features should be mentioned more accurately. The following list might be a better, more accurate fit: (1) "Device modularity" instead of "Device independence." Also, the current title "Write once, run with any device" is inaccurate since the wrapper for multiple devices has not been implemented. (2) "Experiment- and scope-specific layout" instead of "Experiments as programs." (3) "Complex network integration" instead of "Easy implementation of complex systems and scalability" (see reasoning under point a.) (4) "Hardware and software trigger integration" instead of "High performance, " (5) "Developer-friendly programming features" instead of "Simple development tool." c. The authors should avoid using the term "Microscope" when talking about "Python-Microscope." It facilitates the manuscript's readability since it is occasionally not evident in the paper if they refer to the library or a microscope. d. The authors should avoid the phrase "pythonic software platform" in the abstract since Python-Microscope is a library / Python package and not a software platform. Furthermore, the term "pythonic" describes the desired way to write Python code. It means code that does not just get the syntax right but follows the Python community conventions and uses the language in the way it is intended to be used. Instead, it might be advisable to write, "Python-Microscope offers elegant Python-based tools to control microscopes...".
2. Figure 1 should be supported by comments, e.g., #Load packages, #Parameter Initialization, #Create Devices, # Set camera parameters, etc.
3. The paragraph under the section "Experiments as programs" about the advantages of using Python (starting from "We have developed the software in Python, ...") should be moved into the Introduction section.

Significance

The field of microscopy emphasizes more and more openness and transparency of methods and tools being used to accelerate science, but also to guarantee reproducibility.

The authors' library is another step in the right direction. It is open, transparent, tries to satisfy multiple tool developers' needs to make the development of microscopes faster, easier, and more approachable/user-friendly. Although it can not yet be used for arbitrarily complex microscopes, it has the potential to do so in the future. For now, the authors need to manage to incorporate and involve microscopy developers' needs and requirements in the best possible way to be able to design the library as holistic as possible.

I am a physicist and microscope builder and have so far used MatLab, LabView, and Imspector as well as Python scripts to control microscopes, and I will definitely test the authors' library on my own.

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

Learn more at Review Commons

---

Reply to the reviewers

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

We thank the reviewers for their constructive and critical feedback on our original manuscript.

Reviewer #1 (Evidence, reproducibility and clarity (Required)): In this study, the authors explored the tissue-specific regulation of DT size using both global and targeted deletion of Fgf9. They found cell hypertrophy and mineralization dynamics of the DT, as well as transcriptional signatures from skeletal muscle but not bone, were influenced by the global loss of Fgf9. Deletion of Fgf9 in skeletal muscle leads to postnatal enlargement of the DT. However, the innovation of this paper is not enough, the phenotypes of global deletion of Fgf9 were previously reported, most of the data in this paper are mainly descriptive analysis of the phenotypes, and internal cellular and molecular mechanisms were not well investigated.

Here are the major issues:

1.The data showed that fewer osteoclasts were present at both E16.5 and P0 in Figure 2R, V. Whether FGF9 affects both osteogenesis and osteoclast formation?

• Authors’ response to Reviewer: Thank you for your feedback. We revised this manuscript to reflect the concerns of Reviewer 1 related to the lack of cellular and molecular mechanisms as described below. **Based on this question from the Reviewer, we have revised our discussion to clarify our findings as follows: “From our EdU proliferation assays, we observed a decline in cell proliferation in Fgf9null attachments, suggesting an accelerated chondrocyte maturation. Though we saw similar levels of Pthlh expression (a chondrocyte hypertrophy suppressor) in both WT and Fgf9null attachments, we also saw increased expression of Gli1 (a marker of chondrocyte hypertrophy) localized to the attachment in Fgf9null embryos compared to WT embryos. This decrease in proliferation was in parallel with increased hypertrophy of chondrocytes adjacent to the attachment cells within the Fgf9null DT, which may have led to a rapid expansion of matrix in the DT. Even though the DT was enlarged in Fgf9null mutants, we found fewer Sost+ cell clusters in their DTs compared to WT mice. Mature osteocytes express Sost (Winkler et al., 2003), and fewer Sost+ cells may indicate an impaired ability of Fgf9null osteoblasts to embed and mature into osteocytes. Overexpression of FGF9 in the perichondrium has been previously shown to suppress chondrocyte proliferation and limit bone growth in the limb (Karuppaiah et al., 2016); in our study, we found that loss of Fgf9 globally leads to an accelerated enlargement of chondrocytes in the tuberosity. This accelerated enlargement may limit the ability of these cells to deposit matrix and mineral and therefore limit osteocyte differentiation. We also found fewer osteoclasts in the Fgf9null DT which mirrors previous reports using the same mutation to study the length and vascularity of developing limb (Hung et al., 2007). Because the DT is enlarged and resides on the surface of a shortened bone, this phenotype may elucidate a divergent role of FGF9 in patterning of an arrested (e.g., attachment) growth plate compared to a regular (e.g., long bone) growth plate. This includes unexplored roles of FGF9 in vascularity of the tendon attachment and formation of bone ridges that overlap with or deviate from its role in growth plate development that are beyond the scope of the current study.”

1. RNA-sequencing analysis showed the decreased expression of mitochondria/ energy and lipid associated genes in Fgf9 null muscle compared to WT muscle, how does this relate to the enlargement of the DT? What are the detailed molecular mechanisms?

• Authors’ response to Reviewer:
• Based on this question from the Reviewer, we have revised our discussion to reflect the potential molecular mechanisms related to muscle mitochondria, fiber type, and metabolism as follows:

“Fgf9 is expressed in muscle during embryonic stages, which we and others have observed using ISH (Colvin et al., 1999; Garofalo et al., 1999; Hung et al., 2007; Yang and Kozin, 2009). Previous work has established a connection between Fgf9 and muscle, as treatment of muscle and muscle progenitor cells with FGF9 slows maturation, enhances proliferation, and decreases expression of various myogenic genes (Huang et al., 2019). This study found supporting evidence that Fgf9 expression in muscle may be a limiting factor in tuberosity growth. However, it remains unknown how other FGFs and their receptors, FGFRs, regulate superstructure and attachment formation. In this study, we identified potential mediators of skeletal muscle metabolism in Fgf9null muscle, including downregulated mitochondrial-related genes associated with oxidative respiration and proton transport (i.e., Slc36a2 and Ucp1, amongst others). In cultured myoblasts, FGF9 can inhibit myogenic differentiation potentially via increased production of Myostatin (Huang et al., 2019), a well-established mediator of fast glycolytic muscle fibers (Girgenrath et al., 2005; Hennebry et al., 2009). While the role of FGF9 in myoblast fusion has been investigated in vitro, its effect on muscle fiber type and fiber metabolism (i.e., oxidative vs. glycolytic) has not yet been explored. Our findings from bulk RNA-seq of Fgf9null muscle point to potential mechanisms in muscle metabolism that may contribute to the enlarged phenotype that is mimetic of that found in Myostatin deficient mice and other animals (Elkasrawy and Hamrick, 2010; Hamrick et al., 2002). Additionally, further investigations are needed to investigate the potential role of Fgf9 in mitochondrial function and lipid metabolism. Recent work by Huang et al. also identified FGF9 as a potent regulator of calcium signaling and homeostasis in myoblast culture in vitro, and calcium release from the sarcoplasmic reticulum in muscle plays a critical role during embryonic skeletal myogenesis via ryanodine receptor 1 (RYR1). Although Ryr1 was not significantly different in between Fgf9null and WT muscle in the present study, we did find that calmodulin-associated genes (e.g., Calm4, Calml3, Camsap3, Calm5) were all significantly upregulated in Fgf9null muscle compared to WT muscle. Calmodulin interacts with RYR1 and its activation is required for intracellular binding of calcium (Newman et al., 2014, 1). Calmodulin is a crucial component of the calcium signal transduction pathway and also plays an important role in lipid and glucose metabolism (Nishizawa et al., 1988). Taken together, our findings along with recent work by Huang et al. support more mechanistic studies to investigate the metabolic effects of loss and gain of function of Fgf9 on skeletal muscle as well as the muscle secretome.”

Reviewer #1 (Significance (Required)):

R1 The authors compared the phenotypes between globally and muscle-specifically deletion of Fgf9 in mice, and found that Fgf9 secreted by muscle may induced the enlargement of the DT. However, the detailed molecular mechanisms were not well investigated.

**Referees cross-commenting**

R2 I do not disagree with Rev 1, but I do not think such a task is so trial reason why I don't suggest; it could take years to determine molecular mechanisms of anything. The authors could expand the discussion, offer some possibilities. If they had some RNAseq data they maybe could suggest some of the key signaling pathways involved.

**Referees cross-commenting**

R1 We still suggested that the internal cellular and molecular mechanisms should be well investigated in this papaer.

Reviewer #2 (Evidence, reproducibility and clarity (Required)):

• This paper deals with an important topic which is exact molecular mechanisms regulating the growth of bony tuberosities; because this region is essential for force transmission and movement.
• Based on the previous information they had that in the global KO of the gene FGF9 the deltoid muscle is enlarged; and this muscle is in a very important tuberosity; they decided to look at FGF9 as a potential genetic regulator.

• The manuscript is clear, objective, concise. Very clear. Authors used both the global and targeted deletions, very high reproducibility. Reviewer #2 (Significance (Required)):

• This manuscript advances several areas since we know little about the mechanisms controlling local mechanisms of tuberosities. It also advances our knowledge of FGF9. There were several studies before mostly in vitro showing that FGF9 when added to muscle cells could arrest myogenesis, but the types of experiments in vivo had not been performed yet. The authors used an array of methods; the studies are unbiased and very rigorous and also they always show all experimental points, which is excellent. The conclusions are supported by the data.

• The main suggestion for authors: They essentially do not discuss the nature of the potential muscle to bone signaling occurring when they target the deletion of FGF9 in skeletal muscles and muscles enlarge and there is a series of adaptions in the tuberosity. Do the authors believe this to be all the genetic changes or potentially through secreted myokines? In the paper of Huang et al, 2019 the authors document an effect of FGF9 in intracellular calcium homeostasis/signaling; could this be part of the mechanism? Perhaps the authors could propose a model?

Authors’ response to Reviewer:

• Future studies could investigate the secretome of muscle in Fgf9null or muscle-specific knockouts, as well as assess calcium signaling homeostasis in Fgf9 mutant muscles. We did find calcium- and ion-associated genes in the RNAseq and revised the discussion to include this information.
• Based on this question from the Reviewer, we have revised our discussion to reflect the potential molecular mechanisms related to muscle mitochondria, fiber type, and metabolism as follows: “Fgf9 is expressed in muscle during embryonic stages, which we and others have observed using ISH (Colvin et al., 1999; Garofalo et al., 1999; Hung et al., 2007; Yang and Kozin, 2009). Previous work has established a connection between Fgf9 and muscle, as treatment of muscle and muscle progenitor cells with FGF9 slows maturation, enhances proliferation, and decreases expression of various myogenic genes (Huang et al., 2019). This study found supporting evidence that Fgf9 expression in muscle may be a limiting factor in tuberosity growth. However, it remains unknown how other FGFs and their receptors, FGFRs, regulate superstructure and attachment formation. In this study, we identified potential mediators of skeletal muscle metabolism in Fgf9null muscle, including downregulated mitochondrial-related genes associated with oxidative respiration and proton transport (i.e., Slc36a2 and Ucp1, amongst others). In cultured myoblasts, FGF9 can inhibit myogenic differentiation potentially via increased production of Myostatin (Huang et al., 2019), a well-established mediator of fast glycolytic muscle fibers (Girgenrath et al., 2005; Hennebry et al., 2009). While the role of FGF9 in myoblast fusion has been investigated in vitro, its effect on muscle fiber type and fiber metabolism (i.e., oxidative vs. glycolytic) has not yet been explored. Our findings from bulk RNA-seq of Fgf9null muscle point to potential mechanisms in muscle metabolism that may contribute to the enlarged phenotype that is mimetic of that found in Myostatin deficient mice and other animals (Elkasrawy and Hamrick, 2010; Hamrick et al., 2002). Additionally, further investigations are needed to investigate the potential role of Fgf9 in mitochondrial function and lipid metabolism. Recent work by Huang et al. also identified FGF9 as a potent regulator of calcium signaling and homeostasis in myoblast culture in vitro, and calcium release from the sarcoplasmic reticulum in muscle plays a critical role during embryonic skeletal myogenesis via ryanodine receptor 1 (RYR1). Although Ryr1 was not significantly different in between Fgf9null and WT muscle in the present study, we did find that calmodulin-associated genes (e.g., Calm4, Calml3, Camsap3, Calm5) were all significantly upregulated in Fgf9null muscle compared to WT muscle. Calmodulin interacts with RYR1 and its activation is required for intracellular binding of calcium (Newman et al., 2014, 1). Calmodulin is a crucial component of the calcium signal transduction pathway and also plays an important role in lipid and glucose metabolism (Nishizawa et al., 1988). Taken together, our findings along with recent work by Huang et al. support more mechanistic studies to investigate the metabolic effects of loss and gain of function of Fgf9 on skeletal muscle as well as the muscle secretome.

In conclusion, this work established a new role of skeletal muscle derived Fgf9 during skeletal development and tuberosity growth. Additionally, our unbiased transcriptomic approaches and rigorous analyses identified new potential mechanisms associated with muscle development, mitochondrial bioenergetics, and muscle metabolism that warrant further investigation into the role of FGF9 in muscle-bone crosstalk.”

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

This paper deals with an important topic which is exact molecular mechanisms regulating the growth of bony tuberosities; because this region is essential for force transmission and movement. Based on the previous information they had that in the global KO of the gene FGF9 the deltoid muscle is enlarged; and this muscle is in a very important tuberosity; they decided to look at FGF9 as a potential genetic regulator.

The manuscript is clear, objective, concise. Very clear. Authors used both the global and targeted deletions, very high reproducibility.

Significance

This manuscript advances several areas since we know little about the mechanisms controlling local mechanisms of tuberosities. It also advances our knowledge of FGF9. There were several studies before mostly in vitro showing that FGF9 when added to muscle cells could arrest myogenesis, but the types of experiments in vivo had not been performed yet. The authors used an array of methods; the studies are unbiased and very rigorous and also they always show all experimental points, which is excellent. The conclusions are supported by the data.

The main suggestion for authors: They essentially do not discuss the nature of the potential muscle to bone signaling occurring when they target the deletion of FGF9 in skeletal muscles and muscles enlarge and there is a series of adaptions in the tuberosity. Do the authors believe this to be all the genetic changes or potentially through secreted myokines? In the paper of Huang et al, 2019 the authors document an effect of FGF9 in intracellular calcium homeostasis/signaling; could this be part of the mechanism? Perhaps the authors could propose a model?

Referees cross-commenting

I do not disagree with Rev 1, but I do not think such a task is so trial reason why I don't suggest; it could take years to determine molecular mechanisms of anything. The authors could expand the discussion, offer some possibilities. If they had some RNAseq data they maybe could suggest some of the key signaling pathways involved.

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

In this study, the authors explored the tissue-specific regulation of DT size using both global and targeted deletion of Fgf9. They found cell hypertrophy and mineralization dynamics of the DT, as well as transcriptional signatures from skeletal muscle but not bone, were influenced by the global loss of Fgf9. Deletion of Fgf9 in skeletal muscle leads to postnatal enlargement of the DT. However, the innovation of this paper is not enough, the phenotypes of global deletion of Fgf9 were previously reported, most of the data in this paper are mainly descriptive analysis of the phenotypes, and internal cellular and molecular mechanisms were not well investigated.

Here are the major issues:

1.The data showed that fewer osteoclasts were present at both E16.5 and P0 in Figure 2R, V. Whether FGF9 affects both osteogenesis and osteoclast formation?

2.RNA-sequencing analysis showed the decreased expression of mitochondria/ energy and lipid associated genes in Fgf9 null muscle compared to WT muscle, how does this relate to the enlargement of the DT? What are the detailed molecular mechanisms?

Significance

The authors compared the phenotypes between globally and muscle-specifically deletion of Fgf9 in mice, and found that Fgf9 secreted by muscle may induced the enlargement of the DT. However, the detailed molecular mechanisms were not well investigated.

Referees cross-commenting

We still suggested that the internal cellular and molecular mechanisms should be well investigated in this papaer.

Transparent Peer Review

---

Download the complete Review Process [PDF] including:

• reviews
• authors' reply
• editorial decisions

</br>

Transparent Peer Review

---

Download the complete Review Process [PDF] including:

• reviews
• authors' reply
• editorial decisions

</br>

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

Learn more at Review Commons

---

Reply to the reviewers

Dear Dr. Monaco,

Thank you for reviewing our manuscript entitled ‘Discovery of re-purposed drugs that slow SARS-CoV-2 replication in human cells’. We are pleased to see that the reviewers make suggestions that will strengthen the paper. With cases of COVID-19 rising at dramatic levels in some parts of the world, we are anxious to see our results published in a peer-review journal.

Please find below a detailed response to the comments is shown in bold. We can perform the additional experiments and make changes to the manuscript within 3 weeks of a journal agreeing to consider our paper.

Reviewer #1 (Evidence, reproducibility and clarity (Required)):

**Summary:**

Pickard et al. present in the manuscript entitled "Discovery of re-purposed drugs that slow SARS-CoV-2 replication in human cells" a new screen of FDA approved drugs against SARS-CoV-2. The authors based their screen on Vero and HUH7 cell lines. The methods applied for screening including the SARS-Cov-2-ΔOrf7a-NLuc modified virus are properly designed and preformed. This is an interesting study that finds several potential drugs that might be effective as anti SARS-CoV-2 therapies. However, such experiments have been done throughout the last year and the novelty and importance of these findings are questionable.

Regarding this point, there are several studies that have attempted to identify compounds that impact on SARS-Cov2 infection; however, these do not specifically focus on the replication of the virus (studies have used viability markers and staining of viral proteins but many of the compounds identified exert their effects on the virus uptake). Whilst SARS-CoV2-Nluc viruses have been developed these have been used for infection studies to measure the amount of virus taken up by cells and have not further explored how they impact on virus replication. Therefore, we feel that our study shows that a reporter virus can be used to reflect virus replication.

---

**Major comments:**

1. Most the experiments presented are only done twice, while in the screen itself it should not be a problem, for verifying the drugs identified at least three experiments are suggested (Figure 5 and Supplemental Figure 6) At the time of submission there was an urgent need to make our data accessible to the scientific community. Therefore, we performed some experiments with n=2. We used n=2 to validate the screen and each time we got the same experimental outcome. We would perform further repeats for the figures mentioned for publication.

---

To strengthen the results of the screen, the wild type virus should also be tested for plaque reduction assay with these nine drugs.

We will perform these experiments and present these data in the manuscript. We have already performed immunostaining of WT-virus infected cells and could include this as an alternative.

---

Identification of antivirals is important for SARS-CoV-2 and other coronaviruses, regardless of the presence or effectiveness of vaccines. I think the abstract and introduction should be written to emphasize this point (instead of trying to underestimate the vaccine effectiveness). Similarly, the authors ignore the relative failures of known antivirals (known to inhibit SARS-CoV-2 replication in vitro like Remdesavir) in clinical trials and suggest starting clinical trials with their screen results. I think that this suggestion is premature and require several more studies (including animals studies) before initiating clinical trials.

We will re-write this section of the manuscript. We have identified all compounds that have been evaluated in the AGILE clinical trial, and these compounds failed to show a patient benefit and also failed to impact on virus replication in human cells.

---

**Minor comments.**

1. The errors bars are not defined throughout all the figures. I am not sure that error bars are even meaningful if experiments only done twice, I recommend showing the two results for each point. We will add additional repeats or as the reviewer suggests we could add the two points.

Figure 1E and the tables especially supp tables 3 and 4? don't have legends.

Apologies, this will be amended.

---

Most graphs will benefit from presenting the results in logarithmic scale (all Luc counts/ qPCRs).

This can be changed if editors agree.

P6 in the Generation of functional SARS-CoV-2 virus section - a reference is missing "It has been reported that this aids the recovery of replicative virus (Insert ref 3)"

Apologies, this will be amended.

Reviewer #1 (Significance (Required)):

This is a well performed drug screen on two cell lines that identified new potential FDA approved drugs as anti-SARS-CoV-2 inhibitors. There are several studies that already been published or distributed as preprints that have done similar experiments in other cells lines including more relevant lung epithelial cells (for example PMC743673). This study does not verify the screen results by additional methods. However, in the current pandemic situation this study could be important and interesting to follow up.

I am a virologist; my expertise is in viral host interactions within infected cell.

We were unable to identify the paper which is referred to in the reviewer’s comments. We would aim to highlight further in the text that using the reporter virus, we are able to screen and identify compounds that impact on virus replication unlike many of the other studies.

---

**Referee Cross-commenting**

No problem with the other comments

Reviewer #2 (Evidence, reproducibility and clarity (Required)): **Summary:**

In this manuscript, the authors report on the creation of a luciferase-encoding SARS-CoV-2 (deleted for orf7a) and the use of this virus to test infectability of multiple cell lines as well as perform a drug repurposing screen in two cell lines (Vero E6 and Huh7). Of the 35 drugs that blocked the virus replication they further identify 9 drugs that have a (mild) effect on replication when administered 24 hours post infection.

An important note here is that many studies which have identified potential therapeutics for SARS-Cov2 have performed experiments whereby cells are pre-treated with compounds prior to infection. We have been able to performed the same experiments and many of the drugs were unable to prevent replication after infection. The 9 compounds we have identified retain the ability to inhibit replication when applied post-infection. This sets our study apart from other screens that have been conducted for SARS-Cov2.

---

**Major comments:**

1. Figure 2: What's the difference between "Luminescence counts above noise" in Fig 2B and "Luminescence counts per second" in Figure 2C,D ? It seems like there is no difference in luminescence between 1 PFU and 100 PFU (and if anything, the bassline for 1PFU is higher, >1.5M, compared to 100 PFU where is below 1M). One would expect more luminescence in the 100 PFU experiment, as seen in Fig 2B. Also in Fig 2B it does not mention how many replicates, or what does the **** stands for. Thank you for the comment. The difference in “luminescence counts above noise” and “luminescence counts per second” is set out in Figure 2A. When adding more virus the baseline level should increase, as also demonstrated in Supplemental Figure 3. However, the degree of background luminescence varies between virus batches, presumably due to the degree of cell lysis in each sample. You will note in the Supplemental figure that the baseline levels for our P4 viral stock is lower than P1. We performed the experiments in Figure 2C using virus P1 virus stocks and for Figure 2D we used P4 virus. For clarity this information will be included in the figure legend and the data presented at luminescence over background.

The authors do not explain why deleting orf7a was needed to generate the NLuc virus. Was there a rational for this?

Orf7a has been successfully removed from SARS-CoV and SARS-CoV2 in order to incorporate traceable proteins such as fluorescent or bioluminescent proteins. We describe this at the start of the results section. “Orf7a has previously been removed in SARS-CoV and SARS-CoV-2 and yielded infectious and replicative virus particles (Thi Nhu Thao et al, 2020; Xie et al, 2020a; Xie et al, 2020b)”.

Figure 5C - IC50 should be properly determined from compounds where the lowest concentration tested was still inhibitory (such as LY2835219 and panobinostat).

These experiments can be conducted, within 2 weeks. However we do not feel that this would provide additional information to the reader. The aim of these figures is to demonstrate that there are dose dependent effects of these compounds on the replication of SARS-CoV2.

Supplementary tables must be provided in an excel or similar file format. The PDF version is both unreadable and does not allow other researchers to probe the dataset for their own interests.

This would be amended during revision of our submission.

**Minor comments:**

1. Intro: "SARS-CoV-2 infection in patients with COVID-19 can result in pulmonary distress, inflammation, and broad tissue tropism". Broad tissue tropism is not a result of infection, please rephrase. Patients with COVID-19 are reported to have liver and kidney damage. This could be a direct result of SARS-CoV2 infection or indirectly via the cytokine storm. Our data shows that kidney and liver cells are highly susceptible to SARS-CoV2 infection and support replication, in culture. We thank the reviewer for their comment and we will rephrase this statement and cite relevant literature.

Fig S1D - why are the MOI different for WT (moi 0.1) and NLuc mutant (moi 1) ?

This was used to demonstrate the lack of replication of the WT virus in lung epithelial cells, the same MOI used in Vero cells demonstrates that the levels of the nucleocapsid protein increases when compared to other cell types. We have also used an MOI of 10 for the NLuc virus to be able to detect the NLuc protein. This information would be added to the figure legend.

---

Fig S3 - using volume of virus in ul is problematic, as it doesn't allow for proper comparison between the passages. The author would express the virus amount in PFU or MOI.

This will be amended

---

Fig S5 - in panel A - what do the colors represent? What is 0-1?. The number of repetitions for each panel should be indicated.

Apologies, relative expression should have been added alongside the scale. N=3 for this experiments this will be added to the figure legend.

---

The "NLuc activity as a marker of virus replication" and "SARS-CoV-2 replication screen validation" are largely overlapping and should be edited.

We would combine these sections.

---

Methods: "Generation of functional SARS-CoV-2 virus" - the author confuse "virus" with "plasmid". They should also include the reference marked "(Insert ref 3)"

Apologies, this will be amended

---

Reviewer #2 (Significance (Required)):

1. My main concern is that a very similar, if not identical, NLuc encoding virus has been reported in October 2020 (https://www.nature.com/articles/s41467-020-19055-7#Sec9). While the authors cite this paper, they only do so to say that "Orf7a has previously been removed in SARS-CoV and SARS-CoV-2 and yielded infectious and replicative virus particles", without mentioning this was done to generate the same NLuc carrying virus reported in their work. Thus the generation of this virus is not a "new tool" as the authors would seem to suggest. Whilst this is not the first use of a NLuc SARS-CoV2 virus, this is the first time that the virus has been utilised to screen for compounds that effect replication. The study mentioned does not screen nor monitor the replication of the virus, the authors do monitor the capability of the virus to infect cells only during the first 24 hours.

While drug repurposing screens have been performed, the addition validation in Vero E6 and Huh7 cells is of some interest to those working on anti-viral therapies, given that the authors change their supplementary tables to a format that can be accessible by other researchers.

This will be amended for the submission.

---

My expertise: I study virus-host interactions (not coronaviruse). In the last year I have been involved in several drug repurposing efforts against SARS-CoV-2.

**Referee Cross-commenting**

No problem with the other comments.

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

Summary:

In this manuscript, the authors report on the creation of a luciferase-encoding SARS-CoV-2 (deleted for orf7a) and the use of this virus to test infectability of multiple cell lines as well as perform a drug repurposing screen in two cell lines (Vero E6 and Huh7). Of the 35 drugs that blocked the virus replication they further identify 9 drugs that have a (mild) effect on replication when administered 24 hours post infection.

Major comments:

1. Figure 2: What's the difference between "Luminescence counts above noise" in Fig 2B and "Luminescence counts per second" in Figure 2C,D ? It seems like there is no difference in luminescence between 1 PFU and 100 PFU (and if anything, the bassline for 1PFU is higher, >1.5M, compared to 100 PFU where is below 1M). One would expect more luminescence in the 100 PFU experiment, as seen in Fig 2B. Also in Fig 2B it does not mention how many replicates, or what does the ** stands for.
2. The authors do not explain why deleting orf7a was needed to generate the NLuc virus. Was there a rational for this?
3. Figure 5C - IC50 should be properly determined from compounds where the lowest concentration tested was still inhibitory (such as LY2835219 and panobinostat).
4. Supplementary tables must be provided in an excel or similar file format. The PDF version is both unreadable and does not allow other researchers to probe the dataset for their own interests.

Minor comments:

1. Intro: "SARS-CoV-2 infection in patients with COVID-19 can result in pulmonary distress, inflammation, and broad tissue tropism". Broad tissue tropism is not a result of infection, please rephrase.
2. Fig S1D - why are the MOI different for WT (moi 0.1) and NLuc mutant (moi 1) ?
3. Fig S3 - using volume of virus in ul is problematic, as it doesn't allow for proper comparison between the passages. The author would express the virus amount in PFU or MOI.
4. Fig S5 - in panel A - what do the colors represent? What is 0-1?. The number of repetitions for each panel should be indicated.
5. The "NLuc activity as a marker of virus replication" and "SARS-CoV-2 replication screen validation" are largely overlapping and should be edited.
6. Methods: "Generation of functional SARS-CoV-2 virus" - the author confuse "virus" with "plasmid". They should also include the reference marked "(Insert ref 3)"

Significance

1. My main concern is that a very similar, if not identical, NLuc encoding virus has been reported in October 2020 (https://www.nature.com/articles/s41467-020-19055-7#Sec9). While the authors cite this paper, they only do so to say that "Orf7a has previously been removed in SARS-CoV and SARS-CoV-2 and yielded infectious and replicative virus particles", without mentioning this was done to generate the same NLuc carrying virus reported in their work. Thus the generation of this virus is not a "new tool" as the authors would seem to suggest.
   1. While drug repurposing screens have been performed, the addition validation in Vero E6 and Huh7 cells is of some interest to those working on anti-viral therapies, given that the authors change their supplementary tables to a format that can be accessible by other researchers.

My expertise: I study virus-host interactions (not coronaviruse). In the last year I have been involved in several drug repurposing efforts against SARS-CoV-2.

Referee Cross-commenting

No problem with the other comments.

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:

Pickard et al. present in the manuscript entitled "Discovery of re-purposed drugs that slow SARS-CoV-2 replication in human cells" a new screen of FDA approved drugs against SARS-CoV-2. The authors based their screen on Vero and HUH7 cell lines. The methods applied for screening including the SARS-Cov-2-ΔOrf7a-NLuc modified virus are properly designed and preformed. This is an interesting study that finds several potential drugs that might be effective as anti SARS-CoV-2 therapies. However, such experiments have been done throughout the last year and the novelty and importance of these findings are questionable.

Major comments:

1. Most the experiments presented are only done twice, while in the screen itself it should not be a problem, for verifying the drugs identified at least three experiments are suggested (Figure 5 and Supplemental Figure 6)
2. To strengthen the results of the screen, the wild type virus should also be tested for plaque reduction assay with these nine drugs.
3. Identification of antivirals is important for SARS-CoV-2 and other coronaviruses, regardless of the presence or effectiveness of vaccines. I think the abstract and introduction should be written to emphasize this point (instead of trying to underestimate the vaccine effectiveness). Similarly, the authors ignore the relative failures of known antivirals (known to inhibit SARS-CoV-2 replication in vitro like Remdesavir) in clinical trials and suggest starting clinical trials with their screen results. I think that this suggestion is premature and require several more studies (including animals studies) before initiating clinical trials.

Minor comments.

1. The errors bars are not defined throughout all the figures. I am not sure that error bars are even meaningful if experiments only done twice, I recommend showing the two results for each point.
2. Figure 1E and the tables especially supp tables 3 and 4? don't have legends.
3. Most graphs will benefit from presenting the results in logarithmic scale (all Luc counts/ qPCRs).
4. P6 in the Generation of functional SARS-CoV-2 virus section - a reference is missing "It has been reported that this aids the recovery of replicative virus (Insert ref 3)"

Significance

This is a well performed drug screen on two cell lines that identified new potential FDA approved drugs as anti-SARS-CoV-2 inhibitors. There are several studies that already been published or distributed as preprints that have done similar experiments in other cells lines including more relevant lung epithelial cells (for example PMC743673). This study does not verify the screen results by additional methods. However, in the current pandemic situation this study could be important and interesting to follow up.

I am a virologist; my expertise is in viral host interactions within infected cell.

Referee Cross-commenting

No problem with the other comments

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

Learn more at Review Commons

---

Reply to the reviewers

Reviewer response

We thank the reviewers for their response. All reviewers find our study (potentially) interesting and/or a resource to gain further understanding on BAP1 molecular functions. They also have some common comments.

The reviewers would prefer to see further characterization of the interactions and their functional effects. We would have liked to address this but found for COPI that knockdown of these genes is lethal, whereas on the BAP1 side the interactions are mapped to the functionally critical C-terminus, making these experiments technically extremely challenging. These issues, unfortunately, preclude further validation studies at this point. Nevertheless, we do feel that the quality of our interaction dataset is such that it is be worth publishing these finding for this important tumor suppressor.

Most reviewers would like us to place the data more in context. To address this, we have extended the discussion, highlighting the essence of our findings and how we envisage this could impact BAP1 function.

Finally, both reviewer 1 and 3 would like the results section to be more succinct and we have shortened it to improve readability.

Other points are addressed in the point-by-point response to individual reviewers below.

Point-by-point response to reviewers

Reviewer #1 (Evidence, reproducibility and clarity (Required)):

Very interesting study on BAP1 tumor suppressor. The work needs further characterization of the interacting partners identified.

Reviewer #1 (Significance (Required)):

BAP1 is an important tumor suppressor mutated in several malignancies and the mechanism of action of this deubiquitinase are far from being completely understood.

This interesting work aimed at identifying novel cytoplasmic partners of BAP1 which can highly relevant to its tumor suppressor function. BAP1 is predominantly nuclear, but can also be found in the cytoplasm. New insights into the cytoplasmic functions of BAP1 are needed.

The manuscript is overall well written and the data are very solid.

The manuscript would need additional work before acceptance

**Comments**

Reviewer #1 1) The abstract can be improved to reflect the data of the manuscript.

Unfortunately, we do not understand what part of the abstract is meant by the reviewer, which makes it hard to address.

2) The result section, manuscript could emphasize the results rather that the technical aspects

We've improved readability of this part of the manuscript by moving technical parts that are not required for interpretation of the results to the material and method section, a supplementary text and a new supplemental figure 1 (causing the original numbering to shift).

3) It would be interesting to further investigate the significance of some key interactions

We agree that these questions are of importance (see reviewer 2 point 2, reviewer 3 point 1). We have tried to address these questions using gene knockdown techniques. However, the importance of regulation of protein transport and vesicle formation by COPI translates to lethal effects on cell viability upon knockdown of these genes making these experiments technically impossible to execute. Further functional investigation is technically and financially beyond the scope and possibilities of this paper.

4) The discussion is quite short and can put the findings in perspective

We've extended the discussion to place results into perspective, also regarding the potential role of BAP1 activity towards potential substrates (point 8). This will help to highlight the important findings of the research.

5) It would be interesting to test some cancer-associated mutations

The interaction is in the C-terminus of BAP1, which combines several functions[1, 2]. This would dramatically complicate the interpretation of results. Particularly the presence of the NLS, a major regulatory posttranslational modification[3] and the recruitment signal for nucleosomes could all interfere with BAP1 function independently.

6) Figure 4 can be improved

Thanks for this comment. We have increased readability of the figure by addition of schematic representations of the used constructs, a legend that explains the color-coding of the interactors. We have also removed dotted lines to make it less busy.

7) Yu et al MCB 2010 is one of the key papers on BAP1 purification and can be cited

We apologize for omitting this reference and have included it in the revised manuscript.

8) The authors can discuss potential substrates of BAP1 and mechanism of deubiquitination

We've extended the discussion to this extent.

**Referee Cross-commenting**

I agree with the comments of Reviewer #2

Reviewer #2 (Evidence, reproducibility and clarity (Required)):

**Summary:**

BAP1 is a deubiquitinase that mainly functions to control H2Aub levels in nucleus. BAP1 is a tumor suppressor with mutations or deletions in several human cancers. Previous studies have identified many interacting proteins with BAP1, most notably its binding partners involved in PR-DUB complex, such as ASXL1/2, FOXK1/2, HCFC1, OGT etc. In this study, by using FRT-mediated recombination, the authors generated tagged-BAP1 expressed from its endogenous promoter and conducted AP-MS analysis to identify BAP1 interacting proteins in both nuclear and cytoplasmic fractions. These analyses identified several new BAP1-interacting proteins in cytoplasmic fractions, including histone acetyltransferase 1 (HAT1) and the heptameric coat protein complex I (COPI, which is involved in protein sorting and trafficking). The authors further confirmed the interactions between BAP1 and HAT1 (as well as a COPI subunit) at the endogenous level.

**Major comments:**

Overall, the current study has relatively limited data with limited scopes: basically a proteomic study focusing on one single protein (which has already been subjected to several proteomic studies in previous publications). There is a significant room for authors to improve this potentially interesting study (see below for specific comments), although this may take substantial additional efforts.

1. In the current manuscript, there is no data to further characterize the interactions between BAP1 and HAT1 (or COPI). For protein-protein interaction studies, readers generally are interested in information such as whether these bindings are direct or indirect (particularly for COPI because it contains multiple subunits), and which regions mediate the interactions?

Our mass spectrometry data suggests binding of BAP1 to be mediated through its C-terminus. Mutations of the KxKxx domain have shown that this motif is not involved. Mapping the interaction any more specifically is likely to be very difficult as the C-terminus of BAP1 is involved in many different functions and contains many important elements (ULD domain, CTE, NLS) required for its function. Mutational analysis aimed to map the interaction will induce many secondary effects as the protein localization and substrate targeting will be severely affected as shown by us and other research groups. Identifying what subunit of the COPI complex is mediating the interaction requires purification of these proteins along with purified full length active BAP1, for which attempts have been made but were still unsuccessful. Further investigation is technically and financially beyond the scope and possibilities of this paper.

No data to study the functional significance of the identified protein-protein interactions. This is a major weakness of the current study. For example, HAT1 is a histone acetyltransferase and mainly functions in the nucleus. Does the BAP1-HAT1 interaction in cytoplasm suggest that they have functions in cytoplasm independent of their canonical function in regulating transcription in the nucleus? Likewise, does BAP1-COP1 interaction suggest that somehow BAP1 is involved in regulating protein sorting and trafficking?

Please see our general response and reviewer 1 point 3.

The identified interactions appear to be weak. These proteins are located near the edge of the significance curve in volcano plots (Fig. 1C-1D and others). The IP data also appear to be weak; for example, see Fig. 5D, it's hardly to see COPA blot in BAP1 IP. The COPA IB signal from 5% input WCL is probably hundred-fold stronger than that from BAP1 IP. Weak interactions do not necessarily mean they are not important; however, there is no functional data to support this claim (see the point above).

The essence of our paper is that the interactions which are barely visible in figure 1, gain significance in the absence of endogenous BAP1 to the point where all COPI subunits are as confidently identified as previously validated BAP1 interactors like ASXL, FOXK and HCFC proteins (figure 4). Our quantifications indicate that the new interactors have lower stoichiometry, and this may explain why they were harder to identify. This observation is discussed in the discussion section.

It seems that the entire study focuses on one specific cell line. Repeat the analyses in other cell lines can help boost the robustness and significance of the study.

As discussed under the previous point, the removal of endogenous BAP1 was important for significance, and since BAP1 is a common essential gene, we don't have any other cell lines in which this would be possible. However, we have been able to confirm the HAT1 interaction with BAP1 in U2OS on endogenous levels (Fig 1E,F).

The major interacting proteins they identified from the cytoplasmic fraction are still those mainly localized in nucleus (such as HCFC1, FOXK1/2). A western blotting to show nuclear vs cytoplasmic fraction is required.

Our immunoblot containing cytoplasmic and nuclear input samples used for figure 4 show proper separation of fractions without major leakage as shown in supplemental figure 4 (Tubulin and Abraxas lane as cytoplasmic and nuclear markers respectively), while iBAQ data show a substantial amount of protein to be bound (stoichiometry.xls). This is corroborating with the sample correlation data shown in supplemental figure 5B which shows very little correlation between cytoplasmic and nuclear samples in the mass spectrometry experiment. These data show that the interactions of the cytoplasmic partners that are mainly localized in the nucleus are real interaction and are not due to mixing of cellular compartments.

**Minor comments:**

1. page 9 "A GFP coIP experiment of both GFP-BAP1 and the catalytic-dead BAP1 C91S mutant shows coimmunoprecipitation of HAT1 (Figure 1F)." here Fig. 1F should be Fig. 1E.

The Figure was indeed mislabeled. Because of the addition of a new supplemental figure for reviewer 1 point 2 some figure numbers have shifted. The old figure 1E has now become 1C and is now numbered accordingly.

Reviewer #2 (Significance (Required)):

The current study is limited in scope. Without functional data for these interactions, the overall significance of the study is likely limited.

**Referee Cross-commenting**

My overall assessment is similar to the other two reviewers (particularly reviewer 3): the study is rather descriptive, limited in scope, and lacks mechanistic understanding of BAP1 functions: for example, see reviewer 1 comment "it would be interesting to further investigate the significance of some key interactions"; reviewer 3 comment "The present manuscript contained very little information beyond description of BAP1 interactomes and subsequent validation of BAP1-COPI interaction. In the very least, I would recommend for the authors to explore contextual significances and/or regulations of the novel BAP1-COP1 interaction."

Reviewer #3 (Evidence, reproducibility and clarity (Required)):

In the present manuscript, Baas and colleagues seek to identify novel BRCA1-associated protein 1 (BAP1) interacting partners. To do so, the authors performed affinity purifications of different GFP-tagged BAP1 constructs in combination with mass spectrometry from either wild-type HeLa or those, which endogenous BAP1 expression had been knocked-out using CRISPR/Cas9. MS analysis of the pull-downs revealed COPI as a novel cytoplasmic interactor for the full-length GFP-BAP1 in addition to to other previously known BAP1 interactors such as HAT1, ASXL1/2, FOXK1/K2, OGT.

The authors subsequently went on to validate the BAP1-COPI interaction and observed that such interaction was independent of the canonical COPI binding motifs KxKxx present in the BAP1 C-terminus.

**Major comments:**

The present manuscript contained very little information beyond description of BAP1 interactomes and subsequent validation of BAP1-COPI interaction. In the very least, I would recommend for the authors to explore contextual significances and/or regulations of the novel BAP1-COP1 interaction.

Please see our general response and reviewer 1 point 3.

The present manuscript could be written in more concise manner.

We have shortened the results section as discussed under reviewer 1 point 2 to make it more concise.

Reviewer #3 (Significance (Required)):

While the present study may provide a resource to gain further understanding on BAP1 molecular functions, it is very difficult to appreciate the significance of the presence manuscript in the current descriptive form.

We have expanded the discussion to better explain the significance of our findings.

1. Sahtoe, D.D., et al., BAP1/ASXL1 recruitment and activation for H2A deubiquitination. Nat Commun, 2016. 7: p. 10292.
2. Ventii, K.H., et al., BRCA1-associated protein-1 is a tumor suppressor that requires deubiquitinating activity and nuclear localization. Cancer Res, 2008. 68(17): p. 6953-62.
3. Mashtalir, N., et al., Autodeubiquitination protects the tumor suppressor BAP1 from cytoplasmic sequestration mediated by the atypical ubiquitin ligase UBE2O. Mol Cell, 2014. 54(3): p. 392-406.

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

In the present manuscript, Baas and colleagues seek to identify novel BRCA1-associated protein 1 (BAP1) interacting partners. To do so, the authors performed affinity purifications of different GFP-tagged BAP1 constructs in combination with mass spectrometry from either wild-type HeLa or those, which endogenous BAP1 expression had been knocked-out using CRISPR/Cas9. MS analysis of the pull-downs revealed COPI as a novel cytoplasmic interactor for the full-length GFP-BAP1 in addition to to other previously known BAP1 interactors such as HAT1, ASXL1/2, FOXK1/K2, OGT.

The authors subsequently went on to validate the BAP1-COPI interaction and observed that such interaction was independent of the canonical COPI binding motifs KxKxx present in the BAP1 C-terminus.

Major comments:

The present manuscript contained very little information beyond description of BAP1 interactomes and subsequent validation of BAP1-COPI interaction. In the very least, I would recommend for the authors to explore contextual significances and/or regulations of the novel BAP1-COP1 interaction.

The present manuscript could be written in more concise manner.

Significance

While the present study may provide a resource to gain further understanding on BAP1 molecular functions, it is very difficult to appreciate the significance of the presence manuscript in the current descriptive form.

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

Summary:

BAP1 is a deubiquitinase that mainly functions to control H2Aub levels in nucleus. BAP1 is a tumor suppressor with mutations or deletions in several human cancers. Previous studies have identified many interacting proteins with BAP1, most notably its binding partners involved in PR-DUB complex, such as ASXL1/2, FOXK1/2, HCFC1, OGT etc. In this study, by using FRT-mediated recombination, the authors generated tagged-BAP1 expressed from its endogenous promoter and conducted AP-MS analysis to identify BAP1 interacting proteins in both nuclear and cytoplasmic fractions. These analyses identified several new BAP1-interacting proteins in cytoplasmic fractions, including histone acetyltransferase 1 (HAT1) and the heptameric coat protein complex I (COPI, which is involved in protein sorting and trafficking). The authors further confirmed the interactions between BAP1 and HAT1 (as well as a COPI subunit) at the endogenous level.

Major comments:

Overall, the current study has relatively limited data with limited scopes: basically a proteomic study focusing on one single protein (which has already been subjected to several proteomic studies in previous publications). There is a significant room for authors to improve this potentially interesting study (see below for specific comments), although this may take substantial additional efforts.

1. In the current manuscript, there is no data to further characterize the interactions between BAP1 and HAT1 (or COPI). For protein-protein interaction studies, readers generally are interested in information such as whether these bindings are direct or indirect (particularly for COPI because it contains multiple subunits), and which regions mediate the interactions?
2. No data to study the functional significance of the identified protein-protein interactions. This is a major weakness of the current study. For example, HAT1 is a histone acetyltransferase and mainly functions in the nucleus. Does the BAP1-HAT1 interaction in cytoplasm suggest that they have functions in cytoplasm independent of their canonical function in regulating transcription in the nucleus? Likewise, does BAP1-COP1 interaction suggest that somehow BAP1 is involved in regulating protein sorting and trafficking?
3. The identified interactions appear to be weak. These proteins are located near the edge of the significance curve in volcano plots (Fig. 1C-1D and others). The IP data also appear to be weak; for example, see Fig. 5D, it's hardly to see COPA blot in BAP1 IP. The COPA IB signal from 5% input WCL is probably hundred-fold stronger than that from BAP1 IP. Weak interactions do not necessarily mean they are not important; however, there is no functional data to support this claim (see the point above).
4. It seems that the entire study focuses on one specific cell line. Repeat the analyses in other cell lines can help boost the robustness and significance of the study.
5. The major interacting proteins they identified from the cytoplasmic fraction are still those mainly localized in nucleus (such as HCFC1, FOXK1/2). A western blotting to show nuclear vs cytoplasmic fraction is required.

Minor comments:

1. page 9 "A GFP coIP experiment of both GFP-BAP1 and the catalytic-dead BAP1 C91S mutant shows coimmunoprecipitation of HAT1 (Figure 1F)." here Fig. 1F should be Fig. 1E.

Significance

The current study is limited in scope. Without functional data for these interactions, the overall significance of the study is likely limited.

Referee Cross-commenting

My overall assessment is similar to the other two reviewers (particularly reviewer 3): the study is rather descriptive, limited in scope, and lacks mechanistic understanding of BAP1 functions: for example, see reviewer 1 comment "it would be interesting to further investigate the significance of some key interactions"; reviewer 3 comment "The present manuscript contained very little information beyond description of BAP1 interactomes and subsequent validation of BAP1-COPI interaction. In the very least, I would recommend for the authors to explore contextual significances and/or regulations of the novel BAP1-COP1 interaction."

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

Very interesting study on BAP1 tumor suppressor. The work needs further characterization of the interacting partners identified.

Significance

BAP1 is an important tumor suppressor mutated in several malignancies and the mechanism of action of this deubiquitinase are far from being completely understood.

This interesting work aimed at identifying novel cytoplasmic partners of BAP1 which can highly relevant to its tumor suppressor function. BAP1 is predominantly nuclear, but can also be found in the cytoplasm. New insights into the cytoplasmic functions of BAP1 are needed.

The manuscript is overall well written and the data are very solid.

The manuscript would need additional work before acceptance

Comments

1) The abstract can be improved to reflect the data of the manuscript.

2) The result section, manuscript could emphasize the results rather that the technical aspects

3) It would be interesting to further investigate the significance of some key interactions

4) The discussion is quite short and can put the findings in perspective

5) It would be interesting to test some cancer-associated mutations

6) Figure 4 can be improved

7) Yu et al MCB 2010 is one of the key papers on BAP1 purification and can be cited

8) The authors can discuss potential substrates of BAP1 and mechanism of deubiquitination

Referee Cross-commenting

I agree with the comments of Reviewer #2

Transparent Peer Review

---

Download the complete Review Process [PDF] including:

• reviews
• authors' reply
• editorial decisions

</br>

Transparent Peer Review

---

Download the complete Review Process [PDF] including:

• reviews
• authors' reply
• editorial decisions

</br>

Transparent Peer Review

---

Download the complete Review Process [PDF] including:

• reviews
• authors' reply
• editorial decisions

</br>

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

Learn more at Review Commons

---

Reply to the reviewers

Reviewer #1

Summary:

In this work the authors present a simple mathematical model for the distribution of morphogen molecules that travel via cytonemes through a 1- dimensional system. This model is used as a basis for a software package called Cytomorph that takes as an input a set of experimentally measured distributions of cytoneme dynamics as well as experimenter determined parameters such as contact probability and method of cytoneme growth and retraction. The Cytomorph package then outputs spatial and temporal information on the distribution of morphogen as well as cytonemes and their contacts with cells and other cytonemes, all obtained over thousands of simulation runs. A number of in silico experiments are then performed to show that these outputs agree with experimentally measured morphogen distributions of Hedgehog in the imaginal wing disc and abdominal histoblast nest. Further in silico experimentation is done to study how this distribution is affected by a wide array of parameters such as producer row number, cytoneme connection method, and connection probability function. Comparisons to the traditional diffusion based model are also made. The authors find a suite of results based on these experiments and accordingly present the Cytomorph software package as a useful and adaptable tool for the community.

Major comments:

While the various in silico experiments present an expansive and exhaustive study of the different ways in which Cytomorph can be used to examine a cytoneme based distribution system, the machinery behind the software is left notably underdescribed. The authors do not sufficiently make clear what exactly happens within each iteration of the simulations run by Cytomorph, leaving the results irreproducible without the reader going into and deciphering the software code itself.

In order to improve the description of the mathematical and computational steps behind the software, we have created a visual organigram (new Supplementary Figure S.1) with a detailed depiction of the steps. We have also included a short description in the main text and an extended explanation in the Supplementary Material section.

Some of the specific details left undiscussed are how it is determined when and where a cytoneme will spawn or what its maximum length will be, the dynamics of morphogen transport within the cytonemes, the effects of one cytoneme making multiple connections on how much morphogen is delivered through each connection, and where exactly stochasticity is introduced so as to allow for variations between simulation runs; amongst others.

In the new description of the software steps, we have tried to address the Referee’s comments about the dynamics and stochasticity in more detail. In order to help the understanding of the variables, we have also tried to improve their description in the main text.

Additionally, when the authors investigate the diffusion model their stated boundary conditions do not match those presented at the end of the Materials and Methods section. The initial condition u(x,0)=0 and boundary condition du(L,t)/dt=0 represent a perfectly absorbing molecule sink at the x=L end of the system, not the reflecting boundary condition du(L,t)/dx=0 that would correspond to a zero morphogen flux.

We thank the Referee for noticing this annotation mistake since the equation is really dx instead of dt. We have corrected this error and included in the Supplementary Materials the exact lines of code used in Matlab pdepe to certify the conditions used in the resolution of the diffusion equation (new Supplementary Figure S.10).

Finally, while the authors spend a great deal of effort analyzing signal variability between simulation runs, there is no effort made to account for the inherently stochastic nature of molecular production, movement, and degradation. Particularly if molecule numbers are small, fluctuations in these processes could greatly increase signal variability. The authors should either address why these fluctuations are negligible or include them in the modelling.

This work is mainly focused on the transport of the morphogen; other terms as degradation were introduced directly using published experimental data. Regarding the main concern about the negligibility of the fluctuations for cytoneme transport, we agree with the Referee on the importance of this point. Therefore, we have included a detailed description of the variability and fluctuations in a new section of the Supplementary Material. To help its understanding, we have also included a new Supplementary Figure (Supplementary Figure S.11).

The largest fluctuations were found at the tail of the morphogen gradient (last rows of receiving cells). Since this corresponds to the region where the amount of morphogen is low, the absolute fluctuations do not change the activation of the low-threshold target. We then conclude that those fluctuations are biologically negligible for our study.

Minor comments:

The authors should double check all equation and figure references as I noted several instances in which it appeared that the wrong equation or figure was being referred back to. Similarly, the authors should double check the equations themselves, particularly those in the supplemental material.

We thank the Referee for noticing these mistakes. We have reviewed those references in order to fix the wrongly linked ones.

Eqs. SM1.1 and SM1.2 have a plethora of parameters with a wide array of different sub- and superscripts that are left unexplained and possibly incorrectly labelled in some cases,

Equations SM1.1 and SM1.2 described a general form of Triangular and Trapezoidal dynamics and the different sub- and superscripts come from the published experimental data. Nevertheless, in order to make them more intuitive we have simplified the expressions and included a more detailed description of those parameters and their scripts in the revised version.

while the second line of Eq. SM2.2 is nonsensical unless r_I*p=0 and p_i<=1.

We thank the Referee for noticing the uncertainty in this equation, since it was written in an iterative syntax as it is coded in the software. Therefore, in the code we did not have this nonsensical range of data, but we agree that it should be specified with a mathematical syntax as the rest of the equations in the manuscript. Therefore, we redefined the notation and specified better the numerical domains of those variables.

Additionally, the notation used in Figs. 5 and 6 as well as the bottom part of Fig. 7 is confusing. The caption should more explicitly state what the various expressions in the second row of each column represent.

The second row represents the statistical analysis between cases coded in a color matrix, as it is described in the footnote. We thank the Referee for this recommendation because this is not the usual representation. Therefore, we have changed the previous explanation to one hopefully clearer and intuitive; we have also included a specific label in the figures.

In Fig. 5A specifically it is unclear what exactly the variable phi represents.

Phi is a widely used annotation in biology to define cell size diameter and cell position. We didn´t realize it could be unclear. For a better understanding within a multidisciplinary field we have changed this symbol.

Does it have anything to do with the phi that is used as a position variable for the cells, and if it is a ratio of cytoneme length to cell diameter then why does it have units of microns?

We agree that this phi notation is confusing. It has been used to indicate distance position as well as cell diameter. Although these variables are biologically related, in the new version of the manuscript we have changed the notation to separate both concepts and avoid misunderstandings.

Significance:

As the Cytomorph model and software can be applied to a wide variety of systems involving morphogen transport via cytonemes, it provides a technical advance in our ability to analyze and discuss the results of measurements on cytonemes in a more homogenous way. This work and the resulting software is particularly applicable to and build off of studies done by other groups that study the dynamics of cytonemes such as the Kornberg lab (works from which are cited by the authors) and the Scholpp lab (such as Stanganello E, Scholpp

S. Role of cytonemes in Wnt transport. J Cell Sci. 2016; 129(4):665-672), and as such it is experimental labs such as these that will be the most interested in this manuscript and its findings.

My field of expertise lies primarily in stochastic modeling and linear response theory. As such, I feel I do not have sufficient expertise to evaluate the experimental methods outlined in this manuscript and determine their level of scientific rigor.

Reviewer #2

The manuscript "Improving the understanding of cytoneme-mediated morphogen gradients by in silico modelling" addresses the role of in silico modelling in understanding pattern formation via cytonemes: filopodia that transport signalling molecules to and from cells. Investigating the role of cytonemes and, in particular, their dynamics, during development is an important and emerging field in developmental biology, and there is great potential for mathematical modelling to aid in understanding these processes.

The present manuscript attempts to derive a general set of equations describing pattern formation in the context of cytonemes, akin to that of the classic Turing model of morphogenesis. The authors replace the standard diffusion term in the PDE with a non-local term, intended to represent transport via cytonemes. This model is then posed over a one-dimensional domain with a source at one end and no flux boundary conditions at the other and is shown to be able to generate a morphogen gradient profile that could pre-pattern a biological tissue. The model is tested against a key experimental system, namely, Hh signalling in the Drosophila wing imaginal disc and is shown to reproduce some experimental results. Finally, the authors have developed a Matlab-based software package that they claim will be applicable to a wide range of systems. This GUI-based software allows users to input experimentally measured averages of cytoneme properties and explore the effect of these properties on tissue patterning.

My primary concern is that the paper presents itself as a mathematical model of cytoneme formation in general. The authors themselves state in their introduction that the mechanisms for cytoneme generation and maintenance are presently unknown. In fact, it is not even known if they are consistent across biological systems (and in fact, are probably not in general). As such, any present instantiation that connects cytoneme dynamics to tissue patterning can only hope to be specific to a particular system (in this case, the Drosophila wing imaginal disc.

As mentioned in the introduction, the connection of cytonemes with patterning has been described in several works. We had included a list of publications describing the implication of cytoneme-mediated signaling for several morphogens (FGF, Egf, Hh, Dpp, Wnt or Notch) and in many vertebrate and invertebrate systems (Drosophila, chicken, Xenopus, Zebra fish, mouse and human tissue culture cells).

Whilst one may use general models (like the heat equation) to study pattern formation since it requires only specification of parameters, the model here requires specification of families of functions, that are likely to differ from context to context and so the model is not general.

Our model inputs are parameters determined experimentally rather than families of functions. This misunderstanding might derive from the use of triangular and trapezoidal dynamics, which are equations included in the software code but not input functions. To avoid this confusion, we have specified the input data in tables S.1 and S.2 and clarified in the main text that the triangular or trapezoidal family of functions are just the names for the basic dynamics of cytonemes (triangular for elongation and retraction, and trapezoidal when there is a stationary phase in between).

Ultimately, the model is a statistical modelling framework masquerading as a mechanistic one.

In this work, we have not specified the mathematical area to which the model belongs. Furthermore, we always explicitly described the different variables and functions modeled. Therefore, we do not understand what the supposed masquerade is.

As further evidence of the lack of generality of the model, the studied domain is only one dimensional and has signalling sources at one end. This scenario is perfectly adequate for theoretical explorations of pattern-forming systems but is highly unlikely to capture the geometrical intricacies of real-world systems (and I note that even in the diffusive case, boundary conditions are critical for understanding what patterns ultimately arise for a given system).

We agree with the Referee that there are cases in biological systems in which it is required to work in 2D or even 3D to have a full comprehension of the process. Nevertheless, those are mainly related to biological patterns rather than to biological signaling gradients, which usually are studied (experimental and theoretically) in 1D. Therefore, we have limited our model to this case and compared our in silico results with the published experimental data. In any case, we have emphasized in the text that our model is limited to signaling gradients with the source at one end, which is the case of the best studied morphogens: Hh (Sonic-hh), Dpp (BMP) or Wg (Wnt).

Actually, as prove of the generality of the model, we have predicted different properties of Dpp and Wg gradients using our model. We then validated the simulated results using the experimental data obtained from independent publications.

To simulate their model, the authors need to specify triangular and trapezoidal functions, which are unlikely to be generalisable to all contexts. As such, the model is not general and, in particular, there is no way to change the software to make it so.

Cytonemes are filopodial structures based on actin filaments that polymerize and depolymerize to elongate and retract. This is a general process for all filopodial structures and it is why cytonemes were classified in a previous published work as a triangular behavior or, if this dynamic has a stationary phase, as a trapezoidal behavior (Gonzalez-Méndez et al., 2017). Therefore, these functions are just a categorization introduced to better describe the intrinsic dynamics of cytonemes, that could be applied to most of the experimental cases. To attend this Referee’s concern, we have included in the introduction a more detailed description of these behaviors, as well as the references of publications describing the dynamic behaviors of cytonemes for different morphogens and in different organisms.

Trying to make a generalization for all cases, we included in the model those situations in which the cytonemes were static rather than dynamic (detailed simulations comparing dynamic and static cases can be found in the old Supplementary Figure S.5 A (now S.7 A)).

We have concluded that the model can be considered generalizable since it includes the simplest and most general cases in terms of cytoneme dynamics.

Whilst the development of a GUI for this scenario is a nice contribution, I feel that the lack of generalisability will, at best, mean that the software enjoys little use, and at worst, may lead researchers unfamiliar with the modelling context to misuse it in error.

Once we knew the model could be generalized, we were concerned about the misuse of the mathematical model, and that was the reason why we decided to develop a GUI as simple as possible.

Furthermore, in the online repository there is, together with the open software, an user guide of Cytomorph with a full description of parameters, variables and outputs and how to use them properly.

In my opinion, this work would be better suited as a presentation of specific mathematical modelling of tissue patterning in the Drosophila wing imaginal disc. In this case, many of the above concerns would be addressed.

We have rewritten part of the text to indicate the limits of the model and make clear that it has been tested experimentally for the Hh pathway and in two different developing systems: wing imaginal discs and abdominal histoblast nests.

As evidence of a more general use of Cytomorph, we have added in the revised version of the manuscript a new section focused on data prediction for the gradients of Dpp and Wg. We have also included supplementary figures that validate the predictions of our model using published experimental data.

That said, there are still a number of issues with the presentation of the model and results. I shall detail these in the bullet point list below:

1. The domain for Eq. 1 needs to be made explicit. Later, it appears that the domain is a closed one-dimensional interval, but the use of arrows here implies that x is a vector and hence x ∈_ D ⊂ _Rn with n > 1.

We initially described the general equation for morphogens as x ∈ ℝ𝑛 and later we limited it to 1D. This is why at the beginning x, as a vector, contained an arrow, although later it was a scalar variable. Since we were interested in 1D in this work, to avoid this kind of misunderstanding we have rewritten from the beginning the equations as 1D and clearly specified the x domain used: the set of natural numbers x ∈ ℕ0.

1. It is unclear over what the sum in Eq. 2 is being taken.

The sum in Eq. 2 is over the number of producing cell rows. We have changed the notation to clarify this point.

1. The statement "we used the discrete cell position x = φ as spatial coordinate" is vague and does not help the reader understand the discretization._

The number of cell diameters is a widely used discrete unit for position in Developmental Biology. As we expect the readers of this publication to be multidisciplinary, we have changed the notation to avoid misunderstandings and clarify this discretization.

1. p is used both as a probability and as an index for producer cells. This is confusing._

We have changed the notation to avoid misunderstandings.

1. As previously stated, the choice of trapezoidal/triangular cytoneme dynamics is not general. More work needs to be done to showcase how the authors came to the conclusion that this is the best choice, and how the functions (and their associated parameters) describing them were selected.

The names triangular and trapezoidal stand for the published dynamics for elongation and the retraction of cytonemes and we already argued about its generality. As we specified in the manuscript, these types of behaviors have been experimentally observed and, therefore, we considered that the experimental observation was reason enough to include them in the model. If more details are required, the Material and Method section and the Supplementary Table S.3 show that the times measured for triangular and trapezoidal dynamics are statistically different and, consequently, both behaviors have to be considered.

As mentioned in the manuscript, the associated parameters represent the times and velocities for the elongation or retraction that have already been thoroughly analyzed and published (González-Méndez et al., 2017). The question of the Referee about how these functions affect the gradient is answered in the text and in Figure 7 F.

1. I can see how Type 1 and Type 2 cytonemes could be expanded naturally to a higher dimensional case, but it is not clear how Type 3 cytonemes could be, since the probability of any two cytonemes occupying the same space in higher dimensions is likely to be small (if they are imbued with independent dynamics).

We agree with the Referee on this point. It is something that shall be considered for future improvements of the model in higher dimensions. For instance, a complex scenario in 2D will be required of a cytoneme guiding model. Nevertheless, since the present study is limited to 1D, this concern is not applicable for the current model.

1. The statement: "the distance between cells must be smaller than, or equal to, the maximum length of the cytonemes" seems inconsistent with the equations below since λ(t) does not appear to be a maximum length.

The length of the cytonemes is controlled as a dynamic function described by λ(t). Our statement referred to the maximum length for each time step that is given by λ(t). We agree that the initial statement could lead to misunderstanding, so we have suppressed the word “maximum”.

1. I think the authors are confusing probabilities and rates in their discussion of the model. Eq. 1 is a density model and so calling events probabilities here is slightly misleading. As a more general statement, I am currently interpreting contact function C as one defined as a rate, rather than as a set of probabilistic terms. If the latter is true, then Eq. 1 is invalid since it mixes processes at different levels of description._

We thank the Referee for this comment. We have studied in depth this observation but we could not exactly find why the Referee considers that the model is working at different levels. Even though we could not find where in the text we called “probabilities” to the events of eq1, we rewrote the text to make clear what we consider either probability or rate. In addition, in the Supplementary Material section we clarify how the model works and at what levels of modeling we are working.

Significance

In general, the paper is well written, however, the focus of the findings should be on patterning within an epithelium such as the Drosophila wing imaginal disk.

The work will be interesting for the developmental biology community as well as for the upcoming biomathematical modelling community.

Expertise: Developmental biologist with experience in tissue patterning and morphogen gradients

Referees cross-commenting

I agree with Reviewer 3 that the importance of cytoneme-mediated signalling has been described in several systems - invertebrates and vertebrates. However, I think the focus of this work in particular should be on cytoneme signalling in the wing imaginal disc. IMO, this would not limit the conclusion but rather focus it and make it then applicable to epithelial tissues in general. I agree with the other point.

Reviewer #3

There is much to like in this thoughtful and worthwhile study that develops mathematics to describe how cytonemes might generate experimentally observed Hh gradients. Two suggestions:

1. I am not equipped to evaluate the mathematics and as a non-expert would find it helpful if the authors explicitly stated at the outset what assumptions they took, the specific contexts they sought to model, and the parameters that they explored.

We agree with the Referee on the excessively mathematical focus of our interpretation of the results in the old version of the manuscript. We have rewritten part of the text to clarify the biological implications of the variables and simulations explored.

Am I correct that they assume that the Hh gradient correlates with a cytoneme gradient, that all cytoneme contacts have the same duration and exchange equivalent amounts of Hh, and that the variables that were characterized are cytoneme length distributions, cytoneme extension rate, contact duration, and cytoneme density?

Since the mechanism of morphogen exchange is not fully identified, we assumed the simplest case in which all the contacts have the same duration and exchange the same amount of morphogen. Using this approach, we were able to reproduce the gradient and concluded that it is not strictly necessary to propose a more complex mechanism to establish a graded distribution of morphogens. We therefore worked under this assumption.

The variables characterized were the ones pointed out by the Referee, mainly cytoneme features, as the cytoneme length distributions or the different parameters of the temporal dynamics. We tried to define better these variables in the new version of the manuscript.

1. One of the unusual features of the Hh gradient in the wing disc is that the size of the posterior compartment field of Hh-producing cells is large relative to the size and extent of the Hh gradient in the adjacent anterior compartment. Wing discs with large hh mutant clones, wing discs with large smo mutant clones, and wing discs with ttv mutant clones that block Hh uptake provide evidence that the Hh gradient is constituted with Hh that is produced by many cells, some that are far from the compartment border as well as some that are close. Has this been factored into the author's model?

Indeed. Being aware of the importance of the size of the signal source, we simulated how changing the size of the posterior compartment affects the gradient (altering the number of producing cell rows involved, figure 5B). In the old version of the manuscript we had focused on the theoretical approach, so we thank the reviewer for noticing that we should introduce a more biological point of view. Therefore, we included in the revised version of the manuscript a biological interpretation of how our simulations can help to understand the question posed by the reviewer.

Does the fact that the relative size of the posterior compartments and Hh gradients in the histoblasts is not as extreme as it is in the wing disc influence their model?

Following the Referee’s question, we decided to simulate the influence of the relative size of the posterior compartment in the abdominal histoblast nests. We found that in both wing discs and histoblasts, the size of the posterior compartment affects the gradient but in a different scale factor. We have included these data in the revised version of the manuscript (new supplementary figure S.5).

Interestingly, this feature of the Hh gradient in the wing disc is not shared with other gradients in the wing disc such as the Wg, Dpp, and Bnl gradients. I would be interested to know if the author's model can be queried to suggest what properties might contribute to this difference?

In order to answer the reviewer question, we have used our model to tentatively simulate Wg and Dpp gradients. Our preliminary results suggest that considering only cell position and cytoneme length, the Wg and Dpp gradient lengths can be predicted in wing imaginal disc. Nevertheless, each morphogen has its own particularities and further studies are required for a precise simulation of these gradients. We included these results in a new section of the manuscript and in the new Supplementary Figure S.9.

Significance

This is an important contribution to gaining a basic understanding of the role of various properties of dynamic cytonemes to gradient formation.

Referees cross-commenting

I discount the apparently strongly held opinion of Reviewer #2 that "it is not even known if they [cytonemes] are consistent across biological systems (and in fact, are probably not in general)". I do not know where this comes from and do not think that such opinions are appropriate for anonymous reviews.

Cytoneme-mediated signaling has in fact been observed and characterized in many diverse biological systems. I submit that in contrast, mechanisms of dispersion based on diffusion are inferred and lack direct experimental evidence. I do agree that it is fair to ask the authors to carefully describe their work in the context of epithelial signaling, but it is not correct to ask them to limit their conclusions to the wing disc as the authors analyze both wing disc and histoblast signaling. They clearly state that their work is limited to 1D and so we understand that it is inadequate to model 3D morphologies. I do not criticize them for this.

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

There is much to like in this thoughtful and worthwhile study that develops mathematics to describe how cytonemes might generate experimentally observed Hh gradients. Two suggestions:

1. I am not equipped to evaluate the mathematics and as a non-expert would find it helpful if the authors explicitly stated at the outset what assumptions they took, the specific contexts they sought to model, and the parameters that they explored. Am I correct that they assume that the Hh gradient correlates with a cytoneme gradient, that all cytoneme contacts have the same duration and exchange equivalent amounts of Hh, and that the variables that were characterized are cytoneme length distributions, cytoneme extension rate, contact duration, and cytoneme density?
2. One of the unusual features of the Hh gradient in the wing disc is that the size of the posterior compartment field of Hh-producing cells is large relative to the size and extent of the Hh gradient in the adjacent anterior compartment. Wing discs with large hh mutant clones, wing discs with large smo mutant clones, and wing discs with ttv mutant clones that block Hh uptake provide evidence that the Hh gradient is constituted with Hh that is produced by many cells, some that are far from the compartment border as well as some that are close. Has this been factored into the author's model? Does the fact that the relative size of the posterior compartments and Hh gradients in the histoblasts is not as extreme as it is in the wing disc influence their model? Interestingly, this feature of the Hh gradient in the wing disc is not shared with other gradients in the wing disc such as the Wg, Dpp, and Bnl gradients. I would be interested to know if the author's model can be queried to suggest what properties might contribute to this difference?

Significance

This is an important contribution to gaining a basic understanding of the role of various properties of dynamic cytonemes to gradient formation.

Referees cross-commenting

I discount the apparently strongly held opinion of Reviewer #2 that "it is not even known if they [cytonemes] are consistent across biological systems (and in fact, are probably not in general)". I do not know where this comes from and do not think that such opinions are appropriate for anonymous reviews.

Cytoneme-mediated signaling has in fact been observed and characterized in many diverse biological systems. I submit that in contrast, mechanisms of dispersion based on diffusion are inferred and lack direct experimental evidence. I do agree that it is fair to ask the authors to carefully describe their work in the context of epithelial signaling, but it is not correct to ask them to limit their conclusions to the wing disc as the authors analyze both wing disc and histoblast signaling. They clearly state that their work is limited to 1D and so we understand that it is inadequate to model 3D morphologies. I do not criticize them for this.

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 manuscript "Improving the understanding of cytoneme-mediated morphogen gradients by in silico modelling" addresses the role of in silico modelling in understanding pattern formation via cytonemes: filopodia that transport signalling molecules to and from cells. Investigating the role of cytonemes and, in particular, their dynamics, during development is an important and emerging field in developmental biology, and there is great potential for mathematical modelling to aid in understanding these processes.

The present manuscript attempts to derive a general set of equations describing pattern formation in the context of cytonemes, akin to that of the classic Turing model of morphogenesis. The authors replace the standard diffusion term in the PDE with a non-local term, intended to represent transport via cytonemes. This model is then posed over a one-dimensional domain with a source at one end and no flux boundary conditions at the other and is shown to be able to generate a morphogen gradient profile that could pre-pattern a biological tissue. The model is tested against a key experimental system, namely, Hh signalling in the Drosophila wing imaginal disc and is shown to reproduce some experimental results. Finally, the authors have developed a Matlab-based software package that they claim will be applicable to a wide range of systems. This GUI-based software allows users to input experimentally measured averages of cytoneme properties and explore the effect of these properties on tissue patterning.

My primary concern is that the paper presents itself as a mathematical model of cytoneme formation in general. The authors themselves state in their introduction that the mechanisms for cytoneme generation and maintenance are presently unknown. In fact, it is not even known if they are consistent across biological systems (and in fact, are probably not in general). As such, any present instantiation that connects cytoneme dynamics to tissue patterning can only hope to be specific to a particular system (in this case, the Drosophila wing imaginal disc. Whilst one may use general models (like the heat equation) to study pattern formation since it requires only specification of parameters, the model here requires specification of families of functions, that are likely to differ from context to context and so the model is not general. Ultimately, the model is a statistical modelling framework masquerading as a mechanistic one.

As further evidence of the lack of generality of the model, the studied domain is only one dimensional and has signalling sources at one end. This scenario is perfectly adequate for theoretical explorations of pattern-forming systems but is highly unlikely to capture the geometrical intricacies of real-world systems (and I note that even in the diffusive case, boundary conditions are critical for understanding what patterns ultimately arise for a given system). To simulate their model, the authors need to specify triangular and trapezoidal functions, which are unlikely to be generalisable to all contexts. As such, the model is not general and, in particular, there is no way to change the software to make it so. Whilst the development of a GUI for this scenario is a nice contribution, I feel that the lack of generalisability will, at best, mean that the software enjoys little use, and at worst, may lead researchers unfamiliar with the modelling context to misuse it in error.

In my opinion, this work would be better suited as a presentation of specific mathematical modelling of tissue patterning in the Drosophila wing imaginal disc. In this case, many of the above concerns would be addressed. That said, there are still a number of issues with the presentation of the model and results. I shall detail these in the bullet point list below:

1. The domain for Eq. 1 needs to be made explicit. Later, it appears that the domain is a closed one-dimensional interval, but the use of arrows here implies that x is a vector and hence x ∈ D ⊂ Rn with n > 1.
2. It is unclear over what the sum in Eq. 2 is being taken.
3. The statement "we used the discrete cell position x = φ as spatial coordinate" is vague and does not help the reader understand the discretization.
4. p is used both as a probability and as an index for producer cells. This is confusing.
5. As previously stated, the choice of trapezoidal/triangular cytoneme dynamics is not general. More work needs to be done to showcase how the authors came to the conclusion that this is the best choice, and how the functions (and their associated parameters) describing them were selected.
6. I can see how Type 1 and Type 2 cytonemes could be expanded naturally to a higher dimensional case, but it is not clear how Type 3 cytonemes could be, since the probability of any two cytonemes occupying the same space in higher dimensions is likely to be small (if they are imbued with independent dynamics).
7. The statement: "the distance between cells must be smaller than, or equal to, the maximum length of the cytonemes" seems inconsistent with the equations below since λ(t) does not appear to be a maximum length.
8. I think the authors are confusing probabilities and rates in their discussion of the model. Eq. 1 is a density model and so calling events probabilities here is slightly misleading. As a more general statement, I am currently interpreting contact function C as one defined as a rate, rather than as a set of probabilistic terms. If the latter is true, then Eq. 1 is invalid since it mixes processes at different levels of description.

Significance

In general, the paper is well written, however, the focus of the findings should be on patterning within an epithelium such as the Drosophila wing imaginal disk.

The work will be interesting for the developmental biology community as well as for the upcoming biomathematical modelling community.

Expertise: Developmental biologist with experience in tissue patterning and morphogen gradients

Referees cross-commenting

I agree with Reviewer 3 that the importance of cytoneme-mediated signalling has been described in several systems - invertebrates and vertebrates. However, I think the focus of this work in particular should be on cytoneme signalling in the wing imaginal disc. IMO, this would not limit the conclusion but rather focus it and make it then applicable to epithelial tissues in general. I agree with the other point.

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:

In this work the authors present a simple mathematical model for the distribution of morphogen molecules that travel via cytonemes through a 1-dimensional system. This model is used as a basis for a software package called Cytomorph that takes as an input a set of experimentally measured distributions of cytoneme dynamics as well as experimenter determined parameters such as contact probability and method of cytoneme growth and retraction. The Cytomorph package then outputs spatial and temporal information on the distribution of morphogen as well as cytonemes and their contacts with cells and other cytonemes, all obtained over thousands of simulation runs. A number of in silico experiments are then performed to show that these outputs agree with experimentally measured morphogen distributions of Hedgehog in the imaginal wing disc and abdominal histoblast nest. Further in silico experimentation is done to study how this distribution is affected by a wide array of parameters such as producer row number, cytoneme connection method, and connection probability function. Comparisons to the traditional diffusion based model are also made. The authors find a suite of results based on these experiments and accordingly present the Cytomorph software package as a useful and adaptable tool for the community.

Major comments:

While the various in silico experiments present an expansive and exhaustive study of the different ways in which Cytomorph can be used to examine a cytoneme based distribution system, the machinery behind the software is left notably underdescribed. The authors do not sufficiently make clear what exactly happens within each iteration of the simulations run by Cytomorph, leaving the results irreproducible without the reader going into and deciphering the software code itself. Some of the specific details left undiscussed are how it is determined when and where a cytoneme will spawn or what its maximum length will be, the dynamics of morphogen transport within the cytonemes, the effects of one cytoneme making multiple connections on how much morphogen is delivered through each connection, and where exactly stochasticity is introduced so as to allow for variations between simulation runs; amongst others. Additionally, when the authors investigate the diffusion model their stated boundary conditions do not match those presented at the end of the Materials and Methods section. The initial condition u(x,0)=0 and boundary condition du(L,t)/dt=0 represent a perfectly absorbing molecule sink at the x=L end of the system, not the reflecting boundary condition du(L,t)/dx=0 that would correspond to a zero morphogen flux. Finally, while the authors spend a great deal of effort analyzing signal variability between simulation runs, there is no effort made to account for the inherently stochastic nature of molecular production, movement, and degradation. Particularly if molecule numbers are small, fluctuations in these processes could greatly increase signal variability. The authors should either address why these fluctuations are negligible or include them in the modelling.

Minor comments:

The authors should double check all equation and figure references as I noted several instances in which it appeared that the wrong equation or figure was being referred back to. Similarly, the authors should double check the equations themselves, particularly those in the supplemental material. Eqs. SM1.1 and SM1.2 have a plethora of parameters with a wide array of different sub- and superscripts that are left unexplained and possibly incorrectly labelled in some cases, while the second line of Eq. SM2.2 is nonsensical unless r_I*p=0 and p_i<=1. Additionally, the notation used in Figs. 5 and 6 as well as the bottom part of Fig. 7 is confusing. The caption should more explicitly state what the various expressions in the second row of each column represent. In Fig. 5A specifically it is unclear what exactly the variable phi represents. Does it have anything to do with the phi that is used as a position variable for the cells, and if it is a ratio of cytoneme length to cell diameter then why does it have units of microns?

Significance

As the Cytomorph model and software can be applied to a wide variety of systems involving morphogen transport via cytonemes, it provides a technical advance in our ability to analyze and discuss the results of measurements on cytonemes in a more homogenous way. This work and the resulting software is particularly applicable to and build off of studies done by other groups that study the dynamics of cytonemes such as the Kornberg lab (works from which are cited by the authors) and the Scholpp lab (such as Stanganello E, Scholpp S. Role of cytonemes in Wnt transport. J Cell Sci. 2016; 129(4):665-672), and as such it is experimental labs such as these that will be the most interested in this manuscript and its findings.

My field of expertise lies primarily in stochastic modeling and linear response theory. As such, I feel I do not have sufficient expertise to evaluate the experimental methods outlined in this manuscript and determine their level of scientific rigor.

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

Learn more at Review Commons

---

Reply to the reviewers

Response to Reviewers "Cell-cell communication through FGF4 generates and maintains robust proportions of differentiated cell types in embryonic stem cells"

Reviewer #1 (Evidence, reproducibility and clarity (Required)):

In this manuscript Raina et al. use an in vitro model of PE specification based on the transient overexpression of GATA4 in ESCs to show that the acquisition of primitive endoderm (PE) identity is governed at the population levels by cell-cell interactions mediated by FGF signaling. The authors further argue that the specification of a defined proportion of "PE" and "Epiblast" cells in a differentiating population of ESC is an emergent property of a system where paracrine signaling shifts the balance between two alternative stable states. Overall, the work does not reach radically new conclusions: broadly similar models are outlined in several other publications, including from the authors. Yet this study makes use of elegant genetic models and is particularly well executed. In addition, it includes a very accurate characterisation of the spatial range of FGF signaling activity that is original and adds on the existing knowledge. Moreover, the authors show novel evidence suggesting that GATA factors inhibits Fgf4 transcription and the activity of the FGF signaling pathway in ESCs.

We thank the Reviewer for commending the execution of the experiments, and for highlighting the novel insights that they bring. The Reviewer acknowledges that the specification of a defined proportion of PrE-like and Epiblast-like cells in a differentiating population of ESCs is an emergent property which is mediated by paracrine FGF4 signaling. This has not been experimentally demonstrated before. In contrast to the Reviewer’s assertion, we therefore think that our work does reach a conclusion that is radically different from previous experimental studies, a view that is also shared by Reviewer #3 below. In a revised version of the manuscript we will further emphasize the conceptual differences between published models that focus on single cell dynamics, and our experimental and theoretical demonstration of qualitatively different dynamics that emerge at the population level as a consequence of cell fate coupling.

**Two major points deserve further clarification:**

In this manuscript the authors claim that the proportions of cells acquiring PE fate is, at least in the experimental setup adopted, largely independent from the levels of GATA4 induction, and therefore of the initial state of the gene regulatory network regulating this cell fate transition. However, the authors should discuss how the current findings relate to their previous results, showing that the duration/levels of Gata4 induction, in a similar experimental setting, play an important role in determining the final proportion of cells cell acquiring "PE" fate. Absolute expression levels may be crucial for this distinction, but the authors seem to exclude this possibility (see figure S3).

The different roles of GATA4-mCherry induction levels for determining the final proportion of cells acquiring a PrE-like fate reported in our previous (PMID: 26511924) and the current work is because of important differences in the experimental settings between the two studies. In PMID: 26511924, we assayed PrE-like differentiation in medium supplemented with serum and LIF, which provides exogenous signals that promote PrE-like differentiation. These conditions reveal the function of the cell-autonomous circuit, in which GATA4-mCherry levels do control the probability of PrE-like differentiation. In the current work, we likewise observe that cell type proportions depend on GATA4-mCherry induction levels when we supply exogenous FGF4 during the differentiation of wild type cells (Figures S2C and S3D, lower panel). Differentiation in the absence of exogenous factors in contrast reveals the behavior of the coupled system, in which cell type proportions are independent from GATA4-mCherry induction levels.

Furthermore, in the present manuscript, we use new inducible cell lines in which the majority of cells can be induced above the critical GATA4-mCherry threshold required for PrE-like differentiation, in contrast to our previous study where the distribution of GATA4-mCherry induction levels was straddling this threshold.

In a revised version of the manuscript, we will more explicitly emphasize these important differences in the experimental design between the two studies, and discuss how the specific conditions in the present study lead to new conclusions.

Most importantly, the authors incorporate in their model the notion that GATA6 inhibits FGF signaling. It would be interesting to understand how such inhibition is mechanistically mediated. For instance GATA6 has been shown to bind in proximity of the Fgfr2 gene (Wamaitha et al., Genes and Dev., 2015). Alternatively, the authors show a direct effect on Fgf4 expression. The short time window of the reported repressive transcriptional effects (8h, Fig 2 middle), might suggest a direct regulation. The authors should test this possibility, and discuss what alternative modes of regulation could be envisaged (for instance, indirect effects mediated by Nanog). This is a key result that deserves a more detailed mechanistic characterisation.

The regulation of FGF signaling by GATA factors has been pointed out as a central new result of our study by all three reviewers that we will be happy to further expand on in a revised manuscript. Regulation of Fgfr2 expression by GATA6 as suggested by the ChIP-seq data in Wamaitha et al., 2015 (PMID: 26109048) is one possible mechanistic explanation that we will of course discuss.

Most importantly, we will test possible direct effects of GATA factors on Fgf4 expression that are indicated by the short timescales of the transcriptional effects shown in Fig. 2, as noted by the Reviewer. We have already mined the ChIP-seq data from Wamaitha et al., 2015 (PMID: 26109048) and found a GATA6-binding peak approximately 10 kb upstream of the Fgf4 start codon in a region that is highly enriched for GATA6 consensus binding sites. To test the functional role of this binding region, we propose to delete it by CRISPR-mediated mutagenesis in the inducible lines, and to test its ability to regulate reporter gene expression in heterologous assays.

To address the question of alternative modes of regulation of Fgf signaling through NANOG, we have already performed in situ mRNA stainings for Fgf4 expression in cells grown for 40 h in N2B27 medium. While Nanog expression is much reduced under these conditions, Fgf4 mRNA continues to be expressed, indicating that positive regulation through NANOG is not essential for Fgf4 mRNA expression in ESCs. We will add this data to a revised manuscript, and discuss its implications for the regulation of Fgf4 transcription (see also our response to Reviewer #3 below). As a complementary approach to further test the role of indirect effects mediated through NANOG, we will dissect more closely the timing of Fgf4 downregulation reported in Fig. 2B relative to the upregulation of the inducible GATA4-mCherry protein and the downregulation of NANOG protein.

**Minor points:**

Fig S1: The authors should show quantifications of Nanog and GATA6 levels before the beginning of the differentiation protocol.

We will be happy to add this data in a revised version, as part of a more extensive analysis of GATA4-mCherry and GATA6 expression at early stages of the differentiation protocol. See also our response to the next point.

Line 106: The authors write "the initially large proportion of GATA6+; NANOG+ double positive cells". It appears that at 16h of differentiation ESCs have already partitioned between Gata6 or Nanog expressing cells. The authors should rephrase the sentence to reflect what seems to be an almost total absence of truly double positive cells. Possibly, an analysis conducted at earlier time points could clarify these dynamics.

The Reviewer rightly points out that at 16 h of differentiation, most cells are already associated with one of two clusters in the NANOG/GATA6 expression space. The misleading classification of a large number of cells as double positive at 16 h was caused by applying a single gating strategy to the entire experiment, even though the mean expression levels of NANOG and GATA6 in the two clusters change significantly over time. We will update our gating strategy and rephrase this section to more appropriately describe cell clustering and gene expression dynamics over the time course. We will also extend Figure S1 with analysis of GATA6 and NANOG expression levels at earlier time points of the differentiation protocol, to test whether this allows detecting a truly double positive population.

Line 124: The authors write "... concentration dependent downregulation of NANOG expression". The effects may rather depend on the time of doxycycline stimulation.

We agree with the Reviewer that in isolation, the data shown in Fig. 1 and Fig. S2 leave open the possibility that the stronger downregulation of NANOG at higher GATA4-mCherry expression levels is caused by the extended time of doxycycline stimulation rather than GATA4-mCherry concentration. However, in our opinion, this concern is already addressed by the experiments performed in the four clonal lines with independent integrations shown in Figure S3. Here, the time of doxycycline induction is held constant, and a similar relationship between GATA4-mCherry and NANOG expression levels is observed as in the experiments where we modulate induction time in a single clonal line (compare Fig. S2A to Fig. S3B). In a revised version of the manuscript we will describe more clearly how the experiments shown in Figure S3 control for time-dependent effects of doxycycline stimulation.

Line 192: The authors write "...and confined to cells with low GATA4-mCherry expression levels". It would be helpful to have an indication of the cell boundaries, possibly showing localisation of a membrane bound protein.

We agree that more firmly establishing a correlation between GATA4-mCherry expression levels and Fgf4 mRNA expression in single cells would greatly benefit from co-staining with a plasma membrane marker. However, the protocol for mRNA in situ hybridization involves incubation steps with ethanol and formamide and is thus incompatible with staining for commonly used membrane markers. There is one commercially available membrane stain (CellBrite by Biotium) that promises to survive the treatments necessary for in situ hybridization and that we will try to use in our stainings. Should this not be successful, we will resort to identifying a subset of the cytoplasm corresponding to each nucleus by dilating nuclear masks that we will segment based on the DNA stain.

It would be interesting for the authors to discuss how the spatial range of FGF activity measured in culture could affect PE specification in the embryo.

During lineage specification in the embryo, Epi and PrE cells are initially arranged in a salt-and-pepper pattern (PMID: 16678776; PMID: 18725515; PMID: 30514631). In Fig. 4 and Fig. S9 of our manuscript, we show experimentally and theoretically how similar patterns in ESC colonies arise from the short range of FGF activity. In a revised version of the manuscript, we will discuss how the spatial range of FGF activity measured in culture provides a possible mechanistic explanation for the spatial arrangement of cell types in the embryo.

Reviewer #1 (Significance (Required)):

See above.

Reviewer #2 (Evidence, reproducibility and clarity (Required)):

In their manuscript entitled "Cell-cell communication through FGF4 generates and maintains robust proportions of differentiated cell types in embryonic stem cells" Raina et al study the effect of Fgf-signalling based local cell-cell communication for the establishment of PrE-like and Epi-like cells. The authors use an elegant, albeit artificial, system to analyse the effect of Fgf signalling on establishing 'normal' lineage proportions after transient induction of Gata4 expression. The main conclusions of the manuscript are: i) Gata6 positive cells emerge through short range Fgf4 based cell-cell cummunication. ii) Fgf4 signalling can compensate a wide range of initial levels of Gata6 expression and produce properly portioned cell identities. The authors also state that this mechanism could operate in a range of developing tissues.

**Major points:**

1. Fgf4 KOS ESCs are deficient in initiating epiblast lineage differentiation (Kunath 2007). Therefore, the effect studied by the authors might be multifactorial and the general inability of Fgf4 deficient cells to enter differentiation might contribute to the observed differentiation defects and defects of cell fate proportioning. Specifically, it could be expected that Nanog regulation is affected in Fgf4 mutants, although, to my knowledge, the specific phenotype of Fgf4 depletion has not been evaluated in Gata4 induced cell programming towards PrE. What steps have the authors taken to exclude an impact of general cell fate change defects in Fgf4 KO ESCs.

While it is true that Fgf4 mutant cells have a general deficiency in initiating epiblast lineage differentiation, it was already shown in the original publication by Kunath et al. (PMID: 17660198) that general differentiation of Fgf4 mutant cells is restored to wild type levels by supplementing the culture medium with 5 ng/ml recombinant FGF4. This is a concentration that is well within the range of concentrations applied in our study. In initial experiments to characterize our Fgf4 mutant lines, we have measured NANOG expression to test the effectiveness of recombinant FGF4 to restore epiblast lineage differentiation. We found that FGF4 treatment of Fgf4 mutant cells in the absence of doxycycline induction leads to a downregulation of NANOG expression, to levels comparable to those seen in wild type cells grown in N2B27. These data indicate that treatment with recombinant FGF4 rescues defects of general cell fate change in Fgf4 KO ESCs. We will add these data to Figure S4 of a revised manuscript, and explicitly mention the function of recombinant FGF4 to rescue lineage differentiation potential more generally.

Increasing the time of Gata4 expression results in increasing levels of Gata4 levels (Fig 1C). This is shown at the overall mean fluorescence level. However, it is important to also quantify how many cells do actually show some increase in Gata4 levels. Fig1D suggests that the number of Gata4 expressing cells is quite similar between 4h and 8h induction, but this needs to be quantified. An explanation for the apparent dosage independence of Gata4 could then be simple threshold effects, such that there is no additional effect of increased Gata4 levels in WT cells without any further requirement of feedback regulation after a certain threshold level of Gata4 is reached. Have the authors considered such a simple model?

The current version of the manuscript already contains quantifications of GATA4-mCherry expression levels in single cells - see Fig. S2A for the experiments where we vary doxycycline induction time, and Fig. S3B for experiments with independent clonal lines. This analysis confirms the Reviewer’s visual impression of Fig. 1D - the number of GATA4-mCherry expressing cells is similar for different induction times and clonal lines, such that the increase in overall mean fluorescence levels is mainly due to an increase in GATA4-mCherry expression levels in single cells. This analysis therefore rules out the simple model based on threshold effects proposed by the Reviewer. In a revised version of the manuscript, we will more explicitly discuss the quantifications in Fig. S2A and Fig. S3B.

An important point is that in the current setup distinguishing between dosage effects and effects of extended presence of Gata4 cannot be distinguished. Wouldn't titrating the amount of doxycycline used for induction be a more direct way to achieve different initial levels of Gata4 expression?

This concern has also been raised by Reviewer #1, and is addressed in detail in our response to their comment above. Briefly, in our opinion this concern is addressed in the current manuscript by the experiments performed in the four clonal lines with independent integrations (Figure S3). Here, the duration of doxycycline induction and hence time of GATA4-mCherry exposure is held constant, such that the only difference between the conditions is GATA4-mCherry dosage. We will discuss this important function of Fig. S3 in a revised version of a manuscript.

Unfortunately titrating doxycycline does not allow titrating transgene induction levels in a meaningful way, as sub-saturating doses of doxycycline lead to an increased heterogeneity in transgene expression with many non-expressing cells, rather than to reduced expression levels across all cells. See PMID: 17048983 for a possible explanation of this observation.

Another point the authors should appropriately discuss and consider is that a lack of effect of different doses/durations of Gata4 expression could be due to the fact that by the time Gata6 is induced, the levels of Gata4 in cells previously treated for different periods of time are no longer detectably different. Such a regulation would equally result in indistinguishable cell fate proportioning. Can the authors exclude such a regulation? This is an important point at the heart of the authors conclusion.

The Reviewer seems to suggest that by separating the initiation of GATA6 expression from the GATA4-mCherry pulse in time, the decision to initiate PrE-like differentiation could be independent from GATA4-mCherry concentration, thus explaining the robust cell type proportions. The data shown in Figs. S2C, S3D and Fig. 3 A - C clearly exclude such a regulation: In conditions where we supply recombinant FGF4, the proportions of the different cell types scale with GATA4-mCherry expression levels, indicating that GATA4-mCherry dose does indeed affect Gata6 expression. In a revised version of the manuscript we will discuss and consider how these observations argue against a model where the decision to initiate PrE-like differentiation occurs independently from GATA4-mCherry levels.

The authors make some general statements on cell differentiation (e.g. l205). They also claim that the Fgf4-based mechanism of lineage proportioning could act in a range of tissues during development. However, the use of the term differentiation for the induction of PrE-identity (or Gata-factor expression to be exact, see comment below) after Gata4 overexpression is problematic. The system chosen by the authors is entirely artificial. ES cells normally do not differentiate into extraembryonic cell types. It needs to be made clear in the manuscript that they do not study a differentiation process that normally occurs in the embryo or in differentiating ESC cultures. The system the authors are using would, in my opinion, rather qualify as cell programming or transdifferentiation than as differentiation. I suggest presenting the system using clearer unambiguous language and to try to avoid any generalisations based on an artificial transgene-overexpression based system. The results have to be presented with this limitation in mind.

To address the Reviewer’s concerns regarding terminology, we will expand on the relationship of our system to normal ESC differentiation and lineage specification in the embryo, and discuss its possible limitations. We disagree however with the Reviewer’s assertion that using a transgene-based overexpression system precludes drawing any general conclusions. Rather, the system allows mimicking Epi- and PrE-like differentiation in a uniquely accessible context, and thereby to exploit the molecularly simple regulation of this cell fate decision for studying basic principles of cell differentiation. This view is supported by Reviewer #3 in the referees cross-commenting section below, who emphasizes the value of such models and notes that they are very common in developmental biology.

It is unclear how 'PrE-like' (as stated e.g. in the abstract) the cells really are after a short pulse of Gata4 expression. No proper characterisation has been performed but needs to be included, if the authors want to term these cells PrE-like.

A recent study by Amadei et al. (PMID: 33378662) supports the notion that a short pulse of GATA4 expression can trigger bona fide PrE-like differentiation. In this study, the authors induced a similar doxycycline-inducible GATA4 expression system for 6 hours, and observed subsequent differentiation into several PrE derivatives, including the anterior visceral endoderm. In a revised version, we will cite this study to support our claim that the GATA6-positive cells are indeed PrE-like. Additionally, we offer to perform immunostainings with an extended panel of known PrE marker proteins to substantiate the PrE-like character of the GATA6-expressing cells.

How is the statement in l112 that "The clear separation between the two populations suggests that the increase in the proportion of double negative cells at the expense of GATA6+; NANOG- PrE-like cells beyond 40 h is mostly fueled by the downregulation of NANOG expression in the GATA6-negative cell population, combined with a slower proliferation of the GATA6-positive population, rather than by the reversion of PrE-like into double negative cells." supported by the data?

We realize from the comments of all three reviewers that this section was confusing and potentially misleading in the original version of the manuscript. In a revision, we will reword this paragraph to better bring out the major conclusions from the GATA6 and NANOG expression patterns shown in Fig. S1A. These data show that the majority of cells belong to one of two discrete clusters from 16 h onwards. The clear separation of the two clusters furthermore indicates that cells rarely switch their gene expression patterns. Given these observations, the changes of cell type proportions reported in Figure S1B can be explained as a consequence of slower proliferation of cells in the GATA6-positive relative to the GATA6-negative cluster. In addition, NANOG expression in the GATA6-negative cluster declines over time, such that progressively more cells are classified as double negative.

Would the data and modelling performed by the authors be in line with a model in which the decision to express Gata6 is a stochastic choice (with a certain probability based on the levels of Gata4 induction) that is then stabilized and reinforced by Fgf signalling rather than Fgf signalling having an instructive role?

The simulations shown for the Fgf4 mutant case in Fig. 3 D - G, right column, are based on a model in which the decision to express Gata is a stochastic choice with a probability based on the initial levels of GATA expression, and reinforced by FGF signaling. Thus, our data from the Fgf4 mutant, but not the wild type, are perfectly in line with such a model.

We realize from the Reviewer’s comment that we have not made sufficiently clear the conceptual differences between the models for the mutant and the wild type case. We suspect that this lack of clarity stems from the fact that the two models rely on the same circuitry, except for the regulatory link between GATA and FGF. This link however makes a crucial difference: It transforms the simple single cell input-output model of the mutant case, which is common to many previous publications, into a population level model with cell-cell feedback which shows new emergent behavior. And only this population level model, but not the single cell model for the Fgf4 mutant, can recapitulate the experimental data observed in the wild type. In a revised version of the manuscript we will expand on these crucial differences when describing the model and data in Fig. 3.

The statement in line 187 "This indicates that GATA4-mCherry expression negatively regulates FGF4 signaling during cell type specification." is not supported by the data. The authors show only a correlation and actually correctly say so in line 195.

Prompted by the comments of both Reviewer #1 and #3, we will carry out experiments to mechanistically explore the regulation of Fgf4 expression by GATA factors (see our response to Reviewer #1 above for a detailed description). Depending on the outcome of these experiments we will reword this statement.

In Fig 2F statistical analysis between the re-seeded conditions is required for the conclusion that "the proportion of PrE-like cells systematically increased with cell density". Replating itself appears to quite drastically impact lineage distribution. Do the authors have an explanation for this?

The p-value in line 221 of the original manuscript refers to a test for a linear trend between the three conditions following a one-way ANOVA in GraphPad Prism. We apologize that this has not been made clear and will add this information in a revised version.

The observation that replating drastically impacts lineage distribution is perfectly in line with the overall conclusion from this section, namely that FGF signaling is enhanced by cell-cell contacts. Replating strongly reduces the number of direct cell-cell contacts by disrupting the colony structure of the culture. Thus it is expected that the proportion of the PrE-like cells - which require exposure to FGF ligands - is reduced under these conditions compared to the condition that has not been replated. We will discuss this explanation in a revision.

Fig 2G shows a key experiment illustrating the local effect of Fgf4 expression on first and second neighbours. The authors have investigated this effect using a Fgf-signalling reporter. Why did they not assay Gata6 expression in this assay instead of a Spry reporter? This would be the experiment to show that also Gata6 expressing cells (after transient Gata4 induction) are clustered around Fgf4 producing cells and be a strong piece of evidence to show that local Fgf4 signalling and cell-cell communication is indeed involved in cell identity proportioning. The cell lines required for this experiment (including Fgf4 mutant Gata4 inducible ESCs) appear to be available.

We decided to measure the FGF4 signaling range with a Spry4:H2B-Venus reporter because its response time is faster than that of GATA6 expression during differentiation. Furthermore, the Spry4:H2B-Venus reporter provides a quantitative readout for FGF4 signaling, in contrast to a binary read-out that would be expected for GATA6 expression. We will be happy to discuss these considerations in a revised manuscript.

We agree that measuring FGF4 signaling range with Fgf4 mutant Gata4-mCherry inducible cells as suggested by the Reviewer constitutes a complementary approach to further corroborate the role of local FGF4 signaling in cell differentiation. However, we would like to stress that our demonstration of local FGF4 signaling is already supported by two fully orthogonal quantitative experiments, one relying on cell replating and the other one relying on the signalling reporter. The concept of local signaling is further supported by our quantitative analysis of the spatial arrangement of cell types in Fig. 4. The additional experiment suggested by the Reviewer is therefore unlikely to substantially change the paper’s conclusions, as also pointed out by Reviewer #3 in the referees cross-commenting section. Therefore, we offer to perform this experiment for a revision, but would like to seek the editor’s opinion if this is deemed necessary to make the paper acceptable for publication.

The authors conclude from data in Fig 3A that proper cell type proportioning depends on initial Gata4 levels in Fgf4 mutants, in contrast to WT cells where the initial levels appear more irrelevant. Is 10ng/ml too high a dose? Would using a lower concentration (such as ~2ng/ml suggested by Fig 2D to give WT-like distribution) result in a complete rescue of cell lineage proportioning in this assay? Formally a control of adding additional Fgf4 to WT cells will also ne needed to control for a potential effect of exogenous Fgf4 addition.

In our initial characterization of the Fgf4 mutant cell lines, we have performed experiments where we examined cell type proportions upon culture in the presence of different doses of FGF4 following doxycycline induction times between 1 h and 8 h. These experiments confirm the suspicion of the Reviewer that cell type proportions similar to the wild type can be obtained with a lower dose of 2.5 ng/ml FGF4 after 8 h of induction. For shorter induction times followed by differentiation in the presence of 2.5 ng/ml FGF4 however, cell type proportions were strongly skewed towards Epiblast-like cells. These data thus further support the major conclusion from Fig. 3A quoted by the Reviewer: Proper cell type proportioning in Fgf4 mutants depends on GATA4 levels, and this behavior is independent from the FGF4 concentration applied. We offer to add this data to a revised manuscript.

The effects of adding FGF4 to wild type cells are shown in Fig. S2C and S3D in the current version of the manuscript. This control has been performed in all experiments shown in Fig. 3A - C, but we decided to omit it for clarity. We are happy to add this information back in as requested by the Reviewer.

Does the model in Fig 3E consider potentially varying doses of exogenous Fgf4? Can the model also predict what happens if Fgf4 is added to WT cells, as suggested above as control? In general, the value of this model is unclear. Figure 3E is near impossible to understand, no quantitative information is given.

The model in Fig. 3E can of course be simulated with different doses of exogenous FGF4. These simulations recapitulate the experimental results described under point 10 above: Cell type proportions for the Fgf4 mutant case are skewed towards NANOG-positive cells at lower FGF4 doses, and vary with initial conditions irrespective of FGF4 dose. We offer to show the results of these simulations in a revised manuscript alongside the experimental data discussed above.

It is also possible to incorporate into the model addition of exogenous FGF4 to the wild type. Simulations of this condition confirm the experimentally observed increase in PrE-like cells shown in Fig. S2C and S3D of the current manuscript.

To help the reader digest Fig. 3E, we will add separating lines similar to the gates of the flow cytometry data in panel A, and indicate the proportion of cells in the respective quadrants.

The Reviewer’s comment that the value of the model is unclear indicates to us that we have not explained in sufficient detail the conceptual differences between the behavior of the model of the wild type and the mutant case. As detailed in our response to Reviewer’s comment 6. above, we will rewrite the text to bring out more clearly the insight that the model brings.

Fig4A: What were WT and Fgf4 mutant cells treated differently in this assay (8h vs 4h, respectively)?

The spatial arrangement of cell types in Fgf4 mutant cells has been assayed in two conditions that give similar cell type proportions as seen in the wild type, as motivated in lines 366 - 370 of the current manuscript. We decided to show the condition with 4 h induction followed by differentiation in the presence of 10 ng/ml FGF4 in the main Figure 4 because it is most similar to the condition that gives wild-type like cell type proportions in the Fgf4 mutant shown in the immediately preceding main Figure 3, while the condition that uses 8 h induction followed by differentiation in the presence of 2.5 ng/ml FGF4 refers back to the main Figure 2. We show both primary data and the complete analysis for the latter condition in Figures S8D and S10. Fig. S10 provides a direct comparison between the two conditions and clearly demonstrates that they show similar dynamics. We do not think that exchanging the two datasets between main and supplementary Figures will add value to the manuscript.

Does the interpretation that at 24h there is a difference in Fig 4C survive statistical scrutiny? Only few datapoints are shown and any apparent differences seem due to outliers rather than a shift in cluster radii. How often were these experiments independently repeated? This information is missing. In Fig 4B, I cannot appreciate any difference between cell lines.

We will perform statistical testing to assess whether the spatial arrangement of cell types is significantly different between the time points, and mention the results in the text.

To evaluate the spatial arrangement of cell types, we have performed two independent experiments in the wild type, and analyzed two conditions for the mutant case. In each experiment, we have analyzed at least eight positions per condition and control. Spatial clustering of wild type cells at 40 h is also observed in earlier Figures in the manuscript (e.g. Fig. 1D, S2B, S3C).

The similarities between wild type and Fgf4 mutant cells shown in Fig. 4B are not surprising and fully in line with the data shown in panel C, which shows that differences between time points are much more pronounced compared to the differences between genotypes. However, we realize that the micrographs and analysis plots in Fig. 4A and B were perhaps not fully representative for the aggregate behavior shown in panel C. In a revision, we will therefore show data from more representative colonies in panels A and B.

**Minor points:**

a) More information on statistics should be given in the Figures and legends.

To address this concern we will perform statistical tests for differences in proportions of the main cell types in Figures 1D and 3C. In addition, we will perform statistical testing on Fig. 4C as detailed in point 13 above.

b) Percentages should be indicated in the quadrants of the FACS plots of Fig 3A and E.

This is a good suggestion, we will add this information. See also our response to point 11 above.

c) What is the underlying evidence for the statement: "The specification of Epi- and PrE-like cells in ESCs shows both molecular and functional parallels to the patterning of the ICM of the mouse preimplantation embryo."

In the current manuscript, this statement is further substantiated in the subsequent paragraph (lines 483 - 503). We realize that this order is potentially confusing and will change it. We will further modify this section as part of our response to major point 3. above.

d) Fig 5C is difficult to interpret without a comprehensive decoding of colour information.

To facilitate interpretation of this panel, we will add a legend to decode the colour information of the traces (purple: VNPhigh, cyan: VNPlow)

Reviewer #2 (Significance (Required)):

This manuscript provides novel insights into the role of Fgf-mediated cell-cell communication to establish proper ratios of cell identities in a PrE-induction system. The authors provide some interesting data and interpretation. Overall, the significance is slightly impaired by the highly artificial nature of the studied cell fate specification event.

This manuscript will be of interest to readers working on early embryonic cell fate decision as well as researchers working on modelling of cellular processes.

My expertise lies in the field of cell fate decision and pluripotency.

Reviewer #3 (Evidence, reproducibility and clarity (Required)):

It is well established that FGF signalling plays a role in the partitioning of the Primitive Endoderm and Epiblast fates during preimplantation mammalian development. Recent work has shown that this fate decisions is associated with a mechanism that is able to maintain the proportions of the two fates stable in the face of perturbations. Here, the authors address this mechanism and show that it is dependent on FGF signalling and associated with the fate decision. In the process they suggest and test a novel mechanism based on short range FGF signalling. A series of carefully designed and executed experiments, refine and provide evidence for the model. This is an original and important piece of work that will influence the field of pattern formation.

Overall the manuscript is well written but, at least from the perspective of this reviewer, there are places in which clarity can be improved.

Lines 104 and ff: the description of the dynamics of the different populations fater the GATA4 pulse, can be clarified. The reference to the double negative population emerging from the PrEnd population is not clear. It is stated that the proportion of these cells increased continuously and it said to be at the expense of the decrease of the PrEnd population whose variation is referred to as 'slightly declined". How can a slight decline fuel a steady increase in the double negative?

Also, what are these double negative? Could they be cells differentiating into embryonic lineages?

We realize from the comments of all three Reviewers on this paragraph that it was confusing and potentially misleading in the original manuscript. In a revised version we will rewrite this section to clarify our interpretation of the data in Fig. S1. First, the clear separation of the two clusters observed in NANOG-GATA6 expression space indicates that cells rarely switch between the two clusters. Then, a likely explanation for the slow decline in the fraction of GATA6-positive cells is a slower proliferation compared to the GATA6-negative cells. Third, the increase in the proportion of double negative cells is caused by a progressive downregulation of NANOG expression in the GATA6-negative cluster. These NANOG expression dynamics are consistent with NANOG expression dynamics in epiblast cells of the embryo, and could indeed indicate differentiation towards embryonic lineages. We will mention this possibility in a revised manuscript.

See also our response to Reviewer #1 and Reviewer #2, point 5..

In Figure 1 and its discussion, it would be good to see a representation of the stability of the final proportions relative to the different initial conditions, a variation on 1E.

This is a good suggestion. In a revised version, we plan to add a panel to Fig. 1 in which we plot the final proportions of the different lineages versus the GATA4-mCherry expression levels for the different induction times. This will illustrate more clearly that the final proportions of cell types are largely independent from the initial conditions.

Paragraph lines 182 and ff: the report that GATA4 expression is able to suppress FGF4 signalling, autonomously is, at least for this reviewer, a novel and important result and one that impinges on the understanding of the process. The authors should emphasize this.

We agree with the Reviewer that the direct regulation of Fgf4 expression through GATA factors is a new regulatory link suggested by our data that has not been described before and that is crucial for the functioning of the system. Prompted by a similar comment of Reviewer #1 above, we offer to further explore the mechanistic basis of this link through an analysis of published ChIPseq data, functional studies of a GATA binding site upstream of the Fgf4 start codon, or a more detailed temporal dissection of NANOG, GATA and Fgf4 expression dynamics following doxycycline induction (see our response to Reviewer #1 above for more details). These new experiments and analyses will allow us to emphasize this novel result, and thereby significantly strengthen our paper.

Paragraph lines 274 and ff (section on the involvement of FGF4 in the robustness of the process) needs some explanations. The derivation of the conclusion that 'recursive communication vis FGF4 underlies a population-level phenotype ...characterized by the differentiation of robust proportions of cell types..." from the experiments requires some unwrapping. It would be helpful if the authors could reason how the conclusion follows from the experiments.

We realize from this Reviewer’s comment and the comments of Reviewer #2 above that we have not explained well enough how the results shown in Fig. 3 A-C (lines 274 - 283) lead to our conclusion of emergent behavior, which are then further substantiated in the modelling in panels D - G. The central conclusion of this paragraph rests on the observation that cell type proportions are dependent on initial conditions in the Fgf4 mutant, but not in wild type cells. As we had supplied FGF4 externally to the Fgf4 mutant cells, the only difference between these two conditions is that FGF4 dose in wild type cells is regulated by the cell population, i.e. cells can communicate via FGF4, whereas mutant cells cannot. We will expand on this line of reasoning, and also explain in more detail the differences in the models for the mutant case and the wild type, which we believe will help to conceptualize the experimental results. See also our response to Reviewer #2, points 6. and 11..

Their model does not seem to include the commonly agreed regulatory interaction between Nanog and FGF4, at least not directly, and it would be helpful if a reasoning could be provided for this decision.

A discussion of the regulatory interaction between NANOG and Fgf4 has also been requested by Reviewer #1. In our response to their point above, we provide a reasoning why we have omitted it in the current manuscript. Briefly, our decision not to include a direct positive link between NANOG and Fgf4 expression rests on our observation that Fgf4 mRNA continues to be expressed 2 days after switching cells from 2i + LIF medium to N2B27, a time at which NANOG already starts to be downregulated as a consequence of differentiation along embryonic lineages. We will add this data to a revised manuscript. In addition, we propose above to dissect in more detail the temporal sequence of GATA4-mCherry, Fgf4 and NANOG expression upon doxycycline induction. This analysis will provide further information about the role of NANOG for Fgf4 mRNA expression in ESCs.

Reviewer #3 (Significance (Required)):

In this manuscript, Raina and colleagues use an Embryonic Stem (ES) cell based experimental system to address a central problem in developmental biology, namely the emergence of stable scaled populations of different cell fates. The experiments are elegant in design, carefully executed and the effort provides a solution to the problem: a novel mechanism based on short range FGF signalling that provides homeostatic control of relative cell populations. This is an important piece of work with sound conclusions that establishes a new paradigm in pattern formation whose implications are likely to lead to a reassessment of the role of FGF in different patterning paradigms. The experiments are quantitative and supported by a modelling effort based on a theoretical piece of work (Stanoev et al. 2021) which underpins the conclusion.

This manuscript will appeal to a wide audience including developmental and stem cell biologists as well as modellers.

My expertise cover the areas addressed in the manuscript.

**Referees cross-commenting**

It looks as if, with some nuances, we all agree on the value of the work. I do not have any issues with the comments of Reviewer 1, though I disagree that the model tested and improved here is similar to existing ones. While it is true that this work is related to a theory paper by some of the authors, the experimental test and resulting conclusions are very important. On the other hand, I am very surprised by the comments of Reviewer 2 who, after conceding the value and potential significance of the work, raises a list of queries, largely small details and opinions rather than points of substantial concerns, hinting at a need for the authors to perform extra work and analysis that will not change the conclusions of the manuscript. Some of this e.g. #9 would be a nice piece of additional evidence, but more an adornment than a necessary piece of additional evidence. The main problem of this reviewer is the lack of appreciation of what they define as 'highly artificial nature' of the study without providing any reason for why such experiments (very common in developmental biology) can lead to misleading conclusions. It seems to me that most, if not all, of their significant concerns can be dealt with in a rebuttal or by altering the text, either to discuss the issues raised, to clarify the points or qualify the conclusions.

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

It is well established that FGF signalling plays a role in the partitioning of the Primitive Endoderm and Epiblast fates during preimplantation mammalian development. Recent work has shown that this fate decisions is associated with a mechanism that is able to maintain the proportions of the two fates stable in the face of perturbations. Here, the authors address this mechanism and show that it is dependent on FGF signalling and associated with the fate decision. In the process they suggest and test a novel mechanism based on short range FGF signalling. A series of carefully designed and executed experiments, refine and provide evidence for the model. This is an original and important piece of work that will influence the field of pattern formation.

Overall the manuscript is well written but, at least from the perspective of this reviewer, there are places in which clarity can be improved.

Lines 104 and ff: the description of the dynamics of the different populations fater the GATA4 pulse, can be clarified. The reference to the double negative population emerging from the PrEnd population is not clear. It is stated that the proportion of these cells increased continuously and it said to be at the expense of the decrease of the PrEnd population whose variation is referred to as 'slightly declined". How can a slight decline fuel a steady increase in the double negative?

Also, what are these double negative? Could they be cells differentiating into embryonic lineages?

In Figure 1 and its discussion, it would be good to see a representation of the stability of the final proportions relative to the different initial conditions, a variation on 1E.

Paragraph lines 182 and ff: the report that GATA4 expression is able to suppress FGF4 signalling, autonomously is, at least for this reviewer, a novel and important result and one that impinges on the understanding of the process. The authors should emphasize this.

Paragraph lines 274 and ff (section on the involvement of FGF4 in the robustness of the process) needs some explanations. The derivation of the conclusion that 'recursive communication vis FGF4 underlies a population-level phenotype ...characterized by the differentiation of robust proportions of cell types..." from the experiments requires some unwrapping. It would be helpful if the authors could reason how the conclusion follows from the experiments.

Their model does not seem to include the commonly agreed regulatory interaction between Nanog and FGF4, at least not directly, and it would be helpful if a reasoning could be provided for this decision.

Significance

In this manuscript, Raina and colleagues use an Embryonic Stem (ES) cell based experimental system to address a central problem in developmental biology, namely the emergence of stable scaled populations of different cell fates. The experiments are elegant in design, carefully executed and the effort provides a solution to the problem: a novel mechanism based on short range FGF signalling that provides homeostatic control of relative cell populations. This is an important piece of work with sound conclusions that establishes a new paradigm in pattern formation whose implications are likely to lead to a reassessment of the role of FGF in different patterning paradigms. The experiments are quantitative and supported by a modelling effort based on a theoretical piece of work (Stanoev et al. 2021) which underpins the conclusion.

This manuscript will appeal to a wide audience including developmental and stem cell biologists as well as modellers.

My expertise cover the areas addressed in the manuscript.

Referees cross-commenting

It looks as if, with some nuances, we all agree on the value of the work. I do not have any issues with the comments of Reviewer 1, though I disagree that the model tested and improved here is similar to existing ones. While it is true that this work is related to a theory paper by some of the authors, the experimental test and resulting conclusions are very important. On the other hand, I am very surprised by the comments of Reviewer 2 who, after conceding the value and potential significance of the work, raises a list of queries, largely small details and opinions rather than points of substantial concerns, hinting at a need for the authors to perform extra work and analysis that will not change the conclusions of the manuscript. Some of this e.g. #9 would be a nice piece of additional evidence, but more an adornment than a necessary piece of additional evidence. The main problem of this reviewer is the lack of appreciation of what they define as 'highly artificial nature' of the study without providing any reason for why such experiments (very common in developmental biology) can lead to misleading conclusions. It seems to me that most, if not all, of their significant concerns can be dealt with in a rebuttal or by altering the text, either to discuss the issues raised, to clarify the points or qualify the conclusions.