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)): ____ *A significant criticism of the paper is an assumption that readers will be familiar with all of the findings in the author's previous 2016 paper and the PGL-1 papers by Aoki et al. Minimal context is given for each approach. *

To address this concern, we have added a paragraph in the Introduction section of the revised manuscript.

*Some conclusions are not well supported and require further analysis, proper controls, and more extensive descriptions of the experiments performed. *

We have addressed the reviewer’s concerns as detailed below.

Most importantly, the central conclusion and title of the paper is that composition can buffer the dynamics of individual proteins within liquid-like condensates. In other words, in vitro condensation assays often do not recapitulate LLPS behavior in vivo. That said, the findings in this study would be significantly strengthened and complemented by observing endogenously tagged PGL-3 and PGL-3 mutants in living worms, considering the efficiency of using CRISPR in C. elegans to insert tags and make precise mutations.

The original manuscript already contained data where we microinjected wild-type PGL-3 and mutant PGL-3 proteins (recombinantly purified) into adult C. elegans gonads to assay how the P granule phase supports diffusion of these proteins.

In the revised version, we now include additional data which shows “dynamics buffering” in transgenic worms generated using CRISPR/Cas9 technology. Briefly, we used CRISPR/Cas9 to generate transgenic C. elegans which expresses PGL-3-mEGFP or PGL-3(D425-452)-mEGFP from the native pgl-3 locus. In vitro, wild-type PGL-3-mEGFP protein generates liquid-like condensates. On the other hand, the recombinantly purified PGL-3(D425-452)-mEGFP protein generates condensates that are non-dynamic. In contrast to these observations in vitro, both wild-type PGL-3-mEGFP and PGL-3(D425-452)-mEGFP show similar dynamics (half-time of FRAP recovery) within P granules in vivo.

*To improve readability, the introduction to P granules should be expanded, and include the reasons for looking at the nematode-specific PGL-3 protein among all the other known P granule proteins. A recap of previous findings on PGL-3 phase separation, in vivo and in vitro, is warranted, starting with the significant results of Saha et al 2016. Setting up the investigative questions in the context of recent work on PGL-1 (Aoki, et al) is also necessary. *

To address this concern, we have added a paragraph in the Introduction section of the revised manuscript.

The physiological concentration of PGL-3 should be more transparent, including why some experiments in this study are done at physiological concentrations while others are not. Describing why salt concentrations, crowding agents, and protein abundance are similar or different for each experiment is necessary and relevant. For example, after showing in Figure 1 that PGL-3 protein phase separates, the paragraph starting on line 161 says that it was previously shown that PGL-3 doesn't phase separate at physiological concentrations without RNA. One has to go back to Figure 1 to realize it was done differently than Figure 2 and Saha 2016.

The concentrations of PGL-3 protein and use of crowding agents (if any) have already been specified within figures or figure legends. Salt concentrations used are specified within figure legends or materials and methods section.

We have added the following paragraph to the materials and methods section of the revised manuscript.

“Saha et al. 2016 showed that at physiological concentrations (approx. 1 mM), the PGL-3 protein is unable to phase separate into condensates. At these concentrations, mRNA promotes phase separation of PGL-3. To assay for mRNA-dependence of condensate assembly, it is therefore essential to use physiological concentrations of the PGL-3 protein or mutants (e.g. Figure 2). However, these condensates are generally too small to assay rate of internal rearrangement of PGL-3 molecules within condensates using fluorescence recovery after photobleaching experiments. Therefore, to generate large condensates for measuring internal rearrangement of PGL-3 or mutant molecules, we primarily used higher concentrations of these proteins where binding to RNA is not essential for phase separation. However, to mimic the in vivo P granule phase as closely as possible, we generally added constituent proteins in proportion to their in vivo abundance estimated in Saha et al. 2016.”

The added paragraph in the Introduction section of the revised manuscript may be helpful to the readers. * *

*Statements in the same paragraph like "in contrast to full-length PGL-3, mRNA does not support phase separation..." should be qualified by stating the concentration observed, with or without salts or other crowding agents. Similarly, line 230 "suggests that interactions involving the disordered C-terminal region of PGL-3 are not essential for the fast dynamics" and should be qualified with "at non-physiological concentrations and with XX crowding agents or salt concentration." It would be more consistent if physiological concentrations were consistent from figure to figure, as extra variables weaken some of the stated conclusions. *

We thank the reviewer for this suggestion. However, we feel the statements (without full experimental details within main text) help convey the conceptual essence of the findings better. Of course, all these statements contain reference to figures or prior publications which provide relevant details about experimental conditions.

*The 2010 review reference stating that there are 40 P granule enriched proteins is outdated. More recent reviews put the number much higher. This is relevant because the approach to put PGL-3 in a more physiological environment by including just PGL-1, GLH-1 and mRNA with the condensate assays, out of ~100 P granule enriched proteins, may not be sufficient to conclude "that the influence of complex composition on dynamics is modest" (line 223), or imply that the multicomponent nature of the P granule is reconstituted by adding these components (line 355). *

We revised the text to indicate that P granules contain approx. 70 proteins and added appropriate references.

• *

Based on current information of constitutive P granule components (PGL-1, PGL-3, GLH-1, GLH-2, GLH-3, GLH-4, DEPS-1, MIP-1 and mRNA), (Kawasaki et al, 1998, 2004; Spike et al, 2008a, 2008b; Price et al, 2021; Cipriani et al, 2021; Phillips & Updike, 2022) we reconstituted P granule-like phase in vitro with mRNA, PGL- and GLH- proteins that likely constitute the most abundant components within P granules in vivo (based on concentration estimates in Saha et al. 2016).

We do appreciate the reviewer’s comment that more components can be added to our in vitro reconstitution in addition to the limited set of components used in our study. However, we feel it is interesting to observe that a limited set of components can support dynamics buffering (the main message of the paper). Further, the complementary in vivo experiments show that the P granule phase can also support dynamics buffering.

*Figure 1C needs to include PGL-3(370-693) in the analysis. Figure 1E is also incomplete without a comparison of FRAP recovery between PGL-3(1-452) and full PGL-3 as the control.

*

Fig. 1c already includes data with PGL-3 (370-693) [top row, central panel]. FRAP recovery data with full-length PGL-3 is already available in Supplementary Fig. 2c, g.

*Figure 4C is missing an essential control where PGL-3 and S1 FRAP is performed without PGL-1, GLH-1, and mRNA. *

In the revised version, we have added Supplementary Fig. 5f, where FRAP recovery of the following condensates are plotted together: 1) PGL-3 alone, 2) S1 alone, 3) PGL-3 + PGL-1, GLH-1 and mRNA, 4) S1 + PGL-1, GLH-1 and mRNA.

*It would also help show sup Fig4A in the main figure to show concentration dependence. *

We revised Fig. 4 to address the reviewer’s suggestion.

Consider adding subtitles to supplementary figures.

We considered the suggestion but felt it may not be essential.

*M&M should include an explanation for statistical analysis *

We added a paragraph describing statistical analysis within the Materials and Methods section.

*CROSS-CONSULTATION COMMENTS I am also in agreement with the comments and critiques of reviewers 2 and 3.

* Reviewer #1 (Significance (Required)): The paper by Saha and colleagues investigate the in vitro liquid-liquid phase separation propensity of a P granule protein PGL-3 and its structural domains. The findings largely replicate and support the phase-separation properties of a paralogous protein called PGL-1, as recently described by Aoki et al. 2021. Furthermore, they show that the dynamics demonstrated by recombinant PGL-3 may be maintained or buffered by the complex composition of P granules.

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

*Jelenic et al. describe the effect of partner proteins on the FRAP dynamics of recombinant PGL-3 protein and variants in in vitro condensates and C elegans p-granules. The study shows that the N terminal a-helical dimerization domains is required for condensate formation and modulate of it alters aggregation and the FRAP dynamics of its condensates. Interestingly, a construct including the entire IDR region (370-693) by itself does not phase separate on its own at these conditions. The K126E K129E mutant (known previously to disrupt dimerization) and the deletion mutant abrogate llps. A mutant construct that shuffles the sequence in the region 423-453 called S1 here reduces the helicity and the condensate FRAP dynamics but recovered in the presence of a few P granule components. Also, the reduced dynamics of partially unfolded PGL-3 condensates are also rescued by the p-granule components to a certain degree of the unfolded PGL3 concentrations. This threshold concentration for recovering the condensate dynamics is further reduced in the helix reducing S1 mutant, which is also dependent on the number and the nature of P granule components.

Overall, the study aims to probe how "composition can buffer protein dynamics within liquid-like condensates" - yet several underlying aspects of the study do not fully support that conclusion. The introduction does not sufficiently introduce the known structural information of the two dimerization domains in C elegans PGL proteins for which structures are known. The region is discussed as "alpha helical" but really there are two evolutionarily conserved independently folding dimerization domains (referring to the mutants as "reduced alpha helicity" is not helpful - these are mutations that destabilize a folded domain).*

To address this concern, we have added a paragraph in the Introduction section of the revised manuscript.

*Additionally, the abstract and introduction ignore the aspects of aggregation (touched on in discussion) - this is likely what the disruption to the helical region in residue 450 region is doing (the helix is not on the dimer interface based on homology / sequence identity to the crystal structure of PGL-1 central dimerization domain. *

We think elucidating the molecular mechanism of apparent aggregation of PGL-3 (D425-452) could be an interesting direction for future investigation. Here, we focused our analysis predominantly on the mutant S1 since it generates liquid-like condensates with ~20- fold slower dynamics (compared to wild-type) in contrast to non-dynamic condensates/aggregates. Therefore, influence of other P granule components on the dynamics of PGL-3 in liquid-like condensates is easier to address using the mutant S1 rather than PGL-3 (D425-452). We didn’t find evidence that S1 aggregates as we did not detect aggregates of S1 molecules using fluorescence confocal microscopy and the slow dynamics in condensates of S1 does not change significantly over 24 h (Supplementary Fig. 3f).

However, in the revised version, we now include additional in vivo data with C. elegans expressing the aggregation-prone PGL-3 (D425-452)-mEGFP. Briefly, we used CRISPR/Cas9 to generate transgenic C. elegans which expresses PGL-3-mEGFP or PGL-3(D425-452)-mEGFP from the native pgl-3 locus. In vitro, wild-type PGL-3-mEGFP protein generates liquid-like condensates. On the other hand, the recombinantly purified PGL-3(D425-452)-mEGFP protein generates condensates that are non-dynamic. In contrast to these observations in vitro, both wild-type PGL-3-mEGFP and PGL-3(D425-452)-mEGFP show similar dynamics (half-time of FRAP recovery) within P granules in vivo.

Finally, the "dynamics buffering" is not really clearly established and could also be explained as small concentrations of aggregated proteins act like clients while increasing the concentration results in aggregation and "cross linking" in the entire droplet - and this concentration is never achieved in the in worm experiments so it is not clear. In other words, the change in FRAP dynamics not observed in worms is perhaps not surprising if small amount of recombinant proteins are incorporated into the granules. *

*

Data with the S1 mutant establishes that dynamics buffering can be observed in condensates with different sets of additives both in vitro (Fig. 5a, b) and in vivo (Fig. 4a, b). Further, data with condensates of S1 containing the additives PGL-3 (K126E K129E) or S1 (K126E K129E) demonstrate that dynamics (half-time of FRAP recovery) within S1 condensates, and in turn “dynamics buffering” depend on inter-molecular interactions. With respect to the hypothesis proposed by the reviewer, we did not detect aggregates within S1 condensates using confocal fluorescence microscopy.

In contrast to S1 condensates, condensates containing partially unfolded PGL-3-mEGFP together with PGL-1, GLH-1 and mRNA showed spatial inhomogeneities in fluorescence signal throughout the condensate (Fig. 4g). We have not tested if areas with higher fluorescence signal represent aggregates. It is a possibility that the partially unfolded PGL-3-mEGFP fluorescence signal becomes more homogeneous if higher concentrations of additives (PGL-1, GLH-1 and mRNA) are used. However, the presented data demonstrate the significant effect of the P granule components (PGL-1, GLH-1 and mRNA) on the FRAP recovery rate of partially unfolded PGL-3-mEGFP in condensates (compare figures Fig. 3e and Fig. 4g).

However, consistent with dynamics buffering, the P granule phase in vivo supports wild-type dynamics of different PGL-3 constructs over a range of concentrations - PGL-3(D425-452)-mEGFP at physiological concentration (CRISPR transgenic strain, Fig. 4e) or at higher concentrations (microinjected S1 and partially unfolded PGL-3-mEGFP, Fig. 4b).

• *

*It is also not clear what the mechanism of the changes is - is the protein driven to fold more properly (despite S1 disruption of its conserved sequence) inside the condensate? Does it still self interact and act as a dimerization domain? Does this change disrupt interactions? *

We agree with the reviewer that identifying the precise structural changes of the S1 protein within the condensate vs. dilute phase could be an interesting direction for future investigation. However, we have already discussed the issues raised by the reviewer in the original manuscript.

“Our data is consistent with the model that other regions of S1 molecules cooperate with residues 425-452 (shuffled) to generate stronger inter-molecular interactions. For instance, addition of the mutant S1 (K126E K129E) enhances dynamics of S1 within condensates in contrast to maintaining the slower dynamics observed within condensates of S1 alone. This suggests that the interactions disrupted by the mutations K126E and K129E also contribute to slow S1 dynamics. One possibility is that interactions involving the residues K126 and K129 favor S1 conformations that enhance 425-452 (shuffled)-dependent interactions. Indeed, the mutations K126E K129E have been reported to interfere with interactions among N-termini of PGL-3 molecules (Aoki et al, 2021). While two self-association domains within the α-helical N-terminus of PGL-3 have been mapped (Aoki et al, 2021, 2016), structural insights into those associations are limited. However, PGL-3 shares significant sequence similarity with another protein PGL-1. Crystal structures are available for fragments of the PGL-1 protein that show the two self-association domains at the N-terminus are predominantly α-helical and globular in nature (Aoki et al, 2016, 2021). Therefore, one possibility is that shuffling the sequence 425-452 of PGL-3 or heat-induced unfolding of PGL-3 exposes hydrophobic residues that become available to participate in inter-molecular interactions.”

What is the real mechanism by which PGL-3 phase separates if not via the disordered domains? *

*

We agree with the reviewer that elucidating the detailed mechanism of phase separation of PGL-3 is an interesting direction for future investigation. However, we feel this is not required to support the main message of this manuscript.

Throughout the manuscript, the term "dynamics" is used to indicate FRAP, but it would be better to define what is meant (diffusion of PGL-3 in condensates) instead of using dynamics a term that could mean many things. Secondly, FRAP cannot directly measure liquidity etc (see recent critiques by McSwiggen elife 2019, etc) so it is better to be cautious in the claims. Finally, discussing "dyanmics buffering" adds more terminology where it is not needed - perhaps say "changes to diffusion of PGL-3 in condensates".

We feel it is useful to introduce a term that describes our observation. To our knowledge, our observation is novel and therefore requires a new term to describe it.

However, we do appreciate the concern raised by the reviewer. We used a more generic term “dynamics buffering” in contrast to the more specific “diffusion buffering” since we did not directly estimate diffusion behavior at the ‘single-molecule’ level. However, we already described what we mean by “dynamics buffering” in the text as follows.

“We used condensates of similar size for our analysis (average ± 1 SD of diameter of condensates are 6.4 ± 1.7 mm (Fig. 5a) and 5.9 ± 0.4 mm (Fig. 5b)). Therefore, dynamics buffering here is likely to represent similar diffusion rates of S1 within condensates.”

• *

*The "N-terminus" is not 65% of the protein. One could define this as the N-terminal domain, but again there are two clear folded domains in the first 65% of the protein and this needs to be described better. *

We revised the text to replace the terms “N-terminus” and “N-terminal domain” to “N-terminal fragment”.

*The description of "stickers" and the references to tau and hnRNPA1 are confusing as this is a predominantly ordered domain while those are IDRs. *

• *

We feel this is important as it aids discussing our work in the context of current literature describing the mechanisms of macromolecular phase separation.

The suggestion in the discussion that "P granule components support dynamics by participating in intermolecular interactions wth PGL-3-mEGFP molecules" is not well supported because no interaction assays are performed and no mutaitons are made that disrupt these interactions to test this.

Indeed, we have not conducted interaction assays or mutational analysis to directly test this. However, our detailed analysis with the S1 mutant supports this suggestion.

While partially unfolded PGL-3-mEGFP molecules lose 30% of a-helicity, the a-helicity of the S1 mutant is reduced by 15% compared to wild-type PGL-3. Data with S1 and partially unfolded PGL-3-mEGFP molecules show that loss of a-helicity correlates with slower diffusion of protein molecules within condensates. Using the mutants PGL-3 (K126E K129E) and S1 (K126E K129E), we show that diffusion rate of S1 molecules within condensates depend on inter-molecular interactions, and presence of other P granule components support faster diffusion rate of S1 molecules within condensates. Therefore, we feel it is safe to speculate that intermolecular interactions with P granule components can support dynamics of a “more unfolded” (compared to S1) version of PGL-3 molecule. * *

*More detailed analysis of some of the claims: Claim 1: An a-helical region mediates the phase separation of PGL-3, and the C-terminal disordered region by itself does not phase separate. The N-terminal dimerization is essential for LLPS. The C-terminal IDR interactions with mRNA facilitate the LLPS. Comments: The authors show sufficient experimental data using microscopy and FRAP on truncated constructs with the N-terminal and C-terminal regions - but see above regarding how these are described - a proper domain structure with the folded domains shown and the RGG motifs highlighted should be added and integrated throughout the discussion. *

In the revised version of the manuscript, we described the predicted PGL-3 domains within a paragraph in the introduction: “The interactions that support phase separation of the PGL-3 protein remains unclear. Structural studies on the orthologous PGL-1 protein revealed two dimerization domains. This raises the possibility that PGL-3 also contains similar dimerization domains, and phase separation depends on interactions involving these domains.”

Our Fig. 1a already includes the schematic representation of PGL-3 with predicted N-terminal and Central Dimerization domains and RGG repeats.

*They show that the N-terminus is necessary and adequate for LLPS, and the C-terminus by itself does not phase separate. But, how does the N-terminal domains phase separate? This is not explained - what are the interactions? *

• *

Also, a di-mutant (K126E K129E) that is known, and also authors use SEC-MALS to show their N-terminal construct is consistent with the published results. Disrupting the n-terminal dimerization prevents phase separation, suggesting the importance of these residues in the N-terminus for self-assembly and LLPS. The Microscopy data backs the claim that the mRNA-mediated LLPS is facilitated by binding with C-terminus. However, the m-RNA binding to IDR is not sufficient for LLPS. Yet, the authors do not explain how higher salt prevents phase separation - again the mechanism of phase separation is unclear. Is it multivalent interaction of the two dimerization domains? A basic model (that is tested) would be important.

We agree with the reviewer that elucidating the detailed mechanism of phase separation of PGL-3 is an interesting direction for future investigation. However, we feel this is not required to support the main message of this manuscript.

However, our manuscript already provides some relevant insights as follows.

“To investigate the underlying mechanism further, we began by testing if the N-terminal α-helical region of PGL-3 can self-associate. Our analysis using size exclusion chromatography followed by multi-angle light scattering (SEC-MALS) showed that this PGL-3 fragment 1-452 forms a dimer (Supplementary Fig. 2f). Mutation of two residues (K126E K129E) have been shown to interfere with interactions among the N-termini of PGL-3 molecules (Aoki et al, 2021). We mutated these two residues within the full-length PGL-3 protein (K126E K129E) (Fig. 1a) and found that this mutant PGL-3 (K126E K129E) protein cannot phase separate even at high protein concentrations up to ~130 µM (Fig. 1b, c). Addition of mRNA does not trigger phase separation of this protein at physiological concentrations either (Fig. 2a, b). Taken together, our data is consistent with a model where association among folded N-termini of PGL-3 molecules is essential for phase separation.”

A likely possibility is that phase separation of PGL-3 depends on electrostatic inter-molecular interactions among the folded N-terminal fragment of PGL-3 molecules. Therefore, high salt prevents phase separation.

Are the tags removed to ensure that phase separation is not caused by tags or remaining linker regions? Is the protein purified to be without nucleic acid contamination or other purity metrics?

Most of the experiments were done with only 5% of total protein tagged with 6x-His-mEGFP. No additional tags were present on the constructs. For recombinant expression and purification, proteins were cloned such that it is possible to remove the 6xHis-mEGFP tag following treatment with TEV protease. Following removal of the 6xHis-mEGFP tag, the residual linker is just two amino acid residues long. We used 100% tagged-protein for our experiments only in very few cases (indicated in the figure legends).

To demonstrate purity of recombinant proteins, SDS-PAGE gels with all protein constructs used in this study are shown in Supplementary Fig. 1.

To minimize contamination of nucleic acids, we treated samples with Benzonase during the course of purification.

To assess the extent of nucleic acid contamination, the ratio of absorbance at 260 nm and 280 nm (A260/A280) was monitored. In exceptional cases with high A260/A280 values, we analyzed samples further by purifying RNA from the sample using RNA purification kit (Qiagen) and found that RNA represented 1% or less of the sample mass.* *

Claim2: The N-terminal a-helical region modulates the dynamics within condensates. The IDR region has minimal effect on the fast dynamics of PGL-3. Comments: The authors show that the full-length PGL-3 condensates have modest influence of components by comparing the FRAP half times with or without the P granule components, including mRNA. However, have the authors tried this in the presence of mRNAs for the constructs lacking the IDRs as they have several RGG domains and bind with mRNA and are likely to change the dynamics.

We thank the reviewer for this suggestion. However, this experiment is not essential to support the claim made in the context of homotypic condensates of PGL-3 : “The N-terminal a-helical region modulates the dynamics within condensates. The IDR region has minimal effect on the fast dynamics of PGL-3.”

*The authors report the importance of the N-terminal a-helical region by making a construct that lacks/disrupts a part of the helices lowers the thermal stability and significantly lowers the dynamics of the condensates. Also unfolding of helices is shown to reduce the dynamics. One primary concern is whether these "rescued" protein dynamics imply protein functionality. *

An assay of “functionality” e.g. an enzymatic activity of the PGL-3 protein is not available.

However, we compared the fecundity of C. elegans worms expressing from the native pgl-3 locus, PGL-3-mEGFP or the mutant protein PGL-3(D425-452)-mEGFP, to assay the functionality of P granules in these strains. We found that worms of both genotypes produced similar number of offspring (Fig. 4d). This suggests that deletion of residues 425-452 of PGL-3 does not result in significant loss of function of P granules.

Are these semi denatured proteins refolded in the presence of P-granule components?

We feel that identifying the precise structural changes of the semi-denatured PGL-3 proteins within the condensate vs. dilute phase could be an interesting direction for future investigation.

Finally, it is not clear why the authors chose to disrupt folding of the central dimerization domain?

The manuscript included a paragraph to describe the rationale.

“This suggests that interactions involving the disordered C-terminal region of PGL-3 are not essential for the fast dynamics within condensates. Therefore, we addressed the role of the N-terminal α-helical region (1-452) in driving dynamics. In order to avoid engineering mutations that result in significant misfolding of PGL-3 and concomitant loss of its ability to phase separate, we focused our mutational analysis close to the junction of the folded N-terminus and the disordered C-terminus of PGL-3. Surprisingly, we found that a full-length PGL-3 construct (D425-452) that lacks only 27 residues phase separates into condensates that are non-dynamic (Fig. 3a, c). Sequence analysis of the PGL-3 protein predicts that this region 425-452 spans two α-helices (one complete helix and fraction of a second helix) (Supplementary Fig. 3d). We generated a PGL-3 construct (hereafter called ‘S1’) (Fig. 3a) in which the sequence in the region, 425-452, is shuffled while keeping the overall amino acid composition unchanged. We found that S1 phase separates into condensates that are 20- fold less dynamic than with wild-type PGL-3 (Fig. 3d, Supplementary Fig. 3c).”

Saying that "reduced alpha-helicity of PGL-3 correlates with slower dynamics in condensates" may be factual in these assays but "correlation" should be expanded upon to include mechanism and to me it seems that the statement should read "aggregation of PGL-3 causes slower dynamics in condensates" (both the partially destabilized mutant and the fully unfolded WT show similar effects perhaps to different degrees).

We feel that identifying the precise structural changes of the semi-denatured PGL-3 proteins within the condensate vs. dilute phase could be an interesting direction for future investigation.

We did not use the term "aggregation" since we did not detect aggregates of S1 molecules using fluorescence confocal microscopy.

*CROSS-CONSULTATION COMMENTS I agree with the other reviewer's comments and critiques, I have concerns about the biological relevance and also the biophysical mechanisms. Reflecting on the other reviewers' comments, the papers could provide more depth in one or both of these areas to come to firm conclusions that are either revealing about PGL biology or elucidate a (possible) general biophysical mechanism. *

In the revised version, we now include additional data which shows “dynamics buffering” in transgenic worms generated using CRISPR/Cas9 technology. Briefly, we used CRISPR/Cas9 to generate transgenic C. elegans which expresses PGL-3-mEGFP or PGL-3(D425-452)-mEGFP from the native pgl-3 locus. In vitro, wild-type PGL-3-mEGFP protein generates liquid-like condensates. On the other hand, the recombinantly purified PGL-3(D425-452)-mEGFP protein generates condensates that are non-dynamic. In contrast to these observations in vitro, both wild-type PGL-3-mEGFP and PGL-3(D425-452)-mEGFP show similar dynamics (half-time of FRAP recovery) within P granules in vivo.

Reviewer #2 (Significance (Required)): *Hence, although the authors shows how inclusion of other components can alter the one protein component phase separation, this is done with entirely artificial means of destabilizing the fold of one of the domains which likely leads to aggregation. So the true impact of the work is hard to understand because the mutations impact on the basic biophysical properties of the domain (stability, interaction) are not completely characterized and the reason for disrupting this folding is not clear. *

A major impact of our work is elucidation of a novel “dynamics buffering” property within biomolecular condensates in vitro. Our in vivo data is consistent with this finding.

• *

We have chosen two orthogonal ways of perturbing the PGL-3 protein (i.e. mutations and temperature-dependent unfolding) to assay the effect on diffusion rate against different levels of perturbation (e.g. 30% loss of a-helicity in heat-denatured PGL-3-mEGFP vs. 15% loss of a-helicity in the S1 mutant, compared to wild-type PGL-3). Studying the phase separation behavior of these “artificially-generated” constructs provided the understanding that dynamics of PGL-3 in condensates depends on inter-molecular interactions, and slower dynamics generally correlate with stronger inter-molecular interactions. Further, interactions among two or more P granule components can buffer against large change in dynamics / aggregation within the P granule phase. These insights may lay the groundwork for addressing how more “natural” modifications (e.g., post-translational modifications, high local concentration of “sticky” molecules) may influence dynamics within biomolecular condensates in vivo.

Based on current knowledge of P granule composition, chaperone proteins (e.g. heat-shock family proteins) do not show abundant concentration within P granules. However, it is unclear if chaperone proteins are completely excluded from the P granule phase. Therefore, we speculate that weak interactions among two or more non-chaperone proteins contribute significantly to “dynamics buffering” within the P granule phase in vivo.

In the discussion section of the manuscript, we had speculated that “dynamics buffering” may potentially explain observations reported in the nucleolus: “Similarly, interactions among components could be a potential mechanism of storage of misfolding-prone proteins in non-aggregated state within the liquid-like nucleolus under stress in vivo (Frottin et al, 2019).”

Our finding is also relevant in the context of synthetic biology with applications that require steady diffusion rate of macromolecules during biochemical reactions within biomolecular condensates.

• *

My field of expertise is protein phase separation and protein structure. * *

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

Summary: P granules are liquid condensates found in the developing germlines and embryos of C. elegans. Prior work by the authors and others have established P granules as a tractable model to investigate the basic biophysical properties of liquid condensates. Much of the prior published work focused on specific P granule scaffold proteins, PGL-1 and PGL-3. How attributes of these PGL proteins and the effect of other P granule components affect condensate properties is not fully understood. Here, Jelenic, et al. probe the biophysical properties of PGL-3. Using recombinant protein, they show that an N-terminal, alpha-helical region of PGL-3 is sufficient for liquid condensate formation and that N-terminal assembly is required for this formation. Creation of a scrambled alpha-helical region in PGL-3 and heat treatment affects PGL-3 fluidity. This fluidity can be "rescued" in vivo and in vitro with the inclusion of other P granule factors, including wildtype PGL-3, PGL-1, GLH-1 and mRNA. The authors note an inverse correlation between fluidity and mutant PGL-3 fluorescent intensity. They propose a model that heterotypic compositions of condensates can buffer their fluidity against components with stronger multivalent interactions. *

MAJOR: 1. PGL-3 is a fantastic model to study the biophysical properties of a liquid condensate. But as the authors address in their discussion, the S1 mutant will likely affect the central domain folding, at its minimum causing exposure of a hydrophobic surface not typically exposed in biology. These helices are found at the terminal portion of the domain determined in the crystal structure and as depicted in the authors' Figure 1A. While the cause of S1's enhanced molecular interactions does not affect the in vitro work presented in this manuscript, it does affect how the conclusions connect to the biological nature of P granules and liquid condensates more generally. *

We have chosen two orthogonal ways of perturbing the PGL-3 protein (i.e. mutations and temperature-dependent unfolding) to assay the effect on diffusion rate against different levels of perturbation (e.g. 30% loss of a-helicity in heat-denatured PGL-3-mEGFP vs. 15% loss of a-helicity in the S1 mutant, compared to wild-type PGL-3). Studying the phase separation behavior of these “artificial” constructs provided the understanding that dynamics of PGL-3 in condensates depends on inter-molecular interactions, and slower dynamics generally correlate with stronger inter-molecular interactions. Further, interactions among two or more P granule components can buffer against large change in dynamics / aggregation within the P granule phase. These insights may lay the groundwork for addressing how more “natural” modifications (e.g., post-translational modifications, high local concentration of “sticky” molecules) may influence dynamics within biomolecular condensates in vivo.

Based on current knowledge of P granule composition, chaperone proteins (e.g. heat-shock family proteins) do not show abundant concentration within P granules. However, it is unclear if chaperone proteins are completely excluded from the P granule phase. Therefore, we speculate that weak interactions among two or more non-chaperone proteins contribute significantly to “dynamics buffering” within the P granule phase in vivo.

In the discussion section of the manuscript, we had speculated that “dynamics buffering” may potentially explain observations reported in the nucleolus: “Similarly, interactions among components could be a potential mechanism of storage of misfolding-prone proteins in non-aggregated state within the liquid-like nucleolus under stress in vivo (Frottin et al, 2019).”

Our finding is also relevant in the context of synthetic biology with applications that require steady diffusion rate of macromolecules during biochemical reactions within biomolecular condensates.

• Recombinant PGL-3 experiments added PGL-1, GLH-1 and mRNA simultaneously and measured fluidity. It will be interesting to know which components contribute to fluidity and whether fluidity enhancement of each component is dependent on one another. Addition experiments with each component should be included and/or at least discussed in the main text. *

Our data with S1-mEGFP or PGL-3-mEGFP (pre-heated at 50°C) proteins microinjected into C. elegans gonads, and the transgenic strain expressing PGL-3(D425-452)-mEGFP from the pgl-3 locus showed that the P granule phase can support fast dynamics of these mutant PGL-3 constructs. Since P granules have a complex composition, one possibility is that fast dynamics of these constructs is supported by interactions involving many P granule components. We found that using only a limited set of P granule components (PGL-1, GLH-1 and mRNA) can buffer dynamics of S1 in condensates in vitro.

In absence of a systematic analysis investigating the individual role of approx. 70 P granule proteins in buffering S1 dynamics in condensates in vitro, we have claimed in the text that dynamics-buffering of S1 in condensates is supported by interactions among two or more components. However, we do appreciate the reviewer’s comment and feel it would be interesting to investigate the contribution of individual P granule components towards fluidity in future studies. We have discussed this in the ‘Discussion’ section of the manuscript.

• The biological relevance of PGL-1, GLH-1, and mRNA were not discussed in the main text. How these factors contribute to P granule assembly and function should be mentioned in the Introduction or Results. *

To address this concern, we have added a paragraph in the Introduction section of the revised manuscript.

*MINOR: 1. Line 20, "most non-membrane-bound compartments...have complex composition": Are there examples of condensates that do not have complex composition? *

Not all non-membrane-bound compartments may have been characterized. To accommodate this possibility, we refrained from making a more general statement, but stated “most non-membrane-bound compartments…”.

• Lines 40-43, RNA interactions driving LLPS: Please include citations from the Parker Lab (e.g. Van Treeck and Parker, Cell. 2018 doi: 10.1016/j.cell.2018.07.023) *

We added the reference suggested by the reviewer.

• *

• Line 60, condensates contain hundreds of different proteins and RNA: Please cite at least a few examples of condensates with their components identified. *

We added some references following suggestion by the reviewer.

• Lines 82-84, PGL-3 drives assembly: Please cite Kawasaki, et al. Genetics 2004 for the discovery of PGL-3. *

We added the reference suggested by the reviewer.

• Lines 88-89, PGL-3 N-terminal fragment predominantly alpha-helical: The PGL domain structures should be cited here as supporting evidence that these regions are composed primarily of alpha helices (Aoki, et al 2016, 2021) *

• *

To address this concern, we have added a paragraph in the Introduction section of the revised manuscript.

• Lines 158-159, driving forces for phase separation: This statement should be removed or expanded. The authors point regarding the protein concentrations is not clear here but clarified in the Discussion (Lines 691-693). Recommend removing due to its speculative nature. *

We retained the speculative comment in the results section. We feel that this prepares the readers for the discussion later in the manuscript.

• Lines 210: Add commas before and after "PGL-1 and GLH-1"*

We addressed the reviewer’s suggestion.

• Lines 218-219: add "and" instead of comma between PGL-1 and GLH-1 *

We addressed the reviewer’s suggestion.

• Lines 238-239, alpha-helices: The PGL CDD structure should also be referenced here (Aoki, et al 2016). *

To address this concern, we have added a paragraph in the Introduction section of the revised manuscript.

• Lines 680-682, MEG proteins: Please cite accordingly. *

We added the reference suggested by the reviewer.

• Lines 694-695, heterotypic interactions: Please cite Saha, et al. 2016. *

We added the reference suggested by the reviewer.

• Figure 1: Add space between 1 and mM DTT *

We addressed the reviewer’s suggestion.

• Figure 2b: Please provide statistics between condensate numbers. *

We provide statistics between condensate numbers in Fig. 2b.

• Figure 4A: The region of the germline imaged and analyzed should be mentioned in the caption or the main text. *

We revised the Figure legend of Fig. 4a to address this issue.

• Figure 4B,C: Please include statistics between the FRAP curves. *

We have included statistics comparing FRAP curves in Supplementary Fig. 4a-c.

• Figure 4D: It will be helpful to compare this curve to Figure S4A in the same graph. Please also include graph statistics. *

We have revised Fig. 4 to address the reviewer’s suggestion.

• Figure 5: The data points are difficult to resolve. Recommend use of color.*

We considered the suggestion, but felt it works better in the original form.

• Figure 6: This is a very general model that does not highlight the extensive experimental work performed by the authors. Recommend incorporating PGL-3, mutants and P granule factors into this model. *

We thank the reviewer for appreciating our extensive work. However, we retained the original Fig. 6 for the sake of simplicity.

• Methods, Line 939, C. elegans section: What worms were used? TH623? Please describe the genotype. *

We have included a table listing the strains used in the study and their genotype. * CROSS-CONSULTATION COMMENTS While my review was arguably the more favorable of the three, I agree with the other reviewers' comments and evaluation, particularly with Reviewer #1. As written in my review, my primary concern was the biological relevance of the work.*

Reviewer #3 (Significance (Required)):

Overall, the in vitro work presented investigating the biophysical properties of this minimal P granule system was thorough and well-analyzed, and the manuscript was clearly written. Additional citations and statistics will improve the manuscript and the strength of the conclusions, respectively. The biological relevance of this study to P granule form and function in vivo, and to condensates in vivo, is debatable. This work will interest those who study condensate biology, the biophysics of protein-protein and protein-RNA interactions, and RNA biochemists more generally.

A major impact of our work is elucidation of a novel “dynamics buffering” property within biomolecular condensates in vitro. Our in vivo data is consistent with this finding.

We have chosen two orthogonal ways of perturbing the PGL-3 protein (i.e. mutations and temperature-dependent unfolding) to assay the effect on diffusion rate against different levels of perturbation (e.g. 30% loss of a-helicity in heat-denatured PGL-3-mEGFP vs. 15% loss of a-helicity in the S1 mutant, compared to wild-type PGL-3). Studying the phase separation behavior of these “artificially-generated” constructs provided the understanding that dynamics of PGL-3 in condensates depends on inter-molecular interactions, and slower dynamics generally correlate with stronger inter-molecular interactions. Further, interactions among two or more P granule components can buffer against large change in dynamics / aggregation within the P granule phase. These insights may lay the groundwork for addressing how more “natural” modifications (e.g., post-translational modifications, high local concentration of “sticky” molecules) may influence dynamics within biomolecular condensates in vivo.

• *

Based on current knowledge of P granule composition, chaperone proteins (e.g. heat-shock family proteins) do not show abundant concentration within P granules. However, it is unclear if chaperone proteins are completely excluded from the P granule phase. Therefore, we speculate that weak interactions among two or more non-chaperone proteins contribute significantly to “dynamics buffering” within the P granule phase in vivo.

In the discussion section of the manuscript, we had speculated that “dynamics buffering” may potentially explain observations reported in the nucleolus: “Similarly, interactions among components could be a potential mechanism of storage of misfolding-prone proteins in non-aggregated state within the liquid-like nucleolus under stress in vivo (Frottin et al, 2019).”

Our finding is also relevant in the context of synthetic biology with applications that require steady diffusion rate of macromolecules during biochemical reactions within biomolecular condensates.

*I have expertise in P granules, protein/RNA biochemistry, condensate assembly, and C. elegans. *

References

Aoki ST, Kershner AM, Bingman CA, Wickens M & Kimble J (2016) PGL germ granule assembly protein is a base-specific, single-stranded RNase. Proceedings of the National Academy of Sciences of the United States of America

Aoki ST, Lynch TR, Crittenden SL, Bingman CA, Wickens M & Kimble J (2021) C. elegans germ granules require both assembly and localized regulators for mRNA repression. Nat Commun 12: 996

Cipriani PG, Bay O, Zinno J, Gutwein M, Gan HH, Mayya VK, Chung G, Chen J-X, Fahs H, Guan Y, et al (2021) Novel LOTUS-domain proteins are organizational hubs that recruit C. elegans Vasa to germ granules. Elife 10: e60833

Frottin F, Schueder F, Tiwary S, Gupta R, Körner R, Schlichthaerle T, Cox J, Jungmann R, Hartl FU & Hipp MS (2019) The nucleolus functions as a phase-separated protein quality control compartment. Science 365: 342–347

Kawasaki I, Amiri A, Fan Y, Meyer N, Dunkelbarger S, Motohashi T, Karashima T, Bossinger O & Strome S (2004) The PGL family proteins associate with germ granules and function redundantly in Caenorhabditis elegans germline development. Genetics 167: 645–661

Kawasaki I, Shim YH, Kirchner J, Kaminker J, Wood WB & Strome S (1998) PGL-1, a predicted RNA-binding component of germ granules, is essential for fertility in C. elegans. Cell 94: 635–645

Phillips CM & Updike DL (2022) Germ granules and gene regulation in the Caenorhabditis elegans germline. Genetics 220: iyab195

Price IF, Hertz HL, Pastore B, Wagner J & Tang W (2021) Proximity labeling identifies LOTUS domain proteins that promote the formation of perinuclear germ granules in C. elegans. Elife 10: e72276

Saha S, Weber CA, Nousch M, Adame-Arana O, Hoege C, Hein MY, Osborne Nishimura E, Mahamid J, Jahnel M, Jawerth L, et al (2016) Polar Positioning of Phase-Separated Liquid Compartments in Cells Regulated by an mRNA Competition Mechanism. Cell 166: 1572-1584.e16

Spike C, Meyer N, Racen E, Orsborn A, Kirchner J, Kuznicki K, Yee C, Bennett K & Strome S (2008a) Genetic analysis of the Caenorhabditis elegans GLH family of P-granule proteins. Genetics 178: 1973–1987

Spike CA, Bader J, Reinke V & Strome S (2008b) DEPS-1 promotes P-granule assembly and RNA interference in C. elegans germ cells. Development (Cambridge, England) 135: 983–993

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:

P granules are liquid condensates found in the developing germlines and embryos of C. elegans. Prior work by the authors and others have established P granules as a tractable model to investigate the basic biophysical properties of liquid condensates. Much of the prior published work focused on specific P granule scaffold proteins, PGL-1 and PGL-3. How attributes of these PGL proteins and the effect of other P granule components affect condensate properties is not fully understood. Here, Jelenic, et al. probe the biophysical properties of PGL-3. Using recombinant protein, they show that an N-terminal, alpha-helical region of PGL-3 is sufficient for liquid condensate formation and that N-terminal assembly is required for this formation. Creation of a scrambled alpha-helical region in PGL-3 and heat treatment affects PGL-3 fluidity. This fluidity can be "rescued" in vivo and in vitro with the inclusion of other P granule factors, including wildtype PGL-3, PGL-1, GLH-1 and mRNA. The authors note an inverse correlation between fluidity and mutant PGL-3 fluorescent intensity. They propose a model that heterotypic compositions of condensates can buffer their fluidity against components with stronger multivalent interactions.

MAJOR:

1. PGL-3 is a fantastic model to study the biophysical properties of a liquid condensate. But as the authors address in their discussion, the S1 mutant will likely affect the central domain folding, at its minimum causing exposure of a hydrophobic surface not typically exposed in biology. These helices are found at the terminal portion of the domain determined in the crystal structure and as depicted in the authors' Figure 1A. While the cause of S1's enhanced molecular interactions does not affect the in vitro work presented in this manuscript, it does affect how the conclusions connect to the biological nature of P granules and liquid condensates more generally.
2. Recombinant PGL-3 experiments added PGL-1, GLH-1 and mRNA simultaneously and measured fluidity. It will be interesting to know which components contribute to fluidity and whether fluidity enhancement of each component is dependent on one another. Addition experiments with each component should be included and/or at least discussed in the main text.
3. The biological relevance of PGL-1, GLH-1, and mRNA were not discussed in the main text. How these factors contribute to P granule assembly and function should be mentioned in the Introduction or Results.

MINOR:

1. Line 20, "most non-membrane-bound compartments...have complex composition": Are there examples of condensates that do not have complex composition?
2. Lines 40-43, RNA interactions driving LLPS: Please include citations from the Parker Lab (e.g. Van Treeck and Parker, Cell. 2018 doi: 10.1016/j.cell.2018.07.023)
3. Line 60, condensates contain hundreds of different proteins and RNA: Please cite at least a few examples of condensates with their components identified.
4. Lines 82-84, PGL-3 drives assembly: Please cite Kawasaki, et al. Genetics 2004 for the discovery of PGL-3.
5. Lines 88-89, PGL-3 N-terminal fragment predominantly alpha-helical: The PGL domain structures should be cited here as supporting evidence that these regions are composed primarily of alpha helices (Aoki, et al 2016, 2021)
6. Lines 158-159, driving forces for phase separation: This statement should be removed or expanded. The authors point regarding the protein concentrations is not clear here but clarified in the Discussion (Lines 691-693). Recommend removing due to its speculative nature.
7. Lines 210: Add commas before and after "PGL-1 and GLH-1"
8. Lines 218-219: add "and" instead of comma between PGL-1 and GLH-1
9. Lines 238-239, alpha-helices: The PGL CDD structure should also be referenced here (Aoki, et al 2016).
10. Lines 680-682, MEG proteins: Please cite accordingly.
11. Lines 694-695, heterotypic interactions: Please cite Saha, et al. 2016.
12. Figure 1: Add space between 1 and mM DTT
13. Figure 2b: Please provide statistics between condensate numbers.
14. Figure 4A: The region of the germline imaged and analyzed should be mentioned in the caption or the main text.
15. Figure 4B,C: Please include statistics between the FRAP curves.
16. Figure 4D: It will be helpful to compare this curve to Figure S4A in the same graph. Please also include graph statistics.
17. Figure 5: The data points are difficult to resolve. Recommend use of color.
18. Figure 6: This is a very general model that does not highlight the extensive experimental work performed by the authors. Recommend incorporating PGL-3, mutants and P granule factors into this model.
19. Methods, Line 939, C. elegans section: What worms were used? TH623? Please describe the genotype.

CROSS-CONSULTATION COMMENTS

While my review was arguably the more favorable of the three, I agree with the other reviewers' comments and evaluation, particularly with Reviewer #1. As written in my review, my primary concern was the biological relevance of the work.

Significance

Overall, the in vitro work presented investigating the biophysical properties of this minimal P granule system was thorough and well-analyzed, and the manuscript was clearly written. Additional citations and statistics will improve the manuscript and the strength of the conclusions, respectively. The biological relevance of this study to P granule form and function in vivo, and to condensates in vivo, is debatable. This work will interest those who study condensate biology, the biophysics of protein-protein and protein-RNA interactions, and RNA biochemists more generally.

I have expertise in P granules, protein/RNA biochemistry, condensate assembly, and C. elegans.

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

Jelenic et al. describe the effect of partner proteins on the FRAP dynamics of recombinant PGL-3 protein and variants in in vitro condensates and C elegans p-granules. The study shows that the N terminal a-helical dimerization domains is required for condensate formation and modulate of it alters aggregation and the FRAP dynamics of its condensates. Interestingly, a construct including the entire IDR region (370-693) by itself does not phase separate on its own at these conditions. The K126E K129E mutant (known previously to disrupt dimerization) and the deletion mutant abrogate llps. A mutant construct that shuffles the sequence in the region 423-453 called S1 here reduces the helicity and the condensate FRAP dynamics but recovered in the presence of a few P granule components. Also, the reduced dynamics of partially unfolded PGL-3 condensates are also rescued by the p-granule components to a certain degree of the unfolded PGL3 concentrations. This threshold concentration for recovering the condensate dynamics is further reduced in the helix reducing S1 mutant, which is also dependent on the number and the nature of P granule components.

• Overall, the study aims to probe how "composition can buffer protein dynamics within liquid-like condensates" - yet several underlying aspects of the study do not fully support that conclusion. The introduction does not sufficiently introduce the known structural information of the two dimerization domains in C elegans PGL proteins for which structures are known. The region is discussed as "alpha helical" but really there are two evolutionarily conserved independently folding dimerization domains (referring to the mutants as "reduced alpha helicity" is not helpful - these are mutations that destabilize a folded domain). Additionally, the abstract and introduction ignore the aspects of aggregation (touched on in discussion) - this is likely what the disruption to the helical region in residue 450 region is doing (the helix is not on the dimer interface based on homology / sequence identity to the crystal structure of PGL-1 central dimerization domain. Finally, the "dynamics buffering" is not really clearly established and could also be explained as small concentrations of aggregated proteins act like clients while increasing the concentration results in aggregation and "cross linking" in the entire droplet - and this concentration is never achieved in the in worm experiments so it is not clear. In other words, the change in FRAP dynamics not observed in worms is perhaps not surprising if small amount of recombinant proteins are incorporated into the granules. It is also not clear what the mechanism of the changes is - is the protein driven to fold more properly (despite S1 disruption of its conserved sequence) inside the condensate? Does it still self interact and act as a dimerization domain? Does this change disrupt interactions? What is the real mechanism by which PGL-3 phase separates if not via the disordered domains?

• Throughout the manuscript, the term "dynamics" is used to indicate FRAP, but it would be better to define what is meant (diffusion of PGL-3 in condensates) instead of using dynamics a term that could mean many things. Secondly, FRAP cannot directly measure liquidity etc (see recent critiques by McSwiggen elife 2019, etc) so it is better to be cautious in the claims. Finally, discussing "dyanmics buffering" adds more terminology where it is not needed - perhaps say "changes to diffusion of PGL-3 in condensates".

• The "N-terminus" is not 65% of the protein. One could define this as the N-terminal domain, but again there are two clear folded domains in the first 65% of the protein and this needs to be described better.

• The description of "stickers" and the references to tau and hnRNPA1 are confusing as this is a predominantly ordered domain while those are IDRs.

• The suggestion in the discussion that "P granule components support dynamics by participating in intermolecular interactions wth PGL-3-mEGFP molecules" is not well supported because no interaction assays are performed and no mutaitons are made that disrupt these interactions to test this.

More detailed analysis of some of the claims:

Claim 1:

An a-helical region mediates the phase separation of PGL-3, and the C-terminal disordered region by itself does not phase separate. The N-terminal dimerization is essential for LLPS. The C-terminal IDR interactions with mRNA facilitate the LLPS.

Comments:

The authors show sufficient experimental data using microscopy and FRAP on truncated constructs with the N-terminal and C-terminal regions - but see above regarding how these are described - a proper domain structure with the folded domains shown and the RGG motifs highlighted should be added and integrated throughout the discussion. They show that the N-terminus is necessary and adequate for LLPS, and the C-terminus by itself does not phase separate. But, how does the N-terminal domains phase separate? This is not explained - what are the interactions? Are the tags removed to ensure that phase separation is not caused by tags or remaining linker regions? Is the protein purified to be without nucleic acid contamination or other purity metrics? Also, a di-mutant (K126E K129E) that is known, and also authors use SEC-MALS to show their N-terminal construct is consistent with the published results. Disrupting the n-terminal dimerization prevents phase separation, suggesting the importance of these residues in the N-terminus for self-assembly and LLPS. The Microscopy data backs the claim that the mRNA-mediated LLPS is facilitated by binding with C-terminus. However, the m-RNA binding to IDR is not sufficient for LLPS. Yet, the authors do not explain how higher salt prevents phase separation - again the mechanism of phase separation is unclear. Is it multivalent interaction of the two dimerization domains? A basic model (that is tested) would be important.

Claim 2:

The N-terminal a-helical region modulates the dynamics within condensates. The IDR region has minimal effect on the fast dynamics of PGL-3.

Comments:

The authors show that the full-length PGL-3 condensates have modest influence of components by comparing the FRAP half times with or without the P granule components, including mRNA. However, have the authors tried this in the presence of mRNAs for the constructs lacking the IDRs as they have several RGG domains and bind with mRNA and are likely to change the dynamics. The authors report the importance of the N-terminal a-helical region by making a construct that lacks/disrupts a part of the helices lowers the thermal stability and significantly lowers the dynamics of the condensates. Also unfolding of helices is shown to reduce the dynamics. One primary concern is whether these "rescued" protein dynamics imply protein functionality. Are these semi denatured proteins refolded in the presence of P-granule components? Finally, it is not clear why the authors chose to disrupt folding of the central dimerization domain?

Saying that "reduced alpha-helicity of PGL-3 correlates with slower dynamics in condensates" may be factual in these assays but "correlation" should be expanded upon to include mechanism and to me it seems that the statement should read "aggregation of PGL-3 causes slower dynamics in condensates" (both the partially destabilized mutant and the fully unfolded WT show similar effects perhaps to different degrees).

CROSS-CONSULTATION COMMENTS

I agree with the other reviewer's comments and critiques, I have concerns about the biological relevance and also the biophysical mechanisms. Reflecting on the other reviewers' comments, the papers could provide more depth in one or both of these areas to come to firm conclusions that are either revealing about PGL biology or elucidate a (possible) general biophysical mechanism.

Significance

Hence, although the authors shows how inclusion of other components can alter the one protein component phase separation, this is done with entirely artificial means of destabilizing the fold of one of the domains which likely leads to aggregation. So the true impact of the work is hard to understand because the mutations impact on the basic biophysical properties of the domain (stability, interaction) are not completely characterized and the reason for disrupting this folding is not clear.

My field of expertise is protein phase separation and protein structure.

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

• A significant criticism of the paper is an assumption that readers will be familiar with all of the findings in the author's previous 2016 paper and the PGL-1 papers by Aoki et al. Minimal context is given for each approach. Some conclusions are not well supported and require further analysis, proper controls, and more extensive descriptions of the experiments performed. Most importantly, the central conclusion and title of the paper is that composition can buffer the dynamics of individual proteins within liquid-like condensates. In other words, in vitro condensation assays often do not recapitulate LLPS behavior in vivo. That said, the findings in this study would be significantly strengthened and complemented by observing endogenously tagged PGL-3 and PGL-3 mutants in living worms, considering the efficiency of using CRISPR in C. elegans to insert tags and make precise mutations.

• To improve readability, the introduction to P granules should be expanded, and include the reasons for looking at the nematode-specific PGL-3 protein among all the other known P granule proteins. A recap of previous findings on PGL-3 phase separation, in vivo and in vitro, is warranted, starting with the significant results of Saha et al 2016. Setting up the investigative questions in the context of recent work on PGL-1 (Aoki, et al) is also necessary.

• The physiological concentration of PGL-3 should be more transparent, including why some experiments in this study are done at physiological concentrations while others are not. Describing why salt concentrations, crowding agents, and protein abundance are similar or different for each experiment is necessary and relevant. For example, after showing in Figure 1 that PGL-3 protein phase separates, the paragraph starting on line 161 says that it was previously shown that PGL-3 doesn't phase separate at physiological concentrations without RNA. One has to go back to Figure 1 to realize it was done differently than Figure 2 and Saha 2016. Statements in the same paragraph like "in contrast to full-length PGL-3, mRNA does not support phase separation..." should be qualified by stating the concentration observed, with or without salts or other crowding agents. Similarly, line 230 "suggests that interactions involving the disordered C-terminal region of PGL-3 are not essential for the fast dynamics" and should be qualified with "at non-physiological concentrations and with XX crowding agents or salt concentration." It would be more consistent if physiological concentrations were consistent from figure to figure, as extra variables weaken some of the stated conclusions.

• The 2010 review reference stating that there are 40 P granule enriched proteins is outdated. More recent reviews put the number much higher. This is relevant because the approach to put PGL-3 in a more physiological environment by including just PGL-1, GLH-1 and mRNA with the condensate assays, out of ~100 P granule enriched proteins, may not be sufficient to conclude "that the influence of complex composition on dynamics is modest" (line 223), or imply that the multicomponent nature of the P granule is reconstituted by adding these components (line 355).

• Figure 1C needs to include PGL-3(370-693) in the analysis. Figure 1E is also incomplete without a comparison of FRAP recovery between PGL-3(1-452) and full PGL-3 as the control.

• Figure 4C is missing an essential control where PGL-3 and S1 FRAP is performed without PGL-1, GLH-1, and mRNA. It would also help show sup Fig4A in the main figure to show concentration dependence.

• Consider adding subtitles to supplementary figures.

• M&M should include an explanation for statistical analysis

CROSS-CONSULTATION COMMENTS

I am also in agreement with the comments and critiques of reviewers 2 and 3.

Significance

The paper by Saha and colleagues investigate the in vitro liquid-liquid phase separation propensity of a P granule protein PGL-3 and its structural domains. The findings largely replicate and support the phase-separation properties of a paralogous protein called PGL-1, as recently described by Aoki et al. 2021. Furthermore, they show that the dynamics demonstrated by recombinant PGL-3 may be maintained or buffered by the complex composition of P granules.

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)):

The authors have assembled an enormous amount of statistical data on the genomes and phylogeny of Arctic algae, including the genomes of four new species that they sequenced for this study. Their main finding is that horizontal gene transfer has led to convergent evolution in distantly related microalgae.

**Major comments**

Reviewer #1__: The purpose of the study is not clearly stated in the abstract or the introduction. The authors say (line 93) "Defining the genetic adaptations underpinning these small algal species is crucial as a baseline to understand their response to anthropogenic global change (Notz & Stroeve,2016)." Is this their goal? Or are they just quoting another study? The authors state (line 103) "We extend by sequencing the genomes of four distantly related microalgae...". This is not really a question or a hypothesis. I am sure the authors can provide a more compelling reason to embark on such a labor-intensive study.__

Reply: We agree that the aim was lost in the details and the Introduction is now focused towards the original goal of the study, which was to investigate convergent evolution in a biogeographically isolated ocean. Additional references on the formation and history of the Arctic Basin have been added to the Introduction to provide context. “An ocean has been present at the pole since the beginning of the Cretaceous. Shaped by tectonic processes (Nikishin et al., 2021) the Arctic Ocean has been a relatively closed basin since the Masstrichtian at the end of the late Cretaceous epoch (ca. 70 million years before present), with episodic sea-ice cover since that time (Niezgodzki et al., 2019). This long history suggests limited gene flow from the global ocean over vast time scales and Arctic marine species including microalgae could well have unique adaptations to cold arctic conditions.” Line 78-83.

And following this we provide a clear hypothesis “The potential for lineages of ancient Arctic origin and the episodic input of outside species led us to our hypothesis that Arctic microalgae convergently evolved traits or adaptations aiding survival in an ice-influenced ocean. Line 112-117.

We also discuss both the adaptive and distinct physical environment of the Arctic, and its topographical separation from other ocean regions as dispersal limitation would enhance the Arctic-specific genomic signatures. We now cite the recent paper by Sommeria-Kline et al. (2020), which puts eukaryotic plankton biogeography into a global context (Line 72)

Reveiwer #1__: The most prominent shared trait that the authors found are genes for ice-binding proteins. However, in view of their importance, little information is given about their different types and possible functions.__

Reply: We appreciate the comment and have added information on relevant ice binding proteins found in the Arctic Algae. In addition, we discuss how the functional and secretory diversity of IBP would enhance the survivability of pelagic taxa. Lines 534 to 564.

Although ice binding proteins from multicellular animals and plants are outside the scope of this study, there is a recent review; Bar Doley, Braslavsky and Davies 2016 Annual review of Biochemisty 85: 515-542.

.

Reviewer #1__: The HGT of ice-binding proteins is a major focus of this study, but little is said about what previous studies have said about this. What are the previous studies, what are their findings and how do the present findings contribute to this?__

Reply: We agree that this aspect should have been more visible. We incorporated new data to characterize IBPs drawn from MMETSP transcriptomes, and environmental Tara Ocean metagenomes, as well as our Arctic strains. We note that as we take a PFAM-based approach, the IBPs treated are DUF3494/PF11999 domain, which are type 1 IBPs / algal IBPs (Raymond and Remia 2019). As an example of novelty, we identify the position of IBPs from dinoflagellates, within a larger Arctic Clade that included CCMP2293, CCMP2436 and CCMP2097 and Arctic TARA IBP, rendering this a pan-algal IBD clade.

In addition, we were able to resolve the position of anomalous F. cylindrus IBP that fell between two Arctic associated clades (A and B, in our Fig 4). This finding is consistent with F. cylindrus originating in the Arctic as previously suggested and subsequently invading the Southern Ocean.

The recurrent acquisition of multiple diverse IBP isoforms in individual species through HGT events has not been previously reported, and the extent of isoforms in the Arctic was surprising. See for example multiple different IBP forms with separate origins in Pavlovales CCMP2436 (Fig 4). The previous studies are referred to in the context of the phylogeny of the IBD within the results section: Lines 322- 413, and Lines 534-585.

Reviewer #1: Figure 5 on HGT of ice-binding proteins is difficult to follow. It would be clearer if each panel could be described separately, clearly stating its main finding. I doubt that a reader could look at this figure and explain to a colleague what it shows.

Reply: We have revised rearranged the figure (now Fig 4) with Arctic A, B, C and D clearly indicated as well as the two Antarctic dominated clades. The upper schematic includes the deepest phylogeny of algal IBDs to date, incorporating all of UniRef, MMETSP and TARA Oceans. The fasta files underlying the tree and the nexus file used are provided the S1 Data Folder, which is an excel folder with information on the analysis of the data. The callout and order of the clades has been revised to facilitate interpretation of the phylogenies more clearly. The entire section has been completely rewritten.

Reviewer #1: This is also a problem with many of the other figures. For each figure, what is the question being asked and what is its take-home message?

Reply: We agree that the message was lost and have now focused on our original question in our accepted proposal to JGI. “Is there a convergence among arctic microalgae at the genomic level?”. We found some genome properties were common among the Arctic isolates (more unknown PFAMS and several expanded PFAMs). The importance of ice binding proteins in Arctic Isolates and the widespread inter-algal HGT of this important protein among the Arctic strains. The IBP biogeography and phylogeny strongly indicate that the Arctic microalga have acquired IBP locally and that the Antarctic strains have acquired additional isoforms independently from Antarctic bacteria and fungi (Lines 565-585).

Reviewer ____#1____: ____The paper has more data than a reader can absorb. It could be strengthened by reducing the number of figures, simplifying them if possible, and more clearly stating the value of the remaining figures.

Reply. As suggested, we have refocused the paper, removing more speculative statistics based analysis and associated figures. The main conclusions are supported by the 5 main figures. We are now present 5 main figures and 11 supplementary figures (previously 23 downloadable supplementary figures and 40 on-line only figures supporting the support figures). We agree with the reviewer, and we feel the revised version is a more transparent synthesis. Briefly the Figures illustrate the following points. Fig. 1. The multigene tree of available algal genomes and transcriptomes provides a clear framework for judging the divergence of subsequent individual gene and PFAMs phylogenies. Fig. 2 (originally Fig. 3). Indicates the convergence of PFAM domains in the Arctic strains, in contrast to strains from elsewhere. Fig. 3 (originally Figure 4) shows Arctic specific expansions and contraction of PFAM domains, again demonstrating convergent evolution in the Arctic. The figure identifies specific PFAMs that contribute to the within-Arctic convergence. This figure is based on statistical methods independent of Fig 2. Figure 4 is the most extensive IBP phylogeny to date and has been discussed above. Figure 5, which was supplementary in our non-peer reviewed version, shows the biogeographic distribution of IBP, and can be compared to the distributions of the 18S rRNA genes from the four Arctic algae provided as supplementary (S6 Fig.)

**Minor comments**Reviewer #1

1. The figure citations are confusing. E.g., what does "Fig.1- Figure supplement 1" refer to? Does this refer to 1 or 2 figures? Apparently, it refers only to Fig. S1, so many readers will be confused when they look at Fig. 1.

Reply: We apologize for the confusing format; the manuscript had been formatted for the online journal eLife. Our revision follows the more traditional style of PLoS Biology and other Review Commons journals.

.

Multiple citations should be in order of publication date, not alphabetical order.

Reply ; We agree that date of publications is quite standard and recognizes priority of publication. Several on line journals no longer follow this rule and citation order will follow the specific style used by our accepting journal.

Reviewer #1 (Significance (Required)): It is well known that useful genes tend to be shared among microorganisms. The present study strengthens previous studies in showing that gene transfer is an important process in polar regions.

Reply: We thank the reviewer for recognizing the importance of our study.

---

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

This manuscript is the result of a large international collaborative effort, including the US Department of Energy Joint Genome Institute. Its focus is comparative genomics of eukaryotic Arctic algae. The primary data described in the ms are four new genome and transcriptome sequences from diverse Arctic algae, represented by a cryptomonad, a haptophyte, a chrysophyte, and a pelagophyte.

The authors compare these new data to previously published genomic/transcriptomic data from eukaryotic algae with the goal of understanding genome evolution in the Artic. The results of the paper are a series large-scale comparative genomic bioinformatics analyses, including the associated statistical analyses. The key findings center on statistically significant features of Arctic genomes, features that stand out as compared to the genomes of algae that are not primarily found in the Arctic. Together, these findings allow the authors to make various hypotheses and suggestions about genetic adaptations to polar environments.

By far the most significant finding is that the genomes of Arctic algae are enriched in genes encoding proteins with an ice-binding domain, paralleling findings from Antarctic algae. These genes appear to have spread among Arctic algal genomes via horizontal gene transfer, which raises a series of interesting questions. In my opinion, the major conclusions of this paper are supported by the data. Listed below are a few comments that may improve the ms:

Reviewer #2.

1) In today's post-genomics era, everyone seems to be sequencing nuclear genomes. Often what distinguishes high-impact and low-impact genome papers is the number of genomes presented and the quality of the genome assembly. I may have missed it, but reading the main text, the figures/tables, and the supplementary data I was not able to get a sense of the quality of the four genome assemblies from which the main findings are based. I was eventually able to find this information from PhycoCosm (note: some of the links to this site are not working in the ms). My quick scan of the PhycoCosm summary info for the four genomes indicates that the assemblies are highly fragmented, likely because they are based on short-read Illumina sequencing rather than a combination of short and long reads. I think it is important to briefly discuss (and or present) the quality of the assemblies in the ms and to highlight the potential limitations/drawbacks of employing highly fragmented assemblies when carrying out large-scale comparative genomics.

Reply: We agree and the data concerning the genome quality assemblies has been moved to the main text Table 1. The comparison with other paired related strains is provided in an excel folder designated S2 Data Folder.

Reviewer #2.

2) Horizontal gene transfer is undeniably a major driving force in evolution, and one that has shaped genomic architecture across the Tree of Life. I believe the data presented here support a role for HGT in the genome of evolution of Arctic algae, particularly with respect to genes encoding proteins with an ice-binding domain. However, we can all think of numerous instances when authors of genome papers were too quick to point to HGT. Thus, I would urge more caution and balance when presenting the HGT data, including some discussion about factors that could incorrectly lead researchers to conclude a significant role for HGT, such as contamination, gene duplication, mis-assemblies, etc. I'm not suggesting that you change the main conclusions, but just tone down the language in places (e.g., "we reveal remarkable convergence in the coding content ... ").

Reply: We understand the reviewers concerns and now more clearly outline the pipeline we have used to identify HGTs. This included: filtering each genome to remove all possible contaminant sequences first, considering both contig co-presence of vertical- and horizontally-derived genes, and reciprocal and independent annotations of gene sequences in both genome sequences and MMETSP transcriptomes. Retained genes were subjected to simultaneous BLAST analysis and manually curated phylogenies using decontaminated reference datasets. The most parsimonious explanation for our final IBP domain microbial algal clusters (Fig 4) is HGT. On the side of caution, we removed the entire section that identified potential arctic HGT based primarily on a less targeted broad statistical analysis. The focus is now on 3 genes that have clearly identifiable utility in the Arctic, were found to be enriched in Arctic genomes via a separate analysis and had homologs in the Tara Ocean Polar circle data. In addition, we describe more clearly the role of expansion and enrichment of PFAMs and the high proportion genes without an identifiable PFAMs in the Arctic strains as evidence for Arctic convergence separate from potential HGT.

Reviewer #2.

3) The downside of studying protists (as compared to multicellular animals, for instance) is that most are not widely known by the scientific community and even fewer scientists can picture what they actually look like (e.g., Pavlovales sp. CCMP2436). A few more details about the four Arctic algae that make up the focus of this paper might be helpful for the casual reader. My sense is that if at the next departmental meeting I asked my colleagues what a pelagophyte was most would look at me with a blank stare. Moreover, am I right to assume that all four algae are psychrotolerant rather than psychrophilic (Supplement Fig. 1 makes me think otherwise). It might be good to point out the difference in the text.

Reply: High resolution images of each strain are available on the JGI home page for each alga, given the multiple figures we feel photos would not add information.

Reviewer #2

4) I don't think Supp. Table 1 (the Pan-algal dataset) got uploaded correctly during the manuscript submission stage. The first link I click on gives me Supp. Table 2.

Reply: We apologize for this, the format was incorrect for the file designation and there were lost links. We now more actually refer to these as Data Folders as they are excel folders containing multiple sheets, All supplementary links will be verified again on final submission.

.

Reviewer #2 (Significance (Required)):

By far the most significant finding from this paper is that the genomes of Arctic algae are enriched in genes encoding proteins with an ice-binding domain, paralleling findings from Antarctic algae. These genes appear to have spread among Arctic algal genomes via horizontal gene transfer, which raises a series of interesting questions. This is not the first paper to present these types of ideas, but it is arguably the broadest analysis yet, at least with respect to eukaryotic algae. This work will be of great interest to polar scientists, phycologists, protistologists, and the genomics community. I am genome scientist studying protists, including algae.

Reply. We thank the reviewer for their insightful comments.

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

**Summary:**

This manuscript is focused on Arctic microalgae, an important yet understudied community in permanently cold ecosystems. By sequencing the genomes of four phylogenetically diverse and uncharacterized polar algae, the authors seek to elucidate genomic features and protein families that are similar in polar species (and differ from their relatives from temperate environments) This work used high-throughput genomic sequencing and computational analysis to demonstrate significant horizontal gene transfer (HGT) in several gene families, including ice-binding proteins. The authors suggest that this HGT is an effector of environmental adaptation to Arctic environments.

**Major comments and experiment suggestions:**

The authors conclude that HGT between arctic species is a driver of polar adaptation. The authors strongly support the claim that HGT is present more frequently in the polar algae examined here. Whether this is adaptive should be further explored though. For instance, ice-binding domains were one PFAM group found at significantly higher frequencies in the polar species - but are all of these species associated with ice? What would be the benefit of IBDs in an alga that is found in the open ocean. Similar with the other domains (Lns 333-335), its not clear whether these are truly adaptive features. ____This is more speculative.

Reply: We agree that detail was lacking and have considerably expanded our introduction on the character of the Arctic Ocean and have stated the goals and underlying hypothesis. Briefly, all surface water organisms that live in the Arctic encounter ice during the year as the ocean freezes in winter, and surface waters reman around negative 1.7 °C for much of the year. This information has been added to the introduction. We have also expanded the discussion on the multiple effects of different IBPs that would be ecologically beneficial for plankton as well as ice-algae and cite relevant experimental studies and reviews.

Reviewer #3) ____HGT was a major conclusion of this study, putting this in a wider perspective would strengthen the conclusion, especially in the context of HGT from prokaryotes. Are there insights on whether IBDs are present in Arctic prokaryotes?

Reply: This is a good question, and we now point out that there were 91 Arctic bacterial and archaeal IBP sequences in our comparative dataset. In contrast to the Antarctic clades, none were closely related to the Arctic strain IBPs (Fig 4). Line 336.

Reviewer #3) ____The data obtained from the genomic works supports the conclusions stronger that ones from transcriptomes, where what genes/domains are present would depend largely on the sampling conditions. This should be emphasized.

Reply: The main rational for using transcriptomes was that more of these are available and enabled us to detect convergences and HGT across a broader taxonomic range than would be possible with genome-only data, where we had access to a total of only 21 microalgal genomes. In general transcriptome studies are aimed at identifying responses under different conditions and rely on comparative expression data, usually 2-fold differences in up or down expression under different growth conditions, see for example Freyria et al. 2022 (Communications Biology). Unlike a transcriptome expression study, our data mining detected any (constitutive or regulated) expression in these unicellular haploid cells, we would have detected genes used under any condition that an algal happened to be growing. IBD was not detected in any of the temperate genomes, and only detected in transcriptomes of Arctic and Arctic-Boreal groups. However, we agree that there may be some limitation of transcriptomes only studies and mention this. Lines 522-528.

Reviewer #3) ____An experiment to determine whether the species are cold extremophiles (psychrophiles) would be useful here to strongly support the data in Figure 1. The authors state that their species can not survive >6C but this is based on experiments done on older studies. Considering the cultures have been maintained as a continuous culture for decades, confirming that they still have psychrophilic characteristic would be useful. This is a straightforward and low cost experiment that requires simply measuring growth rates at several temperatures to define the optimal and confirm that the cells are not viable above 6C.

Reply: These are interesting points, and the broad “background” statements in the original manuscript would require a separate study,and have been deleted. Temperature tolerance experiments are not so simple for cold adapted algae with slow growth rates. Such experiments require specialized incubators to maintain low temperatures. Temperature experiments have been carried out on the cultures in the context of other studies, see for example, Daugberg et al. 2018, J. Phycol. But this is not within the scope of the present study.

We now restrict our conclusions to the specific question of convergence among Arctic strains. We apologize for the misunderstanding on the history of the cultures. They have not been in “continuous culture” but are cryopreserved. We now simply indicate that they grow below 6 °C, which is sufficient to assume that they are likely cryophiles, our experience is that they do not grow well or at all at higher temperatures, our efforts have been to maintain the cultures that are otherwise easily lost. We now make no claims about optimality or limits. Here we simply examined genomes and available transcriptomes that were generated from algae growing at 4-6 °C.

Reviewer #3) ____**Minor comments:**

Defining the species used here as psychrophiles would put the study in context better. The authors relate their finding to Antarctic species (HGT, ice-binding domains, large genomes) all of which are confirmed psychrophiles.

Reply: The temperature definition of psychrophiles is surprisingly high (optimal growth below 15 °C) and this definition of psychrophiles is now given in the introduction. The point is really that there are few isolates from cold surface waters that have been well studied. We now add. “A handful of polar algal genomes have been extensively studied, with 4 of these from around Antarctica and classified as psychrophiles (not being able to grow above 15 °C (Feller & Gerday, 2003)”. Lines 103-107.

Reviewer #3) ____A short rationale on why these species at all would be useful - are they representative of their classes? Do they have psychrophilic characteristics that might make them useful models in the future? Are they widely used now?

Reply: We appreciate the point as the definition of utility in discovery-based science is an open dialog.

We agree that the study requires context and have added our rational for selecting the species for genome sequencing to the introduction. “To address questions on genetic adaptations to this ice-influenced environment, we sequenced 4 phylogenetically divergent microalgae, from 4 algal classes belonging to 3 algal phyla: Cryptophyceae (Cryptophyta), Pavlovophyceae (Haptophyta), Chrysophyceae and Pelagophyceae (both in the Ochrophyta) isolated from the ca. 77 °N, where surface ice flow persists through June (Mei et al., 2002). The four isolates were selected as representatives of different water and ice conditions and phylogeny from available strains collected in April and June 1998 during the North Water Polynya study”.

Reviewer #3) ____Starting algal cultures were maintained in a continuous culture since 1998 and under continuous light since at least 2015, have the authors confirmed that these algae retain their physiological features even after this long time? The accumulation of mutations is a possibility here.

Reply: We apologize for the misunderstanding of the timeline; the history of the cultures was not given in the manuscript and the inferred history is not quite correct. The 2015 date was the year of publication for the MMETSP data. Our continuous light statement is a record of our standard culture conditions. We now elaborate on the material used in the current study. The cultures were deposited in the Bigelow culture collection (now NCMA) in 2002 and cryopreserved once they had been verified and given a culture designation. We obtained fresh cultures in 2005 and these were used for the MMETSP project. We obtained fresh cultures again in 2011, specifically for the JGI genome project. These algae do not grow fast and most of the DNA was sent to JGI in 2012 for most of the isolates. This history is rather long and not relevant, since one would speculate that over the years the algae would tend to lose the ice associated functionality, e.g. they were not frozen in seawater every year for 4 to 6 months or subject to sudden freshwater exposure, when ice melts. We would encourage other researchers to order the cultures and run experiments. We note that many of the 40 or so algae isolated from the same campaign have been used by others for specific studies and at least 8 are in the MMETSP data set. The presence of 18S rRNA and phylogenetic position of the IBP sequences compared to Tara Arctic circle data confirms long-term Arctic presence of each species and the IBP domains in the Arctic without marked changes over the last 20 years.

Reviewer #3) ____Ln381 - The culture collection IDs for each sequenced species should be included here

Reply: we have added the culture IDs throughout.

Reviewer #3) ____Ln. 389 - Algal cells are harvested and used for nucleic acid extraction, the nucleic acids themselves are not harvested

Reply: we agree and corrected the wording

Reviewer #3 (Significance (Required)):

This study is well places in the current state of research on polar alga and represents a significant and very valuable addition to the current knowledge pool. Algae in general are lagging behind other groups of photosynthetic organisms in the number of sequenced and analyzed genomes, despite algae being one of the main primary producers globally. This is even more strongly felt in polar research, where only 4 species have been sequenced, most of which are restricted to Antarctica. There is a true gap in our knowledge when it comes to Arctic species, and this study fills this gap. As the authors correctly state, we need more knowledge on polar environments and the primary producers that support these important ecosystems in light of current climate change trends.

Reply: we appreciate the succinct summary of our study and thank the reviewer for insights and suggestions that have improved the manuscript.

Reviewer field of expertise: Polar algae, stress responses, plant and algal energetics, cell signalling

Reply: We appreciate the incites and perspective steming from the reviewer's expertise.

Relevant key references cited in the reply:

Daugbjerg N, Norlin A, Lovejoy C. Baffinella frigidus gen. et sp. nov. (Baffinellaceae fam. nov., Cryptophyceae) from Baffin Bay: Morphology, pigment profile, phylogeny, and growth rate response to three abiotic factors. Journal of Phycology. 2018;54(5):665-80

Feller, G. and Gerday, C. (2003) Psychrophilic enzymes: Hot topics in cold adaptation. Nat Rev Microbiol, 1, 200-208.

Freyria NJ, Kuo A, Chovatia M, Johnson J, Lipzen A, Barry KW, et al. Salinity tolerance mechanisms of an Arctic Pelagophyte using comparative transcriptomic and gene expression analysis. Communications Biology. 2022;5(1). doi: 10.1038/s42003-022-03461-2

Mei, Z. P., Legendre, L., Gratton, Y., Tremblay, J. E., Leblanc, B., Mundy, C. J., Klein, B., Gosselin, M., Larouche, P., Papakyriakou, T. N., Lovejoy, C. and Von Quillfeldt, C. H. (2002) Physical control of spring-summer phytoplankton dynamics in the North Water, April-July 1998. Deep-Sea Research Part Ii-Topical Studies in Oceanography, 49, 4959-4982.

Niezgodzki, I., Tyszka, J., Knorr, G. and Lohmann, G. (2019) Was the Arctic Ocean ice free during the latest Cretaceous? The role of CO2 and gateway configurations. Global and Planetary Change, 177, 201-212.

Nikishin, A. M., Petrov, E. I., Cloetingh, S., Freiman, S. I., Malyshev, N. A., Morozov, A. F., Posamentier, H. W., Verzhbitsky, V. E., Zhukov, N. N. and Startseva, K. (2021) Arctic Ocean Mega Project: Paper 3-Mesozoic to Cenozoic geological evolution. Earth-Science Reviews, 217.

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:

This manuscript is focused on Arctic microalgae, an important yet understudied community in permanently cold ecosystems. By sequencing the genomes of four phylogenetically diverse and uncharacterized polar algae, the authors seek to elucidate genomic features and protein families that are similar in polar species (and differ from their relatives from temperate environments) This work used high-throughput genomic sequencing and computational analysis to demonstrate significant horizontal gene transfer (HGT) in several gene families, including ice-binding proteins. The authors suggest that this HGT is an effector of environmental adaptation to Arctic environments.

Major comments and experiment suggestions:

• The authors conclude that HGT between arctic species is a driver of polar adaptation. The authors strongly support the claim that HGT is present more frequently in the polar algae examined here. Whether this is adaptive should be further explored though. For instance, ice-binding domains were one PFAM group found at significantly higher frequencies in the polar species - but are all of these species associated with ice? What would be the benefit of IBDs in an alga that is found in the open ocean. Similar with the other domains (Lns 333-335), its not clear whether these are truly adaptive features. This is more speculative.
• HGT was a major conclusion of this study, putting this in a wider perspective would strengthen the conclusion, especially in the context of HGT from prokaryotes. Are there insights on whether IBDs are present in Arctic prokaryotes?
• The data obtained from the genomic works supports the conclusions stronger that ones from transcriptomes, where what genes/domains are present would depend largely on the sampling conditions. This should be emphasized.
• An experiment to determine whether the species are cold extremophiles (psychrophiles) would be useful here to strongly support the data in Figure 1. The authors state that their species can not survive >6C but this is based on experiments done on older studies. Considering the cultures have been maintained as a continuous culture for decades, confirming that they still have psychrophilic characteristic would be useful. This is a straightforward and low cost experiment that requires simply measuring growth rates at several temperatures to define the optimal and confirm that the cells are not viable above 6C.

Minor comments:

• Defining the species used here as psychrophiles would put the study in context better. The authors relate their finding to Antarctic species (HGT, ice-binding domains, large genomes) all of which are confirmed psychrophiles.
• A short rationale on why these species at all would be useful - are they representative of their classes? Do they have psychrophilic characteristics that might make them useful models in the future? Are they widely used now?
• Starting algal cultures were maintained in a continuous culture since 1998 and under continuous light since at least 2015, have the authors confirmed that these algae retain their physiological features even after this long time? The accumulation of mutations is a possibility here.
• Ln381 - The culture collection IDs for each sequenced species should be included here
• Ln. 389 - Algal cells are harvested and used for nucleic acid extraction, the nucleic acids themselves are not harvested

Significance

This study is well places in the current state of research on polar alga and represents a significant and very valuable addition to the current knowledge pool. Algae in general are lagging behind other groups of photosynthetic organisms in the number of sequenced and analyzed genomes, despite algae being one of the main primary producers globally. This is even more strongly felt in polar research, where only 4 species have been sequenced, most of which are restricted to Antarctica. There is a true gap in our knowledge when it comes to Arctic species, and this study fills this gap. As the authors correctly state, we need more knowledge on polar environments and the primary producers that support these important ecosystems in light of current climate change trends.

Review field of expertise: Polar algae, stress responses, plant and algal energetics, cell signalling

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 is the result of a large international collaborative effort, including the US Department of Energy Joint Genome Institute. Its focus is comparative genomics of eukaryotic Arctic algae. The primary data described in the ms are four new genome and transcriptome sequences from diverse Arctic algae, represented by a cryptomonad, a haptophyte, a chrysophyte, and a pelagophyte.

The authors compare these new data to previously published genomic/transcriptomic data from eukaryotic algae with the goal of understanding genome evolution in the Artic. The results of the paper are a series large-scale comparative genomic bioinformatics analyses, including the associated statistical analyses. The key findings center on statistically significant features of Arctic genomes, features that stand out as compared to the genomes of algae that are not primarily found in the Arctic. Together, these findings allow the authors to make various hypotheses and suggestions about genetic adaptations to polar environments.

By far the most significant finding is that the genomes of Arctic algae are enriched in genes encoding proteins with an ice-binding domain, paralleling findings from Antarctic algae. These genes appear to have spread among Arctic algal genomes via horizontal gene transfer, which raises a series of interesting questions. In my opinion, the major conclusions of this paper are supported by the data. Listed below are a few comments that may improve the ms:

1) In today's post-genomics era, everyone seems to be sequencing nuclear genomes. Often what distinguishes high-impact and low-impact genome papers is the number of genomes presented and the quality of the genome assembly. I may have missed it, but reading the main text, the figures/tables, and the supplementary data I was not able to get a sense of the quality of the four genome assemblies from which the main findings are based. I was eventually able to find this information from PhycoCosm (note: some of the links to this site are not working in the ms). My quick scan of the PhycoCosm summary info for the four genomes indicates that the assemblies are highly fragmented, likely because they are based on short-read Illumina sequencing rather than a combination of short and long reads. I think it is important to briefly discuss (and or present) the quality of the assemblies in the ms and to highlight the potential limitations/drawbacks of employing highly fragmented assemblies when carrying out large-scale comparative genomics.

2) Horizontal gene transfer is undeniably a major driving force in evolution, and one that has shaped genomic architecture across the Tree of Life. I believe the data presented here support a role for HGT in the genome of evolution of Arctic algae, particularly with respect to genes encoding proteins with an ice-binding domain. However, we can all think of numerous instances when authors of genome papers were too quick to point to HGT. Thus, I would urge more caution and balance when presenting the HGT data, including some discussion about factors that could incorrectly lead researchers to conclude a significant role for HGT, such as contamination, gene duplication, mis-assemblies, etc. I'm not suggesting that you change the main conclusions, but just tone down the language in places (e.g., "we reveal remarkable convergence in the coding content ... ").

3) The downside of studying protists (as compared to multicellular animals, for instance) is that most are not widely known by the scientific community and even fewer scientists can picture what they actually look like (e.g., Pavlovales sp. CCMP2436). A few more details about the four Arctic algae that make up the focus of this paper might be helpful for the casual reader. My sense is that if at the next departmental meeting I asked my colleagues what a pelagophyte was most would look at me with a blank stare. Moreover, am I right to assume that all four algae are psychrotolerant rather than psychrophilic (Supplement Fig. 1 makes me think otherwise). It might be good to point out the difference in the text.

4) I don't think Supp. Table 1 (the Pan-algal dataset) got uploaded correctly during the manuscript submission stage. The first link I click on gives me Supp. Table 2.

Significance

By far the most significant finding from this paper is that the genomes of Arctic algae are enriched in genes encoding proteins with an ice-binding domain, paralleling findings from Antarctic algae. These genes appear to have spread among Arctic algal genomes via horizontal gene transfer, which raises a series of interesting questions. This is not the first paper to present these types of ideas, but it is arguably the broadest analysis yet, at least with respect to eukaryotic algae. This work will be of great interest to polar scientists, phycologists, protistologists, and the genomics community. I am genome scientist studying protists, including algae.

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 assembled an enormous amount of statistical data on the genomes and phylogeny of Arctic algae, including the genomes of four new species that they sequenced for this study. Their main finding is that horizontal gene transfer has led to convergent evolution in distantly related microalgae.

Major comments

1. The purpose of the study is not clearly stated in the abstract or the introduction. The authors say (line 93) "Defining the genetic adaptations underpinning these small algal species is crucial as a baseline to understand their response to anthropogenic global change (Notz & Stroeve,2016)." Is this their goal? Or are they just quoting another study? The authors state (line 103) "We extend <previous findings> by sequencing the genomes of four distantly related microalgae...". This is not really a question or a hypothesis. I am sure the authors can provide a more compelling reason to embark on such a labor-intensive study.
2. The most prominent shared trait that the authors found are genes for ice-binding proteins. However, in view of their importance, little information is given about their different types and possible functions.
3. The HGT of ice-binding proteins is a major focus of this study, but little is said about what previous studies have said about this. What are the previous studies, what are their findings and how do the present findings contribute to this?
4. Figure 5 on HGT of ice-binding proteins is difficult to follow. It would be clearer if each panel could be described separately, clearly stating its main finding. I doubt that a reader could look at this figure and explain to a colleague what it shows.
5. This is also a problem with many of the other figures. For each figure, what is the question being asked and what is its take-home message?
6. The paper has more data than a reader can absorb. It could be strengthened by reducing the number of figures, simplifying them if possible, and more clearly stating the value of the remaining figures.

Minor comments

1. The figure citations are confusing. E.g., what does "Fig.1- Figure supplement 1" refer to? Does this refer to 1 or 2 figures? Apparently, it refers only to Fig. S1, so many readers will be confused when they look at Fig. 1.
2. Multiple citations should be in order of publication date, not alphabetical order.

Significance

It is well known that useful genes tend to be shared among microorganisms. The present study strengthens previous studies in showing that gene transfer is an important process in polar regions.

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 are very grateful about the thorough reading and deep understanding of the work that these 4 reviewers have provided. Although it is evident that they still request an improved profiling of some aspects, it is very encouraging that all four think the work is very interesting, original, insightful and adds a new layer of knowledge to the regulation of DNA damage sensing and repair. It is also very rewarding that the four reviewers estimate that this work will sew connections between different fields and interest a broad readership. This is why we have designed here a very deep revision, tailored to satisfy all the raised concerns except one, and this just for technical reasons.

Please find below the original reviewers’ comments and our answers to them preceded by the symbol “>”:

Reviewer #1 (Evidence, reproducibility and clarity (Required)): Ovejero et al. report an increase in lipid droplet (LD) abundance after long (from 120' on) exposure of budding yeast cells to DNA damaging agents zeocin and camptothecin (CPT). Next, they analyze DNA damage signaling in yeast mutants that impair triacylglycerol (TAGs) or sterol (STEs) esterification. They observe a slight anticipation in Rad53/CHK2 phosphorylation (indicative of DDR signaling) in yeast stem mutants, as well as in yeast cells or human cells lines pre-treated with oleate upon zeocin treatment. Yeast stem mutants are sensitive to zeocin and captothecin, but only confer sensitivity to hydroxyurea upon combination with tagD mutations. Authors relate these phenotypes to a somewhat decreases DSB resection in yeh2D mutants (expected to have reduced steryl esters pools) and RPA-foci in steD yeast cells. Next, a reduction in single strand annealing recombination repair events upon zeocin treatment is reported using a genetic reporter in steD mutants and oleate-treated cells. From these data they conclude that inability to process sterols in response to DSBs leads to an exacerbated DDR and prevents DNA repair. Next, it is shown that Flag-tagged Tel1 distinctly interacts with mono-phosphate phosphoinositides, including PI(4)P. An interaction in vivo is also inferred through Proximity Ligation Assays (PLA) using anti-PI(4)P and anti-ATM antibodies in human cell lines, which was moderately downregulated upon treatment with MMS or zeocin. Over-expression of the Osh4/OSBP1 transporter, which consumes PI(4)P, increased the number of Tel1 (nuclear) foci upon zeocin treatment. Conversely Sac1 ablation, in which accumulation of PI(4)P is expected, abrogated nuclear Tel1 foci formation and reduced telomere length (a phenotype related to lack of Tel1 function). From these results authors conclude that Tel1 availability in the nucleus is influenced by PI(4)P availability. Lastly, treatment with an OSBP1 inhibitor led to a cell line and damaging agent -variable reduction of ATM phosphorylation and a mostly non-significant reduction of DNA resection, measured by native BrdU detection, in response to CPT treatment. Overall, authors conclude that i) biding of Tel1/ATM to PI(4)P modulates its functional availability in the nucleus, and that ii) DNA damage elicits the esterification and storage of sterols toward LDs, which contributes to tritate Tel1/ATM away from the nucleus dampening the DDR and affecting long-range resection.

Major comments: While the conclusion that Tel1/ATM binds PI(4)P and this interaction modulates Tel1/ATM functional availability at the nucleus is convincing, the conclusion that DSBs elicit a change in the metabolism of this lipid to "control" Tel1/ATM function is not demonstrated. The notion that sterol processing occurs in response to DSBs is not sufficiently supported by the data presented, as the increase in LD numbers is observed much after activation of the DDR (Rad53 phosphorylation) in Zeozin-treated yeast cells.

We are afraid that we have not been clear enough in explaining the kinetics giving rise to our model. As indicated by the reviewer, our work shows, through kinetic studies, that the storage of sterols within LD occurs at later stages than the activation of the DDR by Tel1 and Rad53 phosphorylation. Tel1 foci decline is necessary for subsequent engagement of downstream DNA long-range resection. Since we propose that sterol storage within LD is a means to attenuate Tel1 engagement at DSBs, it is thus logical (and thus compatible with the data we show) that LD number increase occurs simultaneously with Tel1 foci decrease, at late stages of the reactionWe will include this explanation and graph in the revised version of the work.

In addition, evidence is not provided on the mechanisms by which PI(4)P metabolism would be controlled, which would be expected to be DDR-independent as they are placed upstream of this signaling pathway in the author's model.

The key mechanism through which, in the end, PI(4)P metabolism will be controlled, is the esterification of sterols within LD. Given that, as clarified above, LD formation in response to DSBs occurs “late” (i.e., after 120 min), it is not excluded that the DDR itself can instruct, through phosphorylation of some effector(s), LD formation. In other words, by ordering LD formation, the DDR would be launching a self-limiting mechanism. In support, we now know, although we do not show in this work, that eliminating key DDR proteins prevents the formation of LD in response to DNA damage. Because of this, we have undertaken an educated-guess approach and chosen critical or rate-limiting enzymes in LD biology either possessing an S/T-Q cluster domain (predicted to be a phosphorylation substrate for the DNA Damage Response kinases (1), and/or retrieved in phospho-proteomic screens as specific DDR targets (2,3). This adds up to 28 proteins in S. cerevisiae and 45 proteins in Homo sapiens. Importantly, the emergent candidates fall into two identical categories in both organisms. To provide initial support for their pertinence, we have generated a point mutant in the putative S/T-Q cluster of one of the yeast candidates. Of high relevance, we find that the concerned mutant is impaired in correctly triggering LD formation in response to DNA damage, and we have now obtained a specific funding to pursue this characterization that, as such, constitutes a different work from the one presented in this manuscript. We hope that the reviewer is now convinced yet that she/he agrees in keeping this information for subsequent manuscript(s).

The damaging agents used have been suggested to alter the redox metabolism and even lipid peroxidation (Kitanovic 2009, Mizumoto 1993, Krol 2015, Todorova 2015, Ren 2019, Singh 2014). Hence it is possible that PI(4)P changes are not due to DSBs, but an indirect though relevant effect. In absence of direct evidence supporting an active regulation of PI(4)P dynamics in response to DNA breaks, this conclusion remains speculative and this should be noted in the manuscript.

We fully agree with the reviewer that the used genotoxins are triggering a myriad of effects which could elicit the same phenomenon by indirect means. Yet, we want to stress that the use of camptothecin, which elicits a very robust LD formation phenotype (Figure 1C), is very likely specific, as it is proven as a potent and direct trapper of Top1 onto DNA after having cleaved it. Nevertheless, we propose in the next paragraph two specific experiments to dismiss this problem, please see immediately below.

Authors conclude that LD is specific to DSB induction. This seems an overstatement as they just reported LD increases in response to two agents that also induce other kinds of DNA damage. To also strengthen the link between DSBs and PI(4)P modulation of Tel1 function, authors should analyze LD numbers, Rad53 phosphorylation and Tel1 nuclear re-localization in response to HO-induced DNA breaks (e.g., using the system employed in Figure 3C).

We humbly think that enzymatically-induced DNA breaks will both activate Rad53 phosphorylation and Tel1 nuclear concentration, as this has already been established, thus requiring no further exploration. Yet, it is very important to assess the reviewer’s suggestion concerning whether enzymatically-induced DNA breaks also trigger the formation of LD. To this end, we will perform two complementary studies in which, instead of using HO, which cuts only a few times in the genome, we will:

1. a) exploit the naturally DSB-accumulating mutant rad3-102, which we previously characterized in the past (4), and which we already exploit in this work for recombination analyses (Figure S4A), to evaluate whether it endogenously harbors more LD in comparison with the WT.
2. b) we have recently created a tool in which gRNAs targeted to different subsets of transposons in the genome can drive Cas9 to create DSB in a dose-dependent manner ((9), under revision in Genetics). We will use this system to monitor the LD formation in response to Cas9-triggered cuts. In addition, on figure 5A, significant differences in GFP-Tel1 foci abundance between WT and steD or yeh2D cells are only observed after 210', way after the slight effect on Rad53 phosphorylation is observed. This is at odds with the conclusion that Tel1 association to STEs modulates DDR signaling.

We are afraid that we have not been clear enough in explaining the kinetics giving rise to our model. As indicated by the reviewer, our work shows, through kinetic studies, that the storage of sterols within LD occurs at later stages than the activation of the DDR by Tel1 and Rad53 phosphorylation. Tel1 foci decline is necessary for subsequent engagement of downstream DNA long-range resection. Since we propose that sterol storage within LD is a means to attenuate Tel1 engagement at DSBs, it is thus logical (and thus compatible with the data we show) that LD number increase occurs simultaneously with Tel1 foci decrease, at late stages of the reactionWe will include this explanation and graph in the revised version of the work.

Minor comments:

Figure S1D and E, experiments should be carried out to include time points in which LD accumulation and cell cycle arrest are observed upon zeocin treatment (i.e., up to 210' as in Figure 1A)

We will provide cytometry profiles of cells at 210 min. These data exist already in our laboratory.

How do authors explain increased single strand annealing recombination frequencies in steD and oleate-treated wild type cells (Figure 4A). Should it not be expected that increased STEs also impair recombination induced by endogenous damage?

Only ste∆ (and not +oleate) indeed manifests an increase in basal recombination frequencies, likely arising from endogenous damage. Although the increase is observed, it is not significant. We agree anyway with the reviewer that, was the experiment to be repeated more times, the increase may be found significantly different. We do not have any honest proposal to explain this.

Data presented in figure 4B and 4C are not fully convincing. Performing time course experiments might help concluding if the differences observed represent a relevant defect in DSB processing.

We will perform a Pulsed Field Gel Electrophoresis (PFGE) kinetcis in response to zeocin with or without oleate pre-loading to reinforce the conclusion.

Is Figure 5B referring to Flag-tagged Tel1 or GFP-tagged Tel1 as stated in the figure legend?

There is a misunderstanding here, as the mentioned Figure 5B corresponds to P-ATM immunofluorescences in human cells, not to any tagged Tel1 experiment.

Treatment with the ATM inhibitor AZD0156 increased PI(4)P-ATM PLA signals. From these authors conclude that "association of ATM and PI(4)P inversely correlated with the need for ATM within the nucleus. Do they imply that treatment with ATM-inhibitors reduces the requirement for ATM function in the nucleus? The interpretation of this result should be further elaborated to sustain this conclusion.

We may have conveyed a wrong notion at this point. We do not imply at all that ATM inhibitors reduce the need for ATM in the nucleus. Instead, we imply that, by reinforcing ATM attachment to Golgi-resident PI(4)P, ATM inhibitors end up titrating ATM away from the nucleus. We will clarify our explanation to avoid misunderstandings.

An increase of GFP-Tel1 foci upon OSH4 overexpression is described on Figure 7B. These are described as nuclear in the results, but no reference is made in the figure or legend as to how nucleus positions are addressed in these experiments. This should be clarified.

We systematically combine the tagging of a nucleoplasmic protein (mCherry-Pus1) with the detection of GFP-Tel1 foci, as to unambiguously assess the nuclear position of Tel1 foci. We will include this explanation and the corresponding mCherry-Pus1 channel to clarify this.

Also, WT controls and quantifications should be included in the experiments shown on Figure 7C.

These experiments are quantified from the moment we did them, although we did not include such quantifications in the present version for the sake of space. We will do so in the revised version.

Reviewer #1 (Significance (Required)):

While the conclusion of lipid metabolism responding to DSBs is not convincing, the observation that Tel1/ATM function is modulated by PI(4)P biding is significant and advances the understanding on the function and regulation of this key kinase in promoting genome integrity maintenance. This is an unanticipated result which is highly novel and has implications for the modulation of Tel1/ATM function through pharmacological manipulation of lipid metabolism. This finding would be of broad interest for scientists working on the response to DNA damage and the maintenance of genome integrity. This reviewer belongs to that group and has limited expertise to evaluate the lipid metabolism genetic manipulation in the manuscript.

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

The authors show that cytoplasmic PI4P have a regulatory role on ATM response to DNA double strand breaks. The process involves a balance between exchange of PI4P between Golgi and ER in exchange of esterified sterols. The study is of interest, however provides indirect evidences to support their conclusions.

Major comments : 1). Since the major conclusion relates to PI4P association with ATM in basal conditions to keep ATM outside nucleus and known presence of PI4P, ATM in other organelles of a cell, further experiments such as cell fractionation experimental that show golgi specific interaction would support the main conclusion.

In continuation of 1st comment, since PI4P in substrate of PI4 phosphoinositol kinases, is there a competition between PI4kinases and ATM for PI4P binding should be addressed through immunoprecipitation studies.

First of all, we need to specify here that PI4kinases will phosphorylated PI4 to create PI(4)P. Thus, PI(4)P is the product, and not the substrate, of PI4kinases. We therefore do not expect any competition between such kinases and ATM.

Second, we take good note of the reviewer’s concern that the pool of PI(4)P at the Golgi may be shared, and also that it would be important to reinforce the notion of the relative subcellular localization of ATM under different treatments. To this end, we will perform the following integrative experiment:

Immunoprecipitation of PI(4)P could theoretically be done using our specific antibody, yet the IP efficiency of a lipid cannot be verified by western blot. Further, there are PI(4)P pools elsewhere in the cell that would mess up with interpretations. We therefore dismiss the use of anti-PI(4)P as a tool to perform immunoprecipitations.

Instead, to explore the impact of PI(4)P levels on ATM both at the Golgi and within the nucleus, we will split our cultures in two to either immunoprecipitate specific cytoplasmic Trans-Golgi Network-associated proteins (we will test separately TGN46 and GOLPH3); or the nuclear ATM-interacting factor MRE11 from nuclei, then blot for co-immunoprecipitated ATM. The relative co-immunoprecipitated ATM is expected to vary under different treatments to which the cells will be exposed, namely:

• untreated
• zeocin, to trigger ATM need in the nucleus
• OSBP inhibition (+/- zeocin), to stabilize PI(4)P at the Golgi
• PIK93, an inhibitor of PI4 kinases that prevents PI(4)P synthesis

2). The authors claim that the ATM retention is the main function of PI4P in Golgi. The authors should rule out the possibility that the phenotype observed on DNA damage response is not due to non availability of PI4P substrate for PI4P kinases, that have recently been shown to participate in genome integrity maintenance.

We want to explain that we do not intend to say that PI(4)P main function at the Golgi is ATM retention, as PI(4)P is a molecule binding and modulating multiple proteins, as for example the aforementioned GOLPH3. We will first revise our text to correct it, in case we have conveyed this incorrect notion, as it stems from the reviewer’s comment.

Second, the reviewer evokes the notion that PI(4)P can be the substrate of a second phosphorylation, which could give rise to PI(3,4)P or to PI(4,5)P, which could still undergo remodeling into PI(3)P, for example. Recent work by Dr Michael Sheetz’s lab demonstrated that this set of phosphoinositides serves to drive the nucleation and activation of the ATR-Chk1 branch of the DNA Damage Response upon genotoxic stress, yet was completely inert with respect to the ATM-Chk2 branch (5). To rule out the possibility, as evoked by the reviewer, that the oleate-induced DDR phenomena we describe relate to these other events, we have now explored the response of the ATR-Chk1 branch when comparing the response of zeocin-treated cells that have been pre-loaded or not with oleate. We observe that the ATR-Chk1 branch is unaltered by oleate loading. Thus, we can now propose that the PI(4)P branch exclusively modulates the ATM-Chk2 axis.

3). Does Oleate treatment influences Rad53 protein levels in addition to its phosphorylation that affect DNA damage response may be addressed.

Exponential cultures from three different WT, three different ste∆ and three different yeh2∆ strains have now been taken and pre-loaded for 2 hours with 0.05% oleate, then total levels of Rad53 (without induction of DNA damage) assessed. We can now formally say that basal levels of Rad53 protein are not altered by this incubation. We will include this control in the revised manuscript.

4). Does Yeh2 deletion reduces LDS should be checked.

We frequently use yeh2∆ cells in our studies. In particular, we have recently published work characterizing the phenotype of this strain with respect to the formation of lipid droplets in the nucleus (6). We are currently exploiting those same sets of data to quantify the total number of LD in order to satisfy the reviewer’s concern.

5). Figure 4D representation should show % of phospho reduction of initial activation and a better western blot image should be shown that show equal loading of samples.

We are currently repeating these gels and blots for the sake of clarity, as requested.

6). In immunoprecipitation experiments, kindly include isotypee IgG controls as well to rule out non-specificity.

Of course, this important control will be included every time.

Minor points: 1). Figure S1F do not show oleate treatment as presented in results section.

We will revise the accurate naming.

2). A better gel for S4B should be presented with ponceau of the same gel.

We are currently repeating this gel and associated blot for the sake of clarity, as requested.

3). Nuclear PI4Ps has also been previously reported, an explanation to the specific interaction of ATM and PI4P in the Golgi should be addressed/discussed.

We take it that the reviewer is referring here to the recent work by Fáberová et al (7) in which PI(4)P and PI(4,5)P were described as very dynamic in the nucleus, and mostly related then to mRNA transcription, splicing and export. We will reinforce the connection of our phenomenon to the Golgi-associated pool of PI(4)P thanks to the co-immunoprecipitation experiments proposed above, and will timely contextualize these in light of the paper by Fáberová and co-workers in the revised version. Thank you for reminding us of this work.

Reviewer #2 (Significance (Required)):

The current work definitely adds a layer in our understanding to ATM regulation and cross-talk between different PIKK family of kinases. ATM localisation in extra nuclear regions of a cell has been described earlier with significant impact on cell physiology such as mitochondria etc., ATM retention at golgi and limiting nuclear ATM levels is significant advance at ATM activity regulation, while signifying non canonical function of PI4P.

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

Summary:

In this manuscript, the authors propose that ATM/Tel1 signaling is regulated in a spatiotemporal manner during genotoxic stress both in yeast and mammalian cells. They show that Lipid droplets accumulate in response to genotoxic stress. As a consequence, there is a decrease of exchange of PI4P from the Golgi to ER, thus dampening ATM/Tel1 signaling by sequestering this kinase into the Golgi. The authors combined findings in yeast and mammals showing that this mechanism is conserved throughout eukaryotes. For this purpose, they use a vast number of techniques that support their proposed model.

Major comments:

The conclusions were made based on evidence combining yeast genetics, immunofluorescence, DNA end resection analysis and pharmacological interventions. The hypothesis that ATM is kept away from the nucleus by physically interacting with PI4P at the Golgi, thus allowing processive repair is bold and contributes for a better understanding of the choreography of the DDR kinases during DSB repair. However, many of the experiments in yeast and mammals show only mild phenotypes and there is no evidence that this mode of ATM dampening impact cell viability in mammals.

We agree with the reviewer that the effects associated to the reported phenomenon are indeed mild. This is a fact. We would like to remind that the metabolism of sterols is finely controlled, and at many different levels, in a very complex manner. For example, sterol increases in the cell will immediately be compensated by reduced synthesis, while synthesis inhibition will immediately promote uptake from the medium, and/or release from stores (for example, see (8)). As a natural consequence, the window of manipulation and, more importantly, the strength of the phenotypes we can uncover are small.

Therefore, I have some comments and suggestions of experiments that I think could improve the quality of the manuscript. I believe that most of these new experiments does not require much time and resources.

• Does oleate treatment in RPE-1/Huh-7 cells induce loss of viability? An experiment showing loss of viability like MT-assay or decreased cell proliferation would reinforce the importance of the mechanism proposed.

This experiment was already included in the previous version, yet it may have escaped the attention of the reviewer. We show in Figure S2E that oleate treatment restricts viability in Huh-7 cells alone, and also worsens their tolerance to zeocin. Perhaps we should reconsider moving this result to the main figures so that it does not go unnoticed.

• In yeast there is evidence that a ste delta strain show sensitivity to zeocin/CPT, but there is no experiment showing the same effect on cells lacking Yeh2. Since both strains share similar phenotypes, it would be interesting to show that increased kinetics of Rad53 signaling leads to sensitivity to genotoxins.

We have now performed this experiment, we will include the matching information for yeh2∆ cells, which agrees with the predictions.

• The conclusion that ste delta cells exposed to zeocin leads to unproductive events due to defects in DNA-end resection could be reinforced by a decrease in Rad52 foci. It has been previously shown by the group of Dr. Marcus Smolka, that inhibition of DNA-end resection decreases Rad52 foci (https://doi.org/10.1083/jcb.201607031). Since the authors were able to monitor Rad52-YFP (Figure S1A), it shouldn't consume time and resources.

The reviewer is right that this experiment should not be time- or resources-consuming. We will evaluate the accumulation of Rad52 foci in response to the concerned genotoxin in ste∆ cells.

• Since the authors propose that there is a DNA repair defect due to inhibition of long-range DNA-end resection, it would be important to monitor gamma-H2A(X) signal either in yeast or mammals.

Taking into consideration the reviewer’s suggestion, we have now performed anti-yH2AX immunofluorescence of all the implied conditions (genotoxins +/- oleate pre-load) and will quantify them to answer the concern.

• How do the authors exclude the possibility that yeast mutants or oleate treatment in yeast/mammalian cells change membrane permeability allowing an increase in genotoxin concentration?

Although this is a very reasonable criticism, we want to remind the data we present in Figure S4A in which we use the naturally DSB-bearing rad3-102 cells for recombination analyses, showing that, in the absence of any genotoxin, the same phenotype also applies. Yet, we want to reinforce the notion that LD formation in response to DSB can also occur when the breaks are not chemically, but physically, induced. To this end, and also to match a related request by Reviewer 1, we will:

1. a) exploit the naturally DSB-accumulating mutant rad3-102 (4) to evaluate whether it endogenously harbors more LD in comparison with the WT.
2. b) we have recently created a tool in which gRNAs targeted to different subsets of transposons in the genome can drive Cas9 to create DSB in a dose-dependent manner ((9), under revision in Genetics). We will use this system to monitor the LD formation in response to Cas9-triggered cuts. In addition, on figure 5A, significant differences in GFP-Tel1 foci abundance between WT and steD or yeh2D cells are only observed after 210', way after the slight effect on Rad53 phosphorylation is observed. This is at odds with the conclusion that Tel1 association to STEs modulates DDR signaling.

We are afraid that we have not been clear enough in explaining the kinetics giving rise to our model. As indicated by the reviewer, our work shows, through kinetic studies, that the storage of sterols within LD occurs at later stages than the activation of the DDR by Tel1 and Rad53 phosphorylation. Tel1 foci decline is necessary for subsequent engagement of downstream DNA long-range resection. Since we propose that sterol storage within LD is a means to attenuate Tel1 engagement at DSBs, it is thus logical (and thus compatible with the data we show) that LD number increase occurs simultaneously with Tel1 foci decrease, at late stages of the reactionWe will include this explanation and graph in the revised version of the work.

• It would be interesting to investigate genetic interactions between ste delta (or yeh2delta) and yeast mutants with DNA-end resection problems (exo1delta; sae2delta). For instance, it has been shown that Sae2 antagonizes checkpoint signaling by competing with Rad9 to DSB sites (https://doi.org/10.1073/pnas.1816539115). Also, cells lacking Sae2 show an increase in Rad53 signaling due to increased Tel1 Signaling. Therefore, an epistatic effect between these two pathways would reinforce the hypothesis of the manuscript.

we will build the double mutant sae2∆ yeh2∆ and assess the potential epistatic behavior they may display with respect to some key phenotypes (Tel1 foci formation, Rad53 phosphorylation…).

• The authors showed that Tel1-GFP does not accumulate in the nucleus in cells lacking Sac1 (Figure 7C). Tel1 is important to cope with increased DSBs in the absence of Mec1, thus avoiding genomic instability. Cells lacking both Mec1 and Tel1 show a sick phenotype with an exponential increase in gross chromosomal rearrangements and sensitivity to genotoxins. Therefore, does deletion of Mec1 (and Sml1) in sac1 delta phenocopies a mec1tel1 delta? Alternatively, does pharmacological inhibition of ATR in the presence of the OSBP1 inhibitor causes loss of viability or chromosomal aberrations?

We will delete SAC1 in mec1∆ sml1∆ and compare the fitness, through growth drop assays, with respect to the mutant tel1∆ mec1∆ sml1∆.

We will expose cells either to OSBP1 inhibitor, ATR inhibitor, or both, and assess the phosphorylation of their downstream common effector H2AX. Additionally, we will assess the effect on cell growth of the combination of ATRi and OSBP1i using synergy matrices. We will determine if the combination of both drugs synergizes or not to impair cell proliferation and reduce cell viability.

• Finally, it seems strange to me that ATR/Mec1 signaling is not mentioned throughout the entire manuscript. Does PI4P pathway affect only ATM/Tel1? In Figure 2D, an antibody against phospho-CHK1 could be used to monitor ATR signaling. In line with that, I would like to see in the discussion how these new findings are in line with evidence from a 2019 paper showing that phophoinositides PIP2 and PIP3, but not PI4P are important for ATR signaling (DOI: 10.1038/s41467-017-01805-9). They showed that a nuclear pool of PIP2 increases upon DNA damage induction and rapidly accumulates at DNA lesions. This event is important for the recruitment of ATR. Since PI4P is substrate for PIP2 synthesis and there is a nuclear pool of PI4P and PIP2, I think it is important to discuss if the results presented here are in line with these previous findings.

The reviewer evokes recent work by Dr Michael Sheetz’s lab demonstrating that a different set of phosphoinositides serves to drive the nucleation and activation of the ATR-Chk1 branch of the DNA Damage Response upon genotoxic stress, yet was completely inert with respect to the ATM-Chk2 branch (5). We have now explored, also to satisfy a similar concerned raised by Reviewer 2, the response of the ATR-Chk1 branch when comparing the response of zeocin-treated cells that have been pre-loaded or not with oleate. We observe that the ATR-Chk1 branch is unaltered by oleate loading. Thus, we can now propose that the PI(4)P branch exclusively modulates the ATM-Chk2 axis.

We will of course give the needed credit to this work and contextualize our findings accordingly.

Minor comments:

• Line 124: The correct is Figure S1E, lower panel and not Figure S1F -Lines 127-128: Figure S2A does not show zeocin treatment

Both minor mistakes will be corrected.

Reviewer #3 (Significance (Required)):

Together, these new findings, if corroborated by others, might be important to open new lines of investigation in basic and translational research regarding human diseases as explored in the discussion section. I believe this paper will attract attention not only from the DDR field but also from other areas of research such as nutrient and lipid signaling both in yeast and mammals. I hope I was able to collaborate in this review, since my main expertise is in the area of DNA damage signaling using budding yeast as an organism model.

Reviewer #4 (Evidence, reproducibility and clarity (Required)):

This is a very interesting study where Sara et al. demonstrated a link between lipid metabolism with DNA repair response (DDR). In this study, they have proposed ATM as a novel PI4P-effector. The sterol deposition into lipid droplets impacts the Golgi PI4P level due to lipid exchange machinery facilitated by OSBP1, therefore regulating the cytosolic retention of ATM due to PI4P binding. Although how lipid droplets in the cytosol sense the DNA damage and control the initiation of DDR by regulating ATM is still unclear, this study linked lipid biology/PI signaling to DNA damage repair and showed the evolutionary conservation of PI signaling and DNA repair machinery from yeast to humans. The experiments are well designed, nicely controlled, with a high quality of data presentation. With some improvements, this work could be a very interesting study attracting a broad readership.

In their model, ATM is PI4P-bound and sequestered inside the cytosol under basal conditions. Upon genotoxic stress, activation of OSBP1 removes PI4P and free PI4P-bound ATM for nuclear translocation of DNA repair. This could also be interpreted as genotoxic stress-induced PIP-kinase activity, where PI4P is processed into PIP2 or PIP3, somehow redirecting ATM into the nucleus to initiate its activation for DDR. Those aspects should be discussed and improved.

Both Reviewers 2 and 3 have somehow evoked a similar concern. More precisely, the work by Dr Michael Sheetz’s lab demonstrating that a different set of phosphoinositides serves to drive the nucleation and activation of the ATR-Chk1 branch of the DNA Damage Response upon genotoxic stress, yet was completely inert with respect to the ATM-Chk2 branch (5). We have now explored, to satisfy all reviewers’ concerns, the response of the ATR-Chk1 branch when comparing the response of zeocin-treated cells that have been pre-loaded or not with oleate. We observe that the ATR-Chk1 branch is unaltered by oleate loading. Thus, we can now propose that the PI(4)P branch exclusively modulates the ATM-Chk2 axis.

Additionally, we will of course give the needed credit to this work and contextualize our findings accordingly.

Upon stress, there is nuclear activation of p53-phosphoinositide (PI) signalosomes and PIP-kinases. Also, there is a significant PIP2 pool inside the nucleus with an involvement in DNA damage repair. Those papers and their relevance to the current study need to be discussed. If ATM is a novel PI4P-effector, there is also nuclear PI4P formation or nuclear PI4P accumulation upon stresses based on recent studies; how the ATM interacts with PIPn in the nucleus upon translocation? A know ATM substrate p53 is PIP2/PIP3 bound in the nucleus based on recent studies. Will ATM prefer to interact with other PIPn-bound proteins in the nucleus or PIPn regulate their interaction needs to be discussed.

These additional notions are in line with the previous paragraph presented by the reviewer, and our answers too. We will provide a constructive overview of all these ideas in the revised version of the manuscript.

Major points: 1. The PI4P-ATM complex is supported only by PLA and PIP strips. Need more robust biochemical characterization of the interaction: co-IP, lipid binding, and/or in vitro constitution.

We agree with the need to perform assays in which PI(4)P is embedded in a bilayer, as to confidently assess whether Tel1 can bind it in that context. We have now performed a pilot experiment in which we have confronted purified FLAG-Tel1 to liposomes harboring PI(4)P. Western blot analysis using anti-FLAG antibody shows the encouraging result that FLAG-Tel1 can be found there. As a control, we have performed the same process but in the absence of any liposomes. We observe that a residual fraction of FLAG-Tel1 can nevertheless be found in this control, most probably because the buffer used during the liposome assay makes part of FLAG-Tel1 precipitate.To avoid this type of background and to increase our trust in the results, we propose to perform the liposome assay but on a discontinuous density gradient, so that liposomes will be retrieved in the top layer (and bound FLAG-Tel1 with them, if that is the case), while any precipitated FLAG-Tel1 will be in the bottom fraction (liposome floatation assay). As a further control, we will include the same liposomes but lacking PI(4)P. We expect to be successful in the floatation assays. If we are not, we will repeat the experiment presented above to be confident that the observed increase is reproducible.

1. The use of drug inhibitors only in the final figure is problematic. KD or KO experiments should be performed to confirm that ATM and the exchanger are the relevant targets.

We have now used siRNAs against the exchanger protein, OSBP1, with a very high silencing rate success. We have next monitored the activation status of the chromatin-associated ATM target KAP1, in order to monitor the predicted decrease of ATM activity specifically inside the nucleus. Our results confirm the role of OSBP1, by KD experiments as requested by the reviewer, in attenuating ATM nuclear participation.

1. Poor quality of some WBs (e.g Fig. S1F).

We have now repeated the Western Blot to detect Rad53-P in response to 20 mM HU in WT versus ste∆cells.

1. Lack of statistical analyses for some data (e.g. Fig. 1B-E)

We had already included, in the previous version, the complete statistical analyses corresponding to Figures 1B to E and evoked here by the reviewer. They were indeed included in Figure S1C, and our brief reference to them in the text may have escaped her/his attention. We will make a clear reference to this in the revised version.

Additional clarification points:

Figure 1: No representative images were shown for quantifications in Figure 1C, D, E.

If the reviewer / editor estimates it pertinent, we can of course include them. Yet, they will be very redundant with the images displayed in Figure 1A.

Line 121: Should be Figure S1E, upper panel. Line 124: Should be Figure S1E, lower panel. Figure 2D-E, please show the quantification of the ratio of pCHK2/CHK2 with an N=3

We will correct / include the requested changes.

Figure S2B: needs quantification of NileRed staining to conclude induction in LD formation

We will quantify the LD as requested.

Figure 3C, to show the selectivity of ATM-binding toward PI4P, PLA of ATM with other PIPn species should be assessed, such as PI3P, PI4,5P2, and PI3,4,5P3.

We have provided an overview of the binding preferences of ATM with respect to the full battery of phosphoinositides in the strip-binding assay shown in Figures S5C and 6B. Other than that, we are afraid that PLA studies as the ones we develop in the current manuscript for PI(4)P are not feasible, since no reliable antibodies exist for most of the phosphoinositide species evoked by the reviewer.

Figure S6A, PI4P level could be assessed by IF staining using PI4P antibody besides using PI4P sensor.

We will use our PI(4)P antibody to monitor by immunofluorescence the behavior of this molecule in response to either MMS or zeocin, as suggested.

References

1. Cheung HC, San Lucas FA, Hicks S, Chang K, Bertuch AA, Ribes-Zamora A. An S/T-Q cluster domain census unveils new putative targets under Tel1/Mec1 control. BMC Genomics. 2012;
2. Bensimon A, Schmidt A, Ziv Y, Elkon R, Wang SY, Chen DJ, et al. ATM-dependent and -independent dynamics of the nuclear phosphoproteome after DNA damage. Sci Signal. 2010;
3. BastosdeOliveira FM, Kim D, Cussiol JR, Das J, Jeong MC, Doerfler L, et al. Phosphoproteomics Reveals Distinct Modes of Mec1/ATR Signaling during DNA Replication. Mol Cell. 2015;
4. Moriel-Carretero M, Aguilera A. A Postincision-Deficient TFIIH Causes Replication Fork Breakage and Uncovers Alternative Rad51- or Pol32-Mediated Restart Mechanisms. Mol Cell. 2010;37(5):690–701.
5. Wang YH, Hariharan A, Bastianello G, Toyama Y, Shivashankar G V., Foiani M, et al. DNA damage causes rapid accumulation of phosphoinositides for ATR signaling. Nat Commun. 2017;
6. Kumanski S, Forey R, Cazevieille C, Moriel-Carretero M. Nuclear Lipid Droplet Birth during Replicative Stress. Cells. 2022;11(1390).
7. Fáberová V, Kalasová I, Krausová A, Hozák P. Super-Resolution Localisation of Nuclear PI(4)P and Identification of Its Interacting Proteome. Cells. 2020;9(5):1–17.
8. Luo J, Yang H, Song BL. Mechanisms and regulation of cholesterol homeostasis. Nat Rev Mol Cell Biol [Internet]. 2020;21(4):225–45. Available from: http://dx.doi.org/10.1038/s41580-019-0190-7
9. Coiffard J, Santt O, Kumanski S, Pardo B, Moriel-Carretero M. A CRISPR-Cas9-based system for the dose-dependent study of 4 DNA double strand breaks sensing and repair 5 6. bioRxiv [Internet]. 2021;1–37. Available from: https://doi.org/10.1101/2021.10.21.465387.

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 #4

Evidence, reproducibility and clarity

This is a very interesting study where Sara et al. demonstrated a link between lipid metabolism with DNA repair response (DDR). In this study, they have proposed ATM as a novel PI4P-effector. The sterol deposition into lipid droplets impacts the Golgi PI4P level due to lipid exchange machinery facilitated by OSBP1, therefore regulating the cytosolic retention of ATM due to PI4P binding. Although how lipid droplets in the cytosol sense the DNA damage and control the initiation of DDR by regulating ATM is still unclear, this study linked lipid biology/PI signaling to DNA damage repair and showed the evolutionary conservation of PI signaling and DNA repair machinery from yeast to humans. The experiments are well designed, nicely controlled, with a high quality of data presentation. With some improvements, this work could be a very interesting study attracting a broad readership.

In their model, ATM is PI4P-bound and sequestered inside the cytosol under basal conditions. Upon genotoxic stress, activation of OSBP1 removes PI4P and free PI4P-bound ATM for nuclear translocation of DNA repair. This could also be interpreted as genotoxic stress-induced PIP-kinase activity, where PI4P is processed into PIP2 or PIP3, somehow redirecting ATM into the nucleus to initiate its activation for DDR. Those aspects should be discussed and improved.

Upon stress, there is nuclear activation of p53-phosphoinositide (PI) signalosomes and PIP-kinases. Also, there is a significant PIP2 pool inside the nucleus with an involvement in DNA damage repair. Those papers and their relevance to the current study need to be discussed. If ATM is a novel PI4P-effector, there is also nuclear PI4P formation or nuclear PI4P accumulation upon stresses based on recent studies; how the ATM interacts with PIPn in the nucleus upon translocation? A know ATM substrate p53 is PIP2/PIP3 bound in the nucleus based on recent studies. Will ATM prefer to interact with other PIPn-bound proteins in the nucleus or PIPn regulate their interaction needs to be discussed.

Major points:

1. The PI4P-ATM complex is supported only by PLA and PIP strips. Need more robust biochemical characterization of the interaction: co-IP, lipid binding, and/or in vitro constitution.
2. The use of drug inhibitors only in the final figure is problematic. KD or KO experiments should be performed to confirm that ATM and the exchanger are the relevant targets.
3. Poor quality of some WBs (e.g Fig. S1F).
4. Lack of statistical analyses for some data (e.g. Fig. 1B-E)

Additional clarification points:

• Figure 1: No representative images were shown for quantifications in Figure 1C, D, E.

• Line 121: Should be Figure S1E, upper panel.

• Line 124: Should be Figure S1E, lower panel.

• Figure 2D-E, please show the quantification of the ratio of pCHK2/CHK2 with an N=3

• Figure S2B: needs quantification of NileRed staining to conclude induction in LD formation.

• Figure 3C, to show the selectivity of ATM-binding toward PI4P, PLA of ATM with other PIPn species should be assessed, such as PI3P, PI4,5P2, and PI3,4,5P3.

• Figure S6A, PI4P level could be assessed by IF staining using PI4P antibody besides using PI4P sensor.

Significance

This is a very interesting study where Sara et al. demonstrated a link between lipid metabolism with DNA repair response (DDR). In this study, they have proposed ATM as a novel PI4P-effector. The sterol deposition into lipid droplets impacts the Golgi PI4P level due to lipid exchange machinery facilitated by OSBP1, therefore regulating the cytosolic retention of ATM due to PI4P binding. Although how lipid droplets in the cytosol sense the DNA damage and control the initiation of DDR by regulating ATM is still unclear, this study linked lipid biology/PI signaling to DNA damage repair and showed the evolutionary conservation of PI signaling and DNA repair machinery from yeast to humans. The experiments are well designed, nicely controlled, with a high quality of data presentation. With some improvements, this work could be a very interesting study attracting a broad readership.

In their model, ATM is PI4P-bound and sequestered inside the cytosol under basal conditions. Upon genotoxic stress, activation of OSBP1 removes PI4P and free PI4P-bound ATM for nuclear translocation of DNA repair. This could also be interpreted as genotoxic stress-induced PIP-kinase activity, where PI4P is processed into PIP2 or PIP3, somehow redirecting ATM into the nucleus to initiate its activation for DDR. Those aspects should be discussed and improved.

Upon stress, there is nuclear activation of p53-phosphoinositide (PI) signalosomes and PIP-kinases. Also, there is a significant PIP2 pool inside the nucleus with an involvement in DNA damage repair. Those papers and their relevance to the current study need to be discussed. If ATM is a novel PI4P-effector, there is also nuclear PI4P formation or nuclear PI4P accumulation upon stresses based on recent studies; how the ATM interacts with PIPn in the nucleus upon translocation? A know ATM substrate p53 is PIP2/PIP3 bound in the nucleus based on recent studies. Will ATM prefer to interact with other PIPn-bound proteins in the nucleus or PIPn regulate their interaction needs to be discussed.

Major points:

1. The PI4P-ATM complex is supported only by PLA and PIP strips. Need more robust biochemical characterization of the interaction: co-IP, lipid binding, and/or in vitro constitution.
2. The use of drug inhibitors only in the final figure is problematic. KD or KO experiments should be performed to confirm that ATM and the exchanger are the relevant targets.
3. Poor quality of some WBs (e.g Fig. S1F).
4. Lack of statistical analyses for some data (e.g. Fig. 1B-E)

Additional clarification points:

• Figure 1: No representative images were shown for quantifications in Figure 1C, D, E.

• Line 121: Should be Figure S1E, upper panel.

• Line 124: Should be Figure S1E, lower panel.

• Figure 2D-E, please show the quantification of the ratio of pCHK2/CHK2 with an N=3

• Figure S2B: needs quantification of NileRed staining to conclude induction in LD formation.

• Figure 3C, to show the selectivity of ATM-binding toward PI4P, PLA of ATM with other PIPn species should be assessed, such as PI3P, PI4,5P2, and PI3,4,5P3.

• Figure S6A, PI4P level could be assessed by IF staining using PI4P antibody besides using PI4P sensor.

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, the authors propose that ATM/Tel1 signaling is regulated in a spatiotemporal manner during genotoxic stress both in yeast and mammalian cells. They show that Lipid droplets accumulate in response to genotoxic stress. As a consequence, there is a decrease of exchange of PI4P from the Golgi to ER, thus dampening ATM/Tel1 signaling by sequestering this kinase into the Golgi. The authors combined findings in yeast and mammals showing that this mechanism is conserved throughout eukaryotes. For this purpose, they use a vast number of techniques that support their proposed model.

Major comments:

The conclusions were made based on evidence combining yeast genetics, immunofluorescence, DNA end resection analysis and pharmacological interventions. The hypothesis that ATM is kept away from the nucleus by physically interacting with PI4P at the Golgi, thus allowing processive repair is bold and contributes for a better understanding of the choreography of the DDR kinases during DSB repair. However, many of the experiments in yeast and mammals show only mild phenotypes and there is no evidence that this mode of ATM dampening impact cell viability in mammals. Therefore, I have some comments and suggestions of experiments that I think could improve the quality of the manuscript. I believe that most of these new experiments does not require much time and resources.

• Does oleate treatment in RPE-1/Huh-7 cells induce loss of viability? An experiment showing loss of viability like MT-assay or decreased cell proliferation would reinforce the importance of the mechanism proposed.
• In yeast there is evidence that a ste delta strain show sensitivity to zeocin/CPT, but there is no experiment showing the same effect on cells lacking Yeh2. Since both strains share similar phenotypes, it would be interesting to show that increased kinetics of Rad53 signaling leads to sensitivity to genotoxins.
• The conclusion that ste delta cells exposed to zeocin leads to unproductive events due to defects in DNA-end resection could be reinforced by a decrease in Rad52 foci. It has been previously shown by the group of Dr. Marcus Smolka, that inhibition of DNA-end resection decreases Rad52 foci (https://doi.org/10.1083/jcb.201607031). Since the authors were able to monitor Rad52-YFP (Figure S1A), it shouldn't consume time and resources.
• Since the authors propose that there is a DNA repair defect due to inhibition of long-range DNA-end resection, it would be important to monitor gamma-H2A(X) signal either in yeast or mammals.
• How do the authors exclude the possibility that yeast mutants or oleate treatment in yeast/mammalian cells change membrane permeability allowing an increase in genotoxin concentration?
• It would be interesting to investigate genetic interactions between ste delta (or yeh2delta) and yeast mutants with DNA-end resection problems (exo1delta; sae2delta). For instance, it has been shown that Sae2 antagonizes checkpoint signaling by competing with Rad9 to DSB sites (https://doi.org/10.1073/pnas.1816539115). Also, cells lacking Sae2 show an increase in Rad53 signaling due to increased Tel1 Signaling. Therefore, an epistatic effect between these two pathways would reinforce the hypothesis of the manuscript.
• The authors showed that Tel1-GFP does not accumulate in the nucleus in cells lacking Sac1 (Figure 7C). Tel1 is important to cope with increased DSBs in the absence of Mec1, thus avoiding genomic instability. Cells lacking both Mec1 and Tel1 show a sick phenotype with an exponential increase in gross chromosomal rearrangements and sensitivity to genotoxins. Therefore, does deletion of Mec1 (and Sml1) in sac1 delta phenocopies a mec1tel1 delta? Alternatively, does pharmacological inhibition of ATR in the presence of the OSBP1 inhibitor causes loss of viability or chromosomal aberrations?
• Finally, it seems strange to me that ATR/Mec1 signaling is not mentioned throughout the entire manuscript. Does PI4P pathway affect only ATM/Tel1? In Figure 2D, an antibody against phospho-CHK1 could be used to monitor ATR signaling. In line with that, I would like to see in the discussion how these new findings are in line with evidence from a 2019 paper showing that phophoinositides PIP2 and PIP3, but not PI4P are important for ATR signaling (DOI: 10.1038/s41467-017-01805-9). They showed that a nuclear pool of PIP2 increases upon DNA damage induction and rapidly accumulates at DNA lesions. This event is important for the recruitment of ATR. Since PI4P is substrate for PIP2 synthesis and there is a nuclear pool of PI4P and PIP2, I think it is important to discuss if the results presented here are in line with these previous findings.

Minor comments:

• Line 124: The correct is Figure S1E, lower panel and not Figure S1F
• Lines 127-128: Figure S2A does not show zeocin treatment

Significance

Together, these new findings, if corroborated by others, might be important to open new lines of investigation in basic and translational research regarding human diseases as explored in the discussion section. I believe this paper will attract attention not only from the DDR field but also from other areas of research such as nutrient and lipid signaling both in yeast and mammals. I hope I was able to collaborate in this review, since my main expertise is in the area of DNA damage signaling using budding yeast as an organism model.

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 show that cytoplasmic PI4P have a regulatory role on ATM response to DNA double strand breaks. The process involves a balance between exchange of PI4P between Golgi and ER in exchange of esterified sterols. The study is of interest, however provides indirect evidences to support their conclusions.

Major comments:

1). Since the major conclusion relates to PI4P association with ATM in basal conditions to keep ATM outside nucleus and known presence of PI4P, ATM in other organelles of a cell, further experiments such as cell fractionation experimental that show golgi specific interaction would support the main conclusion.

2). In continuation of 1st comment, since PI4P in substrate of PI4 phosphoinositol kinases, is there a competition between PI4kinases and ATM for PI4P binding should be addressed through immunoprecipitation studies.

3). The authors claim that the ATM retention is the main function of PI4P in Golgi. The authors should rule out the possibility that the phenotype observed on DNA damage response is not due to non availability of PI4P substrate for PI4P kinases, that have recently been shown to participate in genome integrity maintenance.

4). Does Oleate treatment influences Rad53 protein levels in addition to its phosphorylation that affect DNA damage response may be addressed.

5). Does Yeh2 deletion reduces LDS should be checked.

6). Figure 4D representation should show % of phospho reduction of initial activation and a bettier western blot image should be shown that show equal loading of samples.

7). In ammunoprecipitation experiments, kindly include isotypee IgG controls as well to rule out non-specificity.

Minor points:

1). Figure S1F do not show oleate treatment as presented in results section.

2). A better gel for S4B should be presented with ponceau of the same gel.

3). Nuclear PI4Ps has also been previously reported, an explanation to the specific interaction of ATM and PI4P in the Golgi should be addressed/discussed.

Significance

The current work definitely adds a layer in our understanding to ATM regulation and cross-talk between different PIKK family of kinases. ATM localisation in extra nuclear regions of a cell has been described earlier with significant impact on cell physiology such as mitochondria etc., ATM retention at golgi and limiting nuclear ATM levels is significant advance at ATM activity regulation, while signifying non canonical function of PI4P.

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

Ovejero et al. report an increase in lipid droplet (LD) abundance after long (from 120' on) exposure of budding yeast cells to DNA damaging agents zeocin and camptothecin (CPT). Next, they analyze DNA damage signaling in yeast mutants that impair triacylglycerol (TAGs) or sterol (STEs) esterification. They observe a slight anticipation in Rad53/CHK2 phosphorylation (indicative of DDR signaling) in yeast stem mutants, as well as in yeast cells or human cells lines pre-treated with oleate upon zeocin treatment. Yeast stem mutants are sensitive to zeocin and captothecin, but only confer sensitivity to hydroxyurea upon combination with tagD mutations. Authors relate these phenotypes to a somewhat decreases DSB resection in yeh2D mutants (expected to have reduced steryl esters pools) and RPA-foci in steD yeast cells. Next, a reduction in single strand annealing recombination repair events upon zeocin treatment is reported using a genetic reporter in steD mutants and oleate-treated cells. From these data they conclude that inability to process sterols in response to DSBs leads to an exacerbated DDR and prevents DNA repair. Next, it is shown that Flag-tagged Tel1 distinctly interacts with mono-phosphate phosphoinositides, including PI(4)P. An interaction in vivo is also inferred through Proximity Ligation Assays (PLA) using anti-PI(4)P and anti-ATM antibodies in human cell lines, which was moderately downregulated upon treatment with MMS or zeocin. Over-expression of the Osh4/OSBP1 transporter, which consumes PI(4)P, increased the number of Tel1 (nuclear) foci upon zeocin treatment. Conversely Sac1 ablation, in which accumulation of PI(4)P is expected, abrogated nuclear Tel1 foci formation and reduced telomere length (a phenotype related to lack of Tel1 function). From these results authors conclude that Tel1 availability in the nucleus is influenced by PI(4)P availability. Lastly, treatment with an OSBP1 inhibitor led to a cell line and damaging agent -variable reduction of ATM phosphorylation and a mostly non-significant reduction of DNA resection, measured by native BrdU detection, in response to CPT treatment. Overall, authors conclude that i) biding of Tel1/ATM to PI(4)P modulates its functional availability in the nucleus, and that ii) DNA damage elicits the esterification and storage of sterols toward LDs, which contributes to tritate Tel1/ATM away from the nucleus dampening the DDR and affecting long-range resection.

Minor comments:

• Figure S1D and E, experiments should be carried out to include time points in which LD accumulation and cell cycle arrest are observed upon zeocin treatment (i.e., up to 210' as in Figure 1A)

• How do authors explain increased single strand annealing recombination frequencies in steD and oleate-treated wild type cells (Figure 4A). Should it not be expected that increased STEs also impair recombination induced by endogenous damage?

• Data presented in figure 4B and 4C are not fully convincing. Performing time course experiments might help concluding if the differences observed represent a relevant defect in DSB processing.

• Is Figure 5B referring to Flag-tagged Tel1 or GFP-tagged Tel1 as stated in the figure legend?

• Treatment with the ATM inhibitor AZD0156 increased PI(4)P-ATM PLA signals. From these authors conclude that "association of ATM and PI(4)P inversely correlated with the need for ATM within the nucleus. Do they imply that treatment with ATM-inhibitors reduces the requirement for ATM function in the nucleus? The interpretation of this result should be further elaborated to sustain this conclusion.

• An increase of GFP-Tel1 foci upon OSH4 overexpression is described on Figure 7B. These are described as nuclear in the results, but no reference is made in the figure or legend as to how nucleus positions are addressed in these experiments. This should be clarified. Also, WT controls and quantifications should be included in the experiments shown on Figure 7C.

Major comments:

• While the conclusion that Tel1/ATM binds PI(4)P and this interaction modulates Tel1/ATM functional availability at the nucleus is convincing, the conclusion that DSBs elicit a change in the metabolism of this lipid to "control" Tel1/ATM function is not demonstrated.

• The notion that sterol processing occurs in response to DSBs is not sufficiently supported by the data presented, as the increase in LD numbers is observed much after activation of the DDR (Rad53 phosphorylation) in Zeozin-treated yeast cells. In addition, evidence is not provided on the mechanisms by which PI(4)P metabolism would be controlled, which would be expected to be DDR-independent as they are placed upstream of this signaling pathway in the author's model. The damaging agents used have been suggested to alter the redox metabolism and even lipid peroxidation (Kitanovic 2009, Mizumoto 1993, Krol 2015, Todorova 2015, Ren 2019, Singh 2014). Hence it is possible that PI(4)P changes are not due to DSBs, but an indirect though relevant effect.

• In absence of direct evidence supporting an active regulation of PI(4)P dynamics in response to DNA breaks, this conclusion remains speculative and this should be noted in the manuscript.

• Authors conclude that LD is specific to DSB induction. This seems an overstatement as they just reported LD increases in response to two agents that also induce other kinds of DNA damage. To also strengthen the link between DSBs and PI(4)P modulation of Tel1 function, authors should analyze LD numbers, Rad53 phosphorylation and Tel1 nuclear re-localization in response to HO-induced DNA breaks (e.g., using the system employed in Figure 3C)

• In addition, on figure 5A, significant differences in GFP-Tel1 foci abundance between WT and steD or yeh2D cells are only observed after 210', way after the slight effect on Rad53 phosphorylation is observed. This is at odds with the conclusion that Tel1 association to STEs modulates DDR signaling.

Minor comments:

• Figure S1D and E, experiments should be carried out to include time points in which LD accumulation and cell cycle arrest are observed upon zeocin treatment (i.e., up to 210' as in Figure 1A)

• How do authors explain increased single strand annealing recombination frequencies in steD and oleate-treated wild type cells (Figure 4A). Should it not be expected that increased STEs also impair recombination induced by endogenous damage?

• Data presented in figure 4B and 4C are not fully convincing. Performing time course experiments might help concluding if the differences observed represent a relevant defect in DSB processing.

• Is Figure 5B referring to Flag-tagged Tel1 or GFP-tagged Tel1 as stated in the figure legend?

• Treatment with the ATM inhibitor AZD0156 increased PI(4)P-ATM PLA signals. From these authors conclude that "association of ATM and PI(4)P inversely correlated with the need for ATM within the nucleus. Do they imply that treatment with ATM-inhibitors reduces the requirement for ATM function in the nucleus? The interpretation of this result should be further elaborated to sustain this conclusion.

• An increase of GFP-Tel1 foci upon OSH4 overexpression is described on Figure 7B. These are described as nuclear in the results, but no reference is made in the figure or legend as to how nucleus positions are addressed in these experiments. This should be clarified. Also, WT controls and quantifications should be included in the experiments shown on Figure 7C.

Significance

While the conclusion of lipid metabolism responding to DSBs is not convincing, the observation that Tel1/ATM function is modulated by PI(4)P biding is significant and advances the understanding on the function and regulation of this key kinase in promoting genome integrity maintenance. This is an unanticipated result which is highly novel and has implications for the modulation of Tel1/ATM function through pharmacological manipulation of lipid metabolism. This finding would be of broad interest for scientists working on the response to DNA damage and the maintenance of genome integrity. This reviewer belongs to that group and has limited expertise to evaluate the lipid metabolism genetic manipulation in the manuscript.

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 Virant and colleagues have devised well-thought-out experimentation and analysis pipelines to obtain unbiased measurements of kinetochore protein counts and distances from the centromeric histone known as Cnp1 in fission yeast. My concerns with this study are mainly regarding the clarity of some of the data analysis strategies and data presentation. The authors should be able to address these concerns without new experimentation.

We would like to thank reviewer 1 for their valuable comments. We have addressed all comments and highlighted the changed sections in our revised manuscript.

1. Segmentation of individual centromeres: In general, the authors are to be commended for including a detailed description of their procedures and analysis method. However, it wasn't readily clear to me exactly how they segmented individual centromeres. The lack of a consistent offset between the fluorescence spots corresponding to the protein of interest and Cnp1 in image in Figure 1 makes this issue even more confusing. It will help to display representative segmented individual kinetochores either in the main figure or a supplementary figure.

We thank reviewer #1 for this comment and highly appreciate that they value our effort for detailed method descriptions. We strongly agree: Correct segmentation and best-possible visualization are crucial for our analyses. We hope that the following clarifications help to understand our work better:

All images of the manuscript are reconstructed from localization files using Rapidstorm, which linearly interpolates the localizations on a subpixel grid and fills the pixels based on the distance between the localization and the center of the main subpixel bin to avoid discretization errors (Wolter et al., 2012). These images are then overlaid with a Gaussian blur corresponding to their localization uncertainty. In our opinion, this procedure gives the most realistic image impression of the data and results in reconstructed images that as best as possible mimic real fluorescence images. This is in our opinion a very important aspect when presenting SMLM data in reconstructed images as it is crucial that one - when looking at those images - does not over- or under-interpret the SMLM data. We extended the explanation in the manuscript: “For visualization, we aimed to reconstruct SMLM images that neither over- nor under-interpret the resolution of the SMLM data and resemble fluorescence images as closely as possible. Localizations were tracked together using the Kalman tracking filter in Rapidstorm 3.2 with two sigma, and the NeNA value used as sigma. Images were then reconstructed in Rapidstorm 3.2 with a pixel size of 10 nm. Rapidstorm linearly interpolates the localizations and fills neighboring pixels based on the distance between the localization and the center of the main subpixel bin to avoid discretization errors (Wolter et al., 2012). These images were then processed with a Gaussian blur filter based on their NeNA localization uncertainty in the open-source software ImageJ 1.52p (Schindelin et al., 2012). Importantly, images were only used for image representation purposes, all data analysis steps were conducted on the localization data directly (see data analysis).”

Importantly, individual centromeres were segmented not on the images but on the localization data directly which we visualized in a self-written 3D visualization software that has the functionality to e.g. zoom in and out and to switch or overlay channels. This flexible visualization tool allowed us to make the best-possible informed decisions for the cluster selections and pairing of POI/cnp1CENP-A pairs. We extended our explanation by: “For several manual steps, localizations were visualized in a custom software, which allows to zoom in/out flexibly and to switch/overlay between the sad1/POI/ cnp1CENP-A channels. Using this tool, individual localizations could be selected and classified. For channel alignment, localizations belonging to the same fiducial marker in all three channels were grouped together. Cells with visible kinetochore protein clusters in the focal plane were selected and classified as individual region of interests (ROIs) and all clusters were annotated. cnp1CENP-A clusters were paired together with corresponding POI clusters. Whenever there was any doubt whether two clusters belonged to the same kinetochore or whether a cluster represented a single centromere region or several, the clusters were discarded. Two exemplary data sets can be found in the zip-file Supplementary Data 1. The annotation work was quality-checked by cross-checking the annotation of two different persons.” Maybe interesting to add is that we initially and extensively tried several ways to fully automate the annotation. As all tested routines could not reach the quality of manually annotated data and had to a large extend be manually rechecked and corrected, we in the end directly annotated the data manually. We were rigorous in case of doubt: “Whenever there was any doubt whether two clusters belonged to the same kinetochore or whether a cluster represented a single centromere region or several, the clusters were discarded (manuscript page 10, section Visualization and Manual Analytics).”

Furthermore, the reviewer is right, we didn’t include too much visual data/results into our manuscript. We now showcase some examples for our annotation. In the zip-folder “Supplementary Data 1”, the reviewer will find two csv SMLM data examples of annotated localization data of cells with a mitotic spindle that passed the quality checks of drift control and channel overlay. For the first example, a cell with dam1 as POI, all 6 kinetochores are visible and all were grouped and separated from noise. One can also nicely see (due to the large distance of dam1 to cnp1CENP-A) which of the six belong to which spindle by the spatial orientation of the cnp1CENP-A and dam1 cluster to each other, and both groups of three have a pair that is spatially closer to the wrong pole, important for question 2 below. The second example, with spc7 as the example POI, has only 5 visible clusters. A closer look shows that one of them has unusual dimensions and a high number of localizations, representing most likely two overlapping centromeric regions. Thus, for this cell, only 4 kinetochore pairs were annotated for further analysis. Also, the cluster pairs are overlapping much more, so a direct decision to which spindle they belong to gets difficult.

Finally, we realized that the example cell used for figure 1 which we chose a long time ago for representative purposes actually is a dataset that did not pass the final quality controls of drift correction and channel overlay. Thus, it is not part of the data that was used for the results of this work. While it is pretty embarrassing that we did not realize earlier, we are really grateful for the reviewers’ question about it. We now replaced it by another example which nicely represents the data that went into the analysis and also represents the biological heterogeneity, not only in offset but also e.g. in shape: In this work, we simplify by ignoring any shape in our current analysis and only use cluster centroid distances. Kinetochore POI cluster shapes are currently investigated in a more detailed follow-up study.

1. Use the mixture coefficient: The authors use the coefficient λ to create a mixture model for the Bayesian inference of distances. The description provided in the methods section is not sufficient for an average reader to understand how this coefficient is ultimately used (I had to look up the code and then the Stan manual for a superficial understanding of this procedure). It will be very helpful to flesh out this part of the model. Similarly, notation for the model that they use should be included either in the Methods or in supplementary data so that casual readers can get some understanding of the model.

We added the following sentence to explain the significance of the mixture coefficient:

“The coefficient then corresponds to the prior probability that the centromere is attached to the first spindle pole.”

We are not sure whether we understand the last sentence of this comment correctly. We added more explanations and definitions to the Stan code (see kinetochore.stan) to make it easier to understand. However, the section “Distance calculation” is intended to give the reader a full understanding of all relevant parts of the model, a look at the code should therefore not be necessary. The Stan implementation of the model uses non-centered parameterization, which might make it appear more complex than what is described in the text. However, this is an implementation detail that is intended to make the posterior more well-suited for Monte Carlo estimation and does not change the underlying statistical model. It should therefore be of no concern to casual readers. Finally, we prepared a new Supplementary Figure S6 to visualize the model for all readers.

1. The regional centromeres in fission yeast can incorporate varying levels of Cnp1 depending on its expression level (e.g. see Aravamudhan et al. 2013 Current Biology, Joglekar et al. 2008 JCB). Much of this "extra" Cnp1 is likely to be incorporated at sites distal to the Cnp1 molecules that directly nucleate the kinetochore. Therefore, the centroid of Cnp1 molecules is likely to be "shifted" to some extent from the foundation of the kinetochore. Any shift in the Cnp1 centroid will be important especially when comparing the fission yeast measurements with the budding yeast data. The authors should ascertain whether such a shift can be detected by comparing the budding yeast and fission yeast measurements.

The reviewer is absolutely right, there is an active discussion in the field to which extend there are cnp1CENP-A molecules at distal sites and thus not part of the platform for the kinetochore structure. While different quantification methods in the literature partly disagree in cnp1CENP-A numbers, we have no indications that our own assessment by PALM imaging is wrong. In this study, we get a very stable (Suppl. Figure S9 b) cnp1CENP-A read-out by PamCherry1, which agrees with our Lando et al. 2012 study (a single color study with a different, much simpler protocol and using a different microscope with different settings and analysis routines, mainly using mEos2 but also some SI data on PamCherry1 for counting cnp1CENP-A molecules). Furthermore, the recombinant fusions in the native locus of cnp1CENP-A are stable and the strains show no signs of growth or phenotypic defects.

While we therefore can argue that we see a native level and undisturbed distribution of cnp1CENP-A, we nevertheless do not know how much of this cnp1CENP-A is involved in building up the kinetochore. What we believe to know is the following:

1. With our ChipSeq data in Lando et al. 2012 we explored the distribution and read-out hits of cnp1CENP-A within the outer repeats as well as the inner centromeric region of all three chromosomes (Figure 3 and SI in Lando et al. 2012). While cnp1CENP-A is highly populated within the ~ 10 to 15 kb large inner centromeric regions, there are less detections in the outer regions. Thus, while ChipSeq is not a quantitative method, we believe it’s showing the correct trend with some cnp1CENP-A in the outside regions but most cnp1CENP-A localizing in the inner region. We believe that in overexpression studies (like e.g. done both in Joglekar et al. 2008 & Aravamudhan et al. 2013), this most likely will differ (but we did not experimentally explore this).
2. We generally would argue that our distances are rather accurate using a symmetry and compaction argument. The about 10kb inner regions are a roughly 3.4 µm long DNA strands at a linear 1-dimensional scale but in vivo are highly compacted. Total kinetochore sizes as seen in EM data for mammalian cells are “approximately 250 nm wide and 80 nm deep, with an electron-opaque inner plate juxtaposed to the centromeric chromatin, a translucent gap layer, and an electron-opaque, chromatin-distal outer plate apparently embedding the plus ends of spindle microtubules” (Musacchio and Desai, Biology 2017, 6(1), 5; 10.3390/biology6010005) and for * pombe also in the 200 nm range (Ding et al., J Cell Biol (1993) 120 (1); 10.1083/jcb.120.1.141). We evaluated the cnp1CENP-A cluster shapes as seen in our SMLM data. The clusters show a major axis length of 218 ± 88 nm and a minor axis length of 110 ± 29 nm on average. Taking into account our NeNA localization precisions, this is in nice agreement with the EM data measuring the lateral extend of the kinetochore structure. All together, we would argue a) that there is no reason or any indication in the literature that cnp1CENP-A not directly involved in kinetochore nucleation preferably gets incorporated on only one of the distal sites and thus would cause an asymmetry. We rather would argue that they are randomly inserted on both sides at low level and thus keep the symmetry needed to determine the center of the cnp1CENP-A cluster involved as the kinetochore platform. We also would argue b) that the structure is highly compacted and thus errors caused by additional cnp1CENP-A molecules will be small in respect to our resolution. We cannot completely exclude that there is such an effect that would increase the measured distances, however, and given all other sources of error (drift correction, channel alignment, cluster selection etc.) this most likely is not the main factor in defining the widths of the posterior distributions as we obtain them (Suppl. Figure S7). This argument is supported by the fact that we do not see any indication of such an effect for cnp1CENP-A-close proteins. We carefully checked fta2, fta7 and cnp20 data and also included some examples for the reviewer (see innerPOIexamples.zip). We hope that the reviewer agrees with us that there is no indication for a systematic asymmetric offset between the POI and cnp1CENP-A clusters. Finally, our distance numbers nicely agree with the distances the Ries group has measured for S. cerevisiae in the co-submitted manuscript. They a) have a point centromere with presumably only one kMT and b) did not use cnp1CENP-A *as their reference, they used spc7KNL1 (spc105).

Virant and colleagues present a rigorous single-molecule localization-based analysis of the kinetochore protein copy number and organization within the fission yeast kinetochore. Although the fission yeast kinetochore has been extensively studied, the spatial organization of its kinetochore components has remained uncharacterized. This manuscript addresses this deficiency, and in concert with the budding yeast study, highlights the conserved and diverged features of the kinetochore in the two yeast species. Therefore, this manuscript will be of great interest to the kinetochore and cell division field.

We would like to thank the reviewer again for their very helpful and highly constructive review.

Reviewer #2

In this study, Virant and colleagues have applied single molecule localization microscopy to map the positions of proteins in the pombe kinetochore. This has not been reported previously and this study is both well-conducted and the data appear solid. They also use a modification of this technique to assess the stoichiometry of kinetochore proteins. The results that they obtain are broadly in line with several previous studies that use other methodology but not in fission yeast nor to this level of detail. There are some important novel conclusions from this work. I would like the authors to address the following concerns prior to publication:

We would like to thank Reviewer #2 for their appreciation of our work and their helpful remarks regarding our manuscript which we will answer below.

1. It is not clear to me why the sad1-Scarlet-I signal in Figure 1C, displays a grid-like pattern? This must be an artefact of image collection or processing. Could the authors explain this pattern since this may affect the ability to find a centroid position of this signal?

Thanks a lot for this comment. Yes, the grid-like pattern the reviewer observed is an artefact from image processing when compiling the exemplary composite 3-color images. We revisited the raw movie data and changed the procedure to produce the exemplary images to avoid this ugly artifact (which did not influence our data analysis as it is not present in the raw movies used for centroid determinations). Please note: we also changed the example cell for figure 1 due to reasons explained in the answer 1 to reviewer 1.

1. It is my understanding that the distances reported are based on the positions of the proteins in one dimension, along the spindle axis, consistent with other studies and as illustrated in Figure 1b. This should be clearly stated in the results section.

The model underlying our Bayesian inference is that cnp1CENP-A-POI and one of the two sad1 spindles are all on one “mitotic” axis, along the kinetochore microtubule. BUT the orientation of this mitotic axis is NOT necessarily parallel to the spindle axis. See Figure 1a, the in red drawn spindle axis is not necessarily parallel to the green drawn microtubules connecting the kinetochores to the spindle poles. (Please note, thanks to this comment, we found that our original sketch of Figure 1a was misleading and corrected for that.). Figure 1b in this respect is a bit misleading as only one spindle pole is shown. The slight difference between the kinetochore and spindle axis cannot be visualized with only half a spindle. For answering the reviewers comment no. 5, we also now plotted the measured offset from and relative position of cnp1CENP-A cluster centroids to the spindle axis, see below.

1. The distances between proteins in this study are measured during anaphase, whereas the distances measured in cerevisiae previously were in both metaphase and anaphase (Joglekar et al 2009) and in the accompanying manuscript (Cielinski et al) in metaphase. In the comparison of distances, it would be worth describing how the mitotic stage may have affected distances, since Joglekar et al, found significant positional changes in cerevisiae kinetochore proteins from metaphase to anaphase.

We would like to thank the reviewer for this comment. It sparked a longer discussion to precisely disentangle the cell cycle states. We afterwards went carefully through the literature again and added a column to Table S4 stating for which cell cycle phase(s) the individual works were conducted which we believe is highly useful when comparing the different data.

Discussing with the Ries group we made sure that we indeed measured the cells in the same state and that we in our manuscript made mistakes in defining it correctly. Important for S. pombe is, that while it is not possible to decide for 100% between metaphase and anaphase A, we can safely exclude anaphase B so that we can state that we did not image anaphase B: Supplementary Figure S10 shows that all spindle distances measured are smaller than one nuclear diameter which is given by 2-3 µm (MacLean 1964, 10.1128/jb.88.5.1459-1466.1964; Tda et al 1981, 10.1242/jcs.52.1.271) plus we ourselves measured a nup132 strain in early G2 phase and obtained 2.4 µm ± 0.19 µm (data not shown)). This is nicely in line with Joglekar et al. 2009 (this paper actually has a very good SI figure on exactly this topic, see Fig. S1) and corrected the manuscript accordingly. Thanks for pointing this out to remove this lapse in definitions.

__ 4. __It is hard to interpret the POI copy numbers in terms of each kMT. I am assuming that each cluster measured represents a single pombe kinetochore, containing 2-4 kMTs? If we assume that each pombe kinetochore can contain 2, 3, or 4 kMTs, then we might expect to see a trimodal dataset, I am guessing this was not seen in the data? Would it be possible to estimate protein numbers per kMT in Table 2, as done for the Cielinski et al study? I realize this would require an estimate of the number of kMTs per kinetochore. Alternatively, the authors are resolving individual kMTs, in which case this should be made clear.

Yes, the reviewer is absolutely correct, the clusters are associated with 2-4 kMT (as nicely resolved in Ding et al, J Cell Biol (1993) 120 (1); 10.1083/jcb.120.1.141). Thus, we can assume that 2-4 kinetochore structures are also involved per centromeric region. In our current analysis, we have to work with the average of 2-4 kMTs. The shapes of POI and cnp1CENP-A clusters we have in the SMLM data are definitely diverse, and we plan to extract more data on spatial distribution in the future, perhaps even at the level of individual centromeric regions, but we did not systematically explore shapes in this work. Thus currently, we cannot give a precise answer how individual regional kinetochores look like at the level of a single kinetochore, but we strongly agree with the reviewer that this will be highly interesting to explore further. In this manuscript, therefore, to compare the stoichiometries between the point centromere of S. cerevisiae and the regional centromere of S. pombe, we used ratios as given in Suppl. Table S4. These ratios provided us with comparable results across a wide range of literature since the ratios are only calculated internally for each study and not across studies (which would lead to compatibility issues).

For this study, we also labeled MT via atb2. Unfortunately, the SMLM experiments were very difficult as atb2 is also present everywhere else in the nucleus, in particular at the dense central MT bundle (see image below, white sad1-mScarlet-I, blue PamCherry1- cnp1CENP-A, and red mEos3.2-A69T-atb2). Thus, we could not resolve such fine details as single fibers for the kMTs: Most kMTs overlapped with the central fiber and due to the dense central MT bundle of atb2, the data of the atb2 channel could not be read-out neither in a quantitative nor complete way and we could not extract which percentage of atb2 molecules we actually successfully recorded in the SMLM data. Thus, especially visualizing fine fibers was difficult and the images obtained do not meet our quality standards – but the exemplary one below is maybe nevertheless informative in a qualitative way. Our current idea would be to use ExM plus SMLM, but this work would be a stand-alone study requiring the set-up and optimization of such a protocol.

1. The same kMT issue may affect the measurements of distances. Each pombe kinetochore contains multiple kMTs and it is not clear whether these would align perfectly on the spindle axis. Did the authors see anything in their data that would support the notion that individual kMTs are aligned on the axis (as illustrated in Figure 2) or whether they are slightly separated? This is itself a potentially important result.

It is important to note that we did not measure the distance in projection of the spindle axis (defined as sad1-sad1 centroid axis), see also question 2. We can show in our data that the main microtubule bundle between the spindles is angled to the kinetochore microtubules that connect the centroids of our three-color channels for each centromeric region in a sad1-POI-cnp1 axis. For sad1, we cannot simply determine which one of the two spindles is the correct one, thus we implemented a mixture model for our Bayesian model, see also answers to reviewer 1).

While in the budding yeast literature it has been measured by EM that there are only small angles of up to 6° between the two axes present (Joglekar et al. 2009, 10.1016/j.cub.2009.02.056, Figure S1), Ding et al. 1993 showed larger deviations for S. pombe. Our data agrees with these findings. In the new Suppl. Figure S12, we plotted the height of all measured cnp1CENP-A centroids, normalized to the spindle lengths to represent the angular distribution between the spindle and kMT axes and in absolute nanometer distances to show that most kinetochores are in direct vicinity to the central bundle and only few show heights larger than 150 nm (also technically important as our focal z-range is ~600 nm).

1. In all measurements of kinetochore protein intensity (both in this study and previous studies) there seems to be significant variation in the data for individual kinetochores, even for S. cerevisiae, which supposedly has a fixed number of the kMTs. The coefficient of variation is ~ 0.5 in the data shown in Table 2. Could the authors discuss the variability in POI copy numbers since it either reflects an inability to measure protein levels accurately or that there is some flexibility in kinetochore protein stoichiometry (or in this case differing numbers of kMTs per kinetochore - see point above)?

Regarding the variability in our counting data we indeed expect a mixture of biological and technical nature in line with what the reviewer argues above but intuitively would lean towards the latter, being technically limited by the read-out precision of quantitative PALM imaging using fluorescent proteins, which have finite maturation- and read-out efficiencies and possess limited signal-to-noise contrast. When discussing this comment and how we possibly could support our intuition by evidence, we realized that the main argument to be made is that the FtnA oligomer is a biologically highly defined structure and that the variance we obtain for our FtnA calibration standard indeed also directly can serve as a proxy for our technical variance. Using the results of 21.63 counts ± 10.2 STD for the 24mers and of 7.27 counts ± 2.72 STD for the 8mers, we thus can estimate that the technical variance causes a coefficient of variance of 0.35 to 0.5, thus almost completely explaining the experimentally seen coefficient of ~ 0.5. we added this argument to our manuscript by “The POI protein copy numbers as given in Table 2 show a large coefficient of variation. To assess to which extend this variability reflects a technical inability to measure protein levels accurately or some flexibility in kinetochore protein stoichiometry (e.g. due to differing numbers of kMTs per kinetochore), we can use the data of the FtnA oligomer counting standard: The FtnA oligomer is a biologically highly defined structure. Thus, our FtnA measurements can directly serve as a proxy for the contribution of the technical inaccuracy of our PALM imaging and analysis strategy to the variance. Using the results of 21.63 counts ± 10.2 STD for the 24mers and of 7.27 counts ± 2.72 STD for the 8mers, we can estimate that the technical inaccuracy causes a coefficient of variance of 0.35 to 0.5, thus almost completely explaining the experimentally seen coefficient of ~ 0.5 for our POI data (Table 2). Due to this high technical inaccuracy, we cannot resolve sub-populations of possibly different kinetochore structures (and thus POI copy numbers) on 2-4 kMTs in our current counting data (Supplementary Figure S9).”. Secondly, as the reviewer also pointed out earlier, we have a biological variance of 2-4 kMTs and thus must assume smaller and larger kinetochores on individual centromeres. Due to the high technical variance, we nevertheless cannot resolve sub-populations in our current counting data (please also compare to the data in Supplementary Figure S9).

Minor points:

Delete "an" from "...structure at an about 100 nm resolution" (page 3).

Thanks!

In Figure 2 the proteins in the schematic are color coded, but it is not clear what the coloured proteins are in all cases. Would it be possible to color code the adjacent text, e.g. Spc7 in orange.

Thanks for this suggestion, adjusted figure accordingly.

Also in this figure, the POI copy numbers are indicated by color coding of the data points. However, the points will likely be too small in the final figure for these colors to be clearly visible. Perhaps copy numbers could be indicated in another way or the "mean value" boxes could be larger?

We tested several options and decided to adjust the widths of the boxes.

Please define "N" in Table 2. e.g. N = number of kinetochores measured.

We added this information to the caption: “N number of centromeric regions analyzed.”

Reviewer #2 (Significance (Required)):

This manuscript, together with an accompanying one from Cielinski et al., are nice complementary studies that provide the first single molecule localization studies of the yeast kinetochore. Although other labs have used super-resolution methods to study individual kinetochore proteins; both of these new studies map distances between many proteins at the kinetochore and thus are able to produce maps of the overall kinetochore structure. Like the previous study using standard resolution methods (Joglekar et al, 2009. Current Biology 19, 694-699); these studies will likely provide a benchmark for future studies on eukaryotic kinetochore architecture, including those in mammalian systems. Additionally, this work will appeal to super-resolution microscopists.

My expertise is as a yeast kinetochore cell biologist.

We would like to thank Reviewer #2 again for their appreciation of our work and their valuable remarks and discussion points which improved our manuscript substantially during the revision phase.

Additional comments for both reviewers:

As the co-submitted manuscript from Cieslinski et al. 2021 re-analyzed all their distances and numbers during the revision phase, we updated the comparisons in Suppl. Table S4, S5 and our summary text: 1) their cnn1 distance measurement got corrected and no shows no deviation anymore to our data. Also, their protein copy numbers changed slightly. So we changed the summary from “Additionally, two organism-specific differences surfaced: cnp20CENP-T (cnn1) is located between spindle pole and cnp1CENP-A in our case (at similar distance as fta2CENP-P and fta7CENP-Q), whereas Cieslinski et al. position cnn1 (and mif2) behind cnp1CENP-A. Furthermore, the ratio of cnp20CENP-T to COMA is 1:0.9 in our case and 1:2.1 for S. cerevisiae“ to “Importantly, one substantial organism-specific difference for the inner kinetochore strategy surfaced: The ratio of cnp20CENP-T to COMA is 1:0.9 in our case and 1:2.0 for S. cerevisiae.” 2) Additionally, we were able to add a discussion about their new measurement of ask1 of the dam1 complex. Ask1 is a protein of the DASH ring. Their new distance measurement of S. cerevisiae ask1 fits to the distance we measure for S. pombe dam1 and thus supports our discussion that, for S.pombe, the C-terminus of dam1 localizes at the DASH ring and not at the ndc80 heads like for S. cerevisiae. We added this sentence to the summary: “Furthermore, the S. cerevisiae work measured the position of ask1, a protein of the DASH ring. Their positioning of S. cerevisiae ask1 is consistent with the distance we measured for S. pombe dam1 and thus directly supports our reasoning that the C-terminus of S. pombe dam1 is localized to the DASH ring and not to the ndc80 heads (like for S. cerevisiae).”

We found that we – very unfortunately - did a calculation mistake ourselves and used the inverse of the correction factor (multiplied with 0.9 instead of dividing with 0.9) when correcting our SMLM localizations to absolute protein counts. Thus, the numbers we gave in Table 2 and the color bar in Figure 2 were wrongly converted. We now corrected for this lapse.

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 study, Virant and colleagues have applied single molecule localization microscopy to map the positions of proteins in the pombe kinetochore. This has not been reported previously and this study is both well-conducted and the data appear solid. They also use a modification of this technique to assess the stoichiometry of kinetochore proteins. The results that they obtain are broadly in line with several previous studies that use other methodology but not in fission yeast nor to this level of detail. There are some important novel conclusions from this work. I would like the authors to address the following concerns prior to publication:

1. It is not clear to me why the sad1-Scarlett-I signal in Figure 1C, displays a grid-like pattern? This must be an artefact of image collection or processing. Could the authors explain this pattern since this may affect the ability to find a centroid position of this signal?
2. It is my understanding that the distances reported are based on the positions of the proteins in one dimension, along the spindle axis, consistent with other studies and as illustrated in Figure 1b. This should be clearly stated in the results section.
3. The distances between proteins in this study are measured during anaphase, whereas the distances measured in cerevisiae previously were in both metaphase and anaphase (Joglekar et al 2009) and in the accompanying manuscript (Cielinski et al) in metaphase. In the comparison of distances, it would be worth describing how the mitotic stage may have affected distances, since Joglekar et al, found significant positional changes in cerevisiae kinetochore proteins from metaphase to anaphase.
4. It is hard to interpret the POI copy numbers in terms of each kMT. I am assuming that each cluster measured represents a single pombe kinetochore, containing 2-4 kMTs? If we assume that each pombe kinetochore can contain 2, 3, or 4 kMTs, then we might expect to see a trimodal dataset, I am guessing this was not seen in the data? Would it be possible to estimate protein numbers per kMT in Table 2, as done for the Cielinski et al study? I realise this would require an estimate of the number of kMTs per kinetochore. Alternatively, the authors are resolving individual kMTs, in which case this should be made clear.
5. The same kMT issue may affect the measurements of distances. Each pombe kinetochore contains multiple kMTs and it is not clear whether these would align perfectly on the spindle axis. Did the authors see anything in their data that would support the notion that individual kMTs are aligned on the axis (as illustrated in Figure 2) or whether they are slightly separated? This is itself a potentially important result.
6. In all measurements of kinetochore protein intensity (both in this study and previous studies) there seems to be significant variation in the data for individual kinetochores, even for S. cerevisiae, which supposedly has a fixed number of the kMTs. The coefficient of variation is ~ 0.5 in the data shown in Table 2. Could the authors discuss the variability in POI copy numbers since it either reflects an inability to measure protein levels accurately or that there is some flexibility in kinetochore protein stoichiometry (or in this case differing numbers of kMTs per kinetochore - see point above)?

Minor points:

Delete "an" from "...structure at an about 100 nm resolution" (page 3).

In Figure 2 the proteins in the schematic are color coded, but it is not clear what the coloured proteins are in all cases. Would it be possible to color code the adjacent text, e.g. Spc7 in orange. Also in this figure, the POI copy numbers are indicated by color coding of the data points. However, the points will likely be too small in the final figure for these colors to be clearly visible. Perhaps copy numbers could be indicated in another way or the "mean value" boxes could be larger?

Please define "N" in Table 2. e.g. N = number of kinetochores measured.

Significance

This manuscript, together with an accompanying one from Cielinski et al., are nice complementary studies that provide the first single molecule localization studies of the yeast kinetochore. Although other labs have used super-resolution methods to study individual kinetochore proteins; both of these new studies map distances between many proteins at the kinetochore and thus are able to produce maps of the overall kinetochore structure. Like the previous study using standard resolution methods (Joglekar et al, 2009. Current Biology 19, 694-699); these studies will likely provide a benchmark for future studies on eukaryotic kinetochore architecture, including those in mammalian systems. Additionally, this work will appeal to super-resolution microscopists.

My expertise is as a yeast kinetochore cell 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 #1

Evidence, reproducibility and clarity

Virant and colleagues have devised well-thought-out experimentation and analysis pipelines to obtain unbiased measurements of kinetochore protein counts and distances from the centromeric histone known as Cnp1 in fission yeast. My concerns with this study are mainly regarding the clarity of some of the data analysis strategies and data presentation. The authors should be able to address these concerns without new experimentation.

1. Segmentation of individual centromeres: In general, the authors are to be commended for including a detailed description of their procedures and analysis method. However, it wasn't readily clear to me exactly how they segmented individual centromeres. The lack of a consistent offset between the fluorescence spots corresponding to the protein of interest and Cnp1 in image in Figure 1 makes this issue even more confusing. It will help to display representative segmented individual kinetochores either in the main figure or a supplementary figure.
2. Use the mixture coefficient: The authors use the coefficient λ to create a mixture model for the Bayesian inference of distances. The description provided in the methods section is not sufficient for an average reader to understand how this coefficient is ultimately used (I had to look up the code and then the Stan manual for a superficial understanding of this procedure). It will be very helpful to flesh out this part of the model. Similarly, notation for the model that they use should be included either in the Methods or in supplementary data so that casual readers can get some understanding of the model.
3. The regional centromeres in fission yeast can incorporate varying levels of Cnp1 depending on its expression level (e.g. see Aravamudhan et al. 2013 Current Biology, Joglekar et al. 2008 JCB). Much of this "extra" Cnp1 is likely to be incorporated at sites distal to the Cnp1 molecules that directly nucleate the kinetochore. Therefore, the centroid of Cnp1 molecules is likely to be "shifted" to some extent from the foundation of the kinetochore. Any shift in the Cnp1 centroid will be important especially when comparing the fission yeast measurements with the budding yeast data. The authors should ascertain whether such a shift can be detected by comparing the budding yeast and fission yeast measurements.

Significance

Virant and colleagues present a rigorous single-molecule localization-based analysis of the kinetochore protein copy number and organization within the fission yeast kinetochore. Although the fission yeast kinetochore has been extensively studied, the spatial organization of its kinetochore components has remained uncharacterized. This manuscript addresses this deficiency, and in concert with the budding yeast study, highlights the conserved and diverged features of the kinetochore in the two yeast species. Therefore, this manuscript will be of great interest to the kinetochore and cell division field.

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

Please note we have uploaded a PDF with the point to point reply.

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:

Wang et al. present an evaluation of a new generation of time-of-flight-based mass spectrometer that improves on the fraction of ions factually used for detection of peptide analytes, thus boosting the sensitivity of the Zenotof 7600 system when compared to the same instrument with the duty-cycle-enhancing Zenotrap module disabled and also when compared to the previous generation instrument of the same vendor in some of the comparisons.

The authors position the MS acquisition technique as particularly suitable in combination with medium (micro-) and high ('analytical') flow and throughput methods where higher flow rates (vs. conventional nanoflow-LCMS) allow rapid sample turnover and high throughput, yet limit the efficiency of electrospray and ion transfer into the MS system, thus being in dire need for enhanced sensitivity of the MS system employed for detection. The competency of such an MS system for very low input materials as e.g. encountered in emerging single-cell proteomic workflows, typically employing nanoflow chromatography, was thus not part of the study.

Accordingly, a medium- (micro-flow) and very high ('analytical'-flow) throughput LC method were screened on the three MS (parameter) setups using human cell lysate digests typically utilized in such technical evaluations. Well-received, the authors further extended their analysis (for the new instrument) across additional sample types of clinical and extended biological interest and spanning different levels of complexity and dynamic range of contained protein analytes.

In addition, the authors also performed a controlled ratio 2-species mixture experiment which allows detailed benchmarking of proteome coverage as well as the quality of protein quantification in a known differential comparison for the medium throughput (micro-flow) method.

The data quite convincingly demonstrate an increased sensitivity of the instrument based on similar identification performance in DIA bottom up proteomics from ca. 3- to 8-fold lower input peptide mass. However, I see a number of shortcomings mainly in the presentation and in part the completeness of the work, with specific comments below.

Major comments:

• Are the key conclusions convincing?

  • The concluded 10x sensitivity increase is overstating the observed numbers (x5-x8). In addition, the authors should at least discuss other changes than the Zeno trap incurred in the Zeno SWATH vs non-Zeno-SWATH DIA setups, particularly changes in accumulation times per m/z range, with Zeno Swath accumulating ~42 % longer per cycle spanning the same m/z range (85 vs 60 windows with 11ms per window) in the uflow method set and ~ 18 % longer in the high-flow method set (same window number but 13 ms vs. 11 ms dwell time per window). This should be discussed as one of the optimizations/factors contributing to the increased sensitivity observed in Zeno Swath measurements vs conventional SWATH. On that note, it was unclear to me when and where the 40 variable window SWATH method mentioned in the methods was used and where the settings can be found.
  • Since injected material is a critical parameter here, it would be good if it was mentioned also with the key conclusion on the increased number of confidently quantified peptides in microflow (based on the 2-species controlled quantity experiment).
  • Conclusions 'increasing protein identification numbers through the use of analytical-flow-rate chromatography' does not capture the observed data; the use of analytical-flow-rate does not convey an increase in protein identification numbers but enhanced sensitivity rather enables the maintenance of high protein identification numbers / proteome coverage despite/concurrent with analytical-flow chromatography
  • In titration curve experiments like these, probing proteome coverage from relatively small sample amounts, special care

  • Should the authors qualify some of their claims as preliminary or speculative, or remove them altogether?

  • 'Zeno SWATH increases protein identification in complex samples<br /> 5- to 10-fold when compared to current SWATH acquisition methods on the same instrument' - At no point this is shown, a decrease of required input amounts by 5-8-fold (increase in sensitivity) is shown by the data, not a multiplication of protein identification rates by that factor.

  • 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.

  • Figure 1f, Supplemental Figure 1b, Figure 3 and Supplemental Figure 3 lack data for the Zeno SWATH method's performance at higher concentration. Given the fact that there is a clear, continuous trend of significant enhancement of proteomic depth in the highest 3 concentrations sampled by the Zeno SWATH method, I lack an assessment of the upper limit of proteome coverage achievable by the new platform when input material is not limited, or at least learn why injecting more is not advisable on the ZenoTOF 7600 system. It is clear that the region of interest is the lower loads where sensitivity gains are most pronounced, but with the strong trend in IDs per ng injected in the sampled range and discrepant range sampled by the non-Zeno method I feel there is a gap in the dataset and the upper ceiling of proteome coverage could be mapped out more thoroughly (At least for human cell lysate and possibly human plasma where trends appear most (log2-)linear).

  • Similarly, unless constrained for technical or practical reasons, I would suggest to find the ceiling for achievable proteome depth in analytical flow (4, 8 ug?)

  • 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.

  • All these should be re-injections of existing samples on these MS setups and a minor effort provided instrument availability (<1w) and rapid re-analysis via DIA-NN.

  • Are the data and the methods presented in such a way that they can be reproduced?

  • The raw data have not been deposited to a public repository. Reproducibility of the study would benefit significantly by raw data (including search results and spectral libraries with log files of creation) upload/sharing e.g. via ProteomeXchange/PRIDE.

  • If any software versions or firmwares on the hardware are required to perform the measurements on the ZenoTOF on the market today, these versions and prospective release dates should be included or the accessibility of these settings commented on.

  • Are the experiments adequately replicated and statistical analysis adequate?

  • Figures 3 and Supplemental Figure 3 need a clarification in the legend as to the nature and origin of ID numbers (mean? Number of replicates? Add error bars if possible)

  • The usage of DIA-NN for data analysis is somewhat unclear, in particular the in Methods/Spectral libraries "For the analysis of plasma samples, a project-independent public spectral library [29] was used as described previously [15]. The Human UniProt [30] isoform sequence database (UP000005640, 19 October 2021) was used to annotate the library and the processing was performed using the MBR mode in DIA-NN." The authors should address in a revised version whether the identification numbers reported stem from two-pass or single-pass analysis (i.e. when the feature termed Match-between-runs implemented since DIANNv1.8 was enabled and whether all runs, spanning different injection amounts were co-analyzed and data-re-queried for a targeted library containing precursors identified in high load samples in first pass analysis and then queried in low-load samples. In other words, are the low-load IDs independent of high load IDs? If not (i.e. the different loads were co-analyzed with MBR), what proteome coverage to the low sample loads reach bona fide, without the 'guidance' of high-load IDs?

Side note: Turning this around, could a high-load injection e.g. from a pool of limited-amount samples serve as a guiding element in a MBR-enabled analysis of a large cohort with limited sample amounts available per biological condition?

Minor comments:

• Specific experimental issues that are easily addressable.

  • The authors state the impact on dynamic range of identification when comparing ID sets against an external dataset with presumable cellular concentration numbers. I would in addition suggest comparing the dynamic range of the quantititative values observed from the available data which should provide a direct assessment of the dynamic range of quantification of the two methods.

  • Are prior studies referenced appropriately?

  • The statement that conventional DIA methods rely on nanoflow chromatography (p3, paragraph 3) is not accurate as there is previous implementations of data-independent acquisition MS of microflow separations, in part the group's work and referred to later in the text.

o Vowinckel, J. et al. Cost-effective generation of precise label-free quantitative proteomes in high-throughput by microLC and data-independent acquisition. Sci. Rep. 8, 4346 (2018)

o Bruderer, R. et al. Analysis of 1508 plasma samples by capillary-flow data-independent acquisition profiles proteomics of weight loss and maintenance. Mol. Cell Proteom. 18, 1242-1254 (2019).

It is correct that most early implementations of DIA-MS utilized nanoflow separations due to sensitivity and proteome coverage but DIA as such is a chromatography-flow-speed-agnostic principle and the concept to combine microflow LC with DIA not new, yet powerful as demonstrated by the authors and others previously and once again, here. - P.3 paragraph 3 'Moreover, the increased sensitivity of DIA methods has facilitated applications in large-scale proteomics, including system-biology studies in various model organisms, disease states, and species [5-9]' Include Ref 4 where improved sensitivity of DIA was demonstrated (at proteomic breadth..) - Are the text and figures clear and accurate?

-   Text and Figures need to be edited for typos, language, and clarity/accuracy.

1. Abstract 'Zeno SWATH increases protein identification in complex samples<br /> 5- to 10-fold when compared to current SWATH acquisition methods on the same instrument' - At no point this is shown, a drop of required input amounts by 5-8-fold (increase in sensitivity) is shown by the data, not a multiplication of protein identification rates.
2. P. 4 paragraph 3: Use terms 'consensus' or 'shared' identifications or similar to refer to the proteins identified in all 3 replicates, rather than 'reproducible' when discussing the reproducibility of peptide and protein quantification (as contrast to reproducibility of identification).
3. P.3 paragraph 2 'selects and fragments multiple charge ions' -> multiply charged (?)
4. P. 4 p. 1 'leading to under-detection' please clarify (leading to partial ion usage and limited sensitivity?)
5. P. 6 paragraph 3 'The gain in identification number of Zeno SWATH versus SWATH is mostly explained by an increased dynamic range: i.e. more low-abundance proteins are detected' - Reformulate/clarify: Is increased dynamic range of identifications against external quantities an explanation or perhaps simply the increased sensitivity with improved duty cycle?
6. Term 'active gradient' unclear. An inactive gradient is isocratic flow. Omit 'active'. Isocratic/other portions are overhead.
7. Figure 1 panel a) iteration scheme a-d) is redundant with the rest of the figure; use alternative iter scheme within panel a). Panel a) is further contains illegibly small fonts and should be edited for legibility
8. Revisit y-axis labels. Example: Fig. 1f) 'Precursors Identificaiton' -> Precursors identified/Precursor identifications. Correct throughout manuscript
9. ID bar graphs in all Figures: Cumulative IDs shade of grey is not properly visible, suggest alternate color scheme or add black color outline to the bars
10. Figure 1 e) legend 'along gradient length' -> gradient time / retention time
11. Figure 1 d) too small, trend lines mentioned in text invisible in graph. Boxplots very small.
12. There is three different terms used for the high throughput method (analytical-flow, high-flow, and another one.. please align where possible for clarity (i.e. choose 2 names for the 2 methods throughout the manuscript) etc..

13. Do you have suggestions that would help the authors improve the presentation of their data and conclusions?

    • They authors may consider adding a short explanation of the term 'dynamic range coverage of identification' to contrast this from a direct assessment of dynamic range of quantitative values observed in this study.
    • 2-species controlled experiment: The discrepancy of observed vs true mixing ratios suggests the data were scaled during the analysis which, with these mixture ratios, tends to distort the accuracy (i.e. generates offset of observed from true ratio. That's very likely not a pipetting error on a log scale). In other words, you may want to evaluate the raw quantitative ratios (w/o any normalization/scaling applied) which should be more reflective of true/manual pipetting ratios in light of normalization strategy incompatibily with certain species mix scenarios (compare Supplementary Figure 1 a). Note to the editor(s): This will not affect the clear benefit of Zenotrap usage demonstrated by the 2-species benchmark as is but can be considered a minor yet recommended improvement (thus here).
    • The 2-species controlled experiment can reveal more information than currently extracted and I would recommend to show Zeno Swath and Swath xy scatters, including count-scaled density distributions of the observed ratios, side-by side. This would give deeper understanding of the large impact of the Zeno SWATH method. Also, I believe I haven't seen any instrument to date delivering precise quantification over as broad a dynamic range as surmisable from Fig. 1d) which might be worth wile highlighting.

Significance

Wang et al. describe a technical advance in ion usage and sensitivity based on an ion-trap device storing and focusing ions for TOF-based bottom-proteomics measurements. The study demonstrates improved sensitivity relative to previous generation instrumentation and also explores the impact of the specific trap device relative to the general improvements of the remaining MS system. The work outlines a route towards high coverage proteomics at very high throughput and robustness, as desirable in clinical proteomics and prospective personalized medicine approaches. While not all sample types of interest are limited to the amounts where the strongest improvements are seen in the presented data, large scale studies across expansive cohorts will likey be rendered more practical and realistic due to reduced instrument contamination at reduced loads and also further applications beyond those discussed in the manuscript will be rendered feasible on the newer generation instrument.

The improved ZenoTOF system and SWATH method follows a series of innovations in the mass spectrometry instrumentation, most notably and related the drastic improvement of ion utilization by storage e.g in a trapped ion mobility device earlier in the ion stream where, beyond an accumulation-based boost of sensitivity, ion mobility as a further biophysical properties is assessed in addition to the conventional m/z, as reviewed recently (doi: 10.1016/j.mcpro.2021.100138.). While these developments culminated and have been targeting low-flow, ultra-high sensitivity applications such as single-cell proteomics, the present study takes a different angle towards higher throughput measurements from significantly larger than single cell, but also significantly lower than historically required sample amounts that were prohibitive to a range of applications that are now easier to accomplish thanks to this and related work of the authors and others. The presented research appears of broad relevance and interest to the scientific community interested in protein abundance pattern analysis, in particular in larger (clinical) cohorts. Furthermore, the performance metrics on proteomic depth from human cell lysate digests will likely allow researchers with analytical quests other than those exemplified in the manuscript to extrapolate the ZenoTOF and Zeno SWATH suitability for their respective analytical targets.

Reviewer Field of expertise/background:

Quantitative proteomics. DIA mass spectrometry method & algorithm development & heavy usage. Protein Biochemistry. Molecular 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

In this manuscript Wang et al, benchmark the new ZenoTOF with analytical and micro-flow set up and show impressive numbers of proteins identified and quantified. The paper is well written, and I have only a few minor comments:

1. Figure 1. many of the panels are hard to read. Especially 1a
2. Figure 1d. can the human amount not be normalised to log2=0?
3. Please provide in the legend the bin size for the figure in 1e.
4. Page 10 top: did SWATH identify more proteins than Zeno SWATH in plasma? There is something wrong as the figure shows something else. Also: That sentence in brackets is confusing.
5. Typo: page 10: respectively.
6. Please add raw data to PRIDE or a similar repository.

Significance

This is an impressive new technology that has been benchmarked by the Ralser group. It outperforms current state-of-the-art approaches.

The primary audience is the proteomics community.

My expertise is proteomics and quantitative mass spectrometry. I am well qualified to review this paper.

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

1. General Statements [optional]

In our work, we quantified the abundance and positions of major kinetochore proteins within the metaphase kinetochore in budding yeast using single-molecule localization microscopy. Based on these measures, we revised the current model of the kinetochore and provided a nanoscale view of the complex.

We now revised our manuscript according to reviewers’ points. We performed new analyses to quantify the measurement errors and to justify our data analysis workflows. We further exploited the correlation-based analysis and found a correlation between the spreads of kinetochore proteins perpendicular to the spindle axis and their positions along the axis. We also discussed the potential non-centromeric pools and revised our model of the kinetochore. Further information on our analyses was now provided to improve the clarity. Changes to the text were implemented to better reflect our data. Information from relevant works was incorporated to better connect this work to the field.

We thank the reviewers for their points, which help us show the rigorousness of our analyses, further demonstrate the potential of our work, and improve clarity.

2. Point-by-point description of the revisions

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

The authors have developed a rigorous methodology for using single-molecule imaging of exogenously labeled kinetochore proteins to count and estimate their copy numbers and the average distance from the kinetochore protein Spc105. Although the method is technically sound, its application to the kinetochore raises some crucial questions below. My biggest concern is the effect of non-centromeric pools of the centromeric proteins Cse4, Cep3, and Ctf19 on the estimated copy number per kinetochore. The authors should be able to address most, if not all, questions by presenting a more in-depth data analysis.

Major points

1. Accounting for tilt of the yeast spindle relative to the image plane: It is not clear to me how the authors ascertain whether the spindle being imaged is nearly parallel to the image plane. In the companion fission yeast study, spindle poles are used for this purpose, but this study seems to rely only on the labeled kinetochore proteins. The criteria used to select the in-plane spindles should be clearly defined.

We thank the reviewer for pointing this out. We selected the in-plane spindles based on their average PSF size, which informs the z positions of the center of the kinetochore cluster (for simplicity, now all ’half-spindle’ was changed to ‘kinetochore cluster’). To calibrate the z position of kinetochore clusters, we first measured the width of the kinetochore cluster by fitting a cylindrical distribution. Overall, the kinetochores are likely symmetrically distributed around the spindle axes. Therefore, the height and the width of a kinetochore cluster should be the same. We then calibrated the z positions of the PSF size based on fluorescent bead data. Next, we plugged in the cylindrical distribution to the calibration curve to correlate the mean PSF size and position of the kinetochore cluster. We only took the kinetochore clusters with a mean PSF size

1. The effects of PSF depth on counting kinetochore proteins: The authors use a well-characterized nuclear pore protein as the reference to estimate kinetochore protein counts per half-spindle. Although this method appears rigorous in principle, I am unsure about the effect of the spatial distribution of kinetochores on the accuracy of the estimated number. Nuclear pore proteins are all localized within an 100 nm away from the focal plane even when the spindle is perfectly parallel to the focal plane. A discussion of this possibility, its effect on the protein count/distance estimates, and any mitigating factors is essential to highlight the caveats associated with the conclusions.

Based on the cylindrical distribution (see please the reply to point 1) of kinetochore clusters and their positions in z, we calculated the upper and lower boundaries of the distribution of kinetochore proteins in z, given a specific mean PSF size cutoff of a kinetochore cluster. Regardless of how stringent the cutoff is (130 and 135 nm), we made sure the boundaries do not exceed the imaging depth defined by our choice of the PSF size filtering (

1. Presentation of the cross-correlation analysis: The authors use cross-correlation for an unbiased calculation of the axial separation between a protein of interest and Cse4, but I am curious about the structure of the underlying data, and the intensity image in Figure 1 is not easy to examine. It will be helpful to include more analysis of the underlying data for at least a subset of the proteins (e.g., proteins at short, intermediate, and long distances from Cse4) as supplementary data.

2. The authors should include X and Y projections of the cross-correlation function.

3. Do the widths of cross-correlation functions (i.e., their spread perpendicular to the spindle axis) match across all proteins and experiments? This should be an almost invariant characteristic of the measurements, assuming that proteins within each kinetochore tightly cluster around the 25 nm microtubule. This line of thinking makes the large width of the cross-correlation shown in Figure 1 somewhat surprising.

4. It will also be interesting to test if the correlation between the positions of Spc105 molecules, especially perpendicular to the spindle axis, is comparable to the known separations between adjacent microtubules in the yeast spindle (the authors could use Winey et al. 1995 for serial-section EM of yeast spindles for comparison).

The reviewer is interested in the spread, or the size of the distribution, of a protein in a kinetochore along and perpendicular to the spindle axis. This is an interesting idea and can be done practically. However, the information can be more easily obtained based on auto-correlation instead of cross-correlation, due to its better signal-to-noise ratio along the dimension perpendicular to the spindle axis. Cross-correlations in that dimension are convoluted with background localizations and different localization precisions of the two channels. These factors are hard to interpret and disentangled. In auto-correlations, although the background is still present, it can be modeled and then removed easily, as now mentioned on page 15 lines 500-516.

Accordingly, we performed auto-correlation analysis on all the proteins and compared them to simulations representing different sizes. We find that the size of the distribution correlates to the position of the protein along the spindle axis. The results are now included as the new Fig. S5 and discussed on page 6 lines 169-176.

The cross-correlation analysis was based on only the position of the maximum value, not the projections. To keep the figure concise, we decided not to include the projections. However, the auto-correlation analysis was indeed based on projections, which we now included in Fig. S5.

Regarding the correlation between the positions of Spc105 molecules, we believe the reviewer actually refers to the correlation between the positions of kinetochores. Auto-/cross-correlations contain the information of the cluster sizes, based on the first peak (as shown in Fig. S5), and the relative distance (if the pattern is periodic). Unfortunately, the positions of kinetochores perpendicular to the spindle axis are not periodically distributed. Therefore, we cannot comment on the separations between adjacent microtubules.

1. Cse4 count (4 per kinetochore) and the model presented: One of the surprising conclusions of the study is that there are two nucleosomes associated with each microtubule attachment, with Mif2/CENP-C potentially interacting with both nucleosomes. There are two critical issues that the authors must consider.

(1) Fluorescent protein chimeras of Cse4 and CBF3 and COMA complex members do not exclusively localize to kinetochores. Biochemical studies show that both Cse4 and CBF3 proteins interact with non-centromeric DNA, e.g., see work from the Biggins lab regarding Cse4 over-expression and also from the Henikoff group that used ChIP-seq. I can't think of a similar reference for the CBF3 complex, but the DNA-binding proteins are also likely to interact with other parts of the genome. The non-centromeric protein is visible as a significant background fluorescence in wide-field microscopy, e.g., see Cep3 localization here: https://images.yeastrc.org/imagerepo/viewExperiment.do?id=202308&experimentGroupOffset=3&experimentOffset=0&experimentGroupSize=3

Similar background fluorescence can be detected for Cse4 and Ctf19. This extra-centromeric localization of Cse4, Cep3, and Ctf19 makes it possible that the protein counts included by the authors are "contaminated" to some extent by the extra-centromeric protein. The authors should discuss this possibility and how it might affect their counts.

After consideration, we agree with the reviewer that, specifically, a fraction of counted Cse4 molecules should be considered non-centromeric. We agree that the previous data is certainly sufficient to conclude it. The reviewer made a similar suggestion about COMA and CBF3 subcomplexes. In recent years a substantial portion of inner kinetochore components has been reconstituted. In Harrison et al. 2019, the Ctf19 complex structure has been solved. Two copies of the complex were observed. Therefore, the non-centromeric pool of COMA is certainly possible and we now made the adjustments to the text (page 8, lines 219-225) and Fig. 4. Accordingly, we now also modified the abstract (page 1, lines 26-27) and restructured the sections (page 10) to accommodate the different possibility of Cse4 copy numbers. While, fluorescence imaging of CBF3 presents a signal throughout the nuclear region we observed only four copies of Cep3 (part of CBF3). A CBF3 structure also has been resolved by Yan et al. 2018, in which the complex was proposed to exist as a dimer. This translates into four copies of Cep3. Therefore, we find it more suitable to leave all observed Cep3 (CBF3) molecules within a kinetochore model.

(2) The model drawn in Figure 4 makes explicit assumptions about the positioning of the four Cse4 molecules (or two nucleosomes) in each kinetochore relative to the rest of the kinetochore components. Yet, the data shown do not justify this specific arrangement. Lawrimore et al. 2011 claim that the non-centromeric Cse4 nucleosomes must be randomly distributed in the pericentromeric chromatin to evade detection in biochemical tests. Therefore, the nearest-neighbor analysis suggested above will be valuable for gaining new insights into the relative positioning of the centromeric- and non-centromeric Cse4 nucleosomes. A similar analysis for Cep3 and Ctf19 will also be helpful. If stereotypical positioning of these molecules cannot be detected, then the model should be revised accordingly (alternative models that are also consistent with the data can be included).

The reviewer has pointed out that Lawrimore et al. 2011 proposed and justified the existence of a non-centromeric Cse4 pool. This arrangement, also potentially along other inner kinetochore components, makes sense and our data did not indicate it otherwise. Therefore, we now revised our model accordingly by applying changes in the main text on page 10 lines 302-305 __as well as in __Fig. 4.

(3) I suggest one experiment that can help the authors better understand protein organization in one kinetochore. Joglekar et al. 2006 used a dicentric chromosome to isolate single kinetochores on the spindle axis to test the assumption that each kinetochore consists of approximately the same number of molecules of kinetochore proteins. The strains are easy to construct (transform existing strains with a linearized plasmid). Single kinetochores can be seen with a low but reasonable frequency. I leave the decision to perform the experiment to the authors' discretion depending on whether the experiment will be worth the effort in strengthening or enhancing their conclusions.

We performed the suggested experiment using the strain published in Joglekar et al. 2006 (kindly provided by Prof. Kerry Bloom) with Cse4 additionally tagged with mMaple. However, we always observed several super-resolved Cse4 clusters (likely of several kinetochores) overlapping with Nuf2-GFP diffraction-limited signal, therefore unable to assign a single isolated kinetochore to the lagging centromere.

1. Information regarding the degree of correction applied to calculate protein count per half-spindle: It will be helpful to include data regarding the degree of correction applied to the expected and measured numbers of NPC protein as supplementary data so that the readers can see the magnitude of this correction relative to the measured counts.

We would like to clarify that we did not correct the data. Instead, we calibrate the copy number, given that the copy number of Nup188 per NPC is known. We assume the same ratio between localization and copy number applies to both Nup188 and the kinetochore proteins. We now include a new Table S4 listing calibration factors of all experiments shown in Fig. 3.

Minor points:

1. McIntosh et al. JCB 2013 used microtubule plus-ends in serial section electron micrographs of yeast spindles to align the centromeric region and found a disk-shaped structure that roughly corresponds to the size of a single nucleosome ~ 80 nm away from the tip of the microtubule and centered the microtubule axis. The authors should refer to this finding in their discussion of the model that they present with two nucleosomes. In my opinion, this is compelling evidence for a nucleosome-like structure serving as the kinetochore foundation.

We agree with this reviewer's comment. The study, among others, present compelling evidence for a point-centromere. We now included the finding in the discussion on page 10, lines 293-294.

1. As discussed by the authors, the number of Cse4 molecules per kinetochore has been the subject of some controversy. Biochemical data from the Biggins group and ChIPseq data from the Westermann group (Altunkaya et al. 2016 Current Biology) strongly suggest that Cse4 molecules can only be found centered on the centromeric sequence. The latter reference should be included in the discussion.

Thank you for pointing this out. Indeed, this is important. We have now added the relevant reference in the discussion on__ page 10 lines 291-292__.

1. Although microscopy-based methods have estimated anywhere from 1, 2, to 6 Cse4 molecules per kinetochore, these studies generally agree on the stoichiometry between Cse4 and the rest of the kinetochore proteins, e.g., Ndc80 complex proteins are ~ 4-fold more abundant that Cse4, etc. The present study seems to disagree with protein stoichiometry. The authors may find it worthwhile to note this feature of their data.

We now discuss the stoichiometry difference between our results and others on page 11 lines 322-324.

1. Omission of the Dam1 complex from this study is disappointing to me personally, but I am sure that the authors have good reasons for this. They should briefly comment on the absence of the Dam1 complex in this study.

To provide information on the Dam1 complex, we imaged Ask1, a component of the complex. The measured positioning and copy number of the protein are now included in Fig. 2 and Fig. 3 respectively, and described and discussed in respective parts of the manuscript.

Reviewer #1 (Significance (Required)):

Cieslinski and colleagues present a single-molecule localization-based study to define the copy numbers and relative organization of kinetochore proteins in budding yeast. These numbers confirm and significantly refine prior measurements of the same aspects of the kinetochore. They also raise new questions and point to new research directions. The measurements also reveal a model of the protein organization of the budding yeast kinetochore in metaphase. For these reasons, the manuscript is of significant interest to the cell division field.

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

In this study, Cielinski and colleagues have applied single molecule localization microscopy to map the positions of proteins in the yeast kinetochore. This has not been reported previously and this study is both well-conducted and the data appear solid. They also use a modification of this technique to assess the stoichiometry of kinetochore proteins. The results that they obtain are broadly in line with several previous studies that use other methodology. There may be an improvement in accuracy using this new approach that has not been obtained previously and there are some important novel conclusions from this work. I would like the authors to address the following concerns prior to publication:

Major points

1. One interesting finding is that there is a discrepancy in the length of both the MIND and NDC80 complexes (from crystallographic data) with their relative positions. The authors suggest that the outer complexes could be twisted or rotated in respect of the spindle axis. It would be great if the authors could illustrate this in their model (or discuss it in the text), to demonstrate the required angle of twist/rotation of both complexes to account for the discrepancy. A twisted filament structure to the outer kinetochore does have some implications for its response to tension - a key determinant of kinetochore-microtubule attachment. It also may provide some flexibility to the structure under tension.

The discussion about this discrepancy has now been incorporated in the main text, page 9 lines 263-267. For clarity, we only partially reflect this in our schematic model (Fig. 4A; the MIND complex) but we already reflected this in the illustrative structural model in Fig. 4B.

1. For the experiment with cycloheximide, the authors state "Although we observed minor changes in copy numbers, the overall effect of CHX was small." For some proteins, Cse4i for example, there appears to be a significant decrease in intensity (30-40%) after cycloheximide treatment, see Figure S3. While the conclusion that tag maturation does not affect copy number measurements is sound, I suggest modifying this section to reflect the data.

We now modified the section accordingly by pointing out that Cse4i under CHX measurements led to reduction of the signal. The modification can be found on page 8 lines 207-211.

1. Page 5. The statement "These data agree reasonably well with previous diffraction-limited dual-color microscopy studies ..." provides readers with little ability to compare the data. I would like to see a supplementary figure comparing these new data with previous studies, especially those of Joglekar et al 2009, see Figure 3 in this paper.

We thank the reviewer for suggesting such a table. This will allow readers a direct comparison of the data between our study and Joglekar at al. 2009. The comparison can be found in new Table S1 __and __Fig. S4, which are now mentioned on page 5.

1. In terms of the distances quoted, are they in one dimension (as per Jogelkar et al 2009) or in three? The results section is entitled "...positions of kinetochore proteins along the metaphase spindle axis", which suggests a single dimension. Please make this very clear in the results section. In the discussion, is the statement "we mapped the relative positions of 15 kinetochore proteins along the kinetochore axis", which is not entirely clear. It seems from the methods that this is one dimension "...we determined the average distance between the two proteins along the spindle axis. “I suggest clarifying the results section briefly and clearly to indicate that this is a single dimension being measured and also using consistent wording of the axis measured throughout the text.

We agree the previous description may not be clear to the viewers. We now changed the text accordingly in the results section, page 5 lines 129-130.

Minor points:

Abstract: I would drop "all" from "For all major kinetochore proteins...", since full characterisation was performed on 14 proteins (9 in terms of copy number).

We now deleted “all” in the abstract as the reviewer suggested__.__

Page 2: "trough" to through.

Corrected.

Page 2 "S. cerevisiae" to italics

Corrected.

Methods p11. How do the MKY strains relate to common yeast genetic backgrounds? (e.g. are they S288C?).

MKY strains are derivative of S288C. The information was now updated in the Methods section and in Table S2.

Reviewer #2 (Significance (Required)):

This manuscript, together with an accompanying one from Virat et al., are nice complementary studies that provide the first single molecule localization studies of the yeast kinetochore. Although other labs have used super-resolution methods to study individual kinetochore proteins; both of these new studies map distances between many proteins at the kinetochore and thus are able to produce maps of the overall kinetochore structure. Like the previous study using standard resolution methods (Joglekar et al, 2009. Current Biology 19, 694-699); these studies will likely provide a benchmark for future studies on eukaryotic kinetochore architecture, including those in mammalian systems. Additionally, this work will appeal to super-resolution microscopists.

My expertise is as a yeast kinetochore cell 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

In this study, Cielinski and colleagues have applied single molecule localization microscopy to map the positions of proteins in the yeast kinetochore. This has not been reported previously and this study is both well-conducted and the data appear solid. They also use a modification of this technique to assess the stoichiometry of kinetochore proteins. The results that they obtain are broadly in line with several previous studies that use other methodology. There may be an improvement in accuracy using this new approach that has not been obtained previously and there are some important novel conclusions from this work. I would like the authors to address the following concerns prior to publication:

Major points

1. One interesting finding is that there is a discrepancy in the length of both the MIND and NDC80 complexes (from crystallographic data) with their relative positions. The authors suggest that the outer complexes could be twisted or rotated in respect of the spindle axis. It would be great if the authors could illustrate this in their model (or discuss it in the text), to demonstrate the required angle of twist/rotation of both complexes to account for the discrepancy. A twisted filament structure to the outer kinetochore does have some implications for its response to tension - a key determinant of kinetochore-microtubule attachment. It also may provide some flexibility to the structure under tension.
2. For the experiment with cycloheximide, the authors state "Although we observed minor changes in copy numbers, the overall effect of CHX was small." For some proteins, Cse4i for example, there appears to be a significant decrease in intensity (30-40%) after cycloheximide treatment, see Figure S3. While the conclusion that tag maturation does not affect copy number measurements is sound, I suggest modifying this section to reflect the data.
3. Page 5. The statement "These data agree reasonably well with previous diffraction-limited dual-color microscopy studies ..." provides readers with little ability to compare the data. I would like to see a supplementary figure comparing these new data with previous studies, especially those of Joglekar et al 2009, see Figure 3 in this paper.
4. In terms of the distances quoted, are they in one dimension (as per Jogelkar et al 2009) or in three? The results section is entitled "...positions of kinetochore proteins along the metaphase spindle axis", which suggests a single dimension. Please make this very clear in the results section. In the discussion, is the statement "we mapped the relative positions of 15 kinetochore proteins along the kinetochore axis", which is not entirely clear. It seems from the methods that this is one dimension "...we determined the average distance between the two proteins along the spindle axis."I suggest clarifying the results section briefly and clearly to indicate that this is a single dimension being measured and also using consistent wording of the axis measured throughout the text.

Minor points:

Abstract: I would drop "all" from "For all major kinetochore proteins...", since full characterisation was performed on 14 proteins (9 in terms of copy number).

Page 2: "trough" to through.

Page 2 "S. cerevisiae" to italics

Methods p11. How do the MKY strains relate to common yeast genetic backgrounds? (e.g. are they S288C?).

Significance

This manuscript, together with an accompanying one from Virat et al., are nice complementary studies that provide the first single molecule localization studies of the yeast kinetochore. Although other labs have used super-resolution methods to study individual kinetochore proteins; both of these new studies map distances between many proteins at the kinetochore and thus are able to produce maps of the overall kinetochore structure. Like the previous study using standard resolution methods (Joglekar et al, 2009. Current Biology 19, 694-699); these studies will likely provide a benchmark for future studies on eukaryotic kinetochore architecture, including those in mammalian systems. Additionally, this work will appeal to super-resolution microscopists.

My expertise is as a yeast kinetochore cell 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 #1

Evidence, reproducibility and clarity

The authors have developed a rigorous methodology for using single-molecule imaging of exogenously labeled kinetochore proteins to count and estimate their copy numbers and the average distance from the kinetochore protein Spc105. Although the method is technically sound, its application to the kinetochore raises some crucial questions below. My biggest concern is the effect of non-centromeric pools of the centromeric proteins Cse4, Cep3, and Ctf19 on the estimated copy number per kinetochore. The authors should be able to address most, if not all, questions by presenting a more in-depth data analysis.

1. Accounting for tilt of the yeast spindle relative to the image plane: It is not clear to me how the authors ascertain whether the spindle being imaged is nearly parallel to the image plane. In the companion fission yeast study, spindle poles are used for this purpose, but this study seems to rely only on the labeled kinetochore proteins. The criteria used to select the in-plane spindles should be clearly defined.
2. The effects of PSF depth on counting kinetochore proteins: The authors use a well-characterized nuclear pore protein as the reference to estimate kinetochore protein counts per half-spindle. Although this method appears rigorous in principle, I am unsure about the effect of the spatial distribution of kinetochores on the accuracy of the estimated number. Nuclear pore proteins are all localized within an < 100 nm3 volume. Therefore, all proteins within an in-focus nuclear pore will also be in focus. This is not the case with yeast kinetochores, especially in metaphase. A fraction of the kinetochores is likely to be > 100 nm away from the focal plane even when the spindle is perfectly parallel to the focal plane. A discussion of this possibility, its effect on the protein count/distance estimates, and any mitigating factors is essential to highlight the caveats associated with the conclusions.
3. Presentation of the cross-correlation analysis: The authors use cross-correlation for an unbiased calculation of the axial separation between a protein of interest and Cse4, but I am curious about the structure of the underlying data, and the intensity image in Figure 1 is not easy to examine. It will be helpful to include more analysis of the underlying data for at least a subset of the proteins (e.g., proteins at short, intermediate, and long distances from Cse4) as supplementary data.
   • The authors should include X and Y projections of the cross-correlation function.
   • Do the widths of cross-correlation functions (i.e., their spread perpendicular to the spindle axis) match across all proteins and experiments? This should be an almost invariant characteristic of the measurements, assuming that proteins within each kinetochore tightly cluster around the 25 nm microtubule. This line of thinking makes the large width of the cross-correlation shown in Figure 1 somewhat surprising.
   • It will also be interesting to test if the correlation between the positions of Spc105 molecules, especially perpendicular to the spindle axis, is comparable to the known separations between adjacent microtubules in the yeast spindle (the authors could use Winey et al. 1995 for serial-section EM of yeast spindles for comparison).
4. Cse4 count (4 per kinetochore) and the model presented: One of the surprising conclusions of the study is that there are two nucleosomes associated with each microtubule attachment, with Mif2/CENP-C potentially interacting with both nucleosomes. There are two critical issues that the authors must consider.
   • (1) Fluorescent protein chimeras of Cse4 and CBF3 and COMA complex members do not exclusively localize to kinetochores. Biochemical studies show that both Cse4 and CBF3 proteins interact with non-centromeric DNA, e.g., see work from the Biggins lab regarding Cse4 over-expression and also from the Henikoff group that used ChIP-seq. I can't think of a similar reference for the CBF3 complex, but the DNA-binding proteins are also likely to interact with other parts of the genome. The non-centromeric protein is visible as a significant background fluorescence in wide-field microscopy, e.g., see Cep3 localization here: https://images.yeastrc.org/imagerepo/viewExperiment.do?id=202308&experimentGroupOffset=3&experimentOffset=0&experimentGroupSize=3 Similar background fluorescence can be detected for Cse4 and Ctf19. This extra-centromeric localization of Cse4, Cep3, and Ctf19 makes it possible that the protein counts included by the authors are "contaminated" to some extent by the extra-centromeric protein. The authors should discuss this possibility and how it might affect their counts.
   • (2) The model drawn in Figure 4 makes explicit assumptions about the positioning of the four Cse4 molecules (or two nucleosomes) in each kinetochore relative to the rest of the kinetochore components. Yet, the data shown do not justify this specific arrangement. Lawrimore et al. 2011 claim that the non-centromeric Cse4 nucleosomes must be randomly distributed in the pericentromeric chromatin to evade detection in biochemical tests. Therefore, the nearest-neighbor analysis suggested above will be valuable for gaining new insights into the relative positioning of the centromeric- and non-centromeric Cse4 nucleosomes. A similar analysis for Cep3 and Ctf19 will also be helpful. If stereotypical positioning of these molecules cannot be detected, then the model should be revised accordingly (alternative models that are also consistent with the data can be included).
   • (3) I suggest one experiment that can help the authors better understand protein organization in one kinetochore. Joglekar et al. 2006 used a dicentric chromosome to isolate single kinetochores on the spindle axis to test the assumption that each kinetochore consists of approximately the same number of molecules of kinetochore proteins. The strains are easy to construct (transform existing strains with a linearized plasmid). Single kinetochores can be seen with a low but reasonable frequency. I leave the decision to perform the experiment to the authors' discretion depending on whether the experiment will be worth the effort in strengthening or enhancing their conclusions.
5. Information regarding the degree of correction applied to calculate protein count per half-spindle: It will be helpful to include data regarding the degree of correction applied to the expected and measured numbers of NPC protein as supplementary data so that the readers can see the magnitude of this correction relative to the measured counts.

Minor points:

1. McIntosh et al. JCB 2013 used microtubule plus-ends in serial section electron micrographs of yeast spindles to align the centromeric region and found a disk-shaped structure that roughly corresponds to the size of a single nucleosome ~ 80 nm away from the tip of the microtubule and centered the microtubule axis. The authors should refer to this finding in their discussion of the model that they present with two nucleosomes. In my opinion, this is compelling evidence for a nucleosome-like structure serving as the kinetochore foundation.
2. As discussed by the authors, the number of Cse4 molecules per kinetochore has been the subject of some controversy. Biochemical data from the Biggins group and ChIPseq data from the Westermann group (Altunkaya et al. 2016 Current Biology) strongly suggest that Cse4 molecules can only be found centered on the centromeric sequence. The latter reference should be included in the discussion.
3. Although microscopy-based methods have estimated anywhere from 1, 2, to 6 Cse4 molecules per kinetochore, these studies generally agree on the stoichiometry between Cse4 and the rest of the kinetochore proteins, e.g., Ndc80 complex proteins are ~ 4-fold more abundant that Cse4, etc. The present study seems to disagree with protein stoichiometry. The authors may find it worthwhile to note this feature of their data.
4. Omission of the Dam1 complex from this study is disappointing to me personally, but I am sure that the authors have good reasons for this. They should briefly comment on the absence of the Dam1 complex in this study.

Significance

Cieslinski and colleagues present a single-molecule localization-based study to define the copy numbers and relative organization of kinetochore proteins in budding yeast. These numbers confirm and significantly refine prior measurements of the same aspects of the kinetochore. They also raise new questions and point to new research directions. The measurements also reveal a model of the protein organization of the budding yeast kinetochore in metaphase. For these reasons, the manuscript is of significant interest to the cell division field.

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

1. General Statements [optional]

Thank you for the peer review of our manuscript entitled “Hypoxia causes pancreatic β____-cell dysfunction by activating a transcriptional repressor BHLHE40” (RC-2022-01560). We greatly appreciate the reviewers’ constructive suggestions and your invitation to revise the manuscript. Below, we address the comments point-by-point and provide details of the changes we are planning to or have implemented. We believe that the revision plan will meet with the approval of the editor and reviewers. We also would be happy to respond to any further questions and comments that you may have.

2. Description of the planned revisions

Response to comments of reviewer 1

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

The authors examine the role of hypoxia-induced transcriptional repression in mediating loss of β-cell function in type2 diabetes. Transcriptional profiling of mouse and human islets exposed to low oxygen conditions revealed downregulation of β-cell identity and oxidative phosphorylation genes, and upregulation of genes associated with hypoxia. Identification of genes commonly upregulated in Min6 cells, mouse and human islets under hypoxic conditions, revealed induction of two transcriptional repressors BHLHE40 and ATF3. The authors further show that Bhlhe40 deficiency rendered β-cells resistant to hypoxic stress and restored glucose- and KCl-stimulated insulin secretion. This rescue in β-cell function was at least, in part, due to restoration of ATP generation and exocytosis of insulin granules. Furthermore, transcriptional profiling of Min6 cells overexpressing Bhlhe40 indicated down-regulation of key β-cell genes including Mafa. The authors elegantly show that BHLHE40 blocks PDX1 binding to Mafa transcription start site by binding to two E-box sites within the Mafa promoter/enhancer region. Lastly, Cre-mediated β-cell-specific deletion of Bhlhe40 in ob/ob mice restored expression of Mafa and exocytotic genes, accompanied by improvements in ATP generation and insulin secretion.

Major comments 1. The authors conclude that BHLHE40 regulates insulin secretion at two key steps: ATP generation and exocytosis. However, insulin secretory profiles with glucose and KCl seem to be similar with genetic manipulations of Bhlhe40 both in vivo and ex vivo. As the authors indicate in line 176, this suggests a more prominent role of BHLHE40 in regulating exocytotic events downstream of Ca2+ influx. Further experiments are therefore necessary to adequately address the effects on ATP generation. Given the observation that PGC1____α, a regulator of mitochondrial biogenesis is suppressed by BHLHE40, mitochondrial assessments would be crucial. Additionally, the effect on mitochondrial mass in Fig 3K seem to be marginal and need to be confirmed using additional measurements listed below.

We appreciate your constructive suggestion. According to your suggestions, we will explore the role of BHLHE40 in mitochondrial function in more detail.

a. In fig 3F, the authors show no change in KCl stimulated Ca2+ influx. Glucose stimulated Ca2+ influx needs to be examined to confirm regulation of ATP generation.

We thank the reviewer for pointing this out. We performed the experiment according to the suggestion. Please see section 3.

b. OXPHOS subunits, TOM20 levels by western blotting

We thank the reviewer for pointing this out. We will perform Western blotting to check the protein levels of OXPHOS subunits and TOM20 in control (Ctrl) and Bhlhe40 knockdown (KD) MIN6 cells cultured under 20% or 5% O­­­2.

c. mtDNA content, transcript levels by qRT-PCR

We will perform qRT-PCR to check the mtDNA content in Ctrl and Bhlhe40 KD MIN6 cells cultured under 20% or 5% O­­­2.

d. Functional assessments: Changes in mitochondrial membrane potential or oxygen consumption

We will evaluate mitochondrial membrane potential by MitoTracker Red staining in Ctrl and Bhlhe40 KD MIN6 cells cultured under 20% or 5% O­­­2.

2. Data presented in Figure 4 and 5 indicates transcriptional repression of Mafa by BHLHE40 as a mechanism of beta-cell dysfunction under hypoxic conditions. However, additional experiments are necessary to confirm that repression of PDX1-Mafa binding specifically is responsible for defects in GSIS -

a. Fig 5G shows inhibition of PDX1-binding to Mafa with overexpression of Bhlhe40. This needs to be confirmed under hypoxic conditions.

We thank the reviewer for pointing this out. Hypoxia for 16 to 24 hours decreases the expression levels of Pdx1 in mouse islets and MIN6 cells (Figure 1A and Sato Y., PLoS One 2014). In that condition, it is difficult to assess whether the reduction in PDX1 binding to Mafa enhancer is attributed to inhibition of PDX1 binding or PDX1 downregulation. Therefore, we will aim to determine the hypoxia exposure time during which Mafa expression is downregulated but Pdx1 expression is not affected. If we fail to identify the time, we plan to generate Pdx1-overexpressing MIN6 cell lines and evaluate PDX1 binding to Mafa enhancer in hypoxic conditions.

Sato Y. et al. Moderate Hypoxia Induces β-Cell Dysfunction with HIF-1–Independent Gene Expression Changes. PLoS One. 2014;9(12):e114868.

b. Fig 4H and 4I show restoration of insulin secretion normalized to total protein with AAV-Mafa. This needs to be supplemented with insulin content as MAFA has been implicated in regulating insulin gene expression (PMID: 25500951).

We thank the reviewer for pointing this out and agree with the comment on our original manuscript. We will evaluate insulin content by insulin ELISA assay in samples from AAV-Ctrl and AAV-Mafa-overexpressing MIN6 cells cultured under 20% or 5% O2. If the insulin content is affected by Mafa overexpression, insulin secretion will be adjusted by the intracellular insulin content.

c. qRT-PCR of exocytosis genes and ATP generation with hypoxia and AAV-Mafa.

We thank the reviewer for pointing this out. We will evaluate expression of exocytosis genes and ATP generation in AAV-Ctrl and AAV-Mafa-overexpressing MIN6 cells cultured under 20% or 5% O2.

d. Would mutation of A and C E-box sites restore PDX1 binding to Mafa TF region under hypoxia?

To address this question, we plan to introduce mutations in A and C sites of the Mafa gene in MIN6 cells by using CRISPR-Cas9 technology and then to examine PDX1 binding to the Mafa gene by ChIP assay under hypoxic conditions.

3. β-dedifferentiation has been proposed to be involved in loss of insulin secretion in T2D (PMID: 22980982, 16123366). One can speculate that transcriptional repression of Mafa by BHLHE40 is a component of a larger dedifferentiation phenomenon occurring under hypoxia, as other ____β-cell genes were decreased with hypoxia (Fig 1A) and Bhlhe40-OE in Fig 4A. Identifying differences in dedifferentiation and ____β-cell disallowed genes with Bhlhe40 overexpression (RNA seq, qRT-PCR) would therefore potentially reveal a dedifferentiation mechanism.

We thank the reviewer for pointing this out. Please see section 3.

4. The authors identify Atf3 as another transcriptional repressor enriched under hypoxia although to a lesser degree than Bhlhe40. The role of ATF3 in hypoxia-induced apoptosis and adaptive UPR has been previously suggested (PMID: 20519332, 20349223). Additionally, hypoxia represses adaptive UPR in models of T2D and drives ____β-cell apoptosis (PMID: 27039902). The authors discuss the role of ATF3 under hypoxia in the discussion (lines 319-324) and addressing these research gaps regarding ATF3 function would be insightful.

We are grateful for the reviewer’s comment. We will generate Atf3 knockdown MIN6 cell lines and examine the effect of ATF3 on hypoxia-induced apoptosis by PI/AnnexinV staining. If ATF3 is involved in hypoxia-induced apoptosis, we will also measure the mRNA expression levels involved in adaptive UPR.

Minor comments 1. In Fig 2E, increasing replicates would confirm no induction of Bhlhe40 with Thapsigargin.

Thank you for pointing out this issue. We will perform additional experiments to confirm the effect of thapsigargin on Bhlhe40 expression.

2. In Fig 2B, BHLHE40 bands need to be quantified to show time-dependent increase in protein levels.

Thank you for pointing out this issue. Please see section 3.

3. In Fig 3C, insulin content needs to be shown with Bhlhe40-OE as in Fig 3B with hypoxia.

Thank you for pointing out this issue. Please see section 3.

4. In Fid 4E-F, band intensities need to quantified by densitometry to determine degree of downregulation of MAFA.

We performed three independent experiments for Figure 4E. Please see section 3. We plan to perform additional experiments to determine the expression levels of MAFA (Figure 4F).

5. In Fig4H and 6G, insulin content needs to be shown as stated above.

We thank the reviewer for pointing this out. We will perform additional experiments to check the insulin content in these settings.

6. In Supplemental Figure 3C, apoptosis induced by hypoxia was assessed by PI staining that detects late apoptosis. No significant changes were observed with Bhlhe40-KD, but additional cell death assessments can be used to confirm that B40 does not affect ____β-cell death.

We thank the reviewer for mentioning this issue. To address this concern, we plan to investigate the effects of Bhlhe40 KD on the number of Annexin V-positive cells (early apoptosis) and cleaved (activated) caspase 3 expression in Ctrl and Bhlhe40 KD MIN6 cells under hypoxic conditions.

7. It would be interesting to see the rates of diabetes incidence in Bhlhe40KO: ob/ob mice and if Bhlhe40 deficiency protects against or delays development of diabetes.

Thank you for mentioning this issue. Please see section 3.

8. Knockdown efficiency shown in Supplementary figure 3A needs to be estimated by quantifying band intensities.

We plan to perform additional experiments to quantify the band intensities.

9. Line 43 should say "...reversed defects in insulin secretion."

We apologize for our incorrect explanation in the original manuscript. We have corrected this error in the text. Please see section 3.

Reviewer #1 (Significance (Required)):

The data presented provides novel mechanistic insights into the role of hypoxia in β-cell dysfunction. Studies in multiple models of type 2 diabetes (T2D) have shown the loss of signature β-cell genes including Ins1, Pdx1, Mafa, Slc2a2 as a result of excess nutrient stimulation and hypoxia; the precise causal mechanisms, however, still remain to be determined (PMID: 22980982, 28270834). A previous paper from the same group demonstrated downregulation of β-cell signature genes with hypoxia by a HIF1α independent mechanism (PMID: 25503986). Data presented in this report extend those observations and reveal a previously unappreciated role for transcriptional repressor BHLHE40 in the downregulation of a key β-cell gene Mafa. As the authors have identified additional transcriptional repressors including ATF3 and differentially expressed genes in both human and rodent β-cells, this paper would be of great value in understanding the effects of hypoxia. Moreover, studies in mouse models of T2D extend the association of BHLHE40 to clinical β-cell dysfunction and diabetes.

My areas of interest are pancreatic β-cell and mitochondrial physiology. GSE analysis and repression of PGC1α by BHLHE40, as appropriately discussed by the authors, point towards impaired mitochondrial function and ATP generation. Additional experiments would greatly support the role of BHLHE40 in mitochondrial dysfunction under hypoxia.

We thank the reviewer for his/her valuable comments.

Response to comments of reviewer 2

Reviewer #2 (Evidence, reproducibility and clarity (Required)): This study examines the role of BHEH40 in beta-cell function and its role in mediating the changes with 'hypoxia'. Most of the studies use 5% oxygen which is probably close to normal oxygen tension for islets, although it is not great for islet survival once the islets are removed from their normal vasculature. The human islet studies use 2% oxygen which would be actual hypoxia.

Major comments: Why were different oxygen concentrations used for mouse and human islets? What were the effects of 5% oxygen in human islets? Why was 5% oxygen chosen? 5% is close to normal oxygen tension that islets are exposed to in vivo, whereas 2% is not physiological.

We apologize for the lack of explanation in the original manuscript. Please see section 3.

If there was only 1 human donor, are the 2 and 3 RNA-seq technical replicates? If so, they do not show high replicability. Please discuss.

The reviewer is correct. We analyzed human islets from one donor for RNA-seq (Figure 1A). It would be preferable to obtain additional data derived from different human donor samples. Therefore, we plan to analyze human islets from another donor to show the replicability of increased expression of Bhlhe40 under hypoxic conditions.

Are the sequencing results from individual mice? Were the same mouse's islets used for normal and 5% oxygen or are they all different animals?

We apologize for the lack of explanation in the original manuscript. Results of RNA-seq were obtained from different animals. Please see section 3.

The whole gels with appropriate size markers need to be shown for all Westerns - they are not able to be appropriately reviewed in their current formats.

We apologize for the mistake. Please see section 3.

The histology in Figure 1K does not appear to match with the Western blot results in 1B, 1C which show a smaller but still clear band in the control 20% conditions, and in figures 1I and J in ob/ob and db/db controls. Lower power views showing most of the pancreas with a zoom-in shot of an example islet would be more appropriate. The immunofluorescence should be repeated to also include insulin so that beta-cells can be identified.

We thank the reviewer for mentioning this issue. We will repeat the immunohistochemical analysis according to the reviewer’s comments. We also plan to show the data on insulin staining.

What was the ____β-cell deletion efficiency of the knockdown mouse?

We apologize for the lack of explanation in the original manuscript. Please see section 3.

In the setting of hypoxia, would it be 'clinically' beneficial to have increased insulin secretion and thus metabolic demand? Please discuss.

We thank the reviewer for mentioning this important issue. Our results show that BHLHE40 controls at least two steps of insulin secretion: exocytosis and ATP generation. A relatively smaller reduction of Ppargc1a might suggest a more prominent role of BHLHE40 in regulating exocytosis rather than ATP generation (oxygen consumption step). Chronic hyperglycemia induces b-cell damage and impairs insulin secretion, a process known as glucotoxicity (Weir GC, Diabetes 2004, 2020). We believe that inactivation of BHLHE40 may help to reduce glucotoxicity by increasing insulin secretion. However, we would like to discuss this topic in more detail once we have investigated the roles of BHLHE40 in ATP generation (as suggested by reviewer 1).

Weir GC. et al. Five stages of evolving beta-cell dysfunction during progression to diabetes. Diabetes. 2004;53 Suppl 3:S16-21.

Weir GC. Glucolipotoxicity, β-Cells, and Diabetes: The Emperor Has No Clothes. Diabetes. 2020;69(3):273-278.

- Are the key conclusions convincing? Hard to assess the data in some cases - see above. - Should the authors qualify some of their claims as preliminary or speculative, or remove them altogether? Yes, some of the conclusions are too strongly worded. An example is "However, hyperactivation of HIF in ____β-cells impairs insulin secretion by switching glucose metabolism from aerobic oxidative phosphorylation to anaerobic glycolysis (14-16)," this is too broad a statement. Hyperactivation of HIF in ____β-cells BY VHL DELETION impairs insulin secretion by that mechanism. Other ways of increasing HIF in ____β-cells do not all have this effect. So, the following part "suggesting that activation of HIF underlies ____β-cell dysfunction and glucose intolerance in hypoxia." Is not warranted. It would be fair to say "suggesting that unregulated over-activation of HIF may cause ____β-cell dysfunction.

We apologize for our incorrect explanation in the original manuscript. Please see section 3.

The paper is not off to a good start when the author spell Abstract as Abstruct - it suggests a spell-check was not performed. For the last sentence of the abstract, 'and its implication' - what implication?

We apologize for the typo and the poor description in the original manuscript. Please see section 3.

Line 64 High glucose conditions generate RELATIVE, not absolute hypoxia in beta-cells. This statement should also be referenced.

We apologize for our inaccurate explanation in the original manuscript. Please see section 3.

- Would additional experiments be essential to support the claims of the paper? See above.

- Are the data and the methods presented in such a way that they can be reproduced? Not enough detail for methods, but what is presented looks OK.

- Are the experiments adequately replicated and statistical analysis adequate? Unclear, see above.

- Specific experimental issues that are easily addressable.

- Are prior studies referenced appropriately? No, only the body of work on VHL, not HIFs.

- Are the text and figures clear and accurate? See above comments.

- Do you have suggestions that would help the authors improve the presentation of their data and conclusions? See comments about gels etc above.

We thank the reviewer for his/her valuable comments. Please see section 3 regarding to references.

3. Description of the revisions that have already been incorporated in the transferred manuscript

The authors conclude that BHLHE40 regulates insulin secretion at two key steps: ATP generation and exocytosis. However, insulin secretory profiles with glucose and KCl seem to be similar with genetic manipulations of Bhlhe40 both in vivo and ex vivo. As the authors indicate in line 176, this suggests a more prominent role of BHLHE40 in regulating exocytotic events downstream of Ca2+ influx. Further experiments are therefore necessary to adequately address the effects on ATP generation. Given the observation that PGC1____α, a regulator of mitochondrial biogenesis is suppressed by BHLHE40, mitochondrial assessments would be crucial. Additionally, the effect on mitochondrial mass in Fig 3K seem to be marginal and need to be confirmed using additional measurements listed below.

a. In fig 3F, the authors show no change in KCl stimulated Ca2+ influx. Glucose stimulated Ca2+ influx needs to be examined to confirm regulation of ATP generation.

We thank the reviewer for pointing this out. In accordance with the reviewer’s comments, we examined the glucose-stimulated Ca2+ influx and found that the influx stimulated by 22mM glucose in Bhlhe40-overexpressing (OE) MIN6 cells was significantly smaller than that in control MIN6 cells (Figure 3, J and K). We have added this information to the manuscript, as follows: “In addition, glucose-stimulated [Ca2+]i levels were significantly attenuated by Bhlhe40 overexpression (Figure 3, J and K). These results indicate that BHLHE40 suppresses glucose-stimulated ATP generation and the increase of [Ca2+]i levels in MIN6 cells” (lines 197 to 200).

3. β-dedifferentiation has been proposed to be involved in loss of insulin secretion in T2D (PMID: 22980982, 16123366). One can speculate that transcriptional repression of Mafa by BHLHE40 is a component of a larger dedifferentiation phenomenon occurring under hypoxia, as other ____β-cell genes were decreased with hypoxia (Fig 1A) and Bhlhe40-OE in Fig 4A. Identifying differences in dedifferentiation and ____β-cell disallowed genes with Bhlhe40 overexpression (RNA seq, qRT-PCR) would therefore potentially reveal a dedifferentiation mechanism.

We thank the reviewer for pointing this out. To check whether genes involved in dedifferentiation and the expression of b-cell disallowed genes are controlled by BHLHE40, we examined the expression of these genes in Bhlhe40 OE MIN6 cells and found that it was not increased by BHLHE40. However, because the findings were detected under limited experimental conditions, at this point we cannot conclude that BHLHE40 does not cause dedifferentiation of b-cells and induction of b-cell disallowed genes.

Minor comments 2. In Fig 2B, BHLHE40 bands need to be quantified to show time-dependent increase in protein levels.

Thank you for pointing out this issue. We performed three independent experiments and showed statistically significant upregulation of BHLHE40 (Figure 2B).

3. In Fig 3C, insulin content needs to be shown with Bhlhe40-OE as in Fig 3B with hypoxia.

Thank you for pointing out this issue. Bhlhe40 OE did not affect the insulin content in MIN6 cells (Supplemental Figure 3F).

4. In Fid 4E-F, band intensities need to quantified by densitometry to determine degree of downregulation of MAFA.

We performed three independent experiments for Figure 4E. Bhlhe40 OE led to a 67.4% decrease in the expression of MAFA in MIN6 cells (Figure 4E).

7. It would be interesting to see the rates of diabetes incidence in Bhlhe40KO: ob/ob mice and if Bhlhe40 deficiency protects against or delays development of diabetes.

Thank you for pointing out this issue. We found that nonfasting blood glucose concentrations were similar in Ctrl:ob/ob (239.9 ± 19.6 mg/dl; n = 9) and bB40KO:ob/ob mice (215.2 ± 17.1 mg/dl; n = 6) at 12 weeks of age (Supplemental Figure 5F). We have added this information to the revised manuscript, as follows: “In these mice, BHLHE40 deficiency in b-cells had no effect on obesity (Figure 6B), insulin sensitivity (Supplemental Figure 5E), or nonfasting glucose concentrations (Supplemental Figure 5F)” (lines 282 to 285).

9. Line 43 should say "...reversed defects in insulin secretion."

We apologize for our incorrect explanation in the original manuscript and have corrected it accordingly.

Response to comments of reviewer 2

Reviewer #2 (Evidence, reproducibility and clarity (Required)): This study examines the role of BHEH40 in beta-cell function and its role in mediating the changes with 'hypoxia'. Most of the studies use 5% oxygen which is probably close to normal oxygen tension for islets, although it is not great for islet survival once the islets are removed from their normal vasculature. The human islet studies use 2% oxygen which would be actual hypoxia.

Major comments: Why were different oxygen concentrations used for mouse and human islets? What were the effects of 5% oxygen in human islets? Why was 5% oxygen chosen? 5% is close to normal oxygen tension that islets are exposed to in vivo, whereas 2% is not physiological.

We apologize for the lack of explanation in the original manuscript. We previously reported that hypoxic responses occur at 5% to 7% oxygen tension in MIN6 cells and mouse islets and that 3% hypoxia for 24 hours markedly increases MIN6 cell death (Sato Y., J Biol Chem 2011, Sato Y., PLoS One 2014). We also examined whether hypoxic responses occur at the same oxygen tension in human islets. Interestingly, in human islets, exposure to 2% but not 5% oxygen tension induced the upregulation of the SLC2A1 gene without apparent cell death. Another group also reported a hypoxic response in human islets in 2% oxygen (Puri S., Genes Dev. 2013). We have no adequate explanation as to why hypoxic responses occur at different oxygen tensions in mouse and human islets, but because of these findings, we used 5% oxygen in MIN6 cells and mouse islets and 2% oxygen in human islets. We have added this information to the text (lines 98 to 104) and Supplemental Figure 1A.

Sato, Y. et al. Cellular hypoxia of pancreatic beta-cells due to high levels of oxygen consumption for insulin secretion in vitro. J Biol Chem. 2011;286(14):12524-32.

Sato, Y. et al. Moderate hypoxia induces β-cell dysfunction with HIF-1-independent gene expression changes. PLoS One. 2014;9(12):e114868.

Puri S. VHL-mediated disruption of Sox9 activity compromises β-cell identity and results in diabetes mellitus. Genes Dev. 2013;27(23):2563-2575.

Are the sequencing results from individual mice? Were the same mouse's islets used for normal and 5% oxygen or are they all different animals?

We apologize for the lack of explanation in the original manuscript. Each sample was derived from different mice. We have clarified this point by describing “n = 3” as “n = 3 mice/group” and have added this information to the legends of Figure 1 and Supplemental Figure 1.

The whole gels with appropriate size markers need to be shown for all Westerns - they are not able to be appropriately reviewed in their current formats.

We apologize for the inappropriate presentation of the data. We have included whole gel images with molecular weight markers as supplemental material.

What was the ____β-cell deletion efficiency of the knockdown mouse?

We apologize for not including this information. Although we showed the expression levels of Bhlhe40 in the islets of bB40KO mice (original Supplemental Figure 5B), we did not explain the data in the text. The deletion efficiency in the islets was 74.1%. We have added this information to the revised text (lines 274 to 275).

- Should the authors qualify some of their claims as preliminary or speculative, or remove them altogether? Yes, some of the conclusions are too strongly worded. An example is "However, hyperactivation of HIF in ____β-cells impairs insulin secretion by switching glucose metabolism from aerobic oxidative phosphorylation to anaerobic glycolysis (14-16)," this is too broad a statement. Hyperactivation of HIF in ____β-cells BY VHL DELETION impairs insulin secretion by that mechanism. Other ways of increasing HIF in ____β-cells do not all have this effect. So, the following part "suggesting that activation of HIF underlies ____β-cell dysfunction and glucose intolerance in hypoxia." Is not warranted. It would be fair to say "suggesting that unregulated over-activation of HIF may cause ____β-cell dysfunction.

We apologize for our incorrect explanation in the original manuscript. We have corrected this accordingly, as follows: “However, hyperactivation of HIF in b-cells by von Hippel-Lindau (VHL) deletion impairs insulin secretion by switching glucose metabolism from aerobic oxidative phosphorylation to anaerobic glycolysis (15-17), suggesting that unregulated overactivation of HIF may cause b-cell dysfunction (12, 14)” (lines 73 to 77).

The paper is not off to a good start when the author spell Abstract as Abstruct - it suggests a spell-check was not performed. For the last sentence of the abstract, 'and its implication' - what implication?

We apologize for the typo and the poor description in the original manuscript. The spell check was performed by an editing company, and we did not notice the error. We have changed the last sentence of the Abstract, as follows: “Collectively, this work identifies BHLHE40 as a key hypoxia-induced transcriptional repressor in b-cells that negatively regulates insulin secretion by suppressing MAFA expression” (lines 47 to 49).

Line 64 High glucose conditions generate RELATIVE, not absolute hypoxia in beta-cells. This statement should also be referenced.

We apologize for our inaccurate explanation in the original manuscript. We have corrected this accordingly, as follows: “high glucose conditions generate relative hypoxia in b-cells because these cells consume large amounts of oxygen (Sato Y., J Biol Chem 2011; Bensellam M., PLoS One 2012; Bensellam M., J Endocrinol 2018; Ilegems E, Sci Transl Med 2022)” (lines 64 to 65).

Sato, Y. et al. Cellular hypoxia of pancreatic beta-cells due to high levels of oxygen consumption for insulin secretion in vitro. J Biol Chem. 2011;286(14):12524-32.

Bensellam, M. et al. Glucose-induced O₂ consumption activates hypoxia inducible factors 1 and 2 in rat insulin-secreting pancreatic beta-cells. PLoS One 2012;7(1):e29807.

Bensellam M. et al. Mechanisms of β-cell dedifferentiation in diabetes: recent findings and future research directions. J Endocrinol. 2018;236(2):R109-R143

Ilegems, E. et al. HIF-1α inhibitor PX-478 preserves pancreatic β cell function in diabetes. Sci Transl Med. 2022;14(638):eaba9112.

- Are prior studies referenced appropriately? No, only the body of work on VHL, not HIFs.

We apologize for our inadequate references about the involvement of HIFs in hypoxia-induced β-cell dysfunction. We have included the following references in the text (line 77):

Catrina, S.B. et al. Hypoxia and hypoxia-inducible factors in diabetes and its complications. Diabetologia. 2021;64(4):709-716

Gulton JE. Hypoxia-inducible factors and diabetes. J Clin Invest. 2020; 130(10):5063-5073

4. Description of analyses that authors prefer not to carry out

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 study examines the role of BHEH40 in beta-cell function and its role in mediating the changes with 'hypoxia'. Most of the studies use 5% oxygen which is probably close to normal oxygen tension for islets, although it is not great for islet survival once the islets are removed from their normal vasculature. The human islet studies use 2% oxygen which would be actual hypoxia.

Significance

Major comments:

Why were different oxygen concentrations used for mouse and human islets? What were the effects of 5% oxygen in human islets? Why was 5% oxygen chosen? 5% is close to normal oxygen tension that islets are exposed to in vivo, whereas 2% is not physiological.

If there was only 1 human donor, are the 2 and 3 RNA-seq technical replicates? If so, they do not show high replicability. Please discuss.

Are the sequencing results from individual mice? Were the same mouse's islets used for normal and 5% oxygen or are they all different animals?

The whole gels with appropriate size markers need to be shown for all Westerns - they are not able to be appropriately reviewed in their current formats.

The histology in Figure 1K does not appear to match with the Western blot results in 1B, 1C which show a smaller but still clear band in the control 20% conditions, and in figures 1I and J in ob/ob and db/db controls. Lower power views showing most of the pancreas with a zoom-in shot of an example islet would be more appropriate. The immunofluorescence should be repeated to also include insulin so that beta-cells can be identified.

What was the β-cell deletion efficiency of the knockdown mouse?

In the setting of hypoxia, would it be 'clinically' beneficial to have increased insulin secretion and thus metabolic demand? Please discuss.

• Are the key conclusions convincing? Hard to assess the data in some cases - see above.
• Should the authors qualify some of their claims as preliminary or speculative, or remove them altogether? Yes, some of the conclusions are too strongly worded. An example is "However, hyperactivation of HIF in β-cells impairs insulin secretion by switching glucose metabolism from aerobic oxidative phosphorylation to anaerobic glycolysis (14-16)," this is too broad a statement. Hyperactivation of HIF in β-cells BY VHL DELETION impairs insulin secretion by that mechanism. Other ways of increasing HIF in β-cells do not all have this effect. So, the following part "suggesting that activation of HIF underlies β-cell dysfunction and glucose intolerance in hypoxia." Is not warranted. It would be fair to say "suggesting that unregulated over-activation of HIF may cause β-cell dysfunction. The paper is not off to a good start when the author spell Abstract as Abstruct - it suggests a spell-check was not performed. For the last sentence of the abstract, 'and its implication' - what implication? Line 64 High glucose conditions generate RELATIVE, not absolute hypoxia in beta-cells. This statement should also be referenced.
• Would additional experiments be essential to support the claims of the paper? See above.
• Are the data and the methods presented in such a way that they can be reproduced? Not enough detail for methods, but what is presented looks OK.
• Are the experiments adequately replicated and statistical analysis adequate? Unclear, see above.
• Specific experimental issues that are easily addressable.
• Are prior studies referenced appropriately? No, only the body of work on VHL, not HIFs.
• Are the text and figures clear and accurate? See above comments.
• Do you have suggestions that would help the authors improve the presentation of their data and conclusions? See comments about gels etc 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

The authors examine the role of hypoxia-induced transcriptional repression in mediating loss of β-cell function in type2 diabetes. Transcriptional profiling of mouse and human islets exposed to low oxygen conditions revealed downregulation of β-cell identity and oxidative phosphorylation genes, and upregulation of genes associated with hypoxia. Identification of genes commonly upregulated in Min6 cells, mouse and human islets under hypoxic conditions, revealed induction of two transcriptional repressors BHLHE40 and ATF3. The authors further show that Bhlhe40 deficiency rendered β-cells resistant to hypoxic stress and restored glucose- and KCl-stimulated insulin secretion. This rescue in β-cell function was at least, in part, due to restoration of ATP generation and exocytosis of insulin granules. Furthermore, transcriptional profiling of Min6 cells overexpressing Bhlhe40 indicated down-regulation of key β-cell genes including Mafa. The authors elegantly show that BHLHE40 blocks PDX1 binding to Mafa transcription start site by binding to two E-box sites within the Mafa promoter/enhancer region. Lastly, Cre-mediated β-cell-specific deletion of Bhlhe40 in ob/ob mice restored expression of Mafa and exocytotic genes, accompanied by improvements in ATP generation and insulin secretion.

Major comments

1. The authors conclude that BHLHE40 regulates insulin secretion at two key steps: ATP generation and exocytosis. However, insulin secretory profiles with glucose and KCl seem to be similar with genetic manipulations of Bhlhe40 both in vivo and ex vivo. As the authors indicate in line 176, this suggests a more prominent role of BHLHE40 in regulating exocytotic events downstream of Ca2+ influx. Further experiments are therefore necessary to adequately address the effects on ATP generation. Given the observation that PGC1α, a regulator of mitochondrial biogenesis is suppressed by BHLHE40, mitochondrial assessments would be crucial. Additionally, the effect on mitochondrial mass in Fig 3K seem to be marginal and need to be confirmed using additional measurements listed below.
   • a. In fig 3F, the authors show no change in KCl stimulated Ca2+ influx. Glucose stimulated Ca2+ influx needs to be examined to confirm regulation of ATP generation.
   • b. OXPHOS subunits, TOM20 levels by western blotting
   • c. mtDNA content, transcript levels by qRT-PCR
   • d. Functional assessments: Changes in mitochondrial membrane potential or oxygen consumption
2. Data presented in Figure 4 and 5 indicates transcriptional repression of Mafa by BHLHE40 as a mechanism of beta-cell dysfunction under hypoxic conditions. However, additional experiments are necessary to confirm that repression of PDX1-Mafa binding specifically is responsible for defects in GSIS -
   • a. Fig 5G shows inhibition of PDX1-binding to Mafa with overexpression of Bhlhe40. This needs to be confirmed under hypoxic conditions.
   • b. Fig 4H and 4I show restoration of insulin secretion normalized to total protein with AAV-Mafa. This needs to be supplemented with insulin content as MAFA has been implicated in regulating insulin gene expression (PMID: 25500951).
   • c. qRT-PCR of exocytosis genes and ATP generation with hypoxia and AAV-Mafa.
   • d. Would mutation of A and C E-box sites restore PDX1 binding to Mafa TF region under hypoxia?
3. β-dedifferentiation has been proposed to be involved in loss of insulin secretion in T2D (PMID: 22980982, 16123366). One can speculate that transcriptional repression of Mafa by BHLHE40 is a component of a larger dedifferentiation phenomenon occurring under hypoxia, as other β-cell genes were decreased with hypoxia (Fig 1A) and Bhlhe40-OE in Fig 4A. Identifying differences in dedifferentiation and β-cell disallowed genes with Bhlhe40 overexpression (RNA seq, qRT-PCR) would therefore potentially reveal a dedifferentiation mechanism.
4. The authors identify Atf3 as another transcriptional repressor enriched under hypoxia although to a lesser degree than Bhlhe40. The role of ATF3 in hypoxia-induced apoptosis and adaptive UPR has been previously suggested (PMID: 20519332, 20349223). Additionally, hypoxia represses adaptive UPR in models of T2D and drives β-cell apoptosis (PMID: 27039902). The authors discuss the role of ATF3 under hypoxia in the discussion (lines 319-324) and addressing these research gaps regarding ATF3 function would be insightful.

Minor comments

1. In Fig 2E, increasing replicates would confirm no induction of Bhlhe40 with Thapsigargin.
2. In Fig 2B, BHLHE40 bands need to be quantified to show time-dependent increase in protein levels.
3. In Fig 3C, insulin content needs to be shown with Bhlhe40-OE as in Fig 3B with hypoxia.
4. In Fid 4E-F, band intensities need to quantified by densitometry to determine degree of downregulation of MAFA.
5. In Fig4H and 6G, insulin content needs to be shown as stated above.
6. In Supplemental Figure 3C, apoptosis induced by hypoxia was assessed by PI staining that detects late apoptosis. No significant changes were observed with Bhlhe40-KD, but additional cell death assessments can be used to confirm that B40 does not affect β-cell death.
7. It would be interesting to see the rates of diabetes incidence in Bhlhe40KO: ob/ob mice and if Bhlhe40 deficiency protects against or delays development of diabetes.
8. Knockdown efficiency shown in Supplementary figure 3A needs to be estimated by quantifying band intensities.
9. Line 43 should say "...reversed defects in insulin secretion."

Significance

The data presented provides novel mechanistic insights into the role of hypoxia in β-cell dysfunction. Studies in multiple models of type 2 diabetes (T2D) have shown the loss of signature β-cell genes including Ins1, Pdx1, Mafa, Slc2a2 as a result of excess nutrient stimulation and hypoxia; the precise causal mechanisms, however, still remain to be determined (PMID: 22980982, 28270834). A previous paper from the same group demonstrated downregulation of β-cell signature genes with hypoxia by a HIF1α independent mechanism (PMID: 25503986). Data presented in this report extend those observations and reveal a previously unappreciated role for transcriptional repressor BHLHE40 in the downregulation of a key β-cell gene Mafa. As the authors have identified additional transcriptional repressors including ATF3 and differentially expressed genes in both human and rodent β-cells, this paper would be of great value in understanding the effects of hypoxia. Moreover, studies in mouse models of T2D extend the association of BHLHE40 to clinical β-cell dysfunction and diabetes. My areas of interest are pancreatic β-cell and mitochondrial physiology. GSE analysis and repression of PGC1α by BHLHE40, as appropriately discussed by the authors, point towards impaired mitochondrial function and ATP generation. Additional experiments would greatly support the role of BHLHE40 in mitochondrial dysfunction under hypoxia (as discussed under 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

Manuscript number: RC-2022-01594

Corresponding authors: Hidehiko Kawai and Hiroyuki Kamiya

1. General Statements [optional]

We would like to extend our gratitude to the Editor and both Reviewers for their constructive and insightful comments to our manuscript. We deeply appreciate the Reviewers’ careful consideration of our work, in result of which we think the paper has greatly improved. Below, we have responded to all points raised by the Reviewers.

2. Point-by-point description of the revisions

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

The analysis of mutations in mammalian, including human, genomes has been of interest for many decades. Early DNA sequencing technologies enabled direct identification of mutations in target genes provided that the mutant genes could be readily isolated. This requirement stimulated the development of shuttle vector plasmids that carried a mutation marker gene and could replicate in both mammalian and bacterial cells. These were used in experiments in which the plasmids, treated with a mutagen, would be passaged through mammalian host cells after which the progeny plasmids were introduced into an indicator bacterial strain. Colonies with mutant marker genes could be distinguished by color or survival, the plasmids recovered, and the sequence of the mutant gene determined. The shuttle vector plasmid that became the most widely used contained as the marker the supF amber suppressor tyrosyl tRNA gene positioned in the plasmid such that deletion mutations associated with mammalian cell transfection were selected against. Although various improvements have been introduced since its introduction in the mid-1980s, including bar codes to distinguish independent from sibling mutations (in the early 1990s), the basics of the system have been maintained, and it and variations are still in use. The Kamiya group has made several adjustments to the supF shuttle vectors, including the construction of indicator bacterial strains based on survival of bacteria containing mutant supF genes (the initial system relied on colony color). They have published many studies of mutagenesis by various agents, error prone polymerases, etc. In the current submission they describe a comprehensive approach to identifying mutations in the supF gene that exploits Next Generation Sequencing technology that can identify the full spectrum of mutations including those that escape detection in phenotypic screens. The study is exhaustive and presents a methodical validation of each component of their approach. They report UV induced mutations, the mechanism of which has been well characterized in previous literature. They also describe a category of multiple mutations, which had been observed in the early work with the supF plasmids, and whose relationship to UV photoproducts is most likely indirect.

*We thank the Reviewer for their very insightful feedback to our manuscript and their positive assessment. We have added some discussion points based on the essential references mentioned in the Reviewer’s comments, which we believe made the explanation of our study more complete. *

Major comments: This manuscript presents a technical advance on the use of the supF mutation reporter system. The extent of the validation of each component of the system, including the bar code is rigorous. Their data on the nature and location of UV induced mutations are in very good agreement with previous studies with supF and other reporter genes, a further validation of their approach. Their discussion of the mechanism of the UV induced mutations is in accord with prior work from other laboratories. However, their interpretation of the multiple mutations, although reasonable in invoking a role for APOBEC deamination of cytosines (see eLife. 2014; 3: e02001 for another discussion of this issue), overlooks a much earlier study on the same topic that showed that nicks in the vicinity of the marker gene are mutagenic and can induce multiple mutations (Proc Natl Acad Sci 1987 84:4944-8). It would be useful for the authors to consider their data on the multiple mutations in the light of the earlier analysis. Furthermore, a check to verify the covalently closed circular integrity of the plasmid preparations would be an important quality control and could reduce the mutagenesis observed in 0 UV controls.

We thank the Reviewer for the valuable comments that made our manuscript clearer and more emphatic. We are hereby addressing all of the Reviewer’s concerns. The available data accumulated from previous studies have proved the high sensitivity of the supF assay as a mutagenesis assay, which now has been clearly supported by the results in the current study. We believe that this NGS assay will be able to fulfil the data requirements to tackle many questions related to mutagenesis, thanks to the simplicity and cost-effectiveness of the procedure. However, to meet the experimental objectives, the preparation and analysis of the library are crucially important procedures in the stages of initial setting up of the assay. The covalently closed circular integrity of the vector library is definitely one of the important points we should pay attention to when performing this assay. After the construction of the BC12-library, we have to check the quality of the library by agarose gel electrophoresis. The background mutation frequency and the sequence of the library itself (uploaded as described in the DATA AVAILABILITY section of this manuscript) also needs to be analyzed by NGS before the experiment. We are also routinely constructing the double-stranded shuttle vector from a single-stranded circular DNA with a variety of site-specific damaged oligonucleotides. The treatment with T5 exonuclease followed by purification is absolutely essential to decrease the background mutation frequency. Without the treatment with the exonuclease, cluster mutations may be increased under specific experimental conditions. For this study, we carried out the conventional supF assay using the BC12-library purified after T5 exonuclease treatment. However, in this case the process of purification slightly increased the mutant frequency of the BC12-library to about 2 x10-4 (corresponding to 1x10-6/bp).Therefore, when setting up the essay, we have to consider the background control that we will need for the data analysis. In response to the Reviewer’s comments, we have now added the following paragraph in the DISCUSSION section:

Page 16, line 25:

”5) For the supF assay, spontaneous cluster mutations at TC:GA sites were often observed, and it was well illustrated in an earlier study that a nick in the shuttle vector was a trigger for these asymmetric cluster mutations (54). Therefore, we need to be aware of the quality of each library and how it affects the outcome of each analysis, especially for detection of very low levels of mutations. Depending on the purpose of the experiments, in the preparation of covalently closed circular vector libraries it is essential to eliminate the background level of mutations. In fact, the in vitro construction of the library of double-stranded shuttle vectors from single-stranded circular DNA requires the process of treatment with T5 Exonuclease, which drastically decreases background mutations.”

Minor points The authors state that only 30% of the base sequence of the supF gene can be "used for dual-antibiotic selection on the indicator E. coli". An earlier review (Mutation Res 220: 61,1989) indicated that within the mature tRNA region single or tandem mutations had been reported at 87% of sites, using the colony color assay. The direct NGS analyses would be indifferent to phenotype, and one would expect the maximum number of mutable sites would be recovered from this approach. It would be helpful for an explicit statement regarding the number of mutant sites to be in the Discussion, as this should strengthen the case for the NGS strategy.

We thank the Reviewer for the helpful comment. These are important points we should indeed mention. This method will complement previous data, and especially the data from titer plates will provide us with non-biased mutation spectra for the whole analyzed region. We have now explained in detail about the coverage of mutation spectra in the DISSCUSSION section.

Page 14, line 14:

“The mutation spectra of single or tandem base-substitutions for inactive supF genes identified by using the blue-white colony color assays were comprehensively summarized in an earlier review article, and it was noted that the mutations were detected at 86 sites within a 158-bp region covering the supF gene (54%) and at 74 sites within the 85-bp mature tRNA region (87%), thus demonstrating the great sensitivity of the supF assay system for analysis of mutation spectra (19). However, obtaining reliable datasets by the conventional supF assay requires skill and experience, especially for studies where the mutations of interest are induced with low frequency. The method has been advanced by the construction of indicator bacterial strains with different supF reporter genes which allow selection based on survival of bacteria containing mutant supF genes. However, the fact that the supF phenotypic selection process relies on the structure and function of transfer RNAs that may be differently affected by different mutations means that the improvement of the efficiency of the selection process may cause loss of coverage of the mutation spectra, as it is under our experimental conditions, where the coverage is about 30% (19,20).”

Page 15, line 4:

“From this point of view, we believe that we can secure a sufficient number of experiments to improve the accuracy of the analysis and to confirm the reproducibility of the experiments. Furthermore, the data from colonies grown on titer plates provides us, at least in principle, and with the exception of large deletions and insertions, with non-biased mutation spectra for the whole analyzed region.

Supplementary Figure 1 shows the organization of 8 supF reporter plasmids. Were these discussed in the text and employed in the experiments? It was not clear in the text.

We thank the Reviewer for the helpful comment. It was indeed not clear which vectors we used and why we constructed a series of vectors. Now, we have added the vectors we used for the constructions of the library and each experiment in the RESULTS and MATERIALS AND METHODS sections. Since this is quite important for us and, we believe, the readers, we also added the explanations in the DISCUSSION section, detailing why we have constructed a series of shuttle vectors, as follows:

Page 19, line 36:

“Mutational signatures identified in cancer cells are emerging as valuable markers for cancer diagnosis and therapeutics. Innumerable physical, chemical and biological mutagens, including anticancer drugs, induce characteristic mutations in genomic DNA via specific mutagenic processes. The mutation spectra obtained here by using the presented advanced method were in good agreement with accumulated data from previous papers where the conventional method had been used, with the advantage that our method provided less-biased mutation spectra data. As described above, the datasets presented here highlighted novel mutational signatures and also cluster mutations with a strand-bias, which could be associated with the processes of replication, transcription, or repair of DNA-damage, including a single strand break (a nick). In this study, eight series of supF shuttle vector plasmids were constructed, as presented in Supplementary Figure S1; however, the analysis was carried out using N12-BC libraries prepared from either pNGS2-K1 (Figures 1-4) or pNGS2-K3 (Figures 5-10). The pNGS2-K1/-A1/-K4/-A4 and pNGS2-K2/-A2/-K3/-A3 vector series contain an M13 intergenic region with opposite orientations relative to the supF gene, which allows us to incorporate specific types of DNA-damage at specific sites in the opposite strand of the vector library. Also, the pNGS2-K1/-A1/-K3/-A3 and pNGS2-K2/-A2/-K4/-A4 vector series contain the SV40 replication origin, which enables bidirectional replication and transcription, at opposite sides of the supF gene. Although this is still preliminary data, it is notable that the spontaneously induced mutations for the different vectors in U2OS cells were not significantly different. Therefore, the here presented mutagenesis assay with NGS, by using these series of libraries, can be applied in many different types of experiments to address both quantitative and qualitative features of mutagenesis. It is possible to design series of libraries containing DNA lesions or sequences suitable for the investigation of specific molecular mechanisms, such as TLS, template switching, and asymmetric cluster mutations.”

CROSS-CONSULTATION COMMENTS Comment on the issue raised by Reviewer #2 regarding plasmids with unrepaired DNA damage introduced into E. coli after passage through U2OS cells: treatment of the plasmid harvest with Dpn1 eliminates un-replicated plasmid DNA. Also, SV40 T antigen drives run away replication of the plasmids, which contain the SV40 origin of replication. This greatly dilutes plasmids with remaining UV photoproducts.

Reviewer #1 (Significance (Required)):

Significance This is a comprehensive description of a technical advance for the analysis of mutations based on the most widely used system for reporting mutations in mammalian, including human, cells. As costs for NGS decline it is likely to become the approach of choice.

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

In this manuscript, the authors developed a novel mutagenesis assay by combining the conventional supF forward mutagenesis assay with NGS technology. The manuscript is well written, providing design, methods, and results of the experimental system in very much details, which this reviewer highly evaluates. However, the manuscript may be too long and could be more concise. In addition, this reviewer is afraid that main figures seem difficult to fit printed pages (especially multi-paneled figures of large size, such as Fig. 5 through 8). The authors should re-organize the figures by reducing size and/or moving partly to supplementary information.

We thank the Reviewer for the helpful comments to our manuscript. It is true that the multi-paneled figures were too large, and we have now re-analyzed and optimized most of the figures by reducing size, transferring to Supplementary Figures, and separating one figure into two. Although the number of Figures and Supplementary Figures have now increased, we believe that it has become easy to follow for readers and to fit printed pages. *We considered carefully the Reviewer’s remark about the length of the manuscript, but we feel that the text was already as concise as we could make it, and we have already left out some more detailed explanations. *

1. Some UV-induced DNA damage (typically CPD) is repaired only slowly in human cells, so that the replicated plasmid DNAs recovered from U2OS cells may still contain damage and possibly induce mutations in E. coli after transfection. As the result of high sensitivity of NGS analysis, it is worried that such mutations could be also included in the results. To obtain even more accurate mutational characteristics in mammalian cells, the authors could consider to treat the DNA samples with photolyases before transformation of E. coli. The authors could consider to discuss on this point.

*We thank the Reviewer for the helpful comment, indeed Dpn I treatment is one of the very important procedures for avoiding analysis bias. We have now expanded the explanation why the libraries have to be treated with Dpn I, as follows: *

Page 11, line 4:

“the libraries were extracted from the cells, and treated with dam-GmATC-methylated DNA specific restriction enzyme Dpn I to digest un-replicated DNAs that contain UV-photoproducts.”

1. It is quite intriguing that multiple mutations in a single BC clone tend to occur in the same DNA strand. Is there any trend in a distance between the mutated sites? Considering participation of TLS polymerases in the first round of replication, it may be interesting if multiple DNA lesions occur in relatively close positions so that TLS polymerases elongate the DNA strand without switching back to replicative polymerases.

We thank the Reviewer for the valuable and insightful suggestions for this assay. We have analyzed the positions of SNSs in multiple-mutations shown in Supplementary Figures S11 and S12. As the reviewer mentioned, we may be able to address the mechanisms of TLS switching in mammalian cells by using this assay. In this study, the obtained non-biased mutation spectra of multiple mutations may not be enough for the static analysis, but our results indicate that multiple mutations were induced at relatively close positions. It would be interesting if we could address the mechanisms of TLS polymerase switching. We believe that the accumulation of large numbers of non-biased mutation spectra will provide us with growing opportunities to address more questions in mutagenesis. We have now added the Supplementary Figures S11 and S12, as well as the following discussion points:

Page 14, line 6:

“5) The distance between two SNSs in multiple mutations induced by UV irradiation was relatively shorter than the theoretically expected based on the sequence (Supplementary Figures S11 and S12).”

Page 18, line 27:

“In addition, the positions of SNSs in the multiple mutations were closer to each other compared to the theoretically expected positions (Supplementary Figures S11 and S12), which may reflect switching events involving TLS polymerases. It should be noted that the presented data for the distance between two SNSs in the multiple mutations was analyzed from the data from selection plates in order to secure a sufficient number of mutations, and therefore, there may be a bias due to hot spots associated with the selection process. However, the results from the limited number of mutations from the titer plates are similar to these from the selection plates. It can be proposed that this assay may also be applied for analysis of TLS polymerases in mammalian cells.”

1. This reviewer is wondering whether the results of mammalian cells are influenced by transcription-coupled repair in this experimental system. Because the SV40 replication origin functions as bidirectional promoters, the supF region may be transcribed in U2OS cells so that DNA damage on transcribed strands may be removed more efficiently than non-transcribed strands. Please comment on this, if relevant.

*We thank the Reviewer for the insightful comments. This issue is also very important and interesting, and should be addressed in the mutagenesis research. That is exactly the reason why we presented series of vectors for the assay in this paper. The SV40 replication origin has an effect on the background mutations, which this is also dependent on the experimental conditions. However, this needs to be confirmed by further studies. We hope the idea for these constructions will be helpful for many laboratories. We have now added the following parts in the DISCUSSION section. *

Page 18, line 36:

“Mutational signatures identified in cancer cells are emerging as valuable markers for cancer diagnosis and therapeutics. Innumerable physical, chemical and biological mutagens, including anticancer drugs, induce characteristic mutations in genomic DNA via specific mutagenic processes. The mutation spectra obtained here by using the presented advanced method were in good agreement with accumulated data from previous papers where the conventional method had been used, with the advantage that our method provided less-biased mutation spectra data. As described above, the datasets presented here highlighted novel mutational signatures and also cluster mutations with a strand-bias, which could be associated with the processes of replication, transcription, or repair of DNA-damage, including a single strand break (a nick). In this study, eight series of supF shuttle vector plasmids were constructed, as presented in Supplementary Figure S1; however, the analysis was carried out using N12-BC libraries prepared from either pNGS2-K1 (Figures 1-4) or pNGS2-K3 (Figures 5-10). The pNGS2-K1/-A1/-K4/-A4 and pNGS2-K2/-A2/-K3/-A3 vector series contain an M13 intergenic region with opposite orientations relative to the supF gene, which allows us to incorporate specific types of DNA-damage at specific sites in the opposite strand of the vector library. Also, the pNGS2-K1/-A1/-K3/-A3 and pNGS2-K2/-A2/-K4/-A4 vector series contain the SV40 replication origin, which enables bidirectional replication and transcription, at opposite sides of the supF gene. Although this is still preliminary data, it is notable that the spontaneously induced mutations for the different vectors in U2OS cells were not significantly different. Therefore, the here presented mutagenesis assay with NGS, by using these series of libraries, can be applied in many different types of experiments to address both quantitative and qualitative features of mutagenesis. It is possible to design series of libraries containing DNA lesions or sequences suitable for the investigation of specific molecular mechanisms, such as TLS, template switching, and asymmetric cluster mutations.”

1. page 13: Please check whether the description of Fig. 9C is correct (6th line, graph on top; 9th line, bottom graph).

We thank the Reviewer for carefully checking our manuscript, it was mislabeled in the text. Now, following the Reviewer’s comments, most figures have been changed from the figures in the previous submission. We appreciate the careful review.

CROSS-CONSULTATION COMMENTS Reviewer #1 gives quite relevant comments as an expert of the mutagenesis field. It would improve this manuscript greatly for the authors to make appropriate modifications according to his/her suggestions.

Reviewer #2 (Significance (Required)):

It is quite convincing that this method has a great potential to give much more extensive information on mutational characteristics, most importantly, by eliminating the bias caused by phenotypic selection. Therefore, this work certainly must be worth being published in an appropriate journal.

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 developed a novel mutagenesis assay by combining the conventional supF forward mutagenesis assay with NGS technology. The manuscript is well written, providing design, methods, and results of the experimental system in very much details, which this reviewer highly evaluates. However, the manuscript may be too long and could be more concise. In addition, this reviewer is afraid that main figures seem difficult to fit printed pages (especially multi-paneled figures of large size, such as Fig. 5 through 8). The authors should re-organize the figures by reducing size and/or moving partly to supplementary information.

Specific comments

1. Some UV-induced DNA damage (typically CPD) is repaired only slowly in human cells, so that the replicated plasmid DNAs recovered from U2OS cells may still contain damage and possibly induce mutations in E. coli after transfection. As the result of high sensitivity of NGS analysis, it is worried that such mutations could be also included in the results. To obtain even more accurate mutational characteristics in mammalian cells, the authors could consider to treat the DNA samples with photolyases before transformation of E. coli. The authors could consider to discuss on this point.

2. It is quite intriguing that multiple mutations in a single BC clone tend to occur in the same DNA strand. Is there any trend in a distance between the mutated sites? Considering participation of TLS polymerases in the first round of replication, it may be interesting if multiple DNA lesions occur in relatively close positions so that TLS polymerases elongate the DNA strand without switching back to replicative polymerases.

3. This reviewer is wondering whether the results of mammalian cells are influenced by transcription-coupled repair in this experimental system. Because the SV40 replication origin functions as bidirectional promoters, the supF region may be transcribed in U2OS cells so that DNA damage on transcribed strands may be removed more efficiently than non-transcribed strands. Please comment on this, if relevant.

4. page 13: Please check whether the description of Fig. 9C is correct (6th line, graph on top; 9th line, bottom graph).

CROSS-CONSULTATION COMMENTS

Reviewer #1 gives quite relevant comments as an expert of the mutagenesis field. It would improve this manuscript greatly for the authors to make appropriate modifications according to his/her suggestions.

Significance

It is quite convincing that this method has a great potential to give much more extensive information on mutational characteristics, most importantly, by eliminating the bias caused by phenotypic selection. Therefore, this work certainly must be worth being published in an appropriate journal.

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 analysis of mutations in mammalian, including human, genomes has been of interest for many decades. Early DNA sequencing technologies enabled direct identification of mutations in target genes provided that the mutant genes could be readily isolated. This requirement stimulated the development of shuttle vector plasmids that carried a mutation marker gene and could replicate in both mammalian and bacterial cells. These were used in experiments in which the plasmids, treated with a mutagen, would be passaged through mammalian host cells after which the progeny plasmids were introduced into an indicator bacterial strain. Colonies with mutant marker genes could be distinguished by color or survival, the plasmids recovered, and the sequence of the mutant gene determined. The shuttle vector plasmid that became the most widely used contained as the marker the supF amber suppressor tyrosyl tRNA gene positioned in the plasmid such that deletion mutations associated with mammalian cell transfection were selected against. Although various improvements have been introduced since its introduction in the mid-1980s, including bar codes to distinguish independent from sibling mutations (in the early 1990s), the basics of the system have been maintained, and it and variations are still in use.

The Kamiya group has made several adjustments to the supF shuttle vectors, including the construction of indicator bacterial strains based on survival of bacteria containing mutant supF genes (the initial system relied on colony color). They have published many studies of mutagenesis by various agents, error prone polymerases, etc. In the current submission they describe a comprehensive approach to identifying mutations in the supF gene that exploits Next Generation Sequencing technology that can identify the full spectrum of mutations including those that escape detection in phenotypic screens. The study is exhaustive and presents a methodical validation of each component of their approach. They report UV induced mutations, the mechanism of which has been well characterized in previous literature. They also describe a category of multiple mutations, which had been observed in the early work with the supF plasmids, and whose relationship to UV photoproducts is most likely indirect.

Major comments:

This manuscript presents a technical advance on the use of the supF mutation reporter system. The extent of the validation of each component of the system, including the bar code is rigorous. Their data on the nature and location of UV induced mutations are in very good agreement with previous studies with supF and other reporter genes, a further validation of their approach. Their discussion of the mechanism of the UV induced mutations is in accord with prior work from other laboratories. However, their interpretation of the multiple mutations, although reasonable in invoking a role for APOBEC deamination of cytosines (see eLife. 2014; 3: e02001 for another discussion of this issue), overlooks a much earlier study on the same topic that showed that nicks in the vicinity of the marker gene are mutagenic and can induce multiple mutations (Proc Natl Acad Sci 1987 84:4944-8). It would be useful for the authors to consider their data on the multiple mutations in the light of the earlier analysis. Furthermore, a check to verify the covalently closed circular integrity of the plasmid preparations would be an important quality control and could reduce the mutagenesis observed in 0 UV controls.

Minor points:

The authors state that only 30% of the base sequence of the supF gene can be "used for dual-antibiotic selection on the indicator E. coli". An earlier review (Mutation Res 220: 61,1989) indicated that within the mature tRNA region single or tandem mutations had been reported at 87% of sites, using the colony color assay. The direct NGS analyses would be indifferent to phenotype, and one would expect the maximum number of mutable sites would be recovered from this approach. It would be helpful for an explicit statement regarding the number of mutant sites to be in the Discussion, as this should strengthen the case for the NGS strategy. <br /> Supplementary Figure 1 shows the organization of 8 supF reporter plasmids. Were these discussed in the text and employed in the experiments? It was not clear in the text.

CROSS-CONSULTATION COMMENTS

Comment on the issue raised by Reviewer #2 regarding plasmids with unrepaired DNA damage introduced into E. coli after passage through U2OS cells: treatment of the plasmid harvest with Dpn1 eliminates un-replicated plasmid DNA. Also, SV40 T antigen drives run away replication of the plasmids, which contain the SV40 origin of replication. This greatly dilutes plasmids with remaining UV photoproducts.

Significance

Significance:

This is a comprehensive description of a technical advance for the analysis of mutations based on the most widely used system for reporting mutations in mammalian, including human, cells. As costs for NGS decline it is likely to become the approach of choice.

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:

Review of "Identifying novel regulators of placental development using time series transcriptomic data and network analyses."

The authors present a detailed bioinformatic assessment of mouse developmental time series of the placenta. They apply current data mining and analysis methods to identify protein-centred networks that are likely enriched to specific cell types of the placenta. They then translate these findings to humans using statistical comparisons of human single-cell sequencing data of the placenta. Lastly, they use knock-down experiments to validate the conserved functional importance of the hub genes in the mouse protein networks in human cells.

The strengths of this paper are the rigorous data mining methods and the functional translation to humans from mice. There are no critical weaknesses to the article. There is a blend of statistical analysis with anecdotal or hand curation from databases and the literature, but it is unclear if these curated finings are circumstantial or statistically meaningful. In the end, the hypothesis seems to hold in that 4/4 gene knocked down in the human cells gave a migration phenotype.

Comments, questions, critique:

1. Given the translational aims of the paper, more introduction/discussion material on the comparative aspects of mice and humans are needed. Are giant cells and EVT the same? What are the cell equivalents that you are discovering? The Soncin et al. paper is cited, but I think underused. This publication contains time series data on mice and humans and could be used as external validation of clusters, networks, and other analyses. Other publications to consider for context are

2. Cox B, et al. Mol Syst Biol 5: 279.

3. Silva JF, Serakides R. 2016. Cell Adhes Migr 10: 88-110. (specifically discusses migration difference between the species placentae)

We thank the reviewer for the comment and valuable resources. We agree that more information about the similarities and differences between the migratory cells needs to be provided. We have added the following details in the introduction of the manuscript:

“Although there are certain differences between the mouse and human placenta (Hemberger, Hanna, and Dean 2020; Soncin, Natale, and Parast 2015), they do express common genes during gestation, including common regulators and signaling pathways involved in placental development (Cox et al. 2009; Soncin et al. 2018; Soncin, Natale, and Parast 2015; Watson and Cross 2005). For example, Ascl2/ASCL2 and Tfap2c/TFAP2C are required for the trophoblast (TB) cell lineage in both mouse and human models (Guillemot et al. 1994; Kuckenberg, Kubaczka, and Schorle 2012; Varberg et al. 2021). Another example is the HIF signaling pathway, which regulates TB differentiation in both mouse and human placenta (Soncin, Natale, and Parast 2015).”

“Although the structure of the placenta is not identical between mouse and human, certain mouse placental cell types are thought to be equivalent to human placental cell types (Soncin, Natale, and Parast 2015). For example, parietal TGCs and glycogen TBs have been described as equivalent to human extravillous trophoblasts (EVTs) (Soncin, Natale, and Parast 2015). Mouse TGCs are not as invasive as human EVTs (Soncin, Natale, and Parast 2015), and they have different levels of polyploidy and copy number variation (Morey et al. 2021); however, both EVTs and TGCs are able to degrade extracellular matrix to enable TB migration into the decidua (Silva and Serakides 2016).”

Added to discussion:

“These genes were selected primarily based on the network analyses, but also based on expression data from human cells to account for possible differences between mouse and human placental gene expression.”

As the reviewer suggested, we used the Soncin et al., 2015 data for validation. Only 6,317 of the 11,713 protein-coding genes used for hierarchical clustering were detected in the mouse dataset in Soncin et al., 2015. This issue could be because the Soncin data was generated using microarrays.

Nevertheless, we still compared our e7.5 and e9.5 hierarchical groups with: (1) Soncin et al. gene clusters in mouse that were downregulated over time, had highest expression from e9.5-12.5, or were upregulated over time; and (2) Soncin et al. gene clusters in human that were best correlated with mouse clusters and were either downregulated over time or upregulated over time. We observed a general consensus that our e7.5-hierarchical group had the highest percent of agreement with Soncin et al. gene groups that are downregulated over time, and our e9.5-hierarchical group had the highest percent of agreement with Soncin et al. gene groups that either have highest expression at e9.5-e12.5 or genes that are upregulated over time. This data is added below, described in the results section 1, and included in Supplementary Table S1.

Comparison with Soncin et al. mouse data:

Having expression > 0 (in Soncin et al.) and being in any hierarchical clusters

E7.5-hierarchical genes (down-regulation trend)

E9.5-hierarchical genes (up-regulation trend)

Cluster 2, 3 and 7 (Soncin et al., downregulation trend)

1009

800 (79.3%)

279 (27.7%)

Cluster 6 (Soncin et al., highest at e9.5 – e12.5)

120

51 (42.5%)

110 (91.7%)

Cluster 1, 4 and 5 (Soncin et al., upregulation trend)

1019

415 (40.7%)

881 (86.5%)

Comparison with Soncin et al. human data:

Having expression > 0 (in Soncin et al.) and being in any hierarchical clusters

E7.5-hierarchical genes (down-regulation trend)

E9.5-hierarchical genes (up-regulation trend)

HS Cluster 5 (Soncin et al., downregulation trend)

164

92 (56.1%)

52 (31.7%)

HS Cluster 2 and 4 (Soncin et al., upregulation trend)

111

44 (39.6%)

72 (64.9%)

The following statement was added to the result section:

“Second, we compared our hierarchical groups with previously published mouse and human placental microarray time course data from Soncin et al., 2015 (Soncin, Natale, and Parast 2015). Despite the technical differences between the datasets, we observed a consensus that our e7.5 hierarchical cluster had the highest percent of overlap with Soncin et al. gene groups that are downregulated over time, and our e9.5 hierarchical cluster had the highest percent of overlap with Soncin et al. gene groups that either have highest expression at e9.5 - e12.5 or genes that are upregulated over time (Supplementary Table S1).”

Clustering represented in Figure 1B, was this a supervised model? Why only three clusters?) Did you specify that there would be three models and force each gene profile into one of the categories? How robust are the fits? A fitted model might be a better approach as you can specify the ideal models (early high, late high and mid-high), then determine each gene profile that fits each model and only assess those genes with a significant fit to the model. Forcing clustering to the three-model fit likely gives many poorly fitting profiles. While in the end, this works out, it may be due to applying other post hoc methods for gene enrichment, where noise distributes randomly.

We carried out unsupervised transcript clustering using hierarchical clustering (agglomerative approach using Euclidean distance and complete linkage). The resulting dendrogram was cut at the second highest level to obtain three clusters. We have added additional validation with different numbers of clusters (k = 3, 4 and 5) and quantification of agreement between different clustering methods to show the robustness of the hierarchical clusters. We acknowledge that hierarchical clustering could be sensitive to noise and could result in poorly fitted transcripts in each group; however, it was a necessary first step for us to identify genes relevant to the distinct placental processes at the three timepoints. Acknowledging this disadvantage, we only focused the analyses on genes that are differentially expressed over time and were present in the timepoint hierarchical groups.

We added the additional analysis as Supplementary Figure S1, and the following statements were added in the results section:

"First, we used three different algorithms, K-means clustering, self-organizing maps, and spectral clustering, to validate the trends of the expression levels in hierarchical groups, as well as the number of transcript groups (k = 3, 4 and 5). Only with k = 3 did we obtain groups with median expression level trends consistent in all four algorithms (Supplementary Figure S1). Moreover, with k = 3, the maximum percent of agreement (see Materials and Methods) between hierarchical clusters and clusters obtained using each of the different algorithms was 70.34-87.26% (Supplementary Figure S1), while the maximum percent of agreement between hierarchical clusters and clusters obtained from other algorithms decreases to between 55.67-65.72% with k = 4 and 54.81-59.19% with k = 5.”

We agree model-based clustering could be an alternative approach and have added it to the discussion section:

“Combining hierarchical clustering with differential expression analysis, we were able to identify gene groups using an unsupervised approach. It has also been shown that for times-series analyses with fewer than eight timepoints, pairwise differential expression analysis combined with additional methods identifies a more robust set of genes (Spies et al. 2019). Alternatively, model-based clustering using RNA-seq profiles (Si et al. 2014) could also be useful for gene group identification. However, it is still important to evaluate the robustness and functional relevance of the fitted models by carrying out additional downstream analyses.”

Several statements are made about the conservation of importance between mouse and human hub genes. For example, "We predict these highly expressed genes to be generally important for TB function and processes such as cell migration, a term associated with multiple timepoint specific networks (Figure 2A)." While your knock-down assay of migration results shows these hub genes to be necessary to humans, what do they mean to the mouse? You did not use mouse TSC to assess functional importance concurrently. You note a small number of genes as of known importance, "127 hub genes of which 16 have been annotated as having a role in placental development". Were the others knocked out but lack a developmental phenotype or not assessed? Are these functionally redundant in the mouse or not involved in the same processes between the species?

To assess the possible role of hub genes in mouse development more comprehensively, we extended our search for gene functions on the Mouse Genome Informatics (MGI) database to include not only placenta related GO and MGI phenotype terms (defined as “genes with known roles”), but also embryo related GO and MGI phenotype terms (defined as “genes with possible roles”). We included embryo related terms as “genes with possible roles” because embryonic lethal mouse knockout lines frequently have placentation defects, and because defects in placental development can be associated with the development of other embryonic tissues (Brown and Hay 2016; Perez-Garcia et al. 2018; Woods, Perez-garcia, and Hemberger 2018). This change resulted in an increase in the number of genes with relevant functions in mouse, including several annotated as embryonic lethal or with abnormal embryonic growth (see Supplementary Table S6). With the additional annotations:

• 6 out of 17 hub genes of e7.5 networks have known/possible roles.
• 17 out of 28 hub genes of e8.5 networks have known/possible roles.
• 48 out of 127 hub genes of e9.5 networks have known/possible roles. We also carried out randomization tests to determine if the number of known/possible genes we identified were significant. Randomization tests were carried out with the following procedure: for each timepoint, from the respective timepoint-specific groups, we sampled 10,000 gene sets of the same number as the hub gene numbers. Then we counted the number of known/possible genes in each random set. A p-value is calculated as the number of times a random gene set has ≥ known/possible genes than the observed number, divided by 10,000. We found that the number of genes with known/possible roles at each time point are statistically significant (Supplementary Figure S3). This result indicates that the gene sets we identified are significantly associated with relevant phenotypes in mouse.

The remaining hub genes are unannotated as related to placental or embryonic functions in the MGI database. Based on that, it is difficult to determine if they lack a relevant phenotype, or if there has not been a detailed assessment of the placenta.

Added to section 2 of the result section:

“Briefly, genes annotated under any GO or MGI phenotype terms related to placenta, TB cells, TE and the chorion layer are considered as having a “known” role in the placenta. Genes annotated under terms related to embryo are considered as having a “possible” role in the placenta, because embryonic lethal mouse knockout lines frequently have placentation defects, and because defects in placental development can be associated with the development of other embryonic tissues (Brown and Hay 2016; Perez-Garcia et al. 2018; Woods, Perez-garcia, and Hemberger 2018). Hereafter, such genes are referred to as “known/possible genes”. In the e7.5 networks, there were 17 hub genes in which six genes were known/possible. The number of hub genes that are labelled as known/possible is statistically significant when comparing to random gene sets selected from the e7.5 timepoint-specific group (Supplementary Figure S3). In the e8.5 and e9.5 networks, 17 out of 28 and 48 out of 127 hub genes were known/possible, respectively. Similar to e7.5, the number of hub genes labelled as known/possible in e8.5 networks and e9.5 networks were both statistically significant when comparing to random gene sets selected from the corresponding timepoint-specific groups (Supplementary Figure S3). These results indicate that the gene sets we identified are significantly associated with relevant phenotypes in the mouse.”

For the four genes that we tested in HTR-8/SVneo cells, we also added more information about the current known role of the gene in mouse.

Added to the discussion section:

“We identified hub genes and their immediate neighboring genes which could regulate placental development and confirmed the roles of four novel genes (Mtdh, Siah2, Hnrnpk and Ncor2) in regulating cell migration in the HTR-8/SVneo cell line. These genes were selected primarily based on the network analyses, but also based on expression data from human cells to account for possible differences between mouse and human placental gene expression. Previous studies suggested these four candidates are functionally important in mouse. Mtdh has been suggested to regulate cell proliferation in mouse fetal development (Jeon et al. 2010). The Siah gene family is important for several functions (Qi et al. 2013). Of relevance to the placenta, Siah2 is an important regulator of HIF1α during hypoxia both in vitro and in vivo (Qi et al. 2008). Moreover, while Siah2 null mice exhibited normal phenotypes, combined knockouts of Siah2 and Siah1a showed enhanced lethality rates, suggesting the two genes have overlapping modulating roles (Frew et al. 2003). Hnrnpk-/- mice were embryonic lethal, and Hnrnpk+/- mice had dysfunctions in neonatal survival and development (Gallardo et al. 2015) . Ncor2-/- mice were embryonic lethal before e16.5 due to heart defects (Jepsen et al. 2007). According to the International Mouse Phenotyping Consortium database (Dickinson et al. 2016), Ncor2 null mice also showed abnormal placental morphology at e15.5. However, none of these genes have been studied in TB migration function.”

In determining conservation between mouse and human networks, were only 1:1 orthologs examined or did you consider more complex 1:many mapping conditions between the two species?

In this work, we used only one-to-one orthology between mouse and human avoid duplication while sampling in the enrichment tests. We added this detail in the method section. However, as found in Cox et al., 2009, genes with one-to-many orthologs could be highly intriguing and should be investigated in future studies.

Should the migration assay be normalized to survival/adhesion? If 70,000 cells were seeded but had 50% cell death (or reduced adhesion), then it may appear to be poor migration. Should the migration be evaluated as a ratio of top to bottom cell densities to control for poor adhesion or survival?

We thank the reviewer for bringing up this important point. Unfortunately, with the method we used we cannot quantify the densities on top, because the cells on top need to be scraped off prior to measuring the cells at the bottom (the two densities cannot be measured separately). To help with this concern, in a separate experiment we instead counted cell numbers 48-hours post-transfection for cells treated with target gene siRNA and cells treated with negative control siRNA to determine if apoptosis or changes in proliferation rate could be leading to changes in the observed migration. From this data, we determined that none of the siRNA knockdowns resulted in a significant change of cell counts (p-value > 0.05). We do note that Siah2 siRNA #1 has some decrease in counts (p-value = 0.081) and Ncor2 siRNA #1 and #2 have some increase in cell counts (p-value = 0.081 and p-value = 0.077) (Supplementary Figure S7). Additional follow up experiments we have performed with our targets of interest, which are out of the scope of this paper, demonstrate that different pathways and processes could be involved in the resulting decrease in migration we observed (we are following up experimentally in more detail for each gene). Proliferation and other assays could also be used to further examine the increase in Ncor2 cell counts that were observed. We have added the cell count results and additional text to the discussion.

Added to results, section 4:

“When comparing the number of cells 48 hours post-transfection for cells treated with target gene siRNA to cells treated with negative control siRNA, we determined that none of the target gene siRNA treatments resulted in significant changes in cell counts. We do note that Siah2 siRNA #1 has some decrease in cell counts (p-value = 0.081), and Ncor2 siRNA #1 and Ncor2 siRNA #2 have some increase in cell counts (p-value = 0.081 and p-value = 0.077) compared to negative control treated samples (Supplementary Figure S7). This provides evidence that, in general, the reduction in cell migration capacity was likely not due to the target gene impacting the rate of cell death.”

To the discussion:

“Moreover, we observed that cell counts generally were not decreased upon target gene knockdown compared to negative control knockdown. However, more detailed analysis and process specific assays are needed. For example, future studies assessing each gene’s role in cell adhesion, cell-cell fusion, cell proliferation and cell apoptosis can be done to better understand their roles in placental development.”

Reviewer #1 (Significance (Required)):

This significantly advances previous publications on this topic by functionally testing the discovered genes.

This highlights an excellent data mining strategy for a developmental disease using mice and translating to humans.

The audience is likely developmental biologists and reproductive specialists.

My expertise is bioinformatics and developmental biology.

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

The authors used RNA-seq data from mouse fetal placenta at e7.5, e8.5, and e9.5 to create timepoint-specific gene expression interaction networks to find genes that they predicted would regulate placental development. They confirmed four novel candidate genes and showed that in the transfected human trophoblast HTR-8/SVneo cell line, these four candidates reduced cell migration capacity. Additionally, the authors show that bulk RNA-seq data can be used to infer cell-type composition and when used with single-cell RNA-seq, can be a powerful tool to study the biological processes that involve multiple cell-types.

Overall, the authors are rigorous in their analyses, their conclusions appear sound, and the work could be an asset to the broader placental biology field. However, although the authors present an approach that future studies might find useful to replicate and their work has produced numerous novel transcripts/genes that warrant further investigation, the approach is not entirely novel, and could be expanded/improved (as suggested by the authors in the discussion), particularly with regard to validation of the genes/networks identified. Major and minor comments are listed below.

Major comments:

1) The authors used clustering and differential expression analysis to define sets of timepoint-specific genes. However, it was not clear to me the benefits of this approach. Why would using this approach be better than differential expression analysis alone such as in a typical ANOVA?

We have added more discussion on this matter to explain our approach. We believe using hierarchical clustering and pairwise differential expression analysis can help identify gene lists with higher confidence. These are the new details we added to the discussion section:

“Combining hierarchical clustering with differential expression analysis, we were able to identify gene groups using an unsupervised approach. It has also been shown that for times-series analyses with fewer than eight timepoints, pairwise differential expression analysis combined with additional methods identifies a more robust set of genes (Spies et al. 2019). Alternatively, model-based clustering using RNA-seq profiles (Si et al. 2014) could also be useful for gene group identification. However, it is still important to evaluate the robustness and functional relevance of the fitted models by carrying out additional downstream analyses.”

2) Related to number 1 above, although the authors are interested in timepoint-specific transcripts, the author's methods would filter out possibly interesting transcripts that turn on and off during development. The authors might want to check to see if there are transcripts that are up in e7.5 and then down in e8.5 but then up again in e9.5. Also, the author's methods seem to include transcripts that are not exclusive to one timepoint (i.e. are up in e7.5 and e8.5 but not e9.5). It might be interesting to differentiate transcripts that are exclusive to one timepoint from those that are in more than one timepoint.

We thank the reviewer for their valuable comment. We agree genes that turn on and off during the time course could be very interesting. In performing this analysis, we found that the number of such genes is rather small (38 genes that are up-regulated at e7.5 compared to e8.5 and up-regulated at e9.5 compared to e8.5). These genes were not enriched for processes that we observed with timepoint-specific gene groups, such as “trophoblast giant cell differentiation” (e7.5-specific genes), “labyrinthine layer development” (e8.5- and e9.5-specific genes), "blood vessel development” (e7.5- and e9.5-specific genes) and “response to nutrient” (e9.5-specific genes) (Supplementary Table S3). They are generally enriched for processes related to cytokine production and regulation of secretion.

We also agree that it is interesting to differentiate transcripts that are exclusive to one time point from those that are in more than one time point. In the revised manuscript, we added additional analysis for genes that belong to multiple timepoint groups due to different transcripts of the same gene being annotated as timepoint-specific, and genes unique to each timepoint (Added to results section 1):

“It is possible that timepoint-specific groups share genes that have timepoint-specific transcripts. Indeed, we identified 37 genes shared between e7.5 and e8.5, 5 genes shared between e7.5 and e9.5, and 109 genes shared between e8.5 and e9.5 (Supplementary Table S3). We found that genes only present at one timepoint (timepoint-unique genes) were generally enriched for similar terms as the full group of timepoint-specific genes (Supplementary Table S3). However, terms related to the development of labyrinth layer like “labyrinthine layer morphogenesis” and “labyrinthine layer blood vessel development” were only enriched when using all e8.5-specific genes but not when using e8.5 timepoint-unique genes. Moreover, we found that, unlike genes shared between e9.5 and e7.5, genes shared between e9.5 and e8.5 were enriched for processes such as “blood vessel development” and “insulin receptor signaling pathway”. This observation may indicate that different transcripts of the same genes could be expressed at different timepoints for the continuation of certain biological processes.”

3) In the network analysis it would be interesting and helpful to the reader to highlight, if any, nodes or terms that were found to be significant (i.e. hubs or genes that have a high centrality metric etc.) in both the STRING and GENIE3 networks or overlap the networks created by the two different algorithms to compare them. This might help readers better rank genes when using these data to decide what genes are most important at each timepoint.

We observed only one hub gene shared among networks inferred by the two methods (Vegfa in the e9.5 networks). However, hub genes of networks inferred by one method could be nodes in networks inferred by the other method. Hence, we have added lists of such genes in section 2. Interestingly, many of these genes have known roles in placental development. In terms of biological functions shared between the networks at the same timepoints, there were multiple interesting processes such as “positive regulation of cell migration”, “epithelium migration” and “vasculature development”, which we highlighted in Figure 2A.

In the revised manuscript, we have added the following details in different paragraphs of section 2 of the results:

“Although the networks inferred by the two methods did not share any hub genes, hub genes identified with one method could be members of the other method’s networks. These hub genes are Mmp9 (e7.5_1_STRING), Frk, Hmox1, and Nr2f2 (e7.5_2_GENIE3) (Table 1). This observation strengthens the potential roles of Frk gene in placental development.”

“Hub genes identified with one method and present in the other method’s networks are Hsp90aa1, Akt1, and Mapk14 (e8.5_1_STRING), Dvl3 and Msx2 (e8.5_2_GENIE3) (Table 1).”

“Hub genes identified with one method and present in the other method’s networks include important genes such as Rb1 (Sun et al. 2006), Yap1 (Meinhardt et al. 2020) (e9.5_1_GENIE3) and Vegfa (e9.5_2_STRING) (Table 1). Notably, Vegfa is the only hub gene identified with both of the network inference methods.”

4) The author's conclusion that network analysis can be used to identify genes more likely associated with specific placental cell types is very likely true, but I think that the conclusion would be more impactful if the authors reported how the method compares to simply taking a list of differentially expressed genes and looking for cell type enrichments using their favorite enrichment software. For example, if a gene is highly connected in a particular network that has been identified as SCT-specific, but that gene isn't considered an SCT "marker" by the placental biology research community, it would be interesting to highlight that it is prevalent in a previously published scRNA-seq dataset or a dataset that has isolated that particular cell type to show the advantages of using networks to find placental cell type specific genes.

We completely agree with the reviewer’s point and have now added a randomization analysis to compare the enrichment using PlacentaCellEnrich (PCE) with genes in networks and random genes (Supplementary Figure S6). We randomly sampled 10,000 gene sets with the same sizes as the subnetworks from their corresponding hierarchical groups and carried out PCE analysis. These tests showed that the enrichments of cell type-specific genes were only significant with the subnetwork genes but not the random genes. The randomization tests added a valuable highlight that the network genes are highly relevant to cell type-specific genes in the human placenta, and therefore provided more confidence in the gene lists obtained from the network analyses.

We also further checked the expression of the hub genes in other independent data in order to identify hub genes that are potentially cell type specific markers. For example, we observed that Dvl3 (e8.5_2_GENIE3) and Olr1 (e9.5_3_STRING) have been shown to be differentially expressed in SCT compared to other TB subtypes (human trophoblast stem cells, EVT (Sheridan et al. 2021) or endovascular TB (Gormley et al. 2021)).

We added the following detail in the results, section 3:

“Importantly, randomization tests showed that the enrichment of cell type-specific genes were only significant in these subnetworks but not in random gene sets selected from corresponding timepoint hierarchical groups (Supplementary Figure S6), which highlights the biological relevance of the gene network modules.”

Added to the discussion section:

“Moreover, hub genes could be used to identify potential novel markers for the cell types corresponding to their subnetworks. For example, hub genes of subnetworks enriched for SCT-specific genes such as Dvl3 (e8.5_2_GENIE3) and Olr1 (e9.5_3_STRING) are not established SCT marker genes, but are in fact differentially expressed in SCT compared to human trophoblast stem cells, EVT (Sheridan et al. 2021) or endovascular TB (Gormley et al. 2021). In general, combining network analysis with existing gene expression data from single cell or pure cell populations will allow identification of novel cell-specific marker genes to help future studies focused on different TB populations.”

5) While the selection of genes for validation was limited by the model system available for testing, the authors should recognize that the genes/networks identified here should first and foremost be validated in a mouse model (by knockdown/overexpression studies using mouse trophoblast stem cells or by evaluation of placenta/embryo in a KO/transgenic mouse model). Whether or not the data are relevant to human placentation is (at least initially) irrelevant. While we recognize that these are difficult studies requiring significant time and resources, as is, the data and results will have significantly less impact than if even a limited amount of such validation could be performed.

We thank the reviewer for this valuable comment. Based on this comment and the suggestions from reviewer #1, we have added the following points to the manuscript to discuss the relevance of the genes in the mouse models, and further explain our gene choices:

To assess the possible role of hub genes in mouse development more comprehensively, we extended our search for gene functions on the Mouse Genome Informatics (MGI) database to include not only placenta related GO and MGI phenotype terms (defined as “genes with known roles”), but also embryo related GO and MGI phenotype terms (defined as “genes with possible roles”). We included embryo related terms as “genes with possible roles” because embryonic lethal mouse knockout lines frequently have placentation defects, and because defects in placental development can be associated with the development of other embryonic tissues (Brown and Hay 2016; Perez-Garcia et al. 2018; Woods, Perez-garcia, and Hemberger 2018). This change resulted in an increase in the number of genes with relevant functions in mouse, including several annotated as embryonic lethal or with abnormal embryonic growth (see Supplementary Table S6). With the additional annotations:

• 6 out of 17 hub genes of e7.5 networks have known/possible roles.
• 17 out of 28 hub genes of e8.5 networks have known/possible roles.
• 48 out of 127 hub genes of e9.5 networks have known/possible roles. We also carried out randomization tests to determine if the number of known/possible genes we identified were significant. Randomization tests were carried out with the following procedure: for each timepoint, from the respective timepoint-specific groups, we sampled 10,000 gene sets of the same number as the hub gene numbers. Then we counted the number of known/possible genes in each random set. A p-value is calculated as the number of times a random gene set has ≥ known/possible genes than the observed number, divided by 10,000. We found that the number of genes with known/possible roles at each time point are statistically significant (Supplementary Figure S3). This result indicates that the gene sets we identified are significantly associated with relevant phenotypes in mouse.

The remaining hub genes are unannotated as related to placental or embryonic functions in the MGI database. Based on that, it is difficult to determine if they lack a relevant phenotype, or if there has not been a detailed assessment of the placenta.

Added to section 2 of the result section:

“Briefly, genes annotated under any GO or MGI phenotype terms related to placenta, TB cells, TE and the chorion layer are considered as having a “known” role in the placenta. Genes annotated under terms related to embryo are considered as having a “possible” role in the placenta, because embryonic lethal mouse knockout lines frequently have placentation defects, and because defects in placental development can be associated with the development of other embryonic tissues (Brown and Hay 2016; Perez-Garcia et al. 2018; Woods, Perez-garcia, and Hemberger 2018). Hereafter, such genes are referred to as “known/possible genes”. In the e7.5 networks, there were 17 hub genes in which six genes were known/possible. The number of hub genes that are labelled as known/possible is statistically significant when comparing to random gene sets selected from the e7.5 timepoint-specific group (Supplementary Figure S3). In the e8.5 and e9.5 networks, 17 out of 28 and 48 out of 127 hub genes were known/possible, respectively. Similar to e7.5, the number of hub genes labelled as known/possible in e8.5 networks and e9.5 networks were both statistically significant when comparing to random gene sets selected from the corresponding timepoint-specific groups (Supplementary Figure S3). These results indicate that the gene sets we identified are significantly associated with relevant phenotypes in the mouse.”

For the four genes that we tested in HTR-8/SVneo cells, we also added more information about the current known role of the gene in mouse.

Added to the discussion section:

“We identified hub genes and their immediate neighboring genes which could regulate placental development and confirmed the roles of four novel genes (Mtdh, Siah2, Hnrnpk and Ncor2) in regulating cell migration in the HTR-8/SVneo cell line. These genes were selected primarily based on the network analyses, but also based on expression data from human cells to account for possible differences between mouse and human placental gene expression. Previous studies suggested these four candidates are functionally important in mouse. Mtdh has been suggested to regulate cell proliferation in mouse fetal development (Jeon et al. 2010). The Siah gene family is important for several functions (Qi et al. 2013). Of relevance to the placenta, Siah2 is an important regulator of HIF1α during hypoxia both in vitro and in vivo (Qi et al. 2008). Moreover, while Siah2 null mice exhibited normal phenotypes, combined knockouts of Siah2 and Siah1a showed enhanced lethality rates, suggesting the two genes have overlapping modulating roles (Frew et al. 2003). Hnrnpk-/- mice were embryonic lethal, and Hnrnpk+/- mice had dysfunctions in neonatal survival and development (Gallardo et al. 2015) . Ncor2-/- mice were embryonic lethal before e16.5 due to heart defects (Jepsen et al. 2007). According to the International Mouse Phenotyping Consortium database (Dickinson et al. 2016), Ncor2 null mice also showed abnormal placental morphology at e15.5. However, none of these genes have been studied in the context of TB migration.”

Minor comments:

1) In the GO analysis, why not use a combination of hypergeometric and binomial distribution for enrichment decisions?

We used hypergeometric tests as in the default setting of ClusterProfiler. GO enrichment with hypergeometric test for differentially expressed genes was also suggested in Rivals et al., 2007 (Rivals et al. 2007). Combination of hypergeometric and binomial tests will be of great use when carrying out enrichment for cis-regulatory domains where there is a higher chance of sampling a gene randomly (McLean et al. 2010).

We have added this detail in the method section to make the analysis clearer.

2) In Figure 2B, are there any genes that are both hub nodes (diamonds) and annotated as having placental functions (squares)? If so, it might be good to show that in some way.

We agree this is necessary and have altered the presentation in Figure 2. In the revised manuscript, we have added an additional list of hub genes as genes with possible roles. The figure now shows hub genes with known placental functions (diamonds), hub genes with possible functions (hexagons) and hub genes without related annotation (rounded squares). Non-hub genes are now not shown to avoid crowdedness.

3) It might improve the deconvolution analysis to employ more than one method and recent reports have shown that the cell-type signature data is the most important parameter with the main factors influencing performance being biological (such as where the sample was taken) rather than technical (https://doi.org/10.1038/s41467-022-28655-4).

We agree the conclusion would have been further confirmed if we could employ another deconvolution method. Upon literature search, we found another tool, CAM (N. Wang et al. 2016), that had similar approaches to LinSeed which aims to infer cell proportions without reference. However, the tool has been taken down from Bioconductor and is not currently maintained. As a result, to the best of our knowledge, LinSeed is the only deconvolution tool that is completely reference-free.

We also tried carrying out the deconvolution analysis with another method, DSA (Zhong et al. 2013), with a limited number of marker genes obtained through literature review. However, when the marker genes are highly correlated in multiple cell types, the models failed to infer meaningful proportions.

We acknowledge that we need additional single cell RNA-seq data or marker genes obtained from pure cell populations to make more concrete conclusions for the deconvolution analysis. We hope with future studies, there will be more evidence supporting our observations.

We have added this acknowledgement in the results section:

“The identification of these cell groups could have resulted from noise introduced by both biological and technical variation, which is challenging to overcome when using a small sample size or analyzing without prior knowledge in the deconvolution analysis.”

Added to the discussion section:

“Nevertheless, we acknowledge that our deconvolution analysis and cell type annotations were limited due to the absence of matching scRNA-seq data, data from pure cell populations, or extensive cell marker lists. As these types of information are available, deconvolution analysis can be used to identify species-specific cell types or correcting for confounding effects prior to DEA (Sutton et al. 2022).”

4) The above report also shows that there are ways to correct for cell-type composition differences in DEA which might be interesting to look when using bulk data from different timepoints in future studies when focusing on different biological processes and not timepoint-specific transcripts.

We agree correcting for cell proportion prior to differential expression analysis will be interesting for future studies. When single cell RNA-seq data or more extensive marker gene lists are available, deconvolution analysis will be of great use for this purpose.

We have added this in the discussion section (also mentioned in point #3):

“Nevertheless, we acknowledge that our deconvolution analysis and cell type annotations were limited due to the absence of matching scRNA-seq data, data from pure cells, or extensive cell marker lists. As these types of information become more available, deconvolution analysis can be used to identify species-specific cell types or correcting for confounding effects prior to DEA (Sutton et al. 2022).”

5) Could the authors speculate as to possible reason(s) that an siRNA knockdown would give variable results functionally, while the actual gene expression appears to be consistently and sufficiently downregulated? Did the authors evaluate protein levels following siRNA knockdown?

Following the reviewer’s comment, we have evaluated protein levels for each target gene and each siRNA. For the genes that gave variable results between siRNAs (MTDH and NCOR2), we did not observe a change in their ability to reduce protein levels (Supplementary Figure S7). It is therefore possible that there are off-target effects for one of the siRNAs. We considered this possibility in designing the project, which is why we tested two siRNAs per target gene. Although siRNA off-target effects may be present, visual inspection of the migration experiments indicate that transfection with each of the siRNAs reduces migration capacity. We have added the possibility of off-target effects in the discussion section:

“We observed that while all siRNAs were able to decrease cell migration capacity, there was variability in the amount of decrease, even when comparing two siRNAs targeting the same gene. This observation did not seem to be associated with differences in transcript or protein knockdown levels and could be due to different off-target effects for different siRNAs.”

6) As mentioned in the discussion, finding genes that have timepoint dependent isoforms would an interesting and novel addition to the manuscript.

Protein isoforms would be interesting to study. Here we focused on different mRNA transcripts. We carried out additional GO analysis on the genes unique to each timepoint and genes shared among timepoints. This was also done in response to major comment 2:

In the revised manuscript, we added additional analysis for genes that belong to multiple timepoint groups due to different transcripts of the same gene being annotated as timepoint-specific, and genes unique to each timepoint (Added to results section 1):

“It is possible that timepoint-specific groups share genes that have timepoint-specific transcripts. Indeed, we identified 37 genes shared between e7.5 and e8.5, 5 genes shared between e7.5 and e9.5, and 109 genes shared between e8.5 and e9.5 (Supplementary Table S3). We found that genes only present at one timepoint (timepoint-unique genes) were generally enriched for similar terms as the full group of timepoint-specific genes (Supplementary Table S3). However, terms related to the development of labyrinth layer like “labyrinthine layer morphogenesis” and “labyrinthine layer blood vessel development” were only enriched when using all e8.5-specific genes but not when using e8.5 timepoint-unique genes. Moreover, we found that, unlike genes shared between e9.5 and e7.5, genes shared between e9.5 and e8.5 were enriched for processes such as “blood vessel development” and “insulin receptor signaling pathway”. This observation may indicate that different transcripts of the same genes could be expressed at different timepoints for the continuation of certain biological processes.”

7) Although outside the scope of this manuscript, it might be interesting to look at the effects of knocking down network genes on the networks themselves and in combination with a phenotypic readout such as a migration assay. With numerous knockouts and migration assay readouts, one could possibly find a better method to rank the genes within the networks.

We agree with this comment. Upon literature search, we realized this approach has been used in previous studies on other biological contexts such as virus entry (A. Wang et al. 2010; A. Wang, Ren, and Li 2011) and cancer cell growth (Paul et al. 2021). Although these studies used different network inference strategies from ours, their in silico gene knockouts proved to be effective for the candidate selection. However, the knockout process (both computationally and experimentally) may not be trivial; therefore, we agree the approach will be useful for future studies.

CROSS-CONSULTATION COMMENTS

I mostly agree with the other two reviewers.

It is not clear to me that additional KD experiments (i.e. ones that might affect fusion, proliferation, apoptosis), as proposed by Reviewer #3, would be that much more informative. There are many differences between mouse and human placentation, and these model systems (HTR8 and BeWo) are not truly representative of either. The additional data mining/computational work would be more useful and enhance data interpretation.

Reviewer #2 (Significance (Required)):

The authors use RNA-seq of mouse placenta at e7.5, e8.5, and e9.5 to show that timepoint-specific expression patterns are highly correlated with certain biological processes and point to the existence of certain cell types in the sample. While focused on early post-implantation mouse placental development, the author's methods could be transferrable to other timepoints, species, and organs. Furthermore, with their method they uncover what appears to be several novel, early placental, developmentally important genes and their results might be of interest to those in the field studying placental development.

Reviewer #3:

Summary:

This paper is an analysis of RNA-seq data from the mouse human placenta at embryonic day from 7.5 to 9.5 days. Bioinformatics was used to pinpoint genes networks, and tentatively connect with human cell populations. Wet experiments were performed on the HTR8/SV neo trophoblast cell model.

The introduction clearly posits the reasons why mouse models were chosen, and presents some examples of genes that are conserved between human and mouse placentas, before presenting the major steps of mouse placental development at the crucial periods analyzed.

The results are divided into four parts:

1. Identification of genes that are specific of fetal tissues at the three days studied
2. A network analysis of the genes using classical bioinformatics tools (String, Genie3) to identify gene modules
3. A connection with the human placenta at the level of cell-specific expression profile is then analyzed
4. A in vitro validation on a trophoblast cell model using siRNA to Knockdown genes identified in the in silico part of the paper. Three clustering methods were used to classify the genes according to their profile (at which time point they have the highest level). The function associated are dispatched into three logical physiological events (7.5: proliferation and ectoplacental cone development, 8.5 attachment of the placenta -chorioallantoidian at this stage- , and 9.5: syncytiotrophoblast constitution and labyrinth development, structures essential for growth and exchange).

Mostly minor comments:

Quality of the transcriptomics data: 6 replicates per condition (some being pools at E7.5 and 8.5) is a lot, and I congratulate the authors to have make such effort. This says a lot about the technical quality of their results. Nevertheless, there is no comment on the exclusion of two samples in the further analysis based upon the PCA. Could the authors comment upon the reasons why these two samples behave so differently from the others?

We thank the reviewer for the comment. We reviewed the RNA concentration and quality prior to sequencing, and did not observe that the outliers were of lower quality. After sequencing, quality control metrics (obtained with FastQC), also did not indicate that the two outliers were of poor quality. Based on the PCA, it is also unlikely that two samples were swapped. One possibility is that the tissues obtained for these samples were diseased in some way. However, this is difficult to confirm, so we did not want to speculate about this in the manuscript. We did exclude the two samples to ensure the accuracy of our downstream analyses.

Rq: at this stage some statistics of the degree of enrichment in keyword should be provided (such as Enrichment Scores, normalized or not, and False Discovery Rates, to be able to evaluate the actual robustness of the genes network identified. In addition, it seems that the authors supervised the 'keywords' and 'ontologies' toward placental function. A more agnostic approach could be very relevant, such as identifying the ontologies associated to for instance the set of genes that are highest at 8.5 days, by comparing them with preliminary datasets accessible via the GSEA platform of the BROAD institute or similar sites such as Webgestalt. This does not mean that the placental-targeted approach is not useful, but to have a more global overview is in my opinion indispensable.

We agree and this is a good point. We have now added a stringent approach to determine if the placenta-targeted terms are truly relevant to the gene networks. We performed randomization tests using random gene sets sampled from hierarchical groups of the same time point. These tests showed that the selected terms are significant in the networks when compared to gene groups of the same size from the timepoint specific hierarchical groups (Supplementary Figure S3). Moreover, we have added the specific -log10(q-value) of some highlighted enriched terms in the main text, so together with Figure 2A, the degree of enrichment of these terms can be shown in a clearer way.

We have added this detail in the result section:

“Compared to e8.5 and e9.5 networks, e7.5 networks had a higher rank or fold change and were significantly enriched for the GO terms “inflammatory response” (e7.5_1_STRING: -log10(q-value) = 22.82 and e7.5_2_GENIE3: -log10(q-value) = 3.95) and “female pregnancy” (e7.5_2_GENIE3: -log10(q-value) = 4.1) (Figure 2A, Supplementary Table S5). The term “morphogenesis of a branching structure”, which can be expected following chorioallantoic attachment around e8.5, was not enriched at e7.5, but was enriched in multiple e8.5 and e9.5 networks (e8.5_1_STRING: -log10(q-value) = 1.73, e8.5_2_GENIE3: -log10(q-value) = 1.72, e9.5_1_STRING: -log10(q-value) = 4.01, e9.5_1_GENIE3: -log10(q-value) = 1.54, e9.5_2_STRING: -log10(q-value) = 14.33, and e9.5_2_GENIE3: -log10(q-value) = 2.2). After chorioallantoic attachment finishes, nutrient transport is being established. Accordingly, we observed the following enrichments: “endothelial cell proliferation” (highest ranked in e9.5_2_STRING: -log10(q-value) = 15.91), “lipid biosynthetic process” (only significant after e7.5, highest ranked in e9.5_3_STRING: -log10(q-value) = 17.63), “cholesterol metabolic process” (only significant after e7.5, highest ranked in e9.5_2_GENIE3: -log10(q-value) = 2.76 and e9.5_3_STRING: -log10(q-value) = 7.79), and “response to insulin” (only significant after e7.5, highest ranked in e9.5_1_GENIE3: -log10(q-value) = 1.67).”

“Using randomization tests, we observed the majority of these GO terms (10 out of 11 terms) were significantly enriched when using the network genes but not random gene sets (significance level of 0.05; the term “vasculature development” having p-value = 0.0549 and 0.0575 in with subnetwork e9.5_1_GENIE3 and e9.5_3_GENIE3, respectively) (see Materials and Methods, Supplementary Figure S3). This analysis demonstrates that the network genes were highly relevant to the biological functions of interest. Moreover, the observed GO terms strongly aligned with the processes enriched when using the full lists of timepoint-specific genes (Supplementary Table S3), indicating the representative characteristics of the network genes. While the current analysis focuses on the biological processes related to placental development, there are other terms significantly enriched, which can be found in Supplementary Table S5.”

This is partially done in the part 2 of the results, but it would be relevant to do it on the group of highly expressed genes and not only on the clusters found by the algorithm of sting and genie3.

We have added GO analysis for timepoint-specific genes and also observed highly relevant processes being enriched (Supplementary Table S3). This additional analysis has also helped strengthen the relevance of the network genes, as the observed terms with network genes aligned well with the terms enriched with the full lists of genes.

Rq: in the second part of the results, everything is descriptive but no hierarchy is given to facilitate the understanding and to try to generate a few 'take-home messages' for the reader.

We agree with the comment and have adjusted the writing accordingly. We have added the following statements in section 2 of the result section:

“In summary, we identified 18 subnetworks across three timepoints for downstream analyses, some of which were enriched, according to GO analysis and randomization tests, for specific terms relating to placental development (Figure 2A).”

“These results indicate that the gene sets we identified are functionally relevant in the mouse models.”

“In summary, we have identified hub genes in networks at each timepoint. Analyzing the annotations of hub genes using the MGI database demonstrated that the hub genes are biologically relevant to mouse development and will be strong candidates for future investigation.”

The network analysis is well presented in Figure 2. I wonder whether the author could add systematically besides the three examples that are given the network analysis for the other enrichment network that are described (the four at e7.5, the 6 at e8.5 and the 8 at e9.5).

We have added the additional figures in Supplementary Figure S3.

The deconvolution of the 3rd part of the results to try to connect the mouse results to the human cell situation is interesting. I suspect that given the terms of the mouse placentas used, it would be relevant to focus on 1st trimester human placental cells.

The reference dataset we used in the PlacentaCellEnrich analysis was from human 1st trimester placenta samples. For the Placenta Ontology analysis, we were limited to the provided database from (Naismith and Cox 2021); however, it will be interesting to revisit the analysis when the database is extended.

We have specified that the reference data in PlacentaCellEnrich analysis was from human 1st trimester placenta in the methods section:

“For PlacentaCellEnrich, cell-type specific groups were based on the single-cell transcriptome data of first trimester human maternal-fetal interface from Vento-Tormo et al.”

As previously mentioned, this is a highly descriptive paragraph, and two or three sentences at the end of each paragraph of the results would be in my opinion indispensable to present the most important observations of the results in an intelligible way. Overall, the data presented by the authors, are not obviously 'raw data', but an effort of interpretation should be done by the authors to underline the importance of their results, and to stress among these results which are the most important, and which are the most relevant for placental development and human health.

We agree with the comment and have adjusted the writing accordingly. We have added this summary paragraph at the end of section 3 of the result section:

“In summary, we have demonstrated that the identification of timepoint-specific gene groups and densely connected network modules can be used to infer the cellular composition of bulk RNA-seq samples. We used independent human datasets from different sources to annotate the cell types in each timepoint’s samples. As a result, from the bulk RNA-seq data we were able to observe that at e7.5 and e8.5, there was a high proportion of different TB populations, whereas at e9.5, the placental tissues consisted of multiple cell types such as TB, endothelial and fibroblast cells.”

In the last part, which is very important in this type of paper, four genes were selected. A choice of highly expressed genes was made (which can in fact be discussed, some transcriptional factors may have a crucial importance with relatively low levels of expression). The efficiency of the siRNA was overall excellent. The authors showed that each of these siRNA is efficient to inhibit cell migration in the HTR8/SVneo model.

The migration assays are quantified, but there is a inherent limit of the cell model: the authors analyzed only cell migration, but not other very important parameters. One of them is trophoblast fusion, an issue that can be studied in another trophoblast cell model, the BeWo cells, which are induced to fuse under forskolin. It would be highly relevant to test the siRNA identified in this respect, since fusion is a very conspicuous feature of trophoblast cells in mice as well as in humans. Other relevant endpoints such as proliferation markers, apoptosis markers, oxidative stress markers could be studied in the KD cell models. Alternatively, it would have been interesting to evaluate the overall effect of the siRNA by transcriptomics and check whether the modified gene expression leads to specific profiles characteristic of a certain moment of placental development in mice, or proportion of various cells in the human placentas. Without asking for further experiments the authors should mention these limits in their discussion.

We completely agree with this comment and are investigating each of our candidate genes in more detail in ongoing studies. As we have already learned that each gene is involved in different processes and pathways, we feel that these studies are out of the scope of the current paper. However, we have added this point to our discussion section:

“However, more detailed analysis and process specific assays are needed. For example, future studies assessing each gene’s role in cell adhesion, cell-cell fusion, cell proliferation and cell apoptosis can be done to better understand their roles in placental development.”

In sum, I feel that this paper provides an excellent dataset, but that the authors should make an additional effort of redaction to extract the most important conclusions of their paper. This would increase its impact for a wider public.

Thank you. We have attempted to do so in the revised version.

Reviewer #3 (Significance (Required)):

The context is well introduced, but explanatory and synthesis sentences are missing at the end of each paragraph. I am relatively competent in bioinformatics methods, including deconvolution, and rather expert in cell biology. Therefore I feel comfortable to evaluate this paper.

References:

Brown, Laura D., and William W. Hay. 2016. “Impact of Placental Insufficiency on Fetal Skeletal Muscle Growth.” Molecular and cellular endocrinology 435: 69. /pmc/articles/PMC5014698/ (August 24, 2022).

Cox, Brian et al. 2009. “Comparative Systems Biology of Human and Mouse as a Tool to Guide the Modeling of Human Placental Pathology.” Molecular Systems Biology 5: 279. /pmc/articles/PMC2710868/ (July 20, 2022).

Dickinson, Mary E. et al. 2016. “High-Throughput Discovery of Novel Developmental Phenotypes.” Nature 2016 537:7621 537(7621): 508–14. https://www.nature.com/articles/nature19356 (July 20, 2022).

Frew, Ian J. et al. 2003. “Generation and Analysis of Siah2 Mutant Mice.” Molecular and Cellular Biology 23(24): 9150. /pmc/articles/PMC309644/ (July 27, 2022).

Gallardo, Miguel et al. 2015. “HnRNP K Is a Haploinsufficient Tumor Suppressor That Regulates Proliferation and Differentiation Programs in Hematologic Malignancies.” Cancer Cell 28(4): 486–99. http://www.cell.com/article/S1535610815003050/fulltext (August 24, 2022).

Gormley, Matthew et al. 2021. “RNA Profiling of Laser Microdissected Human Trophoblast Subtypes at Mid-Gestation Reveals a Role for Cannabinoid Signaling in Invasion.” Development (Cambridge, England) 148(20). https://pubmed.ncbi.nlm.nih.gov/34557907/ (August 15, 2022).

Guillemot, François et al. 1994. “Essential Role of Mash-2 in Extraembryonic Development.” Nature 371(6495): 333–36. https://www.nature.com/articles/371333a0 (December 21, 2021).

Hemberger, Myriam, Courtney W. Hanna, and Wendy Dean. 2020. “Mechanisms of Early Placental Development in Mouse and Humans.” Nature Reviews Genetics 21(1): 27–43. http://dx.doi.org/10.1038/s41576-019-0169-4.

Jeon, Hyun Yong et al. 2010. “Expression Patterns of Astrocyte Elevated Gene-1 (AEG-1) during Development of the Mouse Embryo.” Gene expression patterns : GEP 10(7–8): 361. /pmc/articles/PMC3165053/ (July 27, 2022).

Jepsen, Kristen et al. 2007. “SMRT-Mediated Repression of an H3K27 Demethylase in Progression from Neural Stem Cell to Neuron.” Nature 450(7168): 415–19. https://www.nature.com/articles/nature06270 (July 27, 2022).

Kuckenberg, Peter, Caroline Kubaczka, and Hubert Schorle. 2012. “The Role of Transcription Factor Tcfap2c/TFAP2C in Trophectoderm Development.” Reproductive BioMedicine Online 25(1): 12–20. http://www.rbmojournal.com/article/S1472648312001010/fulltext (December 21, 2021).

McLean, Cory Y. et al. 2010. “GREAT Improves Functional Interpretation of Cis-Regulatory Regions.” Nature Biotechnology 28(5): 495–501. http://dx.doi.org/10.1038/nbt.1630.

Meinhardt, Gudrun et al. 2020. “Pivotal Role of the Transcriptional Co-Activator YAP in Trophoblast Stemness of the Developing Human Placenta.” Proceedings of the National Academy of Sciences of the United States of America 117(24): 13562–70. https://www.ncbi.nlm.nih.gov/geo/ (April 8, 2022).

Morey, Robert et al. 2021. “Transcriptomic Drivers of Differentiation, Maturation, and Polyploidy in Human Extravillous Trophoblast.” Frontiers in Cell and Developmental Biology 9: 2269.

Naismith, Kendra, and Brian Cox. 2021. “Human Placental Gene Sets Improve Analysis of Placental Pathologies and Link Trophoblast and Cancer Invasion Genes.” Placenta 112: 9–15.

Paul, Abhijit et al. 2021. “Exploring Gene Knockout Strategies to Identify Potential Drug Targets Using Genome-Scale Metabolic Models.” Scientific Reports 2021 11:1 11(1): 1–13. https://www.nature.com/articles/s41598-020-80561-1 (July 27, 2022).

Perez-Garcia, Vicente et al. 2018. “Placentation Defects Are Highly Prevalent in Embryonic Lethal Mouse Mutants.” Nature 555(7697): 463. /pmc/articles/PMC5866719/ (August 11, 2022).

Qi, Jianfei et al. 2008. “The Ubiquitin Ligase Siah2 Regulates Tumorigenesis and Metastasis by HIF-Dependent and -Independent Pathways.” Proceedings of the National Academy of Sciences of the United States of America 105(43): 16713. /pmc/articles/PMC2575485/ (September 20, 2021).

Qi, Jianfei, Hyungsoo Kim, Marzia Scortegagna, and Ze’ev A. Ronai. 2013. “Regulators and Effectors of Siah Ubiquitin Ligases.” Cell biochemistry and biophysics 67(1): 15. /pmc/articles/PMC3758783/ (July 27, 2022).

Rivals, Isabelle, Lé On Personnaz, Lieng Taing, and Marie-Claude Potier. 2007. “Databases and Ontologies Enrichment or Depletion of a GO Category within a Class of Genes: Which Test?” Bioinformatics 23(4): 401–7.

Sheridan, Megan A. et al. 2021. “Characterization of Primary Models of Human Trophoblast.” Development (Cambridge, England) 148(21). /pmc/articles/PMC8602945/ (August 15, 2022).

Si, Yaqing, Peng Liu, Pinghua Li, and Thomas P. Brutnell. 2014. “Model-Based Clustering for RNA-Seq Data.” Bioinformatics 30(2): 197–205. https://academic.oup.com/bioinformatics/article/30/2/197/217752 (July 18, 2022).

Silva, Juneo F., and Rogéria Serakides. 2016. “Intrauterine Trophoblast Migration: A Comparative View of Humans and Rodents.” Cell Adhesion and Migration 10(1–2): 88–110. http://dx.doi.org/10.1080/19336918.2015.1120397.

Soncin, Francesca et al. 2018. “Comparative Analysis of Mouse and Human Placentae across Gestation Reveals Species-Specific Regulators of Placental Development.” Development (Cambridge) 145(2).

Soncin, Francesca, David Natale, and Mana M. Parast. 2015. “Signaling Pathways in Mouse and Human Trophoblast Differentiation: A Comparative Review.” Cellular and Molecular Life Sciences 72(7): 1291–1302.

Spies, Daniel, Peter F. Renz, Tobias A. Beyer, and Constance Ciaudo. 2019. “Comparative Analysis of Differential Gene Expression Tools for RNA Sequencing Time Course Data.” Briefings in Bioinformatics 20(1): 288. /pmc/articles/PMC6357553/ (July 18, 2022).

Sun, Huifang et al. 2006. “An E2F Binding-Deficient Rb1 Protein Partially Rescues Developmental Defects Associated with Rb1 Nullizygosity.” Molecular and Cellular Biology 26(4): 1527. /pmc/articles/PMC1367194/ (February 6, 2022).

Sutton, Gavin J. et al. 2022. “Comprehensive Evaluation of Deconvolution Methods for Human Brain Gene Expression.” Nature Communications 2022 13:1 13(1): 1–18. https://www.nature.com/articles/s41467-022-28655-4 (July 27, 2022).

Varberg, Kaela M. et al. 2021. “ASCL2 Reciprocally Controls Key Trophoblast Lineage Decisions during Hemochorial Placenta Development.” Proceedings of the National Academy of Sciences of the United States of America 118(10). https://www.pnas.org/content/118/10/e2016517118 (December 21, 2021).

Wang, Anyou, S. Claiborne Johnston, Joyce Chou, and Deborah Dean. 2010. “A Systemic Network for Chlamydia Pneumoniae Entry into Human Cells.” Journal of Bacteriology 192(11): 2809–15. https://journals.asm.org/doi/10.1128/JB.01462-09 (July 27, 2022).

Wang, Anyou, Li Ren, and Hong Li. 2011. “A Systemic Network Triggered by Human Cytomegalovirus Entry.” Advances in Virology 2011.

Wang, Niya et al. 2016. “Mathematical Modelling of Transcriptional Heterogeneity Identifies Novel Markers and Subpopulations in Complex Tissues.” Scientific Reports 2016 6:1 6(1): 1–12. https://www.nature.com/articles/srep18909 (July 27, 2022).

Watson, Erica D., and James C. Cross. 2005. “Development of Structures and Transport Functions in the Mouse Placenta.” Physiology 20(3): 180–93.

Woods, Laura, Vicente Perez-garcia, and Myriam Hemberger. 2018. “Regulation of Placental Development and Its Impact on Fetal Growth — New Insights From Mouse Models.” Frontiers in Endocrinology 9(September): 1–18.

Zhong, Yi et al. 2013. “Digital Sorting of Complex Tissues for Cell Type-Specific Gene Expression Profiles.” BMC Bioinformatics 14(1): 1–10. https://bmcbioinformatics.biomedcentral.com/articles/10.1186/1471-2105-14-89 (July 27, 2022).

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:

This paper is an analysis of RNA-seq data from the mouse human placenta at embryonic day from 7.5 to 9.5 days. Bioinformatics was used to pinpoint genes networks, and tentatively connect with human cell populations. Wet experiments were performed on the HTR8/SV neo trophoblast cell model.

The introduction clearly posits the reasons why mouse models were chosen, and presents some examples of genes that are conserved between human and mouse placentas, before presenting the major steps of mouse placental development at the crucial periods analyzed.

The results are divided into four parts:

1. Identification of genes that are specific of fetal tissues at the three days studied
2. A network analysis of the genes using classical bioinformatics tools (String, Genie3) to identify gene modules
3. A connection with the human placenta at the level of cell-specific expression profile is then analyzed
4. A in vitro validation on a trophoblast cell model using siRNA to Knockdown genes identified in the in silico part of the paper.

Three clustering methods were used to classify the genes according to their profile (at which time point they have the highest level). The function associated are dispatched into three logical physiological events (7.5: proliferation and ectoplacental cone development, 8.5 attachment of the placenta -chorioallantoidian at this stage- , and 9.5: syncytiotrophoblast constitution and labyrinth development, structures essential for growth and exchange).

Mostly minor comments:

Quality of the transcriptomics data: 6 replicates per condition (some being pools at E7.5 and 8.5) is a lot, and I congratulate the authors to have make such effort. This says a lot about the technical quality of their results. Nevertheless, there is no comment on the exclusion of two samples in the further analysis based upon the PCA. Could the authors comment upon the reasons why these two samples behave so differently from the others?

Rq: at this stage some statistics of the degree of enrichment in keyword should be provided (such as Enrichment Scores, normalized or not, and False Discovery Rates, to be able to evaluate the actual robustness of the genes network identified. In addition, it seems that the authors supervised the 'keywords' and 'ontologies' toward placental function. A more agnostic approach could be very relevant, such as identifying the ontologies associated to for instance the set of genes that are highest at 8.5 days, by comparing them with preliminary datasets accessible via the GSEA platform of the BROAD institute or similar sites such as Webgestalt. This does not mean that the placental-targeted approach is not useful, but to have a more global overview is in my opinion indispensable.

This is partially done in the part 2 of the results, but it would be relevant to do it on the group of highly expressed genes and not only on the clusters found by the algorithm of sting and genie3. Rq: in the second part of the results, everything is descriptive but no hierarchy is given to facilitate the understanding and to try to generate a few 'take-home messages' for the reader.

The network analysis is well presented in Figure 2. I wonder whether the author could add systematically besides the three examples that are given the network analysis for the other enrichment network that are described (the four at e7.5, the 6 at e8.5 and the 8 at e9.5).

The deconvolution of the 3rd part of the results to try to connect the mouse results to the human cell situation is interesting. I suspect that given the terms of the mouse placentas used, it would be relevant to focus on 1st trimester human placental cells.

As previously mentioned, this is a highly descriptive paragraph, and two or three sentences at the end of each paragraph of the results would be in my opinion indispensable to present the most important observations of the results in an intelligible way. Overall, the data presented by the authors, are not obviously 'raw data', but an effort of interpretation should be done by the authors to underline the importance of their results, and to stress among these results which are the most important, and which are the most relevant for placental development and human health.

In the last part, which is very important in this type of paper, four genes were selected. A choice of highly expressed genes was made (which can in fact be discussed, some transcriptional factors may have a crucial importance with relatively low levels of expression). The efficiency of the siRNA was overall excellent. The authors showed that each of these siRNA is efficient to inhibit cell migration in the HTR8/SVneo model.

The migration assays are quantified, but there is a inherent limit of the cell model: the authors analyzed only cell migration, but not other very important parameters. One of them is trophoblast fusion, an issue that can be studied in another trophoblast cell model, the BeWo cells, which are induced to fuse under forskolin. It would be highly relevant to test the siRNA identified in this respect, since fusion is a very conspicuous feature of trophoblast cells in mice as well as in humans. Other relevant endpoints such as proliferation markers, apoptosis markers, oxidative stress markers could be studied in the KD cell models. Alternatively, it would have been interesting to evaluate the overall effect of the siRNA by transcriptomics and check whether the modified gene expression leads to specific profiles characteristic of a certain moment of placental development in mice, or proportion of various cells in the human placentas. Without asking for further experiments the authors should mention these limits in their discussion.

In sum, I feel that this paper provides an excellent dataset, but that the authors should make an additional effort of redaction to extract the most important conclusions of their paper. This would increase its impact for a wider public.

Significance

The context is well introduced, but explanatory and synthesis sentences are missing at the end of each paragraph. I am relatively competent in bioinformatics methods, including deconvolution, and rather expert in cell biology. Therefore I feel comfortable to evaluate this paper.

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 used RNA-seq data from mouse fetal placenta at e7.5, e8.5, and e9.5 to create timepoint-specific gene expression interaction networks to find genes that they predicted would regulate placental development. They confirmed four novel candidate genes and showed that in the transfected human trophoblast HTR-8/SVneo cell line, these four candidates reduced cell migration capacity. Additionally, the authors show that bulk RNA-seq data can be used to infer cell-type composition and when used with single-cell RNA-seq, can be a powerful tool to study the biological processes that involve multiple cell-types.

Overall, the authors are rigorous in their analyses, their conclusions appear sound, and the work could be an asset to the broader placental biology field. However, although the authors present an approach that future studies might find useful to replicate and their work has produced numerous novel transcripts/genes that warrant further investigation, the approach is not entirely novel, and could be expanded/improved (as suggested by the authors in the discussion), particularly with regard to validation of the genes/networks identified. Major and minor comments are listed below.

Major comments:

1) The authors used clustering and differential expression analysis to define sets of timepoint-specific genes. However, it was not clear to me the benefits of this approach. Why would using this approach be better than differential expression analysis alone such as in a typical ANOVA?

2) Related to number 1 above, although the authors are interested in timepoint-specific transcripts, the author's methods would filter out possibly interesting transcripts that turn on and off during development. The authors might want to check to see if there are transcripts that are up in e7.5 and then down in e8.5 but then up again in e9.5. Also, the author's methods seem to include transcripts that are not exclusive to one timepoint (i.e. are up in e7.5 and e8.5 but not e9.5). It might be interesting to differentiate transcripts that are exclusive to one timepoint from those that are in more than one timepoint.

3) In the network analysis it would be interesting and helpful to the reader to highlight, if any, nodes or terms that were found to be significant (i.e. hubs or genes that have a high centrality metric etc.) in both the STRING and GENIE3 networks or overlap the networks created by the two different algorithms to compare them. This might help readers better rank genes when using these data to decide what genes are most important at each timepoint.

4) The author's conclusion that network analysis can be used to identify genes more likely associated with specific placental cell types is very likely true, but I think that the conclusion would be more impactful if the authors reported how the method compares to simply taking a list of differentially expressed genes and looking for cell type enrichments using their favorite enrichment software. For example, if a gene is highly connected in a particular network that has been identified as SCT-specific, but that gene isn't considered an SCT "marker" by the placental biology research community, it would be interesting to highlight that it is prevalent in a previously published scRNA-seq dataset or a dataset that has isolated that particular cell type to show the advantages of using networks to find placental cell type specific genes.

5) While the selection of genes for validation was limited by the model system available for testing, the authors should recognize that the genes/networks identified here should first and foremost be validated in a mouse model (by knockdown/overexpression studies using mouse trophoblast stem cells or by evaluation of placenta/embryo in a KO/transgenic mouse model). Whether or not the data are relevant to human placentation is (at least initially) irrelevant. While we recognize that these are difficult studies requiring significant time and resources, as is, the data and results will have significantly less impact than if even a limited amount of such validation could be performed.

Minor comments:

1) In the GO analysis, why not use a combination of hypergeometric and binomial distribution for enrichment decisions?

2) In Figure 2B, are there any genes that are both hub nodes (diamonds) and annotated as having placental functions (squares)? If so, it might be good to show that in some way.

3) It might improve the deconvolution analysis to employ more than one method and recent reports have shown that the cell-type signature data is the most important parameter with the main factors influencing performance being biological (such as where the sample was taken) rather than technical (https://doi.org/10.1038/s41467-022-28655-4).

4) The above report also shows that there are ways to correct for cell-type composition differences in DEA which might be interesting to look when using bulk data from different timepoints in future studies when focusing on different biological processes and not timepoint-specific transcripts.

5) Could the authors speculate as to possible reason(s) that an siRNA knockdown would give variable results functionally, while the actual gene expression appears to be consistently and sufficiently downregulated? Did the authors evaluate protein levels following siRNA knockdown?

6) As mentioned in the discussion, finding genes that have timepoint dependent isoforms would an interesting and novel addition to the manuscript.

7) Although outside the scope of this manuscript, it might be interesting to look at the effects of knocking down network genes on the networks themselves and in combination with a phenotypic readout such as a migration assay. With numerous knockouts and migration assay readouts, one could possibly find a better method to rank the genes within the networks.

CROSS-CONSULTATION COMMENTS

I mostly agree with the other two reviewers. It is not clear to me that additional KD experiments (i.e. ones that might affect fusion, proliferation, apoptosis), as proposed by Reviewer #3, would be that much more informative. There are many differences between mouse and human placentation, and these model systems (HTR8 and BeWo) are not truly representative of either. The additional data mining/computational work would be more useful and enhance data interpretation.

Significance

The authors use RNA-seq of mouse placenta at e7.5, e8.5, and e9.5 to show that timepoint-specific expression patterns are highly correlated with certain biological processes and point to the existence of certain cell types in the sample. While focused on early post-implantation mouse placental development, the author's methods could be transferrable to other timepoints, species, and organs. Furthermore, with their method they uncover what appears to be several novel, early placental, developmentally important genes and their results might be of interest to those in the field studying placental development.

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

Review of "Identifying novel regulators of placental development using time series transcriptomic data and network analyses."

The authors present a detailed bioinformatic assessment of mouse developmental time series of the placenta. They apply current data mining and analysis methods to identify protein-centred networks that are likely enriched to specific cell types of the placenta. They then translate these findings to humans using statistical comparisons of human single-cell sequencing data of the placenta. Lastly, they use knock-down experiments to validate the conserved functional importance of the hub genes in the mouse protein networks in human cells. The strengths of this paper are the rigorous data mining methods and the functional translation to humans from mice. There are no critical weaknesses to the article. There is a blend of statistical analysis with anecdotal or hand curation from databases and the literature, but it is unclear if these curated finings are circumstantial or statistically meaningful. In the end, the hypothesis seems to hold in that 4/4 gene knocked down in the human cells gave a migration phenotype.

Comments, questions, critique

1. Given the translational aims of the paper, more introduction/discussion material on the comparative aspects of mice and humans are needed. Are giant cells and EVT the same? What are the cell equivalents that you are discovering? The Soncin et al. paper is cited, but I think underused. This publication contains time series data on mice and humans and could be used as external validation of clusters, networks, and other analyses. Other publications to consider for context are

a) Cox B, et al. Mol Syst Biol 5: 279.

b) Silva JF, Serakides R. 2016. Cell Adhes Migr 10: 88-110. (specifically discusses migration difference between the species placenta)

1. Clustering represented in Figure 1B, was this a supervised model? Why only three clusters?) Did you specify that there would be three models and force each gene profile into one of the categories? How robust are the fits? A fitted model might be a better approach as you can specify the ideal models (early high, late high and mid-high), then determine each gene profile that fits each model and only assess those genes with a significant fit to the model. Forcing clustering to the three-model fit likely gives many poorly fitting profiles. While in the end, this works out, it may be due to applying other post hoc methods for gene enrichment, where noise distributes randomly.

2. Several statements are made about the conservation of importance between mouse and human hub genes. For example, "We predict these highly expressed genes to be generally important for TB function and processes such as cell migration, a term associated with multiple timepoint specific networks (Figure 2A)." While your knock-down assay of migration results shows these hub genes to be necessary to humans, what do they mean to the mouse? You did not use mouse TSC to assess functional importance concurrently. You note a small number of genes as of known importance, "127 hub genes of which 16 have been annotated as having a role in placental development". Were the others knocked out but lack a developmental phenotype or not assessed? Are these functionally redundant in the mouse or not involved in the same processes between the species?

3. In determining conservation between mouse and human networks, were only 1:1 orthologs examined or did you consider more complex 1:many mapping conditions between the two species?

4. Should the migration assay be normalized to survival/adhesion? If 70,000 cells were seeded but had 50% cell death (or reduced adhesion), then it may appear to be poor migration. Should the migration be evaluated as a ratio of top to bottom cell densities to control for poor adhesion or survival?

Significance

This significantly advances previous publications on this topic by functionally testing the discovered genes.

This highlights an excellent data mining strategy for a developmental disease using mice and translating to humans.

The audience is likely developmental biologists and reproductive specialists.

My expertise is bioinformatics and developmental biology.

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

In this study we reveal that, in both mice and humans, the metabolic benefits of caloric restriction (CR) are sex- and age-dependent. Through a systematic review of the literature, we show that sex differences have been largely overlooked by previous CR research, a finding that Reviewer 1 highlights as “an important point”. Our results have critical implications for understanding the fundamental biology linking diet and health outcomes, as well as translational strategies to leverage the therapeutic benefits of CR in humans.

We thank the reviewers for their helpful appraisal of our manuscript, which Reviewer 2 highlights as “a very well written paper”. Reviewer 1 emphasised the translational relevance of our findings and commented on the “systematic” nature of our study. They noted that it “was well performed”, ”is a valuable and important contribution to the field”, and “will elicit great interest in the scientific and public readership.” Indeed, the importance of sex as a biological variable is the focus of a September 2022 news feature in Nature (https://www.nature.com/articles/d41586-022-02919-x), underscoring the timeliness and relevance of our findings. Our response to the reviewers comments is outlined below, including the changes we have incorporated in a revised version of our manuscript.

Reviewer 1 – Major Comments:

__A) The clinical part is definitely the weak spot in the study. I don't think that the data should be omitted, but the authors should be very careful in interpreting the data. Obvious limitations apply to this part, which need to be more directly addressed in the abstract and discussion. It feels like the data from the small-scale clinical trial is exaggerated. __The clinical study was conducted by Prof Alex Johnstone’s group at the Rowett Institute of Nutrition and Health, University of Aberdeen. Her group are experts in the study of dietary interventions for weight loss. The study was conducted to a high standard and therefore we have the utmost confidence in the conclusions drawn from our analysis of this data.

As we discuss in response to the reviewer’s other points below, the clinical study was primarily designed to address other outcomes and we analysed the data retrospectively to investigate if sex and age affect CR-induced weight and fat loss. This explains some of the limitations that the reviewer mentions, e.g. the relatively low numbers of younger males, and the focus on overweight and obese subjects. As requested, we have now addressed these limitations as follows:

1. Updated the abstract to clarify that the data are from overweight and obese subjects.
2. Updated the results to emphasise that we did a retrospective analysis of CR in overweight and obese subjects (lines 396-398).
3. Performed an additional ANCOVA analysis to test if baseline adiposity or BMI contribute to the sex differences in body mass, fat mass or fat-free mass (new Supplementary Figure 11); see Reviewer 1 Major Point D below.
4. Updated the ‘Limitations’ section of the Discussion to highlight the retrospective nature of the human study (lines 746-748).
5. Updated the Methods to again clarify the retrospective nature of the analysis (lines 884-885). __B) It is important to mention in the abstract and the discussion that the human data came from obese participants. This might well influence the findings from human data. __The human subjects were overweight or obese; this was previously stated in the methods section (line 885) and in the discussion (lines 509-511). To further clarify this, we now also mention it in the Abstract (lines 52-53) and have reiterated it in the Discussion (line 744). Importantly, the fact that humans still show age-dependent sex differences in fat loss, even when overweight and obese, supports our conclusion that this age effect in mice is not simply a consequence of aged mice being fatter than younger mice. We refer to this as the ‘baseline adiposity’ hypothesis (lines 500-518 of the Discussion). In response to point D below, we have also analysed if the loss of fat mass or fat-free mass is influenced by adiposity or BMI at baseline (pre-CR). Our analyses show that neither of these parameters explain the sex differences in loss of fat mass or fat-free mass (see new Supplementary Figure 11).

__C) It is very important to calculate the % calorie restriction of the human participants achieved throughout the CR study. This is crucial information to compare it to other studies. __We have updated the Methods (lines 906-909) to explain the basis for the weight loss diet, as follows: “Participants had their basal energy requirements determined and each participant was then fed an individualised diet with a caloric content equivalent to 100% of their resting metabolic rate (Table 3). This approach was taken to standardise the diet to account for individual energy requirements and energy restriction.” We have also updated Table 3 to show the caloric intake for males and females. Note that RMR accounts for ~60-70% of total daily energy expenditure (TDEE) in adults (Martin et al., 2022), so the diet in our study would give a daily caloric deficit of around 30-40% from baseline TDEE.

__D) Since there is quite a wide range in the BMIs of the participants, can the authors also stratify against BMI? __We have done this against both baseline BMI and against baseline fat mass (the latter to further test the ‘baseline adiposity’ hypothesis). We present this data in an updated Supplementary Figure 11. We find that, in males but not in females, baseline BMI or fat mass are significantly associated with the changes in fat mass or fat-free mass: surprisingly, individuals with higher baseline fat mass or BMI show less fat loss and a greater loss of fat-free mass during CR. Importantly, males and females do not significantly differ in the relationships between baseline fat mass (or BMI) and loss of fat mass or fat-free mass. This further supports our conclusion that the sex differences in fat loss are unrelated to differences in baseline adiposity. We report this in lines 409-411 of the Results and lines 513-515 of the Discussion.

__E) There is no mention of any pre-study registration online of the clinical part (e.g. _gov_). Was this done? __This study was done before pre-registration was a requirement for clinical trials. We retrospectively analysed the study data to investigate if sex and/or age influence the outcomes. In the updated manuscript we now state this on lines 884-885 of the Methods, as well as in the Results (line 396) and Discussion (lines 746-748).

__F) In the methods section the authors write "Participants were informed that the study was funded by an external commercial sponsor...". This is important information, and this is not mentioned anywhere else in the paper. Can the authors clarify this point? A commercial sponsor would, in my view, qualify for a conflict of interest that needs to be mentioned. __We have updated the Declaration of Interests section to clarify this as follows: “The human weight loss study was funded by a food retailer; however, the company had no role in the data analysis, interpretation or conclusions presented in this paper.”

__G) How did the authors determine the group sizes for the clinical part? I have some doubts about the sub-group sizes. It would be valuable information if the authors had a statistical analysis plan prior conducting the study. It appears a bit, like the sub-groups were chosen at random, to match findings of the mouse data. Otherwise, there should have been a better allocation within the sub-groups (especially age). __We agree that larger group sizes would have been preferable. This limitation reflects that the study was not originally designed to test age and sex effects on CR outcomes, but instead was analysed retrospectively to investigate the impact of these variables. As mentioned above, we have updated the text of the manuscript to highlight the retrospective nature of the analyses. In the Discussion, under ‘Limitations’, we also highlight the fact that relatively few younger subjects are included in the human study (lines 744-745).

__H) *There's a big problem with the age stratification of the male participants in the clinical data. If I'm correct there are only 5 males 45 groupings.

__I) The applied protocol for CR in mice is known to provoke long fasting phases and probably elicits some effects through fasting alone, rather than the caloric deficit. There are some papers out addressing this (e.g. by deCabo, Lamming). The authors should not dismiss this fact and at least address it in their discussion. Also, given this fact, it would be thoughtful to include a database-search - not only regarding CR - but also regarding various types of intermittent fasting protocols in humans and animal studies (similar to what the authors did in the supplemental figure). __We agree on the importance of highlighting recent studies demonstrating that prolonged daily fasting contributes to the outcomes of typical ‘single-ration’ CR protocols. We have added a new paragraph to the Discussion to address this (Lines 710-719).

Regarding the second point, we feel that including a new literature search that addresses not only CR, but also intermittent fasting, is beyond the scope of the current manuscript. However, this is a very good idea and would be worth addressing in a future standalone review article. We have also updated our source data to include all data from our literature reviews, to help if other researchers wish to analyse according to fasting duration or other variables.

__J) Did the authors monitor the eating time of the mice? __We have since done this in new cohorts of mice fed using the same CR protocol. We find that the mice consume their food within 2-3 hours, consistent with other CR studies. We have now mentioned this in the Methods section (lines 867-868).

__K) While CR certainly has a lot of health benefits in rodents and humans, it should be advised to raise the cautious note that it may not be beneficial for everyone in the general population. For some groups of people and in some cases (e.g. infectious diseases, pregnancy) even CR with adequate nutritional intake of micro/macronutrients might be disadvantageous. This should be mentioned clearly, as the topic gets more and more "hyped" in public media and online. __We now highlight this important point in the opening paragraph of the introduction (lines 65-67).

__L) There is no indication of how the authors dealt with missing data. Statistically this can be very important, especially in cases with a low number of data points. __In the Methods section we previously explained (lines 846-848) that “Mice were excluded from the final analysis only if there were confounding technical issues or pathologies discovered at necropsy.” No data had to be excluded from our human study and we have now stated this in the Methods (lines 897-898). For analyses involving paired or repeated-measures data (e.g. time courses of body mass or blood glucose), if data points were missing or had to be excluded for some mice then we used mixed models for the statistical analysis. We have now updated this information in the ‘Statistical analysis’ section of the Methods (lines 1047-1048). Because of the large numbers of mice used in our studies, analyses remain sufficiently well powered even if some data points were missing or had to be excluded.

__M) Key data from qPCR should be followed up by western blots or other means. If this was done and there was no effect, the authors should report this. Also, is there any evidence or the possibility to support these findings regarding pck1 and ppara in human samples? __As requested, we will next use Western Blotting to assess the expression of proteins encoded by the transcripts that show sex and/or diet differences within the liver (Fig. 6A). These data will be reported in our fully revised manuscript.

Regarding effects of CR on PCK1 and PPARA expression in human liver samples, no human CR studies have taken liver biopsies for downstream molecular analysis. Recent studies of the GTEx database confirm that hepatic gene expression in humans is highly sexually dimorphic (Oliva et al., 2020). We checked PCK1 and PPARA in the GTEx database and found that, in the liver, each of these transcripts is expressed more highly in females than in males (https://www.gtexportal.org/home/gene/PCK1 & https://www.gtexportal.org/home/gene/PPARA). While this is the opposite to what we observe in our ad libitum mice (Fig. 6A), it demonstrates that sex differences in these genes’ hepatic expression do occur in humans. The effect of CR on their hepatic expression, and whether this differs between males and females, remains to be addressed.

N): I think it would be very valuable to analyse the sex-differences in lipolysis directly in fat tissues. The authors concentrated on differences in hepatic mRNA profiles, but there's an obvious possibility and gap in their story. ____We agree that this would be informative. In the Discussion we cite previous research identifying sex differences in adipose lipolysis and lipogenesis and explain how this fits with our findings (lines 567-574). Since submitting our manuscript, we have begun experiments to investigate sex differences in the effects of CR on lipid metabolism and molecular pathways in adipose tissue. However, these analyses are extensive and ongoing, so we feel strongly that attempting to include them in our present paper would not only substantially delay publication, but also overload what is already a very extensive paper. Therefore, we plan to report our findings in future publications.

__O) Given the relatively low n and sometimes small effect sizes I fear that some of their findings won't be reproduced by other labs. Were the (mouse) data collected all at once in one cohort or did the authors pool data from different cohorts/repeats? __We presume the reviewer means ‘relatively high n’, as most of our mouse analyses used large group sizes. The mouse data were pooled from across multiple cohorts, with ANOVA confirming that the same sex-dependent CR effects were observed within each cohort. This reproducibility across multiple cohorts is a clear strength of our study because it demonstrates the robustness of our findings. Importantly, the sex differences in fat loss, weight loss and glucose homeostasis were still observed in our much-smaller cohort of evening-fed mice (Fig. S5-S6) (n = 5-6), demonstrating that large sample sizes are not needed for other researchers to detect these effects.

Reviewer 1 – Minor comments:

__a) The discussion is very extensive, and I suggest compressing the information presented there to make it more easily readable. __We have removed some text that was more speculative, such as the paragraph discussing a possible role for ERalpha. We have also revised wording elsewhere to state things more succinctly. However, given the scope of our study we feel we cannot substantially cut down the Discussion without compromising the interpretation of our findings. We note the Reviewer two’s comment that “This is a very well written paper” and feel that attempting to compress the extensive information in the Discussion would compromise, rather than help, the readability.

__b) There is some confusion present in the literature regarding the nomenclature of CR/fasting interventions. Recently some reviews have summarized the different forms (e.g. Longo Nature Aging, Hofer Embo Mol Med, ...) and the authors should address this briefly. Especially the applied CR intervention in ____mice overlaps with intermittent fasting. __We have updated the Discussion (lines 710-719) to explain how our single-ration CR protocol also incurs a prolonged intermittent fast, and how this fast per se may contribute to metabolic effects.

c): The order of the subpanels in Figure 9 (and other figures where B is below A and so on) is confusing. Please rearrange or indicate in a visual way which panels belong to each other.

We disagree that the order of subpanels is confusing: the panels are clearly labelled, and we find it most logical to have the absolute values shown in the top row (panels A, C and E), with the corresponding graphs of fold changes shown beneath each of these (panels B, D and F). This allows the reader to quickly compare the absolute vs fold-change data for each readout. If we had panels A-C on the top row and D-F on the second row, then the connection between graphs 9C and 9D would be less clear and comparable.

d): Did the authors also measure cardiovascular (e.g. blood pressure) parameters? There is some evidence out there that there is an age/sex dependency during fasting/CR. This would be a nice add-on to the rather small clinical data here.

We did measure various cardiovascular parameters for our mice but find, unlike for the metabolic outcomes, these generally don’t show sex or age differences. In our human study we measured blood pressure and heart rate before starting CR and at weeks 3 and 4 post-CR. For this response to reviewers we have summarized these human data in Figure R1. The data show that CR decreases blood pressure and heart rate in males and females (Figs. R1A-E). In the younger age group (We have decided to not include these data in the current study because we feel it is already extensive and is focused on metabolic outcomes. We instead plan to report the cardiovascular outcomes (from both humans and mice) in a separate paper.

__e) What was the decision basis for stratifying the human data into 45 years? __We used 45 years as the cutoff point because this is the age when, in women, oestrogen levels begin to decline (this point was stated in lines 491-492 of the Discussion, and we now reiterate it in lines 414-415 of the Results).

__f) The part on aging starting in Figure 7 comes quite surprising and it is not clearly linked to the data before. A suggestion here would be to smooth the transition in the text and the authors could again perform a literature search regarding age-of-onset for CR/fasting interventions in mice and humans. __We have added a sentence to smooth the transition to these studies (lines 363-364). We had previously done a literature search to identify the age of onset of CR interventions in mice and humans. We summarise the findings of this search in lines 452-470 and 484-495 of the Discussion. We have also updated the source data so that it includes the our review of the CR literature, allowing other researchers to interrogate this data.

g) At the first mention of HOMA and Matsuda indices, the effect direction should be put into physiological context.

We now mention this in lines 231-232 of the Results.

h) There is no mention of how the PCA analyses were conducted.

We have updated the Methods to explain that the PCA analyses were done using R. We have updated the source data to include the outputs from these analyses, as well as the underlying code. These data and code are now available here https://doi.org/10.7488/ds/3758.

i) Were the mice aged in-house in the authors' facility or bought pre-aged from a vendor? Is it known how they were raised? If bought pre-aged, were female and male animals comparable?

We bred and aged all mice in house. Males and females were littermates from across several cohorts. Therefore, there are no concerns about lack of comparability resulting from environmental differences.

j) Very minor note: I think that "focussed" has become very rarely used, even in British English. I don't know about the journal's language standards, but I would switch to the much more common "focused".

We have updated to ‘focused’ as requested.

k) Figure 6B/F (PCAs) should indicate the % difference of each dimension.

We have updated the figures to show the % variance accounted for by each principal component. We have also updated the figure legend to specify this.

l) Limitations section: Maybe tone down on "world-leading mass spec facility". This sounds like an excuse and this statement is unsupported and doesn't add anything valuable to the section. Other limitations would include the low n, as mentioned above and the mono-centric fashion of the mouse and human experiments.

We have addressed these points as follows:

• Toned down the description of our mass spec facility (they are renowned for expertise in steroid hormone analysis, so we our original text was intended to highlight that our facility are not novices for this).
• Regarding the low n for some of the human groups, we now highlight this on lines 744-745 of the Discussion.
• We have added a new paragraph to the Discussion (lines 710-719) explaining the limitations of our CR protocol, i.e. that includes elements of both CR and intermittent fasting. Reviewer 2:

__Point 1: This is a very well written paper. __We thank the reviewer for this kind comment.

__Point 2: Since the authors fed the animals in the morning, this is likely the reason for energy expenditure to be different in the CR vs ad lib groups. Although the authors do study the effects of night v day feeding and saw no change in the outcomes regarding weight, this fact I think should be mentioned somewhere. Also, figure 4A is expressed a W while all the other graphs are in kJ. I think it would be nice to see it all consistent. __Regarding the first point, we agree that time of feeding can influence when energy expenditure is altered, but most studies show that CR decreases overall energy expenditure regardless of time of feeding. For example, Dionne et al studied the effects of CR on energy expenditure, administering the CR diet during the night phase (Dionne et al., 2016). They found that CR mice have lower energy expenditure in the day but not in the night (Figure 3C in their paper), which is the opposite to our findings (Figure 4C). However, total energy expenditure in their study remains decreased with CR. This goes against the reviewer’s suggestion that feeding the animals in the morning “is likely the reason for energy expenditure to be different in the CR vs ad lib groups”. We have updated our manuscript (Lines 576-581) to clarify this.

Regarding the second point, we have updated Figure 4A to express the data in kJ (showing the average kJ, per hour, at each time point). The figure legend has been updated to reflect this.

__Point 3: For all the graphs, can you make the CR groups bold and not filled as it is hard to see the lighter colours. __We have updated the graphs so that the CR groups are represented by solid lines, rather than dashed lines.

__Point 4: I know many investigators use them, but I am not sure how relevant HOMA-IR and the Matsuda index are in mice since they were specifically designed for humans. __The issue of whether it is ‘correct’ to use HOMA-IR and/or Matsuda index in mice is often debated in the metabolism field. Importantly, we are not using the absolute values for HOMA-IR or Matsuda in the same way that they are used in humans; instead, we are comparing the relative values between groups because these are still physiologically meaningful. We discussed this with Dr Sam Virtue, an expert in mouse metabolic phenotyping (Virtue and Vidal-Puig, 2021), who agrees on their usefulness in this way.

__Point 5: Something also to note is the fact that all the glucose uptake data is under basal conditions. Just because there are no differences in the basal state does not mean that there are no differences after a meal/during an insulin stimulation. I think that this needs to be discussed and the muscle and fat not completely discounted as a player in the differences seen. __We agree that CR can enhance insulin-stimulated glucose uptake but our OGTT data suggest that it is effects on fasting glucose, rather than insulin-stimulated glucose uptake, that contribute to the sex differences we observe. We have now updated the Discussion (lines 608-613) as follows, “CR enhances insulin-stimulated glucose uptake (82) and it is possible that this effect differs between the sexes. However, our second relevant finding is that, during an OGTT, CR decreases the tAUC but not the iAUC, highlighting decreases in fasting glucose, rather than insulin-stimulated glucose disposal, as the main driver of the improvements in glucose tolerance.”

References cited in Response to Reviewers:

Dionne, D.A., Skovso, S., Templeman, N.M., Clee, S.M., and Johnson, J.D. (2016). Caloric Restriction Paradoxically Increases Adiposity in Mice With Genetically Reduced Insulin. Endocrinology 157, 2724-2734. 10.1210/en.2016-1102.

Martin, A., Fox, D., Murphy, C.A., Hofmann, H., and Koehler, K. (2022). Tissue losses and metabolic adaptations both contribute to the reduction in resting metabolic rate following weight loss. Int. J. Obes. 46, 1168-1175. 10.1038/s41366-022-01090-7.

Oliva, M., Muñoz-Aguirre, M., Kim-Hellmuth, S., Wucher, V., Gewirtz, A.D.H., Cotter, D.J., Parsana, P., Kasela, S., Balliu, B., Viñuela, A., et al. (2020). The impact of sex on gene expression across human tissues. Science 369, eaba3066. 10.1126/science.aba3066.

Virtue, S., and Vidal-Puig, A. (2021). GTTs and ITTs in mice: simple tests, complex answers. Nat Metab 3, 883-886. 10.1038/s42255-021-00414-7.

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

This study was investigating sex and age in calorie restriction. In part 1 the authors reviewed the literature to see the percentage of papers that still only use males as their primary animal model, of which it is still the predominant sex being investigated although there are countries that have stipulated the importance of both sexes being involved and responses compared. In human studies, both sexes are often used, but not analysed separately to give sex differences.

In the second section, which was experimental, the authors report that sex does play an important role in the effects of 30% calorie restriction in mice with differences in both metabolic outcomes as well as in adipose tissue - with males responding and females not. Interestingly, this sex effect in mice was not found if the calorie restriction was started later in life, with both sexes responding, suggesting female sex hormones play and important role early in life in the resistance to calorie restriction. This finding was replicated in human studies, with females seemingly resistant to the weight loss effects or CR, especially in the younger age group.

Comments

1. This is a very well written paper
2. Since the authors fed the animals in the morning, this is likely the reason for energy expenditure to be different in the CR vs ad lib groups. Although the authors do study the effects of night v day feeding and saw no change in the outcomes regarding weight, this fact I think should be mentioned somewhere. Also, figure 4A is expressed a W while all the other graphs are in kJ. I think it would be nice to see it all consistent.
3. For all the graphs, can you make the CR groups bold and not filled as it is hard to see the lighter colours.
4. I know many investigators use them, but I am not sure how relevant HOMA-IR and the matsuda index are in mice since they were specifically designed for humans.
5. Something also to note is the fact that all the glucose uptake data is under basal conditions. Just because there are no differences in the basal state does not mean that there are no differences after a meal/during an insulin stimulation. I think that this needs to be discussed and the muscle and fat not completely discounted as a player in the differences seen.

Significance

Due to the lack of studies directly investigating sex and calorie restriction, I believe this is a very important 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

Summary:

In „The effects of caloric restriction on adipose tissue and metabolic health are sex- and age-dependent" the authors systematically studied the effects of sex and age on the response to sustained caloric restriction in mice and humans. The study was well performed and is a valuable addition to the literature. They found overlapping differences between the species when CR-data from mouse experiments and a small-scale clinical trial were stratified for sex and age. It appears that differences in the response to CR in young age in females compared to males depends on the hormonal status and is equalized in aged mice and humans. Since sex has played an underrated role so far in most studies on CR (especially in rodents), as the authors have laid out in their literature search, they make an important point.

Major comments:

• The clinical part is definitely the weak spot in the study. I don't think that the data should be omitted, but the authors should be very careful in interpreting the data. Obvious limitations apply to this part, which need to be more directly addressed in the abstract and discussion. It feels like the data from the small-scale clinical trial is exaggerated.
• It is important to mention in the abstract and the discussion that the human data came from obese participants. This might well influence the findings from human data.
• It is very important to calculate the % calorie restriction of the human participants achieved throughout the CR study. This is crucial information to compare it to other studies.
• Since there is quite a wide range in the BMIs of the participants, can the authors also stratify against BMI?
• There is no mention of any pre-study registration online of the clinical part (e.g. clinicaltrials.gov). Was this done?
• In the methods section the authors write "Participants were informed that the study was funded by an external commercial sponsor...". This is important information, and this is not mentioned anywhere else in the paper. Can the authors clarify this point? A commercial sponsor would, in my view, qualify for a conflict of interest that needs to be mentioned.
• How did the authors determine the group sizes for the clinical part? I have some doubts about the sub-group sizes. It would be valuable information if the authors had a statistical analysis plan prior conducting the study. It appears a bit, like the sub-groups were chosen at random, to match findings of the mouse data. Otherwise, there should have been a better allocation within the sub-groups (especially age).
• There's a big problem with the age stratification of the male participants in the clinical data. If I'm correct there are only 5 males <45 years. Although this looks intriguing, this can easily be a sampling problem.
• The applied protocol for CR in mice is known to provoke long fasting phases and probably elicits some effects through fasting alone, rather than the caloric deficit. There are some papers out addressing this (e.g. by deCabo, Lamming). The authors should not dismiss this fact and at least address it in their discussion. Also, given this fact, it would be thoughtful to include a database-search - not only regarding CR - but also regarding various types of intermittent fasting protocols in humans and animal studies (similar to what the authors did in the supplemental figure).
• Did the authors monitor the eating time of the mice?
• While CR certainly has a lot of health benefits in rodents and humans, it should be advised to raise the cautious note that it may not be beneficial for everyone in the general population. For some groups of people and in some cases (e.g. infectious diseases, pregnancy) even CR with adequate nutritional intake of micro/macronutrients might be disadvanteguous. This should be mentioned clearly, as the topic gets more and more "hyped" in public media and online.
• There is no indication of how the authors dealt with missing data. Statistically this can be very important, especially in cases with a low number of data points.
• Key data from qPCR should be followed up by western blots or other means. If this was done and there was no effect, the authors should report this. Also, is there any evidence or the possibility to support these findings regarding pck1 and ppara in human samples?
• I think it would be very valuable to analyse the sex-differences in lipolysis directly in fat tissues. The authors concentrated on differences in hepatic mRNA profiles, but there's an obvious possibility and gap in their story.
• Given the relatively low n and sometimes small effect sizes I fear that some of their findings won't be reproduced by other labs. Were the (mouse) data collected all at once in one cohort or did the authors pool data from different cohorts/repeats?

Minor comments:

• The discussion is very extensive, and I suggest compressing the information presented there to make it more easily readable.
• There is some confusion present in the literature regarding the nomenclature of CR/fasting interventions. Recently some reviews have summarized the different forms (e.g. Longo Nature Aging, Hofer Embo Mol Med, ...) and the authors should address this briefly. Especially the applied CR intervention in mice overlaps with intermittent fasting.
• The order of the subpanels in Figure 9 (and other figures where B is below A and so on) is confusing. Please rearrange or indicate in a visual way which panels belong to each other.
• Did the authors also measure cardiovascular (e.g. blood pressure) parameters? There is some evidence out there that there is an age/sex dependency during fasting/CR. This would be a nice add-on to the rather small clinical data here.
• What was the decision basis for stratifying the human data into < and >45 years?
• The part on aging starting in Figure 7 comes quite surprising and it is not clearly linked to the data before. A suggestion here would be to smooth the transition in the text and the authors could again perform a literature search regarding age-of-onset for CR/fasting interventions in mice and humans.
• At the first mention of HOMA and Matsuda indices, the effect direction should be put into physiological context.
• There is no mention of how the PCA analyses were conducted.
• Were the mice aged in-house in the authors' facility or bought pre-aged from a vendor? Is it known how they were raised? If bought pre-aged, were female and male animals comparable?
• Very minor note: I think that "focussed" has become very rarely used, even in British English. I don't know about the journal's language standards, but I would switch to the much more common "focused".
• Figure 6B/F (PCAs) should indicate the % difference of each dimension.
• Limitations section: Maybe tone down on "world-leading mass spec facility". This sounds like an excuse and this statement is unsupported and doesn't add anything valuable to the section. Other limitations would include the low n, as mentioned above and the mono-centric fashion of the mouse and human experiments.

Significance

The authors have analysed in great depth the importance of age and sex on the outcomes of CR on fat tissue, body weight and glucose homeostasis. It is a valuable and important contribution to the field. However, some points need to be clarified and some additional analyses/experiments should be performed. The piece will elicit great interest in the scientific and public readership. The authors have collected important evidence that the majority of CR studies in the literature have a drastic sex-bias in both mice and humans. This may be one reason why translational and mechanistic findings from mice to human application have been haltered in the field of nutritional healthy-aging-promoting interventions.

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

The Drosophila oocyte is a classical model to study establishment of cell polarity, and it is known for decades how Bicoid and Oskar define the anterior-posterior axis of the embryo. However, Bicoid and Oskar are not conserved so that these findings cannot be generalized. The situation is different for the Par proteins, which have been identified in C. elegans. They are not only conserved but also their mode of of action seems to be preserved. 20 years ago it was a very surprising finding that the Par proteins contribute to establishment of polarity in the Drosophila oocyte. A fascinating and simple mutual inhibition model emerged over the years, in which the same molecular mechanisms establish cell polarity in the C. elegans one-cell embryo and in the Drosophila oocyte: Anteriorly localised Par-3/Bazooka recruits aPKC kinase, which excludes Par-1 by phosphorylation, whereas posteriorly localised Par-1 kinase excludes Par-3/Bazooka by phosphorylation. The manuscript by Milas et al. challenges this model by closely analysing Par localisation in living Drosophila oocytes. The authors provide strong evidence that the kinetics of Par-1 and Bazooka localisation are not consistent with the model.

Milas et al. first describe a morphological difference between the anterior-lateral and posterior cortex of the oocyte by showing that only the posterior cortex is tightly connected to the overlaying epithelium. This morphological difference correlates with the localisation of Par-1, which is restricted to the posterior, while Bazooka localises only to those regions of the cortex, where there is a gap between the oocyte and the epithelium. This gap expands towards the posterior cortex during stage 10A and encloses it at stage 11. Unexpectedly, Par-1 and Bazooka localisations overlap at the posterior cortex when the gap expands, which contradicts the mutual inhibition model. The authors hypothesised that the close contact of epithelium with the oocyte might influence Par-1/Bazooka localisation. To test this they mechanically detached the epithelium from the oocyte and also ablated groups of epithelial cells. These manipulations resulted in posterior spreading of Bazooka protein within 30-60 minutes. Interestingly, the authors found that in those regions of the posterior cortex, where cells have been ablated, Par-1 and Bazooka colocalise for 30 minutes, which is difficult to reconcile with a model in which Par-1 excludes Bazooka by phosphorylation. The authors also show that Par-1 finally disappeared form the regions where epithelial cells have been ablated. However, aPKC, the kinase that is supposed the exclude Par-1 by phosphorylation, appeared only after Par-1, which argues against the idea that aPKC prevents Par-1 localisation. In summary, the described localisation kinetics are in conflict with the current model, in which direct phosphorylation activities of Par-1 and aPKC orchestrate the mutual exclusive Par domains in the Drosophila oocyte. The data suggest that the mechanisms underlying mutual inhibition are more complex than thought and involve contact with posterior epithelial cells.

The microscopy used by the authors is state of the art, the data are of high quality and the quantitative analysis is convincing. The results are surprising but conclusive since the experiments were performed and presented in a professional way. This combination makes the manuscript very interesting.

Major points:

1. The finding that the posterior cortex is in close contact to the epithelium, while there is a gap between the remaining oocyte cortex and the epithelium is very interesting, and should be quantified and characterised more precisely. When does the gap form and how exactly does it spread posteriorly? Is it possible to distinguish the gap from the attachment zone by using markers for the ECM (e.g. viking-GFP) or adhesion proteins (e.g. Integrin)?
2. The authors suggest that direct contact between the epithelium and the oocyte is required to exclude Bazooka from the posterior oocyte cortex. The polar cells of the follicular epithelium have almost no contact to the oocyte. One would expect that if only the polar cells are ablated, this would not lead to posterior spreading of Bazooka. Such a control experiment could support the author´s model.

Minor points:

1. There are repeatedly double negations which make the text difficult to understand (e.g. "Bazooka exclusion was lost...." (line 104) or "Par-1 does not delocalise from the posterior pole prior to accumulation of Bazooka" (line 163). I see that this follows the logic of the published molecular mechanisms but for the sake of comprehensibility, the authors should try to formulate the results in a positive way (at least in a repeating sentence).
2. Based on the kinetics of Par-1 localisation the authors the conclude that Par-1 binds to diffusible binding sites at the oocyte cortex, which are modulated by posterior epithelial cells. This is one possible explanation for their results but other interpretations are equally possible. Since the authors provide no further evidence for the existence of Par-1 binding sites their interpretation should be formulated more carefully.
3. The authors should mention that they use the Par-1 isoform (N1S) which fully rescues the par-1 mutant phenotype (see Doerflinger et. al, Curr Biol, 2006). What is known about the rescuing activity of the Bazooka transgenes that were used in the manuscript?
4. In principle it is possible that the posterior spreading of Bazooka (after follicle cell detachment or ablation) is caused by premature ooplasmic steaming. However, the movies show that this is not the case. This should be stated in the text.

Significance

The Drosophila oocyte is a classical model to address the fundamental biological question of how cell polarity is established. The current model of mutual Par protein inhibition is a critical part of our understanding of cell polarisation, and was proposed to be conserved between flies and worms. In the case of Drosophila this model mainly relies on a combination of genetic and biochemical data. Milas et al. tested this model by using in vivo imaging, and found that the kinetics of Par localisation do not correspond to the existing model. This suggests that central aspects of the proposed mechanisms controlling mutual Par inhibition in the Drosophila oocyte are not conserved or not fully understood. The work makes therefore a surprising and important contribution to the understanding of cell polarity.

I work for many years on Drosophila oogenesis and my main interest switched from cell polarity to membrane trafficking.

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 examines an important and unsolved question concerning the establishment of the polarity axis in the Drosophila oocyte, namely how the follicle cells located at the posterior of the egg chamber trigger a signal to the oocyte for its subsequent polarization. To address this question the authors, center their studies on the localization of PAR proteins which are distributed along the antero-posterior axis. They more specifically focus on the mutual exclusion of Par1 and Bazooka/Par3 (Baz)at the posterior of the oocyte. The signaling event from the posterior follicle cells toward the oocyte is an essential process however it remains unsolved despite numerous screening and genetic manipulation approaches, leaving open the possibility that classical signaling involving a diffusible ligand emitted by follicular cells with its receptor located at the plasma membrane of the oocyte, would not be applied here.

Here the authors are using original biophysical approaches to address whether signaling between the follicular cells and the oocyte would involve mechanical features.

The authors focus at the dual exclusion between Baz and Par1 between the stages 10 and 11. To specifically follow these two proteins in the oocyte without being disrupted by their expression in the follicle cells, they used the Gal4/UASp system to express Baz and Par1. They found that Baz accumulate again at the posterior of the oocyte at stage 10B following the loss of contact between the posterior follicle cells (PFCs) and the oocyte whereas Par1 is gradually lost at that position. By using a glass micropipette to aspirate and pull on the PFCs they observed a premature Baz accumulation at the oocyte posterior. Then, to spatially improve the targeted area in the PFCs, the authors use a pulsed UV lazer, and show that PFCs are required to locally maintain posterior exclusion of Baz. Using a similar setup, they show that similarly Par1 is eliminated at the oocyte cell cortex region that had been in contact with ablated PFCs. However, Par1, with a kinetic slower to the one of Baz, is never disappearing before Baz appearance. Although difficult to distinguish, the authors report that the disappearance of Par1 is locally connected by an increase in microtubules (see major points). Finally, upon PFCs ablation, the authors show that the posterior reappearance of Baz is followed by the appearance of aPKC. However, the reappearance of PKC is slower than the removal of Par1, suggesting that in this case Par1 is not removed by PKC.

The particularly interesting results of this work show that cellular contacts between PFCs and the oocyte are necessary to maintain Baz exclusion and Par1 localization. Furthermore, the ablation results suggest that individual PFCs are required to maintain local posterior exclusion of Baz. Overall it is an interesting observation, and most of the data are presented in a clean organized manner.

Major comments

1. The authors concentrate their studies on the distribution of Par3 and Par1 at the posterior part of the oocyte, mainly at stage 10 according to the images in the figures and movies. The involvement of Par3 and Par1 on polarized transport to the posterior pole of the oocyte has been well characterized previously between stages 7 and 9. The results of the authors are very interesting but they do not show that beyond the return of Baz and the disappearance of Par1 at the developmental stage they are looking at, the antero-posterior polarization and more particularly the localization of oskar in the posterior is affected. This is an important point as the authors propose that follicle cell contact maintains main body axis polarity. This would be possible by monitoring the impact of PFC ablation on the maintenance of oskar localization by tracking osk RNA with the MCP-MS2 system, or also by visualizing the staufen protein with a stau-GFP transgene.
2. The authors use the Jupiter protein fused to the cherry protein to track MTs. This is perfectly fine to highlight the cytoplasm in the oocyte and to outline the cell-cell contacts between the PFCs and the oocyte. However, with Jupiter-cherry the microtubules are not clearly detected in the oocyte in the data presented.This is a problem because the authors want to make an important point with the potential reappearance of microtubules in the oocyte while Par1 has disappeared in the vicinity of the destroyed PFCs. (Fig5). The authors should use another microtubule reporter that allows better detection of microtubules in the oocyte, Jupiter-GFP, EB1-GFP, Ensconsin MT binding domain (EMTB)-RFP.

Minor comments:

1. The stage of the oocyte is not always indicated, this is particularly the case with the Fig2 with the pulling experiment with a glass micropipette.
2. With the Fig 3E, to highlight the fact that the intensity of Baz increases very quickly after the removal of PFCs (1 mn) the authors should include an insert with a shorter time scale. The authors could also comment on the difference in velocity in baz reappearance when the ablation of PFCs includes or not polar cells.
3. In the discussion line 240, this is not myosin II but myosin V which anchored oskar mRNA at the posterior.
4. For the suppl figure 5, the n is not mentioned in the legend

Significance

Nature and significance of the advance and work in the context of existing literature

This manuscript examines an important and unsolved question concerning the establishment of the polarity axis in the Drosophila oocyte, namely how the follicle cells located at the posterior of the egg chamber trigger a signal to the oocyte for its subsequent polarization (Gonzalez-Reyes et al ; Nature 1995 ;doi: 10.1038/375654a0) and (Roth et al; 1995; Cell; doi: 10.1016/0092-8674(95)90016-0). To address this question the authors, center their studies on the localization of PAR proteins which are distributed along the antero-posterior axis. They more specifically focus on the mutual exclusion of Par1 and Bazooka/Par3 (Baz) at the posterior of the oocyte. The signaling event from the posterior follicle cells toward the oocyte is an essential process however it remains unsolved despite numerous screening and genetic manipulation approaches, leaving open the possibility that classical signaling involving a diffusible ligand emitted by follicular cells with its receptor located at the plasma membrane of the oocyte, would not be applied here. We still know little about the modalities of this signaling between the follicular cells and the oocyte necessary for the polarization of the latter. We know that the first sign of anteroposterior polarization in the oocyte is posteriorly the recruitment of Par1 and subsequently the elimination of Baz. However, we do not know the nature of this signaling. Furthermore, we do not know whether this signaling must be maintained in order to maintain the polarization of the oocyte and more particularly to maintain the localization of oskar RNA, the posterior determinant of the oocyte, Here the authors are using original biophysical approaches to address whether signaling between the follicular cells and the oocyte would involve mechanical features. Important results of this work show that cell contacts between PFCs and the oocyte are necessary to maintain baz exclusion and Par1 localisation. Furthermore, the ablation results suggest that individual PFCs are required to maintain local posterior exclusion of Bazooka.

Audience: These results will be of interest to those interested in the relationship between cell signaling and polarization in particular in a developmental context.

Reviewer's area of expertise: Cell polarity, microtubule-associated transport, oocyte development in Drosophila.

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 formation of mutually exclusive domains of Partition defective (Par) proteins works as a foundation for establishment of cell polarity in a variety of cells. The Drosophila oocyte is a well-known model system to study mechanisms of the asymmetric distribution Par proteins. At stage 6/7 of oogenesis, an unknown signal the posterior follicle cells (PFC) induces the recruitment of the Par-1 kinase to the posterior cortex of the oocyte and the concomitant exclusion of aPKC/Par-6 from this region. By contrast, Bazooka (Par-3) remains at the posterior with Par-1 and only disappears from the posterior at early stage 9. Millas et al. investigate the nature of the PFC signal and whether PFC continue to play a role in keeping Bazooka away from the posterior after the original signal is received by the oocyte. They do so by following the distribution of Bazooka and other Par proteins in living oocytes after pulling away or ablating the PFC at various stages of oogenesis.

Major comments

1. Quality of live imaging Judging from the appearance of the polar follicle cells and the size of the follicle cells, the authors constantly have an issue with maintaining a steady focal plane during live imaging in most movies (Figure 2 and video1; Figure 3 and video 2; Figure 4 and video 4; Figure 5 and video 5, FigureS5 and video 6). The conclusions of the paper are based on measuring changes in fluorescence intensity at the oocyte posterior over time, and this will be undermined by a varying focal plane. Considering the bullet shape of the oocyte, imaging the posterior at different focal planes could also cause artefacts. Supplementary Fig 3D-E and video 3 (a control experiment) are examples where the focal plane did not drift.
2. Mechanical contact of PFC with the oocyte cortex causes the posterior exclusion of Bazooka and maintains oocyte polarity By physically pulling PFC away from the oocyte at stage 10b (Figures 1-2) the authors observed that in some oocytes Bazooka re-localises to posterior and concluded that it is a mechanical contact between PFC and the oocyte cortex that keeps Bazooka away from the posterior. Although this is an interesting observation per se, this is after the polarity of the oocyte has been defined (stages 6-9) and the posterior determinant, oskar mRNA has been localised. Could the authors do the same experiment at stages 6-9 to directly address whether the distance between the PFC and the oocyte cortex actually matters, considering that Bazooka remains at the posterior up to early 9 when the PFC and the oocyte are still at close contact?

The conclusion that the signal between PFC and the oocyte could be mechanical is only one of potential interpretations of the experiment. It still could be a short range/ non-diffusible biochemical signal that is sensitive to the distance between the PFC and the oocyte membrane. The authors do not provide any evidence for or against either interpretation. 3. Figure 5B is supposed to demonstrate that local loss of Par-1 at the posterior causes the re-growth of microtubules from this region. However, the data provided are not convincing. The accumulation of red vesicles at the posterior cortex 150 min post ablation does not look like a specific signal for Jupiter-mCherry-marked microtubules. Similar vesicles start to be visible in the neighbouring follicle cells at the same time.

Minor comments

1. In Figure 4A-C, it is not clear what area has been ablated
2. The authors should provide a simple 1-6 numbering for Video files

Significance

The observation that the PFC are required to maintain oocyte polarity at stage 9 is significant, but not very surprising, given the recent observation by Doerflinger et al that the posterior localisation of Par-1 requires continuous myosin activation, demonstrating that the antagonism between anterior and posterior Par proteins is not sufficient to maintain polarity once established. The authors must improve the quality of the live imaging to support this conclusion.

The conclusion that phosphorylation of Bazooka by Par-1 is not sufficient to exclude Bazooka from the posterior cortex is not novel (see Doerflinger et al 2010, 2022).

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

Comments for Butt et al This study examines the potential role of SHARPIN phosphorylation on cell migration. The role of the phosphorylation sites of SHARPIN, especially S146, is clearly shown by the expression of the mutant in various cell lines. However, the argument of the interaction with the Arp2/3 complex is relatively weak; only the FRET analysis in cells is shown. It would be good if the authors could include more mechanistic insights into the role of the phosphorylation of SHARPIN.

Table 1. the reason for the selection of S131 and S156 is not clear. Why the S165 and T170 were not studied?

The phosphorylation would be better confirmed by the antibody for the phosphorylation peptides. The possible phospho-mimic mutants, with the substitution to the acidic amino acid residues, would be better to be examined.

Significance

The potential role of SHAEPIN in the context of the current understanding of the cell migration machinery would be better to be described. Now the study only contains the mutant expression phenotype.

Referees cross-commenting

I agree with other reviewers and do not have other opinions. The three reviews appear to be reasonable.

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 paper, Butt and co-workers examine the role of SHARPIN phosphorylation in Ser146 as it controls diverse readouts related to cell motility. The main findings of the study are: SHARPIN is robustly phosphorylated by PKCalpha in vitro. Mass spectrometry and in silico approaches indicate that some of these sites may be genuinely phosphorylated in cells. Ser146 phosphorylation seems uninvolved in SHARPIN-mediated integrin inactivation but imitation of the non-phosphorylated state using a Ser to Ala mutant impairs SHARPIN interaction with Arp2/3. S146A and V240A/L242A mutations impair lamellipodia formation. S146E mutation has a much less pronounced effect on SHARPIN-Arp2/3 interaction and lamellipodia formation, being close to the effect of the wild type. The authors conclude, based on this, that the active form of SHARPIN is likely phosphorylated in Ser146. In 3D matrix invasion assay, CRISPR-mediated SHARPIN deletion decreased invasiveness in several tumor cell lines. Perhaps the most striking experiment is that the authors use SHARPIN-KO MDA-MB-231 cells reconstituted with GFP-SHARPIN, WT or 146A, in a xenograft model in zebrafish, and they see that only WT-expressing cells form distal clusters of tumor cells. This is an interesting paper that merits publication, but several issues need to be addressed.

1. Whereas the data are consistent and interesting, some points could be strengthened quantitatively. An example is the quantification of the invasion assays (Fig. 4), which is relatively crude (this reviewer has firsthand experience with these assays).
2. Does SHARPIN control actin polymerization in response to stimulation? For example, growth factors in the cells used, or PMA.
3. PKCalpha has been involved in the control of protrusion dynamics through its local effect on myosin II regulatory light chain (Asokan et al, 2014). Is PKCalpha actually phosphorylating SHARPIN in live cells? Are other kinases involved in vitro? This needs to be clearly and explicitly demonstrated.
4. SHARPIN phosphorylated in Ser146 locally at lamellipodia? This is a hard experiment that requires a phospho-specific antibody, but the subcellular localization of the effect may be critical towards explaining the relative importance and generality of this mechanism.
5. In the same vein, does SHARPIN S146E localize more readily to protrusions that Ser146A?
6. The inference that active SHARPIN is phosphorylated in Ser146 needs to be demonstrated formally for the story to be substantiated. One possible manner could be to immunoprecipitate the Arp2/3 complex and show that most of the associated SHARPIN is phosphorylated in this residue by mass spectrometry. Alternatively, the authors could pull down integrins and show that SHARPIN is not phosphorylated in this case, which is also suggested by their data.
7. The last piece of data is striking, but the xenograft nature of the experiment casts some doubt as to its significance. B16F10 cells similarly treated could be implanted in C57BL/6 mice to make a much stronger case.

Significance

The data presented here, if substantiated as indicated in the previous section, would constitute a significant addition to the state of the art. My enthusiasm for the story is curbed by the technical flaws (relatively minor) and conceptual gaps (more significant) stated above. My expertise is cell and molecular biology with focus on cytoskeletal-related cell motility.

Referees cross-commenting

I mostly concur with the revisions suggested by the other two reviewers. Reviewer #1 raises a significant point that needs to be addressed, which pertains to the overall levels of overexpression. Other than that, I stand by my review, which has many connection points with those of reviewers 1 and 3. I think this paper has been judged fairly and the experiments requested are not overly complicated. I understand if the authors cannot do the in vivo experiment in mice, and this alone should not be grounds for rejection. However, they need to address the rest of our points.

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 SHARPIN protein is involved in multiple cellular pathways, and a number of studies demonstrated that it can interact with many key proteins regulating cellular proliferation, adhesion, motility and other functions that are important in normal cell physiology, but are particularly relevant for cancer metastasis. This study addresses post-translational phosphorylation of SHARPIN in human cells, and identifies two functionally important sites that are specifically involved in the interaction with ARP2/3, as well as in lamellipodia formation and cell motility (invasion). The function of one of these phosphorylation sites, the S146, is further explored in KD, KO and rescue experiments, in several human cancer cell lines, and in zebrafish model, using the directed mutagenesis approach to abolish or mimic the phosphorylation of S146. The authors conclude that Ph-S146 modification of SHARPIN is specifically responsible for interaction with ARP2/3, lamellipodia formation, and cancer cell invasion.

Major comments:

• Are the key conclusions convincing?

Overall, the conclusions are convincing, though the question of wild-type and mutated SHARPIN in rescue experiments should be addressed, given the functional importance of SHARPIN overexpression in cancer. Indeed, throughout the paper, the authors monitor the expression levels of their ectopically expressed SHARPIN, and systematically refer to these molecules as being "overexpressed", without showing their relative levels with regard to normal endogenous levels of SHARPIN in these cell lines, prior to its KD/KO. However, as the typical cancer-related functions of SHARPIN are linked to its expression levels, it is important to understand whether the observed phenomena can be regarded as physiologically relevant, or are the authors operating out of the physiological scale. Given the capacity of HeLa and 293 cell lines to produce very high levels of ectopic proteins, this issue should be systematically controlled. Therefore, in all the experiments with ectopic expression of SHARPIN and its mutants, a western blot should be added to show the relative expression, compared to the endogenous protein.

There are also some major problems with the statistical analysis (please see below). - Should the authors qualify some of their claims as preliminary or speculative, or remove them altogether?

I would modify the text of the article referring to Fig.2A, where the authors qualify a 20-25% reduction of integrin activity as "significant". It is not clear whether they refer to statistical or functional significance, and the statement is generally misleading. - 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.

As mentioned previously, I believe that the expression levels of ectopic SHARPIN have to be systematically monitored in all assays, and compared to the normal, endogenous levels. - 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.

These experiments are fully realistic and should not involve additional costs. - Are the data and the methods presented in such a way that they can be reproduced?

There is some ambivalence about the "n" values in most of the experiments, where it is not clear at all what "n" means (the number of cells? Experimental points?) For example, in Fig.S1D, the legend states that n=1, whereas the panel shows three experimental points per condition, as well as error bars. I would suggest that the authors correct these obvious mistakes and explain more clearly what exactly "n" means, in each case. - Are the experiments adequately replicated and statistical analysis adequate?

I cannot comment on that because I do not understand what exactly "n" means (please see above). For example, if in Fig.3, "n=4" means that only four cells were analyzed per experimental condition, this is clearly not enough. The statistical analysis is not sufficiently described in the Materials and Methods section, and the reasons for attributing one, two or three stars to the results are not stated, either (normally, this information should be equally present in Legends to Figures). The choice of applying the t-test does not appear evident to me, in experiments that clearly require multiple statistical comparisons, and in general, the statistical analysis does not appear adequate.

Minor comments:

• Specific experimental issues that are easily addressable.

Please provide a clear gel for Fig.1A, especially the right-hand panel. The current results look as one big black spot, with two red frames added for no clear reason. - Are prior studies referenced appropriately?

I have no problem with the list of references. - Are the text and figures clear and accurate?

It will be a good idea to proof-read the text. For example, I have noticed a frequent use of the word "lamellApodia" instead of "lamellipodia", as well as other typing errors. - Do you have suggestions that would help the authors improve the presentation of their data and conclusions?

Molecular weight markers should be added to all western blots, and scale bars to all immunofluorescent images.

Significance

• Describe the nature and significance of the advance (e.g. conceptual, technical, clinical) for the field.

Given the functional importance of SHARPIN in cancer, and the fact that it interacts with multiple major regulatory networks, it is important to pinpoint the exact post-translational modification of this protein that is specifically responsible for interaction with ARP2/3, and for the invasion potential of cancer cells. - Place the work in the context of the existing literature (provide references, where appropriate).

I agree with the description of the field and the place of the current study that is provided in the manuscript, and do not have anything significant to add. - State what audience might be interested in and influenced by the reported findings.

The study will undoubtedly be interesting to scientists working with cell motility, ARP2/3-dependent lamellipodia formation, and eventually metastasis growth in cancer. - Define your field of expertise with a few keywords to help the authors contextualize your point of view. Indicate if there are any parts of the paper that you do not have sufficient expertise to evaluate.

I study the RAC1-WAVE-ARP2/3 regulatory pathway in normal and cancer cells. I did not have any problems with evaluating this work, either academically, or with regard to methodology.

Referees cross-commenting

I am very pleased to see that the three reviews have a very similar evaluation of this article, and hope that the raised questions will help the authors to improve the manuscript and successfully publish their work. I do not have any additional 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

1. General Statements

We thank the reviewer for stating that “The detailed analysis uses many state of the art techniques to address the role of ROR1 and is of great interest to a large audience including basic researchers in the field of cancer biology and oncologists in the clinic” and we appreciate the reviewer’s constructive suggestions. We have substantially revised our manuscript and plan to perform new experiments based on these valuable comments.

1. Description of the planned revisions

Three main points: (1) The importance of AURKB as a downstream effector of ROR1 [Reviewer #1: major #2] Based on these suggestions, we plan to perform a colony formation assay using AURKB-overexpressing cells with ROR1-knockdown. We will clarify this point in the revised manuscript.

(2) The link between ROR1 expression and YAP/BRD4 [Reviewer #1: major #5 and Reviewer #3: major #1] Based on the suggestion, we plan to perform the luciferase reporter assay. We will clearly describe this experiment in the revised manuscript.

(3) Single-cell analysis using other models to validate tumor heterogeneity [Reviewer #2: major #1 and Reviewer #3: major #2] Based on your suggestion, we plan to analyze primary human tumors (public data: for example, GSE155698, CRA001160) and examine PDO#1 xenografts (in-house data). We will clearly state this information in the revised manuscript.

For the two minor points suggested by Reviewer #2, we plan to (1) reanalyze TCGA data. (2) perform the organoid or colony formation assay to validate that the siRNA model functionally recapitulates the ROR1low vs. ROR1high phenotype.

Please see the “Authors’ responses to the reviewers' comments” for more details.

1. Description of the revisions that have already been incorporated in the transferred manuscript

As suggested by the reviewer, we have substantially revised our manuscript, and the changes are shown in red. • Reviewer #1: major comments #2, #3, #4, and #5; minor comments #1 and #2 • Reviewer #2: major comments #2, #3, and #4; minor comments #2, #3, #4, #8, and #10 • Reviewer #3: minor comments #1 and #2

Please see the “Authors’ responses to the reviewers' comments” for more details.

1. Description of analyses that authors prefer not to carry out

---

Authors’ responses to the reviewers' comments

Reviewer #1

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

In this manuscript the authors analyzed the role of ROR1 in pancreatic cancer progression and metastasis. They found that ROR1 expression is specifically increased in an partial EMT cell cluster upon scRNA-Seq of tumor cells derived from an orthotopic mouse PDAC model. Moreover, the ROR1 high population in tumors specifies cells with high proliferation and tumor initiation capacities, increased metastatic propensity and chemoresistance, since knockdown of ROR1 shows reduction of these features in vivo. By comparing transcriptomes from several in vivo models the authors identified that ROR1 acts through AURKB and that its expression is regulated by an upstream enhancer that is bound by YAP/TAZ and BRD4 complexes. With this study the authors identified a new targetable pathway that promotes tumor progression and metastasis in PDAC. The detailed analysis uses many state of the art techniques to address the role of ROR1 and is of great interest to a large audience including basic researchers in the field of cancer biology and oncologists in the clinic. However, some of the findings are a bit preliminary and the drawn conclusions are not sufficiently supported by the experimental data. Moreover, some findings seem a bit out of context and do not really help to bring the story forward. At other instances experimental details are missing to mechanistically demonstrate the role of ROR1. In particular it remains elusive how ROR1 is regulated, i.e. which signaling events are crucial to generate ROR1 high vs. low cells. I listed my specific comments below.

[Response] We thank the reviewer for stating that “The detailed analysis uses many state of the art techniques to address the role of ROR1 and is of great interest to a large audience including basic researchers in the field of cancer biology and oncologists in the clinic” and we appreciate the reviewer’s constructive suggestions. We have substantially revised our manuscript and plan to perform new experiments based on these valuable comments.

1. The authors' initial finding is that in the partial EMT cluster ROR1, but also other RTKs (out of 56) are specifically increased. What about the other RTKs? Why was ROR1 chosen to analyze more thoroughly?

[Response 1] We are thankful for the reviewer’s suggestion to clarify why ROR1 was selected. (1) Seven candidate genes (EPHA4, EPHA7, ERBB4, FGFR1, JAK3, LYN, and ROR1) were chosen as surface markers in the partial EMT cluster. (2) The genes were sorted in order of high expression. (3) ROR1 is reported to promote metastasis in breast cancer (Cui et al, 2013). The induction of metastasis is one of the functions of tumor-initiating cells. FGFR1 is already known to enhance the CSC-like phenotype in non-small cell lung cancer (Ji et al, 2016). (4) The antibody against ROR1 was marketed as available for cell sorting using FACS. Therefore, we focused on ROR1 as a potential new marker for tumor-initiating cells with a partial EMT signature.

References Cui B, Zhang S, Chen L, Yu J, Widhopf GF 2nd, Fecteau JF, Rassenti LZ, Kipps TJ. Targeting ROR1 inhibits epithelial-mesenchymal transition and metastasis. Cancer Res. 2013 Jun 15;73(12):3649-60. doi: 10.1158/0008-5472.CAN-12-3832. PMID: 23771907; PMCID: PMC3832210. Ji W, Yu Y, Li Z, Wang G, Li F, Xia W, Lu S. FGFR1 promotes the stem cell-like phenotype of FGFR1-amplified non-small cell lung cancer cells through the Hedgehog pathway. Oncotarget. 2016 Mar 22;7(12):15118-34. doi: 10.18632/oncotarget.7701. PMID: 26936993; PMCID: PMC4924774.

1. The finding of AURKB as crucial target of ROR1 is very weak and needs more in-depth analyses. It is not clear why AURKB was chosen over the other candidates. Is AURKB expression directly regulated by ROR1? Are the two genes directly linked? Can ROR1 deficiency be compensated by AURKB overexpression? Especially the decrease in AURKB protein level in Fig. 4K is not very convincing to account for the different phenotypes in ROR1 high and low cells. Is AURKB and ROR1 expression correlated in TCGA samples (like Fig. 8B)? In Fig. 4L the readout was changed from colony numbers to colony diameter. If AURKB is the crucial player downstream of ROR1, then colony formation efficiency should be affected at first. This needs to be shown. The statement in lines 223,224 that AURKB is a direct downstream target of ROR1 was not shown!

[Response 2-1: changed] We thank the reviewer for noting this issue. We have performed additional experiments to assess the hypothesis that AURKB is a crucial downstream target of ROR1. ROR1-knockdown not only suppressed AKT phosphorylation (Supplemental Figure 9A) but also decreased c-Myc protein levels and the expression of c-Myc target genes (CDK4, CCND1, CDK2, and CCNE1), leading to a reduction in RB phosphorylation (new Supplemental Figure 9B and 9C). Based on these results, ROR1 regulates c-Myc expression through AKT signaling, leading to the activation of the E2F network (new Supplemental Figure 9D). We added some figures and descriptions to the preliminary revision manuscript (new Supplemental Figure 9B–9D, lines 357–363, lines 649–651).

[Response 2-2: the planned revisions] We also plan to perform new experiments with a colony formation assay to determine whether ROR1 deficiency is compensated by AURKB overexpression. We agree that this experiment will confirm that AURKB is an important downstream target of ROR1 in PDAC proliferation.

[Response 2-3] In TCGA-PAAD dataset, AURKB expression was not correlated with ROR1 expression. Since the ROR1high cluster is a minor population in the tumor, a downstream analysis of specific clusters with results from a bulk study such as this TCGA dataset is difficult to perform.

[Response 2-4: changed] We have added a new graph of organoid formation efficiency (new Figure 4L) and changed some descriptions in the preliminary revision manuscript (line 227).

1. Fig. 4 A-E: The ROR1 KD was induced in vitro but not continued in vitro. The transient KD has a strong impact on tumor forming capacity, even though recovery of expression is likely within the first days in vivo. This is very interesting and underscores the role of ROR1 in tumor initiation and presumably independent of differences in proliferation. Would the results be different, if the DOX treatment would start with injection of the cells and continued in vivo? Is then tumor initiation not affected and maybe only tumor growth?

[Response 3: changed] We apologize for the confusing description in the original manuscript. In Fig. 4A–E, we used PDAC cells with stable expression of doxycycline-inducible shROR1. ROR1-knockdown was maintained in vivo by adding doxycycline to the drinking water. Continuous ROR1-knockdown suppressed tumor growth (Fig. 4C–E). Several statements we made were more ambiguous than intended, and we have adjusted the text and the figures for clarity in the preliminary revision manuscript (new Figure 4A and B, lines 203–204).

1. In Fig. 5 the authors show that ROR1 is highly expressed in tumors after gemcitabine treatment and conclude that the ROR1 high cells are a resistant population. However, this statement is too strong, since gemcitabine treatment could also lead to an upregulation of ROR1 in "low" cells during acquisition of chemoresistence. Together with our knowledge on the role of EMT in driving therapy resistance and therapy-mediated induction of EMT, such a scenario is equally likely. Similarly, the statement in lines 370-372 is not supported by experimental evidence.

[Response 4: changed] We appreciate the reviewer’s critical comments. As suggested, we have not clearly determined whether (1) the ROR1high cells survived gemcitabine treatment and/or (2) the ROR1low cells increased ROR1 expression upon exposure to this treatment. We have carefully changed some descriptions in the preliminary revision manuscript (lines 241–242, 382–383).

1. In order to understand how ROR1 is regulated, the authors use ATAC-Seq and cut and run and identified a putative upstream enhancer element (Fig. 7). Although this element increases the activity of the promoter fragment in a reporter construct, the experiments do not help to understand how ROR1 activity is increased specifically in the "high" cells. Are peaks of YAP1 and BRD4 also changed between hi/lo cells? Is YAP OE and KD (BRD4 OE and KD) or the use of the inhibotor JQ1 altering the activity of the reporter constructs (i.e. only of the enhancer-promoter combination but not of the promoter only construct)? This would help to strengthen a direct link between ROR1, YAP and BRD4. Is YAP activity different in ROR1 high vs. low cells?

[Response 5-1: changed] We thank the reviewer for this important comment. We have shown differences in chromatin accessibility and histone modification of the ROR1 enhancer between ROR1high and ROR1low cells using ATAC-seq and CUT&RUN assays (Fig. 7B). Very few ROR1high/low cells are present in xenograft. We were not successful in experiments examining the binding of YAP and BRD4 to enhancers in ROR1high/low cells because of the technical limitations in the ChIP and CUT&RUN assays. Instead, we used public data to examine YAP and BRD4 occupancy at the ROR1 enhancer region of cell lines with low ROR1 expression. In T-47D and MCF7 cells (breast cancer cells, low ROR1 expression), YAP and BRD4 did not bind to the ROR1 enhancer region (new Figure 8D and 8I). We have added figures and some descriptions to the preliminary revision manuscript (new Figure 8D and 8I, lines 304–309, line 768).

[Response 5-2: the planned revisions] We plan to perform new experiments with the reporter assay you suggested. We agree that this experiment will help strengthen the direct link between ROR1, YAP and BRD4.

[Response 5-3] As shown in Figure 8C, GSEA revealed that ROR1high cells in both S2-VP10 xenografts and PDO#1 xenografts expressed higher levels of YAP-regulated genes than ROR1low cells in these xenografts. We have added a description of this result as follows: “Thus, ROR1high cells have higher YAP activity than ROR1low cells.” (lines 304–305).

1. In Fig. 8A the authors identified 202 antigens that match the H3 monomethylation / acetylation pattern. How was YAP etc. chosen?

[Response 6] We apologize for the poor description in the original manuscript. We chose YAP and BRD4 based on the following criteria: (1) these antigens are expressed in S2-VP10 cells and PDO#1 and (2) bind to the ROR1 enhancer region (based on an analysis of public data).

Minor: 1. Fig. 2D,E: What is actually shown here? Is there an overlap between the genes that define ROR1 high vs. low cells in both approaches? The gene list should be provided.

[Response: changed] We apologize for the poor description in the original manuscript. We have added this information to the preliminary revision manuscript (new Supplemental Table 3).

1. Fig. 3G: I suggest to include the images of the tumors from the ROR1 low cells in the main figure as well.

[Response: changed] We appreciate the reviewer’s suggestion. We have moved this information from the supplementary information to the main figure in the preliminary revision manuscript (new Figure 3G, lines 186–189).

Reviewer #1 (Significance (Required)):

PDAC is a very aggressive desease with very low 5-year survival rates. Understanding of the pathobiology is of keen interest. The findings of the authors are of high significance and extremely relevant as they provide a mechanism that can also be targeted by specific drug combinations, i.e. standard care gemcitabine with specific ROR1 inhibition. The findings are of great interest to a large audience including basic researchers in the field of cancer biology and oncologists in the clinic.

[Response] We greatly appreciate the reviewer’s comments.

Reviewer #2

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

In this work Yamazaki and colleagues performed single cell RNA sequencing of one xenograft tumor formed by the S2-VP10 PDAC cell line to explore PDAC intratumor heterogeneity. Using this model they identified ROR1 as heterogeneously expressed in neoplastic cells. Using further in vivo and in vitro models they show that ROR1high cells have higher tumor initiation capacity than ROR1low. By histone and ATAC-seq analyses, they identify a ROR1 enhancer upstream the promoter and show that YAP and BRD4 bind to this genomic region and that BRD4 inhibition by JQ1 reduces ROR1 expression and organoid formation. The data, figures and methods are nicely and clearly presented.

[Response] We thank the reviewer for stating that “The data, figures and methods are nicely and clearly presented”, and we appreciate the reviewer’s constructive suggestions. We have substantially revised our manuscript and plan to perform new experiments based on these valuable comments.

Major comments

1. The authors use one xenograft tumor as starting model and all conclusions are derived from the data generated with this model. To support the existence of identifie heterogeneity in the PDAC neoplastic compartment, I would strongly suggest to validate the existence of the partial EMT population and the ROR1 heterogeneity in single cell data bases generated from primary human tumors.

[Response 1: the planned revisions] We thank the reviewer for the positive suggestion. We plan to perform a new analysis of available public single-cell data from human PDAC tumors. In addition, we also launched a single-cell analysis of PDO#1 xenografts.

1. In Fig. 3G, it is mentioned that tumors grown from ROR1high cells recapitulate the original PDOx histology thus suggesting that ROR1high cells in the tumor are the actual TICs. ROR1low cells could also grow tumors, just with lower incidence. Are these tumors any different to the ROR1high derived ones? Is it just a lower tumor initiation capacity (TIC) or they can not recapitulate the tumor as the ROR1high cell? Can they also give rise to differentiated progeny cells? This should appear in the main text and not only in the discussion. I would suggest to move panel 3G to supplementary figure.

[Response 2: changed] We thank the reviewer for noting this issue and apologize for the confusing description in the original manuscript. ROR1low cells generated tumors at a low frequency, and these tumors showed a hierarchical histology mimicking the original tumor. As suggested, we have added this information to the main text (new Figure 3G, lines 186–189).

1. In line 160 you mention that known CSC markers such as CD44, PROM1 and DCLK1 are not differentially expressed between ROR1 high and low populations. Then, in figure 3H,I you analyze the expression of CD44v6 together with ROR1. I would try to put this information together in the text, or at least in fig. 3 start with something like "we had seen that both ROR1high and low express CD44, however...". In any case, I feel that the experiment with CD44 could be obviated (or at least moved to supplementary), as it brings the question of weather this is also true for DCLK1 or CD133.

[Response 3: changed] We appreciate and agree with the reviewer's comment on this point. Accordingly, we have moved this figure to the supplementary information and changed the description (new Supplemental Figure 5C and 5D, lines 191–196).

1. JQ1 has been described to inhibit PDAC growth by downregulation of MYC. To unequivocally link the effect of JQ1 in the downregulation of ROR1 (Fig. 8M) as discussed in the text it would be important to exclude that other mechanisms such as MYC downregulation are taking place. For example, does JQ1 treatment of ROR1low cells also reduce their colony formation capacity (in an experiment such as the one in fig. 3C). Or does ROR1 re-expression in Fig. 8M rescue the JQ1 effect? These or other experiments could help to establish a stronger link between (BRD4/JQ1) and ROR1.

[Response 4: changed] We thank the reviewer for this important comment. As mentioned in the response to Reviewer #1-major comment #2, we newly found that ROR1 regulates c-Myc expression through AKT signaling, leading to the activation of the E2F network (new Supplemental Figures 9B–9D, lines 357–363).

Minor comments 1. The data are nicely presented (text and figures) and the conclusions are clear. My suggestion to make the story more "catchy" at the beginning would be, if possible, to start from the observation done in primary human data and then move to the PDX model to explore ROR1 as a TIC marker in PDAC. For this, you could use available public single cell data of human PDAC tumors. If this doesn't work (it is of course possible that by unsupervised analysis you don't get the same clusters as in the PDX with the partial EMT cluster popping up), it would be nice if some primary tumor data came early in the story (currently the first figure showing heterogeneity in primary samples is in supplem fig. 4A).

[Response: the planned revisions] We thank the reviewer for these excellent comments. As suggested, we plan to perform several new analyses (please see the previous comment for details: Reviewer #2-major comment #1).

1. It is not clear if the xenografts were subcutaneous or orthotopic. It would be good to include this information in the main text (line 102) and the methods so that the reader knows what is the exact model that has been used.

[Response: changed] We thank the reviewer for this comment and apologize for the poor description in the original manuscript. As suggested, we have added this information to the preliminary revision manuscript (line 101).

1. In Fig. 2F and 2G I would highlight the EMT pathway to help the reader.

[Response: changed] We thank the reviewer for this comment. As suggested, we have changed the relevant figures in the preliminary revision manuscript (new Figure 2F and 2G).

1. In Supp Fig 4B it would be nice to have an amplified view of the staining as in panel C of the same figure.

[Response: changed] We thank the reviewer for this comment. As suggested, we have added high-magnification images of the staining in the preliminary revision manuscript (new Supplemental Figure 4A and 4B).

1. In the same figure (Fig. 4A-D) ROR1 shows an apical staining pattern that doesn't seem to resemble the staining in patient samples. I am not an expert in pathology evaluation but I would recommend a pathologist to give her/his opinion. Possibly, during the PDX process, few cells from the original patient tumor are selected giving a different staining pattern.

[Response] We appreciate the reviewer's comment on this point. Dr. Ito, a coauthor of this paper, is a pathologist. We have changed some images of staining in patient samples (new Supplemental Figure 4A). We agree that ROR1 shows an apical staining pattern in PDX samples. However, some sites show similar apical staining patterns in patient samples (Patient #2 and Patient #4 in the new Supplemental Figure 4A). We propose that PDX mimics the original patient tissue because it has heterogeneity of ROR1 expression and morphological features indicative of a luminal structure.

1. In the analyses of TCGA data, be aware that only 150 from the original dataset are actual PDAC tumors. The dataset contains otherwise data from cell lines, PDX, normal tissue, etc that should be removed for a proper analysis (see DOI: 10.3390/cancers11010126)

[Response: the planned revisions] We thank the reviewer for the careful review of this issue. We are currently reconsidering with the pathologist whether the samples are appropriate based on TCGA data (diagnosis and pathology sections) and the paper you presented. The current data (Figures 3A, 4J, and 8B) were analyzed for samples excluding cell lines, PDX, and normal tissue in the TCGA-PAAD dataset.

1. Does ROR1 correlate with RFS? This would nicely fit with the concept of TIC and metastasis.

[Response] We thank the reviewer for noting this issue. Unfortunately, no correlation was observed between ROR1 expression and RFS.

1. Line 219: ROR1 is not "depleted" in the lines as it is a downregulation model. "ROR1-downregulated" would be more correct.

[Response: changed] We thank the reviewer for this suggestion and agree with your comment. We have corrected this term accordingly in the preliminary revision manuscript (line 223).

1. It would be good to have a supplem figure showing that siROR1 cells show reduction organoid formation, to validate that the siRNA model functionally recapitulates the ROR1low vs high phenotype.

[Response: the planned revisions] We thank the reviewer for this suggestion. We plan to perform a colony formation assay.

1. Some of the supplemental figures are only referred in the discussion although they appear earlier than other in the main text. This is a bit confusing when going through the figures.

[Response] We apologize for the poor description in the original manuscript. We have adjusted the order of the supplemental figures in the preliminary revision manuscript.

CROSS-CONSULTATION COMMENTS I agree with the importance of addressing points 2 (link to AURKB), 4 (selection vs acquisition), 5 (mechanism in high vs low cells) raised by Reviewer 1, and the comments from Reviewer 3. I think that the study of other RTKs (point 1 from Reviewer 1) is not the focus of the story. It would be nice if the authors can comment on why they chose ROR1 but the fact that are other differentially expressed genes does not exclude the validity of the current story. I fell that the in vivo sustained KD experiment (point 3 from Reviewer 1) although interesting, it is not mandatory for a revision of this manuscript in case the adaptation of the animal protocol represents a long process. The experiment provided already in the current version is the best approach to address the role of ROR1 at the early initiation phase.

[Response] We thank the reviewer for these positive comments. As suggested, we have substantially revised our manuscript.

Reviewer #2 (Significance (Required)):

Significance: This is a neat and interesting work with potential implications for the clinical field of pancreatic cancer as the authors identified a new subpopulation with enhanced tumor initiating cell capacity. However, the use of JQ1 for pancreatic cancer has been previously discussed mainly linked to MYC inhibition, but also to stromal reprogramming or DNA damage induction. I missed some discussion in this regard in the discussion section. What is adding the work to the field of JQ1 treatment in PDAC? IN a way, how do the authors foresee that the discovery of ROR1high cells and the regulation of ROR1 by BRD4 and YAP will be beneficial when considering JQ1 in the clinics? Maybe by stratifying patients? Or by following ROR1 upregulation upon initial chemotherapy? These questions are just suggestions. In general, some discussion to put the work into the context of previous works using JQ1 in PDAC would be nice.

[Response: changed] We thank the reviewer for this comment. As you suggested, we have added a description of the proposed use of JQ1 and BRD4 inhibitors in ROR1high PDAC treatment to the Discussion section (lines 412–416).

I believe that this work would be interesting not only to the pancreatic cancer community but also to a more general public working on cancer and/or stemmness as it touches several interesting points in that regard that can be applicable to other systems. My own work is focused on pancreatic cancer, patient heterogeneity and stromal interactions. I am not an expert on histone or ATACseq analyses.

[Response] We greatly appreciate the reviewer’s comments.

Reviewer #3

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

Summary Yamazaki et al investigate partial EMT in pancreatic cancer and provide data that ROR1 marks pancreatic tumor cells that are capable of initiating tumors. The authors exploit scRNAseq of pancreatic tumor xenografts to identify a cluster of cells showing a partial EMT phenotype. The found 7 RTKs expressed more highly in this partial EMT cluster and focus their attention on ROR1, an 'orphan' receptor that has been implicated in WNT signaling and EMT previously. Validation experiments using ROR1-high vs low cells support that ROR1 expression correlates with EMT, poor outcome in human PDA patients, tumor forming and colony forming capacity. They also show that ROR1 high cells form tumors that recapitulate parental tumor histology. The authors show that ROR1 expression is associated with EF2 transcription factor activity, elevated expression of multiple targets including AURKB. Pharmacologic inhibition of AURKB reduces colony formation and genetic loss of ROR1 combined with chemotherapy (gemcitabine) has potent anti-tumor activity in vivo. The authors show that ROR1 expression is elevated in metastatic lesions and identify a novel enhancer element that putatively drives ROR1 expression in tumor cells. They provide evidence that this element is engaged by YAP/BRD4 and show that BRD4 inhibition reduces tumor cell colony formation. The manuscript is a solid combination of techniques with adequate controls and statistics.

[Response] We thank the reviewer for stating that “The manuscript is a solid combination of techniques with adequate controls and statistics”, and we appreciate the reviewer’s constructive suggestions. We have substantially revised our manuscript and plan to perform new experiments based on these valuable comments.

Major Comments: The overall conclusion that ROR1 expression marks a subset of pancreatic cancer cells that have the ability to initiate tumors is supported by the data provided. The correlative data are strong and the demonstration that loss of ROR1 reduces colony formation, reduces metastatic lesions and enhances the efficacy of chemotherapy are compelling. Additionally, the demonstration that ROR1 expression is elevated in metastatic lesions is consistent with many other drivers/markers of EMT in pancreatic cancer.

The conclusion that ROR1 expression is driven by YAP/BRD4 is interesting and provides important mechanistic depth to the study. However, this conclusion could be strengthened by use of a suitable rescue experiment. For instance does overexpression of ROR1 rescue the effect of BRD4 inhibition or loss of YAP?

[Response 1: the planned revisions] We thank the reviewer for this comment. We completely agree with the reviewer’s suggestion. However, the suggested examination to determine whether overexpression of ROR1 rescues the effect of BRD4 inhibition or loss of YAP may not be suitable because BRD4 and YAP act as transcriptional coregulators of various target genes. Instead, as mentioned in response to Reviewer #1-major comments 5-2, we plan to perform new experiments using a reporter assay.

A challenge with the data presented in Figure 1, the scRNA-seq data that lead them to ROR1, is that it is not stated how many tumors are used to generate the scRNA-seq data and the overall number of tumor cells analyzed is relatively low (993). The authors should provide the number of tumors used for the initial scRNA-seq. A general concern with any scRNA-seq data is batch effect, this is mitigated to a degree by the follow on studies that provide functional validation of ROR1 in multiple cell lines.

[Response 2: changed and the planned revisions] We appreciate the reviewer’s comments. As suggested, we have added this information to the preliminary revision manuscript (line 104). In addition, as mentioned in response to Reviewer #2 major comment #1, we plan to perform a new single-cell analysis of PDO xenografts (in-house data) and human PDAC tumors (available public data).

The data and methods are provided in an adequate manner. Reproduction of the experiments is likely. The authors use multiple cell lines and tools that are generally available. The authors note a limitation of the study is that only human tumor xenografts were exploited.

[Response] We thank the reviewer for the positive comment.

Minor comments: Figure 1E and text page 9. The text identifies MERB3 as a gene that marks the partial EMT cluster, I believe this is a type and the gene should actually be MSRB3.

[Response: changed] We apologize for the typo. We have corrected this error accordingly (line 114).

Please provide the dose of gemcitabine in the legend for figure 5

[Response: changed] We apologize for the poor description in the original manuscript. We have added this information.

CROSS-CONSULTATION COMMENTS I think the comments from Referee #2 are pretty reasonable - have no additions

Reviewer #3 (Significance (Required)):

Intratumor heterogeneity is a major challenge for the treatment of many cancers, including pancreatic cancer. The data provided support that ROR1 marks a subset of cancer cells in pancreatic tumors that have the capacity to drive intratumor heterogeneity. If supported these data have the potential to drive significant impact. Identification of a marker and a targetable pathway that supports tumor initiation in pancreatic cancer has the potential to nominate companion therapies that enhance the efficacy of standard of care approaches. Further, identification of a pathway that drives partial EMT in pancreatic cancer provides a substantial increase in baseline knowledge of intratumor heterogeneity.

These data would be broadly interesting to scientists interested in the tumor microenvironment, metastasis, therapy resistance and tumor progression. In addition, oncologists focused on drug development and combinatorial therapy will find this manuscript of interest.

[Response] We greatly appreciate the reviewer’s 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 #3

Evidence, reproducibility and clarity

Summary

Yamazaki et al investigate partial EMT in pancreatic cancer and provide data that ROR1 marks pancreatic tumor cells that are capable of initiating tumors. The authors exploit scRNAseq of pancreatic tumor xenografts to identify a cluster of cells showing a partial EMT phenotype. The found 7 RTKs expressed more highly in this partial EMT cluster and focus their attention on ROR1, an 'orphan' receptor that has been implicated in WNT signaling and EMT previously. Validation experiments using ROR1-high vs low cells support that ROR1 expression correlates with EMT, poor outcome in human PDA patients, tumor forming and colony forming capacity. They also show that ROR1 high cells form tumors that recapitulate parental tumor histology. The authors show that ROR1 expression is associated with EF2 transcription factor activity, elevated expression of multiple targets including AURKB. Pharmacologic inhibition of AURKB reduces colony formation and genetic loss of ROR1 combined with chemotherapy (gemcitabine) has potent anti-tumor activity in vivo. The authors show that ROR1 expression is elevated in metastatic lesions and identify a novel enhancer element that putatively drives ROR1 expression in tumor cells. They provide evidence that this element is engaged by YAP/BRD4 and show that BRD4 inhibition reduces tumor cell colony formation. The manuscript is a solid combination of techniques with adequate controls and statistics.

Major Comments:

The overall conclusion that ROR1 expression marks a subset of pancreatic cancer cells that have the ability to initiate tumors is supported by the data provided. The correlative data are strong and the demonstration that loss of ROR1 reduces colony formation, reduces metastatic lesions and enhances the efficacy of chemotherapy are compelling. Additionally, the demonstration that ROR1 expression is elevated in metastatic lesions is consistent with many other drivers/markers of EMT in pancreatic cancer.

The conclusion that ROR1 expression is driven by YAP/BRD4 is interesting and provides important mechanistic depth to the study. However, this conclusion could be strengthened by use of a suitable rescue experiment. For instance does overexpression of ROR1 rescue the effect of BRD4 inhibition or loss of YAP?

A challenge with the data presented in Figure 1, the scRNA-seq data that lead them to ROR1, is that it is not stated how many tumors are used to generate the scRNA-seq data and the overall number of tumor cells analyzed is relatively low (993). The authors should provide the number of tumors used for the initial scRNA-seq. A general concern with any scRNA-seq data is batch effect, this is mitigated to a degree by the follow on studies that provide functional validation of ROR1 in multiple cell lines.

The data and methods are provided in an adequate manner. Reproduction of the experiments is likely. The authors use multiple cell lines and tools that are generally available.

The authors note a limitation of the study is that only human tumor xenografts were exploited.

Minor comments:

Figure 1E and text page 9. The text identifies MERB3 as a gene that marks the partial EMT cluster, I believe this is a type and the gene should actually be MSRB3.

Please provide the dose of gemcitabine in the legend for figure 5

Referees cross-commenting

I think the comments from Referee #2 are pretty reasonable - have no additions

Significance

Intratumor heterogeneity is a major challenge for the treatment of many cancers, including pancreatic cancer. The data provided support that ROR1 marks a subset of cancer cells in pancreatic tumors that have the capacity to drive intratumor heterogeneity. If supported these data have the potential to drive significant impact. Identification of a marker and a targetable pathway that supports tumor initiation in pancreatic cancer has the potential to nominate companion therapies that enhance the efficacy of standard of care approaches. Further, identification of a pathway that drives partial EMT in pancreatic cancer provides a substantial increase in baseline knowledge of intratumor heterogeneity.

These data would be broadly interesting to scientists interested in the tumor microenvironment, metastasis, therapy resistance and tumor progression. In addition, oncologists focused on drug development and combinatorial therapy will find this manuscript of interest.

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 work Yamazaki and colleagues performed single cell RNA sequencing of one xenograft tumor formed by the S2-VP10 PDAC cell line to explore PDAC intratumor heterogeneity. Using this model they identified ROR1 as heterogeneously expressed in neoplastic cells. Using further in vivo and in vitro models they show that ROR1high cells have higher tumor initiation capacity than ROR1low. By histone and ATAC-seq analyses, they identify a ROR1 enhancer upstream the promoter and show that YAP and BRD4 bind to this genomic region and that BRD4 inhibition by JQ1 reduces ROR1 expression and organoid formation. The data, figures and methods are nicely and clearly presented.

Major comments

1. The authors use one xenograft tumor as starting model and all conclusions are derived from the data generated with this model. To support the existence of identifie heterogeneity in the PDAC neoplastic compartment, I would strongly suggest to validate the existence of the partial EMT population and the ROR1 heterogeneity in single cell data bases generated from primary human tumors.
2. In Fig. 3G, it is mentioned that tumors grown from ROR1high cells recapitulate the original PDOx histology thus suggesting that ROR1high cells in the tumor are the actual TICs. ROR1low cells could also grow tumors, just with lower incidence. Are these tumors any different to the ROR1high derived ones? Is it just a lower tumor initiation capacity (TIC) or they can not recapitulate the tumor as the ROR1high cell? Can they also give rise to differentiated progeny cells? This should appear in the main text and not only in the discussion. I would suggest to move panel 3G to supplementary figure.
3. In line 160 you mention that known CSC markers such as CD44, PROM1 and DCLK1 are not differentially expressed between ROR1 high and low populations. Then, in figure 3H,I you analyze the expression of CD44v6 together with ROR1. I would try to put this information together in the text, or at least in fig. 3 start with something like "we had seen that both ROR1high and low express CD44, however...". In any case, I feel that the experiment with CD44 could be obviated (or at least moved to supplementary), as it brings the question of weather this is also true for DCLK1 or CD133.
4. JQ1 has been described to inhibit PDAC growth by downregulation of MYC. To unequivocally link the effect of JQ1 in the downregulation of ROR1 (Fig. 8M) as discussed in the text it would be important to exclude that other mechanisms such as MYC downregulation are taking place. For example, does JQ1 treatment of ROR1low cells also reduce their colony formation capacity (in an experiment such as the one in fig. 3C). Or does ROR1 re-expression in Fig. 8M rescue the JQ1 effect? These or other experiments could help to establish a stronger link between (BRD4/JQ1) and ROR1.

Minor comments

1. The data are nicely presented (text and figures) and the conclusions are clear. My suggestion to make the story more "catchy" at the beginning would be, if possible, to start from the observation done in primary human data and then move to the PDX model to explore ROR1 as a TIC marker in PDAC. For this, you could use available public single cell data of human PDAC tumors. If this doesn't work (it is of course possible that by unsupervised analysis you don't get the same clusters as in the PDX with the partial EMT cluster popping up), it would be nice if some primary tumor data came early in the story (currently the first figure showing heterogeneity in primary samples is in supplem fig. 4A).
2. It is not clear if the xenografts were subcutaneous or orthotopic. It would be good to include this information in the main text (line 102) and the methods so that the reader knows what is the exact model that has been used.
3. In Fig. 2F and 2G I would highlight the EMT pathway to help the reader.
4. In Supp Fig 4B it would be nice to have an amplified view of the staining as in panel C of the same figure.
5. In the same figure (Fig. 4A-D) ROR1 shows an apical staining pattern that doesn't seem to resemble the staining in patient samples. I am not an expert in pathology evaluation but I would recommend a pathologist to give her/his opinion. Possibly, during the PDX process, few cells from the original patient tumor are selected giving a different staining pattern.
6. In the analyses of TCGA data, be aware that only 150 from the original dataset are actual PDAC tumors. The dataset contains otherwise data from cell lines, PDX, normal tissue, etc that should be removed for a proper analysis (see DOI: 10.3390/cancers11010126)
7. Does ROR1 correlate with RFS? This would nicely fit with the concept of TIC and metastasis.
8. Line 219: ROR1 is not "depleted" in the lines as it is a downregulation model. "ROR1-downregulated" would be more correct.
9. It would be good to have a supplem figure showing that siROR1 cells show reduction organoid formation, to validate that the siRNA model functionally recapitulates the ROR1low vs high phenotype.
10. Some of the supplemental figures are only referred in the discussion although they appear earlier than other in the main text. This is a bit confusing when going through the figures.

Referees cross-commenting

I agree with the importance of addressing points 2 (link to AURKB), 4 (selection vs acquisition), 5 (mechanism in high vs low cells) raised by Reviewer 1, and the comments from Reviewer 3. I think that the study of other RTKs (point 1 from Reviewer 1) is not the focus of the story. It would be nice if the authors can comment on why they chose ROR1 but the fact that are other differentially expressed genes does not exclude the validity of the current story. I fell that the in vivo sustained KD experiment (point 3 from Reviewer 1) although interesting, it is not mandatory for a revision of this manuscript in case the adaptation of the animal protocol represents a long process. The experiment provided already in the current version is the best approach to address the role of ROR1 at the early initiation phase.

Significance

This is a neat and interesting work with potential implications for the clinical field of pancreatic cancer as the authors identified a new subpopulation with enhanced tumor initiating cell capacity. However, the use of JQ1 for pancreatic cancer has been previously discussed mainly linked to MYC inhibition, but also to stromal reprogramming or DNA damage induction. I missed some discussion in this regard in the discussion section. What is adding the work to the field of JQ1 treatment in PDAC? IN a way, how do the authors foresee that the discovery of ROR1high cells and the regulation of ROR1 by BRD4 and YAP will be beneficial when considering JQ1 in the clinics? Maybe by stratifying patients? Or by following ROR1 upregulation upon initial chemotherapy? These questions are just suggestions. In general, some discussion to put the work into the context of previous works using JQ1 in PDAC would be nice.

I believe that this work would be interesting not only to the pancreatic cancer community but also to a more general public working on cancer and/or stemmness as it touches several interesting points in that regard that can be applicable to other systems.

My own work is focused on pancreatic cancer, patient heterogeneity and stromal interactions. I am not an expert on histone or ATACseq analyses.

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 manuscript the authors analyzed the role of ROR1 in pancreatic cancer progression and metastasis. They found that ROR1 expression is specifically increased in an partial EMT cell cluster upon scRNA-Seq of tumor cells derived from an orthotopic mouse PDAC model. Moreover, the ROR1 high population in tumors specifies cells with high proliferation and tumor initiation capacities, increased metastatic propensity and chemoresistance, since knockdown of ROR1 shows reduction of these features in vivo. By comparing transcriptomes from several in vivo models the authors identified that ROR1 acts through AURKB and that its expression is regulated by an upstream enhancer that is bound by YAP/TAZ and BRD4 complexes. With this study the authors identified a new targetable pathway that promotes tumor progression and metastasis in PDAC. The detailed analysis uses many state of the art techniques to address the role of ROR1 and is of great interest to a large audience including basic researchers in the field of cancer biology and oncologists in the clinic. However, some of the findings are a bit preliminary and the drawn conclusions are not sufficiently supported by the experimental data. Moreover, some findings seem a bit out of context and do not really help to bring the story forward. At other instances experimental details are missing to mechanistically demonstrate the role of ROR1. In particular it remains elusive how ROR1 is regulated, i.e. which signaling events are crucial to generate ROR1 high vs. low cells. I listed my specific comments below.

1. The authors' initial finding is that in the partial EMT cluster ROR1, but also other RTKs (out of 56) are specifically increased. What about the other RTKs? Why was ROR1 chosen to analyze more thoroughly?
2. The finding of AURKB as crucial target of ROR1 is very weak and needs more in-depth analyses. It is not clear why AURKB was chosen over the other candidates. Is AURKB expression directly regulated by ROR1? Are the two genes directly linked? Can ROR1 deficiency be compensated by AURKB overexpression? Especially the decrease in AURKB protein level in Fig. 4K is not very convincing to account for the different phenotypes in ROR1 high and low cells. Is AURKB and ROR1 expression correlated in TCGA samples (like Fig. 8B)? In Fig. 4L the readout was changed from colony numbers to colony diameter. If AURKB is the crucial player downstream of ROR1, then colony formation efficiency should be affected at first. This needs to be shown. The statement in lines 223,224 that AURKB is a direct downstream target of ROR1 was not shown!
3. Fig. 4 A-E: The ROR1 KD was induced in vitro but not continued in vitro. The transient KD has a strong impact on tumor forming capacity, even though recovery of expression is likely within the first days in vivo. This is very interesting and underscores the role of ROR1 in tumor initiation and presumably independent of differences in proliferation. Would the results be different, if the DOX treatment would start with injection of the cells and continued in vivo? Is then tumor initiation not affected and maybe only tumor growth?
4. In Fig. 5 the authors show that ROR1 is highly expressed in tumors after gemcitabine treatment and conclude that the ROR1 high cells are a resistant population. However, this statement is too strong, since gemcitabine treatment could also lead to an upregulation of ROR1 in "low" cells during acquisition of chemoresistence. Together with our knowledge on the role of EMT in driving therapy resistance and therapy-mediated induction of EMT, such a scenario is equally likely. Similarly, the statement in lines 370-372 is not supported by experimental evidence.
5. In order to understand how ROR1 is regulated, the authors use ATAC-Seq and cut and run and identified a putative upstream enhancer element (Fig. 7). Although this element increases the activity of the promoter fragment in a reporter construct, the experiments do not help to understand how ROR1 activity is increased specifically in the "high" cells. Are peaks of YAP1 and BRD4 also changed between hi/lo cells? Is YAP OE and KD (BRD4 OE and KD) or the use of the inhibotor JQ1 altering the activity of the reporter constructs (i.e. only of the enhancer-promoter combination but not of the promoter only construct)? This would help to strengthen a direct link between ROR1, YAP and BRD4. Is YAP activity different in ROR1 high vs. low cells?
6. In Fig. 8A the authors identified 202 antigens that match the H3 monomethylation/acetylation pattern. How was YAP etc. chosen?

Minor:

1. Fig. 2D,E: What is actually shown here? Is there an overlap between the genes that define ROR1 high vs. low cells in both approaches? The gene list should be provided.
2. Fig. 3G: I suggest to include the images of the tumors from the ROR1 low cells in the main figure as well

Significance

PDAC is a very aggressive desease with very low 5-year survival rates. Understanding of the pathobiology is of keen interest. The findings of the authors are of high significance and extremely relevant as they provide a mechanism that can also be targeted by specific drug combinations, i.e. standard care gemcitabine with specific ROR1 inhibition. The findings are of great interest to a large audience including basic researchers in the field of cancer biology and oncologists in the clinic.

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

This is a revision plan, the manuscript has not been modified yet as it is being transferred to a journal.

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

This study proposes (and uses) an elegant model of bacteria evolution to study how division of labor can emerge through the interaction between non-random mutations (occurring at some specific ``fragile' genomic sites) and genome architecture. The study is very interesting and the results are convincing. My main concerns are about the presentation of the model and results. Although I am confident about the results, some elements should be clarified for a better understanding and for a correct interpretation of the results. Two points in particular (detailed below as major comments) require clarification.

Major comments:

• the notion of telomere/centromere is used all throughout the paper but I think it is used in a misleading way. First, it seems that here there is only one telomere (but this is actually a detail of the model). More importantly, as long as I know, it is well known that in S. coelicolor the sequence degenerates more rapidly when getting closer to the telomeres (but telomeres are defined independently from this property). But here, the notion of telomere is precisely directly determined by its mutational instability (respectively, the centromere is defined by its stability). Although this is reasonable given the objective of the model, it forbid the use of sentences like "we observed that the genome of the evolved colony founded had two distinct regions: a telomeric [...] and a centromeric [...]" (line 234) or "When bacteria divide, mutations induced at fragile sites lead to the deletion of the part of the genome distal to them, causing large telometic deletions" (line 239 - this is not a result but a hidden description of the model) as this distinction between the two regions is not an outcome of the simulation but rather given a priori as a coded property of the fragile sites that all lead to deletions on the same -- called telomeric -- side (of course, formally if the genome contains no fragile site, there is no distinction but still). Please clarify this in the main text and in the methods. *

Authors response (AR, in the following): we agree with the reviewer that the directionality of the deletions determines centromere and telomere in our model (and the reviewer is correct that we only consider one arm of the chromosome). We will explicitly state both in the main text and in the methods that the model does not include any explicit centromeric and telomeric structure, and that the polarity of the genetic information (and thus centromere and telomere) depends on the choice of directionality of the deletions.

- In most part of the paper (methods, results, figures, sup mat...) antibiotics are considered to have a concentration (or a high/low production) but at least twice in the text (lines 165 and 488) it is said that only the presence/absence of antibiotics is modelled. I was not able to understand how the continuous values are transformed into presence/absence (is there a threshold?) but more importantly, I strongly suspect that this choice has a strong influence on the outcome. For instance, with a diffusion radius equals to 10, it means that an antibiotics producing cell is able to protect 2*\pi*10=~60 replicating cells. Hence, one could conjecture that the fraction of antibiotic-producing mutants should a little more than 2%... which is what is observed by the authors. So (1) please clarify this point (2) discuss (or experiments) the consequences of this choice on the conclusion.

AR: the reviewer is correct that antibiotics are modelled as presence/absence – this was done for computational efficiency. However, the probability that a bacterium deposits an antibiotic at a site within the deposition radius is a continuous number, as it depends on the number of antibiotic genes and growth genes. We will make this clear in the main text and in the methods.

Secondly, we show the effect of varying the deposition radius for the evolutionary dynamics in Supplementary Section S17. We will make this clear in the main text. For the area covered by different radius of antibiotic deposition, please see below.

* Minor comments: - line 262: "We conclude that genome architecture is a key prerequisit for the maintenance of mutation-driven division of labor". Given the model hypotheses you cannot be so affirmative (it is a key prerequisit... in this model!) *

AR: we will modify the statement as suggested. * *

- line 286: "cannot" is probably too strong. It has not been observed...

AR: we will modify the statement as suggested.

- line 288 and following: you seem to consider that there is "selection for diversity". Given the large number of possible antibiotics and given that cells are "automatically" resistant to the antibiotics they produce, could it be simply drift? There is a clear selection pressure to limit the number of growth-promoting genes but no such pressure exist for antibiotics. Hence their number could simply drift (note that figs 2 and SF1 both use a log scale; random variations due to drift could be hidden by the log. Fig. SF2 does use a log scale and shows a dynamics that---to my eyes---claims for drift rather than for selection of diversity).

AR: we agree with the reviewer that drift might contribute to the overall antibiotic diversity. This might be especially true for the antibiotic genes residing downstream of the fragile sites, which have low probability of expression in the wild-type (because of the many growth genes) and are deleted in the mutants. Duplications, deletions and modifications of these genes are effectively neutral, and are therefore likley subject to drift. We will include this discussion in the main text. However, bacteria are highly susceptible to the diverse antibiotics produced by other colonies (i.e. those produced – largely – by the mutants). These antibiotics and their diversity drives colony invasion and is thus selective. The overall number and diversity of antibiotics is therefore, at least in part, under selection.

- line 340: "ends" should be "end" when discussing the model - line 345: "a telomeric region" should be "telomeric regions" when discussing the bacteria - line 359: "S. ambofaciens" should be italic - line 365: same for "Streptomyces"

AR: we will modify the statement as suggested (and thank the reviewer for carefully reading the text).

- line 245 states that colonies begin clonally but methods (lines 434-438) don't support this. Colonies don't begin clonally but they begin without antibiotic-producing spores (see also line 618)

AR: we agree with the reviewer that colonies are not specifically initialised as clonal. We will modify the sentence as: By this process colonies eventually evolve to become functionally differentiated throughout the growth cycle.

- line 442: "their" should be "its" - line 446: "hotspot for recombination" no, for "deletion" - line 449: please remove brackets around the reference.

AR: we will modify the statement as suggested.

- line 458: if I understood it correctly, there is no explicit competition in the model. Competition simply comes from the asynchronous replication. Am I true? Could you clarify that point?

AR: The reviewer is correct that through asynchronous updating only one focal lattice site is update at a time. However, if a site is empty, the bacteria surrounding it are competing based on their replication rate kreplication. Dividing by the neighbourhood size (eta) simply ensures that a bacterium surrounded by a completely empty neighborhood replicates on average alpha_g times (alpha_g being the max growth rate). We will mention this in the methods.

- line 490: "the antibiotic deposited is chosen randomly and uniformly among them". This is not fully clear. I suppose the bacteria is still resistant to all the antibiotics it \it{can} produce?

AR: Yes. This is mentioned in the methods section “Replication”.

- figure SF1: please use the same scales as in figure 2 such that the two plots can be easily compared

AR: we will modify the x-axis to include the number of growth cycles.

- section S3 and figure SF4: What is to be understood from the figure is not clear to me. Seems that WTs win only if generalists produce less AB or replicate slower (?) Is it true?

AR: The reviewer is correct. In other words: when the artificial generalist has the same replication rate and the same antibiotic production rate as the WT, then the competition experiment ends with a near draw (the generalist still wins, but slowly). This means that the fitness cost associated to division of labor, i.e. to having two cell types doing the same work as one generalist – is small.

We will include this description in the section.

The figure is unfortunately complicated by the fact that we do not know a-priori how high the effective antibiotic production rate is (because antibiotics are spatially distributed by the stochastically generated mutants) – and so we had to make a large parameter screen to figure out the parameter values for which the competition experiment made most sense.

- I found it very difficult to draw conclusion from section S4, S5 and S6. These experiments should be analyzed with the help of mathematical analyses of the equations. Moreover, the understanding of these results are rendered difficult due to the lack of clarity regarding the discrete (or not) nature of the antibiotic production/action/diffusion

AR: We hope that we have clarified the distinction between antibiotic production rate and antibiotic presence/absence in the lattice.

The model is not amenable to analytical tractability, which makes it difficult to make exact statements based on the equations that govern it. However, we can check that the model is robust, and identify regions of parameter space where the model behaves in a qualitatively similar way to main text results.

Sections S4, S5 and S6 are essentially parameter screens to verify that the model reproduces the results reported in the main text for a broad range of parameters. The primary conclusion that can be drawn is that the model is robust to parameter changes.

Section S4 explores the model robustness to changes in two key parameters of the model: the antibiotic inhibition due to growth genes beta_g and the parameter h_g, which is the number of growth genes that produces half-maximum growth rate. Section S5 further analyses the relation between these parameters, and how they together determine the strength of the trade-off. Section S6, finally, shows that a strong trade-off is not a necessary requirement for evolution of division of labor as the division also depends (in a counterintuitive way) on the parameter alpha_g, the maximum antibiotic production rate.

We will include and expand these summarizing statements in each section, to make clear what each section achieves.

- S7 and fig SF9. It is unclear to me why the fraction of mutants decrease along time elapsed in the cycle. Please explain.

AR: The reason is that not all mutants are born with the same number of antibiotic genes (Fig. 3A). A mutant with fewer antibiotic genes might be susceptible to some of the antibiotics produced by another mutant, and could be killed by these antibiotics. Once a mutant is killed in the inner colony, a wt will replicate to fill the spot, and likely a wt offspring will take that site rather than another mutant. Thus there is a decline in overall mutant population.

We will include this discussion in Section S7.

- Figure SF14: what are the tin lines? if they correspond to the five repeats, how can it be that the bold line be the median?

AR: we realise that the caption should be clearer. Each of the five lines (both bold and thin) in each pane represents the median number of genetic elements over time. The bold line just highlights one randomly chosen simulation (the same for each genetic element), to better guide the eye.

We will clarify the caption of the figure.

- S13 and figure SF15: given that AB concentration is ON/OFF, is this result really surprising? This also questions about the accumulation of AB genes in the original model. Although the authors regularly claim that this is due to selection for diversity, drift could also be at play (see above)

AR: As mentioned above, we agree with the reviewer and we will mention that drift may co-determine antibiotic gene accumulation.

- S17: for radius 1, 2 and 3, the aliasing is likely to be strong. Hence, the results cannot be interpreted with this sole information. Please give e.g. how many cells are "protected" for each radius (e.g. for r_{alpha}=1, this value can vary between 1 and 9!)

AR: for radius=1, 2, 3, 5 ,8, 10 the area covered by antibiotic production is respectively 5 ,13, 29, 81, 197, 317. We will include this information in the figure.

- L742: "matching the antibiotic bitstring with the bitstring of the antibiotic". True and actually elegant but simpler formulation could ease the reading...

AR: We will change the sentence as follows: “Both antibiotics and antibiotic genes are characterised by a bitstring, which determines their type. Antibiotic resistance in the model is determined by matching these two strings.”

- lines 746-751 and figure SF21: There again, could it be a consequence of the AB ON/OFF diffusion model?

AR: we agree with the reviewer that a continuous diffusion model could affect resistance to antibiotics. We expect that the main effect will come from some antibiotics antibiotics having different concentrations. For instance, we could have a situation in which many deleterious antibiotics are produced in small amount, but have a compounding effect on the susceptible bacterium. This finer model of antibiotic production, diffusion and killing was not included in the model to limit the computational load.

- S18-S19-S20: what should the reader understand from these results? Please better comment the figures.

AR: we agree that figures in Section S18,19 and 20 could have more descriptive captions. Sections S18, 19 and 20 are parameter screen to check that the model is robust to changes in the mutation rates affecting fragile sites activation and de-novo formation. The primary result of Section S18 is that that division of labor evolves over a broad range of fragile site activation rates and de-novo fragile site formation rates (and even when these parameters are decreased by one order of magnitude).

Section S19 shows how these combination of parameters result in quantitative changes in genome composition.

Section S20 shows that the de-novo fragile site formation rate can be zero: as long as the system is initialised genomes that can divide labor, the fragile sites will persist even though no new ones are generated.* *

• CROSS-CONSULTATION COMMENTS Sorry about the confusion about the computation of the number of cells protected by a single AB-producing cell. Of course it is of the order 10*\pi^2 !!! The global argument still holds but the number of cells protected is of course larger than 60 (note that, due to aliasing at the periphery the exact number of cells in the protected area is difficult to determine). *

Author response: We hope the clarifications mentioned above answer the reviewer’s comment.

* Reviewer #1 (Significance (Required)):

First, an very importantly, I must say that I am no familiar with the biological model (Streptomyces coelicolor). So I am not fully able to judge the biological significance of this research (i.e. whether the way division of labor is achieved here enlights---or not---the biology of this bacteria). However, on the computational side, the model and the results (as they are summarized in the conclusion) are very interesting on their own and deserve publication.

Remark: a lots of supplementary results are added to the paper that are not not fully explained or analysed. Please, better discuss all these results and their significance. *

AR: we will extensively check and add detail to the supplementary material, ensuring that results are fully explained (see also response to reviewer 1).*

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

The manuscript "Evolution of genome fragility enables microbial division of labor" presents a model of genetically-based division of labour in bacterial colonies. It is postulated that two essential processes, growth and the important for elimination of competitors production of antibiotics, are poorly compatible in a single cell. The beneficial for a colony cell specialization is assumed to be determined only by genetic differences that appear via deletions of growth- promoting loci. These deletions and production of various antibiotics are mediated by a rather elaborate genetic architecture, which includes position-sensitive "fragile" sites, mutable antibiotic and growth-promoting genes. The model produces rather predictable results that under sufficiently strong incompatibility between growth and antibiotic production, the long-term evolution results in formation of mosaic of colonies, each specialized in production of its specific set of antibiotics. Such production is facilitated by evolving rapidly mutable genomes that constantly generate non-reproducing antibiotic-pumping cells.

The model appears very thoroughly developed and analyzed, and all major conclusion are intuitively appealing. Overall, the manuscript reads as a well-written quantitative proof of the principle of genetically-based division of labour between bacterial cells. The only part of the model that I'm a bit sceptical about is the unwarranted complexity of the genetic architecture. Unless the introduction of "fragile" sites and the directional ordering of genes is strongly justified by empirical data, a simpler and more clear assumption about mutational incapacitation of growth genes would suffice to reproduce the predicted phenomenology. So adding such empirical evidence would boost the relevance of the genetical part of the model. In the present form, all observed adaptations are inevitable simply because the expected division of labour will not evolve without each of them due to the design of the model. *

AR: We agree with the reviewer that a simpler model with a predetermined effect of mutations, such as to incapacitate the growth genes, would suffice to reproduce the phenomenology of the mutation-driven division of labor observed in Streptomyces. Adding the complexity of a genome architecture introduces one more hypothesis: that genome fragility can evolve to organize the division of labor. This hypothesis, supported by the results presented here, can be tested experimentally.

However, there is already some empirical support for our modelling choices: 1) mutation rates along the genome of Streptomyces are highly heterogeneous, 2) the genetic content is partitioned along the chromosome so that some genes are preferentially located in the mutationally quiet centromere, and others are in the mutationally active (sub)telomeric regions, 3) some cis genetic elements in Steptomyces’ genomes readily recombine to produce large-scale duplications and deletions (which we heavily simplified in the model as deletion-inducing fragile sites).

We will extend the introduction to include the references for the empirical support to our model.

* A couple of minor comments...

217 This is achieved when fewer growth-promoting genes are required to inhibit antibiotic 218 production (i.e. lower βg). Shouldn't it be "larger \beta_g"? *

AR: yes. Thanks for catching this!

* Whether in the main text or Supplementary materials, it woud help to add a complete population dynamics equation with all gain and loss terms. *

AR: we agree with the reviewer that it would be interesting to obtain a comprehensive population dynamics equation that captures the spatial dynamics of replication, mutation, and antibiotic production, causing colony formation and between-colony competition. However, deriving such equation would be a very big effort in itself, and we suspect that it would not be analytically tractable. Because of this, we prefer the “procedural” model description we gave – which also mirrors the model implementation (see github repository at github.com/escolizzi/strepto2).

* Strikingly, we find the opposite: division of labor evolves when 224 bacteria produce fewer overall antibiotics (lower αa), under shallow trade-off conditions 225 (hgβg = 5; see Suppl. Section S6).

I don't see why it is"striking". It seems perfectly explicable that a smaller \alpha requires more dedication to antibiotic production, thus favouring specialization. *

* *AR: we agree that we have not conveyed why we found this result surprising. We have set the trade-off shallow enough (h_g beta_g =5) that the generalist wins when alpha_g =1. In addition, lowering alpha_a makes the benefit of creating a mutant smaller, because a highly specialised mutant with zero growth genes makes fewer antibiotics. A generalist is proportionally less affected. Intuitively, we have compunded two benefits for the generalist.

But division of labor evolves, outcompeting the generalist – which surprised us.

We will modify the paragraph to better explain what we expected, and we will tone down the wording, removing the word “strikingly”.

*Reviewer #2 (Significance (Required)):

Due to my relative lack of familiarity with the literature on evolution of genetically-based division of labour, I would rather not comment on the degree of innovation of the manuscript.

The text is well written and is accessible to a wide readership, so it could be recommended to a general biological or evolutionary journal.

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

Summary: In this manuscript the authors explore the co-evolution of genomic architecture and division of labour in antibiotic production, in a model inspired by the bacterium Streptomyces coelicolor. In the model a genetic trade-off is implemented where the having a large number of growths promoting genes (and thus fast growth) leads to a low production of antibiotics. On the other hand, having fewer growth promoting genes allows for a higher production of antibiotics. This trade-off selects for a division of labour, where one sub population specializes in antibiotic production and another sub population specializes in reproduction. This division of labour is achieved by evolving the genome structure, so that growth promoting genes are clustered together, separated from the rest of the genome by several fragile sites (sites that allow for large deletions). This allows a single mutational event to delete a large number of growth-promoting genes, which creates a cell, lacking growth genes and that thus has a high antibiotic production (cell specializing in antibiotic production). In other words, the genome structure evolves to shape evolvability, so as to allow cells with a high growth rate to rapidly and repeatably evolve/mutate into cells with a high antibiotic production. This creates a division of labour where a part of the population specializes in growth/reproduction and another part specializes in antibiotics production. This model provides a tangible mechanism to explain a similar division of labour observed in S. coelicolor. This mechanism also fits well with the large deletions observed in antibiotic-hyperproducing S. coelicolor cells, which are also repeatably generated during colony growth.

Major comments: -Line 69, It would be good to give a bit more information here on the (number of) different types of antibiotics produced by S. coelicolor, to help the reader understand some of the modelling choices later on, such as allowing for the evolution of a large number (16 or higher if I understand correctly) of different antibiotics and a cell automatically being resistant to all antibiotics it produces (instead of having separate resistance genes). *

AR: we agree with the reviewer that adding this information would put the model more in focus. The total number of antibiotics that can be produced by the genus Streptomyces has been estimated to be of the order of 100000 (ten to the fifth, [Watve et al., 2001]). Although we use S. Coelicolor as reference model organism for our computational model, we simulate long-term evolutionary dynamics that diversify the antibiotic repertoire. Each antibiotic is represented by a 16 bits string, meaning that there are 2^16 (= 65536) possible antibiotics in the system – consistent with the number of possible antibiotics in the genus.

This being said, our model genomes evolve to have many more antibiotic genes than typical Streptomyces. Each species in the genus has up to 30 biosynthetic gene clusters [Genilloud, O. (2014)], a fraction of which make antibiotics. We discuss this discrepancy and propose solutions for this in the Discussion (also see below).

Regarding the possibility of separating antibiotic resistance from antibiotic synthesis: we (and most literature on the eco-evolutionary dynamics of antibiotic-producing bacteria) simplified antibiotic production as depending on individual “antibiotic biosynthetic genes”. In reality several genes in a cluster must be expressed to synthesize an antibiotic. A typical biosynthetic gene cluster also encodes resistance genes for the cognate antibiotic, to prevent cell suicide [Mak et al., 2014] – hence antibiotic genes providing resistance in the model. This being said, Streptomyces genomes also host resistance genes to antibiotics for which they have no biosynthetic pathway themselves, including efflux pumps that give some nonspecific resistance [Nag et al 2021].

Modelling antibiotic synthesis in more detail would allow to make a better model of antibiotic evolution, as well as to enrich the social dynamics of the model – because “cheaters” could evolve that are resistant but do not contribute to the antibiotics in the colony. These questions are certainly interesing, but would further complexify the model. They are exciting venues for future model expansions.

We will include the literature mentioned above in the introduction, and use these references to better motivate the model.

* -Lines 127-129 It is mentioned here fragile sites in the genome might represent transposable elements or long inverted repeats. Would both of these types of fragile sites behave the same? Has it been shown that both transposable elements and long inverted repeats can lead to large deletions from a linear chromosome? It would be nice to have a bit more background on how fragile sites might work or what they might look like in an empirical context. I am a bit unsure on this, but depending on their exact empirical nature, should fragile sites not also lead to increased rates of gene duplication near themselves? *

AR: we see that we have not made a clear connection between the introduction, where we introduce the mutational dynamics of Streptomyces, and the methods, where we introduce fragile sites.

Briefly, both duplications and deletions occur in Streptomyces, as well as circularization of the linear chromosome, conjugation, etc. [Hoff et al 2018,Tidjani et al 2019]. However, the outcome of all these mutations is biased towards deletion [hoff et al 2018, Zhang et al, 2020, Zhang et al, 2022]. There are many mechanisms involved in producing these mutations, forming the mutational hotspots, handling DNA breaks, and in the horizontal transfer of genetic material [Tidjani et al 2019; Lorentzi et al, 2021]. As the reviewer suggests – they do not behave all in the same way. To construct the model, we simplified all these mutational mechanisms into one genetic element, the “fragile site”, and assumed that they are solely responsible for the chromosomal-scale mutations that produce deletions.

We will add this information to the introduction (see also response to reviewer 2), and refer to it in the methods.

* -Line 160 As alluded to before, given the introduction provided, two assumptions come about here (lines 160-166) that lack a bit of justification/background/context. First, why does one allow the evolution of such a relatively large number of antibiotics? A bit more empirical in the introduction background would go a long way to making this assumption seem more justified. As far as I can see the genomic architecture leading to division of labour is only demonstrated for values of v that are 6 (i.e. 64 antibiotics) or above. Perhaps it is because I lack empirical background here, but this still seems to be a relatively large antibiotic space. Does the model also work with v=2? Perhaps it would be good to show a simulation with v=2 in supplementary material S16 as well. *

AR: Hopefully the previous comment on the number of possible antibiotics also clarifies this point.

We will carry out a simulation with v=2.

* -Line 166 The assumption is made that if a bacterium produces a certain antibiotic, it is automatically resistant to this antibiotic. Now it could be that this assumption is empirically rooted, in which case it would be good to allude to this empirical justification. I wonder how would the results be impacted if the resistance genes were separated from the antibiotic production genes? (I do not think additional simulations are in any way necessary on this point, but some more context/thoughts on this matter would be helpful, perhaps near lines 306-309) *

AR: Please see response to major comment on the possibility of separating antibiotic resistance from antibiotic synthesis. We will add the discussion there in the Discussion session.

* -Figure 1 In the subscript it becomes evident that the probability of large deletions due to fragile sites is much higher (10 fold) than single gene duplications, it seems to me this should be the other way around, single gene duplications and deletions could be much more probable than fragile site induced large deletions. Would the model still produce the same results if the values for mu-d and mu-f were switched around? (Again, I do not think additional simulations are per se required, some justification for this assumption would already be plenty). *

AR: We chose these parameter values because, empirically, large scale chromosomal rearrangements (deletions) occur more frequently than single gene duplication/deletion in Streptomyces – as they are the primary mechanism for Streptomyces development and division of labor. We now mention this in the caption of Fig. 1.

Still, would we expect results to be affected if mu_d > mu_f? We do not think so, for the following reason: mu_d and mu_f are per-gene probabilities, so the genomic probability of duplication/deletion and of fragile site activation will depend on the evolved number of genes.

in Fig. 5 we show that mu_f can be decreased by more than one order of magnitude and results do not change qualitatively. To compensate for a smaller per-gene deletion rate (mu_f), the evolved number of fragile sites per genome becomes larger (Suppl. Section S19, Fig. SF23). A similar compensatory increase of fragile sites could happen if duplications and deletions rate per gene were larger.

* Minor comments: -Line 36, perhaps replace "must" with " can" as there are other ways to achieve a division of labour that do not hinge on genomic architecture such as those listed in the next sentence. This sentence seems at odds with the next one, which lists ways to achieve cell differentiation that do not per se completely rely on genomic architecture such as gene regulation. Maybe consider moving this sentence to be on line 40 (after "...organized at the genome level remains unclear") *

AR: we will modify the text as suggested by the reviewer

* -Line 48, perhaps remove "disposable" as there is no particular reason the somatic tissue is disposable, furthermore it invokes the disposable soma theory of aging which is not relevant here *

AR: we will remove “disposable”.

* -Line 147-148 Why these particular relationships, as a reader I do not understand how these functions were constructed and how they might influence the results, a bit more justification might be helpful. Perhaps later on (results/discussion) also address what might happen if you were to use different functions? *

AR: we agree that these functions could use a little more explanation. The probability of replication is a function that increases with the number of growth genes. We assume that the function saturates, as growth cannot be arbitrarily large even if the genome hosts many growth genes. So we need at least two parameters: one for the maximum growth rate (alpha_g), and another that controls the curvature of the function (h_g). A simple choice is a Hill function, but other saturating functions would likely work just as well (e.g. an exponential function with a form alpha_g*(1-exp(-g/h_g)). Similarly, antibiotic synthesis inhibition from growth genes should tend to zero for larger numbers of growth genes, hence the exponential (but we expect that a hyperbolic form e.g 1/(1+g/beta_g) would work just the same).

As this discussion is rather technical, we will include it in the methods section.

* -I am clearly biased on this matter, since I work on evolvability. So, the authors should feel free to ignore this comment. Regardless, I think the authors have shown a wonderful example of the evolution of evolvability. Perhaps it would be nice to add a little bit of an evolvability angle in the discussion. In particular thinking about how fragile sites shape evolvability. *

AR: we agree with the reviewer that the work is a clear form of evolution of evolvability. We now explicitly mention this in the discussion.

* -Lines 404-411 It is great to see that the authors consider the wider applicability of their findings. It would be nice to add something here about the broader applicability in bacteria. As a large number of bacteria have circular chromosomes, how would these findings be impacted if circular chromosomes were at play? (I suspect they would largely still work in the same way, but keen to hear what the authors think). Referring to the work of Yona et al. 2012 on transient chromosomal duplications in yeast due to heat stress might also be good here, to show the more general applicability of the authors findings, this is another example where genomic architecture shapes evolvability. Yona AH, Manor YS, Herbst RH, Romano GH, Mitchell A, Kupiec M, Pilpel Y, Dahan O. Chromosomal duplication is a transient evolutionary solution to stress. Proc Natl Acad Sci U S A. 2012 Dec 18;109(51):21010-5. doi: 10.1073/pnas.1211150109. Epub 2012 Nov 29. PMID: 23197825; PMCID: PMC3529009. *

AR: Bacteria show many forms of targeted mutational dynamics (we do already mention CRISPR and HGT). It recently came to our attention that many bacterial and archea genomes host so-called Diversity-Generating Retroelements (DGR) [Macadangdang et al, 2022]. DGRs accelerate microbial evolution at specific sites and generate functional diversity. We will include this reference in the discussion.

We thank the reviewer for pointing us to the work on chromosomal duplication in yeast – we will also incorporate this “dramatic” form of duplication in the discussion.

* -Lines 412 -419 I agree with the authors that in practice the cells specializing in antibiotic production look somewhat like soma, however I would consider not using this term here as strictly speaking the antibiotics producing cells can still reproduce (be it at an extremely low rate, which leads to their loss). *

AR: We tone down both mentions of soma, as follows: “This gives rise to a division of labor driven by mutation, reminiscent of the division between germ and soma in multicellular eukaryotes.”

And, in the last sentence, we write: “...mutant cells *effectively* function as soma by enhancing...”

- Lines 434-438 If I understand correctly authors did not explicitly model the sporulation process (instead selecting random cells from the end of a cycle). I think this is a very good modelling choice that should not be changed; however, I do wonder how the results would be affected if sporulation was more explicitly modelled (for example by adding genes for sporulation, creating a 3 way trade-off between growth, sporulation and antibiotic production). Perhaps something that could be mentioned in the discussion.

AR: we agree with the reviewer that more complex evolutionary problem could be implemented in the system, e.g. through a gene type required for sporulation. They would likely have interesting outcomes. For instance, some bacteria may decide never to sporulate, while others could enhance their antibiotic resistance by turning into spores. Moreover, including additional functions together with an evolvable gene regulation could better capture the developmental dynamics observed through the life cycle of Streptomyces.

* I hope this review is of some use and helps the improvement of this manuscript. *

* Yours sincerely,

Timo van Eldijk

Reviewer #3 (Significance (Required)):

Significance: This study provides a clear conceptual advance by showing and studying how genome structure can evolve to create a division of labor. Thereby mechanistically explaining the division of labor in antibiotic production observed in S. coelicolor. It seems evident to me that whilst this study mainly focuses on S. coelicolor, the mechanism likely plays an important role in microbial evolution in general. Though others have previously theoretically explored such mechanisms, this study provides the first exploration modelled closely after an empirical system and hence provides a significant advance. In a more general sense, the evolution of genome architecture likely governs evolvability not just in microbes but in all life on earth. Therefore, I believe that this paper would be interesting for a general audience interested evolution. It would be of particular interest to those studying microbial evolution. My expertise lies in evolutionary biology, theoretical biology, microbial evolution and palaeontology. *

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 the authors explore the co-evolution of genomic architecture and division of labour in antibiotic production, in a model inspired by the bacterium Streptomyces coelicolor. In the model a genetic trade-off is implemented where the having a large number of growths promoting genes (and thus fast growth) leads to a low production of antibiotics. On the other hand, having fewer growth promoting genes allows for a higher production of antibiotics. This trade-off selects for a division of labour, where one sub population specializes in antibiotic production and another sub population specializes in reproduction. This division of labour is achieved by evolving the genome structure, so that growth promoting genes are clustered together, separated from the rest of the genome by several fragile sites (sites that allow for large deletions). This allows a single mutational event to delete a large number of growth-promoting genes, which creates a cell, lacking growth genes and that thus has a high antibiotic production (cell specializing in antibiotic production). In other words, the genome structure evolves to shape evolvability, so as to allow cells with a high growth rate to rapidly and repeatably evolve/mutate into cells with a high antibiotic production. This creates a division of labour where a part of the population specializes in growth/reproduction and another part specializes in antibiotics production. This model provides a tangible mechanism to explain a similar division of labour observed in S. coelicolor. This mechanism also fits well with the large deletions observed in antibiotic-hyperproducing S. coelicolor cells, which are also repeatably generated during colony growth.

Major comments:

• Line 69, It would be good to give a bit more information here on the (number of) different types of antibiotics produced by S. coelicolor, to help the reader understand some of the modelling choices later on, such as allowing for the evolution of a large number (16 or higher if I understand correctly) of different antibiotics and a cell automatically being resistant to all antibiotics it produces (instead of having separate resistance genes).
• Lines 127-129 It is mentioned here fragile sites in the genome might represent transposable elements or long inverted repeats. Would both of these types of fragile sites behave the same? Has it been shown that both transposable elements and long inverted repeats can lead to large deletions from a linear chromosome? It would be nice to have a bit more background on how fragile sites might work or what they might look like in an empirical context. I am a bit unsure on this, but depending on their exact empirical nature, should fragile sites not also lead to increased rates of gene duplication near themselves?
• Line 160 As alluded to before, given the introduction provided, two assumptions come about here (lines 160-166) that lack a bit of justification/background/context. First, why does one allow the evolution of such a relatively large number of antibiotics? A bit more empirical in the introduction background would go a long way to making this assumption seem more justified. As far as I can see the genomic architecture leading to division of labour is only demonstrated for values of v that are 6 (i.e. 64 antibiotics) or above. Perhaps it is because I lack empirical background here, but this still seems to be a relatively large antibiotic space. Does the model also work with v=2? Perhaps it would be good to show a simulation with v=2 in supplementary material S16 as well.
• Line 166 The assumption is made that if a bacterium produces a certain antibiotic, it is automatically resistant to this antibiotic. Now it could be that this assumption is empirically rooted, in which case it would be good to allude to this empirical justification. I wonder how would the results be impacted if the resistance genes were separated from the antibiotic production genes? (I do not think additional simulations are in any way necessary on this point, but some more context/thoughts on this matter would be helpful, perhaps near lines 306-309)
• Figure 1 In the subscript it becomes evident that the probability of large deletions due to fragile sites is much higher (10 fold) than single gene duplications, it seems to me this should be the other way around, single gene duplications and deletions could be much more probable than fragile site induced large deletions. Would the model still produce the same results if the values for mu-d and mu-f were switched around? (Again, I do not think additional simulations are per se required, some justification for this assumption would already be plenty).

Minor comments:

• Line 36, perhaps replace "must" with " can" as there are other ways to achieve a division of labour that do not hinge on genomic architecture such as those listed in the next sentence. This sentence seems at odds with the next one, which lists ways to achieve cell differentiation that do not per se completely rely on genomic architecture such as gene regulation. Maybe consider moving this sentence to be on line 40 (after "...organized at the genome level remains unclear")
• Line 48, perhaps remove "disposable" as there is no particular reason the somatic tissue is disposable, furthermore it invokes the disposable soma theory of aging which is not relevant here
• Line 147-148 Why these particular relationships, as a reader I do not understand how these functions were constructed and how they might influence the results, a bit more justification might be helpful. Perhaps later on (results/discussion) also address what might happen if you were to use different functions?
• I am clearly biased on this matter, since I work on evolvability. So, the authors should feel free to ignore this comment. Regardless, I think the authors have shown a wonderful example of the evolution of evolvability. Perhaps it would be nice to add a little bit of an evolvability angle in the discussion. In particularl thinking about how fragile sites shape evolvability.
• Lines 404-411 It is great to see that the authors consider the wider applicability of their findings. It would be nice to add something here about the broader applicability in bacteria. As a large number of bacteria have circular chromosomes, how would these findings be impacted if circular chromosomes were at play? (I suspect they would largely still work in the same way, but keen to hear what the authors think). Referring to the work of Yona et al. 2012 on transient chromosomal duplications in yeast due to heat stress might also be good here, to show the more general applicability of the authors findings, this is another example where genomic architecture shapes evolvability. Yona AH, Manor YS, Herbst RH, Romano GH, Mitchell A, Kupiec M, Pilpel Y, Dahan O. Chromosomal duplication is a transient evolutionary solution to stress. Proc Natl Acad Sci U S A. 2012 Dec 18;109(51):21010-5. doi: 10.1073/pnas.1211150109. Epub 2012 Nov 29. PMID: 23197825; PMCID: PMC3529009.
• Lines 412 -419 I agree with the authors that in practice the cells specializing in antibiotic production look somewhat like soma, however I would consider not using this term here as strictly speaking the antibiotics producing cells can still reproduce (be it at an extremely low rate, which leads to their loss).
• Lines 434-438 If I understand correctly authors did not explicitly model the sporulation process (instead selecting random cells from the end of a cycle). I think this is a very good modelling choice that should not be changed; however, I do wonder how the results would be affected if sporulation was more explicitly modelled (for example by adding genes for sporulation, creating a 3 way trade-off between growth, sporulation and antibiotic production). Perhaps something that could be mentioned in the discussion.

I hope this review is of some use and helps the improvement of this manuscript.

Yours sincerely,

Timo van Eldijk

Significance

This study provides a clear conceptual advance by showing and studying how genome structure can evolve to create a division of labor. Thereby mechanistically explaining the division of labor in antibiotic production observed in S. coelicolor. It seems evident to me that whilst this study mainly focuses on S. coelicolor, the mechanism likely plays an important role in microbial evolution in general. Though others have previously theoretically explored such mechanisms, this study provides the first exploration modelled closely after an empirical system and hence provides a significant advance. In a more general sense, the evolution of genome architecture likely governs evolvability not just in microbes but in all life on earth. Therefore, I believe that this paper would be interesting for a general audience interested evolution. It would be of particular interest to those studying microbial evolution. My expertise lies in evolutionary biology, theoretical biology, microbial evolution and palaeontology.

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 "Evolution of genome fragility enables microbial division of labor" presents a model of genetically-based division of labour in bacterial colonies. It is postulated that two essential processes, growth and the important for elimination of competitors production of antibiotics, are poorly compatible in a single cell. The beneficial for a colony cell specialization is assumed to be determined only by genetic differences that appear via deletions of growth- promoting loci. These deletions and production of various antibiotics are mediated by a rather elaborate genetic architecture, which includes position-sensitive "fragile" sites, mutable antibiotic and growth-promoting genes. The model produces rather predictable results that under sufficiently strong incompatibility between growth and antibiotic production, the long-term evolution results in formation of mosaic of colonies, each specialized in production of its specific set of antibiotics. Such production is facilitated by evolving rapidly mutable genomes that constantly generate non-reproducing antibiotic-pumping cells.

The model appears very thoroughly developed and analyzed, and all major conclusion are intuitively appealing. Overall, the manuscript reads as a well-written quantitative proof of the principle of genetically-based division of labour between bacterial cells. The only part of the model that I'm a bit sceptical about is the unwarranted complexity of the genetic architecture. Unless the introduction of "fragile" sites and the directional ordering of genes is strongly justified by empirical data, a simpler and more clear assumption about mutational incapacitation of growth genes would suffice to reproduce the predicted phenomenology. So adding such empirical evidence would boost the relevance of the genetical part of the model. In the present form, all observed adaptations are inevitable simply because the expected division of labour will not evolve without each of them due to the design of the model.

A couple of minor comments...

217 This is achieved when fewer growth-promoting genes are required to inhibit antibiotic 218 production (i.e. lower βg). Shouldn't it be "larger \beta_g"?

Whether in the main text or Supplementary materials, it woud help to add a complete population dynamics equation with all gain and loss terms.

Strikingly, we find the opposite: division of labor evolves when 224 bacteria produce fewer overall antibiotics (lower αa), under shallow trade-off conditions 225 (hgβg = 5; see Suppl. Section S6).

I don't see why it is"striking". It seems perfectly explicable that a smaller \alpha requires more dedication to antibiotic production, thus favouring specialization.

Significance

Due to my relative lack of familiarity with the literature on evolution of genetically-based division of labour, I would rather not comment on the degree of innovation of the manuscript.

The text is well written and is accessible to a wide readership, so it could be recommended to a general biological or evolutionary journal.

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 proposes (and uses) an elegant model of bacteria evolution to study how division of labor can emerge through the interaction between non-random mutations (occurring at some specific ``fragile' genomic sites) and genome architecture. The study is very interesting and the results are convincing. My main concerns are about the presentation of the model and results. Although I am confident about the results, some elements should be clarified for a better understanding and for a correct interpretation of the results. Two points in particular (detailed below as major comments) require clarification.

Major comments:

• the notion of telomere/centromere is used all throughout the paper but I think it is used in a misleading way. First, it seems that here there is only one telomere (but this is actually a detail of the model). More importantly, as long as I know, it is well known that in S. coelicolor the sequence degenerates more rapidly when getting closer to the telomeres (but telomeres are defined independently from this property). But here, the notion of telomere is precisely directly determined by its mutational instability (respectively, the centromere is defined by its stability). Although this is reasonable given the objective of the model, it forbid the use of sentences like "we observed that the genome of the evolved colony founded had two distinct regions: a telomeric [...] and a centromeric [...]" (line 234) or "When bacteria divide, mutations induced at fragile sites lead to the deletion of the part of the genome distal to them, causing large telometic deletions" (line 239 - this is not a result but a hidden description of the model) as this distinction between the two regions is not an outcome of the simulation but rather given a priori as a coded property of the fragile sites that all lead to deletions on the same -- called telomeric -- side (of course, formally if the genome contains no fragile site, there is no distinction but still). Please clarify this in the main text and in the methods.
• In most part of the paper (methods, results, figures, sup mat...) antibiotics are considered to have a concentration (or a high/low production) but at least twice in the text (lines 165 and 488) it is said that only the presence/absence of antibiotics is modelled. I was not able to understand how the continuous values are transformed into presence/absence (is there a threshold?) but more importantly, I strongly suspect that this choice has a strong influence on the outcome. For instance, with a diffusion radius equals to 10, it means that an antibiotics producing cell is able to protect 2\pi10=~60 replicating cells. Hence, one could conjecture that the fraction of antibiotic-producing mutants should a little more than 2%... which is what is observed by the authors. So (1) please clarify this point (2) discuss (or experiments) the consequences of this choice on the conclusion.

Minor comments:

• line 262: "We conclude that genome architecture is a key prerequisit for the maintenance of mutation-driven division of labor". Given the model hypotheses you cannot be so affirmative (it is a key prerequisit... in this model!)
• line 286: "cannot" is probably too strong. It has not been observed...
• line 288 and following: you seem to consider that there is "selection for diversity". Given the large number of possible antibiotics and given that cells are "automatically" resistant to the antibiotics they produce, could it be simply drift? There is a clear selection pressure to limit the number of growth-promoting genes but no such pressure exist for antibiotics. Hence their number could simply drift (note that figs 2 and SF1 both use a log scale; random variations due to drift could be hidden by the log. Fig. SF2 does use a log scale and shows a dynamics that---to my eyes---claims for drift rather than for selection of diversity).
• line 340: "ends" should be "end" when discussing the model
• line 345: "a telomeric region" should be "telomeric regions" when discussing the bacteria
• line 359: "S. ambofaciens" should be italic
• line 365: same for "Streptomyces"
• line 245 states that colonies begin clonally but methods (lines 434-438) don't support this. Colonies don't begin clonally but they begin without antibiotic-producing spores (see also line 618)
• line 442: "their" should be "its"
• line 446: "hotspot for recombination" no, for "deletion"
• line 449: please remove brackets around the reference.
• line 458: if I understood it correctly, there is no explicit competition in the model. Competition simply comes from the asynchronous replication. Am I true? Could you clarify that point?
• line 490: "the antibiotic deposited is chosen randomly and uniformly among them". This is not fully clear. I suppose the bacteria is still resistant to all the antibiotics it \it{can} produce?
• figure SF1: please use the same scales as in figure 2 such that the two plots can be easily compared
• section S3 and figure SF4: What is to be understood from the figure is not clear to me. Seems that WTs win only if generalists produce less AB or replicate slower (?) Is it true?
• I found it very difficult to draw conclusion from section S4, S5 and S6. These experiments should be analyzed with the help of mathematical analyses of the equations. Moreover, the understanding of these results are rendered difficult due to the lack of clarity regarding the discrete (or not) nature of the antibiotic production/action/diffusion
• S7 and fig SF9. It is unclear to me why the fraction of mutants decrease along time elapsed in the cycle. Please explain.
• Figure SF14: what are the tin lines? if they correspond to the five repeats, how can it be that the bold line be the median?
• S13 and figure SF15: given that AB concentration is ON/OFF, is this result really surprising? This also questions about the accumulation of AB genes in the original model. Although the authors regularly claim that this is due to selection for diversity, drift could also be at play (see above)
• S17: for radius 1, 2 and 3, the aliasing is likely to be strong. Hence, the results cannot be interpreted with this sole information. Please give e.g. how many cells are "protected" for each radius (e.g. for r_{alpha}=1, this value can vary between 1 and 9!)
• L742: "matching the antibiotic bitstring with the bitstring of the antibiotic". True and actually elegant but simpler formulation could ease the reading...
• lines 746-751 and figure SF21: There again, could it be a consequence of the AB ON/OFF diffusion model?
• S18-S19-S20: what should the reader understand from these results? Please better comment the figures.

Referees cross-commenting

Sorry about the confusion about the computation of the number of cells protected by a single AB-producing cell. Of course it is of the order 10*\pi^2 !!! The global argument still holds but the number of cells protected is of course larger than 60 (note that, due to aliasing at the periphery the exact number of cells in the protected area is difficult to determine).

Significance

First, an very importantly, I must say that I am no familiar with the biological model (Streptomyces coelicolor). So I am not fully able to judge the biological significance of this research (i.e. whether the way division of labor is achieved here enlights---or not---the biology of this bacteria). However, on the computational side, the model and the results (as they are summarized in the conclusion) are very interesting on their own and deserve publication.

Remark: a lots of supplementary results are added to the paper that are not not fully explained or analysed. Please, better discuss all these results and their significance.

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 thoughtful comments and suggestions and were pleased to note the quality of work and the findings were well received. Both reviewers commented that the datasets and findings represent a valuable resource for the field, and that this is a valuable resource paper. The only major concern related to conceptual advance and we provide a clear plan below that we believe will thoroughly address this issue.

Below, we provide point-by-point responses to each of the reviewer’s comments. These are presented to improve the conceptual advance in section 1 and address all other issues raised in other minor comments in section 2.

SPECIFIC ISSUES:

Section 1: Conceptual A____dvance

One main concern raised by both of the reviewers was that the main biological findings did not represent a major new conceptual advance, which is encapsulated by the comments below:

Reviewer 1:

“Major comments:

The conclusions of the manuscript are convincing. The robust data generated is inherently valuable and is of great interest of the field. However, my impression is that the authors did not utilize the power of their studies. The main message - Prox1 is a key regulator in promoting and maintaining lymphatic cell fate - is well accepted and has been intensively studied. Therefore, the main findings presented in the current manuscript are not conceptually-advancing.

Additional studies focusing on the function of some of the identified hit genes, such as cdh6, slc7a7, fabp11a in lymphatics - either in fish or in vitro - would significantly improve the novelty of the article. Zebrafish is an ideal experimental model that enable a relatively easy and quick way to address these questions. However, considering the time and expense of those experiments, in vitro studies would be also well appreciated instead of fish.”

Reviewer 2:

“While the work presented in this manuscript could be an interesting resource for the researchers in the field, it does not provide significant conceptual advances in the field.”

“CROSS-CONSULTATION COMMENTS

I agree with the excellent technical and statistical comments of Rev. 2. Overall, we are in agreement regarding the strength of the datasets as a resource for the field, but with limited conceptual novelty.”

We appreciate both reviewers’ feedback and take this concern seriously. We believe that the paper would be improved by utilizing the unique and extensive single cell resource to develop a deeper new understanding of the molecular control of lymphatic development. We also believe that the novel and new biology already presented in the paper can be better highlighted by re-writing the paper in key places.

In revision, we will therefore provide two major improvements to address these comments on conceptual advance.

Firstly, we will re-write relevant sections of the paper to provide an improved focus on the new biology uncovered and less of a technical focus on the resource generated. Specifically, the key new biological insights our thorough analysis already made in the manuscript that will be better highlighted include:

1. We define the precise timing of differentiation of lymphatics in vivo. Specified LECs in the cardinal vein are not significantly differentiated from their venous neighbours, rather they become transcriptionally distinct between 40 and 72 hours post fertilisation (hpf), well after the sprouting events by which they emerge from the vein. This was not previously shown in any study.
2. We show in definitive Prox1 null maternal and zygotic double mutant zebrafish for the first time that the sprouting of LECs from the cardinal vein occurs independently of Prox1 function, separating the initial sprouting and fating events. This is different from earlier findings in mouse (Yang, Garcia-Verdugo et al. 2012) and offers a unique new understanding.
3. We define the transcriptional program controlled by Prox1 during the maintenance of LEC fate in vivo at a whole transcriptome level. This has never actually been done before in any published literature. This reveals that Prox1 simultaneously up-regulates lymphatic marker genes and down-regulates blood vascular marker genes (which was known for a small number of markers). However, it also demonstrates the surprising finding that none of the change in fate from blood vascular to lymphatic vascular is regulated independently of Prox1 function. This shows that Prox1 is not “a” master regulator but “the only” master regulator of this fate decision in a definitive manner for the first time.
4. In contrast to how Prox1 maintains lymphatic fate, we also provide a challenging and important analysis of Prox1 function at the earliest stages of lymphatic specification and transdifferentiation from blood vascular to lymphatic vasculature. For the first time in any published literature, we show that the role of Prox1 in vivo in this fate decision is primarily to negatively regulate blood vascular and hematopoietic cell fate and not to positively regulate a lymphatic specification gene network. Importantly, this suggests that lymphatic fate transition begins by blocking what may be a default blood vascular cell fate. This was not previously shown in vivo.
5. Finally, at a molecular level we have demonstrated that Prox1 regulates chromatin accessibility across the genome. Specifically, mutants show a unique and unexpected chromatin signature, whereby chromatin is opened up at many key lymphatic developmental genes but these genes are not transcribed. This discordance in chromatin state and gene transcription appears to be consistent with ectopic activity of early blood and blood vascular transcription factors. This unique finding indicates that Prox1 function negatively regulates blood and blood vascular transcriptional control across diverse enhancers and regulatory elements, and that Prox1 function determines normal chromatin state changes to regulate cell fate. We believe that we did not do a good job of highlighting these important biological observations for the reviewers. Our revision will better emphasise the biological meaning in our data, rather than emphasising the technical aspects of the work.

Secondly, we agree with the reviewers that extracting a new biological finding or understanding from the data will improve the impact of this study. A longstanding question in the field of lymphangiogenesis has been what precise role Notch signalling plays in cell fate decisions and in vessel network growth. The literature is very murky on the role of Notch signalling in the specification of lymphatics. For example:

1. Human in vitro work (Kang, Yoo et al. 2010) showed that increased Notch pathway function repressed expression of key transcription factors Prox1 and CoupTFII and the subsequent induction of LEC fate, but this was not confirmed in vivo.
2. In mice, Murtomaki et al (2013) reported that Notch signalling negatively regulates VEC to LEC transition via suppression of Prox1 expression at the earliest stages of specification of LECs from the cardinal veins (Murtomaki, Uh et al. 2013).
3. This work contradicts the rather definitive observation that endothelial deletion of the core Notch effector Rbpj (Tie2:Cre, Rbpj-f/f) has no effect on the expression of Prox1 in the cardinal veins (Srinivasan, Geng et al. 2010).
4. In zebrafish, it was found that lymphatics don’t form in the absence of Notch signalling (Geudens, Herpers et al. 2010), but in this study we found no evidence of active Notch signalling during LEC specification and sprouting. This was recently explained with the demonstration that it was arterial Notch signalling responsible for the abnormal wiring of zebrafish vessels and loss of lymphatics (Geudens, Coxam et al. 2019). These studies suggest that the role of Notch signalling in zebrafish is not autonomous to the developing veins and lymphatics. Thus, it is currently very unclear if Notch signalling plays a specific role in developing lymphatics or if so, when and how it controls lymphangiogenesis in a cell autonomous manner.

Upon re-evaluating our data, we examined all of the known Notch ligands, receptors and target genes with single cell resolution. To our surprise we found that:

• jag2b expression is a specific marker of the fate-shifted LECs in Zprox1a-/- mutants at 4dpf, switching on when LEC fate is not maintained by Prox1.
• notch1a and notch1b are the key lymphatic expressed receptors for the pathway in zebrafish.
• The main downstream target expressed was her6, which was expressed in a specific manner in vasculature in the maturing LECs.
• Strikingly, there was little to no expression of these key pathway components at specification of LEC fate stages, but rather the Notch pathway is active at later stages when LECs differentiate and grow in the embryo. This prompted us to examine a unique notch1b mutant that we have in the lab. We found that this mutant has clear defects in lymphangiogenesis that impact later stages of development but do not impact early specification.

For our revision, we therefore plan to include one additional large Figure of data. This figure will build of a deeper analysis of Notch signalling in our single cell RNA and ATAC sequencing resource and use our new mutant strain to definitively demonstrate the importance of and timing of Notch signalling in the development of lymphatic vessels. We believe that this will clear up the mystery of when and how Notch controls lymphangiogenesis and will add important new conceptual advance to the paper.

Section 2: Other minor comments

Reviewers’ Comments:

Reviewer #1

An article in 2017 presented abundant expression of fabp11a in zebrafish and suggested its function in brain vessel integrity (PMID: 28443032). In the current manuscript however, the authors did not find fabp11a expression in the head vasculature. Did the authors not detect expression of fabp11a in brain blood vessel endothelial cells at the investigated stages of the zebrafish development? In this case, how would they discuss this seeming contradiction?

We thank the reviewer for pointing out this study. In the paper from Zhang et al. (2017), the authors showed blood vessel expression of fabp11a at earlier stages than we have examined in our images here. In particular, the expression in blood vessels in the head was shown at 1.5, 2 and 3 dpf. We have examined our transgenic line only at 5 dpf. At 5 dpf we do see expression in the trunk veins, which is consistent with the Zhang paper, but we have not looked at the cranial blood vessels at early stages.

In our revision, we will image earlier stage brain blood vessels using our new transgenic line to address this issue and provide additional confidence in our findings.

Minor comments:

In Figure 1a, authors show LEC sprouts in the trunk region at 40 hpf. At 3 dpf however, these LECs sprouts are not shown, but parachordal LECs only. Do these LEC sprouts disappear by 3 dpf? Cartoons on later timepoints suggest that LEC sprouts shown at 40 hpf remain in their location and make connection with parachordal LECs, but the panel in its current form is misleading.

We thank the reviewer for this feedback. We will correct this figure to better indicate these key stages and we will include a full reference at this point of the article to our previous review article Hogan and Schulte-Merker (2017) in which we describe this process in detail and in full (Hogan and Schulte-Merker 2017).

Although I appreciate that the authors were consistent with the colour coding in the graphs, some combinations should be revised. Although the light blue/dark blue colour combination works well in other places, in Figure 4a, it is hard to distinguish those colours. Use of a higher contrast colour combination would be better.

We will correct this by using high-contrast colour combinations as requested.

In Figure 1b, similar colours are used for different purposes. Orange in the upper panel shows 40 hpf cluster, while a very similar colour is used for the VEC_preLEC cluster in the lower panels. Although I recognize the overlay between these clusters, a different colour coding would be more accurate. Maybe, clusters from the upper panel (Stage) should be show individually, just like genes in panel c, to help the reader identifying those clusters at different timepoints.

We will correct this by selecting different colours for VEC_preLEC cluster and cells collected at the 40hpf time point.

Reviewer #2:

Specific Comments:

In general, the authors need to be more precise and cautious in interpreting the RNA velocity analyses. For instance, in Fig 1b, there are two potential regions which could reflect VEC to LEC transition (the one which is connected to LEC sub-cluster and the other which is located in between LEC and VEC/preLEC sub-clusters.) Which trajectory are the authors referring to? In addition, in Fig 3c, the authors claim that RNA velocity analyses showed that the cells within the mutant cluster, however, since cells located within the edge of the clusters tend to have similar trajectory (for instance, cells in the right edge within the LEC_S1 sub-cluster and those in the top left edge within the LEC_S2 sub-cluster), it is difficult to assess whether the trajectory the authors indicated in the mutant sub-cluster is biologically meaningful and relevant. Finally, in Fig 7a, further analyses are needed to support the authors claim which is solely based on RNA velocity analyses.

We thank the reviewer for this feedback and will ensure the size of arrows on our Velocity analysis are increased, to facilitate interpretation of the data. Further to this we will include a second trajectory analysis (Street, Risso et al. 2018) in Extended data figures 1, 2 and 7 that we expect will validate our observations made in Figures 1b, 3c and 7a respectively.

In Fig. 1b, it is not clear whether arterial and venous ECs were excluded from the analyses, if so, the authors need to state how these cell types were identified and excluded. In addition, it would be helpful if the authors show the actual number of cells in each sub-cluster, so the readers could estimate the prevalence of each sub-cluster.

We agree that this information can be more explicitly described, and will include the number of cells per cluster in the legend of Figure 1b, and all single cell RNA-seq UMAPs that define sub-populations. Furthermore, we will include an extra column in Extended Data Table 1a detailing the number of cells per cluster, expand Extended Figure 1 to describe step-wise sub setting of data. We will do this for all 3 single cell datasets. This information will also be written into the Results and Methods.

In Fig 2a, the authors claim that the level of gene expression is different between head and trunk region using cropped fluorescence microscopy images. It would be more convincing if the authors show both head and trunk regions in a single image.

We will address this by using images taken of the entire fish including both head and trunk.

In Fig. 1c, could the authors include an UMAP image showing the expression level of prox1b? It would be helpful for the readers to compare the expressivity of prox1b over time.

We will amend Figure 1c by replacing UMAP images of hexa with prox1b (prox3).

In Fig. 1d, the authors need to explain why the expression of LEC markers diminish at 5dpf.

We thank the reviewer for pointing this out. The 4 dpf single cell RNA-seq libraries are larger than the other libraries included in our developmental time course. While the normalisation (Stuart, Butler et al. 2019) and integration (He, Brazovskaja et al. 2020) approaches have partially corrected this, we believe the higher expression at 4 dpf can be attributed to library size rather than biology. In our revision we will include an analysis that applies down-sampling to larger libraries, that we believe will reduce the contribution of library size to gene expression patterns reported in the developmental time-course.

In Fig. 3a, it would be helpful if the authors show arterial ECs as well, so the readers could assess the characteristics of mutant clusters in a more general context.

We thank the reviewer for this feedback. This information is detailed in Extended Data Figure 2a, which shows UMAPS for all cells in the Zprox1a-/- mutant scRNA-seq dataset. We will expand this figure and include a separate panel with additional UMAP images and dot plots of all endothelial cell types including AEC (arterial endothelial cells), and believe that this will allow readers to better appreciate how different sub-populations of ECs relate to each other.

In the current Figure 3 we focus exclusively on evaluating the LEC, VEC and mutant sub-populations, allowing the reader to hone in on our key points.

In Fig. 3a and 3b, the authors state that Zprox1a null cells generate a peculiar VEC cluster (mutant cluster). Does prox1a influence the transcriptomic profile of VECs as well?

We thank the reviewer for this important question. We will expand the Extended Data Table 2 to include differential expression analyses between Zprox1a-/- mutant and WT AEC (arterial endothelial cells), and VEC (venous endothelial cells). We will include a dot plot in Extended Data Figure 2 that includes cluster specific markers of the mutant cluster with Zprox1a-/- mutant and WT AEC and VEC phenotypes. This demonstrates that the changes in Prox1 mutants are restricted to the cells that normally express Prox1 (i.e. LECs).

It is not clear how the normalization was done in Fig. 3d.

We will include this information in the Results section text more clearly upon revision.

In Fig. 3f, the number of the genes do not match with the extended data table 2b (1034 vs 1107, and 294 vs 326).

We thank the reviewer for picking up these errors. Figure 3f includes all genes that are considered differentially expressed (Wilcoxin Rank Sum adjusted p value

In Fig. 3i and 3k, the authors show the quantification of cdh5/kdrl intensity within the thoracic duct. It would be helpful if the authors could correlate the location of the area used for quantification (whether the quantification represents LEC cluster or mutant cluster).

We thank the reviewer for this suggestion and will add a clear box displaying where measurements were made. We will also amend the text for clarity.

Can the authors specify the unique characteristics of mutant clusters such as the presence of specific markers?

We thank the reviewer for this suggestion, and will amend the text for clarity. We will include a dot plot of top cluster specific markers for all clusters (including the mutant specific cluster) in Extended Data Figure 2.

In Fig. 4g, how prevalent is prox1a/b binding sites and what is the P value?

This is a great question from the reviewer. The Prox1-motif has been problematic but we have now developed robust approaches to identify predicted Prox1-motifs in our snATAC identified peaks. We have now performed a Prox1 motif analysis and will update Figure 4g to include these results. We will include a quantitative comparison of the frequency of Prox1 motifs in LEC, VEC and AEC specific peaks identified in our ATAC analyses.

In Fig. 5a and 5b, the authors assume that the mutant cluster in scRNA-seq data and the mutant cluster in snATAC-seq data are the same population. Is there any validation done?

We thank the reviewer for pointing this out. We will clarify in text that we believe that these are the same cell population for two reasons:

1. They are the only populations in both the scRNA-seq and snATAC-seq data composed almost entirely of Zprox1a-/- mutant cells.
2. Furthermore, all other endothelial cell phenotypes (eg. AEC, VEC, LEC, muLEC, Endocardium) are accounted for in both datasets. At the transcriptional level (in our scRNA-seq) the mutant specific cluster co-expresses LEC and VEC markers, suggesting it is a hybrid cell type that sits between LEC and VEC phenotypes. However, at key lymphatic genes chromatin accessibility and gene expression (comparing snATAC and scRNA-seq) become discordant in the mutant specific clusters, which gives us confidence that we are observing a fate shift due to loss of Prox1 in this specific type of cell. This also suggests that Prox1 is required for concordant chromatin accessibility and gene expression.

In Fig. 5c, figure legend and the extended data table 4a did not match. In Fig 5c, the figure legend says the cut off was set by Wilcoxon Rank Sum, FDRWe thank the reviewer for picking up these errors. As in Figure 3f, Figure 5c includes peaks that are considered differentially accessible (Wilcoxin Rank Sum FDR In Fig. 7d and 7e, it is not clear how the clustering was performed. Based on the image shown in the Fig. 7d/e, three sub-clusters do not seem to clearly separate from one another. It would be helpful if the authors clearly state what was the criteria used for the clustering.

We thank the reviewer for this suggestion. The reason that the clusters sit close together is because these cell types are not yet differentiated from each other. This can be appreciated by looking at clustering of all endothelial cells in 7a. In response to this comment we will no longer show subsetted and re-clustered data in 7d (we will move this to Extended Data), instead will display 7d and 7e using the same UMAP used in 7a with other endothelial cells (AEC, VEC, Endocardium) coloured light grey. We will also expand our description of clustering in the Results and Methods.

Overall, the dot plots should be replaced with the violin plots to better reflect potential heterogeneity within sub-clusters.

We agree that for key points, violin plots could be helpful. We will include violin plots in Extended Data Figures for key data points that include the following: Figures 1c, 3b and 7e. This will ensure that readers have a clear appreciation for heterogeneity within sub-clusters for all key markers that define phenotype in each dataset.

Other comments from reviewers:

Reviewer 1:

Significance:

The manuscript uses state of the art approaches to characterize Prox1-dependent transcriptional and chromatin accessibility changes that define LEC fate and lymphatic sprouting in zebrafish models.

The key role of Prox1 in LEC differentiation and maintenance of lymphatic cell fate and lymphatic development is well known based on previous findings. Strength of the current manuscript is the massive dataset generated, which opens the opportunity to identify downstream players of Prox1 in regulating lymphatic fate and expansion. The authors, however, did not utilize this opportunity for elucidating novel conceptual findings about lymphatic endothelial fate, development or function.

The presented results will be of interest for experts in vascular biology, lymphatic biology, developmental biology and genetics. The generated data may be further used in studies investigating the function of the hit genes highlighted in this manuscript in lymphatic vessels.

Reviewer 2:

Grimm and colleagues analysed developmental lymphangiogenesis in zebrafish embryos using single cell transcriptomics. They identified a number of novel targets of Prox1a, the master regulator for the LEC fate. In addition, the authors have identified a novel mutant-specific sub-cluster in Zprox1a mutant embryos, reiterating the importance of prox1a in the specification and differentiation of LECs.

Significance:

While the work presented in this manuscript could be an interesting resource for the researchers in the field, it does not provide significant conceptual advances in the field. Moreover, there are some technical issues that needs to be resolved prior to the publication of the manuscript.

We thank the reviewers for their positive response and feedback.

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

Grimm and colleagues analyzed developmental lymphangiogenesis in zebrafish embryos using single cell transcriptomics. They identified a number of novel targets of Prox1a, the master regulator for the LEC fate. In addition, the authors have identified a novel mutant specific subclusters in Zprox1a mutant embryos, reiterating the importance of prox1a in the specification and differentiation of LECs.

Specific Comments:

1. In general, the authors need to be more precise and cautious in interpreting the RNA velocity analyses. For instance, in Fig 1b, there are two potential regions which could reflect VEC to LEC transition (the one which is connected to LEC subcluster and the other which is located in between LEC and VEC/preLEC subclsuters.) Which trajectory are the authors referring to? In addition, in Fig 3c, the authors claim that RNA velocity analyses showed that the cells within the mutant cluster, however, since cells located within the edge of the clusters tend to have similar trajectory (for instance, cells in the right edge within the LEC_S1 subcluster and those in the top left edge within the LEC_S2 subcluster), it is difficult to assess whether the trajectory the authors indicated in the mutant subcluster is biologically meaningful and relevant. Finally, in Fig 7a, further analyses are needed to support the authors claim which is solely based on RNA velocity analyses.
2. In Fig. 1b, it is not clear whether arterial and venous ECs were excluded from the analyses, if so, the authors need to state how these cell types were identified and excluded. In addition, it would be helpful if the authors show the actual number of cells in each subcluster, so the readers could estimate the prevalence of each subcluster.
3. In Fig 2a, the authors claims that the level of gene expression is different between head and trunk region using cropped fluorescence microscopy images. It would be more convincing if the authors show both head and trunk regions in a single image.
4. In Fig. 1c, could the authors include an UMAP image showing the expression level of prox1b? It would be helpful for the readers to compare the expressivity of prox1b over time.
5. In Fig. 1d, the authors need to explain why the expression of LEC markers diminish at 5dpf.
6. In Fig. 3a, it would be helpful if the authors show arterial ECs as well, so the readers could assess the characteristics of mutant clusters in a more general context.
7. In Fig. 3a and 3b, the authors state that Zprox1a null cells generate a peculiar VEC cluster (mutant cluster). Does prox1a influence the transcriptomic profile of VECs as well?
8. It is not clear how the normalization was done in Fig. 3d.
9. In Fig. 3f, the number of the genes do not match with the extended data table 2b (1034 vs1107, and 294 vs 326).
10. In Fig. 3i and 3k, the authors show the quantification of cdh5/kdrl intensity within the thoracic duct. It would be helpful if the authors could correlate the location of the area used for quantification (whether the quantification represents LEC cluster or mutant cluster).
11. Can the authors specify the unique characteristics of mutant clusters such as the presence of specific markers?
12. In Fig. 4g, how prevalent is prox1a/b binding sites and what is the P value?
13. In Fig. 5a and 5b, the authors assume that the mutant cluster in scRNA-seq data and the mutant cluster in snATAC-seq data are the same population. Is there any validation done?
14. In Fig. 5c, figure legend and the extended data table 4a did not match. In Fig 5c, the figure legend says the cut off was set by Wilcoxon Rank Sum, FDR<0.05. However, in the extended data table 4a, different cut off was used. Similarly, figure legend for Fig. 5e needs to be revised as well.
15. In Fig. 7d and 7e, it is not clear how the clustering was performed. Based on the image shown in the Fig. 7d/e, three subclusters do not seem to clearly separate from one another. It would be helpful if the authors clearly state what was the criteria used for the clustering.
16. Overall, the dot plots should be replaced with the violin plots to better reflect potential heterogeneity within subclusters.

Significance

While the work presented in this manuscript could be an interesting resource for the researchers in the field, it does not provide significant conceptual advances in the field. Moreover, there are some technical issues that needs to be resolved prior to the publication of 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

Summary:

The authors provide a comprehensive transcriptomic and chromatin accessibility atlas of embryonic lymphangiogenesis in fish using state-of-the-art sc-RNAseq and single cell ATAC sequencing approaches. Furthermore, they present data to prove that Prox1 is a key factor of maintaining LEC identity and promoting lymphatic vascular fate and lymphatic sprouting. Using novel reporter models, they further analyzed the spatial expression pattern of numerous hit proteins in the presence or absence of Prox1 genes.

The manuscript is well written, clear and reproducible based on the given information.

Major comments:

The conclusions of the manuscript are convincing. The robust data generated is inherently valuable and is of great interest of the field. However, my impression is that the authors did not utilize the power of their studies. The main message - Prox1 is a key regulator in promoting and maintaining lymphatic cell fate - is well accepted and has been intensively studied. Therefore, the main findings presented in the current manuscript are not conceptually-advancing.

Additional studies focusing on the function of some of the identified hit genes, such as cdh6, slc7a7, fabp11a in lymphatics - either in fish or in vitro - would significantly improve the novelty of the article. Zebrafish is an ideal experimental model that enable a relatively easy and quick way to address these questions. However, considering the time and expense of those experiments, in vitro studies would be also well appreciated instead of fish.

An article in 2017 presented abundant expression of fabp11a in zebrafish and suggested its function in brain vessel integrity (PMID: 28443032). In the current manuscript however, the authors did not find fabp11a expression in the head vasculature. Did the authors not detect expression of fabp11a in brain blood vessel endothelial cells at the investigated stages of the zebrafish development? In this case, how would they discuss this seeming contradiction?

Minor comments:

In Figure 1a, authors show LEC sprouts in the trunk region at 40 hpf. At 3 dpf however, these LECs sprouts are not shown, but parachordial LECs only. Do these LEC sprouts disappear by 3 dpf? Cartoons on later timepoints suggest that LEC sprouts shown at 40 hpf remain in their location and make connection with parachordial LECs, but the panel in its current form is misleading.

Although I appreciate that the authors were consistent with the color coding in the graphs, some combinations should be revised. Although the light blue/dark blue color combination works well in other places, in Figure 4a, it is hard to distinguish those colors. Use of a higher contrast color combination would be better.

In Figure 1b, similar colors are used for different purposes. Orange in the upper panel shows 40 hpf cluster, while a very similar color is used for the VEC_preLEC cluster in the lower panels. Although I recognize the overlay between these clusters, a different color coding would be more accurate. Maybe, clusters from the upper panel (Stage) should be show individually, just like genes in panel c, to help the reader identifying those clusters at different timepoints.

Referees cross-commenting

I agree with the excellent technical and statistical comments of Rev. 2. Overall, we are in agreement regarding the strength of the datasets as a resource for the field, but with limited conceptual novelty.

Significance

The manuscript uses state of the art approaches to characterize Prox1-dependent transcriptional and chromatin accessibility changes that define LEC fate and lymphatic sprouting in zebrafish models.

The key role of Prox1 in LEC differentiation and maintenance of lymphatic cell fate and lymphatic development is well known based on previous findings. Strength of the current manuscript is the massive dataset generated, which opens the opportunity to identify downstream players of Prox1 in regulating lymphatic fate and expansion. The authors, however, did not utilize this opportunity for elucidating novel conceptual findings about lymphatic endothelial fate, development or function.

The presented results will be of interest for experts in vascular biology, lymphatic biology, developmental biology and genetics. The generated data may be further used in studies investigating the function of the hit genes highlighted in this manuscript in lymphatic vessels.

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 (Evidence, reproducibility and clarity (Required)):

Summary: GlmS, the glucosamine-6-phosphate synthetase in E. coli and related bacteria, is essential, required for synthesis of both peptidoglycan and LPS. It is regulated at various levels, including positive regulation of GlmS translation by the Hfq-binding sRNA GlmZ. GlmZ activation of translation is regulated, indirectly, by the levels of GlcN6P, the product of GlmS. The components of the sensing and regulatory cascade have previously been defined, via genetics, biochemical and molecular biology studies. GlmZ is cleaved by Rnase E, becoming inactive, when GlcN6P levels are high, dependent upon the binding of GlmZ to RapZ. RapZ has been found to directly sense GlcN6P levels; another regulatory RNA, GlmY, also binds RapZ in the absence of GlcN6P, protecting GlmZ from RapZ-mediated processing. The authors of this manuscript performed cryoEM to study the structure of two important complexes in this sensing cascade, RapZ/GlmZ and RapZ/GlmZ/RNase E-NTD, with the aim of clarifying how the RNA binding protein RapZ causes the cleavage of sRNA GlmZ by RNaseE. Some of the predictions for critical residues in the RapZ/GlmZ binary complex structure were investigated by mutagenesis RapZ to define essential resiudes for GlmZ cleavage; the results are consistent with the structure.

Major comments:

• Are the key conclusions convincing? 1) Given that this is basically a structural paper, the major questions would be whether the cryoEM reconstructions are accurate (appear to be consistent with general expectations) and whether there is clear evidence to support the physiological relevance of the structure. The tests of function are of two sorts: a) Effect of RapZ mutants in Fig. 3b-d. These tests show loss of RapZ function with various alleles, likely consistent with model (but as noted below, very difficult for the reader to identify on the structures in 3a). The implication is that these will interfere with GlmZ binding. Possibly direct tests of a couple of these mutants for GlmZ binding (or pull down of GlmZ from in vivo expressed protein) would further support the model. I note that the text says T248A was unaffected in cleavage, but seems to be much reduced in Fig. 3b, even if fusion activity is good.

Our reply. We have made further tests of the mutations for GlmZ binding. Using electrophoretic mobility shift assays, we observe reduced GlmZ binding affinities for RapZ mutants K170A, H190A, C247A, T248A (figure below). We also tested the activity of RapZ variant with 4 substitutions at the proposed RapZ/NTD interface (right lanes in figure below).

We followed the recommendation of the reviewer and performed co-purification experiments (“pull-down”) using StrepTactin affinity chromatography and Strep-tagged RapZ variants as baits. Eluates were assessed for RapZ protein content and co-eluting GlmZ and processed GlmZ* sRNAs using Northern blotting. These new results, which have been incorporated in Fig. S7c, show that all tested RapZ variants except for the wild-type protein are not capable to pull-down GlmZ or GlmZ* in cell extracts. This includes the RapZ-T248A variant, which as noted by the referee is nonetheless still capable to decrease full-length GlmZ to some extent, albeit processed GlmZ* is hardly detectable (Fig. 3b, lanes 23, 24). To address this issue further, we purified the RapZ-T248A variant and some additional variants for comparison and performed EMSA. Globally, the EMSAs confirm the co-purification experiments, i.e. they demonstrate strongly reduced GlmZ binding activity for most tested RapZ variants, but also show that the RapZ-T248 variant kept some residual binding activity. This may explain the weak signal for processed GlmZ in the Northern blot (Fig. 3b) as processed GlmZ* likely binds to RapZ for stabilization. Similar effects were previously seen for the RapZquad and the RapZ 1-279 variants in Durica-Mitic et al. 2020 (Fig. 5). Accordingly, we also changed our wording concerning the RapZ-T248A variant in the text. We have not incorporated the EMSA figure into the updated manuscript.

b) The ternary complex was tested primarily by the BACTH assay of some RapZ mutants (Fig. S11), that show a reduced interaction. This is not a particularly convincing assay for a number of reasons: 1) the effects are relatively modest (2x down, in an assay that is probably not very linear with interaction, 2) some with reduced interaction (S239A, T248A) had good activity (at least all those with full interaction seem to be functional); 3) Ternary complex suggests that RapZ mediates this interaction; this could be tested by deleting glmZ (and maybe glmY as well) from this BACTH strain. 4) the authors suggest that there are also important protein-protein interactions, based on some observed interactions, and support this with similarly difficult to interpret BACTH data from a previous paper for Rnase E-RapZ interaction. Here, too, that is not the most compelling data (is this interaction RNA-independent?).

Our reply: Previous work already indicated that formation of the ternary complex involves multiple interactions – direct protein-protein contacts but also indirect interactions mediated by sRNA GlmZ. For instance, in vitro pull-down signals (RapZ = prey; RNase E = bait) become reduced but not abolished when RNA-free protein preparations are used (Durica-Mitic et al., 2020; Fig. 2E). BACTH signals are reduced 2-fold when using RNase E and RapZ variants that are strongly impaired in their RNA-binding capabilities, respectively (Durica-Mitic et al., 2020; Fig. 2C). As the BACTH assays and in vitro pull-down approaches yield similar trends, we suggest that BACTH experiments represent a useful approach to clarify the questions under study.

Point b1: To demonstrate that removal of multiple interactions is required to disrupt the ternary complex, we combined substitutions of residues making contact to the sRNA as well as residues directly contacting RNase E. According to the structure of the ternary complex presented here, residues T161, Y240, N271 and Q273 in RapZ are proposed to contact RNase E directly. Upon substitution of these four latter residues, resulting in the RapZ variant named RapZ-4 subst., the BACTH signal decreases two-fold – similar to what is observed for the RapZ variants that carry Ala substitutions of residues involved in sRNA-binding, such as H190 or R253. Importantly, when the latter two substitutions are introduced into the RapZ-4 subst. variant – either alone or in combination, the BACTH signal is reduced to almost back-ground levels. These results are in agreement with the features of the ternary complex proposed here and also with data obtained previously: They show that protein-protein and protein-RNA contacts must be concomitantly removed to disrupt the complex completely. We integrated the latter data as Fig. S7a in the revised manuscript and discuss the data at the appropriate positions in the text.

Point b2: In our opinion, the data reporting regulatory activity of the individual RapZ variants (Fig. 3 b-d) correlate well with the BACTH data (Fig. S7a): RapZ variants carrying substitutions of residues I175 and N236 retain regulatory activity and concomitantly a high RNase E interaction potential indistinguishable from the wild-type is observed. In contrast, RapZ variants carrying substitutions affecting sRNA-binding, i.e. H190A, C247A, C247S, T248D, G249W, R253A loose activity completely and concomitantly show a 2-fold decrease in the BACTH signal. The remaining BACTH signal is explained by the remaining (protein-protein) contacts as discussed above (point b1). Therefore, these variants are likely uncapable to present GlmZ in a correct manner to RNase E even though interaction is retained to some degree.

Only the RapZ mutants with exchanges H171A, S239A and T248A do not follow either of these two scenarios: albeit they exhibit reduced interaction with RNase E according to BACTH, they retain the ability to regulate the chromosomal glmS’-lacZ fusion, at least when produced from a plasmid (Fig. 3d). However, inspection of the GlmZ Northern blot signals (Fig. 3b) reveals that full-length GlmZ is decreased as expected, but that processed GlmZ* becomes either not visible or is much reduced when compared to wild-type RapZ. This explains by a reduced sRNA binding affinity, as pointed out above (point 1a), which also provides a rationale for the decreased BACTH signal.

Point b3: We agree that deletion of glmZ in the BACTH strain would be an ideal approach to dissect the contributions of protein-protein and sRNA-protein mediated interactions for formation of the ternary complex in vivo. Unfortunately, construction of the strain is not straight-forward. In our hands, the BACTH reporter strain BTH101 is not amenable to chromosomal manipulations by using engineered recombination tools such as the phage lambda-derived Red system. This may be explained by regulatory elements used by the l Red system that depend on cAMP, which cannot be synthesized in this strain.

__Point b4: __We have addressed this query in the response to point b1.

• Should the authors qualify some of their claims as preliminary or speculative, or remove them altogether? Possibly the importance of RNAse E-RapZ direct interaction, without further proof that this actually is needed for function.

__Our reply: __We partially addressed this issue already in our response to point b1. Additionally, we also tested activity of the RapZ-4 subst variant that lacks the residues making direct contact to RNase E in our structure (Fig. 3b-d, last two lanes/columns). The results that are now described in the last paragraph of the results section show that this variant retains regulatory activity. Interestingly, the level of processed GlmZ* is strongly reduced in this case, similar to what is observed with the RapZ-S239A and RapZ-T248A variants discussed above. Therefore, these direct protein-protein contacts might have a role for GlmZ* decay in a manner that remains to be addressed.

• 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. As noted above, further tests of RapZ mutants for RNA binding would be useful; if this has been done previously, needs to be presented.

Our reply.

This has been addressed in the response above.

Are there Rnase E residues that would be predicted by the model to be critical for the RapZ or GlmZ interaction but are not otherwise needed for activity? Would these disrupt either the BACTH results or activity in vivo?

Our reply.

Please see response to this point above.

• 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. Yes, they are. They are generally extrapolations from what is already in the paper or in previous studies by these groups.
• Are the data and the methods presented in such a way that they can be reproduced? Yes.
• Are the experiments adequately replicated and statistical analysis adequate? Yes.

Minor comments: - Specific experimental issues that are easily addressable. None noted. - Are prior studies referenced appropriately? Yes, they are. However, the paper could more clearly outline what is already known at the level of interactions of the molecules under study here.

Our reply. We have changed the text to better introduce information from previous studies: interprotomer contacts, properties of the isolated RapZ domains, conclusions from the truncation analyses, requirements for interaction for RNase E and for sRNA-binding, stabilization of processed GlmZ through RapZ binding (Göpel et al., 2013; Gonzalez et al 2017; Durica-Mitic and Görke, 2019; Durica-Mitic et al., 2020).

• Are the text and figures clear and accurate?
• In a number of places, the text and figure order/numbers are not correct. See Fig. S1 (p. 4), S2 (legends vs. figure panels).

Our reply. We have corrected these in the revised text.

Better labeling in many figures is needed. Clarify what is shown in Fig. S2d, and make the labels readable. Label the particle types in S3. Use schematics more (as in Fig. 4 and S8) to make it easier for reader to follow structure (for Fig. 2, for instance). It is very difficult to discern RapZ tetramer here. Fig. 3a, it is very difficult to see the residue numbers on the structures. Clearly identify the fructokinase-like domains. Label lanes in Fig. 3b, c, d. Indicate active site for RNase E. in Fig. 4, in schematic at least.

Our reply. We have also corrected these in the revised text.

• Do you have suggestions that would help the authors improve the presentation of their data and conclusions? Overall, clarify and highlight better how the structures here fit with what is already known about important sequences/regions of RapZ, GlmZ, and Rnase E, maybe color-coding parts of GlmZ shown to be important for RapZ recognition, etc.

Our reply. We have added a sequence alignment for RapZ in the supplementary materials section, indicating important residues (Fig. S12).

Page 12, the second last row. Text after 'In this model...' can be simplified or removed because it is just a hypothesis.

Our reply. We have simplified the text.

Our reply:

We believe that the discussion section should also give room for novel ideas and hypotheses. Therefore, we wish to keep the paragraph.

Reviewer #1 (Significance (Required)):

• Describe the nature and significance of the advance (e.g. conceptual, technical, clinical) for the field. Rnase E is a major essential endonuclease in bacteria such as E. coli. How accessory proteins lead to its recognition and cleavage of regulatory RNAs such as GlmZ is not well understood at the structural level, and these structures provide important insight into that process. In addition, the GlmZ/RapZ regulatory circuit plays an important role in bacterial growth and pathogenesis, and understanding it at this level of detail will certainly open up possibilities for targeting this process in the future.

• Place the work in the context of the existing literature (provide references, where appropriate). The components that go into the current structures have been studied previously, with publications on RapZ structure, analysis of critical regions within the GlmZ RNA, and demonstration of the domain of Rnase E involved in interactions with RapZ (Durica-Mitic et al, 2020; Khan et al, 2020, Gonzalez et al, 2017, among others), exactly how these fit together has not been known. Other RNA binding proteins that affect degradation have been reported, but are not fully understood, and ways in which the ribonuclease binds complex RNAs is not fully understood either.

• State what audience might be interested in and influenced by the reported findings. This work should be of broad interested to the field of RNA-based regulation and RNA degradation, with particular interest for those working on these processes in bacteria.

• Define your field of expertise with a few keywords to help the authors contextualize your point of view. Indicate if there are any parts of the paper that you do not have sufficient expertise to evaluate. Our expertise is in RNA-based regulation and microbial genetics; we are not able to critically evaluate the cryoEM analysis itself.

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

Summary:

Islam et al present their characterization of the E. coli RapZ-GlmZ-RNase E ternary complex in this manuscript under review. In E. coli, the RNA binding protein RapZ facilitates cleavage of GlmZ sRNA by RNase E when intracellular concentrations of GlcN-6P are high; when GlcN-6P levels are low RapZ is titrated by GlmY sRNA and GlmZ sRNA promotes an increase in the translation and stability of the mRNA encoding GlcN-6P synthase, GlmS. Via Cryo-EM, the authors of this manuscript solve the structure of the binary RapZ:GlmZ (Fig. 2) and ternary RapZ:GlmZ:RNase Y (Fig. 4) complexes. Based on the apparent RapZ-sRNA binding sites in the solved structure of the binary complex, the authors make substitutions in residues suspected to be involved in RNA binding and measure the impact of these substitutions on cleavage of GlmZ and GlmZ-mediated activation of GlmS expression (Fig. 3). The authors find that some of the residues predicted to be involved in RNA binding based on their structural studies are also important for the cleavage of GlmZ, presumably by RNase E. Finally, the authors show via bacterial two-hybrid assays that some residues of RapZ necessary for GlmZ cleavage are also important for its interaction with RNase E (Fig. S11). I would suggest that the authors measure co-immunoprecipitation of GlmZ with tagged-RapZ with or without substitutions in the proposed RNA binding residues to resolve this issue. Alternatively, EMSAs could be performed.

Our reply. Please see the response above to reviewer 1. We have included results from EMSAs with selected RapZ mutants and for multiple mutations in the BACTH analysis.

Major comments:

Overall, the structural studies our impressive and provide considerable insight into the recognition of substrates by RapZ and RNase E. Given the dearth of solved structures of RNAs with their cognate RNA binding proteins, these results are very significant.

A limitation in this work is the lack of experiments directly testing whether or not the residues of RapZ that appear to be important for its interaction with the GlmZ sRNA in the authors' Cryo-EM structures actually have a significant role in RNA binding. In lieu of measuring GlmZ binding by RapZ, the authors measure GlmZ cleavage in strains expressing RapZ or particular variants harboring substitutions in residues that appear to play a role in sRNA binding (Fig. 3b); however, it is impossible for the authors to determine whether impairment of GlmZ cleavage by RNase E in their assays is due to lack of GlmZ binding to RapZ, extraordinarily tight binding of GlmZ to RapZ, changes in the orientation of GlmZ bound to RapZ, or conformational changes in RapZ that lead to disruption of direct RapZ-RNase E contacts. The lack of this empirical data supporting their structural studies becomes more salient as the authors attempt to test whether RapZ binding of GlmZ is important for its interaction with RNase E via a bacterial two-hybrid assay. Since the authors have not directly examined the importance of particular RapZ residues on GlmZ binding, the authors' interpretation of their results from these assays is very speculative.

Our reply: Reviewer 1 raised a similar point to which we replied above. The role of candidate residues in RapZ for binding GlmZ has been addressed by more direct assays (Pull-down/EMSA).

The authors state on page 7 that "the interaction of RapZ:GlmZ with RNase E does not involve conformational rearrangement of either RapZ or GlmZ". However, the arrangement of SLII relative to SLI appears different between the RapZ:GlmZ and RapZ:GlmZ:RNase E structures presented. Additionally, SLII appears entirely bound by RapZ in the binary complex (Fig. 2b), whereas in the structure of the ternary complex, SLII appears less associated with RapZ (Fig. S4b). A supplementary figure showing side-by-side the structure of GlmZ bound to RapZ solved in the presence or absence of RNase E may make clear whether any differences that exist in the conformation of RapZ and GlmZ between the binary and ternary complex structures.

Our reply: In the revised manuscript, we have included a supplementary figure showing side-by-side comparisons of the structures.

Minor comments: Figure S1 legend. Change "inactivate" to "inactive" or "inactivated"

Figure S2 legend. The description for "(d)" is for S2c and the text for "(c)" refers to the image in S2d.

Figure legend S5a and S9a. If resolution in the key is in angstroms, then it should be indicated.

Our reply: We have now corrected the above points in the revised text.

Figure 1. The model appears to indicate that the apo-form of RapZ binds GlmZ and GlmY, whereas the GlcN-6P bound form does not. Moreover, in the discussion, the authors indicate that GlcN-6P interferes with GlmZ binding to RapZ. How does RapZ bind and cleave GlmZ when GlcN-6P levels are high, if GlcN-6P interferes with GlmZ binding? It would be useful for the authors to address this conundrum in their discussion.

Our reply. We thank the reviewer for pointing out this paradox. Our unpublished work indicates that RapZ may have phosphatase activity for GlcN6P, and we added a comment to this in the discussion section.

Fig. S3B and C. While panels in Fig. S3B and C seemed well aligned, numbering of lanes would provide additional clarity.

We will provide lane numbers, accordingly.

Many bacterial species including Bacillus subtilis, Streptococcus pyogenes, and Clostridium botulinum have RapZ homologs that bear a tyrosine instead of a histidine residue at the position corresponding to H190 in E. coli RapZ. Would you expect this change to reduce GlmZ binding by RapZ or lead to change in RNA specificity based on your structural data? This may be useful to discuss in the manuscript.

We believe that the is more behind this question. Likely, the referee (by inspecting a RapZ sequence alignment) realized that almost all residues proposed to be involved in binding GlmZ are also conserved in RapZ homologs in Gram-positive bacteria, unless His190 and His171, which are replaced by tyrosines in some of these species. However, no RNA-binding activity has been reported for the Gram-positive RapZ homologs. If true, the question arises what is making the difference here? In principle, this could be due to the lacking histidine residues, which are replaced by tyrosines in Gram-positive RapZs. Alternatively, we consider that the positively charged residues at the far C-terminus (K270, K281, R282, K283), which were identified previously to be required for sRNA binding (Göpel et al., 2013; Durica-Mitic et al., 2020), and which could not be resolved in the current structures, are additionally required to obtain RNA-binding activity.

Fig. S10. It is confusing to me that the yellow chain in the structure of RNase E is labeled as the DNase I-domain in the apo structure, whereas in the structure with RprA or GlmZ bound, this colored region is labeled as the 5' sensing domain.

We have changed the figure to make it clearer.

On page 12, the authors appear to indicate that their structural studies of the RapZ-GlmZ-RNase E ternary complex could be informative with regards to how KH domain proteins in Gram-positive bacteria could present their substrates to RNase E. First of all, these bacteria lack RNase E and instead encode an evolutionarily distinct endoribonuclease (RNase Y). Secondly, I think that it is overreaching to state that these structural studies will inform us on how KH domain proteins such as KhpA/KhpB, which may or may not have a chaperoning function akin to Hfq in Gram-positive bacteria, present substrates to RNase Y. Regardless, if this statement is to remain, the authors should make clear that is RNase Y and not RNase E that they are referring to.

We have changed the text to make clear that a different RNase is employed in this case.

Reviewer #2 (Significance (Required)):

In my opinion, the significance of this work is in the achievement of high-resolution structures of the complexes of the RNA binding protein RapZ and the endoribonuclease RNase Y with RNA substrate bound. There are very few structures solved of RNA binding proteins or RNases with their cognate substrates. This is likely due to the difficult in obtaining high resolution data for the bound RNA that may have a large degree of flexibility or many alternative conformations. More structures like this are needed to advance our understanding of RNA-protein interactions.

I believe that these findings would not only be of great interest to those that study small regulatory RNAs, such as myself, but also others more generally interested in RNA binding proteins, RNases, or protein-RNA interactions.

Field of expertise: small regulatory RNAs, RNA chaperones, RNases

**Referees cross-commenting**

1. I agree with Reviewer #1 that the results of the bacterial two-hybrid assay would be more informative, if the authors tested the impact of deletion of glmZ on the ability of the wild type and mutant RapZ proteins to interact with RNase Y by this assay.

As both reviewer #1 and I indicated, I think that it would be useful for the authors to directly assess the effect of key substitutions in RapZ on GlmY binding by a more direct measure of interaction, e.g., CoIP or EMSA.

I do think that it would be nice at some point for the authors to actually provide evidence that GlcN6P binds to the site that they predict as reviewer 3 suggested but this may be be beyond the scope of this manuscript and may be better addressed in another manuscript in which the authors solve the structure of RapZ with GlcN6P bound. In the meantime, the authors could limit their speculation.

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

Summary: The biogenesis of the bacterial cell envelope relies on glucosamine-6-phosphate (GlcN6P), which is mediated by GlmZ and the sRNA-binding protein RapZ. GlmZ stimulates translation of the GlcN6P synthetase. When the levels of the GlcN6P are sufficiently high, RapZ will presents GlmZ to the endoribonuclease RNase E for cleavage and thereby silencing synthesis of the GlcN6P synthetase. However, how RapZ recruit RNase E to GlmZ for degradation is still unsolved. This paper reports the cryoEM structure of the binary complex of RapZ: GlmZ and the ternary complex of the RNase E catalytic domain (RNase E-NTD), RapZ and GlmZ. RapZ interacts with SLI and SLII of GlmZ through complementarity in shape and electrostatic charge to the phosphodiester backbone of the sRNA and presents the sRNA by alignning its SSR comprising the cleavage site into the RNase E active center. This paper suggests a general RNase E recognition pathway for complex substrates, which will help to understand the mechanisms that other RNA chaperones such as Hfq might work in an analogous assembly to present base-paired sRNA/mRNA pairs for cleavage. In total, this is an excellent work. I will support the publication of it until these following points are presented.

Major comments: 1. It was mentioned on Page 5 that "Sulphate and malonate ions were previously seen at these positions in crystal structures of apo RapZ" and pn Page 11 that " Interestingly, the phosphate groups of the RNA backbone occupy positions in RapZ that were previously observed to bind sulphate or malonate ions in the crystal structure of apo-RapZ, suggesting that this pocket could be the binding site for a charged metabolite such as GlcN6P". Is there any following experiments to investigate it further? If possible, I suggest the author to confirm that weather RapZ has the binding activity with GlcN6P or not.

Binding of GlcN6P by the RapZ-CTD was demonstrated previously by SPR as well as by metabolomics of metabolites copurifying with RapZ (Khan et al., 2020), although evidence that the “sulphate/malonate binding sites” in RapZ also bind GlcN6P is still lacking. Crystallization of RapZ+GlcN6P is not straight forward as bound GlcN6P is apparently hydrolyzed over time.

"The kinase-like N-terminal domain of RapZ (NTD) makes only a few interactions with the RNA, and the path of the RNA does not encounter the Walker A or B motifs (Figure 2b). It is possible that this domain could act as an allosteric switch, whereby the binding of an as yet unknown ligand triggers quaternary structural changes that affect RapZ functions." Is there any more structural information supporting it? If the domain act as an allosteric switch, is it possible to make some deletion or substitution to test it?

The properties of the separated NTD and CTD of RapZ were assessed in previous work.

Is there any results to compare the binding affinity of GlmY and GlmZ with RapZ?

Affinities were determined previously using complimentary techniques:

Göpel et al., 2013/EMSA: KD GlmY ~ 30 nM; KD GlmZ ~ 75 nM

Gonzalez et al., 2017/biolayer interferometry: ~ 50 nM for both GlmY/GlmZ (full-length)

Minor comments: 1. Page 8, is it "stabilised" or "stabilized", please check it.

We have changed the spelling to “stabilized”.

The legends for Figure S2 c and d are reversed.

This has now been corrected.

It was suggested to show the RNA molecules in Figure S1a.

We have changed the figure to include single-stranded RNA substrate.

Reviewer #3 (Significance (Required)):

This paper suggests a general RNase E recognition pathway for complex substrates, which will help to understand the mechanisms that other RNA chaperones such as Hfq might work in an analogous assembly to present base-paired sRNA/mRNA pairs for cleavage. In total, this is an excellent work.

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 biogenesis of the bacterial cell envelope relies on glucosamine-6-phosphate (GlcN6P), which is mediated by GlmZ and the sRNA-binding protein RapZ. GlmZ stimulates translation of the GlcN6P synthetase. When the levels of the GlcN6P are sufficiently high, RapZ will presents GlmZ to the endoribonuclease RNase E for cleavage and thereby silencing synthesis of the GlcN6P synthetase. However, how RapZ recruit RNase E to GlmZ for degradation is still unsolved. This paper reports the cryoEM structure of the binary complex of RapZ: GlmZ and the ternary complex of the RNase E catalytic domain (RNase E-NTD), RapZ and GlmZ. RapZ interacts with SLI and SLII of GlmZ through complementarity in shape and electrostatic charge to the phosphodiester backbone of the sRNA and presents the sRNA by alignning its SSR comprising the cleavage site into the RNase E active center. This paper suggests a general RNase E recognition pathway for complex substrates, which will help to understand the mechanisms that other RNA chaperones such as Hfq might work in an analogous assembly to present base-paired sRNA/mRNA pairs for cleavage. In total, this is an excellent work. I will support the publication of it until these following points are presented.

Major comments:

1. It was mentioned on Page 5 that "Sulphate and malonate ions were previously seen at these positions in crystal structures of apo RapZ" and pn Page 11 that " Interestingly, the phosphate groups of the RNA backbone occupy positions in RapZ that were previously observed to bind sulphate or malonate ions in the crystal structure of apo-RapZ, suggesting that this pocket could be the binding site for a charged metabolite such as GlcN6P". Is there any following experiments to investigate it further? If possible, I suggest the author to confirm that weather RapZ has the binding activity with GlcN6P or not.
2. "The kinase-like N-terminal domain of RapZ (NTD) makes only a few interactions with the RNA, and the path of the RNA does not encounter the Walker A or B motifs (Figure 2b). It is possible that this domain could act as an allosteric switch, whereby the binding of an as yet unknown ligand triggers quaternary structural changes that affect RapZ functions." Is there any more structural information supporting it? If the domain act as an allosteric switch, is it possible to make some deletion or substitution to test it?
3. Is there any results to compare the binding affinity of GlmY and GlmZ with RapZ?

Minor comments:

1. Page 8, is it "stabilised" or "stabilized", please check it.
2. The legends for Figure S2 c and d are reversed.
3. It was suggested to show the RNA molecules in Figure S1a.

Significance

This paper suggests a general RNase E recognition pathway for complex substrates, which will help to understand the mechanisms that other RNA chaperones such as Hfq might work in an analogous assembly to present base-paired sRNA/mRNA pairs for cleavage. In total, this is an excellent work.

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:

Islam et al present their characterization of the E. coli RapZ-GlmZ-RNase E ternary complex in this manuscript under review. In E. coli, the RNA binding protein RapZ facilitates cleavage of GlmZ sRNA by RNase E when intracellular concentrations of GlcN-6P are high; when GlcN-6P levels are low RapZ is titrated by GlmY sRNA and GlmZ sRNA promotes an increase in the translation and stability of the mRNA encoding GlcN-6P synthase, GlmS. Via Cryo-EM, the authors of this manuscript solve the structure of the binary RapZ:GlmZ (Fig. 2) and ternary RapZ:GlmZ:RNase Y (Fig. 4) complexes. Based on the apparent RapZ-sRNA binding sites in the solved structure of the binary complex, the authors make substitutions in residues suspected to be involved in RNA binding and measure the impact of these substitutions on cleavage of GlmZ and GlmZ-mediated activation of GlmS expression (Fig. 3). The authors find that some of the residues predicted to be involved in RNA binding based on their structural studies are also important for the cleavage of GlmZ, presumably by RNase E. Finally, the authors show via bacterial two-hybrid assays that some residues of RapZ necessary for GlmZ cleavage are also important for its interaction with RNase E (Fig. S11). I would suggest that the authors measure co-immunoprecipitation of GlmZ with tagged-RapZ with or without substitutions in the proposed RNA binding residues to resolve this issue. Alternatively, EMSAs could be performed.

Major comments:

Overall, the structural studies our impressive and provide considerable insight into the recognition of substrates by RapZ and RNase E. Given the dearth of solved structures of RNAs with their cognate RNA binding proteins, these results are very significant.

A limitation in this work is the lack of experiments directly testing whether or not the residues of RapZ that appear to be important for its interaction with the GlmZ sRNA in the authors' Cryo-EM structures actually have a significant role in RNA binding. In lieu of measuring GlmZ binding by RapZ, the authors measure GlmZ cleavage in strains expressing RapZ or particular variants harboring substitutions in residues that appear to play a role in sRNA binding (Fig. 3b); however, it is impossible for the authors to determine whether impairment of GlmZ cleavage by RNase E in their assays is due to lack of GlmZ binding to RapZ, extraordinarily tight binding of GlmZ to RapZ, changes in the orientation of GlmZ bound to RapZ, or conformational changes in RapZ that lead to disruption of direct RapZ-RNase E contacts. The lack of this empirical data supporting their structural studies becomes more salient as the authors attempt to test whether RapZ binding of GlmZ is important for its interaction with RNase E via a bacterial two-hybrid assay. Since the authors have not directly examined the importance of particular RapZ residues on GlmZ binding, the authors' interpretation of their results from these assays is very speculative.

The authors state on page 7 that "the interaction of RapZ:GlmZ with RNase E does not involve conformational rearrangement of either RapZ or GlmZ". However, the arrangement of SLII relative to SLI appears different between the RapZ:GlmZ and RapZ:GlmZ:RNase E structures presented. Additionally, SLII appears entirely bound by RapZ in the binary complex (Fig. 2b), whereas in the structure of the ternary complex, SLII appears less associated with RapZ (Fig. S4b). A supplementary figure showing side-by-side the structure of GlmZ bound to RapZ solved in the presence or absence of RNase E may make clear whether any differences that exist in the conformation of RapZ and GlmZ between the binary and ternary complex structures.

Minor comments:

Figure S1 legend. Change "inactivate" to "inactive" or "inactivated"

Figure S2 legend. The description for "(d)" is for S2c and the text for "(c)" refers to the image in S2d.

Figure legend S5a and S9a. If resolution in the key is in angstroms, then it should be indicated.

Figure 1. The model appears to indicate that the apo-form of RapZ binds GlmZ and GlmY, whereas the GlcN-6P bound form does not. Moreover, in the discussion, the authors indicate that GlcN-6P interferes with GlmZ binding to RapZ. How does RapZ bind and cleave GlmZ when GlcN-6P levels are high, if GlcN-6P interferes with GlmZ binding? It would be useful for the authors to address this conundrum in their discussion.

Fig. S3B and C. While panels in Fig. S3B and C seemed well aligned, numbering of lanes would provide additional clarity.

Many bacterial species including Bacillus subtilis, Streptococcus pyogenes, and Clostridium botulinum have RapZ homologs that bear a tyrosine instead of a histidine residue at the position corresponding to H190 in E. coli RapZ. Would you expect this change to reduce GlmZ binding by RapZ or lead to change in RNA specificity based on your structural data? This may be useful to discuss in the manuscript.

Fig. S10. It is confusing to me that the yellow chain in the structure of RNase E is labeled as the DNase I-domain in the apo structure, whereas in the structure with RprA or GlmZ bound, this colored region is labeled as the 5' sensing domain.

On page 12, the authors appear to indicate that their structural studies of the RapZ-GlmZ-RNase E ternary complex could be informative with regards to how KH domain proteins in Gram-positive bacteria could present their substrates to RNase E. First of all, these bacteria lack RNase E and instead encode an evolutionarily distinct endoribonuclease (RNase Y). Secondly, I think that it is overreaching to state that these structural studies will inform us on how KH domain proteins such as KhpA/KhpB, which may or may not have a chaperoning function akin to Hfq in Gram-positive bacteria, present substrates to RNase Y. Regardless, if this statement is to remain, the authors should make clear that is RNase Y and not RNase E that they are referring to.

Significance

In my opinion, the significance of this work is in the achievement of high-resolution structures of the complexes of the RNA binding protein RapZ and the endoribonuclease RNase Y with RNA substrate bound. There are very few structures solved of RNA binding proteins or RNases with their cognate substrates. This is likely due to the difficult in obtaining high resolution data for the bound RNA that may have a large degree of flexibility or many alternative conformations. More structures like this are needed to advance our understanding of RNA-protein interactions.

I believe that these findings would not only be of great interest to those that study small regulatory RNAs, such as myself, but also others more generally interested in RNA binding proteins, RNases, or protein-RNA interactions.

Field of expertise: small regulatory RNAs, RNA chaperones, RNases

Referees cross-commenting

1. I agree with Reviewer #1 that the results of the bacterial two-hybrid assay would be more informative, if the authors tested the impact of deletion of glmZ on the ability of the wild type and mutant RapZ proteins to interact with RNase Y by this assay.
2. As both reviewer #1 and I indicated, I think that it would be useful for the authors to directly assess the effect of key substitutions in RapZ on GlmY binding by a more direct measure of interaction, e.g., CoIP or EMSA.
3. I do think that it would be nice at some point for the authors to actually provide evidence that GlcN6P binds to the site that they predict as reviewer 3 suggested but this may be be beyond the scope of this manuscript and may be better addressed in another manuscript in which the authors solve the structure of RapZ with GlcN6P bound. In the meantime, the authors could limit their speculation.

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:

GlmS, the glucosamine-6-phosphate synthetase in E. coli and related bacteria, is essential, required for synthesis of both peptidoglycan and LPS. It is regulated at various levels, including positive regulation of GlmS translation by the Hfq-binding sRNA GlmZ. GlmZ activation of translation is regulated, indirectly, by the levels of GlcN6P, the product of GlmS. The components of the sensing and regulatory cascade have previously been defined, via genetics, biochemical and molecular biology studies. GlmZ is cleaved by Rnase E, becoming inactive, when GlcN6P levels are high, dependent upon the binding of GlmZ to RapZ. RapZ has been found to directly sense GlcN6P levels; another regulatory RNA, GlmY, also binds RapZ in the absence of GlcN6P, protecting GlmZ from RapZ-mediated processing. The authors of this manuscript performed cryoEM to study the structure of two important complexes in this sensing cascade, RapZ/GlmZ and RapZ/GlmZ/RNase E-NTD, with the aim of clarifying how the RNA binding protein RapZ causes the cleavage of sRNA GlmZ by RNaseE. Some of the predictions for critical residues in the RapZ/GlmZ binary complex structure were investigated by mutagenesis RapZ to define essential resiudes for GlmZ cleavage; the results are consistent with the structure.

Major comments:

• Are the key conclusions convincing?
  1. Given that this is basically a structural paper, the major questions would be whether the cryoEM reconstructions are accurate (appear to be consistent with general expectations) and whether there is clear evidence to support the physiological relevance of the structure. The tests of function are of two sorts:
     • a) Effect of RapZ mutants in Fig. 3b-d. These tests show loss of RapZ function with various alleles, likely consistent with model (but as noted below, very difficult for the reader to identify on the structures in 3a). The implication is that these will interfere with GlmZ binding. Possibly direct tests of a couple of these mutants for GlmZ binding (or pull down of GlmZ from in vivo expressed protein) would further support the model. I note that the text says T248A was unaffected in cleavage, but seems to be much reduced in Fig. 3b, even if fusion activity is good.
     • b) The ternary complex was tested primarily by the BACTH assay of some RapZ mutants (Fig. S11), that show a reduced interaction. This is not a particularly convincing assay for a number of reasons: 1) the effects are relatively modest (2x down, in an assay that is probably not very linear with interaction, 2) some with reduced interaction (S239A, T248A) had good activity (at least all those with full interaction seem to be functional); 3) Ternary complex suggests that RapZ mediates this interaction; this could be tested by deleting glmZ (and maybe glmY as well) from this BACTH strain. 4) the authors suggest that there are also important protein-protein interactions, based on some observed interactions, and support this with similarly difficult to interpret BACTH data from a previous paper for Rnase E-RapZ interaction. Here, too, that is not the most compelling data (is this interaction RNA-independent?).
• Should the authors qualify some of their claims as preliminary or speculative, or remove them altogether? Possibly the importance of RNAse E-RapZ direct interaction, without further proof that this actually is needed for function.
• 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. As noted above, further tests of RapZ mutants for RNA binding would be useful; if this has been done previously, needs to be presented. Are there Rnase E residues that would be predicted by the model to be critical for the RapZ or GlmZ interaction but are not otherwise needed for activity? Would these disrupt either the BACTH results or activity 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. Yes, they are. They are generally extrapolations from what is already in the paper or in previous studies by these groups.
• Are the data and the methods presented in such a way that they can be reproduced? Yes.
• Are the experiments adequately replicated and statistical analysis adequate? Yes.

Minor comments:

• Specific experimental issues that are easily addressable. None noted.
• Are prior studies referenced appropriately? Yes, they are. However, the paper could more clearly outline what is already known at the level of interactions of the molecules under study here.
• Are the text and figures clear and accurate?
  1. In a number of places, the text and figure order/numbers are not correct. See Fig. S1 (p. 4), S2 (legends vs. figure panels).
  2. Better labeling in many figures is needed. Clarify what is shown in Fig. S2d, and make the labels readable. Label the particle types in S3. Use schematics more (as in Fig. 4 and S8) to make it easier for reader to follow structure (for Fig. 2, for instance). It is very difficult to discern RapZ tetramer here. Fig. 3a, it is very difficult to see the residue numbers on the structures. Clearly identify the fructokinase-like domains. Label lanes in Fig. 3b, c, d. Indicate active site for RNase E. in Fig. 4, in schematic at least.
• Do you have suggestions that would help the authors improve the presentation of their data and conclusions? Overall, clarify and highlight better how the structures here fit with what is already known about important sequences/regions of RapZ, GlmZ, and Rnase E, maybe color-coding parts of GlmZ shown to be important for RapZ recognition, etc.<br /> Page 12, the second last row. Text after 'In this model...' can be simplified or removed because it is just a hypothesis.

Significance

• Describe the nature and significance of the advance (e.g. conceptual, technical, clinical) for the field.

Rnase E is a major essential endonuclease in bacteria such as E. coli. How accessory proteins lead to its recognition and cleavage of regulatory RNAs such as GlmZ is not well understood at the structural level, and these structures provide important insight into that process. In addition, the GlmZ/RapZ regulatory circuit plays an important role in bacterial growth and pathogenesis, and understanding it at this level of detail will certainly open up possibilities for targeting this process in the future. <br /> - Place the work in the context of the existing literature (provide references, where appropriate).

The components that go into the current structures have been studied previously, with publications on RapZ structure, analysis of critical regions within the GlmZ RNA, and demonstration of the domain of Rnase E involved in interactions with RapZ (Durica-Mitic et al, 2020; Khan et al, 2020, Gonzalez et al, 2017, among others), exactly how these fit together has not been known. Other RNA binding proteins that affect degradation have been reported, but are not fully understood, and ways in which the ribonuclease binds complex RNAs is not fully understood either.<br /> - State what audience might be interested in and influenced by the reported findings.

This work should be of broad interested to the field of RNA-based regulation and RNA degradation, with particular interest for those working on these processes in bacteria.<br /> - Define your field of expertise with a few keywords to help the authors contextualize your point of view. Indicate if there are any parts of the paper that you do not have sufficient expertise to evaluate.

Our expertise is in RNA-based regulation and microbial genetics; we are not able to critically evaluate the cryoEM analysis itself.

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)):

Spinal cord injury (SCI) is a damage to the spinal cord, that causes temporary or permanent changes in its function. While in mammals the regeneration process are very limited zebrafish are able to repair the spinal cord. Based on the hypothesis, that the vascular response might affect the regeneration capacity, the paper by Ribeiro et al addresses the structure and injury response of the spinal cord vasculature. As the growth of zebrafish larvae and juveniles depends a lot on the individual response to the environment, the authors first established comparable body measurement parameters (other than age) and observed the natural spinal cord vascularization process, starting from 6mm body length of the animals. Using transgenic lines the authors describe the formation and patterning of endothelial cells and pericytes up to 9mm length, when a more developed vascular network was present. They observe the processes of vascular regeneration after a contusion based SCI model at different time points (days post injury (dpi)) and in correlation with glial and axonal regrowth, also observing BSCB barrier integrity, angiogenesis, pericyte recruitment and the dependence on Vegf signaling.

The study is interesting and novel, vascular structures in the zebrafish adult spinal cord have not been reported yet and neither has the vascular response to SCI. Currently the study remains very descriptive, although the authors tried to add functional data, by inhibiting Vegf signaling.

Major points for revision: The authors fail to establish whether there is any relationship between spinal cord regeneration and vessel regeneration. While I do very well understand the challenges and limitations the authors should put more effort into functional analyses.

For example: the authors address EC proliferation as a marker for angiogenesis, but do not analyse whether or how much EC proliferation is required for revascularization and regeneration. Pharmacological inhibition of proliferation should be possible and used. From a vascular point of view it would also be interesting whether there is a differential influence of tip or stalk cell proliferation.

Although we agree that it would be interesting to inhibit EC proliferation to assess its role in spinal cord regeneration, the use of proliferation inhibiting drugs would likely have a widespread effect on the lesioned spinal cord, since many cell types proliferate in response to injury. Therefore, a pharmacological approach would not allow us to dissect the specific role of endothelial proliferation.

The same is true for pericyte recruitment: the role of pericytes for the vascular repair or the spinal cord regeneration is not clear. The authors could use use mutants with impaired pericyte development or e.g. nitroreductase mediated ablation of pericytes.

These experiments have been performed in larvae by Tsata et al. (2021). Although it would be interesting to repeat in adults, we believe that these experiments are beyond the focus of our study.

The statements regarding the role of Vegf are too bold. The problem lies in the limitations of assessing the efficiency of Vegf inhibition. The heatshock promotor has been shown to induce transcription for up to 4 hours, depending on the efficiency of heatshock. There are no data on the stability of dnVegfaa protein. Likewise the pharmacological inhibition could be far from complete. A full inhibition of Vegf signaling is expected to stop vessel growth or angiogenesis. While it is a sign of good practice, that the authors combined a genetic model with a pharmacological one, both leave the same unresolved issue. However if we believe a very limites requirement of Vegf-signaling, it would be interesting to look for other signaling pathways, like cxcl, IL, or FGF to regulates regenerative angiogenesis.

We agree that our data does not allow us assess the level of inhibition of the Vegf pathway. Since we are unable to confirm this at the moment, we will be excluding the Vegf inhibition data and make this a descriptive study.

Minor issues

The correlation with spinal cord repair could be stated more clearly throughout the manuscript. For the uninformed reader it is less clear when exactly the spinal cord is functional again.

We will include in Fig. 3 a plot of the swimming capacity in contusion-injured fish until 90 dpi and will explain in the text how the vascular response correlates with the functional recovery.

While I find the model in figure 8 very helpful, it gives 5 to 30 days, for the neuronal regeneration. Maybe a more detailed timeline of EC regeneration and remodeling correlating with neuronal repair would help.

We will update the model in Fig. 8 with a more detailed timeline and a better description of structures important for regeneration (glial bridge, axonal regrowth).

In line with that in figure 4 it is not clear whether the images of different time points are indeed one individual animal at the different time points or representative animals for the stage (also figure 4 lacks panel labels, in my copy I can see A, K and L, but no other letters).

We will detail in the figure legend that the images are of different animals that are representative for each stage.

For understanding the (re)vascularization, the direction of blood flow might be helpful.

We will perform an additional experiment to characterise the direction of blood flow in uninjured fish. For this we will use juvenile fish with a body size of 7-9 mm, in which we expect to be able to perform live imaging. We will use a lighsheet microscope to image circulating cells in the spinal cord blood vessels in fish with labelled thrombocytes (Tg(-6.0itga2b:EGFP); Lin et al., 2005) and endothelial cells (Tg(kdrl:ras-mCherry)). These transgenic lines are already available in our fish facility. Even though the vascular network has not yet reached its mature stage at these body sizes, we expect to have enough intraspinal vessels to describe the blood flow circuit.

Especially for the connection between spinal cord regeneration and vessel regeneration. Does blood flow regulate vessel pruning after 14 dpi?

Although we agree with the reviewer that it would be interesting to understand how blood flow direction is reestablished in repaired vessels and how blood flow levels correlate with vessel remodelling and pruning, this would be difficult to assess in this system. This could be examined using live imaging, but this technique is challenging in adult zebrafish and has only been carried out in more superficial organs than the spinal cord, such as skin (Castranova et al., 2022) and superficial brain structures (Barbosa et al., 2015; Castranova et al., 2021). In addition, SC-injured fish are more sensitive to external conditions and would probably not survive the long-term/repeated anaesthesia required for imaging.

This analysis could be performed in fixed samples, for example using the the Golgi complex position in relation to the endothelial nuclei as a proxy for blood flow direction (Kwon et al., 2016), however: (1) this would require a new transgenic line (Tg(fli1a: B4GALT1-mCherry)) that would take time to import and establish in the lab; (2) the identification of regressing vessels is not straightforward in fixed samples and is usually studied in very well established vascular models, such as the mouse retina and zebrafish ISVs (Franco et al., 2015).

For these reasons, we will not address this question by reviewer 1.

The combined Vegfaa DN and PTK treatment data looks like it could be inhibiting endothelial cell proliferation (Figure7I).However, Supplementary Figure 8B shows endothelial proliferation does not change. Does it mean the number of endothelial cells is same but the volume of endothelial cells decrees?

We will not be addressing the changes in endothelial density in the presence of dn-vegfaa and PTK787, since we will be removing the figures related to Vegf inhibition.

There are also some remaining grammatical errors, for example (but NOT limited to) line 133 to 135.

We will review grammatical errors in the text.

As a personal interest I think evaluating the role of Notch in the SCI model would also be very interesting, especially with regard to the vasculature, however that might be out of the scope of the manuscript.

We agree that Notch signalling may be a player during spinal cord revascularisation. However, mutants for dll4 (the Notch ligand involved in angiogenesis) die between 7-14 dpf and cannot be used for this study. In addition, the use of Notch-inhibiting drugs would likely have pleiotropic effects, since the Notch pathway is also involved in other aspects of spinal cord regeneration, namely in the regulation of regenerative neurogenesis (Dias et al., 2012). To our knowledge, tools that allow the endothelial-specific inhibition of the Notch pathway have not been developed, and therefore we will not be able to address this question.

Reviewer #1 (Significance (Required)):

The study is partially descriptive, but very novel as the aspects of vascularisation in a spinal cord injury model have not been described before. If the major revisions regarding functionality are addressed fully, I would wholeheartedly recommend publication and expect an interest for a broad audience. The presented images and their analyses are of very high quality, and therefore also enhance the impact of the study.

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

The study by Ribeiro et al. investigates the formation of new blood vessels after spinal cord injury in adult zebrafish. The authors initially characterize the extend of spinal cord vascularization during the development of juvenile zebrafish and investigate the association of pericytes with the newly forming vasculature. They then injure the spinal cord and describe the subsequent regeneration of blood vessels. They perform assays to analyze the functionality of the newly forming blood vessels and show that initially blood vessels are leaky. Through EdU labelling the authors show that endothelial cells proliferate. Pericytes similarly increased in numbers. Lastly, the authors inhibited VEGF signaling, which only mildly affected vascular regeneration.

Together, this manuscript describes the re-vascularization of the regenerating spinal cord in adult zebrafish and addresses how blood vessels mature during this process through pericyte recruitment and decrease in leakiness. The manuscript provides some interesting initial insights into spinal cord vascularization, but is mainly descriptive and unfortunately remains superficial in this regard, as specified below:

1. The authors only refer to "blood vessels" without specifying the type of blood vessels they observe (are these veins, arteries, capillaries)? A wealth of markers and transgenic zebrafish lines are available to better characterize spinal cord vessels. This is not necessary in case the authors solely refer to "blood vessels" as they do, but it greatly limits the insights into spinal cord vascularization. For instance, Wild et al. (2017) showed that new vessels apparently sprout from veins in the spinal cord. Is this also true during regeneration?

We will perform RNA in situ hybridisation using probes for arterial and venous markers. We will assay the expression of arterial markers (dll4, dlc, flt1 and efnb2a) and venous markers (flt4 and ephb4a) in uninjured spinal cord (to characterise vessel identity in homeostasis) and in 3 and 7 dpi spinal cords (to investigate the identify of angiogenic vessels during regeneration).

1. The authors state that their characterization revealed a "stereotypic organization of blood vessels". However, the organization does not appear to be stereotypic (as I understand this term as looking the same in each fish) at all. Can the authors compare e.g. 3 or 5 wildtype fish and extract features that all fish share and those that differ between fish? This would greatly enhance our understanding of the vascular variability within the wildtype population.

We will provide an additional figure comparing the spinal cord vasculature in different fish.

1. The authors show an interesting metameric organization of the vasculature with regions of high vascularization interspersed with sparsely vascularized areas. Are there any morphological landmarks that would precipitate these differences?

We will acquire light sheet images of adult spinal cords without removing the vertebrae. This will allow us to determine if the metameric organisation is correlated with the vertebral distribution.

Can the authors check whether they induce a lesion in a highly or poorly vascularized area? This might greatly influence the degree of re-vascularization.

We always perform the spinal cord injury in the region between neural arches (Dietrich et al., 2021). Once we determine how the vasculature is organised in relation to the vertebrae, we will be able to determine if the lesions are performed in a region of high or low vascularisation.

1. The same superficial characterization unfortunately also applies to the cell population the authors refer to as "pericytes". Traditionally, pericytes are characterized as being associated with capillaries and sharing a basement membrane with the endothelium. Is this the case here?

We will further characterise the association between Tg(pdgfrß:citrine)-positive cells and blood vessels using an anti-laminin antibody (#L9393, Sigma) to label the basement membrane. Preliminary results recently acquired indicate that Tg(pdgfrß:citrine)-positive perivascular cells and endothelial cells are both enveloped by the basement membrane, supporting the identity of Tg(pdgfrß:citrine)-positive cells as pericytes. Moreover, pericytes are generally described as solitary mural cells associated with small diameter blood vessels (the type of distribution we observe for Tg(pdgfrß:citrine)-positive cells), whereas vascular smooth muscle cells (vSMCs) form concentric layers around larger blood vessels (a distribution we do not detect with this transgene) (Hellström et al., 1999). For these reasons we believe that this transgene is labelling pericytes. We will explain more clearly in the text how the morphology, localisation and density of Tg(pdgfrß:citrine)-positive cells suggests these cells are pericytes.

In addition, pdgfrb is hardly specific for pericytes, as it also labels a multitude of other cell types (refer to e.g. Tsata et al. (2021)).

The different cell types labelled by the pdgfrb reporter line used in the Tsata et al., 2021 paper were identified not by the use of different cell markers, but by their localisation: perivascular cells (the same cell type that we also detect), myoseptal cells (which we would not expect to detect, since we are only analysing the spinal cord tissue and not the adjacent muscle) and floor plate cells (a reporter distribution that the authors show is lost after 3 dpf and is not present in the adult spinal cord). Moreover, the Tsata et al., 2021 paper also includes a supplementary figure (S1, panel N) showing a restricted perivascular pdgfrb:GFP distribution in the wholemount adult spinal cord, in agreement with our characterisation. By their morphology and density, these perivascular cells are likely pericytes, as argued above.

It is also not clear why the transgenic pdgfrb line the authors use only labels cells next to blood vessels. Tsata et al. show a much broader labelling. The authors need to validate their transgenic line using in situ hybridization showing where pdgfrb is being expressed endogenously and how this overlaps with the fluorescent protein expression of the pdgfrb transgenic line.

We will perform ISH for pdgfrb to confirm if the Tg(pdgfrß:citrine) reporter reproduces the endogenous expression in the uninjured spinal cord and at 3 and 7dpi. The 3-7 dpi period is approximately equivalent to the 1-2 days post-lesion in larvae and, if the non-perivascular pdgfrb:GFP cells observed in the larval spinal cord are present in the adult, we expect to detect them by ISH during this phase of regeneration.

There are also several transgenic lines available that allow for the distinction between smooth muscle cells and pericytes (e.g. Shih,..., Lawson, Development 2021 and Whitesell,..., Childs, Plos ONE 2014). As for the vasculature, this more detailed characterization is not necessary in case the authors refer to the cells as "cells labelled by the pdgfrb transgene and reside next to endothelial cells". However, this would not be reflective of the level of detail currently present in the field.

As we explain above, the morphology and density of the pdgfrb:Citrine-positive cells suggests that these cells are pericytes and not smooth muscle cells (SMCs). To confirm this we will compare the expression of pdgfrb with markers of SMCs (i.e, 𝛼-smooth muscle actin and desmin) using immunohistochemistry and/or ISH.

The reviewer also suggests the characterisation of pericyte subtypes using the lines described by Shih et al., 2021. Although this would be interesting, we do not consider it is essential for our study. It would be very demanding to import the reporter lines and it is not certain that these subtypes are present in the spinal cord.

1. The authors state that "New blood vessels rapidly attracted pericytes, formed through proliferation and possibly migration of existing pericytes". This statement is not supported by the data, as the authors do not perform lineage tracing of pre-existing pericytes. The authors need to specifically label existing pericytes and then follow whether these pre-labelled cells can be found on newly forming blood vessels. Tsata et al. provide some evidence for this in zebrafish larvae, but they also conclude that pdgfrb expressing tenocytes contribute to new mural cells.

We will reformulate the sentence to clarify that we detect pericyte proliferation, but pdgfrb-lineage tracing would be needed to provide evidence that existing pericytes contribute to the generation of mural cells associated to new blood vessels. However, we will not perform the lineage tracing experiment for the revision, as we are unable to currently import this line.

1. The findings that new blood vessel growth only marginally depended on VEGFA signaling is striking. However, it might also point towards an inefficient inhibition of VEGFA signaling. In particular, other publications, for instance Cattin et al. 2015 have shown that inhibiting VEGFA signaling prevents new blood vessel growth during peripheral nerve regeneration in mouse. It will therefore be important that the authors demonstrate that their approach leads to successful inhibition of VEGFA signaling. VEGFAB mutants appear to be homozygous viable and important for spinal cord vascularization (Matsuoka et al., 2017). In addition, heterozygous VEGFAA mutants already have some vascular phenotypes, but are also viable. Can the authors combine these mutants with their inhibitor treatments to achieve a greater reduction in VEGFA signaling?

Since we are unable to confirm the level of inhibition of the Vegf pathway and we are unable to import the suggested lines at the moment, we will be excluding the Vegf inhibition data.

Reviewer #2 (Significance (Required)):

Together, this publication is the first to describe to some extend the regenerating vasculature after spinal cord injury in adult zebrafish. However, both the vascular and regeneration fields are much more advanced than what the authors cover. Both blood vessels and perivascular cells can be characterized in much more detail, as outlined above. Also, studies on nerve regeneration and its dependence on the vasculature, e.g. during peripheral nerve regeneration in mouse have been carried out with a wealth of functional data available. Therefore, the impact of the present study in its current form will be limited. I am an expert on zebrafish blood vessel development.

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

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

Ribeiro et al. described vascular development in the spinal cord from larval to adult stages in zebrafish, and found the dependence of vessel length on body-size. Then, the authors depicted the vascular regeneration process after spinal cord injury (SCI), which includes initial vascularization, angiogenesis, pericyte recruitment, and blood-spinal cord barrier establishment. Although the molecules or signaling pathways that drive the re-vascularization remain unidentified, this study illustrates the cellular processes of spinal cord vascular development and regeneration from the descriptive level, which may facilitate further understandings of mechanisms underlying vascular regeneration in the spinal cord.

Major comments: - Are the key conclusions convincing? The descriptions of spinal cord vascularization during development and vascular regeneration after SCI are convincing. However, inhibition of Vegfaa and Vegfr2 is nearly ineffective. The author might not conclude that the Vegfr2 signaling plays any role.

Since we are unable to confirm the level of inhibition of the Vegf pathway, we will be excluding the Vegf inhibition data.

Should the authors qualify some of their claims as preliminary or speculative, or remove them altogether? - 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.

Major comments: 1) In Figure 3, the exact injured site on the spinal cord is not clear. Please include a schematic illustration of full spinal cord to show where is the injured site. Are all the injury experiments in this study done at the same site? If not, is there any site difference regarding the regenerative capability.

We will include a scheme of the injury site in the spinal cord in Fig.3. All the injury were performed in the same position and this will be clarified in the methods.

2) Figure 2E showed a segmented pattern of spinal cord vasculature. Is this pattern correlated with the position of vertebra?

We will acquire light sheet images of adult spinal cords without removing the vertebrae. This will allow us to determine if the metameric organisation is correlated with the vertebral distribution.

3) In Figure 3, during vascular regeneration after SCI, the author only showed partial regeneration at 30 dpi. Why not show the stage of complete regeneration? At that stage, how about the behaviors of the regenerated animals?

We will add an additional timepoint (90 dpi) to the characterisation of the revascularisation. Moreover, we will include in Fig. 3 a plot of the swimming capacity in contusion-injured fish until 90 dpi and will explain in the text how the vascular response correlates with the functional recovery.

4) Only EdU data is not sufficient to conclude that new vessels come from proliferation of remaining endothelial cells. For example, these new vessels might come from transdifferentiation of lymphatic vessels, or immune cells, or glial cells, in the meantime proliferate. This could also explain why the inhibition of Vegfr2 signaling is ineffective on new vessel formation. Cre/loxP-mediated lineage tracings need to be performed to exactly identify where these new vessels originate.

We will clarify in the text that while the detection of endothelial proliferation suggests existing endothelial cells contribute to new vessels, we cannot exclude that other cell types also give rise to endothelial cells. However, regarding the transdifferentiation of immune and glial cells into endothelial cells, to our knowledge few examples have been described in the literature and generally associated with cancers or in in vitro conditions (Fernandez Pujol et al., 2000; Li et al., 2011; Soda et al., 2011). For this reason we do not expect this rare process to occur during spinal cord repair.

A cell type that has been associated with transdifferentiation into ECs are lymphatic cells (Das et al., 2022). However, we have analysed the expression of a lymphatic marker (Tg(lyve1b:DsRed)) and were only able to detect very few lyve1b:DsRed-positive cells before or after injury, suggesting that any possible lymphatic contribution would likely be very limited. We plan to include these data in the revised submission.

5) To confirm the Tg(hsp70l:dn-vegfaa) did work in this study, the authors need a positive control. For example, the effects on vasculogenesis or angiogenesis during embryonic development after heat shock. If the transgene works, the vascular development at early stages should be blocked (Marín-Juez et al., 2016).

We will be removing the vegf inhibition data, therefore we will not address this question.

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. The suggested experiments are realistic in terms of time and resources.

• Are the data and the methods presented in such a way that they can be reproduced? In the method, the author should describe how to identify the Tg(hsp70l:dn-vegfaa) in more details, because there is no fluorescence before and after heat shock.

We will be removing the vegf inhibition data, therefore we will not address this question.

Are the experiments adequately replicated and statistical analysis adequate? Yes.

Minor comments: - Specific experimental issues that are easily addressable. In Figure 6, from 30 dpi to 90 dpi, the number of pericytes decreased. Did these pericytes undergo apoptosis from 30 dpi on?

We have not investigated pericyte apoptosis during vessel remodelling. However, this experiment would require the acquisition of long-term samples (between 60 and 90 dpi) and we would prefer not to address this question.

Are prior studies referenced appropriately? Yes.

• Are the text and figures clear and accurate? Please clearly labeled the injured region in Figure 6.

We will identify more clearly the site of the injury in Fig.6.

• Do you have suggestions that would help the authors improve the presentation of their data and conclusions? The number of proliferating ECs at 3 dpi is more than those at 5 dpi (Figure 5G). But the number of total EdU+ cells at 3 dpi is less than those at 5 dpi (Figure 5A-D). These data are consistent with Figure S3, which showed ECs were the leading cell type to enter the lesioned site, then were the axons and glial cells at later stages. Please explain and discuss whether the regeneration of other cell types is dependent on the accomplishment of vascular regeneration.

As the reviewer points out, our data suggest that endothelial cells display an earlier peak of proliferation than spinal cord cells in general and colonise the lesioned tissue before new axons and glial cells. Although these observations could point to a role for ECs in the regeneration of other cell types, we would need to inhibit vascular repair to assess this possibility, which we were unable to do using Vegf inhibition. In our discussion we already mention some possible roles for ECs in stem cell proliferation, neurogenesis and axonal regrowth, but can expand this discussion if necessary.

Reviewer #3 (Significance (Required)):

• Describe the nature and significance of the advance (e.g. conceptual, technical, clinical) for the field.

Although this study has characterized the development and regeneration of spinal cord vasculature in details, the significance of the advance needs to be improved due to lack of mechanisms. Obviously Vegfa is not essential for the vascular regeneration after SCI. It is better for the authors to identify one or two factors required for this process, in addition to identify cell origins of new vessels. With those, the significance of this study will be improved because the cell origins and required factors will provide potential therapeutic targets after SCI.

• Place the work in the context of the existing literature (provide references, where appropriate).

• State what audience might be interested in and influenced by the reported findings. The audience includes people who are interested in vascular development and regeneration, and spinal cord clinicians.

• Define your field of expertise with a few keywords to help the authors contextualize your point of view. Indicate if there are any parts of the paper that you do not have sufficient expertise to evaluate. My field of expertise includes brain vascular regeneration, digestive organ development and regeneration. This study reported spinal cord vascular development and regeneration, which fit my expertise.

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:

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

Ribeiro et al. described vascular development in the spinal cord from larval to adult stages in zebrafish, and found the dependence of vessel length on body-size. Then, the authors depicted the vascular regeneration process after spinal cord injury (SCI), which includes initial vascularization, angiogenesis, pericyte recruitment, and blood-spinal cord barrier establishment. Although the molecules or signaling pathways that drive the re-vascularization remain unidentified, this study illustrates the cellular processes of spinal cord vascular development and regeneration from the descriptive level, which may facilitate further understandings of mechanisms underlying vascular regeneration in the spinal cord.

Major comments:

• Are the key conclusions convincing?

The descriptions of spinal cord vascularization during development and vascular regeneration after SCI are convincing. However, inhibition of Vegfaa and Vegfr2 is nearly ineffective. The author might not conclude that the Vegfr2 signaling plays any role. - Should the authors qualify some of their claims as preliminary or speculative, or remove them altogether? - 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.

Major comments:

1. In Figure 3, the exact injured site on the spinal cord is not clear. Please include a schematic illustration of full spinal cord to show where is the injured site. Are all the injury experiments in this study done at the same site? If not, is there any site difference regarding the regenerative capability.
2. Figure 2E showed a segmented pattern of spinal cord vasculature. Is this pattern correlated with the position of vertebra?
3. In Figure 3, during vascular regeneration after SCI, the author only showed partial regeneration at 30 dpi. Why not show the stage of complete regeneration? At that stage, how about the behaviors of the regenerated animals?
4. Only EdU data is not sufficient to conclude that new vessels come from proliferation of remaining endothelial cells. For example, these new vessels might come from transdifferentiation of lymphatic vessels, or immune cells, or glial cells, in the meantime proliferate. This could also explain why the inhibition of Vegfr2 signaling is ineffective on new vessel formation. Cre/loxP-mediated lineage tracings need to be performed to exactly identify where these new vessels originate.
5. To confirm the Tg(hsp70l:dn-vegfaa) did work in this study, the authors need a positive control. For example, the effects on vasculogenesis or angiogenesis during embryonic development after heat shock. If the transgene works, the vascular development at early stages should be blocked (Marín-Juez et al., 2016).
6. 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.

The suggested experiments are realistic in terms of time and resources. - Are the data and the methods presented in such a way that they can be reproduced?

In the method, the author should describe how to identify the Tg(hsp70l:dn-vegfaa) in more details, because there is no fluorescence before and after heat shock. - Are the experiments adequately replicated and statistical analysis adequate?

Yes.

Minor comments:

• Specific experimental issues that are easily addressable.

In Figure 6, from 30 dpi to 90 dpi, the number of pericytes decreased. Did these pericytes undergo apoptosis from 30 dpi on? - Are prior studies referenced appropriately?

Yes. - Are the text and figures clear and accurate?

Please clearly labeled the injured region in Figure 6. - Do you have suggestions that would help the authors improve the presentation of their data and conclusions?

The number of proliferating ECs at 3 dpi is more than those at 5 dpi (Figure 5G). But the number of total EdU+ cells at 3 dpi is less than those at 5 dpi (Figure 5A-D). These data are consistent with Figure S3, which showed ECs were the leading cell type to enter the lesioned site, then were the axons and glial cells at later stages. Please explain and discuss whether the regeneration of other cell types is dependent on the accomplishment of vascular regeneration.

Significance

• Describe the nature and significance of the advance (e.g. conceptual, technical, clinical) for the field.

Although this study has characterized the development and regeneration of spinal cord vasculature in details, the significance of the advance needs to be improved due to lack of mechanisms. Obviously Vegfa is not essential for the vascular regeneration after SCI. It is better for the authors to identify one or two factors required for this process, in addition to identify cell origins of new vessels. With those, the significance of this study will be improved because the cell origins and required factors will provide potential therapeutic targets after SCI. - Place the work in the context of the existing literature (provide references, where appropriate). - State what audience might be interested in and influenced by the reported findings.

The audience includes people who are interested in vascular development and regeneration, and spinal cord clinicians. - Define your field of expertise with a few keywords to help the authors contextualize your point of view. Indicate if there are any parts of the paper that you do not have sufficient expertise to evaluate.

My field of expertise includes brain vascular regeneration, digestive organ development and regeneration. This study reported spinal cord vascular development and regeneration, which fit my expertise.

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 study by Ribeiro et al. investigates the formation of new blood vessels after spinal cord injury in adult zebrafish. The authors initially characterize the extend of spinal cord vascularization during the development of juvenile zebrafish and investigate the association of pericytes with the newly forming vasculature. They then injure the spinal cord and describe the subsequent regeneration of blood vessels. They perform assays to analyze the functionality of the newly forming blood vessels and show that initially blood vessels are leaky. Through EdU labelling the authors show that endothelial cells proliferate. Pericytes similarly increased in numbers. Lastly, the authors inhibited VEGF signaling, which only mildly affected vascular regeneration.

Together, this manuscript describes the re-vascularization of the regenerating spinal cord in adult zebrafish and addresses how blood vessels mature during this process through pericyte recruitment and decrease in leakiness. The manuscript provides some interesting initial insights into spinal cord vascularization, but is mainly descriptive and unfortunately remains superficial in this regard, as specified below:

1. The authors only refer to "blood vessels" without specifying the type of blood vessels they observe (are these veins, arteries, capillaries)? A wealth of markers and transgenic zebrafish lines are available to better characterize spinal cord vessels. This is not necessary in case the authors solely refer to "blood vessels" as they do, but it greatly limits the insights into spinal cord vascularization. For instance, Wild et al. (2017) showed that new vessels apparently sprout from veins in the spinal cord. Is this also true during regeneration?
2. The authors state that their characterization revealed a "stereotypic organization of blood vessels". However, the organization does not appear to be stereotypic (as I understand this term as looking the same in each fish) at all. Can the authors compare e.g. 3 or 5 wildtype fish and extract features that all fish share and those that differ between fish? This would greatly enhance our understanding of the vascular variability within the wildtype population.
3. The authors show an interesting metameric organization of the vasculature with regions of high vascularization interspersed with sparsely vascularized areas. Are there any morphological landmarks that would precipitate these differences? Can the authors check whether they induce a lesion in a highly or poorly vascularized area? This might greatly influence the degree of re-vascularization.
4. The same superficial characterization unfortunately also applies to the cell population the authors refer to as "pericytes". Traditionally, pericytes are characterized as being associated with capillaries and sharing a basement membrane with the endothelium. Is this the case here? In addition, pdgfrb is hardly specific for pericytes, as it also labels a multitude of other cell types (refer to e.g. Tsata et al. (2021)). It is also not clear why the transgenic pdgfrb line the authors use only labels cells next to blood vessels. Tsata et al. show a much broader labelling. The authors need to validate their transgenic line using in situ hybridization showing where pdgfrb is being expressed endogenously and how this overlaps with the fluorescent protein expression of the pdgfrb transgenic line. There are also several transgenic lines available that allow for the distinction between smooth muscle cells and pericytes (e.g. Shih,..., Lawson, Development 2021 and Whitesell,..., Childs, Plos ONE 2014). As for the vasculature, this more detailed characterization is not necessary in case the authors refer to the cells as "cells labelled by the pdgfrb transgene and reside next to endothelial cells". However, this would not be reflective of the level of detail currently present in the field.
5. The authors state that "New blood vessels rapidly attracted pericytes, formed through proliferation and possibly migration of existing pericytes". This statement is not supported by the data, as the authors do not perform lineage tracing of pre-existing pericytes. The authors need to specifically label existing pericytes and then follow whether these pre-labelled cells can be found on newly forming blood vessels. Tsata et al. provide some evidence for this in zebrafish larvae, but they also conclude that pdgfrb expressing tenocytes contribute to new mural cells.
6. The findings that new blood vessel growth only marginally depended on VEGFA signaling is striking. However, it might also point towards an inefficient inhibition of VEGFA signaling. In particular, other publications, for instance Cattin et al. 2015 have shown that inhibiting VEGFA signaling prevents new blood vessel growth during peripheral nerve regeneration in mouse. It will therefore be important that the authors demonstrate that their approach leads to successful inhibition of VEGFA signaling. VEGFAB mutants appear to be homozygous viable and important for spinal cord vascularization (Matsuoka et al., 2017). In addition, heterozygous VEGFAA mutants already have some vascular phenotypes, but are also viable. Can the authors combine these mutants with their inhibitor treatments to achieve a greater reduction in VEGFA signaling?

Significance

Together, this publication is the first to describe to some extend the regenerating vasculature after spinal cord injury in adult zebrafish. However, both the vascular and regeneration fields are much more advanced than what the authors cover. Both blood vessels and perivascular cells can be characterized in much more detail, as outlined above. Also, studies on nerve regeneration and its dependence on the vasculature, e.g. during peripheral nerve regeneration in mouse have been carried out with a wealth of functional data available. Therefore, the impact of the present study in its current form will be limited. I am an expert on zebrafish blood vessel development.

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

Spinal cord injury (SCI) is a damage to the spinal cord, that causes temporary or permanent changes in its function. While in mammals the regeneration process are very limited zebrafish are able to repair the spinal cord. Based on the hypothesis, that the vascular response might affect the regeneration capacity, the paper by Ribeiro et al addresses the structure and injury response of the spinal cord vasculature. As the growth of zebrafish larvae and juveniles depends a lot on the individual response to the environment, the authors first established comparable body measurement parameters (other than age) and observed the natural spinal cord vascularization process, starting from 6mm body length of the animals. Using transgenic lines the authors describe the formation and patterning of endothelial cells and pericytes up to 9mm length, when a more developed vascular network was present. They observe the processes of vascular regeneration after a contusion based SCI model at different time points (days post injury (dpi)) and in correlation with glial and axonal regrowth, also observing BSCB barrier integrity, angiogenesis, pericyte recruitment and the dependence on Vegf signaling.

The study is interesting and novel, vascular structures in the zebrafish adult spinal cord have not been reported yet and neither has the vascular response to SCI. Currrently the study remains very descriptive, although the authors tried to add functional data, by inhibiting Vegf signaling.

Major points for revision: The authors fail to establish whether there is any relationship between spinal cord regeneration and vessel regeneration. While I do very well understand the challenges and limitations the authors should put more effort into functional analyses.

For example: the authors address EC proliferation as a marker for angiogenesis, but do not analyse whether or how much EC proliferation is required for revascularization and regeneration. Pharmacological inhibition of proliferation should be possible and used. From a vascular point of view it would also be interesting whether there is a differential influence of tip or stalk cell proliferation.

The same is true for pericyte recruitment: the role of pericytes for the vascular repair or the spinal cord regeneration is not clear. The authors could use use mutants with impaired pericyte development or e.g. nitroreductase mediated ablation of pericytes.

The statements regarding the role of Vegf are too bold. The problem lies in the limitations of assessing the efficiency of Vegf inhibition. The heatshock promotor has been shown to induce transcription for up to 4 hours, depending on the efficiency of heatshock. There are no data on the stability of dnVegfaa protein. Likewise the pharmacological inhibition could be far from complete. A full inhibition of Vegf signaling is expected to stop vessel growth or angiogenesis. While it is a sign of good practice, that the authors combined a genetic model with a pharmacological one, both leave the same unresolved issue. However if we believe a very limites requirement of Vegf-signaling, it would be interesting to look for other signaling pathways, like cxcl, IL, or FGF to regulates regenerative angiogenesis.

Minor issues

The correlation with spinal cord repair could be stated more clearly throughout the manuscript. For the uninformed reader it is less clear when exactly the spinal cord is functional again. While I find the model in figure 8 very helpful, it gives 5 to 30 days, for the neuronal regeneration. Maybe a more detailed timeline of EC regeneration and remodeling correlating with neuronal repair would help. In line with that in figure 4 it is not clear whether the images of different time points are indeed one individual animal at the different time points or representative animals for the stage (also figure 4 lacks panel labels, in my copy I can see A, K and L, but no other letters).

For understanding the (re)vascularization, the direction of blood flow might be helpful. Especially for the connection between spinal cord regeneration and vessel regeneration. Does blood flow regulate vessel pruning after 14 dpi?

The combined Vegfaa DN and PTK treatment data looks like it could be inhibiting endothelial cell proliferation (Figure7I).However, Supplementary Figure 8B shows endothelial proliferation does not change. Does it mean the number of endothelial cells is same but the volume of endothelial cells decrees?

There are also some remaining grammatical errors, for example (but NOT limited to) line 133 to 135.

As a personal interest I think evaluating the role of Notch in the SCI model would also be very interesting, especially with regard to the vasculature, however that might be out of the scope of the manuscript.

Significance

The study is partially descriptive, but very novel as the aspects of vascularisation in a spinal cord injury model have not been described before. If the major revisions regarding functionality are addressed fully, I would wholeheartedly recommend publication and expect an interest for a broad audience. The presented images and their analyses are of very high quality, and therefore also enhance the impact 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

Reviewer #1.

Reviewer #1 summary:

In this manuscript by Lu et al., the authors describe some CRISPR screens and protein-protein interaction screens to identify novel regulators of wild-type p53 and mutant p53 function and stability. Besides generating a wealth of data, they discover FBXO42-CCDC6 as positive regulators of the some p53 hot-spot mutants, including R273H mutant p53, but not of all p53 mutants tested and also not of wild-type, indicating selectivity. Furthermore, the found C16orf72(TAPR1) as a negative regulator of p53 stability.

Mechanistically, the authors claim a direct interaction between FBXO42 and CCDC6 and p53, but the importance of these interactions has not been shown. On the other hand the authors suggest that the FBXO42/CCDC6 regulate p53 via destabilization of USP28, but also the mechanism has not been worked out. For C16orf72, they show that it interacts with USP7, but no relevance of this interaction is shown either.

Response: We sincerely thank the reviewer for the constructive and thorough review. We have incorporated most of the suggestions into our planned revision, with our major focus on the molecular mechanistic follow-up.

Reviewer #1, major points.

1. One very important point for me is that the authors do not show the levels of expression of p53 in the p53-mClover stable cell lines. It is known that overexpressed p53 is usualy more stable than endogenous levels of wt-p53. Therefore, I think it is necessary that the authors show the levels of the p53-mClover fusion proteins in the stably transduced cell lines compared to endogenous p53 levels in the parental RPE1 cells and also compared to the endogenous levels of R273H mutant in the PANC-1 cells.

Response: We fully agree that the levels of overexpressed p53s are often more than the endogenous ones, due in part to increased expression and stability. In designing the reporter, we first tried to avoid the stabilisation of p53-GFP due to GFP aggregation by using the monomeric mClover-variant. Further, we titrated the WT and R273H clones (similar to our recent work in PMID: 35439056), to select clones with p53 levels closer to endogenous protein, and exhibiting high dynamic response to Nutlin-3a treatment.

In the revised submission, we will include Western blotting comparing the levels of p53-mClover (WT and R273H) expression to the endogenous p53s in RPE1 (WT) and PANC1 (R273H) cell lines, in the presence or absence of Nutlin-3a.

Also the functionality of the wild-type p53-mClover fusion is questionable, at least not shown. One would expect that the overexpression of a functional wt-p53 in p53-KO cells will affect the survival of the RPE1 cells. In Figure 5A the authors show that depletion of MDM2 or C16ORF72 is toxic for the RPE1 cells in a p53-dependent manner, indicating that elevated levels of p53 cannot be handled by these cells. So, experiment(s) showing that the wt-p53/mClover fusion is functional is needed.

Response: We agree that it will be an important point to benchmark the reporter design. The ectopically expressed WTp53 is often observed to have reduced functionality compared to the endogenous WTp53. The WTp53-reporter line behaves similarly to the RPE1 line (p53-proficient), where both chemical (e.g. Nutlin) or genetic perturbation (e.g. depletion of MDM2/C16orf72) would be toxic in a p53-depedent manner. In line with this data, we have observed that the WTp53-reporter line is able to induce a p53 response as demonstrated by induction of p53-target genes such as p21, which is not observed in p53 null RPE cells, albeit the p21 induction is not as dramatic as in RPE1 cells with endogenous WTp53. Together, these data indicate that our WTp53-reporter is functional albeit with a somewhat reduced activity.

In the revised submission, we will better demonstrate the functionality of the WTp53-mClover fusion by probing WTp53 target (e.g. p21), in the presence and absence of Nutlin. This is also performed as a part of the experiment addressing Point #1 above.

A second important point is that the 'verification' of the hits from the screens is only done in one cancer cell line, PANC-1, with mutant p53. I would have like to see at least one other cell line with another p53 mutant endogenously expressed that is also regulated by FBXO42/CCDC6.

Response: we will include validation of the hits (FBXO42, CCDC6) in other 1-2 tumour lines with confirmed R273H endogenous mutation (e.g. MB-MDA-468, etc).

For many of the p53-mutants, a bimodal expression is observed. In the FBXO42- and CCDC6-depleted cells, the equilibrium shifts towards more negative cells but the levels in the two populations itself don’t change (while for example for USP28 depletion also the right peak shifts further up, Fig S4E). Is there any correlation with the cell cycle and p53 expression? And can the authors exclude that FBXO42 and CCDC6 are involved in cell cycle progression and hereby influence p53 indirectly (by combining PI staining with Clover-p53 for example).

Response: we have indeed observed that the “bimodal” levels in the reporters of several mutants, which are also observed in other studies probing the endogenous p53 level (PMID: 29653964); while the population equilibrium shifts, the location of each peak (as a proxy of the level of p53s) are more stable.

Regarding the relation between p53-level and cell cycle stage, indeed, both the authors in the paper above and we have probed this possibility, but were unable to establish a direct connection.

In the revised submission, we will add flow cytometry analysis of the p53-mClover level, and the cell cycle position using Hoechst 33342 (live-cell permeable DNA staining).

The authors claim that the FBXO42-CCDC6 axis regulates stability specifically some p53-mutants, including R273H-mutant, in a manner involving USP28. But USP28 regulates all forms of p53, not just some mutants version. How can the authors reconcile this apparent contradiction?

Response: we thank the reviewer for this critical observation. From our screen (Supplemental Table 1A), we have indeed noticed a pronounced effects (|Z score| >=3) of FBXO42 on R273H and R248Q stability, and a marginal effect on wild-type p53. Similarly, USP28 had pronounced effects on R273H and R248Q and WTp53.

In the discussion of the paper, we noted that USP28 was shown to regulate p53 levels through distinct mechanisms:

‘USP28 was originally implicated as a protective deubiquitinating enzyme counteracting the proteasomal degradation of p53, TP53BP1, CHCK2, and additional proteins68-71. USP28 regulates wild-type p53 via TP53BP1-dependent and -independent mechanisms. Concordantly, our data shows that USP28 and TP53BP1 are strong positive regulators of wild-type p53. However, while USP28 was also a strong hit in the mutant R273H p53 screen, TP53BP1 was not, indicating that the effects we see upon loss of USP28 on R273H p53 are independent of TP53BP1.’

Together, this indicates that the R273H-mutant is regulated by a FBXO42-CCDC6-USP28 axis while wild-type p53 is regulated mainly via a USP28-TP53BP1 axis. We will attempt to address and discuss it in the revision.

On a similar note, the authors show that FBXO42 and CCDC6 interact with p53, but not USP28. Do FBXO42 and CCDC6 interact with each other and with USP28? And is the interaction with p53 specific for the R273H version? This part of the mechanism is very poorly defined and the Co-IPs are not very convincing or relevant for the proposed model.

Response: This comment will be more extensively addressed in the revision. We have indeed observed the interaction between FBXO42 and CCDC6 (via BioID and APMS); however, we failed to recover USP28 as an interactor of either FBXO42 or CCDC6. The interaction between CCDC6/FBXO42 is not specific to R273H; although we were able to IP endogenous R273H with CCDC6 in PANC-1 line, the WTp53 (as in HEK 293 TRex BioID line) was also picked up in the BioID preys of CCDC6/FBXO42. In addition, we have new data to show that FBXO42 directly interacts with WTp53.

In the revised submission, we will improve the molecular underpinning of the FBXO42-CCDC6-USP28-p53 axis we propose. We will specifically address the following.

(1.1.) Biochemically, further support that CCDC6 and FBXO42 regulate p53 via regulating USP28 stability: We will address this by established biochemical assays, e.g. cycloheximide-chase/MG132 experiment. While USP28 is an established WTp53 regulator, little is known about the mechanism, and the “upstream” regulation of USP28; we will attempt to fill this gap:

(1.2.) And to an unbiased systematic approach, how R273H interactome changes upon the loss of CCDC6 or FBXO42.

We will perform R273H-BioID upon loss of CCDC6 and FBXO42 and USP28.

(1.3.) Furthermore, we will specifically exam the interaction of USP28-p53R273H with or without the genetic perturbation of FBXO42/CCDC6.

Through these efforts, we hope to gain further mechanistic insights into this regulatory axis, but hope that the editors and reviewers will agree that a fully annotated mechanistic understanding is probably beyond the scope of this paper.

Reviewer #1, minor points.

The mechanisms of p53 regulation may vary greatly in different cell lines. Can the authors discuss why they choose to do the screen with different mutants, rather than with different cell lines expressing these same mutant endogenously?

Response: While it is certainly very interesting to assess how WT and mutant p53 is regulated in different cell lines, such an approach is confounded by the ‘genetic make-up’ of the respective tested cell lines. For example, TP53BP1 might be a regulator in one cell line but not in another for the simple reason that the later cell line harbors a TP53BP1 deletion or mutation or expression levels. In addition, while working with endogenous p53 mutations certainly has many advantages, comparing different mutants in different cell lines is again very much confounded by the ‘genetic make-up’ of the respective tested cell lines.

Our focus was slightly different, and we wanted to set out and specifically ask what the difference between p53 hotspot mutations are. Are they all the same or are there differences and importantly, are there differences between mutants and WT p53 and this can only be achieved when working in the same cellular background. In designing the screen, we have thus tried to optimise the inclusion of different hotspot mutants in an isogenic screening system. As such, we first depleted the endogenous WTp53 to minimise its interference and built the current isogenic system in the non-transformed RPE1 (“normal”) line.

However, as discussed above, we agree that the screen results will be validated in more cell lines carrying respective endogenous mutants.

Figure 1: Typo in the legends : Nultin ipv Nutlin

Response: We apologise for the typos. This is addressed in the current submission, along with improved figure legends to improve readability.

Figure 1b,1c : Show basal and Nutlin-3 induced MDM2 levels and in the overexpression cell lines; if WT-p53 is functional, MDM2 levels should be higher in WT-transduced cells compared to control or mt-p53 expressing cells.

Response: In the revised submission, we will include Western blotting probing MDM2 levels (antibody permitting); this is a part of the experiment proposed for Points 1 and 2.

Authors should explain which they name USP7 a negative regulator of p53, since it is supposed to de-ubiquitinate p53?!

Response: The effects of USP7 on WTp53 have indeed been difficult to elucidate (by Prof. Vogelstein PMID: 15118411, and PMID: 15058298, and seemingly opposite by Prof. Gu Wei, PMID: 15053880, and PMID: 11923872). However, consistent with Prof. Vogelstein group, the inhibition of USP7 (either by inhibitor or genetically via CRISPR in our studies), has resulted in elevated p53 level.

Figure 2E: the effect of MG132 on p53 seems to be very minimal on this Western blot; it would need quantification to be convincing...Quality of the blot is also not great.The fact that in control cells the levels of p53 R273H are not affected by MG132 treatment fits with Suppl Figure 2E, indicating that the proteasome has no effect on p53 R273H.

Response: We indeed noticed that while the proteasome pathway is largely implicated in the WTp53 screen, it has much reduced effects on R273H. Interestingly, the treatment of MG 132 also has limited effects using PANC-1 line (with endogenous R273H). We will repeat this experiment and provide quantifications and modify the text accordingly.

Suppl figure 3b, 3c, 3d:

Somehow, I have the feeling that the results from the western blots and the FACS do not match fully, although not all the time-points are shown in the various experiments.

For example, the FACS analysis (3b) suggests that in control-transduced cells after 16hr p53 is still increased. However, that is not clear at all in the Western blot (3c)

Is Suppl Figure 3d the quantification of 3c experiment? If so, in the blot also the 24 hrs should be shown.

The blot shown in Suppl Figure 3c suggests that CCDC6 expression increased upon irradiation. Do the authors agree with that? Would that explain why depletion of CCDC6 has more effect upon irradiation?

Suppl Figure S3E: if I am right, this is essentially the same type of experiment as shown in figure 2e, but analysis of p53-expression by Western blot. In that blot no real effect of MG132 on p53 levels could be seen. But here, in the FACS analysis, MG132 clearly increases the p53-Clover fusion levels; for me again that Western blot and FACS data do not neccesarily match.

Response: We apologise for the confusion. In the revised submission, we will improve the figure legends for better readability. Furthermore, in anticipation to the multiple cell lines involved in the revision, we will also clarify the cell lines in the figure.

With regards to the difference between the flow cytometry and WB data, we have generally observed the flow cytometry bimodal shifting to be more sensitive than the WB, e.g. a 50% shift in population (FACS) is reflected by a 15% reduction in WB (which may be partially explained as WB is a measurement across the cell population and FACS determines the p53-GFP levels of every cell and thus the shift of cells between peaks). Similarly, we noticed flow-cytometry based quantification by antibody staining the endogenous p53 yielded similar sensitivity (PMID: 29653964). As such, we will ensure the validation of hits is performed in two modes. For WB experiment, we will do so in two cell lines carrying the endogenous mutants as suggested by Reviewers #1 and 2.

Figure 3B: In the CCDC6 IP a very small amount of p53 can be found. I don't know how much input lysate compared to amount of lysate for IP is used, but the percentage of p53 found interacting with CCDC6 seems so marginal that is difficult to explain the effect of KO of CCDC6 in PANC1 cells.

And, the authors called it a 'reciprocal IP' (Suppl Figure 4a) after transfection of V5-tagged CCDC6 into PANC1 cells, but it actually is the same type of IP. Did the authors try to IP p53 and blot for CCDC6? That would be a reciprocal IP.

Response: We apologise for the confusion. In the revised submission, we will specify the portion of the lysates used for pre-IP (5% lysate) and IP (1 mg). As for the IP, we will also include the true reciprocal IP (IP p53, and blot for CCDC6).

Figure 3H: how can authors explain that basal levels of USP28 in control and CCDC6-KO cells transfected with control plasmid are more or less the same and not reduced in the CCDC6-KO cells?

Response: We will provide a better blot and quantification for this observation. In the current Fig 3H, the CCDC6-KO lane is slightly overladed as seen by the H3 loading control.

Figure 3I: Essentially the whole blot here is of low quality; especially the FBXO42 blot; is deletion of USP28 increasing FBXO42 protein levels, or is it just the quality of the blot? All in all it seems that FBXO42 is very low expressed in the used cell lines.

Response: We apologise for the confusion. In the revised submission, we will repeat and try to include higher quality WB, with more optimised condition for using the FBXO42 antibody.

FBXO42 messenger level is readily detected using qRT.

Figure 4B: I find it a bit surprising that USP7 is also found in the synthetic viability screen, since it has been shown that USP7 has many more essential targets and KO of p53 only partially rescues the development of USP7-KO mouse embryo's.

Response: We thank the reviewer for this critical observation. While the double p53-USP7 knockout line is viable, we acknowledge that it is amongst the top scored hits due to the large differential viabilities between WT and p53-null lines. In the revised submission, we will further clarify the screen analysis and the associated interpretation.

Figure 5: the authors nowhere show the efficacy of the guides targeting c16orf72. A Western blot showing the expression and the reduction upon expressing the guide-RNAs is essential.

Response: We thank the Reviewer for this suggestion. The efficacy of each guide has been verified using ICE (at the genomic level), and in the revised submission, we will include this critical information as part of the Figure S2F.

Figure 5E: First, here probably parental RPE1 cells have been used, but that is not stated. Second, the authors state 'only a slight increase in p53 levels upon siHUWE1'; I would say none compared to scrambled.

I know HUWE1 is a very huge protein, but the blot of HUWE1 is not convincing. I seem to be able to conclude that siMDM2 and siUSP7 reduces HUWE1 levels?

Response: We apologise for the confusion. In the revised submission, we will be specific of the cell line information on the figure, to improve the readability.

We agree with the reviewers that assessment of large protein by WB is often difficult but given that this band almost completely disappears upon HUWE1 knock-down, strongly argues that we are indeed assessing the endogenous HUWE1. We also agree that it is an interesting observation that the levels of HUWE1 seem to be slightly reduced upon knock-down of MDM2 and USP7. We will repeat this experiments and provide quantitative data for HUWE1 and p53. Of note, in the screen, HUWE1 also scored as a negative regulator of wt-p53 and did not quite reach statistical significance for the p53 mutants.

Regarding the relationship between C16orf72 and HUWE1, a newly published work (PMID: 35776542) seems to suggest that siHUWE1 has resulted in an increased C16orf72 level (termed HAPSTR1 in the paper), while siC16orf72 seemed to have no effect on HUWE1 level, although the stability of such a large protein by WB is often difficult to conclude.

Figure 5F, in relation to figure 5D. Here the author overexpress both c16orf72 and USP7, and find an interaction. The implication of that is not clear. If they want to make point of this interaction, they should have looked at endogenous proteins.

Response: We acknowledge the many concerns associated with coIP with ectopically, and especially overexpressed proteins in large quantity. In the revised submission, we will attempt to perform endogenous-based IP experiment (antibody permitting).

It is worrying that USP7 apparently was not one of the hits in the Mass-spec experiment of which results are shown in Figure 5D. Also in that experiment c16orf72 was overexpressed, and USP7 is very highly expressed in essentially all cell lines, so do the authors have an explanation?

Response: We indeed acknowledge this discrepancy. In the revised submission, we will attempt the coIP/IP using endogenous proteins (antibody permitting, or at least using endogenous target for one of the two partners). We also acknowledge that the limitation associated with the APMS for the detection of interactors.

Suppl. figure 5D is missing

Response: We apologise for the confusion. The Figure S5D was inconveniently placed at the top of the figure panel due to space limitation. In the revised submission, we will address this as a part of the overall readability improvement.

Reviewer #1, Significance.

The topic of the paper is of high interest given the relevance of p53 and its gain-of-function mutants in oncology, and the screens are well executed and clearly presented. In terms of novelty, FBXO42 has been linked to p53-degradation before, and c16orf72 was recently shown to be able to destabilize p53. However, the link between CCDC6 and p53 is novel and of interest, since they are both substrates of USP7 and are both regulators of the cell cycle.

We think the manuscript has potential to add something to the field, but would benefit greatly from a better understanding of the molecular underpinnings of their newly described mechanisms, as well as the conditions in which the mechanism is active.

Therefore, it might be advisable to shorten the manuscript, and go more in-depth in finding the mechanisms of regulation.

Response: We sincerely thank the reviewer for all the constructive critiques. We will incorporate them in to our revision.

Reviewer #2.

Reviewer #2 summary:

The paper describes several genome-wide CRISPR screens designed to identify regulators of p53 stability. The authors use a system in which p53 levels are marked by mClover expression, using RFP expression to normalise for gene expression changes.

Reviewer #2, major points.

1. The bimodal distribution of p53 expression levels in some reporter cell lines (G245S, R248Q, R248W and R273H) hampers the implementation of a robust readout and makes correct interpretation of the results challenging. While it is possible that the bimodal distribution indicates dynamic changes in p53 levels within one population, it also seems possible that a subclone of these cells have acquired additional alterations affecting p53 stability, and that the authors are screening a mixed population of two intrinsically different cell populations. This would make it difficult to interpret the results of the screen in these cell lines and may be a challenge when trying to identify something that has not already been highlighted on depmap.

Response: We thank the reviewer for this critical observation. We strongly believe that this bimodal distribution is actually an inherent property of the p53 mutants in these cells for the following reasons: (1) The observation of the similar bimodal appearance in cell lines harbouring corresponding endogenous mutant p53s (PMID: 29653964) suggest that these two populations are of biological significance. (2) We have established 5-10 clonal lines each from the G245S, R248Q, R248W and R273H p53 reporter line and all of them exhibit a bimodal distribution, making it very unlikely that these populations are all through stochastic outgrowth of sub-populations with spontaneous mutations/alterations. (3) The bimodal distribution is stable over several months to years in culture. If it were a spontaneous mutations giving rise to a clone with higher mutant p53 levels, we would likely expect that over time this clone takes over the population. (4) We observed that such a pool of bimodal cells could be “synchronised” (e.g. by Nutlin, or MDM2 knockout) to one population, and later return to and repopulate the other (e.g. Nutlin washoff, Figure 1B). (5) When we sort out a single cells from the upper or the lower peak, expand them, we obtain again populations of cells with the same bimodal distribution, indicating that this is a dynamic process. Thus, we believe that these two populations were rather intrinsic, such that a cell in the population may assume both states.

We also acknowledge the difficulties of screening using a bimodal population; however, we took advantage of these “bimodal” mutants and using FACS assessed the state of a single cell in relation to a genetic perturbation. Each guide has an equal chance of entering a cell that belongs to one of the two populations. If a gene knock-out really affects p53 levels, the cells with the respective guides enrich in one and deplete in the other population and the analysis comparing the guide abundances from these two peaks ensures the experiment are being perfectly internally controlled.

While many of the top scored hits from the resulting screens are known regulators, it is critical that we validate our hits in an independent system, such as the cell lines harbouring endogenous p53 mutations, echoed by both Reviewers #1 and 2.

The coverage of the sgRNA library (200x) is rather low for a negative selection screen, where a coverage of 500x would be more desirable. The FDR threshold is also rather lenient, a more stringent FDR threshold would seem more appropriate and shorten the list of potential hits.

Response: We thank the reviewer for this constructive suggestion. A higher coverage, along with a more stringent FDR, will ensure an even stronger confidence for the remaining individual hits. The present reporter-based enrichment screen and the synthetical viability drop-out screen used four guides per gene, and with 200x coverage for each guide.

In determining the coverage, we tried to reference recent successful screenings and apply earlier titration result for the 200x coverage (e.g. PMID: 26627737, PMID: 33465779, and reviewed in Nat Rev Methods Primers 2, 8 (2022). https://doi.org/10.1038/s43586-021-00093-4). While the threshold of FDR was often arbitrary, we fully agree that a more stringent FDR, which results in shortened hits list, may further boost the confidence of the hits, though also at the cost of losing potential hits due to collateral effects (e.g. guide efficiency).

We agree with this reviewer that a higher FDR, esp. at the hits that result in p53 stabilization, would make sense as any gene whose loss causes cellular or genotoxic stress, would likely lead at least in part to p53 stabilization. In the revised submission, we will adjust the FDR accordingly.

Although the study is focused on the regulation of p53 stability, there are no experiments to show that any of the manipulations alter the ubiquitination or degradation (half-life) of p53. The rescue of expression by proteasome inhibition is very modest (Figure 2E), suggesting the loss of expression may not be a reflection of degradation. A role for endogenous FBXO42 and C16orf72 in regulating the ubiquitination and half-life of endogenous p53 should be confirmed

Response: We thank the reviewer for this suggestion. In the revised submission, we will monitor the ubiquitination status and also degradation (cycloheximide-chase) experiments for R273H cells, with or without the genetic alteration of CCDC6/FBXO42/C16orf72.

Many p53 mutants are used for the initial screens, but very little validation is carried out to show that the apparent differences in factors regulating their stability persists in cells naturally expressing these mutants. For example, FBXO42 is identified as a protein required to maintain the stability of R273H, 248W and R248Q, but not R175H, G245S and R337H. While the authors show an association of CCDC6 and p53 in PANC1 cells (expressing 273H), it would be important to show a panel of R273H, 248W and R248Q expressing tumor cells and the response of p53 to FBXO42 and CCDC6 depletion, compared to similar experiments in a panel of R175H, G245S and R337H expressing tumor cells. Again, it would be important to show that any changes in protein levels are due to changes in protein stability.

Response: We thank the reviewer for this suggestion. In the revised submission, we will include validations in more cell lines carrying endogenous mutant p53s, with a focus on the R273H mutant. We will also try to involve a line with an endogenous p53 mutation that does not respond to FBXO42/CCDC6 alteration.

The potential hits should also be tested in wild type p53 expressing cells to confirm the specificity to mutant p53s.

Response: In the revised submission, we will include WB for WT lines (e.g. RPE1) upon genetic alteration of CCDC6 and FBXO42. This was already performed for C16orf72 (Figure 6D).

(6A) The role of C16orf72 in restraining p53 activity has been reported previously, as has the interaction with HUWE1 (including a new publication PMID: 35776542). The authors suggest an interaction between C16orf72 and USP7, although this should be shown with endogenous proteins. The relative importance of USP7 and HUWE1 binding is not explored. (6B) The effect of C16orf72 overexpression in promoting mammary tumors is impressive, although maybe the more interesting question is whether inhibition of C16orf72 expression can limit tumor development in this system.

Response to 6A: we are excited about the independent observations by other group(s) confirming similar results! As a part of our improvement for mechanistic work-up, in the revised submission, we will attempt to address, whether C16orf72’ regulation of p53 is dependent on USP7 and/or HUWE1, or other known E3s, such as MDM2.

(1) Whether the interaction of C16orf72 and HUWE1 or USP7 is required for the C16orf72 regulation of p53. Specifically, for example, we will perform epistasis experiments to test USP7’ or HUWE1’ ability to rescue the p53 levels in reporters upon ∆C16orf72. Due to the toxicity/lethality in WTp53 lines induced by the loss of C16orf72, we intend to test using R273H-reporter, or RPE1-line with ∆CDKN1A (p21) that is a synthetic viable rescue for ∆*C16orf72. *

(2) In the revised submission, we will attempt to perform endogenous-based C16orf72-USP7 IP experiment (antibody permitting).

6B. The effect of C16orf72 overexpression in promoting mammary tumors is impressive, although maybe the more interesting question is whether inhibition of C16orf72 expression can limit tumor development in this system.

Response: We are also equally excited about the in vivo result supporting the idea that C16orf72 overexpression in tumour-prone mice (Pik3caH1047R) mice harbouring WTp53 may accelerate tumour formations. In the revised submission, we will further support that this effect is specific to WTp53/C16orf72, by including data of the control cohort with p53-null background (LSL-Pi3kH1047R; p53Flox/Flox).

In regard to the effects of C16orf72-depletion in controlling tumour growth - we agree that this would be a very exciting avenue. Conditional C16orf72 mice are being made at the moment and these mice will allow us to comprehensively address this question. However, it will take several more month to generate and validate this line, and then another 2 breeding rounds to generate homozygous C16orf72fl/fl; Pik3caH1047R mice. In addition, the long time required to form tumours in the control mice with WTp53 (~250 days), it becomes not feasible for us to test whether the inhibition of C16orf72 could limit the tumour development, given the revision timeline. As such we respectfully believe that this would be beyond the scope of this manuscript.

Reviewer #2, Minor comments.

Figure 1b: The nutlin concentration stated in the methods section is wrong. Should be 10 µM instead of 10 nM (correct in figure legend).

Figure 6b: y-axis label is missing.

Figure 1e/f Legend: Should be FDR 0.5.

Response: We apologise for typos. The current submission has incorporated the corrections.

Figure 1c: Include results for a mutant that is not regulated by MDM2, such as R175H. Otherwise, as a standalone experiment, this figure doesn't add much.

Response: We thank the reviewer for this suggestion. In the revised submission, we will include R175H/R337H.

Figure 1h: While an UpSet plot is an elegant way to present unique and overlapping hits between different screens, Venn diagrams might be more 'accessible' to many readers and easier to understand.

Response: We thank the reviewer for this feedback. The choice of UpSet blot was largely motivated by the different categories involved, which made the area representation and the intersection of the conventional Venn diagram no longer feasible.

In the revised submission, we will improve our figure legend for the UpSet blot, to improve the readability.

Might be worth stating that mClover is an eGFP variant and can therefore be targeted by eGFP sgRNAs so that it is easier to understand the following:

o Page 5, paragraph 1: "We used the TKOv3 sgRNA library, which contains [...] 142 control sgRNAs targeting EGFP, LacZ and luciferase"

o Page 5, paragraph 2: "As expected, sgRNAs targeting p53 and mClover were the most depleted sgRNAs, [...]

Response: We thank the reviewer for this suggestion. We believe this will also improve the readability and have incorporated this into our current submission.

Reviewer #2, Significance.

Reviewer #2 (Significance (Required)):

This is an interesting concept and the results could provide a useful resource for groups interested in the regulation of p53. The authors chose to focus on candidate genes that could have been identified by looking for the top 30 p53 co-dependent genes on depmap (C16orf72 is #24 in this list and FBXO42 is #28, most of the other genes ranking above are already known as p53 regulators). While this validates the screen, it would have been interesting if the authors had identified and validated new regulators of p53 that were not apparent from previously published work.

Response: We thank the reviewer for all the thorough and constructive comments! In relation to the DepMap dataset, we are excited that many of the top hits from our screens are indeed top WTp53-correlators/anti-correlators (e.g. MDM2, USP28)!

While the DepMap dataset used cell fitness/viability to construct the genetic relation score, this assay may not effectively rule out the many regulators that could otherwise elicit their regulation of p53 via regulating the general cell response to cell cycle, stress, etc. In our screen systems (i.e. protein stability and synthetic viability screens), we attempted to focus on the regulators of p53-stability (post-translational), and further coupled it with the synthetic viability screens to concentrate on hits that have a more direct role in p53 regulation (e.g. MDM2, C16orf72).

One other difficulty to fully couple our screens to the DepMap dataset is due to the limited cell lines harbouring endogenous mutant p53s, e.g. R337H. This may also contribute to the uniqueness of the identified R337H-reporter specific hits (where cell lines harbouring R337H have not yet been included in the DepMap dataset), e.g. several Aminoacyl tRNA synthetases (SARS, YARS, etc) were identified as R337H unique regulators and subsequently verified using different guides in the reporter line, but could not be obtained via DepMap.

We largely see this paper as a resource for the p53 field and would like to publish it as soon as possible. In fact, when we started working on C16orf72 or CCDC6/FBXO42, these hits were not known for their ability to regulate p53. We will work up several other hits, but this would be beyond the scope of this paper and the first author’s Ph.D. thesis that needs to be completed under a timeline.

Reviewer #3.

Reviewer #3 summary:

The manuscript by Lu and coworkers performed genome wide CRISPR screens to search for genes that when knocked out, lead to p53 accumulation or degradation. Wt p53 and a panel of p53 hotspot mutants were chosen as reporter for the screen. The approach reassuringly identified many previously described regulators of p53 degradation, and also found a large set of new hits that many appear to be indirectly affecting p53 level.

A key step of this approach is the follow up functional and mechanistic study of the hits. To this end, the authors chose FBXO42 as a top hit that blocks mutant p53 degradation, and C16orf72 as a top hit that promotes wt/mutant p53 degradation.

Overall the functional data for FBXO42 is disappointing. FBXO42 knockout has quite modest effect on mutant p53 level (~50% reduction). The knockout also showed some effect on p53 mRNA level (~25% reduction), making the determination of mechanism difficult. It does not appear to be a promising targeting for reducing mutant p53 level and gain of function activity in tumor cells.

We thank the reviewer for this constructive comment! We will address this in the revision, as proposed in Point #3.

The C16orf72 finding unfortunately lost some novelty because it was independently identified as a p53 regulator in a recent study using CRISPR screening (PMID: 33660365). However, the repeated identification is reassuring and the current work provides more convincing functional data, showing C16orf72 knockout increase wt p53 level, inhibits cell proliferation specifically in p53+/+ cells, and overexpression of C16orf72 reduce wt p53 level and accelerates progression of a breast tumor mouse model. Their results suggest C16orf72 is a biologically relevant regulator of p53 in cancer development. In order to provide a reasonable amount of new information and set it further apart from the published study, some biochemical analysis looking into the mechanism of C16orf72 will be helpful.

Reviewer #3 Major and Minor comments:

Specific comments:

1. There appears to be a mix up in the figure legend for Fig.1A describing line 1 and 2.

Response: We sincerely apologise for the mix up in the figure legend! In the current submission, this has been fixed.

Fig.2. Data for some p53 mutants mentioned in the text cannot be found in the main figure 2D and supplemental figure S3A.

Response: We apologise for having not included the R175H and R337H mutants in Supplemental Figure S3A. In the revised version, we will include these two mutants.

Fig.2 E-F. The effects of FBXO42 and CCDC6 KO on endogenous mutant p53 level is small (~50% decrease). Given that mutant p53 accumulates at high levels, whether a 50% decrease has meaningful effect on its gain of function activities is questionable. The knockouts also caused a ~25% decrease in p53 mRNA (FigS3F) which makes the mechanism quite difficult to investigate further.

Response: We agree with the reviewer that the current data makes it difficult to conclude the mechanism. Given the design of our reporter, we still believe that the regulations could largely be at the post-translational level. In our revised version, we plan to exam the ubiquitination status of p53 upon losses of CCDC6/FBXO42, and also monitor the p53 degradation via cycloheximide chase.

To further address whether this reduced level of mutp53 has biological impacts, we plan to test it in the tumour cell context. Given the difference in migration capability observed between PANC-1 and PANC-1-∆p53 line (e.g. PMID: 35439056), we plan to also evaluate the migration pattern of PANC-1, with the presence and absence of FBXO42/CCDC6 (controlled by similar FBXO42/CCDC6 loss in PANC-1- ∆p53 background). Furthermore, in tissue culture, although there is only marginal to no difference in cell growth rate between many mutant p53 lines (e.g. PANC-1) and their ∆p53 line, we plan to test whether a reduced serum or nutrient level could exacerbate the difference, and hence further be used to monitor the difference resulted from the loss of FBXO42/CCDC6.

Fig.3B. The IP experiment using p53 shRNA and control shRNA should be done by IP of p53 followed by CCDC6 western blot. If CCDC6 IP is used as in the figure, then a CCDC6 shRNA knockdown sample should be compared to control shRNA. The current data does not rule out the possibility that CCDC6 antibody can nonspecifically pull down some p53.

Response: We apologise for the confusion. In the revised version, we will include the proper reciprocal IP, with IP of endogenous p53 (R273H) followed by blotting of CCDC6.

Fig.3D. The in vitro pull down experiment needs specificity controls such as non affected R175H p53 core domain. The data presented would suggest that MBP-FBXO42c captured more than 1:1 molar ratio of R273H core domain, which is unusual for specific binding unless there is aggregation of p53.

Response: We thank the reviewer for this constructive comment! In the revised version, we will incorporate this, by repeating the in vitro pull-down assay including a non-p53 control protein.

To increase the impact of the current study, the authors could provide more mechanism insight on how C16orf72 regulates p53 level, which was also missing in the other published study. For example, addressing whether C16orf72 effect is dependent on MDM2. Does it cooperate with MDM2 to ubiquitinate p53. Does it promote p53 ubiquitination in the absence of MDM2, since it interacts with HUWE1. Does it act by recruiting usp7 to stabilize MDM2.

Response: we thank the reviewer for this very constructive and thorough comment! In our revised version, we will attempt these assays and incorporate them into the submission.

Together with our response to Reviewer #2, Point #6, in the revised submission, we will attempt to address if C16orf72 regulation of p53 is dependent on MDM2 or HUWE1.

(1) Whether the interaction of C16orf72 and HUWE1, or C16orf72 and USP7 is required for the C16orf72 regulation of p53. Specifically, for example, we will perform epistasis experiments to test HUWE1’ or USP7’s ability to rescue the p53 levels in reporters upon the loss of C16orf72 (∆C16orf72). Due to the toxicity/lethality in WTp53 lines induced by the loss of C16orf72, we intend to test using the R273H-reporter, or RPE1-line with ∆CDKN1A (p21) that is a synthetic viable rescue for ∆*C16orf72. *

(2) Whether C16orf72 dependent upon or cooperate with MDM2 in regulating p53.

We will first probe whether C16orf72 overexpression increased the p53 ubiquitination, and then decide whether overexpression of C16orf72 has additive effects to MDM2 overexpression in regulating p53 levels.

We previously observed that overexpressing C16orf72 could not rescue the R273H level resulted from losing MDM2 (using flow-cytometry in R273H-reporter-∆MDM2), and as such, we plan to test the C16orf72-MDM2 relation in the MDM2-proficient context.

The manuscript is in a form extremely unfriendly to review, text, figures and legends are all split up at multiple locations, the pdf figures are very sluggish to scroll.

Response: We sincerely apologise for the inconvenience. In the current submission, we have split the submission into three separate files, (1) main text, (2) main figures, and (3) supplemental figures, along with (4) supplemental tables as individual EXCELs. We will also reduce the resolution of a few images, so the overall higher resolution is retained, while still fitting into the file size limit.

Reviewer #3 (Significance (Required)):

The work is significant in identifying a functionally relevant regulator of p53 stability.

Response: we thank the reviewer again for the very constructive feedback!

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

The manuscript by Lu and coworkers performed genome wide CRISPR screens to search for genes that when knocked out, lead to p53 accumulation or degradation. Wt p53 and a panel of p53 hotspot mutants were chosen as reporter for the screen. The approach reassuringly identified many previously described regulators of p53 degradation, and also found a large set of new hits that many appear to be indirectly affecting p53 level.

A key step of this approach is the follow up functional and mechanistic study of the hits. To this end, the authors chose FBXO42 as a top hit that blocks mutant p53 degradation, and C16orf72 as a top hit that promotes wt/mutant p53 degradation.

Overall the functional data for FBXO42 is disappointing. FBXO42 knockout has quite modest effect on mutant p53 level (~50% reduction). The knockout also showed some effect on p53 mRNA level (~25% reduction), making the determination of mechanism difficult. It does not appear to be a promising targeting for reducing mutant p53 level and gain of function activity in tumor cells.

The C16orf72 finding unfortunately lost some novelty because it was independently identified as a p53 regulator in a recent study using CRISPR screening (PMID: 33660365). However, the repeated identification is reassuring and the current work provides more convincing functional data, showing C16orf72 knockout increase wt p53 level, inhibits cell proliferation specifically in p53+/+ cells, and overexpression of C16orf72 reduce wt p53 level and accelerates progression of a breast tumor mouse model. Their results suggest C16orf72 is a biologically relevant regulator of p53 in cancer development. In order to provide a reasonable amount of new information and set it further apart from the published study, some biochemical analysis looking into the mechanism of C16orf72 will be helpful.

Specific comments:

There appears to be a mix up in the figure legend for Fig.1A describing line 1 and 2.

Fig.2. Data for some p53 mutants mentioned in the text cannot be found in the main figure 2D and supplemental figure S3A.

Fig.2 E-F. The effects of FBXO42 and CCDC6 KO on endogenous mutant p53 level is small (~50% decrease). Given that mutant p53 accumulates at high levels, whether a 50% decrease has meaningful effect on its gain of function activities is questionable. The knockouts also caused a ~25% decrease in p53 mRNA (FigS3F) which makes the mechanism quite difficult to investigate further.

Fig.3B. The IP experiment using p53 shRNA and control shRNA should be done by IP of p53 followed by CCDC6 western blot. If CCDC6 IP is used as in the figure, then a CCDC6 shRNA knockdown sample should be compared to control shRNA. The current data does not rule out the possibility that CCDC6 antibody can nonspecifically pull down some p53.

Fig.3D. The in vitro pull down experiment needs specificity controls such as non affected R175H p53 core domain. The data presented would suggest that MBP-FBXO42c captured more than 1:1 molar ratio of R273H core domain, which is unusual for specific binding unless there is aggregation of p53.

To increase the impact of the current study, the authors could provide more mechanism insight on how C16orf72 regulates p53 level, which was also missing in the other published study. For example, addressing whether C16orf72 effect is dependent on MDM2. Does it cooperate with MDM2 to ubiquitinate p53. Does it promote p53 ubiquitination in the absence of MDM2, since it interacts with HUWE1. Does it act by recruiting usp7 to stabilize MDM2.

The manuscript is in a form extremely unfriendly to review, text, figures and legends are all split up at multiple locations, the pdf figures are very sluggish to scroll.

Significance

The work is significant in identifying a functionally relevant regulator of p53 stability.

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 paper describes several genome-wide CRISPR screens designed to identify regulators of p53 stability. The authors use a system in which p53 levels are marked by mClover expression, using RFP expression to normalise for gene expression changes.

1. The bimodal distribution of p53 expression levels in some reporter cell lines (G245S, R248Q, R248W and R273H) hampers the implementation of a robust readout and makes correct interpretation of the results challenging. While it is possible that the bimodal distribution indicates dynamic changes in p53 levels within one population, it also seems possible that a subclone of these cells have acquired additional alterations affecting p53 stability, and that the authors are screening a mixed population of two intrinsically different cell populations. This would make it difficult to interpret the results of the screen in these cell lines and may be a challenge when trying to identify something that has not already been highlighted on depmap.
2. The coverage of the sgRNA library (200x) is rather low for a negative selection screen, where a coverage of 500x would be more desirable. The FDR threshold is also rather lenient, a more stringent FDR threshold would seem more appropriate and shorten the list of potential hits.
3. Although the study is focused on the regulation of p53 stability, there are no experiments to show that any of the manipulations alter the ubiquitination or degradation (half-life) of p53. The rescue of expression by proteasome inhibition is very modest (Figure 2E), suggesting the loss of expression may not be a reflection of degradation. A role for endogenous FBXO42 and C16orf72 in regulating the ubiquitination and half-life of endogenous p53 should be confirmed
4. Many p53 mutants are used for the initial screens, but very little validation is carried out to show that the apparent differences in factors regulating their stability persists in cells naturally expressing these mutants. For example, FBXO42 is identified as a protein required to maintain the stability of R273H, 248W and R248Q, but not R175H, G245S and R337H. While the authors show an association of CCDC6 and p53 in PANC1 cells (expressing 273H), it would be important to show a panel of R273H, 248W and R248Q expressing tumor cells and the response of p53 to FBXO42 and CCDC6 depletion, compared to similar experiments in a panel of R175H, G245S and R337H expressing tumor cells. Again, it would be important to show that any changes in protein levels are due to changes in protein stability.
5. The potential hits should also be tested in wild type p53 expressing cells to confirm the specificity to mutant p53s.
6. The role of C16orf72 in restraining p53 activity has been reported previously, as has the interaction with HUWE1 (including a new publication PMID: 35776542). The authors suggest an interaction between C16orf72 and USP7, although this should be shown with endogenous proteins. The relative importance of USP7 and HUWE1 binding is not explored. The effect of C16orf72 overexpression in promoting mammary tumors is impressive, although maybe the more interesting question is whether inhibition of C16orf72 expression can limit tumor development in this system.

Minor comments

• Figure 1b: The nutlin concentration stated in the methods section is wrong. Should be 10 µM instead of 10 nM (correct in figure legend).
• Figure 1c: Include results for a mutant that is not regulated by MDM2, such as R175H. Otherwise, as a standalone experiment, this figure doesn't add much.
• Figure 1e/f Legend: Should be FDR <0.5 not >0.5.
• Figure 1h: While an UpSet plot is an elegant way to present unique and overlapping hits between different screens, Venn diagrams might be more 'accessible' to many readers and easier to understand.
• Might be worth stating that mClover is an eGFP variant and can therefore be targeted by eGFP sgRNAs so that it is easier to understand the following:
  • Page 5, paragraph 1: "We used the TKOv3 sgRNA library, which contains [...] 142 control sgRNAs targeting EGFP, LacZ and luciferase"
  • Page 5, paragraph 2: "As expected, sgRNAs targeting p53 and mClover were the most depleted sgRNAs, [...]
• Figure 6b: y-axis label is missing

Significance

This is an interesting concept and the results could provide a useful resource for groups interested in the regulation of p53. The authors chose to focus on candidate genes that could have been identified by looking for the top 30 p53 co-dependent genes on depmap (C16orf72 is #24 in this list and FBXO42 is #28, most of the other genes ranking above are already known as p53 regulators). While this validates the screen, it would have been interesting if the authors had identified and validated new regulators of p53 that were not apparent from previously published work.

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 by Lu et al., the authors describe some CRISPR screens and protein-protein interaction screens to identify novel regulators of wild-type p53 and mutant p53 function and stability. Besides generating a wealth of data, they discover FBXO42-CCDC6 as positive regulators of the some p53 hot-spot mutants, including R273H mutant p53, but not of all p53 mutants tested and also not of wild-type, indicating selectivity. Furthermore, the found C16orf72(TAPR1) as a negative regulator of p53 stability. Mechanistically, the authors claim a direct interaction between FBXO42 and CCDC6 and p53, but the importance of these interactions has not been shown. On the other hand the authors suggest that the FBXO42/CCDC6 regulate p53 via destabilization of USP28, but also the mechanism has not been worked out. For c16orf72, they show that it interacts with USP7, but no relevance of this interaction is shown either.

Major points

One very important point for me is that the authors do not show the levels of expression of p53 in the p53-mClover stable cell lines. It is known that overexpressed p53 is usualy more stable than endogenous levels of wt-p53. Therefore, I think it is necessary that the authors show the levels of the p53-mClover fusion proteins in the stably transduced cell lines compared to endogenous p53 levels in the parental RPE1 cells and also compared to the endogenous levels of R273H mutant in the PANC-1 cells.

Also the functionality of the wild-type p53-mClover fusion is questionable, at least not shown. One would expect that the overexpression of a functional wt-p53 in p53-KO cells will affect the survival of the RPE1 cells. In Figure 5A the authors show that depletion of MDM2 or C16ORF72 is toxic for the RPE1 cells in a p53-dependent manner, indicating that elevated levels of p53 cannot be handled by these cells. So, experiment(s) showing that the wt-p53/mClover fusion is functional is needed.

A second important point is that the 'verification' of the hits from the screens is only done in one cancer cell line, PANC-1, with mutant p53. I would have like to see at least one other cell line with another p53 mutant endogenously expressed that is also regulated by FBXO42/CCDC6.

For many of the p53-mutants, a bimodal expression is observed. In the FBXO42- and CCDC6-depleted cells, the equilibrium shifts towards more negative cells but the levels in the two populations itself don't change (while for example for USP28 depletion also the right peak shifts further up, Fig S4E). Is there any correlation with the cell cycle and p53 expression? And can the authors exclude that FBXO42 and CCDC6 are involved in cell cycle progression and hereby influence p53 indirectly (by combining PI staining with Clover-p53 for example).

• The authors claim that the FBXO42-CCDC6 axis regulates stability specifically some p53-mutants, including R273H-mutant, in a manner involving USP28. But USP28 regulates all forms of p53, not just some mutants version. How can the authors reconcile this apparent contradiction?

On a similar note, the authors show that FBXO42 and CCDC6 interact with p53, but not USP28. Do FBXO42 and CCDC6 interact with each other and with USP28? And is the interaction with p53 specific for the R273H version? This part of the mechanism is very poorly defined and the Co-IPs are not very convincing or relevant for the proposed model.

Minor points

The mechanisms of p53 regulation may vary greatly in different cell lines. Can the authors discuss why they choose to do the screen with different mutants, rather than with different cell lines expressing these same mutant endogenously? .

Figure 1: Typo in the legends : Nultin ipv Nutlin

Figure 1b,1c : Show basal and Nutlin-3 induced MDM2 levels and in the overexpression cell lines; if WT-p53 is functional, MDM2 levels should be higher in WT-transduced cells compared to control or mt-p53 expressing cells. Authors should explain which they name USP7 a negative regulator of p53, since it is supposed to de-ubiquitinate p53?!

Figure 2E: the effect of MG132 on p53 seems to be very minimal on this Western blot; it would need quantification to be convincing...Quality of the blot is also not great. The fact that in control cells the levels of p53 R273H are not affected by MG132 treatment fits with Suppl Figure 2E, indicating that the proteasome has no effect on p53 R273H.

Suppl figure 3b, 3c, 3d:

Somehow, I have the feeling that the results from the western blots and the FACS do not match fully, although not all the time-points are shown in the various experiments. For example, the FACS analysis (3b) suggests that in control-transduced cells after 16 hr p53 is still increased. However, that is not clear at all in theWestern blot (3c) Is Suppl Figure 3d the quantification of 3c experiment? If so, in the blot also the 24 hrs should be shown. The blot shown in Suppl Figure 3c suggests that CCDC6 expression increased upon irradiation. Do the authors agree with that? Would that explain why depletion of CCDC6 has more effect upon irradiation? Suppl Figure S3E: if I am right, this is essentially the same type of experiment as shown in figure 2e, but analysis of p53-expression by Western blot. In that blot no real effect of MG132 on p53 levels could be seen. But here, in the FACS analysis, MG132 clearly increases the p53-Clover fusion levels; for me again that Western blot and FACS data do not neccesarily match.

Figure 3B: In the CCDC6 IP a very small amount of p53 can be found. I don't know how much input lysate compared to amount of lysate for IP is used, but the percentage of p53 found interacting with CCDC6 seems so marginal that is is difficult to explain the effect of KO of CCDC6 in PANC1 cells. And, the authors called it a 'reciprocal IP' (Suppl Figure 4a) after transfection of V5-tagged CCDC6 into PANC1 cells,but it actually is the same type of IP. Did the authors try to IP p53 and blot for CCDC6? That would be a reciprocal IP.

Figure 3H: how can authors explain that basal levels of USP28 in control and CCDC6-KO cells transfected with control plasmid are more or less the same and not reduced in the CCDC6-KO cells?

Figure 3I: Essentially the whole blot here is of low quality; especially the FBXO42 blot; is deletion of USP28 increasing FBXO42 protein levels, or is it just the quality of the blot? All in all it seems that FBXO42 is very low expressed in the used cell lines.

Figure 4B: I find it a bit surprising that USP7 is also found in the synthetic viability screen, since it has been shown that USP7 has many more essential targets and KO of p53 only partially rescues the development of USP7-KO mouse embryo's.

Figure 5: the authors nowhere show the efficacy of the guides targeting c16orf72. A Western blot showing the expression and the reduction upon expressing the guide-RNAs is essential. Figure 5E: First, here probably parental RPE1 cells have been used, but that is not stated. Second, the authors state 'only a slight increase in p53 levels upon siHUWE1'; I would say none compared to scrambled. I know HUWE1 is a very huge protein, but the blot of HUWE1 is not convincing. I seem to be able to conclude that siMDM2 and siUSP7 reduces HUWE1 levels? Figure 5F, in relation to figure 5D. Here the author overexpress both c16orf72 and USP7, and find an interaction. The implication of that is not clear. If they want to make point of this interaction, they should have looked at endogenous proteins. It is worrying that USP7 apparently was not one of the hits in de Mass-spec experiment of which results are shown in Figure 5D. Also in that experiment c16orf72was overexpressed, and USP7 is very highly expressed in essentially all cell lines, so do the authors have an explanation?

Suppl. figure 5D is missing

Referees cross-commenting

I agree essentially with all comments of Reviewer #2. Especially the major points 3 and 4. The use of more cell lines expressing endogenous mutant p53 is very important. In addition, I can agree with almost all comments of Reviewer #3. The effects especially of FBXO42 ablation are rather minimal, so relevance is questionable.

Significance

Nature and Significance

Compare to existing literature

The topic of the paper is of high interest given the relevance of p53 and its gain-of-function mutants in oncology, and the screens are well executed and clearly presented. In terms of novelty, FBXO42 has been linked to p53-degradation before, and c16orf72 was recently shown to be able to destabilize p53. However, the link between CCDC6 and p53 is novel and of interest, since they are both substrates of USP7 and are both regulators of the cell cycle.

We think the manuscript has potential to add something to the field, but would benefit greatly from a better understanding of the molecular underpinnings of their newly described mechanisms, as well as the conditions in which the mechanism is active.

Therefore, it might be advisable to shorten the manuscript, and go more in-depth in finding the mechanisms of regulation.

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

Manuscript number: RC-2021-01204R

Corresponding author(s): Alexander, Aulehla

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

*The paper by Miyazawa and colleagues addresses a key question: How is changed metabolic activity sensed and to induce changes in developmental programs. In recent years, there is more and more indication that metabolism is not only a dull workhorse synthesizing the building blocks for new cells and providing chemical energy, but that metabolic activity itself has also a regulatory role. How this precisely works is largely unknown and even also unexplored in higher cells. From early insights obtained in microbes, it seems that certain metabolites - possibly reflecting metabolic activity (i.e. flux) - could be metabolic signals that feedback into cellular regulation. *

*The current paper takes this idea now to developmental processes, where the authors found that the glycolytic metabolite fructose-1,6-bisphosphate is a flux-dependent signal that interferes with developmental processes. This is a very exciting finding, as it indicates that this metabolite not only has a regulatory function in microbes but also in mouse during mesoderm development. *

*Answering the question how such a flux-dependent metabolite mechanistically interferes with the developmental processes is an enormously difficult. Compared to other mechanistic studies, where deleting genes, modifying genes, and changing protein expressions will usually do the trick, here, perturbing metabolite levels is extremely challenging, particularly if such perturbations need to be carried out in a way that nothing else is perturbed. Researchers, who are not overly familiar with metabolism, usually underestimate the difficulty with targeted and insightful perturbation of metabolism. *

*To this end, the authors of this paper need to be congratulated for a very well carried out study with very solid data, and excellent control experiments. The authors open up a new path towards understanding how embryo mesoderm development is regulated by metabolic activity. In particular, they show that that glycolytic flux, FBP and important developmental phenotypes as well as protein localization changes are linked. As normal with a complex metabolism-based story as this one, there is always more that could be done. Yet, the results are highly important to be reported now such that the field as a whole can build on these interesting results and to explore the exciting path further that has been opened by the authors. Thus, I strongly recommend publishing these findings: The data generated by the authors are accompanied by the required control experiments. The conclusions drawn are very solid. I do not have any major concerns but just a number of minor suggestions that the authors could consider in a revised version of the manuscript. *

*Minor: *

1. • At the end of the introduction, the authors stated their original goal. As it is phrased, it is unclear whether this goal has been obtained or not. They might want to consider replacing the last introductory sentence by a sentence stating what the reader can find in this paper.*

1. We agree with the reviewer and have rephrased accordingly (line 112–117):

“In this study, our goal was therefore to first determine in vivo sentinel metabolites during mouse embryo PSM development. We then combined genetic, metabolomic and proteomic approaches to investigate how altered glycolytic flux and metabolite levels impact developmental signaling and patterning processes.”

• Data from Fig 3: If you plot the lactate secretion vs the FBP levels of the controls and the overexpression experiment, would the control and the overexpression data lie on one line (maybe if combined with the data shown in Fig 1A)?*

2. As the reviewer suggested, it is of great interest to check whether lactate secretion and FBP levels show a similar correlation in control and cytoPfkfb3 embryos, considering that cytoPfkfb3 overexpression lifts the upper limit of glycolytic capacity and FBP levels (revised Figure 3B, 3E). As the reviewer suggested, we plotted FBP levels against lactate secretion and fitted a linear regression line onto control samples (please see the Figure R1 below). The new plot shows that lactate secretion and FBP levels in cytoPfkfb3 embryos lie on the linear regression line derived from wild-type samples, highlighting that a correlation between lactate secretion and FBP levels is maintained even in cytoPfkfb3 embryos. We now included this new plot in the revised Figure S4C and modified the text accordingly (line 474-477):

“In addition, FBP levels showed a linear correlation with lactate secretion in control explants, and such a correlation was maintained even in cytoPfkfb3 explants (Figure S4C).”

Figure R1. Correlation between lactate secretion and FBP levels in PSM explants. Linear regression line (a grey line) was derived from the data of control samples cultured in 0.5–25 mM glucose (black circles; from Figure 1A and 3E). The data from cytoPfkfb3 embryos cultured in 2.0–10 mM glucose (from Figure 3B and 3E) are shown as red rectangles.

• Maybe the authors could attempt an experiment like the following one: Chose the strongest phenotype observed and test a combination of overexpressing cytoPfkfb3 and reducing extracellular glucose level at the same time? *

3. We agree this suggested experiment is important to show that the phenotype in cytoPfkfb3 embryos is indeed dependent on glycolytic flux and have already addressed this specific point in our manuscript, see results in Figure 4B and 5A in our original manuscript. The results show that the phenotypes in cytoPfkfb3 explants, i.e. reduction in somite formation and downregulation of Msgn mRNA expression occur in a glucose dose-dependent manner. Since in this embryonic context, we show that glucose concentration impacts glycolytic flux (see increased lactate production upon glucose titration in Figure 3B), our findings support the conclusion that the effect of cytoPfkfb3 overexpression is flux-dependent and not due to the overexpression per se. Based on the reviewer's feedback, we have modified the text to clarify and highlight this critical point (line 339–345):

“Combined, these results show that cytoPfkfb3 overexpression results in reduced segment formation, arrest of the segmentation clock oscillations and downregulation of Wnt signaling, in a glucose-dose dependent manner. As glucose concentration impacts, in turn, glycolytic flux (Figure 1A, 3B), these findings suggest that these phenotypes are flux-dependent and are not a mere result of cytoPfkfb3 overexpression.”

• Can the proteomics experiments shown in Fig. 6 be repeated with high and low extracellular glucose? High glucose should yield high FBP levels and one would then expect to see the same as with the experiment where at 2 mM glucose 20 mM extracellular FBP were added. Is this the case? *

4. We agree with the reviewer that based on the findings, one would expect the phenotype, i.e. in this case translocation of proteins, to correlate with FBP levels. Two of our results are of note in this regard.

First, our data indicates that in order to see the effect on protein localization, high levels of FBP have to be reached. Accordingly, we find that Pfkl becomes depleted from the nuclear-cytoskeletal fraction in cytoPfkfb3 explants when cultured in 10 mM glucose but not (visibly) in 2.0 mM glucose (Figure 7D). Corresponding to this, FBP levels in cytoPfkfb3 explants show a significant increase (about 3-fold) from 2.0 to 10 mM glucose conditions (revised Figure 3E).

Second, in control samples, FBP levels saturate in high glucose conditions. FBP levels in control samples do not further increase when glucose concentration is increased from 10mM to 25mM, and thus it does not become as high as in cytoPfkfb3 embryos cultured in 10 mM glucose (revised Figure 3E).

Therefore, in order to reveal the translocation, it requires an experimental strategy that leads to significantly increased FBP levels, such as in cytoPfkfb3 explants with high glucose condition, or alternatively, direct supplementation of FBP.

As also pointed out by the other reviewers, we are experimentally generating controlled conditions that exceed the physiological range which the embryo is exposed to. Accordingly, our data does not constitute evidence that under physiological conditions an alteration of protein localization in response to change in glycolytic flux and FBP levels occurs, at a smaller scale.

We regard our approach as a first step to reveal potential mechanisms and so far hidden possible responses to changes in metabolic flux. In order to see minor changes in translocation upon small changes in glycolytic-flux/FBP levels, more quantitative approaches, such as live-imaging of tagged proteins, will need to be developed. We hence decided to include these discussion in our revised manuscript (line 657-666):

“Of note, the translocation of proteins was observed only when high levels of FBP were reached upon direct FBP supplementation or cytoPfkfb3 overexpression with high glucose (Figure 6, 7). Future studies hence need to investigate whether flux-dependent change in protein localization occurs upon moderate and more physiological changes in glycolytic-flux/FBP levels. To this end, the development of more quantitative approaches, such as live-imaging of tagged enzymes and the development of metabolite biosensors, are needed.”

• While the authors quantified proteins in different compartments, I was wondering whether they also looked for whole-embryo protein expression changes? *

5. We have not done protein expression analysis using whole embryos, or other isolated tissues in this study. This is indeed a potentially interesting future experimental comparison.

• Throughout the manuscript, the authors state the glucose levels or cytoPfkfb3 changes the glycolytic flux. While I tend to agree with this, it is important to note that the authors have not directly measured glycolytic flux, but use the amount of accumulated lactate as a proxy. I think it is important to add this disclaimer at important points in the manuscript, such that readers are aware of this point. *

6. We fully agree with the reviewer and now have added the following sentence in the first result section to make this point clearer to the reader (line 126-128):

"Throughout this study, we used quantification of secreted lactate as a proxy for glycolytic flux due to the inability to directly measure flux in embryonic tissues."

Another aspect for changing FBP levels could be connected on what was found in yeast, where the FBP levels were found to oscillate with the cell cycle (https://pubmed.ncbi.nlm.nih.gov/31885198/). Could this be connected with the pattern formation here?

7. This is indeed an interesting aspect to discuss; in the absence of experimental evidence connecting the observed pattern formation and cell cycle (though some classic work had suggested its existence) we have decided to omit the discussion of this potential link.

• Line 606: The mentioned review article also covers yeast. As such, maybe the authors should replace the term "bacteria" with "microbes"? *

8. We modified our manuscript accordingly.

Reviewer #1 (Significance (Required)):

**Referees cross-commenting**

As I mentioned in my comment, targeted metabolic perturbations are extremely difficult. Perturbing a metabolite level without at the same time perturbing the flux through this pathways is difficult (of not impossible). Also, the opposite is the case.

I am not sure whether experiments as the one suggested by reviewer 2 (comment 1) will really lead to results from which further conclusions can be drawn. Furthermore, there does not need to be a linear correlation between the extracellular glucose concentration and metabolic flux/FBP levels (as my reviewer colleague implies). Thus, I am not sure whether doing this experiment makes sense, or would lead to strengthened conclusions.

Reviewer 2 also states "The lack of proven mechanism for the activity of FBP might restrict the real general impact of this work." I agree that we do not know the downstream targets of FBP, but finding them would likely require many years of additional work. Such work will not be initiated if this paper is not published, and it would be a pity if it would be further delayed. I feel that the evidence is strong enough that FBP has an important role and with this paper published, it will motivate others to look for the downstream targets.

Reviewer 3 makes the point: "Given that FBP levels are highly correlated with extracellular glucose levels (which impact glycolytic flux )(TeSlaa and Teitell, 2014) the authors should elaborate on why progressive increase in extracellular glucose does not affect PSM patterning, in the same way that increasing FBP levels does. " Here, I feel my reviewer colleague might be overlooking that in biochemistry molecular interactions typically reach a saturation at some point. The correlation between extracellular glucose and glycolytic flux has likely only a range where these two measures linearly correlate. Similarily, the correlation between glycolytic flxu and FBP likely also exists only within a certain range, and finally FBP levels and the downstream targets likely also only linearly interact within bounds. Thus, the absence of a correlation at "extremes" does by no mean mean that what the authors propose is incorrect. In fact, it just shows what you expect from biomolecular interactions that there a limits to linear correlations.

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

*Summary. *

*The work described in this paper first searches for potential sentinel metabolites of glycolytic flux, focusing on the process of somitogenesis during mouse embryonic development. By measuring the levels of different metabolites in the presomitic mesoderm (PSM) of E10.5 mouse embryos cultured in the presence of three different glucose concentrations, the authors identify 14 metabolites whose concentration rises with increasing glucose concentration in the culture medium. Among them, they selected fructose 1,6-bisphosphate (FBP) for further analyses, as it showed the highest linear correlation with extracellular glucose concentrations. They then show that addition of FBP to the incubation medium of cultured embryo tails interfere with somitogenesis and tail extension in a concentration-dependent fashion. In addition, they show that this effect is exacerbated when extracellular glucose levels are increased. By analyzing specific targets of Wnt and Fgf signaling, the authors also show that addition of FBP down-regulates both signaling pathways in the PSM. They then use a genetic trick (ubiquitous overexpression of cytoPfkfb3) to increase FBP levels by allosteric activation of Pfk (the enzyme that produces FBP) in developing embryos. When tails from these transgenic embryos were cultured in vitro and exposed to various glucose concentrations somitogenesis was affected in a way resembling the effects of FBP on cultured tails from wild type embryos. The authors then go on to determine the subcellular localization of different proteins in tails incubated in the presence of various FBP concentrations to identify that some enzymes involved in the glycolytic pathway (and they specifically focus on Pfkl and Aldoa) are excluded from nuclear fractions at high FBP concentrations. The authors conclude that FBP functions as a flux-signaling metabolite connecting glycolysis and PSM patterning, potentially through modulating subcellular protein localization. *

*Major comments *

*I think that in general the work described in this manuscript has been performed to the highest technical standards. However, I do not think that I can agree with the authors' conclusions (that FBP connects glycolysis with PSM patterning and that subcellular localization of glycolytic enzymes play a role in this process), which in my opinion go way beyond what can be proven by the data provided. *

*1- Explants incubated with external glucose concentrations up to 25 mM have no obvious defects on somitogenesis or on the segmentation clock as determined by LuVeLu cycling activity. Under these conditions, explants are expected to contain very high FBP levels if this metabolite keeps its linear relationship with external glucose (in this work it was not measured beyond 10 mM glucose in the medium, where FBP concentration was already very high). This contrasts with the phenotypes observed upon exogenous supplementation of FBP, which affects somitogenesis already at 2 mM glucose. These latter results are at odds not only with the lack of phenotypic alterations under high glucose conditions, but also with the observation that exogenous addition of fructose 6-phosphate (F6P), the substrate of Pfk enzymes to generate FBP, does not alter somitogenesis. The authors take the absence of effects by incubation with F6P as a control of the specificity of FBP. However, as F6P is the natural substrate of Pfk, it is possible that supplementation of F6P also leads to an increase of FBP but in a way closer to a physiological condition. Therefore, I find it essential to determine FBP levels in tails incubated in the presence of increasing amounts of F6P, as if it increases FBP levels, similarly to what the authors described for the tails incubated with increasing glucose concentrations, it will have important implications to the interpretation of the work presented in this manuscript. *

9. We agree with the reviewer and to directly address this central point, we have performed an extended, additional experiment, collecting 375 embryos to quantify FBP levels under five conditions with three biological replicates.

There are two major results that we highlight here: First, we found that addition of F6P did not lead to increased FBP levels compared to control samples cultured in 10 mM glucose, which is in stark contrast to cytoPfkfb3 embryos cultured in 10 mM glucose (revised Figure 3E). Second, while increasing glucose concentration is mirrored by elevated FBP levels as we reported, we find clear evidence of saturation above a concentration of 10mM glucose: increasing glucose to 25mM does not increase FBP levels further (revised Figure 3E).

This saturation effect seen in glucose titration, but also the absence of elevated FBP upon F6P addition, might be expected outcomes because, as also the reviewer 1 pointed out in the response, Pfk is commonly considered to be a rate-limiting enzyme in the glycolytic pathway. We now have the direct experimental data supporting this hypothesis and thank the reviewers to have initiated this additional (very involved..) experiment.

This new data allows us to conclude more firmly on the correlation between FBP levels and phenotype: at high FBP levels, which are seen in cytoPfkfb3 samples, we observe PSM patterning defects. These high levels are not reached even at 25mM glucose or upon F6P addition, due to the saturation at the level of PFK enzymatic step. Hence, while glucose titration does elevate FBP significantly until this saturation, FBP levels are not as high as in cytoPfkfb3 samples. As a correlative finding, we see that only those conditions with very high FBP levels, or the direct addition of high levels of FBP, cause the arrest of segmentation clock activity. At moderately elevated FBP levels, observed in control explants with high glucose or in cytoPfkfb3 explants with low glucose, clock activity continues and we find a quantitative effect at the level of gene expression, i.e. Wnt signaling target downregulation (Figure S3, 5A).

The new data has been included in the revised manuscript and the text has been adjusted accordingly:

• (Result Part, line 245–254) "Consistently, we found that cytoPfkfb3 overexpression lifted the upper limit of FBP levels in PSM cells (Figure 3E, S4B, S4C). In control explants, FBP levels did not increase further when glucose concentration was increased from 10 mM to 25 mM. It was also the case when control explants were cultured in 20 mM of F6P (Figure 3E). These results indicate that the Pfk reaction carries a (rate-)limiting role for glycolytic flux and FBP levels, and that cytoPfkfb3 overexpression hinders the flux-regulation function of Pfk."

• (Discussion Part, line 551–573) “Our findings suggest that flux-regulation at the level of Pfk is critical to keep FBP steady state levels within a range compatible with proper PSM patterning and segmentation. In agreement with such a rate-limiting function for Pfk, we found in glucose titration experiments that FBP levels saturated and did not further increase at glucose levels above 10 mM (Figure 3E). Along similar lines, the supplementation of high concentrations of the Pfk substrate F6P did not result in a significant increase of FBP levels, again compatible with a rate-limiting function at the level of Pfk (Figure 3E). The upper limit of glycolytic flux and FBP levels can be experimentally increased by cytoPfkfb3 overexpression (Figure 3B, 3E). We interpret the data as evidence that cytoPfkfb3 overexpression compromises the flux-control function of Pfk and hence much higher FBP (and secreted lactate) levels are reached. Such a drastic increase in glycolytic flux and FBP levels correlates with a severe PSM patterning phenotype (Figure 4), which resembles the phenotype induced by supplementation of high dose of FBP (Figure 2). Our results in mouse embryos hence provides evidence that flux regulation by Pfk, an evolutionary conserved role present from bacteria to humans, serves to maintain FBP levels below a critical threshold.”

*The main difference between the experiments involving FBP supplementation and those involving high glucose concentrations or exogenous F6P addition is that in the later two cases increase in FBP would be restricted to the tissue(s) expressing Pfk, whereas upon FBP supplementation this metabolite would hit any tissue, regardless of whether or not it would ever be physiologically exposed to this molecule. In the case of the PSM, this might be relevant because it has been shown that there is a gradient of glycolysis, being high at the caudal tip and becoming lower at more anterior regions of the PSM, most likely mirroring the distribution of Pfk activity. Exogenous administration of FBP would flatten the gradient, which could lead to alterations in PSM patterning, whereas glucose (and eventually F6P) would not as they would increase FBP locally in the area where it is normally activated, keeping the natural gradient. *

*On the basis of these arguments, to which extent does FBP connect glycolysis and somitogenesis under physiological conditions? *

10. First, we would like to clarify that while indeed glycolytic activity is graded along the PSM, as other and we reported previously (reported in Bulusu et al., 2017 and Oginuma et al., 2017), the baseline expression of the entire glycolytic machinery (from glucose transport to lactate production) is very high, in all PSM cells. Hence, we see that cells all along the entire PSM have very active glycolysis, the posterior PSM being even more active.

For this and related reasons, our interpretation about the difference seen between glucose titration/F6P addition on one side, and FBP addition/cytoPfkfb3 addition on the other side, is based on the role of Pfk in controlling either flux levels or dynamics in all PSM cells.

Hence, while we agree that we generate experimental conditions that allow FBP levels to surpass those found in control embryos, we would like to highlight the fact that even moderate changes in flux does result in very robust functional consequences on gene expression (Figure S3, 5), as we show in this work.

We can currently not fully address the first point raised, i.e. the role of graded flux/graded metabolite levels, due to the experimental limitations. Such a study requires, for instance, the generation of metabolite biosensor reporter lines in order to be able to monitor these changes dynamically, in space and time.

*ESSENTIAL ADDITIONAL EXPERIMENT related to point #1: Measure FBP from PSM explants incubated under various exogenous concentrations of F6P. *

11. We have performed this suggested experiment, which required the collection of n=375 embryos cultured under the various conditions and analysis by LC-MS to quantify metabolites. The outcome was indeed very informative (please refer to our response #9).

*ANOTHER EXPERIMENT THAT COULD BE INFORMATIVE: measure FBP levels in PSM incubated under different glucose concentrations but instead of using the whole PSM together, dividing the PSM in posterior, medium and anterior parts (similarly to what was done in Oginuma et al, 2017, reference in the manuscript) to see if there is a gradient in FBP activation. *

12. While in principle we agree that this experiment could be informative, we consider the proposed experiment beyond the scope of this work and technically very challenging (although possible). With a similar motivation, the development of metabolite biosensors is an alternative route that we are pursuing for future studies (for the detail, please refer to our response #10).

*2- A similar argument could be presented for the results with the cytoPfkfb3 transgenics, as they are based on global artificial overactivation of Pfk, in addition to other possible effects of the ectopic activity of cytoPfkfb3, which were not controlled. Also, while the phenotypic alterations in the PSM in vitro, most particularly in the experiments involving incubation of the tails, are rather strong, the reported effects on somitogenesis in vivo are minor, also questioning the contribution of the in vitro conditions to the final phenotypic effects observed throughout the manuscript. *

13. First of all, we would like to emphasize that the phenotype seen in cytoPfkfb3 embryos, i.e. the reduction of segmentation and downregulation of Wnt-target gene expression, occurs in a glucose dose dependent manner (Figure 4B and 5A). Hence, it is not the overexpression of cytoPfkfb3 per se that can account for the effects seen. But rather, increased glycolytic flux caused by the combination of transgene expression with high glucose results in functional consequences.

In addition, ‘other possible effects’ that the reviewer is referring to should be evident in all transgenic embryos, irrespective of glucose dose. To the contrary, transgenic embryos cultured in low glucose conditions appear unaltered to control embryos.

Second, we agree that we need to distinguish between strong phenotypes, visible at the level of clock arrest, and milder phenotypes, visible at the level of quantitative gene expression changes. It is important to note that the moderate phenotype, i.e. the quantitative gene expression changes seen in posterior PSM, are seen upon the addition of FBP at moderate levels and upon in glucose titration within the physiological concentration range, as well as in cytoPfkfb3 embryos. We take this as evidence that the effects seen in cytoPfkfb3 transgenic embryos reflect a common response also seen under physiological conditions.

To extend this argument to the in vivo setting, we have performed additional experiments using a genetic mouse model for diabetes. As shown in our previous submission, cytoPfkfb3 transgenic animals do not exhibit a drastic in vivo phenotype when dissected at embryonic day 10.5. One interpretation of this finding is that since the cytoPfkfb3 phenotype is glucose and flux-dependent, the in vivo flux is low, reflecting low glucose concentrations described in vivo. To test the effect of increased flux in cytoPfkfb3 embryos in vivo, we therefore crossed the transgenic mice into a diabetic model called Akita, in which a point mutation in the Insulin2 gene causes high maternal glucose levels (Yoshioka et al., 1997; Wang et al., 1999). Using this experimental setup, we tested whether transgenic embryos in Akita diabetic females would manifest in vivo phenotypes.

Indeed, we found that cytoPfkfb3 transgenic embryos developing in Akita diabetic females showed significantly increased cases of neural tube closure defects (50% of cytoPfkfb3 embryos) and developmental delay (control: 38 somites vs. cytoPfkfb3: 34 somites at E10.5), defects not seen in transgenic cytoPfkfb3 embryos from control females (please refer to Figure R2 below). This dependency of the in vivo phenotype on maternal glucose conditions again highlights that the defects observed in cytoPfkfb3 embryos are not due to the expression of cytoPfkfb3 per se, but are rather directly linked to increased/unregulated glycolytic flux.

We included the new in vivo data in the revised Figure S5D-E and modified the text accordingly.

Figure R2. In vivo phenotype of cytoPfkfb3 embryos grown in diabetic Akita females. (A) The number of somites in control (Ctrl) and cytoPfkfb3 (Tg) E10.5 embryos grown in diabetic Akita females. (B) In situ hybridization of Msgn, Uncx4.1, and Shh mRNAs in Ctrl and Tg E10.5 embryos grown in diabetic Akita females (ss, somite stage; scale bar, 500 µm).

In conclusion, combining the arguments in the two previous comments, to which extent the results from the addition of FBP or from the transgenic activation of Pfk are not artefactual phenotypes without real physiological relevance?

14. In our view, two main conclusions, both in vivo and in vitro, can be drawn based on the result we obtained:

First, we find that a moderate increase in glycolytic flux, within the physiological range, leads to a quantitative and consistent change in gene expression, such as downregulation of Wnt target genes (Figure S3, 5). Such a phenotype was the result of either glucose titration or culturing cytoPfkfb3-transgenic embryos in low glucose concentration.

In these conditions, while overall PSM patterning is qualitatively normal, we do find consistent changes at quantitative level, i.e. gene expression changes, which are also mirrored by a reduced rate of segmentation (Figure 4B). A detailed analysis of the quantitative changes at the level of segmentation clock dynamics is being carried out and will be presented in a dedicated follow up study.

Second, we find that a very significant increase in FBP levels, i.e. when cytoPfkfb3 transgenic animals are cultured in high glucose conditions or when samples are cultured in high levels of FBP, PSM patterning is qualitatively altered and segmentation clock ceases to oscillate. In this case, we agree that it is not a physiological condition, as such high levels of flux and FBP are not reached in control samples which have intact flux regulation by Pfk. Nevertheless, such an experimental condition can be insightful, as it very clearly reveals the potential link between glycolysis, clock activity, PSM patterning and the Wnt signaling pathway.

It is the combination between the moderate and the more severe effects, observed both in vitro, and now also in vivo using the Akita model (see above), that we take as evidence for an intrinsic, physiological link between glycolytic activity, PSM patterning and signaling.

*3- The authors seem to give a strong functional meaning to the absence of Pfkl and Aldoa from the nuclear fraction in tails incubated with exogenous FBP, suggesting a "moonlighting" function of these enzymes under FBP regulation. In addition to the purely speculative nature of this interpretation (there is no proof for such activity or even an attempt to test it), the data provided is also difficult to interpret for various reasons. *

15. We fully agree that we do not show a functional role for either the nuclear localization of enzymes or their dynamic change in sub-cellular localization and have tried to express this clearly in the original manuscript:

• (Result Part, line 382-388) “While we have not been able to address the functional consequence of specific changes in subcellular localization, such as the nuclear depletion of Pfkl or Aldoa when glycolytic flux is increased, these results pave the way for future investigations on the mechanistic underpinning of how metabolic state is linked to cellular signaling and functions.”

• (Discussion Part, line 575-577): “While future studies will need to reveal if nuclear localization of glycolytic enzymes is linked to their moonlighting functions or metabolic compartmentalization…”

Based on this comment by the reviewer, we have further emphasised this point in the revised manuscript(line 635-639):

“While we do not have any direct functional evidence so far for a functional role of nuclear localized glycolytic enzymes, our findings do raise the question whether their subcellular compartmentalization is linked to a non-metabolic, moonlighting function.”

The protein levels in nuclear fractions are clearly much lower than those in the cytoplasm (this is best seen in the blots of Figure 6D). Does this represent similar subcellular distribution of these enzymes throughout the tissue or the different levels result from the presence of the enzymes in the nucleus of only a subset of the cells? This might be of importance to understand the possible relevance of the subcellular distribution of those enzymes. All the analyses were done on bulk tissue and, therefore, it is not possible to distinguishing between these possibilities. As the authors have antibodies for these enzymes, they could try to perform immunofluorescence analyses, which would provide spatial data.

16: We agree that a spatially resolved analysis of the subcellular localization of these various enzymes is needed. Unfortunately, the immunofluorescence experiments that we performed did not yield clear, reliable results and hence we can’t provide the answer at this time.

*In addition to this, it would be important to determine Pfkl and Aldoa subcellular localization in explants incubated with different external concentrations of glucose, which in a way reproduces better possible physiological effects (see point 1), to see if under those conditions high FBP also affects subcellular distribution of those enzymes. *

17: Please find our response under #4 (attached below), as this important point was also raised by the reviewer 1.

*(Our response #4) *

*#4. We agree with the reviewer that based on the findings, one would expect the phenotype, i.e. in this case translocation of proteins, to correlate with FBP levels. Two of our results are of note in this regard. *

*First, our data indicates that in order to see the effect on protein localization, high levels of FBP have to be reached. Accordingly, we find that Pfkl becomes depleted from the nuclear-cytoskeletal fraction in cytoPfkfb3 explants when cultured in 10 mM glucose but not (visibly) in 2.0 mM glucose (Figure 7D). Corresponding to this, FBP levels in cytoPfkfb3 explants show a significant increase (about 3-fold) from 2.0 to 10 mM glucose conditions (revised Figure 3E). *

*Second, in control samples, FBP levels saturate in high glucose conditions. FBP levels in control samples do not further increase when glucose concentration is increased from 10mM to 25mM, and thus it does not become as high as in cytoPfkfb3 embryos cultured in 10 mM glucose (revised Figure 3E). *

*Therefore, in order to reveal the translocation, it requires an experimental strategy that leads to significantly increased FBP levels, such as in cytoPfkfb3 explants with high glucose condition, or alternatively, direct supplementation of FBP. *

As also pointed out by the other reviewers, we are experimentally generating controlled conditions that exceed the physiological range which the embryo is exposed to. Accordingly, our data does not constitute evidence that under physiological conditions an alteration of protein localization in response to change in glycolytic flux and FBP levels occurs, at a smaller scale.

We regard our approach as a first step to reveal potential mechanisms and so far hidden possible responses to changes in metabolic flux. In order to see minor changes in translocation upon small changes in glycolytic-flux/FBP levels, more quantitative approaches, such as live-imaging of tagged proteins, will need to be developed. We hence decided to include these discussion in our revised manuscript (line 657-666):

“Of note, the translocation of proteins was observed only when high levels of FBP were reached upon direct FBP supplementation or cytoPfkfb3 overexpression with high glucose (Figure 6, 7). Future studies hence need to investigate whether flux-dependent change in protein localization occurs upon moderate and more physiological changes in glycolytic-flux/FBP levels. To this end, the development of more quantitative approaches, such as live-imaging of tagged enzymes and the development of metabolite biosensors, are needed.”

SUGGESTED ADDITIONAL EXPERIMENTS related to point #3:

*3a- Analysis of subcellular localization of Pfkl and Aldoa by Immunofluorescence. This analysis is not limited by the amount of biological material available, so it could be applied to different experimental conditions. *

18. We addressed this point in our response #15.

*3b- Subcellular distribution of Pfkl and Aldoa in explants exposed to different exogenous glucose concentrations. As this involves wild type embryos, it can be done following similar protocols as in figures 6 and 7 of the manuscript. *

19. We addressed this point in our response #16.

4- The results from the work presented in this manuscript would indirectly indicate a negative relationship between glycolysis and somitogenesis. This contrasts with previous reports indicating the essential role of aerobic glycolysis for the same process. There is no explanation for this apparent (and important) contradiction (the authors only comment the discrepancy between the data provided in this paper and previous reports in what concerns the relationship between glycolysis and Wnt signalling, although they also do not provide an explanation).

19. We cannot resolve this discrepancy, but now offer a more detailed discussion, also based on the additional data we obtained.

First, it is important to point out that we have performed additional experiments to substantiate this part of the work, i.e. a transcriptome analysis with control and cytoPfkfb3 explants cultured in 10 mM glucose. We decided to focus on an early time point, i.e. three-hour after incubation, in order to increase the chance to score the primary response of PSM cells upon changes in glycolytic flux. In addition, our nanostring data in Figure S3 shows that glucose titration can change the expression levels of some Wnt-targets in both directions, i.e. decreasing glucose upregulates their expressions while increasing glucose downregulates their expressions. Again, this analysis was done at short time-scales to score the immediate effect.

One possible explanation regarding the difference to Oginuma et al. could indeed be the late time point of analysis in their study, i.e. 16-hour after culture. This difference in sampling time, i.e. 3-hour vs. 16-hour after culture, is of particular importance given the dynamic nature of metabolic and signaling responses.

We have added a sentence to explain this point in more detail (line 608-617):

“This discrepancy could relate to the time point of analysis: while Oginuma et al. mainly focused on analyzing samples 16-hour after metabolic changes, we chose to score the effects of altered glycolytic flux/FBP levels already after a three-hour incubation, with the goal to capture the primary response of PSM cells. Whether the difference in sampling time underlies the observed difference is yet unknown, but both studies highlight that Wnt signaling is responsive to glycolytic flux, supporting a tight link between metabolism and PSM development.”

Minor comments.

*It was not specified the tissue used for the Western blot analyses (was it the PSM alone, the whole tails including somites, etc). This is of relevance to comment #3. *

20. PSM explants without somites were cultured for one/three-hour and were subjected to subcellular protein fractionation. This information is now included in the revised method section.

Reviewer #2 (Significance (Required)):

-The work described in this manuscript identifies FBP as a sentinel metabolite for the glycolytic flux. This, itself has the potential to be important for different processes in which differences in glycolysis makes a difference, although I do not think that this will be relevant for the developmental process on which the authors focused their study (see major comments #1 and 2). Indeed, the lethality of global transgenic cytoPfkfb3 expression (although it was not analyzed if it was during development of in postnatal stages, or the cause of this lethality) but with very minor effects on somitogenesis in vivo supports this conclusion.

21. Please see our detailed comments also based on the newly added in vivo experiments done with the Akita diabetic mouse model in our responses #9–14.

*- The potential moonlighting activity of Pfk (connected with specific subcellular localization), is an interesting idea but so far does not go beyond pure speculation. This is prone to the typical double edged effect of stimulating research in that direction but also the potential negative effect of being taken for granted without rigorous proof. *

22: We have added a statement to highlight the nature of this finding and the requirement for follow up studies both in this and other contexts. Please refer to our response #15 for the details.

• The importance of metabolism in general and glycolysis in particular for somitogenesis and axial extension has been recently reported (the relevant papers are cited in the manuscript) and therefore the work described in this manuscript extends those studies. Also, the recent observations that metabolic process can influence cell activity beyond their participation on the classical pathways in which they are involved, including processes apparently as distant as epigenetic regulation of gene activity (see for instance Tarazona and Pourquie, 2020, Dev Cell 54, 282-292), is opening new perspectives to the study of the influence of metabolism on physiological and pathological processes (championed by cancer and immunological response). It also provides a link between control mechanisms across large scale phylogeny, from procaryotes to eukaryotes.

-In principle, the potential audience for this work could be wide, as the interest in understanding the involvement of metabolism in the regulation of physiological and pathological processes has been growing over the last years. However, the lack of proven mechanism for the activity of FBP might restrict the real general impact of this work. In this regard, the suggestion that it might control some type of still unknown moonlighting activity of Pfk is so far totally speculative.

• I am a developmental biologist with strong focus on mechanisms of somitogenesis and axial extension in vertebrate embryos. There is no part of this work for which I do not feel competent to evaluate.

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

*Summary - *

*In the present manuscript, Miyazawa and colleagues explore the role of glycolytic flux on embryonic development by using presomitic mesoderm (PSM) patterning as a model. *

*First, the authors examined the steady-state levels of central carbon metabolism metabolites in PSM explants. Explants were cultured in various concentrations of glucose and subjected to gas chromatography mass spectrometry (GC-MS). These experiments allowed the identification of metabolites (such as lactate, 3PG, and FBP) that exhibit a linear correlation with glucose levels and can therefore serve as sentinel metabolites for glycolytic flux in PSM cells. Among the metabolites identified, fructose 1,6-bisphosphate (FBP) showed the strongest linear correlation with glucose levels and was used to inform the design of subsequent experiments. *

*Second, to elucidate the functional role of FBP on PSM patterning, the authors supplement the media used to culture PSM explants with various concentrations of FBP and: *

*- analyze the dynamics of Notch signaling (a critical player in mesoderm segmentation during embryogenesis) using real-time imaging of the LuVeLu reporter; *

*- assess gene expression patterns using in situ hybridization of candidate genes. *

*The authors find that supplementation with FBP, but not F6P or 3PG, impairs mesoderm segmentation and disrupts the activity of the segmentation clock in the posterior PSM. Furthermore, FBP supplementation led to the reduced expression of FGF- and WNT-target genes Dusp4 and Msgn, respectively. *

*Third, the authors generate a conditional cytoPfkfb3 transgenic mouse line in which a cytoplasmic form of the Pfkfb3 enzyme is overexpressed. Pfkfb3 can promote glycolysis, and more importantly, leads to increased levels of FBP in a glucose-dependent manner. The authors find that cytoPfkfb3 transgenic PSM explants contain higher levels of FBP and secrete lactate at higher levels when compared to control explants. Importantly, cytoPfkfb3 transgenic PSM explants exhibit impaired somite formation and reduced expression of Msgn (but not Dusp4) in a glucose-dependent manner when compared to control explants. *

Finally, the authors investigate changes in protein subcellular localization in their pharmacological and genetic models of FBP-driven glycolytic flux activation. This was prompted by previous reports on the changes in subcellular localization of glycolytic enzymes (Hu et al., 2016). To this end, the authors perform proteome-wide cell-fractionation analyses in drug-treated and cytoPfkfb3 transgenic PSM explants and find that certain glycolytic proteins exhibit altered subcellular localization in both cases (albeit in different fractions).

*Major concerns: *

*- (Re: Results from Fig. 2 and Fig. S1.) *

*o Given that FBP levels are highly correlated with extracellular glucose levels (which impact glycolytic flux )(TeSlaa and Teitell, 2014) the authors should elaborate on why progressive increase in extracellular glucose does not affect PSM patterning, in the same way that increasing FBP levels does. This is especially important given the claim that FBP is a sentinel metabolite of glycolytic flux. *

23. This important point was also addressed by the reviewer 2, so please see our responses that are also listed under #9, #10, #14 (attached below).

*(Our response #9) *

*We agree with the reviewer and to directly address this central point, we have performed an extended, additional experiment, collecting 375 embryos to quantify FBP levels under five conditions with three biological replicates. *

*There are two major results that we highlight here: First, we found that addition of F6P did not lead to increased FBP levels compared to control samples cultured in 10 mM glucose, which is in stark contrast to cytoPfkfb3 embryos cultured in 10 mM glucose (revised Figure 3E). Second, while increasing glucose concentration is mirrored by elevated FBP levels as we reported, we find clear evidence of saturation above a concentration of 10mM glucose: increasing glucose to 25mM does not increase FBP levels further (revised Figure 3E). *

This saturation effect seen in glucose titration, but also the absence of elevated FBP upon F6P addition, might be expected outcomes because, as also the reviewer 1 pointed out in the response, Pfk is commonly considered to be a rate-limiting enzyme in the glycolytic pathway. We now have the direct experimental data supporting this hypothesis and thank the reviewers to have initiated this additional (very involved..) experiment.

*This new data allows us to conclude more firmly on the correlation between FBP levels and phenotype: at high FBP levels, which are seen in cytoPfkfb3 samples, we observe PSM patterning defects. These high levels are not reached even at 25mM glucose or upon F6P addition, due to the saturation at the level of PFK enzymatic step. Hence, while glucose titration does elevate FBP significantly until this saturation, FBP levels are not as high as in cytoPfkfb3 samples. As a correlative finding, we see that only those conditions with very high FBP levels, or the direct addition of high levels of FBP, cause the arrest of segmentation clock activity. At moderately elevated FBP levels, observed in control explants with high glucose or in cytoPfkfb3 explants with low glucose, clock activity continues and we find a quantitative effect at the level of gene expression, i.e. Wnt signaling target downregulation (Figure 5A, S3). *

The new data has been included in the revised manuscript and the text has been adjusted accordingly:

- (Result Part, line 245–254) "Consistently, we found that cytoPfkfb3 overexpression lifted the upper limit of FBP levels in PSM cells (Figure 3E, S4B, S4C). In control explants, FBP levels did not increase further when glucose concentration was increased from 10 mM to 25 mM. It was also the case when control explants were cultured in 20 mM of F6P (Figure 3E). These results indicate that the Pfk reaction carries a (rate-)limiting role for glycolytic flux and FBP levels, and that cytoPfkfb3 overexpression hinders the flux-regulation function of Pfk."

- (Discussion Part, line 551–573) “Our findings suggest that flux-regulation at the level of Pfk is critical to keep FBP steady state levels within a range compatible with proper PSM patterning and segmentation. In agreement with such a rate-limiting function for Pfk, we found in glucose titration experiments that FBP levels saturated and did not further increase at glucose levels above 10 mM (Figure 3E). Along similar lines, the supplementation of high concentrations of the Pfk substrate F6P did not result in a significant increase of FBP levels, again compatible with a rate-limiting function at the level of Pfk (Figure 3E). The upper limit of glycolytic flux and FBP levels can be experimentally increased by cytoPfkfb3 overexpression (Figure 3B, 3E). We interpret the data as evidence that cytoPfkfb3 overexpression compromises the flux-control function of Pfk and hence much higher FBP (and secreted lactate) levels are reached. Such a drastic increase in glycolytic flux and FBP levels correlates with a severe PSM patterning phenotype (Figure 4), which resembles the phenotype induced by supplementation of high dose of FBP (Figure 2). Our results in mouse embryos hence provides evidence that flux regulation by Pfk, an evolutionary conserved role present from bacteria to humans, serves to maintain FBP levels below a critical threshold.”

(Our response #10)

*#10. First, we would like to clarify that while indeed glycolytic activity is graded along the PSM, as other and we reported previously (reported in Bulusu et al., 2017 and Oginuma et al., 2017), the baseline expression of the entire glycolytic machinery (from glucose transport to lactate production) is very high, in all PSM cells. Hence, we see that cells all along the entire PSM have very active glycolysis, the posterior PSM being even more active. *

*For this and related reasons, our interpretation about the difference seen between glucose titration/F6P addition on one side, and FBP addition/cytoPfkfb3 addition on the other side, is based on the role of Pfk in controlling either flux levels or dynamics in all PSM cells. *

Hence, while we agree that we generate experimental conditions that allow FBP levels to surpass those found in control embryos, we would like to highlight the fact that even moderate changes in flux does result in very robust functional consequences on gene expression (Figure S3, 5), as we show in this work.

*We can currently not fully address the first point raised, i.e. the role of graded flux/graded metabolite levels, due to the experimental limitations. Such a study requires, for instance, the generation of metabolite biosensor reporter lines in order to be able to monitor these changes dynamically, in space and time. *

(Our response #14)

*In our view, two main conclusions, both in vivo and in vitro, can be drawn based on the result we obtained: *

*First, we find that a moderate increase in glycolytic flux, within the physiological range, leads to a quantitative and consistent change in gene expression, such as downregulation of Wnt target genes (Figure S3, 5). Such a phenotype was the result of either glucose titration or culturing cytoPfkfb3-transgenic embryos in low glucose concentration. *

In these conditions, while overall PSM patterning is qualitatively normal, we do find consistent changes at quantitative level, i.e. gene expression changes, which are also mirrored by a reduced rate of segmentation (Figure 4B). A detailed analysis of the quantitative changes at the level of segmentation clock dynamics is being carried out and will be presented in a dedicated follow up study.

*Second, we find that a very significant increase in FBP levels, i.e. when cytoPfkfb3 transgenic animals are cultured in high glucose conditions or when samples are cultured in high levels of FBP, PSM patterning is qualitatively altered and segmentation clock ceases to oscillate. In this case, we agree that it is not a physiological condition, as such high levels of flux and FBP are not reached in control samples which have intact flux regulation by Pfk. Nevertheless, such an experimental condition can be insightful, as it very clearly reveals the potential link between glycolysis, clock activity, PSM patterning and the Wnt signaling pathway. *

*It is the combination between the moderate and the more severe effects, observed both in vitro, and now also in vivo using the Akita model (see above), that we take as evidence for an intrinsic, physiological link between glycolytic activity, PSM patterning and signaling. *

- (Re: Fig. 2A and Fig. 2B)

*o The authors should be consistent with the glucose concentrations for the experiments where they assess the dynamics of Notch signaling (Figure 2A) and gene expression (Figure 2B) or otherwise elaborate on why different concentrations are used for these assays. *

24: We agree that ideally the experimental parameters should be as consistent as possible. In regards to the control glucose concentration used in this study, both 0.5 mM and 2.0 mM glucose were used. It reflects that over the years, minor adjustments in the experimental protocol were made, i.e. we now use 2.0 mM glucose as standard setting for all experiments, while previously, 0.5 mM glucose was used (see Bulusu et al., 2017). This change is based on the observation of a slightly improved culture outcome, in terms of reporter gene expression. We have confirmed that the developmental outcome and also effects seen upon addition of FBP are consistent at 0.5 mM and at 2.0 mM glucose. We made a note in the methods section to explain this point (line 1082-1084):

“Basal culture condition was 0.5 mM glucose at the beginning of this study but was later switched to 2.0 mM glucose which yields a slightly improved reporter gene expression. No major difference was observed in the effects of FBP between these glucose conditions.”

*- (Re: Results from pharmacological and genetic models of increased FBP levels) *

*o The authors state that FBP-driven impairment of mesoderm segmentation is most pronounced in the undifferentiated PSM cells (in the posterior-most end of the explants) and is, therefore, unlikely to be due to a toxic effect that might otherwise affect the whole explant. While this is a reasonable assumption, it does not discount the possibility that the spatial specificity of the effect of FBP could be driven primarily by increased cell death in the posterior end of the explant. Thus, the authors should test whether cell death underlies the mesoderm patterning defects seen in PSM explants subjected to increased FBP levels. *

25. We have performed immunostaining of active caspase-3 in explants cultured for three-hour in medium containing 0.5 mM glucose and 20 mM FBP and found no difference between control and FBP-treated explants (please refer to the Figure R2 below). This qualitative result does not indicate a major effect via cell death in the tail bud region (i.e. posterior PSM) as the underlying reason for the observed phenotype. We included the new data in the revised Figure S2C and adjusted the text accordingly.

Figure R3. Immunostaining of active caspase-3 in PSM explants. Explants were cultured for three hours in the presence or absence of 20 mM of FBP. Neural tubes were outlined by white dotted lines.

*- (Re: Gene expression experiments/analyses) *

*o This study would benefit greatly from transcriptomic analysis of wt and cytoPfkfb3 transgenic PSM explants (and/or transcriptomic characterization of FBP-treated vs. control PSM explants). The candidate approach used to assess gene expression (through in situ hybridization) may not be sufficient to conclude that cytoPfkfb3 over-expression leads to the downregulation of Wnt signaling (a claim the authors make at the beginning of the manuscript). *

26. We fully agree with the reviewer’s comment. We have now performed RNA-sequencing (RNAseq) analysis using control and cytoPfkfb3 explants cultured in 10 mM glucose, importantly after three hours of incubation in order to score early effects at transcriptome level (please refer to Figure R4).

We found clear evidence that many Wnt-target genes (i.e. Axin2, Cdx4, Dact1, Dkk1, Mixl1, Msgn1, Sp5, Sp8, T) were significantly downregulated in cytoPfkfb3 explants, supporting the conclusion that Wnt signaling activity is downregulated in cytoPfkfb3 explants under high glucose condition.

Furthermore, in order to examine similarities between the effects of cytoPfkfb3 overexpression and FBP supplementation, we also performed RNAseq analysis with explants treated with high dose of FBP or F6P. FBP supplementation resulted in downregulation of Wnt target gene expression (i.e. Dact1, Dkk1, Mixl1, Lef1, Sp5, T, Tbx6), mirroring the effects seen in cytoPfkfb3 samples. Such a response was not detected in F6P-treated explants.

Combined, these new data significantly strengthen our conclusion that an increase in glycolytic flux and FBP levels leads to downregulation of Wnt signaling activity. The new data is now included in the revised Figure 5C–E and adjusted the texts accordingly.

Figure R4. Transcriptome analysis of control (Ctrl) and cytoPfkfb3 (TG) PSM explants. PSM explants were cultured for three hours under different culture conditions. (A) Effects of cytoPfkfb3 overexpression on gene expression under 10 mM glucose condition. (B, C) Effects of 20 mM FBP (B) or F6P (C) on gene expression under 2.0 mM glucose condition. Wnt-target genes that were significantly downregulated in cytoPfkfb3 or FBP/F6P-treated explants are highlighted in blue.

*- (Re: Results related to the neural tube closure defects in cytoPfkfb3 transgenic embryos) *

*o The section of the manuscript describing the neural tube closure defects in cytoPfkfb3 transgenic embryos is superficial, lacks detail, and distracts from the focus of the study. Perhaps the data and text on neural tube closure defects should be included as supplemental information. *

27: We agree with the reviewer that in the previous version, this data appeared isolated. It also connects with the point raised by the reviewer 2 about the in vivo significance of our findings. To address both these points, we have now performed additional in vivo experiments using a diabetic mouse model (Akita) to directly test the in vivo consequence of cytoPfkfb3, which interestingly links to the previous findings of neural tube defects. Please see our response #13 for the details (attached below):

(Our response #13)

*First of all, we would like to emphasize that the phenotype seen in cytoPfkfb3 embryos, i.e. the reduction of segmentation and downregulation of Wnt-target gene expression, occurs in a glucose dose dependent manner (Figure 4B and 5A). Hence, it is not the overexpression of cytoPfkfb3 per se that can account for the effects seen. But rather, increased glycolytic flux caused by the combination of transgene expression with high glucose results in functional consequences. *

In addition, ‘other possible effects’ that the reviewer is referring to should be evident in all transgenic embryos, irrespective of glucose dose. To the contrary, transgenic embryos cultured in low glucose conditions appear unaltered to control embryos.

*Second, we agree that we need to distinguish between strong phenotypes, visible at the level of clock arrest, and milder phenotypes, visible at the level of quantitative gene expression changes. It is important to note that the moderate phenotype, i.e. the quantitative gene expression changes seen in posterior PSM, are seen upon the addition of FBP at moderate levels and upon in glucose titration within the physiological concentration range, as well as in cytoPfkfb3 embryos. We take this as evidence that the effects seen in cytoPfkfb3 transgenic embryos reflect a common response also seen under physiological conditions. *

*To extend this argument to the in vivo setting, we have performed additional experiments using a genetic mouse model for diabetes. As shown in our previous submission, cytoPfkfb3 transgenic animals do not exhibit a drastic in vivo phenotype when dissected at embryonic day 10.5. One interpretation of this finding is that since the cytoPfkfb3 phenotype is glucose and flux-dependent, the in vivo flux is low, reflecting low glucose concentrations described in vivo. To test the effect of increased flux in cytoPfkfb3 embryos in vivo, we therefore crossed the transgenic mice into a diabetic model called Akita, in which a point mutation in the Insulin2 gene causes high maternal glucose levels (Yoshioka et al., 1997; Wang et al., 1999). Using this experimental setup, we tested whether transgenic embryos in Akita diabetic females would manifest in vivo phenotypes. *

Indeed, we found that cytoPfkfb3 transgenic embryos developing in Akita diabetic females showed significantly increased cases of neural tube closure defects (50% of cytoPfkfb3 embryos) and developmental delay (control: 38 somites vs. cytoPfkfb3: 34 somites at E10.5), defects not seen in transgenic cytoPfkfb3 embryos from control females (please refer to Figure R2 below). This dependency of the in vivo phenotype on maternal glucose conditions again highlights that the defects observed in cytoPfkfb3 embryos are not due to the expression of cytoPfkfb3 per se, but are rather directly linked to increased/unregulated glycolytic flux.

We included the new in vivo data in the revised Figure S5D-E and modified the text accordingly.

*Figure R2. In vivo phenotype of cytoPfkfb3 embryos grown in diabetic Akita females. (A) The number of somites in control (Ctrl) and cytoPfkfb3 (Tg) E10.5 embryos grown in diabetic Akita females. (B) In situ hybridization of Msgn, Uncx4.1, and Shh mRNAs in Ctrl and Tg E10.5 embryos grown in diabetic Akita females (ss, somite stage; scale bar, 500 µm). *

• (Re: Conclusions of the study)

o A previous study by Oginuma et al., 2020 provided strong evidence for a mechanism underlying the positive regulation of Wnt signaling by glycolysis (initiated by the elevation of intracellular pH) in the chick embryo tailbud. As mentioned in the discussion, the results of the present study are not consistent with this mode - and this contradiction is not sufficiently resolved. This is a concern, given that the evidence that cytoPfkfb3 inhibits Wnt signaling is sparse (see above).

28: This important point was also raised by the reviewer 2, please see our response as listed under #19 (attached below).

(Our response #19)

*We cannot resolve this discrepancy, but now offer a more detailed discussion, also based on the additional data we obtained. *

*First, it is important to point out that we have performed additional experiments to substantiate this part of the work, i.e. a transcriptome analysis with control and cytoPfkfb3 explants cultured in 10 mM glucose. We decided to focus on an early time point, i.e. three-hour after incubation, in order to increase the chance to score the primary response of PSM cells upon changes in glycolytic flux. In addition, our nanostring data in Figure S3 shows that glucose titration can change the expression levels of some Wnt-targets in both directions, i.e. decreasing glucose upregulates their expressions while increasing glucose downregulates their expressions. Again, this analysis was done at short time-scales to score the immediate effect. *

*One possible explanation regarding the difference to Oginuma et al. could indeed be the late time point of analysis in their study, i.e. 16-hour after culture. This difference in sampling time, i.e. 3-hour vs. 16-hour after culture, is of particular importance given the dynamic nature of metabolic and signaling responses. *

We have added a sentence to explain this point in more detail (line 608-617):

“This discrepancy could relate to the time point of analysis: while Oginuma et al. mainly focused on analyzing samples 16-hour after metabolic changes, we chose to score the effects of altered glycolytic flux/FBP levels already after a three-hour incubation, with the goal to capture the primary response of PSM cells. Whether the difference in sampling time underlies the observed difference is yet unknown, but both studies highlight that Wnt signaling is responsive to glycolytic flux, supporting a tight link between metabolism and PSM development.”

*o Another discrepancy lies in the lack of an observable phenotype when culturing mouse PSM explants at very low glucose concentrations (e.g., 0.5 mM in Fig. 2A). Oginuma et al. observed clear disruptions to embryonic elongation and somite formation at a glucose concentration equal to 0.83 mM. Would this be due to species-specific mechanisms? Furthermore, while the authors focus on sentinel metabolites (such as FBP), experiments involving direct manipulation in glycolysis could resolve some of these inconsistencies. *

29: Indeed species specific differences in the requirement for glucose are to be expected. Our extensive analysis shows that at 0.5mM glucose, segmentation and elongation proceeds (Bulusu et al., 2017).

Regarding the second point, we have outlined several strategies to directly perturb glycolysis, i.e. glucose titration (mirrored by increase in lactate secretion) and by genetic targeting of the rate-limiting enzyme, Pfk. Glucose titration in wild-type embryos corresponds to the experiment the reviewer suggested, and we again found that higher glucose (i.e. higher flux) leads to down regulation of several Wnt-target genes (Figure S3). Of note, also in cytoPfkfb3 explants the effects are glucose-dose dependent (again mirrored by increase of lactate secretion), clearly indicating that we successfully and directly controlled glycolysis.

*References - *

1. • Hu, Hai, et al. "Phosphoinositide 3-kinase regulates glycolysis through mobilization of aldolase from the actin cytoskeleton." Cell 164.3 (2016): 433-446. *
2. • TeSlaa, Tara, and Michael A. Teitell. "Techniques to monitor glycolysis." Methods in enzymology 542 (2014): 91-114. *
3. • Oginuma, Masayuki, et al. "Intracellular pH controls WNT downstream of glycolysis in amniote embryos." Nature584.7819 (2020): 98-101. * *Reviewer #3 (Significance (Required)): *

The experimental results reported in this study enhance our understanding of how cellular metabolic states regulate cellular behaviors during embryonic development. The study provides insight into how PSM elongation is controlled by morphogenetic mechanisms that are modulated by glycolytic flux. One of the strengths of the study is the use of an interdisciplinary approach that includes GC-MS, in vivo imaging and mouse transgenic lines. It should be noted that some of the conclusions of the study diverge from previous papers that examine the role of metabolism in developmental patterning (e.g., Oginuma et al., 2020).

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

Learn more at Review Commons

---

Referee #3

Evidence, reproducibility and clarity

Summary

In the present manuscript, Miyazawa and colleagues explore the role of glycolytic flux on embryonic development by using presomitic mesoderm (PSM) patterning as a model.

First, the authors examined the steady-state levels of central carbon metabolism metabolites in PSM explants. Explants were cultured in various concentrations of glucose and subjected to gas chromatography mass spectrometry (GC-MS). These experiments allowed the identification of metabolites (such as lactate, 3PG, and FBP) that exhibit a linear correlation with glucose levels and can therefore serve as sentinel metabolites for glycolytic flux in PSM cells. Among the metabolites identified, fructose 1,6-bisphosphate (FBP) showed the strongest linear correlation with glucose levels and was used to inform the design of subsequent experiments.

Second, to elucidate the functional role of FBP on PSM patterning, the authors supplement the media used to culture PSM explants with various concentrations of FBP and: - analyze the dynamics of Notch signaling (a critical player in mesoderm segmentation during embryogenesis) using real-time imaging of the LuVeLu reporter; - assess gene expression patterns using in situ hybridization of candidate genes. The authors find that supplementation with FBP, but not F6P or 3PG, impairs mesoderm segmentation and disrupts the activity of the segmentation clock in the posterior PSM. Furthermore, FBP supplementation led to the reduced expression of FGF- and WNT-target genes Dusp4 and Msgn, respectively.

Third, the authors generate a conditional cytoPfkfb3 transgenic mouse line in which a cytoplasmic form of the Pfkfb3 enzyme is overexpressed. Pfkfb3 can promote glycolysis, and more importantly, leads to increased levels of FBP in a glucose-dependent manner. The authors find that cytoPfkfb3 transgenic PSM explants contain higher levels of FBP and secrete lactate at higher levels when compared to control explants. Importantly, cytoPfkfb3 transgenic PSM explants exhibit impaired somite formation and reduced expression of Msgn (but not Dusp4) in a glucose-dependent manner when compared to control explants.

Finally, the authors investigate changes in protein subcellular localization in their pharmacological and genetic models of FBP-driven glycolytic flux activation. This was prompted by previous reports on the changes in subcellular localization of glycolytic enzymes (Hu et al., 2016). To this end, the authors perform proteome-wide cell-fractionation analyses in drug-treated and cytoPfkfb3 transgenic PSM explants and find that certain glycolytic proteins exhibit altered subcellular localization in both cases (albeit in different fractions).

Major concerns:

1. (Re: Results from Fig. 2 and Fig. S1.)
   • Given that FBP levels are highly correlated with extracellular glucose levels (which impact glycolytic flux )(TeSlaa and Teitell, 2014) the authors should elaborate on why progressive increase in extracellular glucose does not affect PSM patterning, in the same way that increasing FBP levels does. This is especially important given the claim that FBP is a sentinel metabolite of glycolytic flux.
2. (Re: Fig. 2A and Fig. 2B)
   • The authors should be consistent with the glucose concentrations for the experiments where they assess the dynamics of Notch signaling (Figure 2A) and gene expression (Figure 2B) or otherwise elaborate on why different concentrations are used for these assays.
3. (Re: Results from pharmacological and genetic models of increased FBP levels)
   • The authors state that FBP-driven impairment of mesoderm segmentation is most pronounced in the undifferentiated PSM cells (in the posterior-most end of the explants) and is, therefore, unlikely to be due to a toxic effect that might otherwise affect the whole explant. While this is a reasonable assumption, it does not discount the possibility that the spatial specificity of the effect of FBP could be driven primarily by increased cell death in the posterior end of the explant. Thus, the authors should test whether cell death underlies the mesoderm patterning defects seen in PSM explants subjected to increased FBP levels.
4. (Re: Gene expression experiments/analyses)
   • This study would benefit greatly from transcriptomic analysis of wt and cytoPfkfb3 transgenic PSM explants (and/or transcriptomic characterization of FBP-treated vs. control PSM explants). The candidate approach used to assess gene expression (through in situ hybridization) may not be sufficient to conclude that cytoPfkfb3 over-expression leads to the downregulation of Wnt signaling (a claim the authors make at the beginning of the manuscript).
5. (Re: Results related to the neural tube closure defects in cytoPfkfb3 transgenic embryos)
   • The section of the manuscript describing the neural tube closure defects in cytoPfkfb3 transgenic embryos is superficial, lacks detail, and distracts from the focus of the study. Perhaps the data and text on neural tube closure defects should be included as supplemental information.
6. (Re: Conclusions of the study)
   • A previous study by Oginuma et al., 2020 provided strong evidence for a mechanism underlying the positive regulation of Wnt signaling by glycolysis (initiated by the elevation of intracellular pH) in the chick embryo tailbud. As mentioned in the discussion, the results of the present study are not consistent with this mode - and this contradiction is not sufficiently resolved. This is a concern, given that the evidence that cytoPfkfb3 inhibits Wnt signaling is sparse (see above).
7. Another discrepancy lies in the lack of an observable phenotype when culturing mouse PSM explants at very low glucose concentrations (e.g., 0.5 mM in Fig. 2A). Oginuma et al. observed clear disruptions to embryonic elongation and somite formation at a glucose concentration equal to 0.83 mM. Would this be due to species-specific mechanisms? Furthermore, while the authors focus on sentinel metabolites (such as FBP), experiments involving direct manipulation in glycolysis could resolve some of these inconsistencies.

References

1. Hu, Hai, et al. "Phosphoinositide 3-kinase regulates glycolysis through mobilization of aldolase from the actin cytoskeleton." Cell 164.3 (2016): 433-446.
2. TeSlaa, Tara, and Michael A. Teitell. "Techniques to monitor glycolysis." Methods in enzymology 542 (2014): 91-114.
3. Oginuma, Masayuki, et al. "Intracellular pH controls WNT downstream of glycolysis in amniote embryos." Nature584.7819 (2020): 98-101.

Significance

The experimental results reported in this study enhance our understanding of how cellular metabolic states regulate cellular behaviors during embryonic development. The study provides insight into how PSM elongation is controlled by morphogenetic mechanisms that are modulated by glycolytic flux. One of the strengths of the study is the use of an interdisciplinary approach that includes GC-MS, in vivo imaging and mouse transgenic lines. It should be noted that some of the conclusions of the study diverge from previous papers that examine the role of metabolism in developmental patterning (e.g., Oginuma et al., 2020).

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 work described in this paper first searches for potential sentinel metabolites of glycolytic flux, focusing on the process of somitogenesis during mouse embryonic development. By measuring the levels of different metabolites in the presomitic mesoderm (PSM) of E10.5 mouse embryos cultured in the presence of three different glucose concentrations, the authors identify 14 metabolites whose concentration rises with increasing glucose concentration in the culture medium. Among them, they selected fructose 1,6-bisphosphate (FBP) for further analyses, as it showed the highest linear correlation with extracellular glucose concentrations. They then show that addition of FBP to the incubation medium of cultured embryo tails interfere with somitogenesis and tail extension in a concentration-dependent fashion. In addition, they show that this effect is exacerbated when extracellular glucose levels are increased. By analyzing specific targets of Wnt and Fgf signaling, the authors also show that addition of FBP down-regulates both signaling pathways in the PSM. They then use a genetic trick (ubiquitous overexpression of cytoPfkfb3) to increase FBP levels by allosteric activation of Pfk (the enzyme that produces FBP) in developing embryos. When tails from these transgenic embryos were cultured in vitro and exposed to various glucose concentrations somitogenesis was affected in a way resembling the effects of FBP on cultured tails from wild type embryos. The authors then go on to determine the subcellular localization of different proteins in tails incubated in the presence of various FBP concentrations to identify that some enzymes involved in the glycolytic pathway (and they specifically focus on Pfkl and Aldoa) are excluded from nuclear fractions at high FBP concentrations. The authors conclude that FBP functions as a flux-signaling metabolite connecting glycolysis and PSM patterning, potentially through modulating subcellular protein localization.

Major comments

I think that in general the work described in this manuscript has been performed to the highest technical standards. However, I do not think that I can agree with the authors' conclusions (that FBP connects glycolysis with PSM patterning and that subcellular localization of glycolytic enzymes play a role in this process), which in my opinion go way beyond what can be proven by the data provided.

1. Explants incubated with external glucose concentrations up to 25 mM have no obvious defects on somitogenesis or on the segmentation clock as determined by LuVeLu cycling activity. Under these conditions, explants are expected to contain very high FBP levels if this metabolite keeps its linear relationship with external glucose (in this work it was not measured beyond 10 mM glucose in the medium, where FBP concentration was already very high). This contrasts with the phenotypes observed upon exogenous supplementation of FBP, which affects somitogenesis already at 2 mM glucose. These latter results are at odds not only with the lack of phenotypic alterations under high glucose conditions, but also with the observation that exogenous addition of fructose 6-phosphate (F6P), the substrate of Pfk enzymes to generate FBP, does not alter somitogenesis. The authors take the absence of effects by incubation with F6P as a control of the specificity of FBP. However, as F6P is the natural substrate of Pfk, it is possible that supplementation of F6P also leads to an increase of FBP but in a way closer to a physiological condition. Therefore, I find it essential to determine FBP levels in tails incubated in the presence of increasing amounts of F6P, as if it increases FBP levels, similarly to what the authors described for the tails incubated with increasing glucose concentrations, it will have important implications to the interpretation of the work presented in this manuscript. The main difference between the experiments involving FBP supplementation and those involving high glucose concentrations or exogenous F6P addition is that in the later two cases increase in FBP would be restricted to the tissue(s) expressing Pfk, whereas upon FBP supplementation this metabolite would hit any tissue, regardless of whether or not it would ever be physiologically exposed to this molecule. In the case of the PSM, this might be relevant because it has been shown that there is a gradient of glycolysis, being high at the caudal tip and becoming lower at more anterior regions of the PSM, most likely mirroring the distribution of Pfk activity. Exogenous administration of FBP would flatten the gradient, which could lead to alterations in PSM patterning, whereas glucose (and eventually F6P) would not as they would increase FBP locally in the area where it is normally activated, keeping the natural gradient. On the basis of these arguments, to which extent does FBP connect glycolysis and somitogenesis under physiological conditions?

ESSENTIAL ADDITIONAL EXPERIMENT related to point #1: Measure FBP from PSM explants incubated under various exogenous concentrations of F6P.

ANOTHER EXPERIMENT THAT COULD BE INFORMATIVE: measure FBP levels in PSM incubated under different glucose concentrations but instead of using the whole PSM together, dividing the PSM in posterior, medium and anterior parts (similarly to what was done in Oginuma et al, 2017, reference in the manuscript) to see if there is a gradient in FBP activation. 2. A similar argument could be presented for the results with the cytoPfkfb3 transgenics, as they are based on global artificial overactivation of Pfk, in addition to other possible effects of the ectopic activity of cytoPfkfb3, which were not controlled. Also, while the phenotypic alterations in the PSM in vitro, most particularly in the experiments involving incubation of the tails, are rather strong, the reported effects on somitogenesis in vivo are minor, also questioning the contribution of the in vitro conditions to the final phenotypic effects observed throughout the manuscript.

In conclusion, combining the arguments in the two previous comments, to which extent the results from the addition of FBP or from the transgenic activation of Pfk are not artefactual phenotypes without real physiological relevance? 3. The authors seem to give a strong functional meaning to the absence of Pfkl and Aldoa from the nuclear fraction in tails incubated with exogenous FBP, suggesting a "moonlighting" function of these enzymes under FBP regulation. In addition to the purely speculative nature of this interpretation (there is no proof for such activity or even an attempt to test it), the data provided is also difficult to interpret for various reasons. The protein levels in nuclear fractions are clearly much lower than those in the cytoplasm (this is best seen in the blots of Figure 6D). Does this represent similar subcellular distribution of these enzymes throughout the tissue or the different levels result from the presence of the enzymes in the nucleus of only a subset of the cells? This might be of importance to understand the possible relevance of the subcellular distribution of those enzymes. All the analyses were done on bulk tissue and, therefore, it is not possible to distinguishing between these possibilities. As the authors have antibodies for these enzymes, they could try to perform immunofluorescence analyses, which would provide spatial data.

In addition to this, it would be important to determine Pfkl and Aldoa subcellular localization in explants incubated with different external concentrations of glucose, which in a way reproduces better possible physiological effects (see point 1), to see if under those conditions high FBP also affects subcellular distribution of those enzymes.

SUGGESTED ADDITIONAL EXPERIMENTS related to point #3: 3a- Analysis of subcellular localization of Pfkl and Aldoa by Immunofluorescence. This analysis is not limited by the amount of biological material available, so it could be applied to different experimental conditions.

3b- Subcellular distribution of Pfkl and Aldoa in explants exposed to different exogenous glucose concentrations. As this involves wild type embryos, it can be done following similar protocols as in figures 6 and 7 of the manuscript. 4. The results from the work presented in this manuscript would indirectly indicate a negative relationship between glycolysis and somitogenesis. This contrasts with previous reports indicating the essential role of aerobic glycolysis for the same process. There is no explanation for this apparent (and important) contradiction (the authors only comment the discrepancy between the data provided in this paper and previous reports in what concerns the relationship between glycolysis and Wnt signalling, although they also do not provide an explanation).

Minor comments.

It was not specified the tissue used for the Western blot analyses (was it the PSM alone, the whole tails including somites, etc). This is of relevance to comment #3.

Significance

• The work described in this manuscript identifies FBP as a sentinel metabolite for the glycolytic flux. This, itself has the potential to be important for different processes in which differences in glycolysis makes a difference, although I do not think that this will be relevant for the developmental process on which the authors focused their study (see major comments #1 and 2). Indeed, the lethality of global transgenic cytoPfkfb3 expression (although it was not analyzed if it was during development of in postnatal stages, or the cause of this lethality) but with very minor effects on somitogenesis in vivo supports this conclusion.
• The potential moonlighting activity of Pfk (connected with specific subcellular localization), is an interesting idea but so far does not go beyond pure speculation. This is prone to the typical double edged effect of stimulating research in that direction but also the potential negative effect of being taken for granted without rigorous proof.
• The importance of metabolism in general and glycolysis in particular for somitogenesis and axial extension has been recently reported (the relevant papers are cited in the manuscript) and therefore the work described in this manuscript extends those studies. Also, the recent observations that metabolic process can influence cell activity beyond their participation on the classical pathways in which they are involved, including processes apparently as distant as epigenetic regulation of gene activity (see for instance Tarazona and Pourquie, 2020, Dev Cell 54, 282-292), is opening new perspectives to the study of the influence of metabolism on physiological and pathological processes (championed by cancer and immunological response). It also provides a link between control mechanisms across large scale phylogeny, from procaryotes to eukaryotes.
• In principle, the potential audience for this work could be wide, as the interest in understanding the involvement of metabolism in the regulation of physiological and pathological processes has been growing over the last years. However, the lack of proven mechanism for the activity of FBP might restrict the real general impact of this work. In this regard, the suggestion that it might control some type of still unknown moonlighting activity of Pfk is so far totally speculative.
• I am a developmental biologist with strong focus on mechanisms of somitogenesis and axial extension in vertebrate embryos. There is no part of this work for which I do not feel competent to evaluate.

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 Miyazawa and colleagues addresses a key question: How is changed metabolic activity sensed and to induce changes in developmental programs. In recent years, there is more and more indication that metabolism is not only a dull workhorse synthesizing the building blocks for new cells and providing chemical energy, but that metabolic activity itself has also a regulatory role. How this precisely works is largely unknown and even also unexplored in higher cells. From early insights obtained in microbes, it seems that certain metabolites - possibly reflecting metabolic activity (i.e. flux) - could be metabolic signals that feedback into cellular regulation. The current paper takes this idea now to developmental processes, where the authors found that the glycolytic metabolite fructose-1,6-bisphosphate is a flux-dependent signal that interferes with developmental processes. This is a very exciting finding, as it indicates that this metabolite not only has a regulatory function in microbes but also in mouse during mesoderm development. Answering the question how such a flux-dependent metabolite mechanistically interferes with the developmental processes is an enormously difficult. Compared to other mechanistic studies, where deleting genes, modifying genes, and changing protein expressions will usually do the trick, here, perturbing metabolite levels is extremely challenging, particularly if such perturbations need to be carried out in a way that nothing else is perturbed. Researchers, who are not overly familiar with metabolism, usually underestimate the difficulty with targeted and insightful perturbation of metabolism. To this end, the authors of this paper need to be congratulated for a very well carried out study with very solid data, and excellent control experiments. The authors open up a new path towards understanding how embryo mesoderm development is regulated by metabolic activity. In particular, they show that that glycolytic flux, FBP and important developmental phenotypes as well as protein localization changes are linked. As normal with a complex metabolism-based story as this one, there is always more that could be done. Yet, the results are highly important to be reported now such that the field as a whole can build on these interesting results and to explore the exciting path further that has been opened by the authors. Thus, I strongly recommend publishing these findings: The data generated by the authors are accompanied by the required control experiments. The conclusions drawn are very solid. I do not have any major concerns but just a number of minor suggestions that the authors could consider in a revised version of the manuscript.

Minor:

1. At the end of the introduction, the authors stated their original goal. As it is phrased, it is unclear whether this goal has been obtained or not. They might want to consider replacing the last introductory sentence by a sentence stating what the reader can find in this paper.
2. Data from Fig 3: If you plot the lactate secretion vs the FBP levels of the controls and the overexpression experiment, would the control and the overexpression data lie on one line (maybe if combined with the data shown in Fig 1A)?
3. Maybe the authors could attempt an experiment like the following one: Chose the strongest phenotype observed and test a combination of overexpressing cytoPfkfb3 and reducing extracellular glucose level at the same time?
4. Can the proteomics experiments shown in Fig. 6 be repeated with high and low extracellular glucose? High glucose should yield high FBP levels and one would then expect to see the same as with the experiment where at 2 mM glucose 20 mM extracellular FBP were added. Is this the case?
5. While the authors quantified proteins in different compartments, I was wondering whether they also looked for whole-embryo protein expression changes?
6. Throughout the manuscript, the authors state the glucose levels or cytoPfkfb3 changes the glycolytic flux. While I tend to agree with this, it is important to note that the authors have not directly measured glycolytic flux, but use the amount of accumulated lactate as a proxy. I think it is important to add this disclaimer at important points in the manuscript, such that readers are aware of this point.
7. Another aspect for changing FBP levels could be connected on what was found in yeast, where the FBP levels were found to oscillate with the cell cycle (https://pubmed.ncbi.nlm.nih.gov/31885198/). Could this be connected with the pattern formation here?
8. Line 606: The mentioned review article also covers yeast. As such, maybe the authors should replace the term "bacteria" with "microbes"?

Significance

Referees cross-commenting

As I mentioned in my comment, targeted metabolic perturbations are extremely difficult. Perturbing a metabolite level without at the same time perturbing the flux through this pathways is difficult (of not impossible). Also, the opposite is the case. I am not sure whether experiments as the one suggested by reviewer 2 (comment 1) will really lead to results from which further conclusions can be drawn. Furthermore, there does not need to be a linear correlation between the extracellular glucose concentration and metabolic flux/FBP levels (as my reviewer colleague implies). Thus, I am not sure whether doing this experiment makes sense, or would lead to strengthened conclusions.

Reviewer 2 also states "The lack of proven mechanism for the activity of FBP might restrict the real general impact of this work." I agree that we do not know the downstream targets of FBP, but finding them would likely require many years of additional work. Such work will not be initiated if this paper is not published, and it would be a pity if it would be further delayed. I feel that the evidence is strong enough that FBP has an important role and with this paper published, it will motivate others to look for the downstream targets.

Reviewer 3 makes the point: "Given that FBP levels are highly correlated with extracellular glucose levels (which impact glycolytic flux )(TeSlaa and Teitell, 2014) the authors should elaborate on why progressive increase in extracellular glucose does not affect PSM patterning, in the same way that increasing FBP levels does. " Here, I feel my reviewer colleague might be overlooking that in biochemistry molecular interactions typically reach a saturation at some point. The correlation between extracellular glucose and glycolytic flux has likely only a range where these two measures linearly correlate. Similarily, the correlation between glycolytic flxu and FBP likely also exists only within a certain range, and finally FBP levels and the downstream targets likely also only linearly interact within bounds. Thus, the absence of a correlation at "extremes" does by no mean mean that what the authors propose is incorrect. In fact, it just shows what you expect from biomolecular interactions that there a limits to linear correlations.

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 all reviewers for their input and suggestions.

In the discussion section, the reviewers agreed on four major points which we have addressed as follows:

In vivo validation of ROP1. We have now carried out mouse infections using the PRUΔKU80, PRUΔROP1, and complemented lines, showing that PRUΔROP1 is completely avirulent. This matches our in vivo CRISPR screen and in vitro IFNγ restriction assays results, and confirms that ROP1 is an important T. gondii virulence factor. We highlight the discrepancy with Soldati et al 1995, and suggest that this important role of ROP1 was previously overlooked in T. gondii RH due to the “hypervirulence” of this strain in laboratory mice. Clarify discrepancies in Irgb6 recruitment compared to published data: We revisited our image analysis pipeline for this data and corrected an error in the host cell segmentation step which was causing erroneously high calling of Irgb6 recruitment. The recruitment we now measure is now less variable and is consistent with published data, confirming that ROP1 does not affect Irgb6 recruitment or rhoptry bulb protein secretion. Conduct specific assays to measure host cell death or remove claims about this. We have carried out kinetic propidium iodide uptake assays as suggested by Reviewer #3. These have clarified that there is minimal parasite-induced host cell death at low MOI which cannot explain the host cell loss observed in the high-content imaging restriction assays. At an MOI of 0.3, ROP1 has no detectable effect on host cell death, while at a much higher MOI of 3, ROP1 knockout moderately increases cell death of PRU-infected BMDMs. Since cell death in macrophages is reported to result from exposure of parasite-derived PAMPS to host cytosolic sensors, we suggest that this increased host cell death at high MOI is a secondary effect of vacuole disrustion. Alternatively, these findings raise the interesting possibility of a differential phenotype for ROP1 at low versus high parasite burden. Queries regarding the restriction assays in C1QBPflox/flox/-/- MEFs. We included these data as we found statistically significant differences between the C1QBPflox/flox and C1QBP-/- MEFs that were dependent on the presence of ROP1 in the parasites. However, given the concerns raised by all three reviewers regarding the apparent lack of T. gondii restriction in these cells, we have withdrawn all conclusions relating to the putative role of C1QBP in parasite restriction. The data are included now in a supplementary figure only as a reference for other researchers working on the role of C1QBP in innate immunity who may use these previously published cell lines. We have made additional attempts to explore the link between ROP1 and C1QBP that have been unsuccessful, which are now mentioned in the discussion section. Although the co-immunoprecipitation of C1QBP with ROP1 is interesting, given the putative role of C1QBP in regulating autophagy and innate immune responses, further exploration of this potential interaction will require new tools and substantially more work that is beyond the scope of this manuscript. Further responses to comments raised by individual reviewers and details of further revisions are described below.

Reviewer #1

The authors identified ROP1 as a significant hit from their in vivo screen. However, they have not done any validation experiments using rop1 KO parasites in mice. Previous studies have shown no virulence defect in mice for rop1 KO in the type I background (PMID: 8719248). The result could be different in the type II strain used here, but this needs to be tested and shown.

Soldati et al 1995 (PMID: 8719248) demonstrated that there was no virulence defect associated with ROP1 knockout in the RH strain. However, the RH strain is extremely virulent in most laboratory mice strains, which can mask phenotypes observed in a type II strain. We have now carried out in vivo infections of C57BL/6J mice using the type II Prugniaud strain, and have shown a severe virulence defect for PRUΔROP1 parasites which is rescued by complementation. This matches our in vivo CRISPR screen and in vitro IFNγ restriction assays results, and confirms that ROP1 is a virulence factor in vivo.

The data in Fig. 2 and S3 do support that reduced parasitemia was due to decrease in number of vacuoles rather than their size or host cell death. However, it is important to control for invasion and/or egress differences of rop1 KO parasites in IFN-g activated cells.

Soldati et al 1995 (PMID: 8719248) demonstrated that ROP1 has no role in invasion in HFFs, therefore it would be very surprising if ROP1 were to have a specific IFNγ-dependent role in invasion in macrophages. There is no involvement of rhoptry bulb-localised proteins in the predominant model of invasion, only rhoptry neck and microneme proteins.

“Natural” egress after ~48 h would not occur here as the cells are fixed after 24 h. IFNγ-induced “early” egress has been documented in HFFs, A549s, and macrophages in vivo, and is apparent through increased host cell death/lysis (Niedelman et al 2013 PMID: 24042117, Rinkenberger et al 2021 PMID: 34871166, Tomita et al 2009 PMID: 19846885). We have now carried out prodium iodide uptake assays to more accurately quantify parasite-induced host cell death, and find no differences between strains at an MOI of 0.3, the targeted MOI we use in our restriction assays. At a higher MOI of 3, we find a moderate increase in host cell death in PRUΔROP1-infected BMDMs, which we suggest results from increased exposure of parasite-derived PAMPs to cytosolic sensors (Fisch et al 2019 PMID: 31268602, Zhao et al 2009 PMID: 19197351). Alternatively, this increased host cell death could result from increased rates of early egress, or from direct inhibition of programmed cell death pathways.

It is important and informative to depict absolute parasite/size when making multiple comparisons. For example, the data in Fig. 4E shows C1QBP-/- MEFs can clear both RH and PRU better than the WT. However the authors do not comment on what is the meaning of > 100% parasite numbers in IFN-g treated MEFs with respect to untreated in WT. Since the data are normalized, it is difficult to appreciate what the actual differences are. Additionally, C1QBP-/- MEFs show close to equal survival in control and IFN-g treated condition (approximately 100%). Is it correct to infer that C1QBP has no effect on parasite survival? This should be considered in light of the comment below on colocalisation of C1QBP.

It is standard practise to show IFNγ-dependent restriction as a percentage of unstimulated cells as this reflects the hypothesis being tested and the statistical tests carried out, as for example in Wang et al 2020 PMID: 33067458, Matta et al 2019 PMID: 31413201, Gay et al 2016 PMID: 27503074, Bando et al 2018 PMID: 30283439, Fleckenstein et al 2012 PMID: 22802726. Absolute numbers are included in the supplementary data for further reference.

The >100% survival observed in the C1QBPflox/flox and C1QBP-/- MEFs is puzzling. We conclude that these cell lines have largely lost the ability to restrict T. gondii parasites as a result of the immortalisation process and/or passage history, and what little restriction we observe is at the limit of detection in our assay. MEFs are otherwise known to restrict both RH and PRU parasites (Niedelman et al 2012 PMID: 22761577). We included these data as we found statistically significant differences between the C1QBPflox/flox and C1QBP-/- MEFs which were dependent on the presence of ROP1 in the parasites. However, after the concerns raised by all three reviewers we agree that it is better to not to draw conclusions from these assays given the lack of parasite restriction. We will include these data only in the supplementary figures as a reference for other researchers working on the role of C1QBP in innate immunity who may use these previously published cell lines.

The authors observed good restriction of both RH and PRU in IFN-g activated THP1s without cell death (Fig S1D). It is important to incorporate this information into the main result and discuss their implications in contrast to a previous report from 2019 (Fisch et al, PMID: 31268602).

As mentioned above, we have carried out propidium iodide uptake assays to address questions regarding host cell death more precisely, which are now included as a main figure in the revised manuscript. While Fisch et al measured host cell death during infection of stimulated THP-1s at an MOI of 3, our restriction assays were carried out at a targeted MOI of 0.3. The Howard lab has shown that host cell death is directly proportional to MOI in BMDMs (Zhao et al 2009 PMID: 19197351), and we also find that at an MOI of 0.3 there is little detectable host cell death in BMDMs. While we are not aware of a similar study in THP-1s, it is likely also the case that at low MOIs there is little detectable cell death.

The authors should consider conducting C1QBP functional assays to explore potential roles in parasite survival/growth within host cells. For example, it would be informative to measure the extent of autophagy or transcriptomic profile of the KO to deduce or suggest possible mechanisms of restriction.

All reviewers have raised concerns regarding the restriction assays in the C1QBPflox/flox/C1QBP-/- MEFs, particularly that they do not appear to restrict parasite growth as would be expected. As a result, we have decided to withdraw conclusions from these assays regarding the role of C1QBP. For the same reason, we feel that further functional assays using these cells would be of limited value. As an alternative, we attempted siRNA-mediated knockdown of C1QBP in primary BMDMs using a pool of three commercial siRNAs, but were able to achieve only

The authors state in their text that "C1QBP localised primarily to the mitochondria (Figure S7A) and therefore did not see any co-localisation with ROP1". The authors should discuss in more detail as this finding seems to contradict the interaction studies. Is there any independent evidence to corroborate the interaction studies and show they are not simply an in vitro artifact?

We have now added immunofluorescence images of C1QBP and ROP1 in infected MEFs and HFFs to the main figure and discuss this in further detail. We find that C1QBP primary localises to the mitochondria, therefore there is some overlap with ROP1 signal at the PVM in RH as type I strains recruit mitochondria to the vacuole via MAF1B (Pernas et al 2014 PMID: 24781109, Adomako-Ankomah et al 2016 PMID: 26920761). However, type II strains do not recruit host mitochondria therefore we see little overlap with ROP1 in PRU. Furthermore, the precise localisation of C1QBP is a matter of some debate, such that it is unclear whether ROP1 and C1QBP are topologically able to interact. One study reported that C1QBP is exclusively localised to the mitochondrial matrix and therefore would not be able to interact with ROP1 even when the mitochondria are recruited to the vacuole (Muta et al 1997 PMID: 9305894). Others have asserted that there is an additional cytosolic pool of C1QBP which can be recruited to the outer membrane of the mitochondria, thus allowing interaction with ROP1 immediately following rhoptry secretion into the cytosol or at the PVM (Xu et al 2009 PMID: 19164550). From these IFAs we are therefore unable to draw any firm conclusions.

We attempted proximity biotinylation to validate this potential interaction in cellulo, but C-terminal fusion of TurboID to ROP1 caused mislocalisation of the protein and prevented secretion of ROP1 to the parasitophorous vacuole. Based on the current evidence, we are unable to exclude that the interaction is an in vitro artefact as a result of cell lysis during the immunoprecipitation, and we clearly state this in the discussion. However, given this pulldown data is highly reproducible and technically sound, we believe it is important to include this result in the manuscript.

The authors state in their text that "Enhanced restriction of Δrop1 parasites is primarily mediated through increased vacuole destruction". Their data is more suggestive of growth restriction. The authors should provide more direct data for destruction of the vacuoles or change the wording to indicate it is due to growth restriction.

In the absence of specific results for vacuole destruction, which is technically challenging to determine quantitatively, we concluded by a process of elimination that restriction of ΔROP1 parasites at low MOI mostly likely occurs primarily through vacuole destruction, as we did not see strong evidence for vacuole size reduction and (now added in revision) do not see differences in parasite-induced host cell death. Reviewer #1 agrees in an earlier comment that vacuole destruction is the most likely mechanism. We note some subtle indications of strain- and host species-dependent differences, namely that RHΔROP1 parasites in THP-1 macrophages appear to have reduced vacuole size compared to RHΔUPRT (but not the complemented strain), but in all other cases the most likely mechanism consistent with our data is destruction of vacuoles. We have revised our conclusions to reflect that this is an inference from the data rather than a definitive finding.

The authors should provide high resolution IFA images and decrease their size to an equivalent size of other sub-figures. Empty white space between sub-figures can be minimized. Font size of the figure labels/axis/titles should be matched and increased slightly.

Thank you for these suggestions. Each IFA image is only 2x2 cm so we feel that decreasing their size further would make them difficult to see.

Reviewer #2

The title indicates a positive role for ROP1 in subverting innate immune restriction, but the data indicate that a deficiency in ROP1 causes susceptibility to innate immune restriction. This might seem like a subtle discord, but in the absence of identifying the mechanism of ROP1 subversion of innate immune restriction it seems more appropriate to provide a title that better reflects the findings i.e., that ROP1 deficient parasites are more susceptible to innate immune restriction.

We recognise this point and have changed the title to reflect both this and new results added in revision: “Toxoplasma gondii virulence factor ROP1 reduces parasite susceptibility to murine and human innate immune restriction”

The reference strains and complement strains lack UPRT, but from the available information it appears the KO strains have UPRT. If this is the case, it is necessary to rule out that the presence of UPRT doesn't render parasites more susceptible to IFNg mediated killing by performing additional experiments comparing RH∆ku80 with RH∆ku80∆uprt and Pru∆ku80 with Pru∆ku80∆uprt.

This is correct, the reference and complemented strains lack UPRT while the ROP1 knockout strains have a functional UPRT. For the high-content imaging assay it was necessary to use a reference strain that expressed a fluorophore, as the knockout and complemented strains do, to allow for accurate segmentation of the parasites. We considered it preferable to insert the mCherry fluorophore at the well-established, non-essential UPRT locus rather than integrate it randomly. Complementation at the UPRT locus is widely used in the literature with no impact on virulence phenotypes: Shen et al 2014 (PMID: 24825012) Fig 4F, Wang et al 2020 (PMID: 33067458) Fig 5F, Wang et al 2020 (PMID: 31908049) Fig 5, Fox et al 2019 (PMID: 31266861) Fig 5, Olias et al 2016 (PMID: 27414498) Fig 6. Moreover, the genome-wide CRISPR knockout screen of Wang et al 2020 (PMID: 33067458) has also demonstrated that knockout of UPRT does not affect growth of RH parasites in IFNγ-stimulated vs. naive BMDMs. UPRT has 100% amino acid identity between RH and PRU so no functional differences are expected.

Although it is understandable why the authors chose MEFs to test the role of C1QBP because MEFs were used for the Co-IP, the MEFs do not appear to be responding to IFNg for parasite growth restriction (-/+ IFNg % survival is at or above 100% for WT MEFs). As such, the authors are potentially blind to the role of C1QBP in the context of IFNg restriction. It would be ideal to repeat these experiments using BMDMs from WT and C1QBP KO mice to assess the potential contribution of C1QBP during IFNg restriction of T. gondii growth and survival.

We agree that it would be preferable to use ΔC1QBP BMDMs for these experiments. However, knockout of C1QBP is lethal in mice (Yagi et al 2012, PMID: 22904065), therefore it is not possible to obtain primary ΔC1QBP BMDMs. We also attempted siRNA-mediated knockdown of C1QBP in wild-type primary BMDMs using a pool of three different siRNAs, but were only able to achieve

Much of the data is analyzed with a paired two-sided t-test, but the authors used Bonferoni correction in some cases and Benjamini-Hochberg adjustment in other cases. It would be helpful to either consistently use the same correction or explain in a short section on stats in the methods the rationale for using different corrections.

We have changed the manuscript to consistently use Benjamini-Hochberg correction for all tests. These correction methods represent different approaches to the multiple-testing problem: Bonferroni correction controls the family-wise error rate, while the Benjamini-Hochberg procedure controls the false-discovery rate. We favour the FDR-based approach as it is less dependent on the somewhat-arbitrary decision as to what constitutes a “family” of tests - for example: should tests in the RH and PRU strains be in the same family; should tests for number of parasites and number of vacuoles/host cells be in the same family; should the tests in BMDMs vs. THP-1s should be the same family. At the alpha level of 0.05, one in twenty significant results are false positives following Benjamini-Hochberg adjustment, as opposed to one in twenty of all tests carried out prior to adjustment. We find this appropriate for our study.

The IFA images in Figure 2B appear to show considerable redistribution of ROP1 from the rhoptries to other parts of the parasite upon short Trition-X100 treatment of formaldehyde fixed samples. Inclusion of a low concentration of glutaraldehyde might help preserve the normal distribution of ROP1. Alternatively, or additionally, permeabilization with saponin or digitonin could help visualize ROP1 associated with the PVM. Improving the imaging is not critical to support the conclusions of the study, but would nevertheless be an asset.

The ROP1 immunofluorescence staining within the parasites is likely from protein that is being trafficked through the ER or Golgi. This staining is more prominent with shorter Triton permeabilisation and in the complement lines. We have also observed PVM localisation of ROP1 using permeabilisation with 0.1% saponin for 15 min, but find that it is clearer and more consistent with the short Triton permeabilisation. We have added additional examples to the supplementary figure validating ROP1 knockout and complement.

Reviewer #3

The IFNg-dependent T. gondii control data as well as the Irgb6 recruitment data are extremely variable and preclude drawing any solid conclusions (Fig 2c-f, 3d, 4e-f, S3a-d, S8). The authors normalize the data by presenting the ratio of +IFNg/-IFNg in % of the measure they are analyzing. However, this does not "clean up" the data. The underlying cause for this problem are the extremely varied input, the time point analyzed and the nature of the microscopy experiment.

We have found that parasite IFNγ-restriction assays are highly variable between biological replicates by all methods that we have tried. We thank Reviewer #1 for sharing their similar experience, as we have heard the same from several colleagues. As it becomes more common to plot individual data points rather than summary statistics, this is increasingly apparent in the literature (e.g. Wang et al 2020 Fig 2, Rinkenberger et al 2021). While we could select the three “best” replicates to publish or instead present only summary statistics to obscure the variation present, we do not feel that this would be in the best interest of the field or of open science. We address Reviewer #3’s specific concerns below.

Presenting the data as a ratio/percentage of IFNγ-stimulated versus naive cells is not an attempt to “clean up” the data, it is standard practice in the field and reflects the hypothesis being tested. For example: Wang et al 2020 PMID: 33067458 (Saeij lab), Matta et al 2019 PMID: 31413201 (Sibley lab), Gay et al 2016 PMID: 27503074 (Hakimi lab), Bando et al 2018 PMID: 30283439 (Yamamoto lab), Fleckenstein et al 2012 PMID: 22802726 (Howard lab).

- Varied input: looking at the Supplementary tables and calculating the input MOI in the -IFNg samples for the BMDM, for example, they range from 0.1-6. Most of the MOIs are in the 0.1-1 range, but this is still a huge variation and not the MOI of 0.3 the authors are aiming to add. Input host cells within one experiment often differ 5 fold between the T. gondii strains analyzed. Consistency in input is not achieved.

The inputs of the restriction assays, as estimated by the parasite and host cell numbers in the unstimulated cells, are variable but not so extreme as Reviewer #3 says. Importantly, variability between strains within a biological replicate is far less than between biological replicates, and it is for this reason that we have analysed the results with paired tests. It is in fact a feature of microscopy-based assays that these sources of variability can be identified. We have attempted to be as open as possible with our data in reporting all of these measured parameters.

MOI is the most appropriate measure of inter- and intra-replicate variability as it accounts for both host cell and vacuole number. We have now added estimated MOI as a column in the supplementary data files (S2B and S3B: MOI = “Vacuole - Minus IFNG Median”/”Host Nuclei - Minus IFNG Median”). Out of 49 data points, there are only three instances in the BMDM restriction assays where MOI >1 (1.26, 1.06, 1.42) and none in THP-1s. Because the MOIs are low, the range of MOI within each replicate is small. Furthermore, for MOI ≤1, we expect there will be little qualitative effect on restriction as the host cells will be infected with either 1 or 0 vacuoles. In our data, we do not find any correlation between MOI and restriction for any individual strain or for all strains combined.

Regardless of the inter-replicate variability, the results for parasite restriction are statistically significant using an appropriate paired test. Our control parasite line PRUΔGRA12 behaves as expected (Fox et al 2019 PMID: 31266861) and differences between the RH and PRU strain in BMDMs dependent on the polymorphic effector ROP18 are apparent, although not tested here. While there is variation in the magnitude of restriction between replicates, there is a qualitative decrease in parasite survival in terms of total parasite number (Fig 2C and 2E) for ΔROP1 parasites compared to both ΔUPRT and the complemented lines in every replicate for both strains in BMDMs (7/7 replicates for each strain), and in all but one replicate in THP-1s (6/7 for RH, 7/7 for PRU). We consider that this restriction phenotype is highly robust.

- Time point analyzed: the authors chose to analyze 24h p.i. IFNg-stimulated MEFs, BMDMs and THP-1s. All will have undergone a significant amount of cell death at this time point. The authors even present the varying host cell numbers in Fig S3: the +IFNg cell numbers vs -IFNg cell numbers in BMDMs range from 35-70% for example in RH; THP-1 seem slightly better but still there is a range of 70-120%.

24 hpi or later is a standard timepoint at which to measure parasite restriction, as at earlier timepoints the dynamic range of the assays are reduced due to lower parasite numbers and therefore lower fluorescence, luminescence, or uracil uptake. For example, the following references all measure parasite restriction at at least 24 hpi: Wang et al 2020 Fig. 2: 24 hpi; Matta et al 2019 Fig. 3: 36-40 hpi; Gay et al 2016 Fig. 8: 36-48 hpi; Bando et al 2018: 24 hpi; Fleckenstein et al 2012 Fig. 4: 48 hpi.

The Howard lab has shown that host cell death is directly proportional to MOI in murine cells (Zhao et al 2009 PMID: 19197351, Lilue et al 2013 PMID: 24175088). At low MOIs of

- Nature of the experiment: Due to rampant host cell death at 24h p.i. analyzing the total parasite or vacuole number or even vacuole size is difficult in a microscopy experiment (dead cells will wash away even after fixation as they do not adhere anymore). Most laboratories (Howard, Yamamoto, Coers, Saeij, Steinfeldt etc), who study these mechanisms employ experimental systems that do not rely on washes/perturbations (plaque assays, plate reader T. gondii luciferase growth assays, uracil incorporation etc), or focus on the "number of parasites/vacuole" measure that are less dependent on host cell numbers (Saeij, Coers etc) or choose host cells that do not undergo IFNg-driven cell death (A549 cells, Sibley recent elife; only overexpression of the host factor induces cell death).

High-content imaging has previously been used to characterise IFNγ-dependent restriction of T. gondii (Gay et al 2016 PMID: 27503074 Fig 8D, Rinkenberger et al 2021 PMID: 34871166, Fisch et al 2021 PMID: 34931666) and for anti-parasitic drug screening (Touquet et al 2018 PMID: 30157171). Certainly different methods have different strengths and weaknesses: for example, fixed imaging-based methods required washing which will remove dead host cells from the analysis, whereas e.g. luciferase or uracil incorporation assays do not; however, with imaging we can measure multiple potential phenotypes simultaneously and identify potential confounding factors that would be unknown in other assays. Acknowledging that every assay has specific weaknesses, we verify the role of ROP1 by orthogonal methods: in vivo CRISPR screen for growth in the mouse peritoneum, in vitro growth in IFNγ-stimulated macrophages, and (now added in revision) in vivo virulence studies. The results of all of these assays concur.

If ROP1 alters the amount of host cell death induced in macrophages, host cell numbers in the +IFNg samples will vary in dependency of ROP1. Hence, in order to assess ROP1-dependent T. gondii restriction it is imperative to know if ROP1 alters host cell death. This can only be measured with cell death assays (total LDH or better a kinetic cell death assay) and not by looking at how many host cells remain after fixation in a microscopy experiment. To compare ROP1-driven parasite restriction, it is important to compare the same input MOI for each strain (-IFNg), determine a time point p.i. before host cell death has taken over and verify host cell numbers stay within a reasonable variation between strains. Input MOI consistency between T. gondii strains is typically measured by plaque viability assay.

We have now carried out kinetic propidium iodide uptake assays to determine formally whether ROP1 affects host cell death in BMDMs, which have been added to the revised manuscript. At an MOI of 0.3, as targeted in our restriction assays, there is very little parasite-induced cell death and no significant differences between parasite strains. This is in line with the results from the Howard lab mentioned above (Zhao et al 2009 PMID: 19197351, Lilue et al 2013 PMID: 24175088), and confirms that the majority of host cell loss observed in the high-content imaging assays at an average MOI of 0.3 is not parasite-induced.

We also carried out propidium iodide uptake assays at an MOI of 3, and found a slight but significant increase in host cell death for PRUΔROP1 in BMDMs of 10-15%. A small increase was also observed for RHΔROP1 compared to the complemented line, but was not significant compared to RHΔUPRT. This differential effect at low and high MOI is interesting, although we cannot rule out that a small phenotype is beyond the limit of detection at low MOI. We conclude that at low MOI host cell death does not have a detectable role in the restriction of ΔROP1 parasites, leaving the most likely mechanism as disruption of vacuoles. At high MOI, vacuole breakage appears to result in host cell death as host cytosolic sensors recognise parasite-derived PAMPs: parasite DNA is sensed by AIM2 in activated THP-1 macrophages leading to apoptosis (Fisch et al 2019 PMID: 31268602), while in murine cells vacuole breakage by the IRGs leads to necrotic cell death (Zhao et al 2009 PMID: 19197351). Increased cell death at high MOI is therefore also consistent with increased disruption of vacuoles.

Fig 2C - Why are there datapoints with more than 100% T. gondii in +IFNg vs -IFNg samples? In these cases, no IFNg restriction was observed for the WT strain. This data is not reliable, as it has been demonstrated before that BMDMs control RH in an IFNg-dependent fashion.

This is one replicate out of seven. Although removing this replicate would not affect the statistical significance of the results in terms of total parasite number or number of vacuoles, we do not consider it appropriate to remove data because it does not match our expectations. Published data also show survival of RH parasites in BMDMs at or close to 100% in some replicates as a result of biological and technical variability (e.g. Wang et al 2020 PMID: 33067458, Fig. 2), which is to be expected in primary cells derived from different donor mice. The average survival we determine for RH across replicates in ~75%, which is in line with published data.

Fig 4C - The description of the IP in the legends and the materials is incomplete. In the materials it only states the loading of the IP fraction. Is the supernatant the post-IP fraction? 202200-HA is not described in the Figure legend. Why is the IP'ed version of C1QBP smaller than the supernatant version?

We will clarify this in the figure legend and methods. The supernatant is the post-IP fraction. It was necessary to load as much material as possible to detect C1QBP in the post-IP supernatant of the RHΔKU80 and PRUΔKU80-infected samples, but as the protein concentration in the supernatant is far higher than in the IP fraction this appears to have caused slight retardation of protein migration in these lanes. This is only apparent for C1QBP as it is the only protein detected in both supernatant and IP.

Fig 4E and F - There is almost no IFNg-dependent restriction in the WT MEFs for any of the parasite strains (4E). Some of the data even shows a dramatic increase of parasite load with IFNg versus without IFNg. Hence, no conclusion can be drawn about the function of C1QBP and making a ratio of KO vs WT (Fig 4F) is not justified as there was no IFNg restriction to begin with in WT cells.

We included these data as we found statistically significant differences between the C1QBPflox/flox and C1QBP-/- MEFs that were dependent on the presence of ROP1 in the parasites. However, after the concerns raised by all three reviewers we agree that it is better to not to draw conclusions from these assays given the lack of parasite restriction that would otherwise be expected in MEFs (e.g. Niedelman et al 2012 PMID: 22761577). We will include these data only in the supplementary figures as a reference for other researchers working on the role of C1QBP in innate immunity who may use these previously published cell lines.

Fig 3D - Something is wrong with the Irgb6 recruitment data. Many labs have shown that RH recruitment of Irgb6 is at most 10% due to the activity of ROP5 and ROP18. The authors get up to 40% recruitment for RH and again the data ranges from 5-42%, that's a huge variation!

Recruitment of IRGB6 was determined by the ratio of the median fluorescence intensity in a 6 pixel radius around the parasites versus the median intensity in the rest of the infected cell cytoplasm. After carefully checking each step of this analysis pipeline, we found that the threshold for host cell cytoplasm segmentation based on CellMask staining was set too low and was including areas of empty space. This resulted in erroneously low median intensity in the host cell and so false-positive calling of IRGB6 recruitment to some vacuoles. We have corrected the cut-off threshold for host cell segmentation and reanalysed all the data with this corrected script. We now find average recruitment of ~5% for RHΔUPRT, ~15% for RHΔROP18, and ~60% for PRUΔUPRT (comparable to e.g. Fentress et al 2011, Fleckenstein et al 2012, Niedelman et al 2012, and Etheridge et al 2014), as well as less variability overall.

Fig S3A and C: The authors conclude that ROP1 does not affect vacuole size. I disagree. The data for THP-1 is significant. However, due to the noisy input with such varied MOI and varied host cell numbers, at the moment, no solid conclusion can be drawn.

Although there is a significant difference between RHΔUPRT and RHΔROP1, this is not rescued by complementation so we are hesitant to claim this as a phenotype of ROP1. There are no significant differences in the PRU strain. We highlight this in the text of the results section, as it may be an important strain- or host species-dependent mechanism.

Fig S3B and D: The data clearly show huge variation in host cell numbers depending on IFNg and T. gondii infection. It is well-known that host cell death occurs in these experimental systems. In order to analyze whether ROP1 impacts host cell death a kinetic cell death assay is needed. Assessment of remaining host cell numbers in a 96 well plate microscopy experiment is not a quantitative assessment of host cell death, so the conclusion is not valid.

Please refer to comments on host cell death above.

T. gondii "type" is always spelt with a lowercase "t" by convention.

We have corrected 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

The authors identify 22 putative T. gondii virulence factors in an in vivo CRISPR screen and aim to analyze ROP1 as one of the hits for its contribution and mechanism in IFNg-driven parasite control mostly in murine and human macrophages. Somewhat disappointingly, most of the data are too variable to draw conclusions. The underlying cause for the variability in the data are the divergent input host cell numbers within the same experiment, variable MOI and most importantly, probably the loss of host cells due to cell death (the infected ones wash away more easily). The authors try to remedy this by pooling and normalizing data, and while there may be a trend to ROP1 counteracting IFNg-driven T. gondii control (Fig 2), the result overall is not convincing. Even if more robust data would confirm the role of ROP1 in innate immune evasion, all mechanistic "tries" the authors undertake in this study show no phenotype (Fig 3 and S6) or are unreliable (Fig 4, see comments below).

Major concern:

The IFNg-dependent T. gondii control data as well as the Irgb6 recruitment data are extremely variable and preclude drawing any solid conclusions (Fig 2c-f, 3d, 4e-f, S3a-d, S8). The authors normalize the data by presenting the ratio of +IFNg/-IFNg in % of the measure they are analyzing. However, this does not "clean up" the data. The underlying cause for this problem are the extremely varied input, the time point analyzed and the nature of the microscopy experiment.

• Varied input: looking at the Supplementary tables and calculating the input MOI in the -IFNg samples for the BMDM, for example, they range from 0.1-6. Most of the MOIs are in the 0.1-1 range, but this is still a huge variation and not the MOI of 0.3 the authors are aiming to add. Input host cells within one experiment often differ 5 fold between the T. gondii strains analyzed. Consistency in input is not achieved.
• Time point analyzed: the authors chose to analyze 24h p.i. IFNg-stimulated MEFs, BMDMs and THP-1s. All will have undergone a significant amount of cell death at this time point. The authors even present the varying host cell numbers in Fig S3: the +IFNg cell numbers vs -IFNg cell numbers in BMDMs range from 35-70% for example in RH; THP-1 seem slightly better but still there is a range of 70-120%.
• Nature of the experiment: Due to rampant host cell death at 24h p.i. analyzing the total parasite or vacuole number or even vacuole size is difficult in a microscopy experiment (dead cells will wash away even after fixation as they do not adhere anymore). Most laboratories (Howard, Yamamoto, Coers, Saeij, Steinfeldt etc), who study these mechanisms employ experimental systems that do not rely on washes/perturbations (plaque assays, plate reader T. gondii luciferase growth assays, uracil incorporation etc), or focus on the "number of parasites/vacuole" measure that are less dependent on host cell numbers (Saeij, Coers etc) or choose host cells that do not undergo IFNg-driven cell death (A549 cells, Sibley recent elife; only overexpression of the host factor induces cell death). If ROP1 alters the amount of host cell death induced in macrophages, host cell numbers in the +IFNg samples will vary in dependency of ROP1. Hence, in order to assess ROP1-dependent T. gondii restriction it is imperative to know if ROP1 alters host cell death. This can only be measured with cell death assays (total LDH or better a kinetic cell death assay) and not by looking at how many host cells remain after fixation in a microscopy experiment. To compare ROP1-driven parasite restriction, it is important to compare the same input MOI for each strain (-IFNg), determine a time point p.i. before host cell death has taken over and verify host cell numbers stay within a reasonable variation between strains. Input MOI consistency between T. gondii strains is typically measured by plaque viability assay.

Other specific major comments:

Fig 2C - Why are there datapoints with more than 100% T. gondii in +IFNg vs -IFNg samples? In these cases, no IFNg restriction was observed for the WT strain. This data is not reliable, as it has been demonstrated before that BMDMs control RH in an IFNg-dependent fashion.

Fig 4C - The description of the IP in the legends and the materials is incomplete. In the materials it only states the loading of the IP fraction. Is the supernatant the post-IP fraction? 202200-HA is not described in the Figure legend. Why is the IP'ed version of C1QBP smaller than the supernatant version?

Fig 4E and F - There is almost no IFNg-dependent restriction in the WT MEFs for any of the parasite strains (4E). Some of the data even shows a dramatic increase of parasite load with IFNg versus without IFNg. Hence, no conclusion can be drawn about the function of C1QBP and making a ratio of KO vs WT (Fig 4F) is not justified as there was no IFNg restriction to begin with in WT cells.

Fig 3D - Something is wrong with the Irgb6 recruitment data. Many labs have shown that RH recruitment of Irgb6 is at most 10% due to the activity of ROP5 and ROP18. The authors get up to 40% recruitment for RH and again the data ranges from 5-42%, that's a huge variation!

Fig S3A and C: The authors conclude that ROP1 does not affect vacuole size. I disagree. The data for THP-1 is significant. However, due to the noisy input with such varied MOI and varied host cell numbers, at the moment, no solid conclusion can be drawn.

Fig S3B and D: The data clearly show huge variation in host cell numbers depending on IFNg and T. gondii infection. It is well-known that host cell death occurs in these experimental systems. In order to analyze whether ROP1 impacts host cell death a kinetic cell death assay is needed. Assessment of remaining host cell numbers in a 96 well plate microscopy experiment is not a quantitative assessment of host cell death, so the conclusion is not valid.

Minor comments:

T. gondii "type" is always spelt with a lower case "t" by convention.

Significance

T. gondii protein ROP1 was identified by the authors in an in vivo CRISPR screen as a potential effector protein mediating parasite resistance to innate immunity. The function of ROP1 is currently unknown. The data presented in this manuscript does not deliver conclusive evidence that ROP1 counteracts IFNg-driven immunity and a potential mechanism has not been uncovered.

Referees cross-commenting

This section contains comments of all reviewers

Reviewer 1

I agree with the comments of review #2 in regard to additional experiments needed for validation. I also agree with reviewer #3 about the variable data with Irgb6 recruitment and IFNg control- we have also done these assays and struggled with the variability. The suggestions for how to reduce variability with modified protocols are good - but is our role really to instruct them how to do the experiments? It seems you are asking them to start over, and I am not sure the situation is that bleak. I would be more inclined to allow them to claim only features that are clearly supported within the variability of the current assays. Perhaps that weakens the conclusions they can make and hence ultimate decision - but I feel that this choice (of fixing it or compromising) should be up to the authors.

Reviewer 3

Thanks for your points, yes I agree, statements can be made about the role of ROP1 in innate immune defence (i.e. it probably contributes to control of vacuole numbers and size in both mouse and human), but the authors should more carefully place them in the framework of their assays.

Rev 2, point 1 is very good (title suggestion). Rev 1, point 1, I agree very much (validate ROP1 phenotype in vivo). Rev 2, minor point 1 (explain your stats) and Rev 1, point 3 (depict absolute numbers in comparisons). This is my point also, the authors try to normalise the data and apply varied statistics to smooth over the variability. All reviewers pointed out the problems with the C1QBP data, specifically the >100% data of +IFNg vs -IFNg and additionally the potential for an in vitro artefact (Rev 1).

My main points remain: - statements on ROP1 and its impact on host cell death without conducting any host cell death assay cannot be made. - the Irgb6 recruitment data is puzzling and contradicts all the many published data on Irgb6 recruitment levels. - C1QBP data is not interpretable with data where IFNg does not restriction WT parasites.

In the end, maybe suggesting in vivo validation of ROP1, remove impact on host cell death claim and the C1QBP data (or conduct further experiments to understand role)?

Reviewer 1

Yes I agree with the points and this slightly reworded final assessment:

Request in vivo validation of ROP1 Clarify discrepancies in Irgb6 data Remove impact on host cell death claim Soften the conclusions about C1QBP data (or conduct further experiments to understand its role).

Reviewer 2

Thanks for leading the discussion R1 and R3. It looks like we are in good agreement. The study has some merit but will require more in vitro and in vivo experiments to support the conclusions and to substantiate the relationship between ROP1 and C1QBP.

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

Toxoplasma gondii secretes a variety of effector proteins that manipulate many aspects of host cell physiology including several that facilitate evasion of cell autonomous immunity due to activation by interferon gamma. The current study features a new CRISPR gRNA library targeting genes encoding proteins that are released from two types of secretory organelles that have previously been strongly implicated in immune evasion, namely rhoptries and dense granules. The authors use the library to identify parasite genes that are more important to in vivo infection than they are suring in vitro culture. That the screen identified several known immune evasion genes suggests that the experiment worked as intended. Among the somewhat marginal hits, the authors identified ROP1, which they pursued further since the function of this protein has remained unknown despite being the first identified ROP protein. As rigorous component of the study, the authors disrupt and restore expression of ROP1 in two different strains (RH and Pru) and identify ROP1 knockout parasite susceptibility to IFNg stimulation in two different types of macrophages (murine BMDMs and human THP-1). Two functional assays were used to indicate that rhoptry secretion is not impaired in parasites lacking ROP1. Co-immunoprecipitation assays suggest an interaction between ROP1 and a multifunctional host protein, C1QBP, inspiring the hypothesis that ROP1 targets C1QBP to evade cell autonomous immunity. However, host cells lacking C1QBP appear to be more restrictive to growth of WT parasites and neutral to ROP1 deficient parasites, implying that C1QBP contributes to T. gondii survival in a manner that is dependent on ROP1. The authors are appropriately cautious when discussing the extent to which the function of ROP1 is mechanistically linked to C1QBP. Overall, the study provides some new evidence for a role of ROP1 in limiting IFNg dependent clearance of T. gondii and identifies some clues to potential mechanism without being able to nail down this aspect.

Main comments

1. The title indicates a positive role for ROP1 in subverting innate immune restriction, but the data indicate that a deficiency in ROP1 causes susceptibility to innate immune restriction. This might seem like a subtle discord, but in the absence of identifying the mechanism of ROP1 subversion of innate immune restriction it seems more appropriate to provide a title the better reflects the findings i.e., that ROP1 deficient parasites are more susceptible to innate immune restriction.
2. The reference strains and complement strains lack UPRT, but from the available information it appears the KO strains have UPRT. If this is the case, it is necessary to rule out that the presence of UPRT doesn't render parasites more susceptible to IFNg mediated killing by performing additional experiments comparing RH∆ku80 with RH∆ku80∆uprt and Pru∆ku80 with Pru∆ku80∆uprt.
3. Although it is understandable why the authors chose MEFs to test the role of C1QBP because MEFs were used for the Co-IP, the MEFs do not appear to be responding to IFNg for parasite growth restriction (-/+ IFNg % survival is at or above 100% for WT MEFs). As such, the authors are potentially blind to the role of C1QBP in the context of IFNg restriction. It would be ideal to repeat these experiments using BMDMs from WT and C1QBP KO mice to assess the potential contribution of C1QBP during IFNg restriction of T. gondii growth and survival.

Minor comments

1. Much of the data is analyzed with a paired two-sided t-test, but the authors used Bonferonni correction in some cases and Benjamini-Hochberg adjustment in other cases. It would be helpful to either consistently use the same correction or explain in a short section on stats in the methods the rationale for using different corrections.
2. The IFA images in Figure 2B appear to show considerable redistribution of ROP1 from the rhoptries to other parts of the parasite upon short Trition-X100 treatment of formaldehyde fixed samples. Inclusion of a low concentration of glutaraldehyde might help preserve the normal distribution of ROP1. Alternatively, or additionally, permeabilization with saponin or digitonin could help visualize ROP1 associated with the PVM. Improving the imaging is not critical to support the conclusions of the study, but would nevertheless be an asset.

Significance

Immune evasion is critical to the survival of pathogens including Toxoplasma gondii, yet the effector proteins responsible for such evasion remain incompletely identified or understood. This study identifies a role for ROP1 in parasite survival in interferon gamma stimulated host cells, thus addressing a long standing question of this protein's function during infection. The study appears suited to a microbiology or infection themed journal. My areas of expertise are T. gondii pathogenesis, virulence, secretion, invasion, egress, resource acquisition, and persistence.

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

Butterworth et al identify a novel function of Toxoplasma ROP1 being involved in subverting host IFN-g restriction of parasite growth. The authors executed a CRISPR-Cas9 sgRNA screen in a type II strain parasite targeting a sub-pool of genes characterized with respect to their localization in rhoptries and dense-granules. They adopted an approach of sequentially sequencing the survivors following growth in vitro in human foreskin fibroblasts followed by in vivo expansion in mouse peritoneum. Measurement of relative depletion of sgRNA sequences produced significant hits of factors involved in parasite survival. They identified many parasite proteins previously known to be involved in different aspects of parasite virulence including those involved in countering murine IFN-g mediated anti-toxoplasma response. They then focused on identifying the function of ROP1 in augmenting virulence and survival of the parasite in mice. Lastly, using immunoprecipitation in mouse cells they identified host C1QBP as a protein that may be involved in facilitating ROP1 mediated parasite resistance against the IFN-g response.

Major Comments:

1. The authors identified ROP1 as a significant hit from their in vivo screen. However, they have not done any validation experiments using rop1 KO parasites in mice. Previous studies have shown no virulence defect in mice for rop1 KO in the type I background (PMID: 8719248). The result could be different in the type II strain used here, but this needs to be tested and shown.
2. The data in Fig. 2 and S3 do support that reduced parasitemia was due to decrease in number of vacuoles rather than their size or host cell death. However, it is important to control for invasion and/or egress differences of rop1 KO parasites in IFN-g activated cells.
3. It is important and informative to depict absolute parasite/size when making multiple comparisons. For example, the data in Fig. 4E shows C1QBP-/- MEFs can clear both RH and PRU better than the WT. However the authors do not comment on what is the meaning of > 100% parasite numbers in IFN-g treated MEFs with respect to untreated in WT. Since the data re normalized, it is difficult o appreciate what the actual differences are. Additionally, C1QBP-/- MEFs show close to equal survival in control and IFN-g treated condition (approximately 100%). Is it correct to infer that C1QBP has no effect on parasite survival? This should be considered in light of the comment below on colocalisation of C1BQ.
4. The authors observed good restriction of both RH and PRU in IFN-g activated THP1s without cell death (Fig S1D). It is important to incorporate this information into the main result and discuss their implications in contrast to a previous report from 2019 (Fisch et al, PMID: 31268602).
5. The authors should consider conducting C1QBP functional assays to explore potential roles in parasite survival/growth within host cells. For example, it would be informative to measure the extent of autophagy or transcriptomic profile of the KO to deduce or suggest possible mechanism of restriction.

Minor Comments:

1. The authors state in their text that "C1QBP localised primarily to the mitochondria (Figure S7A) and therefore did not see any co-localisation with ROP1". The authors should discuss in more detail as this finding seems to contradict the interaction studies. Is there any independent evidence to corroborate the interaction studies and show they are not simply an in vitro artifact?
2. The authors state in their text that "Enhanced restriction of Δrop1 parasites is primarily mediated through increased vacuole destruction". Their data is more suggestive of growth restriction. The authors should provide more direct data for destruction of the vacuoles or change the wording to indicate it is due to growth restriction.
3. The authors should provide high resolution IFA images and decrease their size to an equivalent size of other sub-figures. Empty white space between sub-figures can be minimized. Font size of the figure labels/axis/titles should be matched and increased slightly.

Significance

The work has significance with respect to host-parasite interaction in the Toxoplasma field. The report aims to assign a function to the first identified rhoptry protein in Toxoplasma. It employs CRISPR-Cas9 screens in Toxoplasma to study and identify the function of proteins involved in parasite virulence. The study may be of less interest to groups working in the field of host-pathogen interaction, innate immunity, as the genes studied here are not widely conserved nor do they provide obvious parallels to other systems.

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)): Major comments: 1. The authors mentioned that the heart dysfunction observed upon CHCHD3/6 KD may be mediated via defects in ATP synthase. Then, how does CHCHD3/6 KD affect ATP synthase? Additionally, OPA1 also affects ATP synthase, why does OPA1 KD just reduce fractional shortening (S.T.2) without reducing F-actin staining?

• Since MICOS is in involved ETC assembly/sorting in cristae, ATPase subunits may not be assembled correctly and thus causing defects mitochondrial morphology and function, which is further supported by reduced mito-GFP staining (see Fig. 4B&H) (see discussion, lines 438-41 in red; see also lines 434-438 and 441-44 that are unchanged). F-actin staining with OPA1 KD: it may not be a strong enough KD to cause reduced F-actin staining, since only strong CHCHD3/6 KD shows reduced F-actin.
• To address this issue and ask whether OPA1 interacts with CHCHD3/6 in affecting sarcomeric protein levels, we plan to do an interaction experiment with and OPA1 KD with CHD3/6-C1 KD at 21oC (as in Fig. 6B) and probe for reduced FS; and as well at 25oC and probe for reduced F-actin staining (as with beta-Spectrin KD in Fig. 6D and with Sam50 in Fig. 5D).

1. It has been reported that CHCHD3 KD in HeLa cells causes fragmented mitochondria, so how does CHCHD3/6 KD caused mitochondrial aggregation? What is the mechanism?

• Thank you for pointing that out. It is actually more appropriate to call this phenotype “fission-fusion defects”. See lines 279-86, 416.
• To address this issue further, we propose to do co-KD or co-overexpression (OE) of Drp1 with KD of CHCHD3/6 (as above). Specifically, we will use the strong CHCHD3/6-RNAiA to in conjunction with Drp-1 OE (in the presence of mito-GFP) and see if this can rescue the mitochondrial morphology defects, thus concluding that CHCHD3/6 KD is likely to causes aggregation that is normalized with Drp-1 OE. If this is not the case, but a parallel experiment with Drp-1 KD can rescue, this would support the conclusion that CHCHD3/6 causes increased fragmentation that is counteracted with Drp-1 KD. A complementary experiment would be to use a CHCHD3/6 sensitizer (as in Fig. 5&6). These experiments are intended to address the question whether CHCHD3/6 causes primarily fusion or fission defects.

1. The ultrastructure of mitochondria (especially aggregated mitochondria) in control and CHCHD3/6 KD heart of drosophila should be analyzed by TEM.

• That’s difficult, since cardiac tissue is so thin. It is not clear of fission vs fusion defects can easily be distinguished. Instead, we propose to do genetic interaction experiments (see above 2.).

Reviewer #1 (Significance (Required)): The manuscript partially illustrate the relationship between MICOS complex with Hypoplastic left heart syndrome (HLHS), which is interesting to the reader.

• We thank the reviewer for his/her appreciation of our study

Reviewer #2 (Evidence, reproducibility and clarity (Required)): This study performed whole genome sequencing (WGS) on a large cohort of hypoplastic left heart syndrome (HLHS) patients and their families to identify candidate. Nine candidate genes with rare, predicted damaging homozygous variants were identified. Of the candidate HLHS gene homologs tested, cardiac-specific knockdown (KD) of the mitochondrial contact site and cristae organization system (MICOS) complex subunit dCHCHD3/6 resulted in drastically compromised heart contractility, diminished levels of sarcomeric actin and myosin, reduced cardiac ATP levels, and mitochondrial fission-fusion defects. These heart defects were similar to those inflicted by cardiac KD of ATP synthase subunits of the electron transport chain (ETC), consistent with the MICOS complex's role in maintaining cristae morphology and ETC complex assembly. Analysis of 183 genomes of HLHS patient-parent trios revealed five additional HLHS probands with rare, predicted damaging variants in CHCHD3 or CHCHD6. Hypothesizing an oligogenic basis for HLHS, the authors tested 60 additional prioritized candidate genes in these cases for genetic interactions with CHCHD3/6 in sensitized fly hearts. Moderate KD of CHCHD3/6 in combination with Cdk12 (activator of RNA polymerase II), RNF149 (E3 ubiquitin ligase), or SPTBN1 (scaffolding protein) caused synergistic heart defects, suggesting the potential involvement of a diverse set of pathways in HLHS. General Comments: The authors performed an elegant series of experiments that implicate variants of dCHCHD3/6 in HLHS patients as contributing to mitochondrial and sarcomeric defects and contractile function defects. Demonstrating in Drosphilia the functional and biochemical implications of knocking out dCHCHD3/6 provides some potentially important insights into the functional and biochemical implications of dCHCHD3/6 variants in HLHS patients. The data is also complemented by hiPSC-CM studies in which knockdown of CHCHD6 and CHCHD3 showed similar alterations in ATP synthase and mitochondrial morphology. The authors nicely show that knock down of the subunit dCHCHD3/6 resulted in drastically compromised heart contractility, diminished levels of sarcomeric actin and myosin, reduced cardiac ATP levels, and mitochondrial fission-fusion defects in the Drosphilia. What is not clear is how these changes mirror the phenotype of HLHS in humans. It would helpful to speculate to a greater extent as to how these changes would manifest as a decreased left ventricular development in HLHS.

• This is indeed a very important question we comment on in the discussion (see text revision, lines 415-22 in red; see also lines 402-415 and 423-33 that are unchanged). We want to stress that we focus on genetic interactions in heart development, not on convergent endpoint phenotypes between flies and humans. However, our studies do support the idea that mitochondrial defects could contribute to HLHS. We show that MICOS deficiency causes mitochondrial defects manifest in diminished ATP production in addition to diminished sarcomeric actin and myosin causing diminished contractility. Impaired contractility during development has previously been proposed to contribute to defective human cardiac growth (no flow – no growth, Goldberg and Rychik, 2016; Grossfeld et al., 2019), thereby compounding the potentially polygenic effects from damaging gene variants.

• Why there would a preferential effect on the left ventricle is another interesting question. We speculate that some of the patient-specific variants are in genes preferentially affecting the left ventricle thus preferentially affecting its growth, thus affecting its contractility, then again compounded by impaired blood flow feeding back to diminishing growth. Specific Comments:

  Line 139: Figure 1A does not show echos from the siblings.

• We apologize that the “(Figure 1A)” was in wrong position (after echocardiograms), causing confusion. We moved it to the previous sentence (line 138). In case the reviewers require that echocardiograms are shown as supplemental data, we can provide these.

  Line155: This table is listed as "Table 1" not Supplemental Table 1.

• We apologize for mislabeling. This table is now listed as Supplementary Table 1.

  Reviewer #2 (Significance (Required)): This is a highly significant study. The main audience would be pediatric cardiologists and geneticists.

• We thank the reviewer for his/her appreciation 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 #2

Evidence, reproducibility and clarity

This study performed whole genome sequencing (WGS) on a large cohort of hypoplastic left heart syndrome (HLHS) patients and their families to identify candidate. Nine candidate genes with rare, predicted damaging homozygous variants were identified. Of the candidate HLHS gene homologs tested, cardiac-specific knockdown (KD) of the mitochondrial contact site and cristae organization system (MICOS) complex subunit dCHCHD3/6 resulted in drastically compromised heart contractility, diminished levels of sarcomeric actin and myosin, reduced cardiac ATP levels, and mitochondrial fission-fusion defects. These heart defects were similar to those inflicted by cardiac KD of ATP synthase subunits of the electron transport chain (ETC), consistent with the MICOS complex's role in maintaining cristae morphology and ETC complex assembly. Analysis of 183 genomes of HLHS patient-parent trios revealed five additional HLHS probands with rare, predicted damaging variants in CHCHD3 or CHCHD6. Hypothesizing an oligogenic basis for HLHS, the authors tested 60 additional prioritized candidate genes in these cases for genetic interactions with CHCHD3/6 in sensitized fly hearts. Moderate KD of CHCHD3/6 in combination with Cdk12 (activator of RNA polymerase II), RNF149 (E3 ubiquitin ligase), or SPTBN1 (scaffolding protein) caused synergistic heart defects, suggesting the potential involvement of a diverse set of pathways in HLHS.

General Comments:

The authors performed an elegant series of experiments that implicate variants of dCHCHD3/6 in HLHS patients as contributing to mitochondrial and sarcomeric defects and contractile function defects. Demonstrating in Drosphilia the functional and biochemical implications of knocking out dCHCHD3/6 provides some potentially important insights into the functional and biochemical implications of dCHCHD3/6 variants in HLHS patients. The data is also complemented by hiPSC-CM studies in which knockdown of CHCHD6 and CHCHD3 showed similar alterations in ATP synthase and mitochondrial morphology.

The authors nicely show that knock down of the subunit dCHCHD3/6 resulted in drastically compromised heart contractility, diminished levels of sarcomeric actin and myosin, reduced cardiac ATP levels, and mitochondrial fission-fusion defects in the Drosphilia. What is not clear is how these changes mirror the phenotype of HLHS in humans. It would helpful to speculate to a greater extent as to how these changes would manifest as a decreased left ventricular development in HLHS.

Specific Comments:

Line 139: Figure 1A does not show echos from the siblings.

Line155: This table is listed as "Table 1" not Supplemental Table 1.

Significance

This is a highly significant study. The main audience would be pediatric cardiologists and geneticists.

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 manuscript titled "mitochondrial MICOS complex genes, implicated in hypoplastic left heart syndrome, maintain cardiac contractility and actomyosin integrity", Katja Birker et al. reveal that CHCHD3/6 cardiac-specific KD caused reduced contractility and decreased sarcomeric F-Actin and Myosin staining in fly due to impaired ATP synthase. The findings shown in this manuscript are interesting. However, the additional experiments are needed to confirm the conclusion before publication.

Major comments:

1. The authors mentioned that the heart dysfunction observed upon CHCHD3/6 KD may be mediated via defects in ATP synthase. Then, how does CHCHD3/6 KD affect ATP synthase? Additionally, OPA1 also affects ATP synthase, why does OPA1 KD just reduce fractional shortening (S.T.2) without reducing F-actin staining?
2. It has been reported that CHCHD3 KD in HeLa cells causes fragmented mitochondria, so how does CHCHD3/6 KD caused mitochondrial aggregation? What is the mechanism?
3. The ultrastructure of mitochondria (especially aggregated mitochondria) in control and CHCHD3/6 KD heart of drosophila should be analyzed by TEM.

Significance

The manuscript partially illustrate the relationship between MICOS complex with Hypoplastic left heart syndrome (HLHS), which is intertesing to the reader.

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

Reply to the Reviewers

We thank the reviewers for dedicating time to review our manuscript and providing highly valuable feedback. Please find below a point-by-point answer.

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

Summary:

In the present manuscript, van der Plas et al. compellingly illustrated a novel technique for engendering a whole-brain functional connectivity map from single-unit activities sampled through a large-scale neuroimaging technique. With some clever tweaks to the restricted Boltzmann Machine, the cRBM network is able to learn a low-dimensional representation of population activities, without relying on constrained priors found in some traditional methods. Notably, using some 200 hidden layer neurons, the employed model was able to capture the dynamics of over 40,000 simultaneously imaged neurons with a high degree of accuracy. The extracted features both illustrate the anatomical topography/connectivities and capture the temporal dynamics in the evolution of brain states. The illustrated technique has the potential for wide-spread applications spanning diverse recording techniques and animal species. Furthermore, the prospectives of modeling whole-brain network dynamics in 'neural trajectory' space and of generating artificial data in silico make for very enticing reasons to adopt cRBM.

Major comments:

1. Line 164. The authors claim that conventional methods "such as k-means, PCA and non-negative matrix factorization" cannot be quantitatively assessed for quality on the basis that they are unable to generate new artificial data. Though partly true, in most neuroscience applications, this is hardly cause for concern. Most dimensionality reduction methods (with few exceptions such as t-sne) allow new data points to be embedded into the reduced space. As such, quality of encoding can be assessed by cross-validation much in the same way as the authors described, and quantified using traditional metrics such as percentage explained variance. The authors should directly compare the performance of their proposed model against that of NNMF and variational auto-encoders. Doing so would offer a more compelling argument for the advantage of their proposed method over more widely-used methods in neuroscience applications. Furthermore, a direct comparison with rastermap, developed by Stringer lab at Janelia (https://github.com/MouseLand/rastermap), would be a nice addition. This method presents itself as a direct competitor to cRBM. Additionally, the use of GLM doesn't do complete justice to the comparison point used, since a smaller fraction of data were used for calculating performance using GLM, understandably due to its computationally intensive nature.

PLANNED REVISION #2

We thank the reviewer for the comment, and certainly agree that there are multiple methods for unsupervised feature extraction from data and that they can be validated for encoding quality by cross-validation. However, we stress that reconstructing through a low-dimensional, continuous bottleneck is a different (and arguably, easier) task, than generating whole distributions. Reconstruction delineates the manifold of possible configurations, whereas generative modeling must weigh such configurations adequately. Moreover, none of the methodologies mentioned can perform the same tasks as cRBMs. For instance, NNMF learns localized assemblies, but cannot faithfully model inhibitory connections since, by definition, only non-negative weights are learnt. Also, the connection between the learnt assemblies and the underlying connectivity is unclear. Similarly, rastermap is an algorithm for robustly i) sorting neurons along a set number of dimensions (typically 1 or 2) such that neighboring neurons are highly correlated, and ii) performing dimensionality reduction by clustering along these dimensions. Because Rastermap uses k-means as the basis for grouping together neurons, it does not quantify connections between neurons and assemblies, nor assign neurons to multiple assemblies. Moreover, it is not a generative model, and thus cannot predict perturbation experiments, infer connectivities or assign probabilities to configurations. Therefore, we do not believe that NNMF or Rastermap would be a suitable alternative for cRBM in our study. We nonetheless appreciate the reviewer’s suggestions and agree that we should motivate more clearly why these methods are not applicable for our purposes. Therefore, to emphasize the relative merit of cRBM with respect to other unsupervised algorithms, we now provide a table (Supplementary Table 2) that lists their specific characteristics. We stress that we do not claim that cRBM are consistently better than these classical tools for dimensionality reduction, but focus only on the properties relevant to our study.

Further, we agree that VAEs, which also jointly learn a representation and distribution of data, are close competitors of cRBMs. In Tubiana et al. Neural Computation 2019, we previously compared sparse VAEs with cRBMs for protein sequence modeling, and found that RBMs consistently outperformed VAEs in terms of the interpretability-performance trade-off. In the revised manuscript, we propose to repeat the comparison with VAE for zebrafish neural recordings, and expect similar conclusions.

As for GLM, it is true that the comparison involved subsampling of the neurons (due to the very high computational cost of GLM, where we could estimate the connectivity of 1000 neurons per day). This was already denoted in the relevant figure caption, as the reviewer has seen, but we have now also clarified this point in Methods 7.10.3. Still, we performed our GLM analysis on 5000 neurons (using all neurons as regressors), which is 10% of all neurons, and we believe this is a sufficient number for comparison. This emphasizes the ability of our optimized cRBMs to handle very large datasets, such as the presently used zebrafish whole brain recordings.

Line 26. The authors describe their model architecture as a formalization of cell assemblies. Cell assemblies, as originally formulated by Hebb, pertains to a set of neurons whose connectivity matrix is neither necessarily complete nor symmetric. Critically, in the physiological brain, the interactions between the individual neurons that are part of an assembly would occur over multiple orders of dependencies. In a restricted Boltzmann machine, neurons are not connected within the same layer. Instead, visible layer neurons are grouped into "assemblies" indirectly via a shared connection with a hidden layer neuron. Furthermore, a symmetrical weight matrix connects the bipartite graph, where no recurrent connectivities are made. As such, the proposed model still only elaborates symmetric connections pertaining to first-order interactions (as illustrated in Figure 4C). Such a network may not be likened with the concept of cell assemblies. The authors should refrain from detailing this analogy (of which there are multiple instances of throughout the text). It is true that many authors today refer to cell assemblies as any set of temporally-correlated neurons. However, saying "something could be a cell assembly" is not the same as saying "something is a cell assembly". How about sticking with cRBM-based cell assemblies (as used in section 2.3) and defining it beforehand?

We thank the reviewer for this excellent question. We agree that there is, in general, a discrepancy between computationally-defined assemblies and conceptual/neurophysiological definition of cell assemblies. We have added a clarification in Results 2.1 to clarify the use of this term when it first occurs in Results. However, we still believe that our work contributes to narrowing the gap. Indeed, our RBM-defined assemblies are i) localized, ii) overlapping, iii) rooted in connectivity patterns (both excitatory and inhibitory), and iv) cannot be reduced to a simple partitioning of the brain with full & uniform connectivity within and between partitions. This is unlike previous work based on clustering (no overlaps or heterogeneous weights), NNMF (no inhibition) or correlation network analysis (no low-dimensional representation).

Regarding the specific comments pointed here, we stress that:

• Effective interactions between neurons are not purely pairwise (“First order”), due to the usage of the non-quadratic potential. (see Eqn 12-13). If the reviewer means by “First-order” interactions the lack of hierarchical organization, we agree, to some extent: in the current formulation, correlations between assemblies are mediated by overlaps between their weights. Fully-hierarchical organization, e.g. by using Deep Boltzmann Machines or pairwise connections within the hidden layer is an interesting future direction, but on the other hand may make it hard to clearly identify assemblies as they might be spread out over multiple layers
• Neurons that participate in a given assembly (as defined by a specific hidden unit) are not all connected with one another with equal strength. Indeed, these neurons may participate in other assemblies, resulting in heterogeneity of connections (see Eqn. 15-17) and interactions between assemblies.
• We acknowledge that the constraint of symmetrical connections is a core limitation of our method. Arguably, asymmetric connections are critical for predicting temporal evolution but less important for inferring a steady-state distribution from data, as we do here. In the revised submission, we added a new paragraph in the discussion section (lines 351-357) in which these limitations are discussed, including the imposed symmetry of the connections and the lack of hierarchical structures, copied below. We trust that this addresses the reviewer’s criticism:

---

In sum, cRBM-inferred cell assemblies display many properties that one expects from physiological cell assemblies: they are anatomically localized, can overlap, encompass functionally identified neuronal circuits and underpin the collective neural dynamics (Harris, 2005, 2012; Eichenbaum, 2018). Yet, the cRBM bipartite architecture lacks many of the traits of neurophysiological circuits. In particular, cRBMs lack direct neuron-to-neuron connections, asymmetry in the connectivity weights and a hierarchical organization of functional dependencies beyond one hidden layer. Therefore, to what extent cRBM-inferred assemblies identify to neurophysiological cell assemblies, as postulated by Hebb (1949) and others, remains an open question.

---

I would strongly recommend adding a paragraph discussing the limitation of using the cRBM, things future researchers need to keep in mind before using this method. One such recommendation is moving the runtime-related discussion for cRBM, i.e. 8-12 hrs using 16 CPU from Methods to Discussion, since it's relevant for an algorithm like this. Additionally, a statement mentioning how this runtime will increase with the length of recordings and/or with the number of neurons might be helpful. What if the recordings were an hour-long rather than 25mins. This would help readers decide if they can easily use a method like this.

We thank the reviewer for the suggestion, and agree that it is important to cover the computational cost in the main text. Regarding the runtime for longer recordings, the general rule of thumb is that the model requires a fixed number of gradient updates to converge (20-80k depending on the data dimensionality) rather than a fixed number of epochs. Thus, runtime should not depend on recording length, as the number of epochs can be reduced for longer recordings. While we did not verify this rule for neural recordings, this is what we previously observed when modeling protein/DNA sequence data sets, whose size range from few hundreds to hundreds of thousands of samples (Tubiana et al., 2019, eLife; Tubiana et al. 2019, Neural Computation; Bravi et al. Cell Systems 2021; Bravi et al. PLOS CB 2021; Fernandez de Cossio Diaz et al. Arxiv 2022 Di Gioacchino et al. BiorXiv 2022). We have now added a summary of these points in Methods 7.7.2, also refer to this with explicit mention of the runtime in the Discussion, end of 2nd paragraph:

---

By implementing various algorithmic optimizations (Methods 7.7), cRBM models converged in approximately 8-12 hours on high-end desktop computers (also see Methods 7.7.2).

---

Line 515. A core feature of the proposed compositional RBM is the addition of a soft sparsity penalty over the weight matrix in the likelihood function. The authors claim that "directed graphical models" are limited by the a priori constraints that they impose on the data structure. Meanwhile, a more accurate statistical solution can be obtained using a RBM-based model, as outlined by the maximum entropy principle. The problem with this argument is that the maximum entropy principle no longer applies to the proposed model with the addition of the penalty term. In fact, the lambda regularization term, which was estimated from a set of data statistics motivated by the experimenter's research goals (Figure S1), serves to constrict the prior probability. Moreover, in Figure S1F, we clearly see that reconstruction quality suffers with a higher penalty, suggesting that the principle had indeed been violated. That being said, RBMs are notoriously hard to train, possibly due to the unconstrained nature of the optimization. I believe that cRBM can help bring RBM into wider practical applications. The authors could test their model on a few values of the free parameter and report this as a supplementary. I believe that different parameters of lambda could elaborate on different anatomical clusters and temporal dynamics. Readers who would like to implement this method for their own analysis would also benefit tremendously from an understanding of the effects of lambda on the interpretation of their data. Item (1) on line 35 (and other instances throughout the text) should be corrected to reflect that cRBM replaces the hard constraints found in many popular methods with a soft penalty term, which allows for more accurate statistical models to be obtained.

We thank the reviewer for their analysis and suggestion. Indeed, adding the regularization term - not present in the classical formulation of the RBM (Hinton & Salakhutdinov, 2006, Science) - was critical for significantly enhancing its performance, which allowed us to implement this model on our large scale datasets (~50K visible units). We agree that providing more information on the effect of the regularization term will benefit readers who would like to use this method, and we propose to add this in the revised manuscript, which would implement the reviewer’s suggestion. See “PLANNED REVISION #1”.

The reviewer’s comment on the Maximum Entropy issue calls for some clarification. The maximum entropy principle is a recipe for finding the least constrained model that reproduces specified data-dependent moments. However, it cannot determine which moments are statistically meaningful in a finite-sized data set. A general practice is to only include low-order moments (1st and 2nd), but this is sometimes already too much for biological data. Regularization provides a practical means to select stable moments to be fitted and others to be ignored. This can be seen from the optimality condition, which writes, e.g., for the weights wi,mu:

| i h,mu>data - i h,mu>model | i,mu = 0.

i h,mu>data - i h,mu>model | = lambda sign(wi,mu) if |wi,mu| > 0.

Essentially, this lets the training decide which subset of the constraints should actually be used. Thus, regularized models are closer to the uniform distribution (g=w=0), and actually have higher entropy than unregularized one (see, e.g., Fanthomme et al. Journal of Statistical Mechanics, 2022). Therefore, we believe that a regularized maximum entropy model can still be considered a bona fide MaxEnt model. This formulation should not be confused with another formulation (that perhaps the reviewer has in mind) where a weighted sum of the entropy and the regularization term is maximized under the same moment-matching constraints. In this case, we agree that maximum entropy principle (MaxEnt) would be violated.

The choice of regularization value should be dictated by bias-variance trade-off considerations. Ideally, we would use the same criterion as for training, i.e., maximization of log-likelihood for the held-out test set, but it is intractable. Thus, we used a consensus between several tractable performance metrics as a surrogate; we believe this consensus to be principally independent of the research goal. While the reconstruction error indeed increases for large regularization values, this is simply because too few constraints are retained at high regularizations.

Essentially, the parameters selected by likelihood maximization find the finest assembly scale that can be accommodated by the data presented. Thus, the number and size of the assemblies are not specified by the complexity of the data set alone. Rather, the temporal resolution and length of the recordings play a key role; higher resolution recordings will allow the inference of a larger number of smaller assemblies, and enable the study of their hierarchical organization.

That being said, we fully agree that the regularization strength and number of hidden units have a strong impact on the nature of the representation learnt. In the revised manuscript, we will follow the reviewer’s suggestion and provide additional insights on the effect of these parameters on the representation learnt (please see revision plan).

Minor comments:

From a neuroscience point of view, it might be interesting to show what results are achieved using different values of M (say 100 or 300), rather than M=200, while still maintaining the compositional phase. Is there any similarity between the cRBM-based cell assemblies generated at different values of M? Is there a higher chance of capturing certain dynamics either functional or structural using cRBM? For example, did certain cRBM-based cell assemblies pop up more frequently than others at all values of M (100,200,300)?

This point will be addressed in the future, as detailed in our response to reviewer 2 (see PLANNED REVISION #1).

The authors have mentioned that this approach can be readily applied to data obtained in other animal models and using different recording techniques. It might be nice to see a demonstration of that.

We agree that showing additional data analysis would be interesting, but we feel that it would overburden the supplementary section of the manuscript, which is already lengthy. In previous works, we and collaborators have used cRBMs for analyzing MNIST data (Tubiana & Monasson, 2017, PRL; Roussel et al. 2022 PRE), protein sequence data (Tubiana et al., 2019, eLife; Tubiana et al. 2019, Neural Computation; Bravi et al. Cell Systems 2021; Bravi et al. PLOS CB 2021; Fernandez de Cossio Diaz et al. Arxiv 2022), DNA sequences (Di Gioacchino et al. BiorXiv 2022), spin systems (Harsh et al. J. Phys. A 2020), etc. Many are included as example notebooks - next to the zebrafish data - in the linked code repository. For neural data, we have recently shared our code with another research group working on mice auditory cortex (2-photon, few thousands of neurons, Léger & Bourdieu). Preliminary results are encouraging, but not ready for publication yet.

Line 237. The justification for employing a dReLU transfer function as opposed to ReLU is unclear, at least within the context of neurobiology. Given that this gives rise to a bimodal distribution for the activity of HUs, the rationale should be clearly outlined to facilitate interpretability.

We thank the reviewer for the question. As we detail in the manuscript (Methods), the dReLU potential is one of the sufficient requirements for the RBM to achieve the compositional phase. The compositional phase is characterized by localized assemblies that co-activate to generate the whole-brain neural dynamics. This property reflects neurobiological systems (Harris, 2005, Neuron), which is one of the reasons why we employed compositional RBMs for this study.

As the reviewer points out, the HUs that we infer exhibit bimodal activity (Figure 4). Importantly, the HU activity is not constrained by the model to take this shape, as dReLU potentials allow for several activity distributions (see Methods 7.5.4; “Choice of HU potential”). In fact, ReLU potentials are a special case of dReLU (by $(\gamma_{\mu, -} \to \infty)$), so our model allows HU potentials to behave like ReLUs, but in practice they converge to a double-well potential for almost all HUs, leading to bimodal activity distributions.

Following the suggestion of the reviewer, we have now added this detail for clarity in Methods 7.5.4 and referenced this Methods section at line 237.

Reviewer #1 (Significance (Required)):

van der Plas et al. highlighted a novel dimensionality reduction technique that can be used with success for discerning functional connectivities in large-scale single-unit recordings. The proposed model belongs to a large collection of dimensionality reduction techniques (for review, Cunningham, J., Yu, B. Dimensionality reduction for large-scale neural recordings. Nat Neurosci 17, 1500-1509 (2014). https://doi.org/10.1038/nn.3776; Paninski, L., & Cunningham, J. P. (2018). Neural data science: accelerating the experiment-analysis-theory cycle in large-scale neuroscience. Current opinion in neurobiology, 50, 232-241.). The authors themselves highlighted some of the key methods, such as PCA, ICA, NNMF, variational auto-encoders, etc. The proposed cRBM model has also been published a few times by the same authors in previous works, although specifically pertaining to protein sequences. The use of RBM-like methods in uncovering functional connectivities is not novel either (see Hjelm RD, Calhoun VD, Salakhutdinov R, Allen EA, Adali T, Plis SM. Restricted Boltzmann machines for neuroimaging: an application in identifying intrinsic networks. Neuroimage. 2014 Aug 1;96:245-60. doi: 10.1016/j.neuroimage.2014.03.048.). However, given that the authors make a substantial improvement on the RBM network and have demonstrated the value of their model using physiological data, I believe that this paper would present itself as an attractive alternative to all readers who are seeking better solutions to interpret their data. However, as I mentioned in my comments, I would like to see more definitive evidence that the proposed solution has a serious advantage over other equivalent methods.

Reviewer's expertise:

This review was conducted jointly by three researchers whose combined expertise includes single-unit electrophysiology and two-photon calcium imaging, using which our lab studies the neurobiology of learning and memory and spatial navigation. We also have extensive experience in computational neuroscience, artificial neural network models, and machine learning methods for the analysis of neurobiological data. We are however limited in our knowledge of mathematics and engineering principles. Therefore, our combined expertise is insufficient to evaluate the correctness of the mathematical developments.

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

In their manuscript, van der Plas et al. present a generative model of neuron-assembly interaction. The model is a restricted Boltzmann machine with its visible units corresponding to neurons and hidden units to neural assemblies. After fitting their model to whole-brain neural activity data from larval zebrafish, the authors demonstrate that their model is able to replicate several data statistics. In particular, it was able to replicate the pairwise correlations between neurons as well as assemblies that it was not trained on. Moreover, the model allows the authors to extract neural assemblies that govern the population activity and compose functional circuits and can be assigned to anatomical structures. Finally, the authors construct functional connectivity maps from their model that are then shown to correlate with established structural connectivity maps.

Overall, the authors present convincing evidence for their claims. Furthermore, the authors state that their code to train their restricted Boltzmann machine models is already available on GitHub and that the data underlying the results presented in this manuscript will be made publicly available upon publication, which will allow people to reproduce the results and apply the methods to their data.

One thing the authors could maybe discuss a bit more is the "right" parameter value M, especially since they used the optimal value of 200 found for one sample also for all the others. More specifically, how sensitive are the results to this value?

PLANNED REVISION #1

In the following we jointly address three of the reviewers’ questions (2 from reviewer 1, and 1 from reviewer 2).

Shortly summarized, the cRBM model has 2 free parameters; the number of hidden units M and the regularization parameter lambda. In figures 2 and S1 we optimize their values through cross-validation, and then perform our further analyses on models with these optimal values. The reviewers ask us to examine the outcome of the model for slightly different values of both parameters, in particular in relation to the sensitivity of the cRBM results to selecting the optimal parameters and the change in inferred assemblies and their dynamics.

We thank the reviewers for these questions and appreciate their curiosity to understand the effects of changing either of these two free parameters.

We inspected these models when we performed the model selection (of Figures 2 and S1), but did not formalize our findings into figures for the manuscript. We found that small changes in the parameter setting did not abruptly change the inferred assemblies (e.g., M=100) apart from slightly changing in size, so we expect that the statistics that we intend to include in the proposed supplementary figure would reflect that, and it would definitely benefit the manuscript to include this analysis. Very-low-M settings are interesting to include, because assemblies are much larger - essentially merging smaller assemblies of higher-M models - at the cost of model performance.

We propose to create additional supplementary figures that address these questions. As suggested, we will pick a few example cRBMs with different parameter settings (below-optimal M, above-optimal M, and same for lambda), as well as very low M settings (M~20 or 50). We will then show example assemblies and assembly dynamics, as well as the relevant statistics (assembly size, dynamics time scale etc) that describe them.

And, what happens if one would successively increase that number, would the number of assemblies (in the sense of hidden units that strongly couple to some of the visible units) eventually saturate?

This point will be addressed by inspecting models at different M values (see Revision Plan #1). We would like to further answer this question by referring to past work. In Tubiana et al., 2019, elife (Appendix 1) we have done this analysis, and the result is consistent with the reviewer’s intuition. Because of the sparsity regularization, if M becomes larger than its optimum, the assemblies further sparsify without benefiting model performance, and eventually new assemblies duplicate previous assemblies or become totally sparse (i.e., all weights = 0) to not further induce a sparsity penalty in the loss function. So the ‘effective’ number of assemblies indeed saturates for high M.

Moreover, regarding the presentation, I have a few minor suggestions and comments that the authors also might want to consider:

* In Figure 6C, instead of logarithmic axes, it might be better to put the logarithmic connectivity on a linear axis. This way the axes can be directly related to the colour bars in Figures 6A and B.

We agree and thank you for the suggestion. We have changed this accordingly (and also in the equivalent plots in figure S6).

* In Equation (8), instead of $\Gamma_{\mu}(I)$ it should be $\Gamma_{\mu}(I_{\mu}(v))$.

Done, thank you.

* In Section 7.0.5, it might make sense to have the subsection about the marginal distributions before the ones about the conditional distributions. The reason would be that if one wants to confirm Equation (8) one necessarily has to compute the marginal distribution in Equation (12) first.

We thank the reviewer for the suggestion, but respectfully propose to leave the section ordering as is. We understand what the reviewer means, but Equation (8) can also be obtained by factorizing P(v,h) Equation (7) and removing the v_i dependency. In Equation (8), \Gamma can then be obtained by normalization. We believe this flow aligns better with the main text (where conditionals come first, when used for sampling, followed by the marginal of P(v) used for the functional connectivity inference).

* In Line 647f, the operation the authors are referring to is strictly speaking not an L1-norm of the matrix block. It might be better to refer to that e.g. as a normalised L1-norm of the matrix block elements.

Done, thank you.

* In Line 22, when mentioning dimensionality reduction methods to identify assemblies, it might make sense to also reference the work by Lopes-dos-Santos et al. (2013, J. Neurosci. Methods 220).

Done, thank you for the suggestion.

Reviewer #2 (Significance (Required)):

The work presented in this manuscript is very interesting for two reasons. First, it has long been suggested that assemblies are a fundamental part of neural activity and this work seems to support that by showing that one can generate realistic whole-brain population activity imposing underlying assembly dynamics. Second, in recent years much work has been devoted to developing methods to find and extract neural assemblies from data and this work and the modelling approach can also be seen as a new method to achieve that. As such, I believe this work is relevant for anyone interested in neural population activity and specifically neural assemblies and certainly merits publication.

Regarding my field of expertise, I used to work on data analysis of neural population activity and in particular on the question of how one can extract neural assemblies from data. I have to say that I have not much experience with fitting statistical models to data, so I can't provide any in-depth comments on that part of the work, although what has been done seems plausible.

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

Summary: Understanding the organization of neural population activities in the brain is one of the most important questions in neuroscience. Recent technique advance has enabled researchers to record a large number of neurons and some times the whole brain. Interpreting and extracting meaningful insights from such data sets are challenging. van der Pals \textit{et al} applied a generative model called compositional Restricted Boltzmann Machine (cRBM) to discover neuron assemblies from spontaneous activities of zebra fish brain. They found that neurons can be grouped into around 200 assemblies. Many of them have clear neurophysiological meaning, for example, they are anatomically localized and overlapped with known neural circuits. The authors also inferred a coarse-grained functional connectivity which is similar to known structural connectivity.

The structure of the paper is well organized, the conclusion seems well supported by their numerical results. While this study provides a compelling demonstration that cRBM can be used to uncover meaningful structures from large neural recordings, the following issues limit my enthusiasm.

Major:

1) The overall implication is not clear to me. Although the authors mentioned this briefly in the discussion. It is not clear what else do we learn from discovered assemblies beyond stating that they are consist with previous study. For example, the author could have more analysis of the assembly dynamics, such as whether there are low dimensional structure etc.

First, we will comment on our analysis of the assemblies, before we continue to discuss the main implications of our work, which we believe are the inferred generative model of the zebrafish brain and the perturbation-based connectivity matrix that we discovered. Further, we have implemented the reviewer’s suggestion of analyzing the low-dimensional structure of the hidden unit activity, as further detailed in Question 7.

Indeed, the example assemblies that we show in Figure 3 have been thoroughly characterized in previous studies, which is why we chose to showcase these examples. Previous studies (including our own) typically focused on particular behaviors or sensory modalities, and aimed at identifying the involved neural circuit. Here, we demonstrate that by using cRBM on spontaneous activity recordings, one can simultaneously identify many of those circuits. In other words, these functional circuits/assemblies activate spontaneously, but in many different combinations and perhaps infrequently, so that it is very difficult to infer them from the full neural dynamics that they generate. cRBM has been able to do so, and Figure 3 (and supplementary video 1) serve to illustrate the variety of (known) circuits and assemblies that it inferred, some of which may represent true but not yet characterized circuits, which thus provide hypotheses for subsequent studies.

Further, we believe that the implication of our study goes beyond the properties of the assemblies we’ve identified, in several ways.

We demonstrate the power of cRBM’s generative capacity for inferring low-dimensional representations in neuroscience. Unlike standard dimensional reductionality methods, generative models can be assessed by comparing the statistics of experimental vs in-silico generated data. This is a powerful approach to validate a model, rarely used in neuroscience because of the scarcity of generative models compatible with large-scale data, and we hope that our study will inspire the use of this method in the field. We have made our cRBM code available, including notebook tutorials, to facilitate this.

The generative aspect of our model allowed us to predict the effect of single-neuron perturbations between all ~ 10^9 pairs of neurons per fish, resulting in a functional connectivity matrix.

We believe that the functional connectivity matrix is a major result for the field, similar to the structural connectivity matrix from Kunst et al., 2019, Neuron. The relation between functional and structural connectivity is unknown and of strong interest to the community (e.g., Das & Fiete, 2020, Nature Neuroscience). Our results allowed for a direct comparison of whole-brain region-by-region structural and functional connectivity. We were thus able to quantify the similarity between these two maps, and to identify specific region-pair matches and non-matches of functional and structural connectivity - which will be of particular interest to the zebrafish neuroscience community for developing future research questions.

Further, using these trained models - that will be made public upon publication - anyone can perform any type of in silico perturbation experiments, or generate endless artificial data with matching data statistics to the in vivo neural recordings.

We hope that this may convince the reviewer of the multiple directions of impact of our study. We will further address their comment on analysis of assembly dynamics below (question 7).

2) The learning algorithm of cRBM can be interpreted as matching certain statistics between the model and the experiment. For a general audience, it is not easy to understand $\langle h_{\mu} \rangle_{data}, \langle v_ih_{\mu}\rangle_{data}$. Since these are not directly calculated from experimental observed activities $v_i$, but rather the average is conditioned on the empirical distribution of $p(v_i)$. For example, the meaning of $\langle v_ih_{\mu}\rangle_{data}$ means

\begin{equation}

\langle v_ih_{\mu}\rangle_{data} = \frac{1}{l}\sum_{\mathbf{v}\in S} \mathbb{E}{p(\mathbf{h|v})}(v_ih{\mu}),

\end{equation}

where $S$ is the set of all observed neural activities: $S = {\mathbf{v}^1, \cdots, \mathbf{v}^l}$. The authors should explain this in the main text or method, since they are heavily loaded in figure 2.

We thank the reviewer for their suggestion, and have now implemented this. Their mathematics are correct; and we agree that it is not easy to understand without going through the full derivation of the (c)RBM. At the same time, we have tried not to alienate readers who might be more interested in the neuroscience findings than in understanding the computational method used. Therefore, we have kept mathematical details in the main text to a minimum (and have used schematics to indicate the statistics in Figures 2C-G), while explaining it in detail in Methods.

Accordingly, we have now extended section 7.10.2 (“Assessment of data statistics”) that explains how the data statistics were computed in Methods (and have referenced this in Results and in Methods 7.5.5), using the fact that we already explain the process of conditioning on __v __in Methods 7.5.1. The following sentences were added:

---

[..] However, because (c)RBM learn to match data statistics to model statistics (see Methods

7.5.5), we can directly compare these to assess model performance. [..]

[..]

For each statistic 〈 fk〉 we computed its value based on empirical data 〈 fk〉_data and on the model 〈 fk〉_model, which we then quantitatively compared to assess model performance. Data statistics 〈 fk〉_data were calculated on withheld test data (30% of recording). Naturally, the neural recordings consisted only of neural data v and not of HU data h. We therefore computed the expected value of __ht at each time point t conditioned on the empirical data _v_t, as further detailed in Methods 7.5.1.

[..]

---

3) As a modeling paper, it would be great to have some testable predictions.

We thank the reviewer for the enthusiasm and suggestion. We agree, and that is why we have included this in the form of functional connectivity matrices in Figures 5 and 6. To achieve this, we leveraged the generative aspect of the cRBM to perform in silico single-neuron perturbation experiments, which we aggregated to connectivity matrices. In other words, we have used our model to predict the functional connectivity between brain regions using the influence of single-neuron perturbations.

Obtaining a measure of functional connectivity/influence using single-neuron perturbations is also possible using state-of-the-art neuro-imaging experiments (e.g., Chettih & Harvey, 2019, Nature), though not at the scale of our in silico experiments. We therefore verify our predictions using structural data from Kunst et al., 2019, which we have extended substantially. We provide our functional connectivity result in full, and hope that this can inspire future zebrafish research by predicting which regions are functionally connected, which includes many pairs of regions that have not yet directly been studied in vivo.

Minor:

1) The assembly is defined by the neurons that are strongly connected with a given hidden unit. Thus, some neurons may enter different assemblies. A statistics of such overlap would be helpful. For example, a ven diagram in figure 1 that shows how many of them assigned to 1, 2, etc assemblies.

We thank the reviewer for this excellent suggestion. Indeed, neurons can be embedded in multiple assemblies. This is an important property of cRBMs, which deserves to be quantified in the manuscript. We have now added this analysis as a new supplementary figure 4. Neurons are embedded in an assembly if their connecting weight w_{i, \mu} is ‘significantly’ non-zero, depending on what threshold one uses. We have therefore shown this statistic for 3 values of the threshold (0.001, 0.01 and 0.1) - demonstrating that most neurons are strongly embedded in at least 1 assembly and that many neurons connect to more than 1 assembly.

Updated text in Results:

---

Further, we quantified the number of assemblies that each neuron was embedded in, which showed that increasing the embedding threshold did not notably affect the fraction of neurons embedded in at least 1 assembly (93% to 94%, see Figure S4).

---

2) What does the link between hidden units in Figure 1B right panel mean?

Thank you for the question, and we apologize for the confusion: if we understand the question right, the reviewer asks why the colored circles under the title ‘Neuronal assemblies of Hidden Units’ are linked. This schematic shows the same network of neurons as shown in gray at the left side of Fig 1B, but now colored by the assembly ‘membership’ of each neuron. Hence, the circles shown are still neurons (and not HUs), and their links still represent synaptic connections between neurons. We apologize for the confusion, and have updated the caption of Fig 1B to explain this better:

---

“[..] The neurons that connect to a given HU (and thus belong to the associated assembly), are depicted by the corresponding color labeling (right panel).[..]”.

---

3) A side-by-side comparison of neural activity predicted by model and the experimentally recorded activities would help the readers to appreciate the performance of the model. Such comparison can be done at both single neuron level or assembly level.

We thank the reviewer for this suggestion. The cRBM model is a statistical model, meaning that it fits the statistics of the data, and not the dynamics. The data that it generates therefore (should) adhere to the statistics of the training data, but does not reflect their dynamics. We believe that showing generated activity side-by-side of empirical activity is therefore not a meaningful example of generated data, as this would exemplify the dynamics, which this model is not designed to capture. Instead, in Figure 2, we show the statistics of the generated data versus the statistics of the empirical data (e.g., Fig 2C for the mean activity of all neurons). We believe that this is a better example representation of the generative performance of the model.

4) Definition of reconstruction quality in line 130.

We thank the reviewer for the suggestion, and have added the following sentence after line 130:

---

The reconstruction quality is defined as the log-likelihood of reconstructed neural data v___{recon} (i.e., __v that is first transformed to the low-dimensional h, and then back again to the high-dimensional __v___{recon}, see Methods 7.10.2).

---

Further, please note that Methods describes the definition in detail (Eq 18 of the submitted manuscript), although we agree with the reviewer that more detail was required in the Results section at line 130.

5) Line 165. If PCA is compared with cRBM, why other dimensionality reduction methods, such as k-means and non-negative matrix factorization, can not be compared in terms of the sparsity?

Please see answer to question 1 from R1 and Revision Plan #2.

6) Line 260, please provide minimum information about how the functional connectivity is defined based on assemblies discovered by cRBM.

We apologize if this was not clear. The first paragraph of this section (lines 248-259) of the submitted manuscript, provides the detail that the reviewer asks for, and we realize that the sentence of line 260 is better placed in the first paragraph, as it has come across as a very minimal explanation of how functional connectivity is defined.

We have now moved this sentence to the preceding paragraph, as well as specified the Method references (as suggested by this reviewer below), for additional clarity. We thank the reviewer for pointing out this sentence.

7) Some analysis of the hidden units population activities. Such as whether or not there are interesting low dimensional structure from figure 4A.

We thank the reviewer for their suggestion. In our manuscript we have used the cRBM model to create a low-dimensional (M=200) representation of zebrafish neural recordings (N=50,000). The richness of this model owes to possible overlaps between HUs/assemblies that can result in significant correlation in their activities. The latter is illustrated in Figure 4A-C: the activity of some HUs can be strongly correlated.

The reviewer’s suggestion is similar; to perform some form of dimensionality reduction on the low-dimensional HU activity shown in Fig 4. We have now added a PCA analysis to Figure 4 to quantify the degree of low-dimensional structure in the HU dynamics, and show the results in a new panel Figure 4D.

The following text has been added to the Results section:

---

These clusters of HUs with strongly correlated activity suggest that much of the HU variance could be captured using only a small number of variables. We quantified this by performing PCA on the HU dynamics, finding that indeed 52% of the variance was captured by the first 3 PCs, and 85% by the first 20 PCs (Figure 4D).

---

We believe that further visualization of these results, such as plotting the PC trajectories, would not further benefit the manuscript. The manuscript focuses on cRBM, and the assemblies/HUs it infers. Unlike PCA, these are not ranked/quantified by how much variance they explain individually, but rather they together ‘compose’ the entire system and explain its (co)variance (Figure 2). Breaking up a dominant activity mode (as found by PCA), such as the ARTR dynamics, into multiple HUs/assemblies, allows for some variation in activity of individual parts of the ARTR circuit (such as tail movement and eye movement generation), even though at most times the activity of these HUs is coordinated. We hope the reviewer agrees with our motivation to keep the manuscript focused on the nature of cRBM-inferred HUs.

8) Figure 4B right panel, how did the authors annotate the cluster manually? As certain assembly may overlap with several different brain regions, for example, figure 4D.

We thank the reviewer for this question, and we presume they meant to reference figure 3D as an example? For figure 4, as well as Figure 3, we used the ZBrain Atlas (Randlett et al., 2015) for definition of brain regions. This atlas presents a hierarchy of brain regions: for example, many brain regions are part of the rhombencephalon/hindbrain. This is what we used for midbrain/hindbrain/diencephalon. Further, many assemblies are solely confined to Optic Tectum (see Fig 3L), which we therefore used (split by hemisphere). Then, many brain regions are (partly) connected to the ARTR circuit, such as the example assembly of Figure 3D that the reviewer mentions. These we have all labeled as ARTR (left or right), though technically only part of their assembly is the ARTR. These two clusters therefore rather mean ‘ARTR-related’, in particular because their activity is locked to the rhythm of the ARTR (see Fig 4A). The final category is ‘miscellaneous’ (like Figure 3G).

However we agree that this wasn’t clear from the manuscript text, so we have changed the figure 4C caption to mention that ‘ARTR’ stands for ARTR-related assemblies, which we hope clarifies that ARTR-clustered assemblies can exist of multiple, disjoint groups of neurons, which relate to the ARTR circuit.

9) Better reference of the methods cited in the main text. The method part is quite long, it would be helpful to cite the section number when referring it in the main text.

We thank the reviewer for this helpful suggestion, we agree that it would benefit the manuscript to reference specific sections of the Methods. We have now changed all references to Methods to incorporate this.

10) Some discussion about the limitation of cRBM would be great.

We thank the reviewer for this suggestion, and have now included this. As Reviewer 1 had the same suggestion, we refer our answer to questions 2 and 3 from R1 for more detail.

Reviewer #3 (Significance (Required)):

This work provides a timely new technique to extract meaningful neural assemblies from large scale recordings. This study should be interested to both researchers doing either experiments and computation/theory. I am a computational neuroscientist.

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:

Understanding the organization of neural population activities in the brain is one of the most important questions in neuroscience. Recent technique advance has enabled researchers to record a large number of neurons and some times the whole brain. Interpreting and extracting meaningful insights from such data sets are challenging. van der Pals \textit{et al} applied a generative model called compositional Restricted Boltzmann Machine (cRBM) to discover neuron assemblies from spontaneous activities of zebra fish brain. They found that neurons can be grouped into around 200 assemblies. Many of them have clear neurophysiological meaning, for example, they are anatomically localized and overlapped with known neural circuits. The authors also inferred a coarse-grained functional connectivity which is similar to known structural connectivity.

The structure of the paper is well organized, the conclusion seems well supported by their numerical results. While this study provides a compelling demonstration that cRBM can be used to uncover meaningful structures from large neural recordings, the following issues limit my enthusiasm.

Major:

1. The overall implication is not clear to me. Although the authors mentioned this briefly in the discussion. It is not clear what else do we learn from discovered assemblies beyond stating that they are consist with previous study. For example, the author could have more analysis of the assembly dynamics, such as whether there are low dimensional structure etc.
2. The learning algorithm of cRBM can be interpreted as matching certain statistics between the model and the experiment. For a general audience, it is not easy to understand $\langle h_{\mu} \rangle_{data}, \langle v_ih_{\mu}\rangle_{data}$. Since these are not directly calculated from experimental observed activities $v_i$, but rather the average is conditioned on the empirical distribution of $p(v_i)$. For example, the meaning of $\langle v_ih_{\mu}\rangle_{data}$ means \begin{equation} \langle v_ih_{\mu}\rangle_{data} = \frac{1}{l}\sum_{\mathbf{v}\in S} \mathbb{E}{p(\mathbf{h|v})}(v_ih{\mu}), \end{equation} where $S$ is the set of all observed neural activities: $S = {\mathbf{v}^1, \cdots, \mathbf{v}^l}$. The authors should explain this in the main text or method, since they are heavily loaded in figure 2.
3. As a modeling paper, it would be great to have some testable predictions.

Minor:

1. The assembly is defined by the neurons that are strongly connected with a given hidden unit. Thus, some neurons may enter different assemblies. A statistics of such overlap would be helpful. For example, a ven diagram in figure 1 that shows how many of them assigned to 1, 2, etc assemblies.
2. What does the link between hidden units in Figure 1B right panel mean?
3. A side-by-side comparison of neural activity predicted by model and the experimentally recorded activities would help the readers to appreciate the performance of the model. Such comparison can be done at both single neuron level or assembly level.
4. Definition of reconstruction quality in line 130.
5. Line 165. If PCA is compared with cRBM, why other dimentionality reduction methods, such as k-means and non-negative matrix factorization, can not be compared in terms of the sparsity?
6. Line 260, please provide minimum information about how the functional connectivity is defined based on assemblies discovered by cRBM.
7. Some analysis of the hidden units population activities. Such as whether or not there are interesting low dimensional structure from figure 4A.
8. Figure 4B right panel, how did the authors annotate the cluster manually? As certain assembly may overlap with several different brain regions, for example, figure 4D.
9. Better reference of the methods cited in the main text. The method part is quite long, it would be helpful to cite the section number when referring it in the main text.
10. Some discussion about the limitation of cRBM would be great.

Significance

This work provides a timely new technique to extract meaningful neural assemblies from large scale recordings. This study should be interested to both researchers doing either experiments and computation/theory. I am a computational neuroscientist.

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 their manuscript, van der Plas et al. present a generative model of neuron-assembly interaction. The model is a restricted Boltzmann machine with its visible units corresponding to neurons and hidden units to neural assemblies. After fitting their model to whole-brain neural activity data from larval zebrafish, the authors demonstrate that their model is able to replicate several data statistics. In particular, it was able to replicate the pairwise correlations between neurons as well as assemblies that it was not trained on. Moreover, the model allows the authors to extract neural assemblies that govern the population activity and compose functional circuits and can be assigned to anatomical structures. Finally, the authors construct functional connectivity maps from their model that are then shown to correlate with established structural connectivity maps.

Overall, the authors present convincing evidence for their claims. Furthermore, the authors state that their code to train their restricted Boltzmann machine models is already available on GitHub and that the data underlying the results presented in this manuscript will be made publicly available upon publication, which will allow people to reproduce the results and apply the methods to their data.

One thing the authors could maybe discuss a bit more is the "right" parameter value M, especially since they used the optimal value of 200 found for one sample also for all the others. More specifically, how sensitive are the results to this value? And, what happens if one would successively increase that number, would the number of assemblies (in the sense of hidden units that strongly couple to some of the visible units) eventually saturate?

Moreover, regarding the presentation, I have a few minor suggestions and comments that the authors also might want to consider: - In Figure 6C, instead of logarithmic axes, it might be better to put the logarithmic connectivity on a linear axis. This way the axes can be directly related to the colour bars in Figures 6A and B. - In Equation (8), instead of $\Gamma_{\mu}(I)$ it should be $\Gamma_{\mu}(I_{\mu}(v))$. - In Section 7.0.5, it might make sense to have the subsection about the marginal distributions before the ones about the conditional distributions. The reason would be that if one wants to confirm Equation (8) one necessarily has to compute the marginal distribution in Equation (12) first. - In Line 647f, the operation the authors are referring to is strictly speaking not an L1-norm of the matrix block. It might be better to refer to that e.g. as a normalised L1-norm of the matrix block elements. - In Line 22, when mentioning dimensionality reduction methods to identify assemblies, it might make sense to also reference the work by Lopes-dos-Santos et al. (2013, J. Neurosci. Methods 220).

Significance

The work presented in this manuscript is very interesting for two reasons. First, it has long been suggested that assemblies are a fundamental part of neural activity and this work seems to support that by showing that one can generate realistic whole-brain population activity imposing underlying assembly dynamics. Second, in recent years much work has been devoted to developing methods to find and extract neural assemblies from data and this work and the modelling approach can also be seen as a new method to achieve that. As such, I believe this work is relevant for anyone interested in neural population activity and specifically neural assemblies and certainly merits publication.

Regarding my field of expertise, I used to work on data analysis of neural population activity and in particular on the question of how one can extract neural assemblies from data. I have to say that I have not much experience with fitting statistical models to data, so I can't provide any in-depth comments on that part of the work, although what has been done seems plausible.

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 present manuscript, van der Plas et al. compellingly illustrated a novel technique for engendering a whole-brain functional connectivity map from single-unit activities sampled through a large-scale neuroimaging technique. With some clever tweaks to the restricted Boltzmann Machine, the cRBM network is able to learn a low-dimensional representation of population activities, without relying on constrained priors found in some traditional methods. Notably, using some 200 hidden layer neurons, the employed model was able to capture the dynamics of over 40,000 simultaneously imaged neurons with a high degree of accuracy. The extracted features both illustrate the anatomical topography/connectivities and capture the temporal dynamics in the evolution of brain states. The illustrated technique has the potential for wide-spread applications spanning diverse recording techniques and animal species. Furthermore, the prospectives of modeling whole-brain network dynamics in 'neural trajectory' space and of generating artificial data in silico make for very enticing reasons to adopt cRBM.

Major comments:

Line 164. The authors claim that conventional methods "such as k-means, PCA and non-negative matrix factorization" cannot be quantitatively assessed for quality on the basis that they are unable to generate new artificial data. Though partly true, in most neuroscience applications, this is hardly cause for concern. Most dimensionality reduction methods (with few exceptions such as t-sne) allow new data points to be embedded into the reduced space. As such, quality of encoding can be assessed by cross-validation much in the same way as the authors described, and quantified using traditional metrics such as percentage explained variance. The authors should directly compare the performance of their proposed model against that of NNMF and variational auto-encoders. Doing so would offer a more compelling argument for the advantage of their proposed method over more widely-used methods in neuroscience applications. Furthermore, a direct comparison with rastermap, developed by Stringer lab at Janelia (https://github.com/MouseLand/rastermap), would be a nice addition. This method presents itself as a direct competitor to cRBM. Additionally, the use of GLM doesn't do complete justice to the comparison point used, since a smaller fraction of data were used for calculating performance using GLM, understandably due to its computationally intensive nature.

Line 26. The authors describe their model architecture as a formalization of cell assemblies. Cell assemblies, as originally formulated by Hebb, pertains to a set of neurons whose connectivity matrix is neither necessarily complete nor symmetric. Critically, in the physiological brain, the interactions between the individual neurons that are part of an assembly would occur over multiple orders of dependencies. In a restricted Boltzmann machine, neurons are not connected within the same layer. Instead, visible layer neurons are grouped into "assemblies" indirectly via a shared connection with a hidden layer neuron. Furthermore, a symmetrical weight matrix connects the bipartite graph, where no recurrent connectivities are made. As such, the proposed model still only elaborates symmetric connections pertaining to first-order interactions (as illustrated in Figure 4C). Such a network may not be likened with the concept of cell assemblies. The authors should refrain from detailing this analogy (of which there are multiple instances of throughout the text). It is true that many authors today refer to cell assemblies as any set of temporally-correlated neurons. However, saying "something could be a cell assembly" is not the same as saying "something is a cell assembly". How about sticking with cRBM-based cell assemblies (as used in section 2.3) and defining it beforehand?

I would strongly recommend adding a paragraph discussing the limitation of using the cRBM, things future researchers need to keep in mind before using this method. One such recommendation is moving the runtime-related discussion for cRBM, i.e. 8-12 hrs using 16 CPU from Methods to Discussion, since it's relevant for an algorithm like this. Additionally, a statement mentioning how this runtime will increase with the length of recordings and/or with the number of neurons might be helpful. What if the recordings were an hour-long rather than 25mins. This would help readers decide if they can easily use a method like this.

Line 515. A core feature of the proposed compositional RBM is the addition of a soft sparsity penalty over the weight matrix in the likelihood function. The authors claim that "directed graphical models" are limited by the a priori constraints that they impose on the data structure. Meanwhile, a more accurate statistical solution can be obtained using a RBM-based model, as outlined by the maximum entropy principle. The problem with this argument is that the maximum entropy principle no longer applies to the proposed model with the addition of the penalty term. In fact, the lambda regularization term, which was estimated from a set of data statistics motivated by the experimenter's research goals (Figure S1), serves to constrict the prior probability. Moreover, in Figure S1F, we clearly see that reconstruction quality suffers with a higher penalty, suggesting that the principle had indeed been violated. That being said, RBMs are notoriously hard to train, possibly due to the unconstrained nature of the optimization. I believe that cRBM can help bring RBM into wider practical applications. The authors could test their model on a few values of the free parameter and report this as a supplementary. I believe that different parameters of lambda could elaborate on different anatomical clusters and temporal dynamics. Readers who would like to implement this method for their own analysis would also benefit tremendously from an understanding of the effects of lambda on the interpretation of their data. Item (1) on line 35 (and other instances throughout the text) should be corrected to reflect that cRBM replaces the hard constraints found in many popular methods with a soft penalty term, which allows for more accurate statistical models to be obtained.

Minor comments:

From a neuroscience point of view, it might be interesting to show what results are achieved using different values of M (say 100 or 300), rather than M=200, while still maintaining the compositional phase. Is there any similarity between the cRBM-based cell assemblies generated at different values of M? Is there a higher chance of capturing certain dynamics either functional or structural using cRBM? For example, did certain cRBM-based cell assemblies pop up more frequently than others at all values of M (100,200,300)?

The authors have mentioned that this approach can be readily applied to data obtained in other animal models and using different recording techniques. It might be nice to see a demonstration of that.

Line 237. The justification for employing a dReLU transfer function as opposed to ReLU is unclear, at least within the context of neurobiology. Given that this gives rise to a bimodal distribution for the activity of HUs, the rationale should be clearly outlined to facilitate interpretability.

Significance

van der Plas et al. highlighted a novel dimensionality reduction technique that can be used with success for discerning functional connectivities in large-scale single-unit recordings. The proposed model belongs to a large collection of dimensionality reduction techniques (for review, Cunningham, J., Yu, B. Dimensionality reduction for large-scale neural recordings. Nat Neurosci 17, 1500-1509 (2014). https://doi.org/10.1038/nn.3776; Paninski, L., & Cunningham, J. P. (2018). Neural data science: accelerating the experiment-analysis-theory cycle in large-scale neuroscience. Current opinion in neurobiology, 50, 232-241.). The authors themselves highlighted some of the key methods, such as PCA, ICA, NNMF, variational auto-encoders, etc. The proposed cRBM model has also been published a few times by the same authors in previous works, although specifically pertaining to protein sequences. The use of RBM-like methods in uncovering functional connectivities is not novel either (see Hjelm RD, Calhoun VD, Salakhutdinov R, Allen EA, Adali T, Plis SM. Restricted Boltzmann machines for neuroimaging: an application in identifying intrinsic networks. Neuroimage. 2014 Aug 1;96:245-60. doi: 10.1016/j.neuroimage.2014.03.048.). However, given that the authors make a substantial improvement on the RBM network and have demonstrated the value of their model using physiological data, I believe that this paper would present itself as an attractive alternative to all readers who are seeking better solutions to interpret their data. However, as I mentioned in my comments, I would like to see more definitive evidence that the proposed solution has a serious advantage over other equivalent methods.

Reviewer's expertise:

This review was conducted jointly by three researchers whose combined expertise includes single-unit electrophysiology and two-photon calcium imaging, using which our lab studies the neurobiology of learning and memory and spatial navigation. We also have extensive experience in computational neuroscience, artificial neural network models, and machine learning methods for the analysis of neurobiological data. We are however limited in our knowledge of mathematics and engineering principles. Therefore, our combined expertise is insufficient to evaluate the correctness of the mathematical developments.