Differential regulation of the proteome and phosphosproteome along the dorso-ventral axis of the early Drosophila embryo

  1. Juan Manuel Gomez
  2. Hendrik Nolte
  3. Elisabeth Vogelsang
  4. Bipasha Dey
  5. Michiko Takeda
  6. Girolamo Giudice
  7. Miriam Faxel
  8. Alina Cepraga
  9. Robert Patrick Zinzen
  10. Marcus Krüger
  11. Evangelia Petsalaki
  12. Yu-Chiun Wang
  13. Maria Leptin

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

The initially homogeneous epithelium of the early Drosophila embryo differentiates into regional subpopulations with different behaviours and physical properties that are needed for morphogenesis. The factors at top of the genetic hierarchy that control these behaviours are known, but many of their targets are not. To understand how proteins work together to mediate differential cellular activities, we studied in an unbiased manner the proteomes and phosphoproteomes of the three main cell populations along the dorso-ventral axis during gastrulation using mutant embryos that represent the different populations. We detected 6111 protein groups and 6259 phosphosites of which 3399 and 3433 respectively, were differentially regulated. The changes in phosphosite abundance did not correlate with changes in host protein abundance, showing phosphorylation to be a regulatory step during gastrulation. Hierarchical clustering of protein groups and phosphosites identified clusters that contain known fate determinants such as Doc1, Sog, Snail and Twist. The recovery of the appropriate known marker proteins in each of the different mutants we used validated the approach, but also revealed that two mutations that both interfere with the dorsal fate pathway, Toll 10B and serpin27a ex do this in very different manners. Diffused network analyses within each cluster point to microtubule components as one of the main groups of regulated proteins. Functional studies on the role of microtubules provide the proof of principle that microtubules have different functions in different domains along the DV axis of the embryo.

Article activity feed

  1. Review Commons

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

    Learn more at Review Commons

    Reply to the reviewers

    General response of the authors to the editor and the reviewers:

    We thank the reviewers for their feedback, input and questions as these have helped us to (hopefully) improve the manuscript. We have rewritten several sections of the manuscript, moved methodological descriptions from the Results to the Methods section, and added imaging data for two cytoskeletal proteins, Shot and Cofilin/Twinstar, which confirm the predicted differential DV expression. Because the changes to the text were extensive, we did not mark them by track changes (the manuscript would have been illegible), but would be happy to provide an additional version that includes the tracked changes.

    We provide below the point-by-point response to each question and comment made by the reviewers. Our text is in blue.

    __Reviewer #1 __

    __Evidence, reproducibility and clarity __

    __Summary __

    This manuscript investigated changes in the proteome and phosphoproteome during dorsovental axis specification in the Drosophila embryo. To model the three regions in the embryo that are relevant for DV axis development, the authors used specific mutations to enrich for a single type of cells (ventral, lateral, or dorsal). The detected proteins and phosphopeptides were clustered according to the region of expression. There were differences between the protein and corresponding phosphopeptide abundance, suggesting that phosphorylation is a regulatory modification in DV axis establishment. Two different mutations that both result in a ventralized phenotype were found to change marker protein expression in different ways. Using inhibition of microtubule polymerization, this study also investigated the role of microtubules in epithelial folding.

    __Major comments __

    1. Generally, there is a lack of significance testing throughout the manuscript. Simply reporting fold changes can be misleading, if these changes are not significant. Examples:
    • Rigor of the proteomics evidence showing changes for the expected markers is insufficient because no statistical evaluation is provided. Specifically, in Fig. 1D and Suppl Fig 2: are the fold changes statistically significant?
    • Data in Fig. 4F, 5F need to be assessed for significance. There are other instances in the manuscript where significance should be tested.

    We did ANOVA testing for all proteome and phosphoproteome data, and the outcome of these analyses is reported in Supplementary Tables 2 and 3. We have added references to significance throughout, wherever possible and relevant and have included a table that summarizes all p values for all comparisons in all of the figures (Supplementary Table 2). However, note that we do our clustering independent of statistical significance, i.e., we include all values, as we explain in the manuscript.

    It is difficult to see the value of the obtained dataset for the community, in part because the data are analyzed by a linear model and cluster assignment developed by the authors, which is a somewhat arbitrary representation. Perhaps the authors could explain how their data could be used by other researchers, and maybe even develop an accessible portal for interacting with the data.

    We do provide the entire set of data in a formatted Excel Table as Supplementary Tables 3 and 4, which contain common pairwise comparisons and ANOVA tests that allow a researcher without a strong proteomics background to explore the data, and we also provide the raw proteomics datasets deposited in PRIDE, so any interested colleague can re-analyse them in the manner that suits their purposes best.

    We analysed the data in the way we did because it takes account of the knowledge from genetics that we have of all these cell populations. This also allowed us to include the important set of proteins and phosphosites that are completely absent from all but one mutant genotype, and would therefore have dropped out of the statistical analyses.

    For example, what does it mean biologically that a protein is a member of a specific cluster shown in Fig. 3C? Is there a predictive value in such an assignment, and how does it relate to the main question of DV axis regulation? An example of a novel insight obtained for specific protein(s) would be useful to illustrate the utility of this analysis.

    The clusters represent groups of proteins that are present at higher or lower abundance in subsets of cell populations. So, for example, being present in cluster 5 means (Fig. 3C) that this protein is predicted to be more abundant in the mesoderm than elsewhere (which includes being detected ONLY in the mesoderm, like Snail). This clustering therefore is the way for us to find new proteins that conform to these groups.

    We provide here the immunostainings of two cytoskeleton-associated proteins that our proteomic analyses predicted to be more abundant in the ectoderm (Cluster 6: dorsal+lateral):

    • The actin-microtubule crosslinker Short-stop (Shot), which is seen to be reduced in the mesoderm.
    • The actin-severing protein Cofilin/Twinstar, which was also found downregulated in the mesoderm in the work cited in Ref.:10 Gong L. et al., Development (2004). The staining shows that cofilin-GFP is abundant in the entire subapical region of ectodermal cells, but strongly reduced in ventral furrow cells, where it is only retained in a few apical membrane blebs. These proteins are targets for functional analyses in follow-up work.

    [Imaging Data for Reviewers]

    Figure: Physical cross-sections of fixed embryos showing the enrichment of proteins in the ectoderm (cluster 6: DL). Dorsal is top, ventral is bottom. Scale bar: 50 um Top panel: Staining for short-stop (shot; cyan / grayscale) and snail (yellow) in embryos expressing gap43-mCherry. Bottom panel: staining for discs large (dlg, magenta) and GFP (green / grayscale) in embryos expressing cofilin-GFP (Kyoto protein trap for Cofilin/Twinstar).

    Overall, at present the study appears to have limited novelty and mechanistic insight. The data generally align with prior expectations, but it is unclear how this work advances the field.

    We were reassured that the data align with previous studies, but as we state in the text, they go well beyond these valuable and important studies in several dimensions. We had made the following assumptions:

    1. DV patterning mutants recapitulate biological qualities of DV cell populations and the differential expression of DV fate determinants, as confirmed in Fig. 1 and Fig. 3D.
    2. The differential regulation of the proteomes and phosphoproteomes across DV patterning mutants recapitulates the abundances of proteins and phosphosites within DV cell populations of a wildtype embryo. We confirmed this in Fig. 3A and Fig. 5C with the implementation of a linear model for the abundances of detected proteins and phosphosites. The resulting analysis revealed new avenues for future functional studies, as intended. Most of the work on cell shape regulation at the gastrulation stage has focused on actomyosin and a subset of cell adhesion molecules. We have identified networks of proteins and phosphoproteins that may also control gastrulation (Fig. 6 and Supplementary Fig. 5), including microtubules, which were significantly enriched in networks of phosphoproteins (Fig. 7 and Supplementary Fig. 6).

    For example, the observed differences between marker proteins in Toll10B vs. spn27A data seem to confirm previous suggestions that spn27A has a stronger ventralizing effect.

    This suggestion was made by colleagues who had unpublished observations on a limited number of gene expression patterns that supported their contention. A correlation analysis (see figure below) of our results now shows that proteins with a restricted dorso-ventral pattern change more in spn27Aex mutants than in Toll10B. If we look at the known mesodermal genes such as Snail, Twist, Mdr49 and CG4500 we find them at higher abundance in *spn27Aex *than Toll10B , while the ectodermal genes Egr, Zen, Dtg, Tsg, Bsk, and Ptr are reduced more strongly in spn27Aex than in Toll10B. This takes the prior observation of a stronger ventralization of spn27Aex from an anecdotal to a systematic analysis.

    [Correlation analyses available for reviewers]

    Cross-correlation between the fold changes (FCs) in Toll10B/WT vs. spn27Aex/WT for all proteins detected in wildtype, Toll10B and spn27Aex. Each dot is a protein. The green line is the 'identity' function (slope = 1) that would be expected if the FCs for each protein in both ventralized mutants were exactly the same. A set of proteins with restricted dorso-ventral distribution are highlighted in yellow: mesodermal (ventral) and blue: ectodermal (dorsal).

    The role of microtubules in epithelial folding in the embryo has also been demonstrated before.

       The role of microtubules in epithelial folding in the *Drosophila *embryo has indeed been examined in three previous studies that studied dorsal fold formation (Ref.: 35, Takeda et al. NCB 2018), ventral furrow formation (VFF, Ref.: 36, Ko et al. JCB 2019), and salivary gland invagination (Booth et al. Dev Cell 2014). These data reveal diverse and non-conservative functional requirements, ranging from acto-myosin contractility during apical constriction (Booth et al. 2014), force transmission and repair of the supracellular contractile network (but not apical constriction per se, Ko et al 2019), to the generation of expansile forces during cell shape homeostasis (Takeda et al 2018). In light of this potentially broad functional spectrum, we sought to compare three epithelial folds that form within the context of gastrulation: ventral furrow, cephalic furrow and dorsal folds. We confirmed that the initiation of VFF was normal, but the final invagination failed, as per Ko et al. 2019, while dorsal fold initiation did not occur (extending conclusions from Takeda et al 2018). In contrast, cephalic furrow formation, though delayed, did not require microtubules. We also revealed a novel commonality of MT function. Specifically, prior to the initiation of all three epithelial folds, proper nuclear positioning requires MTs. We additionally discovered novel membrane abnormalities in two distinct types of blebs during ventral furrow and dorsal fold formation, respectively. Thus, our data provide insights into the roles of microtubules during epithelial folding that go beyond prior work.

    The shown phosphorylation changes (if they are significant) for Toll and Cactus are difficult to explain. In Suppl Fig 2B, E: why is Toll more phosphorylated in the lateralized than in ventralized embryos? (the provided reference 20 does not seem to clarify this).

       These changes are indeed significant (Toll-S871: Vtl vs. WT p = 0.01 , Vsp vs. WT p = 0.002; Cactus-S463: Vsp vs WT p = 0.03); see Supplementary Figure 2B and Supplementary Table 2).

   We have corrected Ref. 20 (Shen B. and Manley J.L., Development 1998). Ref. 20 only shows that Tl is phosphorylated by Pelle (Ref 20: Fig. 6A), although neither the exact position of Tl phosphosite(s) nor the function of Tl phosphorylation were explored in this article. A hallmark of Toll Like Receptor (TLR) regulation is these receptors are subject to tyrosine phosphorylation, which has been widely connected to the regulation of the binding of adaptor proteins to the cytoplasmic tail of TLRs. Both our finding of Serine phosphorylation in Tl, and the differential phosphorylation across cell populations is new, but since we do not know what this particular Serine phosphorylation site does in TLRs in general, we cannot speculate on the meaning of it occurring more in lateral than in ventral cells. In Ref. 20, the authors speculate that Tl phosphorylation by Pelle regulates the association between Tl and Pelle, which then enables Dorsal translocation to the nucleus. It might also be part of a feedback regulation loop, but this is entirely speculative.

    Also, certain Cactus phosphorylations appear higher in dorsalized and ventralized embryos, but not in lateralized embryos. Are such changes expected and do they make sense biologically? It is unclear why these phosphorylation data are used to validate the success of the approach.

       The three Cactus phosphosites S463, S467 and S468 were identified and characterised in the work cited in Ref. 19 (Liu Z.P. et al., Genes and Development, 1997), and we used these sites to validate that our approach was sensitive enough to detect known phosphosites in proteins that act on the dorso-ventral patterning pathway specifically at the point of gastrulation (Stage 6 of embryonic development). We also reported in this manuscript the detection of known phosphosites within the Rho-pathway (Fig. 5E,F, Myosin Light Chain: T21, S22; Cofilin: S3).

   Liu Z.P. et al. reported that these three sites map to the Cactus PEST domain, which is required for Cactus degradation in the mesoderm (Belvin M. et al, Genes and Development 1995).  Liu Z.P. et al. also showed that mutating these phosphosites impairs Cactus turnover without affecting the ability of Cactus to bind Dorsal. We can only speculate that the differential phosphorylation across dorso-ventral embryonic cell populations is associated with the regulation of Cactus turnover. Consistent with this, we find Cactus downregulated 1.5 log2 fold in ventralized embryos derived from *spn27Aex/def* mothers. Furthermore, there are a number of signalling pathways that act both in the dorsal and the ventral-lateral domain (e.g., rhomboid/EGF), so it is not surprising to find modifications that are shared by these regions.

    The rationale to use a diffusion algorithm for data analysis is not clear. How would the analysis differ if diffusion was not used?

    Phosphoproteomics data are often sparse and noisy for a number of reasons (technical; low abundance of phosphorylated peptides compared to other peptides in the cell; biological: not all phosphosites are functional). Network diffusion is a common way used for various data types to boost the signal-to-noise ratio. For example, if from a list of 10 phosphosites, 5 all fall in the same network region or process, and the rest are randomly distributed in the network, chances are that the first region is more representative of the regulated process in that dataset. Using network propagation, the signal coming from the first 5 phosphosites would give a higher score to that network region, marking it as the predominant signal. Our specific implementation, which uses the semantic similarity between nodes to model the edges in the network, further boosts the functional signal by preferentially including nodes that have a higher functional similarity to the initial phosphosites. Our approach therefore allows us to identify the processes that are predominantly ‘active’ in our dataset. We refer the reviewer to our recent preprint for more evidence that this strategy boosts the signal-to-noise ratio in phosphoproteomic datasets and further prioritises more functional phosphosites (https://www.biorxiv.org/content/10.1101/2023.08.07.552249v1). If this approach was not used and we based the identification of relevant processes only on the list of phosphosites, we would have acquired more spurious terms in our functional enrichment analysis. The above preprint also shows that different methods such as the Prize Collecting Steiner Forest algorithm perform worse for phosphoproteomics data.

    Generally, the discussion of enriched GO categories presented in Fig. 6 is not rigorous, and it is unclear what biological insight is provided by this figure, probably because the categories are extremely diverse and not clustered in a meaningful way. Despite stating that the work on microtubules came out as a result of proteomic analysis, there is no connection between proteomic data (e.g., data shown in Fig. 6) and microtubule analysis in Fig. 7.

       The connection is between the __phosphoproteomic__ data and the microtubules. The reviewer is correct about the fact there is little connection at the proteomic level with microtubules. Only the diffused network analyses performed on the phosphoproteomic data pointed in this direction. We have improved the writing about this point.

    The Discussion section touches on areas of differential protein degradation and mRNA regulation; however, these data are not presented in Results or Figures and so it is difficult to assess the relevance of this analysis.

         We present these data in Figure 6A,B. The network analyses of the clusters showed significant enrichment of cellular component terms that are connected with protein turnover and mRNA regulation. We have added a reference to figure 6 in the Discussion for clarity.

    There is insufficient citation of prior literature throughout the manuscript: many statements are lacking proper references.

    We have corrected the mistakes and added missing references.

    Proteomics data should be deposited into a standard repository that is a member of ProtomeXchange Consortium, such as PRIDE, etc.

    All proteomics and phosphoproteomics data have been uploaded to PRIDE:

    The raw files for the proteomics and phosphoproteomics experiments were deposited in PRIDE under separate identifiers:

    Proteome: Identifier PXD046050 (Reviewer account details: reviewer_pxd046050@ebi.ac.uk, pw: coJ9otiX).

    Phosphoproteome: Identifier PXD046192 (Reviewer account details: reviewer_pxd046192@ebi.ac.uk, pw: nvkbwClp).

    We have included a statement of raw data availability in the revised version of the manuscript with the PRIDE access information.

    __Minor comments __

    The text has several typos and should be proof-read, and references to figures and tables should be checked, as some of these are not correct.

    We have corrected typos, references to figures and tables in the revised version of the manuscript.

    The genotypes for the mutations used in this study should be accompanied by citation describing identification of these mutations and the resulting phenotypes. It would also be helpful to describe the nature of these alleles (molecular lesion, gain vs loss of function, etc.). Some of this information is included in the Discussion, but it would be useful for the reader to learn this early on, when the chosen genotypes are presented.

    All this information is and was provided in the methods section and in Table 1, including stock numbers and sources of the stocks. Please see 'Methods, *Drosophila *genetics and embryo collections'.

    2G,H - the X axis should be clearly labeled as logarithmic.

    We introduced the log2 label in the X-axis of Fig. 2G,H and any other panel in which this was not expressly made clear.

    In Fig. 2G the locations of lines showing fold changes for Twist and Snail seem incorrect. In Fig. 2H the dotted line does not appear to correspond to 50% of the number of phosphosites.

    We apologise for these errors, both have been corrected in the revised version of the manuscript.

    5D can be improved by adding letters for the coloured clusters.

    We have labelled the clusters in Fig. 3B and Fig. 5D. to ease the identification of biologically relevant clusters.

    It is unclear if any specific additional insight was obtained using SILAC, the authors may want to discuss this approach and outcomes more.

    SILAC has been widely used to deal with the inherent variability of proteomic analyses by introducing a standard that is metabolically labelled, in our case, w1118 flies fed with SILAC yeast were used as the standard. Because the inherent variability is larger in phosphoproteomic experiments (because protein identification is based on phosphorylated peptides only, see Methods), we used SILAC labelling only in the phosphoproteomic experiment.

    __Reviewer #2 __

    Evidence, reproducibility and clarity

    The present article by Gomez et al describes a deep proteomics analysis of the proteome and phosphoproteome of embryos mutated for key genes involved in the dorso-ventral axis in Drosophila melanogaster. Overall, this is a nice article showing new insight in this development process. The results are mainly descriptive, yet identifies potential new players in the definition of the dorso-ventral axis.

    The generation of mutants for genes found up- or down-regulated in each mutant strain would be a significant addition to this manuscript. But I think in its current form the data brings enough new information on this particular developmental step and would be of interest for the fly community.

    My main concern is that the manuscript can be difficult to read and overly convoluted at times even for experts in the field. I would suggest the author move some methodological explanations from the results to the methods section to further detail the goals of some results sections.

    We have followed these suggestions and hope we have made the manuscript more easily readable.

    As an example, the goal of part 3) « A linear model for quantitative interpretation of the proteomes » is not clear to me. Are the authors comparing the abundance of a protein in the WT versus a theoretical WT in order to determine which fractions of mesoderm, lateral ectoderm and dorsal region are actually present in the WT? (...)

    Yes, in part, but the main purpose was to compare how well the theoretical WT, as ‘reconstituted’ from the mutants, corresponds to the observed actual WT (for which we have at least approximate values).

    The question that we faced when we started these calculations was: what is the ‘correct’ fraction (or proportion) we should use to weight each protein (or phosphosite) measurement in the mutants. Theoretically, these values should be those that result in the best match between the theoretical WT and the measured WT abundance of each protein (or phosphosite). We knew from actual measurements only the mesodermal fraction, which was determined to be ~20% of the cross-sectional area (Ref. 21: Rahimi, N., et al Dev. Cell. 2016). The neuroectoderm and ectoderm fractions were estimated to be approx. 40% each (Ref.: 22, Jazwinska, A et al. Development 1999), but we lacked an exact number. The systematic exploration of these proportions led us to conclude that indeed both the neuroectoderm and ectoderm fractions should be around 40% each, provided the mesoderm is fixed at 20%. Thus, we used these fractions: D: 0.4 L: 0.4 V: 0.2 for our follow-up analyses.

    (...) Or are they using it as a reference to obtain a fold change for the different proteins quantified (in this case why not use the WT?)?

    yes, again, in part: as a reference for the EXPECTED fold changes, as would be predicted from the WT.

    Since we have moved some of the details of this approach from the main text to the methods section, we have also revised the remaining text and hope it is now clearer.

    The proteomics data must be deposited in a public repository. I did not see it stated in the methods section.

    All proteomics and phosphoproteomics data have been uploaded to PRIDE; see further comments above in response 13.

    The version of the uniprot database is quite old (2016) so is the version of MaxQuant used in this study. Any reasons for that (other than that the analysis was performed in 2016)?

    That is indeed the reason.

    The data were run on different MS platforms, how did the authors account for the variability in MS signals? What samples were run on which MS platform? Were the WT embryos ran on both?

    We measured three replicates, and all five genotypes (four mutant genotypes plus wildtype) for each of the replicates were measured on the same instrument. Specifically, for the whole proteome analyses, replicate one and three of all genotypes were measured on the QExactive Plus instrument and replicate 2 of all genotypes were measured on a QExactive HF-x instrument, as were the phosphoproteomes. So, indeed, the wildtype was measured on both instruments. We thus did not observe instrument-specific bias in the PCA analysis for the proteome data.

    We have added this in more detail to the method section:

    “Samples of replicate one and three were measured on the QE-Plus system and replicate two was measured on the QE-HF-x system.

    For phosphoproteome analysis, (…) Samples of all three replicates were measured on the QEx-HFx system. We added trial samples measured on the QEx-Plus system to increase the phosphosite coverage using the match between runs algorithm.”

    In the methods section the authors mention that a high-pH reverse phase fractionation was performed? How many fractions of High-pH reverse phase separation were injected per sample? Was this separation performed for all the samples?

    We have adjusted the Methods section regarding the high-pH fractionation by adding the following sentence: “Fractions were collected every 60s in a 96 well plate over 60 min gradient time collecting a total number of 8 fractions per sample.“

    Why did the authors used label-free (proteome) and SILAC (phosphoproteome) quantification methods?

    See our response to reviewer #1, point 19.

    Why is the threshold based on the Q3 of the standard deviation (if I got it right) ? Couldn't they be calculated directly on the distribution of the ratio?

    We could also have done it that way.

    However, we had wanted also to take into account the variation between the replicates, i.e., the quality of the individual measurements, and we therefore devised the procedure we used, by which the standard deviation of the individual technical replicates enters the calculation with the ratio of the averages, the variability between replicates would have been ignored and we considered it more appropriate to take the more conservative approach. But as it turns out, the cut-off would have ended up being very similar had we calculated it the way the referee suggests,

    Page 6: The supplementary figure 2E refers to the protein Cactus and the text to CKII, please modify one or the other to avoid any confusion. Page 7: A dot is missing at the end of the following sentence « if used with the assumed weightings for the populations »

    We have corrected these sentences.

    Page 19: Replace SppedVac by SpeedVac

    We have corrected the error in the manuscript and thank the reviewer for the detailed inspection.

    Page 8: why not using a z-score with thresholds directly instead of a -1/+1/0 system and then using the z-score?

    Because we wanted to compare the relative changes over wt between mutants (i.e. the similarity between 1 0 0 and 0 -1 -1) rather than the relationship of their absolute values to the wt, and to assign proteins with similar relationships into the same dorso-ventral regulation categories.

    The text states this (previously in main text, now in methods):

    “The reason for this is that this method takes into account that value sets that represent similar relative differences between the mutants (for example, 0 -1 -1 vs. 1 -1 -1 or 1, 0, 0) are biologically more similar to each other than the raw values indicate. The z-scores for all of these cases would be 1.1547 -0.5774 -0.5774.”

    In the abstract it is mentioned that 3,399 proteins are differentially regulated at the proteome level versus 1,699 significantly deregulated at a 10 % FDR in the main text (page 5). Is there a reason for this discrepancy? Same comment for the phosphopeptides.

    But we now also see the need to better clarify this point, and we have edited the text accordingly.

    The second number refers to those proteins that show statistically significant changes based on ANOVA (1699 proteins).

    The first number (3398; note that the number 3399 in the abstract was a typo, now corrected) includes all proteins that were detected in at least 1 replicate in the wildtype (5883/6111) minus those that do not change between the genotypes (2156/6111) and minus all those that change in the same direction in all mutants (329).

    This includes proteins that are automatically excluded from ANOVA, i.e., those that are detected only in the wildtype (35/6111 proteins) or in two or more genotypes but only in 1 technical replicate ANOVA negative ones.

    As we stated, we did this because it “allows us to include the important group of proteins that show a ‘perfect’ behaviour, like dMyc and WntD, in that they are undetectable in the mutants that correspond to the regions in the normal embryo where these genes are not expressed.”. This 'regulated' set consists of those proteins that exceed the |0.5| fold threshold.

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

    This review is a list of many individual critiques. It is unclear what the expertise of the reviewer is (they do not provide the answer to that question in the review form, unlike the other referees), but several of the criticisms are unfounded. Three of the PIs of this work are researchers with extensive experience in Drosophila genetics and early development but are nevertheless confounded by some of the comments made by this referee.

    The mutants do not completely "flatten" the embryos.

    We do not claim that they do. Nor are the ventral, lateral and dorsal regions in the normal embryo completely ‘flat’ or homogeneous. But the mutants are good representations of the major fates in these regions, as a wealth of published literature from the last 30 years indicates.

    For instance, Tl10B broadly expresses snail but also expresses sog in the head. (i.e. Fig 1B - sog and sna expression in Figure 1B mutant backgrounds looks odd.) The sog expression likely relates to a deficiency specific effect.

    This ‘sensitive’ area is well known also from other genetic conditions – e.g. partial loss of dorsal and indeed in Spn27A mutants. It is therefore not specific to the Tl10B deficiency but says something about gene interactions in this region. Thus, this cannot be a deficiency-specific effect.

    Is sog seen in a Toll10B/+ mutant background?

    Yes, it is, and more frequently than in Toll10B/Def.

    The deficiency used for the Toll10B experiment is Df(3R)ro80b which is quite large and deletes 14+ genes.

    True. However, this does not matter: the mothers are heterozygous, so the genes are not missing, they are present in one wildtype copy! And these mothers are then mated with wildtype fathers, so if expression of these genes were needed in the embryo, then there would be another full wt copy of each. We appreciate that maternal effect genetics can be difficult to follow, but this is all work that has been done a long time ago, and is not the point of this paper at all.

    The deficiency used for the spn27A experiment is Df(2L)BSC7 and removes 4+ genes.

    Again, this would only matter if these were maternal effect genes that were needed for the establishment of the dorso-ventral axis, and they are not.

    Furthermore, the gd9 allele may not be a complete loss of function.

    It may not be – but what matters is the well characterized phenotype which has been shown to represent dorsal cell types.

    It is possible that the Toll10B allele picked up an accessory dominant mutation.

    This again would only matter if it was a dominant AND maternal effect mutation that affects the DV axis in the embryo – and there are very few of these known. And nothing in our analysis of these embryos, with which we have been working on and off over 3 decades and therefore know very well, indicates that our current stock is any different from those we have seen in the past.

    Unfortunately, these mutant phenotypes that affect DV and AP patterning mean that conclusions cannot be made that changes in protein relate to DV patterning.

    We simply do not understand this statement.

    Why do the mutant phenotypes (gene expression patterns and cell morphologies representative of the ventral, lateral and dorsal cell populations) not mean that the proteins downstream of the fate changes correspond to the cell fates?

    To get a better view of the ventralized phenotype, the authors should repeat the analysis by ectopically expressing Toll10B using the Gal4-UAS system; UAS-activate Toll transgenes are available.

    All Gal4-UAS maternal drivers, even the best and the strongest, result in mosaic expression. Our lab has extensive experience with this system and we know that, for example, the homogeneous, high levels of twist or snail expression that we see in spn or Tl10B embryos cannot be achieved with GAL4.

    Fig 1C-F - due to combined AP and DV effects seen with ventralizing mutants, it is important that the authors confirm that cross-section views relate to the middle to posterior of the embryo.

    We confirm this.

    Costaining with anti-Kr or -Caudal would help to ensure they are assaying the correct AP domain for pure DV effects.

    In our view, this is an unnecessary experiment. I know where the middle of the embryo is. If the reviewer does not believe when we say we are showing a section from the middle, they can see that the sections are not from the end region by, for example, the cell number, and the section angles.

    The authors refer to reference [60] for stages but there is no information regarding morphological criteria used under the microscope to stage the embryos.

    We have now added more detail in the methods section:

    Briefly. using a Zeiss binocular, the embryos were individually hand-selected on wet agar which made the embryos semi-transparent, allowing the assessment of a range of morphological features, of which at least some are visible in each of the mutants:

    • Yolk distance to embryonic surface: distinguishes between early (stage 5a) and late cellularisation (stage 5b).
    • Yolk distribution within the embryo: identification of large embryonic movements of the germ band (e.g.: Initiation of germ band extension, marking the initiation of stage 7). In DV patterning mutants this is seen as twisting of the embryo.
    • Change in the outline of the dorsal-posterior region: polar cell movement from the posterior most region of the embryo (stage 5a/b) to stage 6a/b.
    • Formation of the cephalic and dorsal folds: identification of stage 6 (initiation of cephalic fold) and stage 7 (dorsal folds). The combined use of these morphological criteria, together with the synchronised egg collections allows accurate staging of wild type and mutant embryos.

    Furthermore, what is stage 6a,b? Stage 6 is not typically divided in two stages nor is it clear what a,b relate to.

    We used a generally accepted standard for staging embryos: Campos-Ortega J.A. and Hartenstein V. ‘The embryonic development of Drosophila melanogaster’ book (ref. Nº 60). In this book, they describe the morphological criteria that can be followed in living embryos for proper staging. These stages, with these exact names, are shown on pages 11 and 12 of the 1997 edition (2nd edition).

    According to the published timetable of Drosophila development by Foe et al. 1993 (not cited), gastrulating embryos are 200 min or 3 hr 20'. It's unclear if this is the stage that was assayed.

    Foe is a beautiful paper, but we did not cite it because the commonly used nomenclature predates it (Campos-Ortega and Hartenstein 1985).

    In addition, timing depends on temperature whereas morphological criteria do not.

    The mutant embryos likely develop at different rates relative to wildtype. It seems important to provide details about the staging of embryos. If the mutant embryos take longer to gastrulate, for instance, might that also be a factor that impacts the proteome.

    As described above, we used a combination of criteria to accurately judge staging. DV patterning embryos could in principle develop faster or slower than wildtype. We performed synchronised egg collections (Methods: Embryo collections) for 15’. Therefore, any developmental timing defect would have become evident based on a difference in the number of embryos entering stage 6 and 7 at the point of visual inspection of the collections. This was not the case.

    How many replicates for each genotype? In the text it states, "replicates from the same genotype clustered together (Fig. 2E)....." Similar vague reference for phosphoproteome follows (Fig 2F). It is then stated that it was impossible to determine the experimental source for this variation. Could it relate to differences in timing of samples?

    We had given the numbers of replicates in the figure legend but have now also included them in the methods section for more clarity. We did 3 replicates for each genotype in each experiment, with the exception of gd9 and spn27aex mutants, for which we did 2 biological replicates each with 3 replicates, making a total of 6 replicates for these genotypes in the proteomic experiment. We have included an additional clarification in figure legend 2. The number of replicates per genotype per experiment can also be seen from the correlation matrices shown Fig. 2E and 2F, in which the replicates are shown individually. The measurements for each replicate for each genotype within each experiment were reported in Supplementary Tables 2 and 3, 'description' tabs of the worksheets.

    The lengthy discussion of ratio estimation on page 7 should be streamlined and made more clear. Are the authors throwing out data and only keeping samples that support their model? This seems like overfitting - if I am understanding correctly, you are selecting the samples that support the "majority of proteins fit the linear model" but this isn't necessarily the case.

    No, this is a misunderstanding. We do not select data.

    We have rephrased this section, but to explain here briefly: We do not select any samples, we state that the majority of proteins fit the theoretical model (and that is not even surprising, because any protein that does not change across the populations will automatically fit the model). We then discuss why some might NOT fit the model. The model doesn’t need to be supported, it simply is a calculation that allows us to stratify the data.

    They call this the 'correct' manner (see section 4 page 7) but it seems like a working model and presumptuous to imply that it is the correct way.

    We explained in the text why we refer to this as ‘correct’. It is a matter or definition, not presumption, and we even used quotes to be clear about this. ’Correct’ indicates a combination of values that is consistent with the biological model that the DV mutants are good representations of the corresponding embryonic cell populations in a wild type embryo. We do not in any way ‘throw out’ other data, we just note they don’t fit that model. Clarifications on the concept for the model have been added in various places in the text

    Figure 3C - it is confusing to use a circular diagram to show DV inferred position of the 14 clusters as their position on the circle does not correspond to where they are expressed on the embryos. Perhaps a stacked bar graph for 6 different domains would be better.

    This figure does not show positions of clusters. It is simply a pie chart, as is stated in the figure legend and as can be seen by the numbers and the corresponding sizes of the sectors. We have tried a stacked representation (shown below), but find it no clearer and have therefore stuck with this very common way of representing quantities, and in particular, proportions. We use the same representation with the same colour schemes in all subsequent figures, so proportions can be compared across figures.

    It is very hard to follow the text on page 9.

    We have rephrased this section

    It is very hard to see the gene expression patterns shown in Fig 4A with the color scheme/scale used.

    We appreciate this colour scheme does not correspond to the commonly used dark colour on a light background which would mimic histochemistry to show gene expression. The ‘inferno’ colour scheme was used because it allows better quantitative comparisons between subtly different patterns. However, to make these figures more similar to the types of in situ hybridisations that embryologists are used to seeing, we now use a different representation.

    In general, Figure 4 is uninterpretable - in particular, what do the numbers mean on the greyscale circle plots in panel D?

    We apologize for having failed to explicitly include the explanation for this in the figure legend. The reader will notice that these numbers add up to the number in the circle to the left, and the numbers indicate the number of proteins showing perfect matches (white), partial overlaps (grey) and mismatches (black). We have improved the graphic representation and added an explanation in the figure legend.

    Figure 5A. Why wasn't protein abundance and phosphosites identified from an individual, identical sample?

    This was because of the way the project developed over the course of the research, and the protein part was originally intended only as a proof of concept, with the intended focus being the phosphoproteome. We later decided to include a full analysis of the proteome, but did not consider it worthwhile and necessary to repeat the entire laborious and expensive experiment with both analyses being done from the same samples.

    How can one be sure that the phosphosites were correctly assigned if the proteins were not detected in the proteome but they were only identified in the phosphosite analysis?

    We are not sure we understand this question. The phosphoproteomic analysis identifies phosphopeptides of proteins that in turn allow one to identify the protein itself and the amino acid in that peptide that is phosphorylated. So the identification is done only WITHIN the phosphoproteomic analysis and does not relate directly to the proteomic analysis. This explains why we found some phosphopeptides for which we did not detect the full host protein in the proteomic analysis.

    Thus, if a protein was detected only in either of the experiments, this fact doesn’t modify the validity of the result, because the identification was done individually for each experiment.

    Page 16 - much discussion about the difference between Spn27A and Toll10b/def mutant background. One has half as much Toll receptor. The phenotype of Toll10b/+ should be examined.

    Both genotypes have been extensively examined in the past. Tl10B/def has only one copy of the gene from the mother, and the mutant protein is constitutively active. By putting it over a deficiency, we (and others in the past) made sure that the exclusive source for Tl signalling is from this gain of function Tl allele, and that the wildtype receptor, which would still be activated by the natural ligand in a graded pattern along the DV axis, does not confound the result.

    The Tl10B/+ combination creates a less ventralized phenotype which is not more similar to that of spn27Aex/def but in fact less similar.

    Page 12 - hard to follow the discussion of modeling (?) presented in Figure 6. The results (bottom of page 12 - #1 "most networks are enriched for cellular components associated with regulation of gene expression" and page 13 #2 - "cytoskleeton emerges as a major target of regulation") seem vague and unsubstantiated. Rhabdomere, P granule, micropyle, autophagosome?

    We agree with the reviewer that there are many cellular components that are enriched in the diffused network analyses, many of them unrelated to morphogenesis. We had highlighted this finding on page 12, paragraph 3. Nevertheless, we have rephrased the statements as ‘the heat maps illustrate that most of the enriched cellular components in both experiments were highly enriched with cellular components associated with DNA and RNA metabolism or the regulation of gene expression.’ and have now included numbers.

    We think ‘a major target’ for phosphorylation does in fact apply to the cytoskeleton, and we had already supplied the number to substantiate this in the manuscript (14/62).

    Readers will be able to evaluate these network analyses based on their own fields of interest or particular questions they may wish to address. We haven’t excluded any cellular component terms.

    Figure 7 seems like a separate study.

    Why were the phosphopeptides investigated to determine if they relate to phosphorylated proteins? Phosphoantibodies could have been generated for a subset. Instead the manuscript pivots to analysis of microtubules.

    We are reporting here one example of a proof-of-concept study that we carried out, chosen based on our own research interests and on available tools and reagents. There are clearly many other avenues that could have been explored and that others may want to explore, but that go well beyond this report. We have made this more explicit in the text.

    Page 14 - discussion first paragraph. Please cite ref[10] when discussing the "previous study" otherwise the reader will not understand which study you are referring to until the next paragraph.

    We have moved the reference from its current position to the one suggested by the reviewer.

    • In general, the study would benefit from more attention to references and citations of prior work. A comparison of this work to the Gong et al. Development 2004 study should be made earlier. This work is cited very early on, namely in the introduction.

    • The authors start off saying that no other study has looked at proteins from a spatial perspective. We are unsure what the reviewer refers to. We say precisely the opposite: we indicate that studies have been performed to look at differences in cell populations, including that by the lab of Jon Minden (Gong et al), a highly respected former co-author of one of the current authors (ML). We do state that the technologies at the time did not allow the same depth and temporal resolution as the methods that are available nowadays. For instance, Gong et al. used an excellent and original approach at the time, which however did not detect Snail and Twist in the ventralized mutants.

    The only time we say ‘no other study’ is about ‘region-specific post-translational regulation of proteins’ - though we do state in the discussion that Gong et al would have detected some of these cases because they used 2D gels.

    • Along these lines, there is another more recent proteomic study from Beati et al. Fly 2020 using similarly staged embryos. How do these other experiments compare to the current ones? As they apparently analyzed proteome and phosphopeptides from an identical sample, are the authors' new data using separate samples consistent? This study is actually about a later stage (stage 8 embryos, post-gastrulation). Again, an excellent study, but not directly relevant to our current analysis. It validates the use of SILAC in Drosophila, although it is not the first study to do this. Furthermore, it looks at a different question and biological process using a mutant, htl, to understand the effect of FGF signalling.

    • Furthermore, Adam Martin's lab has been studying microtubule action along the dorsoventral axis (Denk-Lobnig et al 2021) and this work is not cited. Denk-Lobnig et al 2021 is about spatial patterns of myosin and actin and how that is governed genetically on the ventral side of the embryo, pertaining primarily to ventral furrow formation. It does not analyse microtubules nor dorsal-ventral cell populations.

    It is possible there may be some confusion with another excellent study from Adam Martin’s lab, in which the role of microtubules is analysed. But this is exclusively in the ventral furrow, and the study did not look at the effect of microtubule depolymerisation on nuclear positioning nor membrane behaviour. We cite this work extensively (Ref.: 36, Ko et al. JCB 2019) and we compare our results to that paper. However, our work here goes beyond this study in that it looks at all cells along the DV axis.

    General comments:

    Typos throughout. For example, page .4 section heading "dorso-ventral cell..."

    We have scanned the entire document for typos.

    Font size extremely small - for example see Figure 1A gene names, and 1F magnified view.

    We have adjusted the fonts in the main figures.

    Scale bars not shown when showing magnified views. For example, see Fig 1E,

    We have added these.

    Reviewer #3 (Significance (Required)): This study by Gomez et al. uses a proteomic-centered approach to study proteomes associated with cell populations in the embryo that they argue relate to different positions along the dorso-ventral axis. They generate a proteomic resource, though it was unclear how anyone could use the data they produce. There is no searchable database and we have to trust that the authors will ultimately provide such a resource to the community.

    All proteomics and phosphoproteomics data have been uploaded to PRIDE. Also see responses to the other referees’ queries about this point.

    There is the potential for interesting insights but the work is not presented in a way that is accessible or useful. The presentation needs significant improvement.

    We have improved the presentation and way the results are presented as per the suggestion of all reviewers.

  2. Review Commons

    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 mutants do not completely "flatten" the embryos. For instance, Tl10B broadly expresses snail but also expresses sog in the head. (i.e. Fig 1B - sog and sna expression in Figure 1B mutant backgrounds looks odd.) The sog expression likely relates to a deficiency specific effect. Is sog seen in a Toll10B/+ mutant background? The deficiency used for the Toll10B experiment is Df(3R)ro80b which is quite large and deletes 14+ genes. The deficiency used for the spn27A experiment is Df(2L)BSC7 and removes 4+ genes. Furthermore, the gd9 allele may not be a complete loss of function. It is possible that the Toll10B allele picked up an accessory dominant mutation. Unfortunately, these mutant phenotypes that affect DV and AP patterning mean that conclusions cannot be made that changes in protein relate to DV patterning. To get a better view of the ventralized phenotype, the authors should repeat the analysis by ectopically expressing Toll10B using the Gal4-UAS system; UAS-activate Toll transgenes are available.

    • Fig 1C-F - due to combined AP and DV effects seen with ventralizing mutants, it is important that the authors confirm that cross-section views relate to the middle to posterior of the embryo. Costaining with anti-Kr or -Caudal would help to ensure they are assaying the correct AP domain for pure DV effects.

    • The authors refer to reference [60] for stages but there is no information regarding morphological criteria used under the microscope to stage the embryos. Furthermore, what is stage 6a,b? Stage 6 is not typically divided in two stages nor is it clear what a,b relate to. According to the published timetable of Drosophila development by Foe et al. 1993 (not cited), gastrulating embryos are 200 min or 3 hr 20'. It's unclear if this is the stage that was assayed.

    • The mutant embryos likely develop at different rates relative to wildtype. It seems important to provide details about the staging of embryos. If the mutant embryos take longer to gastrulate, for instance, might that also be a factor that impacts the proteome.

    • How many replicates for each genotype? In the text it states, "replicates from the same genotype clustered together (Fig. 2E)....." Similar vague reference for phosphoproteome follows (Fig 2F). It is then stated that it was impossible to determine the experimental source for this variation. Could it relate to differences in timing of samples?

    • The lengthy discussion of ratio estimation on page 7 should be streamlined and made more clear. Are the authors throwing out data and only keeping samples that support their model? This seems like overfitting - if I am understanding correctly, you are selecting the samples that support the "majority of proteins fit the linear model" but this isn't necessarily the case. They call this the 'correct' manner (see section 4 page 7) but it seems like a working model and presumptuous to imply that it is the correct way.

    • Figure 3C - it is confusing to use a circular diagram to show DV inferred position of the 14 clusters as their position on the circle does not correspond to where they are expressed on the embryos. Perhaps a stacked bar graph for 6 different domains would be better.

    • It is very hard to follow the text on page 9.

    • It is very hard to see the gene expression patterns shown in Fig 4A with the color scheme/scale used.

    • In general, Figure 4 is uninterpretable - in particular, what do the numbers mean on the greyscale circle plots in panel D?

    • Figure 5A. Why wasn't protein abundance and phosphosites identified from an individual, identical sample? How can one be sure that the phosphosites were correctly assigned if the proteins were not detected in the proteome but they were only identified in the phosphosite analysis?

    • Page 16 - much discussion about the difference between Spn27A and Toll10b/def mutant background. One has half as much Toll receptor. The phenotype of Toll10b/+ should be examined.

    • Page 12 - hard to follow the discussion of modeling (?) presented in Figure 6. The results (bottom of page 12 - #1 "most networks are enriched for cellular components associated with regulation of gene expression" and page 13 #2 - "cytoskleeton emerges as a major target of regulation" ) seem vague and unsubstantiated. Rhabdomere, P granule, micropyle, autophagosome?

    • Figure 7 seems like a separate study. Why were the phosphopeptides investigated to determine if they relate to phosphorylated proteins? Phosphoantibodies could have been generated for a subset. Instead the manuscript pivots to analysis of microtubules.

    • Page 14 - discussion first paragraph. Please cite ref[10] when discussing the "previous study" otherwise the reader will not understand which study you are referring to until the next paragraph. In general, the study would benefit from more attention to references and citations of prior work. A comparison of this work to the Gong et al. Development 2004 study should be made earlier. The authors start off saying that no other study has looked at proteins from a spatial perspective - but this other study from 2004 did just that. They compared ventralized to lateralized embryos. Along these lines, there is another more recent proteomic study from Beati et al. Fly 2020 using similarly staged embryos. How do these other experiments compare to the current ones? As they apparently analyzed proteome and phosphopeptides from an identical sample, are the authors' new data using separate samples consistent?

    General comments:

    1. Typos throughout. For example, page .4 section heading "dorso-ventral cell..."

    2. Font size extremely small - for example see Figure 1A gene names, and 1F magnified view.

    3. Scale bars not shown when showing magnified views. For example, see Fig 1E,F

    Significance

    This study by Gomez et al. uses a proteomic-centered approach to study proteomes associated with cell populations in the embryo that they argue relate to different positions along the dorso-ventral axis. They generate a proteomic resource, though it was unclear how anyone could use the data they produce. There is no searchable database and we have to trust that the authors will ultimately provide such a resource to the community. Furthermore, Adam Martin's lab has been studying microtubule action along the dorsoventral axis (Denk-Lobnig et al 2021) and this work is not cited. There is the potential for interesting insights but the work is not presented in a way that is accessible or useful. The presentation needs significant improvement.

  3. Review Commons

    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 present article by Gomez et al describes a deep proteomics analysis of the proteome and phsophoproteome of embryos mutated for key genes involved in the dorso-ventral axis in Drosophila melanogaster. Overall this is a nice article showing new insight in this development process. The results are mainly descriptive yet identifies potential new players in the definition of the dorso-ventral axis. The generation of mutants for genes found up- or down-regulated in each mutant strain would be a significant addition to this manuscript. But I think in its current form the data brings enough new information on this particular developmental step and would be of interest for the fly community. My main concern is that the manuscript can be difficult to read and overly convoluted at times even for experts in the field. I would suggest the author move some methodological explanations from the results to the methods section to further detail the goals of some results sections. As an example, the goal of the part 3) « A linear model for quantitative interpretation of the proteomes » is not clear to me. Are the authors comparing the abundance of a protein in the WT versus a theoritical WT in order to determine which fractions of mesoderm, lateral ectoderm and dorsal region are actually present in the WT ? Or are they using it as a reference to obtain a fold change for the different proteins quantified (in this case why not use the WT?) ?

    Other comments:

    • The proteomics data must be deposited in a public repository. I did not see it stated in the methods section.

    • The version of the uniprot database is quite old (2016) so is the version of MaxQuant used in this study. Any reasons for that (other than that the analysis was performed in 2016) ?

    • The data were run on different MS platforms, how did the authors accounted for the variability in MS signals ? What samples were run on which MS platform ? Where the WT embryos ran on both ?

    • In the methods section the authors mention that a high-pH reverse phase fractionation was performed ? How many fractions of High-pH reverse phase separation were injected per sample ? Was this separation performed for all the samples ?

    • Why did the authors used label-free (proteome) and SILAC (phosphoproteome) quantification methods ?

    • Why are the threshold based on the Q3 of the standard deviation (if I got if right) ? Couldn't they be calculated directly on the distribution of the ratio ?

    • Page 6 : The supplementary figure 2E refers to the protein Cactus and the text to CKII, please modify one or the other to avoid and confusion.

    • Page 7 : A dot is missing at the end of the following sentence « if used with the assumed weightings for the populations »

    • Page 19 : Replace SppedVac by SpeedVac

    • Page 8 : why not using a z-score with thresholds directly instead of a -1/+1/0 system and then using the z-score ?

    • In the abstract it is mentioned that 3,399 proteins are differentially regulated at the proteome level versus 1,699 significantly deregulated at a 10 % FDR in the main text (page 5). Is there a reason for this discrepancy ? Same comment for the phosphopeptides.

    Significance

    I think in its current form the data brings enough new information on this particular developmental step and would be of interest for the fly community. My main concern is that the manuscript can be difficult to read and overly convoluted at times even for experts in the field.

    Reviewer experise: Drosophila proteomics

  4. Review Commons

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

    Learn more at Review Commons

    Referee #1

    Evidence, reproducibility and clarity

    Summary

    This manuscript investigated changes in the proteome and phosphoproteome during dorsovental axis specification in the Drosophila embryo. To model the three regions in the embryo that are relevant for DV axis development, the authors used specific mutations to enrich for a single type of cells (ventral, lateral, or dorsal). The detected proteins and phosphopeptides were clustered according to the region of expression. There were differences between the protein and corresponding phosphopeptide abundance, suggesting that phosphorylation is a regulatory modification in DV axis establishment. Two different mutations that both result in a ventralized phenotype were found to change marker protein expression in different ways. Using inhibition of microtubule polymerization, this study also investigated the role of microtubules in epithelial folding.

    Major comments

    • Generally, there is a lack of significance testing throughout the manuscript. Simply reporting fold changes can be misleading, if these changes are not significant. Examples:

    1. Rigor of the proteomics evidence showing changes for the expected markers is insufficient because no statistical evaluation is provided. Specifically, in Fig. 1D and Suppl Fig 2: are the fold changes statistically significant?

    2. Data in Fig. 4F, 5F need to be assessed for significance. There are other instances in the manuscript where significance should be tested.

    • It is difficult to see the value of the obtained dataset for the community, in part because the data are analyzed by a linear model and cluster assignment developed by the authors, which is a somewhat arbitrary representation. Perhaps the authors could explain how their data could be used by other researchers, and maybe even develop an accessible portal for interacting with the data. For example, what does it mean biologically that a protein is a member of a specific cluster shown in Fig. 3C? Is there a predictive value in such an assignment, and how does it relate to the main question of DV axis regulation? An example of a novel insight obtained for specific protein(s) would be useful to illustrate the utility of this analysis.

    • Overall, at present the study appears to have limited novelty and mechanistic insight. The data generally align with prior expectations, but it is unclear how this work advances the field. For example, the observed differences between marker proteins in Toll10B vs. spn27A data seem to confirm previous suggestions that spn27A has a stronger ventralizing effect. The role of microtubules in epithelial folding in the embryo has also been demonstrated before.

    • The shown phosphorylation changes (if they are significant) for Toll and Cactus are difficult to explain. In Suppl Fig 2B, E: why is Toll more phosphorylated in the lateralized than in ventralized embryos? (the provided reference 20 does not seem to clarify this) Also, certain Cactus phosphorylations appear higher in dorsalized and ventralized embryos, but not in lateralized embryos. Are such changes expected and do they make sense biologically? It is unclear why these phosphorylation data are used to validate the success of the approach.

    • The rationale to use a diffusion algorithm for data analysis is not clear. How would the analysis differ if diffusion was not used? Generally, the discussion of enriched GO categories presented in Fig. 6 is not rigorous, and it is unclear what biological insight is provided by this figure, probably because the categories are extremely diverse and not clustered in a meaningful way.

    • Despite stating that the work on microtubules came out as a result of proteomic analysis, there is no connection between proteomic data (e.g. data shown in Fig. 6) and microtubule analysis in Fig. 7. Given the broad range of categories shown in Fig. 6, it is not obvious how the jump to tubulin post-translational modifications and microtubule behavior shown in Fig. 7 was made, which leaves Fig. 7 as a disconnected set of results.

    • The Discussion section touches on areas of differential protein degradation and mRNA regulation, however these data are not presented in Results or Figures and so it is difficult to assess the relevance of this analysis. There is insufficient citation of prior literature throughout the manuscript: many statements are lacking proper references. Proteomics data should be deposited into a standard repository that is a member of ProtomeXchange Consortium, such as PRIDE, etc.

    Minor comments

    • The text has several typos and should be proof-read, and references to figures and tables should be checked, as some of these are not correct.

    • The genotypes for the mutations used in this study should be accompanied by citations describing identification of these mutations and the resulting phenotypes. It would also be helpful to describe the nature of these alleles (molecular lesion, gain vs loss of function, etc.). Some of this information is included in the Discussion, but it would be useful for the reader to learn this early on, when the chosen genotypes are presented.

    • Fig. 2G,H - the X axis should be clearly labeled as logarithmic. In Fig. 2G the locations of lines showing fold changes for Twist and Snail seem incorrect. In Fig. 2H the dotted line does not appear to correspond to 50% of the number of phosphosites. Fig. 5D can be improved by adding letters for the colored clusters.

    • It is unclear if any specific additional insight was obtained using SILAC, the authors may want to discuss this approach and outcomes more.

    Significance

    General assessment

    Strengths: The study uses a good model system (mutations that enrich for a specific type of cells) to investigate the proteome during DV axis establishment. The technical approaches are sound and the raw data are mostly of high quality. Limitations: The lack of significance testing throughout the manuscript makes it difficult to determine whether the stated changes are meaningful. It is unclear how experiments with microtubules are connected to the rest of the story. In its present form, the utility of the data for a broader community is limited, because there is no data analysis portal developed for easy data visualization and interaction, and the data in the supplemental tables are not easily interpretable.

    Advance: Overall, this study may serve as a resource for future functional investigation, however limitations in data analysis and presentation currently limit its impact. At present, the advance of this study appears incremental, as it largely agrees with prior observations and does not show novel mechanistic insights in our understanding of DV axis specification. Providing clear examples of how this analysis may result in new understanding and explaining the biological relevance of the findings would help to address this problem.

    Audience: Researchers working in the fields of dorsoventral axis specification, Drosophila genetics, developmental biology, proteomics.

  5. Version published to 10.1101/2023.08.24.554590v1 on bioRxiv