Glucose-regulated and drug-perturbed phosphoproteome reveals molecular mechanisms controlling insulin secretion

Insulin-secreting beta cells play an essential role in maintaining physiological blood glucose levels, and their dysfunction leads to the development of diabetes. To elucidate the signalling events regulating insulin secretion, we applied a recently developed phosphoproteomics workflow. We quantified the time-resolved phosphoproteome of murine pancreatic cells following their exposure to glucose and in combination with small molecule compounds that promote insulin secretion. The quantitative phosphoproteome of 30,000 sites clustered into three main groups in concordance with the modulation of the three key kinases: PKA, PKC and CK2A. A high-resolution time course revealed key novel regulatory sites, revealing the importance of methyltransferase DNMT3A phosphorylation in the glucose response. Remarkably a significant proportion of these novel regulatory sites is significantly downregulated in diabetic islets. Control of insulin secretion is embedded in an unexpectedly broad and complex range of cellular functions, which are perturbed by drugs in multiple ways.

The phosphoproteomics is expanded to a temporal profiling of dynamic response to glucose stimulation. The authors narrow down interesting p-sites to 29 by comparing significantly regulated sites from the compound analysis to the expanded glucose time course. Of these, 2 sites (on Slc24a2 and Mcoln1) were validated for their importance to GSIS in Ins1e cells upon overexpression of wild-type and S/A mutant. Additionally, it was shown by interaction proteomics that the known HDAC2 and histone 3.1 interactions with Dnmt3a were dependent on phosphorylation of Dnmt3a S7. Furthermore, wt and S/A mutant Dnmt3a overexpression affected the abundance of several proteins.
The presented work comprises a detailed analysis of signaling events related to glucose stimulated insulin secretion (GSIS). The strength of the study is the depth of the analysis, the application of several compounds and the time course analysis. Although the work is comprehensive, intriguing and identifies novel regulatory events of GSIS, the manuscript in its current format presents with many inconsistencies that need to be addressed by the authors.
Several aspects that need attention as listed below. Particularly, many figures and their reference throughout the results section are not complete. Many additional details need to be added to the Materials and Methods. Furthermore the discussion presents merely as a summary of findings rather than a discussion of the data. The section on p. 10 line 5-9 could serve as a great basis for the discussion.
Major comments 1. To establish the role of Dmnt3a in glucose stimulated insulin secretion the authors overexpress the wild-type or S>A mutant isoforms. Unfortunately, overexpression can have many unintended side effects unrelated to the normal physiological role of the protein. A much better experiment to confirm the role of phosphorylation would be to mutate the phosphorylation site to aspartic acid (S>D) or alanine (S>A) endogenously (e.g., with CRISPR). As it stands the overexpression of the WT form, by itself, apparently has an impact on the protein expression levels of a large number of proteins; this affect appears to be independent of glucose stimulation or phosphorylation status.
2. The level of overexpression level of Dmnt3a in HEK293 cell is too high. From the LFQ data on the y-axis, it appears that Dmnt3a is expressed at ~8-fold (in log space) higher in the transfected cells compared to the endogenous. I assuming LFQ is in Log2? This information is not provided.
3. The expression level for Dmnt3a should be confirmed by western blot. 4. Some of the critical interacting proteins should also be confirmed by co-IP with the endogenous Dmnt3a, or by proximity ligation assays.
5. There appears to be some confusion between gene expression and protein expression. Changes in gene expression should be measured by quantifying mRNA levels, while changes in protein expression should be measured by quantifying proteins. Essentially all data supporting the link of Dmnt3a S7 phosphorylation to gene expression (figs 6 ans S8) rely on analyses that show protein abundance changes. The authors should conduct mRNASeq experiments to quantify gene expression changes, or should select a couple of regulated genes and show that the change in protein expression is also occurring on the mRNA level to support any conclusion based on gene expression. 6. The supplementary tables are confusing in their current presentation. The authors have provided data for each analysis, which is good. However, they need to provide the averages and standard deviations for each set of analyses as well. Moreover, the values presented in many of the tables are undefined. Are these Log2 intensity values? 7. Many of the peptides and proteins have missing values in one or more of the columns, with many peptides having missing values in many of the columns. How were these missing data treated? What was the minimum number of measurements needed to include the peptide? Were n=1 values used for comparisons? 8. Perusing the data in Table S3, it is not obvious that Dmnt3a changes significantly in the different conditions relative to the biological variability within each condition. Perhaps the authors could provide a bar plot of this data with error bars? 9. Bacause S7 of Dnmt3a is critical to the manuscript, the authors should provide the annotated MS/MS spectra of this peptide, along with the associated spectra from other Dnmt3a phosphorylation sites. This would help the interested reader to assess the accuracy of the site assignments, especially given the 3 different phospho-isoforms of the same peptide, with closely spaced phosphorylation sites. 10. The authors indicate that phosphorylation of Dnmt3a at S7 is involved in GSIS, but fail to show that mutation of this site affects GSIS. 11. What was the rationale behind the selection of Dnmt3a for further analysis? 12. The introduction and summary should state that the work focus on a murine system. 13. Rework the summary to specify the three signaling pathways, which the compound group into and converge on: PKA, PKC, CK2A. Are the three identified pathways important for GSIS? It would be interesting to know if the use of inhibitors for these three kinases reduce GSIS upon treatment?
14. It is not clear from the results section which concentrations of glucose have been used for low dose versus high dose glucose treatments. Please specify the first time the reader is introduced to these terms. Furthermore, there is inconsistency regarding low dose. Was 2.5 mM (p. 21, legend fig 2) or 3.5 mM (p. 18) used? Additionally please specify if high or low dose glucose was used for the expanded time-course in Min6 cells.
15. There are no details on how the murine pancreatic islets were collected for the analyses.
16. The concentrations used for the seven compounds are not included in the manuscript. Furthermore details are missing on how the compounds were used for in the described phosphoproteomics setup such as pretreatment time.
17. There is no reference to figure S1, S7B, 5F, 6E. 22. p. 8 please specify where the number of "more than 22,000 sites" is derived from? Otherwise please stay consistent with numbers. 23. Figure 3B has to be scaled up to clearly display the motifs. 24. Figure S4 -legend and figure does not match. Furthermore S4B and C is not described in the results.
Other things:

Summary:
The sentence "A high-resolution time course revealed key novel regulatory sites and unexpected connections to epigenetic regulation of gene expression through the methyltransferase DNMT3A" is heavily overstated for the last part. There is no direct evidence of gene expression changes related to DNMT3A.

Reviewer #2 (Remarks to the Author):
In this phosphorproteomics study by Sacco, Schaefer, Mann and colleagues, the authors perform an exhaustive array of glucose and multi-drug perturbation/response experiments on an in vitro model of glucose-mediated insulin secretion using Min6 cells and discover specific drug/target response pathways as candidate avenues for therapeutic intervention. Detailed molecular interactomics analyses of the de novo DNA methyltransferase enzyme Dnmt3a revealed glucosemediated, phospho-dependent functional interactions of this epigenetic regulator with its genomic targets in cells. Type-2 diabetes (T2D) is a major global health burden on society; new therapeutic strategies and approaches are urgently required, which creates significant rationale and potential sustained translational impact for the present work. This is a well-conceived, thoughtful, careful and technically strong research program and manuscript. Its broad general appeal is also well suited to the readership of Nature Communications. I have only a few minor comments intended to refine and polish what is already a very strong study. My recommendations are in no particular order: 1. Some additional functional tests or details of the Min6 system would bolster the manuscript. As-is, the validation of these cells as a surrogate for pancreatic islets is described on a strictly proteomics basis. Even if prior published research evaluating the response of Min6 cells to glucose stimulation or other factors is presented, it would bolster the discussion of these results to make them more balanced between descriptive proteomics and functional analysis. Correlation of rank order of protein copy number alone appears insufficient to describe functional equivalency of two quite different systems.
2. Some balance to the discussion regarding the kinase pathways regulated by the three drug groups would soften a few of the untested and unproven assertions in that section. No discussion of the potential differential regulation of these compounds on phosphatase activity, for example, is provided, in spite of the rather strong language regarding the "unambiguous classification" of these drug-response profiles sorting into three kinase-centric signaling pathways.
3. Some additional details of the brief description of how kinase motifs are developed and credentialed would further support the authors' assertions regarding kinase-specific signaling pathways. The three short sentences in the Supplementary Methods are insufficient to reproduce the authors' results. 4. In the discussion of signaling pathways triggered by drug treatment, it is less clear how the authors linked kinase-substrate motifs to pathways that converge on the activation of cell cyclerelated kinases, such as Cdks. Some additional details and stepwise transitions through this logic would further clarify these important links.

5.
Although it may be beyond the scope of the current work, an ideal analysis of the Dnmt3a phosphosite mutant would include genome-wide methylation array data to assess the direct effects of this phosphorylation site. In addition, the authors could test for rescue of the differential effects of the S7A mutant using a catalytically inactive enzyme. In any case, these data as presented are strongly supportive of the hypothesis that phos-Ser7 is a functional readout of glucose-dependent genomic methylation changes in Min6 cells.

Reviewer #3 (Remarks to the Author):
The work by Sacco et al examines the phosphoproteome of mouse MIN6 insulinoma cells under several conditions that stimulate or augment insulin secretion. The authors have quantified a large number of phospho-sites and show that these cluster into three main signalling groupsdepending on the stimulus used. Some novel pathways are also hinted at in this dataset, which should provide a valuable resource and platform for the investigation of insulin secretory function. The authors have largely focused on insulin secretion (as noted in the title for example), although there is some clear indication of an impact of these treatments on other cellular functions such as cycle/apoptosis that should be discussed in the context of the cell model used (i.e. insulinoma cells). Some care should also be taken in interpretation of the results -for example, does autocrine signalling play a role in some of the observed results? Presumably there should be some conserved signalling between all treatments that increase insulin secretion, since insulin (and other secreted factors) are known to feed-back via cell surface receptors -can this be teased out in the data? Does the data suggest that such feed-back makes only a minor contribution to signalling? Major and minor comments follow: Major: 1. The authors use several different physiological and pharmacological agents, and it's nice to see in Fig. 2 that the cells appear to respond well to these with respect to their insulin secretion (although note: no units are given in the figure legend or Fig 2C). For the most part, the authors have nicely described the mechanism by which most of the compounds used elicit insulin secretion. However, the description of the effects of extracellularly applied ATP seem not to be entirely correct. The authors lump ATP together with glibenclamide and state that these "...do not directly target a kinases..." (page 7) and that "Both these drugs increase insulin secretion by depolarizing the cell membrane. As such, the current thinking seems to be that extracellular ATP will increase insulin secretion partly by raising Ca2+ (or perhaps more correctly, by 'augmenting' the Ca2+ response) and partly by increasing DAG via PLC to promote insulin exocytosis. Is there any evidence for ATP-dependent activation of PLC-PKC (or other) pathways? If not, perhaps this is interesting in itself and merits some discussion.
2. Further to the above point, the authors have very nicely shown that the different agents can be grouped largely according to their presumed mechanism of action. I am somewhat surprised however at how distinctly these groups can be separated. Can the authors comment on cross-talk or overlap between different treatments? Conventional thinking would suggest that autocrine signalling is an important regulatory component of beta-cells (for example, autocrine signalling by several things that might stimulate phosphorylation-dependent signalling within beta-cells are released upon stimulation: insulin, ATP, and GABA are some of these that have been investigated in detail). Would the authors not expect to see some pathways that are always activated by any treatment that promotes secretion -and may be attributed to autocrine activation of signalling pathways (do these represent the 17% overlap mentioned between drugs and glucose-stimulation)? Perhaps this underlies the rationale (which is not well-described in the paper) for using MAPK and PI3K-AKT as indicators of glucose responsiveness (page 7) -since these may be activated by autocrine insulin signalling? 4. The majority of the work performed in this study was in MIN6 insulinoma cells. I recognize that the authors have confirmed similarity in protein expression (but not phosphorylation) between these and primary mouse islets, but the authors should still avoid using the term 'betacells' throughout the paper when referring to the insulinoma cells (including the title, since the 'insulinoma phosphoproteome' rather than 'beta-cell phosphoproteome' was studied). Some additional examples include: "The finding that this site is regulated in the beta cell..." (page 11); "...we decided to investigate the temporal regulation of these sites and of the global beta cell phosphoproteome in a glucose dependent time-course in beta cells." (page 10). Minor: 1. The authors describe isolated islets as the "...in vivo cellular context..." of the MIN6 cells. I think that this needs some clarification. Isolated islets certainly do not re-capitulate all of the invivo context of beta-cells (i.e. loss of innervation and vascularization". The authors may wish to revise this sentence to reflect that the data from primary islets (which as the authors point out could be a useful resource in itself) in comparison to the MIN6 cells suggests that the latter is indeed a reasonable model to use.
2. The first sentence of the Discussion states that "Beta cell dysfunction is a major hallmark of the progression of type 2 diabetes." The authors may want to consider that the beta cell plays not just an important role in disease progression, but is in fact a major contributor to the initiation and genetic susceptibility to diabetes as well.

July 28, 2016
Point-by-point answers to reviewer's comments for "Glucose-regulated and drug perturbed beta-cell phosphoproteome reveals molecular mechanisms controlling insulin secretion" by F. Sacco et al.
We thank all the reviewers for their thoughtful and constructive comments, which were helpful in improving the manuscript.

Reviewer #1 (Remarks to the Author):
In this work the authors focus on cellular response to glucose and other small molecules in a murine cell line Min6. They use a mass spectrometry based approach to analyze the proteome and phosphoproteome related to glucose stimulated signaling events while measuring insulin secretion. Firstly, the authors show that the Min6 cell line resembles beta cells from murine pancreatic islets by a comparative proteome analysis showing overlap in expressed proteins.
Next, the authors perform phosphoproteomics on samples treated with low dose glucose (10 min) and high glucose (10 min and 30 min) including seven different compounds with a known influence on insulin secretion. The authors find that compound treatment group and converge on three main pathways involving PKA, PKC and CK2A activation.
The phosphoproteomics is expanded to a temporal profiling of dynamic response to glucose stimulation. The authors narrow down interesting p-sites to 29 by comparing significantly regulated sites from the compound analysis to the expanded glucose time course. Of these, 2 sites (on Slc24a2 and Mcoln1) were validated for their importance to GSIS in Ins1e cells upon overexpression of wild-type and S/A mutant. Additionally, it was shown by interaction proteomics that the known HDAC2 and histone 3.1 interactions with Dnmt3a were dependent on phosphorylation of Dnmt3a S7. Furthermore, wt and S/A mutant Dnmt3a overexpression affected the abundance of several proteins.
The presented work comprises a detailed analysis of signaling events related to glucose stimulated insulin secretion (GSIS). The strength of the study is the depth of the analysis, the application of several compounds and the time course analysis. Although the work is comprehensive, intriguing and identifies novel regulatory events of GSIS, the manuscript in its current format presents with many inconsistencies that need to be addressed by the authors.
We thank the reviewer for the thorough review and kind comments about our manuscript. We hope the changes and new data we provide further strengthen it in his or her eyes.
Several aspects that need attention as listed below. Particularly, many figures and their reference throughout the results section are not complete. Many additional details need to be added to the Materials and Methods. Furthermore the discussion presents merely as a summary of findings rather than a discussion of the data. The section on p. 10 line 5-9 could serve as a great basis for the discussion.
Major comments 1. To establish the role of Dmnt3a in glucose stimulated insulin secretion the authors overexpress the wild-type or S>A mutant isoforms. Unfortunately, overexpression can have many unintended side effects unrelated to the normal physiological role of the protein. A much better experiment to confirm the role of phosphorylation would be to mutate the phosphorylation site to aspartic acid (S>D) or alanine (S>A) endogenously (e.g., with CRISPR). As it stands the overexpression of the WT form, by itself, apparently has an impact on the protein expression levels of a large number of proteins; this affect appears to be independent of glucose stimulation or phosphorylation status.
We agree with the reviewer that over-expression may have unintended consequences. Therefore, we performed our experiments in two biological systems, Ins1e and HEK293 cell lines, that express the exogenous Dnmt3a at low and high levels, respectively. As shown in Fig We agree with the reviewer that the mutation of endogenous Dnmt3a using CRISPR technology would be an elegant approach. However, embarking on such experiments would be particularly time consuming, since the use of CRISPR-Cas9 for the purposes of point mutation is complicated by numerous factors (such as controlling for clonal selection), which currently limit its widespread application. Furthermore, we feel such studies would go beyond the scope of our work, which is not focused on the identification of Dnmt3a-dependent genes themselves, but rather on investigating the functional consequences of Dnmt3a Ser7 phosphorylation. We hope that the reviewer agrees that our unbiased MS-based proteomics and interactomics strategies demonstrate the impact of Dnmt3a on protein expression levels, and that for a large number of proteins this appears to be dependent on Ser4 phosphorylation.
2. The level of overexpression level of Dmnt3a in HEK293 cell is too high. From the LFQ data on the y-axis, it appears that Dmnt3a is expressed at ~8-fold (in log space) higher in the transfected cells compared to the endogenous. I assuming LFQ is in Log2? This information is not provided.
We have now clearly indicated in the text legend and figure S10F that the Dnmt3a LFQ intensity is on a log2 scale. The high level of Dnmt3a expression is consistent with the high transfection efficiency of HEK293 cells.  4. Some of the critical interacting proteins should also be confirmed by co-IP with the endogenous Dmnt3a, or by proximity ligation assays.
The most critical and abundant Dnmt3a interactors we found are the HDAC2 and histone H3.1 (Fig. 7). These proteins have already been extensively described as Dnmt3a interactors in the literature 1, 2 . We have now added a co-IP experiment, demonstrating that the HDAC2 protein can only bind the wild type form of Dnmt3a, but not the non-phosphorylable mutant (Fig. S10H).
This experiment further confirms our MS-based interactomics results.
5. There appears to be some confusion between gene expression and protein expression. Changes in gene expression should be measured by quantifying mRNA levels, while changes in protein expression should be measured by quantifying proteins. Essentially all data supporting the link of Dmnt3a S7 phosphorylation to gene expression (figs 6 and S8) rely on analyses that show protein abundance changes. The authors should conduct mRNASeq experiments to quantify gene expression changes, or should select a couple of regulated genes and show that the change in protein expression is also occurring on the mRNA level to support any conclusion based on gene expression.
We thank the reviewer for prompting us to clarify this point. To demonstrate the role of Dnmt3a phosphorylation in the beta cell response to glucose, we have indeed used protein levels as an endpoint rather than mRNA levels. To address the reviewers comments, we took advantage of data recently published by group of Anil Bhushan, who demonstrate a key role of Dnmt3a in initiating a metabolic program essential to preserving the glucose-stimulated insulin secretion (GSIS) capacity during the maturation of beta cells 3  In line with these results, our MS-based proteomics dataset revealed that after 6 and 12 hours of glucose stimulation, both Hk1 and Ldha genes are significantly down-regulated. We also demonstrated that the over-expression of WT Dnmt3a significantly downregulates the protein level of Hk1 in Ins1e cells. Importantly, the expression levels of these proteins is not affected by the over-expression of the non-phosphorylable mutant S7A. We hope that the reviewer agrees that these observations further support the importance of Dmnt3a S7 phosphorylation on gene expression regulation. 6. The supplementary tables are confusing in their current presentation. The authors have provided data for each analysis, which is good. However, they need to provide the averages and standard deviations for each set of analyses as well. Moreover, the values presented in many of the tables are undefined. Are these Log2 intensity values?
We regret any confusion the reviewer has faced with regards to the supplementary tables. To improve presentation, we have now revised supplementary data, and clearly state that the presented values are in Log2 scale. We have now also added averages and standard deviations for each dataset. We thank the reviewer for prompting us to clarify this issue. We feel it is important that we provide high quality data in our supplementary materials, and in our phosphoproteome-related tables we therefore only include phosphosites quantified in at least 50% of experimental conditions (12 out of 24). In analyzing the data for this study, we investigated different statistical approaches for reliably identifying glucose-and drug-regulated phosphosites and how to best handle the issue of missing values. Our final choice was to perform the analysis of variance (ANOVA) with FDR control (FDR < 0.05) of the Class 1 sites (Table S3 and Table S4), without thresholding the number of quantified values per peptide. The ANOVA test with FDR control inherently penalizes phosphosites that have missing values in many experimental conditions, and such sites are therefore much less likely to be found to be significant. We believe this strikes a good balance between reporting high quality data while at the same time avoiding inadvertently filtering out sites of low abundance that other researchers may find to be biologically important.
Below we show the distribution of the valid values in our total phosphoproteome (in blue) and in the ANOVA significant phosphosites (in red). As shown, the ANOVA test identifies the highest proportion of significant sites among those phosphopeptides quantified in almost all the experimental conditions. 8. Perusing the data in Table S3 We now provide the associated mass spectra of the three different phospho-isoforms of Dnmt3a.
These data can be found in the new Supplementary Figure S9C. 10. The authors indicate that phosphorylation of Dnmt3a at S7 is involved in GSIS, but fail to show that mutation of this site affects GSIS.
As the reviewer states, we have not shown that mutation of this site affects the ability of the beta cells to secrete insulin upon glucose stimulation. This is consistent with our MS-based proteomics data, which did not show any significant Dnmt3a-dependent modulation of proteins involved in the regulation of any aspects related to insulin secretion (e.g. vesicle trafficking, exocytosis or membrane hyperpolarization).
Upon Dnmt3a over-expression, we found a set of proteins significantly down-regulated. These proteins are involved in the regulation of the cell cycle and hormone response (insulin and Erbb) and importantly are also significantly down-regulated after 6 hours of glucose stimulation. These data support the importance of Dnmt3a as regulator of a transcriptional program triggered by glucose stimulation. 11. What was the rationale behind the selection of Dnmt3a for further analysis?
Dnmt3a is one of the 31 potential regulatory phosphoproteins involved in the glucose response of beta cells. We selected this protein for further characterization experiments for several reasons: Firstly, it plays a crucial role in beta cell differentiation and metabolism. Beta cell-specific deletion of Dnmt3a is sufficient to cause beta-to-alpha-cell reprogramming, driving a metabolic program by repressing key genes to enable the coupling of insulin secretion to glucose levels during beta cell maturation 3,6 . Additionally, a genome-wide association study (GWAS) robustly revealed Dnmt3a as one of the genetic contributors to the pathogenesis of type 1 diabetes 7 . We have now edited the text to further clarify the rationale behind the selection of Dnmt3a for our follow-up investigations. 12. The introduction and summary should state that the work focus on a murine system.
We thank the reviewer for this suggestion. We have now edited the introduction and summary text accordingly. 13. Rework the summary to specify the three signaling pathways, which the compound group into and converge on: PKA, PKC, CK2A. Are the three identified pathways important for GSIS?
It would be interesting to know if the use of inhibitors for these three kinases reduce GSIS upon treatment?
We thank the reviewer for prompting us to clarify this. As suggested, we have now edited the summary and clarified this point, on page 10 of the results section.
As shown in Figure 3, some of the compounds we used are directly linked to the activation of these three master kinases. As an example: PKA is essential for the GSIS 8 . This data is also supported by the MS-based proteomic profile of NOD diabetic islets we now performed for the revision, in which GSIS is impaired and PKA protein levels are concomitantly reduced (Table   S1). We treated beta cells with GLP1, which increases the concentration of cAMP. The GLP1mediated increase of cAMP as well as the treatment with 8-bromo-cAMP directly activates PKA kinase, potentiating the GSIS. We have also treated beta cells with Carbachol and 8-bromo-cGMP, both triggering PKC/PKG activity. These kinases are also believed to be important for GSIS 9 . ATP and Glibenclamide compounds have so far not been linked to the activation of CK2A, and to our knowledge, our data show for the first time that CK2A is involved in GSIS.
Importantly, in agreement with our phosphoproteomics data, the MS-based proteome profiling of NOD diabetic islets also revealed a significant reduction of CK2A levels (Table S1) 15. There are no details on how the murine pancreatic islets were collected for the analyses.
We have now amended the methods section with details of the experimental procedure we applied to isolate islets.
16. The concentrations used for the seven compounds are not included in the manuscript.
Furthermore details are missing on how the compounds were used for in the described phosphoproteomics setup such as pretreatment time.
We thank the reviewer for pointing out this oversight. We have now added the compound concentrations in the methods section of the manuscript.
We now include references to these figures in the manuscript.
18. References to different parts of figure S2 in the text p. 6 do not match the actual figure.
Furthermore, the figure legend does not match.
We have now edited the text and the figure legend accordingly. 2D, S3, 3D, 5E, S1A), which make it hard to interpret the presented data.

Several figures miss the color key (fig
We have now edited these figures to resolve this issue. Fig S3 to have a valid statement/sentence in text p. 7 line 18.

Please include kinase-substrate enrichment analysis for GLP-1 and carbachol in
We now include the kinase-substrate enrichment analysis for GLP1 and Carbachol ( Figure S4 We thank the reviewer for prompting us to clarify this point. In Figure 3C, S7A, S8A we plotted the median of 3-4 biological replicates with the standard deviation (SD). In the text referring to Fig. 2, we state that there is a peak of insulin secretion, compared with the 10-minute stimulation. We chose these two time points for the drug treatment and glucose phosphoproteome profiling in agreement with widely accepted models of GSIS, in which the first peak of insulin secretion occurs after 10 minutes of glucose stimulation, and this is followed by a higher peak of insulin secretion that occurs after 30 minutes of glucose stimulation 10 . We have now added a statistical analysis of the insulin secretion assay. These data are now shown in 23. Figure 3B has to be scaled up to clearly display the motifs.
We have now scaled the figure as suggested.
24. Figure S4 -legend and figure does not match. Furthermore S4B and C is not described in the results.
We apologize for the confusion and we have now corrected the legends in the Supplementary Information accordingly.
25. Figure 5S needs reworking. It is difficult to understand.
We thank the reviewer for this suggestion and we have now edited the figure to make it clearer.

Is a vertical label missing from fig S6C?
We have now added the vertical labeling to Fig. S6C.

Is it correct that only histone H3.1 (and not HDAC2) is confirmed in HEK cells? Fig S8B-C.
In our MS-based interactomics experiments we were not able to detect the HDAC2 proteins.
However, in the new co-IP experiments (Fig. S10H) between the HDAC2 and Dnmt3a WT and S7A proteins, we confirmed the result previously obtained in Ins1e cells (Fig. 6). Fig. S8 and Fig. 6.

What is the actual overlap in significant protein abundance changes between Ins1E cells and HEK293 cells as presented in
In this figure we have represented the actual over-lap between the HEK293 and Ins1e cells.
Among the 2,007 proteins that we found to be expressed in both HEK293 and In1e cells (after homology mapping using their gene names), and among these 133 are up or down regulated in HEK293 and 141 in In1e cells. 9 are significantly up or down-regulated in both these two cellular systems and this overlap is statistically significant (p= 4.5*10 -10 ). The small overlap between these two systems is not surprising given the profound differences of the two cell lines We thank the reviewer for pointing this. We have revised the text, adding the correct reference to the figure. Table S7 refer to? Min6 cells or Ins1e cells? Table S7 contains the Min6 proteomics dataset. We have now clarified this in the legend.

Which cells does the analysis in
Minor comments 1. Fig 1A: numbers in the Venn diagram does not add up and they do not align with the text.
More than 9,000 proteins were identified and 8,661 were quantified in Min6 cells. We have revised the text accordingly.
2. Supplementary material p. 38: is it correct that the murine data was searched against the human Uniprot FASTA database?
We thank the reviewer for pointing out this mistake. Murine data were searched again the Mus Musculus Uniprot FASTA database. We have corrected the supplementary material accordingly.
We have now edited the text accordingly, correcting the legend. Figure 4B, please rework the figure so that p-sites are clearly visible.

4.
We have now edited the figure according to the reviewer's suggestion. 5. P. 10: 6000 sites or 6041 sites?
The ANOVA significant phosphosites are 6,041. We have clarified this in the text. 6. P. 11 how does 38 phosphosites minus 2 discarded sites end up being 29 sites?
We apologize for the confusion and we thank the reviewer for prompting us to look again at our tables. 35 phosphosites are confirmed in the time-course experiment; two are discarded and we end up with 33 potential regulatory sites. We have corrected this in the manuscript.
Other things: Summary: The sentence "A high-resolution time course revealed key novel regulatory sites and unexpected connections to epigenetic regulation of gene expression through the methyltransferase DNMT3A" is heavily overstated for the last part. There is no direct evidence of gene expression changes related to DNMT3A.
We have revised the summary according to the reviewer's suggestion.

Reviewer #2 (Remarks to the Author):
In this phosphorproteomics study by Sacco, Schaefer, Mann and colleagues, the authors perform an exhaustive array of glucose and multi-drug perturbation/response experiments on an in vitro model of glucose-mediated insulin secretion using Min6 cells and discover specific drug/target response pathways as candidate avenues for therapeutic intervention. Detailed molecular interactomics analyses of the de novo DNA methyltransferase enzyme Dnmt3a revealed glucosemediated, phospho-dependent functional interactions of this epigenetic regulator with its genomic targets in cells. Type-2 diabetes (T2D) is a major global health burden on society; new therapeutic strategies and approaches are urgently required, which creates significant rationale and potential sustained translational impact for the present work. This is a well-conceived, thoughtful, careful and technically strong research program and manuscript. Its broad general appeal is also well suited to the readership of Nature Communications. I have only a few minor comments intended to refine and polish what is already a very strong study.
We thank the reviewer for the thorough review and kind comments about our manuscript, and especially for pointing out its broad general appeal. We hope that the changes we have made in response to his or her comments further strengthen this view of our manuscript.
My recommendations are in no particular order: 1. Some additional functional tests or details of the Min6 system would bolster the manuscript.
As-is, the validation of these cells as a surrogate for pancreatic islets is described on a strictly proteomics basis. Even if prior published research evaluating the response of Min6 cells to glucose stimulation or other factors is presented, it would bolster the discussion of these results to make them more balanced between descriptive proteomics and functional analysis. Correlation of rank order of protein copy number alone appears insufficient to describe functional equivalency of two quite different systems.
We agree that the correlation of rank order of protein copy number alone is insufficient to infer functional equivalency of Min6 cells and islets. We have now revised the text, highlighting that Min6 cells were chosen primarily because of their ability to recapitulate the release of insulin after glucose stimulation. Importantly, the Min6 experimental system we used recapitulates the glucose stimulated insulin secretion (GSIS) that occurs in islets: we detected a first peak of insulin secretion after 10 minutes of glucose stimulation, and a second peak after 30 minutes of glucose stimulation (Fig. 2B and S7A). We hope that this functional data together with the comparable expression level of key beta cells proteins, together with the substantial literature, demonstrates to the reviewer that this system is a suitable model for studying some of the signaling processes underlying GSIS.
2. Some balance to the discussion regarding the kinase pathways regulated by the three drug groups would soften a few of the untested and unproven assertions in that section. No discussion of the potential differential regulation of these compounds on phosphatase activity, for example, is provided, in spite of the rather strong language regarding the "unambiguous classification" of these drug-response profiles sorting into three kinase-centric signaling pathways.
We thank the reviewer for this suggestion, and have now revised the discussion text avoiding overstatements. Prompted by the reviewer we have now added a more detailed description of the kinase-substrate motif analysis we performed. The kinase substrate motifs were extracted from the HPRD database 11 and added as annotation to each phosphosite quantified in our phosphoproteome dataset. We subsequently performed two different analyses: -Fisher exact test of the loadings (phosphosites) responsible for the discrimination of the three groups (Fig. 3B).
-1D annotation enrichment analyses to identify statistically significant enriched kinasesubstrates motifs in the various experimental conditions (Fig. 4A and S3).
We have now edited the Supplementary Methods to include this additional information.
4. In the discussion of signaling pathways triggered by drug treatment, it is less clear how the authors linked kinase-substrate motifs to pathways that converge on the activation of cell cyclerelated kinases, such as Cdks. Some additional details and stepwise transitions through this logic would further clarify these important links.
We thank the reviewer for prompting us to clarify this point. The 1D annotation analysis of the kinase substrate motifs in the different experimental conditions (Fig. 4A) revealed a significant enrichment of CDK motifs after treatment with high glucose as well as with all the seven drugs we used. This data is in agreement with the results we obtained by mapping our phosphoproteomics dataset with kinase-substrate networks extracted from the PhosphoSitePlus database 12 . This strategy enables us to "walk" through the signal transduction events triggered by drug treatments, revealing that each drug activates different pathways converging upon activation of cell-cycle related kinases, such as Cdk1, Cdk2, Cdk5 and Cdk7. We have now revised the paper to clarify this.

5.
Although it may be beyond the scope of the current work, an ideal analysis of the Dnmt3a phosphosite mutant would include genome-wide methylation array data to assess the direct effects of this phosphorylation site. In addition, the authors could test for rescue of the differential effects of the S7A mutant using a catalytically inactive enzyme. In any case, these data as presented are strongly supportive of the hypothesis that phos-Ser7 is a functional readout of glucose-dependent genomic methylation changes in Min6 cells.
We agree with the reviewer that the analysis of the genome-wide methylation with the Dnmt3a phosphosite mutant would be extremely interesting, but beyond the scope of the current work.
We thank the reviewer for his or her comment about our follow-up experiments demonstrating the functional role of the phospho Ser7 of Dnmt3a.

Reviewer #3 (Remarks to the Author):
The  Fig 2C). For the most part, the authors have nicely described the mechanism by which most of the compounds used elicit insulin secretion. However, the description of the effects of extracellularly applied ATP seem not to be entirely correct. The authors lump ATP together with glibenclamide and state that these "...do not directly target a kinases..." (page 7) and that "Both these drugs increase insulin secretion by Yelovitch et al, J Med Chem, 2012). As such, the current thinking seems to be that extracellular ATP will increase insulin secretion partly by raising Ca2+ (or perhaps more correctly, by 'augmenting' the Ca2+ response) and partly by increasing DAG via PLC to promote insulin exocytosis. Is there any evidence for ATP-dependent activation of PLC-PKC (or other) pathways? If not, perhaps this is interesting in itself and merits some discussion.
We thank the reviewer for this comment. In figure 4B, we have shown the main signal transduction pathways triggered by the different drugs. Specifically, ATP treatment is connected to an increased intracellular calcium concentration, which in turn activates CaMK2A and CK2A, and to increased activity of PKC kinase, as revealed by the phosphorylation level of its regulatory serine residue. We have now revised the discussion and updated the references, highlighting that our data are in agreement with the accepted mechanism of action of ATP, proposing that this compound increases insulin secretion partly by triggering the Ca2+ response and partly by increasing DAG via PLC and PKC to promote insulin exocytosis.
2. Further to the above point, the authors have very nicely shown that the different agents can be grouped largely according to their presumed mechanism of action. I am somewhat surprised however at how distinctly these groups can be separated. Can the authors comment on cross-talk or overlap between different treatments? Conventional thinking would suggest that autocrine signalling is an important regulatory component of beta-cells (for example, autocrine signalling by several things that might stimulate phosphorylation-dependent signalling within beta-cells are released upon stimulation: insulin, ATP, and GABA are some of these that have been investigated in detail). Would the authors not expect to see some pathways that are always activated by any treatment that promotes secretion -and may be attributed to autocrine activation of signalling pathways (do these represent the 17% overlap mentioned between drugs and glucose-stimulation)? Perhaps this underlies the rationale (which is not well-described in the paper) for using MAPK and PI3K-AKT as indicators of glucose responsiveness (page 7) -since these may be activated by autocrine insulin signalling?
As the reviewer states, the principal component analysis of the phosphoproteomes measured with each treatment condition segregates the drugs into three distinct main clusters. These three groups are enriched in three major kinases reflecting the presumed mechanism of action of each compound. By mapping our based-phosphoproteomics dataset with kinase-substrate network, extracted by the PhosphoSitePlus dabatase 12 , we found that drug treatments overlap in the modulation of different signaling pathways (eg. MAPK1/2 are activated by ATP, glibenclamide, carbachol and 8-bromo-cGMP; while AKT is activated by treatment with all the drugs). This overlap could be explained by the autocrine signaling mentioned by the reviewer, although we would then expect that many proteins involved in the insulin signaling (e.g. RAF-MAPK axis and PI3K-mTOR axis) should be activated by the treatment with all the drugs. As shown in Figure 4B, while the MAPK cascade is differentially modulated by the different drugs, the AKT-mTOR pathway appears to be perturbed by the stimulation with all of the drugs. In addition, drug effects on pathway modulation may overlap because of the highly interconnected nature of signaling networks, and the large number of proteins modulated by the three major kinases (PKA, PKC and CK2A).
We have now revised the discussion, highlighting that from our dataset it is difficult to discriminate signaling events directly due to drug treatment or indirectly to autocrine signaling events. The 17% of sites mentioned by the reviewer are the ones significantly regulated by at least one of the drugs (ANOVA, FDR<0.05). Therefore, we do not think that their phosphorylation may be attributed to autocrine activation. We have used the GSIS as well as the MAPK and mTOR activation to check that the glucose stimulation was working as expected. We have now revised the text highlighting that that the MAPK and mTOR pathways can be activated by autocrine signaling or by calcium and cAMP dependent mechanisms 13 .
3. In the absence of confirmation in primary islets/beta-cells, I would suggest being cautious about conclusions relating to cell cycle and apoptosis (particularly the former -i.e. Fig. 4) obtained from highly proliferative MIN6 cells. Data from primary islets would significantly strengthen this aspect of the paper (along with the Dnmt3a finding).
We agree with the reviewer and we have now revised the discussion, highlighting that one of the most important difference between the islets and Min6 cells concerns the cell proliferation control. We have also edited the text avoiding over-statements regarding the cell cycle and apoptosis-related effects we observed upon drug treatments. 4. The majority of the work performed in this study was in MIN6 insulinoma cells. I recognize that the authors have confirmed similarity in protein expression (but not phosphorylation) between these and primary mouse islets, but the authors should still avoid using the term 'betacells' throughout the paper when referring to the insulinoma cells (including the title, since the 'insulinoma phosphoproteome' rather than 'beta-cell phosphoproteome' was studied). Some additional examples include: "The finding that this site is regulated in the beta cell..." (page 11); "...we decided to investigate the temporal regulation of these sites and of the global beta cell phosphoproteome in a glucose dependent time-course in beta cells." (page 10).
We agree with the reviewer and have added a statement to the second paragraph of the results section, highlighting that Min6 cells are an insulinoma cell line. We think that it is important to consider that Min6 cells are able to secrete insulin after glucose stimulation, which is the key beta-cell relevant feature that we are focusing on in this study. Since our paper aims at the identification of functional mechanisms underlying glucose induced insulin secretion, we would like to keep the "beta cells" instead of "insulinoma" cells in the title. We feel this makes it clearer to readers that our manuscript is focused on the GSIS aspect of beta cell signaling, rather than on cancer-related biology. Minor: 1. The authors describe isolated islets as the "...in vivo cellular context..." of the MIN6 cells. I think that this needs some clarification. Isolated islets certainly do not re-capitulate all of the invivo context of beta-cells (i.e. loss of innervation and vascularization". The authors may wish to revise this sentence to reflect that the data from primary islets (which as the authors point out could be a useful resource in itself) in comparison to the MIN6 cells suggests that the latter is indeed a reasonable model to use.
We agree with the reviewer and we have rephrased this sentence as suggested.
2. The first sentence of the Discussion states that "Beta cell dysfunction is a major hallmark of the progression of type 2 diabetes." The authors may want to consider that the beta cell plays not just an important role in disease progression, but is in fact a major contributor to the initiation and genetic susceptibility to diabetes as well.

Reviewer #1 (Remarks to the Author):
The authors have addressed many of the concerns that were raised in the first round of review either experimentally but mostly through written comments. Also many of the figures have been reworked or polished and are now presented in a reasonable form, which allows for a proper interpretation. Despite these efforts the manuscript still needs some reworking. There is no doubt about the overall quality of the work and the importance for the insulin community. However, a fair amount of inconsistencies need to be addressed in order for the work to be strong, concise and convincing.
Issues Remaining: Although the authors have attempted to rationalize their use of "gene expression profile", the fact remains that gene expression was never actually measured in this manuscript. To minimize confusion for the term "gene expression profile" needs to be removed from legend titles and section subtitles to ensure a precise interpretation of the work presented.
The level of Dnmt3a over-expression in HEK293 cells remains a major issue. The authors point out in the response letter that the LFQ axis is log2 based. An 8-fold change in log2 space ( Figure  S10F) is a 256-fold increase in protein expression! This super-physiological level of expression can have many unintended consequences, especially for a transcriptional regulator. The role of phosphorylation on this massively overexpressed protein might be very different from the same protein expressed at endogenous levels. These experiments need to be repeated with a lower level of transfection and therefore decreased overexpression.
The validation by co-IP of Dnmt3a and HDAC2 (Fig S10H) in the HEK293 cells is weak, at best. Even with the massive overexpression of Dnmt3a, only a barely detectable amount of HDAC2 is being pulled down, despite a high level of expression of HDAC2 in these cells, as apparent from the amount in the WCL fraction. Presumably, if this interaction were real and phospho-dependent, there would be a significant detectable difference in these lanes and HDAC2 would have been detected in the MS analysis.
There seems to be some confusion regarding numbers of identified and quantified proteins in islets and Min6 cells (Table S2) and text p. 8 and figure 2A (venn diagram). Please ensure coherence in the presentation of data. Table S2 reads a total of 9063 rows corresponding to total quantified proteins. Consulting the LFQ columns (Table S2) it seems like in Min6 cells 8466 proteins were quantified and in islets 7555. Please be clear and concise as to numbers referring to "identified" or "identified and quantified".  Please make sure that the correct table S4/S5 is referenced on p. 13.
In the discussion Figure 7 is referenced several times. Please check that this is correct.  Text p. 7, one compound is still listed twice.
Text p. 8, what is the calculation behind the statement 65% of the detected proteome?
p. 15 please specify in the text what "we also validated the interaction results in a human cell line, HEK293" entails. Which or how many proteins were confirmed?

Reviewer #2 (Remarks to the Author):
All of my concerns have been thoroughly addressed in revision.

September 5, 2016
Point-by-point answers to reviewer's comments for "Glucose-regulated and drug perturbed beta-cell phosphoproteome reveals molecular mechanisms controlling insulin secretion" by F. Sacco et al.
The level of Dnmt3a over-expression in HEK293 cells remains a major issue. The authors point out in the response letter that the LFQ axis is log2 based. An 8-fold change in log2 space ( Figure   S10F) is a 256-fold increase in protein expression! This super-physiological level of expression can have many unintended consequences, especially for a transcriptional regulator. The role of phosphorylation on this massively overexpressed protein might be very different from the same protein expressed at endogenous levels. These experiments need to be repeated with a lower level of transfection and therefore decreased overexpression.
As the editor suggested, we have now revised text, acknowledging the caveats that may come with over-expression.
The validation by co-IP of Dnmt3a and HDAC2 (Fig S10H) in the HEK293 cells is weak, at best. Even with the massive overexpression of Dnmt3a, only a barely detectable amount of HDAC2 is being pulled down, despite a high level of expression of HDAC2 in these cells, as apparent from the amount in the WCL fraction. Presumably, if this interaction were real and phospho-dependent, there would be a significant detectable difference in these lanes and HDAC2 would have been detected in the MS analysis.
We thank the reviewer for prompting us to clarify this issue in our manuscript. The interaction between HDAC2 and Dnmt3a has been already described by Fuks et al. 1 . In addition we also observed that this interaction is phospho-dependent in our MS analysis in Ins1e cells ( Fig 7B).
As indicated by the reviewer, in Hek293 cells this interaction is weak. This result is also consistent with the fact that we were not able to detect this association in Hek293 by MS analysis.
There seems to be some confusion regarding numbers of identified and quantified proteins in islets and Min6 cells (Table S2) and text p. 8 and figure 2A (venn diagram). Please ensure coherence in the presentation of data. Table S2 reads a total of 9063 rows corresponding to total quantified proteins. Consulting the LFQ columns (Table S2) it seems like in Min6 cells 8466 proteins were quantified and in islets 7555.
Please be clear and concise as to numbers referring to "identified" or "identified and quantified".
We thank the reviewer for pointing out this oversight. We have now revised the text and Figure   2A accordingly. Table S4 presents the numbers of ANOVA significant phosphorsites and repeatedly 6041 is reported throughout the manuscript. However, consulting the table it has 6042 rows -2 header row=6040 phosphosites. Please correct this error.
We thank the reviewer for pointing out this oversight. We have now revised the text accordingly. We thank the reviewer for pointing out this oversight. We have now revised the text accordingly.
Text p. 13: There is still some confusion regarding number of regulatory sites.
In the response to reviewers the number was corrected to 33. The text still reads 31 or 30 (p. 13 and p. 14, p. 19 reads 30?). When you count the number of listed p-sites in the figure it reads 32.
Please readdress this issue.
We have now revised the text accordingly.
There are 3 legends to supplementary figure S8. Please correct numbering.
We have now edited the numbering of the legends of the three supplementary figures.. We thank the reviewer for pointing out this oversight. We have now revised the text accordingly. We have explained in the figure legend that the Log2 LFQ intensity is Z-scored in Fig. S1C.
FigS3B, C is referenced in the text p. 8 to support a "high degree of overlap" between experimental conditions. Please clarify of show support of this statement.
We have linked this statement to the Fig. S3D. p. 15 please specify in the text what "we also validated the interaction results in a human cell line, HEK293" entails. Which or how many proteins were confirmed?
We thank the reviewer for prompting us to clarify this issue. We have now clarified in the text that in both the experimental system we observed that the S7A mutation impairs the ability to bind almost all the Dnmt3a interactors. This further confirms the role of S7 phosphorylation in the control of the association of the methyltransferase to its partners. In both these two experimental systems we found the H3.1 hystone and HDACs (in Hek293 HDAC8 and in Ins1e   HDAC2). The small overlap between these two systems is not surprising given the profound differences of the two cell lines HEK293 (Human Embryonic Kidney cells), versus Ins1el (Rat Insulinoma cells), and the different levels of Dnmt3a expression observed.

Reviewer #2 (Remarks to the Author):
All of my concerns have been thoroughly addressed in revision.
We are glad that we were able to address the concerns of the reviewer.

Reviewer #3 (Remarks to the Author):
The authors present a very nice revised paper and have adequately addressed my previous concerns, with one exception: The authors maintain that it is appropriate to refer to MIN6 cells as 'beta-cells' so as to 'emphasize that the work focuses on the GSIS aspect of these cells' (this is a good point I think).
However, I disagree with this. MIN6 cells are not beta-cells and to refer to them as such is a bit misleading (one could argue that they are not even the best insulinoma model, but that's beside the point). If the authors with to emphasize the the focus on GSIS, why not refer to "...the drug perturbed phosphoproteome in insulin-secreting cells..."?
We have now revised the text replacing beta cells with "insulin secreting cells".