Polycomb- and REST-associated histone deacetylases are independent pathways toward a mature neuronal phenotype

We submitted previously to eLife a manuscript testing the roles of REST and Polycomb repression in regulating the neuronal lineage in mouse embryonic stem cells. This is an important question because Polycomb has been reported as a general modulator of lineage control in pluripotent cells. As such, many studies focus just on this complex to understand how genes are globally repressed in pluripotency. We provided evidence indicating that the repressor REST is equally efficacious in this role and likely works independently and in parallel for genes expressed in the neuronal lineage. Two of the reviewers indicated that more work was required to substantiate our conclusions, and the editors requested submission of a new manuscript addressing these concerns. We took time to repeat a mass spectrometry study under different conditions, performed new RNA-seq analysis, re-did some of the bioinformatics, and added other new experiments as well. Consequently, this is a much more in-depth and solidified study than the original work and still contains, we feel, the novelty component expected for eLife.

Reviewer #1:

1) During the purification of the REST-associated complexes, the authors do a nice job of validating the candidates by ChIP-qPCR. However, the peptide counts for all of the targets including the bait were relatively low and this may preclude the presence of other interacting protein(s) that may be of interest. More rigorous biochemical purification would benefit the authors in their search for the missing deacetylase, therefore either increasing the number of ESC plates used in the purification should improve the identification of novel factors or the authors should try a different tandem affinity purification strategy to improve the identification of interacting proteins.

Our peptide counts were not dissimilar to other examples in the current literature where novel interacting proteins were identified (Ding, et al, 2012, Cell Res; Costa et al, 2013, Nature), and we identified all known REST interactors using our approach.

However, in response to this concern, we increased the number of plates as suggested and in addition followed the advice in point 4 (below) and performed mass spectrometry analyses using nuclear extracts produced by dounce homogenization, rather than sonicated whole cell extracts as before. Unfortunately, in two biological replicates of the nuclear extract mass spec analysis, we identified primarily nucleolar proteins and no peptides of known REST cofactors, including Rcor1, LSD1/Kdm1a, or G9a, which have been published by several labs as an integral part of a bona fide REST complex and which we identified readily with our previous method. Indeed, to our knowledge no one has identified REST interactors with a REST pull-down using the Dignam method. Therefore, to further test our negative result from the MudPiT analysis we performed a candidate co-immunoprecipitation analysis using nuclear extracts that was not included in the previous submission to confirm these potential interactions. The co-immunoprecipitation analysis confirmed binding of the known interactors of REST but it also identified members of the PRC2 complex Suz12 and Jarid2. However, neither Eed nor Ezh2 were present in the REST pull-downs with this method (new Figure 1–figure supplement 2A). The lack of Eed and Ezh2 is entirely consistent with the lack of REST binding sites marked with H3K27me3, as they are both required for deposition of the H3K27me3 mark. We have included the new results and a discussion of the findings. Finally, we are not sure which deacetylase the reviewer refers to as missing? Perhaps he meant the H3K4me3 demethylase? If the latter, it did not turn up in either of our mass spec analysis methods, indicating that this may be an indirect effect of REST-directed HDAC activity.

2) The use of TSA by the authors as a proxy of histone deacetylation activity should be supported with a figure showing at least the effect of knocking down HDAC1/2, which was identified in from their proteomic analysis.

We have provided a new figure (Figure 4–figure supplement 1) that shows increased levels of H3K9 acetylation at REST binding sites after TSA treatment, which directly inhibits the enzymatic activity of HDACs. We preferred this over the HDAC1/2 knock down experiment because knock down of HDACs is likely to be partial, and we are not sure how we would interpret a negative result. In addition, HDAC1/2 double knockout ESCs shows a profound loss of cell viability (Dovey et al, 2014, PNAS). However, after deletion, these cells show increases in H3K9 acetylation that are similar to those observed in our TSA treated cells. In addition, this report also shows de-repression of several canonical REST targets including neuronal β tubulin, synapsin, and VGF, consistent with our results. We include this new information in the Results section.

With this, we hope the reviewer will give us a pass for the HDAC knock down experiment.

3) In the Discussion, the authors make a statement regarding the link between REST and PRC2. The statement that “Polycomb family proteins were not present in the REST complexes characterized by mass....” This is not valid reasoning as to why REST and PRC2 are acting independently. These two complexes do not need to be associating together in order to function at same targets.

We apologize for our wording. Previous groups had suggested a direct role for REST in recruiting PRC2 to its targets, as well as the possibility that PRC2 directly recruited REST to its targets. Both of these scenarios required biochemical interaction. To better clarify our meaning, we have replaced the sentence in question with the following: “Second, the Polycomb complex member Eed, which is required for H3K27me3 deposition, was absent from the REST complexes characterized by mass spectrometry and co-immunoprecipitation, undermining the likelihood of either repressive complex directly targeting the other.”

4) During the purification of the REST complexes, the authors used sonication. This could be a reason why the REST and PRC2 link was not detected by the authors. The association of these two complexes may be nucleic acid dependent and sonication may be disrupting this interaction. The authors need to repeat their purifications using another method to disrupt the cells and cell fractionated to enrich for the nuclear fractions, not whole cell lysates.

Please see response to point 1, above.

Reviewer #2:

1) Very surprisingly, the authors do not comment on why there is a dramatic increase (of almost 40%!) of H3K27me3 target genes in Rest-/- cells as compared to wild-type ESCs. Additionally, the bioinformatic analyses of REST ChIP-seq are quite poor overall.

Thank you for bringing this to our attention. Addressing this issue further solidified our findings. To address the problem we did the following: First, we went back and completely reanalyzed our results in a manner independent of using Peak Ranger, using MACS only. Secondly, we re-analyzed our H3K27me3 peaks with consideration for how they overlapped with the dataset provided by the ENCODE project’s H3K27me3 ChIP-seq in mESCs, from Bing Ren at LICR. Only those peaks that were called in both our WT dataset and the LICR dataset were deemed ‘true’ H3K27me3 peaks. To ensure that this was an accurate representation of the peaks, we compared the read depth at these sites with four additional published datasets and were satisfied that they were highly correlated and were therefore valid (Figure 1–figure supplement 4). Importantly, with this new analysis, there is no difference in H3K27me3 peaks between WT and REST-/-cells. This result is now entirely consistent with our approach showing lack of correspondence between REST binding sites and PRC2 activity. We have modified the text to state the results and how we did the analysis more clearly.

2) It was previously published that REST and Polycomb are co-recruited via the ncRNA HOTAIR. It is thus possible that the PcG and REST do not necessary occupy the same nucleosome. All the analysis performed in Figure 1 is based on a “peak” overlap rather than looking at regions or genes.

In response to this concern, we extended the REST ‘peak’ 1kb up- and downstream (Broad REST sites) and re-ran the analysis. While the number of overlaps between PRC2 and REST does increase (57 to 270), this quantity is still a very small minority of H3K27me3 sites (2.3%) and REST sites (12.6%). The additional analysis is shown in figure 1A and described in the text.

3) The authors should also analyze PRC1, since CBX proteins have been demonstrated to interact with REST.

In response to reviewer 1 we re-did the proteomics using a different approach and there was still no evidence for PRC1 components in either analysis. We were aware of the CBX result (Ren and Kerppola, 2011, Mol. Cell. Biol.), but it is not interpreted easily because the effects of REST knockdown were opposite depending upon the chromatin context. Further, our current study is focused on PRC2 components and their relationship to REST regulation, a topic of current interest with respect to whether PRC2 alone mediates bivalent chromatin required for pluripotency of all lineages.

4) The data presented in Table 2 are very confusing. The H3K4me3 ChIP-seq is not mentioned and not properly analyzed in this manuscript, yet it is part of this table. It would be interesting to overlap REST ChIP-seq with H3K4me3 ChIP-seq and include this data to Figure 1.

Table 2 is no longer included in the manuscript. This is because we took the revision as an opportunity to update the definitions of H3K4me3 and H3K27me3-marked promoters from Mikkelsen et al (2007, Nature) to Young et al. (2011, Nucleic Acids Res.). After including the new definitions, there is no gene ontology difference within REST targets based upon these marks. Regarding the comment about H3K4me3 ChIP-seq, we have overlapped H3K4me3 peaks (Stamatoyannopoulos et al, 2012, Genome Biology) with our REST ChIP seq results and found that 37% of the REST sites located within 20kb of a TSS were also marked by H3K4me3, and 62% of those within 5kb were marked by H3K4me3. We have included this new information and it strengthens our idea that REST targets in ESCs are associated with H3K4me3, as this reviewer already acknowledges in point 7 below.

5) Data presented in Figure 1 should be normalized for nucleosome density (such as for histone H3).

We have conducted histone H3 ChIP and the normalization is now been included for figures 1D, 2A-C, 3E, and 4A, and we note that it has no effect on our conclusions.

6) The conclusions from Figure 2 are vague and not supported by the data presented. The authors state that “the presence or absence of the 3Me-H3K4 mark is an important part of the chromatin signature orchestrated by REST”. Deletion of Rest affected the H3K4me3 mark of each set of genes in all possible ways.

We apologize for the confusion. Our intent was to demonstrate that there is no correlation between the changes observed in H3K27me3 levels (increased, decreased, or unchanged) and the changes in H3K4me3. As the reviewer noticed, the deletion of REST affected the H3K4me3 mark in these H3K27me3-defined classes in all possible ways, indicating that H3K27me3 has no effect on H3K4me3 levels at REST sites. In addition, we have now included more gene promoters in our analysis of H3K4me3 levels after REST deletion (new Figure 4B) and re-analyzed the data. A majority of them gain H3K4me3 (at least 1.5-fold) after loss of REST. We interpret this result to suggest that H3K4me3 is an important part of the chromatin signature of REST.

7) Similarly, for data presented in Figure 3, Rest deletion affected gene expression of ESCs, EBs and neuronal precursors in all possible ways. I couldn't find a common trend in any set of genes.

The reviewer is exactly right, that there was no trend. We had tried to indicate that the results showed no correlation between changes in H3K27me3 levels and changes in gene expression due to loss of REST. We have now replaced this data with an RNA-seq analysis at the ESC stage showing more clearly this lack of correlation in Figures 3A and 3B.

The authors do not describe properly the results obtained or the model system they used. For a reader not familiar with the ESC differentiation protocols, it would be impossible to understand the rationale behind these experiments. In sum, there is no difference between the 4 groups of genes analysed in Figure 3A.

The reviewer arrived at the correct conclusion, namely that changes in gene expression are not dependent on changes to H3K27me3. To simplify and underscore this result in the resubmission, which will be unexpected to some, we have decided to focus solely on chromatin and gene expression changes at the ESC stage. Indeed, the changes that occur during neuronal differentiation in culture are not likely representative of the in vivo conditions, given the heterogeneity in progenitors and neuronal cell types. For this reason, the differentiation changes might best be studied by single cell transcriptome analysis, which is outside the scope of this study.

If the authors' only conclusion is that deletion of the Rest repressor leads to an increase H3K4me3, this is in my opinion already well-demonstrated. In any case, the author should comment on the fact that there are no differences in expression of the genes analysed in Figure 3 between wild-type and Rest-/- mature neurons, which is quite surprising.

The lack of difference in expression of REST target genes in fully differentiated neurons is not surprising because REST is expressed poorly in these cells; it is down-regulated at terminal differentiation to allow expression of terminal genes. However, as we indicated above, we have now removed this data to make the work more coherent.

It is very far fetched to draw any general conclusion when all the analyses performed in Figures 3B-3E are based on 12 genes. And it is unclear why the authors used only 12 of the 14 genes presented in Figure 3A.

In response to this comment, we did a power analysis to estimate the number of genes we should analyze, given the correlation we observe, to have a falsehood rate < 0.05. Based on this, we extended the analysis to 22 genes and now show that the correlation between H3K4me3 gain and increases in expression are significant (p<0.01, Figure 3E).

8) In my opinion the only interesting (novel) data in this manuscript are those presented in Figure 4B: the authors should start from here and develop this story further!

For example, they should perform this analysis in a genome-wide manner, with overlapping RNA-seq analysis in wild-type and REST -/- cells in the 4 stages (including also mature neurons).

Please see response to point 7 above. We have significantly extended the ESC analysis using RNA-seq, but have eliminated the neuronal differentiation aspect for clarity and in order to focus on PRC2 regulation in stem cells specifically, as the in vitro differentiation may not accurately reflect the in vivo situation.

In regards to the novelty issue, perhaps we obscured some of the novelty for this reviewer in the previous submission. We believe that there is substantial novelty in this work from the point of view of testing the idea, firm in the literature, that Polycomb bivalency is a good proxy for identifying poised genes in all lineages in pluripotent embryonic stem cells, and that PRC2 is required for poising of neuronal genes regulated by REST. In addition, many studies would suggest that simply the presence of polycomb proteins on a site infers repression, and for that reason other repressor mechanisms have been ignored. There are not many repressors that could provide good tests of the bivalent model because the complete set of targets have not been identified, and so it is not easy to discriminate functional significance of the binding. REST is unique: it binds to a unique sequence that allows identity of the binding sites both bioinformatically and by ChIP; the gene targets represent a large set of proteins that are essential for terminal neuronal differentiation, a specific transition during differentiation; and REST is present in ESCs, where these targets are repressed, and minimally expressed at terminal differentiation. Our study shows that Polycomb bivalency is not a general mechanism for poising genes: rather, poising may simply be an active repression mechanism conducted by any repressor (exemplified in this study by REST) balancing RNA Pol II activity, indicated by the H3K4me3 mark. Further, in the neuronal lineage, PRC2 represses proneural genes (Mohn et al, 2008, Mol. Cell, and Burgold et al, 2008, PLoS One), which are not REST targets as identified in our REST ChIP-seq analysis, while REST mediates repression of the later caste of genes required for terminal differentiation, which is a quite considerable number. Our study also demonstrates that in ESCs, the presence of the H3K27me3 mark does not necessarily equate with PRC2-based repression, as we find de-repression after loss of REST even at REST targets that also contain the H3K27me3 mark (e.g. of Calb1 and Glra1 genes). This is an important point and may explain the lack of de-repression observed at some H3K27me3-marked genes in PRC2 mutant cells (Shen et al, 2008, Mol. Cell). We hope the reviewer is more convinced of the novelty component in the new submission.

9) Once again, the conclusions of Figure 5 are vague and not supported by the authors' data: “These results suggest that REST repression in ESCs near a TSS is mediated primarily by recruited HDACs that serve as a counterbalance to basal RNA polymerase II activity, and that there is cross talk between HDACs and H3K4 trimethylation”.

The effect of TSA could be completely independent of REST. The correlation with H3K4me3 has nothing to do with REST. This figure only shows correlation between H3K4me3 and histone acetylation.

We address this concern in several ways. First, we now show a strong correlation between the changes in H3K4me3 due to loss of REST and those due to TSA addition at the specific promoters we have analyzed (new Figure 4B). This correlation suggests that TSA treatment is affecting H3K4me3 levels similarly to loss of REST. Second, we found that a microarray published previously treating ESCs with TSA shows de-repression specifically at REST target genes (new Figure 4C), albeit at lower levels than from loss of REST. Finally, we show that for the genes we analyzed, the change in expression due to TSA treatment was highly correlated with the change in expression due to loss of REST (new Figure 4D). While none of these data prove a direct link between REST, histone deacetylases, H3K4me3, and expression, they all strongly support the model whereby the REST/HDAC complex antagonizes the H3K4me3 mark and subsequent transcription. We have rewritten the Results and Discussion to reflect this idea.

10) The model presented in Figure 5D is wrong. Sox2 is not “bivalent” in ESCs, it is actually expressed. While is not expressed in neurons.

The present focus on ESCs necessitated eliminating this figure.