Transposable elements are regulated by context-specific patterns of chromatin marks in mouse embryonic stem cells

The majority of mammalian genomes are devoted to transposable elements (TEs). Whilst TEs are increasingly recognized for their important biological functions, they are a potential danger to genomic stability and are carefully regulated by the epigenetic system. However, the full complexity of this regulatory system is not understood. Here, using mouse embryonic stem cells, we show that TEs are suppressed by heterochromatic marks like H3K9me3, and are also labelled by all major types of chromatin modification in complex patterns, including bivalent activatory and repressive marks. We identified 29 epigenetic modifiers that significantly deregulated at least one type of TE. The loss of Setdb1, Ncor2, Rnf2, Kat5, Prmt5, Uhrf1, and Rrp8 caused widespread changes in TE expression and chromatin accessibility. These effects were context-specific, with different chromatin modifiers regulating the expression and chromatin accessibility of specific subsets of TEs. Our work reveals the complex patterns of epigenetic regulation of TEs.

10) One impactful conclusion of this work, is that the idea that TEs are uniformly marked by repressive histone modifications is outdated. This might be presented more forcefully in the Abstract and Conclusions.
Minor issues: -What is y axis of Fig5C? Unclear at present.
-Please show genome assembly version together with coordinates, in appropriate Figures.
- Figure S5 -What are the columns? What does the red line represent? Legend is not informative here.
- Fig 1B and others: Where appropriate, it would help to display the n number representing the number of unique TE instances that each figure is displaying.
Reviewer #2 (Remarks to the Author): He et al have performed a large-scale data mining effort to identify chromatin signatures associated with TEs in mouse ES cells, finding a variety of profiles across different subfamilies. Knockdown of a panel of TE-associated chromatin modifiers uncovered class-specific regulators of TE transcription.
This study is most useful in summarising the TE epigenomic landscape and in providing a useful resource in the form of ATAC-seq and RNA-seq data from knockdowns of key TE regulators. What is harder to pick out is which novel results have been uncovered, as many of the highlighted findings are already known, such as: 1) the regulation of L1s an IAPs by PRDM5-mediated arginine methylation, 2) the enhancer profile of RLTR13D6 and other elements, 3) the TE classes that are regulated by SETDB1. That the emerging patterns are complex and context-specific is also not a novel insight, as the independent evolutionary paths of different TEs, along with their age, are known to lead to these different epigenetic profiles. Nevertheless, I appreciate the strength of combining data mining with the mini knockdown screen, and could see this work becoming an important reference point for anyone interested in TE epigenetics.
Technically, however, the study suffers from major flaws in the analysis of ChIP-seq data: 1) Non-uniquely mapped reads were included in all analyses. Whilst this was not spelled out in the methods section, it is clear from the profiles displayed throughout, which do not have the expected decreased mappability across young TEs. The default output from bowtie2 will assign reads with multiple hits of equal quality to one of those locations at random. Therefore, mapping to individual TE copies is ambiguous and no claims can be made about which copies are enriched in which marks. This sort of mapping is only acceptable when generating class-wide profiles, such as those in Figure  1b. In contrast, profiles that display individual TE copies (e.g., heatmaps in Figure 1c and browser snapshots in Figure 1d) are inaccurate and misleading. To demonstrate the difference, I have remapped the H4R3me2 data and extracted uniquely mapped reads. In the attached file I show the profile for the L1Md_T element displayed in Figure 1d (see panel A) and the heatmap for L1Md_F3 elements from Figure 1c (see panel B). Importantly, this has a major impact on any classification that relies on the % of copies that bear a given epigenetic mark, which the authors use throughout. To use this classifier the authors have to: a) use uniquely aligned reads, and b) work out the % only relative to copies that have sufficient mappability to make the claim that a given epigenetic mark is present/absent. This will inevitably affect younger elements more than older ones. Claims that RLTRETN_Mm TEs bear both activating and repressive marks at the same copies are also potentially affected by this.
2) No peak calling algorithm was used. Even though it is missing from the methods section, the legend to Figure 1 states that TEs were classified as marked based on a cut-off of reads per million (but not normalized to TE length?). The problem with this approach is that it dampens the signal from localized peaks within TEs and does not take into account the background from an input sample. For example, the 5' UTR of L1 elements harbours many peaks for proteins associated with its regulation (see, e.g., PMID 29802231). Panel C of the attached figure shows how 5hmC (from the same dataset used by the authors) is highly enriched at full-length L1Md_T elements, yet was not picked up by the authors as an enriched mark. I also would have expected to see enrichment for at least H3K4me3 and H3K27ac at the same elements. Marking of a given TE copy should be based on peak calling, using the appropriate input samples. The ATAC-seq data generated by the authors are an exception and did involve peak calling, but it looks again like this was based on non-uniquely mapped reads.
3) Full-length and truncated elements are analysed together. One cannot generalize the pattern of a TE class when full-length elements are pooled together with much shorter elements that bear only part of the coding or regulatory regions of that TE. This affects L1 elements most prominently, as the vast majority of truncated elements lacks the regulatory 5'UTR. Therefore, 5hmC enrichment can be seen at full-length elements, but not truncated ones (see panel C of the attached figure), and the same will be true of other marks associated with the 5' UTR. This is less of an issue with LTR elements, as the Repeatmasker annotation separates LTR ends from the coding region.
The manuscript therefore needs a major reshuffle of its bioinformatic approaches, to avoid becoming a minefield of artefacts associated with the complexity of analyzing short sequencing data from TEs.

Reviewer #3 (Remarks to the Author):
This manuscript provides data relevant to mechanisms involved in the substantial context-specificity of the chromatin modification system, which marks and regulates TE using a regulatory chromatin code. The authors showed that TEs are marked not only by heterochromatic marks, but are labelled by all major types of chromatin modification in complex patterns in mESC. They knocked-down 41 chromatin modifiers and found 29 significantly deregulated at least one type of TE. Significant effect was found in the loss of Setdb1, Ncor2, Rnf2, Hdac5, Prmt5, Kat5, Uhrf1, Rrp8 and Ash2l, which caused widespread changes in TE expression and matching changes in chromatin opening. These effects were context-specific, as different chromatin modifiers regulated the expression and chromatin accessibility of specific subsets of TEs. Some of the content are relative novel, however, the manuscript covers many problems and needed to be largely and carefully improved.
The novelty is moderate. The epigenetic regulation of transcription, and the relationship between the chromatin marks and the chromatin factors are well known and not novel at all.

2.
The functional validation of TE regulation is lacked. Does the changed of TEs caused by CMs affect genome stability?

3.
The information in the introduction is too simple and incomplete. The information about TE function, regulatory mechanism and TE-related diseases are not included. The previous publications about the relationship between TE and epigenetic regulation are not described. The current achievement and key issues needed to be solved are not clear.

4.
This manuscript shown that TEs are also marked by active chromatin marks, such as H3K4me1, and H3K4me3. Moreover, the expression of the TEs are down regulated when CMs are knocked down. The regulatory mechanisms of these TEs have not been studied and explained. Are they also suppressed in ES cells?

5.
As shown in the manuscript, in agreement with the prevalence of chromatin marks at TEs, CMs were also often bound to TEs in specific patterns. Is the pattern of CMs consistent with chromatin markers? The manuscript gives a few examples, however, the global analysis is lacking.

6.
To rule out the change of cell fate in response to CM KD, the detection of pluripotency and differentiation gene expression is not enough. Protein expression level and ES cell phenotype should be detected.

7.
To prove the relationship between TE expression and chromatin markers, or the relationship between the change of chromatin markers and chromatin state after the depletion of CMs, there is a lack of global wide statistics. The current data are difficult to explain it adequately, because the examples given are different in two parts.

8.
In the part of that chromatin modifiers alter chromatin accessibility at specific sets of TEs, the TEs with up-regulated expression and opened chromatin are described. It is necessary to describe the TEs with down-regulated expression and closed chromatin.

9.
"we chose to KD over a short time period of 4 days……" at page 8. 4 days after KD is the standard time point to check ESC differentiation. The authors should provide the reasons and evidence why they chose the time point.
The manuscript could be written in a more concise manner to transmit clearer its important message. The English should be largely revised, for example at page 2, potential danger to genomic stability or instability? Page 3. Cell's DNA should be instead by cellular genome. Page 10, "although" and "nonetheless" should not coexist in one sentence. Page 11, "looking at" is so informal in manuscript……I hope the author should pay more attention to improve the whole English grammar and the verbal accuracy, not only revise what I mentioned here.

2.
There are many sentences with grammar mistake, which need to be carefully revised by the author. Such as, page 2, "particularly potent" is an adjective and cannot be used as subject. Page 7, "As many of the IAPs had methylated DNA (Fig.1c), and, as previously reported, these IAPs were bound by MBD-family proteins." In page 7, "also" appeared twice in the next sentence. The meaning of "Overall the RLTR46……of the TE RLTR46 copies" is confused. Page 13, "of the KDs, ……" also has serious grammar mistake. Page 14, "Although whether all of these pathways...". "although" cannot be used in the front of an independent sentence. The author should also pay attention to the repeated use of words. Word's diversity should be encouraged. Page 7, "as previously reported" appeared twice nearby.

5.
The author should pay attention to gene versus protein terminology.

6.
Line 168, the protein pol II should be written as Pol II.

7.
Page 3 line 49: A correct punctuation mark is missed.
8. Fig.1C, the authors should provide the color bar and scale bar, as well as for other heatmaps. This work represents an important, fundamental advance in the field. It is ambitious in its scale, tackling the entirety of TEs, dozens of chromatin marks. The data produced will be an invaluable resource, of which none presently exists. The main finding, that TEs are marked by complex and specific histone marks, many of which are activating (not repressive) is an important finding.
The paper is well written (although use of commas is notably lacking, particularly in Abstract).
The figures are attractive and informative. I detect no major conceptual or methodological weaknesses. Nevertheless there are a number of issues that are raised by the paper, often because the scale and fundamental nature of the work raises many interesting questions for follow up work. Therefore, several of the comments below might fall outside the scope of the paper, but nevertheless are interesting to mention.
1) A main concern when working with TEs, is their repetitive nature and consequent difficulty in mapping NGS reads to them, and ambiguity as to precise origin of TE-mapping reads, or even the discarding of such reads due to multi-mapping. Could the authors comment on: how this is dealt with in the analysis / to what extent this affects (or not) this work and conclusions arising from it / what are possible artefacts in the data arising from these effects? This could be mentioned also in the main manuscript, even if briefly.
Response: We thank reviewer for this question. We have added a small part to the results section that briefly mentions how we deal with multimapped reads (marked in red in the text).
This section highlights to the reader how we consider TEs as metagenes, and how the analysis of individual TE copies can be ambiguous. See the reply to reviewer 2 below, which also deals with the issue of multimapping reads.
2) Fig 1A and others: I am missing well-known TEs like B1, B2, MIR, L2 in this Figure and others. Were they included in analysis? Could they be indicated in the relevant figures?
Response: For Figure 1a we only show those TEs with high levels of fold-enrichment. The full set of TEs used, including L2's is in Figure S1b and Supplementary Table 2. We found that the analysis of short <300 bp TEs was not accurate using for ChIP-seq data, using our methods.
Hence B1, B2 and MIR SINEs were not included as we deleted TEs with a short average length. These TEs are included in the RNA-seq analysis, as with the longer reads we are more confident that we can detect them accurately. However, as we did not include them in the ChIP-seq analysis, we do not comment on them in the text. We have added this detail into the methods (marked in red in the text).
3) Fig 1B and  predates the cited papers on Line139. Please cite this, and also, indicate on Figure 2A which TEs REST is here binding to, and whether this supports previous findings.
Response: We thank the reviewer for drawing our attention to this paper, and have cited it in the revision. The Johnson et al., paper indicates REST is binding to L2 LINEs. However, we did not find any evidence of REST binding to L2's in our analysis (Supplementary Table 3).
An explanation for this discrepancy is that the Johnson et al study mainly looked at the human genome, whereas we used the mouse genome. L2 elements are more common in the human genome (~3.3%) versus the mouse (~0.3%) (Waterston et al., 2002), hence this may be a human-specific effect. We do see enrichment of REST on some ERVs (RMER21A, and RLTR44C), and some IAPs (IAPEY2_LTR, IAPEY3_LTR, and IAPEY_LTR). However, all of these TE types are predominantly found in the mouse genome and not the human genome.
These results are in Supplementary  Figure 2b), however, we don't cover many TFs in this manuscript, as we focus on chromatin modifiers which mostly lack sequence-specific DNA binding domains (the list of chromatin modifiers was based on the Epifactors database, which sometimes overlaps with TFs (Medvedeva et al., 2015)). We briefly explore two CMs/TFs that have DNA-binding domains and were in our analysis, YY1 and REST. We detected binding of YY1 to IAPLTR2b and REST to RMER21A, and have added this to Supplementary Figure 2c.
In the course of this analysis, we found an interesting link between NR5A2 and RLTR13B. This link is interesting as NR5A2 is important in reprogramming primed ESCs (EpiSCs) to naïve ESCs (Guo and Smith, 2010). We use previously published ChIP-seq, RNA-seq and ATAC-seq data to show a strong correlation between NR5A2 binding at RTLR13B2 and RNA expression and chromatin opening in naïve ESCs (Supplementary Figure 2e-i). We would like to thank the reviewer for suggesting this idea, which adds a new aspect to our manuscript. 9) An interesting experiment suggested by this work, would be to knock down or overexpress candidate TE RNAs to promote the acquisition of the 2C state in ESCs.
Response: This is an excellent suggestion. However, it is technically challenging as traditional tools for manipulating genes do not work well when applied to TEs. shRNAs are notoriously poor at knocking down TEs, whilst plasmids suitable for efficient overexpression in ESCs are lentiviral-based and contain LTRs, which makes cloning other LTRs, or even ERVs inside these vectors inappropriate or even potentially dangerous. Consequently, this work requires the establishment of genome manipulation systems such as CRISPRa/CRISPRi, and is beyond the scope of the current work.
10) One impactful conclusion of this work, is that the idea that TEs are uniformly marked by repressive histone modifications is outdated. This might be presented more forcefully in the Abstract and Conclusions.
Response: We are grateful for this suggestion. We have added more stress to this in the revised manuscript, and have added new native-ChIP-qPCR data which shows that for a select number of individual genomic TEs, they are simultaneously marked by both the repressive H3K9me3 and activatory H3K27ac (Figure 1c, d).
Minor issues: -What is y axis of Fig5C? Unclear at present.
Response: y-axis is the RNA-seq expression level of the genes at the indicated embryonic stage. We have reworded the figure legend to make this clearer.
-Please show genome assembly version together with coordinates, in appropriate Figures.
Response: We have added the genome assembly to the figures.
- Figure S5 -What are the columns? What does the red line represent? Legend is not informative here.
Response: We thank for reviewers point out this, Figure S5a columns represent the percent of reads mapping to the TEs, for the RNA-seq data for each KD. The redline is the % mapping for the shLuc sample. We have amended the figure legend.
Response: We have corrected these errors.

Reviewer #2 (Remarks to the Author):
He et al have performed a large-scale data mining effort to identify chromatin signatures associated with TEs in mouse ES cells, finding a variety of profiles across different subfamilies.

Knockdown of a panel of TE-associated chromatin modifiers uncovered class-specific regulators of TE transcription.
This study is most useful in summarising the TE epigenomic landscape and in providing a useful resource in the form of ATAC-seq and RNA-seq data from knockdowns of key TE regulators. What is harder to pick out is which novel results have been uncovered, as many of the highlighted findings are already known, such as: 1) the regulation of L1s an IAPs by PRDM5-mediated arginine methylation, 2) the enhancer profile of RLTR13D6 and other elements, 3) the TE classes that are regulated by SETDB1. That the emerging patterns are complex and context-specific is also not a novel insight, as the independent evolutionary paths of different TEs, along with their age, are known to lead to these different epigenetic profiles.
Nevertheless, I appreciate the strength of combining data mining with the mini knockdown screen, and could see this work becoming an important reference point for anyone interested in TE epigenetics.
Response: We would like to thank the reviewer for seeing the merit in our work. to our knowledge, the epigenetic profiles of TEs has never been presented in this detail before.
Additionally, whilst these things may be known amongst researchers that are familiar with TEs, outside of TE focused researchers it is poorly appreciated. When we present this work to non-TE biologists there is surprise that TE chromatin is so complex and dynamic. There remains a view in the wider genomic field that TEs are irritating background 'noise' that make a mess of the analysis of ChIP-seq. Indeed, often the first step in ChIP-seq analysis is the removal of reads mapping to repeats, or the deletion of multimapping reads. We believe our paper can be a useful counterpoint to these views, by collecting and summarising the epigenetic regulation of TEs.
Technically, however, the study suffers from major flaws in the analysis of ChIP-seq data: 1) Non-uniquely mapped reads were included in all analyses. Whilst this was not spelled out in the methods section, it is clear from the profiles displayed throughout, which do not have the expected decreased mappability across young TEs. The default output from bowtie2 will assign reads with multiple hits of equal quality to one of those locations at random. Therefore, mapping to individual TE copies is ambiguous and no claims can be made about which copies are enriched in which marks. This sort of mapping is only acceptable when generating class-wide profiles, such as those in Figure 1b. In contrast, profiles that display individual TE copies (e.g., heatmaps in Figure 1c and browser snapshots in Figure 1d) are inaccurate and misleading. To demonstrate the difference, I have remapped the H4R3me2 data and extracted uniquely mapped reads. In the attached file I show the profile for the L1Md_T element displayed in Figure 1d (see panel A) and the heatmap for L1Md_F3 elements from Figure 1c (see panel B). Importantly, this has a major impact on any classification that relies on the % of copies that bear a given epigenetic mark, which the authors use throughout. To use this classifier the authors have to: a) use uniquely aligned reads, and b) work out the % only relative to copies that have sufficient mappability to make the claim that a given epigenetic mark is present/absent. This will inevitably affect younger elements more than older ones.
Claims that RLTRETN_Mm TEs bear both activating and repressive marks at the same copies are also potentially affected by this.
Response: We thank the reviewer point out this, and agree that our original measure of '% elements marked' is misleading. We have changed our analysis method, and we treat all TEs as 'metagenes' in the paper and generate class-wide profiles. We also reduce the emphasis in the paper on specific TE copies, and have deleted Supplementary Figure 6, for which the conclusions were unreliable without precise mapping of TE copies. This change in technique has resulted in us changing the measure we use to a fold-enrichment score, as used in (Day et al., 2010;Elsasser et al., 2015;Goldberg et al., 2010;Wu et al., 2016). We use a few individual genomic loci to illustrate the data, as we find these are helpful for the reader to visualize the genomic impact of the chromatin regulation of the TEs. This is in line with many other publications that deal with multimapped reads in a similar way to us (Bulut-Karslioglu et al., 2014;Theunissen et al., 2016). We have altered the methods section to make it clearer how we are dealing with multimapped reads (changes marked in red in the text), and we have added text to the results section that describes issues of mapping ambiguity in TEs (changes marked in red in the text).
For simultaneous marking of TEs by both activatory and repressive marks, we have added a new section in which we perform native-ChIP-qPCR with primer pairs that target a single genomic locus (Figure 1c, d). This data highlights that, at least for these examples, TEs can harbor bivalent repressive/activatory marks at the same time.
2) No peak calling algorithm was used. Even though it is missing from the methods section, the legend to Figure 1 states that TEs were classified as marked based on a cut-off of reads per million (but not normalized to TE length?). The problem with this approach is that it dampens the signal from localized peaks within TEs and does not take into account the background from an input sample. For example, the 5' UTR of L1 elements harbours many peaks for proteins associated with its regulation (see, e.g., PMID 29802231). Panel C of the attached figure shows how 5hmC (from the same dataset used by the authors) is highly enriched at full-length L1Md_T elements, yet was not picked up by the authors as an enriched mark. I also would have expected to see enrichment for at least H3K4me3 and H3K27ac at the same elements. Marking of a given TE copy should be based on peak calling, using the appropriate input samples.
The ATAC-seq data generated by the authors are an exception and did involve peak calling, but it looks again like this was based on non-uniquely mapped reads.
Response: Because ChIP-seq peak calling algorithms rely on the patterns at individual TE copies, and due to the unreliably of using individual elements (as highlighted above), we have altered the analysis to only consider TEs as 'metagenes', and to use fold-enrichment (observed/expected) versus the background as our analysis metric. We now mimic peak discovery within the TEs, by taking the maximum signal in any 500 bp bin across the TE, as our enrichment score. In this way it is similar to peak calling, as it allows us to detect localized enrichment within longer TEs (See for example Supplementary Fig S1d, S1e, and this extends to the analysis of all chromatin marks, chromatin modifiers and ATAC-seq data in the manuscript) This change in analysis technique can now correctly detect multiple chromatin marks at L1Md_T (Supplementary Fig S1d, S1e).
For RPM, we would like to apologize as we did not make it clear that all of the measures in the paper were already normalized to TE length, and should have been described as 'RPKM'. We have corrected this in the revised manuscript.
3) Full-length and truncated elements are analysed together. One cannot generalize the pattern of a TE class when full-length elements are pooled together with much shorter elements that bear only part of the coding or regulatory regions of that TE. This affects L1 elements most prominently, as the vast majority of truncated elements lacks the regulatory 5'UTR. Therefore, 5hmC enrichment can be seen at full-length elements, but not truncated ones (see panel C of the attached figure), and the same will be true of other marks associated with the 5' UTR. This is less of an issue with LTR elements, as the Repeatmasker annotation separates LTR ends from the coding region.
Response: We are grateful for reviewer point out this. We have added a new section in the manuscript which specifically discusses chromatin at full length and truncated LINE L1 elements.
During the change of methods from a '% marked' to a 'fold-enriched', we also took the opportunity to change how we detected enrichment. We now take the maximum scoring 500 bp bin window across the metagene for each TE type. This has the effect of more accurately detecting localized enrichments within the TE, particularly for longer TEs.
The manuscript therefore needs a major reshuffle of its bioinformatic approaches, to avoid becoming a minefield of artefacts associated with the complexity of analyzing short sequencing data from TEs.
Response: We would like to thank the reviewer for their insightful comments, which we have used to improve the robustness of the bioinformatic analysis. We have tried to accommodate all of reviewer 1 and 2's comments as best we can, however there is some conflict between the advice given by each reviewer. Reviewer 1 asked us to expand the analysis of individual elements, whilst reviewer 2 cautioned us against the ambiguities of analyzing individual elements. We have tried to reconcile these differences, and our current analysis is in line with common practice in the field, and closely follows analysis techniques published in (Bulut-Karslioglu et al., 2014;Day et al., 2010;Elsasser et al., 2015;Goldberg et al., 2010;Theunissen et al., 2016;Wu et al., 2016), and from other leaders in the field.

Reviewer #3 (Remarks to the Author):
This manuscript provides data relevant to mechanisms involved in the substantial context-specificity of the chromatin modification system, which marks and regulates TE using a regulatory chromatin code. The authors showed that TEs are marked not only by heterochromatic marks, but are labelled by all major types of chromatin modification in complex patterns in mESC. They knocked-down 41 chromatin modifiers and found 29 significantly deregulated at least one type of TE. Significant effect was found in the loss of Setdb1, Ncor2, Rnf2, Hdac5, Prmt5, Kat5, Uhrf1, Rrp8 and Ash2l, which caused widespread changes in TE expression and matching changes in chromatin opening. These effects were context-specific, as different chromatin modifiers regulated the expression and chromatin accessibility of specific subsets of TEs. Some of the content are relative novel, however, the manuscript covers many problems and needed to be largely and carefully improved.
Major points: 1. The novelty is moderate. The epigenetic regulation of transcription, and the relationship between the chromatin marks and the chromatin factors are well known and not novel at all.
Response: This comment was overruled by the editor.

The functional validation of TE regulation is lacked. Does the changed of TEs caused by
CMs affect genome stability?
Response: This is an excellent idea from the reviewer, but we believe this is beyond the scope of the current manuscript. To detail this, we would need knockouts or prolonged CM knockdowns which would need to be followed up with detailed genome wide sequencing, karyotyping and copy number variation measurements. As our current manuscript focuses on the short-term deregulation of TEs, we prefer to leave this work for future studies. We speculate that the answer to this question would be yes, for example, based on the Setdb1-/embryos, which show genome instability. We think this would be an excellent follow up study.
3. The information in the introduction is too simple and incomplete. The information about TE function, regulatory mechanism and TE-related diseases are not included. The previous publications about the relationship between TE and epigenetic regulation are not described.
The current achievement and key issues needed to be solved are not clear.
Response: We have rewritten the introduction according to the reviewer, adding in extra descriptions of TE functions, their regulatory mechanism, and known epigenetic regulation.
We have focused the abstract to more clearly communicate the key issues and the results.
4. This manuscript shown that TEs are also marked by active chromatin marks, such as H3K4me1, and H3K4me3. Moreover, the expression of the TEs are down regulated when CMs are knocked down. The regulatory mechanisms of these TEs have not been studied and explained. Are they also suppressed in ES cells?
Response: We were interested to explore this, it was surprising to us that only a few TEs were down-regulated. One possible explanation for this is that our miniscreen did not include enough activators, or that activators are redundant and several KDs are required to see an effect. Another explanation is that the TEs are marked in a bivalent pattern, with both activatory and repressive marks, and manipulating one or other may not lead to repression.
We have added text to the discussion (changes marked in red in the text).
Due to the lack of a large number of down-regulated TEs, we are unable to perform a global analysis. However, we have briefly explored this issue for Chd4 (see figure below), which has the largest number of down-regulated TEs types. Briefly, we see a selection of TE types that are downregulated, including some LINE and IAP elements, but mostly ERVs. CHD4 acts as a cofactor for the NuRD complex, and does not have direct catalytic activity, so linking CHD4 to specific chromatin marks is difficult. Additionally, Chd4 KD showed strong signs of differentiation ( Supplementary Fig. S4e, S4f), which makes any analysis of this KD challenging. Consequently, we have not added this analysis to the manuscript. We hope to explore the downregulation of TEs more fully in other studies, if we can discover chromatin regulators that lead to the widespread downregulation of TEs. Due to the lack of robust down-regulation of TEs we do not explore this in the current manuscript, beyond the comments we have added into the discussion.  we did ATAC-seq for (Supplementary Fig. S5c) G9A, SUV39H1), are part or multi-protein complexes (RNF2, SETDB1, HDACs), or are co-factors with no catalytic domains (NCOR1/2, ASH2L). Consequently, KD of a single chromatin modifier could affect a wide range of chromatin marks, particularly as compensation mechanisms are known to be activated, for example (Berrens et al., 2017), making their global analysis challenging. Whilst it would be desirable to add ChIP-seq of specific chromatin marks, in practice a large number of both direct and indirect chromatin marks would need to be performed and this would need to be done on a case-by-case basis. This makes ChIP-seq prohibitive if we aim to globally survey chromatin changes. Consequently, we decided to use ATAC-seq as a proxy for changes in chromatin marks, this allowed us to uniformly analyze chromatin changes across the different KDs.
8. In the part of that chromatin modifiers alter chromatin accessibility at specific sets of TEs, the TEs with up-regulated expression and opened chromatin are described. It is necessary to describe the TEs with down-regulated expression and closed chromatin.
Response: This comment is related to comment 4 above. Whilst we find it extremely interesting to probe down-regulated TEs, we do not have enough of them in our dataset to draw meaningful conclusions.
9. "we chose to KD over a short time period of 4 days……" at page 8. 4  8. Fig.1C, the authors should provide the color bar and scale bar, as well as for other heatmaps.
Response: We apologize for these omissions. Due to a reorganization of the methods, the only remaining pileup heatmap is in Supplementary Fig. 1f.