Region-specific H3K9me3 gain in aged somatic tissues in Caenorhabditis elegans

Epigenetic alterations occur as organisms age, and lead to chromatin deterioration, loss of transcriptional silencing and genomic instability. Dysregulation of the epigenome has been associated with increased susceptibility to age-related disorders. In this study, we aimed to characterize the age-dependent changes of the epigenome and, in turn, to understand epigenetic processes that drive aging phenotypes. We focused on the aging-associated changes in the repressive histone marks H3K9me3 and H3K27me3 in C. elegans. We observed region-specific gain and loss of both histone marks, but the changes are more evident for H3K9me3. We further found alteration of heterochromatic boundaries in aged somatic tissues. Interestingly, we discovered that the most statistically significant changes reflected H3K9me3-marked regions that are formed during aging, and are absent in developing worms, which we termed “aging-specific repressive regions” (ASRRs). These ASRRs preferentially occur in genic regions that are marked by high levels of H3K9me2 and H3K36me2 in larval stages. Maintenance of high H3K9me2 levels in these regions have been shown to correlate with a longer lifespan. Next, we examined whether the changes in repressive histone marks lead to de-silencing of repetitive DNA elements, as reported for several other organisms. We observed increased expression of active repetitive DNA elements but not global re-activation of silent repeats in old worms, likely due to the distributed nature of repetitive elements in the C. elegans genome. Intriguingly, CELE45, a putative short interspersed nuclear element (SINE), was greatly overexpressed at old age and upon heat stress. SINEs have been suggested to regulate transcription in response to various cellular stresses in mammals. It is likely that CELE45 RNAs also play roles in stress response and aging in C. elegans. Taken together, our study revealed significant and specific age-dependent changes in repressive histone modifications and repetitive elements, providing important insights into aging biology.


Reviewer#1
The following are included as constructive comments for the authors.
1. The authors introduced ASRRs in the discussion, but I would have found it helpful to have the term introduced earlier in the results, to make it clear which class of peaks are being referred to. For example, it takes some effort to realize that the 578 ASRRs are the same as the differentially-enriched peaks identified in line 234, and to remember that these represent less than 5% of all H3K9me3 peaks called. a. Furthermore, it is not clear whether ASRRs are solely peaks that exist in both young and aged populations (as is mildly indicated by language in lines 233-234), or whether they also include new peaks that are only found in aged populations (as would seem to be indicated by lines [441][442]. This could be resolved by providing more detail in the results section. We appreciate the reviewer's suggestion. In the revised manuscript, we introduced ASRRs (578 upregulated H3K9me peak regions) in the result section (line 286-287) and discussed it throughout the text. To improve clarity, we have also added the label "ASRRs" in the relevant figures (Fig 2B, 3A, 3C, S6A, S7A) and sentences in the main text.
2. If I'm interpreting the figures correctly, ASRRs are never examined as a class (but, as mentioned above, it'd be easier to know that for certain if ASRRs were named earlier and the term used throughout the results section). It seems like a missed opportunity to examine the genomic & transcriptomic landscape of ASRRs more closely. For example, how does gene expression change at these 578 peaks compared to all H3K9me3 peaks? Is it more or less drastic than shown in Fig S5A? Do ASRRs in aged populations experience even higher levels of blurring & spreading of active modifications? a. Is the opposite true for the handful of down-regulated H3K9me3 peaks that make up the rest of the 595 significantly changed peaks? We agree with the reviewer's suggestion and made the suggested revisions. We have made new plots (Fig 2A-B and S7) in which H3K9me3 peak regions were divided into three groups based on age-dependent changes in repressive marks (non-significant, significantly upregulated or significantly downregulated with age). We compared the levels of active histone marks (S7 Fig) and the RNA expression of genes (S2A-B) in the three groups of H3K9me3 peaks and non-peak regions. Fig 2E- We have added the N number in figures and/or figure legends.

It was hard to understand how Fig 2A-D is related to
5. The references in lines 489-490 don't seem to match with the text. We have corrected the errors due to formatting issues.
7. The writing is mostly clear, but there were numerous grammatical errors which would be easily caught by a proofreader. For example, "C. elegans" should be italicized throughout. We have corrected the errors.

Reviewer #2
Major questions: All these ChIP-seq experiments were done using glp-1 to focus on somatic changes. glp-1 mutants are long-lived, in addition to lacking a germline. Could it be that the increased longevity of the glp-1 mutants impacts the age-associated changes in chromatin that the authors observe? Is there any evidence that the age-associated changes in glp-1 mutants are recapitulated in other mutants that lack a germline but are not long-lived (glp-4, fer-1, etc)?
We agree with the reviewer's comment. glp-1 mutants were used to remove the germline and facilitate the harvest of somatic tissues. The age-associated changes observed in the glp-1 mutants could be affected by the altered longevity associated with the glp-1 mutants. As the reviewer suggested, we carried out preliminary investigation using glp-4(bn2) mutant, a temperature-sensitive germlineless mutant that has been reported to not exhibit altered longevity (Syntichaki et al., Nature 2007). Details of these results can be found at the end of this document on pg 9-10.
Line 159-164 the authors claim that the distribution of repressive marks in glp-1 is similar to wild-type and that the profiles for H3K9me3 and H3K27me3 between glp-1 adults and wild-type L3. They should show the data supporting these arguments, as they do for active marks in Fig  S3B. We appreciated the reviewer's suggestion. We have added two plots showing the profiles of H3K9me3 and H3K27me3 in wild-type L3 larvae in S3A Fig. The authors claim that MDS analysis suggests that there are "substantial" age-dependent differences in H3K9me3, but "relatively minor" age-dependent variations in H3K27me3 samples (Fig 1A,B). It is not clear to me that this is an accurate description of the data, as MDS shows a simplified projection of distances between data in only two dimensions. How did the authors evaluate the statistical significance of these differences to make this inference? In the MDS, does 2 dimensions capture the variance in the samples, and how did the authors evaluate this?
We agree with the reviewer's comment. The main purpose of the MDS plot is to visualize the clustering of samples rather than to test for the significance of the changes. We have revised the relevant sections and removed those sentences. Age-dependent changes in repressive marks were examined by differential analysis (line 218-226). Further analysis of the average changes in the enrichment of repressive marks were shown in the metaplots in Fig  The data in Fig S3 seem to show that age-associated changes mostly change the magnitude of enrichment of both H3 modifications rather than a wholesale redistribution. This aspect of the data is lost when only the difference between young and old (as in Fig 1E) is shown. This could be misleading to a reader, so I think the authors should include Fig S3A in the main text. The authors should also include labels on the Y-axis of FIg 1E (and all similar plots) to indicate the magnitude of enrichment that is being shown. It would also be helpful to know which changes are statistically significant when data like these are shown. Finally, the authors should also be more circumspect in claiming "redistribution" in the text of the manuscript.
We agree with the reviewer's comment. We have used "region-specific gain and loss" to replace "redistribution" in the text.
The authors state that "most of the peak regions that lost repressive H3K9me3 or H3K27me3 marks were not statistically significant at FDR cutoff of 0.05" (line 235-236). But then they claim a correlation of increased gene expression at loci with "downregulated" repressive marks ( Fig  2). I don't understand how the authors justify looking for correlations between changes in expression with non-significant changes in repressive marks.
We appreciate the reviewer's comment and have made revisions to improve clarity. In Fig 2E-F, we examined the RNA expression changes of differentially expressed genes that are associated with the repressive peak regions that showed statistically significant changes in H3K9me3 or H3K27me3 (line 250-255).
In Fig 2, the authors claim a statistically significant difference between curves based on the KS test. However, some of the differences are quite small. There are also large differences in the number of genes included in the two groups (especially in C and D). The authors need to explain how they ensured that the statistical test was sufficiently powered to detect the changes they report. If they perform these tests on independent data sets are the conclusions robust? Does the reciprocal comparison support these conclusions (ie looking for enrichment of H3K27me3 peaks in genes that are up-or down-regulated by RNAseq)? With the data provided it is nearly impossible for a reader to appreciate the robustness or biological significance of this conclusion.
We appreciate the review's comment. We have revised the plots in Fig 2A-D. Power analysis helps to determine whether the lack of statistical significance is due to the lack of statistical power. It is not a concern when there is a statistically significant difference. The pairwise KS tests between upregulated peaks, non-significant peaks, and non-peaks are all highly significant (p-value < 0.0001) despite some of the differences appear small. Lack of power did affect the downregulated repressive peak regions, which have small sample sizes and nonsignificant KS tests. The results are described in line 231-243 and the figure legends (Fig 2 and  S4 Fig).
The authors state that "the majority of genes associated with repressive H3K9me3 and H3K27me3 peak regions were silent" but that "significantly upregulated repressive peak regions were preferentially associated with actively expressed genes" (line 276). This sentence is a bit confusing and should probably be edited for clarity; We appreciate the reviewer's comment. We have made revisions to make it clear that the upregulated repressive peak regions only represent a small fraction of the total repressive peak regions. In this small faction of repressive peak regions, genes tend to be actively expressed.
moreover, I can't figure out where they show this with their data? Doesn't this statement contradict the data shown in Fig 2? Also in this paragraph, the authors claim that the fact the 'upregulated' peaks were lacking repressive marks in young animals supports the idea that these genes were expressed in younger animals. This seems like a circular argument. If you start by looking only at "new" peaks then it seems evident that the repressive marks had to be missing in the young animals, by definition. Does the RNAseq data support the assertion these genes are expressed at day 2 and not day 12?
We appreciate the reviewer's comment. We have found that most genes in upregulated repressive peak regions are transcriptionally active at both young and old time points by using our published RNA-seq data results (Line 273-277). We have revised the text to clarify this: The genes that were associated with upregulated repressive peaks and were transcriptionally active at young age remained transcriptionally active at old age.
The authors assert that "most of the actively expressed CELE45 copies in glp-1 adults had been actively transcribed starting at larval stages" and later conclude that the upregulation of this SINE is not associate with loss of repressive marks. It's not clear to me that this has anything to do with the age-associated changes in histone methylation that the authors focused on.
Previous studies in other organisms have found upregulation of transposons at old age correlates with the loss of repressive H3 marks. In this study, we found that despite high levels of repressive H3 markings in the CELE45 repeat regions (S11F and S11G Fig), many CELE45 copies remained transcriptionally active. These active CELE45 copies showed a significant increase in RNA expression at old age, which is not due to the loss of repressive H3 marks. This point is clarified in line 505-518.
Minor points -Line 25 and 27 -consider changing "significant" to something else in order to avoid perception you are making a statistical argument. We have replaced the "significant" in Line 26 with "evident". The "significant" in Line 29 does mean "statistically significant".
-Several different sentences in abstract start with "interestingly". The authors should edit the text to avoid overuse of conjunctive adverbs. We have reduced the use of "interestingly" in the text.
-The authors should consider deleting the first paragraph of the Introduction. The second sentence of the introduction is vague and starts with a reference ("these epigenetic marks") that is confusing. Which marks? The first sentence is about deterioration of chromatin structures and epigenetic information, so the reference to epigenetic marks doesn't really make sense. The third sentence is redundant with the first sentence, and the last sentence doesn't really add any information. We agree with the reviewer's comment and removed the first paragraph of the introduction.
-The text needs to be edited for grammar and clarity. There are several sentences that are awkward throughout the text. For example, the second sentence of the abstract (line 20) starts with "Dysregulated epigenome has been linked..." seems to be missing an article. I also noticed an over-abundance of conjunctive adverbs and a few places with random changes in tense. We appreciate the reviewer's comment and have made revisions.
-It may be more correct to say that the dysregulation of repressive heterochromatin is associated (rather than "linked") with aging (line 69). We appreciate the reviewer's comment and have made revisions (Line 62).
-Line 150, what is the "nevertheless" doing in this sentence?
We have removed the "nevertheless".
-What do the numbers on the top left of each trace in Fig S3 mean? We have revised the figure legend to clarify that those numbers indicate the y-axis range (zscores).
-The authors say that H3K27me3 is enriched in H3K9me3-depleted central regions of chr II and IV, but data in Fig S3 seems to show that there is also more H3K27me3 signal in the middle of chr V. It's not clear why this was not mentioned by the authors.
We agree with the reviewer's comment. We have removed the "II and IV" because H3K27me3 is more enriched in the middle of all chromosomes than H3K9me3.
-The authors merged neighboring peaks if they were within 5 kb (line 152-153). I don't understand the justification for doing this -are there data showing that peaks are generally this large in C. elegans? What is the average size of a peak in the author's data without this manipulation?
The main reason is to adjoin neighboring peaks which in fact represent one long enrichment domain (see S2A Fig, Line 143-148). The threshold (5kb) were empirically determined (S2B/ C  Fig). -The authors say that the majority of peaks marked with H3K9me3 and H3K27me3 are not expressed (paragraph starting at line 216). It is not clear to me how they did this analysis.
(Line 165-170) We first identified the set of genes that overlapped with the repressive peak regions. For these genes, we examined their RNA expression levels by using our previously published RNA-seq data (Pu et al. 2018). We then calculated the fraction of genes that have zero RNA read counts.
-The authors claim that the observation that there is not much overlap in where age-associated changes in H3K27me3 and H3K9me3 occur suggests that "different mechanisms" contribute to the age-dependent changes. I'm not sure I follow this logic. I would expect that modifications at K27 and K9 would use distinct writers and erasers, regardless of effects of aging, so I'm not sure what "mechanisms" they are referring to here.
We appreciate the reviewer's comment and have revised the relevant sentences for clarity. In C. elegans, there is a strong association between H3K9me3 and H3K27me3 in repressive heterochromatin (Ho et al. Nature 2014). Worm-specific co-incidence of H3K9me3 and H3K27me3 is observed in repressive peak regions established at young age but not in repressive peak regions newly formed at old age. There might be specific biological processes regulated the co-localization of repressive H3K9me3 and H3K27me3 which become dysregulated at old age.
-Line 248 there is a reference written as "(2)" in parenthesis instead of brackets; also I don't think the previous findings referred to here should reference to Sen et al.
We have corrected the wrong reference.
-Line 248 it seems there is an extra " Fig" in the parentheses.
We have revised the text.
-In Fig2A the label for "upregulated" is misspelled.
We have corrected the error.
-The authors use "upregulated" and "downregulated" to refer both to gene expression and accumulation or loss of repressive marks, and this gets confusing at points. The authors need to choose another way to refer to changes in ChIP-seq peaks to avoid ambiguity.

increased and decreased expression
We appreciate the reviewer's comment and have made revisions. We have used "upregulate and downregulate" to describe the changes in repressive marks and "increase/decrease" to describe the changes in RNA expression.
-Line 277: the authors refer to genes that are "silent". Do they mean that in RNAseq these transcripts were not detected?
Genes with no detectable RNA reads (zero read count) were referred to as "silent".
-it would be useful to know the experimental conditions for the publically available ChIP-seq and ATAC-seq data sets the authors used to compare to the differential repressive peaks (section starting at bottom of pg 12). A table showing life stage, genotype, reference, and other relevant info would be very helpful here.
We have added a new table (S6 Table).
We have corrected the mistake.

Reviewer #3
Major Comment: The only major comment I have is that, as the experimental setup utilizes day 2 and day12 glp-1 (e2141) animals grown at 25C and lacking a germline, it would be important to know if changes that occur during normal aging at 20C recapitulate what is observed in these germline-less mutants. This can be done by assaying by ChIP-PCR H3K9/K27 me3 occupancy at a few randomly selected loci (e.g., those in Fig S6G) which display significant changes in glp-1 animals. These data would be informative as they are more true to what happens during normal aging, even though they may be complicated by the presence of germline cells in these wildtype animals.
We appreciate the reviewer's comment. In N2 animals, approximately two-thirds of the cells are germline cells and we may not be able to detect the epigenetic changes in somatic cells. We agree that it would be informative to investigate the epigenetic changes during normal aging in wild-type animals. One feasible approach is to isolate somatic cells from wild-type animals by FACS followed by epigenetic profiling. These will be our long-term goals and need further technical development, which is beyond the scope of this study.
We have corrected the typo.

Summary
Here, we compared the H3K9me3 profile produced by CUT&RUN-seq (2 replicates) in the germlineless glp-4(bn2) mutant with the H3K9me3 profile produced by ChIP-seq in the germlineless glp-1(e2141) mutant (Fig 1A). At old age, the average H3K9me3 enrichment across all H3K9me3 peak regions were reduced in both strains (Fig 1B). The loss of H3K9me3 enrichment in peak regions was more prominent in glp-4 than in glp-1 mutants (Fig 1A and 1B).
H3K9me3 peak regions were divided into three groups based on the fold change (old/young) of H3K9me3 enrichment in the glp-1 mutants (>1.25 in group A, 0.75~1.25 in group B and <0.75 in group C). The data suggested an age-dependent increase in H3K9me3 enrichment in the Group A peaks in the glp-4 mutant, similar to that in the glp-1 mutant (Fig 1C). To more clearly demonstrate this, we examined the ASRRs, which are regions that showed significant increases in H3K9me3 at old age in the glp-1 mutants, the data suggested similar age-dependent increase in the H3K9me3 levels in the glp-4 mutants (Fig 1D).
Taken together, glp-4 mutants showed a global loss of H3K9me3 enrichment in peak regions, similar to that of glp-1 mutants. Peak regions (Group A in Fig 1C and ASRRs in Fig 1D) that gained H3K9me3 in the glp-1 mutants also showed increased H3K9me3 enrichment in the glp-4 mutants. We therefore concluded that our findings in the glp-1(e2141) mutant, which are extensively discussed in the manuscript, are likely representative of germlineless strains and not due to the specific mutation in the glp-1 gene. However, due to time constraints and that the primary author Cheng-Lin Li has already begun a new position as a Bioinformatician at SOPHiA Genetics since March, we are not able to conduct additional replicates and therefore do not plan to include the preliminary CUT&RUN data in the manuscript.