H3K9me selectively blocks transcription factor activity and ensures differentiated tissue integrity

The developmental role of histone H3K9 methylation (H3K9me), which typifies heterochromatin, remains unclear. In Caenorhabditis elegans, loss of H3K9me leads to a highly divergent upregulation of genes with tissue and developmental-stage specificity. During development H3K9me is lost from differentiated cell type-specific genes and gained at genes expressed in earlier developmental stages or other tissues. The continuous deposition of H3K9me2 by the SETDB1 homolog MET-2 after terminal differentiation is necessary to maintain repression. In differentiated tissues, H3K9me ensures silencing by restricting the activity of a defined set of transcription factors at promoters and enhancers. Increased chromatin accessibility following the loss of H3K9me is neither sufficient nor necessary to drive transcription. Increased ATAC-seq signal and gene expression correlate at a subset of loci positioned away from the nuclear envelope, while derepressed genes at the nuclear periphery remain poorly accessible despite being transcribed. In conclusion, H3K9me deposition can confer tissue-specific gene expression and maintain the integrity of terminally differentiated muscle by restricting transcription factor activity.

D evelopment in multicellular organisms is governed by a carefully orchestrated programme of gene expression that is controlled by both genetic and epigenetic factors 1,2 . Lysine 9 methylation on histone H3 (H3K9me) is a defining modification of heterochromatin. In multicellular eukaryotes, heterochromatin serves two main functions: it silences the transcription of satellite repeats and transposons 3 and it silences tissue-specific genes during development 4,5 . Consistent with this, the loss of appropriately targeted heterochromatin is associated with a loss of tissue integrity, ageing and cancer [4][5][6][7][8] .
Silencing of repetitive elements (REs) prevents the accumulation of repeat-RNA-driven RNA:DNA hybrids, which in turn leads to replication stress, genome instability and infertility 3,9 . In contrast, the importance of H3K9me-mediated repression at genes is less clear. In mice, the inhibition of H3K9-specific histone methyl transferases (HMTs) boosts the efficiency of pluripotent-stem-cell induction 10,11 , and the deletion of H3K9me3 HMT SUV39H1/H2, coupled with an endoderm-specific deletion of SETDB1 (ref. 4 ) or conditional deletion of G9a 12 , compromises tissue integrity 4 . These findings suggest developmental roles for H3K9me, yet interpretations are complicated by the partial redundancy among the many mammalian H3K9 HMTs and their roles in chromosome segregation. In contrast, Caenorhabditis elegans has only two somatic H3K9 HMTs, and mutant worms that lack all detectable H3K9me have been generated 13 . Surprisingly, these animals develop from embryos to adults 3,14,15 , allowing one to examine how H3K9me controls gene expression in differentiated tissues.
We examined the tissue-specific expression and accessibility of H3K9me-marked sequences following the loss of MET-2, a SETDB1-like HMT responsible for H3K9me1 and H3K9me2 (refs. 13,16 ), and SET-25, a G9a/SUV39H-like enzyme mediating H3K9me3 (refs. 13,17,18 ). Although H3K9me2 is indeed necessary for silencing, gene activation in its absence requires a defined set of transcription factors (TFs). Surprisingly, whereas H3K9me restricts TF binding, the loss of H3K9me does not necessarily result in chromatin decompaction and enhanced sensitivity in the Assay for Transposase-Accessible (Tn5) Chromatin by high-throughput sequencing (ATAC-seq). Subnuclear localization with respect to the nuclear envelope may influence whether or not transcription and accessibility correlate.

Gene derepression following H3K9me loss is tissue specific
We set out to discover whether H3K9me alone controls gene expression in C. elegans and whether this depends on the developmental stage or tissue type. Genome-wide mapping of H3K9me2 and H3K9me3 (H3K9me2/me3) in early C. elegans embryos (up to the 200-cell stage) showed that 16.5% of all genes were marked by H3K9me3 and 3.4% were marked by H3K9me2 (Fig. 1a,b) 3 . Nonetheless, the loss of MET-2 resulted in higher and more widespread gene derepression than the loss of SET- 25 (refs. 18,19 ), and in the set-25 mutant H3K9me2 was retained on the loci that carried H3K9me3 in wild-type worms (Extended Data Fig. 1a) 18 . This indicates that MET-2-mediated H3K9me2 precedes and largely compensates for H3K9me3 in embryos 18 . Nonetheless, even following the loss of all H3K9me2/me3 in the met-2 set-25 double mutant, only about 15% of the genes that carried H3K9me3 in wild-type (WT) embryos were derepressed (Fig. 1c).
Characterization of the genes derepressed following the loss of H3K9me2/me3 revealed that larval genes are expressed prematurely in met-2-deficient early embryos (Extended Data Fig. 1b). We plotted the derepressed genes according to their expression patterns in a WT tissue atlas generated from larval stage 2 (L2) single-cell transcriptomes 20 (Fig. 1d). The genes derepressed in embryos cluster into groups of cell type-specific genes from all somatic tissues except the germline. This provides evidence that H3K9me2/ me3 are needed in embryos to prevent the precocious activation of genes typically expressed at later developmental stages.
To explore H3K9me-mediated gene silencing in differentiated tissues, we optimized the fluorescence-activated cell sorting (FACS)-based isolation of differentiated cell types from a single-cell suspension derived from L3 larvae. By sorting red fluorescent protein (RFP) + cells from a strain with muscle-specific RFP expression (gwIs4; Fig. 1e) 21 , we were able to perform RNA-seq on a highly enriched population of differentiated muscle cells (Extended Data Fig. 1c,d). A comparison of the gene expression in the H3K9 HMT mutants and the WT showed that 310 and 336 genes (false-discovery rate (FDR) < 0.01; log 2 -transformed fold change (FC), log 2 FC > 2), respectively, were upregulated in the met-2-and met-2 set-25-mutant muscle cells (Fig. 1f). Only three genes were upregulated in the set-25 mutant and there was a strong correlation between the transcriptional changes in the met-2-and met-2 set-25-mutant cells (met-2/WT versus met-2 set-25/set-25 log 2 FC, Pearson's correlation coefficient (R) = 0.74; Fig. 1g and Extended Data Fig. 1e). This suggests that MET-2-dependent H3K9me2 precedes deposition of the H3K9me3 mark in WT muscle and that it is sufficient to repress these loci in muscle. In general, the genes derepressed in the met-2-and met-2 set-25-mutant muscle were heterochromatic non-muscle genes with little or no expression in WT muscle ( Fig. 1h and Extended Data Fig. 1e; 65% reads per kilobase of transcript per million mapped reads (RPKM) < 5). The pattern was distinct from that in mutant embryos, for in met-2 muscle many, but not all, derepressed genes were germline-specific (for example, the P-granule component pgl-1; Fig. 1h).
To determine whether the changes in gene expression in H3K9me-deficient animals had overt effects on tissue development, we investigated muscle integrity. In body-wall muscle, the functional equivalent of mammalian skeletal muscle 22 , we found defects in sarcomere morphology in both the met-2 and met-2 set-25 mutants but not in the animals with the set-25 mutation (Extended Data Fig. 1f). The tissue defects manifested as uneven and broken actin-myosin filament structures and defective muscle function (Extended Data Fig. 1f,g). Thus, although muscles form in the absence of H3K9me despite the misexpression of tissue-specific genes, the muscle integrity is compromised.

H3K9me undergoes dynamic changes during differentiation
To confirm a direct role for H3K9me in tissue-specific gene repression, we checked whether the genes derepressed in the muscle of HMT mutants carried H3K9me3 in WT muscle. To this end, we mapped H3K9me2/me3 in isolated muscle cells using a modified CUT&RUN protocol (chromatin immuno-directed cleavage with sequencing, ChIC-seq) [23][24][25] that was adapted for low cell numbers (Fig. 2a,b and Extended Data Fig. 2a,b). As expected, genes derepressed in met-2-or met-2 set-25-mutant muscle cells were enriched for H3K9me3 in WT muscle (Fig. 2b,c; for pgl-1, see Extended Data Fig. 2c). The genes derepressed in H3K9me-deficient embryos were less enriched for H3K9me3 in WT muscle, suggesting that H3K9me3 might not only be gained, but also be lost during development (Fig. 2c). The H3K9me3 ChIC-seq pattern in muscle was indeed substantially different from that of embryos, and only 29% of genes were found methylated at both stages (see results for chromatin immunoprecipitation with sequencing, ChIP-seq; Fig. 2d). The differences corresponded nicely to the distinct transcriptional programmes: genes expressed in WT embryos tended to gain H3K9me2/me3 in muscle (Fig. 2e), whereas genes that gained expression in muscle lost H3K9me2/me3. Genes expressed in other tissues, including germline cells, were tri-methylated at both stages of WT animals (Fig. 2e). Despite specific losses and gains of H3K9me during differentiation, only 12% of all H3K9me3-marked loci were derepressed following the loss of H3K9me in muscle

H3K9me2 requires active maintenance in post-mitotic cells
To determine whether the active deposition of H3K9me continues in non-dividing differentiated cells, we investigated whether there is a requirement for MET-2 protein in post-embryonic tissues. The fusion of a degron tag to the endogenous MET-2 resulted in its rapid degradation in all larval tissues following the addition of auxin (<1 h; Fig. 3a-c). Degradation of MET-2 was accompanied by a progressive loss of H3K9me2, reaching the level of the constitutive met-2 null allele by 24 h post auxin addition (Fig. 3b,c). As the majority of larval cells are post-mitotic 26 , H3K9me2 loss cannot be due to histone replacement during replication. As proof that MET-2 degradation impacts the organismal phenotype, we found that auxin-induced degradation of MET-2 in larvae yields a multi-vulva phenotype in combination with a lin-15A mutant, phenocopying the met-2 deletion mutant (SynMuv assay; Extended Data Fig. 3a) 27 . We conclude that H3K9me requires constant maintenance by MET-2 HMT activity to maintain differentiated tissue identity.
To determine whether the post-embryonic loss of H3K9me2 affected the same genes as the constitutive met-2 null, we treated synchronized early L1 larvae with auxin and isolated messenger RNA from L3 muscle cells. In L1-stage larvae, 81 of the 95 body-wall muscle cells are post-mitotic and seven undergo one final cell division 26 . Of the 134 genes derepressed with auxin treatment, 76% overlapped with the genes derepressed in the met-2-mutant muscle Fig. 1 | limited derepression of H3K9me3-marked tissue-specific genes following the loss of H3K9me2/me3. a, Isolation of early embryos for RNA-seq and ChIP-seq. b, H3K9me2 and H3K9me3 on genes mapped by ChIP-seq in early embryos at 20 °C. Correlation between the mean H3K9me3 and mean H3K9me2 at genes (log 2 FC values normalized to the input). Individual genes are coloured based on enrichment for H3K9me2 (3.4%), or H3K9me3 and Common (16.5%) or neither (black); FC > 2; n = 2 independent biological replicates. c, Correlation between the mean H3K9me3 (as in a) and the mean log 2 -transformed FC in gene expression in the met-2 set-25 double mutant normalized to the WT (n = 3 independent biological replicates). Individual genes (red) are both enriched for H3K9me3 in WT (FC > 2) and derepressed in the met-2 set-25 double-mutant embryos (FDR < 0.01). b,c, IP, immunoprecipitation. d, The expression of genes that are derepressed in met-2-mutant embryos is restricted to specific tissues in WT L2 larvae 20 . Percentage of genes that are derepressed in met-2-mutant embryos (n = 3 independent biological replicates) over the total number of tissue-specific genes for each tissue (left; number of genes: gonad, 0; neurons, 313; hypoderm, 302; pharynx, 247; muscle, 180; glia, 204; and intestine, 313). Heatmap of the expression levels of each derepressed gene (columns) in individual cell types (rows; the coloured bar on the left indicates their parent tissue) in WT L2 larvae. The tissue and cell-type expression data are from 20 . e, Isolation of muscle cells for RNA-seq and ChIC-seq from L3 larvae by FACS of cells expressing myo-3p::RFP (left). Image of L3 larva with muscle-specific RFP expression (right). f, Gene expression (normalized to the WT) in sorted muscle cells of the indicated mutants (two biologically independent replicates are shown). Genes identified as significantly changed are shown in colour. g, Correlation between the gene expression (log 2 FC) between the met-2 and met-2 set-25 double mutants. Significantly changed loci (FDR < 0.01 and FC > 2) are coloured according to the genotype and those that changed in both genotypes are shown in dark blue. h, Genes that are derepressed in met-2 muscle (n = 2 independent biological replicates) as a percentage of all tissue-specific genes in WT L2 larvae (left; gene number: gonad, 172; neurons, 30; hypoderm, 25; pharynx, 13; muscle, 7; glia, 33; and intestine, 28) and as a heatmap of derepressed genes (columns) according to the cell type (right).
( Fig. 3d-f and Extended Data Fig. 3b-d). The auxin-sensitive genes also contain germline-expressed genes (Extended Data Fig. 3c,d); thus, their derepression cannot reflect a failure in germline-to-soma transition as previously proposed 28,29 . We conclude that terminally differentiated post-mitotic cells actively reinforce H3K9me2 to repress genes of other cell lineages, including the germline.

Gene derepression correlates with specific tF motifs
To determine whether the set of genes that becomes derepressed by loss of MET-2 in differentiated larvae is cell-type specific, we examined the transcriptome of hypodermis-derived seam cells from L3 larvae of WT and H3K9 HMT mutants (Extended Data Fig. 4a-c). In FACS-sorted seam cells 30 , as in muscle cells, gene silencing was almost entirely dependent on MET-2, and the genes that were derepressed were characteristic of germline or other tissues (Extended   Data Fig. 4d-f). Interestingly, only 39% of the genes derepressed in H3K9me-deficient muscle were affected in seam cells (35% vice versa; Fig. 4a,b) and although germline genes were affected in both tissues, these were partially distinct sets of germline-specific genes (Extended Data Fig. 4g). Furthermore, there was little overlap between the genes derepressed in embryos and those derepressed in either tissue (Fig. 4a,b). We conclude that the function of H3K9me-marked heterochromatin in gene silencing is both cell-type and developmental-stage specific.
To gain mechanistic insight into what drives gene derepression following the loss of H3K9me, we asked what characterizes the derepressed sets of genes. Although aberrantly expressed REs can affect the expression of proximal genes 31 , we found no preferential proximity of the genes derepressed in met-2 set-25-mutant muscle cells to REs (Extended Data Fig. 5a,b). We examined the enrichment of TF binding motifs 32    midsection of an L1 larva expressing MET-2::degron::3×FLAG::NeonGreen as well as a ubiquitously expressed TIR1 auxin receptor, after being seeded for 1.5 h on plates with (+) or without (−) auxin. Scale bar, 10 µm. Images represent two biologically independent experiments. b, Schematic of auxin addition for the MET-2::degron degradation experiments. For western blots, a synchronized population of worms was split onto different plates, with auxin added at different times during development. All samples were collected and processed simultaneously. For RNA-seq, a synchronized population was grown for 36 h in the presence or absence of auxin to obtain L3 larvae for muscle isolation. c, Example western blot of whole-cell lysates from WT as well as met-2-mutant and met-2::degron L2 larvae after growth at 20 °C with or without auxin for the indicated times (top). The membranes were blotted for FLAG, H3K9me2 and MRG-1 (as a control). The protein levels were normalized to endogenous MRG-1 and the control (0 h auxin) for n = 3 biologically independent experiments. d, Gene expression was determined for auxin treated over untreated sorted muscle cells isolated from L3 larvae expressing met-2::degron, after growth for 36 h at 20 °C with or without auxin. Loci identified as significantly changed are highlighted in colour (FDR < 0.01 and FC > 2, or < −2); n = 2 independent biological replicates. e, Correlation between gene expression (log 2 FC) in sorted met-2::degron muscle cells with auxin over no auxin and met-2 (met-2 null allele) over WT. Loci that were significantly changed (FDR < 0.01 and FC > 2) are coloured according to the genotype and significantly changed loci in both are coloured in dark blue. f, Venn diagram showing numbers of significantly derepressed genes from sorted muscle cells of met-2::degron animals (auxin treatment versus no auxin) and met-2 (met-2 null versus WT) and their overlap.
revealed a unique set of TF motifs enriched in the promoters derepressed in embryos as compared to muscle or seam cells, whereas the enriched TF motifs in muscle and seam cells showed partial overlap (Fig. 4c). Importantly, the expression of the relevant TFs themselves was not altered by HMT mutation (except for atf-8 in embryos; Extended Data Fig. 5c). Interestingly, most of the TFs that bind motifs upstream of derepressed genes in muscle or seam cells are not expressed in embryos, with the exception of EGL-38, ZIP-11 and EFL-1. We found four other TFs to be expressed in both muscle and seam cells, four to be predominantly muscle specific and two that were expressed most strongly in seam cells (Fig. 4d).
The embryo-specific TF motifs enriched on met-2-sensitive genes were recognized by factors that have homologues active in early mammalian development, namely ATF-8/CES-2 (TEF) 33 and KLF-1 (KLF6) 34 , whereas the muscle-and seam cell-specific TFs have homologues involved in tissue determination, for example, CEH-48 (Onecut1/2), which in mammals is a central regulator of both differentiation and the maintenance of the differentiated state of liver and pancreas cells as well as retinal neurons 35,36 . Others involved in tissue determination are HLH-25, -27, -28 and -29 (all homologous to HES1) 37,38 and EGL-38 and PAX-2 (PAX-2/PAX-5) 39 . When we restricted our analysis to the derepressed genes in mutant muscle that were H3K9me3-marked in the WT, we found that the same TF motifs were enriched. In contrast, no TF motifs were enriched for H3K9me3-marked non-derepressed genes (Extended Data Fig. 5d). The fact that the enriched TF motifs are largely shared between muscle and seam cells, even though two-thirds of the expressed genes are different, suggests that these TFs may act together with co-factors that determine tissue-specific gene expression following the loss of H3K9me.

H3K9me restricts specific tF binding and activity
It remained to be shown that the TFs enriched at derepressed genes were responsible for the observed gene derepression following the loss of H3K9me. To achieve this, we knocked down muscle-enriched TFs one by one and monitored the impact of their loss on met-2 targets. One locus we monitored was pgl-1, chosen because it is expressed in mutant muscle and seam cells, carries H3K9me3 in WT larvae (Extended Data Fig. 2c) and has motifs for multiple relevant TFs in its promoter, including HLH-25/-27, HLH-28/-29, CEH-48 and MADF-9 (Supplementary Table 3), like many derepressed germline genes (Extended Data Fig. 5e). of met-2 over WT from sorted muscle cells, sorted seam cells, or early embryos, as marked. Loci that were significantly changed compared with the WT (FDR < 0.01 and FC > 2) are coloured-coded by tissue, and loci that were significantly changed in both tissues in the comparison are coloured in dark blue. b, Overlap of significantly derepressed genes from sorted muscle cells, sorted seam cells or early embryos of met-2 mutants versus WT (as in a). c, Enrichment (log 2 -transformed) of TF motifs in the promoters of the indicated derepressed gene sets (a) versus log 2 -transformed genome-wide occurrence. Significantly enriched motifs (log 2 enrichment > 0.5 and log 2 abundance > 8) are colour-coded with their corresponding TF(s) indicated. Motifs that are enriched in both tissues are shown in bold font. d, Expression levels of TFs that have enriched motifs in the promoters of genes derepressed in muscle (red), seam cells (green) or both (gold). TF expression from RNA-seq, with two biologically independent replicas per tissue shown.
Using quantitative microscopy, we measured expression from the endogenous pgl-1 locus by means of a PGL-1::green fluorescent protein (GFP) fusion 40 . In WT animals, PGL-1::GFP was expressed only in the germline, as expected 41,42 , whereas in animals with the met-2 set-25 double mutation, PGL-1::GFP was expressed in multiple tissues, including muscle and pharynx (Fig. 5a).
In these H3K9me-deficient worms, RNA interference (RNAi) against ceh-48 strongly reduced PGL-1::GFP derepression in both muscle and pharynx cells, while RNAi against hlh-25/-27, hlh-28/-29, zip-11 and madf-9 each had a partial effect (Fig. 5a,b and Extended Data Fig. 6a). Reflecting the absence of their motifs and/or TF expression, pax-2 and egl-38 RNAi had almost no effect  on pgl-1 expression in muscle or pharynx cells. We also detected no reduction in GFP signal in the germline, where pgl-1 is not repressed by H3K9me ( Fig. 5a and Extended Data Fig. 6b).
To document the effects of TF ablation on a genome-wide scale, we performed RNA-seq on whole L1 WT and met-2-mutant larvae treated with RNAi against individual TFs of relevance ( Fig. 5d). We chose L1-stage animals so that we could see the effect of RNAi without potentially confounding developmental defects. In met-2 mutants grown on bacterial carrying an empty RNAi vector we observed derepression of 171 genes representing a broad range of tissues, with enrichment for many TF motifs found in muscle, including HLH-25/-27, HLH-28/-29 and CEH-48 more weakly (Extended Data Fig. 6e-g). Confirming a role for HLH-25/-27 and CEH-48 in H3K9me-target derepression, RNAi against these factors broadly suppressed the upregulation triggered by the loss of H3K9me, whereas HLH-28/-29 had a more selective effect (Fig. 5e). Once again, the levels of pgl-1 mRNA were elevated in the met-2 mutant, and this was sensitive to TF knockdown (Extended Data Fig. 6h). This confirms that specific TFs are required to induce the aberrant gene expression that occurred following the loss of H3K9me in differentiated tissues.
The final step was to examine whether H3K9me blocks the binding of such TFs to their cognate promoters. We performed ChIP in WT and met-2 set-25-mutant L3 larvae expressing GFP-tagged CEH-48 and used similar GFP fusions to the embryo-specific TFs KLF-3 and END-3 as negative controls. Promoter binding was measured by ChIP coupled to detection by quantitative PCR (qPCR) at a subset of target genes that were derepressed in met-2 set-25-mutant muscle cells and are expressed in WT embryos 20 but not in WT larval or adult tissues (Extended Data Fig. 6i, see legend). We have indicated above the graph in Fig. 5f whether binding motifs for the relevant TFs are present in the promoter assayed. We found that CEH-48 gained binding to the promoters of the three derepressed genes bearing CEH-48 motifs in the met-2 set-25 mutants but not to The number of peaks that gain accessibility are indicated. The mean log 2 FC was calculated from n = 3 independent biological replicates. b, Correlation between the log 2 FC for accessibility between met-2/WT and met-2 set-25/set-25. Significantly changed loci (FDR < 0.01) are coloured based on genotype and common loci in maroon. c, Percentage (relative to genome-wide occurrence) of promoters, exons, introns, repeats and intergenic regions with significantly increased accessibility in met-2-(left) and met-2 set-25-mutant (right) muscle cells. d, Correlation between the mean gene expression (log 2 FC) and promoter accessibility (log 2 FC) in muscle cells from met-2 set-25 mutants normalized to the WT. Individual genes are highlighted and divided into groups based on increased expression (Group A), increased accessibility (Group C) or both (Group B). e, H3K9me3 (left) and H3K27ac (right) ChIC-seq enrichment (over the MNAse control) at Group A, B and C genes in WT and met-2 set-25-mutant muscle cells. f, GFP-LacI cassette from the gwIs4 heterochromatic reporter (top). The plots show the normalized average counts of ATAC-seq (middle) and RNA-seq (bottom) reads across the GFP-LacI cassette in WT and met-2 set-25-mutant muscle cells. g, Schematic of muscle and intestine nuclei with GFP-LacI foci from the gwIs4 reporter and emr-1::mCherry marking the nuclear periphery (top left). Images of muscle and intestine cells from WT and met-2 set-25-mutant L1 animals (bottom left). Due to the low level of GFP-LacI expression in the intestine, insets with increased brightness are included. Scale bars, 5 µm. GFP-LacI nuclear intensity, focus size and focal distance from the nuclear periphery (right). Muscle, n = 332 (WT) and 651 (met-2 set-25); and intestine, n = 346 (WT) and 454 (met-2 set-25) pooled foci from two biological independent experiments. NS, not significant (P > 0.05); two-sided Wilcoxon signed-rank test. a.u., arbitrary units. h, Schematic of Emerin DamID, wherein EMR-1 is fused to DAM methylase 49 (top). EMR-1::Dam methylation enriched over GFP::Dam (log 2 -transformed) control in WT and cec-4-mutant muscle cells at Group A, B and C genes (bottom). Data points represent 10 kb bins; n = 3 biological independent replicas. Note that CEC-4 protein tethers H3K9me-modified heterochromatin at the nuclear periphery 49,59 . e,g,h, Boxplots show the median (horizontal line), 25th to 75th percentiles (boxes), and 90% (whiskers) of the group.  Fig. 6i and ref. 21 ). The presence of TF motifs in target genes is indicated. The pals-30 promoter has two 8/9 nucleotide matches to the CEH-48 motif and seven 6/9 nucleotide matches, indicated by half shading; n = 3 biological independent experiments. P values derived from a two-sided t-test are indicated. a promoter lacking its motif or to control genes that were not upregulated (Fig. 5f). No binding was detected in the WT animals, presumably because H3K9me blocked the promoter. As expected, the negative controls KLF-3 and END-3 did not bind in either the WT or met-2 set-25 mutants (Fig. 5f). These data argue that H3K9me prevents the unscheduled binding of at least one of the TFs found to drive gene derepression following the loss of H3K9me in differentiated tissues.

Gene derepression does not require decompaction
H3K9me is associated with condensed chromatin; thus, the reduced accessibility of heterochromatic promoters has been proposed to restrict TF binding [43][44][45][46] . To investigate whether we could detect increased accessibility following the loss of H3K9me, we performed a tissue-specific ATAC-seq assay. Using WT and HMT-deficient muscle cells, we confirmed that H3K9me compacts chromatin: we monitored increased Tn5 accessibility at 3,986 peaks in the met-2 set-25 double mutant ( Fig. 6a; FDR < 0.01). The effects were far less pronounced in the single mutants (met-2, 87 peaks; and set-25, 9 peaks; Fig. 6a). Although the accessibility changes in the met-2-and met-2 set-25-mutant strains showed the expected positive correlation ( Fig. 6b; R = 0.55), there was a striking discrepancy between the changes in gene expression and ATAC-seq. Genes derepressed in muscle were almost identical between the met-2 and met-2 set-25 Intestine Intestine Intestine NS mutants (Fig. 1g), whereas the ATAC-seq signal increase was far more pronounced in the double mutant. This suggested that SET-25 alone (that is, in the met-2 single mutants) is sufficient to restrict ATAC-seq sensitivity, albeit not to repress transcription. The regions that did gain accessibility in the met-2 mutant were equally enriched for REs and promoters, whereas changes in the met-2 set-25 double mutant mapped overwhelmingly to promoters and intergenic regions (Fig. 6c). However, even in the met-2 set-25 mutants, the changes in accessibility did not correlate strictly with changes in gene expression (Extended Data Fig. 7a). This lack of correlation was specific to genes derepressed in the met-2 set-25-mutant muscle, for overall we scored the expected correlation between increased ATAC-seq signals and highly transcribed genes across the entire transcriptome (Extended Data Fig. 7b).
We next plotted the changes in RNA-seq (increased expression levels) versus changes in ATAC-seq (accessibility) triggered by the loss of H3K9me. The genes that were affected in muscle fell into three classes: the majority of derepressed genes (265/336 genes) did not gain ATAC-seq sensitivity at their promoters (Group A), whereas 22% (71/336) gained accessibility coincident with an increase in transcription (Group B; Fig. 6d). Surprisingly, 1,210 loci showed increased accessibility without increased transcription (Group C; Fig. 6d).
Surprisingly, the genes of both Group A and B had on average similar and low absolute levels of accessibility and expression in WT muscle (Extended Data Fig. 7c), whereas the Group C genes started with higher absolute levels of expression and accessibility in the WT muscle, albeit not as high as that scored for the muscle-expressed genes (Extended Data Fig. 7c). In addition, all three groups had comparable levels of H3K9me3 in the WT muscle, which was sensitive in all cases to the loss of met-2 and set-25 (Fig. 6e). However, the Group A genes did not acquire H3K27ac, a mark associated with active loci and enhancers 48 , in the met-2 set-25 mutant, although it increased in Group C. The Group B genes had a weaker or more variable increase in H3K27ac (Fig. 6e).
In addition to endogenous loci, we also tracked the expression and accessibility of the gwIs4 heterochromatic array in muscle (both increased in the met-2 set-25 mutant; Fig. 6f). We could further quantify gwIs4 compaction and subnuclear position by microscopy given that the integrated array contains lacO sites and can bind GFP::LacI (Figs. 5c and 6g) 13 . A comparison of WT and met-2 set-25-mutant muscle cells confirmed that gwIs4 physical chromatin compaction and expression increase coordinately following the loss of H3K9me (Fig. 6g). In contrast, intestinal cells showed derepression without decompaction, resembling the Group A genes (Fig. 6g). This difference seemed to correlate with subnuclear position: measurements of the distances of the GFP::LacI spots to the nuclear periphery confirmed an internal position 21,49,50 for the gwIs4 array in muscle that shifted even more internally in the met-2 set-25 mutant, whereas in intestine the reporter was strongly peripheral and stayed largely peripheral despite derepression (Fig. 6g). Although the repetitive nature of the reporter may be a confounding factor, the behaviour of gwIs4 suggested that genes located in the nuclear interior might be more susceptible to chromatin decompaction (and Tn5 attack) following the loss of H3K9me, whereas genes at the nuclear periphery appear to be derepressed without measureable decompaction.
To determine whether this correlation applies to endogenous single-copy genes, we compared the association of Group A, B and C genes with the nuclear periphery using muscle-specific Emerin DNA adenine methyltransferase identification (DamID; Fig. 6h) 49 . The loci that gained accessibility (Groups B and C) indeed showed a more internal localization overall in WT muscle cells than genes that remained compacted during derepression (Group A). All positioning was sensitive to loss of the H3K9me anchor CEC-4. Together, we conclude that whereas ATAC sensitivity increases dramatically following the loss of H3K9me, the increase does not always correlate with expression. When it does, the genes are generally located away from the nuclear envelope.

H3K9me restricts enhancer activation
Given the distinct behaviours of the loci in Groups A, B and C, we queried whether the promoters were enriched for distinct TF motifs. As expected, the Group A and B gene promoters were enriched for the TF motifs identified among the derepressed genes in muscle ( Fig. 4c and Extended Data Fig. 7d), and the Group B genes also bound basic leucine zipper (bZIP) and zinc-finger TFs (Extended Data Fig. 7d). In the non-derepressed Group C, which nonetheless gained ATAC-seq sensitivity, the only motifs significantly enriched were for EOR-1, DAF-12/NHR-48 and ZTF-3. Importantly, all three of these factors have been shown to bind enhancers during C. elegans development 47,51 .
The presence of enhancer-linked TF motifs, coupled with the fact that Group C sites gain both H3K27ac and ATAC sensitivity following the loss of H3K9me (Fig. 6e), suggested that H3K9me might be limiting enhancer activity in WT tissues. To separate enhancer elements from overlapping promoters, we analysed the intergenic ATAC-seq peaks that gained accessibility in the met-2 set-25-mutant muscle (2,090; Fig. 6c). By comparing ATAC-seq accessibility at these intergenic peaks, we observed that met-2 and set-25 ablations were synergistic and these peaks were absent in WT muscle, where they showed baseline accessibility ( Fig. 7a and Extended Data Fig. 8a). We examined other genomic features in  47 . c, Enrichment of TF motifs (log 2 -transformed) in the intergenic peaks from a versus genome-wide occurrence (log 2 -transformed). Enriched motifs (log 2 enrichment > 0.5 and log 2 abundance > 8) are colour-coded according to TF(s). *Enrichment of a scrambled motif. These motifs characterize enhancers in worms 47,48,52 . d, H3K9me3 (left) and H3K27ac (right) ChIC-seq enrichment (normalized to the MNAse control) at newly accessible intergenic peaks (a) in WT and met-2-and met-2 set-25-mutant muscle. Data points represent 10 kb bins over intergenic peaks. Boxplots show the median (horizontal line), 25th to 75th percentiles (boxes), and 90% (whiskers) of the group. ****P < 2.2 × 10 −12 ; two-way analysis of variance, followed by a post-hoc test. e, Reporter constructs containing a minimal promoter driving GFP (nucleolar localization tag) with either no upstream sequence (control) or the nhr-25-associated enhancer (Enhancer reporter) as previously described 47 (top). Representative images of GFP in the body region of WT and met-2 set-25-mutant animals (middle). Cells with GFP signal are highlighted with yellow arrows, all other signal is intestinal autofluorescence. Scale bar, 5 µm. Level of GFP expression as the proportion of worms with ≥5 GFP + cells, excluding the head and tail (right). Control, n = 30 (WT) and 19 (met-2 set-25) worms; enhancer, n = 47 (WT) and 43 (met-2 set-25) worms from three independent biological replicates. Statistical significance was determined using a two-tailed t-test. f, Continuous deposition of H3K9me2 by MET-2 represses tissue-specific genes and enhancers (left). MET-2 activity is usually followed by SET-25, such that genes accumulate H3K9me3. In the absence of MET-2, genes that are derepressed require specific TFs (right). Perinuclear H3K9me-marked genes seem to be refractory to chromatin decompaction despite activation in a met-2 set-25 mutant, whereas genes that gain both chromatin accessibility and transcriptional activity tend to localize internally. Unlike genes, enhancers can be repressed by either MET-2 or SET-25. the modENCODE datasets from whole-worm lysates 47 for correlation with these regions. The strongest correlation for these intergenic peaks was with larval-specific enhancers (Fig. 7b). Moreover, these peaks were enriched for motifs for EOR-1, DAF-12/NHR-48 and ZTF-3 TF, which typically bind enhancer elements (compare Fig. 7c with Extended Data Fig. 7d). As expected, they were highly enriched for H3K9me3 in WT muscle, and this mark was partially maintained in the met-2 mutant but not in the met-2 set-25 double mutant (Fig. 7d). Moreover, the level of H3K27ac, which characterizes active enhancers 52 , showed a partial increase at these sites in met-2-mutant and was strongly elevated in met-2 set-25 muscle cells (Fig. 7d). This suggests that enhancers, like promoters, are masked by H3K9me3 and can be targeted by both MET-2 and SET-25 redundantly.
Once again, to prove that H3K9me represses enhancer activity as it does gene expression, we made use of a reporter strain containing a transgenic array bearing a minimal promoter driving gfp with or without an upstream enhancer sequence that had been previously identified in proximity to the gene nhr-25 (ref. 47 ). In the absence of the upstream enhancer, the loss of H3K9me had no effect on GFP expression within the body region of L3 larvae. However, when the enhancer element was included, cells in the met-2 set-25  Fig. 8b). Together, we conclude that H3K9me2/me3, deposited by either MET-2 or SET-25, or both, can limit not only promoter but also enhancer accessibility and activity.

Discussion
We demonstrate an important function of H3K9me2/me3 in the silencing of cell type-specific genes in differentiated tissues in C eleganst. We found that following the loss of H3K9me3 (that is, in the set-25 deletion), H3K9me2 serves to maintain gene repression in differentiated cells, highlighting a redundancy that serves to safeguard chromatin-mediated gene silencing. Even in the met-2 set-25 double mutant, which loses both H3K9me2 and H3K9me3, misexpression of genes in differentiated tissues was dependent on the availability of appropriate TFs (Fig. 5). Given the developmental changes in the H3K9me pattern (Fig. 2), we propose that H3K9me reinforces differentiation-linked transcriptional programmes. In other words, tissue-specific target restriction by H3K9me, on both enhancers and promoters, allows TFs to promote distinct expression programmes at different stages of development and/or in different tissues. We confirm that potential binding sites of the conserved TF CEH-48 are occluded in at least three derepressed gene promoters in WT but not met-2 set-25 larvae; importantly, CEH-48 is necessary for the misexpression of genes that lose H3K9me in muscle. Finally, loss of the TFs HLH-25 and HLH-28/-29, which also drive gene derepression in the met-2 set-25-mutant muscle, have pleiotropic defects in multiple tissues, including germline, in otherwise WT worms 38,54 . Our data thus argue that tissue-specific H3K9 methylation restricts the range of targets a factor can bind.
The role of H3K9me in gene silencing seems to not be linked to its function at REs, as we found no preferential proximity between the derepressed genes and specific repeat families. Only 26% of the 331 derepressed genes had REs within 1 kb of the corresponding transcriptional start site compared with 34% for all genes. Nonetheless, we do not exclude that individual REs could influence transcription at some loci. Even if REs were to help target HMTs to genes, TF binding is still likely to be necessary for derepression in met-2 set-25-mutant worms 53 .
Although the loss of H3K9me led to the misexpression of tissue-specific and germline genes, it is not essential for organismal development per se, as 88% of the met-2 set-25 embryos form adult worms 3 . Nonetheless, we see that muscle integrity and function are impaired in the mutant. Whether such defects were caused by a specific misexpressed protein(s) or dysfunction due to imbalanced transcriptional homeostasis is not yet clear.
Surprisingly, although H3K9me seemed to prevent TF binding, a gain in accessibility as measured by ATAC-seq was neither required nor sufficient for gene derepression. One source of this discrepancy arises from the fact that enhancers can lose H3K9me and gain H3K27ac without activating the immediately adjacent genes (for example, Group C). However, approximately 80% of the genes derepressed in the met-2 set-25 mutant (Group A) showed little change in promoter accessibility. At the genes that gain both accessibility and expression (Group B), an analysis of TF motifs indicated that chromatin decompaction may well be a secondary effect of TF binding, as the TFs recruit additional complexes that in turn regulate chromatin opening. Notably, the bZIP TFs that co-map with Group B loci as well as Onecut1/2, the CEH-48 homologue, recruit the CBP/p300 histone acetyltransferase, and EOR-1 cooperates with the SWI/SNF chromatin remodeler 51,54-57 . Thus, depending on the chromatin factors that are recruited, different TFs may trigger different degrees of chromatin opening (Fig. 7f).
Group B genes are further distinct in that they are not enriched at the nuclear periphery, which is in contrast to the majority of H3K9me3-marked loci 13,58 . This would argue that although subnuclear positioning is not the crucial determinant of gene expression 59 , it may determine whether or not a promoter is susceptible to decondensation following the loss of H3K9me. Alternatively, it may be that different combinations of TFs are responsible both for the gene position and its response to a loss of H3K9me. The transient opening of a promoter that allows transcription near the nuclear envelope may allow a derepressed gene to remain ATAC-resistant. Consistent with the idea that subnuclear positioning differentiates levels of accessibility, aberrant bZIP TF activity can induce the relocation of a heterochromatic array towards the nuclear interior, in a manner dependent on the HAT p300/CBP 50 . In summary, our study shows that H3K9me, like DNA methylation 60 , can restrict TF activity at promoters and enhancers and that it does this primarily at tissue-specific genes whose repression is required for tissue robustness.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41556-021-00776-w. Supplementary Table 1.
Except for RNAi experiments, worms were grown at 20 °C on NGM plates and fed OP50 bacteria. For the auxin treatments, the worms were placed on NGM plates containing 1 mM auxin for the indicated times. The RNAi experiments were performed as described previously 63,64 using NGM plates (1 mM IPTG and 100 µg ml −1 carbenicillin), which were prepared with the appropriate RNAi-expressing bacteria and allowed to dry. Synchronized L1s were seeded on the RNAi and either L1 or L3 larvae (as indicated) of the following generation were processed for RNA-seq or microscopy ( Fig. 5 and Extended Data Fig. 6).
Phalloidin staining. Phalloidin staining of muscle fibres and scoring of muscle defects were performed as described previously 65 . Briefly, synchronized L1 larvae were grown to the young adult stage (before the appearance of eggs), and collected and washed with M9 buffer. The worms were then transferred to 1.5 ml low-bind tubes (Eppendorf) and fixed with 4% formaldehyde (Thermo Fischer) in 0.1 M Na 2 HPO4 for 15 min. The worms were pelleted, permeabilized in 500 μl of pre-chilled acetone (5 min at −20 °C) and then washed three times with PBS-TG (PBS, 0.5% Triton X-100 and 30 mM glycine). Rhodamine phalloidin (500 μl; R415, Life Technologies; 1:250 in PBS-TG) was added to the worms for 30 min at room temperature with rotation. The worms were washed three times, settled on poly-l-lysine-coated slides and mounted with ProLong Gold (P36930, Thermo Fischer).
Microscopy and image analysis. Animal morphology images were taken at 20 °C using a Leica M205 FA microscope. All fluorescent microscopy images and movies for the swimming assay were taken at 20 °C on spinning-disk multipoint confocal microscopes using the VisiView software and VisiView v.4.2.0.6. The GFP and NeonGreen, and RFP, mCherry and Rhodamine fluorophores were excited using Toptica iBeam Smart 488 nm and 561 nm lasers, respectively. For live microscopy, worms were collected off plates in M9 buffer and placed on 2% agarose pads. To enrich for L1 larvae, plates containing gravid adults were washed to remove adults and larvae, and newly hatched L1 worms were collected after 3-5 h by washing with M9. To enrich for L3 larvae, gravid adults were seeded for about 1 h on plates and then removed; semi-synchronous L3 larvae were collected after approximately 38-40 h by washing with M9.
Quantification of the GFP-LacI and PGL-1::GFP intensity in larval tissues of RNAi-treated worms was performed using the Fiji/ImageJ software. For PGL-1::GFP, the mean fluorescence intensity was measured in an area with representative signal and in the focal plane of each cell. For GFP-LacI, the GFP intensity was blindly measured as specific tissues and nuclei were identified and the nuclear area selected based on the CEC-4::WmCherry signal.
To perform quantification of the GFP-LacI intensity, spot volume and spot distance from nuclear periphery in muscle and intestine nuclei, individual nuclei were cropped in Fiji and analysed separately using the KNIME Analytics Platform 66 . In summary, nuclei were detected using a seeded watershed segmentation on the EMR-1 (mCherry) channel. For foci detection, we used a Laplacian-of-Gaussian detector from TrackMate 67 (fmiij2-plugins-0.2.5, https:// doi.org/10.5281/zenodo.1173536) on the GFP-LacI (GFP) channel. The distance between the foci and the nuclear periphery was measured by computing a Euclidean distance map on the three-dimensional nucleus mask and measuring its intensity for the coordinates of each spot. Foci outside of a nucleus were ignored in the analysis. The 'distance from periphery' was plotted per condition using the R package ggplot2 (https://ggplot2.tidyverse.org/). Swimming assay. Worms were placed in flat-bottomed 96-well plates in M9 buffer and movies with a duration of 1 min were acquired at 20 °C. For analysis, the movies were slowed fivefold to enable counting of the number of head turns over the 1 min recording time.
Western blotting. Pellets of thoroughly washed L2 worms were flash-frozen in liquid nitrogen and kept at −80 °C until processing. Thawed pellets were lysed using a Fast Prep 24-5 G benchtop homogenizer (MP Biomedicals) and 0.5 mm zirconia/silicon beads (BioSpec) in RIPA buffer in the presence of cOmplete EDTA-free protease inhibitors (Roche) and 1 mM dithiothreitol (DTT) at 4 °C. The lysates were treated with 5 μl benzonase (Sigma) for 1 h at 4 °C with end-over-end rotation. Total protein (10 µg) was separated on Bolt 4-12% bis-Tris plus gels (Thermo Fisher) and transferred to a PVDF membrane (Bio-Rad). The blots were blocked in PBS plus 0.5% Tween-20 with 5% powdered milk. The primary antibodies used were: 1:1,000 mouse anti-FLAG M2-horseradish peroxidase (HRP) conjugated (A8592, Sigma), 1:500 mouse anti-H3K9me2 MABI0317 (MBL) 68 and 1:5,000 rabbit anti-MRG-1 (49130002, Novus Biologicals). The blots were treated with antibodies diluted in blocking buffer overnight at 4 °C. Except anti-FLAG, after three washes, blots were re-blocked and exposed to secondary antibodies coupled to HRP (goat anti-mouse IgG HRP (Jackson ImmunoResearch, 115-035-146) at a 1:10,000 dilution and goat anti-rabbit IgG HRP (Jackson Immuno Research, 111-035-144) at a 1:50,000 dilution). After three subsequent washes, ECL (Millipore) was applied and signal was detected using an Amersham Imager 600 (GE). MRG-1 was specifically chosen for a loading control because it is a euchromatic nuclear protein, its expression remains unchanged in the met-2 set-25 mutant and we have a well-characterized antibody against MRG-1.
Worm dissociation and cell sorting. Isolation of worm tissues for FACS sorting was performed as previously described 69 with minor adjustments. Synchronized L1 worms (200,000) were seeded on 15 cm peptone-rich plates for 32 h at 20 °C before processing. The worms were collected and thoroughly washed in 15 ml Falcon tubes using M9 solution before being transferred into multiple 1.5 ml low-bind tubes to obtain a pellet of approximately 100 ml for each sample (Eppendorf). The worms were resuspended in 200 μl SDS-DTT solution (20 mM HEPES pH 8.0, 0.25% SDS, 200 mM DTT and 3% sucrose) and incubated for exactly 4 min at room temperature before being resuspended in 800 μl of egg buffer (25 mM HEPES pH 7.3, 118 mM NaCl, 48 mM KCl, 2 mM CaCl 2 and 2 mM MgCl 2 with an osmolarity of about 340 mOsm). The worms were washed five times with 1 ml of egg buffer. The worm pellets were then resuspended in 100 μl of 15 mg ml pronase E (Sigma), diluted in egg buffer and pellets from the same sample were pooled. The worms were then vigorously resuspended with a thinned-out Pasteur pipette until most were visibly dissociated. Digestion was stopped with 900 μl of 10% FBS (in M9), the cells were washed twice with 10% FBS, with centrifugation at 9,600g for 5 min at 4 °C. After washing, the cells were resuspended in 1 ml of cold M9 buffer and left to settle for approximately 30 min on ice. The supernatant was collected and filtered into sorting tubes (Becton Dickinson) using 30 µm cell filters (Sysmex) and processed for cell sorting or ChIC-seq.
The filtered cells were kept on ice until sorting; 1 μl DRAQ7 (BioStatus) was added to the samples immediately before sorting to exclude dead cells. The cells were sorted into 1.5 ml low-bind tubes (Eppendorf) following a similar strategy using either a BD FACSAria (Becton Dickinson) or MA900 (Sony) cell sorter. For RNA-seq, approximately 25-50 × 10 3 muscle cells or approximately 1 × 10 4 seam cells per genotype per experiment (three replicas each) were sorted directly into 200 μl of lysis buffer (Norgen) and processed. For ATAC-seq, 50 × 10 3 muscle cells per condition per experiment (three replicas each) were sorted into 200 μl M9 solution and processed. For ChIC-seq, see the 'ChIC-seq' section.

RNA-seq.
For isolated tissues, RNA was purified from lysed muscle or seam cells following the instructions of the Norgen single cell RNA purification kit (51800, Norgen Biotek). Libraries were produced using a Smart-Seq2 mRNA sequencing kit (Illumina). For L1 RNA-seq, worms were removed and thoroughly washed from RNAi plates using M9 solution. The worm pellet was resuspended in TRIzol, flash-frozen in liquid nitrogen and then freeze-cracked four times; the RNA was extracted by phenol-chloroform extraction, followed by isopropanol precipitation. The ribosomal RNA was depleted using a Ribo-Zero gold kit (Epicentre) and libraries were produced using a Total RNA sequencing TruSeq kit (Illumina). Equimolar amounts of indexed libraries were pooled and sequenced on a HiSeq 2500 system (Illumina). For data collection and conversion to fastq format, RTA 1.18.64 (HiSeq2500), RTA 2.4.11 (NextSeqS00) and bcl2fastq2 v2.17 were used.
The reads were analysed as described previously 70 . The reads were aligned to the C. elegans genome (ce10) using the R package QuasR v1.22.0, (www.bioconductor.org/packages/2.12/bioc/html/QuasR.html) 71 . The command 'proj < −qAlign ('samples.txt' , 'BSgenome. Celegans.UCSC.ce10' , 'splicedAlignment = TRUE')' instructs hisat2 (ref. 72 ) to align using default parameters, considering unique reads for genes and genome-wide distribution. Count tables of reads mapping within annotated exons in WormBase (WS220) were constructed using the qCount function of the QuasR package to score the number of reads in each window (qCount(proj,GRange_object, orientation = 'same')) and normalized by division by the total number of reads in each library and multiplied by the average library size. Transformation into the log 2 space was performed after the addition of a pseudo count of eight to minimize large changes in abundance FC caused by low count numbers. The EdgeR package v.3.24 was applied to select genes with differential transcript abundances between indicated the genotypes (contrasts) based on the FDR for genes. Annotation of tissue and cell-type expression is based on annotation tables provided by the tissue atlas 73 .
The cells were thawed on ice (all subsequent steps were performed on ice) and washed once with WB1. The cells were then counted and diluted in WB1 to approximately 5 × 10 5 cells in 200 μl. For each antibody and mutant, 200 μl of cells was transferred to 0.5 ml low-bind tubes (Eppendorf). Primary antibodies (200 μl), diluted to 2× in WB1, were then added to the cells and incubated overnight on a rotator at 4 °C. The antibodies used were to: H3K9me3 (1:2,000; MABI0318, MBL) 68 , H3K9me2 (1:5,000; MABI0317, MBL) 68 and H3K27ac (1:5,000; ab177178, Abcam). The cells were washed once with wash buffer 2 (WB1 with no EDTA), then resuspended in 500 µL of WB2 containing 1 ng ml −1 ProteinA-MNase (recombinant) and 5 µg ml −1 Hoechst 33258, and incubated on a rotator for 1 h at 4 °C. The cells were then washed twice with 500 μl of wash buffer 2 (WB1 with no EDTA) and filtered into sorting tubes with a final volume of approximately 500 μl WB2. For each sample, approximately 5 × 10 3 cells were sorted into 5 μl of wash buffer 3 (WB1 with no EDTA and no protease inhibitor). Hoechst was used as a sorting marker for gating fixed/lysed cells.
After sorting, MNase was activated by adding 10 μl wash buffer 3 containing 4 mM CaCl 2 for 30 min on ice. The reaction was then stopped with the addition of 20 μl Stop solution (40 mM EGTA, 1.5% NP-40 and 2 mg ml −1 proteinase K) and the proteins were digested (65 °C for 6 h, 80 °C for 20 min and hold at 4 °C).
DNA fragments were blunt ended by adding 30 μl of blunt end mix (2.5 U Klenow large fragment, 5 U T4-PNK, 0.4 mM dNTPs, 2.3 mM ATP, 1.7 mM MgCl 2 , 7 μl PNK buffer 10×, 0.24 ng ml −1 BSA and 2.5% PEG8000) and incubating for 30 min at 37 °C, followed by 20 min at 75 °C for enzyme inactivation. The fragments were A-tailed by adding 30 μl tail mix (1 U AmpliTaq 360, 0.7 mM dATP, 167 mM KCl, 0.1 ng ml −1 BSA and 2.5% PEG8000) and incubation at 72 °C for 15 min. The fragments were further ligated to T-tail-containing adaptors with the sequence: forward s tr an d, G GT GATGCCGGTAATACGACTCACTATAGGGAGTTCTACAG TCCGACGATCNNNACACACTAT and reverse strand, /5Phos/TAGTGTGTNNN GATCGTCGGACTGTAGAACTCCCTATAGTGAGTCGTATTACCGGCGAGCTT. In the forward strand, the first six italic basepairs form a fork, the underlined basepairs after that are a T7 promoter, the following italic bases are the binding site (RA5) for the TruSeq Small RNA indexing primers (RPIx) and the underlined three random base pairs represent the unique molecular identifier. Ligation was performed by adding 40 μl ligation mix (1.25 µM adaptor, 2,000 U T4 ligase, 17.5 mM MgCl 2 , 52.5 mM Tris pH 7.5, 26.25 mM DTT, 1.75 mM ATP, 0.1 ng ml −1 BSA and 2.5% PEG8000) and incubating for 20 min at 4 °C, followed by 16 h at 16 °C. The ligase was inactivated by incubation for 10 min at 65 °C.
The ligation reactions were cleaned with Ampure XP beads (Beckman Coulter) at a 0.8× beads-to-sample ratio and resuspended in 8 μl clean water. The cleaned fragments were amplified by adding 12 μl of MEGAscript T7 transcription kit and incubation for 12 h at 37 °C. The template DNA was removed by the addition of 2 μl TurboDNAse (IVT kit) and incubation for 15 min at 37 °C. The produced RNA was further purified using RNA clean XP beads (Beckman Coulter) at a 0.8× ratio, followed by RNA fragmentation for 2 min at 94 °C. After another bead clean-up, 40% (5 μl) of the RNA was primed for reverse transcription by adding 1.5 μl of 3.3 mM dNTPs and 13.3 µM random hexamer RT primer (GCCTTGGCACCCGAGAATTCCANNNNNN), and hybridizing it by incubation at 65 °C for 5 min, followed by direct cool down on ice. Reverse transcription was performed by the further addition of 4 μl RT mix (2.5×first strand buffer, 25 mM DTT, 20 U RNAseOUT and 100 U SuperscriptII) and incubation at 25 °C for 10 min, followed by 1 h at 42 °C. The single-stranded DNA was purified through incubation with 5 µg or 25 U RNAseA for 30 min at 37 °C and PCR amplification to add the Illumina smallRNA barcodes and handles. Next, 40 μl of PCR mix (3.1×NEBNext Ultra II Q5 master mix (5×stock), 0.5 µM RP1 and 0.5 µM RPIx primers) was added. The number of PCR cycles performed were dependent on the abundance of the histone modification assayed, and the PCR products were cleaned twice using Ampure XP beads at a 0.8× ratio. The final libraries were eluted in 7 μl nuclease-free water and the abundance and quality were assessed using the QUBIT and Bioanalyzer systems.
ChIC-seq analysis. Equimolar amounts of indexed libraries were pooled and sequenced on a HiSeq 2500 (Illumina). Adaptors were trimmed using Trimmomatic 0.39 (http://www.usadellab.org/cms/?page=trimmomatic) 74 with parameters set to: CROP:75 HEADCROP:12. Reads were aligned to the ce10 genome assembly using bowtie2 (http://bowtie-bio.sourceforge.net/bowtie2/ index.shtml) 75 with alignment parameters set to sensitive. The reads were then deduplicated using the deduplicate_bismark function in the Bismark program (https://github.com/FelixKrueger/Bismark) 76 . The read density was calculated by tiling the genome into 500 bp non-overlapping windows and using the qCount function of the QuasR package to quantify the number of reads in each window (qCount(proj,GRange_object,orientation = 'any')). H3K9me-positive domains were determined as regions with a consecutive enrichment of H3K9me2 and/ or H3K9me3 over the input. For gene quantification, gene annotation from WormBase was used (WS220). Quantitation for each gene was performed by counting the reads overlapping the exons. Genome annotation was based on the BSgenome.Celegans.UCSC.ce10 package (https://bioconductor.org/packages/ release/data/annotation/html/BSgenome.Celegans.UCSC.ce10.html). Differences in read depths between samples were normalized by dividing each sample by the total reads and multiplying by the average library size. The various count tables used throughout this study were normalized to the total genome count. The log 2 -transformed expression levels were determined after the addition of a pseudo count of eight (y = log 2 (x + 8)) to minimize large changes in FC caused by low count numbers. To control for unspecific activity of the MNase enzyme, results are displayed as: (mean enrichment of antibody + MNase-protein A-treated sample) ÷ MNase-protein A-treated sample. Replicate experiments were combined by averaging the enrichment values. Annotation of tissue and cell-type expression is based on annotation tables provided by the tissue atlas 73  ChIP-qPCR. Synchronized L1 animals were seeded on peptone-rich plates and L3 animals were harvested 36 h later. The worms were washed in M9 solution until clear. The worm pellet was then resuspended in 10 ml of M9 containing 2% formaldehyde and fixed for 30 min at room temperature on a rocker. Fixation was stopped with 500 μl of 2.5 M glycine for 5 min at room temperature on a rocker. The worms were then washed twice with M9. The worm pellet was resuspended in 1 ml of FA buffer (50 mM HEPES pH 7.5, 1 mM EDTA, 1% Triton X-100, 0.1% Na-deoxycholate and 150 mM NaCl) containing 1% sarcosyl, transferred into bead-beater tubes with 200 μl of 0.5 mm zirconia/silicon beads and the worms were lysed in a Fast Prep 24-5 G benchtop homogenizer. The recovered supernatant was transferred into 15 ml polystyrene tubes and sonicated (30 cycles; 30 s on/ off at 4 °C) using the Bioruptor plus sonicator (Diagenode). The samples were transferred to DNA LoBind tubes (Eppendorf), were centrifuged at maximum speed for 5 min at 4 °C, and the supernatant (containing chromatin) was collected. For each sample, 100 μg of chromatin was diluted to 600 μl in FA buffer, and incubated overnight on a rocker at 4 °C with 5 μl of GFP-Trap magnetic particles (ChromoTek). The magnetic particles were washed with the following buffers: 3 × 5 min with FA buffer, 5 min with FA buffer containing 1 M NaCl, 10 min with FA buffer containing 500 μl NaCl, 5 min with TEL buffer (0.25 M LiCl, 1% NP-40, 1% Na-deoxycholate, 1 mM EDTA and 10 mM Tris-HCl pH 8) and 2 × 5 min with TE. The complexes were eluted with 100 μl of elution buffer (1% sarcosyl in TE with 250 mM NaCl) at 65 °C with shaking at 1,100 r.p.m. for 15 min. Cross-links were reversed overnight at 65 °C, followed by addition of 20 μg proteinase K and further incubation for 2 h at 65 °C. The DNA was then purified using AMPure XP beads at a 0.8× ratio.
ATAC-seq. ATAC-seq was performed on sorted muscle cells following the Omni-ATAC protocol 77 . Briefly, sorted cells were centrifuged for 10 min at 1,000g (4 °C) using a swinging-bucket rotor, and the supernatant was carefully removed to avoid disturbing the cell pellet. The cells were lysed for 3 min in 50 μl of cold resuspension buffer (10 mM Tris-HCl pH 7.4, 10 mM NaCl and 3 mM MgCl 2 ) containing 0.1% NP-40, 0.1% Tween-20 and 0.01% digitonin. Lysis was stopped with 1 ml of cold resuspension buffer containing 0.1% Tween-20, and the nuclei were pelleted for 10 min at 1,000g (4 °C). The supernatant was carefully removed and the nuclei were resuspended in 50 μl of transposition mix (25 μl 2×TD buffer, 2.5 μl Tn5, 16.5 μl PBS, 0.5 μl of 1% digitonin, 0.5 μl of 10% Tween-20 and 5 μl double-distilled water) and incubated for 30 min at 37 °C on a thermomixer. The DNA was then purified using a Qiagen MinElute PCR purification kit according to the manufacturer's instructions and eluted in 21 μl of elution buffer. Libraries were amplified as originally described 78 and the final preparations were purified using a 1.6× ratio of AMPure XP beads. Sequencing was performed using NextSeq, with 75 cycles and paired-end reads.
ATAC-seq analysis. The ATAC-seq reads were pre-processed following the ENCODE ATAC-seq pipeline (https://github.com/ENCODE-DCC/ atac-seq-pipeline). Adaptors were trimmed using cutadapt (https://github.com/ marcelm/cutadapt; 19 ) and reads were aligned to the ce10 genome assembly using bowtie2 (http://bowtie-bio.sourceforge.net/bowtie2/index.shtml) 79 with alignment parameters set to sensitive. The mapped reads were marked and removed using the MarkDuplicates function of Picard tools v.2.20.0 (https://broadinstitute.github.io/ picard/). Paired-end reads were converted to single-end reads using bedtools 2.26 (https://github.com/arq5x/bedtools2) 80 and the reads were shifted +4 bp and −5 bp for the positive and negative strands, respectively, to account for characteristics of the Tn5 enzyme. Peaks were called using MACS2 (ref. 81 ) with the following parameters: -g 93260000-nomodel-shift -100-extsize 200. Differential peaks among the genotypes were called using the EdgeR package v.3.24.The read density was calculated by tiling the genome into 500 bp non-overlapping windows or specifically at promoters (defined as 1 kb upstream and 100 bp downstream of the TSS) and using the qCount function of the QuasR package to quantify the number of reads in each window (qCount(proj,GRange_object,orientation = 'any')). Differences in read depths between samples were normalized by dividing each sample by the total number of reads and multiplying by the average library size. The various count tables used throughout this study were normalized to the total genome count. Definitions of characteristic chromatin domains was taken from 82 .
TF-motif analysis. A compendium of TF weight matrices was downloaded (http:// hugheslab.ccbr.utoronto.ca/supplementary-data/CeMotifs/) to examine the set in the file 'TF_Information.txt' . These were used to scan the C. elegans ce10 genome using the matchPWM function in the Bioconductor package Biostrings. Only hits with a minimum score of ten were considered, unless the maximum obtainable score by the weight matrix was lower than ten. In that case, the maximum obtainable score was required. The resulting binding sites were then intersected with gene-promoter annotation (WS220; 1,500 bp 5′ and 500 bp 3′ of the TSS) to determine the number of sites for each TF and gene.

Statistics and reproducibility.
The experiments shown in this study were performed as 2-4 biologically independent experiments, as indicated in the figure legends, and no inconsistent results were observed. Data plotted as boxplots indicate the 25th and 75th percentiles, with the whiskers showing the minima and maxima (5th and 95th percentiles), black circles indicating the outliers and the horizontal line showing the median. Some data are plotted in bar graphs as the mean ± s.d, unless specified otherwise. If not stated otherwise statistical testing was performed using two-sided a Wilcoxon rank-sum test. The FDR index was calculated using the edgeR package (see Methods). Details of the particular statistical analyses used, precise P values, statistical significance, number of independent biological replicates and sample sizes for all of the graphs are indicated in the figures or figure legends. No data were excluded.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
All datasets from this study have been uploaded to the Gene Expression Omnibus (GEO) with the accession code of GSE167168. Previously published EMR-1 DamID and embryo RNA-seq/ChIP-seq data that were re-analysed here are available under the accession codes GSE136577 and SRP080806. All other data supporting the findings of this study are available from the corresponding author on reasonable request. Source data are provided with this paper.      Fig. 5 | Proximity to repeats does not correlate with gene derepression in H3K9me-deficient muscle cells. a) Violin plots of distances (in kb) between derepressed genes and the nearest repetitive element subdivided by repeat family (according to repeat masker annotation), in muscle cells from met-2 set-25 mutants (red). Derepressed gene sets are compared to the same number of random, non-derepressed genes (cyan). Plots have been split to accommodate range of distances. The mean distances for each group are indicated by a dot. Note that repeat elements Chapeav, SNA, TcMar-Tc4, Harbinger and Merlin are slightly closer to the TSS of derepressed genes than the random gene set, but for all except DNA, the average distance to the TSS is >200 kb. b) Correlation of the mean distances in log2 kb between genes derepressed in met-2 set-25 mutants and random set of genes to the nearest repetitive element subdivided by repeat family. We note that the smallest mean distance (DNA repeat family) was still 3.9 kb upstream of the TSSs. Only 87 (26%) of the 331 derepressed genes have a repeat element < 1 kb from the corresponding TSS. In comparison 34% of all genes in the genome have a repeat element < 1 kb in proximity to their TSS. c) Heatmap highlighting log2FC of gene expression for the indicated TFs, in two independent replicas, with indicated mutants over wt. d) Plot showing log2 enrichment of TF binding motifs, versus genome-wide occurrence, in genes marked by H3K9me3 in wt and transcriptionally derepressed (FC > 2; FDR < 0.01) in met-2 set-25 muscle cells (labelled "derepressed") or transcriptionally unchanged (labelled "no change"). Significantly enriched motifs (log2 enrichment>0.5; log2 abundance>8) are colour coded with their corresponding TF(s) indicated. e) As (d) for germline genes that are derepressed in both muscle and seam cells from met-2 set-25 animals. a) The normalized ATAC-seq reads from muscle cells of indicated genotypes are plotted, as boxplots, for all intergenic peaks, or for those that gain accessibility in the indicated genotype. Boxplots show median, boxes 50% and whiskers 90% of the group. N = 3 biological independent replicas. b) Representative images of body region from wt or met-2 set-25 worms expressing the control or enhancer reporter (Fig. 7e). Worms were straightened in ImageJ. Scale bar, 10 µm. The experiment has been performed for three times with similar results, n pooled worms analysed = wt control, 30; met-2 set-25 control, 19; wt enhancer, 47; met-2 set-25 enhancer, 43.