Complete loss of H3K9 methylation dissolves mouse heterochromatin organization

Histone H3 lysine 9 (H3K9) methylation is a central epigenetic modification that defines heterochromatin from unicellular to multicellular organisms. In mammalian cells, H3K9 methylation can be catalyzed by at least six distinct SET domain enzymes: Suv39h1/Suv39h2, Eset1/Eset2 and G9a/Glp. We used mouse embryonic fibroblasts (MEFs) with a conditional mutation for Eset1 and introduced progressive deletions for the other SET domain genes by CRISPR/Cas9 technology. Compound mutant MEFs for all six SET domain lysine methyltransferase (KMT) genes lack all H3K9 methylation states, derepress nearly all families of repeat elements and display genomic instabilities. Strikingly, the 6KO H3K9 KMT MEF cells no longer maintain heterochromatin organization and have lost electron-dense heterochromatin. This is a compelling analysis of H3K9 methylation-deficient mammalian chromatin and reveals a definitive function for H3K9 methylation in protecting heterochromatin organization and genome integrity.

H eterochromatin is gene-poor, repeat-rich regions of the genome that are classically defined to remain condensed during interphase 1 . From these early cytological studies, DAPI staining of A/T-rich repeats to high-resolution microscopy, heterochromatin can be described as an electron-dense chromatin material that shows the concentration in subnuclear compartments and is often associated with the nuclear periphery. Heterochromatin has important functions for chromosome segregation, genome integrity, and gene regulation 2,3 .
A biochemical pathway for heterochromatin formation involves Suv39h-mediated H3K9me3, which is bound by the heterochromatin protein HP1 4,10,11 . HP1 binding promotes compaction and transcriptional repression and stabilizes Suv39h binding to heterochromatin. However, while disruption of Suv39h enzymes results in loss of H3K9me3 and dispersion of HP1, the overall organization of heterochromatin in DAPI-dense foci remains largely intact 5 , indicating that other mechanisms are involved.
Several observations suggest a partial redundancy between H3K9 methylation systems. A sub-stoichiometric complex containing G9a, Suv39h, and Eset1 has been described 12 . Eset1 knockdown in Suv39h double-null MEFs results in partial dispersion of DAPI-dense heterochromatic foci 13 . In C. elegans, double deletion of MET-2 (a Setdb1 homolog) and SET-25 (a KMT with some homology to G9a and Suv39h) results in loss of heterochromatin anchoring to the nuclear lamina 14 . Interestingly, MET-2/SET-25 mutant worms are viable despite lacking all H3K9 methylation, but derepress transposons and satellite repeats, leading to genome instability and sterility 15,16 .
We addressed potential redundancy between H3K9 methylation systems in mouse fibroblast cells by generating compound mutants for all known SET-domain H3K9 lysine methyltransferases (KMT) using the CRISPR/Cas9 technology. We isolated mouse embryonic fibroblast (MEF) cell lines lacking progressively 2, 4, or all 6 H3K9 KMTs. Compound mutant MEF cells lacking all six enzymes lost all H3K9 methylation, derepress multiple families of repeat elements, and display DNA damage and cell death. While mutants retaining the function of some of the H3K9-methylating enzymes maintain heterochromatin clustering, 6KO MEF cells show the complete collapse of heterochromatin foci and loss of electron-dense heterochromatin. These results illustrate the functional redundancy between the three H3K9 KMT systems and provide compelling evidence for an essential function of H3K9 methylation in heterochromatin organization.

Results
Gene pair disruptions identify distinct functions for the Suv39h, Eset, and G9a enzymes in directing H3K9 methylation states. In mouse cells, 6 genes encode SET-domain-containing enzymes with a documented methyltransferase activity toward H3K9: Suv39h1 and Suv39h2, Eset1/Setdb1 and Eset2/Setdb2, and G9a/Ehmt2 and Glp/Ehmt1 (Fig. 1a). Since Setdb1/Eset1-deficient cells display growth defects, we took advantage of existing mouse embryonic fibroblasts (MEFs) with a tamoxifen-inducible conditional deletion of Eset1 6 , hereafter called Eset1cKO, or E1c. We refer to the E1c cells as wild-type (WT) control MEF cells, which, after tamoxifen-induced deletion of Eset1, are converted into E1 (Eset1-null) MEFs. We used the CRISPR/Cas9 system in the E1c MEFs to introduce frameshift mutations within the catalytic domain of all the other SET-domain genes encoding H3K9 KMTs (Fig. 1a).
We first generated cell lines lacking the function of each set of paralogous genes. After transfecting E1c MEFs with plasmids carrying Cas9 and the respective guide RNAs, we isolated clones and confirmed the absence of detectable target proteins as well as the efficiency of tamoxifen-induced Eset1 deletion by western blotting (Fig. 1b). We then assessed the contribution of each enzyme pair to the total levels of H3K9 methylation, by probing chromatin extracts with antibodies specific for H3K9me1, H3K9me2, and H3K9me3 (Fig. 1c). Eset1/Eset2 (E1E2) doublemutant MEFs do not display a significant reduction in overall H3K9 methylation levels. By contrast, G9a/Glp (G9Gl)-deficient cells have greatly lowered H3K9me1 and H3K9me2, while Suv39h1/Suv39h2 (S1S2) mutants lack most H3K9me3, and to a lesser extent, H3K9me2. Thus, the different H3K9 enzyme pairs have distinct functions in establishing H3K9 methylation, in agreement with published observations 5,8 .
Eset-mutant MEF cells display micronuclei but maintain heterochromatic clustering. We visualized pericentric heterochromatin organization in these mutant MEF cells by performing DNA fluorescence in situ hybridization (DNA FISH) with probes specific for major satellite repeats (MSR). In WT (E1c) control cells, MSR sequences are clustered in bright and regular foci overlapping DAPI-dense areas (Fig. 1d). This organization is maintained in Eset1/Eset2 or G9a/Glp double-deficient cells, and only Suv39h1/Suv39h2 mutant cells display very modest aberrations in the definition of heterochromatic foci.
We then determined the integrity of the nuclear envelope by immunostaining with an antibody directed against the nuclear lamina component Lamin B. We did not observe any alteration in the morphology of the nuclear lamina of G9a/Glp and Suv39h1/ Suv39h2 mutant cells when compared to WT. In contrast, Eset1/ Eset2-deficient cells have multiple micronuclei and display deformation of the nuclear lamina/envelope (Fig. 1e), suggesting that they harbor significant DNA damage.
These data indicate distinct functions for the Suv39h1/ Suv39h2, Eset1/Eset2, and G9a/Glp systems in establishing H3K9 methylation states, as well as the different requirements for maintaining nuclear envelope morphology. However, none of these gene pair mutants show significant alterations in heterochromatin organization, suggesting a higher degree of redundancy between these three H3K9-methylating systems.
Compound KMT-mutant MEF cells that lack the six SETdomain H3K9-methylating enzymes have lost all H3K9 methylation states. To address this question, we next generated mutant cell lines with compound inactivation of multiple H3K9 SET-domain genes, in combination with the Eset1-conditional mutation. We proceeded stepwise to establish two series of mutant cell lines that progressively lack increasing numbers of H3K9-methylating enzymes (Fig. 2a). In the first series, we generated Eset1-conditional MEF cells lacking Suv39h1 and Suv39h2 (E1c S1S2) and further inactivated the remaining three H3K9 SET-domain genes to generate cells carrying mutations in all six H3K9-methylating enzymes (E1cE2 S1S2 G9Gl). In a parallel series, we first targeted Eset2 (E1cE2), then G9a and Glp (E1cE2 G9Gl), and finally Suv39h1 and Suv39h2 (E1cE2 G9Gl S1S2). We confirmed the absence of mutant gene products, as well as the efficiency of tamoxifen-induced Eset1 deletion by western blotting (Fig. 2b).
We first characterized the growth properties of these compound H3K9 KMT-mutant cell lines. While all compound H3K9 KMT cell lines are viable and proliferating when growing in the absence of tamoxifen (i.e., when Eset1 is still functional), inducing Eset1 deletion in the continuous presence of tamoxifen leads to cell death after a few days (i.e., more than 4 days) in culture (Fig. 2c). This lethality is more pronounced in cells lacking all six H3K9 KMT (E1E2 S1S2 G9Gl or abbreviated as 6KO1 and E1E2 G9Gl S1S2 or abbreviated as 6KO2) when compared to the single inactivation of Eset1 (E1).
For the following experiments and sample preparations, we, therefore, induced Eset1 deletion by a 2-day incubation with tamoxifen, followed by a 2-day culture in the normal medium (Fig. 2d).
We then examined H3K9 methylation in chromatin extracts from the various H3K9 KMT mutants by western blotting  To validate this observation with an antibody-independent approach, we analyzed the same extracts by mass spectrometry (Fig. 2f). We detected very similar changes in H3K9 methylation states, and confirmed the nearly total absence of H3K9 methylation in chromatin extracted from the 6KO1 and 6KO2  We also investigated possible changes in the abundance of other histone modifications by mass spectrometry, and observed only a moderate increase in H3K9 acetylation or in H3K27me3, together with a decrease in H3S10 phosphorylation, specifically in extracts from the 6KO mutants ( Supplementary Fig. 1A). We also examined changes in total levels of H3 or H4 acetylation, as well as H3K4me3 by western blotting. While H3 acetylation was increased in some of the H3K9 KMT mutants, particularly in extracts from 6KO cells, H4 acetylation and H3K4me3 were not significantly altered ( Supplementary Fig. 1B). Therefore, loss of H3K9 methylation does not lead to a general or bulk increase in histone modifications associated with active chromatin, although the absence of H3K9 methylation might facilitate H3 acetylation. Importantly, these changes may not occur specifically at heterochromatin regions.
The presence of micronuclei in some of the H3K9 KMT mutants prompted us to examine the phosphorylation of the histone variant H2A.X (γH2A.X), a known marker of DNA double-strand breaks 17 . We observed an increase in γH2A.X signal in some of the H3K9 KMT mutants, particularly in extracts from 6KO1 cells ( Supplementary Fig. 1C), but these levels remained low when compared to control cells treated with the topoisomerase inhibitor Etoposide and they were not apparent in extracts from 6KO2 cells. This might reflect different kinetics of DNA repair at heterochromatin regions and/or a somewhat altered involvement of heterochromatin proteins in the activation of DNA repair pathways 18 between the 6KO1 and 6KO2 mutants. In addition, it has been shown that micronuclei can also form without γH2A.X accumulation 19 .
We further assessed global DNA methylation levels in the 6KO mutant cells, either by DNA dot-blot analysis using an antibody recognizing 5-methylcytosine (5mC) (Supplementary Fig. 1D) or by mass spectrometry (Supplementary Fig. 1E). As controls, we used mouse ES cells that were either wild type or mutant for the three DNA methyltransferases establishing 5mC (Dnmt TKO) 20 . Dnmt TKO ES cells have only residual levels of 5mC and of 5hydroxymethylcytosine (5hmC). In contrast, 6KO MEF cells maintain high, although modestly decreased 5mC and 5hmC levels, corresponding to~80% 5mC (P < 0.05) or 60% 5hmC (non-significant) of the signals observed in WT MEFs. These data indicate that loss of H3K9 methylation does not lead to an immediate and overall or genome-wide loss of DNA methylation.
Massive derepression of repeat elements and retrotransposons in compound H3K9 KMT-mutant MEF cells. We determined the transcriptional response of H3K9 KMT disruption by performing RNA sequencing (RNA-seq) in the various H3K9 KMTmutant cell lines. We first focused on all repetitive elements in the mouse genome. As previously reported 6,21 , deletion of Eset1 alone is sufficient to induce derepression of a subset of endogenous retroviruses (ERV) ( Fig. 3b and Supplementary Fig. 2). Notably, the numbers of derepressed ERV elements, as well as the magnitude of their upregulation, are increased in cell lines where additionally also G9a and Glp are deleted (E1E2 G9Gl) and further augmented in the 6KO1 and 6KO2 cells (Fig. 3a, b). Among the distinct ERV subtypes, ERV class I elements of the ERV1 family are most sensitive to Eset1 depletion (Supplementary Fig. 2A). ERV class II elements, such as the ERVK family, show only a modest increase in expression in cells lacking Eset1, or both Eset1 and Eset2 (E1E2), but are strongly upregulated in E1E2 G9aGl mutants and in 6KO1 and 6KO2 cells (Fig. 3c). In contrast, expression of ERV class III elements is either unaltered (ERVL), or only moderately increased (MaLR), even in mutant cells lacking all six KMT (Supplementary Fig. 2A).
Non-long terminal repeat (LTR) containing retrotransposons are classified as short or long interspersed nuclear elements (SINE or LINE). While expressed at relatively high levels in WT cells, SINE is not dysregulated by the lack of H3K9 KMT (Fig. 3b). LINE elements, in particular LINE L1, are overexpressed specifically in H3K9 KMT-mutant cells lacking G9a and Glp (E1E2 G9Gl) and in 6KO1 and 6KO2 cells (Fig. 3a, b).
Satellite DNA forms large tandem repeat arrays at centromeric and pericentromeric regions, but can also be found in a few intergenic regions. In particular, major satellite repeats (MSR) are repressed by Suv39h enzymes in various cell types, including fibroblasts 22 . Satellite repeats are exclusively derepressed in mutant cells lacking Suv39h enzymes, and not in other compound H3K9 KMT mutants (Fig. 3b). In cells depleted for Suv39h1/Suv39h2 together with Eset1, the only upregulated satellite repeats are MSR (annotated as GSAT_MM in Supplementary Fig. 2A). Interestingly, MSR expression is further increased in cells that lack all 6 KMT, indicating that while Suv39h are the chief H3K9-methylating enzymes for MSR repression, other H3K9 KMT can participate in their control ( Fig. 3c and Supplementary Fig. 2A). In addition to MSR, 6KO1 and 6KO2 mutant cells also upregulate MMSAT4 sequences (Supplementary Fig. 2A).
We also assessed changes in chromatin accessibility by ATAC-seq. The disruption of H3K9 KMT leads to a global increase in chromatin accessibility over repeat elements, which is particularly pronounced for MSR sequences (Fig. 3c, d and Supplementary Fig. 2B). Surprisingly, however, changes in Fig. 2 Loss of H3K9 methylation in compound H3K9 KMT-mutant MEF cells. a Generation of compound H3K9 KMT-mutant MEF by successive CRISPR/ Cas9-mediated inactivation of genes encoding H3K9 KMT enzymes. Two series of H3K9 KMT mutants were generated. Nomenclature for the H3K9 KMT mutants is indicated on the scheme. E1c (Eset1-conditional, also WT), E1 (Eset1-null), E2 (Eset2-null), G9 (G9a-null), Gl (Glp-null,) S1 (Suv39h1-null), and S2 (Suv39h2-null). b Western blot analysis for H3K9 KMT gene products in whole-cell extracts of WT and H3K9 KMT mutants. Gapdh was used as a loading control. Representative results of two independent experiments. Source data are provided as a Source Data file. c Proliferation capacity of compound H3K9 KMT mutants. The y axis indicates cumulative doublings after consecutive days in culture (x axis). E1c/WT and E2 were grown in a normal medium, and all other H3K9 KMT mutants in presence of 1 μM 4-OHT in a culture medium (individual data points with line connecting their mean, N = 5 independent experiments). Statistically significant differences in doubling time, when compared to WT (left) or E2 (right) growth curves are shown by asterisks (***P = 1.1 × 10 −10 for E1 and E1 S1S2, P = 7.2 × 10 −11 for 6KO1, P = 2.8 × 10 −8 for E1E2, P = 2.6 × 10 −8 for E1E2 G9Gl and 6KO2, extra-sum-of-squares F test). d Induction of Eset1 deletion. Cells were grown for 2 days in presence of 1 μM 4-OHT in culture medium, followed by 2 days in normal medium without 4-OHT. After cell lysis and fractionation, histones were acid-extracted from the nuclear fraction. e Western blot analysis of H3K9 methylation states in compound H3K9 KMT mutants. Acid-extracted histones were probed with antibodies specific for H3K9me1, H3K9me2, and H3K9me3. Total H3 was used as a loading control. Representative results of three independent experiments. Source data are provided as a Source Data file. f Mass spectrometry analysis of H3K9 methylation states. The y axis indicates the fraction of total H3 carrying H3K9me1, H3K9me2, or H3K9me3 modifications, determined by normalizing intensities obtained for the corresponding peptides in nanoLC-MS (bars represent mean with individual data points overlaid, N = two independent samples).  . From left to right: E1 S1S2, E1E2 G9Gl, and 6KO1. Black dots represent repeat elements whose change in expression is not statistically significant. Red, LINE elements. Light blue, LTR retrotransposons. Green, Satellites. Yellow, SINE elements. Dark blue, DNA transposons. b Global changes in expression for distinct repeat classes (SINE, LINE, LTR, and Satellite), in each of the compound H3K9 KMT mutants. The y axis indicates normalized read counts (mean ± SD, individual data points overlaid, N = 3 independent samples). Satellites are shown on a separate graph due to their lower expression level. Asterisks indicate statistically significant differences when compared to WT levels (LINE: ***P = 1.1 × 10 −4 for 6KO1, **P = 0.006 for E1E2 G9Gl, ***P = 8.0 × 10 −4 for 6KO2; LTR: ***P = 3.7 × 10 −4 for E1, *P = 0.019 for E1 S1S2, ****P = 6 × 10 −5 for 6KO1, *P = 0.026 for E1E2, ****P = 8.4 × 10 −5 for E1E2 G9Gl, ***P = 3.6 ×10 −4 for 6KO2; Satellite, ***P = 1.2 × 10 −4 for 6KO1 and P = 1.4 × 10 −4 for 6KO2, two-sided, unpaired t test). c Coverage plot of major satellite repeat (MSR) transcripts to the consensus MSR sequence in WT and H3K9 KMT mutants (from top to bottom: E1 S1S2, E1E2 G9Gl, and 6KO1). ATAC-seq indicating DNA accessibility in the same mutants is shown below. chromatin accessibility did not strictly reflect changes in repeat expression, and several repeat types, such as ERVL, gained chromatin accessibility without being derepressed (Supplementary Fig. 2B).
In summary, the disruption of H3K9 KMT leads to the derepression of multiple families of repeat elements. These results also reveal that distinct H3K9 KMT have both specific and overlapping functions in the silencing of repeat elements.
Activation of pattern-recognition and DNA damage pathways in compound H3K9 KMT-mutant MEF cells. We next analyzed We determined the functional classification of these deregulated genes with the Ingenuity Pathway Analysis database (Fig. 4b). Among the most significantly enriched pathways, pattern-recognition receptors were consistently upregulated. This included, among other genes, the Tlr3 and Tlr4 genes, encoding Toll-like receptors, as well as Ddx58/Rig-I (Fig. 4c). This is consistent with previous studies where treatment of cancer cells with DNMT inhibitors led to overexpression of ERV, which triggered the activation of the double-stranded RNA sensor TLR3 and, in turn, a type I interferon response 23,24 . Accordingly, H3K9 KMT-deficient MEF cells also display activation of interferon signaling (significantly enriched pathway in all H3K9 KMTmutant cells except Eset2 mutants), including overexpression of type I interferon regulators Irf7, Irf9, and Stat1 (Fig. 4c).
Other significantly altered pathways indicate activation of DNA damage-sensing mechanisms. This includes MAP kinase signaling, with upregulation of Mapk10, Mapk11, and Pik3r5, among others, and of the p53 pathway, with increased expression of Trp53inp1, Rb1, and Cdkn1a (Fig. 4c). In agreement with the growth defect exhibited by the H3K9 KMT-mutant cell lines, we also observed reduced expression of multiple cyclin-dependent kinases (Cdk1, Cdk2, Cdk4, Cdk6, and Cdk7, Fig. 4c), and upregulation of apoptosis-related factors, such as Casp7 and multiple poly (ADP-ribose) polymerases.
Collectively, these results indicate that loss of H3K9 KMT induces a signaling cascade, including DNA damage-sensing pathways, interferon response, attenuation of cell proliferation, and cell death. The activation of pattern-recognition receptors, such as Tlr3 and Rig-I, suggests that a major cause of these defects could be the massive derepression of repeat elements, primarily ERV, that we observed in the H3K9 KMT-mutant cells.
Dysregulation of H3K9me-positive vs. H3K9me-negative genes in H3K9 KMT-mutant MEF cells. We wondered to what extent these changes in gene expression would reflect direct participation of H3K9 methylation. Since comprehensive datasets for all three H3K9 methylation states are not available in mouse fibroblasts, we generated reference maps for H3K9me1, H3K9me2, and H3K9me3 in WT (E1c) MEFs. We first validated the antibodies by ChIP-qPCR at known target loci ( Supplementary  Fig. 3A) and then performed chromatin immunoprecipitation followed by deep sequencing (ChIP-seq). Peak calling identified 42893 regions of enrichment for H3K9me1, 29285 for H3K9me2, and 27281 for H3K9me3. Distinct H3K9 methylation states were often observed in overlapping genomic intervals. However, H3K9me1 peaks had only low signals for H3K9me2 and H3K9me3, H3K9me2 peaks coincided with broad signals for H3K9me3 and H3K9me3 peaks also had some enrichment for H3K9me2 ( Supplementary Fig. 3B, top panels). In total, 559 Mb, corresponding to 21% of the mouse genome were decorated by H3K9 methylation in WT MEFs.
A large fraction of H3K9 methylated regions corresponded to intergenic and intronic sequences, ranging from 55% for H3K9me1 to 69% for H3K9me2 and 70% for H3K9me3 ( Supplementary Fig. 3C). A smaller subset of regions overlapped with gene promoters, defined as sequences within + /− 1 kilobase (kb) of the transcription start site (TSS). This was the case for 19% of H3K9me1 and around 10% for H3K9me2 or H3K9me3-enriched regions. We identified 4289 genes that are decorated by H3K9me1 at their promoters (e.g., Tcap and Cacna1i), 1567 genes with H3K9me2-positive promoters (e.g., Defb40 and Olfr1431), and 1415 genes with promoters enriched for H3K9me3 (e.g., Nkx6-1 and Bmp2) (see Supplementary Fig. 4A). The majority of H3K9me1-positive promoters (3179 genes) were devoid of H3K9me2 and H3K9me3. By contrast, H3K9me2 and H3K9me3 often decorated the same gene promoters and only 270 or 462 genes were predominantly marked by H3K9me2 or H3K9me3 ( Supplementary Fig. 4B). In total, we detected H3K9 methylation at the promoters of 5297 genes.
We then used our RNA-seq data to determine the expression status of these H3K9me-positive genes in WT (E1c) MEFs and in the various H3K9 KMT-mutant cell lines. For this, we applied a meta-analysis that illustrates gene expression in violin plots. Surprisingly, genes with H3K9me1-marked promoters have significantly (P < 2.2 × 10 −16 ) higher expression levels as compared to genes without H3K9 methylation (H3K9me0 genes in the left panel of Fig. 5a), and this elevated expression was largely unaltered in all of the H3K9 KMT mutants (Fig. 5a and browser shot for example gene Bbs9 in Fig. 5b). By contrast and as expected, genes with H3K9me2-or H3K9me3-marked promoters were silenced and had significantly (P < 2.2 × 10 −16 ) reduced or no expression levels as compared to H3K9me-negative (H3K9me0) genes (Fig. 5a). The H3K9me2 mediated gene silencing was weakened (P < 0.01) in E1E2 G9Gl compound H3K9 KMT-mutant cells (Fig. 5a). Both H3K9me2 and H3K9me3 mediated gene silencing was partly lost (P < 0.001) in 6KO1 and 6KO2 cells (Fig. 5a and browser shots for example genes Treml4 and Zfp936 in Fig. 5b). However, it is important to note that loss of H3K9 methylation only caused desilencing in a minor fraction of the H3K9me2-or H3K9me3-positive genes: with a twofold cutoff (and a base mean of >100 reads), 226 of 270 H3K9me2-positive genes, and 390 of 462 H3K9me3-positive genes were not significantly upregulated in any of the H3K9 KMT mutants, not even in the 6KO1 or 6KO2 cells.
We further focused on the 6KO mutants and analyzed changes in gene expression (as compared to WT) separately for H3K9mepositive and for H3K9me-negative genes (H3K9me+ and H3K9me-genes, Fig. 5c). While H3K9me-positive genes represent 11% (5297 genes) of all genes, they account for >20% Fig. 4 Changes in gene expression in compound H3K9 KMT-mutant MEF cells. a MA plots depicting changes in gene expression in compound H3K9 KMT mutants relative to WT, as determined by RNA sequencing (RNA-seq). From left to right: E1 S1S2, E1E2 G9Gl, and 6KO1. Black dots represent genes whose changes in expression are not statistically significant. Yellow dots represent upregulated genes, and blue dots downregulated genes. b Ingenuity pathway analysis (IPA) of genes differentially expressed in compound H3K9 KMT mutants relative to WT. From left to right: E1 S1S2, E1E2 G9Gl, and 6KO1. The x axis indicates fold change, and the color code indicate the z score for each category (legend on the right). Orange indicates upregulation, and purple downregulation of the pathway. Pathways related to pattern-recognition receptors, interferon signaling, MAP kinase signaling, or cell cycle regulation are highlighted by symbols as indicated below the table. c Genome browser tracks showing RNA expression levels (top) or ATAC accessibility (bottom) of example genes in compound H3K9 KMT mutants. From left to right, Toll-like receptor 3 (Tlr3), interferon regulatory factor 7 (Irf7), cyclin-dependent kinase inhibitor 1 A (Cdkn1a), and cyclin-dependent kinase 1 (Cdk1). For both RNA-seq and ATAC-seq tracks, the top profile corresponds to WT, followed from top to bottom by E1 S1S2, E1E2 G9Gl, and 6KO1 mutants.
of the upregulated genes in the 6KO1 cells (701 of 3427 genes) (Fig. 5c) or 6KO2 cells (508 of 2517 genes), indicating an increased response of H3K9me-positive genes to loss of H3K9 methylation. By contrast, 80% of the genes derepressed in the 6KO1 mutants (2726 of 3427 genes) do not have H3K9 methylation at their promoters in WT cells. While some of these H3K9me-negative genes might still be regulated by H3K9 methylation at distal sites, such as enhancers, this result suggests that the majority of the genes dysregulated in the compound H3K9 KMT mutants are not direct targets for the H3K9 KMT systems. Instead, we propose that the failure of repeat repression and the induction of DNA damage and stress signaling pathways in the compound H3K9 KMT mutants leads to widespread changes in gene expression.  (Fig. 6a). As described above (Fig. 1), Eset1, Eset2, or Eset1/Eset2 double-mutant cells display a nuclear clustering of MSR sequences that is indistinguishable from WT cells. MSR clustering is also maintained in E1E2 G9Gl deficient cells, although they display an increased number of smaller heterochromatin foci. By contrast, cells with a compound deficiency of Eset1 and Suv39h1/Suv39h2 enzymes (E1 S1S2) show a partial dispersion of heterochromatin, with a significant fraction of the cells (37%) displaying elongated or generally irregular MSR foci (Fig. 6a). We also stained these cells with antibodies for H3K9me3 and the heterochromatin protein HP1α (Supplementary Fig. 5A, B). Both H3K9me3 and HP1α are completely dispersed in the nuclei of the E1 S1S2 mutant cells. This is not detected in compound H3K9 KMT mutants with functional Suv39h enzymes, even when all of the other four H3K9 KMT are inactivated (E1E2 G9Gl). We verified the specificity of these observations by analyzing the nuclear distribution of Hmga1, a protein that binds AT-rich sequences within pericentric chromatin 25,26 . In all compound H3K9 KMT mutants, Hmga1 signals remain intact and overlap with the DAPI-dense regions ( Supplementary Fig. 5C).
To further investigate nuclear organization in the H3K9 KMT mutants, we visualized the integrity of the nuclear lamina by immunofluorescence detection of Lamin B (Fig. 6b). As we observed for Eset1/Eset2 double-null cells, all compound H3K9 KMT mutants lacking Eset1 function displayed multiple micronuclei and deformation of the nuclear lamina/envelope, in agreement with these cells expressing markers of DNA damage (see Supplementary Fig. 1C). We also analyzed the distribution of Fibrillarin, a protein-labeling nucleoli, and did not observe significant differences between the various compound H3K9 KMT mutants (Fig. 6c). These data suggest that the global organization of subnuclear compartments is not drastically altered in the H3K9 KMT mutants. Together, these observations confirmed that the spatial organization of heterochromatin is mostly dependent on Suv39h enzymes, although an additional inactivation of Eset (and probable defects in the nuclear lamina) in a Suv39h-deficient background is required for a partial breakdown of heterochromatin clustering, as previously reported 13 .
Full collapse of heterochromatin in 6KO H3K9 KMT MEF cells. We next analyzed heterochromatin organization in MEF cells lacking all 6 KMT. Strikingly, FISH staining for MSR sequences revealed a complete collapse of heterochromatin foci in both 6KO1 and 6KO2 mutants (Fig. 6a). While not fully penetrant, this phenotype was observed in the majority of nuclei of both 6KO1 (63%) and 6KO2 (52%) cells. As for the E1 S1S2 cells, H3K9me3 and HP1α were also completely dispersed (Supplementary Fig. 5A, B). Hmga1 remained enriched over DAPI-bright regions, even though these DAPI signals displayed an aberrant and non-clustered organization ( Supplementary Fig. 5C). Since Hmga1 was shown to also contribute to the integrity of heterochromatin foci 26 , its retention at the A/T-rich MSR-repeat sequences might explain, at least in part, the incomplete penetrance of heterochromatin dispersion in the 6KO MEF cells.
Staining for Lamin B recapitulated defects in the nuclear lamina/envelope, as observed in all H3K9 KMT mutants lacking Eset1 function, although the proportion of cells with micronuclei and the size of the micronuclei appeared increased in 6KO1 and 6KO2 cells (Fig. 6b). Fibrillarin staining indicated that the nucleoli structure was not significantly altered (Fig. 6c). Thus, complete loss of H3K9 methylation breaks heterochromatin clustering and reveals a definite function for H3K9 methylation in maintaining heterochromatin organization and genome integrity.
6KO H3K9 KMT MEF cells have lost electron-dense heterochromatin. We analyzed this collapse of heterochromatin at higher resolution by performing transmission electron microscopy (TEM) in the various H3K9 KMT mutants (Fig. 7). In WT (E1c) cells, heterochromatin is readily visualized as small clusters of electron-dense material, often localized close to the nuclear envelope (Fig. 7, white arrows), while larger electron-dense regions correspond to nucleoli. Eset1 (E1), Eset2 (E2), and Eset1/ Eset2 (E1E2) double mutants display a similar distribution of electron-dense heterochromatin as compared to wild type, with >80% of cells displaying clearly defined heterochromatin clusters. Compound H3K9 KMT mutants lacking Eset1 and Suv39h/Suv39h2 (E1 S1S2) and, to some extent, Eset1/Eset2, G9a/ Glp (E1E2 G9Gl) show reduced electron density or lower number of foci, but maintain the distinction between heterochromatin and euchromatin. Strikingly, this is not the case for 6KO1 and 6KO2 MEF cells: in half of these cells analyzed by TEM, no heterochromatin clusters could be detected, although nucleoli are preserved (Fig. 7) and a further fraction of the 6KO1 and 6KO2 cells displayed markedly reduced electron density and only very faint foci. Around 81% (6KO1, n = 26) or 78% (6KO2, n = 14) of the cells have lost the clear distinction of electron-dense heterochromatin. These data provide compelling evidence for the crucial requirement of H3K9 methylation in maintaining heterochromatin organization.  5 Dysregulation of H3K9me-positive vs H3K9me-negative genes in compound H3K9 KMT-mutant MEF cells. a Differential expression of genes with H3K9 methylated promoter regions. Violin plots representing the distribution of gene expression levels in WT and compound H3K9 KMT mutants, as determined by RNA sequencing (RNA-seq, log scale). Red dots indicate mean expression level. The genes are grouped according to H3K9 methylation status over their promoter region, as determined by chromatin immunoprecipitation sequencing (ChIP-seq) in WT MEFs. Promoters are defined as within + /−1 kilobase (kb) from the transcription start site (TSS). From left to right, genes with promoters lacking any H3K9 methylation peak (H3K9me0, 42626 genes), followed by genes with promoters positive for H3K9me1 (3179 genes), H3K9me2 (270 genes), or H3K9me3 (462 genes). It is important to note that this analysis does not include genes with promoter regions that show enrichment for multiple H3K9 methylation states (as summarized in Supplementary Fig. 4B). b Genome browser tracks showing H3K9me1, H3K9me2, or H3K9me3 enrichment (log2 fold change, red) over example gene promoters, as determined by ChIP-seq in WT MEFs. From left to right, Bardet-Biedl syndrome 9 (Bbs9, H3K9me1-positive), Triggering receptor expressed on myeloid cells 4 (Treml4, H3K9me2-positive) and Zinc finger protein 936 (Zfp936, H3K9me3-positive). Bottom shows RNA expression (RNA-seq, blue) in WT and compound H3K9 KMT mutants. Each browser track represents a 2 kb genomic interval surrounding the TSS. c Pie chart of genes classified by H3K9 methylation status at their promoter regions (left). Right, MA plots depicting changes in gene expression (RNA-seq) in 6KO1 cells relative to WT, shown for genes with (H3K9me + ) or without (H3K9me-) H3K9 promoter methylation. Black dots represent genes whose changes in expression are not statistically significant. Yellow dots represent upregulated genes and blue dots downregulated genes. The total number of genes upregulated in 6KO1 cells is indicated above the plots, and the number of genes downregulated is shown below. The numbers of H3K9me+ and H3K9me-up-or downregulated genes are indicated on the corresponding sector of the MA plots. For graphical representation, only genes with base mean >100 are shown on the plots.

Discussion
Heterochromatin forms distinct chromatin domains that can be visualized by DAPI staining of A/T-rich major satellite repeats in mouse cells. Suv39h enzymes are the central KMT for heterochromatic H3K9me3. However, while heterochromatin is weakened in Suv39h double-null cells, its overall organization and clustering remain largely intact. Many mechanisms could contribute to this heterochromatin stability independently of H3K9 methylation, including specific chromosome conformations, such as the Rabl configuration or anchoring to the nuclear membrane or other subnuclear structures 3,27,28 . For example, the threedimensional organization of chromosomes in S. pombe involves both H3K9 methylation, which controls chromatin compaction near centromeres, and cohesin-mediated formation of globular domains along chromosome arms 27 .
Another possibility is that other silencing systems can partially compensate for the loss of Suv39h enzyme function, such as Polycomb Repressive Complexes (PRC) and/or DNA methylation. We indeed observed a moderate increase in H3K27me3 levels in 6KO MEF cells chromatin, while global DNA methylation was only slightly decreased over the examined number of cell divisions. Recruitment of Polycomb-mediated H3K27me3 at heterochromatin has already been observed in other contexts where H3K9 methylation is reduced 5,29,30 .
The only minor decline of overall 5mC levels seems incoherent with the well-documented interdependence between H3K9 chromatin and CpG DNA methylation 31 . However, in lineagecommitted cells, reduction of H3K9 methylation does not cause a global loss of DNA methylation at heterochromatin, as, for example, Suv39h double-null MEF cells keep high levels of 5mC at many repeat elements 32 . Maintenance of DNA methylation by Dnmt1 is indeed not strictly dependent on recruitment to H3K9 methylated nucleosomes 33,34 . Further, CpG methylation is not restricted to heterochromatin regions, but also abundant on transcribed gene bodies, where it is established by H3K36me3mediated recruitment of Dnmt3b 35 . Determining whether and how these other silencing pathways are redirected or adjusted in the H3K9 KMT mutants would require genome-wide ChIP mapping of H3K27me3 and genome-wide bisulfite sequencing for 5mC.
We observed a high incidence of micronuclei and aberrant shape of the nuclear lamina/envelope in compound H3K9 KMTmutant cells that correlated with transcriptional signatures indicating the activation of DNA damage pathways. This is in agreement with the known roles of heterochromatin in the maintenance of genome integrity that include protecting cells from replication stress, chromosome segregation errors, or aberrant recombination of repeats, as it was reviewed recently 3 . Interestingly, these defects were most pronounced in H3K9 KMT mutants where Eset1 is inactivated and they also occurred in chromatin that maintains high levels of H3K9 methylation. These data suggest that in addition to the H3K9 histone, Eset1 has other non-histone substrates, some of which could function in the control of DNA damage and in correcting replication errors. Increased DNA damage was reported following Eset1 deletion in mouse oocytes or in intestinal epithelium 36,37 . The 6KO H3K9 KMT display a collapse of heterochromatin foci and loss of electron-dense heterochromatin. This reveals an essential function for H3K9 methylation in maintaining heterochromatin as a separate nuclear domain. Our data also show a larger degree of cooperation among the 6 SET-domain H3K9 KMT to secure heterochromatin integrity. Conversely, this seems to exclude a contribution to H3K9 methylation and heterochromatin organization of other KMT systems, such as split SETdomain enzymes or "seven-beta-strand" (7βs) KMT 38 . The described dispersion of heterochromatin in Prdm3/Prdm16 shRNA MEF mutants 13 is probably related to concurrent secondary defects in the structuring of the nuclear lamina and not caused by a reduction of H3K9 methylation in nuclear chromatin.
H3K9 KMT mutants of any distinct gene pair of either Suv39h1/Suv39h2, Eset1/Eset2, or G9a/Glp maintain heterochromatin structure, indicating that functional redundancy between these SET-domain H3K9 KMT ensures the robustness of heterochromatin organization. Importantly, the contribution of each of these six SET-domain H3K9 KMT may be different in other cell types. For example, Suv39h1-null T cells start to disintegrate heterochromatic foci, even though the other 5 H3K9 KMT remain functional 39 . Similarly, Eset1/Suv39h compound mutant hepatocytes also have severely reduced EM-dense heterochromatin 40 .
What is the mechanism for the collapse of heterochromatin in the compound H3K9 KMT mutants? Chromatin that is deficient for all H3K9 methylation states will not provide binding for HP1 proteins and may have lost the physicochemical properties to engage in phase separation 3,41,42 . While the dispersion of HP1 binding is not sufficient to break heterochromatin 11 , other H3K9me2 or H3K9me1-binding proteins might substitute for HP1 in Suv39h-deficient cells, but not in compound H3K9 KMTmutant cells that lack all H3K9 methylation states. Possibly, some of these putative H3K9-methyl binders could also assist in phase separation, and thus increase heterochromatin stability. Alternatively, there could also be non-histone substrates for the distinct SET-domain H3K9 KMT, particularly for Eset, that are needed for heterochromatin organization but fail to become methylated in the 6KO MEF cells.
Several proteins were identified as candidate anchoring factors for heterochromatin. In mouse cells, Lamin B receptor (LBR) or Lamin A/C tether heterochromatin to the inner nuclear membrane in a cell-type-specific manner, but the interaction is likely indirect and adaptor proteins remain to be identified 43 . In C. elegans, the chromodomain containing protein CEC-4 binds methylated H3K9 and is required for recruitment of heterochromatin at the nuclear periphery, but dispensable for its silencing 44 . While no clear homolog of CEC-4 is found in mammals, other chromodomain proteins might function in a similar way.
Answering these questions will require future studies and would be facilitated if conditions for compound H3K9 KMTmutant cells can be generated that allow the analysis of collapsed heterochromatin not only at its onset (this study) but also for more prolonged cell divisions. In summary, our work provides compelling evidence for a crucial function of H3K9 methylation in heterochromatin organization.

Methods
Eset1-conditional MEF. Eset1-conditional MEFs were described previously 6 . To optimize deletion efficiency, we sorted single cells in 96-well plates and selected for cell clones with high Cre activity (clone Eset25, referred to as E1c). After CRISPR/ Cas9-mediated gene disruption (see below), we further re-introduced a construct expressing tamoxifen-inducible Cre together with a puromycin resistance gene hydroxytamoxifen (4-OHT, Sigma Aldrich, H7904) for 2 days, followed by 2 days in normal medium, to minimize potential side effects of prolonged growth in presence of tamoxifen. Deletion efficiency was assessed by RT-qPCR with primers specific for the deleted exons, or by western blot for the Eset1 protein. We refer to E1c as WT.
CRISPR/Cas9 mutations. CRISPR/Cas9-mediated gene disruptions were performed by transfection of chimeric sgRNA/Cas9 expression vectors 45 . For each SET-domain KMT target gene, two unique guide RNAs were designed using online tools (CRISPR.mit.edu) and subsequently cloned in the chimeric sgRNA/Cas9 expression vectors pSpCas9(BB)-2A-Puro (pX459) or pSpCas9(BB) (pX330), both from Addgene. Cells were transfected with Xfect (Clontech) and selected for 2 days in puromycin-containing media. After selection, the efficiency of Cas9-mediated cleavage was evaluated by Surveyor assay 45 . Puromycin-resistant cells were further sorted to single cells in 96-well plates, and individual clones analyzed for the absence of functional protein by western blotting when specific antibodies were available, or for the absence of a functional copy of the gene by sequencing the relevant exons (for Eset2). The sequences of the guide RNAs used in this study are listed in Supplementary Table 1.
Gene pair mutations. For gene pair mutants, we transfected combinations of chimeric sgRNA/Cas9 expression vectors in E1c MEFs and selected mutant clones as described above. Conditional Eset1, Eset2-null mutants (E1cE2) were obtained by introducing a single guide RNA specific for Eset2 into E1c MEFs. Clone A4 was selected after sequencing revealed a frameshift mutation within exon 7 of Eset2. G9a/Glp double mutants were generated by co-transfection of two sgRNA/Cas9 vectors targeting G9a and Glp. Clone H7 was selected after western blot analysis confirmed the absence of G9a or Glp proteins. Suv39h-deficient cells were generated by co-transfection of two sgRNA/Cas9 vectors targeting Suv39h1 and Suv39h2 in E1c MEFs. Clone B1 was selected after western blot analysis confirmed the absence of Suv39h1 or Suv39h2 proteins.
Compound SET-domain H3K9 KMT mutants. Compound H3K9 KMT mutants were generated by successive rounds of sgRNA/Cas9 transfection and selection of relevant clones. The nomenclature of the clones refers to null mutation of SETdomain KMT genes and is indicated as E1 (Eset1-null), E2 (Eset2-null), G9 (G9anull), Gl (Glp-null,) S1 (Suv39h1-null), and S2 (Suv39h2-null). Series 1 clones were generated from Eset1-conditional MEF cell, after the first round of selection for high Cre recombinase activity resulting in the isolation of the Eset25 clone, referred to as E1c (see above). E1c cells were transfected with two sgRNA/Cas9 vectors targeting Suv39h1 and Suv39h2 and clone B1 (E1c S1S2) was selected as described above. This clone was further co-transfected with vectors targeting Eset2, G9a and Glp, and clone B1A6 (E1cE2 S1S2 G9Gl) was selected after western blotting (G9a, Glp) and exon sequencing (Eset2).
Histone extraction and nanoLC-MS analysis. After acid extraction of histones, histones samples were separated on 16% SDS Tris Glycine Wedge gels (Thermo Fisher) followed by colloidal coomassie staining (Instant Blue, Expedeon). Histone bands were excised and histone propionylation and subsequent protein digestion using trypsin (Promega) were performed as described 47 , with minor modifications. In-gel propionylation prior to trypsin digestion was conducted two times employing the method "H-42x" 47 followed by the hydroxylamine-mediated reversal of overpropionylation. Peptide samples were desalted using STAGE tips 48 and analyzed using LC-MS as previously described 49 with minor modifications.
Samples were separated within 60 min with a linear gradient from 2% buffer B (80% acetonitril in 0.1% formic acid) to 60% buffer B at a flow rate of 250 nl/min, and this was followed by a washout step (80% buffer B, 5 min, 500 nl/min) and reequilibration (2% buffer B, 5 min, 500 nl/min). Mass spectrometry was carried out using the "sensitive method" (data-dependent acquisition) on Orbitrap QExactive Plus (Thermo Fisher Scientific) 50 .
Mass spectrometry raw data were analyzed using Peaks Studio version 8.5 (Bioinformatics Solution Inc.). MS raw files were searched using the Peaks de novo search algorithm and Peaks DB search against Mus musculus (mouse) UniprotKB protein sequence database (December 2015). Trypsin was selected as a cleaving enzyme, while a maximum of four missed cleavages, a minimum of six amino acids per peptide, 10 ppm MS peptide ion mass tolerance, and 0.2 Da MS/MS tolerance were allowed. Carbamidomethylation of cysteine was set as a fixed modification, whereas the following modifications were set as variable modifications: acetylation at the protein N terminus and at lysine residues, deamidation of asparagine or glutamine, oxidation of methionine, propionylation of lysine and monomethylated lysine residues, mono-methylation of lysine and arginine residues, dimethylation of lysine and arginine residues, trimethylation of lysine residues, and phosphorylation of serine, threonine, and tyrosine. Maximum (allowed) variable PTMs per peptide were set to 4. The peptide and protein false discovery rate (FDR) was set to a maximum of 0.01 using the decoy-fusion database search strategy implemented in Peaks Studio.
Histone peptide intensities were exported and the sum of all peptide intensities (or features), belonging to each particular histone PTM (hPTM) mark, was calculated for histone H3 and histone H4 in each individual biological sample. This was followed by an inter-sample normalization to adjust the total intensity of all hPTMs to the same level in each biological replicate in each condition. To achieve this normalization, the summed peptide intensity of each particular hPTM was divided by the total intensity of a non-modified histone peptide, a histone peptide that is consistently identified and bears no hPTM mark throughout the whole experiment. Peptide "YRPGTVALR" and peptide "VFLENVIR" were used for normalizing mouse histone H3 (sequence: [41][42][43][44][45][46][47][48][49] and mouse histone H4 (sequence: 60-67), respectively.
Data were plotted using GraphPad Prism version 8.3.0.
DNA methylation analysis. Genomic DNA was purified by phenol/chloroform extraction and ethanol precipitation according to standard procedures. DNA dot blot for 5mC was performed according to standard procedures 51 55 . For analysis of repeat element expression, library size factors as computed from the gene expression data were used rather than recomputing these using the expression metrics of the repeat elements themselves. BigWig files using in visualization tracks were constructed using deepTools 56 . For pathway analyses, normalized counts as computed by DESeq2, as well as fold changes and adjusted P values, were loaded into Ingenuity Pathway Analysis and the results subsequently plotted in R. For visualization of coverage over repeat regions (Fig. 3c) reads where aligned to the repeat sequence extracted from Repbase 57 using bwa mem 58 version 0.7.17 with default settings, bigWig tracks were constructed using deepTools and IGV 59 .
ATAC sequencing. For ATAC library preparation 60 , 50,000 cells were lysed in cold lysis buffer (10 mM Tris-HCl pH 7.4, 10 mM NaCl, 3 mM MgCl 2 , 0.1% v/v Igepal CA-630) and centrifuged at 4°C for 10 min at 500g. The pellet was resuspended in a transposition reaction mixture provided by the Nextera DNA Library Prep kit (Illumina, 15028212) and incubated at 37°C for 30 min. After transposition, the samples were purified with a Qiagen MinElute PCR purification kit (28604) according to the manufacturer's instructions. For library preparation, each sample was subjected to PCR amplification with a pair of barcoded primers provided by the Nextera Index kit (Illumina 15055289). The reaction was monitored by qPCR and further cycles were performed when necessary, based on the amplification plot. Paired-end, 75 bp reads were generated with Illumina HiSeq3000 sequencer with a depth of 50 million reads per sample. Two biological replicates were sequenced per cell line.
ATAC-seq analysis. Unprocessed reads were first quality and adapter trimmed using fastp 61 version 0.19.4, using a mean window quality threshold of 3 and excluding low-complexity sequences. These were then aligned to the GRCm38 genome from Ensembl using STAR 53 version 2.6.0c allowing no spliced alignments and with other options as described by TEtranscripts 54 . For extraction of the openchromatin subset of reads, deepTools 56 version 3.1.2 was used with a fragment length threshold of 150 bases. Alignments were subsequently quantified across repeats using TEtranscripts version 2.0.3 and the repeatmasker file from UCSC as the source for repeat locations. For visualization, bigWig tracks were constructed using deepTools and IGV 59 . Differential accessibility was determined using DESeq2 55 version 1.20.0. Counts across gene bodies for genes showing unchanged expression in the RNA-seq data were used to determine sample-specific size factors, as other normalization methods would wash away global accessibility shifts. Pathway analysis was performed as with the RNA-seq data.
Chromatin immunoprecipitation sequencing (ChIP-seq). Ten million cells were trypsinized, washed with PBS, and cross-linked in 1% formaldehyde for 10 min at room temperature. After quenching with glycine (0.125 M), cells were pelleted, washed with cold PBS, and lysed in 1% SDS, 10 mM EDTA, 50 mM Tris-HCl, pH 8, and protease inhibitors (Roche). Lysates were sonicated (Covaris S220) to fragment sizes of 200-500 bp. For each ChIP, chromatin corresponding to 10 μg of DNA was diluted in 1% Triton X-100, 2 mM EDTA, 167 mM NaCl, 20 mM Tris-HCl, pH 8, and protease inhibitors, and incubated overnight at 4°C with 4 μg of antibody against H3K9me1 (Abcam, ab8896), H3K9me2 (Abcam, ab1220) or H3K9me3 (Abcam, ab8898), followed by 4 h at 4°C with 25 μl of Protein G dynabeads (Invitrogen). Beads were washed three times in low salt (150 mM NaCl) and once in high salt (500 mM NaCl) before elution in 1% SDS, 50 mM NaHCO 3 , RNAse A and Proteinase K treatment, and DNA purification with Qiagen MinElute PCR purification kit (28604). Enrichment was validated by qPCR for control loci ( Supplementary Fig. 3A: Major satellite repeats, L1MdA 5'UTR, IAP-EY3, Zinc finger protein 180 and Oct4 promoter as positive controls, as well as Actin B promoter as a negative control). Primer sequences are listed in Supplementary Table 1. Libraries were prepared with NEBNext Ultra II DNA Library Prep Kit for Illumina (NEB, E7645L). Paired-end, 100 bp reads were generated with Illumina NovaSeq 6000 with a depth of 100 million reads per sample. Two biological replicates were sequenced per ChIP as well as for input material.
ChiP-seq analysis. The quality of raw Fastq reads was checked using FastQC (http:// www.bioinformatics.babraham.ac.uk/projects/fastqc). Reads were aligned to mm10 (GRCm38) with bwa version 0.7.1. 58 . Peaks were called using histoneHMM 62 . Peaks were annotated using ChiPseeker software 63 version 1.18.0. Peaks that are located within + /− 1 kb of the TSS were extracted for downstream analysis. Heatmaps and profiles were generated using deeptools version 3.5.0 56 . Marks were aligned around the center of the peak (computeMatrix reference-point). Flanking regions were added to show the signal extending outside enriched regions. ComputeMatrix was executed with the following parameters:-skipZeros-binSize 10 -a 10000 -b 10000-refer-encePoint center. Heatmaps were generated using plotHeatmap. Coverage and normalization to the input were obtained using deepTools-3.5.0 bamCompare. FRiPs scores were obtained using deepTools/plotEnrichment. Violin plots represent the distribution of gene expression levels for each H3K9 methylation state. Normalized values of RNA-seq expression were obtained using DESeq2 version 1.26.0 55 on the count matrices output from snakePipes version 2.1.0 64 . Red dots represent the mean expression level. Data visualization was performed using ggplot2 65 version 3.3.2. P values comparing the means were calculated using the t test.
Transmission electron microscopy (TEM) sample preparation and imaging. Cells grown on Thermanox coverslips (Nunc) were prepared as described 66 and adapted for cell monolayers. Cells were fixed with 2.5% glutaraldehyde and 2% paraformaldehyde (Electron Microscopy Sciences EMS, 16400 and 15700) in 0.1 M Hepes buffer pH 7.4 for 1 h at room temperature and then overnight at 4°C. After washes in cold 0.15 M cacodylate buffer pH 7.4 (Sigma Aldrich, CO250) the cells were post-fixed in 1.5% potassium ferrocyanide (Sigma Aldrich, 60280) and 2% osmium tetroxide (EMS, 19110) in 0.15 M cacodylate buffer pH 7.4 on ice for 30 min. The cells were then washed with double-distilled water (ddH2O) and immersed in filtered 1% thiocarbohydrazide (TCH, EMS, 21900) solution for 10 min. After this step, the cells were stained in 2% osmium tetroxide for 30 min, rinsed again in ddH 2 O, and stained with 1% uranyl acetate at 4°C overnight. After washes with ddH 2 O, cells were stained with Walton's lead aspartate (Sigma A9256 and EMS 17900) solution at 60°C for 30 min and dehydrated in a graded alcohol series. After flat-embedding in Epoxy resin (SERVA Electrophoresis GmbH 21045) samples were cured at 60°C for 12 h. Coverslips were detached from the resin using repetitive freeze−thaw in liquid nitrogen. A region of interest containing cells was selected (~1 × 1 mm), cut, and mounted on an aluminum pin. For transmission electron microscopy (TEM), serial thin sections (60 nm) were cut using an ultramicrotome (Ultracut 7, Leica, Vienna) and imaged using a Spirit 120 kV (FEI, now THF, Eindhoven) at magnifications between 1700× and 8200× equipped with a side-mounted camera (Veleta, EMSIS Gmbh).

Data availability
The proteomics datasets generated in this study are available at ProteomeXchange via the PRIDE partner repository with record number PXD018175. The UniProtKB database is accessible at https://ftp.uniprot.org/pub/databases/uniprot/previous_major_releases/ release-2015_12/knowledgebase/. A database filtered for mouse peptides is available as a FASTA file at ProteomeXchange, together with the datasets generated in this study. The RNA-seq, ATAC-seq, and ChIP-seq datasets generated in this study are available in the GEO repository with record number GSE142105. The Ingenuity Pathway Analysis (IPA) database is accessible at https://digitalinsights.qiagen.com/products-overview/discoveryinsights-portfolio/analysis-and-visualization/qiagen-ipa/. The data and any unique materials are available from the corresponding author upon reasonable request. Source data are provided with this paper.