Genomic and Proteomic Resolution of Heterochromatin and its Restriction of Alternate Fate Genes

SUMMARY

Heterochromatin is integral to cell identity maintenance by impeding the activation of genes for alternate cell fates. Heterochromatic regions are associated with histone 3 lysine 9 trimethylation (H3K9me3) or H3K27me3, but these modifications are also found in euchromatic regions that permit transcription. We discovered that resistance to sonication is a reliable indicator of the heterochromatin state, and we developed a biophysical method (Gradient-seq) to discriminate subtypes of H3K9me3 and H3K27me3 domains in sonication-resistant heterochromatin (srHC) versus euchromatin. These classifications are more accurate than the histone marks alone in predicting transcriptional silence and resistance of alternate-fate genes to activation during direct cell conversion. Our proteomics of H3K9me3-marked srHC and functional screens revealed diverse proteins, including RBMX and RBMXL1, that impede gene induction during cellular reprogramming. Isolation of srHC with Gradient-seq provides a genome-wide map of chromatin structure, elucidating subtypes of repressed domains that are uniquely predictive of diverse other chromatin properties.

INTRODUCTION

To maintain cell differentiation, cells must activate genes specific to their lineage while simultaneously repressing genes of alternative lineages. Among regions of chromatin that are transcriptionally silent, heterochromatin is distinguished by its tight physical compaction (Fussner et al., 2011; Gilbert et al., 2004; Wallrath and Elgin, 1995) and its ability to maintain gene repression despite the presence of transcriptional activators (Feldman et al., 2006; Epsztejn-Litman et al., 2008). Mammalian heterochromatin suppresses recombination among repeat-rich sequences, silences expression of cell type-inappropriate genes, and impedes changes in cell identity during development and reprogramming (Becker et al., 2016; Beisel and Paro, 2011; Hawkins et al., 2010; Peters et al., 2001). Heterochromatic regions are typically marked by histone 3 lysine 9 trimethylation (H3K9me3) or lysine 27 trimethylation (H3K27me3), yet these same marks can be found in transcribed gene bodies (Blahnik et al., 2011; Riddle et al., 2012; Vakoc et al., 2005) and at accessible gene promoters (Breiling et al., 2001; Dellino et al., 2004). At present, no method exists for the definitive mapping of heterochromatic regions across the genome, and the subsets of H3K9me3- and H3K27me3-marked chromatin that most strongly impede changes in cell identity remain unclear.

We previously described megabase-scale, H3K9me3-enriched domains that are resistant to the initial binding of the pluripotency transcription factors Oct4, Sox2, Klf4, and cMyc in human fibroblasts, but where the factors can bind in human embryonic stem (ES) cells (Soufi et al., 2012). These Differentially Bound Regions (DBRs) contain pluripotency genes whose activation during iPS reprogramming is restricted to rare cells late in the reprogramming process (Buganim et al., 2012). Reducing H3K9me3 levels improves iPS reprogramming efficiency (Onder et al., 2012; Soufi et al., 2012) as well as reprogramming by somatic cell nuclear transfer (Matoba et al., 2014). Several regulators of mammalian heterochromatin—the enzymes SUV39H1/SUV39H2 and SETDB1 that catalyze H3K9me2 and H3K9me3, G9a and GLP that catalyze only H3K9me2, and the three isoforms Heterochromatin Protein 1 (HP1) that bind H3K9me3—have been shown to impede reprogramming to pluripotency (Becker et al., 2016; Onder et al., 2012; Sridharan et al., 2013). Yet in addition to their decreased abundance in pluripotent cells, large H3K9me3 domains also show dynamics when compared across differentiated lineages (Becker et al., 2016; Soufi et al., 2012). An open question, then, is whether heterochromatin also impedes direct cell conversion between mature lineages, similar to pluripotency reprogramming.

H3K27me3, elicited by the Polycomb Repressive Complex 2 (PRC2), marks the inactive X chromosome and promoters for developmental transcription factors (Beisel and Paro, 2011; Hawkins et al., 2010; Trojer and Reinberg, 2007). However, PRC2 components do not impede iPS reprogramming (Onder et al., 2012), and H3K27me3-marked sites can be accessible to general transcription factors and RNA polymerase (Breiling et al., 2001; Dellino et al., 2004), leaving the literature conflicted on whether the term “heterochromatin” includes H3K27me3-marked regions (Beisel and Paro, 2011; Trojer and Reinberg, 2007).

To discover proteins embedded in H3K9me3-marked heterochromatin, mass spectrometry (MS) has been performed after H3K9me3-directed chromatin immunoprecipitation (ChIP) (Engelen et al., 2015; Ji et al., 2015; Soldi and Bonaldi, 2013) or pulldown with an H3K9me3 bait (Eberl et al., 2013; Vermeulen et al., 2010). However, the strong signal for H3K9me3 in certain transcribed regions, as described for Drosophila chromosome 4q and human zinc finger (ZNF) clusters on chromosome 19 (Blahnik et al., 2011; Riddle et al., 2012; Vakoc et al., 2005), indicates that such datasets are contaminated with euchromatin. Although RNA-binding proteins appear abundant after H3K9me3 ChIP (Soldi and Bonaldi, 2013) or among readers of the H3K9me3 mark (Vermeulen et al., 2010), the proteins could be associated with the repressed or active forms of H3K9me3-marked chromatin.

We found that H3K9me3 domains, more than H3K27me3 domains, impede the activation of liver genes during direct reprogramming of fibroblasts to hepatic cells, thus emphasizing the generality of H3K9me3 domains as being most restrictive for cell fate control. We found that H3K9me3 domains can be resistant to sonication, causing their under-representation in ChIP-seq samples and allowing us to isolate sonication-resistant heterochromatin (srHC) fragments on sucrose gradients, a method we term “Gradient-seq.” Gradient-seq mapping of srHC predicted which subsets of H3K9me3 and H3K27me3 domains are transcriptionally silent and reprogramming-resistant, and gradient sedimentation enabled a proteomic analysis of the repressed, heterochromatic form of H3K9me3 domains. Depletion of numerous identified H3K9me3 heterochromatin proteins enhanced gene activation during direct cell reprogramming. Overall, srHC mapping provides a distinctly structural means to characterize the genome, independent of histone modifications and other surrogates, and is highly predictive of diverse other chromatin features as well as gene dysregulation in disease.

RESULTS

H3K9me3 domains impede direct reprogramming

Human fibroblasts can be reprogrammed to human induced hepatocytes (hiHeps) by the ectopic expression of liver transcription factors (Huang et al., 2014). The resulting hiHep cells express hepatic markers and recapitulate some features of liver metabolism, but expression profiling reveals a failure to fully express many hepatocyte genes. We analyzed microarray data comparing hiHeps to fibroblasts and cultured hepatocytes (Huang et al., 2014), classifying genes according to histone modification ChIP-seq data (Bernstein et al., 2010). We found that, among hepatic genes that are silent in fibroblasts, genes marked by H3K9me3 in fibroblasts have a profound failure to activate during cell conversion to hiHeps, with the median induction barely above 0% (Figure 1A). Such H3K9me3-marked genes encode the bile acid receptor FXR/NR1H4, FOXA2, metabolic enzymes, cytochrome P450s, secreted plasma proteins, and adhesion proteins (Table S1). Silent genes marked by H3K27me3 show only a modest defect in activation, compared to genes lacking either mark (Figure 1A). We conclude that H3K9me3-marked chromatin is the most refractory to gene activation by hepatic factors, similar to such domains impeding reprogramming to pluripotency (Matoba et al., 2014; Soufi et al., 2012), leading us to investigate the properties and regulation of H3K9me3-marked heterochromatin.

Figure 1. H3K9me3-marked heterochromatin impedes direct reprogramming and is resistant to sonication.

(A) Violin plots showing levels of gene activation during hiHep reprogramming. Microarray data (Huang et al., 2014) was filtered for genes silent in fibroblasts and expressed in cultured hepatocytes. These genes were classified by whether they overlapped at least 50% with fibroblast H3K9me3 or H3K27me3 domains (Table S1; ChIP-seq from GSE16368). RNA levels in hiHep cells (n=3) were plotted on a relative scale ranging from fibroblast levels (0%) to hepatocyte levels (100%), using log2-transformed values. White circles indicate median values. P values by Wilcoxon rank sum test, compared to unmarked genes.

(B) Differentially Bound Regions (DBRs, black bars) (Soufi et al., 2012) compared to H3K9me3 ChIP-seq in IMR90 fibroblasts (GSM469974) with and without input-subtraction (GSM521926). Red arrows: depletion in input signal.

(C) Agarose gel of purified DNA isolated from sucrose gradient, after loading with crosslinked, sonicated chromatin from BJ fibroblasts. Boxes indicate fractions pooled in later analyses.

(D) qPCR on sucrose gradient fractions (as in C), with equal DNA loading per PCR, using validated qPCR sites in DBR and non-DBR chromatin (Soufi et al., 2012) (see text and STAR Methods). Error bars: SEM, two biological replicates.

(E) Diagram of samples used for DNA sequencing.

(F) Browser view of sequencing signal for srHC and euchromatin (“euchr.”) fractions compared to histone mark ChIP-seq and mRNA-seq. “srHC+H3K9me3” denotes H3K9me3 IP performed from srHC fraction. H3K27me3 data from GSE16368 (n=3); all other data generated in this study (n=2 per track, average shown). Horizontal bars above each track indicate domains enriched in that sample. Arrows: srHC domains marked by H3K9me3 alone (black), H3K27me3 alone (green), or both marks (red). See also Figure S1.

Heterochromatic regions are resistant to sonication

Using ChIP-seq data from the Epigenomics Roadmap (Bernstein et al., 2010), we find that the input-normalized H3K9me3 signal (Figure 1B, purple track) is strongly enriched over DBRs, which are domains refractory to binding by iPS reprogramming factors in fibroblasts (Soufi et al., 2012). However, without normalizing to input, the H3K9me3 signal is only modestly elevated in these regions (Figure 1B, yellow track). Surprisingly, the input signal alone shows depletion over most H3K9me3 domains, including DBRs (Figure 1B, blue track, red arrows). Sequencing of both input and ChIP is typically performed after size-selecting short DNA fragments, thereby depleting regions that are more resistant to sonication. Sites of high sonication susceptibility map to active promoters, which is the basis of the techniques Sono-seq (Auerbach et al., 2009) and FAIRE (Torres et al., 2016), but resistance to sonication has not previously been used to identify heterochromatic regions.

We prepared crosslinked, sonicated chromatin from human BJ foreskin fibroblasts and used sucrose gradient ultracentrifugation (Gilbert et al., 2004) to sediment the larger, sonication-resistant fragments (Figure 1C). By qPCR, we found that sites inside DBRs (Soufi et al., 2012) (Figure 1D, red line) are markedly enriched in fractions from the middle of the gradient, which contain longer DNA fragments normally excluded from ChIP-seq (Figure 1C, red box) and some smaller fragments that co-sediment due to crosslinking or compaction. Control sites outside DBRs, shown to lack H3K9me3 and allow iPS factor binding in fibroblasts (Soufi et al., 2012), are depleted from the sonication-resistant material in the middle of the gradient (Figure 1D, blue line). The enrichment for DBR over non-DBR sites in middle gradient fractions is comparable to that achieved by conventional H3K9me3 ChIP-qPCR (Figure S1A). Thus, gradient sedimentation of sonication-resistant chromatin allows biophysical enrichment of heterochromatic regions without reliance on histone modifications or antibodies.

Gradient fractions with the highest enrichment of DBR sites (red bracket in Figure 1D) were pooled and termed the “sonication-resistant heterochromatin (srHC) fraction.” Fraction #2 at the top of the gradient, termed the “euchromatin fraction,” was used for comparison. Secondary H3K9me3 ChIP, performed from the srHC chromatin, resulted in additive enrichment of DBR sites (Figure S1A) and a much higher rate of chromatin recovery, compared to IP from bulk chromatin or the euchromatin fraction (Figure S1B), confirming that the gradient step enriches for H3K9me3-marked chromatin.

Gradient-seq reveals the landscape of sonication-resistant heterochromatin

After further shearing the isolated srHC DNA (Figure S1C), we sequenced DNA from the srHC and euchromatin gradient fractions, a method we call Gradient-seq. We also performed conventional input-normalized H3K9me3 ChIP-seq, H3K9me3 IPs from both gradient fractions (Figure 1E), and mRNA-seq in BJ fibroblast cells. Of the 258 DBRs (Soufi et al., 2012), 256 have higher sequencing signal in the srHC fraction than in input chromatin, in contrast to regions flanking DBRs (Figure S1D). The srHC sequencing signal forms megabase-scale domains that correspond closely to H3K9me3 ChIP-seq domains, including DBRs, while the euchromatin fraction follows an inverse pattern overlapping the mRNA-seq signal (Figure 1F, S1E). Many H3K27me3-marked regions are enriched in the euchromatin fraction, while others have strong srHC signal even in the absence of H3K9me3 (Figure 1F, green arrows). Genome-wide, the srHC sequencing strongly correlates with H3K9me3 ChIP-seq from both this study and the Epigenomics Roadmap, correlates weakly with H3K27me3, and anti-correlates with the euchromatin fraction (Figure 2A).

Figure 2. Gradient-seq maps the H3K9me3- and H3K27me3-marked forms of repressive heterochromatin.

(A) Spearman correlation and unsupervised clustering of ChIP- and Gradient-seq datasets, using input-normalized tag density per 10-kb sliding window. Data from the Epigenomics Roadmap is labeled with “(R)”.

(B) ChIP-seq levels for fibroblast histone marks (Bernstein et al., 2010; Chandra et al., 2012), normalized for input, plotted over the width of srHC domains (average of 28,807 domains weighted by domain length). An additional 15 acetyl marks depleted in srHC (see STAR Methods) are not shown.

(C) Overlap of H3K9me3 and H3K27me3 domains with chromatin categories defined by Gradient-seq.

(D) Fibroblast mRNA-seq tag counts, normalized by DESeq2 and divided by gene length, for genes in each chromatin category (whiskers: 5^th and 95^th percentiles).

(E) Top 15 non-redundant Gene Ontology categories of Refseq genes inside srHC domains (Table S2 for full list). FDR by Benjamini-Hochberg.

(F) Frequency with which each CpG is methylated, by whole-genome bisulfite sequencing (GSM1127120). Whiskers: 10^th and 90^th percentiles among CpGs. The thousands (“K”) or millions (“M”) of CpGs per category are indicated. P values by Wilcoxon rank sum test.

We used a 10-kb sliding window algorithm (see STAR Methods) to call genomic regions enriched in the srHC fraction versus input; these “srHC domains” (horizontal red bars in Figure 1F, S1E) cover 997 MB of the human genome and have weighted average length 135 kb. “Euchromatin domains” enriched in the euchromatin fraction compared to the srHC fraction total 1240 MB. Remaining regions with similar abundance in the srHC and euchromatin fractions were called “intermediate domains” (all domain coordinates are provided in Table S1). All three domain classes have similar GC content (Figure S1F). Of 28 histone marks profiled in human fibroblasts (Bernstein et al., 2010; Chandra et al., 2012), H3K9me3 is most strongly enriched over srHC domains, while H3K27me3 and H3K9me2 have modest enrichment (Figure 2B); input-paired data for H4K20me3 in fibroblasts was not available for analysis. All other marks tested, including those associated with active promoters/enhancers or transcriptional elongation, are strongly depleted (Figure 2B). Consistently, 607 MB of the srHC domains (60.8%) are also called as H3K9me3 domains, while 327 MB (32.8%) are H3K27me3 domains (Figure 2C), with a low rate of overlap between the two marks (7.0%, 70 MB), as seen elsewhere (Chandra et al., 2012; Hawkins et al., 2010).

Genes falling within srHC domains (Table S2) are strongly transcriptionally repressed compared to genes in euchromatin domains, with the intermediate domains nearly as repressed as srHC (Figure 2D). Gene classes enriched in srHC domains relate to mature, non-fibroblast lineages (neurons, immune cells, epithelial cells) (Figure 2E, Table S2). As expected, srHC domains contain the majority of telomeric and satellite repeat sequences (82% and 68%, respectively) and are enriched for diverse LINE elements and endogenous retroviruses (Table S2). Whereas euchromatin domains are highly demethylated at CpG islands, srHC domains have high rates of methylation inside and outside of CpG islands (Figure 2F). These findings support the heterochromatic nature of srHC domains.

Euchromatic subtypes of repressive histone mark domains

Gradient-seq revealed 31 MB marked by H3K9me3 (3.2% of the total H3K9me3) and 193 MB (28.7%) marked by H3K27me3 that, surprisingly, were enriched in the euchromatin fraction (Figure 2C, Table S3 and S4). Genes in these euchromatic subtypes of H3K9me3 and H3K27me3 domains are consistently transcribed (Figure 3A), validating their classification as euchromatic. Furthermore, euchromatic H3K9me3 domains are enriched for the transcriptional elongation mark H3K36me3 (Figure S2A).

Figure 3. A subset of H3K9me3 and H3K27me3 domains are structurally euchromatic and permissive to transcription.

(A) Boxplots show mRNA-seq tag counts (normalized by DESeq2, divided by gene length) for genes inside each kind of domain (whiskers: 5^th and 95^th percentiles). Labels: “euchr” for euchromatin, “int” for intermediate. Parentheses denote number of genes per category. P values by Wilcoxon rank sum test.

(B) Browser view of euchromatic H3K9me3 domains over expressed ZNF gene family cluster, which is depleted from srHC. The Gradient-seq track is shown as the difference in sequencing signal between the srHC and euchromatin fractions. The “euchr + K9 IP” track shows the H3K9me3 ChIP from the euchromatic (“euchr”) fraction of the gradient.

(C) Non-redundant gene categories (InterPro database) significantly enriched for the euchromatic subtypes of H3K9me3 and H3K27me3 domains. FDR by Benjamini-Hochberg. See also Table S3, S4.

(D) Genomic sites reported to lose H3K9me3 after depletion of the HUSH complex/SETDB1 (Tchasovnikarova et al., 2015).

(E) Browser view of euchromatic H3K27me3 (“K27”) domain over HOXA gene cluster.

(F) Fraction of satellite repeat types overlapping with srHC versus euchromatic H3K9me3 domains. Asterisks indicate significant enrichment (FDR < 0.05) in srHC (red) or euchromatic H3K9me3 (blue) based on 1000 simulations of randomly shuffled domains. See also Figure S2.

The euchromatic form of H3K9me3 domains is remarkably selective for a few gene families, notably genes encoding KRAB domain-containing zinc fingers (KRAB-ZNFs) (Figure 3B-C), which are expressed bi-allelically (Blahnik et al., 2011) and have H3K9me3 enriched over gene bodies and depleted over promoters (Figure S2B). The euchromatin sequencing signal at KRAB-ZNF genes is further increased by H3K9me3 IP performed off of the euchromatin fraction (Figure 3B, “euchr. + K9 IP” track), confirming that the same chromatin fragments, from the same cells, are exhibiting both sonication sensitivity and presence of H3K9me3. Euchromatic H3K9me3 domains also contain a majority of sites regulated by the HUSH complex (Tchasovnikarova et al., 2015), an enrichment of 39-fold compared to the srHC subtype (Figure 3D). Indeed, the sites reported to have the greatest dependence on HUSH/SETDB1 for H3K9me3 (Tchasovnikarova et al., 2015) are in euchromatin (Figure S2C). Meanwhile, euchromatic H3K27me3 domains are enriched for transcription factor families (Figure 3C), including all four human HOX gene clusters (e.g., Figure 3E), and are transcriptionally permissive (Figure 3A, E). Thus, Gradient-seq reveals euchromatic gene classes and targets of chromatin complexes that are clearly distinct from transcriptionally silent heterochromatin, despite bearing the same histone modifications.

While most satellite repeats are within with srHC domains, the silent repeat HSATII is almost entirely within euchromatic H3K9me3 domains (Figure 3F), which we corroborate by finding it to be DNase I-sensitive and depleted for Lamin B1 (Dou et al., 2015) (Figure S2D). The euchromatic structure of HSATII could explain why it is the most overexpressed repeat in diverse human cancers (Ting et al., 2011). Cancers also have an elevated mutation rate in H3K9me3 domains (Lawrence et al., 2013), which we find to be more pronounced in the srHC and intermediate subtypes compared to euchromatic H3K9me3 (Figure S2E). Thus, Gradient-seq resolves subtypes of H3K9me3 domains with distinct roles in cancer. These subtypes are not distinguished by FAIRE-seq (Torres et al., 2016), which does not show increased signal in euchromatic H3K9me3 domains compared to the srHC form (Figure S3A).

Structural subtypes reflect diverse properties of chromatin

Our finding that regions with the same repressive histone marks could be partitioned into srHC, intermediate, and euchromatic subtypes (Figure 4A) led us to investigate how these subtypes related to other properties traditionally associated with heterochromatin. Our classifications are strongly supported by DNA methylation rates at CpG islands (Figure 4B), which are high in all srHC subtypes and low in all euchromatic subtypes, whereas outside CpG islands methylation is high in all domain types (Figure S3B). CpG island methylation was actually highest in the 133 MB of srHC lacking enrichment for H3K9me3 or H3K27me3 (Figure 4B, far right), showing that srHC domains have heterochromatic features even where they diverge from histone mark-based approaches. Lamin B1 binding by ChIP-seq (Dou et al., 2015) is higher in H3K9me3 domains compared to H3K27me3 domains (Figure 4C), as seen previously (Guelen et al., 2008). However, within H3K9me3 and H3K27me3 domains, the srHC and intermediate subtypes have higher Lamin B1 binding (Figure 4C) and greater overlap with published Lamina Associated Domains (Guelen et al., 2008) (Figure S3C), compared to euchromatin subtypes. Thus, Gradient-seq parses out subtypes of chromatin with more heterochromatic or euchromatic features that are not distinguished by the histone marks.

Figure 4. Classification of chromatin by sonication resistance and histone mark predicts diverse properties of heterochromatin.

(A-G) Gradient-seq was used to classify fibroblast chromatin into three categories: srHC, intermediate (“int”), and euchromatin (“euchr”). Here we use diverse datasets to compare these categories on a genome-wide basis (“total genome”) and also for the regions inside H3K9me3 domains, inside H3K27me3 domains, or outside of both (“neither mark enriched”). Asterisks: P<0.05 by Wilcoxon rank sum test. Whiskers: 5^th and 95^th percentiles, unless specified below.

(A) Gradient-seq data per 10-kb window plotted as the sequencing signal in the srHC fraction divided by that of euchromatin fraction. This data was used to classify the domains and serves as a reference for comparison to the other panels.

(B) The frequency of DNA methylation per CpG (Bernstein et al., 2010) at CpG islands (whiskers: 10^th and 90^th percentiles among CpGs). Data on far left is also shown in Figure 2F, left. See also Figure S3.

(C) Lamin B1 ChIP-seq signal (Dou et al., 2015), divided by corresponding input signal, per 10-kb window.

(D) DNase-seq reads (Thurman et al., 2012) per million mapped per 10-kb window (n.s., not significant).

(E) Timing of DNA replication. For each cell cycle phase, the number of Repli-seq reads (Pope et al., 2014) in each domain category was normalized for sequencing depth and expressed as a fraction of the total signal for that domain type, so that each column sums to 1.

(F) Extent of gene activation during hiHep reprogramming. Microarray data (Huang et al., 2014) was curated for genes expressed in native hepatocytes more than fibroblasts (“hepatic genes”), and hiHep expression levels (log2-transformed) are plotted on a relative scale between fibroblast and hepatocyte levels. Violin plots use width to show where values are concentrated; median is indicated by white circles.

(G) As in F, but for neuronal (hiCN) reprogramming (Liu et al., 2013), with genes filtered for those expressed in spinal cord over fibroblasts (“neural genes”).

Euchromatin domains were consistently more DNase-sensitive (Thurman et al., 2012) than srHC or intermediate domains, even where they overlapped with H3K27me3, supporting the findings of Gradient-seq (Figure 4D). However, euchromatic H3K9me3 domains were DNase-resistant (Figure 4D). This difference between the euchromatic forms of H3K9me3 and H3K27me3 was similar inside and outside of gene bodies (Figure S3D). Similar to DNase, MNase accessibility (MACC) measured in human K562 cells (Mieczkowski et al., 2016) was elevated in euchromatin domains, both as a whole and in H3K27me3 domains, but not in euchromatin overlapping H3K9me3 (Figure S3E). In regions with similar H3K9me3 or H3K27me3 patterns in both K562 cells and foreskin fibroblasts, MACC closely mirrored the pattern of srHC versus euchromatin domains (Figure S3F, left). This indicates that there is widespread similarity of sonication-resistance and MNase-resistance, and that the Gradient-seq results are not dependent upon crosslinking or gradient sedimentation. However, regions bearing both H3K9me3 and H3K36me3 in both cell types (Figure S3F, right) are euchromatic by Gradient-seq (and transcribed) but MNase-resistant (negative MACC). Thus, Gradient-seq but not nuclease-based methods can distinguish transcriptionally active H3K9me3 domains from inactive ones.

With regard to replication timing (Pope et al., 2014), both srHC and euchromatin forms of H3K9me3 replicate late, while H3K27me3 domains replicate in early-to-mid S phase (Figure 4E), as seen previously (Chandra et al., 2012). We conclude that late replication is more closely associated with the H3K9me3 histone mark than with chromatin structure as measured by Gradient-seq.

We curated all genes expressed at higher levels in hepatocytes than fibroblasts and analyzed their activation during hiHep reprogramming (Huang et al., 2014). Hepatocyte-expressed genes within the srHC or intermediate subtypes of fibroblast H3K9me3 domains, but not in euchromatic H3K9me3 domains, have a profound failure to activate during cell conversion to hiHep cells (Huang et al., 2014) (Figure 4F). Genes within H3K27me3-marked srHC were resistant to gene activation during hiHep reprogramming, though not as resistant as H3K9me3-marked srHC (Figure 4F). Importantly, similar results were obtained using data where fibroblasts were converted to human induced cholinergic neurons (hiCNs) (Liu et al., 2013). Neuron-expressed genes in the srHC subtypes of H3K9me3 and H3K27me3 domains, compared to the euchromatin subtypes, were more refractory to activate during hiCN reprogramming (Figure 4G). Thus, combining sonication-resistance with histone marks accurately predicts which genes are most difficult to activate for multiple forms of direct cell conversion.

The proteome of H3K9me3-marked srHC

To identify proteins embedded in heterochromatic H3K9me3 domains, separate from active H3K9me3 domains, we performed label-free quantitative proteomics (Cox and Mann, 2008; Schwanhausser et al., 2011) on three biological replicates of srHC+H3K9me3 chromatin (immunoprecipitations of H3K9me3 from srHC fractions, as in Figure 1F, Figure S1A-B), as well as proteomics of the srHC fraction itself (Figure 5A). Results were compared to replicate euchromatin-containing fractions (“gradient top”) that contain both euchromatic fragments and soluble proteins from the lysate. Of 1,864 proteins detected in the srHC fraction, 217 “srHC-enriched proteins” were enriched over the gradient top (T-test, P < 0.05) (Figure S4A). Of 716 proteins in the srHC+H3K9me3 sample, 172 “H3K9me3 heterochromatin proteins” were enriched over the gradient top (Figure 5B, orange dots), with substantial overlap with the srHC-enriched proteins (Figure S4B). An additional 429 “shared” proteins were detected similarly in srHC+H3K9me3 and the gradient top (Figure 5B, grey dots). Of 3,097 total proteins with two or more peptides, 1,474 were unique or significantly enriched in the gradient top (Table S5 provides all categories).

Figure 5. Proteomic analysis of purified H3K9me3-marked heterochromatin.

(A) Strategy for quantitative proteomics study of 3 purified fractions.

(B) Proteins with a higher average rank in the srHC+H3K9me3 sample than the gradient top sample (x-axis) and a significance less than 0.05 (y-axis) were classified as “H3K9me3 heterochromatin proteins”. See also Table S5.

(C) (left) List of selected H3K9me3 heterochromatin proteins that were previously shown to contribute to heterochromatin and/or gene repression; (right) the 172 H3K9me3 heterochromatin proteins were sorted and plotted by their fold-enrichment in the srHC+H3K9me3 (relative to gradient top), with the selected proteins on the left indicated by orange dots. RBMX and RBMXL1, a focus of subsequent studies, are indicated in red (RBMXL1 is srHC-enriched but not one of the 172 H3K9me3 heterochromatin proteins). References: (van Dijk et al., 2010; Dong et al., 2009; Hayashihara et al., 2010; Mathur et al., 2001; Thompson et al., 2015; Vermeulen et al., 2010).

(D) Percentage of each proteomic category (eg, H3K9me3 heterochromatin proteins) that is found in each published dataset. The raw number of proteins found in common is listed below the bars. Significance was computed relative to the background overlap for the total set of 3,097 MS-detected proteins: *p<0.05, **p<0.001.

(E) Protein categories from proteomics analysis were compared to lists of iPS repressors (knockdown increases reprogramming) and iPS effectors (knockdown inhibits reprogramming) from a genome-wide screen (Toh et al., 2016). The ratio of repressors to effectors is plotted (numbers above the bar). Asterisk indicates P < 0.05 by simulation test. See also Figure S4.

(F) The 172 H3K9me3 heterochromatin proteins were sorted and plotted by their fold-enrichment in the srHC+H3K9me3 (relative to gradient top), with proteins recurrently mutated in ALS (Cirulli et al., 2015) indicated in green.

(G) Percentage of proteins that overlap with each dataset, for all H3K9me3 heterochromatin proteins and those with RNA-binding activity according to (Gerstberger et al., 2014).

Many of the 172 H3K9me3 heterochromatin proteins are known to generate or associate with compacted chromatin, including linker histones H1.1, H1x, and H1.0, histone variant macroH2A, Lamin B1, HDAC2, HNRNPK (Bao et al., 2015; Thompson et al., 2015), the co-repressors NONO and SFPQ (Dong et al., 2009; Mathur et al., 2001; Vermeulen et al., 2010), and HP1 interactors (HP1BP3, THRAP3, and BCLAF1) (Vermeulen et al., 2010; Hayashihara et al., 2010) (Figure 5C). HP1γ (a “shared” protein) and HP1α were partially and non-significantly enriched in srHC versus gradient top, consistent with our finding that HP1-associated ZNF domains (Vogel et al., 2006) are enriched in euchromatin. Of the 172 H3K9me3 heterochromatin proteins, 64 (37%) were previously identified by H3K9me3-directed ChIP-MS in HeLa cells (Soldi and Bonaldi, 2013) (Figure 5D), and 124 of 208 (60%) proteins discovered in the latter occur in H3K9me3 heterochromatin or shared categories. Proteins found to bind the H3K9me3 mark (Vermeulen et al., 2010) or pericentromeric satellites in murine ES cells (Saksouk et al., 2014) are enriched among H3K9me3 heterochromatin and srHC proteins, compared to gradient top proteins (Figure 5D). Using the STRING v10 database, the H3K9me3 heterochromatin proteins exhibited far more direct binding interactions among each other (Figure S4C, right) than with euchromatin proteins (Figure S4C, left) (Table S5). H3K9me3 heterochromatin proteins are enriched for repressors of iPS reprogramming over factors that enhance reprogramming, including 22 known iPS repressors (Bao et al., 2015; Han et al., 2013; Qin et al., 2014; Toh et al., 2016) (Figure 5E, S4D).

Unexpectedly, six H3K9me3 heterochromatin proteins are encoded by genes recurrently mutated in amyotrophic lateral sclerosis (ALS) (Cirulli et al., 2015) (Figure 5F), a higher number than expected (Figure 5D, right). Of such proteins, TDP-43 (TARDBP), which forms neuronal inclusions in ALS (Taylor et al., 2016), has been shown to impede iPS reprogramming (Qin et al., 2014; Toh et al., 2016) (Figure S4D). Using transcriptome data from ALS patient fibroblasts (Highley et al., 2014), we find that germline mutations of TARDBP are associated with a widespread upregulation of genes in fibroblast srHC, compared to healthy controls (Figure S4E-G). Such upregulation is absent in sporadic ALS cases, where the fibroblast copies of TARDBP are normal (Figure S4E-G). These findings may explain activation of repeat elements and the altered organization of compacted chromatin in ALS models (Amlie-Wolf et al., 2015; Li et al., 2012). TDP-43 and FUS mediate liquid-liquid phase separation in stress granules via their low-complexity domains (Taylor et al., 2016), raising the possibility that they may perform similar functions at heterochromatin domains, which also form via phase separation in an HP1-dependent manner (Larson et al., 2017; Strom et al., 2017). We found that 20 of 172 H3K9me3 heterochromatin proteins contain FUS-like RGG domains, as curated by (Thandapani et al., 2013), versus 1 in 1,474 Gradient Top proteins (Table S5), indicating that a motif capable of mediating phase separation is enriched among H3K9me3 heterochromatin proteins.

While HP1 localization to heterochromatin is facilitated by RNA (Muchardt et al., 2002), RNA binding is not known as a common feature among mammalian heterochromatin proteins, yet 119 of the 172 H3K9me3 heterochromatin proteins have annotated RNA-binding activity (Gerstberger et al., 2014). The RNA-binding proteins in H3K9me3 heterochromatin, compared to the total heterochromatin proteins, have even higher agreement with published H3K9me3 ChIP-MS proteins or H3K9me3 readers (Figure 5G). Gradient sedimentation prior to IP ensures that such RNA binding proteins are associated with transcriptionally repressed H3K9me3 domains.

Screen of heterochromatin proteins reveals functional impediments to gene activation

To monitor changes in heterochromatin, we selected three hepatic transcripts—DSC2, NR1H4, and CRP—that are H3K9me3-marked and non-euchromatic in fibroblasts (Figure 6A) and do not properly activate in hiHeps. We transduced fibroblasts with the transcription factor (TF) cocktail of FOXA3, HNF1A, and HNF4A for making hiHeps (Huang et al., 2014) (Figure S5A), and screened for siRNAs that allow activation of the hepatic transcripts in the presence of these factors and hepatocyte maintenance medium (Figure 6B, STAR Methods). The tested siRNAs targeted 50 H3K9me3 heterochromatin and srHC-enriched proteins (Table S6), in addition to positive controls (SUV39H1 and SETDB1) and non-targeting siRNA. The transcripts DSC2, NR1H4, and CRP were all upregulated by SUV39H1 knockdown during reprogramming, whereas SETDB1 knockdown was more selective (Figures S5B-C). Among the 50 screened knockdowns, 36 upregulated at least one of the genes in the presence of hepatic TFs, 22 upregulated two genes, and 10 upregulated all 3 genes (Figure 6C-D, Table S6). The 22 causing upregulation of at least two transcripts included known repressors or HP1 interactors, such as GATAD2A, HNRNPK, NONO, BCLAF1, and HP1BP3 (Figure 5C, 6C), but the majority targeted novel regulators (including TARDBP), providing strong validation of the heterochromatin proteomics dataset as a resource for discovery. For DSC2 and NR1H4, there was only 1 protein out of 50 whose knockdown caused downregulation more than two-fold (EWSR1 and NCOA5, respectively; Figure 6C), whereas 17 did for CRP, suggesting that this inflammatory gene is more labile and subject to secondary effects.

Figure 6. A screen of heterochromatin proteins reveals functional impediments to gene activation.

(A) Browser views comparing fibroblast (“fib.”) and liver chromatin state at hepatic genes monitored for functional screen. All ChIP-seq data (n=2 for fib. H3K9me3, otherwise n=3) and mRNA-seq data (n=1) were obtained from the Epigenomics Roadmap. Red arrows indicate liver-specific mRNA-seq signal.

(B) Experimental setup of siRNA screen.

(C) Heatmap showing fold-upregulation of indicated transcripts after siRNA treatment in the presence of hepatic TFs (average of two siRNAs and two replicates), relative to control siRNA and GAPDH endogenous control. See also Figure S5, Table S6.

(D) Fold-change in expression of hepatic genes in fibroblast H3K9me3 domains (DSC2, NR1H4, and CRP) in siRNA screen performed in the presence of hepatic TFs. Each dot indicates a different siRNA targeting one of the 50 screened genes (two siRNAs per gene) or targeting the H3K9me3 methyltransferases SUV39H1 or SETDB1. Fold-changes are calculated relative to an average of negative control siRNAs, and the axes compare the effect seen in two independent screen replicates. Red dots indicate significant upregulation by T-test in the two replicates of that siRNA compared to control. Labeling is applied selectively to significant siRNAs of interest. Indicated “RBMX” siRNAs co-target both RBMX and RBMXL1.

RBMX/L1 maintains heterochromatin to impede reprogramming

The strongest induction of heterochromatic genes was seen with siRNAs co-targeting RNA Binding Motif Protein, X-linked (RBMX) and the related protein RBMXL1 (Figure 6C-D, Figure S5D-E). RBMX had the second highest proteomic enrichment among H3K9me3 heterochromatin proteins, while RBMXL1 protein was srHC-enriched (Fig 5C, right). RBMX binds chromatin independent of its RNA recognition motif and promotes pericentromeric cohesion (Matsunaga et al., 2012), similar to SUV39H1 (Peters et al., 2001). RBMX was also found to associate with H3K9me3 by ChIP-MS (Soldi and Bonaldi, 2013), but there was no evidence of a functional role in gene silencing or heterochromatin. We found that two cycles of siRNA against both RBMX and RBMXL1 (RBMX/L1), in the presence but not absence of hepatic TFs, enhanced induction of six of eight liver-genes in fibroblast H3K9me3, similar the SUV39H1 knockdown (Figure 7A). The heterochromatic genes at this early reprogramming stage were expressed lower than normal liver (Figure S5F), as expected.

Figure 7. RBMX and RBMXL1 maintain the sonication resistance and reprogramming resistance of hepatic genes in heterochromatin.

(A) RT-PCR for hepatic genes marked by H3K9me3 in fibroblasts, after two cycles of siRNA transfection, in fibroblasts expressing hepatic TFs (left) or not (right). Error bars, SEM of two biological replicates; P values by Student’s T test.

(B) Relative enrichment of sequences in gel-extracted srHC DNA versus sonication-sensitive DNA, by qPCR, after treatment with the indicated siRNA. PCR sites include gene promoters in srHC domains (left) and euchromatic sites (far right; primers: 1-Ch3_nonDBR_1 and 2-Chr20_nonDBR_3). Error bars: SD of two biological replicates. See also Figure S6.

(C) Comparison of expression fold-changes induced by SUV39H1 siRNA versus RBMX/L1 siRNA, in the presence of hiHep factors, at genes inside srHC domains, by mRNA-seq (n=2 per condition). See also Table S7.

(D) Browser views of genes in srHC that are upregulated by si-SUV39H1 and si-RBMX/L1 in the presence (left) or absence (far right) of hepatic factors.

(E) Top: flow cytometry comparison of hiHep reprogramming efficiency at day 14 (percent of cells double-positive for albumin and alpha-1-antitrypsin/AAT) after indicated siRNA treatments (*p<0.05, **p<0.005, Student’s T test, n=2, error bars: SD). Bottom: representative immunofluorescence of siRNA-treated hiHeps at day 10 of reprogramming. See also Figure S7.

We next investigated whether RBMX/L1 controls sonication-resistant heterochromatin, in the absence of hepatic TFs. RBMX/L1 siRNA reduced nuclear staining for H3K9me3 to an extent correlating with the extent of knockdown and similar to SUV39H1 siRNA (Figure S6A-C). We purified DNA from sonicated chromatin, performed gel extraction of large versus small fragments and used PCR to assess the sonication-resistance of specific sites (Figure S6D-E). Importantly, depletion of SUV39H1 and RBMX/L1 significantly reduced the sonication-resistance of the promoters of DSC2, NR1H4, and CRP (Figure 7B), even though these genes were not upregulated without hepatic factors (Figure 7A, right), and of two out of three additional srHC H3K9me3 sites (Figure S6F). SUV39H1 siRNA also reduced sonication resistance at non-heterochromatic sites, whereas RBMX/L1 siRNA specifically reduced sonication resistance at heterochromatic sites (Figure 7B, far right; Figure S6G).

To understand the role of RBMX/L1 in impeding hepatic TF activity genome-wide, we performed mRNA-seq in fibroblasts treated with RBMX/L1 siRNA, compared to control siRNA and siRNA against SUV39H1, after 7 days of hepatic TF expression (as in Figure 6B). Unsupervised clustering of the mRNA-seq data confirmed the similarity of biological replicates and showed that the RBMX/L1 siRNA treatment clustered with the SUV39H1 knockdown (Figure S7A, green box). Of 1333 genes upregulated by RBMX/L1 knockdown (compared to control siRNA) in the presence of hepatic TFs, 65% were also upregulated by SUV39H1 (P < 10⁻¹⁰⁰) (Figure S7B). The SUV39H1 knockdown revealed 281 genes in srHC that are responsive to heterochromatin disruption in the presence of hepatic TFs; for these genes, the effect of RBMX/L1 siRNA and SUV39H1 siRNA was well-correlated (Figure 7C, top), including for CRP and NR1H4 (Figure 7D). However, some srHC genes were differentially affected (Figure 7C, bottom; Figure S7C, Table S7). In the absence of ectopic transcription factors, fewer genes in srHC were upregulated (Figure S7D-E), consistent with our qPCR studies (Figure 7A). Importantly, knockdown of RBMX/L1 without added TFs still activated 67 genes within srHC (Table S7), such as TMEM178 (Figure 7D, right).

When hiHep reprogramming was performed for 14 days, the initial treatment with RBMX/L1 siRNA significantly increased the overall efficiency of cellular reprogramming, as assessed by the number of cells double-positive for albumin and alpha-1-antitrypsin (Figure 7E, Figure S7F-G). Thus, RBMX, one of the most highly enriched proteins in H3K9me3 heterochromatin, together with RBMXL1, maintains the sonication resistance of heterochromatin to impede the activity of transcription factors and serve as a barrier to direct cell conversion.

DISCUSSION

The fields of cell differentiation and reprogramming have focused on activators of new cell fates, while mechanisms by which cells resist fate changes have been less studied. Prior work on the latter focused on H3K9me3 and H3K27me3 marked chromatin (Ezhkova et al., 2009; Hawkins et al., 2010; Mansour et al., 2012; Matoba et al., 2014; Soufi et al., 2012; Xu et al., 2014), yet there was ample evidence, which we affirmed, that H3K9me3 and H3K27me3 domains in mammalian cells could be transcriptionally active or competent for activation (Blahnik et al., 2011; Breiling et al., 2001; Riddle et al., 2012; Vakoc et al., 2005). We find that using sucrose gradients to recover sonication-resistant chromatin, usually excluded from ChIP-seq studies, effectively purifies the subsets of H3K9me3 and H3K27me3 domains that are transcriptionally silent, nuclease resistant, hypermethylated, associated with Lamin B1, and most refractory to cellular reprogramming. We thus suggest that the parameter of sonication resistance is a reliable predictor of the functional properties associated with heterochromatin, including repression and resistance to gene induction. Our findings highlight the importance of cross-referencing H3K9me3 or H3K27me3 maps with transcriptional data, H3K36me3 marks, and structural assays like Gradient-seq (or other measure of sonication-resistance) before inferring heterochromatin status.

Previous studies applied gradient sedimentation to MNase-digested or sonicated chromatin (Gilbert et al., 2004; Ishihara et al., 2010), but focused on fragments that differed in buoyancy rather than fragmentation resistance. Gradient-seq allows for heterochromatin enrichment genome-wide and avoids an electrophoresis step, enabling subsequent IP or proteomics. Regions of the genome resistant to sonication are also resistant to MNase (Mieczkowski et al., 2016) and DNase (Figure 4D, Figure S3E-F). Thus, normalization to input controls is important for both crosslinked and native ChIP. However, the ability to distinguish the euchromatic and srHC subtypes of H3K9me3 domains, which differ dramatically by gene expression (Figure 3A) and resistance to reprogramming (Figure 4F-G), is unique to Gradient-seq and not seen with nuclease digestion or FAIRE (Figure S3A).

We identified 172 proteins within the heterochromatic form of H3K9me3 domains, having removed contaminating proteins from transcribed H3K9me3 domains, in contrast to previous studies (Engelen et al., 2015; Ji et al., 2015; Soldi and Bonaldi, 2013). The fact that RNA-binding proteins remain strongly enriched in H3K9me3-marked chromatin even after depletion of euchromatic H3K9me3 domains (Figure 5G), with many acting to functionally impede gene activation (Figure 6C), provides strong support for a role of RNA-binding proteins in mammalian heterochromatin. In particular, the 172 H3K9me3 heterochromatin proteins are enriched for RNA-binding RGG motifs, which are capable of mediating liquid phase separation (Thandapani et al., 2013), a process shown to be integral to heterochromatin formation (Larson et al., 2017; Strom et al., 2017).

Two related proteins, RBMX and RBMXL1, although not previously implicated in gene repression, were among the strongest hits in both our biochemical and functional studies of H3K9me3 heterochromatin. We find that removing RBMX/L1 in the absence of TFs elicits sonication sensitivity at target heterochromatic genes, while usually not inducing transcription until TFs are added, similar to results with SUV39H1 knockdown. Thus, alternative-lineage TFs help test the functionality of heterochromatin proteins. The loosening of heterochromatin upon RBMX/L1 depletion, as assayed by sonication sensitivity, explains the enhanced efficiency of hiHep reprogramming in these cells. Following the example of our work with RBMX/L1, we suggest that the H3K9me3 heterochromatin protein dataset can be explored further to understand how different subtypes of heterochromatin are formed and can be destabilized. As we have shown, relevant contexts for such studies include cellular reprogramming, cancer, and neurodegenerative disorders.

STAR METHODS

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for reagents may be directed to and will be fulfilled by the corresponding author Ken Zaret (zaret@upenn.edu)

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Human BJ foreskin fibroblasts were obtained from Stemgent (08-0027) at passage 6 and cultured in Eagle’s Minimum Essential Medium (EMEM) (Sigma-Aldrich M2279) supplemented with 10% fetal bovine serum (FBS, Hyclone SH30071) and 2mM L-glutamine (Gibco) at 37°C and 5% CO₂.

METHOD DETAILS

Preparation of crosslinked chromatin lysates

BJ fibroblast cells were grown to ~80% confluence in 150-mm dishes (5 - 20 plates per chromatin batch, ~4 × 10⁶ cells per plate). Cells were crosslinked directly in the culture dishes, at room temperature, by addition of 2 ml formaldehyde solution (50 mM HEPES-KOH, pH 7.5, 100 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 11% formaldehyde) to 20 ml media, for 1% final formaldehyde concentration. After 10 min, crosslinking was quenched by addition of glycine to a final concentration of 125 mM, followed by incubation for 5 min at room temperature. Cells were harvested from the plate with a plastic cell lifter, pelleted at 200 g for 4 min (4°C), and washed three times with ice-cold PBS. All subsequent steps were performed on ice or in centrifuges cooled to 4°C. To enrich for nuclei, cells were allowed to swell for 10 minutes in 4 ml Hypotonic Lysis Buffer (20 mM HEPES-KOH pH 7.5, 20 mM KCl, 1 mM EDTA, 10% glycerol, 1% IGEPAL CA-630, 0.25% Triton-X, 1 mM DTT, 0.2 mM PMSF, Protease Inhibitor Cocktail) and were mechanically dounced for 50 strokes. Nuclei were pelleted at 1,350 g for 4 min, washed with hypotonic buffer, and pelleted again at the same speed. Pellets were resuspended in 10 ml Nuclear Wash Buffer (10 mM Tris-HCl pH 8.0, 200 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 1 mM DTT, 0.2 mM PMSF, Protease Inhibitor Cocktail) and incubated for 10 min, rocking. Nuclei were pelleted at 1,350 g for 4 min and resuspended in 0.5 - 1 ml Sonication Lysis Buffer (10 mM Tris-HCl pH 8.0, 100 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 0.5% N-lauroylsarcosine, 0.1% sodium deoxycholate, 1 mM DTT, 0.2 mM PMSF, Protease Inhibitor Cocktail) and transferred to 15-ml polystyrene tubes. Sonication was performed in polystyrene tubes using a Diagenode Bioruptor UCD-200 (power HI, cycles of 30s on/30s off) for at least 25 cycles. After sonication, lysates were transferred to microcentrifuge tubes, supplemented with Triton-X to 1% final concentration (to promote chromatin solubility), and centrifuged at 20,000 g to pellet debris. Supernatants were transferred to new tubes, snap-frozen in liquid nitrogen, and stored at −80°C while chromatin shearing was assessed by agarose gel electrophoresis of purified DNA (as described for chromatin immunoprecipitation, below). Sonication was repeated as necessary until the majority of DNA was 200–400bp, with only a faint trail of larger material.

Chromatin immunoprecipitation

Immunoprecipitation was performed using Dynabead Protein G magnetic beads (Thermo Fisher, 10004D) saturated with the antibody of interest, such as anti-H3K9me3 (Abcam ab8898) or anti-mouse IgG control (Abcam ab46540). 5 μg of antibody and 25 μl of Dynabead slurry were used per 25 μg of chromatin (according to mass of purified DNA measured by nanodrop), with scaling as necessary. Dynabeads were first washed twice with 200 μl PBS, using a magnetic rack, and then resuspended in a volume of PBS that is 4X the original slurry volume. This suspension was supplemented with the desired amount of antibody, mixed, and incubated on a rotating rack at 4°C for 2 - 6 hours, to allow antibody conjugation. Sonicated, crosslinked chromatin lysates (see above) were thawed on ice, and the desired mass of chromatin (25 μg or more) was aliquoted into low-retention 1.5-ml (Axygen MCT-150-L-C) tubes. Chromatin was diluted to 1 ml final volume with ice-cold Chromatin IP Buffer (10 mM Tris-HCl pH 8.0, 100 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 0.1% N-lauroylsarcosine, 1% Triton X-100, 1 mM DTT, 0.2 mM PMSF, Protease Inhibitor Cocktail - Roche #11873580001). A separate aliquot of chromatin (one-fifth the mass) was reserved as the input sample. Antibody-conjugated beads were washed twice with Chromatin IP buffer and then resuspended in the 1-ml sample of chromatin. Immunoprecipitations were incubated for 16 hours at 4°C on a rotating rack. Using a magnetic rack, the unbound lysate was aspirated, and the beads were washed 5 times with 900 μl ice-cold ChIP RIPA Buffer (50 mM HEPES, pH 7.5, 500 mM LiCl, 1 mM EDTA, 1% IGEPAL CA-630, 0.7% sodium deoxycholate, Protease Inhibitor Cocktail) and once with 900 μl ice-cold TE buffer (10 mM Tris-HCl, pH 8.0, 1 mM EDTA). Chromatin was eluted in 200 μl ChIP Elution Buffer (50 mM Tris-HCl, pH 8.0, 10 mM EDTA, 1% SDS) at 65°C for 30 min, shaking, and the eluate was transferred to a new tube. Reserved input sample was similarly diluted at least 3-fold in ChIP Elution Buffer, to 200 μl final volume. To purify DNA from ChIP eluate and Input, samples were decrosslinked by heating at 65°C for 18 hours. Chromatin was diluted with 200 μl TE and treated with 8 μl RNase A (10 mg/ml stock, Roche #10109169001), followed by incubation for 2 hours at 37°C. Protein was degraded by addition of 4 μl Proteinase K (20 mg/ml stock, Roche #03115828001) and incubation for 2 hours at 55°C. DNA was purified by two rounds of extraction with 400 μl phenol-chloroform-isoamyl alcohol. The extracted aqueous phase was supplemented with 16 μl of 5 M NaCl and 1.5 μl glycogen (20 mg/ml stock, Roche #10901393001). DNA was precipitated by addition of 2 volumes (800 μl) 100% EtOH and overnight incubation at −20°C. DNA was pelleted at 20,000 g for 10 min (4°C), washed with 500 μl 80% EtOH, and pelleted again. The DNA pellet was air-dried and dissolved in 200 μl TE buffer. DNA yield was quantified by Quant-iT PicoGreen dsDNA Assay (Thermo Fisher P7589). Five-fold serial dilutions of input DNA were prepared in TE and were used as a standard curve when analyzing ChIP eluates by qPCR, in order to quantify sequence recovery as percent input.

Sucrose gradient sedimentation of chromatin

For preparative gradients used for proteomic and sequencing studies, crosslinked and sonicated chromatin was purified (see above) from near-confluent BJ fibroblasts in 20 150-mm plates (at least 8 × 10⁷ cells, 0.5 mg DNA) per gradient. Chromatin lysates were prepared in 0.5 mL Sonication Lysis Buffer (see above) to allow the majority of the sample to be loaded on a single gradient. Prior to running gradients, chromatin shearing efficiency was verified by purifying DNA from a 5 μl chromatin aliquot (as described above for ChIP input DNA), while the remaining lysate was snap-frozen in liquid nitrogen and stored at −80°C. Empirically, we found that achieving sufficient levels of DNA shearing, comparable to standard ChIP, was important to achieve robust heterochromatin enrichment via sucrose gradient sedimentation.

6–40% linear sucrose gradients in Chromatin IP buffer were poured into 12-mL Ultra-Clear centrifuge tubes (Beckman #344059, 14 × 89 mm), using a Hoefer SG-15 gradient maker fitted with a two-way stopcock (Bio-Rad #7328102). Gradients were prepared by loading the chambers of the gradient maker with two solutions of approximately equal weight: 5.7 mL of 40% Sucrose Solution (40% sucrose, 10 mM Tris-HCl, pH 8.0, 100 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 1% Triton-X, 0.1% N-lauroylsarcosine, 1 mM DTT, 0.2 mM PMSF, Protease Inhibitor Cocktail) and 6.4 mL of 6% Sucrose Solution (6 mL of 40% Sucrose Solution, plus 34 mL of: 10 mM Tris-HCl, pH 8.0, 100 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 1% Triton-X, 0.1% N-lauroylsarcosine, 1 mM DTT, 0.2 mM PMSF, Protease Inhibitor Cocktail), which are gradually mixed as the solutions exit the gradient maker. Gradients were filled bottom-up, beginning with the heavier 40% Sucrose Solution. To load the gradients, 455 μl of thawed chromatin lysate was mixed with 65 μl of the 40% Sucrose Solution, for a 5% final concentration of sucrose, and then the resulting sample was layered gently on the 6–40% gradients with a 1-mL pipet. Gradients were spun on a Beckman SW 41 Ti rotor at 41,000 rpm for 3 hours (4°C), using conditions similar to those developed by Bickmore (Gilbert et al., 2004). Slow acceleration and deceleration settings were used. After sedimentation, gradients were fractionated top-down using a micropipet, generating 24 fractions of 500 μL. 30 μL of each fraction was used for DNA purification (performed as for ChIP samples, see above) to allow fraction-specific qPCR studies; the remaining samples were snap-frozen and stored at −80°C.

The fractions found by qPCR to have the strongest enrichment for heterochromatic regions (fractions #10–17) were pooled as the “sonication-resistant heterochromatin (srHC) fraction”, while fraction #2 (second fraction from the top) was used as the “euchromatin fraction.” Both samples were dialyzed to Chromatin IP buffer (10 mM Tris-HCl pH 8.0, 100 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 0.1% N-lauroylsarcosine, 1% Triton-X, 1 mM DTT, 0.2 mM PMSF, Protease Inhibitor Cocktail) to remove sucrose. Dialysis was performed for two rounds of five hours each using Slide-A-Lyzer G2 Cassettes 7K MWCO (Thermo Fisher). 5% of the dialyzed chromatin samples (approximately 250 ng DNA for srHC fraction) were used for DNA purification, using the same procedure as for ChIP eluates above, to enable subsequent qPCR and sequencing studies. For purification of proteins from gradient fractions, equal DNA equivalents were used for both the euchromatin and srHC fractions, based on Quant-iT PicoGreen dsDNA Assay (Thermo Fisher P7589). Prior to protein precipitation, the dialyzed fractions were buffered by adding Tris-HCl pH 8.0 to 100 mM final concentration and then decrosslinked by heating at 65°C overnight and then at 99°C for 30 minutes. The proteins were precipitated in 6 volumes acetone (overnight at −20°C), pelleted at 3,600 g, washed once with ice-cold acetone, resuspended in 8M urea, and quantified by Bradford Assay (Bio-Rad #500-0006). Alternatively, the dialyzed srHC and euchromatin fractions were used for chromatin IP against H3K9me3 or IgG control (using above ChIP protocol). For these IP eluates in ChIP Elution Buffer, one-twelfth of the sample was used for DNA purification, as above. For protein analysis, the remaining eluate was run in a centrifugal evaporator to reduce sample volume to ~25 μl, mixed with 4X NuPAGE LDS Sample Buffer (Thermo Fisher NP0007) and β-mercaptoethanol (2.5% final concentration), and decrosslinked by heating at 99°C for 30 min.

Proteomics analysis of chromatin samples

Proteins were prepared for mass-spectrometry by in-gel digestion. Samples were mixed with 4X NuPAGE LDS Sample Buffer (Thermo Fisher NP0007) and β-mercaptoethanol (2.5% final concentration) and loaded into NuPAGE Novex 4–12% Bis-Tris Protein Gels (Thermo Fisher NP0335). Gels were run at 100V in NuPAGE MOPS SDS Running Buffer (Thermo Fisher NP0001) and washed three times with dH₂O for five minutes each. Gels were stained with SimplyBlue SafeStain (Thermo Fisher LC6060) for 1 hour at room temperature, photographed, and destained in dH₂O overnight. Gels were then incubated in 40% ethanol, 10% acetic acid for 1–2 hours to fix and further destain. Lanes were excised and cut into five pieces that were digested in separate tubes (but pooled prior to nLC-MS/MS). Each piece was further diced into ~1 mm³ cubes and transferred to clean microcentrifuge tubes.

Gel slices were reduced with 10 mM DTT for 1 hour at 56°C, alkylated using 55 mM iodoacetamide for 45 minutes (room temperature in dark) and digested overnight with trypsin at 37°C at a concentration of 12.5 ng/μl. Digested peptides were collected and extracted with two rounds of 5% formic acid alternating with acetonitrile. Peptides were then desalted using in-house prepared C₁₈ microcolumns, and the five peptide samples from the same gel lane were combined. Nano liquid chromatography was performed using a Thermo Scientific Easy nLC 1000 equipped with a 75 μm × 18 cm in-house packed column using Reprosil-Pur C18-AQ (3 μm, Dr. Maisch GmbH). Buffer A was 0.1% formic acid and Buffer B was 0.1% formic acid in acetonitrile. Peptides were resolved using a 165 min gradient from 2 to 28% B at a flow rate of 300 nL/min. The HPLC was coupled online to an Orbitrap Elite mass spectrometer (Thermo Scientific) for the first replicate and a Q-Exactive (Thermo Scientific) for the second and the third replicate operating in positive mode. Spectra were acquired using a data dependent acquisition (DDA) method, performing the full MS scan in the orbitrap at 60,000 (Elite) and 70,000 (Q-Exactive) resolution. The MS/MS was performed for both instruments at 17,500 resolution in the orbitrap mass analyzer. Loop count was set to 15 (Elite) and 12 (Q-Exactive) and collision energy was set to 20. Database searching was performed using MaxQuant v1.5.2.8 (Cox and Mann, 2008), using all standard settings unless otherwise stated. Database used was Human UniProt (v July 2015). For protein quantification the iBAQ option (Schwanhausser et al., 2011) was enabled and adopted.

qPCR analysis of gradient fractions

Purified DNA samples from gradient fractions were quantified with Quant-iT PicoGreen dsDNA Assay (Thermo Fisher P7589) and diluted with TE buffer to 0.1 ng/μl, to allow loading of equal DNA mass per qPCR reaction. DNA concentrations were verified after dilution by repeat PicoGreen assay and were adjusted as needed. 10 μl qPCR reactions were prepared in 384-well optical plates with 2 μl DNA sample, 0.1 μl primer mix (10μM each primer), and 5 μL Power SYBR Green PCR Master Mix (Thermo Fisher #4367659). Plates were in a 7900HT Real-Time PCR machine (Thermo Fisher #4329001), using the following thermal cycler protocol: 50°C for 4 min, 95°C for 10 min, followed by 40 cycles of 95°C for 15 s then 54°C for 15 s then 72°C for 45 s, with a dissociation curve generated to verify that a single PCR product was amplified. qPCR results for each fraction were normalized to the input sample.

Primer sequences for detecting sites inside and outside DBRs were from (Soufi et al., 2012) and are listed below. All PCR amplicons correspond to Oct4/Sox2 binding sites: bound in fibroblasts and ES cells for non-DBR sites, and bound only in ES cells for DBR sites (Soufi et al., 2012). There is also enrichment for H3K9me3 at the DBR sites but not the non-DBR sites by ChIP in fibroblasts (Soufi et al., 2012).

primer name	forward primer (5′ - 3′)	reverse primer (5′ - 3′)	location
Ch3_DBR3	TGGTCTTGAATTCCTGGCTG	GCTTAAGAATCGTCCGGAGG	DBR site
Ch3_DBR4	ACCGCCATACCCAACTTG	GATGGCCCTAGGTCTTTAATGG	DBR site
Ch20_DBR2	ATCAAGTGCCAGGAATGGAG	ATGGAGCCCGAATTTCTCAG	DBR site
Ch20_DBR5	AATTTCAAGCGGAGCCCTAG	TCAGAAACCCTATTGAAGCCTC	DBR site
Ch22_DBR2	GCCATTCGTGTGCAGAAAAG	CTGTCCATAGTCAGCGTTCC	DBR site
Ch22_DBR5	CCTCAAGGGATTGGAAGATCTC	GGTGCCCAGATTAAATGTTCC	DBR site
DPPA4	TCCACCTCACCTCTTCTT	GTATTAGTAATTCAACCCAGACAA	DBR site
NANOG	TGTTGAACCATATTCCTGAT	TCTACCAGTCTCACCAAG	DBR site
Ch3_nonDBR1	CATGGAGCAATTGTGAATAAATGTG	ATTAGGCTGGGGCTTTCTG	intergenic
Ch3_nonDBR2	CCTCCAGTATCAACCGAAGAG	TCCGAAGACTCCTACTCACAC	promoter
Ch3_nonDBR3	AAATGCTAAGAGGGTGTGGG	GAGAGTTGCCAGGAACAGAG	gene body
Ch20_nonDBR1	CCCCGCAGACAATGACTATTAG	AGGTGTGAGCGTTCGATATG	intergenic
Ch20_nonDBR3	GGACCACAGCACGGAAAC	CCTTCTCACTCCTCTTCTCCG	promoter
Ch22_nonDBR2	GGGCTTGCATAGTGAAAACATG	ACGGTAGAGGACAGGGAAG	gene body
Ch22_nonDBR3	CAGATTAATGTTTGCCAGGGC	AATATTTCCATTGCTCCAAAATTTCC	gene body
Ch22_nonDBR4	CCCCTATCATTGTGAGAGTGTG	CAATTTACCCGCCACATCAC	promoter

Measuring sonication resistance by gel extraction

Chromatin was prepared as described above for preparation of crosslinked chromatin lysate and sonication was performed in a Covaris S220 sonicator using snap-cap microTUBEs (Covaris #520045) for 12 min (settings: 200W peak power, 10% duty factor, 200 cycles/burst). Sonicated chromatin was run on a 1% agarose gel at 114V for 30 minutes, stained with 0.5 μg/ml Ethidium bromide in TAE buffer for 30 minutes and de-stained in water for 5 minutes twice. Imaging was performed in a Gel Logic 212 Pro. Excision of gel fragments was performed with a scalpel using a UV fluorescent ruler to ensure consistency of gel excision across samples. For each replicate sample a lower molecular weight sonication sensitive and a higher molecular weight sonication resistant portion of the gel was excised as shown in Figure S6D. DNA purification was done using a QIAquick gel extraction kit (Qiagen). Recovered DNA was quantified by Quant-iT PicoGreen dsDNA Assay (Thermo Fisher P7589). qPCR was performed in triplicate on DNA purified from gel fractions as described above for qPCR analysis of gradient fractions. Additional primers used in this analysis are listed in the table below. Quantification of enrichment in the sonication resistant fraction was performed by calculating the fold enrichment of the target sequence in the sonication resistant fraction relative to the sonication sensitive fraction.

primer name	forward primer (5′ - 3′)	reverse primer (5′ - 3′)	location
CRP_Site1	CCAGATGGCCACTCGTTTAAT	AGAAATCTCTCAGGGCTCCA	Promoter
CRP_Site2	ACCTCCTCCTGCCTGGATTTA	TAGTGGCGCAAACTCCCTTA	Promoter
DSC2_Site1	GCTTGTGGGTGTGTGTAGGA	GGCCCACCTGACAGAAAGTT	Promoter
DSC2_Site2	TAGCCCACGACTAGGACCTC	TAGGGAATCAAGGCGACTGC	Promoter
NR1H4_Site1	TCAGTCCCGGGGAAGTGATA	CTCCTGGGCACCCGTATTTC	Promoter
NR1H4_Site2	GTACAGAAATACGGGTGCCCA	GCATTTCTCCTCTGGCCGTT	Promoter

Preparation of Gradient-seq and ChIP-seq libraries

Approximately 50 ng of purified DNA was used for each library, and two biological replicates (independent gradients) were sequenced per sample type. For samples containing large DNA fragments (including the sonication-resistant heterochromatin fraction, IPs off of this fraction, and the input to the gradient), the pure DNA was first sheared in a Covaris S220 sonicator using snap-cap microTUBEs (Covaris #520045) for 5 min (settings: 175W peak power, 10% duty factor, 200 cycles/burst) to produce 150–300 bp fragments. Libraries were then prepared using the NEBNext Ultra DNA Library Prep Kit for Illumina (New England Biolabs E7370S), and amplified using 8 - 9 cycles of PCR using NEBNext Multiplex Oligos for Illumina (New England Biolabs E7335S and E7500S). For H3K9me3 ChIP-seq and ChIP Input libraries, the adapter-ligated DNA was size-selected using Agencourt Ampure XP beads (Beckman Coulter A63881), following the protocol in the NEBNext Ultra kit for a 200-bp average insert size. The libraries for Gradient-seq were not size-selected. Library yield and fragment size distribution was assessed on a Bioanalyzer 2100 instrument (Agilent Technologies), using the DNA 1000 kit (Agilent Technologies 5067-1504).

Western blotting

Whole-cell protein extracts were prepared by resuspending cells in RIPA extraction buffer (25 mM Tris, pH 7.5, 150 mM NaCl, 1% Na-deoxycholate, 1% IGEPAL CA-630, 0.1% SDS) supplemented with Protease Inhibitor Cocktail (Roche #11873580001). Suspensions were incubated on ice for 10 minutes and sonicated for 15 s on HI using a Diagenode Bioruptor UCD-200. Samples were centrifuged at 20,000 g for 10 min (4°C) to pellet debris, and the supernatant was transferred to new tubes. Protein content was quantified by BCA assay (Thermo Fisher Scientific #23227). Protein samples were mixed with 4X NuPAGE LDS Sample Buffer (Thermo Fisher Scientific NP0007) and 10X NuPAGE Sample Reducing Agent (Thermo Fisher Scientific NP0009), and were denatured at 70°C for 10 min. Samples were loaded in NuPAGE Novex 4–12% Bis-Tris Protein Gels (NP0335), and run using NuPAGE Running Buffers (NP0001; NP0002). Wet transfer to PVDF membranes (100V for 1.5 hr) was performed using NuPAGE Transfer Buffer (NP0006) containing 20% methanol, and membranes were blocked overnight in 5% nonfat dairy milk in TBS-T (20 mM Tris, pH 7.5, 150 mM NaCl, 0.1% Tween-20). Primary antibodies were diluted in 1% milk/TBS-T at the following concentrations: anti-GAPDH (1:1000, Santa Cruz Biotechnology sc-365062), anti-SUV39H1 (1:1000, Bethyl Laboratories A302-127A), anti-RBMX (1:1000, Cell Signaling Technology #14794), anti-H3K9me3 (1:1000, Abcam ab8898), and anti-Histone H3 (1:3000, Abcam ab1791). HRP-conjugated secondary antibodies (Santa Cruz Biotechnology sc-2004, sc-2005) were diluted 1:5,000 in 1% milk/TBS-T. Blots were developed using SuperSignal West Pico Chemiluminescent Substrate (Thermo Fisher Scientific #34080) and visualized with an Amersham Imager 600 (GE Healthcare Life Sciences).

Immunostaining

Cells were grown on plates coated with collagen I (Corning #354236), washed twice briefly with PBS, and fixed in 4% paraformaldehyde in PBS for 20 minutes at room temperature. Fixed cells were washed 3 times with PBS (all washes 5–10 minutes at room temperature, rocking), permeabilized with ice-cold 0.1% Triton-X in PBS for 10 minutes, and washed twice with TBS-T (20 mM Tris-HCl pH 7.4, 150 mM NaCl, 0.05% Tween-20). Samples were blocked with 4% donkey serum (Sigma-Aldrich D9663) in PBS for 1–2 hours at room temperature or overnight at 4°C. Primary antibodies added in blocking solution and incubated overnight at 4°C, using the following concentrations: anti-human-albumin (1:100, Bethyl Laboratories A80-229A), anti-alpha-1-antitrypsin (1:200, Thermo Fisher Scientific RB-367), anti-FOXA3 (1:300, Santa Cruz Biotechnology sc-5361), anti-HNF1A (1:150, Santa Cruz Biotechnology sc-6547), anti-HNF4A (1:150, Santa Cruz Biotechnology sc-8987), anti-H3K9me3 (2 μg/ml, Abcam ab8898), anti-RBMX (1:25, Santa Cruz Biotechnology sc-14581). Cells were washed 3 times with TBS-T and then incubated with AlexaFluor 488- or 594-conjugated secondary antibodies raised in donkey (Thermo Fisher Scientific, see Key Resources Table) at a 1:500 dilution in PBS, for 45 minutes at room temperature, protected from light. Samples were then washed 3 times with PBS, counterstained with 1 μg/mL DAPI (Thermo Fisher, D1306) in PBS for 10 minutes, or instead with 10 μM DRAQ5 (Thermo Fisher Scientific #62251) in PBS for 20 minutes. Cells were then washed once more with PBS. Fluorescence images were taken using a Nikon eclipse TE2000-U microscope controlled by Nikon elements software and equipped with the appropriate filters.

Lentivirus production

Lentiviral plasmids pWPI.1-FOXA3, pWPI.1-HNF1A, and pWPI-HNF4A were kindly provided by the laboratory of Lijian Hui (Huang et al., 2014). 293T cells were grown in DMEM High Glucose (Thermo Fisher Scientific #11995) supplemented with 10% FBS (Hyclone SH30071) and were seeded in 10-cm dishes at a density of 8 × 10⁵ cells per plate. After 24 hours, transfection mixtures were prepared by mixing 2.5 μg lentiviral vector, 1.7 μg packaging plasmid psPAX2 (Addgene #12260), 0.8 μg envelope plasmid pMD2.G (Addgene #12259), and 30 μl Fugene 6 Transfection Reagent (Promega E2691) with 570 μl OptiMEM-I Reduced Serum Medium (Thermo Fisher Scientific #31985070) per plate. Transfection mixtures were vortexed, incubated for 15 min, and added drop-wise to plates containing 10 mL media. Media was changed 16 hours after transfection. 60 hours later, media containing viral particles was collected. Debris was pelleted at 800 g for 10 min (4°C), and the supernatant was passed through a 0.45 μm syringe filter (Millipore SLHV033RS). Viral particles were concentrated by ultracentrifugation at 25,000 rpm for 1.5 hr (4°C) with an SW-32 swinging bucket rotor (Beckman Coulter), and the viral pellet was resuspended in pure DMEM for 1/100^th the original supernatant volume. Viral titer was determined by immunostaining for FOXA3, HNF1A, and HNF4A in fibroblasts three days after infection with serial dilutions of concentrated virus. Dilutions of virus that produced 10–35% transgene-expressing cells were used to calculate the multiplicity of infection (MOI), and in turn the titer, using the relationship MOI = (−1) * ln(1 - [proportion infected]).

hiHep reprogramming

Transdifferentiation of fibroblasts to hiHep cells was carried out as previously described (Huang et al., 2014). BJ fibroblasts growing in complete EMEM (see above) were plated on collagen I-coated plates at a density of 3 × 10⁴ cells per well in 12-well format. One day after plating, cells were infected with a cocktail of three lentiviruses (pWPI.1-FOXA3, pWPI.1-HNF1A, and pWPI.1-HNF4A), with an MOI of 1.25 per virus, in media containing 4.5 μg/ml polybrene. One day after infection, virus-containing media was removed, cells were washed twice with PBS, and fresh complete EMEM was added. On the second day after infection, medium was switched to Hepatocyte Maintenance Medium (HMM): DMEM/F12 (Hyclone SH30023.01) supplemented with 0.544 mg/L ZnCl₂, 0.75 mg/L ZnSO₄·7H₂O, 0.2 mg/L CuSO₄·5H₂O, 0.025 mg/L MnSO₄·1H₂O, 2 g/L Bovine serum albumin (Sigma-Aldrich), 2 g/L galactose (Sigma-Aldrich G0750), 0.1 g/L ornithine (Sigma-Aldrich O2375), 0.03 g/L proline (Sigma-Aldrich P5607), 0.61 g/L nicotinamide (N0636), 1X Insulin-transferrin-sodium selenite media supplement (Sigma-Aldrich I1884), 40 ng/ml TGF-α (Peprotech AF-100-16A), 40 ng/ml EGF (Peprotech AF-100-15), 10 μM dexamethasone (Sigma-Aldrich D4902), and 1% FBS (Hyclone SH30071). Media was changed with fresh HMM every 48 hours. Cells were analyzed for hepatic markers at 10–14 days after lentiviral infection.

siRNA transfection experiments

All knockdown experiments were performed with two cycles of siRNA transfection, three days apart. The following Silencer Select siRNAs from Thermo Fisher Scientific were used: Negative Control No. 1 siRNA (#4390843), SUV39H1 siRNA (s13660), RBMX siRNA “#1” (s56033), RBMX siRNA “#2” (s56035), and RBMX siRNA “#3” (s223747). Transfections were performed using 10 nM final concentration of siRNA and 2.5 μl/ml final concentration Lipofectamine RNAiMAX (Thermo Fisher Scientific #13778150). 10X transfection mixtures were prepared by adding 100 nM siRNA and 25 μl/ml Lipofectamine to OptiMEM-I Reduced Serum Medium (Thermo Fisher Scientific #31985070) and incubating for 10 min at room temperature. On day 0, cells were reverse transfected by seeding a 0.9X volume of cells/media into wells containing 0.1X volume of 10X transfection mixture. After 24 hours, siRNA-containing media was replaced with normal culture media. After 3 days, forward transfections were performed on adherent cells by adding 0.1X volume of 10X transfection mixture drop-wise to 0.9X volume of culture media. Media was again replaced 24 hours after transfection. RNA or protein was harvested on day 6, three days after the second siRNA transfection. For experiments performed with uninfected, proliferating fibroblasts, cells were passaged over the course of this six-day time-course, as needed to prevent over-growth.

Knockdown experiments in cells expressing hepatic transcription factors, including the siRNA screen presented in Figure 6 (see Table S6 for list of siRNAs used), were performed as follows. First, on day −1, cells were infected as a batch (single well or plate) with pWPI.1-FOXA3, pWPI.1-HNF1A, and pWPI.1-HNF4A lentiviruses (MOI of 1.25 per virus), in media containing 4.5 μg/ml polybrene. 24 hours after infection, on day 0, virus-containing media was removed, cells were washed twice with PBS, and cells were detached from the plate and split for reverse 10 nM siRNA transfection. The remainder of the six-day siRNA time-course was performed as described as above, except that, on day 4, media was changed to HMM (see above under “hiHep reprogramming”) to promote hepatic induction. By infecting cells with factors prior to splitting for transfection with specific siRNAs, we ensure similar dosage of factors across the conditions being compared.

RNA isolation and Reverse Transcription-PCR

Total RNA was isolated using the ZR-96 Quick-RNA kit (Zymo Research, R1052), which includes an on-column DNase I treatment step, and eluted in 30 μL RNase-free dH₂O. cDNA was prepared using the High Capacity cDNA Reverse Transcription Kit (Thermo Fisher #4368814). For detection of transcripts from liver genes in fibroblast heterochromatin, TaqMan-based quantitative PCR was performed using the TaqMan Gene Expression Master Mix (Thermo Fisher #4369016), and data was normalized to GAPDH (for siRNA screen in Figure 6, S5C), or an average of GAPDH and 18SRNA endogenous controls (all other experiments). See Key Resources Table for TaqMan primers/probes used. For verification of siRNA knockdown efficiency, qPCR was performed using Power SYBR Green PCR Master Mix (Thermo Fisher #4367659), and data was normalized using the GAPDH primer as an endogenous control (see primers listed below). Both SYBR and TaqMan qPCR reactions were run in 384-well format on an 7900HT Real-Time PCR machine (Thermo Fisher #4329001), using the following thermal cycler protocol: 50°C for 2 min, 95°C for 10 min, followed by 45 cycles of 95°C for 15 s then 60°C for 1 min. For SYBR-based qPCR reactions, a dissociation curve was generated to verify that a single PCR product was generated.

Primers for SYBR-based RT-PCR experiments to verify knockdown efficiency:

transcript	forward primer (5′ - 3′)	reverse primer (5′ - 3′)
hGAPDH	CCAGGTGGTCTCCTCTGACTTC	TCATACCAGGAAATGAGCTTGACA
hSUV39H1	GTCATGGAGTACGTGGGAGAG	CCTGACGGTCGTAGATCTGG
hSUV39H2	TCGATACGGCAATGTGTCTC	ACAATGCTATTCGGGGAAGA
hRBMX	CAGTTCGCAGTAGCAGTGGA	TCGAGGTGGACCTCCATAA
hRBMXL1	AGCAGCTCACGTGATGGATA	GATCACTTCGGCTGCTTGAG

Flow cytometry studies

Two biological replicates of day 14 hiHep cells for each treatment condition were washed with PBS and dissociated from the plate into a single cell suspension with Accutase (Stem Cell Technologies) at 37°C for 5 minutes. Cells were washed twice with PBS and fixed in 4% paraformaldehyde in PBS for 15 minutes at room temperature. Cells were permeabilized with ice-cold 0.1% Triton in PBS for 10 minutes. Staining was performed in blocking solution for 1 hour at room temperature with primary antibodies anti-human-albumin (1:100, Bethyl Laboratories A80-229A), anti-alpha-1-antitrypsin (1:200, Thermo Fisher Scientific RB-367). Cells were washed three times with PBS then incubated with AlexaFluor 488- or 647-conjugated secondary antibodies raised in donkey (Thermo Fisher Scientific). Cells were washed three times with PBS, resuspended in water then analyzed on an Accuri C6 and data were collected for all cells. All gating and quantification of populations was performed in FlowJo (V10.1). Initial FSC/SSC gating was performed and applied uniformly to all samples. For all antibodies used the boundaries between positive and negative staining was established using a secondary antibody only control and the gating strategy was applied uniformly across all samples for quantification.

Preparation of polyA-selected RNA libraries

Two biological replicates were sequenced per experimental condition. Purified total RNA was diluted to 50 μl in BTE buffer (10 mM Bis-tris, pH 6.7, 1 mM EDTA), denatured by heating at 65°C for 5 minutes, and place immediately on ice. Oligo(dT)₂₅ Dynabeads (Thermo Fisher Scientific #61002) were washed three times in 2x Oligo-dT Binding Buffer (2xOBB: 20 mM Tris, pH 7.5, 1 M LiCl, 2 mM EDTA), resuspended in 50 μl 2xOBB, and mixed with an equal volume of denatured RNA. RNA and beads were incubated at room temperature for 10 min, shaking. Beads were washed three times with Oligo-dT Washing Buffer (10 mM Tris, pH 7.5, 150 mM LiCl, 1 mM EDTA), and eluted in 10 μl BTE buffer by heating at 80°C for 2 min. Strand-specific cDNA libraries were generated using the NEBNext Ultra Directional RNA Library Prep Kit for Illumina (New England Biolabs, E7420S). This protocol includes a heat-based mRNA fragmentation step (15 min at 94°C in first-strand cDNA synthesis buffer), and actinomycin D (Sigma-Aldrich A1410) is added during first-strand cDNA synthesis to inhibit DNA-dependent DNA polymerase activity and prevent template switching. Adapter-ligated cDNAs were amplified by 12 cycles of PCR with NEBNext Multiplex Oligos for Illumina (New England Biolabs E7335S and E7500S). Library yield and fragment size distribution was assessed on a Bioanalyzer 2100 instrument (Agilent Technologies), using the DNA 1000 kit (Agilent Technologies 5067-1504).

Next-generation sequencing

Libraries were quantified by qPCR using the KAPA Library Quantifcation Kit for Illumina (KAPA Biosystems KK4824). Libraries were diluted to 8 nM concentration and pooled for multiplexing, and then their diluted concentrations were checked a second time using the KAPA kit, adjusting as necessary. Diluted libraries were denatured in 0.2 M NaOH, loaded into the cartridge of the NextSeq 500/550 High Output v2 kit (Illumina FC-404-2005, 75 cycles) at a concentration of 3.2 pM in the kit’s Hybridization Buffer, and sequenced in an Illumina NextSeq 500 machine.

QUANTIFICATION AND STATISTICAL ANALYSIS

Tests of statistical significance

Repeated measurements being compared between two samples were analyzed for significance by two-tailed Student’s T-test. Distributions of unequal sizes, such as gene expression values for two different sets of genes, were compared by Wilcoxon rank sum test, implemented in R using the wilcox.test() function (paired=FALSE). Paired distributions (equal size) were analyzed by Wilcoxon signed rank test, using R’s wilcox.test() function (paired=TRUE). Correction for multiple comparisons, where noted, were performed using the Benjamini-Hochberg procedure, implemented via DAVID Bioinformatics tools or the p.adjust() function in R, and a 5% FDR cutoff was applied. For testing the significance of overlaps between two sets drawn from a common pool, the hypergeometric test was used, implemented with the dhyper() function in R.

Alignment and visualization of Gradient-seq and ChIP-seq data

For sequencing data generated in this study, sequencer output was demultiplexed (bcl2fastq using BaseSpace) to produce FASTQ files for individual samples. Sequenced reads were aligned to the hg19 genome assembly using bowtie2 v2.1.0 (parameters: –very-sensitive). Bowtie output files were converted to .bam files using samtools v1.1, and then to .bed files using bedtools v2.20.1 (bamtobed). For each sample, reads mapping to the same genomic position (duplicate reads) were collapsed to a single entry (unique reads). 75-bp sequencing reads were extended to 200 bp, to match the average size of DNA prior to library preparation. Meanwhile, ChIP-seq data obtained from public consortia (see table below) was downloaded from GEO as aligned, unique reads.

To generate input-normalized genome coverage tracks, BED files were converted to BedGraph files using genomeCoverageBed (bedtools v2.20.1) and normalized to the number of millions of reads sequenced (rpm), to correct for lane or sample biases. For each sample’s normalized BedGraph, the normalized BedGraph for the corresponding input sample was subtracted on a basepair-by-basepair basis. The resulting subtracted BedGraph was converted to a bigWig file using bedGraphToBigWig (v4).

Calling enriched genomic domains

All code used for calling domains is available upon request. Enriched genomic domains were called using a 10-kb sliding window algorithm, with a sliding step of 500 bp, using custom scripts. For every 10-kb window in the genome, the number of reads falling in that window for a given sample (normalized to the number of reads sequenced) was divided by the number of reads in that window for the corresponding input file (also normalized to the number of reads sequenced). Divide-by-zero errors were avoided by spiking in a small value into both numerator and denominator (0.25 reads per million reads sequenced). Enriched domains were initially formed by taking all 10-kb windows whose signal-over-input value exceeds a threshold, after averaging all replicates. For human H3K9me3 ChIP-seq (produced by our lab or the Epigenomics Roadmap), the signal-over-input values form a bimodal distribution; consequently, we used kmeans clustering (via the kmeans() function in R) to find the partition between the “unenriched” and “enriched” modes of the distribution. This led to the selection of 1.23 as the threshold for the BJ fibroblast H3K9me3 ChIP-seq produced in this study and 1.41 as the threshold for the foreskin fibroblast H3K9me3 ChIP-seq produced by the Epigenomics Roadmap – values that reflect the different dynamic ranges of the two datasets and yield highly concordant domain maps (86% overlap). Similarly, kmeans clustering was used to select a threshold of 1.30 for the human liver H3K9me3 ChIP-seq from the Epigenomics Roadmap. For other genomic datasets, fixed enrichment thresholds were applied (instead of kmeans clustering) to ensure fair comparisons. For example, the srHC sequencing has similar dynamic range as our H3K9me3 ChIP-seq data, and thus the same threshold of 1.23 was applied. For H3K27me3 ChIP-seq data from the Epigenomics Roadmap, the same threshold was used as the Roadmap H3K9me3 data (1.41). Rates of overlap between these datasets were similar across a wide range of chosen threshold values and were further corroborated by threshold-free correlation analyses.

Once enriched 10-kb windows were chosen, all adjacent enriched 10-kb windows were merged into contiguous domains. To increase the local resolution of the domain calls, an edge-pruning step was applied, whereby 500-bp steps were removed from either end of each domain until a 500-bp step is encountered with a signal-over-input value of at least 1.2. (This step serves to eliminate “overhangs” where a local region of strong signal causes an entire 10kb window to be called as enriched. Empirically, this pruning step removes 5% or less of the total domain coverage.) Finally, any remaining overlaps among domains were merged, to produce the final domain calls.

For calling euchromatin domains, a small modification was made because of the close similarity of the euchromatin fraction (containing the majority of chromatin fragments) to the gradient input. To increase contrast for domain-calling, the euchromatin signal was normalized to the srHC signal (instead of the input). A threshold of 1.2 was applied, based on the bimodal distribution of euchromatin-over-srHC values. The remainder of the domain-calling procedure was as described above.

Intermediate signal domains were defined by taking all regions outside of srHC and euchromatin domains that also had sequencing signal in both gradient replicates. Thus, uninformative regions without sequencing data were not assigned a sonication-resistant type.

The srHC subtype of H3K9me3 or H3K27me3 domains are simply regions of overlap between the srHC domains and histone mark domains. Among the remaining regions for each histone mark, the euchromatic subtype was called by selecting regions where the exact same 10-kb window met enrichment criteria for both euchromatin and the histone mark. (In other words, it was not sufficient for there merely to be overlap between one euchromatic 10-kb window and a nearby 10-kb window enriched for the histone mark; both properties had to co-occur in the same window, ensuring the stringency of the euchromatic calls.) Finally, remaining regions enriched for the histone mark that were neither srHC nor euchromatic were called as intermediate.

Correlation analysis of Gradient-seq and ChIP-seq datasets

As for domain calling (see above), the genome was divided into 10-kb sliding windows (500-bp sliding step). For each window, the number of reads for the genomic sample of interest (normalized to the number of reads sequenced) was divided by the number of reads in that window for the corresponding input file (also normalized to the number of reads sequenced). As above, divide-by-zero errors were avoided by spiking in a small value into both numerator and denominator (0.25 reads per million reads sequenced). Using the cor() function in R, we computed the pairwise Spearman correlation among the samples in terms of their signal-over-input values, across all 10-kb windows genome-wide. Spearman correlation values were used to construct a heatmap using the “pheatmap” package in R. The dendrogram distances and clustering were set according to dissimilarity, where dissimilarity = [ 1 - (Spearman correlation) ]. Similar correlations and identical clustering were obtained by Pearson correlation.

Defining genes sets overlapping chromatin domains

The hg19 Refseq gene table was downloaded from the UCSC Table Browser. Refseq genes were defined as “inside” a chromatin domain of interest if at least 50% of the gene body overlapped that domain class. If there were multiple overlaps of the Refseq gene with different domains in that category, these overlaps were summed together, and the gene was said to overlap the domains as long as the sum met the 50% cutoff. Many Refseq genes have the same gene symbol; for analyses at the gene symbol level (such as gene expression by mRNA-seq), genes symbols were said to overlap the domain if any of their associated Refseq genes met the 50% cutoff. Results regarding gene expression in chromatin domain types were highly invariant to the percentile cutoff chosen.

Analysis of microarray data for hiHep and hiCN reprogramming

Microarray data for human fibroblasts, cultured human hepatocytes, hiHep cells, and immortalized hiHeps were downloaded from (Huang et al., 2014) (GSE42643). The microarray data was analyzed and quantile-normalized using the Partek Genomics Suite. For Refseq genes with multiple probes, only the probe with the highest variance across all samples was used. The full microarray was filtered down to genes expressed significantly higher in normal hepatocytes compared to fibroblasts (P < 0.05, at least 2-fold). For analyses specifically related to silent genes in fibroblasts (eg, Figure 1A), the list was further reduced to genes expressed in the bottom 40% among in fibroblasts, which corresponded to the lower mode of a biomodal distribution. Log2-normalized gene expression for hiHep cells was calculated on a relative scale, with 0% representing the log2-normalized fibroblast expression, and 100% representing the log2-normalized hepatocyte expression. Negative values (hepatic genes expressed lower in hiHeps than fibroblasts) were rounded up to 0%. Expression values on this scale were plotted as violin plots using a modified version of the vioplot package in R.

For analysis of gene expression in human induced cholinergic neurons (hiCNs) (Liu et al., 2013), quantile-normalized microarray data was downloaded from GSE45954 for IMR90 lung fibroblasts, spinal cord tissue, and IMR90-derived hiCN cells. Analysis was performed as for hiHep cells, except that genes were filtered for those expressed at least 2-fold higher in spinal cord than fibroblasts (“neural genes”), genes downregulated during hiCN reprogramming were removed from consideration, and the relative expression scale was such that 100% corresponded to the log2-normalized expression in spinal cord.

Computing histone mark enrichment over domains

Each domain was divided into 100 equally sized bins. The read pileup for a given histone mark was counted per bin, and normalized for sequencing depth (number of millions of reads sequenced) and the length of the bin in kb. For each bin in each domain, the results were averaged across all replicate ChIP-seq datasets for that mark, and then the results for the corresponding input samples (same number of replicates, also normalized for sequencing depth) were subtracted. This yielded a topography of histone mark enrichment or depletion across each individual domain, broken into a vector of 100 values for the 100 bins. An average of these vectors was then taken across all domains in the genome, weighted by domain length. Histone mark ChIP-seq data for this analysis was obtained from the resources below (the Roadmap data was downloaded as aligned reads, while the data from Chandra et al, 2012, was downloaded as raw reads and aligned using bowtie):

ChIP type	Cell type	Source	GEO accession #
H3K4me1	foreskin fibroblasts	Roadmap	GSM817234, GSM941717, GSM958164
H3K4me3	foreskin fibroblasts	Roadmap	GSM817235, GSM941718, GSM958158
H3K9me3	foreskin fibroblasts	Roadmap	GSM817236, GSM817239
H3K27ac	foreskin fibroblasts	Roadmap	GSM1127076, GSM1127060, GSM958163
H3K27me3	foreskin fibroblasts	Roadmap	GSM817237, GSM817240, GSM958154
H3K36me3	foreskin fibroblasts	Roadmap	GSM817238, GSM817241, GSM958149
Input	foreskin fibroblasts	Roadmap	GSM817246, GSM817247, GSM958168
H3K4me2	IMR90 fibroblasts	Roadmap	GSM521899, GSM521900
H3K9ac	IMR90 fibroblasts	Roadmap	GSM469973, GSM521912
H3K9me1	IMR90 fibroblasts	Roadmap	GSM752986, GSM752987
H3K79me1	IMR90 fibroblasts	Roadmap	GSM521904, GSM521906, GSM521907, GSM521908
H3K79me2	IMR90 fibroblasts	Roadmap	GSM521909, GSM521911
H4K20me1	IMR90 fibroblasts	Roadmap	GSM521915, GSM521917
Input	IMR90 fibroblasts	Roadmap	GSM521926, GSM521927, GSM521928, GSM521929, GSM521930, GSM521931, GSM521932, GSM521933
H3K9me2	IMR90 fibroblasts	Chandra et al, 2012 (Chandra et al., 2012)	GSM942082, GSM942084
Input	IMR90 fibroblasts	Chandra et al, 2012 (Chandra et al., 2012)	GSM942119
H2AK5ac	IMR90 fibroblasts	Roadmap	GSM521866, GSM521868
H2AK9ac	IMR90 fibroblasts	Roadmap	GSM818012, GSM818013
H2BK5ac	IMR90 fibroblasts	Roadmap	GSM818017, GSM832837, GSM832838
H2BK12ac	IMR90 fibroblasts	Roadmap	GSM521871, GSM521873, GSM521874
H2BK15ac	IMR90 fibroblasts	Roadmap	GSM521875, GSM521877, GSM521878
H2BK20ac	IMR90 fibroblasts	Roadmap	GSM521879, GSM521880
H2BK120ac	IMR90 fibroblasts	Roadmap	GSM521869, GSM521870
H3K4ac	IMR90 fibroblasts	Roadmap	GSM521893, GSM521894
H3K14ac	IMR90 fibroblasts	Roadmap	GSM521881, GSM521883
H3K18ac	IMR90 fibroblasts	Roadmap	GSM469965, GSM521884
H3K23ac	IMR90 fibroblasts	Roadmap	GSM521885, GSM521886
H3K56ac	IMR90 fibroblasts	Roadmap	GSM521902, GSM521903
H4K5ac	IMR90 fibroblasts	Roadmap	GSM469975, GSM521918
H4K8ac	IMR90 fibroblasts	Roadmap	GSM521919, GSM521921, GSM521922, GSM521923
H4K91ac	IMR90 fibroblasts	Roadmap	GSM521924, GSM521925

Analysis of repetitive elements enriched in chromatin domains

The Repeat Masker table for hg19 was downloaded from the UCSC table browser and was intersected with BED files listing non-overlapping domains, using Bedtools. For each type of repeat, the total number of base pairs falling within domains was computed as a percentage of the total genomic coverage of that repeat. This analysis was performed for each repeat class, repeat family, and individual repeat name listed in the Repeat Masker table. P-values were determined by permutation test, using 1000 simulations of domains randomly shuffled across the genome by Bedtools (preserving domain number and size). For each repeat, the P-value was the proportion of domain simulations that produced an equal or greater overlap with that repeat. To correct for multiple hypothesis testing, P values were used to calculate the FDR using the Benjamini-Hochberg procedure. Enriched repeats with an FDR < 0.05 were treated as statistically significant.

Analysis of DNA methylation in chromatin domains

Processed whole-genome bisulfite sequencing (WGBS) data for human foreskin fibroblasts was downloaded from the Roadmap Epigenomics Mapping consortium (GSM1127120) (Bernstein et al., 2010). This data file reports the percent methylation per CpG. CpGs were then divided into two categories based on whether they fell in annotated CpG islands (UCSC Table Browser). Both categories of CpG sites (inside and outside of CpG islands) were intersected with chromatin domains of interest, and the distribution of percent-methylation values was plotted.

Analysis of datasets for Lamin B1 ChIP-seq, DNase-seq, FAIRE-seq, and MNase sensitivity

Each dataset was analyzed so as to obtain values per 10-kb sliding genomic window (500-bp slide), which could then be plotted for chromatin regions of interest. Lamin B1 ChIP-seq reads and input reads for IMR90 fibroblasts (Dou et al., 2015) (two replicates) were downloaded from GSE63440, aligned using bowtie2 v2.1.0 (parameters: –very-sensitive), and extended to 200 bp. PCR duplicates were collapsed to unique reads. Using custom scripts, the enrichment of Lamin B1 ChIP reads to input reads (after normalizing for sequencing depth) was computed over 10-kb genomic windows, with a 500-bp sliding step, and values were averaged across the two replicates.

ENCODE DNase-seq data for BJ fibroblasts (two replicates) were downloaded as aligned reads from GSM736518 and GSM736596 (Thurman et al., 2012). The pileup of aligned reads was computed per 10-kb window (500-bp sliding step) and normalized based on the number of millions of reads sequenced.

FAIRE-seq reads for hTERT-immortalized foreskin fibroblasts (two replicates) were downloaded from GSM1898780 and GSM1898783 (Torres et al., 2016) and aligned using STAR. PCR duplicates were collapsed to unique reads. The pileup of aligned reads was computed per 10-kb window (500-bp sliding step) and normalized based on the number of millions of reads sequenced.

MACC scores, which represent sensitivity to MNase from an enzymatic titration, were downloaded for human K562 cells from GSE78984 (Mieczkowski et al., 2016). MACC scores were published at 500-bp resolution, and these scores were converted to 10-kb sliding window scores (500-bp slide) by averaging the component 500-bp values within each 10-kb window. During this averaging, missing values were ignored (rather than counting them as zero), and data was not reported for the 10-kb window if fewer than five of the constituent 500-bp bins contained data.

To plot each of these datasets over chromatin domains of interest, the chromatin domains were converted into the set of non-overlapping 10-kb windows that they contain (windows that start and stop on multiples of 500 bp). These 10-kb windows were then matched to their corresponding values for Lamin B1 enrichment, DNase/MNase-sensitivity, and FAIRE signal.

Analysis of DNA replication timing data

ENCODE Repli-seq data for BJ fibroblasts (Pope et al., 2014) were downloaded as aligned reads from ENCSR894LZX. These files contain sequencing of newly replicated DNA in six cytometry-fractionated cell populations, with two replicates per timepoint. The downloaded alignments were intersected with each domain BED file using bedtools, and the number of intersecting reads was divided by the number of millions of reads in the sequencing file. These normalized intersection scores were then averaged between the two sequencing replicates per timepoint, and then they were expressed as a fraction of the sum of the intersection scores across the six timepoints. Thus, the final values for each domain type sum to 1.0 across the six cell cycle fractions.

Analysis of microarray data for ALS patient-derived fibroblasts

Exon microarray data were downloaded from GSE33855 (Highley et al., 2014) and quantile-normalized with median polish using the Partek Genomics Suite. This dataset includes fibroblasts from ALS patients, some with germline mutations and some with sporadic disease, in addition to healthy controls. To extract whole gene-level expression from exon-directed probes (multiple per gene), we analyzed the microarray as described in the original publication (Highley et al., 2014): exon-specific probes were removed from consideration if their log2-normalized signal (averaged across all biological samples) was 3 standard deviations above or 1 standard deviation below the average signal for all the probes for that gene. This served to remove probes with nonspecific hybridization or affected by alternative splicing, respectively. An average was then taken of the remaining probes for each gene, for each patient sample. These log2-transformed were then converted back to a linear scale and averaged across the patients for a given genotype – TARDBP-mutated (n=3), sporadic ALS (n=6), or control (n=6) – in order to calculate, for each gene, the fold-change versus control.

Alignment and visualization of mRNA-seq data

Sequencer output was demultiplexed (bcl2fastq using BaseSpace) to produce FASTQ files for individual samples. Sequenced reads were aligned to the hg19 Refseq gene model, in a strand-specific manner, using TopHat v2.0.11 (parameters: –b2-very-sensitive –library-type fr-firststrand). To generate genome coverage tracks, BED files were first pooled between biological replicates. Reads aligning to genomic regions longer than the sequencing length (75 bp), due to spanning of splice junctions, were discarded for genomic visualization purposes. Pooled BED files were then converted to BedGraph files using genomeCoverageBed (bedtools v2.20.1) and normalized to the number of millions of reads sequenced (rpm), to correct for lane or sample biases. The resulting normalized BedGraphs were converted to bigWig files using bedGraphToBigWig (v4).

Gene expression analysis for mRNA-seq data

Sequenced reads were assigned in a strand-specific manner to genes from the hg19 Refseq table using HTSeq v0.6.1 (parameters: –stranded=reverse –mode=intersection-nonempty –type=exon -i gene_id), such that all exons for all Refseq genes with the same official gene symbol were considered to belong to the same feature. The unnormalized HTSeq tables, after removal of lines for unassigned reads, were analyzed by the DESeq2 v1.11.45 package in R to produce normalized count tables. All samples for both the “hepatic-TF” and “no-TF” RNA-seq studies were normalized in DESeq2 at the same time. For quantification of fibroblast gene expression across chromatin categories, the DESeq2-normalized counts for the control siRNA-transfected fibroblasts in the no-TF condition were used, after dividing each value by the length of that gene’s exon model in kb and averaging the two biological replicates. Differentially expressed genes were determined in a pairwise manner using DESeq2, with a significance cutoff of p.adjust < 0.05. For analysis of genes upregulated by siRNAs compared to control siRNA in the “hepatic TF” condition, genes significantly downregulated by hepatic TFs alone (hepatic-TF control siRNA compared to no-TF control siRNA) were removed from consideration – 3,487 out of 26,839 total genes. This was to ensure that upregulated genes could be interpreted as being truly upregulated by the siRNA, rather than the siRNA treatment inhibiting the downregulation of the gene by the hepatic TFs. For mRNA-seq sample correlation analyses, a regularized log2-transformation was performed on the normalized counts using DESeq2, and Euclidean distances were calculated using the dist() function in R and visualized as a heatmap/dendrogram using the “pheatmap” package.

Defining protein sets enriched in H3K9me3 heterochromatin and gradient top

Intensity-based absolute quantification (iBAQ) values were ranked in order of their iBAQ score for each of three biological replicates of each sample, to facilitate comparison among samples with very different numbers of identified proteins. Technical replicate MS runs were averaged. Proteins with only a single detected peptide were removed from consideration, and common contaminants (immunoglobulin chains, skin keratins) were filtered out. Protein ranks were compared between samples by two-tailed Student’s T test. Proteins with a significantly higher rank (P<0.05) in the srHC+H3K9me3 sample (H3K9me3-directed IP using srHC fraction chromatin) compared to the Gradient Top fraction were used to define the “H3K9me3 heterochromatin proteins.” Proteins detected in srHC+H3K9me3, but without significant enrichment or depletion compared to the Gradient Top sample, were used to define the “shared” proteins. Meanwhile, for the srHC whole fraction sample, proteins with significantly higher rank in the srHC fraction compared to the Gradient Top fraction were defined as “srHC-enriched.” Finally, remaining proteins were classified as “Gradient Top” proteins if they were either significantly enriched in the Gradient Top fraction over srHC+H3K9me3, significantly enriched in Gradient Top over srHC, or unique to the Gradient Top fraction. Note that proteins not detected in the Gradient Top fraction were assigned a rank equal to the total number of proteins detected for that replicate; thus, proteins detected uniquely in srHC+H3K9me3 or the srHC fraction could still be quantified as significantly enriched. Note also that the significance by T-test requires at least two values per sample being compared, and thus the “H3K9me3 heterochromatin” and “srHC-enriched” proteins by definition had to have been detected in at least two out of three replicates of the srHC+H3K9me3 and srHC samples, respectively.

Protein interaction network analysis

Protein interaction datasets were downloaded from the STRING v10 database (string-db.org). For all downstream analysis, only experimentally determined direct-binding interactions were considered from these datasets. All direct-binding interactions identified in these datasets between proteins determined by our mass-spectrometry as enriched in a specific fraction (Table S5) were utilized to generate a network of direct interactions. Dijkstra’s algorithm for identifying the shortest distance on a graph between two nodes was utilized to calculate the minimum pairwise distance between all proteins. Hierarchical clustering was performed on the proximity matrix composed of pairwise minimum distances between proteins.

Image quantification

Image quantification was performed using MATLAB (version 9.0.0.34) and the image processing toolbox (version 2.2.23). To define the nuclear region, images taken in the DRAQ5 channel were first converted to grayscale, contrast adjusted using the included adapthisteq function and thresholded using Otsu’s method as implemented in the included graythresh function. The thresholded data was further processed using the included imfill and imopen functions before removing objects containing fewer than 40 pixels using the bwareaopen function. Nuclear perimeter was defined by running the included function bwperim on the processed image. Quantification of H3K9me3 and RBMX signal was performed for each cell by calculating the average pixel intensity in the corresponding channel for the region defined as the nuclear perimeter based upon the DRAQ5 channel perimeter call.

DATA AND SOFTWARE AVAILABILITY

All custom scripts used for domain-calling and other computational analyses are available at request. All next-generation sequencing data generated by this study (FASTQ files of sequenced reads and bigwig files for browser visualization) were uploaded to GEO accession number GSE87041. All proteomic raw data files are available on the Chorus database under Project ID 1172, Experiment ID 2585.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit polyclonal anti-Histone H3 (tri methyl K9)	Abcam	ab8898
Rabbit polyclonal anti-Histone H3K27me3	Active Motif	#39155
Rabbit polyclonal anti-Histone H3	Abcam	ab1791
Rabbit polyclonal anti-mouse IgG H&L	Abcam	ab46540
Rabbit monoclonal anti-RBMX	Cell Signaling Technology	#14794
Goat polyclonal anti-RBMX	Santa Cruz Biotechnology	sc-14581
Rabbit polyclonal anti-SUV39H1	Bethyl Laboratories	A302-127A
Mouse monoclonal anti-GAPDH	Santa Cruz Biotechnology	sc-365062
Goat polyclonal anti-Human Albumin	Bethyl Laboratories	A80-229A
Rabbit polyclonal anti-Alpha-1-Antitrypsin	Thermo Fisher Scientific	RB-367
Goat polyclonal anti-FOXA3/HNF3-γ	Santa Cruz Biotechnology	sc-5361
Goat polyclonal anti-HNF1A	Santa Cruz Biotechnology	sc-6547
Rabbit polyclonal anti-HNF4A	Santa Cruz Biotechnology	sc-8987
Goat anti-rabbit IgG-HRP	Santa Cruz Biotechnology	sc-2004
Goat anti-mouse IgG-HRP	Santa Cruz Biotechnology	sc-2005
Donkey anti-Goat IgG (H+L), Alexa Fluor 594 conjugate	Thermo Fisher Scientific	A-11058
Donkey anti-Rabbit IgG (H+L), Alexa Fluor 594 conjugate	Thermo Fisher Scientific	A-21207
Donkey anti-Rabbit IgG (H+L), Alexa Fluor 488 conjugate	Thermo Fisher Scientific	A-21206
Chemicals, Peptides, and Recombinant Proteins
Fetal bovine serum, characterized	Hyclone	SH30071
Protease Inhibitor Cocktail (cOmplete, EDTA-free)	Roche	#11873580001
RNase A	Roche	#10109169001
Proteinase K	Roche	#03115828001
Glycogen	Roche	#10901393001
4X NuPAGE LDS Sample Buffer	Thermo Fisher Scientific	NP0007
10X NuPAGE Sample Reducing Agent	Thermo Fisher Scientific	NP0009
20X NuPAGE MOPS SDS Running Buffer	Thermo Fisher Scientific	NP0001
20X NuPAGE MES SDS Running Buffer	Thermo Fisher Scientific	NP0002
20X NuPAGE Transfer Buffer	Thermo Fisher Scientific	NP0006
OptiMEM-I Reduced Serum Medium	Thermo Fisher Scientific	#31985070
FuGENE 6 Transfection Reagent	Promega	E2691
Lipofectamine RNAiMAX Transfection Reagent	Thermo Fisher Scientific	#13778150
Actinomycin D	Sigma-Aldrich	A1410
Collagen I, rat tail	Corning	#354236
Insulin-transferrin-sodium selenite media supplement	Sigma-Aldrich	I1884
D-(+)-galactose	Sigma-Aldrich	G0750
Nicotinamide	Sigma-Aldrich	N0636
L-ornithine	Sigma-Aldrich	O2375
L-proline	Sigma-Aldrich	P5607
Recombinant human TGF-α	Peprotech	AF-100-16A
Recombinant human EGF	Peprotech	AF-100-15
Dexamethasone	Sigma-Aldrich	D4902
Donkey serum	Sigma-Aldrich	D9663
DAPI	Thermo Fisher Scientific	D1306
DRAQ5	Thermo Fisher Scientific	62251
Accutase	Stem Cell Technologies	07920
Critical Commercial Assays
Dynabead Protein G Beads	Thermo Fisher Scientific	10004D
Quant-iT PicoGreen dsDNA Assay	Thermo Fisher Scientific	P7589
NuPAGE Novex 4–12% Bis-Tris Protein Gels	Thermo Fisher Scientific	NP0335
SimplyBlue SafeStain	Thermo Fisher Scientific	LC6060
ZR-96 Quick-RNA purification kit	Zymo Research	R1052
High Capacity cDNA Reverse Transcription Kit	Thermo Fisher Scientific	#4368814
TaqMan Gene Expression Master Mix	Thermo Fisher Scientific	#4369016
Power SYBR Green PCR Master Mix	Thermo Fisher Scientific	#4367659
Agencourt AMPure XP beads	Beckman Coulter	A63881
NEBNext Ultra DNA Library Prep Kit for Illumina	New England Biolabs	E7370S
NEBNext Multiplex Oligos for Illumina	New England Biolabs	E7335S; E7500S
KAPA Library Quantification Kit for Illumina	KAPA Biosystems	KK4824
DNA 1000 Kit for Bioanalyzer	Agilent Technologies	5067-1504
Dynabeads Oligo(dT)25	Thermo Fisher Scientific	#61002
NEBNext Ultra Directional RNA Library Prep Kit for Illumina	New England Biolabs	E7420S
NextSeq 500/550 High Output v2 kit (75 cycles)	Illumina	FC-404-2005
QIAquick gel extraction kit	Qiagen	28704
EndoFree Plasmid Maxi Kit	Qiagen	#12362
BCA Protein Assay Kit	Thermo Fisher Scientific	#23227
Bradford Protein Assay	Bio-Rad	#500-0006
SuperSignal West Pico Chemiluminescent Substrate	Thermo Fisher Scientific	#34080
Deposited Data
DNA Sequencing of gradient-isolated chromatin samples (Gradient-seq)	This study	GSE87039
Human BJ fibroblast H3K9me3 ChIP-seq	This study	GSE87039
Raw mass spectrometry data files for proteomic analysis of heterochromatin and euchromatin	This study	Chorus Project ID 1172
mRNA-seq in siRNA-transfected BJ fibroblasts, with and without hepatic factor expression	This study	GSE87040
Raw image files	This study	Mendeley Data doi:10.17632/n9b7b6s789.1
Human foreskin fibroblast H3K9me3 ChIP-seq	Roadmap consortium (Bernstein et al., 2010)	GSM817236; GSM817239
Human foreskin fibroblast Input for ChIP-seq	Roadmap consortium (Bernstein et al., 2010)	GSM817246; GSM817247; GSM958168
Human foreskin fibroblast H3K27me3 ChIP-seq	Roadmap consortium (Bernstein et al., 2010)	GSM817237; GSM817240; GSM958154
Differentially Bound Regions (DBRs)	(Soufi et al., 2012)	GSE36570
Human foreskin fibroblast H3K4me3 ChIP-seq	Roadmap consortium (Bernstein et al., 2010)	GSM817235; GSM941718; GSM958158
Human foreskin fibroblast H3K36me3 ChIP-seq	Roadmap consortium (Bernstein et al., 2010)	GSM817238; GSM817241; GSM958149
Human liver H3K9me3 ChIP-seq	Roadmap consortium (Bernstein et al., 2010)	GSM537695; GSM537710; GSM669986
Human liver Input for ChIP-seq	Roadmap consortium (Bernstein et al., 2010)	GSM670008; GSM669910; GSM621629
Human IMR90 H3K9me3 ChIP-seq	Roadmap consortium (Bernstein et al., 2010)	GSM469974
Human IMR90 Input for ChIP-seq	Roadmap consortium (Bernstein et al., 2010)	GSM521926
IMR90 H3K9me2 ChIP-seq	(Chandra et al., 2012)	GSM942082; GSM942084
IMR90 ChIP-seq Input (for H3K9me2 normalization)	(Chandra et al., 2012)	GSM942119
27 histone marks profiled in IMR90 and foreskin fibroblasts for computing enrichment over chromatin domains	(Bernstein et al., 2010)	GSE16368, GSE16256 (see STAR methods for specific samples)
K562 H3K9me3 ChIP-seq	ENCODE Project	GSM733776
K562 H3K27me3 ChIP-seq	ENCODE Project	GSM733658
K562 H3K36me3 ChIP-seq	ENCODE Project	GSM733714
K562 Input for ChIP-seq	ENCODE Project	GSM733780
Percent CpG methylation by whole-genome bisulfite sequencing in human foreskin fibroblasts	Roadmap consortium (Bernstein et al., 2010)	GSM1127120
Lamin B1 ChIP-seq in IMR90 fibroblasts	(Dou et al., 2015)	GSE63440
Lamina Associated Domains in IMR90 fibroblasts	(Guelen et al., 2008)	Supplementary Data 1
DNase I-sequencing in BJ fibroblasts	(Thurman et al., 2012)	GSM736518; GSM736596
MNase titration data (MACC) for K562 cells	(Mieczkowski et al., 2016)	GSE78984
Repli-seq data for BJ fibroblasts	(Pope et al., 2014)	ENCSR894LZX
FAIRE-seq in hTERT-immortalized foreskin fibroblasts	(Torres et al., 2016)	GSM1898780; GSM1898783
Microarray of hiHep cells, cultured hepatocytes, fibroblasts	(Huang et al., 2014)	GSE42643
Microarray of hiCN cells, human spinal cord, IMR90 fibroblasts	(Liu et al., 2013)	GSE45954
Microarray from ALS patient-derived fibroblasts and healthy controls	(Highley et al., 2014)	GSE33855
Loci with decreased H3K9me3 after deletion of HUSH complex/SETDB1, and their H3K9me3 fold-changes	(Tchasovnikarova et al., 2015)	Table S2
Noncoding cancer mutation rate per gene	(Lawrence et al., 2013)	Table S5
H3K9me3 interactors by ChIP-MS	(Soldi and Bonaldi, 2013)	Table S1
H3K9me3 readers by peptide bait pulldown	(Vermeulen et al., 2010)	Table S1
Proteins enriched at murine major satellites	(Saksouk et al., 2014)	Table S1
List of genes recurrently mutated in ALS	(Cirulli et al., 2015)	Table 1
List of human proteins with Tri- and Di-RGG motifs	(Thandapani et al., 2013)	Table S1, S2
Census of 1,542 human RNA-binding proteins	(Gerstberger et al., 2014)	Supplementary information S3 (table)
Effectors and repressors of iPS reprogramming	(Toh et al., 2016)	Table S1
Genes whose knockdown promotes iPS reprogramming	(Qin et al, 2014)	Data S1
Experimental Models: Cell Lines
BJ human fibroblasts (early passage, p6)	Stemgent	08-0027
293T cells		N/A
Recombinant DNA
pWPI.1-FOXA3	(Huang et al., 2014)	N/A
pWPI.1-HNF1A	(Huang et al., 2014)	N/A
pWPI.1-HNF4A	(Huang et al., 2014)	N/A
pMD2.G	Addgene	#12259
psPAX2	Addgene	#12260
Sequence-Based Reagents
Silencer Select Negative Control No. 1 siRNA	Thermo Fisher Scientific	#4390843
SUV39H1 siRNA (Silencer Select)	Thermo Fisher Scientific	s13660
RBMX siRNA (Silencer Select) #1	Thermo Fisher Scientific	s56033
RBMX siRNA (Silencer Select) #2	Thermo Fisher Scientific	s56035
RBMX siRNA (Silencer Select) #3	Thermo Fisher Scientific	s223747
Silencer Select siRNAs used for siRNA screen of 50 proteins	Thermo Fisher Scientific	Table S6
qPCR primers used for detecting DBR and non-DBR sites: see table in STAR Methods	(Soufi et al., 2012)	N/A
qPCR primers for measuring sonication-resistance of gene promoters: see table in STAR methods	This study	N/A
RT-PCR primers for verifying siRNA knockdown efficiency: see table in STAR Methods	This study	N/A
human GAPDH TaqMan primers and probe	Thermo Fisher Scientific	Hs02758991_g1
human RNA18S5 TaqMan primers and probe	Thermo Fisher Scientific	Hs03928990_g1
human FOXA2 TaqMan primers and probe	Thermo Fisher Scientific	Hs00232764_m1
human NR1H4 TaqMan primers and probe	Thermo Fisher Scientific	Hs01026590_m1
human DSC2 TaqMan primers and probe	Thermo Fisher Scientific	Hs00951428_m1
human DSG2 TaqMan primers and probe	Thermo Fisher Scientific	Hs00170071_m1
human ONECUT1 TaqMan primers and probe	Thermo Fisher Scientific	Hs00413554_m1
human CYP2C9 TaqMan primers and probe	Thermo Fisher Scientific	Hs04260376_m1
human CYP2C19 TaqMan primers and probe	Thermo Fisher Scientific	Hs00426380_m1
human SERPINA7 TaqMan primers and probe	Thermo Fisher Scientific	Hs02384980_m1
Software and Algorithms
Bowtie2 v2.1.0		http://bowtie-bio.sourceforge.net/bowtie2/index.shtml
TopHat v2.0.11		https://ccb.jhu.edu/software/tophat/index.shtml
HTSeq v0.6.1		https://pypi.python.org/pypi/HTSeq
DESeq2 v1.11.45		http://www.bioconductor.org/packages/release/bioc/html/DESeq2.html
Bedtools v2.20.1		http://bedtools.readthedocs.io/en/latest/
Samtools v1.1		http://www.htslib.org/
DAVID Bioinformatic Resources		https://david.ncifcrf.gov/
MaxQuant v1.5.2.8	(Cox and Mann, 2008)	http://www.biochem.mpg.de/5111795/maxquant
Partek Genomics Suite	Partek, Inc.	http://www.partek.com/pgs
GraphPad Prism 6.0	GraphPad Software	http://www.graphpad.com/scientific-software/prism/
R v3.3.0	The R Foundation	https://www.r-project.org/
RStudio v0.99.896	RStudio	https://www.rstudio.com/products/rstudio/download3/
MATLAB v9.0.0.34	MathWorks	https://www.mathworks.com/products/matlab.html
Image Processing Toolbox v2.2.23	MathWorks	https://www.mathworks.com/products/image.html
FloJo v10.1	FloJo, LLC	https://www.flowjo.com/
STRING v10 database	STRING Consortium 2017	https://string-db.org/
Other
Bioruptor Sonicator	Diagenode	UCD-200
Gradient maker	Hoefer	SG15
Two-way stopcock	Bio-Rad	#7328102
SW 41 Ti rotor and Bucket Set	Beckman Coulter	#331336
Ultracentrifuge tubes, Thinwall, Ultra-Clear, 14 × 89 mm	Beckman Coulter	#344059
Optima L-90K Ultracentrifuge	Beckman Coulter	#365670
Low-retention microcentrifuge tubes	Axygen	MCT-150-L-C
Slide-A-Lyzer G2 Dialysis Casettes 7K MWCO	Thermo Fisher Scientific	#87727; 87728
EASY-nLC 1000 Liquid Chromatograph	Thermo Fisher Scientific	LC120
Orbitrap Elite mass spectrometer	Thermo Fisher Scientific	IQLAAEGAAPFADBMAZQ
Q-Exactive mass spectrometer	Thermo Fisher Scientific	IQLAAEGAAPFALGMAZR
Amersham Imager 600	GE Healthcare Life Sciences	N/A
Syringe filter unit, 0.45 μm, PVDF	Millipore	SLHV033RS
7900HT Fast Real-Time PCR System with 384-Well Block Module	Thermo Fisher Scientific	#4329001
Accuri C6 flow cytometer	BD Biosciences	C6
Gel Logic 212 Pro	Carestream	212 Pro
S220 Focused Ultrasonicator	Covaris	S220
microTUBE AFA Fiber Pre-Slit Snap-Cap	Covaris	#520045
2100 Bioanalyzer	Agilent Technologies	G2939AA
NextSeq 500 Sequencing System	Illumina	N/A

Supplementary Material

Supplemental figures

Table S1. Gradient-Seq & ChIP-Seq domains; related to Figures 1, 2, and 6

Tabs 1–7: Genomic coordinates in hg19 for all the domain types called in this study: srHC domains, intermediate signal domains, euchromatin domains, and BJ fibroblast H3K9me3 domains (ChIP-seq from this study), foreskin fibroblast H3K9me3 domains (Roadmap consortium data), foreskin fibroblast H3K27me3 domains (Roadmap), and human liver H3K9me3 domains (Roadmap). Domains are listed in BED file format and can be uploaded directly to genome browsers.

Tab 8: List of “H3K9me3-marked Hepatic Genes” used for the plot in Figure 1A. These are RefSeq genes that meet two criteria: they are expressed in cultured hepatocytes and silent in fibroblasts in microarray data from (Huang et al., 2014), and at least 50% of the gene overlaps with H3K9me3 domains for foreskin fibroblasts (Roadmap consortium data, GSE16368).

Tab 9: List of “H3K27-marked Hepatic Genes” used for the plot in Figure 1A. As in Tab 8, but these genes overlap at least 50% with H3K27me3 domains for foreskin fibroblasts (Roadmap data, GSE16368).

Table S2. Features of srHC domains; related to Figures 2 and 3D

Tab 1: List of RefSeq genes with at least 50% overlap with srHC domains. Official gene symbol, gene name, fraction overlap with srHC domains, and whether the transcriptional start site (TSS) falls inside srHC domains, are indicated.

Tab 2: Gene Ontology (GO) analysis for RefSeq genes in srHC domains, searched using the DAVID Bioinformatics tool. The top of the table shows significant GO terms (Benjamini-Hochberg FDR < 0.05), manually filtered to remove redundant terms. All enriched terms (p < 0.10) are listed below.

Tab 3: Annotation of RefSeq genes in srHC domains by enriched InterPro protein domain terms, searched using the DAVID Bioinformatics tool. The top of the table shows significant InterPro terms (Benjamini-Hochberg FDR < 0.05), manually filtered to remove redundant terms. All enriched terms (p < 0.10) are listed below.

Tab 4: For each repeat name from the hg19 Repeat Masker, we report the fraction of the total repeat coverage that falls in srHC domains. Repeats that are significantly enriched in srHC domains (FDR < 0.05, base on 1000 simulations of randomly shuffled domains of same number and size as srHC domains) are highlighted in yellow.

Tab 5: As in tab 4, but for each repeat family in the hg19 Repeat Masker.

Tab 6: As in tab 4, but for each repeat class in the hg19 Repeat Masker.

Table S3. Euchromatic H3K9me3 domains; related to Figure 3

Tab 1: Genomic coordinates in hg19 for the euchromatic subtype of H3K9me3 domains, in BED file format.

Tab 2: Genomic coordinates in hg19 for the srHC subtype of H3K9me3 domains.

Tab 3: Genomic coordinates in hg19 for the intermediate subtype of H3K9me3 domains.

Tab 4: List of RefSeq genes with at least 50% overlap with Euchromatic H3K9me3 domains. Official gene symbol, gene name, fraction overlap with Euchromatic H3K9me3 domains, and whether the transcriptional start site (TSS) falls inside Euchromatic H3K9me3 domains, are indicated.

Tab 5: Annotation of genes in Euchromatic H3K9me3 domains by enriched InterPro protein domain terms, searched using the DAVID Bioinformatics tool. The top of the table shows significant InterPro terms (Benjamini-Hochberg FDR < 0.05), manually filtered to remove redundant terms. All enriched terms (p < 0.10) are listed below.

Tab 6: For each repeat name from the hg19 Repeat Masker, we report the fraction of the total repeat coverage that falls in Euchromatic H3K9me3 domains. Repeats that are significantly enriched in Euchromatic H3K9me3 domains (FDR < 0.05, base on 1000 simulations of randomly shuffled domains of same number and size as Euchromatic H3K9me3 domains) are highlighted in yellow.

Tab 7: As in tab 6, but for each repeat family in the hg19 Repeat Masker.

Tab 8: As in tab 6, but for each repeat class in the hg19 Repeat Masker.

Table S4. Euchromatic H3K27me3 domains; related to Figure 3

Tab 1: Genomic coordinates in hg19 for the euchromatic subtype of H3K27me3 domains, in BED file format.

Tab 2: Genomic coordinates in hg19 for the srHC subtype of H3K27me3 domains.

Tab 3: Genomic coordinates in hg19 for the intermediate subtype of H3K27me3 domains.

Tab 4: List of RefSeq genes with at least 50% overlap with Euchromatic H3K27me3 domains. Official gene symbol, gene name, fraction overlap with Euchromatic H3K27me3 domains, and whether the transcriptional start site (TSS) falls inside Euchromatic H3K27me3 domains, are indicated.

Tab 5: Annotation of genes in Euchromatic H3K27me3 domains by enriched InterPro protein domain terms, searched using the DAVID Bioinformatics tool. The top of the table shows significant InterPro terms (Benjamini-Hochberg FDR < 0.05), manually filtered to remove redundant terms. All enriched terms (p < 0.10) are listed below.

Table S5. H3K9me3 heterochromatin proteins; related to Figure 5

Tabs 1–4: There is one tab for each of the protein classes identified by proteomics, including H3K9me3 heterochromatin proteins (n=172), srHC-enriched proteins (n=217), shared proteins (n=429), and gradient top proteins (n=1,474). For each of these tabs, the fold-enrichment of each protein (by rank) is listed for the srHC+H3K9me3 and srHC samples, compared to the Gradient Top samples, as well as the significance of these comparisons. Columns K-M indicate whether the protein was previously found in association with H3K9me3 or heterochromatin in the following studies (Saksouk et al., 2014; Soldi and Bonaldi, 2013; Vermeulen et al., 2010).

Tab 5 (“All Proteins”): All proteins detected by MS in any sample (n=3,097), along with the sample-specific protein ranks that were used for statistical comparisons (starting in column L).

Tab 6 (“RGG motif proteins”): List of all proteins detected by MS that were in a curated database of Tri-RGG of Di-RGG motif-containing proteins (Thandapani et al., 2013), along with their classification in our study.

Tab 7: (“StringDB Interaction Matrix”): The StringDB protein interaction distance among the 172 H3K9me3 heterochromatin proteins, ordered by hierarchical clustering based on interaction distance (key: self/black cell=0; interaction with another protein/red cell=2; interaction via another protein/orange cell=3, interaction via two other proteins/white cell=4).

Table S6. Functional screen of 50 srHC proteins; related to Figure 6

Tab 1: Fold-upregulation in each of the tested transcripts (DSC2, NR1H4, and CRP) for each siRNA and each replicate of the screen, relative to the average of negative control siRNAs. Asterisks indicate significant upregulation versus control (P < 0.05 by T-test).

Tab 2: List of genes with at least one of two siRNAs causing significantly upregulation of at least one of three heterochromatic transcripts. Also indicated are which transcripts were significantly upregulated.

Table S7. srHC gene regulation by SUV39H1, RBMX/L1; related to Figure 7

Tab 1: List of genes in srHC domains that are significantly upregulated (adjusted p < 0.05) by SUV39H1 siRNA or RBMX/L1 siRNA, or both, in the presence of hepatic transcription factors. Fold-changes relative to control non-targeting siRNA (also with hepatic factors) are shown. Related to Figure 7C.

Tab 2: List of genes in srHC domains that are significantly upregulated by RBMX/L1 siRNA alone (no hepatic transcription factors). Columns indicate the fold-change relative to control siRNA, as well as the normalized expression scores (by DESeq2) for reach sample replicate.

Tab 3: full RNA-seq differential gene expression analysis of RMBX/L1 siRNA compared to control siRNA, in cells expressing hepatic transcription factors. Genes significantly (adjusted p < 0.05) upregulated or downregulated by RBMX/L1 siRNA are indicated in red and blue, respectively.

Tab 4: As in tab 3, but for the RNA-seq experimental condition without introduction of hepatic transcription factors.

Tab 5: Full RNA-seq differential gene expression analysis of SUV39H1 siRNA compared to control siRNA, in cells expressing hepatic transcription factors. Genes significantly (adjusted p < 0.05) upregulated or downregulated by SUV39H1 siRNA are indicated in red and blue, respectively.

Tab 6: As in tab 5, but for the RNA-seq experimental condition without introduction of hepatic transcription factors.