PRC2 proteins EZH1 and EZH2 Regulate Timing of Postnatal Hepatocyte Maturation and Fibrosis by Repressing Gene Expression at Promoter Regions in Euchromatin in Mice

Abstract

BACKGROUND & AIMS:

Little is known about mechanisms of postnatal hepatocyte maturation or chromatin regulation of genes that control fibrogenesis. We investigated transcription of genes that regulate fibrosis and the effects of chromatin compaction and the polycomb repressive complex 2 (PRC2) in postnatal hepatocytes of mice, focusing on the roles of the histone methyltransferases EZH1 and EZH2.

METHODS:

Hepatocytes were isolated from C57BL/6J and C3H mice, as well as mice with liver-specific disruption of Ezh1 and/or Ezh2, at postnatal day 14 (P14) and 2 months after birth (M2). Liver tissues were collected and analyzed by RNA-seq, H3K27me3 chromatin immunoprecipitation-seq, and sonication-resistant heterochromatin-seq (a method to map heterochromatin) analyses. Liver damage was characterized by histologic analysis.

RESULTS:

We found more than 3000 genes differentially expressed in hepatocytes during P14 to M2 liver maturation. Disruption of Ezh1 and Ezh2 in livers caused perinatal hepatocytes to differentiate prematurely and express genes at P14 that would normally be induced by M2. This resulted in liver fibrosis. Genes with H3K27me3-postive and H3K4me3-positive promoter regions in euchromatin were prematurely induced in hepatocytes with loss of EZH1 and EZH2—these genes included those that regulate hepatocyte maturation, fibrosis, and genes not specifically associated with the liver lineage.

CONCLUSIONS:

The PRC2 proteins EZH1 and EZH2 regulate genes that control hepatocyte maturation and fibrogenesis and genes not specifically associated with the liver lineage by acting at promoter regions in euchromatin. EZH1 and EZH2 thereby promote liver homeostasis and prevent liver damage. Strategies to manipulate PRC2 proteins might be used to improve hepatocyte derivation protocols or developed for treatment of patients with liver fibrosis.

Graphical Abstract

Due to the shortage of fully differentiated cells for transplantation and disease modeling of various tissue types, there is interest in generating replacement cells. Directed differentiation and cell reprogramming can generate early stage cells, but with a failure to activate a terminally differentiated transcriptional program and a failure to repress genes of the starting cell type^1-4. In directed differentiation of hepatocyte-like cells, many fetal genes can be activated, but often there is failure to induce mature CYP P450 enzymes and repress fetal markers such as Afp^5-9. When fibroblasts are reprogrammed to hepatic cells, the cells express liver markers and perform some liver metabolic functions but fail to activate various mature liver genes and cannot consistently rescue liver damage^10-12 The literature on embryonic liver development^13,14 rarely covers postnatal hepatic maturation^15-17, though there are major physiological changes in liver size, diet¹⁸, microbiota and microbial metabolites^19,20, sexual maturation, and polyploidization²¹. Thus, there is a need to understand postnatal hepatic maturation.

Components of Polycomb Repressive Complexes 1 or 2 (PRC1 or PRC2) regulate multiple gastrointestinal cell types. The PRC2 protein EED promotes intestinal stem cell proliferation and inhibits differentiation²², while the PRC1 protein BMI1 regulates intestinal stem cell proliferation and renewal^23(p1). PRC1 proteins BMI1 and MEL18 contribute to colitis-associated cancer²⁴. PRC2 protein JARID2 is needed for late differentiation of pancreatic beta-cells²⁵. In liver development, EZH2, one of the two H3K27me3 histone methytransferases, modulates the cell fate choice of embryonic endoderm to become pancreatic or hepatic buds ²⁶ and is required for hepatoblast proliferation²⁷. Loss of both EZH1 and EZH2 leads to chronic liver damage in adult mice²⁸. Genes marked by the PRC2 repressive histone modification H3K27me3 exhibit activation defects in human fibroblast-to-hepatocyte reprogramming protocols, with genes in heterochromatin being the most resistant to activation^29-33. This leaves open a role for Polycomb-based regulation in postnatal hepatic maturation.

By genetically ablating both Ezh1 and Ezh2, we investigated the relationship between postnatal transcriptional, H3K27me3, and chromatin compaction dynamics in postnatal hepatocyte maturation and homeostasis. We found that PRC2 represses three main classes of genes in postnatal hepatocytes; hepatic maturation genes, non-liver lineage genes, and fibrosis genes. Genes that derepress in response to Ezh1/2 loss have a unique chromatin signature. These findings impact our understanding of the means by which hepatocytes mature postnatally, how hepatocyte lineage fidelity is maintained, and how fibrosis genes are primed for a transcriptional response.

Results

Postnatal hepatic maturation involves differential expression of thousands of genes

Using RNA-seq on hepatocytes isolated by liver perfusion, we found 1215 upregulated and 2011 downregulated genes in hepatic maturation between P14 and M2 (alpha ≥ 0.05, FC ≥ 2) (Figure 1A, Supplementary Figure 1A-D, Supplementary Table 2,3,4). Upregulated genes include the key liver metabolic enzymes Ces1f and Ces3b, and are enriched by Gene Ontology for the Uniprot liver tissue expression category, P450 enzymes, and genes for metabolism of retinol, xenobiotics, and bile acids (Figure 1AB, Supplementary Table 5). As expected, fetal liver genes Afp and Gpc3 are among the top downregulated genes (Figure 1A)^16,34. Downregulated maturation genes enrich cell cycle, cell adhesion, differentiation, and signaling categories (Figure 1B). The cell cycle-associated categories reflect diminished proliferation in adult hepatocytes, as assayed by staining for proliferation marker PCNA (Supplementary Figure 1E). PubMed searches for a random sampling of genes in the cell adhesion and signaling categories and “liver” or “hepatocyte” often return no results, indicating that these may be new targets to study in liver maturation. Thus, P14 hepatocytes are transitioning extensively from a fetal to a mature transcriptional program.

Figure 1. Postnatal hepatic maturation during the P14 to M2 transition involves differential expression of thousands of genes and H3K27me3 dynamics.

(A) log₂(M2/P14 fold change) for genes differentially expressed in maturing hepatocytes.

(B) Gene Ontology for genes up and downregulated from P14 to M2 in Wt hepatocytes.

(C) H3K27me3 dynamics at genes up- (top, grey shading) and downregulated (bottom, orange shading) in maturation

(D) Percent H3K27me3 gene body coverage of “retain” H3K27me3 genes. Monte Carlo simulation and estimated p-value. Whiskers: 5^th/95^th percentiles.

P14 and M2 H3K27me3 states correlate with postnatal hepatic maturation

We profiled H3K27me3 in P14 and M2 hepatocytes (Supplementary Table 2,3). As expected, H3K27me3 was absent from the expressed liver gene Alb and present at the silent Hoxd cluster, and this anticorrelation of expression and H3K27me3 was observed genome-wide (Supplementary Figure 1FG). H3K27me3 genomic coverage increased from 608 MB at P14 to 827 MB at M2 (Supplementary Figure 1H).

We called promoters and gene bodies as “lacking”, “losing”, “gaining”, or “retaining” H3K27me3 from P14 to M2. Acquisition of H3K27me3 is characteristic of many genes that are downregulated in the maturation transition from P14 to M2, with 44% of such promoters gaining (n=179) or retaining (n=909) H3K27me3 and 40% gene bodies gaining (n=172) or retaining (n=821) H3K27me3 (Figure 1C). In the case of gene bodies that retain H3K27me3, the percent of coverage increases from a median of 62% at P14 to 78% at M2; thus increasing H3K27me3 presence at these “retain H3K27me3 genes” was not discernible in categorical lack/lose/gain/retain calls (Figure 1D). This increase was not solely due to increased H3K27me3 coverage in M2 (p-value at 0.001, Monte Carlo simulation). About 9.4% (140) of promoters and 8.6% (129) of gene bodies belonging to maturation upregulated genes lose H3K27me3 during the P14 to M2 transition (Figure 1C). The H3K27me3 and transcriptional dynamics between P14 and M2 indicate a role for PRC2 proteins in regulating maturation gene expression.

EZH1 and EZH2 restrain premature postnatal hepatic maturation

To assess the role of PRC2 repression in postnatal hepatocytes, we performed RNA-seq on Alb-Cre/Alb-Cre; Ezh1^−/−; Ezh2^fox/flox (“Ezh1/2”) hepatocytes, in which both histone methyltransferases for H3K27me3 are ablated. Loss of Ezh1 or Ezh2 alone does not result in H3K27me3 loss (Supplementary Figure 2A). Ezh1 single knockout have transcriptomes highly similar to Wt, with only 203 differentially expressed genes (Supplementary Figure 2B, Supplementary Tables 2, 3). With a homozygous Alb-Cre transgene, whose expression starts around birth, P14 is the earliest age we observed quantitative H3K27me3 loss in Ezh1/2 hepatocytes; hence this time was used to minimize secondary effects (Supplementary Fig 2CD). Notably, P14 Ezh1/2 hepatocytes have expression profiles between Wt hepatocytes P14 and M2 hepatocytes (Supplementary Figure 1D). Strikingly, of the 1215 genes upregulated during hepatocyte maturation at M2, 263 (22%) are prematurely upregulated in P14 Ezh1/2 hepatocytes (Figure 3A, Supplementary Table 3). Of the 2011 genes downregulated in hepatocyte maturation, 128 (6.3%) are prematurely downregulated in P14 Ezh1/2 hepatocytes. Expression changes do not preferentially affect zonated genes (Supplementary Figure S2E). Taken together, the results indicate that the PRC2 proteins EZH1/2 normally restrain maturation of postnatal hepatocytes until the appropriate postnatal time.

Figure 3. Stable global chromatin compaction in maturation and P14 Ezh1/2.

(A)srHC-seq and RNA signal for positive controls sites for heterochromatin, Zfp936, euchromatin, Alb locus, or a dynamic gene, Cux2.

(B) srHC-seq scores at promoters and gene bodies for genes binned into expression quantiles. 0-30 represents genes without any RNA signal. Note the with increasing expression, srHC scores become more euchromatic. Whiskers: 5^th/95^th percentiles.

(C) Megabases of the genome called as heterochromatic, intermediate, or euchromatic and the overlap with H3K27me3 domains.

(D) srHC-seq domains and signal, H3K27me3, and RNA. Note that H3K27me3 occurs in both heterochromatic (red) and euchromatic (green) regions.

Non-hepatocyte lineage genes are derepressed in P14 Ezh1/2 hepatocytes

There are 665 genes upregulated and 90 genes downregulated in P14 Ezh1/2 hepatocytes that are not normally changed during P14 to M2 maturation (Figure 3A, Supplementary Table 3). Upregulated genes include many transcriptional regulators, DNA-binding factors, and genes expressed in non-liver tissues, but GO analysis did not reveal specific biological pathways that were deregulated (Supplementary Table 5). These data are consistent with work from others that Polycomb proteins broadly repress alternate lineage-specific identity regulators^35-38, but are insufficient upon deletion to result in outright hepatocyte identity change at this postnatal maturation stage.

A subset of maturation and alternative lineage genes are repressed by EZH1/2

As expected, given the repressive functions of PRC2, the 263 prematurely upregulated “late maturation” genes have high H3K27me3 at promoters and gene bodies at P14, but not at M2, as compared to genes only upregulated in maturation (Figure 2B). There are 148 genes upregulated in P14 Ezh1/2 hepatocytes (Figure 2A), although they are normally downregulated in maturation and H3K27me3-marked, indicating that these are “early maturation” genes that are downregulated in the course of maturation by PRC2. The 665 “alternative lineage” genes upregulated in P14 Ezh1/2 mutants have high P14 and M2 H3K27me3 levels and the majority are lowly expressed in Wt hepatocytes (Supplementary Figure 2H). These data support a model where PRC2/H3K27me3 repression is used to repress early and late maturation genes at the appropriate times and non-hepatocyte alternative lineage genes.

Figure 2. PRC2 proteins EZH1/2 restrain postnatal hepatic maturation.

(A) Relative expression of genes differentially expressed in maturation or in P14 Ezh1/2 hepatocytes.

(B) P14 and M2 H3K27me3 density at promoters, genes, or enhancers. Whiskers: 5^th/95^th percentiles. Wilcoxon rank-sum test.

(C) srHC-seq (heterochromatic-enriched in red, euchromatic-enriched in green), H3K27me3, and RNA signal at genes prematurely upregulated in P14 Ezh1/2 hepatocytes (Slc15a5, Cyp26a) and a control gene (Pax3).

P14 and M2 H3K27me3 was low at liver enhancers, regardless of whether the enhancers were centered by DNase hypersensitivity or if windows were varied around enhancer centers from 200 bp to 5 kb (Supplementary Figure 2G). Specifically, H3K27me3 levels were low at enhancers associated with genes upregulated in P14 Ezh1/2, except for the early maturation genes, which had only 33 associated enhancers (Figure 2B). From this general lack of H3K27me3 at liver enhancers, we conclude that the Ezh1/2 premature maturation phenotype is not functioning primarily through altered repression of enhancers.

In contrast, of the 263 prematurely upregulated maturation genes, 99 have P14 H3K27me3 at promoters and 85 have P14 H3K27me3 on gene bodies, for a total of 117 genes with at least one type of H3K27me3 marking. These genes include Slc13a5, which plays a key role in importing citrate into liver cells³⁹, Pcsk9, which is associated with liver cholesterol (LDL) uptake⁴⁰, and Cyp26a1, a key enzyme in the clearance of retinoic acid from the liver⁴¹ (Figure 2C, Supplementary Figure 2F). In comparison, Pax3, a silent muscle gene, retains H3K27me3 marking at the promoter and was not upregulated in maturation. GO analysis on the 117 H3K27me3-marked genes did not reveal grouped biological process (data not shown). Taken together, we conclude H3K27me3 is used to restrain expression of genes in postnatal hepatocytes specifically at promoters and gene bodies, and minimally so at enhancers.

srHC-seq reveals that Ezh1/2 sensitive genes have euchromatic and bivalently-marked promoters

To assess whether there are global changes in chromatin compaction in hepatic maturation, we characterized P14 and M2 Wt hepatocytes using an enzyme- and antibody-independent assay termed sonication-resistant heterochromatin sequencing (Supplementary Figure 3A). srHC-seq utilizes the physical property of crosslinked chromatin to be differentially sensitive to sonication; that is, structurally compact, heterochromatic regions are sonication-resistant while structurally open, euchromatic regions are sonication-sensitive²⁹ srHC-seq involves fractionation of large and small DNA fragments and analyzes the ratio of the two to identify both heterochromatic and euchromatic regions.

srHC-seq was highly reproducible in replicates (Supplementary Table 2, Supplementary Figure 3BC). We plotted srHC data as log₂(large fragments/small fragments), with heterochromatic regions as y > 0 (Figure 3A, red) and euchromatic regions as y < 0 (Figure 3A, green). As expected, the silent, non-hepatocyte gene Zfp936 has a heterochromatic profile in liver and genes at the highly expressed, hepatocyte-specific Alb/Afp/Afm locus have euchromatic profiles. Importantly, srHC-seq scores for promoters and gene bodies shows the inverse correlation between expression and chromatin compaction holds true genomically (Figure 3B).

In contrast to silent and highly expressed loci, Cux2 expression is temporally- and sex-specific^42(p2) and srHC profiles reflect these dynamics (Figure 3A). While sex-specific differences were ascertainable, the differences represent only a fraction of the genome. We therefore merged male and female srHC data for the remaining analyses.

Both P14 and M2 Wt hepatocytes have about 1200 Mb of srHC heterochromatin and 1100 Mb of euchromatin, with H3K27me3 occurring in both domains (Figure 3CD, Supplementary Figure 3DE). We did not observe a global increase of euchromatin in P14 Ezh1/2 hepatocytes or by confocal microscopy of DAPI staining of 6-week-old Ezh1/2 samples (data not shown). We conclude that large scale euchromatic and heterochromatic domains are generally stable in postnatal hepatic maturation, while a subset of genes are dynamic.

While the PRC2 complex is classically thought to repress expression by eliciting chromatin compaction⁴³, Polycomb-bound or -marked chromatin can be accessible to binding by some factors and transcribed^29,44-49. PRC2 loss leads to derepression of only a fraction of H3K27me3-marked genes in diverse tissues, with promoter H3K4me2/3 predicting derepression^28,36,50. To investigate the basis by which a subset of H3K27me3-marked promoters genetically respond to Ezh1/2 loss, we plotted srHC-seq profiles at 8121 silent and 2221 highly expressed in P14 and M2 Wt animals (Figure 4A). Significantly, promoters of genes from all three classes of genes upregulated in P14 Ezh1/2 hepatocytes are already euchromatic in P14 Wt hepatocytes, including those with promoter H3K27me3 signal (Figure 4A, note extensive green in P14 Wt). To focus on apparent direct PRC2 targets, we plotted srHC signal of genes with H3K27me3-marked promoters in P14 Wt samples (Figure 4B). These promoters become further euchromatic in Ezh1/2 hepatocytes and the open regions increase in width (Figure 4B). Using published P12 liver ChIP data⁵¹, we found that the three classes of P14 Ezh1/2 upregulated genes have promoters with low levels of H3K4me3, RNA Polymerase II, and general transcriptional factors (Figure 4A, Supplementary Figure 4B).

Figure 4. srHC-seq reveals that Ezh1/2 sensitive genes have euchromatic and bivalently-marked promoters.

(A) ChIP and srHC-Seq signal around transcriptional start sites (−/+ 5kb) for silent genes, genes highly expressed in P14 and M2 (top 10% in expression in P14 and M2), three classes genes upregulated in P14 Ezh1/2, and the remaining P14 H3K27me3+ promoters. Note that the three classes of genes upregulated in P14 Ezh1/2 hepatocytes have euchromatic promoters (green srHC signal) in P14 Wt hepatocytes. In comparison promoters that have H3K27me3 but are not upregulated (bottom, orange) have heterochromatic signals (red srHC signal) in P14 Wt hepatocytes.

(B) srHC-seq metaplots for panel (A) groups except highly expressed genes. Filtered for genes with P14 H3K27me3-marked promoters. Note that late maturation genes, early maturation genes, and alternative lineage genes (green, purple, and red lines) have more euchromatic signal in P14 Wt hepatocytes than the genes which are not upregulated in P14 Ezh1/2 and have H3K27me3+ promoters (orange line).

By plotting signal at promoters of genes that are not differentially expressed in maturation or in P14 Ezh1/2 hepatocytes and do have P14 promoter H3K27me3 (Figure 4A, “remaining P14 H3K27me3+ promoters”), we found that H3K4me3 does indeed help predict genes that become upregulated in Ezh1/2 mutants, as does RNAP2, general TFs, and a pre-existing euchromatic chromatin state (Figure 4A, Supplementary Figure 4B). Thus, promoters that are open and poised yet restrained by PRC2 are regulated during hepatocyte maturation.

Chronic Ezh1/2 loss leads to liver damage

It has previously been reported that Ezh1/2 in liver leads to chronic liver damage and fibrosis in mice by 8 months and increased sensitivity to liver damaging agents²⁸. We observe that apoptosis occurs 4.6-fold more in Ezh1/2 hepatocytes by 1 month postnatal with very rare ductular reactions (Figure 5A). Yet by M2, but not at M1, liver fibrosis can be detected by Sirius Green/Fast Red staining (Figure 5B, Supplementary Figure 5A). Additionally, ductular responses can occur by H&E and CK19 staining by M2, with some larger diameter, malformed ducts than were reported previously (Figure 5B, Supplementary Figure 6A). Macroscopic regenerative nodules occur in nearly half of Ezh1/2 livers by M2, visible in H&E sections or in the most severe cases, by the naked eye as lumps on the liver (Figure 5C). While the Ezh1/2 model loses H3K27me3 in nearly 100% of hepatocytes at P14 (Supplementary Figure 2C), by M2 these nodules stain for H3K27me3 similar to Wt and also stain for EZH2 (Figure 5DE, Supplementary Figure 6AC), which was not observed previously in Bae et al.²⁸. As the liver is a regenerative organ and nodular hepatocytes in Ezh1/2 livers have increased proliferation (Supplementary Figure 6C), we conclude that the EZH2+ nodules originate by selection for rare cells which escape Alb-Cre recombination. Thus, loss of PRC2-based repression results in loss of hepatic homeostasis, chronic liver damage, and fibrosis.

Figure 5. Ezh1/2 loss leads to chronic liver damage.

(A) Apoptosis as assayed by TUNEL staining.

(B) Fibrosis as assayed Sirius Green (total protein) and Fast Red (collagen) staining.

(C) Examples of a normal Wt liver and 2 Ezh1/2 livers with macroscopic liver nodules. Quantification of animals with liver nodules as assessed by histology.

(D)H3K27me3 immunohistochemistry in M2 livers. Regions inside and outside of regenerative nodules are outlined by the dotted circle. Wt: 2034 hepatocytes counted. Ezh1/2: 2440 hepatocytes counted. Ezh1/2 nodule: 1056 cells counted in 7 nodules. Two-sided student’s t-test with standard error.

Genes involved in liver fibrosis are primed by euchromatic H3K27me3+ promoters

A 232 gene signature predicting fibrosis before histopathological detection was identified using a non-alcoholic steohepatitis (NASH) mouse model, and the relevance to humans was confirmed by finding 71 of 123 human NASH genes in the mouse 232 gene datasets^52,53. RNA-seq on liver biopsies from a mixed cohort of chronic liver disease patients with Hepatitis C and/or fatty liver disease identified 121 genes upregulated in advanced human liver fibrosis as compared to early fibrosis⁵⁴. We found that genes from both datasets are normally downregulated in postnatal hepatic maturation, are upregulated in P14 and M2 Ezh1/2 hepatocytes, and have the predictive euchromatic H3K27me3+/H3K4me3+ promoter chromatin state that we identified as sensitizing genes to Ezh1/2 loss (Figure 6A). Genes significantly upregulated in P14 Ezh1/2 samples in mouse and human datasets were each 2.6-fold more likely to have P14 H3K27me3+ promoters (permutation test, p=0.003 and p=0.009). Included in fibrosis gene sets with H3K27me3 promoter-marked genes are Fbn1, an inflammation and chemotaxis gene that is also upregulated in response to damaging chemicals⁵⁵, Fstl1, which is upregulated in humans with HCV-induced fibrosis and steatosis⁵⁶, and Col1a1, which is upregulated in mice and humans with liver fibrosis⁵⁷ (Figure 6B, Supplementary Figure 7BC). Interestingly, genes proposed to be involved hepatocyte epithelial-mesenchymal transition (EMT) in response to liver damage include Tgfb1, Vim, and S100a458 also have H3K27me3-marked promoters and may be upregulated in M2 Ezh1/2 hepatocytes (Figure 6B, Supplementary Figure 7BC). Together, the failure to repress fibrosis-related genes starting at two weeks postnatal is associated with the 2-month-old fibrotic phenotype in Ezh1/2 livers.

Figure 6. PRC2 represses liver fibrosis signature genes.

(A) Genes upregulated in a murine model of NASH-related fibrosis (top) and genes upregulated in human advanced fibrosis patients as compared to low level fibrosis (bottom). Murine hepatocyte expression, H3K27me3 and srHC-seq, and H3K4me3 signal around transcriptional start sites −/+ 5 kb. Note that many fibrosis genes have H3K27me3+ promoters that are also euchromatic (green srHC signal) and have promoter H3K4me3. M2 Ezh1/2 RNA-seq represents a mix of hepatocytes from inside and outside of nodules.

(B) srHC-seq, H3K27me3, and RNA signal at fibrosis-related or EMT-related genes.

Discussion

To be effective therapeutically, newly generated hepatocyte-like cells must perform the complex metabolic functions of native hepatocytes that are naturally induced postnatally^59-61, but these late maturational functions are limited in many hepatocyte-like cells^5-9. Here we define a P14 to M2 transitional hepatocyte profile with more than 3000 transcripts exhibiting differential expression. The RNA maturation dataset can be a useful resource for assessing the stage of hepatocytes generated from stem cells or by directed reprogramming. Perinatal loss of EZH1/2 in hepatocytes leads to premature differentiation, suggesting that modulation of Polycomb components may enhance maturation in protocols for generating new hepatocytes.

In our study, altering PRC2 repression led to impaired liver function and fibrosis, and previous reports have shown impaired ability to respond to liver damaging agents²⁸. Interfering with splicing factors alters liver maturation in mice and also leads to susceptibility to liver damage¹⁵. Considering the rising worldwide prevalence of NASH and hepatitis B⁶² and that fibrosis is a key indicator of chronic liver injury of any etiology⁶³, the common thread of impaired maturation predisposing the adult liver to damage suggests that studying mechanisms of hepatic maturation may help us identify environmental factors that alter maturation in humans and predispose them to liver diseases. Here we have identified a H3K27me3+/H3K4me3+ euchromatic promoter signature that primes many genes to be upregulated in response to maturation or other signaling events, and we propose that this is a chromatin-level mechanism leading to liver fibrosis in the Ezh1/2 model. Given the euchromatic state of these promoters, it may be that compaction-independent repressive activities of Polycomb proteins, such as PRC2 inhibiting elongation⁶⁶ and PRC1 interfering with RNAPII recruitment^64,65, are key to regulating transcription in postnatal hepatocytes, though a full catalog of precise mechanisms by which Polycomb proteins repress transcription is still being worked out. It will be interesting to see how agonists or antagonists of different aspects of Polycomb Protein function may help reverse fibrosis, considering that there is growing evidence of fibrotic reversal after treatment for hepatitis B, C, and autoimmune hepatitis⁶³.

There is conflicting evidence suggesting that hepatocytes can acquire a fibroblastic phenotype and expression of mesenchymal markers through EMT during liver fibrosis or in response to TGBβ treatment^58,67-69. We find that expression of these mesenchymal markers is slightly upregulated in M2 Ezh1/2 hepatocytes, but expression is highly variable (Supplementary Figure 6). This may reflect variability of multiple different sources, including variable levels of liver damage and fibrosis, variable contributions of H3K27me3+ nodular hepatocytes, and variable numbers of cells undergoing EMT at possibly different stages of EMT. The mesenchymal genes tend to be marked by H3K27me3 and have euchromatic promoters (Figure 6, Supplementary Figure 7). While we cannot conclude from these results whether EMT is occurring, it will be interesting to assess whether PRC2 repression maintains an epithelial state in hepatocytes by repressing mesenchymal genes and whether liver damage elicits their derepression.

Our results emphasize the role of PRC2 proteins in regulating maturation and how that may affect fibrosis, but not all maturation genes in our dataset are prematurely up- or downregulated in P14 Ezh1/2 mutants. There must be other mechanisms that regulation maturation genes in response to changing diet, microbiota, and sexual maturation during the two week to two month period. These results highlight the need to identify maturation factors, both for the purposes of enhancing in vitro hepatic maturation and for understanding factors which predispose humans to disease.

Methods

Liver perfusion and hepatocyte isolation

Mice are anesthetized using isoflurane, the abdominal cavity exposed, venae cavae cannulated, and the portal vein severed. 37°C liver perfusion media (Invitrogen 17701-038) and then liver digest media (Invitrogen 17703-034) are perfused (25 mL for P14, 45 mL for M2). Dissociated livers in William’s E are strained through a 100 μm filter (Figure S1a) and pelleted at 50g for 5 min at 4°C. Hepatocyte enrichment was confirmed by depletion of RNAs from contaminating cells types by comparing whole perfused liver, isolated hepatocytes, and supernatant (pelleted at 500g) fractions.

RTqPCR

RNA was isolated from TRIzol, cDNA generated (Biorad 170-8891), and expression analyzed with Power SYBR Green (Thermo 4368577).

Western blotting

Whole cells or nuclei (isolated by douncing in RSB) were resuspended in Buffer C (200 mM Tris pH 7.9, 400 mM NaCl, 1 mM EDTA, 1 mM EGTA, 0.8% SDS, PIC (Roche #11873580001), 1mM PMSF). Nuclear extracts were sonicated to reduce the viscosity from high DNA content. Extracts were denatured for 30 min at 99°C in 100mM DTT/Sample Buffer (Thermo NP0007) and run with the NuPAGE system (NP0335,NP0002) at 80V. Wet transfer to PVDF membranes (100V for 3 hr, transfer buffer NP0006) and membranes were blocked for an hour in 5% NFDM-TBST. Primary antibodies were incubated overnight in 5% NFDM-TBST. Secondary antibodies (Santa Cruz Biotechnology sc-2004, sc-2005) were incubated in 5% NFDM-TBST for 1 hour. Blots were developed using Thermo #34080. Primary antibodies: H3: 1/5000 Millipore 05-928. H3K27me3: 5ug/mL Abcam 6147. H3K27me3: 1/1000 Active Motif 39155.

Chromatin preparation and immunoprecipitation

Hepatocytes were fixed in 25 mL 1% formaldehyde in PBS for 10 min at room temperature, quenched with 2.3 mL 2.5 M glycine for 5 min, pelleted at 4°C at 50g for 5 min, resuspended in 10 (P14) or 20 mL (M2) ice-cold RSB (10 mM Tris pH7.4, 10 mM NaCl, 3 mM MgCl2, 0.5% NP40, PIC, 0.1 mM PMSF, 1% Triton-X 100), dounced, pelleted for 10 min at 4°C at 100g, and resuspended in 2 mL ice-cold AS sonicationlysis buffer (10 mM Tris-HCl, pH8, 100 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 0.1% Na-deoxycholate, 0.5% N-lauroylsarcosine, 1 mM DTT, PIC, 0.1 mM PMSF).

For srHC experiments, samples were sonicated using a Diagenode Biorupter UCD-200 (settings: 30 sec on high, 30 sec off for 30 minutes), Triton-X 100 added to 1%, debris pelleted for 15min at 4°C, and supernatant collected. DNA was extracted from 50 μL chromatin by adding 150 μL TE/1% SDS and decrosslinked overnight at 65°C. Next, 200 μL TE and 8 uL 10 mg/mL RNAse A added for 2 hours at 37°C shaking. Next, 4 μL 20 mg/mL Proteinase K was added for 2 hours at 55°C shaking. DNA was extracted by phenol:chloroform:isoamyl separation and ethanol precipitation.

For ChIP experiments, hepatocytes in AS-sonication lysis buffer proceeded to the extensive sonication and ChIP protocol in Nicetto et al., (manuscript accepted, Science). P14 and M2 hepatocyte H3K27me3 (Millipore 07-449) libraries were then generated using the ThruPlex DNA-seq kit (Rubicon #R400428). The M2 H3K27me3 ChIP was previously published Nicetto et al. (GSE114198).

RNA-seq

RNA was isolated using TRIzol, polyA-selected I(nvitrogen (dT)25-61002), libraries prepped (NEB 7420), and sequenced with 75 bp single-end reads. For P14 Wt versus Ezh1 libraries, libraries were prepped with the NEBNext E6110.

Immunohistochemistry

Mice were prepped as for liver cannulation as described above and blood blanched from the liver with liver perfusion media. Livers were rinsed in PBS, fixed in 4% PFA-PBS for 1 hour at 4°C, washed in PBS, dehydrated, embedded in paraffin, 12 μm sections taken, dried, rehydrated in H2O, and washed. Antigen retrieval was performed in citrate buffer (10 mM Na Citrate, 0.05% Tween20, pH 6) by microwaving for 15 min. Slides were rinsed in water and PBS, quenched in 3% H2O2 in PBS for 15 min, washed, blocked for 15 min in avidin, washed, blocked for 15 min with biotin, washed, and serum blocked (10% FBS in PBS) for 30 min at room temperature. Primary antibodies were incubated in 10% FBS in PBS overnight at 4°C, washed in PBST (PBS, 0.1% Tween 20). Secondary antibodies (1/200 Santa Cruz-2004, −2005) were incubated in 10% FBS in PBS for 45 min at 37°C. Slides were washed in PBST and developed with DAB. Slides were then dehydrated and mounted with Cytoseal. PCNA antibody: 1/50 Santa Cruz 7907 Lot0402. H3K27me3 antibody: 1/500 Active Motif 39155 Lot 01613015. CK19 antibody: 1/100 from Ben Stanger’s Lab. For statistics, the two-sided student’s ttest with ftest were used on the average percent staining per animal.

TUNEL staining:

Trevigen TACS TdT-DAB kit (Cat #4801-30-K) after avidin/biotin blocking.

H3K27me3 gene body coverage increase statistical analysis

Monte Carlo simulation was employed to assess the significance of the gene body coverage by H3K27me3 domains (Figure 1D): the selected genes were measured for a median P14-M2 increase in gene body domain coverage, a random set of genes of equal size was sampled, and the median difference in gene body domain coverage was measured 1,000 times, with the p-value estimated as the number of samples in which the median difference met or exceeded the observed difference divided by 1,000.

srHC

10 μg DNA from sonicated chromatin were resuspended in 50 μL TE. For large fragments: 25 μL beads (0.5 volumes) (Beckman Coulter A63881) were added to the 50uL of DNA, incubated, and beads were removed and large DNA isolated from them as described by the manufacturer. For medium fragments: 10 μL beads (0.2 volumes) were added to the supernatant from the large beads/DNA slurry, incubated, and beads were removed and medium DNA isolated from the beads. For small fragments: 35 μL beads (0.7 volumes) were added to the supernatant from the medium beads/DNA slurry, incubated, and beads were removed and small DNA isolated from the beads. Size selection efficacy was confirmed (Agilent 5067-4626) (Figure S3a).

For library preparation considerations, large DNA was sonicated after size selection with a Covaris S220 with the following settings PP-175 W, DF-10, CB-200, 4-9°C, 5 minutes, then ethanol precipitated.

Libraries were prepped (NEB E7370) per the manufacturer’s recommendations. For size selection of small libraries: 55 μL beads for the first step and 25 μL beads for the second step. Libraries were sequenced with 75 bp single-end reads.

Enhancer-promoter unit and DHS processing

EPUs were downloaded from Shen et al, 2007⁷⁰. 1 bp was added to enhancers that loop to multiple genes. Enhancers were centered by intersecting with concatenated liver DHS sites (GSM1014195 replicates 1-14)⁷¹. The first DHS event in each enhancer was used.

Domain calling

We adapted a previously described algorithm²⁹ to call domains. H3K27me3 parameters: 2kb windows, 1kb slide, top 30% cutoff. srHC parameters: 10kb windows, 2kb slide, and 40% cutoff. Intermediate domains include regions not called as hetero/euchromatic and heterochromatin/euchromatin double positive regions. 50% or 75% coverage by domains were used for marked gene and promoter calls, respectively.

DNA-sequencing and processing

Reads were aligned to the NCBI v37/mm9 genome using STAR2.4.2a with the following arguments: --outFilterMultimapNmax 20, --alignIntronMax 1, then filtered for unique alignments to avoid PCR duplication artifacts. RPM-normalized bedgraphs of alignments were generated, then values log2(large/small) or ChIP minus Input subtracted calculated for every block. To avoid dividing by zero, a small addend was added to every block. Biological replicate values were merged using bedtools unionbedg and averaged.

Read density (or srHC-seq scores) were calculated using the Bioconductor Genomation v1.6.0 package using ChIP minus Input (negative values converted to 0) or log2(large/small) bigwigs. The resulting values were quantile-normalized across all P14 and M2 individual biological replicates. On box and whisker plots, whiskers indicate 5^th and 95^th percentiles and Wilcoxon statistical testing was used.

RNA-sequencing and processing

Reads were aligned to the mm9 genome using STAR2.4.2a with the following arguments: --outFilterType BySJout, --outFilterMultimapNmax 20, --alignSJoverhangMin 8, --alignSJDBoverhangMin 1, --outFilterMismatchNmax 999, --alignIntronMin 20, --alignIntronMax 1000000. HTSeq version 0.6.1p1 count was used to assign reads to genes with the NCBI v37/mm9 genome file (RefSeq genes, refFlat), then normalized and DE genes called using DESeq2 (alpha ≤ 0.05, FC ≥ 2).

For browser views, alignments greater that 75 bp were filtered out to avoid showing spurious intronic signal, and converted to RPM-normalized strand-specific bigWigs. Replicates were averaged at the RPM-normalized bedGraph stage with the tool bedtools unionbedg.

Zonated genes were kindly provided by Dr. Itzkovitz⁷².

Experimental model and subject details

All animal studies were performed with the University of Pennsylvania IACUC approval. Genetics include the Alb-Cre transgene⁷³, and Ezh1^−/−37 and Ezh2^f/f (Ezh2tm1Tara)⁷⁴ alleles in a mixed C57BL/6J and C3H background.

Abbreviations:

ChIP

chromatin immunoprecipitation

DHS

DNase hypersensitive sites

NAFLD

nonalcoholic fatty liver disease

NASH

nonalcoholic steohepatitis

PRC2

Polycomb Repressive Complex 2

RNA-seq

RNA-sequencing

RNAP2

RNA polymerase 2

srHC-seq

sonication-resistant heterochromatin sequencing

TSS

transcriptional start site