Acquired Tissue-Specific Promoter Bivalency Is a Basis for PRC2 Necessity in Adult Cells

Abstract

Bivalent promoters in embryonic stem cells (ESCs) carry methylation marks on two lysine residues, K4 and K27, in histone3 (H3). K4me2/3 is generally considered to promote transcription, and Polycomb Repressive Complex 2 (PRC2) places K27me3, which is erased at lineage-restricted genes when ESCs differentiate in culture. Molecular defects in various PRC2 null adult tissues lack a unifying explanation. We found that epigenomes in adult mouse intestine and other self-renewing tissues show fewer and distinct bivalent promoters compared to ESCs. Groups of tissue-specific genes that carry bivalent marks are repressed, despite the presence of promoter H3K4me2/3. These are the predominant genes de-repressed in PRC2-deficient adult cells, where aberrant expression is proportional to the H3K4me2/3 levels observed at their promoters in wild-type cells. Thus, in adult animals, PRC2 specifically represses genes with acquired, tissue-restricted promoter bivalency. These findings provide new insights into specificity in chromatin-based gene regulation.

Figure 1. H3K27me3 in adult mouse intestinal stem and villus cells

See also Figure S1. (A) H3K27me3 distribution in ISC and villus cells over the 2,328 genes most heavily marked in intestinal cells (clusters 1 and 2 from Figure S1D). H3K27me3 on gene bodies (5 equal-sized bins) ±2 kb is represented on the brown scale and RNA levels on the blue-red scale. Each row represents the same gene in ISCs and villus cells, with typically undetectable to low mRNA. (B) Scatter plot of 3,879 transcripts differentially expressed (≥3-fold, q<0.05) in purified villus cells (y-axis) and Lgr5⁺ ISC (x-axis), represented by grey dots outside the dotted lines. H3K27me3 gains and losses in villus cells compared to ISCs (determined using DiffReps, q<0.001) are overlaid in red and blue, respectively. (C) ChIP- and RNA-seq traces at Cdkn2b, showing significant H3K27me3 loss accompanying increased mRNA expression in villus cells compared to ISC. In contrast, Pkd2I2 illustrates gain of H3K27me3 in villus cells, with insignificant change in mRNA expression. (D) Genes that lose H3K27me3 during differentiation among those that show higher mRNA (≥3-fold, q<0.05) in villus cells over ISC (arranged in order of log₂ fold-gain of mRNA). For all 1,932 genes upregulated in villus cells, the pie charts show the fraction of genes with SICER-identified peaks in ISC or villus cells. (E) Gains of H3K27me3 in villus cells (arranged in descending order, reads/bp X1000) were concentrated among genes expressed selectively in ISC, including Lgr5, as illustrated in the adjoining ChIP- and RNA-seq data tracks.

Figure 2. Differential bivalent domains in adult mouse tissues

See also Figure S2. (A) H3K27me3 (amber), mRNA (green), and promoter H3K4me3 (blue) profiles in adult intestinal villus, blood (Hasemann et al., 2014) and skin (Lien et al., 2011) cells. Genes strongly marked in any of the 3 tissues were collated (Figure S2A–B) and clustered on the basis of H3K27me3 levels in one or more tissues. H3K27me3 is represented on gene bodies (5 equal-sized bins) ±2 kb and H3K4me3 is represented ±1.5 kb from the TSS (center). (B) Average H3K27me3 (left) and H3K4me3 (right) levels in intestinal villus cells at genes that are strongly marked with H3K27me3 in all 3 tissues (Group1, dotted brown); carry H3K27me3 in the intestinal epithelium and up to one other tissue (Group 2a, solid brown); or lack H3K27me3 in intestinal cells (Group 2b, blue). (C) ChIP- (H3K27me3 and H3K4me3) and RNA-seq tracks of representative genes. (D) mRNA levels among genes from the three groups in villus cells and ISC. Boxes demarcate the 25^th and 75^th percentiles, whiskers represent the 1.5X interquartile range.

Figure 3. Acquired tissue-specific bivalency in adult tissues

See also Figures S2 to S4. (A) H3K27me3 (amber) and corresponding H3K4me2 (blue) profiles in enterocyte and secretory progenitors in intestinal crypts. Clustering is based on genes heavily marked with H3K27me3 in the 3 adult tissues (Figure 2A). H3K27me3 is shown on gene bodies (5 equal-sized bins) ±2 kb and H3K4me2 is represented ±1.5 kb from the TSS (center). (B) H3K27me3 profiles in undifferentiated ESCs (data taken from (Subramanian et al., 2013)) of the same genes as in A. Pie charts show the fraction of genes bearing H3K27me3 in ESCs in each of the 3 groups denoted by brackets. (C–D) Overlap of genes carrying H3K27me3 (C) or regarded as bivalent in ESCs (D, www.dailab.sysu.edu.cn/bgdb), with those marked in the adult mouse duodenum. ChIP-seq data tracks in C illustrate one of 1,035 genes - Casc4 - with intestinal bivalency (promoter H3K27me3 and H3K4me3) and no evidence of PRC2 activity in ESCs.

Figure 4. Intestinal epithelial defects in the absence of PRC2

See also Figure S5. Tissue sections from Eed^+/+ and Eed^−/− (various days after tamoxifen exposure) mouse duodenum. All scale bars, 100 μm. (A) H3K27me3 immunostains, showing epithelium-specific loss (outside the dotted line), with preservation of signal in the lamina propria. (B) Ki67 immunostains, showing reduced crypt cell replication. Arrowheads indicate EED-proficient (H3K27me3+, replication-competent) crypts. (C) Hematoxylin & eosin (H&E) stains, showing stunted, dysplastic villi amidst rare, scattered EED-proficient crypts and intact villi. (D) Alkaline phosphatase stains, showing mature enterocytes confined to Eed^−/− villus tips.

Figure 5. Gene deregulation in PRC2-null intestinal villi and other mouse tissues

(A) Differential gene expression in purified Eed^+/+ (x-axis) and Eed^−/− (y-axis) villus epithelium, determined from triplicate RNA-seq analysis 4 days after tamoxifen exposure. Dots represent unaffected (grey) and altered (q <0.05, blue) transcripts. (B) Relation of mRNA changes in Eed^−/− villus cells (all blue dots in A) to basal expression and H3K27me3 marks in wild-type villi. Most dysregulated genes that are expressed in wild-type intestinal villi lack H3K27me3 and are as likely to rise as to fall (pie chart) in mutant villi; hence, these effects are likely indirect. In contrast, genes that are barely or not expressed in wild-type villi (e.g., Barx1, Tbx15, etc.) typically carried significant H3K27me3 in wild-type villi and these effects are likely direct. Box plot shows H3K27me3 ChIP-seq signal density (TSS ±500 bp) at the genes we consider to be affected directly or indirectly by PRC2 loss. Boxes represent the 25^th and 75^th percentiles, and whiskers represent the 10^th and 90^th percentiles. (C) Transcripts absent in the wild-type tissue and increased >3-fold in PRC2-null intestine (this study), blood (Xie et al., 2014) or skin (Ezhkova et al., 2011). Cdkn2a and Tbx15 are the only genes that meet these criteria in all 3 tissues. Violin plots show H3K27me3 signal density (black bar = median) in the 3 tissues at genes activated specifically in each. (D) Activation in Eed^−/− intestinal villus cells of up-regulated genes (direct targets, very few of which showed reduced expression) marked with H3K27me3 in all wild-type tissues or only in the intestine. Boxes represent the 25^th and 75^th percentiles and whiskers represent the 1.5X interquartile range. Significance in B and D was determined using the Mann-Whitney test.

Figure 6. Basal H3K4me3 levels at bivalent promoters are crucial determinants of PRC2 dependency in the adult intestine

See also Figure S6. (A) RNA-seq tracks illustrating the wide range of gene activation in Eed^−/− villus epithelium. Note the different y-axis scales for 3 representative examples. Small gains in absolute mRNA levels (e.g., Emx2, others in Figure S6) are consistent across triplicate samples. (B) Wild-type H3K27me3, H3K4me3 and H3K4me2 signal densities at promoters (TSS ±1.5 kb) of genes activated in PRC2-null intestinal villi. Genes are arrayed in descending order of mRNA gain in mutant cells, as reflected in the left heatmap. All these genes show similar H3K27me3 marking in the wild-type intestine and the degree of gene activation is proportional to the level of promoter H3K4me3. (C) Illustrative ChIP- and RNA-seq tracks from points on the spectrum of gene deregulation.

Figure 7. Relation of basal H3K4me3 levels and RNA Pol-II binding to the transcriptional consequences of PRC loss in adult mouse tissues

See also Figure S7. (A) Average H3K4me3 and H3K27me3 profiles at genes activated in Eed^−/− intestinal villi, separated into quintiles of absolute increase in mRNA levels. (B) Average H3K4me3 profiles for the top, middle, and bottom quintiles of genes activated in PRC2-null blood (Xie et al., 2014) and skin (Ezhkova et al., 2011), determined from published microarray data. Heatmaps are shown in Figure S7. (C) Gene deregulation in PRC2-null villus epithelial cells (absolute mRNA gain, y-axis) is highly correlated with the level of promoter H3K4me3 in wild-type intestine (x-axis). Genes in Group 1 carry little to no H3K4me3 and accordingly concentrate toward the left (violin plots); few such genes (examples named) are activated. In contrast, genes in Group 2a express in nearly linear relation to basal H3K4me3 levels. The dotted line shows curve fit using non-linear regression. Figure S7D shows the same analysis restricted to transcripts increased at P <0.05. (D) H3K27me3 (amber), mRNA (green – from Figure 2A), H3K4me3 (blue – from Figure 2A), and Pol2 (purple) ChIP-seq signal densities (centered on TSSs ±10 kb or 1.5 kb) in wild-type mouse intestinal villus cells. Genes are clustered and grouped exactly as in Figure 2A. Genes active in villus cells (Group 2b) show Pol2 occupancy, whereas genes with H3K27me3 acquired in the intestine and up to two (Group 1) or one other tissue (Group 2a) lack RNA expression or Pol2 binding. (E) Average Pol2 occupancy at genes in Groups 1 (high H3K27me3 in all 3 tissues), 2a (high H3K27me3 in the intestine and up to 1 other tissue, and 2b (no intestinal H3K27me3). ChIP- and RNA-seq tracks from representative genes in Groups 2a and 2b.