Genome-Wide Characterization of H3K27me3 Plasticity.

We obtained the ChIP-seq datasets of H3K27me3 in 19 human cell lines from the ENCODE consortium (17) (SI Appendix, Table S1). The raw sequence reads were normalized and mapped to nonoverlapping bins of 200 bp in size. We initially observed that the H3K27me3 patterns were highly variable across different cell types, with certain regions associated with high-degree of plasticity, such as the homeobox (HOX) gene clusters (Fig. 1A). We quantified the plasticity of H3K27me3 for each bin based on a metric known as the index of dispersion (IOD), and ranked the bins according to their associated plasticity scores (Materials and Methods). We selected 133,980 bins whose plasticity scores were significantly higher than the genome background (P < 0.01) (Fig. 1B). After merging neighboring selected bins, we obtained 46,755 contiguous regions that we refer to as the HPRs in the rest of the article. The average length of the identified HPRs is 507 bp (SI Appendix, Fig. S1). For comparison, we also defined low plasticity regions (LPRs) by using a similar criterion on the opposite end of distribution (Materials and Methods).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Overview of the properties of HPRs. (A) ChIP-seq signals of H3K27me3 in 19 human cell lines from ENCODE (blue tracks) and corresponding plasticity scores (red track). Locations of HPRs are marked by the black segments. Predicted plasticity values (based on the N-score model) are shown as the green track. (B) Genome-wide distribution of the plasticity scores. Shaded areas show the location of HPRs and LPRs. (C) PhastCons scores surround HPRs (red) and LPRs (light blue). Baseline genome-wide average is also shown for reference (dashed black line). (D) Enrichment score of various annotated functional elements in HPRs. The stars indicate statistically significant enrichment (P < 1E-4, Fisher’s exact test). (E) Boxplot showing the distribution of the variance of DNase-Seq signal in HPRs, LPRs, and genome-wide regions across the 19 cell lines. (F) Boxplot showing the distribution of the variance of the expression levels of the genes harboring either HPRs or LPRs in their promoters across the 19 cell lines. Genome-wide distribution is also shown for reference.

The HPRs are highly conserved compared with the genome-background (Fig. 1C), suggesting that they are enriched with functional elements. To rule out the possibility that the elevated conservation can be simply explained by an enrichment of exons, we divided the HPRs into three groups corresponding to exon, intron, and intergenic regions, respectively, and repeated the analysis within each group. For each group, the conservation scores are significantly higher at the HPRs compared with the genomic background and LPRs (SI Appendix, Fig. S2), suggesting the association is not an artifact. We further investigated the overlap between HPRs and various annotated functional elements, and found significant enrichment (P < 1E-4, Fisher’s exact test) in CGIs, CGI shores, promoters, enhancers, and DNase I hypersensitivity sites, suggesting that variation of H3K27me3 is strongly associated with transcriptional regulatory activities (Fig. 1D). A closer examination suggested that the HPRs were associated with elevated variation of DNase-seq signal across cell types (Fig. 1E). We also observed a moderate enrichment in exons (Fig. 1D). Although the function of H3K27me3 in gene bodies remains unclear, recent studies have suggested that it may be related to alternative promoter use (18), alternative splicing (19, 20), or monoallelic gene expression (21). Long noncoding RNAs have recently been implicated in gene regulation and Polycomb recruitment (22, 23). Consistent with these studies, our analysis shows a moderate but statistically significant enrichment of long noncoding RNAs in the HPRs. In addition, we investigated the enrichment of HPRs in all of the repetitive classes of the RepeatMasker annotations. We found that the HPRs were significantly depleted in SINE, LINE, and LTR elements but enriched in Satellite elements. Similar biases were previously observed in mouse embryonic stem (ES) cells (24) (SI Appendix, Fig. S3).

We next investigated the correlation between H3K27me3 and DNA methylation variability. Previous studies have suggested that the transition from H3K27me3 to DNA methylation is a precursor event for cancer (25). High-throughput bisulphite sequencing of multiple cancer samples identified hundreds of differentially methylated regions (DMRs) in cancer among different cancer samples (26). These DMRs are highly colocalized with the HPRs defined herein based on the H3K27me3 patterns (P < 1E-40, Fisher’s exact test; ES = 32.4) (Fig. 1D), confirming the close correlation between these two epigenetic marks. To test whether this colocalization can be simply explained by their common association with CGIs, we repeated our analysis by further constraining the distance with respect to CGIs. By analyzing CGI proximal (<2 kb) and distal (>2 kb) regions separately, we found that the HPRs were strongly associated with DMRs in both groups, with a much higher enrichment score in CGI distal DMRs (SI Appendix, Fig. S4).

If variation of H3K27me3 levels at HPRs plays a significant role in transcriptional regulation, we should expect that the expression levels of their target genes are also highly variable. To avoid the ambiguity associated with target gene identification, we considered only the subset of HPRs located within promoter regions, and compared the variance of the expression levels of their downstream genes relative to that of the genomic background. We found an increased variance in expression levels associated with HPR-associated genes (Fig. 1F). We applied GREAT (27) to search for enriched functional categories associated with HPRs and identified “transcription regulatory region sequence-specific DNA binding” (P < 1E-40) and “sequence-specific DNA binding” (P < 1E-40) as among the most enriched categories, further supporting a role of HPRs in establishing cell-type–specific gene expression programs.

Recently, a number of groups (3, 28) have identified domain-like histone modification patterns by analyzing ChIP-seq data. We compared the locations of HPRs identified by our model with the domains detected in these studies. We observed that 58% of the H3K27me3 domains previously identified by ref. 28 (total 2,498, 100-kb size) overlap with at least one HPR (P < 1E-4, permutation test) (SI Appendix, Fig. S5A). Similarly, of the 9,397 H3K27me3 domains identified by ref. 3 in ES cells, 57% overlap with at least one HPR (P < 1E-4, permutation test) (SI Appendix, Fig. S5B). A smaller percentage (26%) of the IMR90-associated H3K27me3 domains overlap with HPRs (SI Appendix, Fig. S5C), probably because IMR90 was not included in the 19 cell lines based on which the HPRs were identified.

Taken together, the above results strongly suggest that the HPRs mark critical regulatory regions for establishment of cell-type–specific gene regulatory programs. Further characterization of variation pattern within the HPRs to identify the underlying sequences and associated factors will likely provide important mechanistic insights into the establishment and maintenance of the cell-type specific epigenetic patterns.

Prediction of H3K27me3 Plasticity from DNA Sequences.

To understand the molecular determinants of cell-type–specific HPRs, we analyzed DNA sequence signatures that were predictive of HPRs. Previous studies have identified a number of DNA sequence features associated with H3K27me3, including CGIs (10), TF sequence motifs (12, 13), and short RNA hairpins (29). However, these aforementioned studies were focused on one cell type (primarily ES cells) at a time; thus, to what extent the DNA sequence impacts the overall plasticity across cell-types remains poorly understood.

As an initial evaluation, we applied motif independent metric, which is a Kullback–Leibler distance-based method, to quantify DNA sequence specificity (30). We found that the HPRs were associated with a significantly higher degree of sequence specificity compared with the LPRs (P < 1E-4, permutation test), suggesting that DNA sequence information indeed contributes to the modulation of H3K27me3 variability. The 20 most informative k-mers are GC-rich, as expected (SI Appendix, Table S2).

The enrichment of CGIs in HPRs represents an extension of previous studies to multiple cell lines. In previous work, it was shown that a simple model based on the CGI distance and frequency alone was sufficient to predict many Polycomb targets in mouse ES cells (11, 12), although improved accuracy could be achieved by more sophisticated models (13). To test whether this property was applicable to H3K27me3 plasticity, we built a simple model using only the C+G and CpG density sequence features to distinguish HPRs from LPRs (named CG-only model; see Materials and Methods for details). As anticipated, the CG-only model already has substantial prediction power [area under the receiver operating characteristic curve (AUC) = 0.74] (Fig. 2A).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Prediction of H3K27me3 plasticity using DNA sequence patterns. (A) Receiver operating characteristic (ROC) curve for classification of HPRs vs. LPRs using either the N-score model (in red) or the CG-only model (in blue). The corresponding AUC scores are shown next to the ROC curves. (B) Genome-wide correlation between the observed and N-score model predicted plasticity scores. (C) Average profiles of observed and N-score predicted plasticity score and at high CpG promoters. (D) Average profiles of observed and N-score predicted plasticity score and at low CpG promoters. (E) Average profiles of observed and N-score predicted plasticity score and at enhancer regions.

We then assessed to what extent the prediction accuracy could be improved by incorporating additional sequence features. By applying the previously developed N-score model (31, 32) (Materials and Methods), we obtained an improved AUC score at 0.82 (Fig. 2A; see SI Appendix, Table S3 for the list of the most informative sequence features). Although this improvement seems moderate, it enabled us to predict not only CGI proximal, but also distal HPRs (SI Appendix, Figs. S6 and S7). For example, the improved performance of the N-score model is evident at the flanking regions of the HOXC locus (Fig. 1A). The genome-wide correlation between predicted and observed plasticity score is statistically significant (ρ = 0.29, P value < 1E-40) (Fig. 2B). Importantly, our model correctly predicted the overall pattern at promoters and enhancers (Fig. 2 C–E), including the significant differences between GC-rich and GC-poor promoters. It is important to note that the overall plasticity score at enhancers was low because of the presence of nonvariable H3K27me3 regions within all of the enhancers. As expected, when using only enhancers overlapping with the HPRs in the analysis, we found that the plasticity scores were much higher (SI Appendix, Fig. S8).

Classification of H3K27me3 Variability Within the HPRs.

We applied the k-means clustering method to identify distinct subpatterns of H3K27me3 variability within the HPRs (Fig. 3A). These patterns can be broadly classified into three groups: a “continuous” group (C8 and C9), where the H3K27me3 mark is present in most cell-types but its intensity is highly variable; a “binary” group (C1, C2, C3, C4, and C6), where the H3K27me3 mark is present mainly in a small number of cell types; and a “mixed” group that contains all other clusters. We observed a strong association between the continuous group and the overall GC-content (Fig. 3A). The proximal HPRs tend to be spatially extended, whereas the distal HPRs are more focal (Fig. 3B). These differences suggest that the CGI proximal and distal HPRs are regulated through distinct mechanisms.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Distinct variation patterns between subsets of HPRs. (A) k-means clustering of HPRs shows distinct patterns of variation. The color bar on the left shows H3K27me3 ChIP-seq read counts, and the color bar on the right shows CpG density. (B) Average plasticity scores around CGI proximal and distal HPRs. (C) Decomposition of the total HPRs into CGI proximal and distal subsets.

Of note, 44% of the total HPRs are CGI distal (Fig. 3C). Although the role of promoter-associated H3K27me3 in gene silencing has been extensively studied, its role in distal regions remains less well understood. Previous studies suggested that H3K27me3 may work together with H3K4me1 in marking poised enhancers, which could be activated by switching to the H3K27ac mark (15, 16), To quantitatively assess the degree of colocalization between distal HPRs and enhancers, we applied chromHMM (33), a recently developed method for chromatin-state segmentation, to identify putative enhancers based on the combinatorial pattern of nine histone marks. Using this approach, we were able to annotate the enhancer regions in 12 of the 19 cell-lines, including 9 cell-lines for which the enhancers have been annotated previously (5). We further excluded regions that are within 2 kb of an annotated transcription start site and CGIs, resulting in a remaining set of 57,815 HPRs that overlapped with chromHMM identified enhancers. Fourteen percent of the distal HPRs are located in the enhancer regions, which is highly statistically significant (9% expected by chance; P < 1E-40).

Identification of Distal HPR-Associated Cell-Type–Specific TFs.

To systematically identify candidate TFs that may modulate H3K27me3 variability in a cell-type–specific manner, we developed a unique bioinformatic pipeline that consists the following five steps (Fig. 4A and SI Appendix, Fig. S9). First, we defined a z-score by comparing the H3K27me3 intensity in each cell-type relative to the other 18 cell-types, and selected a subset of HPRs (which we named z-HPRs) with the highest z-scores (>2) for further analysis. Second, we generated a genomic background with a matching C+G content. Third, we scanned the DNA sequences at z-HPRs for enrichment of known TF motifs in the TRANSFAC (34), JASPAR (35), and FACTORBOOK (36) databases, by comparing with the matched genomic background (P value cutoff = 1E-4). Fourth, we compared the motif site frequency in the center of z-HPRs with the flanking regions, and selected the motifs that were enriched in the center regions (fold-change > 1.2). Fifth, we further integrated gene expression data to select for candidate TFs that were specifically expressed in the cell-type (z-score > 0.75) whereas its target genes neighboring the z-HPRs were repressed (z-score < -0.75).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Identification of cell-type–specific z-HPR regulators. (A) A flowchart of the computational pipeline for identification of cell-type specific z-HPR regulators (details shown in the main text and SI Appendix, Fig. S5). (B–F) Application of the above pipeline identifies PAX5 as a candidate regulator for GM12878/GM06990 z-HPRs. (B) The PAX5 motif logo in the JASPAR database. (C) Average enrichment profile of PAX5 motif sites around distal z-HPRs. (D) Correlation between the expression levels of PAX5 and target genes neighboring z-HPRs across different cell lines. Expression levels are normalized by z-scores. The red star corresponds to the GM12878/GM06990 cell lines. Average profile (E) and heatmap (F) of ChIP-seq signal for H3K27me3 and PAX5 around distal z-HPRs.

In total, our analysis identified 41 cell-type–specific associations between TFs and z-HPRs (Table 1). Several of these associations were implicated in the literature. For example, our analysis identified paired box 5 (PAX5) as a candidate Polycomb recruitment factor in B-lymphoblastoid cells (based on GM12878/GM06990 cell lines) (Fig. 4 B–D). PAX5 is known to be essential for normal B-cell development by activating B-cell commitment genes, while concomitantly repressing non–B-cell lineage genes (37). Depletion of PAX5 in pro-B cells resulted in significantly reduced levels of H3K27me3 at V_H genes, suggesting a role of PAX5 in the recruitment of PRC2 complex (38). To validate our predictions, we analyzed the colocalization between z-HPRs and PAX5 binding sites, using a ChIP-seq dataset generated by the ENCODE consortium. This analysis demonstrated that PAX5 binding signal was indeed highly enriched at z-HPRs (Fig. 4 E and F).

Furthermore, our analysis identified GATA binding protein 1 (GATA1) to be strongly associated with z-HPRs in the human erythroleukemia cell line K562 (SI Appendix, Fig. S10 A–C). GATA1 is a master regulator for erythroid and megakaryocyte development (39) and its mutation has been linked to Down syndrome and acute megakaryoblastic leukemia (40). Previous work showed that GATA1 employs the PRC2 complex in gene repression in erythroid cells (41). To validate our predictions, we obtained a GATA1 ChIP-seq dataset in K562 cells generated from the ENCODE consortium. Similar to PAX5 in B cells, we found that the GATA1 ChIP-seq signal was highly enriched at the z-HPRs in K562 cells (SI Appendix, Fig. S10 D and E).

In addition, our analysis suggested that AP2-α [or transcription factor AP-2 alpha (TFAP2A)] may recruit Polycomb in normal human epidermal keratinocytes (NHEK) (SI Appendix, Fig. S11), consistent with a role of AP2-α in regulating keratinocyte-specific gene expression (42). Similarly, ZFX (zinc finger protein, X-linked) was associated with z-HPRs in SK-N-SH_RA cells (a neuroblastoma cell line) (SI Appendix, Fig. S12), consistent with an implicated role of ZFX in aggressive gliomas (43). These analyses not only validated the computational predictions of our bioinformatics pipeline, but also provided links between lineage-restricted TFs and the Polycomb regulators that may be functionally important for their roles in respective cell lineages.

Computational Analysis Identifies TAL1 in Regulating H3K27me3 Variability.

Recently we have applied genomic methods to query the regulatory mechanism underlying developmental-stage changes of gene activities, using ProE cells as a model system (6). We found that these changes were mainly mediated by the enhancer activities, whereas the promoter activities were nearly identical between developmental stages. In addition, we observed a small fraction but still significant number of enhancers to be transcriptionally “poised” (1,249 of 12,960 total enhancers), as defined by the presence of both H3K4me1 and H3K27me3 marks (15, 16).

To gain mechanistic insights into the cell-type–specific H3K27me3 activity in the ProE cells, we applied the computational pipeline described above to determine ProE-specific H3K27me3 modulators. To this end, we adjusted the choice of HPRs by merging the H3K27me3 data in ProE and the other 19 cell lines. Of the total HPRs, 78% were unaffected by this adjustment, suggesting that incorporating a new dataset had only a moderate effect on the overall HPR selection. On the other hand, the adjustment is useful for identifying variability signal that is specific to the ProE cells. Among the 20,174 total ProE z-HPRs, 33% are located in distal regions. Importantly, we observed a statistically significant overlap between the ProE-specific z-HPRs and poised enhancers (160 common regions, as opposed to 5 expected by chance, P < 1E-4 permutation test) (Materials and Methods). We then identified eight TFs that are associated with these distal z-HPRs (Table 2). Of note, among these identified TFs, four were previously reported to physically interact [TAL1-LIM domain only 2 (LMO2), nescient helix loop helix 1 (NHLH1)-LMO2, and TAL1-epididymal sperm binding protein 1 (ELSPB1)], suggesting that they cooperate in regulating H3K27me3 variability and Polycomb activities (44). Importantly, our analysis identified TAL1, a principal regulator of hematopoietic development (39), as a candidate regulator of ProE-specific H3K27me3 variability (Fig. 5 A–C). TAL1 is commonly known as a transcriptional activator in various hematopoietic lineages (39), although a role for gene repression has recently been described in a different cellular context (embryonic endothelium) (45). Our analysis suggested that TAL1 might also contribute to transcriptional repression in hematopoietic cells by modulating the activity of poised enhancers. To test this hypothesis, we examined genome-wide TAL1 chromatin occupancy in ProE cells by ChIP-seq analysis. Consistent with the predicted association between TAL1 motifs and distal HPRs (Fig. 5 A–C and Table 2), we observed a significant enrichment of TAL1 ChIP-seq signal within distal HPRs in ProE cells (Fig. 5 D and E).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

TAL1 is predicted to be associated with z-HPRs in ProE cells. (A) TAL1 motif logo from the TRANSFAC database. (B) Average enrichment profile of TAL1 motif sites around distal z-HPRs. (C) Correlation between the expression level of TAL1 and target genes neighboring distal z-HPRs. The red star corresponds to the ProE cells. (D) Average ChIP-seq signals surrounding H3K27me3 and TAL1 around distal z-HPRs. (E) Heatmaps of ChIP-seq signals surrounding H3K27me3 and TAL1 around distal z-HPRs. Each row represents a different z-HPR region. The color bar represents the level of ChIP-seq signal.

To gain functional insights, we applied GREAT (27) to identify enriched Gene Ontology (GO) categories associated with the poised enhancers. One of the most enriched GO biological processes was “regulation of heart contraction” [false-discovery rate (FDR) q-value = 1.1E-3], and one of the strongly associated mouse phenotypes was “congestive heart failure” (FDR q-value = 7.4E-4) (SI Appendix, Table S4). Interestingly, in the aforementioned study (45), the genes repressed by TAL1 were also associated with cardiomyogenesis. Despite the differences between the two model systems, we observed a significant overlap (58 of 965, P = 5.2E-6) between the genes harboring a poised enhancer in ProE cells and those up-regulated by TAL1 depletion in embryonic endothelium (45). In contrast, TAL1-H3K27ac genes were enriched with biological processes such as “regulation of erythrocyte” (P = 1.5E-16), “negative regulation of cell-cell adhesion” (P = 5.4E-11), and “response to iron ion” (P = 6.2E-6), consistent with the role of TAL1 in activating genes required for normal erythroid development (46, 47).

TAL1 Modulates H3K27me3 Variability at Poised Enhancers in Erythroid Cells.

The above findings suggest that TAL1 plays a dual role in regulating gene-expression programs in primary human erythroid cells. We next investigated the context differences between TAL1–z-HPR and TAL1-H3K27ac regions using motif analysis. The most enriched motif in TAL1–z-HPR regions corresponds to the consensus binding sequence for GFI1B (Fig. 6A), a transcriptional repressor critical for normal hematopoietic development and oncogenesis (48, 49). The transcriptional corepressor complex lysine (K)-specific demethylase 1A (LSD1)/CoREST was previously shown to be required for GFI1B-mediated transcriptional repression in erythroid cells (50). In contrast, the most significantly enriched motif in TAL1-H3K27ac regions was the TAL1:GATA1 composite motif (Fig. 6A and SI Appendix, Fig. S13A). Gene-expression analysis showed that GFI1B was up-regulated, whereas its target genes were down-regulated in ProE cells (Fig. 6B). An opposite trend was associated with GATA1/TAL1 common targets (Fig. 6B). These differences are consistent with our computational prediction based on sequence analysis.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 6.

GFI1B is a cofactor for TAL1-mediated H3K27me3 variation in ProE cells. (A) GFI1B and TAL1::GATA1 motif logos from the TRANSFAC database. (B) Correlation between the expression level of GFI1B (Left) or GATA1 (Right), and target genes neighboring TAL1+z-HPRs (or TAL1+H3K27ac). (C) Average ChIP-seq signal for multiple factors surrounding TAL1+z-HPR (Left) or TAL1+H3K27ac (Right) regions. (D) Sequential ChIP analysis in ProE cells with the indicated pairs of antibodies. ChIP DNA was analyzed by quantitative PCR for selected TAL1-GFI1B-H3K27me3 cooccupied z-HPRs and negative control regions (GAPDH and ACTB). The percent (%) of input is shown. Results are mean ± SD from at least three technical replicates. (E) Co-IP experiments for EZH2, TAL1, GATA1, and GFI1B using nuclear extracts from ProE cells. MTA2 and SIN3A are analyzed as negative controls.

To experimentally test whether TAL1 and GFI1B cooperate to regulate H3K27me3 at poised enhancers, we generated GFI1B and GATA1 genome-wide binding profiles in ProE cells by ChIP-seq analysis. Importantly, the ChIP-seq signal for GFI1B was substantially higher at TAL1–z-HPRs than the TAL1-H3K27ac regions, whereas the GATA1 ChIP-seq signal was stronger at the TAL1-H3K27ac regions than TAL1–z-HPRs (Fig. 6C). This difference was highly consistent with the computational predictions, suggesting that TAL1 and GFI1B may function cooperatively within z-HPRs. Additionally, the target genes of GFI1B-TAL1-H3K27me3 cooccupied regions were expressed at lower levels compared with those corresponding to TAL1-H3K27me3 or TAL1-H3K27ac, respectively (SI Appendix, Fig. S13B). To further determine whether TAL1, GFI1B and H3K27me3 cooccupy on the same allele in ProE cells, we performed sequential ChIP analysis. Sequential ChIP analysis of TAL1 and H3K27me3 demonstrated colocalization at nine out of ten representative z-HPRs (Fig. 6D and SI Appendix, Fig. S14A). Similarly, sequential ChIP analysis of GFI1B and H3K27me3 showed colocalization at 8 of 10 z-HPRs (Fig. 6D). The reverse direction of TAL1-H3K27me3 and GFI1B-H3K27me3 sequential ChIP also indicated colocalization at the majority of z-HPRs (SI Appendix, Fig. S14 B and C). Taken together, these results are consistent with the colocalization of TAL1, GFI1B, and H3K27me3 at a genomic scale (Fig. 6C), suggesting that TAL1 and its cofactor GFI1B cooperate on the same alleles to modulate H3K27me3 in ProE cells.

To further establish the causality of TAL1 occupancy and H3K27me3 variability, we ectopically expressed TAL1 in the human lymphoid (REH) and embryonic kidney (293T) cell lines, followed by ChIP-quantitative PCR analysis of H3K27me3 at the representative ProE-specific z-HPRs (SI Appendix, Fig. S15). Notably, a modest increase of H3K27me3 and a significant increase of TAL1 binding were observed at three loci (UGT2B4, INTS4, and ATP4A), respectively. Similarly, ectopic expression of TAL1 in 293T cells led to modest increase of H3K27me3 at two loci (PID1 and INTS4). Taken together, these results suggest that TAL1 can gain access to at least a subset of ProE-specific z-HPRs in other cellular context, accompanied by increased H3K27me3 at the same genomic location (SI Appendix, Fig. S15).

The colocalization of TAL1 and GFI1B within z-HPRs suggests that TAL1, GFI1B, and the PRC2 complex may physically interact to modulate H3K27me3 in a context-specific manner. To test this hypothesis, we performed a coimmunoprecipitation (co-IP) experiment using nuclear extracts prepared from ProE cells. Importantly, EZH2, the histone methyltransferase subunit of the PRC2 complex, could efficiently pull-down both TAL1 and GFI1B in ProE cells (Fig. 6E). Similarly, both TAL1 and GFI1B could pull-down the endogenous EZH2 protein, respectively, whereas no detectable interactions were found for MTA2 (a core subunit of the Mi-2β/NuRD complex) and SIN3A (a core subunit of the transcriptional corepressor and histone deacetylase complex SIN3A) (Fig. 6E). These results suggest that TAL1, GFI1B, and PRC2 physically interact and cooperate within the same complex that functions to regulate H3K27me3 marks at a subset of gene-distal regulatory elements.

Taken together, these data strongly support a model in which TAL1 may mediate gene repression by modulating H3K27me3 variation through interaction with transcriptional cofactors GFI1B and the Polycomb regulators. Such context-dependent regulation enables a single master regulator to simultaneously control multiple gene-expression programs, which may in turn enhance the precision of cell-fate decisions during development.