Brg1 modulates enhancer activation in mesoderm lineage commitment Open Access

The emergence of individual cell types during development relies on the correct sets of genes becoming activated, while inappropriate sets of genes are simultaneously repressed. This process is achieved in large part by modifying chromatin structure, which packages the genome within the nucleus and affects multiple facets of gene regulation (Ho and Crabtree, 2010).

Histone modifications annotate the genome by marking active, repressed and other functional domains singly or in combination (Zhou et al., 2011). For example, trimethylation of histone H3 lysine 27 (H3K27me3), deposited by the Polycomb repressive complex 2 (PRC2), is associated with gene silencing (Surface et al., 2010). In embryonic stem cells (ESCs), PRC2 targets a broad group of developmental regulators for silencing to ensure appropriate lineage-specific gene expression (Surface et al., 2010). Conversely, acetylated histone H3 lysine 27 (H3K27ac) is a hallmark of active chromatin and, in particular, of enhancers, non-coding DNA elements that regulate tissue-specific gene expression patterns (Calo and Wysocka, 2013). Enhancer occupancy by H3K27ac is highly dynamic during cellular differentiation, yet the factors that modulate H3K27ac occupancy at promoters and enhancers remain poorly understood.

Eukaryotes use ATP-dependent chromatin remodelers to unwind, slide and/or evict individual nucleosomes (Ho and Crabtree, 2010; Wu et al., 2009). In mammals, the Swi/Snf-like Brg1/Brm-associated factor (BAF) chromatin-remodeling complexes include 10-12 interchangeable subunits and function in regulating cell-cycle progression, DNA repair and development (Ho and Crabtree, 2010). Remodeling activity is mediated by the ATPase subunit, which is encoded by either Brg1 (also known as Smarca4) or the related gene Brm (also known as Smarca2). Brg1 is essential for embryonic development and maintenance of pluripotency, whereas Brm knockout mice are viable (Bultman et al., 2000; Ho et al., 2009; Reyes et al., 1998), arguing that Brg1 is the consequential ATPase during development. Brg1 is required in numerous tissue and cell types in vivo (Ho and Crabtree, 2010), including multiple cell types within the cardiac lineage (Hang et al., 2010; Stankunas et al., 2008; Takeuchi and Bruneau, 2009; Takeuchi et al., 2011). How BAF complexes function to regulate gene expression within these cell lineages is still poorly understood.

A comprehensive picture of Brg1 function during development necessitates detailed understanding not only of the regulatory loci bound by Brg1 but also its functional activity at these regions, including how this activity facilitates the transitions between distinct chromatin states – marked by histone modifications – that occur during cell differentiation. Such questions bear on fundamental features of chromatin regulation, namely how epigenetic modification and chromatin remodeling are deployed during development to modify the chromatin template in a coordinated fashion. Brg1 is found at distal regulatory regions (Euskirchen et al., 2011; Hu et al., 2011; Morris et al., 2014; Rada-Iglesias et al., 2011; Yu et al., 2013), but the role of Brg1 at these regions remains poorly understood. Here, we investigate the function of Brg1 during embryonic stem cell differentiation. We find that loss of Brg1 leads to disruption of cardiomyocyte differentiation and dysregulation of lineage-specific gene expression during mesoderm induction. Furthermore, we find that Brg1 is required for robust H3K27 acetylation, predominately at distal enhancers that transition from inactive to active states during mesoderm induction. Brg1 is also required to maintain Polycomb-mediated repression of non-mesodermal developmental regulators through deposition of H3K27me3, suggesting cooperativity between Brg1 and Polycomb complexes.

To gain deeper insight into how chromatin is regulated during cardiac differentiation, we analyzed our published expression datasets (Wamstad et al., 2012) to identify the expression patterns of known chromatin regulators. We selected genes annotated with involvement in chromatin remodeling (Gene ontology category GO0006338) or covalent chromatin modification (GO00016569), and clustered their gene expression patterns across four stages of cardiomyocyte differentiation: ESCs, mesodermal precursors (MES), cardiac precursors (CP) or functional cardiomyocytes (CMs) (supplementary material Table S1). This analysis classified chromatin regulators into three expression patterns (Fig. 1A). We identified one cluster with high expression in ESCs and reduced expression upon exit of the pluripotent state. Within this group were the de novo DNA methyltransferase Dnmt3b, which modulates DNA methylation levels in the early mouse embryo (Okano et al., 1999), and Dpy30, a member of MLL family complexes that is required for differentiation of ESCs (Jiang et al., 2011) (Table 1). Expression in a second cluster peaked at the MES stage, followed by downregulation in differentiated CMs. Many BAF complex subunits demonstrated this expression pattern, including the essential core subunits Baf57 (Smarce1 – Mouse Genome Informatics) and Baf47 (Smarcb1 – Mouse Genome Informatics) and the enzymatic subunit Brg1 (Ho and Crabtree, 2010; Wu et al., 2009). The last expression cluster demonstrated increased CM expression; within this group were Smyd1, a known regulator of cardiac development (Gottlieb et al., 2002); Hdac9, which modulates the hypertrophic response (Zhang et al., 2002); and Jmjd3 (Kdm6b – Mouse Genome Informatics), a histone H3 lysine 27 demethylase. This cluster also contained multiple components of BAF complexes, including Baf60c (Smarcd3 – Mouse Genome Informatics), Brm, Baf170 (Smarcc2 – Mouse Genome Informatics), Baf45c (Dpf3 – Mouse Genome Informatics), Baf45d and Baf250b (Arid1b – Mouse Genome Informatics), suggesting that BAF complexes undergo subunit switching during cardiomyocyte differentiation analogous to that observed in the nervous system (Ho and Crabtree, 2010; Wu et al., 2009).

Dynamic expression of BAF complex subunits suggested that these complexes play distinct roles at different stages of cardiac differentiation. We confirmed that Brg1 protein is more abundant at early stages of cardiac differentiation than in late-stage cultures enriched for differentiated CMs (Fig. 1B). The enrichment of Brg1 in mesoderm precursors suggested that BAF complexes perform a broad and uncharacterized function in the early progenitors of the cardiac lineage.

To investigate the function of Brg1 at distinct stages of cardiac differentiation, we used directed CM differentiation (see supplementary material Methods) in a mouse ESC line with two floxed alleles of Brg1 and a constitutively expressed Cre recombinase-estrogen receptor fusion protein (Brg1^f/f;Actin-CreER) (Ho et al., 2011,, 2009). Adding 4-hydroxytamoxifen (4-OHT) led to efficient deletion of the floxed allele and loss of Brg1 protein, which was near-complete 48 h after 4-OHT treatment and undetectable by 72 h (Fig. 1C). Treatment of differentiating cultures with 4-OHT allowed for controlled deletion of Brg1 at specific stages of differentiation and allowed for the comparison of control and treatment groups within a single differentiation, which limited confounding effects from differentiation variability. We added 4-OHT at three time points: during the induction of MES (day 2), as mesodermal precursors are differentiating towards CMs (day 4), and after the appearance of beating CMs (day 8) (Fig. 1D). Cultures treated with 4-OHT at day 8 showed no discernable differences from control cultures: they continued to contract many days after addition of 4-OHT and had comparable numbers of cardiac troponin T (cTnT; Tnnt2 – Mouse Genome Informatics)-positive CMs (Fig. 1D,E). By contrast, cultures treated with 4-OHT at day 2 or day 4 had fewer cTnT⁺ CMs than controls. This was most striking in cultures treated with 4-OHT at day 2, which had a near-complete loss of cTnT⁺ CMs. Whereas control-treated cultures expand to generate dense layers, containing multiple fibers of interconnected CMs after mesoderm induction, cultures treated with 4-OHT at day 2 failed to expand in these conditions, giving rise to a sparse monolayer of differentiated cells (supplementary material Fig. S1A,B). Treatment of wild-type ESCs undergoing the same differentiation protocol with 4-OHT did not affect their ability to differentiate (supplementary material Fig. S1C). Thus, Brg1 is required for the differentiation of embryonic stem cells to CMs.

As addition of 4-OHT at day 2 would lead to deletion of Brg1 during mesoderm induction, we examined mesodermal markers during differentiation of ESCs. We differentiated Brg1^f/f;Actin-CreER ESCs for 2 days as embryoid bodies (EBs) in serum-free medium and induced MES by treating these cultures with Vegf, activin A (Inhba – Mouse Genome Informatics) and Bmp4 in the presence of 4-OHT or vehicle control for 40 h, analogous to the first 4 days of our directed CM differentiation protocol. We measured the induction of Flk-1 (also known as Kdr) and Pdgfra, receptor tyrosine kinases expressed on cardiogenic mesodermal cells (Kattman et al., 2011). Whereas control cultures showed robust induction of Pdgfra by flow cytometry, Brg1-deleted cultures showed a clear reduction in the number of Pdgfra- and Flk1-expressing cells (n=3; Fig. 2A). Similarly, Brg1-deleted cultures showed reduced expression of the mesodermal marker Mesp1 (Fig. 2B). The remaining expression still observed for Pdgfra, Flk-1 and Mesp1 in these experiments might be the result of residual Brg1 activity, as mesoderm is induced concomitant with addition of 4-OHT, leading to a gradual loss of Brg1 protein during mesoderm induction. Taken together, these data demonstrate that Brg1 is required in ESCs for robust induction of molecular markers of mesoderm.

To identify the Brg1-dependent transcriptional program during mesoderm differentiation, we collected cultures before mesoderm induction (day 2) and cultures 40 h after treatment with Vegf, activin A and Bmp4 (day 4) that had been treated with either 4-OHT or control, and measured global gene expression by RNA-seq (Fig. 3A). This allowed identification of the transcriptional changes that occur normally during mesoderm differentiation, in addition to genes differentially expressed between 4-OHT and control. In this way, we could determine whether Brg1-dependent genes demonstrate a common expression pattern during mesoderm induction. Using stringent criteria (FDR=1%, fold change ≥2), we identified 350 downregulated and 502 upregulated genes in Brg1-deleted cultures (Fig. 3B; supplementary material Table S2). This analysis demonstrated that Brg1 was downregulated more than tenfold in Brg1-deleted cultures, confirming the efficacy of the genetic deletion. Among the downregulated genes were Flk1 and Mesp1, mesodermal markers that demonstrated reduced induction by flow cytometry or quantitative PCR. We also found other genes essential for mesoderm development, of which the expression was reduced by loss of Brg1, including Cxcr4, Cyp26a1 and Snai1. Cxcr4 is expressed in Flk-1/Pdgfra-expressing mesoderm and mediates differentiation towards the cardiac lineage. Cyp26a1 modulates retinoic acid signaling, a crucial regulator of mesodermal patterning (Aulehla and Pourquié, 2010; Chiriac et al., 2010; Nelson et al., 2008). Snai1 is a transcriptional repressor that controls epithelial-to-mesenchymal transition (EMT) and mesoderm morphology in vivo (Carver et al., 2001). Gene ontology (GO) analysis revealed that downregulated genes were enriched for genes involved in cell adhesion as well as those associated broadly with development (multicellular organismal process) and signaling (molecular transducer activity) (Table 2; supplementary material Table S3). These findings are consistent with defective induction of mesodermal genes in Brg1-deficient cultures. Pdgfra, which had reduced expression, as measured by flow cytometry (Fig. 2A), was downregulated 1.93-fold in our RNA-seq analysis and thus fell slightly below our stringent criteria for significance. Therefore, our statistical cutoff is probably a conservative estimate of Brg1-dependent gene expression.

Examination of the genes upregulated by loss of Brg1 (Fig. 3B) revealed Wnt signaling ligands (Wnt8a, Wnt9a, Wnt7b, Wnt4, Wnt10a and Wnt6), Wnt antagonists (Sfrp1 and Frzb), and numerous developmental transcription factors (TFs). The latter included TFs from the Fox, Tbx, Mef2, Dlx, Runx, Pax, Lhx, Six, Nkx, Sox, Pou, Cdx, Irx and Hox TF families. GO analysis was consistent with this finding; upregulated genes demonstrated strong enrichment for genes associated with development, morphogenesis and transcription factor function (Table 2). Investigation of expression data from a broad range of murine tissues and cell types confirmed that upregulated TFs are expressed in many distinct and non-overlapping lineages (supplementary material Fig. S2), demonstrating that loss of Brg1 does not result in differentiation of ESCs towards a single, non-mesodermal lineage. Strikingly, many (18 of 38) Hox genes, representing all four Hox clusters, were upregulated in Brg1-deleted mesoderm.

Within genes significantly upregulated by loss of Brg1 were numerous TFs that function during heart development, despite the striking defect for these cultures to generate cardiomyocytes at subsequent stages. This group included the well-characterized regulators of skeletal and cardiac myogenesis Myocd and Mef2b as well as conserved cardiogenic factors Tbx5 and Nkx2-5. Of particular note was Nkx2-5, the expression of which increased more than sevenfold in Brg1-deficient mesoderm. Our previous studies (Wamstad et al., 2012) have shown that these factors are expressed at low levels in mesodermal cultures, becoming robustly expressed only later, concomitant with the onset of cardiomyocyte differentiation. We did not observe broad upregulation of markers of cardiomyocytes in these cultures, suggesting instead that the upregulation of these cardiogenic TFs reflects the broader misexpression of inappropriate developmental regulators in Brg1-deficient mesodermal cultures.

Brg1 might function to facilitate dynamic changes in gene expression that occur during mesoderm induction or might be required to maintain active and repressed transcriptional states. To better understand the function of Brg1 in transcriptional regulation of mesoderm differentiation, we investigated the expression patterns of Brg1-dependent genes during this process. We rank-ordered genes based on fold change in gene expression during normal mesoderm differentiation (day 4 control versus day 2) and compared normal and Brg1-deleted mesoderm differentiation. We limited our analysis to only those genes measured by RNA-seq in all three experimental conditions. We found that the majority (82%) of genes downregulated by loss of Brg1 are activated during normal mesoderm induction (Fig. 4A, left bar); Brg1 deletion during mesoderm induction led to less robust activation for these genes (Fig. 4A, right bar). By contrast, genes upregulated by loss of Brg1 showed a tendency for repression during normal mesoderm differentiation. We found that upregulated genes were generally expressed at very low levels in normal mesodermal cultures (supplementary material Fig. S3). Loss of Brg1 leads to derepression of these genes during mesoderm differentiation. Collectively, our global gene expression analysis supports broad roles for Brg1 in gene activation and maintaining gene repression of key developmental regulators during mesoderm differentiation of ESCs.

We next investigated to what extent Brg1 is required for transcriptional change during mesoderm induction. We identified genes differentially expressed between day 4 control and day 2 (FDR=1%, fold change ≥2), categorized these genes as either activated or repressed during mesoderm induction, and overlapped these gene sets with Brg1-dependent genes. This analysis revealed that 18% of genes activated during normal mesoderm differentiation (i.e. significantly higher expression at day 4) had reduced expression in Brg1-deleted mesodermal cultures, compared with just 2% of repressed genes. Mesoderm-activated genes were considerably enriched for those dependent on Brg1 for expression (Fig. 4B). This demonstrates that a substantial proportion of the mesoderm transcriptional program requires Brg1 for proper activation.

To better understand the mechanism by which Brg1 affects the mesodermal transcriptional program, we defined the genomic regions bound by Brg1 in mesodermal cultures. To this end, we used ESCs harboring a Brg1 allele that encodes a 3×-FLAG epitope tag fused to the C-terminal end of Brg1 targeted to the endogenous Brg1 locus (Attanasio et al., 2014). We confirmed the expression of Brg1-FLAG in cultured pluripotent ESCs (supplementary material Fig. S4). Purification of Brg1-FLAG using an anti-Flag column yielded a staining pattern that closely resembled those published for BAF complexes (Ho et al., 2009; Wang et al., 1996) (supplementary material Fig. S4), consistent with Brg1-FLAG incorporation into BAF complexes. Mass spectrometry of isolated complexes revealed a composition highly similar to previously reported esBAF complexes (Ho et al., 2009) (data not shown). Mice homozygous for the Brg1-FLAG allele are viable, further indicating that the allele is fully functional, and ChIP-seq data, obtained in mouse tissues using this Brg1-FLAG allele, strongly correlated with published Brg1 ChIP-seq data obtained with antisera (Attanasio et al., 2014). We differentiated Brg1-FLAG ESCs to mesodermal precursors and performed chromatin immunoprecipitation and deep sequencing (ChIP-seq) on biological duplicate samples. FLAG ChIP-seq replicates overlapping H3K27ac-enriched regions were well correlated (r²=0.71) and showed modest enrichment that probably reflects the transient and dynamic nature of chromatin remodeler binding. To identify high-confidence Brg1-bound regions from these data, we overlapped statistically enriched peaks identified through an input-corrected Poissonian model across both replicates (Marson et al., 2008) and identified 3027 bound regions distributed throughout the mouse genome (see supplementary material Methods, Table S4). Given the modest enrichment of our FLAG ChIP-seq dataset, we expect these regions, statistically enriched in both biological replicates, to represent a conservative estimate of Brg1 occupancy.

To validate Brg1-FLAG-bound regions, we performed ChIP-exo, using an antibody against endogenous Brg1 (Morris et al., 2014) in Brg1-FLAG ESCs differentiated to mesodermal precursors in biological duplicates. As expected, our 3027 Brg1-FLAG-bound regions demonstrated modest but clear enrichment for Brg1 in the ChIP-exo dataset (Fig. 5A), with occupancy characteristics similar to published results (Morris et al., 2014; Shi et al., 2013). Correlation between anti-FLAG and anti-Brg1 ChIP over H3K27ac-enriched regions is 0.52. To further validate these regions and the specificity of the antibody, we performed Brg1 ChIP-exo on differentiating Brg1^f/f;Actin-CreER ESCs treated with either THF (control) or 4-OHT (deletion of Brg1) for 48 h. Brg1 enrichment, modest but clear in THF-treated samples, was greatly reduced in Brg1^f/f;Actin-CreER ESCs treated with 4-OHT. Taken together, we have defined Brg1-bound regions consistently enriched by different methods, which are sensitive to genetic deletion of Brg1. We acknowledge that the modest enrichment only allows the identification of high-confidence high-enrichment regions; therefore, our conclusions regarding the direct function of Brg1 are limited to these regions.

Comparison of Brg1-bound regions with genomic annotations revealed Brg1 binding within gene promoters, introns, exons and intergenic regions (Fig. 5B). We identified 691 genes with reproducible binding of Brg1 within 2.5 kb of the transcriptional start site. Some genes in this group were also significantly changed in our RNA-seq dataset. However, an intersection of Brg1-bound promoters with Brg1-dependent genes revealed little overlap (Fig. 5C), as observed for other chromatin remodelers (Gelbart et al., 2005; Sala et al., 2011). We find the majority of genes with Brg1 promoter enrichment do not show significant changes in gene expression. Moreover, most Brg1-dependent genes lacked robust Brg1 binding within the promoter. This suggests that Brg1 does not predominately modulate gene expression through promoter regulation in differentiating mesoderm.

Brg1 localizes to well-characterized enhancers (Bultman et al., 2005) and predicted enhancers genome-wide (Euskirchen et al., 2011; Hu et al., 2011; Rada-Iglesias et al., 2011; Yu et al., 2013). Given that only 23% of Brg1 peaks were found within promoter regions, we hypothesized that Brg1 functions at distal enhancers. To test this, we generated genome-wide maps of H3K27ac in control and 4-OHT-treated mesodermal cultures in biological duplicates. H3K27ac marks active enhancers and can be used to identify putative distal regulatory elements genome-wide (Calo and Wysocka, 2013). Comparison of Brg1-FLAG and H3K27ac ChIP-seq signals at Brg1-enriched loci showed substantial enrichment for H3K27ac at Brg1-bound regions (Fig. 5A; see also Attanasio et al., 2014). Moreover, our ChIP-seq data revealed a high correlation between Brg1 and H3K27ac signals throughout the genome (Fig. 5D; supplementary material Fig. S5). Using our genome-wide maps of H3K27ac, we identified 16,724 putative enhancer regions distal (>2.5 kb) from the transcriptional start site. Strikingly, we found that 68% of Brg1-bound regions distal to transcriptional start sites overlapped with predicted enhancer regions (P<0.0001; 10,000 permutations). Thus, Brg1 associates with a proportion of H3K27ac-marked enhancers genome-wide in mesodermal cultures.

We searched Brg1-bound enhancers for enriched transcription factor DNA binding motifs that might predict a mechanism for site-specific recruitment of Brg1. H3K27ac⁺ Brg1-bound regions were scanned using the ‘match’ algorithm of TRANSFAC. A number of motifs were significantly enriched (q<0.001), with many belonging to well-known regulators of mesodermal differentiation, including T-box, GATA and Fox factors, which function in a highly interactive network (Fig. 5E). Although enrichments were highly significant, fold enrichment in motif abundance between Brg1-associated enhancers and all enhancers was modest (between 1.11- and 1.95-fold enrichment). We conclude that Brg1-bound enhancers are enriched for specific developmental TFs, but it is unlikely that these factors alone direct Brg1 occupancy, and might instead reflect bias in Brg1 recruitment to developmentally regulated enhancers.

The presence of Brg1 at enhancers suggests that Brg1 regulates transcriptional activation during mesoderm induction through modulation of enhancer activity. We therefore asked whether H3K27ac levels were altered in Brg1-deleted mesodermal cultures. To this end, we compared H3K27ac genome-wide maps from control and 4-OHT-treated mesodermal cultures, and rank-ordered promoter and enhancer regions based on fold change in H3K27ac in Brg1-deleted cultures. We found that levels of H3K27ac were largely unchanged at promoter regions (median log₂-fold change=0.07), although downregulated genes showed clear reductions in H3K27ac levels proximal to the TSS, probably reflecting decreased transcriptional activity at these genes (supplementary material Fig. S6). In contrast to promoter regions, we observed a global reduction in H3K27ac levels at predicted enhancer regions in Brg1-deleted cultures (median log₂-fold change=−0.39) (Fig. 6A). Consistent with a functional role for enhancer activity in the transcriptional changes seen in Brg1-deleted cultures, we observed a correlation between changes in H3K27ac seen at an enhancer and changes in expression of its nearest gene (Fig. 6B). This relationship was observed despite limitations of computational approaches in predicting enhancer-gene regulatory pairs. Furthermore, we found that enhancers proximal to significantly downregulated genes showed greater reductions in H3K27ac compared with all putative enhancers (Fig. 6C). These findings support a functional role for Brg1-dependent enhancer activity in the transcriptional control of mesoderm differentiation.

To investigate whether Brg1 regulates enhancer activity directly through its recruitment to these loci, we partitioned mesodermal enhancers into Brg1-bound or Brg1-unbound cohorts. We found that Brg1-bound enhancers showed greater losses in H3K27ac than those without Brg1-enrichment, providing evidence that Brg1 directly modulates H3K27ac at enhancers (Fig. 6D). We also observed greater occurrence of Brg1-bound enhancers proximal to significantly downregulated genes than expected by chance alone (P=0.0002, hypergeometric test). A potential indirect role for Brg1 in enhancer regulation through transcriptional control of histone modifying enzymes was discounted, as expression of histone acetyltransferases responsible for depositing H3K27ac at enhancers was not affected by loss of Brg1 (supplementary material Fig. S7). These data support a direct role for Brg1 in control of enhancer activity.

Enhancer usage is highly cell-type specific and dynamic during cell differentiation, and H3K27ac enrichment distinguishes active enhancers from other enhancer states (Calo and Wysocka, 2013). Given the requirement for Brg1 for H3K27ac levels at enhancers proximal to downregulated genes and that many of these genes are induced during mesoderm differentiation (Fig. 4A), we hypothesized that Brg1 might be required to activate quiescent enhancers during mesoderm differentiation. To test this, we used our published enhancer predictions in directed cardiac differentiations of ESCs to distinguish enhancers that are dynamically activated during mesoderm differentiation from those that remain active from undifferentiated cell states (Wamstad et al., 2012). We overlapped our 16,725 predicted enhancers with those identified at an analogous stage of ESC differentiation and divided this cohort into ‘activated’ or ‘static’ enhancers, based on whether these regions were uniquely marked by H3K27ac in mesodermal cultures or marked in both mesodermal cultures and ESCs, respectively. As expected, genes proximal (nearest gene) to activated enhancers are transcriptionally activated during mesoderm induction (data not shown). Whereas static enhancers showed no changes in H3K27ac on average (median log₂-fold change=0.012), activated enhancers had reduced H3K27ac in Brg1-deleted mesoderm (median log₂ fold change=−1.02) (Fig. 6B,E). Genes proximal to activated enhancers showed greater reductions in gene expression upon loss of Brg1 than those proximal to static enhancers (Fig. 6F). Moreover, we found that activated enhancers were significantly enriched for Brg1 occupancy compared with static enhancers (Chi-squared test, P=3.043e⁻¹¹). Thus, our data reveal that Brg1 activity is most important at regulatory regions that are transitioning in activation status.

Consistent with a role for Brg1 in activation of mesodermal enhancers, we found multiple Brg1-bound enhancers near the mesodermal genes Flk1 and Cyp26a1 that showed marked loss of H3K27ac in Brg1-deleted mesoderm (Fig. 6G; supplementary material Fig. S5). This included an experimentally validated regulatory region ∼30 kb upstream of the Flk1 TSS that directs early mesodermal expression in the mouse embryo (Ishitobi et al., 2011). Furthermore, we detected a clear reduction in H3K27ac within a Brg1-bound region roughly 5 kb upstream of the Mesp1 TSS, which functions as a regulatory enhancer for Mesp1 expression (Haraguchi et al., 2001) (Fig. 6G). These enhancers are not marked by H3K27ac in ESCs and, thus, are activated during mesoderm differentiation to coordinate the transcriptional activation of nearby genes. We propose that Brg1 regulates the transcriptional induction of mesodermal gene expression through binding to distal regulatory regions and facilitating the recruitment of these regions towards the activation of nearby genes.

Finally, Brg1 has been observed to associate with large enhancer collectives that have been dubbed ‘super’ or ‘stretch’ enhancers (Hnisz et al., 2013; Parker et al., 2013; Whyte et al., 2013). Based on correlation with transcriptional activity, these large stretches of H3K27ac have been proposed to be associated with highly cell type-specific gene regulation. We identified 4894 ‘super-enhancers’, of which 594 were bound by Brg1. Although these had significant reductions in H3K27ac occupancy in the absence of Brg1, the loss of H3K27ac was significantly less pronounced than smaller dynamic enhancers (supplementary material Fig. S6). Thus, Brg1 is important for activating initially silent enhancers and is less important at larger enhancers, perhaps due to redundant mechanisms of enhancer activation (Hnisz et al., 2013).

We next investigated the mechanism by which Brg1 regulates the repression of developmental regulators in Brg1-deleted mesoderm. Intersection of our RNA-seq analysis with published ChIP-seq datasets of H3K27me3 and Polycomb subunit occupancy in undifferentiated ESCs demonstrated that upregulated genes were highly enriched for Polycomb targets (Ku et al., 2008) (supplementary material Fig. S8). Given that Brg1 positively regulates PRC2 repression of Hox loci in undifferentiated ESCs (Ho et al., 2011), we hypothesized that Brg1 is broadly required for PRC2-mediated silencing in differentiating mesoderm.

To test this, we measured genome-wide occupancy of H3K27me3 in control and 4-OHT-treated mesodermal cultures by ChIP-seq in biological triplicate. We analyzed the enrichment of H3K27me3 at the promoters of genes upregulated by loss of Brg1 in normal mesodermal cultures and found that a subset of these genes (termed group I) were substantially enriched for H3K27me3 (Fig. 7A). This subset included nearly all derepressed developmental TFs. In agreement with the exclusivity of the two marks, group I genes were relatively low in H3K27ac, which instead marked a second subset of upregulated genes with few developmental regulators (group II).

We focused on group I genes and compared H3K27me3 genome-wide maps from control and 4-OHT-treated mesodermal cultures, to determine whether loss of Brg1 led to reduced levels of H3K27me3. Whereas most developmental TFs were still marked by H3K27me3 in Brg1-depleted cultures, we observed clear, reproducible reductions in H3K27me3 at group I genes (Fig. 7B,C; supplementary material Fig. S5C). Clear examples of this are Irx1 and Nkx2-5, two homeodomain TFs that are upregulated upon loss of Brg1 (Fig. 7B). We did not observe reduced expression of PRC2 subunits or H3K27 demethylases in our RNA-seq analysis, arguing against an indirect effect on H3K27me3 levels through Brg1 transcriptional regulation of these chromatin regulators (supplementary material Fig. S7). We next considered that reduced levels of H3K27me3 might result from abrogated recruitment of PRC2 or reduced activity of recruited complexes. To distinguish between these two possibilities, we measured genome-wide occupancy of Suz12, an essential subunit of PRC2 complexes (Surface et al., 2010), in control and 4-OHT-treated mesodermal cultures in biological duplicate. Suz12 demonstrated clear enrichment at group I genes in both normal and Brg1-deleted cultures. Composite analysis of all group I genes revealed a modest, albeit statistically significant, reduction in Suz12 occupancy (Fig. 7D); however, this reduction was small in comparison to the reduction in H3K27me3. Thus, Brg1 probably modulates PRC2 repression independently of PRC2 recruitment.

Our findings support a role for Brg1 in balancing lineage-specific gene expression (summarized in Fig. 8). In particular, Brg1 is essential for transcriptional activation of essential mesodermal genes during mesoderm induction. Our genome-wide occupancy data support a primary role for Brg1 at distal enhancers rather than at promoters. The absence of a strong correlation between Brg1 promoter occupancy and gene regulation might reflect the greater stability of chromatin states at promoter regions seen across cell-types and during differentiation (Ernst et al., 2011; Wamstad et al., 2012).

Brg1-bound loci distal to TSSs largely overlapped putative enhancer regions marked by H3K27ac, consistent with findings in other cell types (Euskirchen et al., 2011; Hu et al., 2011; Rada-Iglesias et al., 2011; Yu et al., 2013). Our data show that H3K27 acetylation depends on Brg1 at a number of loci. Of particular interest, our findings demonstrate that differentiating ESCs are most sensitive to Brg1 function at dynamic enhancer regions, pointing to an essential role for Brg1 in the transition of developmental enhancers from inactive to active. This might reflect the importance of chromatin remodeling in the conversion of inaccessible chromatin to open chromatin by facilitating TF binding and histone acetyltransferase recruitment. A similar function for BAF complexes has been proposed downstream of Cer1-mediated activation of Nkx2-5 (Cai et al., 2013). BAF and the related yeast SWI-SNF complexes mediate TF recruitment (Hu et al., 2011; Kwon et al., 1994; Takeuchi and Bruneau, 2009), but the functional interplay between SWI-SNF family complexes and histone acetyltransferases is less clear (Agalioti et al., 2000; Narlikar et al., 2002). Our data suggest that Brg1 enhances the function of histone acetyltransferases at transitioning enhancers, but the mechanism for this interaction is not clear. Once enhancer regions acquire characteristics of open chromatin, such as H3K27ac, they appear less dependent on Brg1 in maintaining these characteristics. Thus, our findings predict that, whereas Brg1 might be recruited broadly to enhancer regions in many cell types, Brg1-dependent gene expression is likely to reflect those regions undergoing dynamic chromatin remodeling.

Brg1 is also required for repression of a diverse group of developmental regulators during mesoderm differentiation. BAF complexes have classically been annotated as Trithorax group (TrxG) complexes, which counteract Polycomb-mediated repression, based on studies in Drosophila (Tamkun et al., 1992). Indeed, brm knockdown leads to increased H3K27me3 in addition to reduced H3K27ac in flies (Tie et al., 2012). In mouse ESCs, loss of Brg1 is linked to reduced H3K27me3 levels at Hox clusters, classic Polycomb targets (Ho et al., 2011). Our study demonstrates that loss of Brg1 disrupts Polycomb repression of Hox clusters, as well as a broad range of other developmental regulators, during ESC differentiation. Thus, cooperativity between PRC2 and BAF complexes is not unique to the pluripotent state and is probably a crucial function for BAF complexes in lineage commitment. The nature of PRC2/BAF cooperativity is unclear. Our ChIP-seq analysis of Brg1 occupancy revealed few clear peaks within H3K27me3-marked domains, whereas Brg1 was found to co-occupy more clearly PRC2-regulated loci in developing organs, including heart (Attanasio et al., 2014). Therefore, in mesoderm Brg1 might associate with Polycomb-repressed genes in rare, transient interactions that are below our threshold for detection. Our ChIP-seq analysis revealed little change in Suz12 occupancy at derepressed genes, suggesting that Brg1 is dispensable for Suz12 recruitment and might regulate PRC2 activity at bound loci. Nucleosome density affects PRC2 activity in vitro (Yuan et al., 2012). Thus, chromatin remodeling by BAF complexes could increase PRC2 efficiency by augmenting nucleosome fluidity. This model requires further exploration.

Mouse ESCs were cultured in feeder-free conditions and serum containing media with leukemia inhibitory factor. Directed differentiations were performed as described previously (Wamstad et al., 2012). For Brg1 deletion, Brg1^fl/fl;Actin-CreER ESCs (Ho et al., 2009; Ho et al., 2011), cultures were treated with 200 nM 4-hydroxytamoxifen (4-OHT) diluted from a 5 mg/ml stock solution in tetrahydrofuran (THF) or with only THF for control. Additional details are provided in the supplementary material Methods.

RNA was extracted using TRIzol and reverse transcribed using a High-Capacity cDNA Reverse Transcription kit (Applied Biosystems). Quantitative PCR was performed in technical triplicate using Taqman probes and expression was normalized to Gapdh. The following probes were used: Mesp1 – Mm00801883_g1, Gapdh – 4352932E.

For cell surface staining, cells were trypsinized, quenched with serum and washed in FACS buffer. Cells were stained with biotinylated anti-Flk-1 (Hybridoma Clone D218; 1:10,000) antibody, washed and stained with PE-conjugated anti-Pdgfra (eBioscience, 12-1401-81; 1:400) and APC-Streptavidin (1:200). Cells were analyzed on an LSRII flow cytometer (BD). For intracellular staining, cultures were trypsinized, fixed and stained with anti-cTnT (Thermo Scientific #MS295, Clone 13-11; 1:100) antibody, followed by secondary antibody. All steps were performed in D-PBS with 0.5% saponin and 4% FBS. Western blotting was performed using standard techniques. Briefly, protein lysate was sonicated and cleared by centrifugation. Supernatant was diluted and boiled. Following electrophoresis, protein was transferred to a PVDF membrane. Membranes were incubated with desired antibody in 5% milk TBST overnight at 4°C, then washed in TBST and stained with secondary antibody. After antibody staining, membranes were washed, incubated in SuperSignal chemiluminescence substrate (Thermo Scientific) and visualized. Antibodies used were anti-Brg1 (Santa Cruz, sc-10768; 1:2000), anti-actin (Sigma, A1978; 1:2000) and anti-FLAG (Sigma, M2; 1:2000). For immunofluorescence, cultures were fixed, and, after blocking, were incubated with primary antibody at 4°C overnight. Slides were washed and incubated in secondary antibody. Slides were stained with Hoechst 33342 (10 μg/ml) in D-PBS, and immediately imaged in 50 μl D-PBS. The anti-cTnT (Thermo Scientific #MS295, Clone 13-11; 1:100) antibody was used.

Total RNA was isolated from 1.5-2×10⁶ cells using TRIzol reagent in biological duplicates for each experimental condition. 8 μg of total RNA was used as input material for the preparation of the RNA-Seq libraries, according to Illumina RNA Seq library kit with minor modifications. Briefly, mRNA was isolated using Dynabeads mRNA Purification Kit (Invitrogen), followed by fragmentation (Ambion) and ethanol precipitation. First- and second-strand synthesis were performed followed by end repair, A-tailing, adapter ligation and size selection on a Beckman Coulter SPRI TE nucleic acid extractor. 200-400 bp dsDNA was enriched by 13 cycles of PCR with Phusion High-Fidelity DNA Polymerase (NEB). Amplified libraries were sequenced on an Illumina HiSeq 2000.

Single-end 40-bp reads were aligned to the mouse genome (mm9) using Bowtie (Langmead et al., 2009). Differential gene expression between conditions was determined using the USeq package (Nix et al., 2008) considering all Refseq genes. Genes with an FDR of ≤1% and twofold expression change were considered significantly differentially expressed unless otherwise noted. USeq was also used to calculate reads per kilobase exon per million reads (RPKM) and fold change values between conditions. Mapped reads were filtered to allow a maximum of 50 identical reads, and genes expressed <0.5 RPKM in all conditions were excluded from subsequent analysis. GO analysis was performed using Go Elite (http://www.genmapp.org/go_elite/), with all genes having an RPKM >0.5 in at least one condition serving as the gene universe. Graphical representation of upregulated and downregulated genes was performed in R.

Chromatin immunoprecipitation of histone modifications were performed according to Lee et al. (2006) with minor modifications, in biological duplicate for H3K27ac, Suz12 and FLAG. H3K27me3 ChIP-seq was performed in biological triplicate. Additional details are provided in the supplementary material Methods. Antibodies used were anti-FLAG (Sigma, M2 F1804; 10 µg), anti-H3K27ac (ActiveMotif, #39134; 5 µg), anti-H3K27me3 (Millipore, 17-622; 5 µg) and anti-Suz12 (Bethyl, A302-407A; 5 µg). ChIP-seq analysis pipeline and statistical methods are provided in the supplementary material Methods.

Brg1 ChIP-exo was performed as previously described (Serandour et al., 2013) using anti-Brg1 antibody (Abcam, 110641; 3 µg). Briefly, Brg1 ChIP was performed, and, while still on magnetic beads, the immunoprecipitated DNA was polished, ligated with P7 adapter and nicks were repaired. The resulting DNA was digested with Exo I and RecJ_f exonucleases (NEB). Exonuclease-digested DNA was eluted from the beads, cross-links were reversed and the bound protein was digested with proteinase K at 65°C overnight. DNA was purified using Agencourt Ampure XP beads (Beckman Coulter), denatured, and the single-stranded DNA was used to synthesize the second strand using P7 primer, followed by ligation of P5 adapter. The resulting DNA fragment was PCR-amplified, gel-purified and sequenced using an Illumina HiSeq 2500 sequencer at a minimum depth of 25 million mapped reads, with most exceeding 30 million.

All sequencing data have been deposited in GEO (accession number GSE45448).

We thank T. Sukonnik for immunofluorescence, J. Wylie for western blots and G. Crabtree for use of the Brg1^fl/fl;Actin-CreER ESCs. We thank P. Devine, A. Holloway and L. Boyer for input on the manuscript and G. Howard for editorial assistance.