BET bromodomain inhibition suppresses the functional output of hematopoietic transcription factors in acute myeloid leukemia

Summary

The bromodomain and extraterminal (BET) protein BRD4 is a validated drug target in leukemia, yet its regulatory function in this disease is not well understood. Here, we show that BRD4 chromatin occupancy in acute myeloid leukemia closely correlates with the hematopoietic transcription factors (TFs) PU.1, FLI1, ERG, C/EBPα, C/EBPβ, and MYB at nucleosome-depleted enhancer and promoter regions. We provide evidence that these TFs, in conjunction with the lysine acetyltransferase activity of p300/CBP, facilitate BRD4 recruitment to their occupied sites to promote transcriptional activation. Chemical inhibition of BET bromodomains was found to suppress the functional output each hematopoietic TF, thereby interfering with essential lineage-specific transcriptional circuits in this disease. These findings reveal a chromatin-based signaling cascade comprised of hematopoietic TFs, p300/CBP, and BRD4 that supports leukemia maintenance and is suppressed by BET bromodomain inhibition.

Introduction

Deregulation of chromatin structure and function is a major mechanism that drives the pathogenesis of human cancer, which often occurs through somatic mutation of genes encoding chromatin-modifying enzymes or factors that interact with modified histones (Dawson and Kouzarides, 2012). Chromatin state can also become altered in cancer as a secondary effect of hyperactive signal transduction or as a result of metabolic changes that occur during tumorigenesis (Lu and Thompson, 2012; Wajapeyee et al., 2013). One of the major consequences of these epigenetic changes is that cancer cells become reliant on certain chromatin regulators to maintain a malignant phenotype (Mair et al., 2014). This observation has led to a widespread interest in targeting chromatin as a therapeutic approach in cancer, with several chromatin-modulating drugs under active clinical investigation (Dawson and Kouzarides, 2012; Mair et al., 2014).

Bromodomain-containing protein 4 (BRD4) is a prominent example of a chromatin regulator that has been widely validated as a therapeutic target in cancer (Shi and Vakoc, 2014). BRD4 belongs to the bromodomain and extraterminal (BET) family of chromatin reader proteins, which includes BRD2, BRD3, and BRDT. BET proteins feature two conserved N-terminal bromodomains that interact with acetylated lysine residues on histones and other nuclear proteins, resulting in the localization BET proteins to hyperacetylated chromatin locations (Dey et al., 2003; Dhalluin et al., 1999). One key attribute of BET bromodomains is their preferential interaction with peptides that are acetylated on multiple lysine residues within a span of 2–5 amino acids, which occurs on histone tails and on certain transcription factors (TFs), such as GATA-1 and TWIST (Filippakopoulos et al., 2012; Lamonica et al., 2011; Moriniere et al., 2009; Shi et al., 2014). BET proteins can also bind to TFs in a bromodomain-independent manner to function as transcriptional cofactors (Wu et al., 2013). One of the important effectors recruited by BRD4 is P-TEFb, which promotes elongation of RNA polymerase II through its kinase activity (Jang et al., 2005; Yang et al., 2005). Notably, BET bromodomains can be selectively targeted with small-molecule inhibitors, such as JQ1 and I-BET, which compete with acetyl-lysine recognition to displace BET proteins from chromatin (Filippakopoulos et al., 2010; Nicodeme et al., 2010).

While BRD4 has been characterized as a general transcriptional regulator, BET bromodomain inhibition leads to potent therapeutic effects in several cancer models. Acute myeloid leukemia (AML) was one of the first neoplasms in which BRD4 was shown to perform a cancer maintenance function, which was identified through a negative selection RNAi screen performed in a mouse model of MLL-rearranged leukemia (Zuber et al., 2011c). Furthermore, several studies have shown that small-molecule BET inhibitors interfere with BRD4 function in AML and extend the survival of leukemia-bearing mice (Dawson et al., 2011; Mertz et al., 2011; Zuber et al., 2011c). While hematopoietic malignancies are highly heterogeneous, it is noteworthy that multiple subtypes of myeloid and lymphoid malignancy are sensitive to BET inhibition at doses that exhibit little toxicity to normal hematopoiesis (Delmore et al., 2011; Ott et al., 2012; Zuber et al., 2011c). One of the cellular effects of BET inhibition in AML is the induction of myeloid differentiation and the suppression of self-renewal (Dawson et al., 2011; Zuber et al., 2011c). Based on the promising results of pre-clinical studies of BET inhibitors in animal models, clinical trials investigating the efficacy of BET inhibition were recently initiated in human AML and in several other malignancies (Clinicaltrials.gov Identifiers: NCT01713582, NCT02158858, and NCT02308761).

Since BRD4 is not deregulated via mutation or overexpression in AML, it remains unclear at present why leukemia cells are hypersensitive to BET inhibition relative to normal hematopoietic cells. It has been found that transcription of certain proto-oncogenes, including MYC, BCL2, and CDK6, is rapidly suppressed by BET inhibition in AML (Dawson et al., 2011; Mertz et al., 2011; Zuber et al., 2011c). A subset of BRD4-dependent genes exhibit high levels of BRD4 occupancy at distal enhancer elements, which can exist in clusters termed super-enhancers (Loven et al., 2013). As an example, MYC is among the most sensitive transcripts to BET inhibition in MLL-rearranged AML and harbors one the largest clusters of BRD4-occupied enhancers in this cell type, which is found 1.8 megabases downstream of the MYC promoter (Shi et al., 2013). Super-enhancers have also been identified near other BET-dependent genes in AML (Dawson et al., 2014; Groschel et al., 2014). However, the targeting mechanisms that localize BRD4 to promoter and enhancer regions in AML are unknown. In particular, it is unclear which lysine acetyltransferases are responsible for supporting the cancer maintenance functions of BRD4 and whether interactions with transcription factors and/or histones are relevant for BRD4 recruitment.

Here, we show that hematopoietic TFs perform a critical role in directing BRD4 recruitment to enhancer and promoter regions in AML. These effects are mediated by the p300/CBP acetyltransferases, which are recruited by TFs to their occupied sites to maintain lysine acetylation that is critical for BRD4 occupancy. We provide evidence that interactions with both histones and TFs are contributory to the recruitment mechanism of BRD4 in this disease. In addition, we identify a histone-like motif on the transcription factor ERG, which is acetylated by p300 to mediate a direct interaction with BRD4. Finally, we show that chemical inhibition of BET proteins suppresses the functional output of hematopoietic TFs in leukemia cells. Since hematopoietic TFs and p300/CBP are essential to maintain the leukemia cell state, our findings suggest that BET inhibitors exert therapeutic effects in this disease by perturbing lineage-specific transcriptional programs.

Results

BRD4 occupancy in acute myeloid leukemia correlates with hematopoietic transcription factors flanked by acetylated nucleosomes

To gain mechanistic insight into the leukemia maintenance function of BRD4, we performed ChIP-seq analysis in RN2 cells, which is a line derived from a mouse model MLL-AF9/Nras^G12D acute myeloid leukemia (Zuber et al., 2011a). Importantly, this model is dependent on BRD4 for disease maintenance and is highly sensitive to the BET inhibitor JQ1 (Zuber et al., 2011c). Consistent with prior observations in other cell types (Loven et al., 2013; Zhang et al., 2012), BRD4 colocalized with histone H3K27 and H4K8 acetylation at the majority of active promoters and enhancers in RN2 (Figure 1A–B). We also identified super-enhancer regions exhibiting higher levels of BRD4 enrichment, with nearby genes being associated with sensitivity to BET inhibition (Figure S1A–B). Unexpectedly, we noticed that BRD4 and histone acetylation existed in overlapping, yet largely distinct patterns (Figure 1A–B). Histone acetylation was found preferentially on nucleosomes that flank a nucleosome-depleted site, in a characteristic ‘valley’ of acetylation at enhancers and promoters (Ramsey et al., 2010). In contrast, BRD4 was enriched as a single peak, with its summit coinciding with the local minimum of both histone acetylation and overall histone density (Figure 1A–B). Instead, the breadth of BRD4 peaks tended to correlate with levels of flanking histone acetylation (Figure 1B). This pattern of BRD4 occupancy and histone acetylation was also observed in datasets obtained from other cell types (Figure S1C). Since the nucleosome-free region at promoters and enhancer regions is known to be the preferential site of TF binding to DNA (Ramsey et al., 2010), we considered the role of TFs in supporting BRD4 function in AML.

Figure 1. BRD4 occupancy in acute myeloid leukemia correlates with hematopoietic transcription factors flanked by histone acetylation.

(A) ChIP-Seq meta-profiles for BRD4, H3K27ac, H4K8ac and H3 representing the average read counts per 20bp bin for 5,135 BRD4 occupied regions. Data were normalized to max signal. (B) Density plot of different ChIP-Seq datasets centered on 1,950 BRD4-occupied promoters and 3,185 BRD4-occupied enhancers. Each row represents a single peak. (C) MEME suite motif analysis performed on BRD4-occupied sites. A 400 bp region centered on BRD4 peaks was used for motif discovery. The distribution of each motif relative to the BRD4 peak summit is indicated. (D) ChIP-Seq occupancy profiles for BRD4, p300, H3K27ac, H4K8ac and hematopoietic transcription factors at the Myc locus. (E) ChIP-Seq occupancy profiles at Myc enhancer E5. See also Figure S1.

To identify TFs that colocalize with BRD4, we analyzed the enrichment of DNA motifs that were present at BRD4-occupied enhancers and promoters using the MEME suite of motif-finding algorithms (Bailey et al., 2009). This revealed a strong association of BRD4 with the CAAT box motif recognized by TFs of the C/EBP family and various motifs recognized by ETS family TFs (Figure 1C). By examining the expression of C/EBP and ETS family members in RN2, we identified C/EBPα, C/EBPβ, FLI1, PU.1 and ERG as candidates, which are all TFs known to play a role in hematopoietic stem cell self-renewal and/or myeloid cell differentiation (Kruse et al., 2009; Rosenbauer and Tenen, 2007) (Figure 1C, S1D). Although less enriched in the MEME analysis, we also identified a motif recognized by the hematopoietic TF MYB as being associated with BRD4-occupied enhancers (Figure 1C). These results prompted us to perform ChIP-seq analysis of C/EBPα, C/EBPβ, FLI1, PU.1, ERG, and MYB in RN2 to evaluate their degree of overlap with BRD4. We found that essentially all BRD4-occupied sites coincided with occupancy of one or more hematopoietic TFs, with more than half of all BRD4-enriched regions exhibiting combinations of 4 or more TFs (Figure 1B, Figure S1E). As highlighted in the examples of Myc and Cdk6, the locations of TF occupancy coincided with the summit of BRD4 enrichment, with a lower level of BRD4 enrichment at flanking regions harboring histone acetylation (Figure 1D–E, and Figure S1F). Collectively, these findings suggest that BRD4 preferentially associates with chromatin regions harboring hematopoietic TFs flanked by acetylated nucleosomes.

Hematopoietic transcription factors promote BRD4 recruitment to their occupied sites

The findings above suggest that a combination of hematopoietic TFs might be responsible for recruiting BRD4, either directly or indirectly, to regulatory regions. To evaluate this hypothesis, we ectopically introduced a combination of hematopoietic TFs (PU.1, FLI1, ERG, C/EBPβ and MYB) via retrovirus into NIH3T3 fibroblasts and evaluated whether this was sufficient to alter BRD4 occupancy. Since C/EBPα and β are capable of functionally substituting for one another in certain settings, we only used C/EBPβ for these experiments (Jones et al., 2002). At day 3 post-transduction, all TFs were highly expressed but had not yet caused any phenotypic effects in the fibroblasts, suggesting this was a suitable timepoint for studying the primary effect of TFs on chromatin (Figure 2A). By RT-qPCR analysis, we identified a set of hematopoietic genes (Btk, Ccl4, Fcgr2b, and Pecam1) that were upregulated by the TFs at this timepoint (Figure 2B). Transcription of these four genes was suppressed by JQ1 in leukemia, which was also recapitulated in the TF-transduced fibroblasts (Figure 2B, S2A). This indicates a BET protein requirement for TF-dependent transcriptional activation. We also found that these TFs occupied enhancer and promoter regions of Btk, Ccl4, Fcgr2b, and Pecam1 when introduced ectopically in fibroblasts, in a similar pattern as observed in RN2 (Figure 2C–H, S2B–E). Importantly, we found that co-expression of these TFs was sufficient to promote BRD4 occupancy at their occupied sites, but not at a control region that lacks TF occupancy (Figure 2I). These effects occurred without a change in BRD4 expression (Figure S2F). TF-dependent BRD4 occupancy in fibroblasts was also eliminated by JQ1, indicating that this effect was bromodomain-dependent (Figure S2G). Although Myc was already highly expressed in fibroblasts and was not altered by introducing these TFs, we nonetheless observed TF-dependent BRD4 recruitment at the Myc E2 and E3 enhancers (Figure 2C–I and S2H). Taken together, these experiments demonstrate that hematopoietic TFs can stimulate BRD4 recruitment to their occupied sites.

Figure 2. Hematopoietic transcription factors can facilitate BRD4 recruitment to their occupied sites.

(A) Western blotting and light microscopy of TF-transduced NIH3T3 fibroblasts 72 hours following retroviral transduction. Cells were transduced at ~100% efficiency for each of the individual TF-expressing retroviral vectors. (B) RT-qPCR analysis of RNA prepared from TF-transduced fibroblasts. A 6-hour exposure to DMSO or 500nM of JQ1 was performed following 72 hours of TF retroviral transduction. Each mRNA level was normalized to Gapdh. Data are represented as mean ±SEM, and n=3. (C–I) ChIP-qPCR with indicated antibodies after 72 hours of TF transduction in fibroblasts. Neg refers to a negative control region in a gene desert region. ChIP-qPCR primers for Btk, Ccl4, Fcgr2b, and Pecam1 were designed based on BRD4 ChIP-Seq profile from RN2 cells. Data are represented as mean ±SEM, and n=3. See also Figure S2.

Hematopoietic TFs recruit the lysine acetyltransferase activity of p300/CBP to support BRD4 occupancy

Since the effects described above were bromodomain-dependent, we considered whether TF-dependent BRD4 recruitment required the activity of a specific lysine acetyltransferase (KAT). We evaluated this hypothesis using a genetic strategy, reasoning that a specific KAT that functions upstream of BRD4 should be essential in leukemia cells and regulate a similar gene expression program as BRD4. Using 3–5 shRNAs per gene, we systematically targeted 17 mammalian KATs, and identified ELP3, p300, TAF1, MYST2, NCOA1, and CBP as essential for RN2 cell proliferation in culture, based on multiple independent shRNAs exhibiting negative selection (Figure 3A). We next performed an RNA-seq analysis in RN2 that compared the global mRNA changes induced upon KAT knockdown with those observed upon knockdown of BRD4. Using an unsupervised clustering analysis of gene expression changes, we identified p300 and CBP as KATs that performed the most similar function to BRD4 in controlling gene expression (Figure 3B). Interestingly, the other KAT enzymes appeared to perform a regulatory function in opposition to BRD4 (Figure 3B). p300 and CBP are homologous KAT enzymes which are known to function as general TF coactivators (Goodman and Smolik, 2000). We focused here on p300, as knocking down this KAT led to a stronger anti-proliferative phenotype in leukemia cells than targeting CBP (Figure 3A), however prior studies would suggest these two KATs are likely to perform overlapping functions in this system (Blobel, 2000).

Figure 3. The lysine acetyltransferase p300 is recruited by hematopoietic TFs to support BRD4 occupancy in leukemia.

(A) Summary of competition-based negative selection shRNA experiments targeting the indicated KAT enzymes performed in RN2 cells. The average fold-decrease in GFP percentage over 10 days for three to six independent shRNAs is plotted, which represents the relative degree of growth inhibition conferred by the indicated shRNA. Red bars indicate KATs having a greater than two-fold average decrease in the percentage of GFP-positive cells. Data are represented as mean ±SEM. (B) Heat map of unsupervised hierarchical clustering of RNA-Seq data performed using the GENE-E software. Two independent shRNAs against each candidate KATs, BRD4, or Renilla were induced with doxycycline for 48 hours using the TRMPV-Neo vector in RN2 cells. Each row represents the row normalized expression value of an individual gene. (C) p300 ChIP-qPCR following 72 hours of TF transduction in fibroblasts as described in Figure 2. (D and E) BRD4 ChIP-Seq meta-profiles for 1,950 BRD4-occupied promoters and 3,185 BRD4-occupied enhancers following 10 uM C646 treatment for 2 hours. (F–H) ChIP-Seq occupancy profiles of BRD4 at Myc, Cdk6 and Pecam1 locus, in the presence or absence of C646. (I) RT-qPCR following 6 hours of JQ1 or C646 exposure in RN2 cells. (J) Western blotting of RN2 lysates following JQ1 or C646 exposure. (K) RT-qPCR following 6 hours of JQ1 or 10 uM C646 exposure in MM1.S cells. (L) Gene Set Enrichment Analysis (GSEA) evaluating a JQ1-sensitive gene signature in the RNA-seq analysis of 6 hour C646 exposure in RN2. (NES) Normalized enrichment score; (FDR), false discovery rate. For C, I, and K, data are represented as mean ±SEM, and n=3. See also Figure S3.

Using ChIP-seq analysis, we found that p300 occupancy globally coincided with BRD4 at promoter and enhancer regions in leukemia, consistent with prior observations (Figure 1B) (Zhang et al., 2012). Similar to BRD4 and hematopoietic TFs, p300 also preferentially occupied the nucleosome-depleted site at promoter and enhancer regions (Figure 1B). Interestingly, C/EBPα, C/EBPβ, PU.1, and MYB have all been shown previously to interact with p300 or CBP (Dai et al., 1996; Mink et al., 1997; Yamamoto et al., 1999). Consistent with these known interactions, ectopic expression of hematopoietic TFs in fibroblasts stimulated p300 occupancy at the same sites in which BRD4 was recruited, without influencing p300 expression levels (Figure 3C and S2F). We next evaluated whether the KAT activity of p300/CBP was necessary for BRD4 occupancy in leukemia. For these experiments, we utilized C646, a selective inhibitor of the p300/CBP KAT activity (Bowers et al., 2010). Remarkably, exposing RN2 cells to C646 for 2 hours led to a genomewide reduction of BRD4 occupancy (Figure 3D–E). This effect occurred with about equal potency at enhancer and promoter regions, with slightly greater effects at super-enhancers (Figure 3D–E and S3A). As examples, we observed substantial reductions in BRD4 occupancy at the Myc, Cdk6, and Pecam1 loci upon C646 exposure (Figure 3F–H). ChIP-qPCR experiments validated that BRD4 was evicted from the Myc E1–E5 enhancers in the range of 5- to 20-fold upon C646 treatment (Figure S3B). p300 knockdown via shRNA also led to reduced BRD4 occupancy at Myc E1–E5 enhancers in RN2 (Figure S3C). In addition, treating the human MM1.S multiple myeloma cell line with C646 also triggered loss of BRD4 from multiple enhancer regions, indicating that these effects were not unique to RN2 (Figure S3D). In contrast, knockdown or chemical inhibition of p300/CBP failed to influence TF occupancy (Figure S3E–G). These findings together suggest that p300 is recruited by hematopoietic TFs to promote acetylation-dependent BRD4 occupancy.

One implication of the findings described above is that C646 should lead to similar transcriptional changes in RN2 as BET inhibition with JQ1. In agreement with this hypothesis, we found that p300/CBP KAT inhibition with C646 led to pronounced suppression of Myc mRNA and protein levels, to an extent that is similar to the effects of JQ1 (Figure 3I–J). In addition, C646 and JQ1 suppressed MYC expression in MM1.S cells (Figure 3K). Using gene set enrichment analysis (GSEA) of RNA-seq data (Subramanian et al., 2005), we found that C646 exposure and p300 knockdown globally suppressed gene signatures shown previously to be sensitive to BET inhibition (Figure 3L and S3H). Collectively, these data support TF-dependent recruitment of p300/CBP as a major determinant of BRD4 localization and function in leukemia.

p300/CBP maintains local histone acetylation near BRD4-occupied sites

We next investigated the acetylated substrates of p300/CBP that contribute to BRD4 recruitment in leukemia cells. p300 and CBP are known to acetylate several lysine residues on core histones, which includes H3K27 and H4K8 (McManus and Hendzel, 2003; Szerlong et al., 2010). Since BRD4 occupancy tends to encompass acetylated nucleosomes that flank TF binding sites, we considered whether p300/CBP maintains histone acetylation near BRD4-occupied enhancers and promoters. For this purpose, we performed ChIP-Seq analysis of H3K27ac and H4K8ac following a 2 hour exposure of RN2 cells to C646. These experiments revealed a reduction of both acetylation marks at BRD4-occupied promoters and enhancers genomewide, without an effect on global acetylation levels (Figure 4A–G and S4). In addition, the reduction in H3K27/H4K8 acetylation correlated linearly with the reduction in BRD4 occupancy across the genome (Figure 4H–I). Consistent with histone acetylation being driven by TF-dependent p300/CBP recruitment, we found that ectopic expression of hematopoietic TFs in fibroblasts led to induced H3K27 and H4K8 acetylation at their occupied sites, thus paralleling the recruitment of p300 and BRD4 (Figure 4J–K). Collectively, these findings are consistent with p300/CBP-dependent histone acetylation contributing to BRD4 recruitment to nucleosomes that flank TF binding sites.

Figure 4. p300/CBP maintains local histone acetylation near BRD4-occupied sites.

(A–D) H3K27ac and H4K8ac ChIP-Seq meta-profiles of BRD4-occupied promoters or enhancers following 2 hours of C646 exposure. (E–G) ChIP-Seq occupancy profile of H3K27ac and H4K8ac at Myc, Cdk6 and Pecam1 loci following C646 exposure for 2 hours. (H and I) Comparison of fold change for H3K27ac, H4K8ac, and BRD4 tag counts at BRD4-enriched regions following C646 treatment. A R² value was calculated using linear regression analysis. (J and K) ChIP-qPCR analysis of H3K27ac and H4K8ac following 72 hours of TF transduction in fibroblasts as described in Figure 2. See also Figure S4.

Physical interactions between hematopoietic transcription factors and BRD4

While histone acetylation is likely to contribute to the breadth of BRD4-occupied regions, it is noteworthy that C646 ablated the summit of BRD4 peaks, which preferentially occurs over nucleosome-depleted regions (Figure 3D–E). Since p300 and CBP are known to acetylate TFs (Gu and Roeder, 1997), we considered whether hematopoietic TFs might be capable of interacting with BRD4 in an acetylation-dependent manner. For this purpose, we overexpressed FLAG-tagged TFs in HEK293T cells and evaluated their association with endogenous BRD4 using anti-FLAG immunoprecipitation. These experiments revealed that each TF was capable of associating with BRD4, albeit with varying efficiency (Figure 5A). The addition of competitive JQ1 in the binding reaction failed to dislodge the C/EBPβ-BRD4 association, which is consistent with a prior report showing that C/EBPβ and C/EBPα bind directly to BRD4 in an acetylation-independent manner (Wu et al., 2013) (Figure 5B). In contrast, we found that the association of ERG with BRD4 was reduced by the addition of JQ1, suggesting a contribution of lysine acetylation to this interaction (Figure 5C).

Figure 5. BRD4 interactions with hematopoietic transcription factors.

(A) FLAG-TF immunoprecipitation performed in nuclear extracts prepared from HEK293T cells, transfected with the indicated pcDNA3 expression plasmids. (B and C) Immunoprecipitation experiments, as in (A), with 10 uM JQ1 or vehicle included the lysate prior to IP. (D) Schematic depiction of the functional domains of ERG and fragments utilized for KAT assays. (E) In vitro KAT assay with purified p300 KAT domain (1135–1810) and indicated ERG protein fragments. Purified p53 C-terminus (309–393) was used as a positive control. Acetylated products were detected by pan-acetyl lysine in Western blots. (F) Mapping of ERG (1–208) acetylation sites by mass spectrometry using reaction conditions as shown in (E). Relative abundance of acetylation was calculated by acetylation signal intensity of Acetyl-CoA (+) / Acetyla-CoA (−) from the same peptide. (G) Sequence alignment of histone H4, ERG, TWIST, and GATA-1. (H) Peptide-pulldown assay was carried out by mixing FLAG-BRD4 (1–722) purified from E. coli with indicated biotinylated peptides using streptavidin beads. The bound FLAG-BRD4 was analyzed by western blotting. (I) FLAG-ERG (wild-type or K96R/K99R mutant) IP-Western blotting as shown in (A). (J) RT-qPCR analysis in NIH3T3 fibroblasts following transduction with the indicated retroviral constructs. Data are represented as mean ±SEM, and n=3. See also Figure S5.

These results prompted us to examine whether ERG is a substrate of for p300. In vitro reactions performed with the p300 KAT domain revealed robust acetylation of an N-terminal 208 amino acid fragment of ERG, a region lacking the DNA binding domain but containing a regulatory SAM/PNT domain (Figure 5D–E). Using iTRAQ-coupled mass spectrometry, we mapped these ERG acetylation sites to K74, K96, and K99 (Figure 5F). It has been noted previously that BET bromodomains preferentially interact with di-acetylated peptides, in particular when separated by amino acids with small side chains, such as glycine (Filippakopoulos et al., 2012). Such a sequence feature is present in the histone H4 tail and in the TFs GATA-1 and TWIST, which all engage in acetylation-dependent interactions with BET proteins (Lamonica et al., 2011; Shi et al., 2014) (Figure 5G). Interestingly, lysines 96 and 99 of ERG are separated by two glycine residues, a motif resembling the histone H4K5/K8 di-acetylation site (Figure 5G). Based on this, we evaluated whether ERG K96/K99 di-acetylation contributes to an interaction with BRD4. We synthesized biotinylated ERG peptides harboring K96/K99 in either an unmodified or di-acetylated state and evaluated direct binding to recombinant BRD4 in pulldown assays. This revealed a robust interaction of di-acetylated ERG, but not the unacetylated form, with BRD4, which was only slightly weaker than the positive control di-acetyl H4-BRD4 interaction (Figure 5H). To examine the significance of this interaction in a cellular context, we transfected HEK293T cells with wild-type or K96R/K99R mutant forms of ERG and evaluated the BRD4 interaction by IP, which revealed a diminished interaction between BRD4 and ERG carrying the lysine-to-arginine substitutions (Figure 5I). Finally, we examined the functionality of the K96R/K99R ERG mutant by assaying transcriptional activation upon its ectopic expression in fibroblasts. When expressing ERG alone we observed a modest induction of Pecam1 and no effect on other hematopoietic genes, implying that the full complement of hematopoietic TFs is required for full activation in this heterologous system (Figure 5J, data not shown). Nonetheless, the K96R/K99R ERG mutations led to impaired Pecam1 activation as compared to wild-type ERG, despite equivalent levels of DNA occupancy for wild-type and mutant protein (Figure 5J and S5). While multiple TF and histone interactions are likely to contribute to BRD4 recruitment to chromatin, these findings suggest that p300-dependent acetylation of functionally important lysine residues on ERG facilitates its direct association with BRD4.

Hematopoietic transcription factors are essential in acute myeloid leukemia and are functionally suppressed by BET inhibitors

The findings above support a model in which hematopoietic TFs and p300/CBP provide gene-specificity to BRD4 function in leukemia cells. Two predictions from this model would be that hematopoietic TFs, like BRD4, should be essential to maintain the leukemia cell state and that BET inhibitors should suppress the functional output of each TF. To evaluate these possibilities, we validated two independent shRNAs that effectively suppressed the expression of each hematopoietic TF and evaluated the impact on RN2 proliferation (Figure 6A). We also used validated MYB shRNAs described previously (Zuber et al., 2011b). This analysis revealed that PU.1, FLI1, ERG, C/EBPβ, and MYB were all essential for proliferation of RN2 cells, similar to p300 and BRD4 (Figure 6B). Knockdown of C/EBPα failed to influence RN2 cell growth (data not shown), which is consistent with a role for this TF in the initiation but not the maintenance of MLL-fusion AML (Ohlsson et al., 2014). The PU.1 and MYB requirement in MLL-fusion leukemia is also consistent with prior studies (Aikawa et al., 2014; Jin et al., 2010; Zuber et al., 2011b). These results confirm that MLL-AF9/Nras^G12D AML cells are dependent on several hematopoietic TFs for their aberrant proliferation.

Figure 6. Hematopoietic transcription factors are essential in acute myeloid leukemia and are functionally suppressed by BET inhibitors.

(A) Western blotting in whole cell lysates prepared from RN2 cells following TRMPV-shRNA induction with dox for 48 hours. (B) Competition-based negative selection experiments that track the relative abundance of shRNA+/GFP+ cells over time. Percentages were normalized to day 2 values. Data are represented as mean ± SEM, and n=3. (C) TF gene signatures, defined by RNA-seq analysis following shRNA-based knockdown for 48 hours with two independent shRNAs. The 100 top down-regulated genes were identified by comparison to shRen.713 control. Each row represents row-normalized unit based on linkage-analysis mediated hierarchical clustering using Cluster 3.0 software. (D) GSEA of each TF signature following JQ1 treatment for 6 hours. RNA-seq was performed following 500nM of JQ1 exposure for 6 hours in RN2 cells. (E and F) Western blotting in RN2 lysates following indicated intervals of 500 nM JQ1 exposure. (G) RT-qPCR analysis of primary transcripts of three representative TF signature genes following JQ1 exposure. See also Figure S6.

We next examined whether BET inhibition with JQ1 suppressed the functional output of each hematopoietic TF. To this end, we knocked down each TF with two independent shRNAs in RN2, followed by RNA-seq analysis to define the top 100 down-regulated genes, hereafter referred to as a TF signature (Figure 6C). MYB and C/EBPα signatures were defined by analyzing previously published microarray data obtained in RN2 or in normal hematopoietic cells, respectively (Ohlsson et al., 2014; Zuber et al., 2011b). Upon defining each TF gene signature, we evaluated whether each gene set was suppressed following a 6 hour exposure of RN2 cells to JQ1. Using GSEA, we found that the signatures of each hematopoietic TF were significantly suppressed by BET inhibition (Figure 6D). In this RNA-seq analysis, however, we noticed that the genes encoding PU.1, FLI1, ERG, and MYB were also suppressed by JQ1 at the 6 hour timepoint, which we also confirmed by Western blotting and RT-qPCR (Figure 6E, S6A–B). While this effect was much weaker than the suppression of Myc, it nevertheless raised the possibility that suppression of TF signatures by JQ1 might be due to suppression of TF expression rather than TF function. To distinguish between these two possibilities, we analyzed representative TF signature genes by RT-qPCR at 30 and 60 minutes following JQ1 exposure, a timepoint that precedes reductions in TF protein levels (Figure 6F–G). In this analysis, we analyzed short half-life primary transcripts using PCR amplicons that span exon-intron junctions, thereby aiding the detection of immediate transcriptional effects (Figure 6G). For all of the TF signature genes tested, we found that JQ1 rapidly suppressed expression prior to reductions in TF protein levels (Figure 6G). These results, together with the findings above using JQ1 in the fibroblast system (Figure 2B), suggest that BET inhibition suppresses the activation function of hematopoietic TFs. Notably, hematopoietic TFs are known to positively auto-regulate their own expression and this may explain why hematopoietic TF expression becomes reduced upon longer durations of JQ1 exposure (Chen et al., 1995) (Figure S6C).

Discussion

The major objective of our study was to define a molecular mechanism that underlies the therapeutic effects of BET inhibition in acute myeloid leukemia. Using a combination of genetic, genomic, and biochemical approaches, we have found that an ensemble of hematopoietic TFs utilizes BRD4 and p300/CBP as linked cofactors to maintain the leukemia cell state. In this system, the lysine acetyltransferase activity p300/CBP is critical for BRD4 interactions with chromatin, which may involve acetylated TF and histone proteins as potential BRD4 bromodomain ligands. The major implication of these findings is that a primary effect of BET inhibition in AML is the blockade of TF-dependent transcriptional activation. Since all of these TFs are known regulators of hematopoietic differentiation, this mechanism may explain why BET inhibitors preferentially suppress self-renewal and promote myeloid differentiation in mouse models of MLL-fusion AML.

It has been previously proposed that BET inhibitors directly displace the MLL-fusion oncoprotein from chromatin, however this model is contradicted by a lack of effect for BET inhibition on the expression of primary MLL-fusion target genes at the HOXA cluster (Dawson et al., 2014; Dawson et al., 2011; Wang et al., 2013). Furthermore, our own ChIP experiments have failed to identify an effect of JQ1 on MLL-AF9 chromatin occupancy (unpublished observations). Instead, the findings presented here suggest that BET inhibitors would only indirectly block the oncogenic effects of MLL-AF9, through the suppression of TF-dependent transcriptional activation. MLL-AF9 is known to directly transactivate expression of Myb to promote aberrant self-renewal, with leukemia cells being more sensitive to Myb suppression than normal myeloid progenitors (Calabretta et al., 1991; Zuber et al., 2011b). Based on this, we speculate that leukemia cells exhibit hypersensitivity to BET inhibition because of an oncogene-driven reliance on hematopoietic TFs to sustain aberrant self-renewal programs that propagate disease. We also observed that BET inhibitors suppress the expression of hematopoietic TFs at later timepoints of drug exposure, which has also been observed in a prior study performed in diffuse large B cell lymphoma (Chapuy et al., 2013). Decreases in TF expression are also likely to contribute to the therapeutic effects of BET inhibition in hematological malignancies.

PU.1, FLI1, ERG, C/EBPα, C/EBPβ, and MYB all perform essential functions during normal hematopoiesis. This then raises the question as to why normal hematopoiesis is largely insensitive to pharmacological BET inhibition in animal models. However, it should be noted that therapeutic studies of BET inhibition generally rely on pulsatile drug exposure, which allows a >12 hour recovery period between doses (Dawson et al., 2011; Zuber et al., 2011c). While normal hematopoiesis is unaffected during such drug trials, a recent study demonstrates that sustained BRD4 inhibition using shRNAs results in impaired hematopoiesis, with our study now providing a mechanism to account for this phenotype (Bolden et al., 2014). Hence, a therapeutic window for BET inhibition may rely on dosing regimens that exploit TF-addiction in leukemia cells while allowing an opportunity for normal hematopoiesis to recover between cycles of drug exposure.

Our study also supports consideration of p300 and CBP as therapeutic targets in AML. It was previously shown that p300 acetylates the AML1-ETO oncoprotein to support leukemogenesis, with this genetic subtype of leukemia being sensitive to p300 inhibition (Wang et al., 2011). Our findings suggest a broader indication for p300/CBP inhibition in multiple AML subtypes, which may resemble the pattern of BET inhibitor sensitivity. Drug-like derivatives of p300/CBP KAT inhibitors are currently under development, which would be expected to perturb proto-oncogene expression in a manner similar to chemical inhibition of BET bromodomains. Moreover, chemical inhibition of p300 and CBP bromodomains may present an additional therapeutic opportunity in AML (Borah et al., 2011; Hay et al., 2014).

One unexpected finding in our study was the preferential occupancy of BRD4 at nucleosome-depleted regions, as detected by ChIP-seq analysis. While we cannot rule out that this observation was influenced by crosslinking biases, this pattern suggests a role for TFs in recruiting BRD4 to enhancer and promoter regions. Acetylated ERG would be a candidate TF that facilitates BRD4 recruitment at such sites, however, it is likely that a multitude of different TFs and cofactors cooperatively recruit BRD4 to regulatory elements. Another possibility is that acetylated histone tails can extend into the nucleosome-free region to support BRD4 recruitment to flanking elements or that specific patterns of histone di-acetylation might be rare at these regulatory elements, yet still promote high-affinity BRD4 binding. It is also possible that different BET proteins will localize in a distinct manner across the genome and may preferentially associate with specific TFs or combinations of histone marks (Stonestrom et al., 2015). Nonetheless, our study highlights the central role of TFs in directing BRD4 recruitment, which is likely to involve p300/CBP-dependent acetylation histone and non-histone proteins.

Experimental Procedures

RNA-Seq and ChIP-Seq library construction

Total RNA was extracted using TRIzol reagent (Invitrogen) according to the manufacturer’s instructions. TruSeq sample Prep Kit V2 (Illumina) was used for all the RNA-Seq libraries construction according to the manufacturer’s instructions. Protocols for performing ChIP have been previously described (Shi et al., 2013). For ChIP-Seq library construction, 50–100e06 of RN2 cells were crosslinked using 1% formaldehyde for 20 minutes at room temperature. After purifying immunoprecipitated DNA using QIAquick Gel Extraction Kit (QIAGEN), ChIP-Seq library was constructed with TruSeq ChIP Sample Prep Kit (Illumina) following the manufacturer’s instructions with the following exceptions; following adapter ligation the library was amplified with 15 cycles of PCR. Both the quantity and quality of library was determined by using Bioanalyzer with High Sensitivity chips (Agilent). The main GEO accession number for the raw and processed sequencing data reported in this paper is GSE66123.

A full description of experimental procedures can be found in the Supplemental Information.

Highlights.

• BRD4 occupancy in leukemia correlates with hematopoietic transcription factors

• The p300/CBP lysine acetyltransferases are essential for global BRD4 recruitment

• p300 acetylates ERG to promote an interaction with BRD4

• Hematopoietic transcription factors are functionally suppressed by BET inhibitors