Abstract
Alveolar rhabdomyosarcoma is a life-threatening myogenic cancer of children and adolescent young adults, driven primarily by the chimeric transcription factor PAX3–FOXO1. The mechanisms by which PAX3–FOXO1 dysregulates chromatin are unknown. We find PAX3–FOXO1 reprograms the cis-regulatory landscape by inducing de novo super enhancers. PAX3–FOXO1 uses super enhancers to set up autoregulatory loops in collaboration with the master transcription factors MYOG, MYOD, and MYCN. This myogenic super enhancer circuitry is consistent across cell lines and primary tumors. Cells harboring the fusion gene are selectively sensitive to small-molecule inhibition of protein targets induced by, or bound to, PAX3–FOXO1-occupied super enhancers. Furthermore, PAX3–FOXO1 recruits and requires the BET bromodomain protein BRD4 to function at super enhancers, resulting in a complete dependence on BRD4 and a significant susceptibility to BRD inhibition. These results yield insights into the epigenetic functions of PAX3–FOXO1 and reveal a specific vulnerability that can be exploited for precision therapy.
Significance: PAX3–FOXO1 drives pediatric fusion-positive rhabdomyosarcoma, and its chromatin-level functions are critical to understanding its oncogenic activity. We find that PAX3–FOXO1 establishes a myoblastic super enhancer landscape and creates a profound subtype-unique dependence on BET bromodomains, the inhibition of which ablates PAX3–FOXO1 function, providing a mechanistic rationale for exploring BET inhibitors for patients bearing PAX-fusion rhabdomyosarcoma. Cancer Discov; 7(8); 884–99. ©2017 AACR.
This article is highlighted in the In This Issue feature, p. 783
Introduction
Transcription factors (TF) recognize specific noncoding sequences across the genome, recruiting epigenetic machinery to regulate key cell identity genes, and are sequentially exchanged during development and differentiation (1). Oncogenic fusion genes involving TFs are predicted to profoundly alter normal developmental progression and cell identity in many malignancies (2).
Rhabdomyosarcoma (RMS) is a cancer of childhood and adolescence characterized by its inability to exit the proliferative myoblast-like state. Genomic and transcriptomic characterization implicates either chromosomal translocation resulting in the oncogenic fusion transcription factor PAX3/7–FOXO1 (fusion-positive alveolar subtype, FP-RMS) or mutations in receptor tyrosine kinase/RAS pathways (fusion-negative embryonal subtype, FN-RMS; refs. 3, 4). FP-RMS is characterized by a strikingly low somatic mutational burden indicating that the fusion gene is the primary oncogenic driver. Many other transcription/chromatin factor fusion gene–driven sarcomas have similar low mutational burdens (5–7). Importantly, patients with RMS who harbor a PAX3 fusion are more likely to be metastatic at presentation, relapse despite aggressive therapy, and have very poor survival (8), underscoring the critical need to develop therapeutic strategies for this subset of patients.
Early (PAX family TFs) and late (MYOG) regulators of normal myogenesis are temporally mutually exclusive in normal muscle development, yet FP-RMS tumors concurrently express high levels of PAX3–FOXO1 and the myogenic MYOD, MYOG, as well as MYCN (9, 10). Although the transcriptional perturbation caused by the PAX fusions has been previously reported (11, 12), the chromatin mechanisms by which PAX fusions dysregulate the myogenic program are unknown. In this work, we interrogated the underlying epigenetics that enforce the myogenic and oncogenic transcriptional program of cell lines and clinical tumor samples with PAX3 fusions. In charting the genome-wide landscape of histone modifications, we discovered that PAX3–FOXO1 drives expression of its target oncogenes by creating large deposits of active histone marks exclusively at enhancers, collaboratively with myogenic TFs, and by recruiting chromatin reader bromodomain-containing protein 4 (BRD4), which function at looped enhancer–promoter pairs within topological domain boundaries. Integrating epigenetic and mechanistic drug screening data exposed multiple biological nodes of chemical vulnerability, including BET bromodomains. The BRD4 inhibitor JQ1 has recently shown efficacy in RMS (13), but no mechanistic connection has been shown between BRD4 and PAX3–FOXO1. Here, we report that BRD4 inhibition disrupts a hitherto undiscovered PAX3–FOXO1 interaction with BRD4, causes a rapid degradation of the fusion gene, and ablates its transcriptional output, thus revealing a subtype-selective therapeutic vulnerability to BRD4 inhibition.
Results
PAX3–FOXO1 Establishes Active Chromatin at Distal Enhancers
The hallmark reciprocal translocation of chromosomes 2 and 13 (14) has coding potential for two fusion proteins (Supplementary Fig. S1A), but the only expressed allele has the 5′ end of PAX3 (DNA binding domain) and the 3′ transactivation domain of FOXO1 (Supplementary Fig. S1B). To gain insight into the epigenetic consequences of PAX3–FOXO1, we mapped the landscape of active and repressive histone marks by sequencing DNA enriched by chromatin immunoprecipitation (ChIP-seq) from a patient-derived fusion-positive FP-RMS cell line, RH4. Genome-wide, PAX3–FOXO1 resided predominantly (99%) in sites more than 2.5 kb distal from the nearest transcriptional start site (15), all of which harbored active enhancer marks (Fig. 1A), including acetylation at histone 3 lysine 27 (H3K27ac) and H3 lysine 4 mono/dimethylation (H3K4me1 and H3K4me2), but not the active promoter-associated mark H3K4me3 (Fig. 1B). PAX3–FOXO1 sites showed no evidence of poised and repressed chromatin, as demarcated by Polycomb-deposited trimethylation of H3 lysine 27 (H3K27me3), and, conversely, regions marked by H3K27me3 lacked both PAX3–FOXO1 and H3K27ac (Fig. 1B). Because histone marks are deposited in a combinatorial fashion, we defined reoccurring patterns associated with various chromatin functional states (16), providing the first epigenomic map upon which to overlay PAX3–FOXO1 occupancy (Fig. 1C). We found PAX3–FOXO1 most frequently occupied the strong enhancer chromatin state (Fig. 1C), exemplified by known PAX3–FOXO1 target FGFR4 (Supplementary Fig. S1C) and oncogenes MYC, ALK, and MET (Supplementary Fig. S1D).
Binding of PAX3–FOXO1 to Enhancers of High Disease and Biological Relevance
To identify which of these PAX3–FOXO1 sites are of high disease relevance, we first performed ChIP-seq in two additional cell FP-RMS lines, RH3 and SCMC, to identify a broader set of targets and recurrent sites. There were 1,783 peaks shared among 2 and 555 peaks for all 3 cell lines, and these showed the most statistical significance and largest signal per peak (Supplementary Fig. S1E–S1F). We next mapped a key histone marker of enhancers H3K27ac in a panel of FP-RMS cell lines and tumor samples, and found 1,107 high-confidence PAX3–FOXO1 sites occupying enhancers in one or more of the cell lines (Fig. 1D; Supplementary Table S1). We then generated a similar map of enhancers in FN-RMS cell lines and tumors, which shared 334 enhancer loci. Four hundred forty-six PAX3–FOXO1-bound enhancers were also present in myogenic samples. The enhancers unique to FP-RMS were enriched, by GREAT ontology (17), in pathways involved in early development, whereas shared enhancers were enriched for late muscle differentiation (Supplementary Fig. S1G).
PAX3–FOXO1 acting across large one-dimensional sequence distances has confounded target gene identification. Previous reports to identify these targets were based on either changes in gene expression (11, 12, 18) or proximity of a gene to a PAX3–FOXO1 peak in a single-cell line with no consideration of expression or chromatin context (15). We therefore identified high-confidence PAX3–FOXO1 target genes by a series of criteria (see Supplementary Fig. S2A–S2B and Supplementary Methods). In brief, we (i) used only PAX3–FOXO1 bound to enhancers that were recurrent in cell lines and tumors (from Fig. 1D), (ii) selected for expressed genes, as PAX3–FOXO1 was found only in active chromatin states, (iii) excluded nearby expressed genes if they were not found within the same topologically associated domain (TAD; predicted by HiC data, ref. 19) as the PAX3–FOXO1 bound enhancer, and (iv) included the maximally expressed gene within each TAD harboring PAX3–FOXO1. Using this approach, we found 1,010 high-confidence targets, 678 of which were novel, and, of note, 439 were significantly reduced by short hairpin RNA (shRNA) knockdown of PAX3–FOXO1 for 48 hours (Supplementary Fig. S2C–S2G). Novel targets included oncogenes (n = 24), TFs (n = 53), and several imprinted genes (n = 7; Supplementary Table S2).
Thus, our data support the hypothesis that PAX3–FOXO1 enables transcription by directing active chromatin marks to distal enhancers surrounding oncogenes, imprinted and myogenic genes.
FP-RMS Tumors and Cell Lines Possess a Myogenic Transcriptional Program
TFs act on enhancers in a stoichiometric manner, and those that are expressed at unusually high concentrations and are able to bind a majority of enhancers are defined as master transcription factors (MTF) and key determinants of cell fate (1). Although PAX3–FOXO1 is the primary driver, as an MTF it is likely to not work alone. We identified other MTFs that were overexpressed compared with normal tissues (P < 10−30), had consistently high levels (average FPKM >20), were super enhancer regulated, and had significant motif enrichment in enhancers (Fig. 2A). To do this, we first compared the RNA-sequencing (RNA-seq) profiles of RMS primary tumors (n = 103) and cell lines (n = 37; ref. 3) with normal human organ tissues (n = 188). This identified a consistently overexpressed core of 170 TFs, suggesting a convergent underlying epigenetic state (Fig. 2B; Supplementary Table S3). The TFs, of these RMS tumors resembled those of myoblasts and myotubes (20), and all have remarkably high levels of the lineage-determining TFs MYOD1 and MYOG as compared with other normal tissues (Fig. 2B and C). Unlike myoblasts, MYF5 is typically missing from RMS, although when present it appears to be mutually exclusive with MYOD1 (21). Another important divergence from normal myogenesis was high expression of the transcriptional amplifier MYCN (Fig. 2C), a known target of PAX3–FOXO1 (15). MYCN expression was generally higher in FP-RMS tumors whereas MYC expression was higher in FN-RMS; however, many tumors had expression of both, such that the sum was consistently high for all patients with RMS (Supplementary Fig. S3A and S3B). RMS cell lines faithfully recapitulated the transcriptional commitment of the primary tumors to these candidate MTFs.
Super Enhancer Analysis in RMS Cell Lines and Primary Tumors Implicates MYOD, MYOG, and MYCN as Master Regulators
A small fraction of active enhancers acquire a large fraction of transcriptional machinery and active chromatin marks (frequently demarcated by H3K27ac), and have been defined as super enhancers (22, 23). These regions are cell type–specific and control expression of cell identity genes in normal tissues and oncogenes in cancer. Although super enhancers may simply be clusters of additive enhancers (24), they nevertheless capture the most active enhancers associated with the core regulatory MTF circuitry (25). In RMS cells bearing PAX3–FOXO1 (RH4), we identified 776 super enhancers (ranked as the top 4% of all enhancers with 38% of the total H3K27ac signal) that were associated with hallmark RMS genes such as IGF2, FGFR4, ALK, MYOD1, MYOG, and MYCN (Fig. 2D). To evaluate the clinical relevance of our cell lines as models for enhancer architecture, we mapped super enhancers by H3K27ac signal in a set of 3 FP-RMS and 5 FN-RMS primary tumors. We found concordance of super enhancers among PAX3–FOXO1-driven cell lines and tumors (Fig. 2D). MYCN had the highest ranked super enhancer in MYCN-amplified cell lines and tumors. A common epigenetic landscape was found in FN-RMS cell lines and tumors (Fig. 2E; Supplementary Fig. S3C), which differed from FP-RMS at super enhancers such as FOXO1, MYCN, and ALK (Supplementary Fig. S3D). MYCN itself possessed a remarkable 5 super enhancers within the surrounding TAD structure (HiC data from ref. 19), which circularized chromatin conformation capture followed by sequencing (4C-seq) revealed all physically interact not only with MYCN, but also with each other (Supplementary Fig. S3E). Motif analysis at super enhancer sites revealed a highly significant enrichment of MYOD, MYOG, PAX3–FOXO1, and MYCN recognition sequences (Fig. 2F; Supplementary Table S4). Importantly, we observed a consistent reduction in cell viability in 2 FP-RMS cell lines when PAX3, MYCN, or MYOD1 were targeted by siRNAs (Fig. 2G). The suppression was greater in FP-RMS than in FN-RMS, further implicating them as essential master regulators of FP-RMS.
MYOD alone can reprogram cells into myogenesis (26), and it is intriguing that FP-RMS tumors remain undifferentiated despite extremely high levels of the myogenic TFs MYOD and MYOG. Given this, we hypothesized that PAX3–FOXO1 acts as a strong chromatin activator, preventing decommissioning of distal enhancers that are lost during normal muscle differentiation. To study this, we analyzed active enhancer mark H3K27ac (27) around PAX3–FOXO1 peaks, in myoblasts, myotubes, and muscle tissue. We found that enhancers controlling MYOD1 decreased during muscle differentiation and were absent in other tissues and cell lines, whereas in FP-RMS cells the MYOD1 enhancer locus was expanded and contained multiple PAX3–FOXO1 peaks (Fig. 2H). Genome-wide, myogenic enhancers found at PAX3–FOXO1 peaks decreased during the transition from myoblasts to myotubes to skeletal muscle (Fig. 2I). Genes associated with these decommissioned enhancers included TFs (MYCN, MYC, MSC, and MYOD1 itself) and epigenetic modulators which have known involvement in regulating self-renewal, embryonic development, muscle development, and chromatin organization (Supplementary Table S5), and imprinted genes involved in mesoderm development (MEST, IGF2). These data suggest that PAX3–FOXO1 may induce a myoblastic state by maintaining active chromatin at enhancers controlling these genes. This was further supported by the observation that after 48 hours of PAX3–FOXO1 knockdown these enhancers lost substantial H3K27ac signal as measured by ChIP with reference exogenous genome (ChIP-Rx; ref. 28; Fig. 2J).
PAX3–FOXO1 Collaborates with MYOG, MYOD, and MYCN at Super Enhancers
Our results indicated that PAX3–FOXO1 activates the MTFs MYOD, MYOG, and MYCN, and together with these, establishes the epigenome and transcriptome of FP-RMS (Fig. 3A). To validate this, we performed ChIP-seq on MYOD, MYOG, and MYCN and studied the extent of their collaboration with PAX3–FOXO1 to shape the enhancer landscape. We observed widespread binding of MYOD, MYOG, and MYCN, each covering 9 to 15 times as much genomic space as PAX3–FOXO1 (Supplementary Fig. S4A). When all four MTFs were colocalized in the genome they harbored greater signal of active histone marks, especially H3K27ac (Supplementary Fig. S4B–S4C). Enhancers with all 4 MTFs were frequently super enhancers (Fig. 3B). Almost every super enhancer was occupied by 3 or more of these MTFs, unlike typical enhancers (Fig. 3C). Super enhancers spanned a median of 23.5 kb (compared with 1.2 kb for typical enhancers) and exhibited a higher load of MTFs only when considering their constituent peaks (Fig. 3C; Supplementary Fig. S4D). PAX3–FOXO1 occupied only a small fraction of typical enhancers (5%) but many super enhancers (47%). MYOG, MYOD, and MYCN are bound to almost every super enhancer, and although PAX3–FOXO1 was found in only half of all super enhancers, its preference for super enhancers over typical enhancers was profound (Fig. 3D and E). Super enhancer–associated genes were transcribed at significantly higher levels than typical enhancer genes, and enhancers occupied by 4 MTFs were found to be most highly transcribed compared with enhancers with fewer MTFs (Fig. 3F).
The genome-wide placement of super enhancers was strongly determined by the myogenic MTFs recognizing the underlying DNA sequence CAGCTG (Fig. 2F), but interestingly there was low enrichment of the canonical MYCN motif CACGTG (Supplementary Table S4). More than 80% of MYCN peaks were distal (Supplementary Fig. S4E), and de novo motif analysis discovered MYCN prefers a myogenic E-box sequence (RRCAGCTG) nearly identical to that of MYOG and MYOD (Supplementary Fig. S4F). Thus, MYCN may behave in a manner akin to MYC acting as a general transcriptional amplifier, following to locations opened by more sequence-specific (and lineage determining) TFs (29, 30).
Master (or lineage determining) TFs are predicted to maintain cell identity by mutual and self-reinforcement, creating autoregulatory feed-forward loops (1, 31). We found this to be the case in FP-RMS, where the super enhancers controlling PAX3–FOXO1, MYOD, and MYCN contain all four of these MTFs (Fig. 3G). The MYOG super enhancer is bound by all except the PAX fusion, consistent with the logic and timing of normal myogenesis which successively progresses from dominance of PAX3 to MYOD to MYOG (32). To investigate the contribution of interconnection to gene expression, we used shRNA against each factor followed by RNA-seq. PAX3–FOXO1 depletion disrupted both enhancer acetylation and RNA expression at MYOD and MYCN, but an indirect increase at MYOG (Supplementary Fig. S4G). MYOG expression was the most profoundly reduced by knockdown of either MYOD1 or MYCN (Supplementary Fig. S4H and S4I). All MTFs were sensitive to MYCN depletion, and MYCN expression was sensitive to depletion of any one factor (Supplementary Fig. S4G and S4H).
BRD4, MED1, and p300 Occupy Key PAX3–FOXO1-Established Super Enhancers
Because PAX3–FOXO1 binds between 10Kb and 1Mb away from nearby promoters, it may mediate its transcriptional impact through chromatin factors looping over long distances. The transactivation domain of FOXO1 is known to recruit the coactivator p300 (33), which enzymatically acetylates histones, leading to binding of additional factors, including BRD4 and Mediator (34). We therefore hypothesized that PAX3–FOXO1 recruits p300 and the other cofactors leading to chromatin remodelling. To test this, we performed ChIP-seq of these components and analyzed their co-occupancy at PAX3–FOXO1-bound enhancers. Our results confirmed that these proteins co-occupy enhancers with PAX3–FOXO1 and were sites of open chromatin as determined by DNase hypersensitivity (Fig. 4A). We observed that p300 followed PAX3–FOXO1 at virtually every site, and BRD4 was co-occupant at the majority of these enhancers (72%), especially those with super enhancer architecture (95% of super enhancers; Fig. 4B). PAX3–FOXO1 locations lacking BRD4 showed no evidence of looping machinery MED1, CTCF, and RAD21, whereas sites bound by BRD4 did (Supplementary Fig. S5A). We predicted that these two PAX3–FOXO1 modes (with and without BRD4; Supplementary Fig. S5B–S5C) would have divergent functional consequences, and found that sites with BRD4 had greatly increased expression from associated genes (Supplementary Fig. S5D), and GREAT ontology analysis showed only BRD4-containing peaks were enriched for FP-RMS gene sets (Supplementary Fig. S5E).
It is unknown if PAX3–FOXO1 itself is capable of inducing de novo myogenic enhancer formation. We thus stably expressed PAX3–FOXO1 in a human fibroblast cell line (7250) and studied changes in chromatin and the corresponding changes in gene expression. We found that PAX3–FOXO1 opened the chromatin landscape, as evidenced by an increase in DNA hypersensitivity at enhancers compared with control parental cells (Fig. 4C). Furthermore, PAX3–FOXO1-bound enhancers saw an increase in H3K27ac and recruitment of the acetylated lysine reader BRD4 (Fig. 4C). These sites of opening (n = 836) are active, PAX3–FOXO1 bound, and BRD4 loaded in FP-RMS cells (Fig. 4D). Many of the enhancers, such as the super enhancers upstream of MYOD1 (Fig. 4E), MYOG, and FGFR4 (Supplementary Fig. S5F), are faithfully reconstituted with a size and shape similar to those in RMS. Of the 568 high-confidence PAX3–FOXO1 super enhancers, 349 are recapitulated in these fibroblasts reprogrammed by exogenous PAX3–FOXO1 (Supplementary Fig. S5G).
The proximity of the PAX3–FOXO1-directed super enhancers near MYOD1 and their coappearance with MYOD1 transcription led us to predict three-dimensional looping to bring these super enhancers to the promoter. We confirmed these cis interactions in RH4, using 4C-seq from two viewpoints (one at the most distal PAX3–FOXO1-bound super enhancer and the other at the MYOD1 promoter; Fig. 4F). These loops enable the physical interaction of the super enhancer–bound proteins MED1, p300, and BRD4 with the promoter of MYOD1 to facilitate transcription. The 4C interactions were restrained to the TAD predicted by HiC data (19) in other human cells (Fig. 4F). Thus, PAX3–FOXO1 acts as a pioneering factor, opening chromatin, recruiting coactivators, and driving transcription through looped myogenic super enhancers.
Molecular Sensitivities of PAX3–FOXO1 Tumor Cells Are Associated with Super Enhancers
Our results thus far showed that PAX3–FOXO1 establishes super enhancers not only at myogenic genes but also at multiple druggable oncogenic drivers. Because super enhancers are cell-type restricted, we hypothesized that FP-RMS would be selectively vulnerable to inhibition of these super enhancer–driven pathways. Hence, we determined the landscape of molecular sensitivities in PAX3–FOXO1-positive patient-derived cell lines (RH41, RH5) by dose responses at 48 hours for 1,912 compounds. To deprioritize compounds with nonselective cytotoxicity, we also treated fibroblast cell lines (NIH3T3, 7250, and T9195). Our small-molecule library MIPE4 (35) was assembled to have high mechanistic diversity with an emphasis on clinically relevant compounds (Supplementary Fig. S6A–S6B; Supplementary Table S6). Area under the dose response curve (AUC) was used as the measure of potency, as this metric captures both dose dependence and maximum response (see Supplemental Methods and Supplementary Fig. S6C–S6E). PAX3–FOXO1-expressing cells were selectively sensitive to super enhancer–driven RTKs (FGFR4, IGF2/IGF1R, and ALK) and downstream kinases (PI3K, AKT, and mTOR). Furthermore, transcriptional cofactors (HDACs, BRD4) originally identified as super enhancer–associated proteins (23) were also selective for FP-RMS cells (Fig. 5A–B; Supplementary Fig. S6F). Thus, the identification of super enhancer–associated genes highlighted multiple candidate targets for therapy, which may be a useful approach for other cancers. Of note, FP-RMS was sensitive to inhibiton of BRD4 (36), which has recently shown promising results in RMS tumor models (13), although no molecular explanation in connection with PAX3–FOXO1 has previously been made. Thus, with these new data reported herein that BRD4 lies at an important node in the PAX3–FOXO1 circuitry (Supplementary Fig. S6G), we next sought a mechanistic explanation for the sensitivity.
Cells with PAX3–FOXO1 Are Selectively Sensitive to BET Bromodomain Inhibition
Given the sensitivity of FP-RMS cells to chemical BRD inhibition, we sought to determine if this was attributable to PAX3–FOXO1, or if the FN-RMS subtype was also vulnerable. Thus, we tested an expanded panel of RMS cell lines against 5 structurally diverse BET bromodomain inhibitors (BRDi) and 1 pan-bromodomain inhibitor, Bromosporine. Thienodiazepine inhibitors, JQ1 and the clinical analogue OTX015, were the most potent, with dose response consistently in the nanomolar IC50 range for FP-RMS cell lines, whereas most often in the micromolar range in fusion-negative (mutant RAS) RMS lines (Fig. 5C; Supplementary Fig. S7A–S7C). BRD4 inhibition dramatically reduced proliferation over time in PAX3–FOXO1-driven cells, whereas mutant RAS-driven RMS cells were relatively unhindered (Fig. 5D; Supplementary Fig. S7D). JQ1 action was mediated by programmed cell death in a dose-dependent manner (Supplementary Fig. S7E). Importantly, although MYCN amplification clearly confers special sensitivity to BRD4 inhibition in neuroblastoma (37), FP-RMS cells with (RH5 and SCMC) or without (RH3, RH4, and RH41) MYCN amplification were all sensitive (Fig. 5C). PAX3–FOXO1 also conferred 11-fold increased BRDi sensitivity to fibroblasts (Fig. 5E). Patient-derived xenografts grown in culture further confirmed PAX-fusion vulnerability, and FN-RMS resistance, to JQ1 (Supplementary Fig. S7F).
The JQ1 targets BRD2/3/4 (but not BRDT) are expressed in RMS, but not overexpressed compared with normal tissues (Supplementary Fig. S8A). RNAi screening of bromodomain- containing proteins revealed greatest dependence on EP300, KAT2A, BRD3, and BRD4 (Supplementary Fig. S8B). Among the BET family members, BRD4 was the most sensitive to genetic depletion, which incurred apoptotic events (Supplementary Fig. S8C–S8E).
PAX3–FOXO1 Requires BRD4 for Function and Stability
The sensitivity of PAX3–FOXO1-driven cell lines to BET inhibition and our data showing co-occupancy of BRD4 in all PAX3–FOXO1-bound super enhancers led us to the hypothesis that PAX3–FOXO1 is dependent on BRD4 to mediate transcription of its target genes. This is consistent with the known role of BRD4 in stimulating transcriptional elongation (38) and previous reports that BRD4 inhibition causes rapid decommissioning of super enhancers and selective inhibition of super enhancer–driven genes (39). To test this, we compared the fold change in super enhancer–associated versus typical enhancer–associated genes before and after treatment with JQ1 for 6 hours by RNA-seq. Indeed, we found that super enhancer–associated genes in FP-RMS cells were especially sensitive to JQ1, and that this selectivity was also seen upon genetic depletion of PAX3–FOXO1 itself (Fig. 6A; Supplementary Table S7). Gene set enrichment analysis (GSEA) revealed that JQ1 was able to selectively downregulate PAX3–FOXO1 target genes, with enrichment mirroring knockdown of PAX3–FOXO1 (Fig. 6B; Supplementary Fig. S9A–S9B). Many key super enhancer–driven TFs and PAX3–FOXO1 targets were suppressed whereas cell-cycle arrest and apoptosis genes were upregulated by JQ1 (Fig. 6C). The known sensitivity of MYC family proteins to BRD inhibition was seen at both the transcript and protein levels (Fig. 6C; Supplementary Fig. S9C). Coordinately, master regulators MYOD and MYOG were also reduced at the protein level upon JQ1 treatment in a dose- and time-dependent fashion (Fig. 6D; Supplementary Fig. S9D). These effects were not a consequence of reducing PAX3–FOXO1 transcription, as evidenced by exon-level expression in RNA-seq data (Fig. 6E).
Although ChIP-seq evidenced that BRD4 and PAX3–FOXO1 are co-occupant spatially, it was unclear whether they bind chromatin cotemporally. To study this, we performed tandem chromatin immunoprecipitations (re-ChIP) for PAX3–FOXO1 followed by BRD4, and vice versa, in RH4 cells treated with either DMSO or JQ1 for 6 hours. Quantitative PCR of re-ChIP DNA at enhancer sites bound by both PAX3–FOXO1 and BRD4 (MYOD1, PIPOX) revealed strong enrichment regardless of ChIP order (Fig. 6F), which was almost completely ablated in JQ1-treated cells. This was not observed at BRD4-only enhancers near MYOG. To corroborate this, we performed coimmunoprecipitation with BRD4 and PAX3–FOXO1 in the presence of DMSO or JQ1 (Fig. 6G), and found that JQ1 indeed ablated this endogenous interaction. The PAX3–FOXO1 and BRD4 interaction was seen in both directions using exogenous, tagged versions of these proteins (Supplementary Fig. S9E). Given no alteration in PAX3–FOXO1 mRNA levels, and only modest reduction in protein levels (Fig. 6H), we suspected JQ1 caused destabilization of the PAX3–FOXO1 protein. Remarkably, the half-life of PAX3–FOXO1 was reduced from >8 hours to 28 minutes with JQ1 compared to DMSO (Fig. 6I–J; Supplementary Fig. S9F) in the presence of cycloheximide to inhibit protein translation. This appears to be an on-target effect, as shRNA against BRD4 also caused PAX3–FOXO1 to decrease at the protein level, but not the transcript level (Supplementary Fig. S9G). Thus, PAX3–FOXO1 interacts with BRD4 at enhancers, and treatment with JQ1 leads to loss of this interaction with rapid degradation of PAX3–FOXO1 protein.
JQ1 Selectively Disrupts PAX3–FOXO1-Driven Transcription to Suppress Tumor Growth In Vivo
The in vitro sensitivity of FP-RMS to BRD4 inhibition by the rapid and the specific inhibition of PAX3–FOXO1 function indicated this may be an effective therapeutic strategy. To test this, we developed an imagable readout to monitor in vivo activity of PAX3–FOXO1 super enhancer affected by drugs administered to mice. We engineered RMS cells to express luciferase (and GFP), controlled either by a constitutively active cytomegalovirus (CMV) promoter or a PAX3–FOXO1-dependent super enhancer (cloned from the intronic ALK super enhancer; Supplementary Fig. S10A). This ALK super enhancer was consistent in FP-RMS cell lines and tumors, while completely absent in FN-RMS (Supplementary Fig. S3D). No activity was seen from the ALK super enhancer construct in FN-RMS cells lacking PAX3–FOXO1 (Supplementary Fig. S10B). The ALK super enhancer was suppressed by knockdown of PAX3–FOXO1 or by point mutation of the PAX3–FOXO1 binding motif (Supplementary Fig. S10C–S10E). BRD inhibition with JQ1 suppressed only the ALK super enhancer– but not the CMV-driven reporter in a dose-dependent manner in vitro (Fig. 7A), whereas the general transcription CDK7 inhibitor THZ1 suppressed both. Use of these reporter lines in vivo demonstrated a selective inhibition of PAX3–FOXO1 activity with subsequent significant suppression of tumor growth (Fig. 7B and C; Supplementary Fig. S10F). The tumor-suppressive activity of JQ1 was confirmed in a second FP-RMS cell line xenograft, SCMC (Supplementary Fig. S10G–S10H).
Discussion
Pediatric sarcomas that harbor fusion oncogenes are reported with relatively few genomic alterations despite their clinically aggressive behavior (7). Here, we report that the PAX3–FOXO1 fusion gene directs a profound epigenetic chromatin remodeling in cooperation with the master regulators MYOD, MYOG, and MYCN by the establishment of super enhancers.
By comprehensively charting the first epigenetic landscape of RMS in cell lines and primary tumors, we identified a core commitment to super enhancer regulation of MYOD, MYOG, and MYCN, which in turn drive virtually all densely acetylated enhancer clusters that specify RMS cell identity. Although the concept of a “super” enhancer remains controversial, and may simply be a cluster of additive enhancers (24), we found the classification useful in that the most uniquely and highly expressed candidate MTFs were all driven by super enhancers, often more than one. Our results indicate that PAX3–FOXO1 directly establishes super enhancers to drive itself, MYOD1, and MYCN, and indirectly establishes a super enhancer (through MYOD and MYCN) to drive MYOG. Downstream, these factors are acting in concert, whereas in normal development they act sequentially (32). Virtually all of the discovered super enhancers were bound by 3 or 4 of these MTFs, and associated with the highest levels of gene expression. Remarkably, the presence of MYCN at these hyperactive regions resembles the function of MYC as a general transcriptional amplifier (29) and may underlie transcriptional addiction in FP-RMS (40). Together, these four MTFs cause a profound epigenetic reprogramming, freezing the cells in a myoblastic state with PAX3–FOXO1 as the conductor.
Cell type–specific distribution of BRD4 is accomplished by TF recruitment to their enhancers (41). Our study showed that PAX3–FOXO1 can direct de novo recruitment of BRD4 to specific chromatin sites when introduced into human fibroblasts, accompanied by opening of chromatin and acetylation of H3K27 at enhancers proximal to key RMS genes. This purely activating function of PAX3–FOXO1 in pediatric RMS is in sharp contrast to the dual functionality of EWS–FLI1 in pediatric Ewing sarcoma (5), which has both activating and repressive functions depending on genomic context.
The extent to which epigenetic profiling can aid prediction and interpretation of chemical sensitivities remains unclear. Using mechanistically informed drug screening, we observed that molecules targeted to super enhancer–bound coactivators (BRD4 and HDAC; ref. 22) and super enhancer targets (such as the receptor tyrosine kinases FGFR4 and ALK) are the most selectively potent in these FP-RMS cells. These data reinforce previous strategies that showed FP-RMS sensitivity to FGFR and IGF1R inhibition (42, 43) and HDAC inhibitors (4, 44, 45), and add previously unknown mechanistic insights to recently discovered BRD4 vulnerability (13). The observation that super enhancer–associated targets represent key vulnerabilities may be broadly applicable to cancers driven by epigenetic reprograming through super enhancer networks and may be critical to prioritizing combination strategies as well.
The same translocation which causes FP-RMS also creates an Achilles' heel by addicting cells to BRD4. Our data add to the pleiotropic utility of BRD4 inhibitors, with a recent wave of studies in a diverse group of cancers (37, 46–50). The transcriptional impact of BRD4 inhibition appears to be context dependent, where its antitumor effect is linked to dampening one or more master regulators, often including MYC in hematologic malignancies (51, 52), MYCN in neuroblastoma (37), POU2AF1 and PAX5 in DLBCL (47), or FOSL1 in adenocarcinoma (53), among others. JQ1 was shown recently to be effective in reducing RMS tumor growth by antiangiogenic properties, but the underlying chromatin-based mechanisms or FP-RMS subtype selectivity were not explored (13). Here, we find BRDi caused selective downregulation of MYC, MYCN, MYOD1, MYOG, and many other downstream PAX3–FOXO1- and super enhancer–driven genes. Surprisingly, JQ1 did not decrease expression of PAX3–FOXO1 mRNA at 6 hours, yet it rapidly decreased PAX3–FOXO1 protein stability. This may result from the disruption of interaction between BRD4 and PAX3–FOXO1, discovered by both coimmunoprecipitation and re-ChIP experiments, which JQ1 abrogated within hours of drug exposure. Thus, BRD inhibition by JQ1 leads to significant tumor suppression in vitro and in vivo, ablating the transcription-driving function of the fusion gene. Indeed, this mechanism may partially explain the antitumor effects seen with BI-2536 in FP-RMS mouse models (54), as this PLK1 inhibitor was recently found to possess a nanomolar inhibition of BET bromodomains (55). Excitingly, this provides a means of selectively drugging PAX3–FOXO1, a long-standing goal of FP-RMS research (56), and provides a new precision therapeutic for treatment of the aggressive PAX fusion–bearing RMS.
Methods
Cell Lines and Primary Tumors
All cell lines were routinely tested for Mycoplasma within one or two passages of each experiment herein, and cell line identities have been ensured by RNA-seq and genotyping. RH4, RH3, RH5, and RH41 were kind gifts from Dr. Peter Houghton (obtained between 2008 and 2010), SCMC from Dr. Janet Shipley (obtained between 2013 and 2015), and RD, CTR, and Birch from Dr. Lee Helman (obtained between 2008 and 2010). CRL7250 and NIH3T3 were obtained from the ATCC (obtained between 2008 and 2010). Validation was performed by DNA fingerprinting AmpFlSTR Identifiler PCR Amplification Kit (Catalog Number 4322288) by Life Technologies. Cell lines were grown in DMEM, 10% FBS, and supplemented with penicillin/streptomycin. Primary RMS cultures established from patient-derived tumor xenografts were collected at the St. Jude Children's Research Hospital (57). Cells were maintained in Neurobasal-A medium (ThermoFisher Scientific) supplemented with 2xB-27 supplement (ThermoFisher Scientific), 20 ng/mL bFGF, and 20 ng/mL EGF (both from PeproTech), and cultured on gelatin-coated plates in 5% CO2 at 37°C. Primary tumors were acquired via the NCI-coordinated ClinOmics protocol as previously described (58).
ChIP-seq and RNA-seq
ChIP-seq (59) and RNA-seq (3) were performed as previously described. Raw sequencing data and processed files have been made available through Gene Expression Omnibus (GEO) SuperSeries accession number GSE83728, which is comprised of SubSeries accession numbers GSE83724, GSE83725, GSE83726, and GSE83727. Details, including Illumina sequencing and bioinformatic methods, are available in Supplementary Methods.
DNase-seq with 10,000 Cells
Sites of DNase-sensitive chromatin were captured from 10,000 cells as recently described (60) with slight modifications. Briefly, freshly trypsinized cells were resuspended in DMEM, counted in duplicate (Nexcelom Automated Cell Counter), pelleted and resuspended in lysis buffer to achieve 120 μL of 60K cells, which was then divided into 6 replicates (10K cells per tube). DNaseI (Roche 04-716-728-001) was added to the cells (0.25–0.5 units) and incubated for 5 minutes at 37°C. The digestion was halted with 50 μL of stop buffer (9.5 mL H2O + 100 μL 1M TrisHCl pH 7.4 + 20 μL 5 mol/L NaCl + 200 μL 0.5 mol/L EDTA, with 150 μL 10% SDS and 125 μL proteinase K added just before use). Proteinase K activation at 55°C for 1 hour was followed by DNA purification by column (MiniElute PCR purification kit, Qiagen). Library preparation was performed as with ChIP-seq samples, except that paired-end was used rather than single-end sequencing on the NextSeq 500 (Illumina).
Small-Molecule Compounds
All molecules were dissolved in DMSO to a final concentration of 10 mmol/L and diluted to a final DMSO concentration of <0.03% by volume in DMEM for cell culture experiments. JQ1 was a gift from Jay Bradner (Novartis) and Jun Qi (Dana-Farber Cancer Institute). Bromosporine was provided by Peter Brown of the Structural Genomics Consortium. THZ1 was supplied by Nat Gray (Dana-Farber Cancer Institute). Other bromodomain inhibitors (OTX015, I-Bet-151, I-Bet-762, and I-Bet-726) were generously supplied by the Developmental Therapeutics Program (NCI, NIH).
Time Course of Dose-Response Cell Growth Assay
Dose responses were performed by quantifying percent cell confluence from phase contrast images taken every 4 hours using the IncuCyte ZOOM in 384-well plate format. Dose response was achieved using a range of 12 concentrations from 30 μmol/L to 0.17 nmol/L (dilutions divided by 3) and were performed in triplicate. Cells were plated to achieve 15% confluence at time of drug dosing, and monitored until control (DMSO) wells reached >95% confluence. IC50 values were calculated for each time point using the R statistical package drc (https://cran.r-project.org/web/packages/drc/drc.pdf).
Luciferase-Expressing Cells
pGreenFire vector from Systems Biosciences was modified by insertion of a cis-regulatory element surrounding the PAX3–FOXO1 binding site within the super enhancer found within the intronic region of the ALK gene (chr2:29880537–29880842). Cloning was performed using pCR2.1–TOPO system from PCR amplification product from genomic DNA of RH4 FP-RMS cells, and shuttled into the pGF1 vector at the EcoRI restriction site upstream of a minimal CMV promoter, which was completely inactive on its own in RMS cell lines. Viral particles were produced in HEK293T cells, harvested, filtered, and pelleted. pGF1 cloning vectors developed at SBI are self-inactivating as a result of a deletion in the U3 region of 3′-LTR. Upon integration into the genome, the 5′ LTR promoter is inactivated, which prevents formation of replication-competent viral particles. Pooled cells were selected using puromycin.
Animal Studies
Animal studies were approved by the National Cancer Institute's Animal Care and Use Committee, and all animal care was in accordance with institutional guidelines. Complete details are reported in Supplementary Methods.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Authors' Contributions
Conception and design: B.E. Gryder, M.E. Yohe, H.-C. Chou, J.S. Wei, C.J. Thomas, J. Khan
Development of methodology: B.E. Gryder, H.-C. Chou, Y. Song, R. Rota, X. Wen, S. Sindiri, M. Ferrer, J.F. Shern, C.J. Thomas, J. Khan
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): B.E. Gryder, M.E. Yohe, X. Zhang, J. Marques, M. Wachtel, N. Sen, Y. Song, R. Rota, X. Wen, J.S. Wei, F.G. Barr, S. Das, T. Andresson, M. Lal-Nag, M. Ferrer, C.J. Thomas, J. Khan
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B.E. Gryder, H.-C. Chou, J. Marques, M. Wachtel, B. Schaefer, N. Sen, A. Cleveland, X. Wen, S. Sindiri, J.S. Wei, F.G. Barr, R. Guha, M. Lal-Nag, M. Ferrer, J.F. Shern, K. Zhao, C.J. Thomas, J. Khan
Writing, review, and/or revision of the manuscript: B.E. Gryder, M.E. Yohe, H.-C. Chou, B. Schaefer, N. Sen, R. Rota, J.S. Wei, F.G. Barr, R. Guha, M. Ferrer, J.F. Shern, K. Zhao, C.J. Thomas, J. Khan
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): B.E. Gryder, H.-C. Chou, X. Wen, J. Khan
Study supervision: R. Rota, C.J. Thomas, J. Khan
Other (writing original manuscript): B.E. Gryder
Other (wet-lab investigation, providing materials): A. Gualtieri, S. Pomella
Acknowledgments
The authors are very grateful to C. Thiele, B. Stanton, J. Waterfall, L. Cao, T. Misteli, and C. Vakoc for helpful discussions and review of the manuscript. We are thankful to J. Qi, J. Bradner, N. Gray, P. Brown, the Structural Genomics Consortium, and the Developmental Therapeutics Program (NCI, NIH) for providing small molecules. We also appreciate the help of S. Li, N. Shivaprasad, S. Michael, and M. Skarzynski in providing reagents, cell lines, and antibodies. We thank C. Lin and R. Patidar for sharing scripts and bioinformatic advice.
Grant Support
This project has been funded with federal funds from the National Cancer Institute, NIH, contract no. HHSN261200800001E. A. Gualtieri is a recipient of a Fondazione Veronesi fellowship. R. Rota has been supported by Associazione Italiana Ricerca sul Cancro (AIRC) and PE-2013-02355271.
The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. government. Frederick National Laboratory is accredited by AAALAC International and follows the Public Health Service Policy for the Care and Use of Laboratory Animals. Animal care was provided in accordance with the procedures outlined in the “Guide for Care and Use of Laboratory Animals” (National Research Council, 2011; National Academies Press, Washington, DC).