Targeting the BRD4-HOXB13 Coregulated Transcriptional Networks with Bromodomain-Kinase Inhibitors to Suppress Metastatic Castration-Resistant Prostate Cancer

Prostate cancers can be slow growing and nonlife threatening or aggressive and lethal. Prognosis of castration-resistant prostate cancer (CRPC) is bleak; the castration resistant forms relapse in 2 to 3 years despite treatments with the second-generation androgen receptor (AR) blockers, abiraterone (ABR) or enzalutamide (ENZ). Thus, CRPC remains a leading cause of cancer-related deaths in American men, and the 5-year survival rate is only about 29% due to the limited therapeutic options (1, 2). Metastatic CRPC is considered an “epigenetic disease” as mutations are found at a much lower frequency in comparison with other solid tumors (3–5). Epigenetic deregulation can lead to aberrant activation of tissue-specific and developmentally regulated transcription factors and may be one mechanism driving castration resistance. Thus, identification of factors driving metastasis and antiandrogen resistance is critical to target, treat the resistant forms of the disease, to improve patient outcomes.

HOXB13 is a lineage-specific homeodomain containing transcription factor predominantly expressed in the prostatic tissues. Despite the concurrent expression with AR during prostate development in mice, HOXB13 expression is largely independent of androgen (6, 7). Moreover, HOXB13 regulates AR function in a context-dependent manner to promote or suppress AR transactivator function at various target genes (8). This may underlie its differential roles in regulating hormone-sensitive versus hormone-refractory prostate cancers (9–11). Not only the full-length AR, HOXB13 also regulates the genomic recruitment of AR-V7, a splice variant of AR that lacks the ligand binding domain, to define the AR-V7 cistrome in CRPCs (12). Consistently, HOXB13 expression is a feature of a majority of AR-positive prostate cancers and bone mCRPCs and correlates with poor prognosis (13). Further, a rare germline mutation in HOXB13 (G84E) was uncovered, which was not only associated with an increased risk of familial and hereditary prostate cancer in different ethnic populations, but male carriers develop the aggressive form of the disease with an earlier onset (14–17). Consistently, a recent study revealed poor prognosis and an early death for metastatic CRPC patients positive for HOXB13 circulating tumor cells (CTC), following treatment with the ABR (18). Although highly correlative, it is unclear whether HOXB13 is essential for CRPC growth, as well as the identity of its key effectors driving metastatic progression is unknown. Importantly, germline mutations in HOXB13 are rare; we reasoned that CRPCs may epigenetically promote deregulated expression that may underlie its role in malignancy.

A notable AR transcriptional coregulator is the bromodomain and extraterminal (BET) domain containing protein BRD4 (19, 20). The members of the BET family, BRD2 and BRD4, have essential functions during embryonic development and also regulate the pluripotency of embryonic stem cells (21). BRDs bind acetylated lysine residues at the N-terminus of histone H3/H4 or the nonhistone proteins such as the AR, and the prototype BET inhibitor JQ1 blocks this recognition to suppress the expression of target genes, such as c-MYC and PSA (19, 22). Although c-MYC expression is suppressed, it does not appear to be a major target of BRD4 inhibition in CRPCs (19, 20). We report for the first time that the BET domain protein, BRD4, binds the enhancer of HOXB13 gene upregulating its expression, and this BRD4-HOXB13 epigenetic axis activates AR-independent cell-cycle programs to promote CRPC proliferation. Combined, our study uncovers a conserved BRD4-HOXB13 transcriptomic network in mCRPCs that promotes cancer cell proliferation despite androgen deprivation and is targetable with novel small-molecule bromodomain-kinase inhibitors.

The human prostate cell lines 22Rv1, DU145, LNCaP, C4-2, PC3, RWPE-1, and VCaP were directly purchased from ATCC that have been authenticated by short tandem repeat profiling and grown as recommended by ATCC. All cell lines in the current study were used within 3 months or ∼6 to 8 passages upon receipt and replenished from frozen stocks. C4-2B was grown as described earlier (2). HOXB13 CRISPR/Cas9 (GFP expressing) gene-editing and HOXB13 HDR plasmid (RFP expressing) constructs were purchased from Santa Cruz. Cultures are routinely tested for mycoplasma contamination with a sensitive PCR-based screening using the PCR Mycoplasma Test Kit I/C from Promokine once in 2 months (PK-CA91-1048).

Anti-Actin (Sigma-Aldrich, A2228), Anti-Vinculin (Sigma-Aldrich, V9131), Histone H3K27ac pAb (Active Motif, 39133), Anti-Acetyl-Histone H4K16 (EMD Millipore, 07-329), Anti-Acetyl-H4K12 (EMD Millipore, 07-595), H4K5Ac (CST 8647), H4K8Ac (Active Motif), H3K37me3 (CST 9733S), H3K4me1 (Active Motif, 39398), H3K4me3 (CST 9727S), RNA Pol II (Active Motif, 101307), Normal Rabbit Ig (CST, 2729), HoxB13 (H-80), Santa Cruz Biotechnology sc-66923, HoxB13 (H-9), sc-28333, TAF1 Rab mAb (CST, D6J8B), AR (C-19), sc-815, AR (N-20), sc-816, BRD4 Antibody Bethyl Laboratories A301-985A50, Cleaved PARP (Asp214; D64E10) XP Rabbit mAb, CST, 5625, c-MYC (CST, D84C12), IgG (Active Motif 101226), AR chromatin immunoprecipitation (ChIP) Ab (Abcam, ab3509) were purchased from commercial sources. MA4-022-1 (Compound 3), MA4-022-2 (Compound 4), SG3-179 (Compound 5), MA3-068-1 (Compound 1), and MA6-082 (a tetherable analogue of MA4-022-1) synthesis and structures are described (23). Mass spectrometry and HNMR were performed to confirm the purity of each of the above compounds. The following compounds were all purchased from Selleck Chemicals and structures for the non–FDA-approved compounds are shown in Supplementary Table S1; ENZ (S1250), JQ1 (S1047), barasertib (AZD1152-HQPA; S1147), abiraterone acetate (S2246), GSK-126 (S7061), fedratinib (S2736), Ruxolitinib (S1378) and iBET-762 (S7189). siRNAs were purchased from Santa Cruz; human BRD4 siRNA (SC43639), BRD3 siRNA (SC60284), BRD2 siRNA (SC60282), BRDT siRNA (SC60286), control siRNA (SC37007), and AR siRNA (SC29204). HOXB13 siRNA (SR307118c from Origene) and BRD4 siRNAs were purchased from Dharmacon (Supplementary Table S1).

Drug-affinity chromatography experiments were conducted as described (24). Briefly, MA6-082, a tetherable analogue of MA4-022-1, and ampicillin were immobilized on NHS-activated Sepharose for Fast Flow resin (GE Healthcare) and blocked with ethanolamine overnight. C4-2B cells lysate containing 1 mg of protein was added to the affinity matrix for 2 hours and processed. Competition experiments were conducted by incubating total cell lysates with 20 μmol/L MA4-022-1 for 30 minutes prior to affinity chromatography.

A nanoflow ultra high-performance liquid chromatograph (RSLC, Dionex) coupled to an electrospray bench top orbitrap mass spectrometer (Q-Exactive plus, Thermo) was used for tandem mass spectrometry peptide sequencing experiments. Data were searched by Mascot (v2.4.1) using the Swiss-Prot human database. Following protein ID, the data were filtered (95% minimum peptide threshold, 95% protein threshold, 10 ppm parent tolerance, 0.05 Da fragment tolerance; 0.6% peptide False Discovery Rate (FDR), 3.4% protein FDR) using Scaffold 4.6.1 (Proteome Software). A maximum of two missed cleavages were allowed. The data were then exported and imported into Galaxy (25–27) for analysis with the affinity proteomics analysis tool APOSTL (http://apostl.moffitt.org; ref. 28). The data were preprocessed into inter, bait, and prey files and analyzed by SAINTexpress (29) and the CRAPome (30) within APOSTL.

RNA was prepared and quantified as described earlier (2). Raw data (reads) were quality controlled by FastQC algorithm. Mapping to human genome (hg38) was done by tophat2 (31). SAMTools (32) was used to select only mapped reads. Alignment files (.bam) were then imported into Partek NGS (Partek, Inc.) computing read counts per gene and transcript and converting them to RPKM (reads per kilo base per million) data. Feature summarization step in Partek uses the expectation-maximization (EM) approach to estimate transcript abundance. Obtained RPKM signals were further log₂ transformed. Box and whiskers plot were used to verify that sample distributions had no outliers. After the import and quality control (QC) were performed, we created phenotypic groups and defined differentially expressed genes (DEG) between biological groups of interest using DEseq procedure in R software (33). MetaCore and NextBio systems were used for enrichment analysis to identify which Gene Ontology (GO) processes, pathway maps, process networks, diseases and metabolic networks were represented in the analysis data with greater statistical significance. Heat maps and Venn diagrams were produced with Partek (Partek, Inc.) and GENE-E software from the Broad Institute (https://software.broadinstitute.org/GENE-E/). Primers used for qRT-PCR for detection of HOTBIN10 genes, HOXB13 and Actin, as well as for ChIP analysis of BRAH1/BRAH2, and IGX1A (control primers) are listed in Supplementary Tables S2 and S3.

Moffitt Information Shared Services/Collaborative Data Services Core provided the deidentified gene-expression data for the 194 primary and 29 metastatic tumors (Rosetta/Merck Human RSTA Custom Affymetrix 2.0 microarray; HuRSTA_2a520709.CDF) from the Moffitt TCC data set. Studies that fall under the TCC protocol have been obtained with patients' written consent. The metastatic samples were obtained from multiple sites (lung, liver, bone, lymph nodes, and soft tissue), while primary samples were obtained from anterior/posterior apex/mid/base of prostate. All samples passed multiple internal QC filters. All probe set expression values were normalized using IRON (34). In case of multiple probe sets, the gene-expression value was represented by the probe set with the median intensity. The GSE21034 series MSKCC data set (35) was used for validation studies.

Evince multivariate analysis software version 2.7.0 (UmBio AB) was used for this purpose. After PCA plots were generated using all genes in a given list, the initial gene lists were reduced to genes with the highest loading scores in order to maximize the fraction variability explained by the first principal component.

All animal experimentation was performed using the standards for humane care in accordance with the NIH Guide for the Care and Use of Laboratory Animals and conducted under a protocol approved by the University of South Florida IACUC Committee (IACUC 2095R). Four-week-old intact or castrated male SCID mice (n = 10–12/group) were obtained from Charles Rivers Laboratories. Mice were implanted subcutaneously with 2 × 10⁶ C4-2B or VCaP cells suspended in 100 μL of PBS with Matrigel (BD Biosciences) into the dorsal flank. Mice were randomized when the tumors reached 100 to 200 mm³ and treated with subcutaneous injections of either the vehicle (control) or MA4-022-2 or JQ1 at 50 mg/kg body weight for 5 days/week for 4 weeks. Measurements of the tumors were made using digital calipers twice a week and tumor volumes were calculated. Tumors, blood, and organs were harvested from the vehicle and treated animals after euthanasia for histopathology analysis and sampled for RNA expression profiling.

Differences in means between individual groups were analyzed by Student t test or one-way analysis of variance (ANOVA). ANOVA was followed by Sidak multiple comparison tests to correct for multiple comparisons. Statistical assumptions of homogeneous variance and normality were tested using the Levene F test and the Shapiro test, respectively. The Wilcoxon-rank sum test was used to compare the groups when the assumptions were violated. All analyses were made using the Graph Pad Prism 6.0 software. Two-sided P values < 0.05 were considered statistically significant.

Statistical analyses were done using R 3.4.1. A stepwise variable selection method based on Akaike information criterion was used to select the optimal logistic regression model. Area under receiver-operating characteristic curve (AUC) was used to evaluate overall performance of the logistic regression model (36). Wilcoxon-rank sum test was used to compare Gleason scores of two groups.

RNA-sequencing accession numbers are GSE112298 and GSE112300.

A finer analysis for potential BRD4 binding sites revealed three specific peaks in the HOXB13 upstream region in ChIP-sequencing data in VCaP cells (19). Of these, two of the major BRD4 binding peaks located at the 268th nucleotide position (BRAH1-BRD4 recruitment At HOXB13 1) and at the 799-nucleotide position (BRAH2) upstream from the HOXB13 transcription start site (TSS) were abolished following treatment with the prototype BRD4 inhibitor JQ1 (Fig. 1A). Other BET bromodomain proteins such as BRD2 were recruited to the HOXB13 gene suggesting a certain degree of redundancy (Supplementary Fig. S1A). We recently reported the development of novel small-molecule inhibitors (MA4-022-1, MA4-022-2, and SG3-179) that have the dual ability to inhibit the BET bromodomain proteins and a subset of kinases of the JAK family (refs. 23, 37; Supplementary Fig. S2A). To validate HOXB13 as a bona fide epigenetic target of BRD4, chromatin extracts prepared from the metastatic prostate cancer cell line C4-2B were treated with vehicle (DMSO), JQ1 (BET inhibitor), MA4-022-2 (BET-kinase inhibitor; ref. 23), the AR antagonists ENZ, or abiraterone acetate (ABR; ref. 38). ChIP-qPCR was performed with anti-BRD4 or IgG control antibodies for BRAH1, BRAH2, and IGX1A (control) sites. BRD4 binds both BRAH1 and BRAH2 sites but not the control site IGX1A, which was abolished in JQ1- and MA4-022-2 (MA4-2)–treated cells (Fig. 1B–D). Consistently, BRD4 was recruited to the BRAH1/2 sites in another metastatic CRPC model cell line, 22Rv1, but not at the IGX1A control site (Fig. 1E–G). Further, analysis of ChIP-sequencing data revealed binding of BRD4 was reduced in DHT-stimulated VCaP cells treated with ENZ and to a lesser extent with bicalutamide (ref. 38; Supplementary Fig. S1A).

To evaluate whether AR has a role in BRD4 recruitment at the HOXB13 genomic locus, we first analyzed ChIP-sequencing data (GSE55064) and confirmed AR recruitment to the BRAH1/BRAH2 sites (Supplementary Fig. S1B). Next, we validated AR recruitment by ChIP-qPCR analysis in C4-2B cells and 22Rv1 cells (Fig. 1E; Supplementary Fig. S1C). Consistently, a decrease in BRD4 recruitment at the BRAH1/2 sites was diminished following AR blockade therapies, but not completely abolished in the CRPC model C4-2B (Fig. 1B and C). In contrast, BRD4 was not recruited to the FOXA1 enhancer binding site previously reported to restrict HOXB13 expression to the prostatic lineage (ref. 39; Supplementary Fig. S1D).

As acetylation of histone H3 and H4 at lysine residues is known to regulate the assembly of the transcriptional coactivator complex at enhancer sites (40), we performed ChIP-qPCR with histone H3K27, H4K5/K8/K12, and H4K16 acetylation-specific antibodies to examine the epigenetic landscape at the HOXB13 gene locus (Fig. 1E–I; Supplementary Fig. S1E–S1G). We observed a significant enrichment of the histone H3K27 acetylation marks, characteristic of promoter/enhancers of transcriptionally active genes as well as H4K12ac marks at the BRAH1/BRAH2 enhancer but not at the IGX1A site (Fig. 1E–I). Intriguingly, the H4K12ac and to some extent H4K8ac, but not the H4K5ac, marks were enhanced in ENZ- as well as in JQ1-treated CRPC cells at the BRAH1/BRAH2 sites, but not at the transcriptionally silent site (IGX1A; Supplementary Fig. S1E–S1G). Further, the HOXB13 enhancer region and TSS were enriched for RNA Pol II and Transcription Initiation Factor IID subunit 1 (TAF1) consistent with an actively expressed gene (Fig. 1E–F and I). These results suggest that the HOXB13 gene is in a transcriptionally permissive state despite antiandrogen treatment.

To ascertain the molecular targets of the novel small-molecule BET-kinase inhibitors in CRPCs, we performed affinity proteomics (23, 37). Chemical-affinity proteomic analysis confirmed BRD4 as one of the top binding proteins to the MA4-022-1 beads (Supplementary Fig. S2A). In addition, because the scaffold for the MA/SG compounds is based on the diaminopyrimidine scaffold, kinases such as NEK9 showed significant binding to the beads in the C4-2B cell extract (Supplementary Fig. S2A). However, RNA interference experiments revealed that knockdown of NEK9 had minimal impact on the growth of prostate cancer cells. In contrast to the short half-life of JQ1, making it unsuitable for clinical purpose, the MA4-022-1 compound and analogues developed by our group are potent, highly soluble, and stable in aqueous media (23). To compare the sensitivity of CRPCs of the novel BET-kinase inhibitors versus ENZ, we analyzed the antiproliferative activity of the prototype BET inhibitor JQ1, the three novel BET-kinase inhibitors, ENZ and a control compound, MA3-068-1 (23). AR-expressing CRPC cell lines [C4-2B, C4-2B ENZR (ENZ-resistant derivative of C4-2B; ref. 2), 22Rv1], the androgen-responsive CRPC cell line VCaP, a normal prostate epithelial cell line, RWPE-1 as well as the AR-negative cell lines DU145 and PC3 were evaluated (Supplementary Fig. S2C–S2F and Supplementary Fig. S3A–S3C). We observed that compared with ENZ, the three BET-kinase inhibitors displayed potent antiproliferative activity at submicromolar concentrations against C4-2B, C4-2B ENZR, VCaP, and the metastatic CRPC cell line 22Rv1 (Supplementary Fig. S2C–S2F). Importantly, 22Rv1, which is highly resistant to the antiandrogen ENZ (IC₅₀ 32 μmol/L), retained high sensitivity to the novel BET-kinase inhibitors (IC₅₀ 1–2 μmol/L), suggesting that these inhibitors can be used to overcome antiandrogen resistance (Supplementary Fig. S2D).

Subsequently, we analyzed HOXB13 mRNA levels in C4-2B cells depleted of BRD4 with specific siRNA or control siRNA (Fig. 2A). A specific decrease in HOXB13 mRNA levels was observed in BRD4 and HOXB13 knockdown cells but not in the control siRNA- or AR siRNA-transfected cells. However, HOXB13 depletion significantly decreased AR levels consistent with earlier reports (1). Importantly, an increase in AR and HOXB13, but not in BRD4, was observed following treatment with the antiandrogen ENZ (Fig. 2A). Decrease in HOXB13 had no effect on BRD4 expression in the absence of ENZ with some effects in its presence (Fig. 2A).

To validate HOXB13 as a therapeutically relevant target of the novel BET-kinase inhibitors, we performed two independent RNA-sequencing analysis of C4-2B cells treated with MA4-022-1 and analogues (Supplementary Fig. S4A and S4B). A reduction in HOXB13 expression was observed in MA4-022-1–treated cells (Supplementary Fig. S4A) and in C4-2B cells treated with JQ1 or BET-kinase inhibitors in a repeat RNA-sequencing analysis (Supplementary Fig. S4B).

Quantitative reverse transcriptase PCR (qRT-PCR) in CRPC lines revealed that HOXB13 mRNA expression was suppressed following treatment with JQ1, MA4-022-1, and analogues (Fig. 2B–D). In contrast, FOXA1 mRNA expression was not affected by the BET inhibitors (Fig. 2B). As a control, we analyzed the expression levels of PSA and c-MYC and shown to be affected by BET inhibition and observed them to be suppressed as reported (ref. 19; Fig. 2C and D; Supplementary Fig. S4C). Notably, HOXB13 expression was also suppressed in LNCaP cells grown in charcoal-stripped media following treatment with BET inhibitors independent of DHT stimulation (Fig. 2E). Treatment of prostate cancer cells with GSK-126, an inhibitor of the catalytic subunit of the polycomb repressive complex 2 (PRC2), did not decrease HOXB13 mRNA expression (Fig. 2E and F) or negatively affect C4-2B cell proliferation (Supplementary Fig. S5A and S5B). Thus, HOXB13 gene activation is likely dependent on BRD4-mediated epigenetic regulation. To determine whether inhibition of BET bromodomain protein affects the expression of other HOXB family members in cis or trans, we examined the expression of HOX genes in cis on chromosome 17 as well as HOXA13, HOXC13, and HOXD13 located in trans. We did not detect the expression of HOXB1 to HOXB5 and HOXB8 in VCaP cells. A marginal effect on HOXB6 and HOXC13 was observed. In contrast, HOXA13 expression was significantly upregulated while HOXB7 and HOXD13 were downregulated following BET inhibition (Supplementary Fig. S4F). As HOXB13 was found to be expressed in the BT474 breast cancer cell line and associated with tamoxifen resistance (41), we examined the effect of BET-kinase inhibitors on HOXB13 mRNA levels and found it to be suppressed (Supplementary Fig. S4G). Collectively, our results indicate the epigenetic activation of HOXB13 by bromodomain proteins may be a conserved regulatory mechanism in cancers of different tissue specificities.

To further confirm BRD4-mediated control of HOXB13 protein expression, we silenced BRD2-4 and BRDT expression with selective silencing RNAs (siRNA; Fig. 3A). A significant reduction in HOXB13 protein expression was observed when prostate cancer cells were transfected with BRD4 and BRDT siRNAs (Fig. 3A). Analysis of the kinetics of HOXB13 expression revealed that the prototype BET inhibitor JQ1 as well as the novel BET-kinase inhibitors MA4-022-2 or SG3-179 caused a rapid reduction in HOXB13 protein expression (15 hours in MA4-022-2–treated cells vs. 48 hours in JQ1-treated VCaP cells; Fig. 3B). Moreover, a rapid induction of apoptosis was observed as seen by c-PARP cleavage that was concomitant with HOXB13 downregulation (Fig. 3B, third panel) and to a much lesser extent in ENZ (Fig. 3C). c-MYC was observed to be downregulated with JQ1 as well as by MA4-022-2 and SG3-179, consistent with earlier reports (Fig. 3B). Immunoblotting also revealed a significant reduction in HOXB13 protein expression with various BET inhibitors in VCaP, C4-2B, and E8 (a mouse prostate cancer model), but was not affected in cells treated with ENZ (Fig. 3C–E).

To determine whether BRD4 loss phenocopies the effect of HOXB13 depletion, we performed siRNA-mediated knockdown of BRD4 or HOXB13 or NEK9. Loss of BRD4 or HOXB13 but not NEK9 or control siRNA led to a significant decrease in cell proliferation (Fig. 3F). We performed a rescue experiment with exogenously expressed wild-type HOXB13 that is not under the epigenetic control of BRD4. HOXB13-WT could significantly rescue the cytotoxicity induced by the treatment of C4-2B cells with JQ1 at 0.5 μmol/L concentration (P < 0.001, Fig. 3G). A complete lack of rescue could be attributed to the low transfection efficiency characteristic of prostate cancer cells. Moreover, we observed that migration of 22Rv1 was significantly affected following treatment with MA4-022-2 and to a lesser extent with JQ1 but not ENZ or ABR (Fig. 3H and I). Collectively, these results suggest that HOXB13 is a critical transcriptional target of BRD4 in CRPCs.

To delineate whether HOXB13 function is indeed essential in CRPC progression and promoting antiandrogen resistance, we generated isogenic WT (parental) and HOXB13 pKO cell lines (partial knockout of HOXB13) in the C4-2B genetic background using the CRISPR/Cas9 gene-editing technology, as complete deletion was lethal (Fig. 4A). We observed that haploinsufficiency of HOXB13 altered the sensitivity of C4-2B cells to BET inhibitors by 2-fold and ENZ sensitivity by 3-fold (Supplementary Fig. S5A). To further examine whether reduced HOXB13 expression mitigates CRPC xenograft tumor growth in vivo, we injected the isogenic C4-2B and C4-2B HOXB13pKO cells subcutaneously in intact and castrated male SCID mice. Compared with the isogenic parental cells, the HOXB13pKO C4-2B cells were significantly impaired in their ability to form tumors in intact and particularly in castrated male mice (Fig. 4B and C). In contrast to other HOX13 paralogs, HOXD13 levels were significantly increased following the reduction in HOXB13 expression in residual HOXB13 pKO tumors harvested from castrated mice (Supplementary Fig. S5B). In addition, we also observed an increase in AR mRNA and c-MYC levels, which was exacerbated following castration (Fig. 4D). In addition to the C4-2B model, we also assessed the effect of genetic ablation of HOXB13 in VCaP cells. Genetic reduction of HOXB13 abrogated VCaP xenograft tumor growth, suggesting an extreme transcription factor dependency (Fig. 4E and F). This result demonstrates that not only does HOXB13 collaborate with androgen-dependent AR signaling to establish tumors in intact male mice, but it can also compensate for the lack of androgen to promote CRPC growth.

To determine whether the BET inhibitors phenocopy the genetic ablation of HOXB13 to affect growth of CRPC tumors, we implanted castrated male immunocompromised mice subcutaneously with C4-2B or intact mice with VCaP cells, which were castrated following tumor initiation (Fig. 4G; Supplementary Fig. S5C). Mice were treated with vehicle or the experimental compound for 5 days a week for 2 weeks (C4-2B CRPC xenograft tumors) or 4 weeks (VCaP CRPC xenograft tumors) at 50 mg/kg body weight. A significant inhibition of C4-2B growth was observed with JQ1 and/or MA4-022-2 (Fig. 4G and H) and to a lesser extent in VCaP (Supplementary Fig. S5C). Histologic examination of the liver, kidney, and brain did not reveal any gross morphologic abnormalities and the weights remained constant (Supplementary Fig. S5D and S5E), suggesting that the novel compound MA4-022-2 or its derivatives could be potential therapeutics for CRPCs for preclinical development.

To evaluate the therapeutic potential of BET inhibitors in preventing the dissemination of HOXB13-positive prostate cancers from the primary site, we analyzed for the presence of CTCs. To obtain prostate-specific CTCs, we collected blood from castrated male SCID mice harboring VCaP xenograft tumors that were either treated with vehicle (DMSO) or JQ1 (n = 5 in each group) for 4 weeks. In contrast to the control (DMSO) group, there was a significant drop in the number of CTCs, as well as HOXB13-positive CTCs with the BET inhibitor treatment (677 in the DMSO group vs. 63 in the JQ1 group; Supplementary Fig. S5F).

We performed integrative bioinformatics analysis to identify the targets that are potentially regulated by HOXB13. As a first step to filter nonspecific candidate genes, we isolated DEGs that were commonly affected by the three novel closely related BET-kinase inhibitors. Integration of RNA-sequencing data sets for the three analogues revealed 195 genes (Fig. 5A). Subsequently, we performed a second round of integrative analysis by comparing the 195 DEGs against the DEGs isolated from the HOXB13pKO/parental C4-2B (DMSO; 2012) list of DEGs, using a stringent P_adj values cutoff = 0.001 to identify HOXB13-specific targets that are responsive to the BET inhibitors (Fig. 5B). This analysis revealed 124 differentially expressed HOXB13 target genes that were also significantly affected by BET inhibition (Fig. 5B and C). Gene Ontology analysis of these HOXB13 target genes (HOXTAR124) revealed an overrepresentation of genes involved in chromatin structure and organization, cell-cycle progression, DNA replication, and repair pathways (Fig. 5D). Consistently, cell-cycle analysis revealed that the BET-kinase inhibitor caused a G₁–S inhibition similar to JQ1 at low concentrations and arrested cells at G₂–M at higher concentrations in all three CRPC models (Fig. 5E–G). In contrast, ENZ had no impact on cell-cycle progression of CRPCs and was similar to the vehicle control.

Compound mutant mice (NKX3.1^CreERT2/+/PTEN^flox/p53^flox; NPp53) with luminal prostate epithelial specific genetic deletion of PTEN and p53 tumor suppressor develop CRPC that harbors molecular features of metastatic human CRPCs (42). To determine whether HOXB13-regulated proproliferative function is conserved in multiple organisms, we analyzed HOXB13 target gene expression in public gene-expression data set of mCRPC tumors derived from the NPp53 mice (GSE92721; Supplementary table S4). Gene set enrichment analysis (GSEA) revealed a positive enrichment for ∼109 of the 124 HOXB13 target genes (HOXTAR109) originally identified (Fig. 6A–C), specifically in tumors derived from NPp53 compared with NKX3.1^CreERT2/+ mice (N; NES = 1.644, P = 0.0; Fig. 6B) or in mice with PTEN deletion (NP; NES = 2.064, P = 0.0; Fig. 6C). In contrast, a significant enrichment was not observed in NP versus N mice (NES = 1.215, P = 0.156; Fig. 6A). Notably, analysis of HOXB13 target gene expression from NPp53CRPC mice treated with abiraterone, a subset referred to as exceptional nonresponders due to their propensity to develop accelerated tumor phenotypes and metastasis, revealed a significant enrichment (NES = 2.115, P = 0.0), suggesting that activation of the HOXB13 pathway correlates with severity of disease progression (Supplementary Fig. S5G; Supplementary Table S5).

To further ascertain the clinical significance of HOXTAR124, we analyzed their expression in microarray data obtained from deidentified 194 primary and 29 metastatic tumors from the Moffitt Cancer Center Tissue Core (TCC) Data Set (Fig. 6D–F; Supplementary Tables S6–S8). Heat maps were generated for 124 unique genes (Fig. 6D), which revealed that a 10-gene core set (HOXB13 target genes responsive to BET inhibitors) HOTBIN10 could effectively distinguish primary from metastatic prostate tumors by first component of PCA at 86.3% (Fig. 6E). Violin plots also revealed a clear stratification of HOTBIN10 in primary versus metastatic tumors with lower expression segregating with primary and higher expression with metastatic prostate cancers (Fig. 6F). The HOTBIN10 gene sets were independently validated for their ability to separate primary from metastatic tumors by first component of PCA at 70.4% in the GSE21034 data set (ref. 35; Fig. 6G–I; Supplementary Table S9).

To confirm the association of HOTBIN10 with prostate cancer progression, we analyzed expression in the different mouse cohorts that represent different stages of progression. The subset of the NPp53CRPCs treated with abiraterone and described as exceptional nonresponders due to their distinct pathologic features showed an elevated HOTBIN10 gene expression (Supplementary Fig. S5H). Combined, these results indicate HOTBIN10 genes as a hallmark of lethal metastatic CRPCs.

To determine whether HOTBIN10 gene expression can be modulated by genetic or pharmacologic ablation of HOXB13, we silenced HOXB13 expression by siRNA in CRPCs (C4-2B or VCaP) or treated cells with the BET-kinase inhibitor MA4-022-2 and found a corresponding decrease in the expression of HOTBIN10 genes (Fig. 7A and B). Importantly, HOTBIN10 gene expression remained significantly high in C4-2B despite ENZ treatment compared with LNCaP (Fig. 7C). Importantly, c-MYC was not affected by the loss of HOXB13 (Fig. 7A). Search Tool for the Retrieval of Interacting Genes (STRING) analysis revealed that the genes comprising HOTBIN10 may form an evolutionarily conserved protein–protein interaction network (Fig. 7D). Ablation of BRD4, HOXB13, and AR using siRNAs revealed that HOTBIN10 gene expression was not affected following control and AR silencing in DMSO (vehicle)- or ENZ-treated cells (Fig. 7E and F). Consistently, knockdown of BRD4 or HOXB13 suppressed expression of the HOTBIN10 genes (Fig. 7E and F).

Given the association of HOTBIN10 gene expression with AR independence, we built gene-expression prediction models to stratify patients into those with primary versus metastatic prostate cancer. Each prediction model was trained on the basis of 150 patients in the MSKCC validation data set and came out with a threshold score (AUC 0.711; Supplementary Fig. S6A). The optimal models and corresponding thresholds were then tested on Moffitt TCC data. Analysis of the HOTBIN10 gene set revealed an overall accuracy of 95.4% and AUC of 0.99 in classifying primary and metastatic prostate cancer in the Moffitt TCC data set (Supplementary Fig. S6B). A recent study of 27 cases of mCRPC patients treated with first-line abiraterone revealed that a subset of patients (6/6) with HOXB13-positive expression in CTCs died within a year (18). To further examine the significance of HOTBIN10 in CRPC dissemination, we examined the expression of HOTBIN10 in single-cell RNA-sequencing data from these patients (GSE67980; Supplementary Table S10). We observed significant clonal heterogeneity in the single CTCs based on their HOXB13 and HOTBIN10 gene-expression profiles (Supplementary Fig. S6C). Of these, the expression of two genes, AURKB and MELK, correlated with HOXB13, suggesting that AURKB inhibitors could be used to target high-risk HOXB13-positive metastatic prostate cancers (Supplementary Fig. S6D and S6E). We therefore tested the activity of AURKB inhibitor barasertib currently in phase I/II clinical trials for lymphoid and leukemia. Although barasertib showed significant antiproliferative activity in CRPC models, the BET-kinase inhibitors performed better in comparison (Supplementary Fig. S7A–S7C). We observed that VCaP cells treated with ENZ had increased expression of AURKB, suggesting a novel mechanism of resistance (Supplementary Fig. S7D). Combined, our studies not only uncovered a role for the BRD4–HOXB13–AURKB axis in antiandrogen resistance of CRPCs but small-molecule inhibitors to target this regulatory circuit (Supplementary Fig. S7E).

Prostate cancer remains a leading cause of cancer-related deaths among men worldwide. The prognosis for patients diagnosed with metastatic CRPC is bleak as the disease ceases to respond to the current first- and second-generation antiandrogen therapies, indicating that the cancer cells rewire their transcription programs to survive in the absence of their fuel, androgen. Importantly, the molecular mechanisms by which CRPCs evade androgen deprivation therapies (ADT) are varied, suggesting that identifying and targeting AR-dependent and -independent pathways are critical to improve patient outcomes. In this report, we demonstrate that HOXB13 gene expression is epigenetically regulated by the BET family of bromodomain proteins that is largely androgen independent. The BRD4–HOXB13 nexus in turn promotes an androgen-independent transcriptional network of genes implicated in cell-cycle, DNA-damage response, nucleotide metabolism, and metastasis. Thus, a nexus of transcription factor (HOXB13) and an epigenetic regulator (BRD4) encoded by two developmental genes that are reexpressed in CRPCs weans cells of androgen requirement and drives new dependency.

In silico analysis revealed that in addition to BRD4, peaks for BRD2 and BRD3 were detected at the HOXB13 enhancer region in VCaP cells, suggesting potential recruitment (Supplementary Fig. S1A). Moreover, we found that silencing of individual BRD members with gene-specific siRNAs (BRD 2, 3, or 4) reduced HOXB13 expression to a significant extent but not completely indicating potential redundancy in functionality. Similarly, BRD2, 3, and 4 have been identified as regulators either of AR target gene expression or of AR splicing in CRPCs, underscoring the importance of targeting them to overcome antiandrogen resistance (19, 20). Importantly, the novel bromodomain-kinase inhibitors due to their pan-BRD inhibitory activity should be very useful to inhibit multiple BRDs that may compensate for each other to thwart the development of resistance.

The molecular role of HOXB13 in prostate cancer has been uncertain. HOXB13 was initially reported to prevent transactivation of hormone-regulated AR target genes, through its interaction and sequestration of AR from binding its cognate responsive elements (8, 9). Subsequently, Kim and colleagues have also reported that HOXB13 is overexpressed in hormone-refractory tumors and promotes androgen-independent growth of LNCaP cells by regulating the RB-E2F signaling to inhibit the tumor suppressor p21Waf and promote proliferation (11). In addition to full-length AR, HOXB13 has been shown to interact with the AR-V7, a splice variant that lacks the ligand binding domain and promotes the establishment of AR cistrome in CRPCs (12). The function of HOXB13 in prostate cancer may also be regulated via its interaction with the MEIS proteins. Lower expression of these potential tumor suppressors MEIS1 and 2 is associated with increased likelihood of metastasis (43). We observed that CRPCs deploy HOXB13 to upregulate a proproliferative BRD4-HOXB13–dependent transcriptional network in response to ADT. Key members of the proproliferative network that we identified in this study include a core HOXB13 effector 10 gene set (HOTBIN10-HOXB13 target genes response to BET inhibitors) that can remarkably stratify human prostate tumors into the normal, primary, and metastatic groups and has been validated in independent data sets (∼400 biopsies). In addition, our study reveals that the HOXB13 transcriptional network was insensitive to AR depletion or blockade but responsive to the inhibition of the BRD4–HOXB13 epigenetic axis. Importantly, identification of BRD4 as an epigenetic regulator of HOXB13 even under conditions of androgen deprivation uncovers a novel mode of targeting and inhibiting HOXB13 expression in prostate cancers. Our studies therefore unravel for the first time not only an epigenetic mechanism underlying HOXB13 gene regulation in prostate cancer, but also the therapeutics to target the HOXB13 pathway, a promoter of CRPC growth.

N.J. Lawrence, E. Schonbrunn, and H.R. Lawrence are named inventors of a patent (assigned to the Moffitt Cancer Center) describing the dual activity bromodomain-kinase inhibitors.

Conception and design: M. Ayaz, J. Puskas, U. Rix, R. Perera, N.J. Lawrence, E. Schonbrunn, K. Mahajan

Development of methodology: N. Nerlakanti, D.T. Nguyen, A.K. Patel, M. Ayaz, N. Agarwal, U. Rix, N.J. Lawrence, K. Mahajan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N. Nerlakanti, D.T. Nguyen, M. Ayaz, B.M. Kuenzi, R.M. Karim, N. Berndt, J. Puskas, D. Coppola, J. Zhang, S. Shymalagovindarajan, U. Rix, E. Schonbrunn, K. Mahajan

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N. Nerlakanti, J. Yao, D.T. Nguyen, A.K. Patel, A.M. Eroshkin, H.R. Lawrence, M. Ayaz, B.M. Kuenzi, Y. Chen, A.M. Magliocco, D. Coppola, J. Zhang, Y. Kim, E. Schonbrunn, K. Mahajan

Writing, review, and/or revision of the manuscript: A.K. Patel, A.M. Eroshkin, M. Ayaz, A.M. Magliocco, D. Coppola, J. Dhillon, J. Zhang, U. Rix, Y. Kim, R. Perera, N.J. Lawrence, K. Mahajan

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N. Nerlakanti, D.T. Nguyen, A.K. Patel, M. Ayaz, R.M. Karim, A.M. Magliocco, U. Rix, Y. Kim, N.J. Lawrence, K. Mahajan

Study supervision: Y. Kim, R. Perera, N.J. Lawrence, K. Mahajan

Other (synthesis of MA4-022-1, MA4-022-2, and SG3-179 molecules): S. Gunawan

Other (performed assays for the study): S. Shymalagovindarajan

We thank Devon DeLoach for assistance with animal studies, Zachary Thompson for statistical assistance, Dr. Vonetta Williams and Margaret Penichet for prostate cancer data curation. We thank Drs. Srikumar Chellappan, Jose Conejo-Garcia and Nupam Mahajan for suggestions. This work was supported in part by the Cancer Center Support Grant P30 CA076292 from NCI, by the NIH/NCI R01CA181746 (U. Rix) and NIH/NCI F99/K00 CA212456 (B. M. Kuenzi), by SBPM Discovery Institute Bioinformatics Core and SBP CCSG P30 CA030199 (A.M. Eroshkin), by Florida Department of Health, Bankhead-Coley Cancer Research Program 5BC08 (R. Perera), by R50CA211447 (H.R. Lawrence), by Moffitt Team Science Award (E. Schonbrunn and N.J. Lawrence) and Department of Defense Grants (W81XWH-14-1-0251 and W81XWH-15-1-0059), and Moffitt and Washington University in St. Louis Startup Funds to K. Mahajan.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.