Acetylated HOXB13 Regulated Super Enhancer Genes Define Therapeutic Vulnerabilities of Castration-Resistant Prostate Cancer

Prostate cancer afflicts ∼1.3 million men worldwide and is a leading cause of global cancer-related deaths (1). Patients diagnosed with hormone-sensitive prostate cancer who undergo treatment with androgen deprivation therapy (ADT) invariably develop castration-resistant prostate cancers (CRPC; refs. 2–4). Following disruption of the primary androgen/androgen receptor (AR)-mediated regulatory axis, cancer cells activate bypass mechanisms for survival (5, 6). Notably, cancer cells that activate pioneer transcription factor (TF) regulated programs appear to not only tolerate hormone deprivation better but are also characterized by an increased metastatic potential (7–12). The mechanism by which developmental TFs direct prostatic luminal epithelial cells away from differentiation is currently unclear.

Epigenetic deregulation of developmentally regulated chromatin readers and pioneer transcription factors co-ordinates reactivation of quiescent gene expression programs to transform cells permanently into new subtypes (13–15). However, tissue-specific TFs may themselves function as epigenetic regulators during this cellular transformation. Identifying this de novo epigenetic activity of TF is critical to understand their role in overcoming the programming of normal cells for differentiation. Pioneer TFs, HOXB13 and FOXA1, regulate prostate differentiation through their interactions with the AR and enhancers, largely under the influence of testosterone (16–19). Particularly, the HOXB13 Glycine 84 to Glutamic acid (G84E) mutation is associated with an increased risk for prostate cancer (20). Moreover, increased expression of HOXB13 correlates with biochemical recurrence and metastatic progression after radical prostatectomy and promotes resistance to ADT (18, 21–23). Whether HOXB13 functions as an epigenetic regulator of prostate cellular transformation to the malignant CRPC state is unknown.

Acetylation of histone and nonhistone proteins is known to play a major role in eukaryotic transcriptional regulation (24, 25). Although enhancers and super enhancer (SE)-regulated transcriptional networks are emerging features in many tumor types (26, 27), how cells mark pathogenic SEs in the genome is unclear. In this study, we identified K13-acetylated (acK13)-HOXB13 mediated by the histone acetyl transferase CBP/p300 as a critical regulator of SE selectivity prior to CRPC development. We uncovered that acK13-HOXB13 synergizes with H3K27 acetylation at lineage-specific and tumor-promoting SEs of critical CRPC targets, thereby functioning as an epigenetic regulator of tumor growth. Our studies indicate prior treatment with a selective-kinase inhibitor (R)-9b may prevent the establishment of the lethal CRPC state by eliminating cells primed for castration resistance.

RWPE-1, LNCaP, VCaP, PC-3, 22Rv1, and HEK293T were obtained from ATCC. C4-2B and LAPC4 were cultured as described previously (28). All cell lines were used within 3 months or six to eight passages before being replenished from frozen stocks. The HOXB13K13A site-directed mutant C4-2B and 22Rv1 clones described in this study were generated by the Genome Engineering and iPSC Center (GEIC core) at the Washington University in St. Louis. Single-cell clones were verified for nucleotide substitution and SNPs by DNA sequencing. All cultures were tested for mycoplasma contamination every 2 months using the PCR Mycoplasma Test Kit I/C (PromoKine). Identities of all cell lines were confirmed by short tandem repeat (STR) profiling.

HA-tagged or GFP-tagged full-length HOXB13, HOXB13-K13A, HOXB13-K13R, and HOXB13-G84E constructs were generated by Genscript and verified by sequencing. Flag-tagged CBP, His-tagged p300, and additional Flag-tagged HATs (GCN5, TIP60) and GFP-TAF1 were purchased from Addgene.

Biotinylated modified and unmodified HOXB13 peptides were synthesized by Genscript. High-titer rabbit polyclonal antibodies directed against acK13-HOXB13 were generated by Genscript and affinity-purified. mAb targeting acetylated HOXB13 were generated by ProMab. acK13-HOXB13–specific antibodies were screened by ELISA against the modified and unmodified HOXB13-K13 peptides. Hybridoma supernatants from the 2B7C1 clone were purified using the Capturem Protein G Maxiprep Columns (Takara Bio).

Pan-HOXB13 (F-9) monoclonal, SCBT catalog no. sc-28333 (RRID: AB_627744); pan-HOXB13 (H-80) polyclonal, SCBT catalog no. sc-66923 (RRID: AB_2233136); pan-HOXB13 (D7N8O) CST catalog no. 90944 (RRID: AB_2734734); AR (F39.4.1), BioGenex catalog no. AM256 (RRID: AB_2687514); CBP (D6C5) CST, catalog no. 7389 (RRID: AB_2616020); p300 (F-4) SCBT catalog no. sc-48343 (RRID: AB_ 628075); CTCF Diagenode catalog no. C15410210 (RRID: AB_2753160); total H3 (96C10) CST catalog no. 3638 (RRID: AB_1642229); H3K27ac CST catalog no. 4353 (RRID- AB_10949503); H3K27ac Active Motif catalog no. 39133 (RRID: AB_2561016); total H4 (L64C1) CST, catalog no. 2935 (RRID: AB_1147658); H4 pan-acetyl; Active Motif, catalog no. 39925 (RRID: AB_2687872); HA (C29F4), CST catalog no. 3724 (RRID: AB_1549585); FLAG (M2), Sigma-Aldrich catalog no. F1804 (RRID: AB_262044); histidine (RM146), Sigma-Aldrich, catalog no. SAB5600227 (RRID: AB_2810125); β-actin (AC-74), Sigma-Aldrich catalog no. A2228 (RRID: AB_476697); BRD9, bethyl laboratories (A303-781A) RRID: AB_11218396; HRP-conjugated anti-mouse, Promega catalog no. W4021 (RRID: AB_430834); HRP-conjugated anti-rabbit, Promega catalog no. W4011 (RRID: AB_430833).

C4-2B cells were transfected with siRNAs using the Nucleofector Kit R from Lonza. Plasmid or siRNA transfection was performed on 22Rv1, VCaP, LAPC4, and HEK293T cells using the X-tremeGENE transfection reagent (Sigma). Transfected cells were harvested for analysis after 2 days. siRNAs and primer sequences are provided in Supplementary Data.

Co-immunoprecipitation was performed as described previously (28) or with Capturem beads from Takara Bio. Cells were sonicated in RLB buffer containing 250 to 500 mmol/L NaCl (28). Protein extracts (0.5–1 mg) were immunoprecipitated using an anti-HA affinity gel or 3 to 4 μg of antibody directed against pan- or acK13-HOXB13 coupled to protein A/G agarose beads. After overnight incubation, the beads were washed with RLB and 1× PBS, boiled in sample buffer, electrophoresed (SDS-PAGE), and immunoblotted with the respective antibodies.

ChIP-IT Express or ChIP-IT High Sensitivity Kit (Active Motif) were used for ChIP experiment. Fresh human prostate tissue specimens were minced while on dry ice and fixed immediately with 1% formaldehyde for 10 minutes. Cell lines were grown in charcoal stripped medium for 48 hours prior to crosslinking with 1% formaldehyde upon reaching 70% to 80% confluency. DNA was purified using a PCR Purification Kit (Qiagen) followed by analysis by qPCR or high-throughput sequencing. To identify the maximum number of HOXB13 acetylation binding sites from ChIP-sequencing, we used a HiSeq 3000 sequencing system which results in ∼350 million reads per lane at a 1 × 50 read length. The DNA sequencing data were processed using a pipeline that was generated for use in the Encyclopedia of DNA Elements (ENCODE) project.

Two human HOXB13 peptides (amino acids 1–25) were synthesized with K13 at the center as shown below. Both the peptides were biotinylated at N-terminus. Peptide sequences are as follows:

Unmodified HOXB13 (1–25): NH₂-MEPGNYATLDGAKDIEGLLGAGGGR-COOH

K13ac-HOXB13 (1–25): NH₂-MEPGNYATLDGAK(ac)DIEGLLGAGGGR-COOH

The biotinylated peptides were immobilized on streptavidin-coupled magnetic beads for 30 minutes at room temperature. After several washes, the peptide-bound beads were incubated with precleared cell lysates in IP-MS–compatible cell lysis buffer (Pierce, catalog no. 90409) at 4°C overnight. After extensive washing, the captured peptide/protein complexes were isolated via an on-bead digestion method, and nano-LC/MS-MS was performed. Competition experiments were performed with acetylated histone H4K5ac peptide; SGRG-Lys-Ac-GGKGLGKGGA.

Cells were seeded in 96-well plates containing 200 μL of growth medium per well. Eight replicates were included for each concentration. Four days after the addition of the drug, 25 μL of cell titer-Glo 2.0 (Promega) was added to each well, and luminescence was recorded after 15 minutes using a Synergy HTX multi-mode reader (BioTek). The IC₅₀ were calculated using GraphPad Prism version 8.4.3.

For colony-forming unit (CFU) assay, parental or HOXB13K13A mutant C4-2B or 22Rv1 single cells were plated in 6-well plates in three replicates and cultured for 2 weeks in C4-2B medium or for 22Rv1 in charcoal stripped medium, respectively. Cells were fixed with 4% crystal violet solution in 1% methanol and 4% paraformaldehyde solution. Macroscopic colonies visible were counted using ImageJ and the area of individual colonies was plotted using Graph Pad Prism version 8.4.3.

Parental and HOXB13K13A mutant cells (C4-2B or 22Rv1) were seeded in a 24-well plate at approximately 5,000 cells per well in 40 μL of growth factor–reduced and phenol red–free Matrigel medium (Corning CB-40234) and maintained in organoid culture medium. 3D-images were captured on an EVOS M5000 microscope (Invitrogen) and analyzed using ImageJ.

All animal studies were performed under approved Institutional Animal Care and Use Committee protocols at Washington University in St. Louis. Mice were randomized prior to the injection of the cells. Two million C4-2B cells (parental or HOXB13K13A mutants) in 200 μL of growth media in 50% Matrigel (Corning) were implanted subcutaneously into the dorsal flanks of 6-week-old male SCID mice (Charles River Laboratories; n = 6–8 mice per group). Tumor growth was monitored over a 10-week period. Tumor volumes were measured twice weekly using calipers. At the end of the study, all mice were euthanized humanely. Xenograft tumors were harvested, weighed, and photographed and obtained after euthanasia.

Prostate tissues (fresh or formalin fixed) were collected after obtaining informed written consent from prostate cancer patients and the studies were conducted in accordance with the recognized ethical guidelines pertaining to the Declaration of Helsinki after approval by the Institutional Review Board at the Washington University in St. Louis (IRB; HRPO #201411135; 202010061). The patients were men at 50 to 75 years of age. Each of the normal and tumor specimens (∼3 mm) were obtained fresh after radical prostatectomy, based on MRI-guided biopsy collection and pathology report. A board-certified GU pathologist reviewed ∼0.5 to 1 mm hematoxylin and eosin (H&E)-stained FFPE specimens from each case to confirm tumor cellularity. Supplementary Table S1 and Fig. S3 include clinical annotation for all the specimens used in this study.

Freshly isolated prostate tissues were placed in sterile 60 mm dishes, washed twice with PBS, and minced into nearly 0.1 to 0.5 mm pieces. Minced tissues were transferred into 2 mL of cell dissociation media containing collagenase (29) and incubated at 37°C for 45 minutes to 1 hour with continuous gentle rotation. The dissociated prostate tissues were centrifuged at 500 × g for 5 minutes at 4°C. The tissues were dissociated further with 5 mL of TrypLE enzymes (Gibco, no. 12605-028) with Rho-associated, coiled-coil containing protein kinase (ROCK) inhibitor (Y27632) for 20 minutes. Approximately 10 mL of ice-cold human prostate organoid culture media was added followed by centrifugation at 500 × g for 5 minutes at 4°C. The dissociated cell suspension was again filtered through a 40-μmol/L mesh filter. Cells were resuspended in 75% Matrigel plus organoid culture media and plated in a single 40 μL drop in the middle of a well in a 24-well plate to create a dome shape. The plate was inverted and incubated for 30 minutes at 37°C and 5% CO₂. Complete human prostate organoid culture medium (0.5 mL) was added after the Matrigel had solidified. Cultures were maintained until organoids (PDOs) reached 300 μm in size. Medium was replenished every 3 to 4 days to maintain the integrity of PDOs. Organoid images were captured using an EVOS M5000 microscope (Invitrogen) and analyzed using ImageJ or Adobe Photoshop.

Prostate organoids were grown in matrigel domes in a 24-well plate with 600 μL organoid culture medium and treated with vehicle or inhibitor. After 6 days, media was withdrawn, cells were recovered by incubating in 200 μL of cell recovery solution (Corning 354253) at 4°C on a rotator for 60 minutes. Fifty μL of CellTiter-Glo 2.0 (Promega) was added to each well and the contents were transferred to a 96-well microtiter plate. Luminescence was recorded after 15 minutes using a Synergy HTX multimode reader (BioTek). Percent cell viability was calculated by normalizing the readings obtained in response to each inhibitor concentration to the readings in vehicle (DMSO).

Single-cell suspensions from 6,750 cells were generated from prostate tissue organoids for library preparation as described in the 10× genomics single-cell suspension protocol. All RNA samples were quantitated twice on the bioanalyzer and checked to ensure integrity. Sequencing was performed at 25K depth per cell using the NovaSeq S4 flow cell. Cell Ranger software was used to align the reads to genome reference GRCh38. Loupe browser analysis was performed to generate t-SNE plots.

All data are presented as means ± SEM. Statistical analyses included unpaired Student t test to compare two groups; ANOVA was used for three or more group comparison by GraphPad Prism software. All P values <0.05 were considered to be statistically significant.

HOXB13-K13 acetylation antibodies, C4-2B and 22Rv1 HOXB13K13A mutants, reagents, and recombinant DNA constructs will be made available upon request to Dr. Kiran Mahajan (E-mail: [email protected]).

GEO accession numbers are GSE167506, GSE169134, GSE169132, and GSE169133. Human prostate ChIP-sequencing data are maintained in the Washington University server and will be provided upon request.

To investigate the mechanism by which HOXB13 promotes lethal prostate cancer, we affinity-purified HOXB13 from the metastatic CRPC cell line, C4-2B. Mass-spectrometry (MS) analysis revealed a novel acetylation site at an evolutionarily-conserved Lys 13 (K13) at the N-terminus of HOXB13 (Fig. 1A; Supplementary Fig. S1A), which is absent in other HOX13 paralogs. Secondary protein structure prediction revealed that this lysine 13 is located within a 6-residue helical region (Supplementary Fig. S1A; ref. 30). To evaluate the functional relevance of this modification in prostate cancer, we developed and characterized a high-affinity monoclonal (2B7C1) and polyclonal antibodies (Abs) against acetylated lysine 13. Both antibodies specifically recognize only the acetylated K13-HOXB13 but not the unmodified HOXB13 (Supplementary Fig. S1B). Subsequently, to identify the lysine acetyltransferase that promotes this specific modification, we co-expressed hemagglutinin (HA)-tagged HOXB13 in human embryonic kidney cells (HEK293T) together with known histone acetyltransferases (HAT; Supplementary Fig. S1C). Among these, two closely related and evolutionary conserved HATs, cyclic adenosine monophosphate-responsive element-binding protein (CREB)-binding protein (CBP) and adenovirus E1A-associated 300-kDa protein (p300), acetylated the wild-type HOXB13 but not the HOXB13 mutated at K13 to the nonmodifiable amino acids, alanine (K13A) or the arginine (K13R; Fig. 1B and C). In addition, the HOXB13-G84E prostate cancer risk variant was modified by p300 (Fig. 1B). Consistently, we detected an interaction of p300 with the wild-type but not with the recombinant HOXB13-K13A mutant (Supplementary Fig. S1D). Significantly, we detected neither the acK13-HOXB13 nor the total HOXB13 protein in the normal prostate cell line, RWPE-1 (Fig. 1D). As opposed to RWPE-1, both AR-positive cell lines (VCaP, C4-2B, 22Rv1), and negative cell line (PC3) showed acK13-HOXB13, which is sensitive to a p300/CBP inhibitor (A-485; Fig. 1D). Likewise, CBP-mediated nuclear localized wild-type HOXB13 shows loss of acetylation in response to A-485 treatment; in contrast, mutant HOXB13K13A is not acetylated (Fig. 1E). Consistently, C4-2B and VCaP showed a dose-dependent decrease in HOXB13-K13 acetylation following treatment with A-485 or GNE-049 (Supplementary Fig. S1E and S1F). Moreover, A-485–sensitive HOXB13 acetylation is less susceptible to deprivation of glucose or acetate than CBP/p300-mediated histone H3K27 acetylation (Supplementary Fig. S1G). To confirm CBP/p300 as the HAT, C4-2B, VCaP, and LAPC4 prostate cancer cell lines were transfected with either CBP or p300 silencing RNA alone or in combination. The ratio of acetylated to total HOXB13 is decreased when transfected with either CBP or p300 siRNA alone or in combination and is comparable with HOXB13 silencing (Fig. 1F–I; Supplementary Fig. S1H). A significant amount of acK13-HOXB13 expression is observed in human prostate cancers (Supplementary Fig. S1I). These results reveal that HOXB13 is a bona fide CBP/p300 substrate.

To examine the genome-wide distribution of acK13-HOXB13 in prostate cancer, we first validated the cognate antibodies for ChIP (Supplementary Fig. S2A–S2F). Chromatin extracts prepared from C4-2B cells were transfected with either control or two independent CBP siRNAs followed by immunoprecipitation with acK13-HOXB13, pan-HOXB13, or IgG antibody. We detected acK13-HOXB13 binding at selective targets in the control siRNAs but not at IGX1A (control) in the CBP siRNA-transfected cells (Supplementary Fig. S2A–S2F). Consistently, the expression of these targets was decreased following CBP silencing (Supplementary Fig. S2G). acK13-HOXB13 is also enriched at known CRPC targets in prostate tumors compared with normal validating the antibodies for ChIP (Supplementary Fig. S2H and S2I).

To understand the relevance of this novel HOXB13-K13 acetylation in human prostate cancer, we characterized its binding, expression, and function in well-annotated clinical specimens (n = 38; Supplementary Table S1; Supplementary Fig. S3A and S3B; Materials and Methods). As a first step, we performed ChIP sequencing of matched human prostate normal and primary tumor tissues taken after radical prostatectomy (n = 5: 2N and 3T; subset 1) with H3K27ac or acK13-HOXB13 antibodies (Fig. 2) or with pan-HOXB13, H3K27ac, acK13-HOXB13, or IgG antibodies (n = 4; 2N and 2T; subset 2; Supplementary Fig. S4). This analysis revealed an increased acK13-HOXB13 signal at the transcription start site (TSS) in tumor tissues compared with normal prostate and is consistent across multiple patient samples in both subsets (Fig. 2A; Supplementary Fig. S4A). In contrast, the H3K27ac signal enrichment at peaks varied and were either higher or lower at TSSs in normal tissue compared with tumor depending on the subset (Fig. 2B; Supplementary Fig. S4A). A heatmap depicting acK13-HOXB13 and H3K27ac peaks within 1 kb of the TSSs for the first subset is shown in Fig. 2C and D. Compared with the normal prostate, we observed a variation in the number of promoters with acK13-HOXB13 signals in tumors (27.6% tumor vs. 36.06% normal in subset 1 and 19% tumor vs. 16% normal in subset 2). However, there is consistent enrichment of acK13-HOXB13 signal within the transcribed portion of the gene in tumors, specifically at introns (24%–39%) and at distal intergenic regions (22%–39%) across both subsets. H3K27ac-enriched promoters in prostate tumors ranged from 14% to 18% in normal to 9% to 35% in tumors. In all cases, the distal intergenic regions had the highest proportion of pan-HOXB13 or acK13-HOXB13 signals at 39% to 45% for acK13-HOXB13 or pan-HOXB13 antibody, respectively. To determine whether tumor-specific sites harbor increased acetylated HOXB13 signals compared with pan HOXB13 or H3K27ac, we analyzed the ChIP sequencing data from subset 2. Our analysis revealed several thousand peaks for each antibody ChIP (Supplementary Fig. S4A). A majority of the peaks were unique for each antibody. Percentage (4.6%) of acK13-HOXB13 peaks in normal were retained in the tumor, whereas 16% of the acK13-HOXB13 peaks in tumors overlapped with the normal (Supplementary Fig. S4B). In contrast, 1.76% of pan-HOXB13 peaks in normal were retained in the tumor, whereas 0.95% of the pan-HOXB13 in tumor overlapped with normal. Only 2.7% of the H3K27ac tumor peaks overlapped with the normal (Supplementary Fig. S4B). These results suggest an increased occupancy of acetylated HOXB13 at the tumor-specific sites even in the normal prostate.

As acK13-HOXB13 signals were higher than those expected at a typical enhancer, we determined whether the acK13-HOXB13 signal is differentially enriched at SEs (27, 31). We stratified SEs according to the strength of acK13-HOXB13 or pan-HOXB13 or H3K27ac signals (Fig. 2E–H; Supplementary Fig. S4C and S4D). Even in the normal prostates, we detected increased acK13-HOXB13 signals at specific SEs (Fig. 2E) and were ∼2-fold higher than pan-HOXB13 (Supplementary Fig. S4D). In contrast, H3K27ac signals were same or 2-fold more intense at SEs in normal prostate than in tumor (Fig. 2F and H). In some cases, acK13-HOXB13 was further increased up to 3-fold at SE peaks in tumors, tracked with H3K27ac and higher than pan-HOXB13 (Fig. 2G; Supplementary Fig. S4D). Subsequently, we analyzed ChIP-sequencing data to identify acK13-HOXB13–target genes regulated by SEs in tumors (Supplementary Tables S2A and S2B). We identified 8% of genes overlapped in pan-HOXB13 versus acK13-HOXB13 in normal and tumor-enriched SEs; this cohort included HOXB13 as well as ACK1/TNK2, SPON2, KLK3, BMPR1B, and AURKB (Fig. 2I–K; Supplementary Fig. S4E–S4H). ChIP-qPCR analysis confirmed acK13-HOXB13 enrichment at ACK1 and FOLH1 in primary prostate tumors compared with normal. Gene set enrichment analysis (GSEA) revealed that in contrast to the normal prostate, which showed enrichment for both basal and luminal subtypes, the luminal epithelial subtype dominated in tumors. Importantly, we also observed a negative enrichment of tumor-specific SE-proximal genes in normal prostate (normal: NES = −1.563, P = 0.007 vs. tumor: NES = 2.695, P = 0.000; Fig. 2L and M). Our analysis of TF motifs at acK13-HOXB13 binding sites revealed enrichment of SMAD2 and MAFA binding motifs in normal prostate (Fig. 2N). In contrast, the most common TF motifs at acK13-HOXB13 binding sites in prostate tumors were those of factors associated with epithelial–mesenchymal transition (Fig. 2O).

Subsequently, we analyzed the expression of acK13-HOXB13 regulated SE targets (HOXB13, AR, ACK1, FOLH1, VEGFA, and SPON2) in matched human normal and prostate tumors (n = 38; total: 19N and 19T); acetylated HOXB13 signals were significantly upregulated in tumors compared with normal (Fig. 3A). To investigate the role of acK13-HOXB13–regulated SE-linked genes in prostate cancer pathogenesis, we focused on a representative CRPC target, ACK1 tyrosine kinase, which had acquired a tumor-specific SE (SE rank 677 and 548 in RP #1 and rank 146 in RP #2; Fig. 3B; Supplementary Fig. S4E), correlating with increased recruitment of acetylated HOXB13 (Supplementary Fig. S4E) and expression of ACK1 mRNA and protein (Fig. 3A and C). We detected increased phosphorylated ACK1 (pY284-ACK1) associated with kinase activation HOXB13 and AR expression in tumors compared with normal and benign prostatic hyperplasia (BPH; Fig. 3D).

To corroborate the role of HOXB13-K13 acetylation in prostate cancer, we first examined its role in C4-2B followed by validation in additional prostate cancer lines (VCaP and 22Rv1). ChIP followed by deep sequencing of androgen-deprived C4-2B revealed a significant enrichment of acK13-HOXB13 at ∼46,684 sites comprising distal intergenic regions and promoters across the genome (Fig. 4A). Approximately 85% of the acK13-HOXB13 binding sites were located within ∼2 kb of (TSS) and another 10% were identified at distal enhancers (Fig. 4B). The acK13-HOXB13 binding motif is enriched for the GTAAACA sequence that differs from unmodified HOXB13 motifs CCAATAAA and CTCGTAAA (Fig. 4C; ref. 32). Other most frequently detected DNA motifs enriched include Zinc finger protein and X-linked factor (ZFX), and AR element (ARE)-half sites (Fig. 4C; Supplementary Table S3).

We uncovered broad acK13-HOXB13 peaks in the vicinity of HOXB13, as well as several lineage and prostate tissue-specific genes NKX3-1, PMEPA1, SLC45A3, PSCA, KLKs, oncogenes MYC, BCL6, mitotic kinase AURKB, AR, and ACK1/TNK2 and novel targets FOLH1, RXRA, NFIX, and ZBTB16 (Fig. 4D; Supplementary Fig. S5A). At proximal TSSs, acK13-HOXB13 enrichment signals overlapped with H3K27ac, H3K4me2/3, and RNA Pol II binding peaks (Supplementary Fig. S5A). At distal TSS sites, the acK13-HOXB13 peaks overlapped with H3K27ac and H3K4me2 marks, suggesting that acK13-HOXB13 marked regions may represent spatially-distinct enhancers (Fig. 4D; Supplementary Fig. S5A; Supplementary Table S4). Furthermore, Hi-C chromatin conformation capture prediction analysis revealed H3K27ac and acK13-HOXB13 chromatin loop interactions at the ACK1 genomic locus (Supplementary Fig. S5B; ref. 33). Approximately 50% of acK13-HOXB13 binding sites overlap with those of genome-wide H3K27ac peaks and occupied long tracks of several kilobases in length (Supplementary Fig. S5C). Consistently, we confirmed that acetylated HOXB13 interacts with the chromatin looping factor CTCF (Supplementary Fig. S5D). ROSE analysis revealed 619, 592, and 632 acK13-HOXB13 or H3K27ac peak-enriched SEs (Fig. 4E–G). Furthermore, 388 out of the 592 (65.5%) H3K27ac signal-enriched SEs overlapped with acK13-HOXB13, whereas 204 SEs were unique (Supplementary Table S5). We validated acK13-HOXB13 at BMPR1B and AR by directed ChIP-qPCR (Supplementary Fig. S5E and S5F). In contrast, we found that AURKB, is targeted by both HOXB13 and acK13-HOXB13 (17).

To determine whether acK13-HOXB13 at TSS and SEs might be due to increased HOXB13 expression in prostate cancers, we performed ROSE analysis which revealed 632 SEs with acK13-HOXB13 after normalizing for input compared with 675 SEs after normalizing for pan-HOXB13 with a majority enriched for acK13-HOXB13 (Supplementary Fig. S6A; Supplementary Table S6A). A representative enrichment at ACK1/TNK2 and FOLH1 is shown (Supplementary Fig. S6B). We conducted ChIP-sequencing using isogenic parental (C4-2B and 22Rv1) and their corresponding HOXB13K13A mutants so as to determine how much the acK13-antibody is picking up peaks as well as SEs in the HOXB13K13A mutant expressing cells. Bioinformatics analysis revealed acK13-HOXB13 unique peaks in both C4-2B and 22Rv1 cell lines at ∼70%, whereas there were fewer than 5% unique peaks in the corresponding HOXB13K13A mutant expressing cells (Supplementary Fig. S6C–S6D). ROSE analysis revealed unique acK13-HOXB13–specific signal-enriched SEs in parental at ∼76% versus HOXB13K13A mutant at ∼1.4% (Supplementary Fig. S6E–S6F; Supplementary Table S6B). Treatment of prostate cancer lines with the CBP/p300 inhibitor reduced mRNA expression of representative acK13-HOXB13 SE-regulated targets (Supplementary Fig. S6G–S6H). Overall, these results indicate that a majority of the SEs were specifically enriched for acK13-HOXB13 signals.

To investigate whether the high-intensity acK13-HOXB13 signal-enriched SE target genes are associated with pathogenesis, we performed GSEA analysis. GSEA revealed a significant enrichment for acK13-HOXB13 SE-proximal targets in mCRPCs (n = 260) compared with primary PC [n = 1,061; NES = 1.392; FDR q = 0.022] and in primary PCs (n = 1,061) compared with normal (n = 794; NES = 1.955; FDR q = 0.000; Fig. 4H; ref. 34). To examine the functional impact of the enriched CRPC-SEs, we integrated the acK13-HOXB13 peak-enriched SE proximal genes with the genes that were differentially expressed (DEG) following HOXB13 ablation (17, 23). A significant enrichment of SE-regulated genes of the primary luminal epithelial subtype was observed in primary PCs (NES = 1.998; FDR q = 0.000; Fig. 4I). Of the 2,540 HOXB13-associated DEGs, 1,307 (51.4%) harbored an acK13-HOXB13 enhancer within 500 kb of the TSS; 194 (7.6%) had an H3K27ac signal-enriched SE and 186 (7.3%) had an acK13-HOXB13 signal-enriched SE (Supplementary Table S7).

To validate the activation of acK13-HOXB13–enriched SE-proximal genes in a larger cancer dataset, we performed t-SNE analysis of these genes in The Cancer Genome Atlas Prostate Adenocarcinoma (TCGA-PRAD) dataset. A heatmap examining this expression revealed no specific pattern in the normal prostate dataset, but uncovered two prominent clusters in tumors (Supplementary Fig. S7A and S7B). These clusters are enriched for genes associated with angiogenesis ANG, VANGL1, and ANGPTL 3–4, enzyme regulators such as ACK1/TNK2 and PPP1R9B (Supplementary Fig. S7C). Directed ChIP-qPCR validated acK13-HOXB13 binding at these targets, which also was not impacted following treatment with enzalutamide (Supplementary Fig. S7D and S7E). Subsequently, we confirmed increased expression of ANGPTL3 and ANGPTL4 in primary tumors by quantitative gene expression analysis (Supplementary Fig. S7F).

To further validate the actionable targets of acK13-HOXB13 epigenome, we examined two critical targets, ACK1 and FOLH1. ACK1 autophosphorylation at Tyr284 leading to its kinase activation is increased in CRPCs, enriched in stem-like cells, and is linked to poor prognosis (7, 35, 36). A potent ACK1 small molecule inhibitor (R)-9b is also reported (Fig. 4J; ref. 28). The acK13-HOXB13 and H3K27 acetylation signal-enriched SE of ACK1 starts upstream of TSS and extends into Intron I (Fig. 4J). Directed ChIP-qPCR validated acK13-HOXB13 binding at the SEs of the ACK1 genomic locus in C4-2B, but not at the control IGX1A locus (Fig. 4K).

FOLH1 encodes the prostate-specific membrane antigen (PSMA; Supplementary Fig. S8A). The radioligand of PSMA, ⁶⁸Ga-PSMA 11-PET (RLT), and the recently approved ¹⁷⁷Lu-PSMA-617 are FDA-approved diagnostic or therapeutic for metastatic CRPCs (37–39). We validated the binding of acK13-HOXB13 to a genomic region proximal to FOLH1 by directed ChIP-qPCR (Supplementary Figs. S2F and S8B). Furthermore, siRNA-mediated depletion of HOXB13 led to a decrease in FOLH1 expression in C4-2B, VCaP, and 22Rv1 cells (Supplementary Fig. S8C–S8E). Moreover, the level of FOLH1 mRNA expression correlated with HOXB13 mRNA levels in both primary and metastatic PCs (Supplementary Fig. S8F and S8G). These results collectively confirm ACK1 and FOLH1 as direct transcriptional targets of acK13-HOXB13.

To investigate the molecular mechanism by which acK13-HOXB13 drives CRPC SE selection, we performed affinity pulldown with biotinylated unmodified and lysine13 modified HOXB13 peptides (1–25 amino acids) followed by MS (Supplementary Fig. S9A). Proteomic analysis revealed a differential enrichment of proteins binding the acK13 peptide compared with the unmodified control HOXB13 peptide (Supplementary Fig. S9B). Proteins enriched in the acK13-HOXB13 peptide pulldown included the SWI/SNF chromatin remodeling proteins SNF2L1/SMARCC1, the bromodomain containing proteins (BAZ2B, SMARCA2, BRD7, and BRD9; ref. 40), DAXX, as well the chromatin looping protein CTCF (Supplementary Fig. S9C; Supplementary Table S8). In contrast, both modified and unmodified HOXB13 N-terminal peptides equally bound members of the transcription pre-initiation complex, including TAF1A/B/C/D and TAF3 associated with RNA polymerase II (Supplementary Fig. S9C–S9D; refs. 41, 42). To confirm BRD9 as a direct reader of K13 acetylation, we tested binding with modified and unmodified biotinylated HOXB13-K13 peptides. BRD9 binds specifically to acetylated HOXB13 but not the unmodified peptide and persists in the presence of H4K5ac peptide (Supplementary Fig. S9E). Co-immunoprecipitation studies revealed interaction of acetylated HOXB13 with BRD9 in C4-2B and VCaP, and this interaction is abolished in the HOXB13K13A mutant (Supplementary Fig. S9F and S9G).

As acetylated K13-HOXB13 is enriched at SEs proximal to CRPC associated target genes, we examined the impact of effect of HOXB13K13A mutation in C4-2B and 22Rv1 cell lines. HOXB13K13A point mutants had undetectable levels of K13-acetylation in both cell lines and with a significant decrease in the expression of HOXB13, ACK1, and PSMA proteins (Fig. 5A; Supplementary Fig. S10A). Functionally, HOXB13K13A exhibited reduced cell proliferation, comparable with the C4-2B HOXB13-pKO deletion mutant (Fig. 5B; Supplementary Fig. S10B; ref. 17). HOXB13K13A point mutants display decrease in number and size of the colonies (Fig. 5C; Supplementary Fig. S10C), reduced 3-dimensional (3D) spheroid formation that indicated a failure to self-renew and self-organize (Fig. 5D; Supplementary Fig. S10C–S10E). Next, to examine resistance to AR-targeted therapies, we treated isogenic parental and HOXB13K13A mutants with enzalutamide (ENZ). Treatment with the ENZ caused a rounding up and decreased colony formation of the HOXB13K13A mutant (Fig. 5E; Supplementary Fig. S10F and S10G). Although the C4–2B parental cells are highly resistant to ENZ (IC₅₀ = 64 μmol/L), the isogenic HOXB13K13A (IC₅₀ = 10–19 μmol/L), and the HOXB13pKO deletion mutants (IC₅₀ = 12.15 μmol/L) had lowered resistance, with a 3- to 5-fold increase in the sensitivity to ENZ (Fig. 5F). The levels of acK13-HOXB13 increased in response to ENZ and decreased in response to the Bromodomain and Extra-Terminal (BET) bromodomain inhibitor, JQ1 (Fig. 5G). Moreover, C4-2B HOXB13K13A mutant displayed reduced H3K27 acetylation but not the H4K12 acetylation (Fig. 5H). Notably, the HOXB13K13A mutants (Fig. 5I–K; Supplementary Fig. S11A–S11C) formed significantly smaller xenograft tumors in male SCID mice with downregulation of ACK1 and PSMA proteins (Supplementary Fig. S11D) and angiopoietin gene expression (ANG, ANGPT2, ANGPTL3, and ANGPTL4) compared with isogenic parental control (Supplementary Fig. S11E and S11F). In addition, flow cytometry analysis confirmed downregulation of PSMA in the HOXB13K13A mutant compared with the parental cells (Fig. 5L). Differential gene expression (DEG) analysis revealed 103 of 113 genes harbored at least one acK13-HOXB13 peak within 500 kb of their TSS (Supplementary Table S9). Furthermore, Gene Ontology (GO) analysis revealed enrichment of chromatin remodeling and self-renewal genes as top targets.

To determine how the loss of acK13-HOXB13 impacts CRPC xenograft tumor growth, we analyzed its recruitment to CRPC-associated targets (Fig. 6A). Directed ChIP-qPCR analysis revealed acK13-HOXB13 binding persisted at the SEs following treatment with the anti-androgen enzalutamide (Fig. 6B–D). Consistently, acetylated HOXB13 recruitment is reduced in the HOXB13K13A mutant but not in the parental cells at the representative SE targets (Supplementary Fig. S12A). Subsequently, we tested whether restoring HOXB13-WT rescues some of the phenotypes observed in the acetylation-defective K13A mutant (Supplementary Fig. S12B). Individual clones from GFP-tagged vector, HOXB13-WT or the HOXB13K13A mutant were flow-sorted, confirmed by Western blotting (Supplementary Fig. S12C), and analyzed for spheroid formation and enzalutamide sensitivity (Supplementary Fig. S12D and S12E). We recovered significantly larger spheroids in the wild-type HOXB13 compared with the HOXB13K13A mutant expressing cells (Supplementary Fig. S12D) that were also resistant to enzalutamide (Supplementary Fig. S12E). Consistently, HOXB13K13A mutants show reduced expression of SE-associated targets (Supplementary Fig. S12F).

As TFs are not directly druggable, we pursued targets amenable to therapeutic intervention. To investigate HOXB13-ACK1 axis in prostate tumor development, we performed single-cell RNA sequencing of human prostate organoids (43, 44). HOXB13, AR, and ACK1 expression was seen in 11 sub clusters in the normal prostate organoids by t-SNE analysis (Supplementary Fig. S6E–S6G). Further delineation of the sub clusters based on the expression of NKX3–1 (luminal epithelial stem cell marker), ZEB 1 (basal epithelial stem cell marker) revealed the presence of progenitor cells expressing both NKX3-1 and ZEB1 (Fig. 6H; Supplementary Fig. S13; refs. 13, 45). As ACK1/TNK2 is expressed within the multiple sub clusters (Fig. 6F), we compared the sensitivity of the normal and tumor-derived prostate organoids to the selective ACK1 inhibitor (R)-9b and the AR antagonists ENZ or abiraterone (Fig. 6I–K). Human prostate tumor organoids display significant sensitivity to (R)-9b, are fairly resistant to enzalutamide, and are modestly sensitive to abiraterone (Fig. 6K). Collectively, our studies underscore for the first time ACK1 tyrosine kinase as an actionable target of HOXB13-deregulated prostate cancers.

This study reveals that acetylation of HOXB13 at lysine 13 transforms it into a pro-CRPC transcription factor. Although pioneer TFs, including HOXB13, can access their target sites within compacted chromatin, and thus are capable of initiating the earliest steps of transcription, whether their chromatin binding activity is regulated is not known (32). p300 and CBP are two closely-related lysine acetyltransferases that function as transcriptional coactivators by engaging lineage-specific TFs at SEs to control cell and tissue identity (24, 25). Our studies elucidate for the first time a close co-operation between HOXB13 and p300/CBP to promote epigenetic marking at enhancers and SEs in cancer cells. These results also suggest that although normal levels of gene expression are maintained by histone H3K27 acetylation, the high levels of gene expression observed in tumors is dependent on the activity of K13 acetylated HOXB13.

Chromatin binding studies indicate that binding of acK13-HOXB13 overlaps with sites of H3K27 acetylation at several characterized SEs, notably those controlling the expression of the AR pathway genes (AR and KLK3), tissue-identity genes (SLC45A3/Prostein, SPON2, PSCA, and FOLH1), and angiopoietins (ANG1, ANGPTL3, and ANGPTL4). As a group, these genes confer prostate-tissue type characteristics at SEs and some occur de novo such as the oncogene ACK1 tyrosine kinase, which is a critical regulator of CRPC growth. In CRPCs, ACK1 expression is high, which in turn promotes the expression of the AR and androgen-independent AR-mediated transcriptional activity (7, 28). Consistently, activation of ACK1 and downstream signaling protects the cells from lethal irradiation, replicative stress, and DNA damage (46, 47). ACK1 is also a target of the E3 ubiquitin ligases SIAH1 and SIAH2 in hormone-associated breast cancers (48). However, prior to this study, it was not clear how ACK1 is upregulated in prostate cancers. Our studies revealed that acetylated HOXB13 has a direct role in ACK1 gene regulation in prostate cancer via its capacity to establish a CRPC-specific SE. Increased ACK1 expression overrides the loss of androgen stimulation consistent with the sensitivity to (R)-9b (28). In summary, our data demonstrate that acetylated HOXB13-mediated CRPC-SEs are critical mediators of anti-androgen resistance.

N.P. Mahajan reports grants from NIH/NCI, Prostate Cancer Foundation, and Department of Defense during the conduct of the study, as well as other support from TechnoGenesys, Inc. outside the submitted work; N.P. Mahajan also has patents 9,850,216 and 10,017,47 issued and licensed to TechnoGenesys, Inc. K. Mahajan reports grants from Department of Defense, NCATS Clinical and Translational Sciences Award, Siteman Comprehensive Cancer Center Support Grant, and Phi Beta Psi Sorority during the conduct of the study, as well as other support from Technogenesys outside the submitted work; K. Mahajan also has a patent for WU Reference: T-019682 pending. No disclosures were reported by the other authors.

D.T. Nguyen: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. W. Yang: Data curation, formal analysis, validation, investigation, methodology, writing–review and editing. A. Renganathan: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. C. Weimholt: Resources, data curation, formal analysis, validation, investigation, visualization, methodology. D.H. Angappulige: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. T. Nguyen: Data curation, formal analysis, validation, investigation, methodology. R.W. Sprung: Data curation, formal analysis, validation, investigation, methodology, writing–review and editing. G.L. Andriole: Resources, investigation, methodology, writing–review and editing. E.H. Kim: Resources, validation, investigation, methodology, writing–review and editing. N.P. Mahajan: Resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–review and editing. K. Mahajan: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

K. Mahajan acknowledges support from the Phi Beta Psi Sorority, Department of Defense W81XWH-21-1-0203, NCATS Clinical and Translational Sciences Award, #UL1 TR002345, and the Department of Surgery at Washington University. N.P. Mahajan is a recipient of NIH/NCI grants (1R01CA208258 and 5R01CA227025), Prostate Cancer Foundation (PCF) grant (17CHAL06) and Department of Defense grant (W81XWH-21-1-0202). The WU-PSR is supported in part by the WU Institute of Clinical and Translational Sciences (NCATS UL1 TR000448), the Mass Spectrometry Research Resource (NIGMS P41 GM103422; R24GM136766), and the Siteman Comprehensive Cancer Center Support Grant (NCI P30 CA091842). The Siteman Cancer Center is supported in part by an NCI Cancer Center Support Grant No. P30 CA091842.

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.