MAP3K7 Loss Drives Enhanced Androgen Signaling and Independently Confers Risk of Recurrence in Prostate Cancer with Joint Loss of CHD1

The majority of men diagnosed with prostate cancer will exhibit indolent tumors, but about 20% will eventually develop aggressive, metastatic disease (1). Although some men diagnosed with early-stage prostate cancer choose active surveillance, most patients will receive some form of definitive treatment including surgery, radiation, chemotherapy, and/or hormonal therapy, all of which have side effects that can be significant (2, 3). Therefore, identification of aggressive subtypes of prostate cancer is of the utmost importance to identify men who would benefit most from aggressive treatment strategies and spare those who are less likely to succumb to their disease. Numerous studies have examined molecular features of primary prostate tumors to define prostate cancer subtypes, but this strategy has yet to identify drivers that predict worse disease-free survival (DFS) in patients (4, 5). Translocations of ERG, for example, are the most common genomic rearrangement, occurring in about 50% of primary prostate cancer; however, ERG rearrangements by themselves do not predict lethality (6).

We recently identified and validated a lethal subtype of prostate cancer that harbors deletions of MAP3K7 and CHD1 (6). Codeletion of CHD1 and MAP3K7 occurs in 15% of primary tumors and associates with poor DFS (6). In multiple cohorts of patients with prostate cancer, loss of CHD1 is significantly associated with loss of MAP3K7, and both are mutually exclusive of ERG translocations (4, 6, 7). In our tissue recombination mouse model, which recapitulates prostate cancer development and pathology through a mix of fetal urogenital mesenchyme and mouse prostate epithelial stem cells, cosuppression of MAP3K7 and CHD1 led to histopathologic evidence of invasive, aggressive disease with signs of abnormal differentiation and increased focal expression of the androgen receptor (AR; ref. 6). Cosuppression of MAP3K7 and CHD1 in a LNCaP human xenograft model also led to increased tumor growth and decreased survival of the mice (6).

MAP3K7 encodes TGFβ activated kinase (TAK1), a serine/threonine kinase downstream of many cytokine signaling receptors including TNF and IL1. TAK1 is known to signal to p38, JNK, NF-kβ, and others (8, 9). CHD1, chromodomain helicase DNA-binding protein 1, binds histone 3 lysine 4 trimethylation marks on chromatin as a reader to facilitate a more open chromatin structure (10). Multiple recent studies have evaluated the role of CHD1 loss in prostate cancer including its contribution to the “normal” AR chromatin landscape, or cistrome, and its relation to resistance to the AR-targeted therapy enzalutamide (11, 12). However, CHD1 is rarely lost without codeletion of MAP3K7, and few studies have evaluated its biological role in the context of MAP3K7 loss.

The AR pathway is a major therapeutic target in the clinic, and many therapies have been developed to disrupt its activity (13, 14). Despite initial response, most patients relapse with a castration-resistant form of the disease for which there is no definitive cure (15). Current efforts are aimed at identifying molecular profiles that can indicate cases with greater potential for treatment resistance (16). In the current study, we determined that a molecular subtype of prostate cancer with loss of MAP3K7 and CHD1 exhibits an increase in many features of AR activity including greater chromatin binding, target gene expression, and resistance to enzalutamide. Low protein expression of MAP3K7 in particular confers significant risk for disease recurrence in prostate cancer patient samples, highlighting an aggressive prostate cancer subtype with potential for resistance to common hormonal therapies.

LNCaP cells (ATCC Cat# CRL-1740, RRID:CVCL_1379) were obtained from the ATCC, and LAPC4 (ATCC Cat# PTA-1441, RRID:CVCL_4744) and 22RV1 (ATCC Cat# CRL-2505, RRID:CVCL_1045) cell lines were acquired from the University of Colorado Cancer Center cell line repository. Cell line identity was verified by spectral karyotyping every 6 months, and cells were used below passage 50. Each batch of frozen cells is tested for Mycoplasma by DAPI staining and PCR, and once thawed, cells are passaged a maximum of 4 times for use in experiments. Production of lentivirus for shRNA plasmids and generation of LNCaP shControl, shMAP3K7, shCHD1, and shDouble cell lines was described previously (6). LAPC4 and 22RV1 cells were generated in the same manner. Cells were kept in shRNA antibiotic selection with 1 μg/mL puromycin (Sigma #P8833) and 800 μg/mL Geneticin (InvivoGen, G-418, #ant-gn-5). ShRNA targeting constructs are listed in Supplementary Table S1. Cells were maintained in RPMI1640 (LNCaP, 22RV1) or IMDM (LAPC4) medium (Gibco #21875034 and #12440053) with 1% penicillin–streptomycin (PS; Thermo Fisher Scientific #15140122) and 10% or 7.5% FBS (Gemini Bio-Products #100–106), respectively. LAPC4 cells were also supplemented with 1 nmol/L R1881 (PerkinElmer, #NLP005005MG). Where indicated, cells were cultured in steroid-depleted medium, which consisted of phenol red-free RPMI1640 (Gibco #11835030) or IMDM (Gibco #21056023) supplemented with 10% (LNCaP, 22RV1) or 7.5% (LAPC4) charcoal-stripped FBS (Gibco #12676029) and 1% PS. For indicated assays, cells were treated with varying doses of R1881 and enzalutamide (Selleckchem #MDV3100).

LNCaP monolayer growth assays were quantified by trypan blue exclusion. LAPC4 and 22RV1 growth assays and enzalutamide growth assays for all cell types were measured with IncuCyte ZOOM Live-Cell Analysis System using percent well confluence.

RNA was isolated using TRIzol Reagent (Invitrogen #15596018). Real-time quantitative PCR (qRT-PCR) was carried out using diluted cDNA, Power SYBR Green PCR MasterMix (Applied Biosystems #A25742) and the primers listed in Supplementary Table S1. The primers for AR were described previously (17). Samples were run on an ABI 7500 Real-time PCR Instrument (Applied Biosystems).

LNCaP cells were transfected with Lipofectamine RNAiMAX Reagent (Invitrogen #13778150). AR-variant-targeting siRNA constructs have been described previously (18, 19). RNA was collected at 24 hours for qRT-PCR verification, and immunoprecipitation (IP) of AR(N20) was done after 72 hours.

LNCaP cells were plated in 96-well plates in complete growth medium for 2 days. Medium was changed to steroid-depleted medium, and cells were transfected with AR-variant–targeting siRNA constructs. Twenty-four hours after transfection, Annexin V fluorescent conjugate (Biotium #29005) and propidium iodide (PI) fluorescent dye (Biotium #40016) were added to culture medium and Annexin V (green), PI (red), and cell confluence were measured with the IncuCyte Zoom Live-Cell Analysis System.

LNCaP cells were grown in steroid-depleted medium ± 1 nmol/L R1881 for 48 hours. Staining of cells for cell-cycle analysis was performed as described previously (20).

Standard protocols were used for protein isolation, quantification, and immunoblotting. Images were acquired using LI-COR Odyssey imaging systems and quantified using Image Studio software (RRID:SCR_015795). Antibodies used were MAP3K7 (Cell Signaling Technology #5206S, RRID:AB_ 10694079), CHD1 (Bethyl # A301–218A, RRID:AB_890568), AR N-20 (Santa Cruz Biotechnology #sc-816, RRID:AB_1563391), AR C-19 (Santa Cruz Biotechnology #sc-815, RRID:AB_ 630864), AR-v7 (Precision Antibody #AG10008, RRID:AB_2631057), Lamin A/C (Cell Signaling Technology #4777, RRID:AB_10545756), α-tubulin (Cell Signaling Technology #3873, RRID:AB_1904178), β-actin (Cell Signaling Technology #3700, RRID:AB_2242334), PARP (Millipore #AB16661, RRID:AB_ 90869), cleaved PARP (Cell Signaling Technology #5625, RRID:AB_10699459), cleaved caspase-3 (Cell Signaling Technology #9661S, RRID:AB_2341188), goat anti-rabbit-HRP (Santa Cruz #sc-2004, RRID:AB_631746), and goat anti-mouse-HRP (Santa Cruz Biotechnology, #sc-2005, RRID:AB_631736).

Nuclear and cytoplasmic extracts were prepared as described previously (21). Aliquots of whole-cell lysates were taken from each sample after hypotonic lysis and before centrifugation.

LNCaP cells in triplicate were grown in steroid-depleted medium for 3 days, then treated with either 0.1% ethanol (EtOH) or 1 nmol/L R1881 for 4 hours. RNA was extracted using the PerfectPure RNA Cultured Cell Kit (5Prime #2302350) including an additional DNase treatment using Turbo DNA-Free Kit (Ambion #AM1907). Samples were submitted to the University of Colorado Anschutz Medical Campus Genomics and Microarray Core. Libraries were prepared with the Illumina TruSEQ Stranded Total RNA with Ribo-Zero TM Library Preparation Kit. Sequencing was performed on Illumina HiSEQ 2000 High-throughput mode using V3 Chemistry, PE100. Gene expression was quantified using default parameter of RSEM (v1.2.1, RRID:SCR_013027) with Bowtie2 (v2.1.0, RRID:SCR_016368) as the read aligning program and differential expression was calculated using EBSeq (v1.1.5, RRID:SCR_003526). Reads were aligned to hg19 RefSeq (RRID:SCR_003496) annotated genes. LAPC4 and 22RV1 shControl and shDouble cells were collected as described for LNCaP cells, and samples were submitted at a later date to Novogene. Libraries were prepared using Novogene's eukaryotic RNA sequencing (RNA-seq) method, and sequencing was performed on the Illumina NovaSeq 6000 Platform, PE150. Gene expression was quantified as described for LNCaP cells. Heatmaps and further comparison of gene expression changes for all cell lines were generated with Rstudio (v1.2.5033, RRID:SCR_000432).

Details for chromatin immunoprecipitation (ChIP) sample collection are outlined in Supplementary Methods. For ChIP sequencing (ChIP-seq), 1 ng of ChIP or input DNA per sample was used for library construction (NuGEN). Samples were pooled and sequenced on Illumina HiSEQ 2000 high-throughput mode using V3 Chemistry, PE100 at the University of Colorado Anschutz Medical Campus Genomics and Microarray Core. The fastq file quality was accessed using FastQC (RRID:SCR_014583; http://www.bioinformatics.babraham.ac.uk/projects/fastqc/) and MultiQC (RRID:SCR_014982; https://multiqc.info/). Illumina adapters and low-quality reads were filtered out using BBDuk (RRID:SCR_016969; http://jgi.doe.gov/data-and-tools/bb-tools). Bowtie2 (v.2.3.4.3, RRID:SCR_016368) was used to align reads to the hg38 reference human genome. Samtools (v.1.9, RRID:SCR_002105) was used to select the mapped reads. Optical and PCR duplicates were removed using Picard (RRID:SCR_006525) MarkDuplicates tool (Broad Institute, http://broadinstitute.github.io/picard/). The same number of reads for all samples was randomly extracted using samtools view. Bedtools (RRID:SCR_006646) genomecov was used to create bedgraph files, and Igv (RRID:SCR_011793) tools toTDF was used to create tdf files. Peaks were called using MACS2 (v2.1.1.20160309, RRID:SCR_013291; ref. 22) with default parameters. Average profiles of AR ChIP peaks to their binding sites were generated using ngs.plot (RRID:SCR_011795; ref. 23).

The Stanford University database of surgically removed prostates (1986–2003) was previously subjected to a comprehensive histopathologic review (24). In this current study, we selected high-grade cases (containing Gleason grade 4 or 5) with 5+ years of follow-up, 125 with biochemical recurrence (BCR) and 125 without BCR that were matched by age at surgery and by percent total high-grade cancer. Two hundred and six of these matched cases, 112 with BCR and 94 without, had sufficient tumor tissue for all histologic analyses. The pathologist was blinded to patient clinical features including BCR status during IHC scoring of samples. The investigators did not have access to patient identifiers and the Institutional Review Board of the University of Colorado (Aurora, CO) considered this as nonhuman subject research. Complete details regarding selection of this cohort and analysis of clinical and histopathologic features are outlined in Supplementary Methods. Complete information for all cases is included in Supplementary Data.

mRNA expression data, genomic CNA data, and clinical data for MSKCC (25), and TCGA patient datasets (4) were downloaded from the cBioPortal for Cancer Genomics (RRID:SCR_014555; ref. 26). Data were further processed using GraphPad Prism (RRID:SCR_002798) and RStudio (RRID:SCR_000432), and pathway annotations were generated with gene set enrichment analysis (GSEA, RRID:SCR_003199; ref. 27). Metascape (RRID:SCR_016620) was used for cell line RNA-seq and ChIP-seq pathway enrichment analysis (28). For the meta-analysis of patient genomic data, CNA data from the MSKCC/DFCI (29), Broad/Cornell (30), TCGA PanCan (31), and SU2C (32) datasets were downloaded from cBioPortal. Duplicate patient IDs were filtered to produce a final list of unique patient IDs, along with discretized calls for deep and shallow deletions for MAP3K7 and CHD1, and fusions for ERG.

Statistical analyses were performed with GraphPad Prism 7 and 8 (RRID:SCR_002798). Data were plotted as means ± SEM. Two-tailed P < 0.05 was considered statistically significant. Asterisks denote levels of statistical significance as determined by one or two-way ANOVA with Tukey or Dunnett multiple comparisons test, unpaired two-tailed t test, one-sided Fisher exact test, log-rank Mantel–Cox test, Wilcoxon rank sum test, or χ² test for trend. Specific tests used are indicated in the figure legends. All experiments were performed in triplicate unless noted otherwise.

Accession codes: The RNA-seq and ChIP-seq data are available at the Gene Expression Omnibus (GEO), reference series GSE168671. The code used for our patient dataset analysis is available at GitHub: https://github.com/jillsonl/MCR_2021_publication.

We first sought to explore the functions of MAP3K7 and CHD1 in androgen signaling using LNCaP, an AR-responsive prostate cancer cell model. To determine whether dual loss of MAP3K7 and CHD1 alters androgen-stimulated proliferation, we performed a monolayer growth assay using LNCaP cells stably expressing shRNAs against MAP3K7 and/or CHD1 (Fig. 1A) in the absence [castration condition, EtOH] or presence of androgen. Suppression of MAP3K7 and CHD1 (shDouble) led to a robust increase in the growth of LNCaP cells in the absence and presence of a minimally stimulatory dose (10 pmol/L) of the synthetic androgen R1881 (Fig. 1B). This was true for seven different combinations of shRNAs against both MAP3K7 and CHD1.

Suppression of MAP3K7 alone was also sufficient for increased growth (Fig. 1B and C) while knockdown of CHD1 was not (Fig. 1B and C), suggesting MAP3K7 suppression contributes more to this growth effect. A dose–response curve of R1881 showed that MAP3K7 knockdown increased relative androgen-stimulated growth at many doses compared with shControl cells (Fig. 1D). In addition, shDouble cells exhibited a shifted dose–response curve, reaching higher doses of R1881 before a growth-inhibitory effect occurred (Fig. 1D). In LAPC4 cells, another androgen-sensitive prostate cancer model, dual knockdown of MAP3K7 and CHD1 (Supplementary Fig. S1A) also increased androgen-independent and stimulated growth compared with shControl cells (Supplementary Fig. S1B). shMAP3K7 and shDouble LAPC4 cells similarly displayed increased growth at various R1881 doses compared with shControl (Supplementary Fig. S1C). In the castration-resistant cell line 22RV1, knockdown of MAP3K7 and CHD1 did not provide a growth advantage, likely due to the advanced stage and androgen independence of this cell line (Supplementary Fig. S1D).

We next examined the effects loss of MAP3K7 and/or CHD1 had on gene expression of AR canonical target genes. Expression of all genes was induced by R1881 as expected, but to varying degrees with the different knockdowns (Fig. 1E). TMPRSS2, NKX3–1, and PMEPA1 were strongly induced in shMAP3K7 and shDouble cells compared with shControl cells, while KLK3 was most strongly induced in shCHD1 and shDouble cells (Fig. 1E). This suggests that knockdown of MAP3K7 and CHD1 each contribute to the increased expression of various AR target genes seen in shDouble cells. Similar gene expression trends were seen in LAPC4 and 22RV1 cells (Supplementary Fig. S1E and S1F), demonstrating that the combined knockdown of MAP3K7 and CHD1 increases AR transcriptional activity.

AR protein levels remained largely consistent across multiple medium conditions in shDouble LNCaPs cells relative to shControl cells (Fig. 1F). Because translocation of AR from the cytoplasm to the nucleus upon ligand binding is a critical step in AR signaling, we then assessed nuclear localization of AR. A slight increase in AR nuclear localization was observed in shDouble cells in the absence of ligand, while a robust increase occurred in response to androgen stimulation in shDouble cells (Fig. 1G; Supplementary Fig. S1G). The effects seen on AR nuclear localization along with increased target gene expression suggest significant alterations in AR activity induced by the combined knockdown of MAP3K7 and CHD1.

To globally assess changes in AR-regulated gene expression, we performed RNA-seq and AR (N-terminal) ChIP-seq on shControl, shMAP3K7, shCHD1, and shDouble LNCaP cells grown in castration conditions and supplemented with vehicle (EtOH) or 1 nmol/L R1881 for 4 hours (Fig. 2A). We also performed RNA-seq on 22RV1 and LAPC4 shControl and shDouble cells ± 1 nmol/L R1881. Initially, we analyzed the mRNA expression profiles of all 3 cell lines against a set of verified AR target genes (Fig. 2B; ref. 33). Strikingly, the shDouble LNCaP, LAPC4, and 22RV1 cells all had a higher basal level of AR target gene expression than shControl cells in castration conditions, which increased even further upon R1881 administration (Fig. 2B). The increases in gene expression seen in shDouble LNCaP cells were much stronger than those induced by shMAP3K7 or shCHD1 individually, with some target genes being increased specifically by knockdown of either MAP3K7 or CHD1. Consistently, a similar pattern of expression was observed when the RNA-seq data was analyzed against the “Hallmark Androgen Response” gene set from the Molecular Signatures Database (MSigDB; Supplementary Fig. S2A; ref. 34).

Using GSEA, we evaluated an AR signaling gene set (MSigDB: M15861; ref. 34) in a TCGA dataset of primary prostate cancer stratified by MAP3K7 and/or CHD1 deletions (4). When comparing cancers with deletions of MAP3K7, CHD1, or both genes to those with loss of neither, only those with the dual deletion exhibited a relative enrichment in the AR signaling signature (Fig. 2C). This was also true within a Memorial Sloan Kettering Cancer Center (MSKCC) prostate cancer dataset (Supplementary Fig. S2B; ref. 25).

Based on the increased AR gene signature and target gene expression, we suspected an increase in AR chromatin binding in shDouble cells. Using AR ChIP-seq data from LNCaP cells, we compared AR binding events across the genome in ethanol and R1881-treated conditions. In ethanol conditions, only shDouble cells exhibited a significant increase in AR binding compared with shControl cells (Fig. 2D, top). In R1881-stimulated conditions, shDouble cells maintained this increased AR binding (Fig. 2D, bottom). Interestingly, knockdown of CHD1 did not significantly change overall AR binding intensity compared with shControl cells while knockdown of MAP3K7 decreased AR binding (Fig. 2D, bottom). Examples of AR peaks at two target genes are shown in Supplementary Fig. S2C. At the ELL2 locus, AR binding is slightly elevated in shCHD1 cells and significantly increased in shDouble cells (Supplementary Fig. S2C, top). Conversely, both individual knockdowns decrease AR binding at the FKBP5 locus while the double knockdown greatly increases the AR peak (Supplementary Fig. S2C, bottom). These findings suggest a mechanism by which the combination of MAP3K7 and CHD1 suppression cooperatively increases AR chromatin binding.

To further investigate the respective contributions of MAP3K7 and CHD1 suppression to changes in the AR cistrome, we compared all ChIP binding events in shControl, shMAP3K7, shCHD1, and shDouble R1881-stimulated cells to shControl EtOH-treated cells as a relative baseline. We compared the overlap of differential AR binding events and identified many peaks that were shared among the 3 cell types (5453). Suppression of CHD1 induced many more differential peaks overall, as well as contributed more unique peaks than shMAP3K7 to the shDouble AR cistrome (1769 vs. 182; Fig. 2E). Interestingly, shDouble cells had many unshared sites (1138) that were either binding events unique to shDouble cells, or only exhibited a significant increase over shControl EtOH cells upon dual suppression of MAP3K7 and CHD1 (Fig. 2E).

We then overlapped the differential AR ChIP binding events with RNA-seq differential expression to identify functional gene expression changes potentially induced by AR. Using the ChIP peaks from Fig. 2E, we further filtered for genes that showed significant expression changes (FDR < 0.05) in R1881-treated cells relative to shControl (EtOH-treated) cells (Fig. 2F). Although shDouble cells still exhibited many unique ChIP sites with significant gene expression changes (749), the relative contribution of shCHD1 and shMAP3K7 to effects seen in shDouble cells shifted to a more even distribution (372 and 321, respectively; Fig. 2F). shMAP3K7 cells displayed significantly more unique AR-bound genes with altered gene expression (562), implying that some of the shared AR-bound sites between all 3 conditions from (Fig. 2E) only exhibited significant expression changes upon loss of shMAP3K7 alone.

The individual points from Fig. 2F, including data from shControl R1881 versus shControl EtOH not shown previously, are shown in Fig. 2G. These results illustrate that while knockdown of CHD1 contributed more to chromatin-binding changes, knockdown of MAP3K7 induced greater expression changes for genes bound by AR. In addition, suppression of MAP3K7 significantly contributed to the pathway enrichments seen among genes from Fig. 2G in shDouble cells (Supplementary Fig. S2D). Similarly, when comparing differential AR ChIP binding events induced by R1881 within each genetic background, a stepwise increase in AR ChIP binding intensity (fragment reads) was evident from shControl, shMAP3K7 and shCHD1 to shDouble cells (Supplementary Fig. S2E).

Given the observed alterations in androgen signaling, we sought to determine whether loss of CHD1 and MAP3K7 expression altered the cellular response to enzalutamide, a common AR antagonist administered in the clinic (35). When LNCaP cells were treated with varying concentrations of enzalutamide, we observed a significant increase in resistance in shDouble cells compared to shControl cells (Fig. 3A). Knockdown of either MAP3K7 or CHD1 was sufficient for increased resistance at several doses, highlighting the impact that both genes have on AR activity. shDouble LAPC4 cells demonstrated a similar trend of increased resistance at multiple doses compared with shControl cells, while the single knockdowns were more sensitive to enzalutamide at some doses (Supplementary Fig. S3A).

A number of mechanisms of enzalutamide resistance have been proposed. These include alternative splicing of the AR gene which results in AR splice variants (AR-Vs) lacking the ligand-binding domain. This allows AR to function in a ligand-independent, constitutively active manner (19, 36). Expression of AR-v7, the most abundant AR variant, is predictive of BCR, upregulated in castration-resistant prostate cancer (CRPC; refs. 19, 37), and correlated with decreased survival following enzalutamide treatment (38).

We evaluated expression of AR-v7 in our models and found AR-v7 to be weakly expressed in shControl LNCaP cells in castration conditions and significantly increased in shDouble cells (Fig. 3B, black bars). The increase in AR-v7 expression is likely induced by loss of MAP3K7 as AR-v7 levels in shDouble cells are not much higher than in shMAP3K7 cells (Fig. 3B, green bars). Because LNCaP cells express AR-full length (AR-FL) at much higher levels than AR-v7, we also calculated the ratio of AR-v7 to AR-FL mRNA. We observed that multiple shDouble combinations had higher AR-v7/AR-FL ratios relative to shControl cells (Supplementary Fig. S3B). Suppression of MAP3K7 and CHD1 in LAPC4 cells also resulted in increased expression of AR-v7 (Supplementary Fig. S3C). 22RV1 cells already have high levels of AR-v7 expression, and shDouble cells did not exhibit a further increase. However, an increase in a similar variant AR-v567es was observed, suggesting MAP3K7 and CHD1 may regulate AR splice variant expression more broadly (Supplementary Fig. S3D).

Whether or not AR-v7 transcriptional targets differ from those of AR-FL is heavily debated (39, 40), but some putative specific targets of AR-v7 include UGT2B17, UBE2C, and AKT1 (17). UGT2B17 and AKT1 were significantly upregulated in shDouble LNCaP cells relative to shControl cells, with UGT2B17 also being significantly increased in shMAP3K7 and shCHD1 cells (Fig. 3C). There was no difference in UBE2C expression. We evaluated the association of MAP3K7 and CHD1 with expression of these AR-v7 target genes in the MSKCC patient dataset (25). Expression of all three genes, UGT2B17, AKT1, and UBE2C, was significantly inversely correlated with expression of MAP3K7 and CHD1 (Fig. 3D–F).

To determine whether increased AR-v7 expression contributed to the enhanced growth of shDouble cells, we utilized AR isoform-specific siRNA targeting constructs to knockdown AR-FL, AR-v7, or total AR capturing both variants (Supplementary Fig. S3E and S3G; refs. 18, 19). We evaluated cellular staining of PI and Annexin V as markers of apoptosis in LNCaP cells transfected with the siRNAs in castrate medium conditions. In both shControl and shDouble cells, total AR knockdown significantly increased PI staining by 48 hours (Fig. 3G). shControl cells were also somewhat sensitive to siAR-FL and siAR-v7 while shDouble cells were only sensitive to siAR-v7. Annexin V detection displayed similar results (Supplementary Fig. S3H). These patterns of increased PI and Annexin V staining were recapitulated by detection of cleaved PARP. shControl cells exhibited cleaved PARP induction with siAR-v7, siAR-FL, or most strongly with siAR-total (Fig. 3H). shDouble cells instead exhibited PARP cleavage only with siAR-v7 or siAR-total (Fig. 3H).

We know from previous studies that gene expression of CHD1 and MAP3K7 correlate with each other and with poor DFS (6). We recently found that decreased MAP3K7 protein (TAK1) alone is correlated with prostate cancer brain metastases (41), and others showed that copy number loss of MAP3K7 and CHD1 is associated with cribriform pattern prostate cancer in the TCGA (42). Cribriform patterning is a sub-variant of Gleason pattern 4 and is an independent predictor of poor outcome (43). Although these recent studies support the hypothesis that the dual loss of MAP3K7 and CHD1 drives aggressive prostate cancer, the evaluation of CHD1 and MAP3K7 protein expression in primary prostate cancer and its association with patient outcome has not been evaluated.

In this study, we evaluated the expression of CHD1 and MAP3K7 in a cohort of well-characterized radical prostatectomy specimens with clinical follow-up to assess the impact of reduced expression of one or both proteins on BCR. From the Stanford database, we identified 206 high-grade cases, 112 with BCR and 94 without BCR matched for age at surgery and percent total high-grade cancer (Gleason grade 4 or 5). This cohort is outlined in detail in the Supplementary Methods. Supplementary Table S2 shows characteristics of these cases illustrating that patients with BCR had higher Gleason scores, PSA, total cancer volume, clinical stage, and involvement of lymph nodes and seminal vesicles. Cases with BCR also more often had cribriform patterns present.

H-scores for CHD1 and MAP3K7 were dichotomized into high expressors (≥150) and low expressors (<150) for each case. Examples of samples with high and low expression of CHD1 and MAP3K7 are shown (Fig. 4A; Supplementary Fig. S4A). An initial comparison of H-scores across all cases illustrated that expression of MAP3K7 and CHD1 were each significantly lower in the BCR group compared with the non-BCR group (Fig. 4B; Supplementary Table S3). We further stratified patients into 4 groups based on the combination of high and low CHD1 and MAP3K7 H-scores to assess their individual and combined associations with BCR. As expected, cases with low H-scores of both proteins demonstrated the highest frequency of BCR (Fig. 4C, 84.6%). Interestingly, cases with low MAP3K7 H-scores alone also demonstrated a high percentage of BCR (Fig. 4C, 74.4%), much higher than that of cases with only low CHD1 H-scores (Fig. 4C, 46.2%). This suggests that MAP3K7 protein expression is a potential driving factor of aggressiveness in the MAP3K7/CHD1 subtype.

Because our earlier analyses of patient datasets showed that coordinate loss of CHD1 and MAP3K7 was associated with ERG-negative tumors (6), we stratified BCR cases and non-BCR cases by ERG expression (Fig. 4D; Supplementary Table S4). The mean H-scores for CHD1 and MAP3K7 were significantly lower in BCR cases compared with non-BCR cases in both ERG-positive and ERG-negative cases (Fig. 4D). Despite the fact that MAP3K7 and CHD1 deletions are generally mutually exclusive of ERG-fusions, 60% of ERG-positive cases had low H-scores for either MAP3K7, CHD1, or both (Fig. 4E). In both ERG-positive (Fig. 4E) and ERG-negative (Fig. 4F) subgroups, cases with only low MAP3K7 H-scores again had a higher proportion of BCR than cases with only low CHD1 H-scores, suggesting MAP3K7 expression as a key factor regardless of ERG status. It is important to note that there are very few ERG-negative cases with a low CHD1 H-score in absence of a coordinately low MAP3K7 H-score (Fig. 4F), similar to what is seen at the genomic level in patient datasets.

If cases with cribriform morphology were removed from the analysis, low MAP3K7 was significantly associated with BCR regardless of ERG status while CHD1 expression was only different between BCR and non-BCR groups in ERG-positive tumors (Supplementary Fig. S4B and S4C). In tumors with cribriform morphology, however, H-scores for both CHD1 and MAP3K7 were only significantly lower in ERG-negative cases with BCR (Supplementary Fig. S4D). Unlike in noncribriform cases (Supplementary Fig. S4C), independent low H-scores of CHD1 or MAP3K7 both resulted in high frequencies of BCR, similar to that of their combined effect (Supplementary Fig. S4E). This suggests that low CHD1 expression has more impact on BCR in cases with cribriform cancer than those without.

Because of the frequency of low MAP3K7 and CHD1 H-scores we see in ERG-positive cases, we performed a meta-analysis of patient genomic data from multiple published cohorts to better evaluate the frequency of these genomic alterations. Interestingly, we found that at the primary stage, 19.4% of ERG-positive cases exhibited MAP3K7 deletions (homozygous and heterozygous together), and 11.6% had CHD1 deletions (Fig. 4G; Supplementary Table S5). These frequencies are recapitulated well in our IHC cohort with 9% of the ERG-positive cases having truly low (<100) H-scores of MAP3K7 and CHD1 together. The proportion of overlap between MAP3K7 and CHD1 deletions and ERG fusions increases even more in metastatic tumors (Supplementary Fig. S5; Supplementary Table S5).

We performed univariate and multivariate logistic regression analyses to determine key variables that predict BCR. In the univariate analysis, CHD1 H-scores, MAP3K7 H-scores, total cancer volume, and clinical stage were all significantly, positively associated with BCR (Table 1). We then performed a multivariate regression analysis using an optimized combination of clinical factors with the highest predictive value as the base model, and BCR as the output. We tested all clinical variables in our study that would be available postsurgery, and the strongest combination included total cancer volume, PSA, seminal vesicle invasion, lymph node involvement, Gleason scores, age, date of surgery, and prostate weight. MAP3K7 and CHD1 H-scores were then added individually and together to determine whether they added to or subtracted from the predictive value of this base model (Table 2). Interestingly, MAP3K7 H-scores added to the model while CHD1 H-scores did not. In addition, we used an established model of post radical prostatectomy risk assessment, the CAPRA-S Score, as the base model for the multivariate analysis. The CAPRA-S score considers multiple postoperative factors including PSA, Gleason score, seminal vesicle invasion, lymph node invasion, surgical margins, and extracapsular extension (44). MAP3K7 H-scores also significantly added to the predictive value of the CAPRA-S score (Supplementary Table S6). This demonstrates that MAP3K7 expression has a strong association with BCR that significantly adds to the predictive value of established postsurgical clinical factors.

We also performed a multivariate regression analysis using MAP3K7 H-scores as the base model. Only the addition of seminal vesicle invasion, total cancer volume, and lymph node involvement notably added to the predictive power of this model (Supplementary Table S7).

Because we determined that low MAP3K7 protein expression is a major driving force behind the clinical impact of the MAP3K7/CHD1 subtype, we further investigated the consequences of loss of MAP3K7 expression. We performed Metascape analysis on significant gene expression changes in LNCaP shMAP3K7 cells compared with shControl cells (both R1881-treated) to identify pathways altered by MAP3K7 loss (28). Pathways enriched in genes upregulated by suppression of MAP3K7 included ephrin signaling, cell adhesion, and mesenchymal cell differentiation (Fig. 5A, green bars). Genes downregulated by shMAP3K7 were heavily involved in the cell cycle, DNA repair, cytoskeletal organization, and response to cytokines (Fig. 5A, blue bars).

We next compared enrichments seen in our cell line model to patient data. TCGA primary prostate cancer cases were stratified by mRNA expression quartiles to perform enrichments in cancers with low (quartile 1) versus high (quartile 4) MAP3K7 expression, CHD1 expression, or combined expression (4). GSEA enrichments for these three comparisons are summarized by normalized enrichment scores showing shared and unique pathway enrichments (Fig. 5B) (27).

Many of the pathways enriched in cancers with low MAP3K7 or low MAP3K7 and CHD1 expression are similar to those seen in LNCaP cells. Some of the strongest enrichments observed in TCGA MAP3K7 low cases are highlighted (Fig. 5C). The negative enrichment scores for axon guidance and adherens junction suggest cytoskeletal and cell adhesion changes, and we have previously shown that suppression of MAP3K7 induces changes in actin cytoskeleton and increases cell motility (45). The spliceosome pathway is also enriched in cancers with low MAP3K7 expression (Fig. 5C), which could explain alterations seen in AR splice variant expression. Further studies will be needed to elucidate the specific mechanisms by which MAP3K7 may regulate splicing.

Alterations in DNA repair pathways and the cell cycle (Fig. 5B and C) were of particular interest as we have recently shown that loss of MAP3K7 induces sensitivity to the combination of the cell-cycle checkpoint inhibitor dinaciclib and DNA-damaging agents (46). One of the strongest negative enrichments seen in TCGA prostate cancer data was that of the G₂–M transition (Fig. 5B), and GSEA analysis of LNCaP shMAP3K7 versus shControl cells showed a similar enrichment (Fig. 5D). We performed cell-cycle analysis in LNCaP cells after 48 hours ± 1 nmol/L R1881 to evaluate potential changes in cell-cycle transitions. Androgen-treated shControl cells became growth-inhibited relative to vehicle-treated cells at this R1881 concentration and time point, demonstrated by a decreased G₂–M ratio (Fig. 5E) along with an increase in percentage of cells in G₁ (Supplementary Fig. S5A). ShMAP3K7 cells were not significantly affected following R1881 treatment while shCHD1 cells were growth inhibited by this dose of R1881 (Fig. 5E; Supplementary Fig. S5A). In contrast, androgen-treated shDouble LNCaP cells displayed a significantly higher G₂–M ratio (Fig. 5E) and lower percentage of cells in G₁ (Supplementary Fig. S5A) relative to untreated cells. This was also significant or trending across multiple shDouble shRNA construct combinations (Supplementary Fig. S5B).

These results indicate that shDouble cells move rapidly through the cell cycle at a dose of androgen that is normally growth inhibitory to LNCaP cells. This finding suggests that suppression of MAP3K7 is significantly contributing to the shift in response to androgen. The corresponding enrichment of the G₂–M transition across many TCGA cases with low expression of MAP3K7 suggests a robust alteration in genes regulating the cell cycle.

We have made a number of significant findings in the current study. We are the first, to our knowledge, to evaluate MAP3K7 and CHD1 expression by IHC in association with BCR in a clinical cohort of prostate cancer patients. We showed that tumors with BCR had significantly lower expression of CHD1 and MAP3K7 than those without BCR, and that low MAP3K7 expression was the most significant independent predictor of BCR. This is a critical finding as other studies in the field have focused primarily on the loss of CHD1, which we demonstrated is a factor secondary to MAP3K7 expression for predicting BCR. However, evaluation of MAP3K7 and CHD1 expression in primary prostate cancer and its correlation with subsequent metastasis and therapeutic response is also needed to fully understand the predictive value of this subtype. Recent studies have evaluated such clinical cohorts, but they did not include the prevalent losses of MAP3K7 and CHD1 in their analysis (47).

In our previous study, we evaluated suppression of MAP3K7 and CHD1 in our mouse prostate stem cell, tissue recombination model. We found that grafts of dual knockdown cells exhibited aggressive pathology, as well as focal areas of increased AR expression and evidence of lineage plasticity (6). Using human prostate cancer cell line models in our current study, we have expanded our findings to show that combined loss of MAP3K7 and CHD1 has striking effects on AR activity. Importantly, these effects are far greater than those seen with suppression of either gene alone. Although it is important to evaluate the individual effects of CHD1 and MAP3K7 for mechanistic insight, the dual loss of these genes is the condition most often seen in patients.

Suppression of MAP3K7 resulted in decreased AR chromatin binding along with increased AR target gene expression, suggesting a potential repressive role of MAP3K7 (TAK1) on normal AR activity. We also noted that the addition of MAP3K7 suppression to that of CHD1 drives larger changes in expression of AR-targeted genes. The synergistic changes in AR activity we see upon loss of both CHD1 and MAP3K7 may be due to compound effects on the AR signaling axis. MAP3K7 loss may be increasing AR-v7 levels along with relieving repression of AR while CHD1 loss alters the chromatin, dictating where AR and its variants can bind. Further studies on this potential mechanism are warranted. Previous studies evaluating effects of CHD1 loss alone may therefore underrepresent the significance of its impact on the AR pathway in patients.

Remarkably, loss of MAP3K7 significantly impacted the cell cycle and related pathways. In prostate cancer, there is significant interplay between AR and the cell cycle with feedback that can in turn affect AR signaling (48). UBE2C, a putative AR-v7-regulated gene, is known to be involved in controlling cyclin levels, and a significant increase in UBE2C expression was observed in patients with low expression of MAP3K7 and CHD1 (49). We previously reported that suppression of MAP3K7 sensitizes cells to the combination of a cell-cycle inhibitor dinaciclib and DNA-damaging agents (46). It will be interesting to evaluate how these drugs affect the increased AR signaling we see in cells with suppression of MAP3K7 and CHD1.

The significant differences in the genetic phenotype and disease stage of our prostate cancer cell lines did pose limitations in our ability to identify consistent mechanisms to describe the increased AR activity seen upon suppression of MAP3K7 and CHD1. The specific mechanisms may vary slightly depending on the presence of different cofactors and confounding genetic alterations of the cell lines. In addition, the LNCaP cell line possesses an ETV1 rearrangement, which is typically mutually exclusive from loss of MAP3K7 and CHD1 in genomic databases (50). However, our meta-analysis in Fig. 4G (Supplementary Fig. S5) and findings in our clinical cohort in Fig. 4 demonstrate that these features are not as mutually exclusive as previously presented. We have repeated some of our studies in other cell lines to ensure the results were not specific to LNCaP cells. Correlating AR-v7 levels with genetic alterations in public datasets has also proven difficult, and extensive IHC studies are likely necessary to successfully profile the correlation of AR-v7 expression with the loss of MAP3K7 and CHD1.

The findings in this study could have significant implications in the clinical management of prostate cancer. Measurement of low protein expression of MAP3K7 alone or in conjunction with CHD1 could be employed to stratify patients who are likely to have recurrent disease and therefore might benefit from more aggressive therapy. In addition, the alterations seen in AR signaling along with increased AR-v7 expression and enzalutamide resistance suggest that patients with this subtype may have limited response to some AR-targeting therapies commonly used in the clinic. Based on our current and previous findings regarding changes in cell-cycle progression upon loss of MAP3K7, patients with loss or lower expression of MAP3K7 and CHD1 may instead benefit from cell-cycle inhibitors alone or in combination with hormonal therapies.

L.K. Jillson reports grants from NIH during the conduct of the study. L.U. Rodrigues reports grants from NIH during the conduct of the study, and is currently employed at and receives stock grants from Epizyme. D.M. Peehl reports grants from NIH during the conduct of the study. J.C. Costello reports grants from Boettcher Foundation and NIH/NCI during the conduct of the study. S.D. Cramer reports grants from NIH during the conduct of the study. No disclosures were reported by the other authors.

L.K. Jillson: Conceptualization, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. L.C. Rider: Conceptualization, validation, investigation, visualization, methodology, writing–review and editing. L.U. Rodrigues: Conceptualization, validation, investigation, visualization, methodology, writing–review and editing. L. Romero: Validation, investigation, project administration. A. Karimpour-Fard: Formal analysis, methodology. C. Nieto: Investigation. C. Gillette: Investigation. K. Torkko: Formal analysis. E. Danis: Formal analysis, methodology. E.E. Smith: Investigation. R. Nolley: Investigation. D.M. Peehl: Conceptualization, resources, writing–review and editing. M.S. Lucia: Resources, investigation. J.C. Costello: Formal analysis, funding acquisition, visualization, methodology, writing–review and editing. S.D. Cramer: Conceptualization, supervision, funding acquisition, investigation, methodology, writing–original draft, writing–review and editing.

This project was supported by National Institutes of Health (NIH) grants R21-CA187354 and R01-CA199741 to S.D. Cramer, U01-CA231978 to S.D. Cramer and J.C. Costello, American Cancer Society (126572-PF-14-223-01-CCE) to L.C. Rider, Boettcher Foundation Webb-Waring grant to J.C. Costello, and University of Colorado Cancer Center Fellowship to L.K. Jillson. The University of Colorado Anschutz Medical Campus Genomics and Microarray Core is supported in part by the Genomics and Microarray Shared Resource of Colorado's NIH/NCI Cancer Center Support Grant P30CA046934 and by NIH/NCATS Colorado CTSA Grant Number UL1 TR001082. This study was also partly supported by the National Institutes of Health P30CA046934 Bioinformatics and Biostatistics Shared Resource Core. Contents are the authors' sole responsibility and do not necessarily represent official NIH views.

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.