Abstract
Castration-resistant prostate cancer can be treated with the antiandrogen enzalutamide, but responses and duration of response are variable. To identify genes that support enzalutamide resistance, we performed a short hairpin RNA (shRNA) screen in the bone-homing, castration-resistant prostate cancer cell line, C4-2B. We identified 11 genes (TFAP2C, CAD, SPDEF, EIF6, GABRG2, CDC37, PSMD12, COL5A2, AR, MAP3K11, and ACAT1) whose loss resulted in decreased cell survival in response to enzalutamide. To validate our screen, we performed transient knockdowns in C4-2B and 22Rv1 cells and evaluated cell survival in response to enzalutamide. Through these studies, we validated three genes (ACAT1, MAP3K11, and PSMD12) as supporters of enzalutamide resistance in vitro. Although ACAT1 expression is lower in metastatic castration-resistant prostate cancer samples versus primary prostate cancer samples, knockdown of ACAT1 was sufficient to reduce cell survival in C4-2B and 22Rv1 cells. MAP3K11 expression increases with Gleason grade, and the highest expression is observed in metastatic castration-resistant disease. Knockdown of MAP3K11 reduced cell survival, and pharmacologic inhibition of MAP3K11 with CEP-1347 in combination with enzalutamide resulted in a dramatic increase in cell death. This was associated with decreased phosphorylation of AR-Serine650, which is required for maximal AR activation. Finally, although PSMD12 expression did not change during disease progression, knockdown of PSMD12 resulted in decreased AR and AR splice variant expression, likely contributing to the C4-2B and 22Rv1 decrease in cell survival. Our study has therefore identified at least three new supporters of enzalutamide resistance in castration-resistant prostate cancer cells in vitro.
Introduction
Early-stage prostate cancer is an androgen-dependent disease, and advanced prostate cancer is largely treated with androgen deprivation therapy, which targets the androgen receptor (AR). Inevitably, tumors progress to androgen deprivation therapy–resistant prostate cancer, clinically referred to as castration resistant. In castration-resistant prostate cancer, AR signaling continues through multiple mechanisms including AR full-length (AR-FL) over-expression, increases in androgen production, and induction of constitutively active AR splice variants (AR-V) (1). Currently, abiraterone acetate and enzalutamide are standard therapies in metastatic castration-resistant prostate cancer (2, 3). Enzalutamide is well tolerated and effective, extending castration-resistant prostate cancer patient survival by 5 months versus placebo after chemotherapy (4) and delaying initiation of chemotherapy by 18 months versus placebo (5). Nevertheless, tumors progress to enzalutamide-resistant castration-resistant prostate cancer. How these tumors progress to resistance is an area of active investigation, but both AR-dependent and AR-independent mechanisms have been proposed.
The purpose of our studies was to identify genes that support resistance to enzalutamide and identify gene products (proteins) that once targeted could resensitize tumors to enzalutamide. We first selected a bone metastasis–derived castration-resistant prostate cancer model, C4-2B (6), which upregulates AR-FL and AR-V7 in response to continuous enzalutamide treatment (7), recapitulating an enzalutamide-resistance mechanism observed in human patients. We then performed a short hairpin RNA (shRNA) screen, comparing the response of vehicle-treated and enzalutamide-treated C4-2B cells and assessed which genes, when knocked down, resulted in increased cell death. Through this approach, we identified 11 genes (TFAP2C, CAD, SPDEF, EIF6, GABRG2, CDC37, PSMD12, COL5A2, AR, MAP3K11, and ACAT1), of which ACAT1, MAP3K11, and PSMD12 appear the most promising in validation studies.
Materials and Methods
shRNA screen
The 27k Module 1 DECIPHER lentiviral shRNA library (Cellecta, DHPAC-M1-P) was transduced into C4-2B cells, targeting approximately 5,043 genes, whose gene products represent members of signal transduction pathways and approved drug targets. Cells were transduced with a pooled lentiviral shRNA library, generated per the manufacturer's instructions, and then split into three groups. Cells in the first group were collected and represent the initial population of shRNA quantity (two replicates, 3095 YY1–2). The cells from the second group served as an enzalutamide-negative control group [cell culture in the enzalutamide vehicle control (DMSO) for 6 days; three replicates, 3095 YY3–5), and cells from the third group were cultured in the presence of 100 nmol/L enzalutamide for 6 days (three replicates, 3095 YY6–8). We chose the 100 nmol/L dose as the IC50 of enzalutamide in LNCaP cells, the androgen-dependent precursor to C4-2B, is 36 nmol/L. Thus, an induction of sensitivity at 100 nmol/L would reflect a significant induction of sensitivity to enzalutamide in this castration-resistant cell line. Genomic DNA was isolated, and shRNA was quantified by deep sequencing. The ratio of the abundance of each shRNA in the third group versus both control groups was calculated. A given shRNA was considered a ‘‘hit’’ if it showed at least a 2-fold abundance decrease relative to both enzalutamide-negative control and initial samples. Data have been deposited in GEO as GSE156816.
Computational workflow
We performed quality control on the sequenced data using FastQC (8). We exported barcode and gene annotation information for each of the targeted 5,043 genes from DECIPHER's Bar-code Analyzer and Deconvoluter software (http://www.decipherproject.net/software/) and wrote in-house scripts to count the instances of each barcode and to match the barcode to annotation information. The scripts are given in the Supplementary Methods, section 3. Next, we used EdgeR (9, 10) and Limma (11, 12) to test for differential expression based on gene (barcode) count. After verifying that all of our samples had a similar sequencing depth, we used the Limma-trend test to identify which barcodes are decreased in enzalutamide-treated versus vehicle-treated and untreated cells. We filtered the genes for log fold-change differences of −0.2 or less, and FDR-corrected P < 0.2, in both Enza (EnzPos) versus Initial, and EnzPos versus DMSO (EnzNeg). Eleven genes satisfied those criteria in both comparisons and were selected for later validation. Additional information can be found in the bioinformatics supplement (Supplementary Methods).
Human clinical data
To assess the expression of our screen-identified genes during prostate cancer progression, we used the Prostate Cancer Transcription Atlas web tool (http://www.thepcta.org/), which includes 1,321 clinical specimens from 38 prostate cancer cohorts, detailed here (13). Publication-ready images were downloaded. To evaluate the frequency of alterations of our putative drivers in metastatic castration-resistant prostate cancer, we used cBioPortal (14, 15) to interrogate a set of publicly available data (16) with associated outcome and gene-expression data (17). This data set also includes an AR activity score, which is a measure of how active AR is, as well as a neuroendocrine prostate cancer (NEPCa) score, a measure of how neuroendocrine-like the gene-expression pattern is. For the (i) gene-expression comparison between naïve and exposed patients and (ii) correlation with AR and NEPCa activity scores, raw data were downloaded and plotted using GraphPad Prism. All other plots were exported as is from cBioPortal.
Cell culture
LNCaP clone FGC, 22Rv1, PC-3, and DU145 cells were purchased from ATCC (CRL-2876, CRL-2505, CRL-1435, and HTB-81). C4-2B (6) cells were provided by Drs. Ruoxiang Wu and Leland Chung (Cedars-Sinai); C4-2B CTRL and C4-2B MDVR (7) cells were provided by Dr. Allen Gao (University of California, Davis); CWR-R1 CTRL and CWR-R1 EnzR (18) cell lines were provided by Dr. Donald Vander Griend (University of Illinois, Chicago). All cell lines were short tandem repeat validated prior to use and were Mycoplasma tested using LookOut Mycoplasma PCR Detection Kit (Sigma) following expansion of stocks. LNCaP, 22Rv1, PC-3, and DU145 cells were cultured as previously described (19). CWR-R1-Ctrl cells were maintained in 10% FBS RPMI-1640 + L-glutamine, whereas CWR-R1-EnzR cells were similarly maintained with the addition of 10 μmol/L enzalutamide. C4-2B-Ctrl and C4-2B-MDVR cells were maintained in 10% FBS RPMI-1640 + L-glutamine (Gibco).
siRNA transient transfections
Cells were transfected with two siRNAs (Ambion Silencer Select, Thermo Fisher; see Table 1) per gene in Lipofectamine RNAi Max transfection reagent (Invitrogen) and OptiMEM medium (Gibco, Life Technologies). For cell survival assays, cells were transfected with 1 pmol/L siRNA for 24 hours and then cells were treated with either 1:1,000 DMSO (vehicle) or 10 μmol/L enzalutamide for 6 days in RPMI medium, with a media change at day 3 with eight biological replicates. For protein expression studies, cells were transfected with 30 pmol siRNA for 3 days or 6 days.
siRNA constructs.
Quantitative PCR (qPCR)
Total RNA was extracted from cells using the RNeasy Plus Mini kit (Qiagen) according to the manufacturer's instructions. RNA was reverse transcribed to cDNA using the Bio-Rad iScript cDNA Synthesis Kit following the manufacturer's instructions. Following reverse transcription, qPCR was performed to quantitate RNA levels in C4-2B and 22Rv1 cells and to establish the efficacy of siRNA constructs. qPCR was performed using the SYBR GREEN PCR Master Mix kit (Bio-Rad). Relative expressions of total RNA were normalized to endogenous control GAPDH. All samples were run with two technical replicates and six biological replicates. No reverse transcriptase controls and no transcript controls were run in duplicate along with each primer in every experiment. Unpublished primers were designed using Primer-BLAST (20) to span exon–exon junctions when possible and identify as many transcript isoforms as possible. All primers were tested for amplification efficiency by using a standard curve generated from four 1:10 serial dilutions of cDNA prior to experiments. Efficiency was calculated using the formula E = −1 + 10(−1/slope), where E is the efficiency. All primers used reported >80% efficiency according to this method. Primers are listed in Table 2.
qPCR primers.
Cell survival assays
Cell survival in response to knockdowns and enzalutamide was assessed using crystal violet staining. The crystal violet staining protocol was used as previously described (21). In brief, 1 × 104 cells were seeded per well in 100-μL RPMI-1640 with L-glutamine in 10% fetal bovine serum and underwent described drug treatments with eight biological replicates. At the end of the treatment course, wells were fixed in 4% paraformaldehyde in PBS (Alfa Aesar) and stained with 0.05% crystal violet stain (Ricca). Cells were washed with deionized water and dried for 16–24 hours before being imaged. Destaining for quantification was performed using 10% acetic acid and absorbance was read at 590 nm using a spectrophotometer (SpectraMax iD3, Molecular Devices). An average value of nontargeting vehicle was generated, and each value was divided by this average value to generate a normalized fold-change cell survival reading.
Western blotting
Protein isolation was performed using working RIPA buffer (120 mmol/L NaCl, 50 mmol/L pH 8.0 Tris, 0.5% NP-40, 1 mmol/L EGTA) containing protease and phosphatase inhibitors (100 μg/mL PMSF, 1 mmol/L NaOrVa, 50 μg/mL aprotinin, 50 μg/mL leupeptin). Protein concentration was determined using a Bio-Rad protein concentration assay dye and absorbance read at 595 nm on a spectrophotometer (SpectraMax ID3, Molecular Devices). Total protein (20 μg) in β-mercaptoethanol–containing loading buffer was added per well onto 4%–12% SDS-PAGE gels (NuPAGE) and transferred to PVDF (Bio-Rad) or nitrocellulose (Amersham) membrane. Nitrocellulose membranes were stained for total protein using Ponceau S (Boston BioProducts). Membranes were blocked with 10% BSA or milk fat in TBST for 1 hour at room temperature. Primary antibodies from Abcam for AR (ab74272), AR-V7 (ab198394), phosphorylated AR-Ser650 (ab47563), ACAT1 (ab168342), MAP3K11 (ab51068), PSDM12 (ab229930), and GAPDH (Invitrogen, 6C5) were used at recommended concentrations in 2.5% milk fat or 2.5% bovine serum albumin overnight at 4°C. Anti-rabbit and anti-mouse secondary antibodies (GE Healthcare) were used in 2.5% milk fat for 1 hour at room temperature. Membranes were imaged via chemiluminescence reagents (Super Signal West Pico Plus, Thermo Fisher) on a digital imager (ChemiDoc Touch, Bio-Rad).
Drug treatments
Cell response to the MAP3K11 inhibitor CEP-1347 was assessed in crystal violet cell survival assays. 1 × 104 cells were seeded per well in 100-μL RPMI-1640 with L-glutamine in 10% fetal bovine serum and allowed to settle. CEP-1347 (Tocris, cat. No. 4924) reagent was resuspended in DMSO for 1 mmol/L stock. Working solutions of 80, 40, 20, and 10 μmol/L were made with DMSO, then diluted 1:100 in RPMI-1640 with L-glutamine in 10% fetal bovine serum for final concentrations of 800, 400, 200, and 100 nmol/L, respectively. Enzalutamide (MDV3100; Selleckchem CAS No. 915087-33-1) was resuspended in DMSO for a working concentration of 100 mmol/L and then diluted 1:1,000 in media. DMSO control media contained equal amounts of solvent as drug treatment samples. Drug-containing media (100 μL) were plated per well for 6 days with new drug treatments performed on the third day. Cell survival at the end of drug treatment course was assessed via crystal violet staining as described.
Statistical analysis
Statistical analysis was performed using GraphPad Prism. For analysis of cell survival experiments, following the removal of outliers using the ROUT method, vehicle-to-vehicle, enzalutamide-to-enzalutamide comparisons were made using the Kruskal–Wallis test as data were not always normally distributed. Comparisons between gene expression of abiraterone and/or enzalutamide exposed and naïve patients used the nonparametric Kruskal–Wallis test as data were not normally distributed. qPCR ddCT values were log transformed and analyzed by one-way ANOVA analysis. Correlations between gene expression (FPKM capture) and AR and NEPCa activity scores were calculated using nonparametric Spearman correlations. Clinical data from the Prostate Cancer Transcription Atlas web tool included statistical analysis.
Results
To identify genes that support enzalutamide resistance in castration-resistant prostate cancer in vitro, we transduced C4-2B cells with a lentiviral shRNA library that targeted 5,043 genes whose products act in signal transduction and are drug targets. The products of these genes include kinases, enzymes, receptors, and other proteins that already have targeted therapies. C4-2B cells were transduced with the shRNA library and split into three groups, one collected at day 0, one treated with vehicle for 6 days, and one treated with enzalutamide for 6 days. We used high-throughput sequencing to quantitate the number of shRNA barcodes in each sample. We were predominantly interested in the identity of the shRNA barcodes that had decreased in abundance in the enzalutamide-treated cells versus nontreated and vehicle-treated cells, as the loss of these barcodes reflected the death, cell-cycle arrest, growth inhibition, or senescence of cells in which these genes were essential. Through this analysis, we identified 11 genes whose silencing resulted in increased cell death in response to enzalutamide: TFAP2C, CAD, SPDEF, EIF6 (ITGB4BP), GABRG2, CDC37, PSMD12, COL5A2, AR, MAP3K11, and ACAT1 (Fig. 1A). As prostate cancer is driven by AR, and enzalutamide targets AR, the presence of AR in our gene list was an encouraging sign. The fold-change differences were more dramatic between shRNA counts from enzalutamide-treated cells versus the initial population of shRNAs (Fig. 1B) as compared with shRNA counts from vehicle-treated cells (Fig. 1C). This likely reflects the importance of some of these genes in cell survival in general and/or in response to DMSO.
Identification of genes that support enzalutamide resistance in C4-2B cells. A, Log-scaled shRNA barcode counts of differentially expressed shRNA barcodes (genes) in C4-2B cells in response to enzalutamide (Enza) and vehicle (DMSO) with fold change ≤ −0.2 and FDR-corrected P < 0.2. Cells collected at the start of the experiment (initial) represent the starting pool of shRNA barcodes. ITGB4BP is now EIF6. B and C, Underrepresented shRNA barcodes (genes) between enzalutamide-treated and untreated (initial, B) and vehicle-treated (DMSO, C) cells, comparing log fold change in shRNA counts and −log10(FDR-corrected P value). Volcano plots of 11 genes that support enzalutamide resistance. D, Most supporters of enzalutamide resistance are amplified in metastatic castration-resistant prostate cancer. Oncoprint of 429 patients from Abida et al. (16) in cBioPortal (14, 15) evaluating frequency of amplifications, deletions, and mutations. E and F, AR does not strongly regulate ACAT1, MAP3K11, or PSMD12. Knockdown of AR for 3 or 6 days in C4-2B (E) and 22Rv1 (F) cells. G, Comparison of AR, AR-V7, ACAT1, MAP3K11, and PSMD12 expression in androgen-dependent (LNCaP), castration-resistant [C2-2B (LNCaP-derivative) and 22Rv1], matched enzalutamide-sensitive (CWR-R1 CTRL, C4-2B CTRL) and resistant (CWR-R1 EnzR, C4-2B MDVR), and androgen-independent (DU145, PC-3) prostate cancer cell lines.
Our first step was to evaluate whether our 11 genes were expressed in castration-resistant prostate cancer. First, we used cBioPortal (14, 15) to interrogate a set of publicly available data from metastatic castration-resistant prostate cancer patient samples (16) with outcome and gene-expression data. Unsurprisingly, AR was the most frequently altered of these genes, with 59% of samples bearing amplifications and mutations (Fig. 1D). For the other genes, alterations were rarer, occurring in 2.8% to 6% of cases, depending on the gene. With the exception of ACAT1, which has more deletions, most of our gene list alterations were amplifications.
We also examined whether our putative resistance supporter genes increased after abiraterone and enzalutamide treatment. For this analysis, we included patients who were exposed or naïve to abiraterone and enzalutamide, and excluded patients on treatment or whose exposure status was unknown from Abida and colleagues (16). Through this analysis, we determined that AR expression was dramatically increased in patients who had been exposed to abiraterone and enzalutamide (Supplementary Fig. S1), as reported (16). However, there was no difference in gene expression of our putative resistance drivers in patients who had received abiraterone and enzalutamide versus patients who had not (Supplementary Fig. S1). With the exception of AR, all of our genes exhibited gains and losses in both patient groups (Supplementary Figs. S2 and S3).
To validate our putative supporters of enzalutamide resistance in vitro, we turned to siRNA knockdown studies. To complement validation studies in C4-2B cells, we also used 22Rv1 cells, which express AR-V, including AR-V7 and AR-V9, and are enzalutamide resistant (22, 23). We used two siRNA constructs per gene and evaluated how knockdown of each gene affects enzalutamide sensitivity and full-length AR (AR-FL) and AR-V7 levels. Through this analysis, we validated three candidate genes (ACAT1, MAP3K11, and PSMD12) in two cell lines as supporters of enzalutamide resistance in vitro. The remaining genes (TFAP2C, CAD, SPDEF, EIF6, GABRG2, CDC37, and COL5A2) did not reproducibly inhibit survival in one or both cell lines or had no effect in our validation assay.
In our validation studies, we first assessed whether AR regulated expression of our genes of interest (Supplementary Fig. S4). Analysis of previously published R1881-induced gene expression in LNCaP cells (24) revealed only SPDEF is androgen regulated, and none of our 11 genes were identified as induced by enzalutamide treatment (25). Our two AR constructs both efficiently knocked down expression of AR, with construct No. 2 being more efficient (Fig. 1E and F; Supplementary Fig. S4A). Importantly, both of our knockdown constructs targeted AR-V7. Although AR-V7 expression has been reported in C4-2B cells (26), under our conditions, the expression was minimal. We next evaluated the consequences of AR knockdown on our genes of interest. Knockdown of AR reduced gene expression of MAP3K11, but it did not reduce the expression of ACAT1 or PSMD12 (Supplementary Fig. S4B–S4D). We also evaluated whether knockdown of AR altered protein expression of ACAT1, MAP3K11, and PSMD12, and observed no significant changes (Fig. 1E and F). Interestingly, on a gene-expression level, knockdown of many of our genes resulted in dramatic changes in expression of other members of our gene list (Supplementary Figs. S4 and S5).
To examine the expression of ACAT1, MAP3K11, and PSMD12 during prostate cancer progression, we examined a panel of androgen-dependent, castration-resistant, and androgen-independent prostate cancer cells (Fig. 1G). LNCaP cells are the androgen-dependent parental cell line from which castration-resistant C4-2B cells are derived (6). The C4-2B CTRL and CWR-R1 CTRL lines are the approximately passage-matched controls for the enzalutamide-resistant C4-2B MDVR (7) and CWR-R1 EnzR (18) cell lines. ACAT1 expression decreases slightly between matched androgen-dependent and castration-resistant cells, and enzalutamide-sensitive and -resistant cell lines. AR-independent cell lines express ACAT1 at variable levels. MAP3K11 expression is similar between LNCaP and C4-2B cells, but it does increase between cell lines as they progress from enzalutamide-sensitive to resistant. MAP3K11 expression is higher in PC-3 versus DU145 androgen-dependent prostate cancer cells. Finally, PSMD12 is uniformly expressed between the cell lines.
Our first gene of interest, ACAT1, encodes Acetyl-CoA acetyltransferase 1 (ACAT1). Although gene expression for ACAT1 does not increase between benign and primary prostate cancer, ACAT1 expression decreases in metastatic castration-resistant prostate cancer versus primary prostate cancer (Fig. 2A–C). Although ACAT1 message appears to decrease between primary and metastatic castration-resistant prostate cancer, on a protein level, ACAT1 expression reportedly increases (27, 28). For ACAT1, both siACAT1 constructs were efficient, and knockdown of ACAT1 did not dramatically alter AR or AR-V7 protein expression (Fig. 2D and E). Over a 6-day treatment time course, knockdown of ACAT1 resulted in a dramatic reduction in cell survival in C4-2B cells (55%–76% siACAT1 enzalutamide-treated vs. nontargeting enzalutamide-treated cells; Fig. 2F). In 22Rv1 cells, this effect was more modest, achieving a 14% survival reduction versus nontargeting enzalutamide-treated cells (Fig. 2G). To evaluate how ACAT1 could be supporting enzalutamide resistance, we again turned to the Abida and colleagues data set (16). Using cBioPortal (14, 15), we selected patients with metastatic castration-resistant prostate cancer who were naïve to abiraterone and enzalutamide treatment and compared their AR activity and neuroendocrine (NE/NEPCa) activity scores. In clinical samples, ACAT1 expression correlates with the AR activity score (Fig. 2H) but not the NE/NEPCa score (Fig. 2I).
ACAT1 supports enzalutamide resistance in vitro. A–C, ACAT1 expression decreases during progression to metastatic castration-resistant prostate cancer. Lollipop (A), box plot (B), and line plot of mean trend (C) of ACAT1 log2 median-centered and quantile scaled normalized gene-expression values in benign prostate, prostate cancer, and metastatic castration-resistant prostate cancer patient samples. GS: Gleason score; mCRPC: metastatic castration-resistant prostate cancer. D and E, Knockdown of ACAT1 does not strongly affect AR expression. C4-2B (D) and 22Rv1 (E) cells were transfected with one of two nontargeting siRNA (siNT) or ACAT1-targeting (siACAT1 No. 1 or No. 2) and ACAT1, AR, AR-V7 (in 22Rv1 cells) was evaluated with GAPDH used as a loading control. F and G, Knockdown of ACAT1 in castration-resistant prostate cancer cells increases cell death in response to enzalutamide. C4-2B (F) and 22Rv1 (G) were transfected with one of two nontargeting siRNA (siNT) or ACAT1-targeting (siACAT1) siRNAs and challenged with 10 μmol/L DMSO (vehicle) or enzalutamide for 6 days. Comparisons between DMSO and DMSO, enzalutamide to enzalutamide using Kruskal–Wallis test with Dunn multiple comparisons test. *, P < 0.05; **, P < 0.01; ***, P < 0.001; **, P < 0.0001. H, ACAT1 expression positively correlates with AR activity metastatic castration-resistant prostate cancer patient tissues. Correlation between ACAT1 mRNA expression and the AR activity score evaluated by Spearman correlation in 106 abiraterone- and enzalutamide-naïve metastatic castration-resistant prostate cancer patients. I, ACAT1 expression does not correlate with NEPCa activity metastatic castration-resistant prostate cancer patient tissues. Correlation between ACAT1 mRNA expression and the NEPCa activity score was evaluated by Spearman correlation.
Our second gene of interest was MAP3K11, which encodes MAP3K11, also known as mixed lineage kinase 3 (MLK3). MAP3K11 expression increases dramatically during prostate cancer progression (Fig. 3A–C). Both of the MAP3K11 targeting constructs knocked down MAP3K11 expression, with construct No. 1 being more efficient (Fig. 3D and E). The more efficient MAP3K11 knockdown resulted in more cell death in response to enzalutamide (Fig. 3F and G). In C4-2B cells, knockdown of MAP3K11 with construct No. 1 and treatment with enzalutamide resulted in a 73% decrease in cell survival versus nontargeting enzalutamide-treated cells, whereas in 22Rv1 cells this reduction was more modest with a reduction of 28%. Consistent with previous reports, when we knocked down MAP3K11, we observed no decrease in AR (29) or AR-V7 expression, but we did see a loss of AR-Ser650 phosphorylation in 22Rv1 cells; this decrease was more pronounced with greater MAP3K11 ablation (Fig. 3E). Phosphorylation of AR-Ser650 promotes maximal AR transactivation activity (30), suggesting MAP3K11 could be supporting enzalutamide resistance through activation of AR. Interestingly, in C4–2B cells, we did not observe AR-Ser650 phosphorylation under normal cell culture conditions, consistent with previous reports where AR-Ser650 phosphorylation is induced by phorbol 12-myristate 13-acetate (PMA; ref. 29). In the abiraterone- and enzalutamide-naïve patients, MAP3K11 expression was not statistically correlated with the AR activity score (Fig. 3H), but it was negatively correlated with the NE/NEPCA score (Fig. 3I).
MAP3K11 supports enzalutamide resistance in vitro. A–C, MAP3K11 expression increases during progression to metastatic castration-resistant prostate cancer. Lollipop (A), box plot (B), and line plot of mean trend (C) of MAP3K11 log2 median-centered and quantile scaled normalized gene-expression values in benign prostate, prostate cancer, and metastatic castration-resistant prostate cancer patient samples. GS: Gleason score; mCRPC: metastatic castration-resistant prostate cancer. D and E, Knockdown of MAP3K11 does not strongly affect AR expression but it does reduce AR-Serine 650 (AR-Ser650) phosphorylation. C4-2B (D) and 22Rv1 (E) cells were transfected with one of two nontargeting siRNA (siNT) or MAP3K11-targeting (siMAP3K11 No. 1 or No. 2) and ACAT1, AR, AR-V7 (in 22Rv1 cells), and phosphorylated AR-Ser650 was evaluated with GAPDH used as a loading control. F and G, Knockdown of MAP3K11 in castration-resistant prostate cancer cells increases cell death in response to enzalutamide. C4-2B (F) and 22Rv1 (G) were transfected with one of two nontargeting siRNA (siNT) or MAP3K11-targeting (siMAP3K11) siRNAs and challenged with 10 μmol/L DMSO (vehicle) or enzalutamide for 6 days. Comparisons between DMSO and DMSO, enzalutamide to enzalutamide using Kruskal–Wallis test with Dunn multiple comparisons test. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. H, MAP3K11 expression does not correlate with AR activity metastatic castration-resistant prostate cancer patient tissues. Correlation between MAP3K11 mRNA expression and the AR activity score evaluated by Spearman correlation in 106 abiraterone and enzalutamide-naïve metastatic castration-resistant prostate cancer patients. I, MAP3K11 expression inversely correlates with NEPCa activity metastatic castration-resistant prostate cancer patient tissues. Correlation between MAP3K11 mRNA expression and the NEPCa activity score evaluated by Spearman correlation. J, Knockdown or inhibition of MAP3K11 reduces AR-Ser650 phosphorylation. Comparison of AR-Ser650 phosphorylation in 22Rv1 cells transfected with nontargeting siRNA (siNT) or MAP3K11 targeting siRNA (siMAP3K11) to vector or MAP3K11 inhibitor (CEP-1347) treated cells. Importantly, AR and AR-V levels do not change. K and L, Inhibition of MAP3K11 with CEP-1347 potentiates enzalutamide treatment. Treatment of C4-2B (K) and 22Rv1 (L) with increasing concentrations of CEP-1347 with 10 μmol/L DMSO (vehicle) or enzalutamide. Analysis by two-way ANOVA with Sidak's multiple comparisons test comparing 0 nmol/L CEP-1347 (DMSO) to varying CEP-1347 concentrations (black to black) and 0 nmol/L CEP-1347 (DMSO) plus enzalutamide to varying CEP-1347 concentrations plus enzalutamide (red to red). For both C4-2B and 22Rv1 cells, there is a significant interaction between the drug and concentration (P < 0.0001 and P = 0.0074, respectively). *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.
As several inhibitors have been developed for the MLK family, we next evaluated whether one of these could be used in combination with enzalutamide. CEP-1347 was initially developed to prevent HIV-1–associated neurocognitive disorders (HAND) and Parkinson disease (31–33). Treatment of 22Rv1 cells with CEP-1347 did not reduce AR or AR-V7 expression, but it did reduce AR-Ser650 phosphorylation to similar levels as MAP3K11 knockdown (Fig. 3J). Indeed, combination therapy with increasing concentrations of CEP-1347 induced increased cell death in response to 10 μmol/L enzalutamide- versus vehicle-treated cells in C4-2B and 22Rv1 cells in a dose-dependent manner (Fig. 3K and L).
Our final gene is PSMD12, which encodes Proteasome 26S Subunit, Non-ATPase 12, a component of the 26S Proteasome. PSMD12 expression does not change between benign, primary, and metastatic prostate cancer (Fig. 4A–C). In our validation studies, both constructs were highly efficient at knocking down PSMD12 gene expression in both cell lines (Fig. 4D and E). In cell survival assays, we observed that C4-2B cells were exquisitely sensitive to PSMD12 knockdown, with knockdown driving an 85%–91% decrease in cell survival in C4-2B cells and a 28% reduction in survival in 22Rv1 cells (Fig. 4F and G). This increased sensitivity was likely due to a loss of AR and AR-V7 expression, as when PSMD12 expression decreased, AR and AR-V7 expression decreased (Fig. 4C and D). In the clinical data, PSMD12 expression was not correlated with AR activity or NE/NEPCa activity scores (Fig. 4E and F).
PSMD12 supports enzalutamide resistance in vitro. A–C, PSMD12 expression is similar in benign, primary, and metastatic castration-resistant prostate cancer. Lollipop (A), box plot (B), and line plot of mean trend (C) of PSMD12 log2 median-centered and quantile scaled normalized gene-expression values in benign prostate, prostate cancer, and metastatic castration-resistant prostate cancer patient samples. GS: Gleason score; mCRPC: metastatic castration-resistant prostate cancer. D and E, Knockdown of PSMD12 decreases AR and AR-V expression. C4-2B (D) and 22Rv1 (E) cells were transfected with one of two nontargeting siRNA (siNT) or PSMD12-targeting (siPSMD12 No. 1 or No. 2) and PSMD12, AR, AR-V7 (in 22Rv1 cells) was evaluated with GAPDH used as a loading control. F and G, Knockdown of PSMD12 in castration-resistant prostate cancer cells increases cell death in response to enzalutamide. C4-2B (F) and 22Rv1 (G) were transfected with one of two nontargeting siRNA (siNT) or PSMD12-targeting (siPSMD12) siRNAs and challenged with 10 μmol/L DMSO (vehicle) or enzalutamide for 6 days. Comparisons between DMSO and DMSO, enzalutamide to enzalutamide using Kruskal–Wallis test with Dunn multiple comparisons test. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. H and I, PSMD12 expression does not correlate with AR activity or NEPCa activity in 106 abiraterone and enzalutamide-naïve metastatic castration-resistant prostate cancer patient tissues. Correlation between PSMD12 mRNA expression and the AR activity score (H) or NEPCa activity (I) evaluated by Spearman correlation.
Discussion
Previous approaches to understanding enzalutamide resistance have focused on identifying genes that are altered by enzalutamide treatment (7, 18, 23, 34–37) or predict response (38). Another approach has been to select genes that are frequently deleted in primary or metastatic prostate cancer and determine whether their loss alters enzalutamide response in LNCaP xenografts (39). Conversely, we undertook a functional screen of a castration-resistant prostate cancer cell line in order to identify genes whose expression is required for surviving enzalutamide exposure.
Our shRNA screen in C4-2B cells identified 11 putative supporters of enzalutamide resistance (TFAP2C, CAD, SPDEF, EIF6, GABRG2, CDC37, PSMD12, COL5A2, AR, MAP3K11, and ACAT1). Although AR expression was frequently increased in response to abiraterone and enzalutamide, we observed no statistically significant changes in the other genes in patient tissues. It is plausible that there is no increase in gene expression, but there may be other mechanisms by which these proteins increase in expression, such as increased stability or activity via phosphorylation. It is also possible that if we compared enzalutamide responder to nonresponders or matched patient samples pre- and post-treatment, we would observe an increase in these genes. Additional studies are planned to address these questions.
In our in vitro validation studies, our siRNA constructs efficiently targeted all of our genes of interest, and revealed that, at least on a gene-expression level, there appears to be considerable cross-talk between these genes. For example, knockdown of MAP3K11 results in decreased expression of genes that are regulated by AR (like SPDEF), likely by reducing its transcriptional efficiency, or genes affected by AR knockdown. Whether this cross-talk between our gene products occurs, which interactions are mediated by AR, and whether this translates to alterations in protein expression remains to be determined. Nevertheless, through our validation studies, we verified that transient knockdown of ACAT1, MAP3K11, or PSMD12 could sensitize castration-resistant prostate cancer cells to enzalutamide.
In general, C4-2B cells were much more sensitive to knockdown of these three genes in combination with enzalutamide treatment. There are two possible explanations for this observation. The first is that we identified these genes as supporting enzalutamide resistance in C4-2B cells through our shRNA screen, and thus these cells are more dependent on these pathways. Alternatively, it is possible that the high level of AR-V in 22Rv1 cells, which confers enzalutamide resistance by removing dependency on ligand binding, simply makes 22Rv1 cells much more resistant. As knockdown of PSMD12 reduces AR and AR-V7 expression, and 22Rv1 cells do not become dramatically more sensitive to enzalutamide, this would suggest some of these genes might only be required for enzalutamide resistance in C4-2B cells, and by extension, only a subset of prostate cancers.
One of our enzalutamide-resistance supporting genes, ACAT1, encodes the protein ACAT1 that has several roles in the mitochondria. First, it serves as the final enzyme in isoleucine metabolism, converting 2-methylacetoacetyl-CoA into propionyl-CoA and acetyl-CoA (40). ACAT1 also functions in ketone body metabolism in which its substrate is acetoacetyl-CoA (40). Produced acetyl-CoA is shuttled into the Krebs cycle to be oxidized for energy production. More recently, ACAT1 has been implicated in regulating the pyruvate dehydrogenase complex, comprised of pyruvate dehydrogenase and PDH phosphatase, through acetylation (41). Importantly, knockdown of ACAT1 activity is sufficient to push cancer cells from their preferred aerobic glycolysis, favored by cancer and proliferating cells, and into oxidative phosphorylation, the energy production mechanism favored by differentiated cells. Beyond its role in metabolism, ACAT1 dysregulation could alter acetyl-CoA levels. Notably, AR can be acetylated at AR-K618 (42) in the DNA binding domain and AR-K632, AR-K633 (43) and AR-K630 (44) in the hinge region, which supports increased AR transcriptional activity and increased tumor growth in xenograft models (42, 44). Which of these mechanisms supports enzalutamide resistance remains to be determined.
In prostate, ACAT1 is associated with more aggressive prostate cancer and castration-resistant disease (27, 28). These observations, combined with our experimental data, suggest that ACAT1 is expressed in castration-resistant prostate cancer and may support enzalutamide resistance, making it an attractive therapeutic target. ACAT1 activity can be inhibited by arecoline hydrobromide (41) or the FDA-approved drug sulfasalazine (45). Unfortunately, neither arecoline hydrobromide nor sulfasalazine are specific to ACAT1.
Another gene of interest, MAP3K11, encoding MAP3K11, perhaps holds the most therapeutic potential. MAP3K11 is a serine/threonine kinase (46), which via phosphorylation of MAP2Ks (MKKs) is a regulator of the mitogen-activated protein kinases (MAPK), including JNK, ERK, and p38 (reviewed in ref. 47). MAP3K11 can act both as a kinase and as a scaffold for other kinases (47). Earlier studies identified MAP3K11 as a potent regulator of AR transcriptional activity in C4-2B cells through an RNAi phenotypic screen (29). The link between MAP3K11 and AR activity appears to be driven by phosphorylation of serine 650, located in the hinge region of AR (43). The hinge region is absent in most AR-Vs, including AR-V7, but it is present in a subset, including ARv567es (43), suggesting this posttranslational modification could regulate the activities of some AR-Vs. Based on alanine-for-serine mutation studies using human AR in COS cells, phosphorylation of AR-Ser650 is required for optimal AR transactivation activity, as without AR-Ser650 phosphorylation, transcription of an AR-responsive mammary tumor promoter was reduced by 30% (30). Although it is possible that AR-Ser650 is a direct target of MAP3K11, in vitro kinase assays revealed AR-Ser650 is phosphorylated by JNK and p38 (48), which are downstream of MAP3K11. Which of these pathways are effectors of MAP3K11 in castration-resistant prostate cancer is currently under investigation.
Of our three genes, MAP3K11 has the most specific and tested therapeutics. There are two inhibitors that are somewhat selective for MAP3K11 versus other mixed lineage kinase family members: CEP-1347 and URMC-099. CEP-1347 was initially developed to prevent HAND and Parkinson disease (31–33), and while safe and well tolerated it failed to prevent Parkinson disease progression in a phase II clinical trial (49). CEP-1347 is a fairly selective MAP3K11 inhibitor (IC50 = 23 nmol/L, ref. 32; IC50 = 6 nmol/L, ref. 50), with limited off-target effects on other MLK family members (MLK1 IC50 = 38 nmol/L and MLK2 IC50 = 51 nmol/L, ref. 32; MLK1 IC50 < 1 nmol/L and MLK2 IC50 = 2 nmol/L, ref. 50). URMC-099 is a more specific MAP3K11 inhibitor (IC50 = 14 nmol/L), with limited off-target effects on other MLK family members (MLK1 IC50 = 19 nmol/L; MLK2 IC50 = 42 nmol/L; and DLK IC50 = 150 nmol/L; ref. 50). Our current studies focus on evaluating these compounds as a therapy in conjunction with enzalutamide in enzalutamide-resistant castration-resistant prostate cancer.
Our final gene, PSMD12, has not been previously implicated in prostate cancer, but dysregulation of the proteasome has been associated with prostate cancer. Inhibition of the proteasome with therapeutics like bortezomib has been shown to increase the efficacy of first-generation antiandrogens such as bicalutamide by decreasing the expression of AR and AR-V (51) or induce sensitivity of AR-independent prostate cancer cells to etoposide (52). Unfortunately, proteasome inhibitors, to date, have proven acutely toxic, and targeting PSMD12 with currently available therapeutics is unlikely to provide a favorable risk-to-benefit ratio.
Although not all of our candidate genes validated in our follow-up studies, it should be noted that many of these hold promise and may be better evaluated in other cell lines and assays. For example, GABRG2 encodes the Gamma-Aminobutyric Acid Type A Receptor Gamma2 Subunit protein, which is a GABA receptor. Previous studies by Jin and colleagues discovered GABRG2 as a member of a prognostic 21-gene panel (NARP21) that predicted decreased overall cancer-specific survival and metastasis-free survival of patients with prostate cancer (53). Similarly, CAD, which encodes carbamoyl-phosphate synthetase II, aspartate transcarbamylase, and dihydroorotase, is associated with the synthesis of pyrimidine nucleotides, necessary for cell proliferation. CAD is regulated by the MAPK cascade (54), and knockdown of MAP3K11 reduces CAD gene expression. CAD also interacts with AR and fosters AR translocation into the nucleus, and is posited as an early marker of prostate tumor recurrence (55). Finally CDC37, which directs kinases to the HSP90 complex, has been implicated in supporting prostate cancer cell growth via increasing AR activity and activation of kinases (56). The failure to validate some of these potential supporters of enzalutamide resistance, therefore, is likely to be due to the limitations of our assay rather than their lack of importance in prostate cancer biology.
These studies have several limitations. First, we have focused on genes that are already expressed in castration-resistant prostate cancer cell lines rather than those that have emerged because of treatment with enzalutamide. We have also only focused on castration-resistant prostate cancer treated with enzalutamide, as at the initiation of our studies, enzalutamide treatment was not yet approved in the metastatic hormone-sensitive setting (57). In addition, our study has been limited to 5,043 genes that are involved in signal transduction and are drug targets, which has left out a considerable number of potential drivers, like transcription factors, that likely play an important role in enzalutamide resistance. We have also performed our validation experiments in the context of transient knockdown experiments in vitro and focused exclusively on plausible AR-centric resistance mechanisms. We hypothesize these gene products affect AR activity and AR-target gene expression in the presence of enzalutamide, which we will evaluate in future studies. It is also likely that ACAT1, MAP3K11, and PSMD12 act beyond AR to support enzalutamide resistance. Subsequent studies will focus on defining these enzalutamide-resistance drivers more holistically, including both loss-of-function and gain-of-function experiments to delineate how these genes support enzalutamide resistance both in vitro and in vivo. In summary, our studies have identified 11 genes (TFAP2C, CAD, SPDEF, EIF6, GABRG2, CDC37, PSMD12, COL5A2, AR, MAP3K11, and ACAT1) that support enzalutamide resistance in castration-resistant C4-2B cells, and we have validated three of these genes in enzalutamide-resistance in vitro (ACAT1, MAP3K11, and PSMD12).
Authors' Disclosures
S.E. Kohrt reports grants from NIH/NIGMS T32 GM008056 during the conduct of the study. T.C. Case reports grants from NIH during the conduct of the study. M.M. Grabowska reports personal fees from Synchronicity Pharma outside the submitted work. No disclosures were reported by the other authors.
Disclaimer
The content of this manuscript solely the responsibility of the authors and does not necessarily represent the official views of the NIH.
Authors' Contributions
S.E. Kohrt: Conceptualization, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. W.N. Awadallah: Conceptualization, formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing. R.A. Phillips: Data curation, formal analysis, validation, methodology, writing–original draft, writing–review and editing. T.C. Case: Resources, methodology, writing–review and editing. R. Jin: Conceptualization, resources, supervision, methodology, writing–review and editing. J.S. Nanda: Investigation, methodology, writing–review and editing. X. Yu: Conceptualization, resources, supervision, methodology, writing–review and editing. P.E. Clark: Conceptualization, supervision, funding acquisition, project administration, writing–review and editing. Y. Yi: Conceptualization, resources, supervision, funding acquisition, methodology, project administration, writing–review and editing. R.J. Matusik: Conceptualization, resources, supervision, funding acquisition, project administration, writing–review and editing. P.D. Anderson: Data curation, formal analysis, supervision, validation, visualization, methodology, writing–original draft, project administration, writing–review and editing. M.M. Grabowska: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, visualization, methodology, writing–original draft, project administration, writing–review and editing.
Acknowledgments
We would like to thank Jianghong Zhang, PhD, formerly of Vanderbilt University Medical Center, for technical assistance with the shRNA screen. The authors would like to acknowledge funding from the William L. Bray and Joe C. Davis Foundation (to R.J. Matusik) and the Vanderbilt Institute for Clinical and Translational Research (VICTR, to Y. Yi, P.E. Clark, and R.J. Matusik). The Vanderbilt Institute for Clinical and Translational Research (VICTR) is funded by the National Center for Advancing Translational Sciences (NCATS) Clinical Translational Science Award (CTSA) Program, Award Number 5UL1TR002243. We would also like to acknowledge the Case Research Institute, a joint venture between University Hospitals and Case Western Reserve University, start-up funds (to M.M. Grabowska), the Cell and Molecular Biology Training Program (T32 GM 008056 to S.E. Kohrt), and the Molecular Therapeutics Training Program (T32 GM 008803 to S.E. Kohrt).
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.
Footnotes
Note: Supplementary data for this article are available at Molecular Cancer Therapeutics Online (http://mct.aacrjournals.org/).
Mol Cancer Ther 2021;20:398–409
- Received March 29, 2020.
- Revision received September 11, 2020.
- Accepted November 17, 2020.
- Published first December 9, 2020.
- ©2020 American Association for Cancer Research.
References
Volume 20, Issue 2
