Androgen Receptor Signaling Regulates DNA Repair in Prostate Cancers

Multiple clinical trials comparing radiotherapy plus androgen deprivation therapy (ADT) versus radiotherapy alone for high-risk, and more recently intermediate-risk, prostate cancer show significant improvement in disease-free and overall survival with the addition of ADT (1, 2). Furthermore, posttreatment biopsy series demonstrate improved local control when ADT is added to radiotherapy (3). In light of these and other studies, combining ADT with radiotherapy for high-risk prostate cancer has been the standard of care for nearly 20 years, yet the mechanism by which ADT improves patient outcomes remains unknown. Furthermore, it is unknown whether ADT benefits a subset of patients substantially or all patients with prostate cancer to a smaller degree.

Given the compelling body of clinical evidence, many have sought to elucidate how inhibiting the androgen receptor (AR) potentiates ionizing radiation. Using in vitro and in vivo models, the addition of ADT to ionizing radiation has been shown to increase prostate cancer cell death (4). A number of mechanisms have been explored to explain the increase in cell death when ADT is combined with ionizing radiation, including decreased tumor cell hypoxia (5), decreased DNA repair (6), or simply decreased AR-mediated cell growth independent of direct synergy (7). Surprising light has recently been shed on additional interrelationships between AR and DNA repair, including a role for AR in mediating prostate cancer–specific translocations following high-dose ionizing radiation (8) and the discovery that the DNA repair protein, PARP1, is an important cofactor for AR transcriptional activity (9).

Defining the mechanism by which ADT increases prostate cancer radioresponse has never been more clinically relevant given the recently demonstrated success of second-generation antiandrogens in the treatment of castration-resistant patients (10, 11). Given the clinical potential of deploying these new agents as part of radiotherapy for primary disease coupled with the ability now to leverage prostate cancer genomics data to help define genetic mechanisms in a less biased way, it is critical to reexamine the basic biologic question of how AR signaling promotes prostate cancer radioresistance.

We began with an unbiased query of how gene expression is perturbed when a clinically validated xenograft model of castration-resistant prostate cancer, LNCaP-AR, is treated with the second-generation antiandrogen ARN-509 (12, 13). After 4 days of treatment with ARN-509, transcriptome analysis was performed by Illumina HT12 expression array. Standard gene set enrichment analysis (GSEA) was performed, and, to our surprise, three out of the top 10 gene sets that were enriched in the control versus ARN-509–treated groups represented DNA repair gene sets; in total, six DNA repair gene sets comprised the top 50 enriched gene sets (Fig. 1A and Supplementary Table S1).

Given the unexpected result of ADT decreasing DNA repair gene expression in the castration-resistant model, we next sought to determine whether there was an association of AR transcriptional output with DNA repair genes in primary, castration-sensitive human prostate tumors (14). This set of primary prostate cancer tumors represents the appropriate group to analyze, as this is the disease state treated with ADT and radiotherapy. First, we determined the variance of canonical AR transcriptional output, using a well-known signature derived by Hieronymus and colleagues (14, 15) across primary human prostate cancer tumors. Again, to our surprise, we observed a large spectrum of canonical AR output (Fig. 1B). Given this variance of AR output across primary prostate cancer tumors, we next asked whether there was a correlation between a composite score of canonical AR output, as previously calculated (16), and a composite score of the enriched DNA repair genes from the previous xenograft experiment. Upon this analysis, we indeed found a significant correlation (P < 0.001) between canonical AR output and the enriched DNA repair genes (Fig. 1C). Next, to define the most robust signature of AR-associated DNA repair genes without limiting ourselves to the enriched genes from the initial xenograft experiment, we more broadly asked which of the DNA repair genes in all of the six combined DNA repair gene sets (294 genes) were most associated with canonical AR output in the primary human prostate cancer dataset (Fig. 1D and Supplementary Table S2). Filtering for an association with canonical AR output (P < 0.01 and r > 0), we defined an “AR-associated DNA repair gene” signature of 144 DNA repair genes that were significantly associated with canonical AR output.

Given the association of canonical AR output with DNA repair gene expression in primary prostate cancers, we next sought to determine using an in vitro model which, if any, of these DNA repair genes are transcriptionally regulated by androgen. We selected the most widely studied prostate cancer cell line, LNCaP, to model primary prostate cancer. After treating the LNCaP prostate cancer cell line with synthetic androgen (R1881) for 2 days and comparing with vehicle-treated [dimethyl sulfoxide (DMSO)] cells, we measured gene expression level changes by RNA-seq and identified AR target genes by AR chromatin immunoprecipitation sequencing (ChIP-seq). We first confirmed that our newly defined, 144-gene AR-associated DNA repair gene signature was in fact enriched by GSEA in the androgen-treated samples, in addition to other well-known AR signatures such as Nelson and colleagues (ref. 17; Fig. 2A). Of the 144 genes that comprise the AR-associated DNA repair signature, we next identified those genes whose expression was increased by androgen and that contained AR binding sites (Fig. 2B). Thirty-two genes were both induced by androgen and exhibited AR peaks in their enhancers (32) or promoter (1), suggesting that these represented bona fide AR target genes (Fig. 2B and C and Supplementary Fig. S2). Motif analysis of the AR binding peaks of these 32 genes revealed the classic consensus AR binding site (Fig. 2D).

We next sought to determine whether the reduction in DNA repair gene expression observed with androgen deprivation was associated with (i) reduced DNA repair and (ii) increased DNA damage. Using the same in vitro LNCaP model, we exposed prostate cancer cells, pretreated for 2 days with either synthetic androgen (1 nmol/L R1881) or mock (DMSO), to 2 Gy of ionizing radiation, and γ-H2AX foci were quantified. Comparing the γ-H2AX foci in the two conditions, we found that in androgen-depleted conditions the foci peaked later and higher and resolved more slowly, consistent with decreased repair (Fig. 3A). Because γ-H2AX foci represent a surrogate, indirect marker of unrepaired breaks, we next sought to measure DNA damage itself. To do so, we used the neutral Comet assay that directly measures DNA double-strand breaks (DSB). The findings of the Comet assay under the same conditions recapitulated the results obtained by analyzing γ-H2AX foci (Fig. 3B). Of note, at baseline after 48 hours of pretreatment (time 0), mock-treated cells demonstrated increased γ-H2AX foci and increased DNA damage by Comet assay. However, after normalizing the findings from both assays to each condition's respective time 0, the post-ionizing radiation findings of relative decreased repair and increased damage persisted (Supplementary Fig. S1). Given the surprising result that even in the absence of ionizing radiation androgen deprivation alone exhibited increased DNA damage as compared with androgen-treated cells, we next asked whether treatment with ARN-509 would also increase DNA damage relative to control. After 48 hours of ARN-509 (1 μmol/L) treatment versus mock (DMSO), we found increased DNA damage in LNCaP cells, a finding we then demonstrated in two additional cell lines, LNCaP-AR and VCaP (Fig. 3C).

Given these findings, we next asked whether antiandrogen-treated cells compared with mock-treated cells exhibited decreased cell viability following ionizing radiation. To do so, we used the classic clonogenic assay using the LNCaP cell line and demonstrated that treatment with ARN-509 (1 μmol/L) compared with mock (DMSO) resulted in decreased cell survival (Fig. 3D). To further address the possibility that the enhanced radiosensitivity observed with ADT is not due to partial synchronization of rapidly dividing cells into more radiosensitive phases of the cell cycle, we measured the percentage of LNCaP cells in G₁, S, and G₂–M in androgen-deprived cells (DMSO) compared with a range of doses of R1881 in charcoal-stripped serum, each for 48 hours. The percentage of cells across the phases of the cell cycle did not change, yet increasing concentrations of androgen resulted in increased cell growth except for the highest concentration (10 nmol/L; Supplementary Fig. S2). We repeated this analysis for LNCaP cells treated with ARN-509 versus control for 48 hours, and again found no meaningful change in cell-cycle distribution (Supplementary Fig. S3).

Given these findings that AR transcriptionally regulates a network of DNA repair genes and that antiandrogen treatment increases DNA damage and radiosensitizes cells, we next sought to determine which DNA repair pathways are functionally abrogated by ARN-509 (Fig. 4A). Because the DNA DSBs induced by ionizing radiation are thought to be repaired by classical nonhomologous end-joining (C-NHEJ) and homologous recombination, we focused upon these repair pathways.

To assess the effects of antiandrogen treatment on C-NHEJ, we used the transient V(D)J recombination assay as previously described (Fig. 4B; refs. 18, 19). The V(D)J recombination substrate along with RAG1 and RAG2 expression vectors were transfected into LNCaP cells that had been pretreated for 48 hours with either mock (DMSO) or ARN-509 (1 μmol/L). In this assay, rescued substrate plasmids that fail to undergo C-NHEJ–dependent recombination only express the ampicillin (Amp)-resistant gene, whereas those that successfully undergo recombination express both Amp- and chloramphenicol (Cam)-resistant genes. Therefore, after transforming bacteria with the rescued substrate, a recombination frequency can be calculated by the ratio of colonies grown on the Amp + Cam plates compared with the ampicillin plates. Compared with control treatment, ARN-509–treated cells demonstrated significantly decreased C-NHEJ–mediated recombination (>60%), comparable with that seen with stable LIG4 knockdown (Fig. 4C and Supplementary Fig. S4). Next, to assess the effects of antiandrogen treatment on homologous recombination, we used the transient DR-GFP assay, a widely used repair reporter assay for studying homologous recombination (Fig. 4D; ref. 20). DR-GFP contains direct repeats of two defective GFP genes with the upstream repeat containing a recognition site for the I-SceI endonuclease. A DSB induced by I-SceI followed by gene conversion with the downstream copy results in a functional GFP gene, and GFP⁺ cells can be scored by flow cytometry. Using this assay, no significant difference in homologous recombination was observed between the treated and control cells, in contrast to the 2.8-fold reduction in homologous recombination observed with expression of BRC3, a peptide known to interfere with homologous recombination (Fig. 4E; ref. 21).

Androgens, acting through AR, regulate a complex transcriptional program for both prostate cancer growth and differentiation, with recent data demonstrating unique sets of AR target genes in castration-sensitive versus -resistant tumors (22). In this study, we discover a network of DNA repair genes that comprises part of the complex AR-regulated transcriptome. Our data establish that AR signaling increases the expression of DNA repair genes and, in parallel, promotes prostate cancer radioresistance by accelerating repair of ionizing radiation–induced DNA damage. Collectively, these data provide a strong mechanistic rationale for the observed synergy between ADT and radiotherapy.

Clinical trials of ADT plus radiotherapy have been unable to answer the question of whether ADT benefits all patients modestly or a subset to a greater degree. The surprising spectrum of AR output that we observe in primary prostate cancer tumors suggests that ADT may preferentially benefit those patients with high AR transcriptional output and, consequently, high expression of DNA repair genes. Therefore, in a similar manner that patients with breast cancer are selected for antiestrogen therapy by a molecular determinant of response (estrogen receptor/progesterone receptor status), perhaps it is possible to select patients with prostate cancer who will receive ADT along with radiotherapy (as opposed to radiotherapy alone) based on a tumor's AR output. This hypothesis could be tested using pathologic specimens from one of the landmark clinical trials comparing ADT and radiotherapy with radiotherapy alone.

The discovery that AR activity, among its other known biologic effects, also regulates DNA repair could have implications beyond the question of how ADT synergizes with radiotherapy. First, the finding that C-NHEJ can be dynamically modulated via inhibition of a nuclear hormone receptor/transcription factor represents a new kind of mechanism by which to abrogate this important pathway. Second, it remains to be determined whether the other identified AR-regulated DNA repair genes play a functional role in their respective pathways (e.g., mismatch repair, base excision repair, and Fanconi pathway) when their expression levels are modulated by AR (Fig. 4A). Finally, recent genomic studies demonstrate that in comparison with primary tumors, one of the hallmarks of castration-resistant prostate cancer seems to be widespread genomic instability (14, 23). The possibility that the therapeutic intervention of ADT itself during the several-year period of castration sensitivity may contribute to the genomic instability observed in castration-resistant tumors raises a number of important future directions for investigation.

LNCaP/AR xenografts were established in castrate mice as described previously and, once established, were treated for 4 days with either 10 mg/kg ARN-509 plus vehicle (1% carboxymethyl cellulose, 0.1% Tween-80, and 5% DMSO) or vehicle alone (12). RNA was isolated as per the standard protocol and expression profiling performed using IlluminaHT-12 array (Arora and colleagues; in preparation). GSEA (24) was used to test for pathway level differential expression. Discovery analysis of the xenograft transcriptomes used all gene sets listed in both the curated (c2) and Gene Ontology (c5) of the Broad Institute's Molecular Signature Databases (MSigDB).

Canonical AR output was quantified using an unweighted summed score of the n-gene AR-responsive gene signature defined by Hieronymus and colleagues (15). The initial set of DNA repair candidates was taken as the union of the six DNA repair–associated signatures resulting from the xenograft experiment. The correlation between the AR signature and each candidate gene was calculated. Results were filtered to identify all DNA repair–associated genes with significant (Pearson correlation, P < 0.01) positive correlations. The resulting candidates were summarized to the within sample mean as a relative measure of the signature. Spearman correlation was used to test for significant relationship between the AR and DNA repair signatures.

LNCaP cells were grown in described conditions in triplicate, and RNA was isolated by the RNeasy Mini Kit (Qiagen) with the additional steps of lysate homogenization using QIAshredder (Qiagen) and DNase digestion using the RNase-Free DNase Set (Qiagen). Samples were prepared and libraries were created using TruSeq RNA Sample Preparation Kit v2 (Illumina), which included a poly-A selection step. Libraries were pooled at 2 nm concentration and the samples were then subject to cBot cluster generation using TruSeq Rapid PE Cluster Kit (Illumina). The amplified libraries were sequenced using the TruSeq Rapid SBS Kit on the HiSeq 2500 (Illumina). mRNA-seq data were aligned with Mapsplice (25) and genes were quantified with RNA-Seq by Expectation Maximization (26). Gene expression estimates were log transformed and upper quartile normalized. Differential expression was measured using unequal variance t test to identify candidate genes. GSEA analysis of the cell line data was used to test whether the 144 AR-associated DNA repair genes were associated with R1881 treatment relative to DMSO. AR ChIP-Seq was performed according to Cai and colleagues (in preparation).

All described doses of ionizing radiation were delivered using Cs-137 irradiator (Shepherd Mark, Model 68) at a dose rate of 184 cGy/min and at a turntable speed of 6 rpm.

LNCaP cells were grown in described conditions for 2 days on CC2-coated, four-chamber slides (Nunc Lab-Tek II; Thermo Scientific), approximately 20,000 cells per well in 500 μL total volume, and following ionizing radiation were washed and fixed with 4% paraformaldehyde and 0.2% Triton X-100. Slides were blocked with 10% FBS and 0.5% Triton X-100, and the primary antibody (γ-H2AX; Millipore) was incubated overnight at 4°C. Slides were then washed, incubated with secondary antibody (Alexa Fluor 488 Dye; Life Technologies) for 1 hour at room temperature, and stained for 4′,6-diamidino-2-phenylindole. Slides were scanned by confocal microscopy (LSM 5 LIVE) with a 20×/0.8NA objective and foci were counted by MetaMorph image analysis software (Molecular Devices). For each time point, on average, foci of 2,000 separate nuclei were counted.

LNCaP, LNCaP-AR, and VCaP cells were grown in described conditions for 2 days and neutral Comet assay was performed using Comet-Assay Electrophoresis System (Trevigen) as per assay protocol (27).

LNCaP cells were grown in 10-cm tissue culture–treated polystyrene dishes (BD Falcon) in described conditions for 2 days before irradiation. Cells received either 0, 2, 5, or 10 Gy of ionizing radiation, and were subsequently replated in respective conditions into 6-well, tissue culture–treated polystyrene plates (BD Falcon) in a series of 1:3 dilutions (24,000, 8,000, 2,667, 867, 289, and 96 cells/well) at each dose. Lethally irradiated HeLa feeder cells (400,000) were added to each well to promote colony formation. Plates were incubated for 14 days, then washed and fixed with methanol, and stained with 0.2% crystal violet (Sigma) in 10% formalin (Sigma). Plates were scanned by GelCount (Oxford Optronix), and colonies were counted using GelCount software. Clonogenic survival curves were generated as previously described.

Whole-cell lysates were prepared using 10% M-PER lysis buffer and clarified by centrifugation. Proteins were separated by 4% to 12% SDS-PAGE gel and transferred onto polyvinylidene difluoride (PVDF) membranes (Invitrogen). After primary antibody incubation (LIG4 and GAPDH; Abcam) for 1 hour at room temperature, washings, and incubation with secondary antibodies, blots were developed with a chemiluminescence system (Pierce).

The transient V(D)J recombination assay was performed as described according to Gauss and Lieber (18, 19). The V(D)J recombination substrate plasmid along with RAG1 and RAG2 expression vectors were transfected into LNCaP cells (2 × 10⁶) pretreated for 48 hours with either mock (DMSO) or ARN-509 (1 μmol/L), or the same cell line infected with shNT (nontargeting) or shLIG4 lentiviral vectors (Sigma-Aldrich). Plasmids were retrieved 48 hours following transfection, and electrocompetent bacteria (Mega-X DH10B; Invitrogen) were then transformed with the rescued substrate plasmids and plated on both ampicillin and Amp + Cam plates. Bacteria were plated at 1:1,000–5,000 dilution for ampicillin plates and 1:10–50 dilution for Amp + Cam plates to grow colonies at numbers that could be quantified by manual counting. Substrate plasmids that failed to undergo recombination by C-NHEJ only expressed the ampicillin-resistant gene (Amp^r), whereas plasmids that underwent successful recombination expressed both Amp^r and chloramphenicol-resistant (Cam^r) genes. Recombination efficiency was calculated by the ratio of the colonies grown on the Amp + Cam plates to the ampicillin plates.

To measure DSB repair by homologous recombination, 2 × 10⁶ LNCaP cells were nucleofected with 0.7 μg of DR-GFP plasmid and 2 μg I-SceI expression vector (pCBASce) and 1.3 μg empty vector (pCAGGS) or 3.3 μg pCAGGS or for a positive control 2 μg NZEGFP and 2 μg pCAGGS (20). After transfection, cells were immediately plated in regular medium containing 1 μmol/L ARN-509 or control (DMSO). For BRC3 experiments, 2 × 10⁶ LNCaP cells were nucleofected with 0.7 μg of DR-GFP plasmid, 2 μg I-SceI expression vector (pCBASce), and 2 μg of BRC3 expression plasmid or empty vector (pCAGGS; ref. 21). Flow cytometry was performed 48 hours after transfection to analyze GFP expression.

LNCaP and VCaP cell lines used in this article were purchased directly from American Type Culture Collection and cultured according to specifications. LNCaP-AR is an AR-overexpressing (wild-type) cell line originally derived from parental LNCaP; the LNCaP-AR cell line was authenticated for AR overexpression by immunoblot.

J. Wongvipat has ownership interest (including patents) in a UCLA patent for ARN-509. C.L. Sawyers has ownership interest (including patents) in ARN-509 (co-inventor) and is a consultant/advisory board member of Aragon Phamaceuticals. No potential conflicts of interest were disclosed by the other authors.

Conception and design: W.R. Polkinghorn, P.J. Iaquinta, N.P. Shah, H. Hieronymus, M.J. Zelefsky, M. Jasin, S.N. Powell, C.L. Sawyers

Development of methodology: W.R. Polkinghorn, J.S. Parker, E.M. Kass, D.E. Spratt, D. Zheng, B.S. Carver, Y. Chen, A.G. Goglia, J. Chaudhuri, S.N. Powell

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): W.R. Polkinghorn, E.M. Kass, D.E. Spratt, V.K. Arora, W.-F. Yen, L. Cai, S. Fujisawa, A.G. Goglia, A. Gopalan, J. Wongvipat

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): W.R. Polkinghorn, J.S. Parker, M.X. Lee, D.E. Spratt, P.J. Iaquinta, V.K. Arora, D. Zheng, B.S. Carver, Y. Chen, N.P. Shah, S. Fujisawa, A.G. Goglia, H. Hieronymus, M. Jasin, S.N. Powell, C.L. Sawyers

Writing, review, and/or revision of the manuscript: W.R. Polkinghorn, D.E. Spratt, W.-F. Yen, B.S. Carver, Y. Chen, A. Gopalan, P.T. Scardino, M. Jasin, C.L. Sawyers

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): W.R. Polkinghorn, M.X. Lee, D.E. Spratt, P.A. Watson, J. Wongvipat

Study supervision: M.J. Zelefsky, C.L. Sawyers

The authors thank M.R. Lieber (University of Southern California, Los Angeles, CA) for graciously providing the plasmids for the transient V(D)J recombination assay. In addition, W.R. Polkinghorn wishes to thank T.G. Bivona for his encouragement and friendship and C.L. Sawyers both for the opportunity to pursue this important question and for teaching him to become a scientist along the way.

This work was supported by the Prostate Cancer Foundation, Creativity Award (2011).