Androgen Receptor Overexpression in Prostate Cancer Linked to Purα Loss from a Novel Repressor Complex

Increased androgen receptor (AR) expression and activity are pivotal for androgen-independent (AI) prostate cancer (PC) progression and resistance to androgen-deprivation therapy. We show that a novel transcriptional repressor complex that binds a specific sequence (repressor element) in the AR gene 5′-untranslated region contains Purα and hnRNP-K. Purα expression, its nuclear localization, and its AR promoter association, as determined by chromatin immunoprecipitation analysis, were found to be significantly diminished in AI-LNCaP cells and in hormone-refractory human PCs. Transfection of AI cells with a plasmid that restored Purα expression reduced AR at the transcription and protein levels. Purα knockdown in androgen-dependent cells yielded higher AR and reduced p21, a gene previously shown to be under negative control of AR. These changes were linked to increased proliferation in androgen-depleted conditions. Treatment of AI cells with histone deacetylase and DNA methylation inhibitors restored Purα protein and binding to the AR repressor element. This correlated with decreased AR mRNA and protein levels and inhibition of cell growth. Purα is therefore a key repressor of AR transcription and its loss from the transcriptional repressor complex is a determinant of AR overexpression and AI progression of PC. The success in restoring Purα and the repressor complex function by pharmacologic intervention opens a promising new therapeutic approach for advanced PC. [Cancer Res 2008;68(8):2678–88]

Recurrent prostate cancer (PC) remains a therapeutic challenge in part because the mechanisms of progression to androgen independence (AI) and the reasons for the development of resistance to androgen deprivation therapy in hormone-refractory (HR) PC patients are still unknown. A large proportion of patients with high-grade localized cancer (1), metastatic disease (2), and HR-PC (3) show increased expression of the androgen receptor (AR), suggesting that this plays a key role in disease progression. Approximately 30% of cases have AR gene amplification (4) and 10% have mutations (5), but the vast majority of cases have increased AR gene transcription and/or altered receptor protein stability (6). Although PC cells can derive proliferative signals through several pathways, it is apparent that the AR is essential to PC progression in the presence or absence of the androgen ligand (7).

Increased AR expression is also a feature of the emergent AI phenotype in PC models. To mimic the AI phenotype that emerges during androgen deprivation therapy, we chronically deprived an androgen-dependent (AD) cell line (LNCaP) of androgen (8). We found that, similar to some advanced PC xenografts (9, 10), the emergent AI cell line had 2-fold to 4-fold more AR than the parental AD cells (8) resulting from transcriptional up-regulation of the AR. Increased AR was also described in another pair of LNCaP-AD cells and their AI derivatives (9). More recently, a microarray analysis of seven HR/hormone-sensitive isogenic pairs of human PC xenografts showed that among 12,500 gene probe sets tested, only AR mRNA was differentially expressed in all seven cases (10). In the latter study, the HR-PC had more AR protein, and even a modest change in AR level was able to shift the relative abundance of coactivators and corepressors assembled on the promoters of androgen target genes. Moreover, an increase in the AR level of a magnitude similar to that observed in our AI-LNCaP cells caused AR antagonists to function as agonists (10).

Overall, this body of evidence indicates that AR expression level is an important determinant of biological outcomes and provides a compelling reason to study the mechanisms responsible for AR overexpression (10–12). Indeed, although multiple alterations at the gene and protein level have been described in AI PC (13), there is only limited information regarding alterations in the regulation of AR transcription that may account for increased AR mRNA levels (14). We found that a suppressor element (ARS), originally identified in a mouse AR gene and located in the 5′-untranslated region (5′-UTR) of the AR promoter (15, 16), is also present in the human AR gene and that it malfunctions in AI cells (17). A promoter/reporter construct with a deleted ARS element produced an 8-fold increase in AR promoter activity after transfection into AD cells. We also found that nuclear extracts of AD cells contain a repressor protein complex that binds the ARS in a gel shift assay, although this complex was significantly reduced in AI cells (17).

We have now identified the ARS-binding transcriptional repressor complex as containing Purα and hnRNP-K (18–20). Based on the insight gained from our mechanistic studies of the Purα-containing repressor in AD and AI cell lines, and on observed changes in its expression, localization, and binding to the ARS that take place during human PC progression, we conclude that Purα is an important regulator of AR transcription and of AI growth. We also show that Purα expression and function can be restored by agents that relieve epigenetic silencing, suggesting that regulation of the transcriptional levels of AR may provide a novel therapeutic strategy to control PC progression and to enhance the efficacy of existing systemic therapies.

Cells and tissues. LNCaP cells were maintained in RPMI 1640 with 10% heat-inactivated fetal bovine serum. The AI-LNCaP derivative was isolated by growing LNCaP cells in 10% charcoal-stripped fetal bovine serum and 5 μg/mL of insulin for several months. An initial reduction in the total cell population was followed by a gradual outgrowth of AI cells (8). Formalin-fixed, paraffin-embedded and frozen PC tissues were obtained from the New York University Prostate Cancer Tissue Resource, the Stadtische Kliniken, Offenbach, Germany, and the Mount Sinai School of Medicine. Twelve hormone-naïve (HN), Gleason score 6 frozen PCs were microdissected for quantitative PCR (Q-PCR), four were used for chromatin immunoprecipitation (ChIP), seven HR-PC metastases were processed for Q-PCR: two bone marrows, one bone metastasis protruding into the brain, one cervical lymph node, one epidural metastasis, one malignant ascitis, and one malignant pleural effusion. The latter three were used for ChIP. All patients signed Institutional Review Board–approved consent.

Tissue selection, tissue microarray, immunohistochemistry, and image analysis. Four core samples from 18 HN radical prostatectomy and 18 HR transurethral resection specimens of the prostate after progression following several months of androgen deprivation, matched for Gleason score, were cut in 4-μm-thick sections, histology verified every 10th section and was assembled for tissue microarray (TMA; Beecher Instruments; ref. 21). For immunohistochemistry, TMA slides were soaked in H-3300 solution (Vector Laboratories), microwaved and treated with streptavidin peroxidase. The Purα monoclonal antibody 10B12 (IgG₁ isotype; 1:1,000; ref. 22) and AR polyclonal antibody, which have been previously described (23), were horseradish peroxidase–conjugated with secondary antibody and diaminobenzidine. Negative controls for Purα included an isotype-matched irrelevant monoclonal antibody (mouse IgG₁, X0943; Dako); and for AR, a rabbit immunoglobulin fraction (solid-phase absorbed; X09036; Dako) diluted 1:1,000 in PBS. All sections were incubated with horseradish peroxidase–conjugated secondary antibody and diaminobenzidine. Quantitative immunohistochemical analysis was performed using Kodak Molecular Imaging (ver. 4.0) as previously described (24, 25). Positively stained pixel regions-of-interest and the percentage of staining were calculated using Microsoft Excel. Two independent observers determined the percentage of positive Purα nuclear staining based on two 40×-magnified TMA images.

Protein isolation, electrophoretic mobility shift assay, and immunoblotting. Nuclear AD and AI cell proteins were processed for electrophoretic mobility shift assay (EMSA) as previously described (17). For EMSA with purified GST-Purα, the protein was reacted with [γ-³²P]-labeled double-stranded (ds) wt-ARS oligonucleotide 5′-ACC CCG CCT CCC CCC ACC CT-3′ +323 and +342 nucleotides; ref. 17, or single-stranded (ss) C-rich or G-rich ARS, or its three mutants: mARS1, ACC CAA CCT CCT TCC ACC CT; mARS2, ACC CCT TCT CAA CCC ACC CT; and mARS3, ACC TTG CCT AAC CCC ACC CT (Bio-synthesis, Inc.), in binding buffer and subjected to 8% PAGE. For the supershift assay, the nuclear proteins were preincubated with anti-Purα monoclonal antibody and 10-fold excess mARS3. Anti-Purα monoclonal antibody, Purα fusion protein pGPur4, and mutant Purα proteins (amino acids 167-322/m-GST-Purα-1 and amino acids 216-322/m-GST-Purα-2) were previously described (22, 26). Southwestern blot was performed as previously described (17).

Transfection, luciferase assay, and Western blot. Cells were cotransfected with 1 μg of “empty” vector DNA (pVector), AR luciferase reporter pLARS-1, or its ARS-deleted mutant pLARS-del (17), with or without expression plasmids for phnRNP-K (27, 28), or Purα vector pHApur1 (26) or Purα-small interfering RNA (siRNA; 1–2 μg/well) and 0.5 μg of pSV-β-galactosidase (Promega) control, using Effectene (Qiagen). Luciferase and β-galactosidase activity (Promega) of cell lysates were measured 48 h later. Total and subcellular proteins from parallel cultures were processed for Western blotting 48 h after cotransfection using Purα 10B12 (22), AR, β-actin (Santa Cruz), hnRNP-K monoclonal (ImmunQuest, Ltd.), and p21 (Dako) antibodies as described (29).

Pull-down assays using biotinylated DS-ARS-streptavidin beads. Nuclear extracts of AD cells were incubated with streptavidin/magnetic beads (BioLab) bound to three repeats of DS-ARS [sense, 5′-CCC GCC TCC CCC CAC CCG CCT CCC CCC ACC CGC CTC CCC CCA-3′; antisense, (BioTEG)5′-TGG GGG GAG GCG GGT GGG GGG AGG CGG GTG GGG GGA GGC GGG-3′; and its mutant, sense, 5′-CCA ACC TCC TTC CAC CAA CCT CCT TCC ACC AAC CTC CTT CCA-3′; and antisense (BioTEG)5′-TGG AAG GAG GTT GGT GGA AGG AGG TTG GTG GAA GGA GGT TGG-3′]. The beads were then eluted and the protein fractions processed for Western blot.

ChIP assays. The protein/DNA complexes from intact AD and AI cells or from minced frozen PC tissues were cross-linked and then lysed as previously described (30). Equal amounts of protein or tissue lysates (50–130 mg total weight) were incubated with the primary antibody, Purα 5B11 and 1A12 (31), followed by Protein G PLUS-Agarose beads (Santa Cruz), washed twice with high-salt buffer (0.5 mol/L NaCl), followed by LiCl/detergent solution and TE buffer. The DNA-protein immunoprecipitates were eluted, proteinase K–treated, and the DNA extracted. PCR (Expand High-Fidelity PCR System; Roche) was performed using primers to sequence +248 to +487 nucleotides of the AR 5′-UTR region and the Sp1 sites: 5′-AGC TGC TAA AGA CTC GGA GG-3′ and 5′-GGA GTT ACC TCT CTG CAA AC-3′. The images of PCR products stained with SYBR Gold (Molecular Probes) in agarose gels were captured with Fotodyne, imported into Photoshop, and the black bands scanned and quantified using NIH Image. For clinical samples, Purα binding was corrected for input DNA.

Purα knockdown. AD cells seeded at 5,000 cells/well were either fixed 24 h later (day 0) with 10% trichloroacetic acid or transfected with Purα-siRNA (85 ng/well), AR siRNA (85 ng/well), Purα-siRNA plus AR siRNA (85 ng each/well), or control (scrambled) siRNA (labeled NC, 85 ng/well; Ambion) using Effectene in serum-free medium. Cell growth was measured at baseline, 48, and 72 h posttransfection by sulforhodamine B assay. AI cells in AI-medium served as positive control. AD cells seeded at 0.5 to 1 × 10⁶ cells/well were transfected with 1 μg of siRNA per well and processed for protein extraction and Western blotting with AR and Purα.

Analysis of AR and Purα expression using Oncomine. We searched Oncomine⁵

for PURA cDNA and PC. The original raw data and results including the median, 75/25 percentiles, t test, and P values for each study were collected and linked to the original publications. Only one (32) of two studies described HN PC and metastatic HR-PC as defined in our study and probed for Purα.

Q-PCR. Frozen PC specimen were microdissected and processed for RNA extraction, cDNA, and Q-PCR as previously described (33). Primers and VIC-labeled probe for glyceraldehyde-3′-phosphate dehydrogenase, AR, and PURA (ABI) were used to measure copy numbers by reference to log-linear standard curves (2 to 2 × 10⁷ copy range) of serial dilutions of linearized plasmid DNAs containing the respective gene inserts. The positive controls were LNCaP cells, the negative control lacked a cDNA template. The mean copy number of triplicate Q-PCR assays were used to compare samples.

Statistical analysis. Average values of multiple assessments of AR and Purα regarding the percentage of positive staining intensities and expression levels by Q-PCR and ChIP among HN HR-PC's were compared by the Mann-Whitney test. The P values were exact significance levels and two-sided, except where indicated, and statistically significant at <0.05 by SAS version 9.0 (SAS Institute) analysis.

Identification of ARS nuclear binding complex. We previously reported that high AR expression in LNCaP-AI cells might be due to loss of binding of a repressor complex to an ARS element in the 5′-UTR of the AR gene (17). Gel blotting analysis of nuclear extracts from LNCaP-AI and AD cells with a labeled dsARS probe (Southwestern) revealed several bands (∼105 to ∼30 kDa; results not shown). To further identify these bands, nuclear extracts of AD cells were affinity-purified on an ARS-oligonucleotide column (dswt-ARS, 20 bp), the protein was eluted with a linear gradient of KCl and separated on duplicate SDS-PAGE gels; one gel was analyzed by Southwestern analysis whereas the other was used to elute proteins from the areas corresponding to the bands (a total of five) detected by Southwestern analysis. One of the bands was identified as the nuclear protein hnRNP-K (CAA51267.1; ref. 34), by Bio-Mass Spectrometry (results not shown). This protein has multiple functions (35), including the ability to bind ssDN (28, 36) and to repress transcription (34). Because hnRNP-K can shuttle between the nuclear and cytoplasmic compartments (37), we tested nuclear and cytoplasmic extracts of AD and AI cells for proteins that bind to the ssARS. In Southwestern analysis with a C-rich oligonucleotide probe, a single, equal-intensity band of ∼55 to 65 kDa (Fig. 1A,, left) was produced by nuclear extracts of AD and AI cells. When the membrane was stripped of the radioactive probe and tested by immunoblotting with specific anti–hnRNP-K antibody, a doublet, with molecular weight corresponding to hnRNP-K was detected (Fig. 1B,, left). Several additional faint bands were also noted (Fig. 1A,, left). Using G-strand ARS as a probe, several distinct bands, some of high intensity, were noted (Fig. 1A,, right) including a weak 40 kDa band, which was more intense in AD extracts. Immunoblotting with anti–Purα antibodies revealed a band of similar molecular weight (Fig. 1B,, right). Purα was reported to cooperate with hnRNP-K in transcriptional repression (27). Direct immunoblotting of whole cell extracts showed that AR is elevated and confirmed that Purα is reduced in AI cells, whereas hnRNP-K was equally expressed in AI and AD cells (Fig. 1C,, left). Moreover, Purα was shown to be reduced in the cytoplasm of AI cells and was almost undetectable in their nuclei (Fig. 1C,, right). Finally, analysis of AD and AI cells using an Affymetrix gene expression array (U133A GeneChip) showed in three independent arrays that AI cells express ∼2.5-fold more AR mRNA, similar levels of hnRNP-K mRNA, and ∼4-fold less Purα mRNA compared with AD cells (Fig. 1D).

Purα and hnRNP-K form an ARS-binding complex in vitro. We previously showed that nuclear extracts of AD cells incubated with a dsARS probe produced a retarded complex in gel shift (EMSA) assays that were greatly diminished in AI cells (17). We show here that the intensity of this complex is strongly reduced by anti–hnRNP-K blocking antibody, whereas nonimmune IgG have no effect (Fig. 2A,, lanes 3 and 4). Anti-Purα antibody “supershifted” most of this complex (Fig. 2A,, lane 7), as well as a complex produced by purified GST-Purα fusion protein (Fig. 2B,, lane 3), and this complex could be competed off with an excess of unlabeled dsARS (ref. 17; results not shown). Two Purα truncation mutants (mGST-Purα-1 and mGST-Purα-2, amino acids 167–322 and 216–322, respectively), shown previously to bind only very weakly to a Purα element in the c-MYC promoter (30), did not produce retarded complexes with the ARS (Fig. 2B,, lanes 6 and 7). The complex pattern was different when three mutated ARS oligonucleotides (mARS1–3) were used as probes: AD nuclear extracts did not bind mARS1 and mARS3 at all, and with mARS2, produced patterns of bands different from those observed with wtARS (ref. 17; results not shown). Purified GST-Purα protein was bound to neither mARS1 nor to mARS3 (Fig. 2C) and was indistinguishable from the wtARS when bound to mARS2. Moreover, excess of unlabeled dsARS reduced the ARS-Purα complex (Fig. 2C) indicating specificity of interaction. These results suggest that Purα and hnRNP-K are present in the complex and that they directly bind to the ARS.

Both Purα (38–40) and hnRNP-K (34, 41) can strand-separate dsDNA and subsequently bind to the individual DNA strands. Nuclear extracts of AD and AI cells retarded the single C- and G-rich ssARS probes (Fig. 2A,, lanes 10–14 and 15–20, respectively), and the anti-Purα antibody supershifted the G-rich band (Fig. 2A,, lanes 18 and 19), whereas the anti–hnRNP-K antibody reduced the intensity of the C-rich band (Fig. 2A , lane 13). Thus, it seems that when faced with dsARS, each of the proteins could strand-separate it and bind to the C-rich (hnRNP-K) or G-rich (Purα) strand, producing a retarded protein-DNA complex.

A pull-down experiment in which AD cell extracts were incubated with biotinylated wt-dsARS oligonucleotide bound to streptavidin beads showed that both hnRNP-K (Fig. 2D,, top) and Purα (middle) were eluted at 0.25 and 0.75 mol/L KCl, respectively. Beads with mutated dsARS bound much less hnRNP-K and no Purα (Fig. 2D). Sp1, also present in the nuclear extracts (see below), did not bind to the ARS probe (bottom). This suggests that both hnRNP-K and Purα interact specifically with dsARS.

hnRNP-K and Purα bind to the ARS in vivo. We used ChIP assays to test whether Purα and hnRNP-K interact with the ARS in vivo. As the PCR primers used for ChIP were positioned to amplify the ARS and the two nearby SP1 binding sites of the AR gene, anti-Sp1 antibody was used as a positive control. A sequence located between +248 to +487 nucleotides, which contains the putative ARS (+323 to +342 nucleotides), was amplified from the DNA immunoprecipitated by two different anti-Purα antibodies, anti–hnRNP-K antibody or anti-Sp1 antibody in AD and AI cells (Fig. 3, top). No DNA amplification was found with irrelevant IgG, beads alone or lysates alone. To quantify the amount of Purα associated with the ARS, we performed quantitative ChIP (Q-ChIP), using strictly controlled reaction inputs for AD and AI cells. With the two anti-Purα antibodies (22), 1.9-fold to 2.5-fold more Purα was associated with the ARS-containing 248- to 487-nucleotide sequences in the AD cells (Fig. 3, bottom).

Down-regulation of AR transcription by Purα. To test whether Purα and hnRNP-K affect transcription of the AR gene, AD and AI cells were cotransfected with an ARS-containing AR promoter/luciferase reporter (pLARS-1) or with a reporter containing a deleted ARS (pLARS-del; ref. 17) and with expression plasmids for Purα or hnRNP-K or both. Forced expression of Purα in pLARS-transfected AI cells produced a >50% decrease in luciferase activity, and a significantly (P < 0.05) lesser effect in pLARS-del–transfected cells (Fig. 4A). Importantly, this transcriptional effect was accompanied by a reduced expression of endogenous AR protein (Fig. 4A,, right). Forced expression of hnRNP-K, which produced only a very slight increase in the protein level, did not further enhance the effect of Purα (Fig. 4A), suggesting that the level of hnRNP-K was not rate-limiting.

As an alternative experimental approach, we tested Purα and hnRNP-K knockdown by siRNAs in AD and AI cells for their potential to derepress AR transcription, using the AR promoter/luciferase assay or AR protein levels. Transfection of AD cells, which have higher endogenous levels of Purα, with 1 or 2 μg of Purα/siRNA increased AR/luciferase activity by 6-fold and 8-fold, respectively (Fig. 4C), and transfection with 1 μg of Purα/siRNA induced a strong increase in endogenous AR protein (Fig. 4D). Transfection with two individual hnRNP-K siRNAs produced only an ∼1.5-fold increase in AR/luciferase activity (Fig. 4B), and a slight enhancement of the siPurα effect in AI cells (Fig. 4D). As expected, the endogenous Purα protein was very low in AI cells, and was further reduced to a barely detectable level by Purα-siRNA (Fig. 4D), causing a 2.0-fold increase in luciferase activity (Fig. 4B) and a slight increase in AR protein (Fig. 4D). These results reveal the fine transcriptional tuning of AR by Purα expression.

Reduction of Purα level converts AD cells to androgen independence for growth. We reasoned that AI proliferation and higher AR expression (AI phenotype), similar to that achieved by chronic maintenance of LNCaP-AD cells in androgen-poor medium (8, 12), might be mimicked by acute reduction of Purα levels. Thus, we transfected AD cells with Purα/siRNA and cultured the cells in androgen-depleted medium for 2 and 3 days. Compared with cells transfected with a scrambled siRNA, the reduced Purα led to increased AR protein and resulted in a 1.6-fold and 2.5-fold increase in cell growth on days 2 and 3, respectively (Fig. 5A). Importantly, proliferation was entirely AR-dependent because reduction of AR by knockdown in Purα siRNA-treated cells blocked androgen-induced growth (Fig. 5B). We and others have previously shown (12, 42) that high AR levels inhibit p21/WAF1 expression. A similar effect was then found in cells in which AR expression was increased by Purα-siRNA treatment (Fig. 5A and B), indicating that the restored AR is functional. These results indicate an inverse relationship between Purα and AR levels and cell growth.

To assess the potential clinical value of this dependence, we tested whether the repressor complex was amenable to up-regulation by pharmacologic agents known to have epigenetic effects on gene expression. AI cells were treated with suberoylanilide hydroxamic acid (SAHA), 5-azacytidine (5-AzaC), or the two drugs together, and examined using a gel shift assay for the presence of a functional (ARS binding) repressor complex. Individual treatments increased the intensity of the bands (Fig. 5C,, lanes 5 and 6 compared with lane 4), and the combination of the two drugs produced a more intense band (lane 7), which was supershifted by anti-Purα antibody (Fig. 5C,, lane 8), in a pattern similar to that of AD cells (Fig. 5C,, lanes 2 and 3). Treatment of AI cells with the same dose of SAHA (7 μmol/L) increased Purα and strongly reduced the AR mRNA and protein levels (Fig. 5D,, inset). Treatment with 5-AzaC (5 μmol/L) also restored Purα expression and in vitro binding to ARS (gel shift) but had no effect on AR levels. Most importantly, SAHA and 5-AzaC individually, and in combination, had a dose-dependent growth-inhibitory effect on AI cells (Fig. 5D). The isobologram showed that the combination was synergistic (results not shown).

Higher AR levels and lower or delocalized Purα characterize HR human PCs. To examine whether human PCs recapitulate the reciprocal relationship between Purα and AR levels observed in the AI and AD cell lines, we compared the AR and Purα content and/or their subcellular localization in TMA. Gleason score–matched 18 localized HN and 18 HR-PCs, were immunostained for AR and Purα. Figure 6A shows the results of representative sections of HN and HR tumors. In all tumors, AR was found to be predominantly nuclear (Fig. 6A,, top), whereas Purα was both nuclear and cytoplasmic in HN tumors (Fig. 6A,, bottom left) but was mostly cytoplasmic in the majority of HR tumors (Fig. 6A , bottom middle). The mean staining intensity of nuclear AR was significantly lower (P = 0.011) in HN tumors (mean ± SD, 15.4 ± 16.4) than in HR tumors (24.4 ± 17.8). The overall Purα level in the immunohistochemical analysis was not significantly different between HN and HR tumors (P = 0.52), possibly due to the difficulty in direct comparison of immunostaining intensities of nuclear and cytosolic antigens. However, Purα was localized almost exclusively to the nucleus in the majority (12 of 15) of the HN tumors, but in only 5 of 15 HR tumors, and the percentage of Purα-positive nuclei per tumor was significantly lower (P = 0.005) for HR tumors (mean ± SD, 34.1 ± 37.9; median, 12.0) than for HN tumors (mean ± SD, 80.5 ± 37.2; median, 100).

We also compared AR and Purα mRNA levels by Q-PCR in frozen tissues from 12 localized PCs (HN tumors) and 7 HR metastatic tissues (Fig. 6B). HN tumors had significantly (P = 0.0317) lower AR mRNA levels (mean ± SD, 4,172.1 ± 5,992.0 and 52,341.0 ± 102.570.0, respectively) and significantly (P = 0.0317) higher Purα-mRNA levels (5,282.3 ± 6,498.4 and 1,764.5 ± 1,055.6, respectively) than HR metastases. Moreover, a study published in Oncomine reported a significant (t = 10.924, P = 8.7 × 10⁻¹³) decrease of Purα mRNA level in metastatic HR-PC compared with HN primary PC (Fig. 6C; ref. 32). The box plot shows that the median for Purα was −1.28788 in the HR group and 0.50424 in the HN group. No data were available for AR and hnRNP-K expression for this study, but published evidence supports increased AR expression in advanced tumors (32, 43).

Finally, we performed Q-ChIP analyses of Purα on tissue samples from four primary HN PCs and three HR metastases. In Fig. 6D (top), we show three independent PCR amplifications for a representative HN and HR tumor. Similar results were obtained for each of the tumors tested (data not shown), and the gel bands were scanned, quantified using NIH Image, and corrected to account for the DNA input (Fig. 6D , bar graph). These results show that the amount of Purα bound to the ARS-containing DNA sequence isolated from the HR tumors was nearly 4-fold lower (one-sided P = 0.029) than Purα bound to HN-derived DNA. Together with the gene expression data, these results are consistent with the role of Purα in AR regulation in human cancer and provide evidence for a causal relationship between decreased Purα binding and increased AR expression in the AI progression of human PC.

We have now identified a mechanism to explain the previously confirmed association of increased AR and PC progression to hormone resistance. Results of gel shift assays of nuclear extracts and in vivo ChIP analyses in an AD and AI LNCaP cell model indicate that a previously identified cis-acting suppressor element in the 5′-UTR of the human AR gene (17) binds a novel transcriptional repressor complex that contains Purα and hnRNP-K (Fig. 1). Compared with the parental LNCaP-AD cells, the level of expression (Fig. 1C and D) and the binding of Purα to ARS was markedly decreased in the LNCaP-AI cells, which are androgen-independent for growth, and which overexpress AR (Fig. 1A–C; ref. 8). HnRNP-K levels were similar in both cell lines and the binding of hnRNP-K to the ARS was similar in both cell types (Fig. 1A and B), suggesting that deregulation of Purα activity is the pathogenic event.

Indeed, we have determined that regulation of Purα, and not hnRNP-K, is crucial for the repression of AR levels in this model. SiRNA knockdown of Purα in LNCaP-AD cells produced higher AR levels and activation (Fig. 4C and D), resulting in the inhibition of its downstream target p21 (Fig. 5A; refs. 12, 42) and AI growth (Fig. 5A). Increased expression of an active AR was shown to be the direct cause of AI growth of the LNCaP-AD Purα knockdown cells because concomitant knockdown of AR abrogated AI growth in these cells (Fig. 5B). Conversely, forced expression of Purα in the AI-derivative of LNCaP cells, which had markedly decreased expression and binding of Purα to ARS, was sufficient to reduce AR expression and to decrease transcriptional activation of an AR promoter/reporter construct (Fig. 4A). Our studies also suggest that the interaction between Purα and AR is quite specific. Forced expression of Purα in AI cells reduced ARS-containing AR promoter/reporter activity only when the ARS was intact (Fig. 4A), indicating that the cis-element is specific in mediating this effect and in lowering the expression of endogenous AR. Moreover, Purα mutants, in which two or three of the five central repeat modules involved in binding to a single-stranded Purα response element (26) were deleted, lost their ability to bind to the dsARS (Fig. 2C). The specificity and the dependence of this effect on AR regulation are noteworthy in view of the observation that overexpressed Purα suppressed growth through other mechanisms (24). Thus, it is likely that decreased binding of Purα to the ARS, which allows overexpression of active AR, provides the AI-derivative of LNCaP-AD cells with a mechanism for thriving in the androgen-deprived conditions, a requirement for AI progression in patients on hormonal treatment.

It is remarkable that the interrelations and alterations observed in this model seem to be directly relevant to HN and HR-PC specimens from patients. Compared with HN specimens, HR-PCs with higher AR levels had reduced Purα-mRNA, cytosolic rather than nuclear localization of Purα protein, and occupancy by Purα of the suppressor element in the 5′-UTR of the human AR gene (P = 0.029; Fig. 6A–D). Although the number of HR-PC specimens available for analysis by both Q-PCR and ChIP was small, the consistency of the results in this sequential, nonselected set of samples, and their correlation with similar results from an independent gene expression array (Oncomine) on a similar group of patients (Fig. 6C), supports the notion that this mechanism is a frequent determinant of AI progression in patients with PC.

The mechanisms that regulate Purα expression and its localization in PC cells have not yet been definitively elucidated. The fact that a histone deacetylase inhibitor or an inhibitor of DNA methylation restored the expression and function of the Purα-containing repressor complex indicates that the genes coding for proteins constituting the repressor complex are epigenetically silenced. Also, there are reports that Purα can down-regulate its own expression, and that it can be transactivated by E2F-1 (44). Our data suggest that both a reduced level of Purα and its exclusion from the nucleus can account for the loss of Purα from the AR 5′-UTR in HR-PCs. This conclusion is further strengthened by the results of biochemical analyses of Purα levels, which showed an overall reduction with absence in the nucleus in LNCaP-AI cells (Fig. 1C), and of the ChIP assays which showed a major decrease in Purα binding in HR tumors compared with HN tumors (Fig. 6D). It has been previously shown that the nuclear form of Purα migrate slower in PAGE and that a domain within Purα, required for nuclear transport or retention, might be a substrate for phosphorylation (45). There is also evidence that Purα shuttles between the nucleus and the cytoplasm during the cell cycle (39, 46). Based on these observations, we speculate that the loss of nuclear localization is crucial for up-regulation of AR transcription and expression during hormone-refractory progression. Therefore, identification of the mechanism that regulates the subcellular localization of Purα in human PC might emerge as an important future outcome of this work.

Overall, our finding of the combined presence of Purα and hnRNPK in the repressor complex of the AR gene is similar to that recently reported for the CD43 gene promoter (27). In this instance, however, both proteins bind upstream of the transcription initiation site to the same DNA strand with hnRNP-K binding to a cruciform DNA structure with a single-stranded loop. In contrast, our findings show the binding of two proteins to opposing strands of the AR promoter. Using the Gene Runner algorithm, several secondary structures were found surrounding putative binding sites of Purα and hnRNP-K in the CD43 promoter, including the structure previously reported (27), but a similar analysis of the 60mer oligo surrounding the ARS detected no secondary structures. Because three Sp1-binding sites surround the ARS, it is possible that the repressor complex interferes with the proper function of Sp1 or other transactivators (47).

Sequencing of the Purα transcript from AI cells did not reveal any mutations, suggesting that epigenetic changes might define the altered function of Purα and, under the selective pressure of androgen deprivation, lead to AR overexpression and AI growth. As mentioned above, we found that normal function of the Purα-containing repressor can be restored by treatment of AI cells with the HDAC inhibitor SAHA alone or combined with 5-AzaC (Fig. 5C). This was not achieved by treatment with 5-AzaC alone suggesting that its effect on Purα expression and binding to the ARS element is insufficient to restore full activity of the repressor complex. Therefore, its growth inhibition of AI cells must be mediated via a different mechanism which, when combined with SAHA, becomes synergistic. Epigenetic regulators are known to affect the activity of multiple genes in diverse cells (48), but our data suggests that in PCs, inhibition of AR transcription (17) through Purα may be the dominant effect that determines the AD phenotype. Therefore, although further studies are needed to understand the effect of these agents on the regulation of Purα expression and trafficking, we have identified a potential therapeutic strategy to reverse the effect of Purα loss on AR overexpression and AI growth of HR-PC (Fig. 5D). Furthermore, because the activity of the AR protein can be altered by several posttranslational modifications through a variety of signaling pathways (13), reducing the levels of AR (17) may enhance the efficacy of agents targeted to ligand-independent pathways that activate AR, inhibitors of androgen-binding, or cytotoxic agents (12, 49).

In summary, this is the first report that links the loss of a defined transcriptional repressor complex to increased AR levels in an AI PC cell line and human HR-PC tissues. Specifically, we have established that decreased Purα level and/or its displacement from the nucleus to the cytoplasm are critical to the loss of transcriptional repressor complex function and AI growth. Furthermore, we have shown that therapeutic restoration of Purα repressor function with agents that relieve epigenetic silencing can reduce AR transcription and inhibit AI growth, providing a novel strategy to control HR-PC progression.

Grant support: USPHS Research grants CA-98135-04 (A.C. Ferrari), the Chemotherapy Foundation (A.C. Ferrari and L.G. Wang), CA55219 (E. Johnson), USPHS Research grant CA-40578, and the Samuel Waxman Cancer Research Foundation (L. Ossowski).

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.

We thank Drs. Charles Hauer and X. Ding from the Wadsworth Center, New York State Department of Health (Albany, NY), and Dr. R. Wang at Mount Sinai School of Medicine (New York, NY) for partial mass spectrometry analysis of purified protein samples; and Drs. Robert Gallagher and Arthur Zelent for critical reading of the manuscript.