Mutation of the androgen receptor causes oncogenic transformation of the prostate

Abstract

Recent evidence demonstrates that the androgen receptor (AR) continues to influence prostate cancer growth despite medical therapies that reduce circulating androgen ligands to castrate levels and/or block ligand binding. Whereas the mutation, amplification, overexpression of AR, or cross-talk between AR and other growth factor pathways may explain the failure of androgen ablation therapies in some cases, there is little evidence supporting a causal role between AR and prostate cancer. In this study, we functionally and directly address the role whereby AR contributes to spontaneous cancer progression by generating transgenic mice expressing (i) AR-WT to recapitulate increased AR levels and ligand sensitivity, (ii) AR-T857A to represent a promiscuous AR ligand response, and (iii) AR-E231G to model altered AR function. Whereas transgenes encoding either AR-WT or AR-T857A did not cause prostate cancer when expressed at equivalent levels, expression of AR-E231G, which carries a mutation in the most highly conserved signature motif of the NH₂-terminal domain that also influences interactions with cellular coregulators, caused rapid development of prostatic intraepithelial neoplasia that progressed to invasive and metastatic disease in 100% of mice examined. Taken together, our data now demonstrate the oncogenic potential of steroid receptors and implicate altered AR function and receptor coregulator interaction as critical determinants of prostate cancer initiation, invasion, and metastasis.

The work by Huggins and Hodges in the early 1940s demonstrated that prostate cancer, like the gland from where it arises, is initially dependent on androgens for growth and survival (1). Since then, androgen ablation has been used for the treatment of locally advanced and metastatic prostate cancer (2). Despite initially favorable responses, androgens are not the prostate cancer Achilles heel, as most patients receiving this therapy ultimately progress to develop hormone ablation therapy-resistant disease (2). Moreover, studies in a autochthonous mouse model have directly demonstrated that androgen ablation can frequently select for more aggressive and metastatic disease (3).

Because androgen receptor (AR) is expressed in almost all clinical and murine prostate cancer both before and after androgen ablation therapy (4, 5), AR activity likely contributes to all stages of prostate cancer progression. Global gene expression profiling demonstrated AR as the only gene consistently up-regulated in emerging therapy-resistant human prostate cancer xenografts (6), and AR gene amplification after androgen ablation therapy has been reported in almost one-third of prostate tumors (7). Furthermore, many recurrent prostate cancers overexpress coactivators TIF2 and SRC1 that can increase AR activity at physiological concentrations of adrenal androgen (8). As well, IGF-I, KGF, EGF, and Her-2/neu growth factor pathways activate AR in the absence of androgen in prostate cancer (9–11). Collectively, inappropriate AR activation might provide cancer cells with a growth survival advantage after androgen ablation; despite progression in an androgen-depleted environment, the cancers may still be AR dependent.

Somatic missense mutations in AR have been identified in primary, recurrent, and metastatic forms of clinical prostate cancer and cell lines (12–14). The prototypical example is AR-T877A, a missense mutation identified in the LNCaP cell line originally isolated from the lymph node metastasis of a hormone refractory prostate cancer patient. Located in the ligand-binding domain (LBD) T877A confers promiscuity to AR allowing activation by progesterone, estrogen, adrenal androgens, and hydroxyflutamide in addition to androgens (15). Although >60 missense mutations in the AR have been identified from clinical specimens (16), their association with disease progression and emergence of a hormone-independent phenotype has only recently been appreciated (4). Much like somatic mutations in p53, the spontaneous mutations in AR cluster to discrete hot spots that collectively cover only 8% of the receptor coding sequence, implying they define important structural and functional regions (17). We have reported that spontaneous somatic AR mutations identified in prostate tumors derived from a genetically engineered mouse model also collocate to the same hot spots (17), supporting the conservation of AR structure–function relationships in man and mouse.

An important finding of our previous analysis in a genetically engineered mouse model was that androgen ablation resulted in a predominance of AR gene mutations in the N-terminal domain (NTD) of the receptor compared with untreated animals (5). In particular, two mutations were of interest because they resulted in amino acid substitutions (A229T, E231G) in a short AR NTD signature motif (ARNSM; Fig. 1A and ref. 18). The ARNSM is unique to AR and is the region of the AR-NTD most highly conserved during evolution (18), suggesting it is critical for NTD structure and recruitment of transcription factors (19, 20). Both AR-A229T and AR-E231G display increased ligand-independent basal activity, whereas AR-E231G has increased responsiveness to coactivators ARA160 and ARA70 (5). Recently, the AR-NTD was shown to facilitate direct interaction with the Hsp70-interacting protein (CHIP) that functions as a negative regulator of AR transcriptional activity (21). Indeed, the AR-A229T and AR-E231G mutations reduced the interaction between CHIP and AR by 16% and 43%, respectively. These data confirm that the AR NTD contains an evolutionarily conserved motif that likely has a critical role in modulating AR action.

Fig. 1.

Generation and identification of transgenic mice. (A) The AR-N-terminal signature (ANTS) sequence was identified from multiple AR sequence alignments by using a clustalw algorithm (42) with human AR as the profile sequence. Alignment revealed a conserved region of amino acids corresponding to mouse residues 229–242. Blue, identical; yellow, conserved; white, nonhomologous. (B) The AR-E231G transgene. The E231G mutation (GAG > GGG) was introduced into a construct carrying an HA epitope-tagged WT mouse AR cDNA (AR-WT) between the rPB with a rabbit β-globin fragment with a small intron and a bovine growth hormone poly(A) signal sequence. Arrows indicate the primer pairs. The DNA sequences were confirmed in germ-line DNA. (C) The AR-T857A transgene. The T857A mutation (ACT > GCT) was created as described in B. (D) Expression analysis of AR-E231G and AR-WT transgenic mice. Tissue RNA at 12 weeks of age was analyzed by RT-PCR with P2 primers. Primers for L-19 were included as internal controls. Nontransgenic littermates were the controls (NT). Transgene plasmid DNA was the positive control (+). Transgenes were expressed predominantly in ventral prostate in independent lines. Expression of AR-E231G was primarily detected in the ventral prostate, but some expression was detected in the dorsolateral prostate. TE, testis; SV, seminal vesicle; SP, spleen; LU, lung; LV, liver; KD, kidney; BL, bladder; H, heart; MU, muscle; TH, thymus; BR, brain; M, DNA molecular weight markers. (E) RNase protection analysis. Probe was hybridized with RNA from VP lobes of AR-T857A, AR-E231G, AR-WT, and NT mice. Protected probe was separated by acrylamide gel. (F) Expression of AR-WT, AR-E231G, and AR-T857A protein. Extracts prepared from the DLP, VP, and AP lobes of transgenic and NT littermates were fractionated by SDS/PAGE and probed with anti-HA (Upper) or anti-GAPDH (Lower) antibodies. Transgene protein was detected only in the VP lobes of transgenic mice.

Owing to a paucity of appropriate experimental systems, it has been difficult to prove a direct and/or causal relationship between AR expression, activation, or function, and the initiation or progression of prostate disease beyond the fact that genes like prostate-specific antigen can be detected in tumors growing independent of normal physiologic levels of testicular androgens. Therefore, the current study was undertaken to directly assess the consequences of prostate-restricted expression of (i) wtAR to mimic increased receptor levels and androgen sensitivity, (ii) the LNCaP AR to model promiscuous receptor activation by nonclassical ligands, and (iii) AR-E231G to represent alterations in AR function.

Methods

Construction of the Transgene. The cDNA fragments encoding WT mouse AR, E231G (human E251G), and T857A (human T877A) bearing N-terminal hemagglutinin (HA) epitope tags were excised from pCDNA3/HA-AR-WT, pCDNA3/HA-mAR/E231A, and pCDNA3/HA-mAR/T857A (5) with KpnI and PstI, converted to blunt ends with T4 polymerase, and ligated into a Pb-kbpa vector with a –426 to +28 fragment of rat probasin (rPB) digested with EcoRI and blunt ended with T4 polymerase to yield rPB-AR-WT, rPB-AR-E231G, and rPB-AR-T857A, respectively. For microinjection, plasmids were digested with NotI and XhoI and subjected to agarose gel electrophoresis, and transgenes were recovered by QIAEX II (Qiagen, Valencia, CA). Linear fragments were introduced by pronuclear injection into FVB embryos (22).

Screening of Transgenic Mice. Mouse-tail DNA isolation and PCR screening were as described (23). The P1 primers used to screen for positive transgenic mice were P1F (5′-CTTGTCAGTGAGGTCCAGATACCTACAG-3′) and P1R (5′-ATCCGGCACATCATAAGGGTATCCCATG-3′). The point mutation E231G was confirmed by sequencing fragments amplified by P2 primers P2F (5′-ACCCTTATGATGTGCCGGATTATGCC-3′) and P2R (5′-GGCGTAACCTCCCTTGAAAGAGGA-3′). The point mutation T857A was confirmed by sequencing fragments amplified by P3 primers P3F (5′-TGCTGCTCTTCAGCATTATTCCAGT-3′) and P3R (5′-GGTTTTGGGTATTAGGGTTTCCAAA-3′).

RT-PCR. Tissue RNA was extracted with the RNeasy kit (Qiagen) and digested with DNase I (Invitrogen). RT-PCR used 1 μg of total RNA as described (5). The P2 primers were used for transgene-specific transcripts. Primers specific for ribosome protein L19 were used for controls (24).

Ribonuclease Protection Assay. The antisense probe was a 215-bp fragment containing the HA sequence and 170-bp AR N-terminal sequence amplified with 5′-CGCGGATCCCGGCTACCACCATGGGATACCC-3′ (forward) and 5′-CGGGATCCTGCCTCTGCTGTAAACAGGCG-3′ (reverse) primers. The fragments were digested with KpnI and BamHI and cloned into KpnI- and BamHI-digested pDP18-T7/T3 (Ambion). The [α-³²P]UTP-labeled probe was synthesized by in vitro transcription (BD Biosciences, Franklin Lakes, NJ) with T7 polymerase. RPA was performed with 5 μg of total ventral prostate RNA. Protected probes were separated by 10% acrylamide gel and identified by exposure to XAR5 film (Kodak).

Histology. The genitourinary tract consisting of the bladder, urethra, seminal vesicles, and prostate was harvested as a complex. The prostate dorsal lobe (DP lobe), prostate lateral lobe (LP lobe), prostate ventral lobe (VP lobe), and prostate anterior lobe (AP lobe) were dissected under a microscope. Tissues collected at necropsy were fixed in 4% paraformaldehyde in PBS for 4 h and transferred to 70% (vol/vol) ethanol overnight. Specimens were processed through graded alcohols, embedded in paraffin, sectioned at 5 μm, and mounted on ProbeON-Plus slides (Fisher). Sections for histological analysis were stained with hematoxylin/eosin (H&E).

Immunohistochemistry. Paraffin sections were baked overnight at 55°C and dewaxed. Antigens were retrieved by boiling in 10 mM sodium citrate (pH 6.0) for 15 min and cooled for 1 h at 25°C. Endogenous peroxidase activity was quenched by a 10-min immersion in a solution of 3% H₂O₂ in methanol. Blocking was performed with the Dako LSAB system for 5 min at 25°C. Primary antibodies and working conditions were as follows: HA antibody (Covance, Richmond, CA), 1:500, overnight at 4°C; Ki-67 antibody (Vector Laboratories), 1:1,000, 1 h at 25°C; and AR antibody (Upstate Biotechnology, Lake Placid, NY), 1:60, overnight at 4°C. Products were visualized with Dako LSAB kit and the chromogen 3′ 3′-diaminobenzidine tetrahydrochloride (BioGenex Laboratories, San Ramon, CA). Primary antibodies were replaced with normal rabbit serum for negative controls. Sections were counterstained with methyl green.

Immunoblotting. Mouse tissues were homogenized in ice-cold RIPA buffer (40 mM Tris-HCl, pH 7.0/1 nM EDTA/4% glycerol/10 mM DTT/0.2% SDS/2 mM PMSF/20 μg/ml aprotinin/5 μg/ml leupeptin/5 μg/ml pepstatin/1 mM NaF), by using a 9-mm Polytron homogenizer (Brinkmann). Protein concentrations were determined by the Bradford assay. Briefly, 50 μg of protein was resolved through a 7.5% SDS-polyacrylamide gel and transferred to nitrocellulose. Membranes were probed with antibodies specific for HA at 1:500 (H-11, Covance) and subsequently with a goat anti-mouse IgG horseradish peroxidase conjugate. AR protein was detected by Supersignal chemiluminescent substrate (Amersham Pharmacia). Membranes were stripped and probed with rabbit polyclonal antibody for AR at 1:500 (N-20, Santa Cruz Biotechnology) and anti-GAPDH antibody at 1:5,000 (Ambion). Immunoblots were developed as before and quantified with the AlphaImager system (Alpha Innotech, San Leandro, CA).

Structural Analysis of AR Mutations. Molecular model 3-dimensional structures representing WT or mutant AR-NTD peptide sequences were constructed by using chemsite pro (ChemSW, Fairfield, CA). Nonpolar hydrogen atoms were added, and the structure was solvated in a box of single-point charge water molecules and subjected to energy minimization (20 steps of steepest descent with a cutoff for nonbonded interactions of 10 Å). For each peptide, 10 independent molecular dynamic simulations were performed at constant temperature (300 K) over 100 ps. The rms deviation (rmsd) values for all atoms were calculated by using the relevant minimized starting structures as templates, and solutions were colored according to displacement from the starting minimized structure. Existing structures for the human AR (Protein Data Bank ID code 1I37) and its LNCaP variant (Protein Data Bank ID code 1I38) were analyzed by using spdbv (Version 3.7).

Results

Generation of Transgenic Mice Harboring AR Variants. To investigate the role of the AR NTD and test the hypothesis that changes in AR signaling could cause spontaneous prostate cancer in vivo, we generated independent lines of transgenic mice with prostate-specific constructs encoding WT-AR, AR-E231G, and AR-T857A (Fig. 1 B and C). We identified multiple independent founders carrying AR-WT (three lines), AR-E231G (seven lines), and AR-T857A (five lines) and confirmed the presence of the E231G (Fig. 1B) and T857A (Fig. 1C) mutations in the germ line of respective founder mice.

Tissue-specific expression of all transgenes was confirmed in independent lines by RT-PCR. Consistent with previous data, the minimal probasin promoter reproducibly directed transgene expression to the epithelial compartment of the ventral prostate (Fig. 1D). We were unable to detect transgene expression in other organs (Fig. 1D). As shown in Fig. 1E, steady-state transcript analysis demonstrated all transgenes were expressed at similar levels. Analysis of tissue extracts procured from independent lines of transgenic mice confirmed VP lobe-restricted AR protein expression (Fig. 1F and data not shown). We were unable to detect expression in prostate tissues of nontransgenic littermates under identical conditions.

Consequence of AR Variant Expression in Transgenic Mice. The consequence of AR-WT, AR-E231G, and AR-T857A expression was first examined at 12 weeks of age. As shown in Fig. 2, the DP, LP, VP, and AP lobes appeared normal in the nontransgenic littermates (Fig. 2 A–D) and the AR-WT mice (Fig. 2 E–H). Whereas the DP, LP, and AP lobes of AR-E231G mice were unremarkable (Fig. 2 I, J, and L), we readily observed epithelial hyperplasia and dysplasia in the VP lobes of all AR-E231G mice (Fig. 2K). The lesions in AR-E231G mice typically presented as disorganized layers of epithelial cells in contrast to the single regular layers observed in the nontransgenic and AR-WT mice. In contrast, we were unable to detect a significant phenotype in the AR-T857A mice (Fig. 2 M–P).

Fig. 2.

Pathobiology of transgenic mice. Paraffin sections (5 μm) were prepared from NT (A–D), AR-WT (E–H), AR-E231G (I–L), and AR-T857A (M–P) mice at 12 weeks of age. Representative sections stained with H&E are shown for DP lobe (A, E, I, and M), LP lobe (B, F, J, and N), VP lobe (C, G, K, and O), and AP lobe (D, H, L, and P). Histological features consistent with prostatic intraepithelial neoplasia were observed in the VP lobes of AR-E231G mice (K, Q–T). The anti-HA antibody was used to probe serial sections of VP (T) and NT (U), AR-WT (V) and AR-T857A mice (W). Brown nuclei indicate immunoreactivity. Sections were counterstained with methyl green. All ×20 except S and T, which were ×40 original magnification. (X) Increased proliferation in the VP lobes of AR-E231G transgenic mice. An anti-Ki67 antibody was used on sections of VP lobes from NT, AR-WT, and AR-E231G mice at 12 weeks of age. Quantitation of Ki-67 positive cells was from at least three mice for each group. *, Significant difference between AR-E231G and NT and AR-WT, P < 0.05 by Student's t test.

The histopathologic features observed in AR-E231G mice were consistent with the mouse prostatic intraepithelial neoplasia phenotype and resembled the prostatic intraepithelial neoplasia lesions in other transgenic models (25). The frequent papillary lesions with tufting and intraepithelial lumen formation, cellular atypia with prominent enlarged and hyperchromatic nuclei of variable sizes and shapes were reproducibly observed to be perfectly concomitant with transgene expression (Fig. 2 Q–T). Whereas expression of AR-WT (Fig. 2V) and AR-T857A (Fig. 2W) were readily detected in the VP lobe, these glands appeared normal and unremarkable up to 50 weeks of age. As shown in Fig. 2X, the level of proliferation was very low in the epithelial compartment of the nontransgenic and AR-WT mice at 12 weeks of age, consistent with the proliferation normally observed in mouse prostate epithelium (26). In contrast, the Ki-67 index in the VP lobe of AR-E231G was statistically greater (P < 0.05) than in nontransgenic littermates. We were unable to detect a significant difference in proliferation index in the DLP between any of the genotypes. These observations confirm that AR-E231G specifically promoted epithelial proliferation leading to the rapid development of prostatic intraepithelial neoplasia in the ventral prostate.

The consequence of expression of AR-WT, AR-E231G, and AR-T857A was then examined at ≈50 weeks of age. Although we observed no significant difference in pathologic grade between AR-WT, AR-T857A mice, and control littermates at 1 year, all (5/5; 100%) of the AR-E231G mice demonstrated primary prostate tumors with associated lymphocytic invasion (Fig. 3). The epithelial origin of these tumors was confirmed by analysis for E-cadherin (Fig. 3B), and expression of the transgene was concomitant with the adenocarcinoma (Fig. 3 C and D). Moreover, metastatic deposits were detected in the lungs of all of the mice (Fig. 3 E and F). These deposits were found to express the transgene (Fig. 3G) but not NKX2.1, a specific marker of lung epithelial differentiation (Fig. 3H). Inflammation associated with the adenocarcinoma in AR-E231G mice was not related to the NH₂-HA epitope because AR-WT and AR-T857A also carried this sequence. In all, these observations support a direct causal relationship between AR-E321G expression and malignant prostate cancer. To our knowledge, this is the first report of a somatic mutation able to convert the AR, or any other steroid receptor, into a potent oncogene sufficient to cause invasive and metastatic prostate cancer in vivo. These data also functionally demonstrate that the ARNSM is a critical region of AR and provide an alternative to the hypothesis that increased AR sensitivity alone initiates cancer.

Fig. 3.

Pathobiology of prostate cancer in AR-E231G mice. (A–D) Adjacent sections (5 μm) were prepared from AR-E231G mice at 50 weeks of age. (A) Representative section of VP lobes stained with H&E shows adenocarcinoma with lymphocytic infiltration. (B) Analysis with anti-E-cadherin antibodies demonstrates E231G tumors to be of epithelial origin. (C) Analysis with anti-AR-specific antibodies demonstrates collocation of AR expression in adenocarcinoma. (D) Analysis with anti-HA antibody demonstrates expression of transgene collocates with tumor and AR expression (compare with C). (E–H) Sections of lungs of AR-E231G mice at 50 weeks stained with H&E demonstrating minimal residual glandular structure of a metastatic deposit (E) and a poorly differentiated lung deposit from an independent AR-E231G mouse (F). (G) Stained section adjacent to that in E with anti-HA specific antibodies demonstrates expression of the AR-E231G transgene in the metastatic lesion. (H) Stained section adjacent to that in G with anti-NKX2.1-specific antibody demonstrates the metastatic lesion is not of lung origin.

Modeling the WT and Variant AR NTD. Because the AR NTD has not been crystallized or induced to form an ordered structure suitable for structural determination via NMR, dynamic simulation was used to further study the E231G mutation. Modeling the A229T and E231G substitutions was performed on the local helical structure of a peptide encompassing the predicted NTD signature sequence (²²⁸NAKELCKAVSV²³⁸) and assumed that the peptide was exposed to solvent and adopted an alpha-helical structure as previously predicted (19). Comparison of 10 independent molecular dynamic simulations for each AR peptide solvated in explicit single-point charge water demonstrated a significantly (P < 0.05) higher positional fluctuation (rmsd) for the WT than either Thr-229 or Gly-231 mutant peptides (Fig. 4 A–C). The position of these two substitutions at the N terminus of the helix may reinforce the helix-generated dipole with subsequent stabilization as has previously been demonstrated for substitution of glutamate from glutamine at the N terminus in a model helix (27). Therefore, both Thr-229 and Gly-231 mutations may stabilize the secondary structure in the vicinity of the ARNSM that could enhance accessory protein recruitment (19). This is supported by in vitro analysis demonstrating enhanced ligand-specific responses of AR-E321G to ARA70 and ARA160 (5). It will be important to fully elucidate how AR-E231G induces deregulated growth given the evolutionary pressure to maintain the encompassing ARNSM sequence.

Fig. 4.

Structural analysis of AR mutations. (A–C) Three-dimensional molecular model structures representing WT or mutant AR-NTD peptide sequences constructed by using chemsite pro. Nonpolar hydrogen atoms were added, and the structure was solvated in a box of single-point charge water molecules and subjected to energy minimization (20 steps of steepest descent with a cutoff for nonbonded interactions of 10 Å). Ten independent molecular dynamic simulations, performed at constant temperature (300 K) over a period of 100 ps, are shown for each peptide. rmsd values for all atoms were calculated by using the relevant minimized starting structures as templates, and solutions are colored according to displacement from the starting minimized structure (blue, minimal displacement; red, maximum displacement). For each mutant, the WT peptide is shown in pink. Average rmsd (+SD) of the 10 solutions for each peptide is indicated. Solvent is not shown. (D) Existing structures for the human AR (Protein Data Bank ID code 1I37) and its LNCaP variant (Protein Data Bank ID code 1I38) were analyzed by using spdbv (Version 3.7). The environment within 6 Å of bound ligand is shown in magenta or aqua, with bound ligand shown in blue and hydrogen bonds represented by dashed green lines.

The LBD of mouse AR-WT and AR-T857A have identical amino acid sequences to the human WT and LNCaP variants, respectively, and can be represented by the solved crystal structures of these receptors (28, 29). Detailed analysis of these structures demonstrated that the T857A substitution had very little effect on overall LBD structure with an rmsd of only 0.41 Å for all atoms within6Åof bound ligand (excluding Thr/Ala-877) and a peptide backbone deviation of only 0.81 Å over the whole molecule (data not shown). However, the LNCaP mutation causes a significant rearrangement of the hydrogen bonding network between receptor and bound ligand (Fig. 4D), resulting in a lowered constraint on the ligand D ring and lowering selectivity of the AR-LBD for ligands with differing D ring substituents at the 17β position. In addition, the LNCaP mutation results in an ≈5% increase in volume of the ligand-binding pocket volume, which may explain the ability to bind progesterone.

Discussion

Although mechanisms have been proposed to explain how AR may facilitate prostate cancer (for review, see refs. 30 and 31), our new data support the hypothesis that AR is a protooncogene and that abrogation of the classical AR signal pathway by mutation or hormonal perturbation can facilitate the transformed state. Indeed, these data are consistent with a report that immunization against luteinizing hormone-releasing hormone proteins was unable to decrease prostate cancer in a transgenic model (32) as well as the Prostate Cancer Prevention Trial report showing that androgen signal titration by finasteride-mediated inhibition of 5α-reductase type II could select for more advanced disease (33). Although the AR status in the patients receiving finasteride therapy was not determined, our data suggest the possibility that mutations will be found in the AR gene or other components of the androgen signaling axis. Observations that tumors in patients receiving intermittent androgen blockade and cyclic androgen ablation and testosterone replacement therapy progress slower than for continuous androgen ablation (34–36) further support the hypothesis that maintaining an intact and functional AR signaling axis can be beneficial.

The paradoxical ability of the AR to both drive prostate growth and limit prostate cancer progression may represent compartment-specific roles within the prostate microenvironment (37). We have proposed that the consequence of androgen action in the prostate stroma is to elaborate polypeptide growth factors critical for the growth and survival of the epithelial compartment and that abrogation or supplementation of such signals in vivo can lead to profound phenotypes (38). Conversely, the consequence of androgen action in the epithelial compartment is to suppress proliferation and maintain terminal differentiation and cellular function. By our scheme, systemic androgen ablation not only abrogates the elaboration of stromal factors that support epithelial growth, but also removes a differentiation signal that constitutes selective pressure for the proliferation of epithelial cells or progenitors that had, through somatic mutation or other mechanisms, acquired the ability to grow in a stromal independent fashion. This model not only explains the initially dramatic response observed in prostate cancer patients to hormone withdrawal, it also provides a paradigm for the subsequent emergence of hormone therapy-resistant disease from either a non-AR stem population or one already expressing the AR. Given that hormone ablation selects for AR mutations and these mutations can cause the development of aggressive and metastatic disease, it could be argued that maintaining an intact androgen signaling axis would be of benefit to most patients. Collectively, these studies suggest that prevention trials based solely on androgen titration paradigms should proceed cautiously.

Because AR protein is overexpressed in advanced prostate cancer, increased AR signaling may contribute to tumor progression and emergence of the hormone-independent phenotype (39). However, studies on the LNCaP cell line indicate that although low-level activation of (albeit mutated) AR can facilitate cell growth, hyperactivation can lead to cell death (40), consistent with our hypothesis that AR has properties consistent with a tumor suppressor and that abrogation of AR action is associated with proliferative disease. This is reminiscent of the case of p53. Indeed, we show that enforced expression of AR-WT did not cause significant prostate disease in vivo. Our current data show that increased expression of WT AR or increased AR sensitivity through mutation (AR-T857A) in the presence of endogenous AR cannot initiate the prostate cancer pathway.

The findings of this study also address the functional role of mutations such as AR-T857A. It is known that the LBD of mouse AR-WT and AR-T857A have identical amino acid sequences to the human WT and LNCaP variants, respectively, and can therefore be represented by the solved crystal structures of these receptors (28, 29). As shown in Fig. 4, the T857A substitution causes a significant rearrangement of the hydrogen bonding network between receptor and bound ligand, resulting in a lowered constraint on the ligand D ring. This consequently lowers selectivity of the AR-LBD for ligands with different D ring substitutions at the 17β position, which together with a ≈5% increase in volume of the ligand-binding pocket volume, explains the capacity of the mutant receptor to bind ligands such as progesterone. However, analysis of these structures indicates that T857A has very little effect on overall LBD structure withanrmsdofonly0.41Åforallatomswithin6Åof bound ligand (excluding Thr/Ala-857) and a peptide backbone deviation of only 0.81 Å over the whole molecule (data not shown). We therefore propose that AR-T857A did not promote prostate cancer in mice because this substitution does not alter receptor function per se, but allows it to be activated indiscriminately. In contrast, we have previously demonstrated that E231G provides AR with a true gain of function, namely an increased baseline activity in the absence of hormonal signals and an enhanced functional response to the coregulators (5). That the AR-NTD adopts a loose flexible arrangement with limited structural content has so far precluded crystal analysis of this region of the receptor. However, chemical cleavage has defined a few small regions of structure within the AR-NTD, including an alpha helix that spans the highly conserved ARNSM and the E231 residue, as critical determinants of AR transcriptional activity (19). Changes in structural order across the ARNSM, as suggested by modeling of the E231G substitution, are predicted to result in an altered capacity of the AR-NTD to assemble components of the transcription complex (19), including the CHIP E3 ligase that was recently shown to interact directly with this conserved region (21). That enforced expression of AR-E231G but not AR-T857A in the mouse prostate results in tumor formation underscores the more dramatic consequence of the NTD substitution with respect to receptor function. Our results support other evidence that programs regulated by transcription factors, including steroid receptors, are critically dependent on the nature of multiprotein complexes recruited during all phases of activation (41).

In summary, we have presented an example that a variant nuclear steroid receptor is sufficient for the development of cancer, confirmed a critical domain in the AR-NTD that may act through coregulators to modulate AR function, and distinguished true gain of function AR variants from those that act to facilitate promiscuous ligand binding. Importantly, the AR-E231G transgenic mouse represents a new model of prostate cancer that is independent of potent oncogenes or supraphysiological levels of steroids. This model will facilitate preclinical testing of emerging strategies that aim to inhibit the growth of prostate tumors by directly targeting the AR.