|Home | About | Journals | Submit | Contact Us | Français|
Progression of prostate cancer following castration is associated with increased androgen receptor (AR) expression and signaling despite AR blockade. Recent studies suggest that these activities are due to the generation of constitutively active AR splice variants, but the mechanisms by which these splice variants could mediate such effects are not fully understood. Here we have identified what we believe to be a novel human AR splice variant in which exons 5, 6, and 7 are deleted (ARv567es) and demonstrated that this variant can contribute to cancer progression in human prostate cancer xenograft models in mice following castration. We determined that, in human prostate cancer cell lines, ARv567es functioned as a constitutively active receptor, increased expression of full-length AR (ARfl), and enhanced the transcriptional activity of AR. In human xenografts, human prostate cancer cells transfected with ARv567es cDNA formed tumors that were resistant to castration. Furthermore, the ratio of ARv567es to ARfl expression within the xenografts positively correlated with resistance to castration. Importantly, we also detected ARv567es frequently in human prostate cancer metastases. In summary, these data indicate that constitutively active AR splice variants can contribute to the development of castration-resistant prostate cancers and may serve as biomarkers for patients who are likely to suffer from early recurrence and are candidates for therapies directly targeting the AR rather than ligand.
The androgen receptor (AR) is a principal driver of prostate cancer progression (1). This concept was first established by Huggins et al., with the demonstration that castration slowed, albeit temporarily, the progression of prostate cancer (2). Subsequent castration-resistant growth of prostate cancer has been attributed to a variety of mechanisms that include activation by receptor tyrosine kinases from growth factors, loss of cell cycle regulators, and rarely, genomic mutations in the AR allowing response to nonspecific AR ligands, such as progesterone or glucocorticoids (3–6). More recently, it was demonstrated that increased expression of the AR was the most common event associated with castration-resistant growth (7). Other studies support a process of metabolic adaptation, involving intracrine androgen synthesis (8–10). However, even when agents are used that decrease the tumoral androgen concentrations to very low levels, increased AR expression and signaling persists (11). In a percentage of tumors, the progression of prostate cancer is associated with activating AR mutations, but these events are infrequent (12). These observations suggest the possibility of alternative mechanisms, independent of androgenic ligands that maintain AR program activity in castration-resistant prostate cancers (CRPCs).
Recently, studies of cell lines and prostate cancers have identified several alternative splice forms of the AR (12–14). These AR variants have somewhat different structures, although each variant lacks portions of the ligand-binding domain (LBD), a feature predicted to produce a constitutively active receptor. Interestingly, the elevated expression of AR splice variants was found to be associated with more rapid disease recurrence following radical prostatectomy for localized disease, when compared with patients with lower expression of the variant (13, 15). Of additional interest, the splice forms were not expressed in the nucleus of normal prostate epithelium and rarely at substantial levels in primary prostate cancer. These data suggest that the presence of constitutively active splice variants of the AR arises following castration and plays a role in the progression of prostate cancer.
In this study, we report the identification and characterization of what we believe to be a previously unrecognized AR splice variant that comprises the full sequences of exons 1–4 and the full sequence of exon 8, skipping exons 5, 6, and 7 (hereafter referred to as ARv567es, in which “v” denotes variant, and “es” denotes exons skipped). Because of the alternative splicing of exon 4 to the beginning of exon 8, a frame shift occurs that generates a new stop codon after the first 29 nucleotides of exon 8. Thus, the ARv567es protein is not only smaller than the wild-type AR, but it terminates in a 10–amino acid sequence that we believe to be unique. We determined that ARv567es is not only constitutively active but also increases expression of full-length AR (ARfl) in the absence of ligand.
To identify alterations in the AR that could contribute to the growth of CRPC, we used RT-PCR to measure AR transcript size in a panel of 25 different prostate cancer xenografts, termed the LuCaP series. Most of the LuCaP xenografts were derived from metastases obtained from men with CRPC, after prolonged exposure to androgen-deprivation therapy (ADT); however, their responses to castration, when grown in SCID mouse hosts, vary (Supplemental Table 1; supplemental material available online with this article; doi: 10.1172/JCI41824DS1). We used 2 sets of primers to amplify exons 1–3 and exons 2–8 of the ARfl cDNA (NCBI accession number NG_009014).
As shown in Figure Figure1A,1A, 2 of the LuCaP xenografts, 86.2 and 136, express shorter AR transcripts in the region spanning exons 2–8, compared with ARfl amplified in the remaining xenografts. We sequenced the short AR transcripts from LuCaP 86.2 and 136 and found identical cDNA sequences. In comparison to ARfl, the variant AR mRNA from the LuCaP xenografts 86.2 and 136 lacks exons 5, 6, and 7, which encode the LBD of AR (ARv567es; GenBank GU208210) (Figure (Figure1A).1A). While the full nucleotide sequence of exon 8 is present, due to the splicing of exon 4 to exon 8, a frame shift occurs in the ORF of ARv567es. This frame shift results in a stop codon after the first 30 nucleotides, and thus, the amino acid sequence of exon 8 in ARv567es is shortened to a 10–amino acid sequence, when compared with the amino acid sequence of exon 8 in the ARfl (Supplemental Figure 1). The ARv567es protein, therefore, is predicted to be 739 amino acids, compared with the 920–amino acid protein for ARfl. The 2 xenografts, 86.2 and 136, were derived from 2 different patients who had undergone ADT during their clinical course. To address whether the ARv567es variant was derived from a genomic mutation or alternative mRNA splicing, we extracted genomic DNA from the LuCaP 86.2 and 136 xenografts and determined the sequence of each AR exon and the intron flanking regions and found no differences when compared with the reference AR entry in GenBank (NM_000044). These results suggest that the ARv567es variant transcript is a result of alternative mRNA splicing. We then designed primers to specifically detect the exon 4–8 junction present in ARv567es, reexamined all the xenografts for the presence of the variant, and detected some level of variant AR in almost all xenograft samples (Figure (Figure1,1, A and B). Interestingly, xenograft pairs comprising an androgen-sensitive derivative and a castration-resistant derivative expressed higher mRNA levels of ARv567es in the castration-resistant sample (denoted by AI) compared with the androgen-sensitive sample (35 vs. 35AI and 96 vs. 96AI; P < 0.05).
To investigate the function of ARv567es in both benign and cancer prostate cells, we cotransfected the P69 benign immortalized prostate epithelial cell line and the AR-null M12 human prostate cancer cell line, with the AR activity reporter construct pGL3-probasin ARE-ARR3-luciferase (ARR3-Luc) and a construct expressing ARv567es (referred to as P69 ARv567es cells and M12 ARv567es cells, respectively). We also transfected the AR-null M12 cells with a wild-type ARfl (referred to as M12 ARfl cells). The pcDNA-ARv567es expression vector expressed a protein of appropriate size for the ARv567es cDNA in both benign and cancer cells (Figure (Figure2A2A and Figure Figure3A).3A). Interestingly, expression of ARv567es in the benign P69 cells also resulted in expression of ARfl (Figure (Figure2A).2A). Normally, ARfl is only seen in these cells following dihydrotestosterone (DHT) treatment. Expression of the variant AR did not have this effect in the M12 cells (Figure (Figure3A),3A), which remained AR negative even after DHT treatment. As shown in Figure Figure2B2B and Figure Figure3B,3B, expression of ARv567es resulted in marked activity of the ARR3 reporter in the absence of the AR ligand DHT. The P69 ARv567es cells, unlike the M12 ARv567es cells, demonstrated a further enhancement in reporter activity following DHT treatment (Figure (Figure2B2B and Figure Figure3B).3B). The AR antagonist flutamide had no affect on the ARR3 reporter activity in M12 ARv567es cells, whereas flutamide completely blocked DHT transactivation in the M12 ARfl cells (Figure (Figure3B).3B). These data are consistent with a constitutively active ARv567es without a functional LBD. In contrast, cells expressing an ARfl required DHT to transactivate the ARR3 reporter construct.
To further evaluate the mechanism involved in constitutive ARv567es activation, we stained the M12 cells transiently transfected with ARfl or ARv567es with an antibody specific to the N terminus of AR (sc441). In the absence of DHT, ARfl was localized predominantly in the cytoplasm but translocated into the nucleus in the presence of DHT. In contrast, cells expressing ARv567es consistently showed AR localized in the nucleus, regardless of the presence of DHT (Figure (Figure3C). 3C).
When ARv567es was stably transfected into the ARfl-positive LNCaP cells (LNCaP ARv567es cells), we observed not only expression of the variant AR but also an increase in the amount of endogenous ARfl protein compared with mock-transfected LNCaP cells (LNCaP pc cells) (Figure (Figure4A4A and Figure Figure5A).5A). This finding was consistent with observations in P69 cells engineered to express ARv567es (Figure (Figure2A).2A). LNCaP ARv567es cells grown in medium containing charcoal-stripped (CS) serum had the cuboidal appearance of control LNCaP pc cells grown in the presence of DHT (Figure (Figure4B),4B), whereas the control LNCaP pc cells grown in CS medium exhibited the expected elongated, stressed appearance of LNCaP cells grown without DHT. This alteration in morphology in the absence of DHT suggested that ARv567es might function dominantly as a constitutively active AR in prostate cells that also express a ARfl.
Following the introduction of the ARv567es construct into LNCaP cells (referred to as LNCaP ARv567es cells), an increase in basal ARR3 luciferase reporter activity was observed as well as a marked increase in reporter activity in response to DHT (Figure (Figure4C),4C), a finding that is concordant with the responses measured in P69 cells expressing ARv567es. We then evaluated the DHT-induced changes in the expression of several androgen-induced genes (kallikrein-related peptidase 3 [PSA, also known as KLK3], transmembrane protease, serine 2 [TMPRSS2], FK506 binding protein 5 [FKBP5], NK3 homeobox 1 [NKX3.1], insulin-like growth factor 1 receptor [IGF1R]) using RT-PCR in the LNCaP pc cells versus the LNCaP ARv567es cells (Figure (Figure4D).4D). The LNCaP ARv567es cells had significantly higher mRNA levels for PSA, TMPRSS2, and FKBP5 compared with those of LNCaP pc cells with and without DHT (P < 0.001 and P < 0.0001 compared with LNCaP pc cells with same treatment.). NKX3.1 mRNA levels in the LNCaP ARv567es cells were equally high, regardless of DHT exposure. Interestingly, IGF1R mRNA levels decreased in cells expressing ARv567es compared LNCaP pc cells (Figure (Figure4D),4D), suggesting there is a functional difference between activation of ARfl by ligand compared with the constitutively active ARv567es. We expect that the difference in IGF1R mRNA levels is due to the fact that a nongenomic AR pathway regulates IGF-IR transcription; since ARv567es primarily resides in the nucleus, the nongenomic pathway is not activated in these cells. Proliferation assays demonstrated that the LNCaP ARv567es cells proliferated in response to lower concentrations of DHT than control LNCaP pc cells (Figure (Figure4E).4E). Together, these data suggest that the presence of ARv567es enhances the transcriptional response of the endogenous ARfl to androgens.
Because AR can autoregulate its own transcription and we saw an increase in ARfl protein in the LNCaP ARv567es cells, we wanted to determine whether the increased effect of the variant on androgen-regulated gene expression was simply through an increase in ARfl protein levels. We overexpressed ARfl in LNCaP cells and then performed ARR3-Luc reporter assays. Overexpression of ARfl did result in increased ARfl protein expression and increased reporter activity compared with that of LNCaP pc cells (Figure (Figure5,5, A and B). But more importantly, even though LNCaP ARv567es cells had a similar increase in ARfl protein expression, they had significantly increased reporter activity compared with LNCaP ARfl cells, indicating that the variant AR is affecting transcriptional activity to a greater degree than when ARfl alone is amplified (Figure (Figure5,5, A and B).
To further explore the mechanisms by which ARv567es increases the activity of ARfl, we expressed both ARfl and a HA-tagged ARv567es in the AR-null M12 cells and immunoprecipitated ARv567es from cell lysates with an anti-HA antibody. Western blots were done on the immunoprecipitates using an N terminus–directed AR antibody (sc441), which recognizes both ARfl and ARv567es, and a C terminus AR–specific antibody (C-19), which recognizes only ARfl, to detect any immunocomplex of the 2 ARs. As shown in Figure Figure6A,6A, ARfl coprecipitated with ARv567es in the presence or absence of DHT, indicating a physical association of ARv567es with ARfl. M12 cells transfected with the empty pcDNA vector were used as a negative control, since these cells lack an endogenous AR. As a positive control, we cotransfected both an untagged and a Flag-tagged ARfl into M12 cells, immunoprecipitated AR using a Flag antibody, and then immunoblotted with AR C-19 antibody (Figure (Figure6A).6A). In these cells, dimerization of ARfl required ligand as opposed to the association of ARfl and ARv567es observed in the M12 ARfl and ARv567es cells in the absence of DHT. We then transfected M12 cells with either the ARfl construct alone or both the ARfl and ARv567es constructs and performed immunofluorescence staining using the AR C-19 antibody, which detects only ARfl protein. When both ARfl and ARv567es were present, ARfl translocated to the nucleus in the absence of ligand, whereas it remained in the cytoplasm in cells expressing only ARfl (Figure (Figure6,6, B and C). To determine whether an interaction between ARfl and ARv567es occurs in tumors expressing both AR types, we immunoprecipitated ARfl from lysates taken from castrate-resistant xenografts using the AR C-19 antibody and then immunoblotted them with AR sc441. We detected both full-length and variant AR in LuCaP 136 (strong ARv567es band) and LuCaP 35 (weak ARv567es band) xenografts (Figure (Figure6D).6D). These results suggest that in cells that endogenously express AR splice forms, the ARv567es can functionally interact with ARfl.
Because we observed an increase in ARfl protein levels without a long-term increase in ARfl mRNA levels (Figure (Figure7A)7A) and there appeared to be an interaction between ARv567es and ARfl, we sought to determine whether the interaction between the 2 ARs had an effect on mRNA stability or AR protein degradation. We examined AR mRNA stability following treatment with actinomycin D and found no differences between LNCaP pc and LNCaP ARv567es cells (Figure (Figure7B).7B). We then determined whether the interaction of ARv567es with ARfl affected AR protein degradation. When translation was halted with cycloheximide, there was a slowing in degradation of the ARfl in cells containing both receptors, particularly in the presence of DHT, when compared with cells with only the endogenous AR receptor (Figure (Figure7,7, C and D). These data are also consistent with studies showing that AR activation increases AR protein through increased translation efficiency and a decrease in the rate of AR degradation (16).
Since the ARv567es does not contain a LBD, androgen ablation has no effect on signaling. Thus, in tumor cells expressing ARv567es, alternative approaches for abrogating AR signaling will be required. The histone deacetylase inhibitor SAHA has been shown to alter levels of AR protein by inhibiting AR transcription (15). Therefore, we grew LNCaP pc and LNCaP ARv567es cells in vitro to 80% confluence with either CS serum or with CS serum plus DHT. SAHA was then added to the medium at the concentrations indicated in Figure Figure8A.8A. In the absence of DHT, SAHA markedly decreased ARfl expression in LNCaP pc cells and ARv567es and endogenous ARfl in the LNCaP ARv567es cells. However, in the presence of DHT, SAHA only effectively decreased ARfl expression and had a minimal effect on ARv567es expression. In order to determine whether this same effect of androgens would occur on tumors that were resistant to castration and in which the primary AR consisted of ARv567es, we cultured LuCaP 86.2 tumors in vitro and treated the cells with SAHA or SAHA plus DHT. SAHA was effective at decreasing ARv567es in the presence and absence of DHT (Figure (Figure8,8, B and C). Although SAHA decreased ARv567es protein levels in the LuCaP 86.2 cells, the resulting decrease in growth (Figure (Figure8D)8D) cannot be conclusively attributed to loss of ARv567es, since histone deacetylases have multiple effects on cells that result in suppression of cell proliferation (17, 18).
In order to compare and contrast the gene expression program regulated by ARv567es and ARfl, we first measured transcript abundance changes in LNCaP cells expressing ARfl following exposure to androgen. Cells were grown in CS serum, with or without 10–9 M DHT, for 24 hours in triplicate. Following RNA isolation, transcript abundance levels were measured using whole-genome microarrays. Because the magnitude of changes was quite high between the ARfl CS and ARfl DHT-treated groups, t test significance was set at a stringent q value of less than 0.01%. A separate experiment compared transcript abundance in LNCaP cells expressing ARfl or ARv567es grown in CS medium. Gene expression differences between ARfl CS and ARv567es CS groups were more subtle, and genes exhibiting q values of less than 10% were included in order to cross-compare similar numbers of gene expression changes between the experiments (Figure (Figure9).9). We performed an analysis of Gene Ontology (GO) using the EASE software tool, which calculates overrepresentation statistics for GO terms in the significant list, with respect to all genes represented in the data set (Supplemental Tables 2 and 3, GO analysis statistics).
We determined that well-known androgen-regulated genes, such as PSA, TMPRSS2, and FKBP5, altered by DHT treatment in cells expressing ARfl, were also altered in the context of ARv567es expression in the absence of exogenous androgens, confirming the hypothesis that ARv567es is capable of activating the AR-regulated gene expression program in the setting of castration. We further found that ARv567es regulates a subset of genes that we believe to be unique that are not influenced by androgens in the context of the ARfl. Analysis of GO terms enriched specifically in ARv567es cells revealed that GO molecular function “transcription factor activity” is significantly increased in the absence of androgen, potentially signifying activation of other growth and survival pathways. Among the transcription factors upregulated by ARv567es that are known to induce a proliferative program of gene expression were STAT3, which has antiapoptotic as well as proliferative effects, and JUN, which behaves as a positive regulator of cell growth by protecting cells from p53-dependent senescence and apoptosis (19–21). ARv567es also activated genes involved in the metabolism of androgens. Interestingly, even in the absence of exogenous AR ligands, cells expressing ARv567es demonstrated a signature of androgen metabolism, with enriched GO terms of biological processes “steroid biosynthesis” and “sterol metabolism.” Such ligand-independent stimulation of steroidogenic pathways in ARv567es cells may provide a survival advantage in a low-androgen environment.
In addition to differences in the genomic signaling programs between ARv567es and ARfl, our data also indicate that these variant receptors may differentially regulate other AR activities. In this context, ARv567es altered components of IGF pathway signaling in a manner distinct from that of the ARfl. IGF-IR expression was enhanced by ligand stimulation of ARfl, whereas ARv567es suppressed IGF-1R expression (Figure (Figure4D4D and Figure Figure9B).9B). Pandini et al. have shown that the AR stimulates IGF-IR expression through a nongenomic pathway, involving binding of ARfl to Src and downstream activation of the transcription factor MAPK (22). Our cell-based localization studies indicated that ARv567es was rapidly targeted to the nucleus and, during this process, facilitated the movement of ARfl as well. Thus, ARv567es may have a primary effect through enhanced genomic AR activity and the concomitant suppression of nongenomic AR activity exerted by ARfl.
Since ARv567es is constitutively activated and also transactivates ARfl in the absence of androgen, we next sought to determine whether ARv567es influences prostate cancer responses to ADT in vivo. We implanted LNCaP ARv567es cells or control LNCaP pc cells into immunocompromised nu/nu mice in subcutaneous locations. In the initial androgen-sensitive growth phase in eugonadal animals, there was no difference in tumor growth rate among the cell lines inoculated (data not shown). After tumors attained a size of approximately 0.2 cc, the mice were castrated, and tumor volumes were measured over a 13-week time period. We did not observe significant growth of the grafts comprised of LNCaP pc cells. However, grafts comprised of LNCaP ARv567es cells were measurably larger than those of control cells by the 10-week time point (tumor volume of LNCaP pc vs. LNCaP ARv567es cells; P < 0.01), and subsequent rapid growth required animal sacrifice by 13 weeks due to tumor size (Figure (Figure10A).10A). We examined the resulting tumors, resected at the study endpoint for AR transcript levels and splice variants. In the small LNCaP pc tumors surviving castration, transcript levels encoding ARfl were not significantly different relative to levels in cells prior to castration, and the ARv567es remained undetectable. In contrast, in LNCaP ARv567es tumors the ratio of ARv567es transcripts to those encoding ARfl increased significantly from 0.5 to 4.3 (P < 0.05). These results support evolution of a tumor cell population favoring the selection of cells expressing higher levels of the ARv567es variant.
We next examined the response of prostate cancers expressing endogenous ARv567es to ADT. We chose 3 representative LuCaP xenografts, which have distinct differences in the expression of ARfl and ARv567es (Figure (Figure10E).10E). The LuCaP 35 xenograft expressed ARfl, with little detectable ARv567es. In addition, we have previously published that LuCaP 35 xenografts maintain high levels of intratumoral DHT following castration (1.7 ± 0.27 ng/g of tumor tissue precastration versus 1.5 ± 0.48 ng/g of tumor tissue after castration; ± SD; P = NS) (8, 9). The LuCaP 136 xenograft expressed ARfl and had high intratumoral DHT levels (2.4 ± 0.9 ng/g) in eugonadal animals, but following castration there was a significant decrease in intratumoral DHT (0.15 ± 0.03 ng/g; P < 0.001) and increased expression of ARv567es (Figure (Figure10E).10E). Finally, the LuCaP 86.2 xenograft, which had low levels of intratumoral DHT regardless of castration status (0.4 ± 0.04 ng/g precastration vs. 0.1 ± 0.04 ng/g after castration; P < 0.05), expressed primarily ARv567es (Figure (Figure10E).10E). We implanted replicate xenografts subcutaneously into SCID mouse hosts, 24 animals per group, and after achieving a tumor volume of 0.2 cc, half of the animals in each group were subjected to ADT by surgical castration. Tumor responses correlated with the level of ARv567es expression. No response to ADT was seen for the LuCaP 86.2 xenograft, which predominately expressed ARv567es (Figure (Figure10,10, B and E). The growth of LuCaP 136 tumors, which express a mix of ARfl and ARv567es following castration, was significantly suppressed following ADT, relative to eugonadal animals (14-week time point; P < 0.01), although there was a slow progressive increase in tumor volumes over time (Figure (Figure10,10, C and E). LuCaP 35 tumors, which predominately express ARfl, exhibited a sustained response to ADT, with minimal tumor growth measured over 14 weeks (Figure (Figure10D). 10D).
Since ARv567es is a splice variant and generation of the variant is posttranscriptional, acquisition of ARv567es by the prostate could be considered an adaptive mechanism rather than a selected event, such as those encoded by a mutation in AR DNA. This would appear to be the case for the LuCaP 136 and 86.2 xenografts. When we evaluated the original tissue specimens from which these xenografts were derived, ascites cells and a bladder metastasis, respectively, only ARfl mRNA was present in the original LuCaP specimen, whereas both ARfl and ARv567es mRNA were present in the LuCaP 86.2 original specimen (Supplemental Figure 2). This finding was consistent with the LuCaP 136 xenograft initially responding to castration in the mouse, but the 86.2 xenograft having no response. However, after castration and continued passage in SCID mice, ARv567es mRNA became the dominant AR mRNA in the LuCaP 86.2 xenograft, and ARv567es mRNA was generated in the LuCaP 136 xenograft (Supplemental Figure 2 and Figure Figure10E).10E). Further, as mentioned earlier, in those LuCaP xenografts with pairs of castrate-sensitive and castrate-resistant tumors, there was a significant increase in ARv567es expression in the castrate-resistant tumors compared with castrate-sensitive tumors (Figure (Figure1).1). These data indicate that the generation of the splice variants is a dynamic process that occurs in response to castration pressures exerted on the AR signaling program.
To further evaluate the contribution of the ARv567es variant to prostate cancer, we investigated the distribution of the AR splice variants directly in normal and neoplastic human prostate epithelium. We first acquired benign prostate epithelium by laser-capture microdissection (LCM) from prostate biopsies of 36 normal men, without evidence of prostate cancer, enrolled in a study of male contraception (acycline or DHT gel). Using quantitative RT-PCR (qt-RT-PCR) to quantitate transcripts isolated from benign epithelium, we were able to detect mRNAs encoding ARfl in 35 subjects, and 6 out of 36 (17%) men expressed ARv567es, 2 out of 36 (6%) men expressed the recently described AR splice variant termed AR-V7, and 1 out of 36 (3%) men expressed the AR splice variant termed AR3 (13, 14) (Table (Table11 and Figure Figure11A).11A). The men who expressed AR variants, while they did not have prostate cancer, had either been castrated using the LHRH receptor antagonist acycline or had tissue androgens suppressed by administration of DHT gel.
Since we found ARv567es in our group of normal men, we next evaluated whether in primary (untreated) prostate cancer specimens the variant was in the malignant epithelium, benign epithelium, or both; if primarily in the malignant epithelium, this might suggest that the presence of a variant was etiologic in cancer development. Therefore, we examined cDNA from laser-captured, matched benign and malignant epithelium of primary prostatectomy tissue. We found that the variants were expressed in both benign and malignant prostate epithelial tissue (Figure (Figure11A).11A). These data suggest that the presence of AR variants is not necessarily etiologic in the initiation of prostate cancer.
To determine the frequency of AR splice variant expression in advanced prostate cancers, we successfully amplified cDNA from 69 metastases, derived from 13 patients undergoing a rapid necropsy for tumor acquisition. All of these patients were documented to have either surgical or chemical castration, with tumor progression to a clinical state of CRPC. Tumor tissue was isolated by laser-assisted microdissection, and RNA integrity was verified. Despite amplification of GAPDH, no full-length or variant AR expression was detected using RT-PCR in 23 samples. These 23 metastases were from patients whose primary tumors had a neuroendocrine phenotype and would not be expected to depend on functional AR. Of the remaining 46 metastases, 37 out of 46 (80%) expressed ARfl, 20 out of 46 (43%) expressed ARv567es, 11 out of 46 (24%) expressed AR-V7, and 3 out of 46 (6%) expressed AR3 (Table (Table11 and Figure Figure11B).11B). Interestingly, 9 out of 46 (20%) metastases expressed only the ARv567es variant, but expression of AR-V7 or AR3 occurred concomitantly with ARfl expression. Overall, 27 out of 46 (59%) AR-positive metastases expressed one or more AR splice variants. When evaluating all of the metastases from each individual patient collectively, 12 out of 13 patients had at a minimum one metastasis that was positive for at least one AR splice variant, and 10 out of 13 patients had at a minimum one metastasis that was positive for ARv567es. Further, the presence of an AR variant was clustered among patients. For example, the majority of metastases for one patient may have been positive for a variant, whereas a different patient may have very few metastases that were positive for that variant. No patient had all samples positive for the same AR variant.
The progression of prostate cancer to an AR-independent state has recently become regarded as a misnomer (8–10, 23). Transcriptional activation of the AR has been suggested to occur through several nonsteroidal factors, such as IL-6, EGF, integrins, and insulin-like growth factors as well as by increased expression of AR cofactors (3, 24–29). Another mechanism for AR-regulated prostate cancer progression following castration is an alteration in AR protein. This alteration has been reported to occur by genomic mutations: either ligand-dependent activating mutations or ligand-independent activating mutations (12). Genomic mutations in the AR are not common, but those that result in an AR that functionally increases AR activity and lead to CRPC are even less common. Recently, constitutively active posttranslational splice variants of the AR have been reported in human prostate cell lines, xenografts, and human tissues (13, 14, 30).
In this paper, we describe a splice variant of the AR that we believe to be new, in which exon 4 has been spliced to exon 8, and thus, exons 5, 6, and 7 have been deleted. This alternative splicing event leads to a frame shift, so that 10 amino acids that we believe to be unique are encoded for and followed by a stop codon, resulting in an AR variant protein that is 180 amino acids shorter than the ARfl protein. Thus, this variant, unlike the previously reported AR splice variants, retains the hinge region, which is necessary for complete nuclear translocation of the AR (31). Similar to previously reported AR splice variants, we show that the AR splice variant produced by this event is constitutively active (13, 14, 30). Unlike other reports, though, in which variants were cloned using a 3′ RACE method in a prostate cancer cell line or in which the authors created potential deletion mutants based on the different splicing sites in the AR, we identified ARv567es through the analysis of 25 distinct human prostate xenografts. We originally sequenced the ORF of the AR in the xenografts using 2 sets of PCR primers, as described. Since these primers detect the predominant AR mRNA species, 2 of the xenografts, 86.2 and 136, had the splice variant as the majority of their AR. When specific primers were designed to detect the 4–8 junction present in ARv567es, the variant was found in several additional xenografts. We also noted that in the same xenograft, the level of the ARv567es was not constant and varied depending on passage number. When we examined the primary human tumors from which the LuCaP 86.2 and 136 xenografts were developed, we noted that the variant AR had always been present in LuCaP 86.2, while in LuCaP 136 the ARfl was the predominate receptor. However, the AR variant became more predominant in the LuCaP 136 xenograft following long-term growth in mice as well as after it was grown in castrated mice. These data suggest that the level of splice variant AR is not fixed but is dynamic and likely regulated by environmental factors, including decreasing androgen levels (12, 13).
We expressed this new AR variant in both benign and prostate cancer cell lines and found that ARv567es functions like a constitutively active AR in both types of cells and is resistant to the effects of flutamide on AR activity. Further, we report that in cells expressing endogenous ARfl, ARv567es increases the level of endogenous receptor by slowing the rate of protein degradation.
Since a constitutively active AR could lead to the progression of the tumor after castration, we examined expression of the ARv567es variant in specimens of prostate cancer from men who had died from their cancer. All of these men were castrated at the time specimens were obtained; while a third of the samples did not have any detectable AR, over 50% of the specimens that contained detectable AR transcripts were positive for one or more of the constitutively active variants. These included biopsies from the prostate as well as lymph node, lung, liver, and bone. Interestingly, the presence of ARv567es was not distributed randomly among the patients but rather the variant clustered within a patient’s specimens. Further, not all specimens within a patient were positive for an AR variant, consistent with a hypothesis of ongoing splicing activity that generates AR variants. We also examined specimens for the presence of 2 other AR variants recently reported to affect survival in prostate cancer, AR3 and AR-V7 (13, 14). ARv567es was the most common of the 3 variants examined and several specimens contained more than one AR variant. Additionally, nearly all of the specimens that contained one or more of the variants also contained a ARfl. These findings are not surprising, since there are multiple mechanisms by which the AR continues to be activated following castration.
Since AR splicing is determined in part by differences in androgen levels, we asked whether the splice variants might be present in men without cancer and in benign epithelium in men with cancer. In young men, ages 35–55, with normal PSA levels of less than 2.0 ng/ml and no clinical evidence of prostate cancer, who had a prostate biopsy as part of their participation in a male contraceptive study, we found that the ARv567es variant was occasionally expressed in prostate epithelium of men receiving treatment but not in men receiving placebo. In a separate set of non-castrate older men, ages 59–70 years, who had a radical prostatectomy for prostate cancer, analysis of paired laser-captured benign and malignant prostate epithelial samples showed that AR splice variants could be found in benign, malignant, or both specimens from the men. These data suggest that expression of AR variants is not necessarily etiologic in cancer development, but if AR variants are present, they might serve to facilitate progression when hormone suppression is applied. This phenomenon has been demonstrated in a mouse model (32) as well in our current data, using various human xenografts grown in castrated versus intact SCID mice.
In this paper, we have shown that some prostate tumors may express only the variant AR receptor. However, while expression of AR splice variants is of interest and their functions have been examined as specific entities, the vast majority of splice variants are expressed in cells, along with a full-length receptor. Dimerization of full-length receptors occurs through both AF1 and AF2 sites as well as other N-terminal/C-terminal (N/C) interactions (33). Whereas most steroid receptors require the N/C interaction for binding of cofactors and subsequent receptor activation, the AR is unique in that the AR AF2 domain has evolved such that it preferentially interacts with FxxLF-like motifs contained in the AR N-terminal domain (23FQNLF27 and 433WHTLF437) over LxxLL-like motifs in the coregulatory proteins’ amino termini. As such, an AR lacking the LBD is still active, but now ligand cannot induce dimerization (33).
Although we and others have shown the constitutive function of the AR variants independent of ARfl, the majority of tumors express both ligand-dependent and constitutively active splice receptors. In this study, we believe that we have shown for the first time that a constitutively active AR splice variant, ARv567es, interacts with ARfl. We demonstrated that we could pull down variant AR with ARfl, not only in prostate cancer cells lines in which we overexpressed the variant, but also in human xenografts that endogenously express both full-length and variant AR protein. Further, we show that the variant causes translocation of the ARfl in the absence of ligand, which results in enhanced AR transcriptional activity and increased cell proliferation in response to very low concentrations of ligand. However, the characterization of where the splice variant binds and interacts with ARfl has yet to be determined.
Recent studies and data we presented here have shown that after standard androgen ablation, such as castration, levels of androgen in the prostate are still measurable. In our studies, using a panel of human xenografts that we believe to be unique, we further show a strong correlation between increased expression of the different AR variants in tumors and low intratumoral levels of androgens. In particular, the ARv567es is expressed most abundantly in tumors with the lowest levels of androgens. Thus, the expression of the AR variants, including the ARv567es variant that we believe to be unique, provides a mechanism to enable cancer progression in the face of decreased androgens.
ARv567es contains the entire DNA-binding domain and nuclear localization sequence (Supplemental Figure 1). It is clear that ARv567es is constitutively active and can function without ligand or ARfl. On the other hand, we have also shown that ARv567es can form a heterodimer with ARfl and enhance its activity (Figure (Figure4).4). Thus, for the first time to our knowledge, we show that a constitutively active AR variant generated by castration in human tumors may function in a ligand-independent manner or enhance the ligand-dependent function of the ARfl. Either one or both of these mechanisms may be active in a tumor cell when both variant and ARfl coexist.
Since ARv567es does not bind ligand and has no N/C interaction, it might be expected that its interaction with AR coregulators and its binding to androgen response elements (AREs) may be altered. We have shown the latter in this study, since we saw enhanced activation of ARE-containing reporter constructs in cells overexpressing ARv567es compared with cells simply overexpressing ARfl. Further, in cells expressing ARv567es, we observed a decrease in genes that are upregulated via the nongenomic pathway, including IGF-1R. This would suggest that the repertoire of genes activated by the splice variant receptors will have similarities and differences with that of the full-length ligand-dependent AR. In this study, we have seen that there are significant differences in profiles of androgen-regulated genes in LNCaP cells that only express the endogenous AR and those that express the splice variant, even when androgens are added to the culture medium of the control LNCaP cells. These differences in gene expression by cells that express the AR splice variants may be significant, because the AR has tumor-promoting and -suppressing properties. Since the AR splice variants are associated with cancer progression, the differences in gene expression may help to delineate those AR-regulated genes involved in tumor progression.
Using data from the gene arrays comparing xenografts with high and low levels of expression of the ARv567es variant, we were able to identify a panel of splicing factors that were differentially expressed. Specific regulation of these factors by androgens has not been determined, although several of the factors have been implicated in generating variability of prostate tumors (34–37). If specific splicing factors can be identified that generate these AR splice variants, inhibition of these factors at the time of androgen ablation could significantly enhance the effects of androgen ablation and prevent or delay development of castration-resistant disease.
In summary, we demonstrate in this report that AR splice variants that contribute to castration resistance are common in prostate cancer. There appears to be a family of AR variants, generated after androgen ablation, that results in a constitutively active AR. Identification of these variant receptors may serve as a biomarker for men destined for early recurrence and who are candidates for therapy directly targeting the AR rather than ligand.
The generation and characterization of the M12 prostate cell line has been described previously (38–41). M12 cells were cultured in RPMI 1640 medium supplemented with 5% FCS, 10 ng/ml EGF, 0.02 mM dexamethasone, 5 μg/ml insulin, 5 μg/ml transferrin, 5 ng/ml selenium, fungizone, and gentamicin at 37°C with 5% CO2. The androgen-sensitive LNCaP line was a gift from Robert Sikes (University of Delaware, Newark, Delaware, USA). These cells were grown in T-medium (Invitrogen) supplemented with 10% FBS, 100 units/ml penicillin, and 100 μg/ml streptomycin at 37°C with 5% CO2. The generation of several of the LuCaP xenografts has been described previously (8, 14, 30, 42–45). The origin and characteristics of all 25 xenografts are included in Supplemental Table 1. These xenografts fail to grow as cell lines and do not survive freezing, thus they are maintained by serial passage in SCID mice. The LuCaP 86.2 xenograft is the exception and will grow in vitro for up to a month. LuCaP 86.2 xenografts were removed from animals and digested using collagenase digest buffer, as previously described (46). Dissociated cells were then plated down in RPMI medium plus 10% FBS.
Human primary and metastatic prostate tissues were obtained as part of the University of Washington Medical Center Prostate Cancer Donor Autopsy Program, which is approved by the University of Washington Institutional Review Board. Details of this program have been described previously (47, 48). The Institutional Review Board of the University of Washington Medical Center approved all procedures involving human subjects, and all subjects signed written informed consent. Prostate tissue samples from 36 healthy subjects without prostate cancer were obtained from men enrolled in a clinical trial of medical castration (n = 8) or DHT administration (n = 28) performed at the University of Washington (9) (ClinicalTrials.gov numbers NCT00161486 and NCT00161486, respectively). These studies were comprised of young men, ages 35–55 years, with normal PSA levels of less than 2.0 ng/ml and no clinical evidence of prostate cancer, who had a prostate biopsy as part of their participation in the respective studies. Transrectal ultrasound-guided prostate biopsies of the peripheral zone were immediately embedded in OCT compound (TissueTek OCT Compound, Sakura Finetek) and snap frozen in liquid nitrogen. Under a University of Washington institutional protocol for the use of excess tissue after surgery, 8 matched samples of benign and tumor prostate tissue were obtained from eugonadal patients undergoing radical prostatectomy for localized prostate cancer (49, 50). In addition, benign prostate tissue was obtained from 2 patients undergoing cystoprostatectomy unrelated to prostate cancer. An H&E-stained section of each sample was examined microscopically and none of these demonstrated cancer.
To detect AR splice variants in human prostate tissues, prostate biopsy samples embedded in OCT were used for LCM. Approximately 2,000 to 3,000 epithelial cells per sample were collected from 8-μm sections using the Arcturus Veritas Laser Capture Microdissection System, according to the Arcturus HistoGene LCM Frozen Section Staining Kit protocol. Total RNA was isolated using the RNeasy kit (Qiagen), followed by treatment with DNase using the Qiagen RNase-Free DNase Set. RNA was quantitated in a Gene-Spec III spectrophotometer (Hitachi), and RNA integrity was evaluated using gel electrophoresis. Total RNA from xenografts and cell lines was isolated with RNA STAT-60 (Tel-Test). cDNA was synthesized using SuperScript First-Strand Synthesis System (Invitrogen). PCR was performed using Takara High Fidelity Taq polymerase with the following primers: hAR1113 forward, 5′-AGGATGGAAGTGCAGTTAGGGCT-3′; hAR2974 reverse, 5′-CATTTCCGAAGACGACAAGATGG-3′; hAR2695 forward, 5′-GGATGGATAGCTACTCCGGACCTTACG-3′; and hAR3973 reverse, 5′-CAAGGCACTGCAGAGGAGTAGTGCAGAG-3′. PCR products were subjected to sequencing analysis using the Applied Biosystems 3730XL DNA Analyzer. Betaine (Sigma-Aldrich) was added in PCR and sequencing reactions to a final concentration of 1 M to facilitate the amplification of the AR GC-rich region.
PCR of ARfl was performed using the following primer pair: hAR3619 forward, 5′-ACATCAAGGAACTCGATCGTATCATTGC-3′, and hAR3832 reverse, 5′-TTGGGCACTTGCACAGAGAT-3′. The AR splicing variant, ARv567es, was detected with the following primer pair: hARv567es forward, 5′-CCAAGGCCTTGCCTGATTGC-3′, and hAR3832 reverse, 5′-TTGGGCACTTGCACAGAGAT-3′. All primers were designed using the human AR mRNA reference sequence (GenBank NM_000044). We have submitted the novel AR splice variant mRNA sequence to GenBank (ARv567es; GenBank GU208210).
cDNA of ARfl (AR20) was cloned into pcDNA3 as described previously (41). cDNA of the entire ARv567es variant was amplified by PCR from the LuCaP 86.2 xenograft cDNA, using primer pair hAR1113 forward, 5′-AGGATGGAAGTGCAGTTAGGGCT-3′, and hAR3973 reverse, 5′-CAAGGCACTGCAGAGGAGTAGTGCAGAG-3′, and subcloned into pcDNA3.1 after sequence confirmation. The human influenza hemagglutinin–AR (HA-AR) variant construct was generated by inserting a HA linker into the BamHI site of the pcDNA-ARv567es construct. The HA linker sequences are as follows: HA-s, 5′GATCCATGTACCCATACGATGTTCCAGATTACGCT3′, and HA-as, 5′GATCAGCGTAATCTGGAACATCGTATGGGTACAT3′. These expression constructs were transfected into the human prostate cancer cell lines M12 and LNCaP and the benign immortalized human prostate cell line P69 with Lipofectin 2000 (Promega), according to the manufacture’s protocol. Stable clones were obtained with G418 selection (400 mg/ml), and the expression of ARfl or ARv567es was confirmed by RT-PCR and Western blotting.
M12 cells were grown on glass coverslips on 12-well plates and transfected with pcDNA-Flag-ARfl alone or cotransfected with pcDNA-Flag-ARfl and pcDNA-HA-ARv567es. Twenty-four hours after transfection, cells were treated for an hour with 10–9 M DHT, then fixed with cold 4% formaldehyde, and stained with anti-AR (C19) primary antibody and subsequently with Alexa Fluor 594–conjugated secondary antibody. Cells were counterstained with DAPI. Cells were imaged with an Applied Precision DeltaVision Microscope. AR nuclear translocation was measured by using software Cyteseer 2.0 (Vala Sciences Inc). AR translocation was determined by the ratio of total integrated intensities of the red channel on the nuclear mask (DAPI-stained area) divided by total integrated intensities of the red channel on the whole cell mask. At least 100 cells were analyzed in each group. All measurements were subtracted from the baseline condition (the population with the smallest percentage of cells demonstrating nuclear localization) and then compared with the population demonstrating the highest percentage of cells with nuclear localization.
Cells were lysed in cold lysis buffer (50 mM HEPES, 150 mM NaCl, 1.5 mM MgCl2, 1 mM EGTA, 1% Triton X-100) with Complete Protease Inhibitors (Roche Applied Science). Precleared cell lysate was incubated with the anti-HA antibody (Santa Cruz Biotechnology Inc.) and ultralink immobilized protein A/G plus (Pierce). Immune complexes were separated by SDS-PAGE and transferred onto a nitrocellulose membrane for Western blotting detection of AR. Antibodies AR sc441 and C-19 (Santa Cruz Biotechnology Inc.), specific for the N terminus and C terminus, respectively, of ARfl, were used for detection of AR. For Western blot analysis of cell lysates, cells were washed with PBS and lysed with cold lysis buffer containing Phosphatase Inhibitor Cocktail II (Sigma-Aldrich) and proteinase inhibitors (Complete Mini Tablets, Roche). Twenty-five micrograms of protein were resolved on 4%–15% SDS-PAGE, transferred onto a nitrocellulose membrane, and probed with respective antibodies. The blot was washed and incubated with a horseradish peroxidase–conjugated secondary antibody (Pharmacia Biotech) for 1 hour. Immunoreactive proteins were detected by ECL (Pharmacia Biotech). The membranes were stripped for 30 minutes in Stripping Buffer (Pierce) and reprobed with anti-GAPDH antibody as a loading control (Chemico) as described above. Independent experiments validated that this stripping procedure did not lead to loss of signal.
LNCaP pc and LNCaP ARv567es stable cells were treated with actinomycin D (10 mmol/l) for 0, 2, 6, or 16 hours. Equal amounts of RNA were analyzed for ARfl and ARv567es by real-time PCR (normalized using internal GAPDH control).
To determine whether ARv567es changes ARfl degradation, we used cycloheximide to inhibit new protein synthesis and assess AR protein stability. LNCaP pc cells and LNCaP ARv567es stable cells were treated with 20 μg/ml cycloheximide for 0, 2, 6, or 16 hours. Cell lysates were collected and Western blotted with AR C-19 antibody, which recognizes the AR C terminus (Santa Cruz Biotechnology Inc.).
Total RNA from experimental samples was isolated using the RNeasy Maxi Kit (Qiagen) and amplified for 1 round using the Ambion MessageAmp aRNA Kit (Ambion Inc.), incorporating amino-allyl UTP into amplified antisense RNA. Sample quality and quantity were assessed by agarose gel electrophoresis, and absorbance was measured at A260 using the Nanodrop 1000 spectrophotometer (Thermo Fisher Scientific). Probes were labeled with Cy3 or Cy5 fluorescent dye and hybridized to Agilent 44K whole human genome expression oligonucleotide microarray slides (Agilent Technologies Inc.), following the manufacturer’s suggested protocols. Fluorescence array images were collected for Cy3 and Cy5 using the Agilent DNA microarray scanner G2565BA (Agilent Technologies Inc.), and Agilent Feature Extraction software was used to grid, extract, and normalize data. The data were filtered to exclude Feature Outliers and Population Outliers, as defined by the software, and spots with average signal below 300 were also removed. The Statistical Analysis of Microarray (SAM) program ( http://www-stat.stanford.edu/~tibs/SAM/) was used to analyze expression differences between groups. Unpaired, 2-sample t tests were calculated for each probe and controlled for multiple testing by estimation of q values using the false discovery rate (FDR) method. These results were reduced to unique genes by eliminating all but the highest scoring probe for each gene.
To study the in vivo effect of the ARv567es variant, 3 × 106 LNCaP cells transfected with control pcDNA (LNCaP pc) or ARv567es (LNCaP ARv567es) were injected 1:1 with Matrigel s.c. in the flank of athymic nude-Foxn1nu mice (Harlan Sprague Laboratories). Ten male mice, aged 4–6 weeks, were used in each experimental group. Animals were weighed twice a week. Tumors were measured twice weekly, and tumor volume was estimated by the formula volume = (l × w2)/2, where l stands for length and w stands for width. Following a University of Washington IACUC-approved animal protocol, some animals were euthanized at specified time points, or when tumors reached a volume of 1,000 mm3, or when animal weight loss exceeded 20% of initial body weight. After euthanization, tumors were collected. A portion of the tumor was fixed in 10% neutral buffer formalin and embedded in paraffin. Five-micrometer sections were prepared for immunohistochemistry staining. The remaining portions of the tumor were separated into single cells mechanically by mincing and filtering through 70-μm nylon sieves.
To determine whether the ratio of ARv567es to ARfl may predict the tumor response to castration, we selected LuCaP xenografts with “low,” “medium,” and “high” ratios of ARv567es to AR; these were LuCaP lines 35, 136, and 86.2, respectively. Clinical assessment of the subjects from which LuCaP 35 and 136 primary tumors were obtained suggests that the patients were responsive to androgen ablation, whereas the 86.2 donor was castrate resistant. These lines were placed s.c. into SCID mice, 24 mice per xenograft. Tumor volume was measured twice weekly. Orbital blood was collected biweekly for PSA measurements. When tumor volumes reached 100–200 mm3, half of the mice were castrated. All mice were followed until tumor volumes began reaching 1,000 mm3. At this point, all the mice were euthanized in accordance with guidelines from the University of Washington IACUC. At the study end point, tumors were removed and portions were saved for histology, RNA, and protein analysis by Western blot.
Tissue androgens, including testosterone and DHT, were quantified in xenograft tissues by mass spectrometry, using methods we have published recently (9).
All data are displayed as mean ± SEM, except for PCR data in Figure Figure1,1, which are displayed as mean ± 1 SD. When 3 or more groups were compared, 1-way ANOVA was used, followed by Bonferroni-Dunn post-hoc test. (Statview 5.0). When 2 groups were compared, 2-tailed Student’s t test was used (Excel 2003). A P value of 0.05 or less was considered significant. For microarray data, gene expression differences between groups were analyzed with the SAM software by unpaired, 2-sample t tests, controlled for multiple testing by estimation of q values using the FDR method. Genes identified as significantly altered by t test were compared using Venn diagrams, and the functional relevance of those gene sets was determined by GO analysis. The EASE software ( http://david.abcc.ncifcrf.gov) was used to calculate overrepresentation statistics for genes in the Venn diagram lists, with respect to all genes represented in the data set. Essentially, for each term within the categories tested (GO biological processes, GO cellular components, GO molecular functions, KEGG pathways, and GenMAPP pathways) the Fisher exact probability of overrepresentation was calculated and used to generate the EASE score by weighting significance of those terms supported by many genes. The EASE score was further controlled for multiple testing, using the FDR with those with a P value of less than 0.05.
We would like to thank Jennifer Wu and Marco Marcelli for critical review of the manuscript and Martín Escandón for assistance with manuscript preparation. Grants from the NIH-NCI (PO1 CA 85859 to S.R. Plymate, R. Vessella, and P.S. Nelson), the Pacific Northwest Prostate Cancer SPORE (P50 CA 097186 to S.R. Plymate, R. Vessella, and P.S. Nelson), and the Veterans Affairs Research Program (to S.R. Plymate) supported this work.
Conflict of interest: The authors have declared that no conflict of interest exists.
Citation for this article: J Clin Invest. 2010;120(8):2715–2730. doi:10.1172/JCI41824.