ETV1 directs androgen metabolism and confers aggressive prostate cancer in targeted mice and patients

ERG and ETV1 regulate distinct programs in immortalized nontumorigenic prostate cells

The genetically engineered knock-in mice distinguished ETV1 from ERG in supporting invasive adenocarcinoma. To gain mechanistic insights into this difference, we performed an integrated genomic analysis to identify their respective target genes. First, we ectopically expressed ETV1 or ERG in immortalized human nontumorigenic prostate epithelial cells, RWPE-1 cells. Thus, RWPE-1 cells were engineered to express full-length ERG or ETV1 cDNA carrying a substrate tag that permits in vivo biotinylation by coexpressed Escherichia coli biotin ligase BirA (bioERG and bioETV1) (Supplemental Fig. S8A,B). RWPE-1 cells overexpressing bioERG (R.ERG) or bioETV1 (R.ETV1) proliferated normally compared with controls (Supplemental Fig. S8C). Microarray expression profiling data strongly overlapped with those previously reported (ERG [Gupta et al. 2010] and ETV1 [Tomlins et al. 2007] in RWPE-1 cells [Supplemental Fig. S8D,E]) and yielded two findings. First, a set of genes was uniquely regulated by either ERG or ETV1. Second, both factors also regulated a common set of genes but in a largely opposing fashion (Fig. 3A). This latter observation is depicted by the quantitative two-dimensional comparison of the ERG and ETV1 data sets, which illustrates a statistically significant correlation between ERG-driven up-regulated genes and ETV1-driven down-regulated genes and vice versa (Fig. 3B). Gene ontology (GO) and Ingenuity pathway analysis (IPA) implicated genes up-regulated upon ERG expression and down-regulated upon ETV1 expression that are associated with cell proliferation. In contrast, genes down-regulated on ERG expression and up-regulated by ETV1 are correlated with cell motility and lipid metabolism (Supplemental Fig. S9). As revealed by quantitative RT–PCR of select common genes, ETV1 expression induced up-regulation of genes involved in AR signaling (TMPRSS2 and SOX9) or invasion and lipid metabolism (VIMENTIN, ADRB2, and ACSL3) as well as down-regulation of cell cycle genes (E2F1 and BRCA1) (Fig. 3C). In contrast, these genes exhibited largely an opposite expression pattern in ERG-expressing RWPE-1 cells (Fig. 3C). Thus, these data point to distinct regulatory programs driven by ERG and ETV1 in nontumorigenic prostate cells.

Open in a separate window

Figure 3.

ERG and ETV1 regulate a common program in immortalized nontumorigenic RWPE-1 prostate cells but in an opposing fashion. (A) Expression profiling of ERG-overexpressing (R.ERG) and ETV1-overexpressing (R.ETV1) RWPE-1 cells compared with BirA-expressing controls (CTL). Heat map generated by hierarchical clustering and by applying Pearson correlation and the complete linkage rule. The heat map shows differentially expressed genes (fold change, >1.5; false discovery rate [FDR], <0.05). (Red) Highest expression; (blue) lowest expression. (B) Bidimensional plot comparing expression profiles of genes differentially expressed (fold change, >1.5) in R.ERG versus R.CTL and in R.ETV1 versus R.CTL RWPE-1 cells. The red line represents the distribution of genes. The dotted line corresponds to a gene density fold change of 1. (C) RT–PCR analysis of select genes associated with prostate cancer pathways upon ERG or ETV1 overexpression in RWPE-1 cells. n = 3 per group. Error bars, SEM; t-test: (**) P < 0.01. If no P-value is indicated, P > 0.05.

ERG and ETV1 have shared and distinct chromatin targets

To interrogate further similarities and differences of ERG and ETV1 transcriptional programs, we performed ChIP (chromatin immunoprecipitation)-on-chip analysis. We identified ERG ChIP targets in VCaP cells (harboring TMPRSS2-ERG) by an anti-ERG antibody. As a ChIP-quality antibody for ETV1 is not available, we used the bio-ChIP approach (Supplemental Fig. S10A; de Boer et al. 2003; Wang et al. 2006; Kim et al. 2008) on LNCaP cells (harboring ETV1 fusions) ectopically expressing bioETV1 and the E. coli biotin ligase BirA (Supplemental Fig. S10B). On comparison of ERG and ETV1 ChIP targets (Supplemental Table S1), we identified three subsets: ERG–ETV1 common targets, ERG-only targets, and ETV1-only targets (Fig. 4A), which are consistent with reported ChIP-seq (ChIP combined with deep sequencing) data in RWPE-1 cells (Hollenhorst et al. 2011). Bound regions for each factor were typically in close proximity to the transcription start sites (TSSs) (Supplemental Fig. S10C,D). We confirmed that ERG–ETV1 common target regions were occupied by both ERG and ETV1. As anticipated, ETS-binding motifs are the most statistically significant enriched motifs within the predicted targets (Fig. 4B). Further analysis of the defined target subsets revealed clear differences of motif enrichment at common versus unique targets (Fig. 4B; Supplemental Fig. S10E), consistent with expression profiling data indicating distinct biological processes regulated by ERG and ETV1 in prostate cells (Supplemental Figs. S9A,B, S10F).

Open in a separate window

Figure 4.

ERG and ETV1 drive specific transcriptional programs. (A) Venn diagram of targets occupied by ERG and ETV1. The intensity of binding at each probe was calculated by model-based analysis of tiling array (MAT) (Johnson et al. 2006). MAT scores were then normalized by quantile–quantile normalization (Bolstad et al. 2003) between ETV1 and ERG ChIP-on-chip experiments. Target loci were defined as the peaks associated with P-value < 10⁻⁴. (B) Enrichment of ETS-binding motifs and other indicated motifs in all ChIP target subsets. The Fisher exact test was applied. (C) IPA analysis of ChIP-defined target gene sets implicating common target genes in nuclear receptor signaling pathways and ETV1 unique targets in lipid metabolism network. The significance of enrichment of each gene set is shown as −Log (P-value).

Interestingly, IPA analysis indicated that nuclear receptor signaling pathways, including those associated with estrogen, androgen, and glucocorticoid receptor signaling, were significantly enriched in ERG–ETV1 common targets (Fig. 4C). In contrast, the ERG-only subset correlated with the cell cycle network. Intriguingly, the lipid metabolism biological network as well as the Oncostatin M and IL-3 signaling pathways, which have been correlated with increased cell motility and invasiveness (Dentelli et al. 1999; Holzer et al. 2004), were enriched in the ETV1-only subset (Fig. 4C). Taken together, our combined gene expression and ChIP-on-chip analyses argue that ERG and ETV1 control distinct transcriptional programs in prostate cells.

ERG and ETV1 interact differentially with the AR signaling pathway

The AR pathway is a critical driver of tumorigenic prostate development in both androgen-dependent (AD) and castration-resistant stages (Wang et al. 2009). Our data suggest that genes associated with AR signaling belong to the ERG–ETV1 common target category (Fig. 4C; Supplemental Fig. S11A). To address potential cross-talk of ERG and ETV1 common targets with the AD program, we defined an improved androgen-driven signature from AD VCaP and LNCaP cell lines that mitigates differences among diverse expression-based gene sets (Supplemental Fig. S12). Gene set enrichment analysis (GSEA) (Subramanian et al. 2005) revealed that this signature was significantly depleted after ETV1 knockdown in LNCaP cells (Fig. 5A) but enhanced in VCaP cells upon ERG silencing (Fig. 5B). This finding was confirmed by quantitative RT–PCR for select AR-driven genes in LNCaP and VCaP cells (Figure 5C,D). ChIP analysis further demonstrated that, while androgen stimulated AR and ETV1 binding to the enhancers and promoters of two established AR targets, PSA and TMPRSS2, androgen decreased ERG occupancy to these AR target genes (Supplemental Fig. S11B,C). These data indicate that ETV1 cooperates with activation of AR signaling, while ERG negatively modulates the AR transcriptional program.

Open in a separate window

Figure 5.

ERG and ETV1 regulate AR signaling in an opposite manner. (A) Androgen-induced genes are depleted in ETV1-silenced LNCaP cells upon 16-h androgen stimulation ([left] no androgen stimulation; [right] with androgen stimulation). The androgen-induced signature was obtained from the common AR ChIP targets in LNCaP and VCaP cells that were up-regulated in them upon androgen stimulation. (B) Androgen-induced genes are significantly enriched in ERG-silenced VCaP cells upon 16-h androgen stimulation compared with controls ([left] no androgen stimulation; [right] with androgen stimulation). (C,D) ETV1 silencing specifically decreases expression of AR-associated genes (C), whereas ERG silencing increases their expression (D). Mean, n = 3; error bars, SEM; t-test: (*) P < 0.05; (**) P < 0.01. If no P-value is indicated, P > 0.05. (E) Flow cytometry analysis demonstrating robust GFP⁺ population in the T-ETV1 prostates but not in the T-ERG prostates. However, in the presence of the Pb-AR transgene, GFP expression can be readily detected in Pb-AR;T-ERG prostates; in addition, GFP expression in Pb-AR;T-ETV1 prostates appear to be further elevated. (F) IHC staining showing weak ERG staining in the ventral lobe of a T-ERG knock-in male (blue arrow; compared with strong Erg staining in the endothelial cells [black arrow]) but much stronger ERG staining in the ventral lobe of a Pb-AR;T-ERG male (blue arrow; almost comparable with ERG staining in endothelial cells in the same section [black arrow]). Bars, 50 μm. (G) Real-time PCR quantification showing up-regulation of most AR target genes in Pb-AR;T-ETV1 prostates and slight down-regulation of them in Pb-AR;T-ERG prostates in relation to those of Pb-AR-alone prostates. Mean, n = 3; error bars, SEM; t-test: (*) P < 0.05, (**) P < 0.01. If no P-value is indicated, P > 0.05.

To determine whether ERG and ETV1 also regulate AR signaling differentially in vivo, we examined GFP expression in T-ERG and T-ETV1 males, as the reporter serves as a surrogate for the in vivo activity of the Tmprss2 promoter in these identically engineered mice. GFP expression was readily detected in T-ETV1 prostates but barely detected in T-ERG prostates (Fig. 5E). In contrast, GFP expression was detected in estrogen receptor-positive mammary epithelial cells (Sleeman et al. 2007) of both T-ERG and T-ETV1 females (Supplemental Fig. S13). These data indicate that Tmprss2 is indeed an estrogen and androgen dual-responsive promoter and that its promoter activity is down-regulated in T-ERG prostate cells.

To ascertain whether elevated AR signaling up-regulates the Tmprss2 promoter activity in vivo, we bred T-ERG and T-ETV1 mice to transgenic mice that express a mutated version of AR (E231G) expressed from the Pb promoter (Pb-AR) (Han et al. 2005). In Pb-AR;T-ERG double-transgenic males, we detected a strong GFP signal in the prostate (Fig. 5E). The Pb-AR transgene is most active in the ventral prostate lobes (Han et al. 2005). In accord with this, we observed stronger staining for ERG in the ventral lobes of Pb-AR;T-ERG prostates, compared with barely detectable ERG staining in T-ERG prostates (Fig. 5F). Furthermore, we measured expression levels of select AR targets in mouse prostates. In the Pb-AR background, most AR targets were down-regulated in T-ERG males, whereas AR targets were typically up-regulated in T-ETV1 males, thus illustrating the opposite regulation of AR signaling by ERG and ETV1 in vivo (Fig. 5G). Despite elevated AR signaling, prostates of Pb-AR-alone as well as those of Pb-AR;T-ERG and Pb-AR;T-ETV1 males appeared largely normal. Taken together, human prostate cancer cell and mouse model data indicate that differential regulation of the AR pathway by ETV1 and ERG occurs not only in vitro, but, importantly, also under the physiological setting.

ETV1 directs androgen metabolism in prostate epithelial cells

In addition to the opposing regulation of common targets by ETV1 and ERG, we hypothesized that unique targets controlled by ETV1 might contribute to the aggressive phenotype seen in association with ETV1 expression. To gain mechanistic insights into programs selectively regulated by ETV1 we sorted GFP⁺ (thus, ETV1-expressing) prostate luminal cells from T-ETV1 knock-in males and compared them with luminal cells from wild-type prostates by microarray expression profiling (Fig. 6A,B). We confirmed the luminal cell expression pattern in both sorted samples (Supplemental Fig. S14). By GSEA, we identified several cancer-associated metabolic pathways that were enriched in T-ETV1 luminal cells (Supplemental Fig. S15A). Of note, cholesterol and steroid biosynthesis pathways, both of which are strongly related to prostate tumorigenesis (Twiddy et al. 2010; Zadra et al. 2010), were most highly enriched (Fig. 6C,D). On analysis of a patient cohort with CRPCs metastatic to bone (Stanbrough et al. 2006), we observed that genes associated with the steroid hormone biosynthesis pathway and androgen and estrogen metabolism are significantly enriched in tumors with higher ETV1 expression (Supplemental Fig. S15A). HSD17B7, a gene shared by steroid biosynthesis and steroid hormone biosynthesis pathways, was up-regulated in both T-ETV1 luminal cells and ETV1-high bone metastases (Supplemental Fig. S15B). HSD17B7 as well as other HSD17B enzyme genes (HSD17B4 and HSD17B10) are ChIP targets of ETV1 in prostate cancer cells and are components of the lipid metabolism network enriched in ETV1-only ChIP targets (Fig. 4C; Supplemental Fig. S15C). By independent ChIP analysis, we validated ETV1, but not ERG, binding to the HSD17B7 and HSD17B4 promoters (Fig. 6E; Supplemental Fig. S15D). Expression of HSD17B7 in LNCaP cells was reduced upon ETV1 depletion (Fig 6F). In ETV1-overexpressing RWPE-1 cells, HSD17B7 expression trended upward (although not statistically significant) (Supplemental Fig. S16A). We also confirmed higher Hsd17b7 expression in lineage-depleted T-ETV1 prostate cells (Fig. 6G). Since HSD17B7 is critical in converting less active forms of estrogen and androgen to more active forms (Fig. 6H; Krazeisen et al. 1999), and Tmprss2 is an androgen and estrogen dual-responsive gene, we reasoned that up-regulation of the steroid biosynthesis pathway by ETV1 may provide prostate cells with an intrinsic source of steroids. If this supposition were correct, such cells might be intrinsically castration-resistant. To test this prediction, T-ETV1 knock-in males as well as T-ERG and wild-type control males were castrated. We observed that almost half of prostate cells from the castrated T-ETV1 mice were GFP⁺. In addition, we also detected a small population of GFP^low prostate cells from the castrated T-ERG males (Supplemental Fig. S16B), consistent with a recent study and possibly reflecting the existence of a subpopulation of Tmprss2⁺ prostate cells that are intrinsically castration-resistant (Casey et al. 2012). As expected, the castration-resistant GFP⁺ cells in T-ETV1 males exhibited higher Hsd17b7 levels than controls (Supplemental Fig. S16C).

Open in a separate window

Figure 6.

ETV1 regulates steroid metabolism in prostate cells. (A, right plot) Flow cytometry profiles and gating strategies showing GFP⁺ prostate luminal cells (Lin⁻Sca-1⁻CD49f^med) sorted from T-ETV1 knock-in males used for microarray analysis. (Left plot) The Lin⁻Sca-1⁻CD49f^med prostate luminal cells sorted from wild-type (WT) control males were used as the control. (B) Real-time PCR quantification confirming ectopic ETV1 expression in sorted GFP⁺ prostate luminal cells from T-ETV1 knock-in mice (mean, n = 3 samples per group; error bars, SEM). (C,D) Steroid and cholesterol biosynthesis pathways are the top pathways significantly enriched in T-ETV1 knock-in prostate luminal cells compared with controls. Note that a critical enzyme in the steroid biosynthesis pathway, HSD17B7, is also a key enzyme in the steroid hormone biosynthesis pathway, which is enriched in metastatic prostate cancers (Supplemental Fig. S15). (E) ChIP-PCR validation of ETV1 binding to HSD17B7 (pB7), HSD17B4 (pB4), and HSD17B10 (pB10) promoters (mean, n = 5; error bars, SEM). (pCTL1 and pCTL2) Nonspecific promoter control regions. (F) Only HSD17B7 levels significantly decreased upon knockdown of ETV1 (k/d) (mean, n = 3, error bars, SEM) under both the androgen-deprivation condition ([CH-T] charcoal-treated) and the regular condition in the presence of serum (FBS). Conversely, HSD17B7 expression increased upon ETV1 overexpression in RPWE-1 cells (Supplemental Fig. S16A). t-test: (**) P < 0.01. If no P-value is indicated, P > 0.05. (G) T-ETV1 knock-in prostate cells exhibit increased Hsd17b7 expression levels compared with wild-type controls (mean, n = 3; error bars, SEM). t-test: (*) P-value < 0.05. (H) Schematic diagram showing the key role of 17-β HSD enzymes, including HSD17B7, in converting androgen and estrogen from their less active forms to active forms. (I) ETV1 overexpression in RWPE-1 cells promoted the elevation of the endogenous testosterone level, while no changes were observed upon ERG overexpression (mean, n = 4; error bars, SEM). Testosterone levels per 10⁶ cells (R.ETV1 mean = 642.16 pg/μL; R.ERG mean = 0.49 pg/μL; R.CTL mean = 1.89 pg/μL). t-test: (***) P < 0.001. (J) Testosterone levels were reduced in androgen-deprived (charcoal-treated) LNCaP cells upon stable ETV1 silencing (k/d) as compared with controls (mean, n = 3; error bars, SEM). (NSC) Nonsilencing shRNA control. Testosterone levels per 10⁶ cells (NSC mean = 74.69 pg/μL; R.ERG mean = 0.49 pg/μL; ETV1k/d mean = 0.56 pg/μLr). t-test: (***) P < 0.001.

To determine whether increased ETV1 expression enhances steroid production, we directly measured by liquid chromatography/mass spectrometry (LC/MS) the levels of endogenous testosterone in control, ETV1-expressing, and ERG-expressing human RWPE-1 cells. Remarkably, ETV1-expressing RWPE-1 cells showed much higher levels (>300-fold higher) of endogenous testosterone compared with ERG-expressing and control RWPE-1 cells (Fig. 6I; Supplemental Fig. S17A,B). Conversely, ETV1 knockdown in LNCaP cells reduced testosterone production under conditions of androgen deprivation (Fig. 6J). To further characterize this observation, we quantified the intraprostatic levels of testosterone in wild-type, T-ETV1, and T-ERG mice under noncastrated and castrated conditions. As expected, noncastrated mice showed very low concentrations of testosterone in their prostate cells. Interestingly, castrated T-ETV1 mice indeed exhibited higher testosterone levels than those of castrated wild-type and T-ERG prostate samples (Supplemental Fig. S17C). These results indicate that ETV1 expression directly regulates androgen production in prostate cells.

AR signaling is a common pathway regulated by ERG and ETV1 but in an opposite manner

AR signaling is central to prostate development and tumorigenesis. Indeed, AR has recently been implicated in double-strand breaks that favor the formation of translocations involving androgen-driven promoters and ETS family members (Lin et al. 2009; Haffner et al. 2010). We observed a complex relationship between ERG- or ETV1-regulated networks and AR signaling. In agreement with prior findings, we observed negative regulation of AR signaling by ERG (Yu et al. 2010). In contrast, ETV1 cooperates with AR signaling by favoring activation of the AR transcriptional program (Fig. 5). Upon androgen stimulation, ETV1 recruitment to the established PSA and TMPRSS2 regulatory elements correlates with AR binding, suggesting coordinate regulation of androgen-driven genes by ETV1 and AR.

We validated divergent regulation of AR signaling by ERG and ETV1 in vivo in knock-in mice. As Tmprss2 is an AR target, the transcriptional output from the Tmprss2 promoter serves as a reporter for AR activity. In T-ERG knock-in mice, ERG expression appears to down-regulate AR target genes, including Tmprss2, which would lead to down-regulation of its own expression, thus forming a negative regulatory loop, consistent with prior findings (Yu et al. 2010). In contrast, in T-ETV1 knock-in males, ETV1 expression positively cooperated with AR signaling, leading to further enhancement of expression of AR targets, including Tmprss2, which would then support robust expression of ETV1 and the GFP reporter, reflecting a positive regulatory loop. These mechanistic differences account for the striking difference in GFP intensity in the two knock-in strains, which were identically engineered (Fig. 5E). Of note, the negative loop in the T-ERG knock-in is overridden by elevated AR signaling in vivo. Indeed, the only consistent association between the TMPRSS2-ERG fusion and clinical parameters identified thus far is the association of this fusion (in particular when ERG expression was measured by IHC) with a high level of AR signaling (Minner et al. 2011; Hoogland et al. 2012). Last, in human prostate tumors with ETV1 fusions, several 5′ fusion partners other than TMPRSS2 are also involved. Since the majority of these 5′ fusion partners are also androgen-responsive genes (Tomlins et al. 2007; Attard et al. 2008b), we reason that a similar positive regulatory loop may be operative in such tumors to enhance AR signaling and ETV1 expression. In addition, AR has been reported to regulate the endogenous ETV1 locus as well (Cai et al. 2007), suggesting a positive feedback between both the endogenous and rearranged ETV1, thus predisposing affected prostate cells for prostate cancer development.

ETV1, but not ERG, reprograms metabolic pathways in normal prostate epithelial cells

Among the unique genes up-regulated by ETV1, those encoding for enzymes involved in cholesterol and steroid biosynthesis and in cancer-associated metabolic pathways (e.g., glycolysis) (Supplemental Fig. S15A) are of particular interest. In particular, we were able to demonstrate increased production of endogenous testosterone upon ETV1 expression in both human prostate cells and T-ETV1 knock-in mice (upon castration) (Fig. 6I,J; Supplemental Fig. S17C). Due to the increased testosterone level, we anticipated observing higher levels of androstenedione or dihydrotestosterone as well; however, none were detected, possibly due to technical limitations. However, decreased levels of estrone were detected in ETV1-expressing RWPE-1 cells, probably due to the promotion of testosterone synthesis (Supplemental Fig. S17B).

Although ETV1-expressing prostate cells appear ostensibly normal (Fig. 1E), their metabolic programs resemble those of cancer cells. Increased aerobic glycolysis has been observed only in advanced disease, whereas increased sterol and protein synthesis are common features of both primary and advanced prostate cancer (Swinnen et al. 2000; Rossi et al. 2003; Ettinger 2004; Bauer et al. 2005). In particular, activation of lipid metabolism has been described in most localized and metastatic prostate tumors, underscoring its potential role in prostate cancer progression (Twiddy et al. 2010; Zadra et al. 2010). Arachidonic acid metabolism and Toll-like receptor signaling inflammatory pathways (De Marzo et al. 2007) are enriched in both CRPCs with higher ETV1 expression and T-ETV1 mouse prostate cells, correlating with the extended areas of inflammation observed in T-ETV1 prostates (Fig. 1E). Thus, we speculate that this unique ETV1-controlled program, in concert with positive cooperation with AR signaling, may reprogram prostate cells for malignant progression in association with additional oncogenic events (Fig. 7F).

TMPRSS2-ETV1 and TMPRSS2-ERG fusions make distinct contributions to prostate tumorigenesis

Although both Tmprss2-ERG and Tmprss2-ETV1 cooperate with loss of a single copy of Pten in leading to localized PIN lesions, similar to those demonstrated previously in mice overexpressing ERG or ETV1 under the Probasin or viral promoters (Carver et al. 2009; King et al. 2009; Zong et al. 2009) or in mice expressing ERG from an extended human TMPRSS2 promoter (Casey et al. 2012), our mechanistic and animal model studies suggest they do so differently (Fig. 7F). Ectopic ERG expression likely represses the differentiation program of prostate cells (e.g., represses AR and AR targets and up-regulates Ezh2 and its targets, as suggested previously) (Yu et al. 2010). Interestingly, it has been reported that Pten loss leads to a castration-like phenotype by suppressing androgen-responsive gene expression through modulation of AR transcriptional activity. Moreover, conditional deletion of AR further promotes proliferation of prostate cells with Pten loss (Mulholland et al. 2011). We speculate that TMPRSS2-ERG may act in a similar fashion by down-regulating AR and, consequently, promoting cell proliferation. Indeed, the ERG expression pattern in our murine models (Figs. 1C, 2B,D) suggests that ERG may be required primarily at early stages of the disease but may be not as functionally relevant at late stages. This may explain our observation that ERG cooperates with Pten haploinsufficiency (i.e., under a more sensitized Pten^+/− background), whereas its contribution under the total Pten loss background appears far less. The high levels of ERG expression often observed in localized fusion-positive human prostate cancers may be a secondary consequence of high activity of AR signaling in such tumors in general (i.e., similar to what we observed in Pb-AR;T-ERG prostates) (Fig. 5E–G) rather than a critical requirement of ERG overexpression at this stage. In contrast, ectopic ETV1 expression appears to enhance androgen signaling and reprogram the metabolism of prostate cells, processes critical for both early and advanced stages of the disease. Activation of the PI3K/AKT pathway drives anabolic metabolism and tumorigenesis (Ward and Thompson 2012). We propose that TMPRSS2-ETV1 cooperates with Pten loss by further enhancing metabolic reprogramming, in particular, by favoring steroid biosynthesis, a pathway critical for invasive adenocarcinoma cells. The cooperation between ETV1 and Pten loss is also consistent with the recent finding that combined loss of Pten and COP1, a ubiquitin ligase that negatively regulates ETV1 levels, leads to more invasive prostate adenocarcinomas (Vitari et al. 2011).

In aged PbCre;T-3Mb-Erg;Pten^L/L males, we also observed invasive prostate cancer (Fig. 2E). Interestingly, in such invasive cancers, we observed a mosaic pattern of Erg expression (Fig. 2E). Several possibilities can explain this observation. Erg-expressing prostate cells may send signals to Erg-negative cells so that high levels of Erg expression are not needed in all cells of the invasive cancer. Alternatively, Erg may not be critical for the development of advanced cancer, and another genetic or epigenetic change may contribute to advanced disease. One potential genetic change is haploinsufficiency of one or more deleted genes in the interstitial region. Interestingly, ETS2, a gene residing within the interstitial region, was recently proposed to be a tumor suppressor contributing to aggressive prostate cancer cases carrying TMPRSS2-ERG fusions with deletion (Grasso et al. 2012). Whether haploinsufficiency of the deleted genes (e.g., ETS2) contributes to the advanced cancer phenotype awaits further investigation.

Pathology, immunostaining, and flow cytometry

Standard protocols were followed.

Cell lines

Cell lines were obtained from American Type Culture Collection and cultured accordingly. ERG or ETV1 overexpression and silencing experiments were performed by standard protocols.

RT–PCR

Real-time RT–PCR was performed according to standards protocols. All primer sequences are listed in Supplemental Table S2.

Gene expression microarray analysis

RWPE-1 stable cell clones (R.ERG, R.ETV1, and R.CTL) were grown under normal conditions. VCaP and LNCaP cells, 24 h after ERG or ETV1 RNAi, respectively, were grown in hormone-depleted conditions for 2 d, and then in the presence or absence of 10 nM DHT for 16 h. Mouse primary prostate cells were FACS-sorted and processed according to standard procedures. Affymetrix HG133 plus 2.0 or Mouse Genome 430 2.0 expression arrays were used for human or mouse samples, respectively. Gene Pattern software (Reich et al. 2006) was used for data normalization, extraction of expression values, and generation of GTC files for GSEA (Subramanian et al. 2005). A bidimensional comparison plot was used to compare differentially expressed genes (P < 0.05 by two-tailed t-test) in RWPE-1 cells upon overexpression of either ETV1 or ERG.

ChIP and ChIP-on-chip

BioChIP–chip for ETV1 was performed as described (Kim et al. 2008), and conventional ChIP–chip reaction for ERG was as described (Kim et al. 2004). Affymetrix Human Promoter 1.0R array hybridization was performed at the Dana-Farber Cancer Institute Microarray Core Facility. Peak identification was calculated by MAT score (Johnson et al. 2006). For ChIP-PCR experiments, conventional ChIP reactions were performed. Antibodies used were as follows: anti-AR (N20X), anti-ERG (C17X) and anti-rabbit IgG from Santa Cruz Biotechnology, and anti-ETV1 kindly provided by Dr. Litovchick. The online DAVID functional annotation tool (Huang et al. 2009) and the IPA tool (Ingenuity Systems, Inc.) were used to determine the enrichment for all “FAT” GO terms and canonical pathways/networks in each gene set.

Patient tumor data analysis

Gene sets associated with indolent and aggressive prostate cancer were extracted from the Swedish, MSKCC, Sharma, and Glinsky cohorts (Glinsky 2004; Setlur et al. 2008; Taylor et al. 2010; Sharma et al. 2013) and analyzed for their mutual overlap between tumor cohort-derived signatures (differentially expressed genes: fold change, >1.5; false discovery rate [FDR], <0.05) and ERG- and ETV1-associated gene sets obtained in our studies. The overlap between gene sets was represented by a connectivity network, where the width of the connector edge was −log10(P-value). The P-value was derived from a hypergeometric distribution by using Fisher exact test to analyze the significance of the mutual overlap. Cytoscape software version 2.8 (Cline et al. 2007) was used for the visualization of gene sets overlapping the network.

Statistics

All statistics were based on a Student's t-test, unless otherwise indicated. Dot plots and histograms show data means, and error bars are standard error of the mean (SEM). All statistics were performed using the data analysis package within Microsoft Excel or the analysis tool within GraphPad Prism 5.0. Kaplan-Meier survival analysis was performed using GraphPad Prism 5.0.

Steroid metabolism measurement

Steroids from RWPE-1 and androgen-deprived LNCaP cells were extracted following Lemmen et al. (2002) and quantified by LC/MS at the Harvard FAS Center for Systems Biology.

Testosterone measurement

The intraprostatic testosterone levels were measured by a mouse testosterone ELISA kit (Calbiotech) based on the manufacturer's instructions. Briefly, mouse prostates were microdissected in cold PBS and lysed in RIPA buffer. Testosterone levels were calculated as the total amount per gram of total protein.

We thank Drs. Larisa Litovchick, Carmen Priolo, and Fatima Al-Shahrour for reagents, mouse surgical techniques, analysis support, and helpful discussions, and Dr. Maaike van Bragt for FACS analysis of knock-in mammary glands. This work was supported by a Career Development Award from Dana-Farber/Harvard Cancer Center Prostate Cancer SPORE (to Z.L.), a post-doctoral award from the Spanish Ministry of Science (EX2006-0958 to E.B.), and grants from the Department of Defense (PC060492 to S.H.O., PC100704 to Z.L., and PC080097 to E.B.) and NIH (U01CA105423 to S.H.O.). S.H.O. is an Investigator of Howard Hughes Medical Institute.