ETS factors reprogram the androgen receptor cistrome and prime prostate tumorigenesis in response to PTEN loss.

Studies of ETS-mediated prostate oncogenesis have been hampered by a lack of suitable experimental systems. Here we describe a new conditional mouse model that shows robust, homogenous ERG expression throughout the prostate. When combined with homozygous Pten loss, the mice developed accelerated, highly penetrant invasive prostate cancer. In mouse prostate tissue, ERG markedly increased androgen receptor (AR) binding. Robust ERG-mediated transcriptional changes, observed only in the setting of Pten loss, included the restoration of AR transcriptional output and upregulation of genes involved in cell death, migration, inflammation and angiogenesis. Similarly, ETS variant 1 (ETV1) positively regulated the AR cistrome and transcriptional output in ETV1-translocated, PTEN-deficient human prostate cancer cells. In two large clinical cohorts, expression of ERG and ETV1 correlated with higher AR transcriptional output in PTEN-deficient prostate cancer specimens. We propose that ETS factors cause prostate-specific transformation by altering the AR cistrome, priming the prostate epithelium to respond to aberrant upstream signals such as PTEN loss.

Introduction

Translocations of ETS transcription factors ERG, ETV1, ETV4, ETV5 and FLI1 occur in half of all prostate cancer and the TMPRSS2-ERG translocation is the most common molecular alteration[1-4]. Evidence from human tumor analysis strongly implicates aberrant ETS expression as an early if not initiating event[5-7].

Transgenic mouse models have shown that neither ERG nor ETV1 is sufficient to initiate prostate cancer[8-12]. Unfortunately, the existing probasin based transgenic ETS models are poorly suited for further mechanistic exploration, especially when combined with other genetic events that turn off probasin expression[9,13]. To overcome these shortcomings, we constructed a knock-in model of prostate-specific ERG expression that gives robust, uniform ERG expression throughout the mouse prostate. This model led us to the discovery that ERG reprograms the AR cistrome. These effects, in the context of Pten loss which suppresses AR, restore AR transcriptional activity and activate transcriptional targets involved in cell death, inflammation, migration and angiogenesis that result in rapid onset, widely invasive prostate cancer. Similarly, ETV1 also alters the AR-cistrome and AR-transcriptional activity in ETV1-translocated, PTEN-negative prostate cancer cells. The findings reveal a previously unappreciated role for chromatin context in ETS-mediated transformation and offer a potential explanation for the tissue-restricted nature of ETS translocations.

Results

A robust mouse model of ERG-driven prostate cancer

We generated a conditional mouse of the TMPRSS2-ERG transgene knocked into the Rosa26 locus (R26) (Supplementary Fig. 1). We crossed R26 with prostate specific Pb-Cre4 mice to express ERG specifically in the prostate [14]. IHC against ERG or the IRES-linked EGFP showed that ERG was uniformly expressed in the ventral and dorsolateral lobes by 8 weeks and the anterior lobes by 3 months and that ERG did not affect AR expression (Fig. 1a, 2c, Supplementary Fig. 2). We did not appreciate any differences in prostate histology or cellular proliferation (Ki67 staining) in either heterozygous Pb-Cre4;R26/+ or homozygous Pb-Cre4;R26 mice up to 1 year of age. Approximately 50% of ERG mice older than 1 year exhibited focal ventral lobe hyperplasia (Fig. 1b, c). We conclude that ERG alone, even in the context of robust and high level protein expression is insufficient to cause prostate cancer[15-17].

Figure 1

ERG expression induces minimal histological phenotype in mouse prostates

(a) Representative H&E histology, ERG IHC, and AR IHC of the anterior, ventral and dorsolateral (AP, VP and DLP) lobes in a 3-month old Pb-Cre4;R26(R26) and a littermate control Cre-negative (WT) mouse prostates. Scale bars: 50 μm. (b) A representative low-power H&E histology image of VP hyperplasia in the prostate of a 13-month old R26 mouse is shown on the left. High power magnification of boxed region, including H&E and EGFP IHC, is shown on the right is shown on right. Scale bars: 50 μm. (c) Summary of histological findings of WT (Cre+), ERG heterozygous (Cre+;R26+) and ERG homozygous (Cre+; R26) mouse prostates examined at 8 and 12 weeks, 6, 12 and 18 months respectively.

Figure 2

ERG robustly cooperates with Pten loss in prostate tumorigenesis

(a) Comparison of H&E prostate histology, ERG IHC, phosphorylated AKT (pAKT) IHC, AR IHC and Ki67 IHC of representative 6-month old Cre+;Pten (Pten) and Cre+;Pten (Pten) prostate. Scale bars: 50 μm. (b) A Representative example of gross appearance of anterior (A), ventral (V), dorsolateral (DL) lobes of Pten and Pten mice euthanized at 6 months. (c) Quantification of Ki67 (3 mice, 3 20× fields per mouse, mean ± SD) of 6-month old WT, R26, Pten and Pten mouse prostates. For Pten mice, we separately quantified the PIN which is histologically similar to that of Pten mice, and adenocarcinoma. (d) Summary of histological findings of Pten and Pten mouse prostates examined at 8 and 12 weeks, 6, 9 and >10 months respectively. Mice were characterized by the most advanced finding found on histology. (e) Kaplan-Meier survival analysis of Pten and Pten mice.

Previously reported transgenic models of ERG expression in a Pten germline heterozygous background show prostatic intraepithelial neoplasia (PIN) that is patchy and variably penetrant[8,10,18]. We crossed R26 to Pten mice to generate double homozygous mice (R26 mice developed highly penetrant and homogenous PIN that does not progress to grossly invasive disease. In Pten mice, invasive adenocarcinoma characterized by small irregular glandular structures comprised of malignant cells with large, pleiomorphic nuclei and pale cytoplasm developed adjacent to PIN by 8 weeks (Supplementary Fig. 3).

By six months, approximately 80% of Pten mice contained regions of adenocarcinoma with enlarged, hardened prostates (Fig. 2a, b). The tumor cells uniformly express nuclear ERG and AR, and display Akt activation (pAkt). While the invasive regions are highly proliferative, the proliferative index of PIN lesions in Pten prostates is only slightly higher than those in Pten prostates (Fig. 2c), suggesting that ERG expression within PIN does not significantly affect proliferation. Instead, ERG expression may facilitate invasion and progression, as suggested in earlier in vitro studies[16,18]. Human PIN retains a basal layer of p63 and cytokeratin 5 (CK5)-positive cells beneath a luminal layer of cytokeratin 8 (CK8)-positive cells whereas adenocarcinoma is characterized by irregular glandular structures that have lost the basal layer. In Pten PIN, p63 and CK5 are maintained in the basal cells and CK5 is ectopically expressed in some luminal cells, consistent with prior reports (Supplementary Fig. 4)[19]. Pten adenocarcinoma is invariably positive for CK8 and negative for p63 consistent with human adenocarcinoma. CK5 expression is variable and, when present, coincides with CK8 expression. CK5/8 double positive cells, coined as the “intermediate cells” are detectable in a subset of human prostate cancers[20].

By 12 months, some mice develop foci of poorly differentiated carcinoma that maintain expression of AR, ERG, and CK8 and display patchy neuroendocrine differentiation demonstrated by Nestin staining (Fig. 2d, Supplementary Fig. 5). These patterns are reminiscent of human Gleason 5 cancer with focal neuroendocrine differentiation, in contrast to “small cell” cancer characterized by loss of AR and uniform neuroendocrine staining. Pten mice have shortened survival relative to Pten mice (Fig. 2e), with early deaths due to increased abdominal girth and penile prolapse.

ERG reprograms the AR cistrome

The robust and uniform expression of ERG in R26 and Pten mice provides an ideal model to explore the ERG cistrome and transcriptome under controlled conditions. ChIP-seq analysis identified 24,665 ERG peaks in prostate tissue from R26 mice (Supplementary Table 1). While most ERG peaks reside in the enhancer regions, they were enriched at promoters (30% versus 3% in genome-wide background, p<2.2×10-16) (Supplementary Fig. 6), consistent with prior ERG ChIP-seq[21,22]. ERG peaks were present in homologs of many well-characterized human ERG and ETV1 target genes such as Dusp6, Tmprss2, and Fkbp5 that were defined in ERG-positive VCAP and ETV1-positive LNCaP cells (Fig. 3a, Supplementary Fig. 7, 8), giving us further confidence in the physiologic relevance of the mouse model. We observed similar distribution of ERG binding sites in the Pten loss background (Supplementary Fig. 9, 10).

Figure 3

ERG localizes to pre-defined H3K4me1 marked regions and reprograms genome-wide localization of AR

(a) Representative ChIP-seq profiles of ERG, AR, and H3K4me1 at the Dusp6 gene locus in WT and R26 mouse prostates and at the human DUSP6 gene locus in the ERG-positive VCAP and the ETV1-positive LNCaP human prostate cancer cell lines. (b) Overlap of AR peaks in WT, R26 and Pten mouse prostates. Number of peaks is in parenthesis. (c) Bar graph of percentage of conserved AR peaks in ERG-negative mice and new AR peaks only in ERG-positive mice that have nearby FOXA1 (red, left y-axis) and GATA2 (blue, right y-axis) motifs in Pten and Pten prostates. (d) Venn diagram of ERG ChIP-seq peaks in R26 mice, AR-ChIP-seq peaks in WT mice, and AR-ChIP-seq peaks in R26 mice. Overlap of ERG and AR peaks in both WT and R26 mice are significant (P < 2.2×10-16). (e) Overlap of mapped genes of ERG and AR peaks in WT and R26 mouse prostates. (f) LEFT: Graph of the percent of conserved AR peaks in WT mice and new AR peaks in R26 mice that overlap with ERG peaks in R26 mice. RIGHT: Graph of percent of mapped genes of conserved AR peaks, of new AR peaks, and of all Refseq genes that overlap with mapped genes of ERG peaks. (g) Profiles of H3K4me1 ChIP-seq in WT and R26 mouse around ERG binding sites of R26 mice. (h) Profiles of H3K4me1 ChIP-seq associated with the conserved AR binding sites (left panel) and new AR binding sites (right panel) induced by ERG expression in R26 compared to WT mouse prostates.

We next examined the genome-wide localization of AR by ChIP-seq. Strikingly, there was a >4-fold increase in the number of AR peaks in R26 (14,889) compared to WT prostates (3,476) (Fig. 3b, Supplementary Table 1). The number of AR peaks was also increased in Pten prostates but to a lesser extent than in Pten mice. We validated increased AR binding at several enhancers using ChIP-qPCR of independent samples (Supplementary Fig. 11). We confirmed that the change in the AR cistrome was not due to difference in AR protein or circulating testosterone (Fig. 1a, Supplementary Fig. 12, 13).

De novo MEME motif analysis[23] of ERG peaks identified the ETS core consensus GGAA motif with 5′-CC and 3′-GT bias identical to human prostate cancer cells. AR motif analysis showed that pre-existing sites in WT mice and new sites in R26 mice contained the identical AR motif (Supplementary Fig. 14). To discover potential cooperating transcription factors for AR binding, we performed DREME analysis[24]. One distinguishing feature between conserved and new AR peaks was greater enrichment for GATA motifs in the new AR peaks and FOXA1 motifs in conserved AR peaks (Fig. 3c, Supplementary Table 2). This is of potential interest because GATA2 is essential for prostate development and cooperates with AR to modulate gene expression[25,26].

The percentage of AR and ERG peaks that physically co-localize in R26 mouse prostates was ∼44% (Fig. 3d, left), which is highly significant and comparable to VCAP cells. Yet, new AR peaks have less overlap with ERG peaks than the conserved AR peaks (∼40% versus ∼60%, Fig. 3e, f), making it unlikely that ERG directly recruits AR to new sites. However, a large fraction of new AR sites (77%, p < 1×10-20) map to genes containing ERG sites, raising the possibility of an ERG-mediated field effect that promotes AR binding, perhaps by functioning as a pioneer factor.

Several classes of pioneer factors have been defined. One class, exemplified by the ETS factor PU.1, binds to closed chromatin regions and generates de novo enhancers, characterized by H3K4me1, which in turn guide recruitment of other transcription factors[27,28]. Another class, exemplified by FOXA1, binds to pre-established H3K4me1 enhancer regions and instructs binding of additional transcriptional factors (e.g. AR) to both adjacent and distant pre-existing enhancer regions[25,29,30]. To determine if ERG resembles one of these classes of pioneer factors, we examined the distribution of H3K4me1. The collective distribution of H3K4me1 around ERG binding sites was similar between wild-type and R26 prostates, suggesting that ERG binds to pre-existing enhancers (Fig. 3g). Further, the collective distribution of H3K4me1 around both conserved and new AR peaks were also similar between wild-type and R26 prostates, suggesting that ERG expression helps guide AR to pre-existing enhancers (Fig. 3h). Interestingly, the normalized H3K4me1 ChIP-signal was consistently slightly higher in R26 mice for all three types of peaks (ERG, new AR and conserved AR), raising the possibility that ERG may strengthen pre-existing enhancers. Collectively, these data support a model whereby ERG reprograms the AR cistrome without significant changes in the K3K4me1 landscape.

ERG expression induces robust transcriptome changes in Pten loss background

We examined ERG-induced transcriptome changes in prostates from 4-month old R26 mice and littermate controls and found only 20 genes changed by 1.5-fold with a FDR<0.3 cutoff, as appreciated by a volcano plot (including a 16-fold increase in ERG transgene linked EGFP) (Supplementary Fig. 15a, Supplementary Table 3).

However, in the Pten background, ERG expression induced robust transcriptome changes, with greater than 800 genes significantly changed using the same criteria (Supplementary Fig. 15b, Supplementary Table 4). Principle component analysis of the 4 groups of mice showed that the first component is determined by Pten status and the second component is determined by ERG status (Fig. 4a). Unsupervised hierarchical clustering showed that Pten prostates are clearly distinguished by ERG status. Pten intact prostates also clustered by ERG status though less robustly, consistent with the subtle transcriptome alterations (Fig. 4b). Closer examination of ERG-induced gene expression changes (ERG Δ) in the Pten versus the Pten background revealed that ERG generally induced the same directional transcriptome changes that are greatly amplified in magnitude by Pten loss (Fig. 4a, b).

Figure 4

ERG expression primes prostate to respond to Pten loss

(a) Principle component analysis of expression profile of 4-month old WT, R26, Pten, and Pten mouse prostates. Component 1 is determined by Pten status and component 2 by ERG status. (b) Hierarchical clustering of genes significantly changed either between WT and R26 or between Pten and Pten mouse prostates (FDR<0.3, fold-change > 1.5). Clustering groups the genes by effect of Pten loss [Pten loss up (P Up), Pten loss unchanged (P Unc), Pten loss down (P Dn)] and ERG expression [ERG up (E Up), ERG down (E Dn)]. The three vertical heatmaps on the right show the fold-change of ERG expression in WT mice (between WT and R26) and in Pten mice (between Pten and Pten) and effect of Pten loss in ERG-negative mice. (c) GSEA of the ERG expression profile in Pten loss mouse prostates (Pten vs. Pten) showing that a gene set defined by ERG-positive vs. ERG-negative human prostate cancers[33] (Taylor_PCA_ERG_UP) and a gene set defined by genes down-regulated after ERG knockdown in VCAP cells1616 (VCAP_siERG_DN) are positively enriched. (d) Scatter plot of ERG vs. TFF3 and ERG vs AMPD3 expression in human prostate cancer and normal prostate tissue (left) and scatter plot of EGFP (linked to ERG via IRES) vs Tff3 and EGFP vs Ampd3 expression in mouse prostate (right). (e) Normalized expression of genes that belong to cell death, inflammation, migration, and angiogenesis functional groups that are regulated by ERG and Pten loss.

In Pten prostates, both ERG up-regulated and ERG down-regulated genes were enriched with ERG and AR peaks (Supplementary Fig 16a, b). When we divided genes into those with only AR peaks, only ERG peaks, or both, only those with both were significantly enriched for regulation by ERG (Supplementary Fig. 16c–e). Among genes with AR peaks, those with “old” pre-existing AR peaks and those with “new” peaks found only in the setting of ERG expression are both enriched (Supplementary Fig. 16f, g). This data suggests that AR binding facilitates ERG-mediated transcriptional regulation. ERG-mediated upregulation of gene expression was also associated with increasing and widening of the H3K4me3 profile toward the gene body, a chromatin mark associated with active transcription[31], whereas ERG-mediated downregulation was associated narrowing and decreasing of the H3K4me3 peak (Supplementary Fig. 17a).

To determine whether ERG-induced transcriptome changes in the context of Pten loss resemble those observed in ERG-positive human prostate cancer, we used gene-set enrichment analysis (GSEA)[32] (Supplementary Table 5). Two ERG related human gene sets, one defined by genes upregulated in ERG-positive compared to ERG-negative human prostate cancer samples[33] and a second defined by genes down-regulated after ERG knockdown in VCAP cells[16], were highly enriched (Fig. 4c). Further evidence in support of the human relevance of the model comes from examination of specific genes such as adenosine monophosphate deaminase 3 (AMPD3), which is upregulated in both human and mouse ERG-positive prostate cancers. Conversely, trefoil factor 3 (TFF3)[34] is highly expressed in ERG-negative human prostate cancer and Pten mouse prostate and down-regulated in ERG-positive human and Pten mouse prostate tumors (Fig. 4d, Supplementary Fig 17b).

The transcriptome heatmap reveals six distinct blocks of genes whose expression patterns vary as a group across genotypes. 2 groups can be primarily defined by genes upregulated by Pten loss (Pup), one of which is downregulated by ERG (Pup/Edown) and the other further upregulated by ERG (Pup/Eup). Similarly, there are 2 groups primarily defined by genes downregulated by Pten loss (Pdown), one of which is further downregulated by ERG (Pdown/Edown) and the other upregulated by ERG (Pdown/Eup). Lastly, two additional groups of genes unchanged by Pten loss are primarily defined by up or downregulation by ERG (Eup or Edown). To understand the functional consequences of ERG expression, we performed pathway analysis of these distinct groups of genes using IPA and DAVID GO[35] (Supplementary Tables 6, 7). Among the most enriched processes in these groups are cell death (Pup/Edown), inflammation and migration (Pup/Eup group) and angiogenesis (Eup) (Fig. 4e). Identification of migration but not proliferation pathways is consistent with our histological evidence that ERG induces invasion without a change in Ki67 staining. The cell death finding agrees with prior work suggesting that Pten loss induces oncogenic stress[36], which may be alleviated by ERG expression.

ERG restores the AR transcriptome in mouse and human prostate cancers with PTEN loss

Having demonstrated that ERG induces dramatic changes in the AR cistrome, we asked if this resulted in changes in the AR transcriptome. GSEA revealed that two gene sets, one defined by genes down-regulated in mouse prostate with castration and another by genes induced by androgen in ERG-positive VCAP cells, were highly enriched by ERG expression in the Pten background (Fig. 5a). As expected from prior work, Pten loss resulted in decreased expression of AR-regulated genes, as defined by castration experiments[9,37]. ERG expression had no significant effect on AR target genes in Pten intact mice but significantly restored AR transcriptional output in the setting of Pten loss (Fig. 5b, c). Since AR may regulate a distinct transcriptome in the setting of Pten loss, we performed a castration experiment of Pten and Pten mice (Supplementary Fig. 18). Expression profiling of the four groups of prostates show that genes downregulated by castration in the setting of Pten loss are upregulated by ERG expression, including established AR target genes such as probasin (Pbsn), Nkx3-1, and β-microseminoprotein (Msmb), all of which also correlate with the H3K4me3 profile that marks active transcription (Supplementary Fig. 19).

Figure 5

ERG increases androgen receptor signaling in Pten loss prostate cancer

(a) GSEA of the ERG expression profile in Pten loss mouse prostates (Pten vs. Pten) showing that the mouse prostate specific AR-dependent gene set (defined by changes from mouse castration) and human AR-dependent gene set (defined by genes upregulated by DHT in ERG-positive VCAP cells) are both significantly and positively enriched. (b) Hierarchical clustering of mouse androgen upregulated genes (Castration DN) and androgen downregulated genes (Castration UP) in mouse prostates. The data shows that many androgen upregulated genes are downregulated by Pten loss and restored by ERG expression and many androgen downregulated genes are upregulated by Pten loss and decreased by ERG expression. (c) The sum of the normalized expression of mouse androgen-regulated genes, defined as genes downregulated by castration, by genotype (mean ± SD). (d) Sum of normalized expression of human androgen-regulated genes from University of Michigan (UM) rapid autopsy series and MSKCC prostatectomy series. Black dots, blue dots, and red dots represent ETS-negative, ERG-positive, and ETV1-positive samples respectively (mean ± SEM). For UM series, significant comparisons are: PTEN high vs. PTEN low (P < 0.0001) and PTEN low;ETS neg vs PTEN low;ETS pos (P = 0.001). For MSKCC series, significant comparisons are: PTEN high vs. PTEN low (P < 0.0001) and PTEN low;ETS neg vs PTEN low;ETS pos (P = 0.043).

To determine the interaction of Pten loss and ETS expression in human prostate cancers, we turned to two large scale human prostate cancer gene expression data sets, one from the University of Michigan rapid autopsy series and one from the MSKCC prostatectomy series[7,33]. In both datasets, tumors with PTEN loss had a significantly decreased AR signature, consistent with prior reports. Further, in the setting of PTEN loss, ETS-positive tumors showed partial restoration of the AR signature (Fig. 5d). Thus, these analyses corroborate with experimental data that ETS overexpression positively regulates the AR- transcriptome in PTEN loss prostate cancer.

ETV1 modifies the AR cistrome and AR transcriptional activity

We next asked if the effects of ERG on AR DNA binding and transcriptional output described here are observed with other ETS family proteins targeted by prostate cancer translocations. The LNCaP prostate cancer cell line harbors an ETV1 translocation and PTENloss[11]. We performed ETV1 ChIP-seq in LNCaP cells and compared the binding sites with published ERG ChIP-seq data in VCAP cells[21]. ETV1 and ERG binding sites were highly similar, as 91% of ETV1 sites in LNCaP were bound by ERG in VCAP (Fig. 6a, see examples on 3a, Supplementary Fig. 7, 8). Next, we examined the role of ETV1 in AR binding by performing AR ChIP-seq in LNCaP cells expressing an ETV1-specific shRNA (ETV1sh2) or a scrambled control[38]. ETV1 knockdown resulted in a striking ∼90% decrease in the number of AR binding peaks (Fig. 6b, c, Supplementary Fig. 20). We validated this result by independent ChIP-qPCR experiments of the PSA and TMPRSS2 enhancers with two different ETV1 shRNAs and in the presence or absence of R1881 treatment (Supplementary Fig. 21). We next performed transcriptional profiling of ETV1 knockdown using two shRNAs. ETV1 knockdown reduced AR transcriptional activity measured both by AR output score and by GSEA which showed that the AR signature is the most enriched gene set (Fig. 6d, e, Supplementary Fig. 22, Supplementary Table 8). Thus, both ETV1 and ERG positively regulate AR binding and AR transcription in the context of PTEN loss.

Figure 6

ETV1 alters the AR cistrome and the AR-dependent transcriptome LNCaP cells

(a) Overlap of ETV1 ChIP-seq peaks in LNCaP cells with published ERG ChIP-seq peaks in VCAP cells[21]. (b) Overlap of AR ChIP-seq peaks in LNCaP cells 72-hours after infection with scrambled shRNA and ETV1sh2 shRNA. (c) Representative ChIP-seq profile at the TMPRSS2 locus showing ERG and AR profiles in VCAP cells and ETV1 baseline profile in LNCaP cells and AR profiles LNCaP cells infected with scrambled and ETV1sh2 shRNA. (d) The sum of normalized expression of genes in an AR signature from expression profiling in LNCaP cells 72-hours after infection with scrambled and two ETV1 shRNAs (n = 3, mean ± SD). Significant comparisons are Scr vs ETV1sh1 (P = 0.0026) and Scr vs ETV1sh2 (P = 0.0033). (e) GSEA profile showing that the AR signature gene set is highly enriched among genes downregulated by ETV1 knockdown.

Discussion

Previous transgenic models, while critical in establishing ERG's oncogenic potential, are limited by the subtle phenotype and variable penetrance. Here we report a novel ERG knock-in model with uniform transgene expression in all prostate epithelial cells that gives highly predictable, early onset invasive prostate cancer when combined with Pten deficiency. These characteristics make this a broadly useful new model for the prostate cancer research community for mechanistic and therapeutic studies of ERG-driven prostate cancer.

An unanticipated finding is that ERG appears to function as a pioneer factor, causing dramatic changes in the AR cistrome. This property is not unique to ERG because knockdown of ETV1 results in a similarly dramatic loss in AR binding sites. The large increase in AR binding sites seen in ERG-expressing mice cannot be explained solely by co-recruitment of AR by ERG to adjacent binding sites. However, the fact that ∼80% of the new AR sites map to genes that also contain ERG sites support a pioneer model whereby ERG causes local chromatin changes that facilitate AR binding to nearby but not adjacent sites. ChIP-seq studies of the H3K4me1 enhancer mark in prostate tissue from wild type mice establish that ERG primarily binds to pre-defined enhancers, presumably established during development of the genitourinary tract.

At the transcriptional level, the primary consequence of the ERG-driven changes in the AR cistrome is an increase in AR output, which is particularly evident in Pten deficient mice and in PTEN-negative human tumors. This requirement for additional signaling input to enhance the transcriptional effects of ERG is reminiscent of earlier work on the related ETS family protein ETV1 showing that KIT kinase activity amplifies ETV1 transcriptional output in gastrointestinal stromal tumors[38].

Our conclusion that ERG enhances AR function contrasts with previous work showing that ERG suppresses AR function[21,22]. Potential explanations for these apparently contradictory findings include the fact that the earlier studies were conducted in VCAP cells, which are PTEN-intact and therefore may be less “sensitized” to the ERG-effect on AR transcriptome. Even in VCAP cells, there is evidence that ERG serves as a pioneer factor at a number of sites for AR binding including at the SOX9 gene[39]. Further, ERG expression may impair differentiation, which can manifest as reduced expression of classic AR target genes associated with terminal prostate differentiation. Further work that comprehensively assesses the AR transcriptome in various contexts should clarify this.

In summary, ERG and ETV1 mediate context-dependent oncogenic transformation by influencing the prostate lineage-specific master regulator, AR, and priming the prostate epithelium to cooperate with aberrant upstream signaling in prostate oncogenesis. Based on the increasing number of mutant transcription factors and chromatin modifiers emerging from cancer genome sequencing efforts, we speculate that the mechanism of oncogenesis described here for prostate cancer may be generalizable to other cancer types.

Supplementary Methods

Gene Targeting and mouse breeding

All mouse studies are approved by MSKCC IACUC under protocol 06-07-012. Pb-Cre4[9], and Pten[9] mice were previously described and all in the C57B6 background. Rosa26 targeting is described in Srinivas et. al.[40] with modifications (Supplementary Fig. 1). We started with with pBigT-invloxP (kind gift from Juan Pedro Martinez-Barbera) where the loxP sites were inverted in orientation to remove two sense ATG's that may aberrantly initiate translation prior to the ERG gene[41], and subsequently cloned the TMPRSS2-ERG cDNA containing non-coding exon 1 of human TMPRSS2 with exon 4 of ERG[8] and an IRES-nlsEGFP (Addgene plasmid 15037)[42] into the polycloning site respectively. This vector, pBigT-invloxP-ERG-IRES-nlsEGFP was cloned into Rosa26-Pam1 (Addgene 15036)[42]. The vector was targeted into 129 ES cells and injected into C57B6 blastocysts. Genotyping was performed using the following primers: R26-TA-WT-3F (5′-TCCCGACAAAACCGAAAATC-3′), R26-WT-3R (5′-AAGCACGTTTCCGACTTGAG-3′), ERG Ex7F (5′-CAAAACTCTCCACGGTTAATGC-3′), ERG Ex10R (5′-GCACTGTGGAAGGAGATGGT-3′) with wild-type band of 468 bp and targeted band of 205 bp.

To initially characterize gene expression, Pb-Cre4;R26+ and Pb-Cre4;R26 mice were generated through standard mouse breeding. To facilitate breeding, upon generation of R26 homozygous mice, subsequent crosses involved Pb-Cre4;R26 males with R26 females that generated a 1:1 ratio of Cre+ and Cre mice. ChIP-seq and gene expression analysis were carried out in R26 homozygous mice. To cross into the Pten conditional background, we crossed Pb-Cre4;R26 with Pten mice and after two generations, obtained Pb-Cre4;Pten/+, Pten+;R26 mice. From these mice, subsequent breeding between Pb-Cre4;Pten+ males and Pten females generated Pb-Cre4;Pten males (abbreviated as Pten in text and figures) and breeding between Pb-Cre4;Pten+;R26 males and Pten females generated Pb-Cre4;Pten males (abbreviated as Pten in text and figures).

Mouse procedures

To measure serum testosterone level, blood was obtained immediately after CO2 euthanasia via cardiac puncture and testosterone ELISA was performed using a KIT from ALPCO. Mouse prostate dissection was performed as described[13]. For isolation of chromatin and RNA for Pten wild-type mice, all prostate lobes were pooled. For Pten prostates, the ventral, lateral, and dorsal lobes were pooled as the anterior lobe was more cystic and frequently highly fibrotic. Mouse castration was performed as described[9] and RNA isolation was carried out 48 hours after castration.

Chromatin immunoprecipitation and sequencing

To isolate chromatin from mouse prostate, prostates from 6-month old mice were minced using scissors, and crosslinked using 1% paraformaldehyde for 15 minutes. Sample was washed in PBS, resuspended in lysis buffer, and dounced in a Tenbroeck style tissue grounder prior to sonication and IP as previously described[38]. Chromatin isolation from LNCaP cells growing in FBS was performed as previously described[38]. For ETV1 knockdown experiments, chromatin was isolated 72-hours after lentiviral infection.

ChIP-qPCR primers for human KLK3, TMPRSS2 were described[43]. Mouse ChIP-qPCR primers pairs are: Tmprss2 enhancer (5′-GAGGCACTTTTTGCCCAGTG-3′, 5′-CCAGGATGTGTCTGGGGAAC-3′), Fkbp5 enhancer (5′-TGTGGCTGGCACATGAACTCGA-3′, 5′-GCTGTATGCTCCCCACCCCC-3′), and Nkx3-1 enhancer (5′-TGTTGACATGGCTTCCTCGT-3′, 5′-TGGTTTATCGCCGTACCTTT-3′).

Next generation sequencing was performed either on an Illumina Genome Analyzer 2 or HiSeq2000 with 50 base-pair single reads. Reads were aligned to either the mouse genome (mm9) or the human genome (hg 18) using the ELAND alignment software within the Illumina Analysis Pipeline and duplicate read were eliminated for subsequent analysis. Peak calling was performed using MACS 1.4[44] comparing IP chromatin with input chromatin. Based on RefSeq gene annotation, the resultant peaks were separated into promoter peaks (located within ± 2 kb of a transcription start site (TSS)), promoter distal peaks (located from -50 kb of a transcription start to + 5kb of a transcription end), and otherwise intergenic peaks. Genes with one or more peaks in their promoter or promoter distal (also referred to as “enhancer”) regions were considered as AR or ERG targets. The MEME software suite[23] and DREME[24] were applied for motif analysis using 300-bp sequences centered on either AR or ERG peaks. ChIP-seq profiles are presented using Integrated Genome Browser software of SGR format files. For overlapping analysis of AR, ERG, H3K4me3, and H3K4me1 peaks, we defined two peaks overlap if they shared at least one base pair across their full peak spans, as detected by MACS.

RNA analysis

To isolate RNA from mouse prostates, freshly dissected 3-month old prostate tissue was immediately placed into Trizol and homogenized using a FastPrep-24 instrument with Lysing Matrix A (MP Biomedicals). After phase separation, the RNA fraction was further purified using RNAEasy Mini kit.

Gene expression profiling was performed as described[38] using the Illumina MouseWG-6 v2.0 Expression BeadChip (4-month old WT, R26, Pten and Pten) and Illumina MouseRef-8 v2.0 Expression BeadChip (6-month old intact and castrate Pten and Pten). Expression analysis was performed using Partek Genomics Suite. Significantly changed probes induced by ERG expression were defined as genes with a >1.5-fold change and a Benjamini–Hochberg FDR < 0.3.

For gene expression profiling of LNCaP cells, cells were infected using a scrambled or two ETV1-specific shRNAs in the PKO.1 backbone[38] in triplicates. RNA was harvested 72-hours after infection and profiled using Illumina HT-12 microarray. To obtain the AR signature score, the normalized expression of a set of canonical AR-regulated genes was summed[9].

Gene ontology analysis of ERG-regulated gene sets were performed using DAVID[35] and Ingenuity IPA (http://www.ingenuity.com). Gene set enrichment analysis was performed using the JAVA program (http://www.broadinstitute.org/gsea) as described[38]. For ERG profile in Pten loss mouse prostate, genes were ranked from most upregulated to most downregulated in Pten mice compared to Pten mice. For ETV1 profile in LNCaP cells, genes were ranked by the mean changed by two shETV1 hairpins compared to scrambled hairpin. We used the gene sets in the Molecular Signatures Database (MSigDB) and added the following: CARVER_CASTRATION_UP and CARVER_CASTRATION_DN defined by genes upregulated and downregulated >1.5-fold in the mouse prostate (GSE24691), TAYLOR_PCA_ERG_UP defined by genes with greater than 1.5-fold higher expression in ERG-positive compared to ERG-negative tumors (GSE21034)[33], VCAP_siERG_DN defined by genes downregulated >2-fold by siRNA against ERG, VCAP_R1881_UP defined by genes upregulated >3-fold by R1881 treatment[16], and AR_SIG is a set of canonical AR upregulated genes[9].To integrate transcriptome with cistrome, we analyzed the gene that was both present in the Illumina MouseWG-6 v2.0 microarray and mouse RefSeq (19,260 genes) and compared the percent of all 19,260 of these genes with ERG and AR peaks with perturbed genes with ERG or AR peaks.

We analyzed two human prostate cancer datasets, a rapid autopsy series from University of Michigan (GSE35988) employing Agilent 44K microarray and MSKCC prostatectomy series employing Affymetrix Human Exon 1.0 ST Array (GSE21034). PTEN loss is defined, ERG and ETV1 status are defined as shown (Supplementary Fig. 23). AR signature score of each sample is the sum of normalized expression of each gene in a canonical AR signature[9] with the exception of TMRPSS2 because many ETS-positive tumors contain a genomic deletion of TMPRSS2 due to genomic fusion with the ETS partner.

Protein Analysis

The following antibodies were used for IHC, WB, and ChIP: rabbit AR (Epitomics clone ER179(2)), rabbit ERG (Epitomics clone EPR3864), rabbit ETV1[38], rabbit histone H3K4me1 (Abcam ab8895), rabbit AKT (Cell Signaling #4691), rabbit phospho-AKT (Cell Signaling #4060), rabbit Ki67 (Vector Labs #VP-K451), rabbit mouse smooth muscle actin (Sigma A5228), chicken EGFP (Abcam ab13970), mouse β-Actin (clone AC-15, Sigma), mouse GAPDH (clone 1D4, Santa Cruz), rabbit p63 (Epitomics clone EPR5701), rabbit CK5 (Covance), rabbit CK8 (Covance), and rabbit nestin (Abcam).

Tissue paraffin embedding, sectioning, and H&E staining was performed by MSKCC core facility. Immunohistochemistry was performed by the MSKCC molecular cytology core using a Ventana Discovery XT.

To generate lysates for Western blotting, tissue was homologized in RIPA buffer using FastPrep-24 system with Lysing Matrix A (MP Biomedicals).

Statistics

All statistical comparisons between two groups were performed by Graphpad Prism software used two-tailed unpaired t-test. Kaplan Meier analysis was performed by Graphpad Prism using Log-rank Mantel-Cox test.

Accession Codes

Gene expression and ChIP-seq data can be found online at the Gene Expression Omnibus with the following accession numbers.

Gene expression of LNCaP cells infected with scrambled and two different lentiviral shRNA against ETV1: GSEXXXX.

Gene expression comparing four cohorts of mouse prostates (WT, R26, and Pten): GSEXXXX.

Gene expression comparing the effect of castration on Pten, and Pten mouse prostates: GSEXXXX.

ChIP-seq of mouse prostates of the 4 cohorts (WT, R26) using the following antibodies (AR, ERG, H3K4me1, H3K4me3): GSEXXXX.

ChIP-seq of steady state LNCaP cells (ETV1 and H3K4me1) and LNCaP cells infected with scrambled and ETV1sh2 (AR): GSEXXXX.