TMPRSS2:ERG Fusion by Translocation or Interstitial Deletion Is Highly Relevant in Androgen-Dependent Prostate Cancer, But Is Bypassed in Late-Stage Androgen Receptor–Negative Prostate Cancer

Prostate cancer is the most frequent cancer in men in countries with a Western lifestyle and the second cause of male cancer death (1). Surgery and radiation are standard therapy of localized prostate cancer. Palliative therapy of metastatic prostate cancer aims at blocking androgen receptor (AR) function. A better understanding of the molecular mechanisms of tumorigenesis is essential for the development of novel therapies. Additionally, a knowledge of the mechanism of prostate cancer development will improve prediction of the clinical course of the disease.

Recently, overexpression of the ETS gene ERG has been described in clinical prostate cancer (2). Subsequently, it was shown that overexpression of ERG and related ETV1 was due to fusion of the TMPRSS2 gene to either ERG or ETV1 (3). This important finding adds gene fusion to the mechanisms of gene overexpression in epithelial tumors. At low frequency, TMPRSS2 might also be fused to ETV4 in prostate cancer (4). Expression of TMPRSS2 that maps to 21q22 is androgen regulated and prostate specific (5). ERG is also located on 21q22, ∼3 Mbp proximal to TMPRSS2. ETV1 maps to 7p21 and ETV4 to 17q21. Together with FLI1, modified ERG, ETV1, and ETV4 are well-known oncogenes involved in translocations in Ewing sarcoma and acute myeloid leukemia (6).

We investigated the ETS genes ERG, ETV1, ETV4, and FLI1 in human prostate cancers transplanted on nude mice. Xenografts are powerful models for dedicated genetic and molecular studies because they lack normal cells of human origin. The xenografts used represent a variety of clinical stages of prostate cancer, ranging from primary tumors and local metastases to recurrent disease and distant metastases, and from androgen-dependent to androgen-independent cancers (7–9). Our data reveal both interstitial deletion and gene translocation as mechanisms of fusion between TMPRSS2 and ERG. Further, our results show high overexpression of two splice variants of the TMPRSS2:ERG fusion gene in all androgen-dependent xenografts and absence of ERG overexpression in late-stage, AR-negative xenografts, even if they contain a TMPRSS2:ERG fusion gene. The latter observation was also made in AR-negative clinical prostate cancer. In almost all late-stage, AR-negative prostate cancer samples, apparently androgen-independent expression of wild-type ETV4 and FLI1 is detected. These findings show a key role of TMPRSS2:ERG in androgen-dependent prostate cancer, which might be bypassed by other ETS factors in late-stage, AR-negative disease.

Prostate cancer samples. The in vivo growing xenografts PCEW, PC82, PC133, PC135, PC295, PC310, PC324, PC329, PC339, PC346, and PC374 were propagated by serial transplantation on male nude mice as described (7–9). PCEW, PC82, PC295, PC310, and PC329, derived from primary tumors or local metastases, are androgen-dependent. PC133, PC324, PC339, PC346, and PC376 are derived from distant metastases or local progressive disease and are androgen-independent (PC133, PC324, and PC339) or androgen-sensitive (PC346 and PC374). PC135 is androgen independent and is derived from a lymph node metastasis (see Supplementary Table S1).

Clinical prostate tumor samples were obtained from recurrent disease by transurethral resection (TUR-P) after informed consent, following approval of the institutional ethical committee.

DNA and RNA preparation. Genomic DNA was isolated using the Puregene system from Gentra Systems (Minneapolis, MN) according to the procedure described by the manufacturer. Xenograft RNA was isolated according to the LiCl protocol (10). For isolation of RNA from clinical samples, the Illustra mini RNA kit (General Electric Healthcare, Fairfield, CT) was used.

Array-based comparative genomic hybridization. Arrays were produced from the human 3600 bacterial artificial chromosome (BAC)/P1-derived artificial chromosome genomic clone set of the Welcome Trust Sanger Institute, covering the full genome at ∼1 Mb spacing. Degenerated oligonucleotide PCR products were prepared for spotting on CodeLink slides (General Electric Healthcare) according to published protocols (11) with some modifications (12). DNA labeling and hybridization were done essentially as described (11) with minor modifications (13). After hybridization arrays were scanned in a ScanArray Express HT (Perkin-Elmer, Freemont, CA). The resulting images were analyzed with GenePix Pro 5.0 software (Axon Instruments, Foster City, CA) and subsequently visualized with an excel macro (12).

Quantitative PCR. Analysis of mRNA expression was done by quantitative PCR (QPCR). Two micrograms RNA were reverse transcribed using 400 units M-MLV RT (Invitrogen Life Technologies, Carlsbad, CA) and an oligo-dT12 primer. QPCR was done in an ABI Prism 7700 Sequence Detection System (Applied Biosystems, Foster City, CA). QPCR reactions were done in Power SYBR Green PCR Master Mix (Applied Biosystems) containing 330 nmol/L forward and reverse primer in a total volume of 25 μL. Thermocycling conditions were according to the recommendations of the manufacturer. Amounts of specific RNAs for each sample were determined relative to posphobilinogen deaminase (PBDG) by the standard curve method (Applied Biosystems). Primer combinations used were as follows: PBGD-F (5′-CATGTCTGGTAACGGCAATG-3′) and PBGD-R (5′-GTACGAGGCTTTCAATGTTG-3′). PSA-4A (5′-ACGTGTGTGCAAGTTCACC-3′) and PSA-5B (5′-TGTACAGGGAAGGCCTTTCG-3′). TMPRSS2-F (5′-CCTCTGGTCACTTCGAAGAAC-3′) and TMPRSS2-R (5′-GTAAAACGACGTCAAGGACG-3′). AR-7/8A (5′-TGACTCCGTGCAGCCTATTG-3′) and AR-8B (5′-ATGGGAAGCAAAGTCTGAAG-3′). TMPRSS2-E1:ERG-E4F (5′-AGCGCGGCAGGAAGCCTTA-3′) and ERG-E4/5R (5′-CATCAGGAGAGTTCCTTGAG-3′). TMPRSS2-E2:ERG-E4F (5′-GATGGCTTTGAACTCAGAAGC-3′) and ERG-E4/5R. TMPRSS2-E3F (5′-CCACCAGCTATTGGACCTTA-3′) and ERG-E4/5R (5′-CATCAGGAGAGTTCCTTGAG-3′). TMPRSS2-E1F (5′-GAGCTAAGCAGGAGGCGGA-3′) and ETV1-E5R (5′-TGACTGCAGGCAGAGCTGAT-3′). TMPRSS2-E2F (5′-CCTATCACTCGATGCTGT-3′) and ETV1-E5R. ERG-F (5′-TGCTCAACCATCTCCTTCCA-3′) and ERG-R (5′-TGGGTTTGCTCTTCCGCTCT-3′). ETV1-F (5′-CATACCAACGGCGAGGATCA-3′) and ETV1-R (5′-TGGAGAAAAGGGCTTCTGGA-3′). ETV4-F (5′-ACCGGCCAGCCATGAATTAC-3′) and ETV4-R (5′-GAGAGCTGGACGCTGATTG-3′). FLI1-F (5′-GAGGAGCTTGGGGCAATAAC-3′) and FLI1-R (5′-AGAGCAGCTCCAGGAGGAAT-3′).

Interphase fluorescent in situ hybridization. Nuclear suspensions of the prostate cancer xenografts were prepared essentially as described by Vindelov et al. (14). Interphase fluorescent in situ hybridization (FISH) was done as described (15). BAC clones RP11-164E1, RP5-1031F17 (both flanking ERG, see Fig. 2A), RP11-113F1 (TMPRSS2, see Fig. 2A), RP11-79G16 (ETV1), RP11-268E15 (ETV4) and RP11-44O2 (FLI1) were purchased from BacPac Resources (Oakland, CA). Specificity of BACs is shown in Supplementary Figs. S1 and S2. BAC DNA clones were biotin-16-dUTP or digoxigenin-11-dUTP labeled using a nick translation reagent kit (Vysis, Downers Grove, IL) according to the directions of the manufacturer. Biotin-labeled probes were visualized with FITC-conjugated avidin (Vector Laboratory, Burlingame, CA) and digoxigenin-labeled probes with rhodamine-conjugated antidigoxigenin antibody (Roche, Mannheim, Germany). Cells were 4′,6-diamidino-2-phenylindole counterstained. Images of the three fluorochromes were collected on an epifluorescence microscope (Leica DM, Rijswijk, the Netherlands) equipped with appropriate filter sets (Leica) and a CCD cooled camera (Photometrics, Tucson, AZ).

Sequence analysis. PCR products were purified using SAP/Exonuclease I (USB Corporation, Cleveland, OH) according to the instructions of the manufacturer. Purified PCR fragments were labeled using the ABI BigDye Terminator Ready Reaction kit v3.1 (Applied Biosystems) according to instructions of the manufacturer. In the sequence reactions, the same primers were used as for fragment amplification. Sequence samples were run on the ABI 3100 genetic Analyzer (Applied Biosystems).

Eleven xenografts derived from various stages of clinical prostate cancer (Supplementary Table S1; refs. 7–9) were used to decipher the role of ETS genes. PCEW, PC82, PC295, PC310, PC329, PC346, and PC374 are AR positive. PCEW, PC82, PC295, PC310, and PC329 grow androgen dependent on male nude mice; PC346 and PC374 are, to a varying extent, androgen sensitive. PC133, PC135, PC324, and PC339 are androgen independent.

First, expression of AR, PSA, and TMPRSS2 in xenografts was assessed by QPCR. There is a good correlation between the expression of AR and the well-known, androgen-regulated PSA and TMPRSS2 genes in androgen-dependent and androgen-sensitive xenografts (Fig. 1A; refs. 5, 9, 16). An exception is PC310, which shows AR and PSA expression, but clearly is TMPRSS2 negative (see below). All androgen-independent xenografts are AR negative or express an inactive AR, as deduced from lack of PSA and TMPRSS2 expression (see also ref. 17).

Next, we investigated expression of ERG and TMPRSS2:ERG fusion transcripts using an ERG specific primer set and primer sets spanning TMPRSS2 exons 1, 2, or 3 and ERG exon 4, respectively, combined with an ERG exon 4/5 primer (Fig. 1B). In all five androgen-dependent xenografts, ERG overexpression corresponded with AR and PSA expression, linking ERG to a functional AR and to strict androgen-dependent tumor growth. Overexpression of ERG correlated with the presence of TMPRSS2:ERG fusion transcripts. As confirmed by sequencing, due to alternative splicing in all five xenografts, two transcripts were present, one containing TMPRSS2 exon 1 linked to ERG exon 4 and a second, linking TMPRSS2 exons 1 and 2 to ERG exon 4, respectively (Fig. 1D). PCEW contained a third transcript linking part of TMPRSS2 exon 3 to ERG exon 4 due to use of a cryptic splice donor site in TMPRSS2 exon 3. The open reading frame (ORF) of transcript 1 is predicted to start at an internal ATG in ERG exon 4. In transcripts 2 and 3, the ORF will begin at the start codon of TMPRSS2 and continues in-frame with the indicated part of ERG (Fig. 1D). The reason of the high frequency of alternative splicing is at present unclear. Possibly expression of a truncated ERG protein from the TMPRSS2(exon 1):ERG(exon 4) fusion transcript favors tumor growth.

Androgen-sensitive PC374 showed high ETV1 overexpression (Fig. 1C). High ETV1 expression correlated with TMPRSS2:ETV1 gene fusion, as shown by interphase FISH on nuclear suspensions using ETV1 and TMPRSS2 specific BACs (Fig. 1C), and the presence of two splice variants of TMPRSS2:ETV1 (Fig. 1C and D). ERG and ETV1 were not overexpressed in PC346 or in late-stage, AR-negative prostate cancer xenografts. Interphase FISH indicated that low-level ETV1 expression in PC135 could not be correlated with TMPRSS2 gene fusion (Fig. 1C). 5′-Rapid amplification of cDNA ends (5′-RACE) confirmed that ETV1 expressed in PC135 was wild type (data not shown).

TMPRSS2 and ERG map in the same orientation at short distance on chromosome band 21q22.2-q22.3 (Fig. 2A). To determine the mechanism of TMPRSS2:ERG gene fusion, genomic DNA from the five xenografts overexpressing the fusion gene was investigated by 1-Mbp-spaced array-based comparative genomic hybridization (array-CGH). Two different representative array-CGH profiles of chromosome 21 are depicted in Fig. 2B and C. In PC295, the region between ERG and TMPRSS2 was lost, as indicated by the low T/R ratio of the four BACs mapping in this chromosomal region (Fig. 2B). A similar profile was present in PC329 (data not shown). Although PC82 contains the fusion transcript, the region between ERG and TMPRSS2 was largely present (Fig. 2C). A comparable profile was found in PCEW and PC310 (data not shown). In PC310, the profile was accompanied by a small homozygous deletion of the last exons of TMPRSS2 (data not shown), explaining total absence of TMPRSS2 transcripts in this xenograft. We extended the array-CGH data by interphase FISH of PC82, PC295, and PC310 nuclei. We used as hybridization probes BACs RP11-164E1 and RP5-1031P17, which map at a distance of ∼1 Mbp, flanking ERG at positions indicated in Fig. 2A. Both BACs exclusively stained chromosome band 21q22.2 and showed two spots on interphase nuclei from normal cells (Supplementary Fig. S1A and B). Representative nuclei of PC295 and PC82 are presented in Fig. 2B and C, respectively. In nuclei from PC295 cells, we found a higher number of green spots than red spots (average ratio 1.6), indicative of loss of the region between ERG and TMPRSS2 in one copy of chromosome 21. In PC82 and PC310, we observed an identical number of green and red spots (average ratios 1.0 and 1.1, respectively). In PC82 that contains four copies of chromosome 21, three pairs of red and green spots were always closely linked and one pair was clearly separated, as illustrated in Fig. 2C. Both array-CGH and FISH data strongly suggest two different mechanisms of TMPRSS2:ERG fusion: one by an ∼3 Mbp interstitial deletion of one copy of chromosome 21, and a second more complex mechanism by chromosomal translocation.

Array-CGH of genomic DNA from two of the six xenografts that did not overexpress TMPRSS2:ERG showed a remarkable pattern. In the androgen-independent, AR-negative xenografts PC133 and PC339, we detected a similar interstitial deletion as in PC295 and PC329. The chromosome 21 profile of PC339 is depicted in Fig. 3A. Long-range PCR followed by sequencing confirmed the fusion between TMPRSS2 and ERG in PC133, mapping the chromosomal breakpoints in ERG intron 3 and in TMPRSS2 intron 1, respectively (Fig. 3B). Similarly, PCR plus sequencing identified TMPRSS2:ERG fusion in PC324 that does not show a 21q22 interstitial deletion. This adds a third AR-negative xenograft to those with TMPRSS2:ERG gene fusion without expression of the fusion gene. Like in PC133, the breakpoint in PC324 is in TMPRSS2 intron 1. We also mapped the breakpoints in AR-positive PC82, PC295, PC310, and PC329 (Fig. 3B). As expected from expression data (Fig. 1B), these breakpoints were in intron 2 of TMPRSS2 and in intron 3 of ERG. All six breakpoints in ERG were in the last part of intron 3, suggesting a preferred region of recombination in this part of the gene. It remains to be investigated whether the difference in TMPRSS2 introns involved in ERG fusion between AR-positive and AR-negative xenografts, introns 2 and 1, respectively, is coincidental or of functional importance. The absence of TMPRSS2:ERG expression in PC133, PC324, and PC339 (Fig. 1B) indicates that it is not involved in the androgen-independent growth of these xenografts. Importantly, however, the presence of TMPRSS2:ERG in genomic DNA strongly suggests that the fusion gene has been instrumental in an earlier androgen-dependent stage of tumor growth.

We postulated that in PC133, PC324, and PC339, androgen-regulated ERG expression is bypassed and subsequently down-regulated by other mechanisms of progressive tumor growth. One mechanism to become independent of androgen-regulated ERG overexpression might be by androgen-independent increased expression of a member of the ETS transcription factor gene family. As shown in Fig. 1, we had no evidence that this was the case for ERG or ETV1. Therefore, we investigated in the xenografts expression of two other ETS transcription factors known to be involved in oncogenesis, ETV4 and FLI1. ETV4 was highly expressed in PC133 and FLI1 in PC324 and PC135 (Fig. 3C and D). Also, in PC339, we observed expression of FLI1. ETV4 and FLI1 were not or were hardly expressed in androgen-dependent or androgen-sensitive xenografts. Overexpression was not the result of fusion to TMPRSS2 as illustrated by interphase FISH (Fig. 3C and D). Additional 5′-RACE experiments confirmed that ETV4 and FLI1 mRNA in AR-negative xenografts was wild type and not the result of fusion to other genes (data not shown).

In clinical prostate cancer, many recurrent tumors still express a functional AR. However, a substantial proportion of recurrent tumors is heterogeneous for AR expression or is AR negative (18, 19). We investigated whether, like in xenografts, TMPRSS2:ERG gene fusion without expression of the fusion gene was present in clinical samples. Array-CGH showed in 4 of 11 recurrent tumors the interstitial deletion at 21q22 indicative of TMPRSS2:ERG fusion (see, e.g., Fig. 4A and B). From three of these tumors (T1-1, T1-8, and T3-7), RNA was available. Importantly, we also had available DNA and RNA from recurrent tumor T1-7 of which AR-negative xenograft PC324 was derived. These four tumors and a control recurrent tumor without TMPRSS2:ERG fusion (T6-9) were investigated by QPCR for specific gene expression patterns (Fig. 4C and D). T1-1, T3-7, and T6-9 expressed AR and its target genes PSA and TMPRSS2, although expression in T1-1 was low. In contrast, T1-8 and, as expected, T1-7 showed hardly any AR, PSA, and TMPRSS2 expression. All tumor samples had a basal level of ERG expression. ERG overexpression, combined with TMPRSS2:ERG fusion transcripts, was clearly detected in T3-7 and absent in T1-8 and T1-7. T1-1 showed a low level of TMPRSS2:ERG expression, which might indicate tumor heterogeneity. The data confirm and extend the findings in xenografts showing that AR-negative tumors can carry a TMPRSS2:ERG fusion gene without expression of the gene. Background expression levels and presumed heterogeneity of tumors hampered accurate investigation of ETV4 and FLI1 in the clinical samples (Supplementary Fig. S3). ETV4 expression was highest in AR-negative T1-8 and T1-7, but differences with other tumor samples were small. Like in PC324, expression of FLI1 was high in T1-7, but T1-1 also showed high expression of FLI1. The latter might be explained by tumor heterogeneity, as proposed previously (Fig. 4C and D). Obviously, more detailed immunohistochemical studies, including AR and ETV4 or FLI1 double staining, are needed to substantiate the latter observation.

In conclusion, our xenograft data extend previous observations in clinical prostate cancer (3, 4) and shed new light on the role of ETS transcription factors in prostate cancer. First, we detected two mechanisms of gene fusion between ERG and TMPRSS2. Second, we observed that TMPRSS2:ERG overexpression is functionally correlated with AR expression. Both in xenografts and clinical samples, we showed that the TMPRSS2:ERG fusion gene can be present in absence expression of the gene in AR-negative tumors. Furthermore, our data suggest that other members of the ETS family, possibly wild-type ETV4 or FLI1, might take over the role of androgen-regulated TMPRSS2:ERG in late-stage, AR-negative prostate cancer.

Grant support: Erasmus University Medical Center and the Dutch Cancer Society (Koningin Wilhelmina Fonds).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.