Main

The role of recurrent chromosomal translocations in the pathogenesis of many hematopoetic malignancies is well established.1 In lymphomas and leukemias these translocation events lead either to the activation of the 3′ partner gene—as in the activation of MYC following fusion to IgH in Burkitt's lymphoma—or to the formation of fusion proteins, such as the BCR-ABL fusion protein resulting from the t(9;22) translocation in chronic myeloid leukemia. In recent years, a variety of recurrent chromosomal translocations, most of which result in the formation of fusion proteins, have been described in sarcomas.2 It is now known that approximately one-third of all soft tissue sarcomas are characterized by specific recurrent translocations, and in many cases these are the only detectable cytogenetic abnormality.2 In both sarcomas and hematopoetic malignancies, many of these recurrent translocations have proven to be pathogenetically important.1, 2

In contrast to lymphomas/leukemias and sarcomas, carcinomas are generally believed to arise via multi-step carcinogenesis, and recurrent chromosomal trasnslocations, with the exception of MECT1-MAML2 t(11;19) translocation in low-grade mucoepidermoid carcinoma, are very uncommon in carcinomas.3, 4, 5 However, a recently described gene fusion between TMPRSS2 and ETS family genes in prostate carcinoma stands to challenge the old paradigm that epithelial malignancies arise by nonspecific chromosomal aberrations. Through bioinformatic analysis of DNA microarray data and subsequent fluorescence in situ hybridization (FISH) and reverse transcription polymerase chain reaction (RT-PCR) analysis, Tomlins et al6 identified TMPRSS2-ERG fusion in 16/29 (55%) and TMPRSS2-ETV1 fusion in 7/29 (27%) cases of prostate cancer. ETV4, another member of the ETS family, was later identified as another minor partner in this gene fusion.7 The high prevalence of prostate cancer in western societies, in combination with the reported frequency of fusion between TMPRSS2 and ETS family genes, would make this one of the most common genetic alterations identified in human malignancy.

While the concept of a pathogenetically important recurrent genetic aberration in prostate carcinoma may be novel, the 3′ genes involved in this fusion event are well known. ERG, ETV1 and ETV4 are all members of the ETS gene family, and ERG and ETV1 have been previously characterized as minor fusion partners (<10% cases) of EWS-ETS gene fusions in Ewing's sarcoma/primitive neuroectodermal tumors.8 In addition, the ERG gene has also been shown to be the 3′ fusion partner in translocation events seen in a subset of acute myeloid leukemia, with the 5′ partners being FUS, EWS or ELF4.9 The 5′ partner of this fusion, TMPRSS2, encodes a transmembrane serine protease that is constitutively expressed in prostate under the transcriptional control of androgens.10, 11

In prostate cancer, multiple TMPRSS2-ERG fusion variants have been identified at the transcript level, most of them involving only the 5′ untranslated sequence of TMPRSS2.6, 12, 13, 14, 15 The TMPRSS2-ETS fusions thus lead to overexpression of the ERG or ETV protein induced by the TMPRSS2 promoter. In contrast to other translocations involving the ERG gene, for example, EWS-ERG fusion, no fusion protein is expected in most TMPRSS2-ERG fusions. Unlike the TMPRSS2-ETV1 fusion, which involves a t(7;21) translocation, the TMPRSS2 and ERG genes are located close to each other on chromosome 21, approximately 3 Mb apart.15 Thus, the TMPRSS2-ERG fusion could result either from balanced or unbalanced translocation, or from deletion of the intervening DNA segment, and the literature to date suggests deletion as the more frequent event.16, 17

In the present study, we evaluated a series of 82 formalin-fixed paraffin-embedded prostate cancer specimens, by RT-PCR and by FISH analyses. TMPRSS2-ERG fusion was observed in 43% of the cases, more commonly resulting from deletion than translocation. Novel fusion transcript variants were identified, and overexpression of ERG downstream sequence was seen in most fusion-positive cases. Possible correlation between the fusion status, histological features and grade of the tumor was also observed. In comparison to the frequent TMPRSS2-ERG fusion, only one TMPRSS2-ETV1 fusion was identified, indicating that this fusion is less common than previously reported.

Materials and methods

Tissue Specimens

Tissue specimens, derived from radical prostatectomy, were obtained from the Department of Pathology at the Weill Medical College of Cornell University, following a protocol approved by the Institutional Review Board.

RNA Extraction

One representative block was identified from each case, and four 8 μm sections were used for RNA extraction, using Optimum FFPE RNA isolation kit (Ambion, Austin, TX, USA). The non-tumor areas on the slides were manually removed with surgical blades, and the remaining tissue was scraped into an Eppendorf tube for RNA extraction.

RT-PCR and DNA Sequencing

All primer sequences are listed in Table 1. TMPRSS2-ERG and TMPRSS2-ETV1 gene fusions were evaluated by nested RT-PCR. For TMPRSS2-ERG fusion, two sets of RT-PCR were performed, coupling a 5′TMPRSS exon 1 primer to a 3′ERG primer located at either exon 4 or exon 5. For TMPRSS2-ETV1 fusion, the 5′TMPRSS2 exon 1 primer was coupled to a 3′ETV1 exon 6 primer. Although fusion products were often detectable by gel electrophoresis after the first PCR, nested PCRs were performed on all cases with corresponding internal primers, and PCR products were identified by 1% agarose gel electrophoresis and ethidium bromide visualization. In cases where discrete PCR products were detected, direct PCR sequencing was performed on the purified PCR products. DNA elution from the gel fragments was performed if more than one DNA species was present. Overexpression of downstream ERG or ETV1 sequences was evaluated by quantitative RT-PCR assays, using ERG exon 5–exon 6 primer pair or ETV1 exon 6–exon 7 primer pair, coupled to corresponding TaqMan FAM-labeled ERG or ETV1 probe (Applied Biosystems gene expression assay ID Hs00171666 and Hs00231877, respectively). 18S ribosomal RNAs were used as endogenous controls for RNA quality. ETV1 exon 2–exon 3 primer pair amplification (Applied Biosystems assay ID Hs00951945) was also found to be a reliable endogenous control and was used to normalize the ERG amplification results. PCR were performed for 45 cycles as previously described.18

Table 1 RT-PCR primers for detection of TMPRSS2-ERG and TMPRSS2-ETV1 fusionsa
Full size table

FISH

Interphase FISH was performed on the same set of cases analyzed by RT-PCR. Tissue microarray slides were prepared from the blocks used for RNA extractions, with each case represented by three 1 mm tissue cores. TMPRSS2-ERG fusion was evaluated using break-apart probes, consisting of rhodamine-labeled 5′ERG probe (BAC RP11-95I21) and FITC-labeled 3′ERG probe (BAC RP11-476D17). Using this break-apart approach, a cell with two normal copies of chromosome 21 would have two yellow signals, due to the close proximity of the two probes. A cell with a translocation would have one yellow, one green and one red signal, and a cell in which TMPRSS2-ERG fusion was due to a deletion of intervening DNA would have one green and one yellow signal—the 5′ (red) signal would be lost. TMPRSS2-ETV1 fusion was evaluated with a two-color two-signal approach, using rhodamine-labeled TMPRSS2 probe (BAC RP11-35C4) and FITC-labeled ETV1 probe (BAC RP11-124L22). TMPRSS2-ETV1 translocation results in fusion of these two signals, generating a yellow signal. BAC clones were obtained from Children's Hospital of Oakland Research Institute (CHORI) and from Invitrogen (Carlsbad, CA, USA). Specificity and quality of the probes were confirmed by hybridization to the metaphase spread of normal peripheral lymphocytes. In a previous study,6 the same 5′ERG probe that we used showed a weak signal on chromosome 2 due to cross-hybridization. However, we detect this nonspecific signal neither in our metaphase spread nor in the interphase of non-neoplastic cells on tissue sections. Tissue pretreatment was performed using Paraffin Pretreatment kit I (Vysis, Des Plaines, IL, USA), and hybridization and washing were performed using Vysis hybridization reagents, following the manufacturer's protocols. An average of 100 cells were evaluated, and the FISH results were independently scored by a cytogeneticist (S Mathew) and a pathologist (S Rohan).

Results

Frequency of TMPRSS2-ETS Fusions

Of 82 cases, eight cases showed poor RNA quality, evidenced by suboptimal 18S rRNA amplification and lack of amplification with the ETV1 exon 2/exon 3 primer pair, and these eight cases were evaluated by FISH only. Excluding these cases, TMPRSS2-ERG fusion transcripts were detected in 33 of 74 cases (45%). Examples of RT-PCR results are shown in Figure 1.

Figure 1
figure 1

Representative TMPRSS2-ERG fusion transcripts by RT-PCR. The arrows indicate the following three transcript variants in this panel: T1/E4 (lanes 2, 3, 6, 8), T1-4/E4 (lane 4) and T1/E2 (lane 9). Lanes 1, 5 and 7 were negative for gene fusion. The higher-molecular-weight DNA bands in lanes 1 and 7 were sequenced and shown to be unrelated to TMPRSS2 and ERG, representing nonspecific amplifications.

Full size image

FISH analysis for ERG-related translocations and/or deletions was successfully performed on tissue microarray in 59 of the 82 cases. Gene fusion signals were detected in 27 cases (46%), resulting from either translocations or deletions (see below). Combining both assays, 35 of 82 cases (43%) were shown to have TMPRSS2-ERG fusion by one of the two assays. Both RT-PCR and FISH data were available in 52 cases, including 25 concordant positive and 25 concordant negative cases. Two discordant cases were seen, one PCR-negative, FISH-positive, and the other PCR-positive, FISH-negative. RT-PCR was repeated and confirmed in both cases. Using FISH as the gold standard, RT-PCR using FFPE thus had 93% sensitivity and 93% specificity.

In contrast to the frequent TMPRSS2-ERG fusion, the TMPRSS2-ETV1 fusion was not found in any case by RT-PCR. However, one case did show evidence of TMPRSS-ETV1 fusion by FISH analysis (Figure 2d).

Figure 2
figure 2

TMPRSS2-ETS fusion analysis by FISH. (a–c) TMPRSS2-ERG fusion was analyzed using break-apart ERG probes. TMPRSS2-ERG fusion resulting from deletion of the intervening sequence led to an isolated 3′ERG signal (green) and no 5′ERG (red) signal (a). In contrast, a red signal is detected in cases of TMPRSS2-ERG fusion due to translocation (b). Low-copy ERG amplification coexisted with TMPRSS2-ERG fusion in one case (c), which in most cells showed an intact copy of ERG (yellow) and two copies with only 3′ERG signal (green), likely due to duplication of the copy carrying TMPRSS2-ERG fusion. (d) The only case of TMPRSS-ETV1 fusion, represented by the fused (yellow) signal.

Full size image

Deletion, Translocation and Other FISH Findings

Of the 27 FISH-positive cases, evidenced by split ERG signals, 22 (81%) cases showed loss of the split 5′ ERG signal (red), indicating a deletion of the intervening sequence between TMPRSS2 and the ERG genes (Figure 2a). The remaining five cases, in contrast, retained both 5′ (red) and 3′ (green) split signals, indicating a translocation instead of deletion (Figure 2b).

In addition to deletion and translocations, one case showed additional ERG copies, with up to five 3′ (green) ERG signals seen in the tumor cells. This case was TMPRSS2-ERG fusion positive by RT-PCR. Three ERG signals (one yellow and two green) were seen as the most common karotype in this case, indicating one intact copy (yellow) and two copies with the 5′ERG sequence deleted, presumably resulting from duplication of the copy containing TMPRSS2-ERG fusion (Figure 2c).

Correlation of Gene Fusion to Downstream ETS Overexpression

The downstream ERG mRNA expression level was evaluated using qRT-PCR assay and ERG exon 5 and exon 6 primers, 3′ to all fusion junctions (see below). PCR amplification using ETV1 exon 2/exon 3 primers, 5′ to any potential ETV1-related gene fusion, was found to produce consistent amplification results across most specimens, and this was used as endogenous control to evaluate potential ERG and ETV1 mRNA overexpression. Figure 3a shows the correlation between the TMPRSS2-ERG gene fusion status and the expression levels of the downstream ERG sequences. Setting a normalized –ΔCt(ERG exon 5/6-ETV1 exon 2/3) value of 4 as the cut-off threshold, ERG overexpression was found in 32 of 75 cases with valid RT-PCR data, among which 29 were positive for TMPRSS2-ERG fusion by RT-PCR and/or by FISH. FISH was negative in two of the remaining three cases, and failed in the third case.

Figure 3
figure 3

Correlation between mRNA expression of ETS downstream sequences (ERG (a) and ETV1 (b)) to TMPRSS2-ETS fusion status. Cases positive for TMPRSS2-ERG or TMPRSS2-ETV1 fusion (most by RT-PCR, two by FISH only, see text) are shown in solid circles, whereas negative cases are in open circles. The y-axis depicts relative expression levels to ETV1 5′mRNA, the endogenous control, and higher numbers indicate higher mRNA levels. By setting the threshold for overexpression at 4 (dashed lines), most TMPRSS2-ERG fusion-positive and -negative cases can be separated ((a), see text). All but one case were negative for TMPRSS2-ETV1 fusion, and similar ETV1 downstream mRNA levels were seen in all but three fusion-negative cases.

Full size image

Conversely, of the 33 TMPRSS2-ERG RT-PCR fusion-positive cases, 28 showed ERG overexpression, with five cases showing ERG levels similar to those of fusion-negative cases. Using fusion-negative, ERG overexpression-negative cases as control and assuming 100% PCR amplification efficiency, the calculated downstream ERG mRNA levels in the fusion-positive group was on average 11-fold greater than that of the control group.

Downstream ETV1 overexpression was similarly investigated, and only three cases showed a clear increase of ETV1 downstream transcript level (Figure 3b). Both RT-PCR and FISH were negative in these cases. The only TMPRSS2-ETV1 translocation-positive case detected by FISH showed no detectable ETV1 overexpression.

TMPRSS2-ERG Transcript Variants

Six TMPRSS2-ERG fusion transcript variants were identified in this study. Table 2 shows a compilation of these six and additional 13 variants reported in the literature. The most predominant variant, seen in 29/33 (88%) of fusion-positive cases, fused TMPRSS2 exon 1 (5′ untranslated exon) to exon 4 of ERG, and this transcript was designated T1/E4. Using this terminology, the other transcript variants found in this study were T1/E5 (six cases), T1/E2 (two cases), T2/E4 (one case), T1-4/E4 (one case) and T1-4/E5 (one case). All six cases positive for T1/E5 also expressed T1/E4 transcript. In contrast, the T1/E2 and T2/E4 cases did not express T1/E4 or T1/E5. One single case showed two novel transcripts consisting of TMPRSS2 exon 1 alternatively spliced to exon 4, which was fused to the ERG exon 4 or exon 5 (T1-4/E4 and T1-4/E5).

Table 2 TMPRSS2-ERG fusion transcript variants identified in current study and in the literature
Full size table

Of the different variant transcripts described, five variants were reported by at least three groups and constituted the main variants isolated to date (Figure 4). As the TMPRSS2 native translational initiation site is located in exon 2, only the T2/E4 transcript would result in a TMPRSS2-ERG fusion protein. The T2/E4 variant potentially encodes a fusion protein of 454 amino acids, of which the N-terminal five residues being derived from TMPRSS2, a region outside of any known functional domain. Among the other four common transcripts, only the T1/E2 variant can encode the full-length ERG protein (462 amino acids), the other three variants can only encode truncated ERG proteins using an internal methionine as translational initiation site, with predicted putative proteins of 423 (T1/E4) and 363 (T1/E5 and T2/E5) residues, both still containing the ETS functional domain.

Figure 4
figure 4

Major TMPRSS2-ERG fusion transcript variants and their putative protein products. T1/E4 is by far the most commonly detected transcript (see text). Exons are depicted in open boxes and protein coding sequences are in color boxes. Except for T1/E2 variant, all other variants code for proteins with truncated ERG sequences. The predicted translational initiation sites are represented by solid arrows, with the ERG amino-acid numbers of the full-length ERG protein indicated. Although the T2/E5 variant could initiate from the TMPRSS2 sequence (dashed arrow), this would not be in frame with the ERG and likely not translated.

Full size image

Correlating Genetic Changes to Pathological Stages and Histological Parameters

The 36 fusion-positive cases, by either RT-PCR or FISH, included 23 of 54 (43%) stage 2 (T2) tumors, 13 of 28 (46%) stage 3 (T3) tumors and 2 of 5 (40%) lymph node-positive (N1) tumors. Therefore, the fusion status did not correlate to the pathological stages. In contrast, these 36 fusion-positive cases contained eight of 12 (67%) Gleason's 6, 22 of 52 (42%) Gleason's 7, one of five Gleason's 8, and five of 13 of Gleason's 9 carcinomas. The frequency of TMPRSS2-ERG fusion was lower in poorly differentiated carcinoma (Gleason's 8 and above, 6/18, 33%) than in the better differentiated ones (Gleason's 6 and 7, 30/64, 47%), but the difference was not statistically significant (P=0.42). The differences between tumors of individual grades were also not statistically significant.

Possible morphological-genetic correlations were explored, and the results are shown in Table 3. Tumors with TMPRSS2-ERG fusions more frequently had focal intraluminal mucin, amphophilic cytoplasm and cribriform architecture. In contrast, tumors without TMPRSS2-ERG fusions more often had focal foamy gland change, signet ring-like feature or ductal differentiation. However, the only statistically significant difference was focal intraluminal mucin, seen at a much higher frequency in tumors with the fusion (78% vs 46%, P=0.004).

Table 3 The correlation of the histological features and TMPRSS2-ERG fusion
Full size table

Discussion

The clinical and pathological significance of recurrent chromosomal translocation in hematopoetic malignancies and certain sarcoma subtypes has been recognized for many years. The fusion of TMPRSS2 to ETS family genes in prostate cancer, first described by Tomlins et al,6 raised the possibility that similar recurrent genetic alterations may also be important in prostate carcinogenesis. However, a few substantial differences have now emerged based on our present study and other recent studies.6, 12, 13, 14, 16, 19

First, the frequency of gene fusion in prostate cancer is lower than in translocation-positive sarcomas. In synovial sarcoma, for instance, the fusion of SYT-SSX is present in essentially all cases, to the extent that the absence of SYT-SSX fusion would raise serious doubt about the diagnosis. In the initial study of prostate cancer, a very high TMPRSS2-ETS fusion frequency (79.3%, 23/29) was observed, including TMPRSS2-ERG fusion in 55% (16/29) and TMPRSS2-ETV1 in 24% (7/29) of cases. Such a high fusion frequency was similarly observed by Soller et al,13 who found TMPRSS2-ERG fusion in 14/18 cases (78%), but no TMPRSS2-ETV fusions. In comparison to these two studies, we observed a significantly lower frequency (43%) of TMPRSS2-ERG fusion in our larger series, and only a single TMPRSS2-ETV1 fusion was detected by FISH. To exclude the possibility of false-negative results due to RT-PCR assays on paraffin-embedded tissues, we analyzed the same set of cases for ERG overexpression and by FISH, and both showed similar results. This lower frequency of gene fusion was indeed comparable to that observed in the more recent studies,14, 16 indicating TMPRSS2-ERG fusion as the crucial event in about 40–55% of prostate cancer cases. Additionally, similar to the study by Soller et al,13 we found TMPRSS2-ETV1 fusion to be a rare event. Although it is possible that the remaining half of the cases might have involved currently unknown gene fusion partners, the finding that ERG and ETS genes are the only genes found to be overexpressed so far6, 20 argues against such a possibility and implies the existence of more than one pathway in prostate carcinogenesis.

Another difference between ERG fusions in prostate cancer and gene fusions in other types of malignancies is the lack of a fusion protein in the former. The ERG gene is also involved in chromosomal translocations of Ewing's sarcoma and acute myeloid leukemia, with EWS being the most common 5′ partner. The EWS-ERG fusion, however, leads to fusion proteins that include functional DNA/RNA binding domains of both partners, presumably resulting in a synergistic biological effect. In contrast, the TMPRSS2-ERG fusion will create fusion proteins in rare cases, for example, in the T2/E4 transcript variant. Even in these cases, the TMPRSS2 sequence is minimal and not anticipated to be biologically active. The result of the TMPRSS2-ETS fusion is thus analogous to the IgH-MYC fusion in B-cell lymphomas, in that the 5′ partner serves the function of providing a tissue-specific promoter, resulting in constitutive expression of the 3′ gene.

One unique feature of the TMPRSS2-ERG fusion is the vast diversity of fusion transcripts. We identified six variants in the present study, including two novel variants T1-4/E4 and T1-4/E5. Compiling our and other recent RT-PCR studies, a total of 19 transcript variants have been identified to date, variably and confusingly named types I–VIII,14 1–14,12 etc, by investigators. In light of this complexity, we would propose that a generic nomenclature system be used, and these variants would be named as T1/E4, T1/E5, T1-4/E4, etc, as was first used by Clark et al.12 From Table 2, it is clear that T1/E4 is the predominant transcript observed in all studies, and other common variants are T1/E5, T1/E2, T2/E4 and T2/E5. Intriguingly, except for the T2/E4 variant, all other fusion transcripts, including the most common T1/E4 variant, can only encode truncated ERG proteins. The translational efficiency of these aberrant RNAs and the biological activity of the truncated proteins in prostate cancer remain to be investigated. In addition, it is unclear how these transcript variants would correspond to different fusion junctions at the genomic DNA level. The observation that T1/E5 coexisted with T1/E4 in most cases (Clark et al12 and this study) indicates that these two variants are presumably alternatively spliced products from a single fusion. On the other hand, T1/E4 does not appear to coexist with T1/E2, T2/E4 or T1-4/E4, suggesting that these might represent different gene fusion junctions. It is of interest that Wang et al14 proposed that the T2/E4 variant might be associated with a more aggressive phenotype of prostate cancer. This variant was only observed in one case in our series that was of Gleason's grade 7 and lymph node negative. Elucidation of the fusion junctions by long-range PCR, chromosomal walking or other cloning methodology should shed light on this issue.

Another unique feature of the TMPRSS2-ERG gene fusion is that unlike TMPRSS2-ETV1 or other gene fusions, TMPRSS2 and ERG are located 3 Mb apart on chromosome 21, and deletion instead of translocation appears to be the main mechanism of gene fusion. First demonstrated by Yoshimoto et al15 with three-color FISH, Perner and co-workers showed deletion in ∼70% of cases, similar to our observation (21/27, 81%). This finding raises the possibility that the TMPRSS2-ERG fusion event, in addition to ERG activation, is often accompanied by variable losses of intervening genes. It is currently unclear whether this microdeletion of chromosome 21q would carry additional biological significance beyond the ERG activation, a view that was recently entertained.16, 17

The finding that TMPRSS2-ETS fusion is seen in approximately half of the prostate cancer also raises the issue of whether the fusion-positive and fusion-negative cases would differ biologically, for instance, in their histomorphology, response to treatment or prognosis. Previous studies have shown no correlation to Gleason's grades, lymph node status or clinical stages,16 and Demichelis et al21 suggested that TMPRSS2-ERG fusion-positive carcinomas might represent a more aggressive phenotype. In contrast to this notion, we confirmed that the gene fusion did not correlate with the tumor's stages. Although a trend of more frequent TMPRSS2-ERG fusion was observed in moderately differentiated tumor (vs poorly differentiated tumor) in this study, this finding was not statistical significant and needs to be further evaluated. We also found carcinomas with focal visible intraluminal mucin to have a significantly higher frequency of TMPRSS2-ETS fusion. Other histologically features that we examined, for example, amphophilic cytoplasm, cribriform architecture, focal foamy gland change, etc, were not statistically different between the fusion-positive and -negative groups. Such correlations between the fusion status and morphologic features could impact on pathology practice if the presence (or absence) of gene fusion carries therapeutic or prognostic implications. Since the TMPRSS2-ETS fusions would place the ERG/ETV genes under the regulation of androgen control, one might indeed speculate that the fusion status might predict the response to hormonal treatment, and studies to address this possible correlation, such as the recent in vitro study by Hermans et al,22 would clearly be clinically important.