Discovery of non-ETS gene fusions in human prostate cancer using next-generation RNA sequencing

Validation of candidates in the read-through and cis categories

To begin with, we sought to validate selected read-through and cis candidate chimeras. Nine of the 11 highest scoring read-through chimeras were validated by RT-PCR (Supplemental Table S3). Read-through transcripts form from similarly oriented neighboring genes. While the mechanism is unknown, evidence from several studies using high-resolution oligonucleotide SNP array analysis suggest that there is neither gross genomic rearrangement nor loss of intervening genomic DNA required to the formation of this specific class of fusion transcripts (for review, see Gingeras 2009). Our results suggest that read-through events are not cancer-restricted, as several positively validated chimeric read-through transcripts were nominated in both prostate cancer and benign prostate tissues (Supplemental Table S3) including the previously described ZNF649–ZNF577 and SLC45A3–ELK4 read-through events (Maher et al. 2009a; Rickman et al. 2009). Although this analysis does not completely rule out possible contamination by epithelial cancer cells in the benign tissue, our data provide evidence for an extended complexity of the cellular transcriptome and support a broader definition of a gene that might include utilization of exons from neighboring genes (Gerstein et al. 2007; Gingeras 2009). It is still unknown whether the formation of read-through transcripts is part of a normal biological activity or if it rather defines a cancer-related process that could appear in adjacent benign tissue of a cancerous prostate. None of the two cis chimera candidates were validated by RT-PCR, suggesting that they might be nominated due to an incomplete gene annotation, as exemplified by PICK1–SLC16A8 (Supplemental material; Supplemental Fig. S2).

Identification of novel gene fusions in prostate cancer

Seven high-scoring inter/intrachromosomal cancer-specific chimeric candidates were nominated for experimental validation (Fig. 1B,C). From two TMPRSS2–ERG fusion negative tumor samples (STID3023_B62 and STID2620_D), we identified and validated two novel gene fusions involving ETS family members, KLK2–ETV1 and FKBP5–ERG (Fig. 2; Supplemental Fig. S3; Supplemental material). Both of these 5′ partners are known androgen-regulated genes (Makkonen et al. 2009), and KLK2 has been reported as a 5′ gene fusion partner to ETV4 (Hermans et al. 2008). To our knowledge, this is the first description of a KLK2–ETV1 fusion event. At the transcript level, the fusion was successfully validated using RT-PCR followed by Sanger sequencing (Fig. 2B). We confirmed the KLK2–ETV1 rearrangement at the DNA level by FISH break-apart assays (Fig. 2C). In the second case, empirical validation of fusions involving FKBP5 led to the discovery of a complex triple fusion event with FKBP5 joined to TMPRSS2 and ERG (Supplemental Fig. S3). Interestingly, two novel fusion candidates (CDKN1A–CD9 and TNPO1–IKBKB) were nominated in a tumor sample that was also TMPRSS2–ERG gene fusion positive (STID2858_C) (Figs. 3, ​,4).4). Both fusions were validated by RT-PCR (Figs. 3B, ​,4B),4B), and the presence of genomic rearrangements was confirmed by FISH assays (Figs. 3C, ​,4C).4C). The gene fusions only occur in the cancer focus, not in the adjacent benign tissue. The fusion event between CDKN1A and CD9 appears to influence the expression of CDKN1A, which is low in this sample (Fig. 3D). Evaluation of CD9 protein expression in the tissue derived from this patient showed partial loss of plasma membrane staining in prostate cancer cells as compared to adjacent benign prostate epithelial cells or to a prostate cancer without the CDKN1A–CD9 rearrangement (Fig. 3E). Examination of the CDKN1A–CD9 fusion open reading frame (ORF) sequence reveals that the fusion brings into frame a downstream initiating methionine of the CD9 and therefore is predicted to produce an N-terminally truncated CD9 that lacks one transmembrane domain and the small extracellular loop (Fig. 3F). To determine the basis underlying altered expression of CD9, we transiently transfected HEK293 cells with either wild-type (WT)–CD9 or CDKN1A–CD9 fusion expression vector and examined the expression of the protein products. Western blot analysis indicated lower protein expression from the CDKN1A–CD9 fusion construct in HEK293 cells as compared with the WT–CD9 construct (Supplemental Fig. S4A). This is most likely due to the utilization of an AUG start site that has a weaker Kozak consensus sequence than the one used to produce WT–CD9. Immunofluorescence staining revealed that in contrast to WT–CD9, truncated CD9 exhibits weak to absent membranous expression (Fig. 3G). In a prostate model (benign prostate epithelial RWPE), we confirmed the lack of membranous expression in cells transfected with truncated CD9 (Supplemental Fig. S4B). The truncated CD9 protein was unable to transform cells in foci formation assays using NIH3T3 cells (data not shown). We screened various prostate cancer cell lines for endogenous expression of CD9 (Fig. 3H). The lowest expression was found in DU145 cells. The androgen receptor (AR) negative DU145 line is derived from a brain metastasis and represents an aggressive stage of prostate cancer (van Bokhoven et al. 2003). DU145 is normally invasive in vitro, and, therefore, we set out to further characterize the contribution of CD9 loss to the induction of invasion. Stably reintroducing high WT–CD9 levels in DU145 resulted in a significant reduction in the invasive behavior of DU145 cells (Fig. 3I–K).

Open in a separate window

Figure 2.

Identification of KLK2 as novel 5′ fusion partner of ETV1. (A) Circos plot of KLK2–ETV1 rearrangement. Outer ring: (purple) chromosome 19; (pink) chromosome 7. Inner ring: (blue) genes; (green) exons. Within the inner ring, lines denote PE reads with both reads belonging to KLK2 (gray) and reads connecting ETV1 and KLK2 (red). (B) RT-PCR and Sanger sequencing of the resulting fusion transcript reveals the expression of two different fusion transcripts. PCR products were sequenced using the reverse primer, so sequence traces are given in reverse orientation. (C) ETV1 and the KLK locus are rearranged as determined by FISH break-apart assays. Of note, KLK2-specific FISH is not feasible due to the small size of the KLK2 gene and its close location to neighboring genes of the KLK locus. To address this, genomic rearrangement of the KLK2 gene is inferred from a rearrangement in the genomic locus of the KLK gene family.

Open in a separate window

Figure 3.

Characterization of the CDKN1A–CD9 gene fusion candidate. (A) Mapping of PE RNA-seq reads to CDKN1A with either both ends mapping to CDKN1A (gray) or one end mapping to CD9 (red) and thereby connecting CDKN1A exon 1 to 3′ exons of CD9. (B) Experimental validation of the gene fusion transcript by RT-PCR and subsequent Sanger sequencing. (C) FISH validation of the CDKN1A–CD9 fusion in the index case (right) but not in another control cancer (left). (D) The fusion positive prostate cancer sample (black) has the lowest CDKN1A expression levels across 25 prostate cancers. Expression levels are indicated in reads per kilobase of exon model per million mapped reads (RPKM) as defined previously (Mortazavi et al. 2008). (E) Immunodetection of CD9 by immunohistochemistry. (Left) Strong CD9 membranous expression detected in malignant glands from a prostate cancer sample without a detectable CD9 rearrangement. (Right) Malignant glands in CDKN1A–CD9 fusion positive case (arrow) showing weak membranous expression in comparison to adjacent benign areas (arrowhead). (F) Schematic illustration of Flag-tagged WT–CD9 and Flag-tagged CDKN1A–CD9 fusion protein. The fusion event leads to a truncated CD9 protein with loss of two transmembrane domains and the small extracellular domain (EC1). (Bottom) The amino acid sequence of WT–CD9 protein with the truncated CD9 protein version underlined. (G) Immunofluorescence staining revealed the expression of WT–CD9 and truncated CD9 in HEK293 cells stained with anti-Flag (red) antibody. Nuclei were stained with DAPI (blue). (H) Expression levels of CD9 mRNA transcripts in prostate cell lines. (I) CD9 expression from control or stably CD9-expressing DU145 cells. (J) Bar graph comparing the invasiveness of the two lines as assessed by Boyden chamber assays. (K) Representative images of invaded cells for each cell line.

Open in a separate window

Figure 4.

Characterization of the TNPO1–IKBKB gene fusion candidate. (A) Mapping of PE RNA-seq reads to TNPO1 with either or both ends mapping to TNPO1 (gray) or one end mapping to IKBKB (red). (B) Experimental validation by RT-PCR and Sanger sequencing verifies the expression of a TNPO1–IKBKB fusion transcript. (C) Fusion specific FISH assays confirm the existence of TNPO1–IKBKB fusion in cancer tissue (right) but not in adjacent benign tissue (left). (D) IKBKB expression levels in a set of 25 prostate cancers. IKBKB levels are highest in the fusion positive sample (black). (E) Dose response curve assessing the effect of the IKK-beta inhibitor BMS-345541 on viability of LNCaP and 22Rv1 prostate cancer cells. (F) Immunoblots depicting the effect of BMS-345541 exposure on RelA phosphorylation in LNCaP and 22Rv1 cells.

We also detected a fusion placing IKBKB (IKK-beta) next to TNPO1 (transportin 1) regulatory sequences (Fig. 4). The presence of the TNPO1–IKBKB fusion was associated with an IKK-beta gene expression level more than ninefold higher than the median expression level of other prostate cancer samples (Fig. 4D). The fusion gene maintains the native translation start site in exon 2 of IKBKB and therefore is predicted to encode for wild-type IKK-beta (the predicted ORF resulting from the fusion is provided in the Supplemental material). We examined whether IKK-beta expression is essential to maintain the viability of human prostate cancer cells LNCaP and 22Rv1 by incubating them with BMS-345541, a potent selective allosteric site-binding inhibitor of IKK-beta (Burke et al. 2003). Within 72 h, cell viability in the BMS-345541-treated cells was significantly compromised compared to vehicle-treated cultures (Fig. 4E). Apoptotic cells were manifest as early as 24 h following treatment at concentrations of 1 μM and 5 μM, respectively (data not shown). Within 14 h, reduced levels of phospho-RelA were observed indicating that BMS-345541 can inhibit NFKB signaling in these two prostate cancer cell lines through its ability to antagonize IKK-beta activity (Fig. 4F).

Interestingly, another potential oncogene, PIGU, was nominated in one case harboring TMPRSS2–ERG rearrangement (STID580_Ba). Paired-end reads connecting PIGU (chr 20) and ALG5 (chr 13) (Fig. 5A) indicated the presence of a reciprocal balanced translocation event between ALG5 and PIGU giving rise to two novel gene fusions confirmed by RT-PCR and FISH (Fig. 5B,C). As none of the common prostate cancer models (i.e., LNCaP, DU145, VCaP, 22Rv1) harbor this translocation (as determined by RT-PCR), we transiently expressed ALG5–PIGU and PIGU–ALG5 in HEK293 cells. While both mRNA transcripts were detectable (Fig. 5D), only the ALG5–PIGU fusion protein is produced with a molecular weight of <50 kDa (Fig. 5E; Supplemental Fig. S5A). A sequence-based prediction of protein domains further indicates that the functional domain of PIGU is maintained in the ALG5–PIGU fusion protein (Supplemental material). The expression of PIGU in the tumor with this fusion event did not, however, appear significantly different from other samples that did not harbor the fusion (Supplemental Fig. S5B). In vitro experiments using siRNA (Fig. 5F,G; Supplemental Fig. S5C) demonstrate that PIGU expression is required for maintaining anchorage-independent growth and proliferative capacity of prostate cancer cells. Since a previous PIGU translocation was described as a germline event in a patient with bladder cancer (Guo et al. 2004), we also examined this case for a similar germline alteration by using FISH, but no alterations were observed in normal adjacent prostate tissue (Supplemental Fig. S5D).

Open in a separate window

Figure 5.

Characterization of the reciprocal balanced translocation event involving PIGU and ALG5. (A) Circos plot of ALG5–PIGU rearrangement. In the outer ring, (gray) chromosome 20 and (blue) chromosome 13. Genes are represented in purple (PIGU) and orange (ALG5) on the inner ring. Within the inner ring, each line denotes paired-end reads with either both ends mapping to PIGU (gray) or one end mapping to PIGU and the other end to ALG5 (red). The red lines connect 5′ ALG5 to 3′ PIGU and vice versa, thereby indicating a balanced translocation event. (B) Experimental validation of the two resulting gene fusions by RT-PCR and subsequent Sanger sequencing of the resulting PCR products. Two different primer pairs were used for verification of each gene fusion transcript indicated as 1 and 2. (C) FISH validation of the ALG5–PIGU fusion in the index case by break-apart assays (left and middle) and a fusion assay (right). (D) Expression of ALG5–PIGU and PIGU–ALG5 messages, as determined by quantitative PCR following transfection of the indicated constructs in HEK293 cells. (E) Immunoblot analysis on the transfected HEK293 cells showing protein expression only in ALG5–PIGU transfected cells. (F,G) LNCaP cells were treated with nontargeting (NT) siRNAs or siRNAs against PIGU to assay for effect of anchorage-independent growth in soft agar. (F) Bar graph showing expression of PIGU after siRNA treatment. (G) LNCaP cells treated with PIGU siRNAs show reduced colony formation ability.

Interestingly, none of the newly identified 5′ gene fusion partners (i.e., ALG5, PIGU, CDKN1A, and TNPO1) are androgen-regulated (Supplemental Fig. S6) in contrast to the most common 5′ prostate cancer fusion partners (i.e., TMPRSS2, SLC45A3, and NDRG1). Finally, we detected a fusion between MIER2 and RSRC2 in a third TMPRSS2–ERG fusion positive sample (STID581_Dt). RSRC2 has been proposed previously as a tumor suppressor gene in the setting of esophageal cancer (Kurehara et al. 2007). RT-PCR validation and subsequent Sanger sequencing revealed a fusion transcript connecting exon 3 of MIER2 with exon 11 of RSRC2 (Supplemental Fig. S7A). The predicted open reading frame resulting from the MIER2–RSRC2 fusion encodes a 114-amino-acid fusion protein that might retain 81 amino acids of the N-terminal portion of MIER2 and 59 amino acids from the small C-terminal portion of RSRC2 (Supplemental Fig. S7B). Thus, this fusion likely impairs the proper function of these two genes. In addition, examination of mRNA expression levels showed that this sample had relatively low levels of RSRC2 and MIER2 (Supplemental Fig. S7C,D). We screened additional prostate cancer samples by FISH on tissue microarrays for the presence of rearrangements involving CDKN1A (n = 87), CD9 (n = 77), IKBKB (n = 83), or PIGU (n = 74). In all, we identified one additional case displaying CD9 rearrangement and one with IKBKB rearrangement. Interestingly, these two fusions occur in ETS rearrangement positive cases. Neither CDKN1A nor TNPO1 were the respective gene fusion partners in these cases. The lack of availability of frozen material for this sample prevented us from nominating any potential 5′ fusion partners. However, these data suggest that, as exemplified by the plethora of ETV1 gene fusion partners (Tomlins et al. 2007), there may be additional unknown 5′ fusion partners to CD9 and IKBKB. In another set of 110 prostate cancer cases from the University of Michigan, we failed to identify additional instances of the novel fusions presented here.

RNA-seq sample prep and sequencing

Total RNA was prepared in accordance with Illumina's sample preparation protocol for PE sequencing of mRNA unless described otherwise. In brief, 5–10 μg of total RNA was fragmented by heat between 2 and 3 min based on the desired insert size, reverse-transcribed using Superscript II (Invitrogen), and transformed to double-stranded cDNA. To improve PE RNA-seq data quality, we introduced an additional gel-based size selection step after cDNA double-strand synthesis and before the ligation of the PE adapters. This was postulated by Quail et al. (2008) as a means to reduce the inclusion of artifactual chimeric transcripts that are composed of two cDNA fragments into the sequencing library. We also integrated the use of T4 ligase (Enzymatics Inc.) to improve the efficiency of adapter ligation. Over the course of the study, we increased the library size range from 250 bp to 450 bp. The gel dissolutions of all gel-based purification steps were conducted at room temperature under slight agitation as described by Quail et al. (2008). After the enrichment of cDNA template by PCR, the concentrations and the sizes of the libraries were measured using a Qubit fluorometer (Invitrogen) and DNA 1000 kit (Agilent Technologies) on an Agilent 2100 Bioanalyzer, respectively. PE RNA-seq was performed with the Genome Analyzer II (Illumina) increasing the read size of the PE reads from 36 to 54 bp over the course of the study. Additionally, Illumina introduced improved sequencing reagents and upgraded imaging software over time to increase data quality and sequencing coverage.

Processing of RNA-seq data through FusionSeq

PE reads were aligned to the human genome (hg18) using ELAND, part of the standard software suite from Illumina. PE reads that mapped with only up to two mismatches were then processed through FusionSeq. In brief, FusionSeq consists of a compendium of modules that:

1. Classify the reads in groups depending on if the respective PE reads belong to the same gene, to different genes, or if they are not mappable.
2. Filter all the chimeric PE reads, that is, those where the two reads map to different genes, in order to discard artifactual candidates such as those involving homologous genes as well as those joining highly expressed genes, such as ribosomal, that might have been generated from sample prep. It also filters candidates that may result from PCR artifacts as well as gene annotation issues. This module also categorizes candidates and attributes various scores in order to sort and prioritize them for experimental follow-up.
3. Identify the sequence of the junction between the mRNAs coming from the two different genes with the aim to find overall support for the chimeric transcript.

Validation experiments by RT-PCR

PCR validation has been performed using primers outside of the region that is covered by PE reads to introduce a second step of quality control for the further reduction of the possibility of detecting false-positive chimeric transcripts due to a regional high homology between partner genes. A second round of PCR, using another primer pair, was employed if the results from the first primer pair were inconclusive (Supplemental Table S6).

Validation experiments by FISH

All FISH experiments were performed as described (Tomlins et al. 2005; Perner et al. 2006) on the same cancer focus that was subjected to RNA extraction for RNA-seq and PCRs. In a FISH break-apart assay, a nucleus without gene rearrangement shows two pairs of red–green signals, often forming two yellow signals, indicating two intact alleles. A nucleus with gene rearrangement shows the split of one yellow signal into two distinct red and green signals for the rearranged allele and one remaining juxtaposed red–green (yellow) signal indicating the intact allele. In a FISH fusion assay, a nucleus with fusion shows one red and one green distinct signal coming together to form a yellow signal and indicating the fusion allele. The FISH probes are listed in Supplemental Table S7. For screening analysis, FISH experiments were carried out on two tissue microarrays (TMA) containing 41 and 47 samples, respectively.

Immunohistochemistry

Immunohistochemical staining of CD9 (clone 72F6, 1:250 dilution; Abcam) was performed using the Bond Max Autostainer (Leica Microsystems). Formalin-fixed, paraffin-embedded tissue sections were deparaffinized, and endogenous peroxidase was inactivated. Antigen retrieval was accomplished using the Bond Epitope Retrieval Solution 2 (ER2) for 20 min at 99°C–100°C (Leica Microsystems). Following retrieval, the sections were incubated sequentially with the primary antibody for 25 min, post-primary for 15 min, and polymer for 25 min ending with colorimetric development with diaminobenzidine (DAB) for 10 min (Bond Polymer Refine Detection; Leica Microsystems).

We thank S. Pond and Y. Liu for outstanding technical assistance in performing RNA-seq and IHC, respectively. We thank D. Chen for building the wrapper for Circos and J. Rozowsky for contributing to the abnormal insert size concept of FusionSeq. We thank H. Beltran, M. Bacharach, and K. Park for technical assistance and advice, and R. Kim, R. Leung, and the translational research program for providing high-quality prostate samples. The Yale University High Performance Computing Center and National Institutes of Health funded the computer cluster instrumentation where the FusionSeq analysis was performed. This work was supported by the National Cancer Institute R01-CA116337 and RO1-CA125612 (F.D., M.A.R.); the Prostate Cancer Foundation (M.A.R.); the Heinrich-Warner-Foundation (D.P.); the National Institutes of Health/National Human Genome Research Institute R44HG004237 (M.S.C.); and by a grant from the Spanish Government “Rio Hortega” record no. GMO8/2006 (R.E.). A.M.C., F.D., and M.A.R. are coinventors on a patent filed by the University of Michigan, Ann Arbor and the Brigham and Women's Hospital, Boston covering the diagnostic and therapeutic fields of ETS gene fusions in prostate cancer. The diagnostic field has been licensed to Gen-Probe Inc. and sublicensed to Ventana/Roche. Gen-Probe and Ventana/Roche have not played a role in the design and conduct of the study; in the collection, analysis or interpretation of the data; or in the preparation, review, or approval of the article.