Summary
One of the hallmarks of cancer is the formation of oncogenic fusion genes as a result of chromosomal translocations. Fusion genes are presumed to occur prior to fusion RNA expression. However, studies have reported the presence of fusion RNAs in individuals who were negative for chromosomal translocations. These observations give rise to “the cart before the horse” hypothesis, in which fusion RNA precedes the fusion gene and guides the genomic rearrangements that ultimately result in gene fusions. Yet RNA-mediated genomic rearrangement in mammalian cells has never been demonstrated. Here we provide evidence that expression of a chimeric RNA drives formation of a specified gene fusion via genomic rearrangement in mammalian cells. The process is (1) specified by the sequence of chimeric RNA involved, (2) facilitated by physiological hormone levels, (3) permissible regardless of intra-chromosomal (TMPRSS2-ERG) or inter-chromosomal (TMPRSS2-ETV1) fusion, and (4) can occur in normal cells prior to malignant transformation. We demonstrate that, contrary to “the cart before the horse” model, it is the antisense rather than sense chimeric RNAs that effectively drive gene fusion, and that this disparity can be explained by transcriptional conflict. Furthermore, we identified an endogenous RNA AZI1 that acts as the ‘initiator’ RNA to induce TMPRSS2-ERG fusion. RNA-driven gene fusion demonstrated in this report provides important insight in early disease mechanism, and could have fundamental implications in the biology of mammalian genome stability, as well as gene editing technology via mechanisms native to mammalian cells.
Introduction
Fusion genes are among the most cancer-specific molecular signatures known. They are important for understanding cancer mechanisms and developing useful clinical biomarkers and anti-cancer therapies (Mitelman et al., 2007). Fusion gene formation as a result of chromosomal translocations is presumed to occur prior to fusion RNA expression. However, several studies have reported the presence of fusion transcripts in individuals without detectable fusion genes at the genomic DNA level. For instance, the AML1-ETO fusion transcript, associated with a subtype of acute myeloid leukemia, was present in some patients who were negative for chromosomal translocations (Langabeer et al., 1997). Other fusion RNAs, such as BCR-ABL, MLL-AF4, TEL-AML1, PML-RARα, and NPM-ALK, were reported in healthy individuals (Janz et al., 2003). Although the discrepancy between the presence of fusion transcripts and the absence of fusion genes could result from detection limitations of the methodologies employed, fusion transcripts in normal cells could also arise from RNA trans-splicing in the absence of chromosomal translocations (Zaphiropoulos, 2011). Indeed, JAZF1-JJAZ1 fusion transcripts are expressed in normal human endometrial tissue and an endometrial cell line in the absence of chromosomal translocation (Li et al., 2008). Furthermore, trans-splicing between JAZF1 and JJAZ1 was demonstrated to occur in vitro using cellular extracts, resulting in a fusion RNA similar to that transcribed from the JAZF1-JJAZ1 fusion gene in endometrial stromal sarcomas (Li et al., 2008). These observations raise the possibility that cellular fusion RNAs created by trans-splicing act as guide RNAs to mediate genomic rearrangements. A precedent for RNA-mediated genomic arrangements is found in lower organisms such as ciliates (Fang and Landweber, 2012; Nowacki et al., 2008). Rowley and Blumenthal (Rowley and Blumenthal, 2008) coined this as “the cart before the horse” hypothesis, in that “RNA before DNA” defies the normal order of the central dogma of biology: DNA → RNA → protein (Crick, 1970). Despite important implications in biology and human cancer, RNA-mediated genomic rearrangement in mammalian cells has not been directly demonstrated. In this report, we provide the first evidence that expression of a specific chimeric RNA can lead to specified gene fusion in mammalian cells.
Results
To test whether the expression of a fusion RNA in mammalian cells can lead to a specific gene fusion, the TMPRSS2-ERG fusion (Perner et al., 2006; Tomlins et al., 2005), found in ~50% of prostate cancers, was selected as a model. Both the TMPRSS2 and ERG genes are located on chromosome 21, an intra-chromosomal configuration prone to rearrangements. To recapitulate TMPRSS2-ERG fusion gene formation, we used the LNCaP prostate cancer cell line that lacks the TMPRSS2-ERG fusion (Horoszewicz et al., 1980; Tomlins et al., 2005). Furthermore, treating LNCaP cells with androgen increases the chromosomal proximity between the TMPRSS2 and ERG genes (Bastus et al., 2010; Lin et al., 2009; Mani et al., 2009), which was thought to increase the possibility of gene fusion. To test “the cart before the horse” hypothesis (Rowley and Blumenthal, 2008; Zaphiropoulos, 2011), we transiently expressed a short fusion RNA consisting of two exons, TMPRSS2 exon-1 joined to ERG exon-4, which is a short fragment of a full-length TMPRSS2-ERG fusion RNA that is most common in prostate cancer (Fig. 1A, upper panel). This short fusion RNA mirrors the presumptive trans-spliced fusion RNA product that is generated only in the sense orientation because the correct splice sites are absent in the antisense orientation. However, because the ‘antisense’ sequence should, in theory, contain the same template information for guiding genomic rearrangements, we tested both the sense and antisense short fusion RNA. Each was individually expressed using either a CMV or a U6 promoter (Fig. 1A, upper panel) and designated as ‘input RNA’ to distinguish them from the ‘endogenous’ full-length fusion RNA transcribed from the genome.
We transiently transfected LNCaP cells with either plasmid and treated the cells with dihydrotestosterone (DHT, a metabolite of testosterone) for 3 days. If the expression of an input RNA leads to a TMPRSS2-ERG gene fusion, it is expected that the endogenous full-length fusion RNAs would be transcribed from the newly induced fusion gene. Specific RT-PCR assays were designed to distingwish between endogenous full-length fusion RNAs and the input RNAs exogenously expressed from the plasmids (see Fig. S1A for primer designs). As shown in Fig. 1A, expression of the sense short fusion RNA resembling the trans-spliced product, either by the CMV or U6 promoter (Fig. 1A lower panel, lane 1 and 2 respectively), led to no detection of an induced endogenous fusion transcript. Expression of a longer version of sense fusion RNA consisting of four exons (TMPRSS2 exon-1 joined to ERG exon-4/5/6) also failed to induce the endogenous fusion transcript (Fig. S2). In contrast, expression of antisense short fusion RNAs induced a clear band of 721 bp (Fig. 1A lower panel, lane 3 and 4). Sanger sequencing revealed that the induced band contains TMPRSS2 exon-1 fused to ERG exons-4/5/6/7 (Fig. S3), and that the exons are joined by annotated splice sites, which would be expected of mature endogenous fusion mRNA derived from the TMPRSS2-ERG fusion gene. This induced fusion transcripts cannot possibly arise from the sequence of input RNAs as the expression plasmids used contain only TMPRSS2 exon-1 and ERG exon-4 without the ERG exon-5/6/7 sequence. Notably, the induction was more pronounced when the antisense input RNA was driven by the U6 promoter (Fig. 1A lower panel, antisense-2, lane 4) compared to the CMV promoter (antisense-1, lane 3). These differences (antisense vs. sense, U6 vs. CMV) are not caused by differing amounts of input RNA because all input RNAs were expressed at relatively equal levels (Fig. 1A, lower panel). Transfection with a parental plasmid containing mCherry sequence (Fig. 1A, lane 5), DHT treatment without plasmid transfection (Fig. 1A, lane 6), and PCR reaction without cDNA served as RT-PCR controls (Fig. 1A, lane 8), all resulted in no endogenous fusion transcript. In addition, all experiments were performed independently at least four times and results were identical. Together, the data suggest that expression of an input RNA with chimeric sequence can lead to the induction of a specified endogenous fusion transcript in human cells. Surprisingly, the antisense, rather than the sense version of input RNA, exhibits the capacity of induction.
Antisense input RNAs described above contain 218 nt against the entire ERG exon-4 and 78 nt against the entire TMPRSS2 exon-1 (Fig. 1A), suggesting that 78 nt is sufficient to specify a parental gene for gene fusion. Furthermore, because the effective input RNAs are of the ‘antisense’ orientation, the data imply that the input RNAs may not require an RNA junction resembling that of the TMPRSS2-ERG fusion transcript generated by splicing in the sense orientation. To further analyze the sequence requirement, we used the U6 promoter to express a series of antisense input RNAs with 75 nt complementary to ERG exon-4 joined to various segments (33, 52, 67, 82 nt) that are complementary to TMPRSS2 near the exon-1/intron-1 boundary (Fig. 1B). A parallel set of sense input RNAs were also tested as controls (Fig. 1C). As shown in Fig. 1B (lower panel), all antisense RNAs, with the exception of antisense-3, induced fusion transcripts even though their target regions span the exon/intron boundary. The level of induction peaked for antisense-5, which contains 52 nt designed to anneal with TMPRSS2, suggesting that this number of nucleotides might be optimal to specify a parental gene for induction. The results were confirmed using a different, but more efficient, primer pair (Fig. 1B, lower panel; primer design in Fig. S1B) followed by Sanger sequencing of the induced band (Fig. S4). In contrast to the antisense input RNA, all corresponding sense input RNAs failed to induce endogenous fusion transcripts (Fig. 1C, lower panel). This was true even when the sense input RNA was intentionally expressed at a much higher level than the antisense RNA (Fig. S5). Additional experiments using plasmids with a severed U6 promoter (Fig. S6), to eliminate input RNA expression, also confirmed that it is the antisense input RNAs expressed from plasmids, not the DNA sequence of plasmids, that induce the observed TMPRSS2-ERG fusion transcripts.
As shown in Fig. 1D, the amount of endogenous fusion transcript induced by antisense-5 (the most effective antisense input RNA) appears to correlate with the concentration of DHT used, presumably because the hormone increases the chromosomal proximity between the TMPRSS2 and ERG genes (Bastus et al., 2010; Lin et al., 2009; Mani et al., 2009). Antisense-5 was effective at DHT concentrations as low as 20 nM as revealed by sensitive nested PCR (Fig. 1D lane 3), indicating that fusion events induced by input RNA can occur under physiologically relevant androgen conditions (Boyce et al., 2004). As a control, DHT treatment alone up to 2 μM failed to induce fusion (Fig. S7). Titration of DHT showed that the induction by antisense-5 reaches the 50% of maximal level (EC50) at 0.9 μM DHT (Fig. S7). Under this standard EC50 condition, we estimated that the percentage of LNCaP cells induced by antisense-5 to express the TMPRSS2-ERG fusion transcript is approximately 1 in 103 or 104 cells (see assay in Fig. S8). Together, these results demonstrated that the induction by input RNA can occur at physiologically relevant hormone levels, but does not represent a high frequency event.
Although induced fusion is infrequent, all antisense RNAs described in Fig. 1B successfully induced endogenous fusion RNA except antisense-3, which is only 30 nt longer than antisense-5 in the arm targeting TMPRSS2 intron-1 (Fig. 1B, upper panel). To test whether its inability to induce was due to input RNA length or the specific target sequence in TMPRSS2 intron-1, we constructed a hybrid antisense (antisense-7) that shifted the 52 nt recognition window of antisense-5 to target the TMPRSS2 intron-1 region covered by antisense-3 (Fig. 1E). This alteration resulted in the loss of induction (Fig. 1E, lane 1 vs. 3), implying that the inability of antisense-3 to induce is not reflective of input RNA length. Rather its targeting arm may interfere with a motif important for the fusion process. BLAST alignment of the genomic DNA sequence revealed an imperfect stem (named stem A) potentially formed by the sense genomic TMPRSS2 sequence complementary to the sense genomic ERG sequence (Fig. 2A, left). We reasoned that this genomic DNA stem (Tm= ~44°C) could potentially stabilize a three-way junction that involves an RNA/DNA duplex formed by the antisense-5 RNA and its targeted genomic DNA in a sequence-specific manner (Fig. 2A, left). If correct, then the formation of this putative three-way junction would be disrupted by antisense-3 because its recognition sequence invades the genomic DNA stem (Fig. 2A, left). Consistent with the idea that induction requires bringing TMPRSS2 and ERG gene in close proximity, expression of antisense-5 as two separate halves (Fig. 2A right panel, antisense-5A and -5B) severed the link between TMPRSS2 (52 nt) and ERG (75 nt) sequence within the input RNA, resulting in the loss of induction (Fig. 2B, lane 1 to 3).
To test whether the proposed three-way junction formation could facilitate induction, we used BLAST alignment to identify several intron locations where genomic DNA stems could be formed by the sense genomic TMPRSS2 sequence paired with the sense genomic ERG sequence (stems B to G in Fig. 2C and 2D; genomic coordinates in Fig. S9; sequences flanking the stems in Fig. S10). Matching antisense input RNAs (termed antisense-B1 to G1) were then designed to facilitate the formation of a three-way junction with the possible intron stems (Fig. 2D) that would mirror the three-way junction formed by antisense-5 on stem A as postulated in Fig 2A. Because these input RNAs target the introns (Fig. 2C) and contain no exon sequence, any observed induction of endogenous fusion transcripts composed of exons cannot arise from the sequence of input RNAs or plasmids used to express them. As shown in Fig. 2E, targeting genomic DNA stem B, C, and D that exhibit higher DNA stem stability (Tm= 40°C, 40°C, and 44°C, respectively) by the corresponding antisense input RNAs clearly induced fusion transcripts (Fig. 2E, lanes 2 to 4). In contrast, targeting less a stable stem E, F, and G (Tm= 30°C, 24°C, and 16°C, respectively) failed to induce fusion transcripts (Fig. 2E, lanes 5 to 7). To disrupt the three-way junction invovling stem B, C, and D, six additional antisense RNAs (antisense-B2, B3, C2, C3, D2, and D3) were designed with one side of their recognition sequence altered to invade each of the respective genomic DNA stems on the TMPRSS2 side or the ERG side (Fig. S11). These modifications were chosen to mirror the interference on stem A by antisense-3. Similar antisense RNAs were also designed to invade stem A (Fig. S12). In all cases, invasion of the genomic DNA stems by the modified input RNAs resulted in the complete loss of induction (Fig. 2F). While these results by no means necessitate that a three-way junction is required for fusion transcript induction, they nevertheless suggest that such transiently stabilized structures may ‘facilitate’ the process and could have important implications in developing gene editing technologies via mechanisms native to mammalian cells. Consistent with earlier observations, the corresponding sense version of the effective antisense input RNAs (sense-B1, C1, D1) all failed to induce fusion transcripts (Fig. 2G, lanes 2 to 4).
The fact that antisense input RNAs, but not their sense counterparts, induce fusion transcripts, raises the possibility that the former act as a docking station to mediate trans-splicing between endogenous sense TMPRSS2 and ERG pre-mRNAs. Because that the antisense, but not the sense input RNAs, are complimentary to both sense TMPRSS2 and ERG pre-mRNAs, they can base-pare with both parental pre-mRNAs, thus resulting in spliced fusion transcripts without the requirement of genomic rearrangement. This mechanism, however, is unlikely as the major contributor to the observed induction for the following reasons. First, although TMPRSS2 is expressed in LNCaP cells (Fig. 3A, upper panel), endogenous ERG mRNA is not detected in LNCaP cells (Tomlins et al., 2005) in the presence or absence of DHT or before and after transfection of antisense-5 (Fig. 3A, middle and lower panel with different primer pairs). In fact, parental ERG mRNA was not detected even using three rounds of nested RT-PCR using various primer sets (Fig. S13). Therefore prior to and during induction, no or an insufficient number of parental ERG mRNAs are available in LNCaP cells as raw material for trans-splicing to account for the level of induced fusion transcript. Second, after initial transient transfection and DHT treatment for 3 days, we continued to propagate and enrich the induced LNCaP population for 52 days in the absence of DHT (experimental procedures described in Fig. S14). As shown in the lower panel of Fig. 3B, antisense-5 RNA transiently expressed by plasmids was degraded and completely absent beyond day 17. In contrast, the induced fusion transcript was continuously expressed and enriched up to day 52 in the absence of antisense input RNA and DHT (Fig. 3B upper panel), indicating the persistent nature of the induced fusion product. Taken together, these results strongly suggest that the induced expression of the TMPRSS2-ERG fusion transcript is the consequence of gene fusion at the DNA level, which has a permanent nature. This is in contrast to the result of induced trans-splicing at the RNA level mediated by antisense input RNA, which is transient and requires the continuous presence of input RNAs.
To provide definite evidence of gene fusion via genomic rearrangement, we used genomic PCR to identify the genomic breakpoint induced by antisense-5 in the enriched LNCaP population (primer designs in Fig. S15A and Fig. 3C). As shown in Fig. 3D, the un-rearranged wildtype TMPRSS2 and ERG alleles were amplified by gene-specific primer pair A/B and C/D both in untransfected cells (Fig. 3D, lane 1 and 2) and enriched LNCaP cells (Fig. 3D, lane 4 and 5). In contrast, a genomic fusion band of ~862 bp amplified by fusion-specific primer pair A/D was present only in the enriched LNCaP population (Fig. 3D, lane 6) and absent in untransfected LNCaP cells (lane 3). Sanger sequencing of the excised fusion band (Fig. 3D, lane 6) revealed the exact genomic breakpoint located within TMPRSS2 intron-1 (chr21:41502038, GRCh38/hg38) and ERG intron-3 (chr21:38501207, GRCh38/hg38) (Fig. 3E; full-length Sanger sequence shown in S16). Intriguingly, within TMPRSS2 intron-1 the induced breakpoint lies within an Alu, a transposable element known to contribute to genomic arrangements (Rudiger et al., 1995). In ERG intron-3, the breakpoint resides in a hot spot clustered with genomic breakpoints previously identified in prostate cancer patients (Fig. S15B) (Weier et al., 2013). There is no obvious sequence homology between TMPRSS2 and ERG at the genomic breakpoint except for a three nucleotide ‘CTG’ microhomology (Fig 3E and S16), suggesting that this gene fusion may be mediated by non-homologous break repair mechanisms (Lieber, 2010; Zhang et al., 2009).
To test whether antisense input RNA can cause TMPRSS2-ERG fusion in non-malignant cells prior to cancerous transformation, we performed experiments using immortalized normal prostate epithelium cells (PNT1A), that express very low levels of androgen receptors (Coll-Bastus et al., 2015). As shown in the lower panel of Fig. 3F, prolonged expression of antisense-5 for 12 days induced fusion transcripts (Sanger sequencing confirmation in Fig. S17). This induction was not due to prolonged exposure to DHT because continuous treatment of 0.9 μM DHT alone for up to 2 months resulted in no detectable fusion transcripts in PNT1A cells (Fig. 3F, lane 8). Thus, our results indicate that the induction of TMPRSS2-ERG fusion by antisense input RNA can occur in normal prostate epithelial cells prior to malignant transformation and is not restricted to the pathological cellular context of malignant cells.
To test whether an input RNA can specify a pair of genes to undergo fusion other than TMPRSS2-ERG in a sequence-specific manner, we designed a series of input RNAs to induce TMPRSS2-ETV1, an inter-chromosomal fusion gene found in approximately 1% of prostate cancers (Rubin et al., 2011; Tomlins et al., 2005). Eight antisense RNAs (Fig. S18) were designed to target different chosen regions in the introns where three-way junctions potentially can be forged between the genomic DNA and input RNAs (Fig. S18 and S19). Again, because these input RNAs target introns and contain no exon sequence, it rules out the possibility that induced endogenous fusion transcripts composed of exons arise from the sequence of input RNAs or the plasmids. As shown in Fig. 4A, targeting stem TMPRSS2-ETV1-A, which has the highest genomic DNA stem stability (Tm= 72°C) among this group, led to clear induction of the TMPRSS2-ETV1 fusion transcript (Fig. 4A, lane 1). Sanger sequencing validated that the induced transcript contains TMPRSS2 exon-1 joined with ETV1 exon-3 (uc003ssw.4) by annotated splice sites (Fig. S20). Similar to earlier observations, targeting with sense versions of input RNAs (Fig. 4B, lane 1 vs. 2), or using antisense input RNAs designed to form three-way junctions with lower genomic DNA stem stabilities (Fig. 4A, lanes 2 to 8; Fig. S18), resulted in no detectable induction. Furthermore, the input RNA designed to target TMPRSS2 and ETV1 induced TMPRSS2-ETV1 fusion but not TMPRSS2-ERG fusion (Fig. 4C, lane 2). Conversely, antisense-5 targeting TMPRSS2 and ERG induced TMPRSS2-ERG fusion but not TMPRSS2-ETV1 fusion (Fig. 4C, lane 1), indicating that fusion formation is specified by the sequence of input RNA and not secondary effects such as global genomic stability.
To verify that TMPRSS2-ETV1 as a second example of induced fusion that is indeed the consequence of genomic translocation, we propagated and enriched the induced LNCaP population for 47 days after the initial transfection of input RNA and DHT treatment (experimental procedures same as described for TMPRSS2-ERG enrichment in Fig. S14). The transiently expressed antisense input RNA had been degraded and was absent by day 47 (Fig. 4D, lane 1 vs. 2). The induced TMPRSS2-ETV1 fusion transcript, however, was continuously expressed beyond day 47 (Fig. 4D, lane 2). Once again this observation indicated that the sustained expression of an induced fusion does not require the continuous presence of input RNA. Moreover, genomic PCR assays identified three distinct genomic breakpoints between TMPRSS2 and ETV1 gene (labeled as x, y, z in Fig. 4E) that were present only in the enriched LNCaP population but absent in untransfected LNCaP cells (Fig. 4E, lane 3 vs. 6, lane 9 vs. 12). Similar to earlier observations, no obvious sequence homology between TMPRSS2 and ETV1 was observed at the genomic breakpoints except for a few nt of microhomology (Fig. 4F, and Fig. S22 to S24), indicating that the gene fusion is mediated by non-homologous break repair mechanisms (Lieber, 2010; Zhang et al., 2009).
Unlike TMPRSS2 and ERG that are located near each other on the same chromosome, TMPRSS2 and ETV1 are located on different chromosomes. Thus, gene fusion as a result of chromosomal translocation could be confirmed unequivocally by evidence of chromosomal co-localization of the latter pair. Using probes specific to TMPRSS2 and ETV1, we performed fluorescence in situ hybridization (FISH) followed by deconvolution microscopic imaging of 3301 cells from the enriched LNCaP cell population and 620 cells from the control untransfected LNCaP population. Analyses of constructed 3D images showed that approximately 0.9% of the enriched population (30 out of 3301 cells) were positive for co-localization of TMPRSS2 and ETV1 gene in the cellular nucleus (Fig. 4G, examples of constructed 3D images are shown as movies in S25). In contrast, none of the cells from the untransfected population showed co-localized FISH signals as determined by the same 3D image criteria. These results, together with the genomic breakpoints identified by genomic PCR at single base resolution (Fig. 4E & 4F), provide strong evidence that the induced expression of the TMPRSS2-ETV1 fusion transcript represents the consequence of gene fusion caused by chromosomal translocation.
One important observation emerging from our study is that all sense input RNAs failed to induce gene fusion. In particular, of the ten antisense input RNAs that were demonstrated to induce fusion (Fig. 1A, 1B, 2E, 4A), all of their corresponding sense counterparts failed to induce fusion (Fig. 1A, 1C, 2G, 4B). This specificity was oberserved despite the fact that sense input RNAs in theory can anneal to the same genomic sites targeted by their antisense counterparts and form similarly stable DNA/RNA hybrids. To test whether the disparity between antisense and sense is due to transcriptional activity of parental genes, we expressed the input RNAs by U6 (a pol-III promoter) for one day, followed by α-amanitin-mediated inhibition of pol-II transcription for various time periods to shut down parental gene transcription. α-amanitin was then removed to resume cellular transcription and the induction by sense vs. antisense input RNA were compared. As shown in Fig. 5 upper panel, the corresponding sense input RNAs that previously failed to induce fusion began to induce TMPRSS2-ERG after 12 hours of α-amanitin treatment (lane 9 and 10), and TMPRSS2-ETV1 fusion (lane 20) after 24 hours of α-amanitin treatment. This latent induction is not a property of general cellular toxicity of α-amanitin because the toxicity caused by the same treatment actually reduced the induction by antisense input RNAs (lane 1 vs. 5 for TMPRSS2-ERG, lane 11 vs. 15 for TMPRSS2-ETV1). Furthermore, the input RNAs designed to target TMPRSS2 and ERG, regardless of their sense or antisense nature, induced TMPRSS2-ERG fusion but not that of TMPRSS2-ETV1 (lane 1 to 10). Conversely, input RNAs targeting TMPRSS2 and ETV1, regardless of their directional orientation, induced TMPRSS2-ETV1 fusion but not TMPRSS2-ERG fusion (lane 11 to 20). Additional control experiments using a parental plasmid vector lacking the input RNA sequences, DHT treatment without plasmid transfection, and PCR reactions without cDNA, all induced no endogenous fusion transcript under the same α-amanitin treatment (Fig. 5, lower panel). The specificity exhibited by these experiments argues against general toxicity effects. The results suggest that the antisense versus sense disparity is largely due to transcriptional conflict.
With the plausibility of RNA-mediated gene fusion established, we then sought evidence that specific endogenous cellular RNAs can act as the ‘initiator’ to induce TMPRSS2-ERG fusion, which is found in ~50% of prostate cancers. To identify candidate cellular initiator RNAs, we analyzed an available mRNA-seq database consisting of prostate tumors and matched benign tissues (Kannan et al., 2011). However, we found no evidence of perfect endogenous antisense chimeric RNAs in which the TMPRSS2 sequence was joined to any ERG sequence by discernable 5’ and 3’ splice sites in the antisense orientation. This suggests that if endogenous initiator RNAs do exist, they might arise from unrelated genomic sources that coincidentally resemble an imperfect chimeric RNA antisense to both TMPRSS2 and ERG. In a detailed study to be presented elsewhere (manuscript in preparation), we have performed thermodynamic calculations of RNA/DNA hybrids to identify cellular RNAs with partial sequence complementarity to the TMPRSS2 and ERG genes. We identified that AZI1 mRNA (also known as CEP131) (Aoto et al., 1997; Aoto et al., 1995) could form high affinity RNA/DNA hybrids with TMPRSS2 and ERG genomic sequences. As shown in Fig. 6A, overexpressing full-length AZI1 mRNA (3619 nt, uc002jzn.1) induced the TMPRSS2-ERG fusion transcript in LNCaP cells. The induction was observed at a physiologically relevant concentration (40 nM) of DHT (Fig. 6A, lane 4). Furthermore, expression of exon16-17 of AZI1, a short 220 nt segment containing an imperfect sequence antisense to TMPRSS2 and ERG, was sufficient to induce TMPRSS2-ERG fusion (Fig. 6B, lane 2). This result suggests that the induction of gene fusion is mediated by an RNA sequence that resides in exon16-17 and requires no AZI1 protein. Consistent with previous observations that sense input RNAs are ineffective for the fusion process, the expression of exon16-17 in the antiparallel orientation also failed to induce TMPRSS2-ERG fusion (Fig. 6B, lane 3).
Discussion
In summary, this report provides the first evidence that expression of a chimeric RNA can drive the formation of gene fusions in mammalian cells. Hence, we propose that “the cart before the horse” hypothesis concerning fusion gene causation is mechanistically plausible. Our data support a model (shown in Fig. 7) where the initiator RNA with chimeric sequence invades chromosomal DNA to stabilize a transient RNA/DNA duplex using DNA sequences located in two distant genes. Resolution of such an RNA/DNA duplex by DNA repair mechanisms might yield the final gene fusion through recombination in regions prone to DNA breaks (Fig. 7A). Such events were rare in the initial population of transfected cells (1 in 103 or 104 cells occurred within 3 days). However, the necessary machinery is clearly present in normal prostate epithelial cells prior to malignant transformation. If the resulting gene fusion (such as TMPRSS2-ERG) provides a growth advantage, a single affected cell among billions of cells in a normal prostate tissue may proliferate abnormally and eventually contribute to cancer formation. Identifying such initiator RNAs might provide novel insights into early disease mechanisms, as well as the discovery of new preventive and therapeutic strategies to combat cancer.
Contrary to the previous “cart before the horse” model (Rowley and Blumenthal, 2008; Zaphiropoulos, 2011), our results do not support the postulation that a sense fusion mRNA derived from trans-splicing between two pre-mRNAs effectively directs gene fusion. Expressing sense input RNAs mirroring the trans-spliced mRNA failed to induce fusion in LNCaP cells (Fig. 1A and S2). Of ten antisense RNAs that were demonstrated to be capable of inducing fusion (Fig. 1A, 1B, 2E, 4A, 6B), all of their corresponding sense RNAs failed to induce fusion (Fig. 1A, 1C, 2G, 4B, 6B). This occurred even though the sense RNAs could, in theory, anneal to the same genomic sites targeted by their antisense counterparts and form similar DNA/RNA hybrids when paired with the antisense strand of genomic DNA. As demonstrated in Fig. 5, this antisense versus sense disparity can be explained by transcriptional conflict (Fig. 7B and 7C). Because the TMPRSS2 promoter is highly active in LNCaP cells, sense chimeric RNAs forming DNA/RNA hybrids with antisense strands of genomic DNA (the template strand used for transcription) would be frequently “bumped” off and unable to stabilize the structures required for initiating genomic arrangements. The mechanistic basis for the antisense versus sense RNA, and whether these phenomena can be generalized, remain to be investigated.
Our results also do not support the hypotheses that antisense input RNAs, acting as a docking station, mediate trans-splicing by base-pairing with both endogenous sense parental pre-mRNAs, or by bringing the parental genes in close proximity thus facilitating trans-splicing of parental pre-mRNAs transcribed from two genomic loci. Both mechanisms would require the continuous presence of antisense input RNAs to sustain the expression of induced fusion transcripts. Yet we showed that the induced fusion expression has a permanent nature and requires no continuous presence of input RNAs (Fig. 3B and Fig. 4D). Furthermore, in the case of TMPRSS2-ERG there is no detectable ERG parental RNA as raw material (Fig. 3A) to account for the trans-splicing models. Moreover, sense input RNAs, which are not complimentary to the sense parental pre-mRNAs thus cannot act as their docking station, are able to induce the fusion transcripts after a brief period of transcriptional inhibition, again arguing against the docking model. On the contrary, the genomic breakpoints identified by genomic PCR and chromosomal co-localization provided by FISH, strongly support that the induced expression of fusion transcript is largely the consequence of gene fusion resulting from chromosomal translocation. While prior works have shown that infrequent TMPRSS2-ERG fusions can by induced through genotoxic stress such as gamma radiation in the presence of DHT that increases double-stranded DNA breaks (Lin et al., 2009; Mani et al., 2009), such mechanisms of general genotoxicity fail to account for the specificity of gene fusion partners found in cancer. This report is the first to demonstrate RNA-mediated gene fusion in mammalian cells, and provides an RNA-driven mechanism that can account for the ‘specificity’ of gene fusion partners that were selected to undergo gene fusion in early disease stages. The results may represent a pathological example in a broad spectrum of potential RNA-mediated genome rearrangements and could have fundamental implications in the biology of mammalian genome stability, as well as gene editing technology via mechanisms native to mammalian cells.
Author contributions
SKG and LL conducted the experiments, SKG and LY designed the experiments, SKG and LY conceived the project and wrote the manuscript.
Materials and Methods
LNCaP cell culture
LNCaP cells were routinely cultured in RPMI 1640 medium (RPM1 1640, 1X, with L-glutamine, #10-040-CV, CORNING cellgro) containing 10% fetal bovine serum (premium grade FBS, #1500-500, Seradigm) and 1% penicillin/streptomycin (#15140-122, Gibco) in a 5% CO2 humidified incubator. For experiments involving the induction of fusion gene by input RNA, regular fetal bovine serum in the culture medium was replaced by Charcoal:Dextran stripped fetal bovine serum (catalog#100-119, Gemini Bioproducts) to remove hormones present in serum. LNCaP cells were cultured in this special medium for 24 hrs prior to plasmid transfection.
PNT1A cell culture
PNT1A cells were routinely cultured in RPMI 1640 medium containing 10% fetal bovine serum (premium grade FBS, #1500-500, Seradigm) and 1% penicillin/streptomycin (#15140-122, Gibco) in a 5% CO2 humidified incubator.
Transient transfection of plasmids for expressing the input RNAs
Twenty hours prior to transfection, LNCaP cells were seeded in 12-wells plate (BioLite 12 Well Multidish, #130185, Thermo Fisher Scientific) with a density of 5x105cells/well and 1 ml/well of culture medium containing Charcoal:Dextran stripped fetal bovine as described above. Transfection was performed using Turbofect transfection reagent (Thermo Scientific, #R0531) according to manufacturer’s protocol. Briefly, 1μg of a particular plasmid was first diluted in 100μl of the serum-free DMEM followed by immediate mixing by pipetting. 4μl of the transfection reagent was then added to the diluted DNA followed by mixing and incubation for 20 min. The DNA/transfection reagent mixture was then added drop wise to a well containing LNCaP cells in 1ml medium.
For transfection in PNT1A cells, 5x105 cells/well were plated in 12-wells plate in 1 ml/well of cultured medium 24 hrs prior to transfection. Transfection was performed using the same formula described for LNCaP cells. For repetitive transfections, initially transfected PNT1A cell population were split every three days, half was processed for RT-PCR assay and half was seeded again in a new well for the next transfection.
DHT preparation and treatment
DHT (Dihydrotestosterone) was purchased from Sigma Adlrich (5α-Androstan-17β-ol-3-one, #A8380). Concentrated stock of 1500μM was prepared by dissolving 4.3566 mg of DHT powder in 10 ml of 100% ethanol (200 proof ethanol, Koptec, #V1016) and then aliquoted in 1ml tubes and stored at -80°C.
For treating cultured cells, concentrated DHT stock was diluted as 10x working solutions (for example, for 0.9 μM final concentration, 10x is prepared as 9.0 μM) with the appropriate complete culture medium and used immediately. Complete media for LNCaP cells: RPMI 1640 + 10% Charcoal:Dextran stripped fetal bovine serum + 1% penicillin/streptomycin.
Complete media for PNT1A: RPMI 1640 + 10% fetal bovine serum + 1% penicillin/streptomycin. Six hrs post transfection, 111μl of fresh 10x DHT working solutions was added to each well of 12-wells plate containing 1ml medium and transfected cells.
For long-term treatment, medium was changed with fresh DHT every three days.
RNA isolation
Total RNA from cultured cells was extracted using High Pure RNA isolation Kit according to manufacturer’s instructions (#11828665001, Roche). Briefly, cells were suspended in 200μl of PBS buffer and were then lysed with 400μl of lysis buffer. The sample was then passed through the filter assembly resulting in the binding of the nucleic acids to the filter. The filter containing nucleic acids was then incubated with DNase I dissolved in DNase incubation buffer to degrade genomic and plasmid DNAs. The column was then rinsed with wash buffer and total RNA then eluted in a new tube for further analysis.
For detection of residual genomic and plasmid DNA, eluted RNA was subject to PCR reaction with primers specific to intron regions of house-keeping gene GAPDH, and with primers specific to plasmid transfected. Total RNA was converted to cDNA only if it is validated as free of DNA contamination.
Reverse transcription reaction
1 μg of total RNA was used for each reverse transcription reaction according to manufacturer instruction (superscript III RT, # 18080-051, Invitrogen). RNA was converted to cDNA either with Oligo dT primer (for induced fusion transcripts) or with random hexamers (for input RNAs expressed by U6 promoter). After the addition of dNTPs, the mixture was denatured at 65°C for 5 minutes. This was followed by the addition of a master-mix containing 1× superscript buffer, 10 mM DTT, 5 mM Magnesium chloride, RNaseOUT and Superscript III reverse transcriptase. Reactions were carried out at 50°C for 50 minutes and then terminated by incubation at 85°C for 5 minutes. cDNA was then treated with RNase-H for 20 minutes at 37°C to degrade RNA in DNA/RNA hybrid. 1 μl of cDNA was used as template for each subsequent PCR reaction.
RT-PCR for detecting induced fusion transcripts
The majority of induced fusion RNAs in this manuscript were detected using one-round RT-PCR. The following cases were assayed using three-round nested PCR: (1) the results of DHT treatment at physiological concentrations as shown in Fig. 1D and 6A, (2) the induction of TMPRSS2-ERG fusion transcript in non-malignant PNT1A cells as shown in Fig. 3F, (3) the specificity of input RNAs assayed in Fig. 4C, and (4) endogenous ERG level detection in LNCaP cells in Fig. S13. The following cases were assayed using two-round nested PCR: the induction of TMPRSS2-ETV1 fusion RNA as shown in Fig. 4A and 4B, and the detection of TMPRSS2-ETV1 fusion in the enriched population in Fig. 4D.
PCR was done with a standard three-step protocol using REDTaq DNA polymerase (#D5684-1KU, Sigma) according to manufacturer instruction.
Reaction was set as follows:
PCR reaction:
Standard one-round PCR conditions for TMPRSS2-ERG:
PCR conditions for three-round nested PCR for TMPRSS2-ERG:
1st round : PCR with TMPRSS2 ex-1 F1 and ERG ex-4 R1 on 1μl of cDNA
2nd round : PCR with TMPRSS2 ex-1 F2 and ERG ex-4 R2 on 1μl of 1st round product, PCR conditions same as 1st round.
3rd round : PCR with TMPRSS2 ex-1 F3 and ERG ex-4 R3 on 1μl of 2nd round product, PCR conditions same as 1st round.
PCR conditions for two-round nested PCR for TMPRSS2-ETV1:
1st Round: Top down PCR with TMPRSS2 ex-1 F1 and ETV1 ex-6 R1
2nd Round: PCR with TMPRSS2 ex-1 F2 and ETV1 ex-5 R1 on 1μl of 1st round.
3rd round (for Fig. 4D): PCR with TMPRSS2 ex-1 F3 and ETV1 ex-5 R2 on 1μl of 2nd round, PCR conditions same as 2nd round.
Long range PCR for detecting genomic DNA fusion junction
Nested long-range PCRs according to the manufacturer’s protocols using LA PCR kit (Takara, # RR002M). 200 ng of genomic DNA was used in each reaction and PCR was performed with annealing and extension at 68°C for 20 minutes. 1μl from the above reaction (1st round PCR) was used as template for the 2nd round PCR.
For the genomic breakpoint identified in this manuscript, 1st round long range PCR was done using primers TMPRSS2 genomic bk-F1 and ERG genomic bk-R1 shown in primer list below. 1μl from the above reaction (1st round PCR) was used as template for the 2nd round PCR using inner primers TMPRSS2 genomic bk-F2 and ERG genomic bk-R2.
Cloning and Sanger sequencing of induced fusion transcripts
PCR amplified cDNA bands were excised from the gel and eluted using QIAquick Gel Extraction Kit (#28706, Qiagen). The eluted bands were then cloned to pGEM-T vector (pGEM-T vector system I, # A3600) following manufacturer instruction. Sanger sequencing was performed using the service of Beckman Coulter Genomics.
Tm calculations
Melting temperature (Tm) of putative genomic DNA stems were calculated using the following formula (Rychlik and Rhoads, 1989):
A high energy G·T and A·C wobble pair known to have Watson-Crick like geometry in DNA double helix (Kimsey and Al-Hashimi, 2014; Watson and Crick, 1953) are considered as having the same stability as an A·T pair.
Fluorescent in situ hybridization (FISH)
Enriched population carrying TMPRRS2-ETV1 fusion events were first grown on 18 mm round #1 coverglass in a 12-well cell culture plate at the initial density of 200-400k/well. Cells were then fixed with 4% (vol./vol.) formaldehyde followed by denaturation of DNA with 0.1 N HCl for 5 min and with 70% formamide at 85°C for 7 min. Hybridization of target DNA with probes were done at 37°C for 16hr in a humidified chamber. Cells were then washed, stained with DAPI and imaged with microscope. FISH probes for TMPRSS2 (RP11-35C4, red) and ETV1 (RP11-769K2, green) were purchased from Empire Genomics.
α-Amanitin assay
Twenty hours prior to transfection, LNCaP cells were seeded in 12-wells plate (BioLite 12 Well Multidish, #130185, Thermo Fisher Scientific) with a density of 5x105cells/well and transfection was performed using Turbofect transfection reagent (Thermo Scientific, #R0531) as described earlier. DHT was added at the final concentration of 0.9μM six hours post transfection. Following overnight incubation, cells were then treated with 4μg/ml α-amanitin for various time periods (0, 2, 6, 12 and 24 hours). Cells were then revived in fresh medium containing 0.9μM DHT without α-amanitin and RT-PCR was performed for either TMPRSS2-ERG or TMPRSS2-ETV1 fusion.
Input RNA sequences
Sense-2 long
Antisense-1
Antisense-2
Antisense-3
Antisense-4
Antisense-5
Antisense-6
Antisense-7
Antisense-8
Antisense-9
Antisense-5A
Antisense-5B
Antisense-B1
Antisense-B2
Antisense-B3
Antisense-C1
Antisense-C2
Antisense-C3
Antisense-D1
Antisense-D2
Antisense-D3
Antisense-E1
Antisense-F1
Antisense-G1
Sense-B1
Sense-C1
Sense-D1
Antisense-TMPRSS2-ETV1-A1
Antisense-TMPRSS2-ETV1-B1
Antisense-TMPRSS2-ETV1-C1
Antisense-TMPRSS2-ETV1-D1
Antisense-TMPRSS2-ETV1-E1
Antisense-TMPRSS2-ETV1-F1
Antisense-TMPRSS2-ETV1-G1
Antisense-TMPRSS2-ETV1-H1
Sense-TMPRSS2-ETV1-A1
List of primers used
RT-PCR primers for amplifying induced fusion RNAs:
Primers used in three-round PCR for amplifying induced fusion RNAs:
RT-PCR primers for amplifying endogenous parental mRNAs:
RT-PCR primers for amplifying BCAM-AKT2 chimeric RNA:
RT-PCR primers for amplifying input RNAs:
PCR primers used for amplifying the identified TMPRSS2-ERG genomic DNA breakpoint:
The rest of PCR primers used for genomic breakpoint analyses are not listed, but their locations are shown in Fig. S14A.
PCR primers used for amplifying the identified TMPRSS2-ETV1 genomic DNA breakpoint:
Acknowledgments
We thank Richard Kelley, Michael Ittmann, Richard Sifers and Tom Cooper for critical suggestions, Christine Shiang for cloning AZI1 ex16-17, Olga Dakhova, Jianghua Wang, and Michael Ittmann for providing the cell lines. Sachin Kumar Gupta has been supported by CPRIT training grant RP160283. Laising Yen has been supported by Duncan Cancer Center Pilot Grant, CPRIT HIHRRA RP160795, and NIH R01EB013584. We would also like to thank Radhika Dandekar, Fabio Stossi, and Michael Mancini for assisting with microscopy, with the support by the Integrated Microscopy Core at Baylor College of Medicine with funding from the NIH (DK56338, and CA125123), CPRIT (RP150578) the Dan L. Duncan Comprehensive Cancer Center, and the John S. Dunn Gulf Coast Consortium for Chemical Genomics.