Recurrent TRIO Fusion in Nontranslocation–Related Sarcomas

Sarcomas represent a heterogeneous group of rare tumors accounting for approximately 1% of adult cancers with more than 50 histologic subtypes. These tumors, derived from mesenchymal tissues (i.e., soft tissue, bone, or muscle), are classified on the basis of their line of differentiation, following the World Health Organization (WHO) recommendations (1). Along with this histologic heterogeneity, sarcoma genetics is also highly heterogeneous in terms of oncogene-specific driver alterations and whole-genome instability. Sarcomas can be classified into two groups depending on genome stability. One group consists of sarcomas with a relatively stable genome with specific oncogenic alterations (rare gains and losses involving mainly full chromosomes). Among these, gastrointestinal stromal tumors (GIST; KIT or PDGFRA mutations; ref. 2) and sarcomas with a specific translocation (e.g., Ewing sarcoma, synovial sarcoma, and alveolar rhabdomyosarcoma; refs. 3–5) represent 20% and 30% of sarcomas, respectively. The second group consists of nontranslocation-related sarcomas with higher chromosomal instabilities and pleomorphic histologic patterns. There is no specific genetic alteration described so far in those sarcomas. Nevertheless, we can distinguish well-differentiated and dedifferentiated liposarcoma (20% of sarcomas) with recurrent MDM2 and CDK4 amplification within a moderately rearranged profile (6, 7). The remaining 30% mainly comprise leiomyosarcoma, undifferentiated pleomorphic sarcoma, myxofibrosarcoma, pleomorphic liposarcoma, and pleomorphic rhabdomyosarcoma. Although their genomic instability has not been deciphered yet, some recurring genetic alterations have been identified in these tumors: the 13q14-21 region is frequently lost in leiomyosarcomas and undifferentiated pleomorphic sarcoma targeting RB1 deletion/inactivation (8); the TP53 pathway is consistently inactivated, mainly by TP53 deletions/mutations (9); and a gain of 5p region is also frequent and triggers TRIO amplification (10, 11). Extensive analysis of the tumor genome by molecular cytogenetic approaches showed many breakpoints [array comparative genomic hybridization (aCGH)] and translocations per cell (chromosome painting; refs. 12, 13). Breakpoints located in coding regions are supposed to produce chimeric transcripts with potent oncogenic effects. It has been demonstrated that RNA sequencing (RNA-seq) is an efficient approach to identify fusion transcripts in tumors, particularly in sarcomas (14–16). Hence, we subjected 117 nontranslocation-related sarcomas to RNA-seq to identify fusion transcripts.

The samples used in this study are part of the Biological Resources Center of Institut Bergonié (CRB-IB; Bordeaux, France). In accordance with the French Public Health Code (articles L. 1243-4 and R. 1243-61), the CRB-IB has received the agreement from the French authorities to deliver samples for scientific research (number AC-2008-812). The samples come from care and are requalified for research. The project was approved by the Institut Bergonié ethics committee (scientific advisory board). Every case was histologically reviewed by the pathologist subgroup from the French Sarcoma Group and classified according to the 2013 WHO classification by histology, IHC, and molecular genetics and cytogenetics when needed.

The screened series was composed of 112 sarcomas with complex genomics and five cell lines (CL; CL1 derived from an undifferentiated pleomorphic sarcoma T1; CL2 derived from a dedifferentiated liposarcoma; CL3 derived from a myxofibrosarcoma; and CL4 and CL5 derived from two leiomyosarcomas) for the first cohort.

The second and the third cohorts are composed of 27 synovial sarcomas and 74 GIST, respectively. All tumors have been histologically reviewed by a pathologist panel from the French Sarcoma Group.

Cell line establishment was performed as described previously (17). Authentication of cell line is analyzed by aCGH by comparing the cell line with the corresponding tumor origin.

The culture medium was composed of RPMI1640 with GlutaMAX (Gibco BRL, Life Technologies) supplemented with 10% FCS and 1% antibiotics (penicillin/streptomycin, Gibco BRL, Life Technologies). All cell lines were tested for mycoplasma by PCR (Sigma; LookOut Mycoplasma PCR Detection Kit) according to the manufacturer's recommendations.

RNA extraction from frozen and formalin-fixed paraffin-embedded (FFPE) samples, sample preparation, and RNA-seq were performed as described previously (18).

Bioinformatics analysis was performed as described previously (18).

Briefly, deFuse (v0.6.1; ref. 19) was used on curated FastQ files with ENSEMBL GRCh37.74 annotations. We kept candidate fusions with the following criteria: probability >50%, non-read-through, non-read-through like (same chromosome, indicated as deletion, coding/coding gene region, and genomic distance <20,000) and breakpoint homology <30.

TopHat-Fusion (v2.1.0; ref. 20) was used on curated FastQ files with the same annotations as described by Lesluyes et al. in 2016 (18). Fusion minimum distance was set to 10,000, and fusion anchor length was set to 13. Other parameters were similar to the expression processing.

ChimeraScan (v0.4.5; ref. 21) algorithm was used on raw FastQ files (equal-length reads restriction) with UCSC (fixed on 2016/02/09) annotations. Multiple hits parameter was set to 50. We kept fusions with the following criteria: non-read-through, score >5, spanning reads >2, and non-HLA/HLA fusions (due to sequence homology).

For the three detection algorithms, duplicated entries (same genes) and non-ORF conservation fusions were discarded.

Unsupervised clustering method was performed with Wald algorithm and Euclidean distance available in the cluster R package (v2.0.3). Differential gene expression analysis was performed with DESeq2 R package (1.8.1; ref. 22). Gene interaction was performed using BioGRID (23). Gene Set Enrichment Analysis (GSEA; v2.2.0; refs. 24, 25) was used with Hallmarks gene sets from MSigDB (v5.0), gene set permutation type, RefSeq chip platform without gene symbol collapse, and a minimum set size of 10. Differentially exonic expression was performed by DEXSeq (6 vs. 40; ref. 26).

Total RNA was reverse transcribed using High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems) according to the manufacturer's instructions.

For PCR, primers were designed using Primer3 program (http://frodo.wi.mit.edu/primer3/) and are presented in Supplementary Table S1. Touchdown 60°C program has been used (TD 60°C; two cycles at 60°C, followed by two cycles at 59°C, two cycles at 58°C, three cycles at 57°C, three cycles at 56°C, four cycles at 55°C, four cycles at 54°C, five cycles at 53°C, and finally 10 cycles at 52°C). PCR was performed on 50 ng of cDNA using AmpliTaqGold DNA Polymerase (Applied Biosystems). PCR products were then purified using ExoSAP-IT PCR Purification Kit (GE Healthcare), and sequencing reactions were performed with the Big Dye Terminator V1.1 Kit (Applied Biosystems) according to the manufacturer's recommendations. Samples were then purified using the Big Dye XTerminator Purification Kit (Applied Biosystems) according to the manufacturer's instructions, and sequencing was performed on a 3130xl Genetic Analyzer (Applied Biosystems). Sequence analysis was performed with SeqScape software v2.5 (Applied Biosystems).

Expression of different TRIO-TERT and TRIO transcripts was studied with probes and primers listed in Supplementary Table S2. Normalization of expression was carried out with two reference genes: ACTB and RPLP0. Probes and primers (TaqMan Gene Expression Assay, Life Technologies) of these genes are Hs99999903_m1 and Hs99999902_m1, respectively. Reaction was performed with TaqMan Universal PCR Master Mix (Life Technologies) using StepOnePlus (Life Technologies) according to the manufacturer's recommendations. Expression was calculated with the formula: |$\Delta {C_{\rm{t}}}{\rm{ }} = {\rm{ }}{C_{\rm{t}}}_{_{TRIO - TERT}} - {\rm{ }}{C_{\rm{t}}}_{_{{\rm{reference \ gene}}}}$|(27). Subsequently, difference of expression between both transcripts was calculated with the formula |${2^{\Delta {C_{\rm{t}}}}}_{{\rm{transcript}}\,{\rm{1}} - {\rm{transcript}}\,{\rm{2}}}$|⁠.

FISH assay on FFPE cases T1, S867, and S909 was performed using the Histology FISH Accessory Kit (Dako) as described previously (28) on a 4-μm paraffin-embedded section. Regarding metaphase chromosome spreading on CL1, cells were incubated with colchicine overnight and resuspended in isotonic KCl buffer. Cells were finally fixed in standard 3:1 methanol:acetic acid final fixative. After several washes, cells were spread on a glass slide. Denaturation (5 minutes at 82°C for paraffin-embedded tissues and 1 minute at 73°C for chromosome spreading) and hybridization (overnight at 37°C) were achieved by placing the slides into a hybridizer (Dako). For the identification of the TRIO fusion, one BAC clone covering 5′ part of TRIO (RP11-1134M22; red signal) and two BAC clones covering 3′ part of TRIO (RP11-81P9, RP11-1079G4; green signals) were used; green and red fluorescent signals were analyzed in tumors using a Nikon Eclipse 80i fluorescence microscope with appropriate filters. Pictures were captured using a Hamamatsu C4742-95 CCD camera and analyzed with the Genikon software. A TRIO rearrangement was detected when red and green signals were separated in the nucleus. A TRIO rearrangement was considered as present if at least 10% of tumor cells showed a rearrangement pattern. Unbalanced rearrangements (fusion signal associated with at least one supernumerary red signal) are considered as positive case when chimeric transcript has been validated by RT-PCR and Sanger sequencing and with at least 10% rearranged signals.

Genomic DNA was isolated using a standard phenol–chloroform extraction protocol. Affymetrix CytoScan HD Array (Affymetrix) was used according to the manufacturer's instructions. The test was conducted on DNA samples from TRIO translocated cases. CEL files obtained by scanning the CytoScan arrays were analyzed using the Chromosome Analysis Suite software (Affymetrix) and the annotations of the genome version GRCH37 (hg19).

Tissues were deparaffinized in xylene and rehydrated in a series of ethanol baths. For antigen retrieval, slides were incubated in DAKO Target Retrieval Solution, pH 9 (DAKO), for 20 minutes in a microwave oven. The primary antibodies and dilutions (dilution in DAKO REAL antibody diluent, DAKO) used in this study are as follows: anti-PML (PG-M3, 1:200, sc-966, Santa Cruz Biotechnology) and anti-TERF2 antibody (1:200, HPA001907, Sigma). All primary antibodies were incubated for 1 hour at room temperature. Secondary antibodies and dilutions used are as follows: anti-mouse immunoglobulins/FITC (1/400, Dako) and anti-rabbit IgG (H+L) Alexa Fluor 594 conjugate (1/500, Invitrogen). Slides were mounted with VECTASHIELD/DAPI medium (Vector Laboratories) and were then analyzed using a Nikon Eclipse 80i fluorescent microscope with appropriate filters. Pictures were captured using a Hamamatsu C4742-95 CCD camera.

Short hairpin RNA (shRNA) sequences targeting TRIO₃₃-TERT₂ and a nontargeting control sequence (Supplementary Table S3) were cloned into the doxycycline-inducible pLKO.1neo lentiviral shRNA vector (Addgene). VSV-G–pseudotyped lentiviral particles were produced by cotransfection of 293T cells with pLKO.1 constructs and the compatible packaging plasmids psPAX2 and pVSVg. Cell lines were incubated overnight with lentiviral supernatants in the presence of 8 μg/mL polybrene (Sigma H9268), and stably transduced cells were selected with neomycin (0.1–1 mg/mL, Sigma 4727894001) for 10 days. Downregulation of TRIO₃₃-TERT_2/3 mRNA expression was controlled by qRT-PCR.

Proliferation measurement was performed using the IncuCyte Live Cell Imaging System (Essen BioScience). CL1s were seeded at 3,000 cells per well in a 96-well plate with eight replicates for each conditions. Cells were pretreated with or without doxycycline (1 μg/mL, Sigma D3072) for 48 hours before proliferation analysis. Treatment was maintained during 5 days of experiment. Proliferation was monitored according to the manufacturer's recommendations (two pictures every 2 hours for each well).

RNA-seq raw files (FastQ) are available on Sequence Read Archive under the accession numbers SRP057793 (contains 112 tumors) and SRP059536 (contains the five sarcoma cell lines: CL1 is M965, CL2 is M969, CL3 is M961, CL4 is M963, and CL5 is M964).

Expression data for the 117 samples are available on Gene Expression Omnibus under accession number GSE75885.

To identify fusion transcripts, we performed RNA-seq on 112 sarcomas with complex genetics (Table 1) and 5 sarcomas cell lines. Means of 78.34 (± 18.71) and 74.97 (± 17.83) million reads per sample were obtained after library sequencing following low quality 5′ and 3′ trims and overlap correction, respectively. DeFuse (19) analysis predicted a total of 10,382 fusion transcripts for all samples with a fusion mean of 88.74 (± 52.7) per sample (Fig. 1). Among them, 3,610 detected chimeric transcripts were actually read-throughs (chimeric transcripts between two adjacent genes and induced by transcription; ref. 29) according to deFuse and were consequently discarded from further processing. Following this selection, 6,772 potential fusion transcripts were kept. To identify in-frame fusion transcripts, we filtered this selection and focused on the 420 in-frame predicted fusions with conserved open reading frame (ORF) for each partner. Among these transcripts, eight presented a potential recurrence: CTSC-RAB38 (13 samples), GSE1-RP11-680G10.1 (12 samples), NSUN4-FAAH (8 samples), CTBS-GNG5P2 (8 samples), TRIO-TERT (4 samples), FARSA-SYCE2 (2 samples), RPL11-TCEB3 (2 samples), and TFG-GPR128 (2 samples). Among these eight transcripts, five are known to be read-throughs but evaded the deFuse filter: CTSC-RAB38 and TFG-GPR128 have been reported as read-throughs (30, 31), whereas CTBS-GNG5P2, RPL11-TCEB3, and NSUN4-FAAH are referenced in the ConjoinG database (32), which contains detailed information about identified read-throughs in the human transcriptome. FARSA-SYCE2 and GSE1-RP11-680G10.1 fusions have been identified in normal human tissues (RT-PCR and Sanger sequencing; Supplementary Fig. S1) and the two genes implicated in the fusion are on the same chromosome and transcribed on the same genomic strand. As they share similar characteristics as read-throughs, we discarded them from further analysis. The last potential in-frame recurrent chimeric transcript predicted by deFuse, TRIO_33/34-TERT_2/3 (fusion between TRIO either exon 33 or 34 and either TERT exon 2 or 3), occurred in four samples (CL1, S867, S829, and S822) and has not been detected in normal tissues (Fig. 2A and B). In addition, TRIO-TERT gene fusion has been validated by RT-PCR and Sanger sequencing in T1, the primary tumor of CL1 (Fig. 2A). No reciprocal transcript has been detected for these fusions (data not shown). Same analyses have been performed with two supplementary algorithms, TopHat-Fusion (20) and ChimerScan (21). TRIO-TERT is also the only recurrent in-frame chimeric transcript detected with these algorithms.

Consequently, TRIO-TERT fusion is the only recurring in-frame fusion identified in this cohort.

aCGH analysis of three translocated cases (T1, S829, and S822) with high quality DNA available demonstrated that the TRIO (5p15.2) and TERT (5p15.33) loci (Fig. 2C) were rearranged (gain and/or amplification) in two of three cases (e.g., presented in Supplementary Fig. S2A and S2B). Interestingly, in the TRIO gene, the copy number variation break is located in the region of fusion. We therefore hypothesized that the frequently observed chromosome 5p/TRIO amplification in sarcomas (11) could be associated, at least in some cases, to TRIO rearrangements. We thus reexamined the genomic profile of the 112 sarcomas tumors of RNA-seq cohort and 24 supplementary cases of nontranslocation-related sarcomas (Supplementary Table S4). Three cases presented a comparable TRIO amplification profile (T2, S909, and S921). Specifically, in T2, TRIO₃₄-TERT₃ fusion was detected by RT-PCR and Sanger sequencing (Fig. 2B). The two new remaining cases with TRIO amplification did not express TRIO-TERT gene fusion. We performed RNA-seq on these two additional samples, and deFuse analysis indicated that they both expressed a fusion gene involving TRIO, but not TERT. In sample S909, TRIO₃₃ is fused to LINC01504, and in sample S921, TRIO₃₄ is fused to ZNF558; both were subsequently validated by RT-PCR and Sanger sequencing (Fig. 2D–F).

To validate the genomic origin of the fusion, we performed FISH analysis in four cases with available and good enough quality FFPE blocks. A break-apart approach targeting TRIO validated rearrangement in the three interpretable cases (CL1/T1, S867, and S909) with 84%, 36%, and 58% of positive cells, respectively (Fig. 3).

In all cases, regardless of the partner, either TRIO exon 33 or exon 34 is fused. In five cases, fusion joined either TRIO exon 33 (TRIO₃₃-TERT: T1 and S867) or exon 34 (TRIO₃₄-TERT: cases S829, S822, and T2) to TERT, an exception being case S822, in which CDH18 exon 2 is inserted in between TRIO exon 34 and TERT. Whatever the TRIO (or CDH18 for case S822) exon, fusions occurs either with TERT exon 2 or exon 3 (TRIO-TERT₂ or TRIO-TERT₃).

As it has been reported that TERT exon 2 is subjected to alternative splicing (33), the expression of both isoforms should be due to alternative splicing instead of two translocation events (a much rarer genomic event).

Among these six distinct chimeric transcripts of the TRIO-TERT fusion, four (TRIO₃₃-TERT₃, TRIO₃₄-TERT₂, TRIO₃₄-CDH18₂-TERT₂, and TRIO₃₄-CDH18_2-TERT₃) did not conserve the TERT ORF with a stop codon after the last fused TRIO exon at 77, 353, 23, and 23 bp, respectively (Fig. 2A and B).

Given that all cases expressed both isoforms (TRIO_33/34-TERT₂ and TRIO_33/34-TERT₃), we sought to determine which isoform is preferentially expressed. Isoform-specific quantitative RT-PCR analysis showed that chimeric transcripts with TERT exon 3 were at least 15 times more expressed regardless the TERT ORF conservation (Supplementary Fig. S3).

In case S909, TRIO exon 33 is fused to LINC01504, leading to the expression of three isoforms: TRIO₃₃-LINC01504_intron2, TRIO₃₃-LINC01504_intron3, or to TRIO₃₃-LINC01504₄, resulting in the loss of the ORF after the breakpoint with a stop codon after the last TRIO exon at 146, 84, and 113 bp, respectively (Fig. 2D). Different transcripts observed in S909 could be explained by alternative splicing mechanisms as sequences near the fusion break-points in LINC01504 introns 2 and 3 are predicted to be splicing sequences by Human Splicing Finder (34) with 66.9% and 74% of consensus value, respectively (Supplementary Fig. S4).

In case S921, TRIO exon 34 is fused to ZNF558 exon 8 (TRIO₃₄-ZNF558) but on the ZNF558 opposite strand resulting in the nonconservation of ORF of ZNF558 with a stop codon after the breakpoint at 287 bp (Fig. 2E).

Seven cases have been identified: five expressing two isoforms of TRIO-TERT and two others cases expressing TRIO gene fusion with out-of frame interchromosomal gene partner. For TRIO-TERT cases, the preferential expressed isoform is TRIO_33/34-TERT₃ and not the isoform with TERT ORF conservation. All these results indicated that TRIO gene fusion lead to the formation of truncated TRIO protein.

Unfortunately, we could not detect these fusion proteins by Western blot analysis of sarcoma cell line lysates, due to the lack of sensitivity/specificity of the TRIO antibody available that recognize these fusion proteins (data not shown).

The seven identified translocated cases belong to different histotypes: three undifferentiated pleomorphic sarcomas, two dedifferentiated liposarcomas, one pleomorphic rhabdomyosarcoma, and one myxofibrosarcoma. According to the Fédération Française des Centres de Lutte Contre le Cancer grading system, five of them were classified as grade 3 (T1, S867, S829, S822, and S909) and two as grade 2 tumors (T2 and S921). Three tumors presented a metastatic relapse and two presented a local recurrence (Table 2). These results indicate a trend toward tumor aggressiveness without reaching significance considering the number of cases.

Specificity of all TRIO chimeric transcripts has been tested on two independent cohorts of 74 GISTs and 24 synovial sarcomas by RT-PCR. None have been identified in these nonpleomorphic sarcoma cohorts associated to a specific oncogenic event (KIT/PDGFRA mutations and t(X;18) translocation, respectively; refs. 2–5).

TRIO gene fusion appears, thus, specific so far to nontranslocation-related sarcomas.

To test whether the presence of chimeric TRIO transcripts could be associated with a peculiar transcriptional profile of the fused TRIO samples (FTS), we applied unsupervised clustering method on 117 sarcomas with RNA-seq expression, including the six fused TRIO cases and the 111 nonfused TRIO samples (non-FTS; Supplementary Fig. S5). Our results indicate that FTSs do not have similar transcriptomic profiles as they did not cluster together. FTSs are localized within a subgroup of nonmuscular sarcomas, located in the extremities (adjusted Fisher P = 0.014 for both tumor differentiation and location).

We then performed a supervised differential gene expression analysis comparing fused and nonfused cases and identified 423 significant genes (Supplementary Fig. S6), out of which 69 were overexpressed in the FTS (Supplementary Table S5) and 354 overexpressed in non-FTS (Supplementary Table S6). We observed that TRIO and TERT were significantly more expressed (4.88 and 16.20 times, respectively) in FTS (adjusted Wald's P = 4 × 10⁻⁷ and 4.66 × 10⁻³, respectively). To establish whether TRIO overexpression is due to wild-type or fused TRIO, we measured TRIO differential exonic expression of FTS versus non-FTS using DEXSeq (Supplementary Fig. S7; ref. 26). As the analysis could not be done on the whole non-FTS cohort (n = 111), we selected the 40 (maximum analysis capacity) samples that highly expressed TRIO. This analysis did not measure any differentially expressed exon in FTS versus non-FTS. However, we can see a slight exonic usage switch in layer “exon usage” starting at exon 34. Exons before this point are a bit more expressed in FTS, where TRIO expression is due to wild-type and fused TRIO. The trend is inversed after the breakpoint; exons 34 at 58 are a bit more expressed in non-FTS. In addition, this result shows that as no reciprocal fusions were observed in FTS and exons beyond 34 are expressed, wild-type TRIO is expressed in FTS. Genes interacting with TRIO and TERT were not identified as differentially expressed according to BioGRID (23).

GSEA (24, 25) was then applied on the Hallmarks new collection included in the MSidDB v5.0 database (containing 50 referenced gene sets). In fused TRIO samples, 16 significantly enriched gene sets were identified, mainly implicated in immunity/inflammation (6/16 gene sets), regulation of cell cycle (3/16), and cell proliferation and migration (3/16; Supplementary Fig. S8). Immunity gene set enrichment due to immune cell infiltration has been ruled out following a histologic review indicating that there was no significant difference between translocated and nontranslocated cases in terms of immune cell infiltration. In non-FTS, overexpressed gene sets are implicated, notably in myogenesis and adipogenesis (Supplementary Fig. S9).

These data suggest that TRIO fusion events are associated with overexpression of genes involved in immunity and inflammation, regulation of cell cycle, cell proliferation, and migration.

In a tumoral context, telomere length is maintained by two exclusive ways, either by telomerase reactivation or by the alternative lengthening of telomeres mechanism (ALT mechanism), and both mechanisms appear to be mutually exclusive (35, 36). In pleomorphic sarcomas, the ALT mechanism is very frequent (60% of cases; ref. 37). To test whether TRIO-TERT fusion could be associated with TERT reactivation, we evaluated the colocalization of PML and TERF2 as a hallmark of the ALT mechanisms, which, if negative, means that it is the telomerase reactivation pathway (38). The four TRIO-TERT interpretable cases were negative for PML/TERF2 colocalization and, consequently, negative for ALT mechanism. Moreover, S921 (TRIO₃₄-ZNF558) presented the same results and was also declared ALT negative (Supplementary Fig. S10).

To uncover the role of TRIO fusions, we performed TRIO₃₃-TERT_2/3 knockdowns in CL1 using lentivirally delivered shRNA contructs. Two shRNAs for each isoform were tested at day 0 and day 4 and only one allowed the specific decrease of TRIO₃₃-TERT₂ (50%) at 4 days without decreasing TRIO and TRIO₃₃-TERT₃ expressions (Fig. 4A). With respect to transcriptomic analysis, we thus used it to run proliferation assays in CL1. As shown in Fig. 4B and C, contrary to the nontargeting shRNA, knocking down TRIO₃₃-TERT₂ significantly reduces cell proliferation.

Sarcomas with complex genetics are tumors that harbor a high number of chromosomal rearrangements. To better understand the biology of these tumors, we searched for recurrent chimeric transcripts.

In the current study, 10,382 potential chimeric transcripts have been predicted, and 3,604 of them were considered read-throughs by the detection algorithm deFuse. Although read-throughs were not considered in this study, it has been reported that tumor-specific read-through expression could be involved in oncogenesis (39).

More than 80 filtered fusion transcripts were predicted per case in nontranslocation-related sarcomas, and most of the chimeric transcripts detected arose from intrachromosomal rearrangements and were not in-frame (96%). This is in agreement with data reported by Yoshihara and colleagues, regarding fusion transcript prediction in a large series of 4,366 cancers (various types excluding sarcomas; ref. 40). They showed that the high complexity of tumor genome is associated with a higher number of predicted fusion transcripts. Our results are consistent with the high genomic complexity levels of nontranslocation-related sarcomas. On the contrary, according to the data presented by Yoshihara and colleagues, approximately 36% of detected fusion transcripts were predicted to be in-frame in their cohort compared with 4.13% in our study. This suggests that chimeric transcripts could have other role(s) in nontranslocation-related sarcomas as they would mainly produce a truncated protein or no protein product at all.

Here, we report recurrent TRIO fusions identified by RNA-seq and validated in 6 poorly differentiated pleomorphic sarcomas out of 117 (5.1%) and one dedifferentiated liposarcoma (T2) identified in aCGH analysis out of 26 nontranslocation-related sarcomas (3.8%). Three different TRIO fusion partners have been identified in this study, TERT, LINC01504, and ZNF558. Whatever the fusion partner, breakpoints in TRIO are consistently either after exons 33 or 34. This likely suggests that either these conserved domains, in the fused TRIO proteins, are essential or that genomic location of the breakpoint is specifically targeted by an unknown mechanism. Furthermore, two of the partners (LINC01504 and ZNF558) are fused out-of-frame, suggesting that translocation leads to a truncated TRIO protein and that this truncated form is of a selective advantage for tumor cell. This hypothesis is strengthened by the isoform-specific quantitative analysis, showing that chimeric transcripts with an out-of-frame TERT fusion are the most expressed. It is even more strengthened by the differential expression analysis. For statistical consideration, GSEA (50 cancer Hallmarks) was applied instead of Gene Ontology (41) analysis, and we evidenced that TRIO-translocated cases significantly overexpressed genes involved in immunity/inflammation, cell proliferation, and migration.

TRIO protein has two GEF domains (42). While GEF1 mediates GDP to GTP exchange, leading to Rac1 and RhoG activation, GEF2 domain triggers RhoA activation. In addition to these domains, TRIO has a SEC14 domain, several spectrin repeats, two SH3 domains, an Ig-like domain, and a serine kinase domain. Unlike the GEF domains, the roles of these latter domains are still unclear. Rac1 and RhoG are members of Rho proteins controlling several pathways leading to cytoskeletal rearrangements, kinase activation, and gene transcription. They are implicated in cell proliferation and motility (43).

All TRIO fusions encode for TRIO N-terminal domains Sec14, spectrin, and GEF1 domain (Supplementary Fig. S11), which triggers Rac1 activation. Rac1 has been demonstrated to be directly involved in transformation and tumor progression (44). Rac1 acts by promoting anchorage-independent growth (45), proinflammatory pathway (46), proliferation, cell spreading, and migration by regulating lamellipodia formation (47, 48). Interestingly, TRIO gene fusions lead to the loss of the C-terminal GEF2, suggesting that the GEF2 domain is not associated to the process of tumor progression in sarcoma. This is consistent with the observation that most processes induced by TRIO depend on the activation of Rac1 by GEF1 and not on the activation of RhoA by GEF2 (49). Moreover, the depletion of TRIO₃₃-TERT₂ isoform demonstrates that TRIO fusion has an impact on cell proliferation. As this isoform is the least expressed in all cases, we hypothesize that the effect would be stronger if the TRIO₃₃-TERT₃ isoform was downregulated concomitantly (no shRNA available).

TRIO encodes at least seven different isoforms (49) and two of them (B and C), containing only the Sec14, spectrin, GEF1, and SH3 domains, are exclusively expressed in nervous system (50). Interestingly, the TRIO exons involved in the fusions (exons 1 to 33-34) are exactly the same as the ones observed in the two neuronal isoforms B and C, and as a consequence, they share the same protein domains. The role(s) of these two isoforms is unknown. Salhia and colleagues (51) have demonstrated that the transcriptional activation of the nervous system–specific TRIO isoform, through transfer in a different cellular context, could be an efficient way to hijack and remodel existing metabolic pathways for the benefit of tumor cells as demonstrated in glioblastoma (51).

TRIO gene has been identified in fusion with three partners: TERT, LINC01504, and ZNF558. Although TERT's role on elongation of telomeres is well known, the role of LINC01504 and ZNF558 is unknown. Stransky and colleagues (52) identified TRIO₃₃-TERT₂ fusion in two dedifferentiated liposarcomas. The authors suggest that this gene fusion could permit the expression of TERT in these tumors. In our study, the length of telomeres is maintained by telomerase reactivation and not by ALT mechanism in the TRIO gene fusions. TERT is overexpressed in these samples regardless of the TRIO fusion partner (i.e., TERT, LINC01504, and ZNF558), suggesting that, in these cases at least, telomerase activation is not associated with TRIO rearrangements. Moreover, in-frame TERT chimeric transcript loses the TEN domain of TERT permitting the fixation of TR RNA, the matrix for telomeres elongation (32). This prompts us to conclude that TRIO-TERT fusions do not trigger TERT reactivation in these tumors.

TRIO fusion transcripts have been detected exclusively in a cohort of sarcomas with complex genetics. We did not detect any TRIO chimeric transcript in large cohorts of GISTs and synovial sarcomas, which are both sarcoma subtypes associated with a specific genetic alteration and a very low genomic complexity. Moreover, in Yoshihara and colleagues' study (40), chimeric RNA-seq data of 4,366 tumor samples have been submitted to fusion detection, and TRIO fusion transcripts have not been detected in the whole cohort of various tumors (excluding sarcomas). TRIO fusions could therefore be quite specific to nontranslocation-related sarcomas, with an incidence of around 5%. It is the unique recurrent gene fusion so far identified in these aggressive nontranslocation-related sarcomas. Within this large sarcoma category, TRIO fusions are not limited to a specific histologic subtype. This suggests that, unlike sarcomas with simple genetics and a recurrent specific translocation (e.g., Ewing sarcoma, synovial sarcoma, and alveolar rhabdomyosarcomas), which are associated to a specific transcriptomic profile (14), this chromosomal event is certainly not an initiating one but a secondary one caused by genetic instability of nontranslocation-related sarcomas and involved in tumor progression via cell proliferation enhancement. However, we did not have sufficient statistical evidence with seven cases to highlight this point, and we cannot exclude the possibility that more cases would permit to refine the biological and/or clinical context associated with these TRIO rearrangements.

No potential conflicts of interest were disclosed.

Conception and design: L. Lartigue, J.-M. Coindre, F. Chibon

Development of methodology: S. Schmidt, F. Chibon

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Delespaul, T. Lesluyes, G. Pérot, J. Baud, P. Lagarde, S. Le Guellec, A. Neuville, P. Terrier, A. Debant, J.-M. Coindre, F. Chibon, D. Vince-Ranchère

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Delespaul, T. Lesluyes, G. Pérot, C. Brulard, L. Lartigue, J. Baud, J.-M. Coindre, F. Chibon

Writing, review, and/or revision of the manuscript: L. Delespaul, T. Lesluyes, L. Lartigue, S. Le Guellec, S. Schmidt, A. Debant, F. Chibon

Study supervision: F. Chibon

Computer time for this study was provided by the computing facilities MCIA (Mésocentre de Calcul Intensif Aquitain) of the Université de Bordeaux and of the Université de Pau et des Pays de l'Adour. The authors thank Andreas Bikfalvi laboratory for granting access to IncuCyte. We are grateful to the people involved in the French Sarcoma Group (GSF) for clinical annotations and access to their samples: L. Guillou, F. Collin, A. Leroux, Y-M. Robin, M. C. Chateau, I. Peyrottes, and F. Mishellany. The authors would also like to thank Dr. Ravi Nookala of Institut Bergonié for the medical writing services.

This work was supported by euroSARC (FP7-HEALTH-2011), Fondation ARC pour la recherche Contre le Cancer (to A. Debant), Ligue Contre le Cancer (Comité régional Gironde to F. Chibon and Comité régional Languedoc Roussillon to A. Debant) Inserm, and Integrated Cancer Research Site (SIRIC) of Bordeaux (BRIO-Bordeaux Oncology research).

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