Our study highlights the prevalence of gene fusions as one of the major genomic abnormalities in GBM. Fusions occur in approximately 30-50% of GBM patient samples. In the Ivy Center cohort of 24 patients, 33% of samples harbored fusions that were validated by qPCR and Sanger sequencing. We were able to identify high-confidence gene fusions from RNA-seq data in 53% of samples in a TCGA cohort of 161 patients. We identified 13 cases (8%) with fusions retaining the tyrosine kinase domain in the TCGA cohort and one case in the Ivy Center cohort. Recent advances in the development of tyrosine kinase inhibitors (TKIs) have demonstrated that these drugs can provide significant benefit to patients whose tumors have a specific genetic abnormality. We also identified a novel class of fusions (14%) that result in the C-terminal truncation of its 5′ partner due to fusion with non-coding RNA genes. One such case was also present in the Ivy Center cohort. This study reveals the diversity of gene fusions in GBM samples. The majority of the fusions are private fusions occurring in one patient. There are a few fusions that recur at low frequency in GBM.
Our study is the first to provide a comprehensive view of the gene fusion landscape in GBM by examining sequences from 185 patients from two independent cohorts. We successfully utilized our in-house pipeline for fusion discovery using SOLiD single-end, 50 bp RNA-seq data with a 100% validation rate. For the TCGA cohort, we used two different gene fusion detection software packages to comprehensively identify fusions from Illumina paired-end, 75 bp RNA-seq data. Ours is the first study to describe recurrent fusions involving non-coding genes. We combined copy number data with gene fusion discovery to elucidate mechanisms of the formation of gene fusions in GBM. All of the fusions detected in this study can be further visualized and analyzed on our website (http://ivygap.swedish.org/fusions).
We were able to validate all of the fusions in our SOLiD single-end RNA-seq data by using strict filtering criteria. It is likely that we may have underestimated fusions for Ivy Center data. Due to lack of access to the tissue samples, we could not determine the validation rate for our set of curated fusions in the TCGA cohort. The curated fusion set did have a significantly higher percentage of fusions associated with copy number changes relative to the low-confidence set. We applied filters to discard likely passenger fusions [34, 39], but the functional significance of these fusions still needs to be evaluated.
Singh et al. was the first study to describe multiple fusions of FGFR-TACC in GBM, reporting this phenomenon in 3 of the 97 tumors examined. They showed that the fusion protein has oncogenic activity when introduced into astrocytes and oral administration of an FGFR inhibitor prolongs the survival of mice harboring intracranial FGFR-TACC-initiated glioma . A second study by Parker et al. showed that the fusion gene is overexpressed by escaping miR-99a regulation due to loss of the 3′ UTR of FGFR3 . In their cohort, 4 out of 48 samples harbored the FGFR3 → TACC3 fusion. In our Ivy Center cohort, the FGFR3 → TACC3 fusion was detected in one out of 72 samples. We tested for this fusion in an additional 48 samples in addition to the 24 RNA-seq samples, but did not detect any fusion events. In the TCGA cohort, 2 of 161 samples harbored the FGFR3 → TACC3 fusion. Fusions of FGFR genes are identified in other cancers, including bladder cancer, cholangiocarcinoma, squamous lung cancer, breast cancer, thyroid cancer, oral cancer, head and neck squamous cell carcinoma and prostate cancer [15, 23]. Tropomyosin-Receptor Kinases (Trk) are known to play a role in cancer biology. Rearrangements of the NTRK1 gene are consistently observed in a small fraction of papillary thyroid carcinomas . We identified two cases of NTRK1 fusions in the TCGA cohort. Frattini et al.  screened 248 samples for NFASC-NTRK1 fusion but did not find any. We identified a single case of a CEP85L-ROS1 fusion in the TCGA patient samples. In a recent study by Giacomini et al. , a CEP85L-ROS1 fusion was detected for angiosarcoma. There have been two more reported cases, one angiosarcoma and one epithelioid hemangioendothelioma, with ROS1 rearrangements. ROS1 rearrangements also define a unique molecular subclass of lung cancer that may respond to an ALK inhibitor . We identified fusions of EGFR in nine patient samples from the TCGA cohort, out of which six retained the tyrosine kinase domain and resulted in a carboxyl-terminal truncation. A study by Cho et al. has shown that cetuximab prolonged the survival of intracranially xenografted mice with oncogenic EGFR carboxyl-terminal deletion mutants compared with untreated control mice . It is likely that patients with fusions of EGFR leading to carboxyl-terminal truncation will show sensitivity to EGFR inhibitors. Frattini et al.  showed that EGFR-SEPT14 fusions which occur in about 4% of GBMs was a functional gene fusion in GBM and confers mitogen independence and sensitivity to EGFR inhibition. A total of 13 cases from both cohorts have fusions of genes involved in chromatin remodeling and modification. These genes include ARID1A, ARID1B, ASH1L, CHD4, HDAC1, HMGA2, JMJD1C, KDM4B, RERE, SETD1B and YEATS4. ARID1A-MAST2 fusion has been shown to be a critical driver fusion in an MDA-MB-468 breast cancer cell line . In 27 samples, the 5′ partner gene fuses with non-coding RNA. These fusions are predicted to have a C-terminal truncation. These cases also have highly expressed non-coding RNAs that are not expressed in other samples. A recent study by Zhang et al.  discovered a signature comprising of six long non-coding RNA that predicts survival in GBM. There is now growing evidence of an oncogenic and tumor suppressive role for long, non-coding RNAs in tumor biology . Their identification in gene fusion events has thus far been neglected, as most studies focus on fusions of the coding genes.
Even though gene fusion events in GBM are abundant with scarce recurrent events, they are not random events. Majority of the fusion events occur at 7p11, 12q14-15, 1q32 and 4q12 which are also recurrently amplified regions in GBM. These fusion hotspots are consistent in both Ivy and TCGA cohorts. Also majority of the fusion events are due to unbalanced genomic rearrangements. Analysis of whole genome sequencing data also showed that 88% of genic rearrangements in GBM are associated with copy number alterations . Some of the key genes implicated in GBM biology within these hotspots are EGFR, MDM2, CDK4, PIK3C2B, MDM4 and PDGFRA. A recent study  identified a dense breakpoint pattern on 12q14-15 indicative of local chromosome instability and defined this region as “breakpoint enriched region” (BER). They showed that patients with BER pattern had poor survival and this pattern was associated with MDM2/CDK4 co-amplification. There are two other cancers, dedifferentiated liposarcomas and lung adenocarcinomas that also show MDM2/CDK4 co-amplification in 90% and 4% of cases respectively [38, 47]. All three types of cancer display distinct genomic aberration patterns in 12q14-15 region in spite of having MDM2/CDK4 co-amplification. GBM samples show shattering of the region with alternate high level deletions and gains, lung adenocarcinomas mostly contain large amplified segments and dedifferentiated liposarcomas contain multiple amplified segments (see Additional file 6). Whole genome sequencing, copy number and RNA-seq datasets show that GBMs contain deletion bridges that connect these amplified segments and generate a large number of fusion transcripts. Such complex genomic rearrangements are more prevalent on chromosome 12 but not limited to as shown in the study by Malhotra et al.  where they analyzed whole genome sequencing data of 18 GBMs. About 40% of fusion transcripts are formed due to such complex genomic rearrangements. With the advent of RNA-seq technology the list of fusion sequences in solid tumors is growing exponentially but little is known about the mechanisms that facilitate fusion events. The formation of the TMPRSS2-ERG gene fusion that occurs in about 50% of prostate cancers has been shown to be facilitated by androgen signaling which induces proximity of the TMPRSS2 and ERG genomic loci and then exposure to gamma irradiation which causes DNA double-strand breaks . The overview of the fusion landscape in GBM leads to questions about what mechanisms are responsible for generating highly site specific DNA double-strand breaks and then joining of these breaks that result in complex genomic rearrangements.