Landscape of gene fusions in epithelial cancers: seq and ye shall find

Enabled by high-throughput sequencing approaches, epithelial cancers across a range of tissue types are seen to harbor gene fusions as integral to their landscape of somatic aberrations. Although many gene fusions are found at high frequency in several rare solid cancers, apart from fusions involving the ETS family of transcription factors which have been seen in approximately 50 % of prostate cancers, several other common solid cancers have been shown to harbor recurrent gene fusions at low frequencies. On the other hand, many gene fusions involving oncogenes, such as those encoding ALK, RAF or FGFR kinase families, have been detected across multiple different epithelial carcinomas. Tumor-specific gene fusions can serve as diagnostic biomarkers or help define molecular subtypes of tumors; for example, gene fusions involving oncogenes such as ERG, ETV1, TFE3, NUT, POU5F1, NFIB, PLAG1, and PAX8 are diagnostically useful. Tumors with fusions involving therapeutically targetable genes such as ALK, RET, BRAF, RAF1, FGFR1–4, and NOTCH1–3 have immediate implications for precision medicine across tissue types. Thus, ongoing cancer genomic and transcriptomic analyses for clinical sequencing need to delineate the landscape of gene fusions. Prioritization of potential oncogenic “drivers” from “passenger” fusions, and functional characterization of potentially actionable gene fusions across diverse tissue types, will help translate these findings into clinical applications. Here, we review recent advances in gene fusion discovery and the prospects for medicine. Electronic supplementary material The online version of this article (doi:10.1186/s13073-015-0252-1) contains supplementary material, which is available to authorized users.


Introduction
Recurrent chromosomal rearrangements in cancers have been described for over half a century [1,2]. The characterization of the oncogenic fusion BCR-ABL1 at t (9,22) translocation loci in chronic myeloid leukemia, which culminated in the development of a molecularly targeted therapy, provides a compelling "bench to bedside" paradigm for cancers [3,4]. Numerous gene fusions have since been defined at cytogenetically distinct loci of recurrent chromosomal aberrations in hematological malignancies and sarcomas, as well as in solid cancers, albeit much less frequently, arguably owing to technical limitations in resolving karyotypically complex, heterogeneous sub-clones in solid tumor tissues [5,6]. The serendipitous discovery of ETS family gene fusions in common prostate carcinoma [7,8], and of ALK and ROS kinase fusions in lung cancer [9,10] through transcriptomic and proteomic approaches, bypassing chromosomal analyses, provided a strong fillip to the search for gene fusions in common solid cancers and pointed to alternative approaches to gene fusion discovery. Developments in high-throughput sequencing techniques over the past decade [11] have made possible a direct, systematic discovery of gene fusions in solid cancers [12][13][14], rapidly revealing a diverse genomic landscape. Gene fusions have now been identified in several common carcinomas, including those of the prostate, lung, breast, head and neck, brain, skin, gastrointestinal tract, and kidney, which alongside the widely documented gene fusions in thyroid and salivary gland tumors support the notion that gene fusions are integral to the genomic landscape of most cancers.
Here, we review the emerging landscape of gene fusions across solid cancers, focusing on the recent spurt of discoveries made through sequencing. We review common features of "driver" fusions (those that contribute to tumor progression), the major functional classes of fusions that have been described, and their clinical, diagnostic and/or therapeutic implications.

Detection of gene fusions in carcinoma
The first gene fusions to be defined in solid cancers, RET/PTC [15] and NTRK1 [16] rearrangements in papillary thyroid carcinoma were identified through a "transformation assay" using cancer genomic DNA transfected into murine NIH3T3 cells, followed by retrieval and analysis of human genomic DNA from transformed cells [17]. More typically, karyotyping and cytogenetic analysis of recurrent translocations helped define early gene fusions in solid cancers, such as CTNNB1-PLAG1 [18] and HMGA2 fusions [19] in salivary gland pleomorphic adenomas, PRCC-TFE3 in renal cell carcinomas [20], and ETV6-NTRK3 fusion in secretory breast carcinoma [21]. Incorporating more molecular approaches, a recurrent 2q13 breakpoint locus, t(2;3)(q13;p25), in follicular thyroid carcinoma was fine mapped using yeast artificial chromosomes, and cloned through 3′ rapid amplification of cDNA ends (RACE) of the candidate PAX8 cDNA, leading to characterization of the PAX8-PPARγ gene fusion [22]. Anticipating high-throughput genomics approaches, an expressed sequence tag (EST) mapping to the recurrent chromosomal breakpoint at t(15;19)(q13;13.1) in midline carcinoma was identified from an EST database and cloned through RACE to identify the pathognomonic gene fusion BRD4-NUT [23]. The gene fusions defined in solid cancers thus far were localized at cytogenetically distinct, recurrent chromosomal aberrations, and were largely confined to relatively rare subtypes of solid cancers [5].
However, between 2005 and 2007, independent of a priori evidence of genomic rearrangements, recurrent gene fusions involving ETS family genes were discovered in prostate cancer, based on analysis of genes displaying outlier expression [7,8,24]. Around the same time, a transformation assay with a cDNA expression library (not genomic libraries [17]) from a lung adenocarcinoma sample led to the discovery of EML4-ALK fusions [10], and a high-throughput phosphotyrosine signaling screen of lung cancer cell lines and tumors identified SLC34A2-ROS1 fusions in non-small-cell lung carcinoma (NSCLC) [9]. Thus, analyses of cancer RNA and proteins provided a critical breakthrough in the identification of oncogenic gene fusions in common carcinoma. In Fig. 1, we summarize the timeline of gene fusion discoveries, 100 years since Boveri's prescient hypothesis that malignant tumor growth is a consequence of chromosomal abnormalities, including "combinations of chromosomes" [25].

Next-generation sequencing
High-throughput sequencing of tumor samples provides a direct readout of chimeric sequences corresponding to putative gene fusions, and the available depth of coverage helps uncover even relatively minor, sub-clonal events. In a proof of principle study, high-throughput genomic sequencing was used to identify several gene fusions in a panel of breast cancer cell lines and tissues [14]. However, considering that only a small subset of genomic breakpoints correspond to gene fusions encoding fusion transcripts or proteins, alternative approaches were explored. In a directed approach, focusing on chimeric transcripts as the readout of "expressed" gene fusions, Maher and colleagues used coupled short-and long-read transcriptome sequencing [12] and paired-end transcriptome sequencing [13] to detect chimeric RNAs that could be analyzed to characterize gene fusions. RNA sequencing has since been widely used in the discovery of numerous gene fusions in diverse epithelial cancers. Additionally, paired-end tag [26] and chromatin interaction analysis by paired-end-tag sequencing have been employed for gene fusion discovery [27], as well as phosphoproteome analysis, as in the discovery of a SND1-BRAF fusion in a gastric carcinoma sample [28]. 4. High-throughput sequencing enables systematic discovery of gene fusions with high sensitivity and precision. 5. High-throughput sequencing often identifies multiple gene fusions in individual samples, presenting a challenge to distinguish oncogenic "driver" from unimportant "passenger" aberrations. 6. Chimeric RNAs expressed independent of chromosomal rearrangements are frequently observed in cancer (and benign) tissues. 7. Functionally recurrent gene fusions provide clinically relevant molecular subclassifications of existing morphological categories of tumors.
8. Functionally recurrent gene fusions that are seen across tissue types define functionally distinct molecular subtypes of cancers. 9. Gene fusions represent personalized therapeutic targets and prognostic and diagnostic markers.
The DNA-or protein-based methods, however, are not as commonly used as RNA sequencing, likely owing to several additional, specialized steps that are involved.
Interestingly, RNA sequencing has also identified a class of chimeric RNAs that do not involve chromosomal aberrations. For example, "read-through" chimeric SLC45A3-ELK4 transcripts, such as those detected in prostate cancer, result from runaway transcription of the androgen-inducible, prostate-specific gene SLC45A3 into ELK4, the adjacent ETS family gene in the same orientation [12,[29][30][31]. Similarly, the VTI1A-TCF7L2 fusion, originally identified through genomic sequencing of colorectal carcinoma (CRC) samples [32], was found in a follow-up study using RNA analyses to be quite prevalent in other cancers, as well as in benign samples [33]. Chimeric transcripts not associated with genomic translocation have also been observed between non-contiguous genes. Guerra and colleagues identified CCND1-TACSTD2 (TROP2) chimeric mRNA that involves genes located on different chromosomes in subsets of ovarian, breast, gastrointestinal, and endometrial cancers [34]. The functional significance of these RNA chimeras is not clear at present, as their expression is typically seen to be relatively non-specific. Fig. 1 Timeline of gene fusion discoveries. A timeline representation of salient gene fusion discoveries starting with 1914, the year that marked the publication of Boveri's monograph "Zur Frage der Entstehung maligner Tumoren", in which he proposed that aberrant "combinations of chromosomes" underlie malignant transformation [25]. The top bar shows recurrent chromosomal rearrangements or gene fusions in hematological (purple) and soft tissue (green) malignancies, and the bottom bar shows gene fusions in relatively rare (blue) and those in common (red) epithelial cancers. ACC adenoid cystic carcinoma, AML acute myeloid leukemia, ALL acute lymphoblastic leukemia, APL acute promyelocytic leukemia, cholangio cholangiocarcinoma, CML chronic myeloid leukemia, CRC colorectal carcinoma, MLL mixed lineage leukemia, PLGA pediatric low grade astrocytoma, Ph Philadelphia chromosome

Driver and passenger gene fusions
High-throughput sequencing of cancer samples frequently identifies multiple gene fusions in individual samples, often presenting a challenge for identifying potentially oncogenic driver fusions among irrelevant passenger aberrations. Some useful generalizations have emerged from multiple analyses: first, driver fusions are typically marked by a continuous open reading frame (ORF) that retains functional domains, such as the kinase domain in gene fusions involving oncogenic kinases, or DNA-binding domains in the case of transcription factors; second, some fusions display loss of auto-inhibitory domains (for example, loss of the N-terminal inhibitory domain in the product of BRAF fusions, or loss of 3′ UTR sequences in FGFR or HMGA2 fusions that serve as binding sites for inhibitory microRNAs). Yet other types of fusions juxtapose the promoter of certain tissue-specific, inducible or highly expressed genes; for example, the prostate-specific, androgen-inducible genes TMPRSS2 or SLC45A3 fused in frame with the proto-oncogenes ERG or BRAF, respectively, generate the TMPRSS2-ERG and SLC45A3-BRAF gene fusions in prostate cancer.
In the case of novel gene fusions involving less characterized genes, distinguishing candidate driver fusions from random events is complicated by the many false positive candidates resulting from alignment artifacts, such as multi-mapping of reads owing to homologous (pseudogenes) and/or repetitive sequences, and sequencing artifacts due to errors in library generation (particularly ligation and PCR artifacts) and sequencing. Incorporating these considerations, and additional bioinformatics filters, various bioinformatics pipelines have been developed to help prioritize fusion candidates from next-generation sequencing (NGS) data, including Chimerascan [35], FusionSeq [36], DeFuse [37], TopHat-Fusion [38], PRADA [39], and JAFFA [40]. While useful to help reduce the number of false candidates, the output from bioinformatics pipelines needs to be further validated, preferably followed by functional assays, before designating candidate gene fusions as novel driver aberrations. Recurrence of fusions, fusion partners or partner gene families in gene fusion databases also helps to prioritize candidate fusions. Once validated, screening for novel gene fusions in larger cohorts of samples employs quantitative RT-PCR or more recent techniques such as nano-string-based detection [41][42][43].
These gene fusions in solid cancers encompass the diversity of fusion architectures, as shown in Fig. 2 and Box 2, and represent a spectrum of functional categories, including those described earlier such as kinases and transcription factors, as well as those involving newer pathways and loss-of-function fusions (discussed later). Notably, even as numerous novel gene fusions are being discovered fairly rapidly, most of these are either non-recurrent singletons, or are seen to recur at exceedingly low frequency in tumor subtypes or to recur across tumor types (Table 1). Incidentally, gene fusions displaying molecular recurrence involving both 5′ and 3′ partner genes, as in TMPRSS2-ERG, EML4-ALK, and BRD4-NUT, are relatively few. A large number of fusions display recurrence of a fusion gene in combination with multiple different partners; for example, BRAF/RAF1 [76,79,82,83] and FGFR1/2/3 [88][89][90][91][92][93][94] are fused to several different 5′ partners across different tissue types (Additional file 1). This heterogeneity is likely reflective of the diverse tissue-physiological milieu in which these oncogenes impart selective advantage to the cancer cells. Conversely, some lineage-specific genes are seen to serve as 5′ partners across multiple different 3′ genes; for example, TMPRSS2 and SLC45A3 in prostate cancer have been observed as 5′ partners of ERG, ETV1, ETV4, ETV5, BRAF, and ELK4 (Table 1 and Additional file 1). Another type of observed "recurrence" involves isoforms of a gene familyfor example, ETV1/2/3/4/5, FGFR1/2/3, BRAF/RAF1, BRD3/4, CRTC1/CRTC3, and NTRK1/3as fusion partners. Considering that individual fusions may be observed relatively rarely (even uniquely), the potential functional consequences of gene fusions assumes priority over considerations of recurrence.

Functional consequences of gene fusions
Functionally distinct molecular classes of gene fusions that are shared across tumor types can be identified in solid cancers.

Transcription factors
Gene fusions involving dysregulated expression of transcription factors include ETS family gene fusions, seen in approximately 50 % of all prostate cancers and probably one of the most prevalent transcription factor gene fusions in common epithelial cancers. Among these, ERG represents the most common fusion partner and ETV1 the most promiscuous, with a dozen or more different fusion partners described to date (Additional file 1) [24,96].
Other gene fusions involving transcription factors include NUT (or NUTM1), POU5F1, MAML2, NFIB, PLAG1, TFE3, NOTCH, and PAX8 fusions, imparting spatially and/or stochastically dysregulated expression in multiple different cancer types. NOTCH1 and NOTCH2 fusions result in dysregulated transcriptional outcomes, because after ligand activation, the NOTCH intracellular domain (NICD) forms a transcriptional activator complex, activating genes involved in differentiation, proliferation and apoptosis, and those associated with carcinogenesis. MAML2 acts as a transcriptional coactivator for NOTCH proteins by amplifying NOTCHinduced transcription of HES1. TFE3, which belongs to the MITF/TFE family of basic helix-loop-helix leucine zipper transcription factors, is involved in TGF-βinduced transcription, and has important roles in cell

Box 2. Mechanisms of generation of gene fusions
An overview of the genomic architecture of gene fusions reveals that fusions may result from insertion, deletion, inversion, or tandem duplication or amplification, and may involve the same chromosome (intra-chromosomal) or different chromosomes (inter-chromosomal) ( Fig. 2). A majority of chromosomal rearrangements have been associated with intra-chromosomal tandem duplications and amplifications in multiple whole-genome sequencing studies [14,26,80,150]. Micro-homologies and repeat elements have been associated with loci of recurrent break points [151]. In an analysis of RAF family gene fusion breakpoints in low-grade astrocytomas, tandem duplications generated by microhomology-mediated break-induced replication were identified as the mechanism of generation of fusions [74].
Spatial proximity between distant chromosomal loci has been associated with chromosomal rearrangements, as observed between RET and the H4 genes located 30 megabases (Mb) apart on chromosome 10, involved in RET gene fusions in papillary thyroid carcinoma [152]. This proximity may be induced by genotoxic stress; for example, androgen stimulation coupled with the genotoxic stress of radiation was shown to generate gene fusions through "induced proximity" between TMPRSS2 and ERG (located on chromosome 21q22.2, approximately 3 Mb apart) as well as between TPMRSS2 and ETV1 (located on chromosome 7) [153,154] (Fig. 3a).
Another phenomenon, called chromothripsis, describes the frequent occurrence of massive chromosomal aberrations localized to only one or two chromosomes, with fragments of chromosome joined randomly [155,156]. Chromothripsis may be responsible for the generation of numerous, apparently random passenger gene fusions that are retained in the multiclonal cells of epithelial cancers, as well as loss-offunction fusions involving tumor suppressors, likely involving the non-homologous end-joining DNA repair system (Fig. 3b).
Several cancer-causing viruses, such as Epstein-Barr virus (EBV), Kaposi's sarcoma-associated herpesvirus (KSHV), human papilloma virus (HPV), hepatitis B and C viruses (HBV and HCV), and Merkel cell polyomavirus (MCV), integrate into human genomic DNA at defined hotspots as well as seemingly randomly [157]. Viral integration events have been associated with chromosomal aberrations, such as MYC amplification in HPVpositive genital carcinoma [158], and not uncommonly, loss of gene function [159,160] or gene fusions involving viral-human sequences have been reported [161,162]. The recent report of a recurrent gene fusion of UBR5 on 8q22.3 and ZNF423 on 16q12.1 (UBR5-ZNF423) in 8 % of EBVassociated primary nasopharyngeal carcinomas suggests a driver function of this gene fusion in a subset of nasopharyngeal cancers [163]. The fusion retains all the functional domains of HMGA2, and removes the 3′ UTR sequence that contains several inhibitory let7 microRNA binding sites. Absence of the Let-7-regulated 3′ UTR in the fusion transcript results in overexpression of HMGA2 that is sufficient for neoplastic transformation [19] FGFR-PLAG1 FGFR is the 5′ partner, which, without its kinase domain, provides the promoter to drive the expression of the 3′ partner, PLAG1 This FGFR fusion product does not include the FGFR kinase domain, and therefore is not a target for FGFR inhibitors [91] Adenoid cystic carcinomas (salivary glands, lacrimal glands, ceruminal glands; also breast)     NFIB binds viral and cellular promoters activating transcription and replication. POU5F1 and PAX8 are homeobox-containing transcription factors, a family of genes that play a role in cell fate and differentiation programs, and whose role in cancer is well recognized, particularly PAX8 in thyroid cancer [22].

Other functional classes
Metabolic enzymes CD44-SLC1A2/EAAT2 gene fusions are detected in 1-2 % of gastric cancers involving the glutamate transporter SLC1A2 [85], and cause intracellular accumulation of glutamate, a growth-promoting amino acid associated with oncogenic functions [97,98]. Thus, this gene fusion may be establishing a pro-oncogenic metabolic milieu, akin to the increased levels of sarcosine reported in prostate cancer [99]. involving R-spondin family genes, EIF3E-RSPO2 in two cases and PTPRK-RSPO3 in five cases [84]. Both these gene fusions retained the functional domain of the Rspondins that are known to be agonists of the canonical Wnt/β-catenin signaling pathway. Additionally, the LACTB2-NCOA2 chimeric transcript detected in 6 of 99 (6.1 %) colorectal cancer cases led to disruption of NCOA2 expression, thus activating the Wnt/β-catenin pathway [100]. Recently, R-spondin fusions such as GRHL2-RSPO2 were described in prostate cancer as well [83].

TGF-β pathway
Recently, fusions involving SKIL (which encodes a SMAD inhibitor) 3′ to androgen-regulated promoters such as TMPRSS2, SLC45A3, and ACPP, were found in 6 of 540 (1.1 %) prostate cancers and one cell line xenograft, LuCaP-77 [101]. SKIL overexpression in these tumors was associated with upregulation of the TGF-β pathway, likely providing the oncogenic mechanism in these tumors.

Chromatin modifier genes
In an analysis of fusion transcripts observed in TCGA data across multiple tumor types, fusions involving chromatin modifier genes, including histone methyltransferase and histone demethylase genes, were identified in 111 samples (2.5 %) [89]. Chromatin modifier genes are potential therapeutic targets and these gene fusions thus represent a novel class of potentially actionable aberrations.

Loss-of-function gene fusions
While most reported gene fusions pertain to gain-offunction aberrations imparting neoplastic phenotypes, with high-throughput sequencing, fusions resulting in loss of function of tumor suppressors such as TP53 and PTEN have been identified as well [102]. The LACTB2-NCOA2 fusion in colorectal cancer leads to disruption of NCOA2, which encodes an inhibitor of the Wnt/β-catenin pathway [100], thus acting to promote carcinogenesis.

Gene fusion signatures in personalized medicine of epithelial cancers
Some gene fusions are associated with distinct subtypes of carcinoma, while others have been detected across different tissues or lineages, defining molecular subsets of cancers transcending morphological distinctions.

Recurrent gene fusions as biomarkers of subtypes of solid cancers
Some of the salient gene fusions that define molecular subtypes of epithelial cancers within specific organs or tissue types are summarized in Table 1. The ETV6-NTRK3 fusion is a diagnostic biomarker of secretory breast carcinoma, as well as the acinic cell carcinoma or cystadenocarcinoma recently designated as "mammary analog secretory carcinoma of salivary glands" (MASC) [21,103]. BRD-NUT fusions define NUT midline carcinoma [104,105]. CRTC-MAML2 fusions are the defining molecular aberration of mucoepidermoid carcinoma (MEC) [106,107]; translocation-negative MECs are proposed to be designated as a distinct subgroup of adenosquamous carcinoma [108]. CRTC-MAML fusions are also found in MEC of the lung [109][110][111][112], cervix [113], thyroid glands and oral cavity [114], as well as in clear cell hidradenoma of the skin [115,116]. In all cases, MAML2 fusions characterize benign or low-grade tumors, and for reasons not described so far have been associated with a favorable prognosis [117]. Interestingly, pulmonary MECs have shown clinical response to gefitinib in the absence of sensitizing EGFR mutations, suggesting a potential connection with CRTC-MAML2 and the possibility of therapeutic application in other MECs harboring this fusion [110,118]. The diagnostic subclass of adenoid cystic carcinomas, including salivary gland and breast cancer, is characterized by MYB-NFIB gene fusions [119,120]. Fusions defining subtypes within a cancer include RET and NTRK gene fusions in subsets of papillary thyroid carcinoma [121], while PAX8-PPARγ fusions characterize subsets of follicular thyroid carcinoma [22,122]. ETS family gene fusions, primarily including ERG (and less frequently, ETV1, ETV4, ETV5 or FLI1), are found in approximately 50 % of prostate cancers, the most common fusion being TMPRSS2-ERG.
The EWSR1-ATF1 fusion found in hyalinizing clear cell carcinoma of the salivary glands, a rare and indolent tumor, can potentially be used as a molecular marker of this subtype that is histologically similar to the more aggressive MEC [123]. Gene fusions or fusion partners found across tissue types are common in solid cancers. The EML4-ALK fusion, initially identified in lung cancer [9,10] has since been reported in breast cancer [124], colorectal carcinomas [66,124], and in pediatric renal medullary carcinoma that affects young African-Americans with the sickle cell trait [125,126]. Similarly, RET fusions, first characterized in thyroid cancer, are widely observed in lung cancers, and the EWSR1-POU5F1 fusion was detected in two rare epithelial tumors, hidradenoma of the skin and MEC of the salivary glands [127].
Fusions involving FGFR tyrosine kinase family genes have also been observed across diverse cancers [88]. The first FGFR fusion observed in epithelial cancers, FGFR1-PLAG1, was found in a subset of pleomorphic salivary gland adenomas, and involves FGFR1 as the 5′ partner upstream of PLAG1, the known driver of salivary gland tumors [91]. Curiously, this fusion excludes the tyrosine kinase domain of FGFR. Fusions that retain the tyrosine kinase domain of FGFR include FGFR3-TACC3 in glioblastoma [92,129]. Subsequently, diverse FGFR fusions, all retaining the tyrosine kinase domain, have been observed in bladder, lung, breast, thyroid, oral, and prostate cancers, involving FGFR1, 2, or 3 either as the 5′ or 3′ partners [88,94].

Some gene fusions provide personalized therapeutic targets
In Additional file 2 we summarize recent clinical trials involving gene fusions in epithelial cancers. The RET inhibitor vandetanib shows antiproliferative activity in RET-mutant medullary thyroid cancer (MTC) [130], and was recently approved by the US Food and Drug Administration for treatment of metastatic MTC. Sensitivity to vandetanib was also observed in RET-fusionpositive papillary thyroid carcinoma [131] and lung cancer cells [68,132]. Treatment with Pfizer's kinase inhibitor crizotinib (PF02341066) led to a dramatic clinical response in EML4-ALK-positive NSCLC patients [133,134], as well as in one patient with an SLC34A2-ROS1fusion-positive tumor [58]. Unfortunately, resistance is inevitably observed, owing to mutations in the kinase domain [134,135], or ALK gene fusion amplification, KIT amplification or increased auto-phosphorylation of EGFR [136]. This is representative of the challenge of treating solid cancers and argues for the development of combinatorial therapeutic approaches from the start rather than sequentially, as is the practice currently. RAF or MEK inhibitors represent potential precision therapeutic options for several solid cancers with the diverse RAF family gene fusions described earlier. Several FGFR inhibitors currently in clinical trials represent potential therapeutics for cancers harboring FGFR fusions across multiple cancer types, including bladder cancer, prostate cancer, and others [88,90,94,137]. The rare PIK3C family gene fusions in prostate cancer (for example, TBXLR1-PIK3CA and ACPP-PIK3CB) show overexpression of the PI3KC genes and may be sensitive to PIK3CA inhibitors [83].
For treatment of secretory breast carcinoma expressing the ETV6-NTRK3 fusion, therapeutic targeting of the downstream signaling axis of IGF1R, using the IGIFR/INSR kinase inhibitors BMS-536924 and BMS-754807 that are currently in clinical trials, was found to be effective [138]. Breast cancer cells expressing NOTCH fusion products that retain the γ-secretase cleavage site were sensitive to γ-secretase inhibitor (GSI) in culture, and treatment with GSI reduced tumor growth in vivo [86]. On the other hand, breast cancer cells harboring NOTCH fusions that encode NICD independent of the γ-secretase cleavage site were insensitive to GSI.
In a recent clinical sequencing study of 102 pediatric cancers, among 37 non-sarcoma solid cancers, several functional gene fusions were identified, including TFE3 fusions in a colorectal cancer (SFPQ-TFE3) and renal cell cancer (ASPSCR1-TFE3)both cases were treated with pazopanib, the latter displaying stable disease for 10 months [139].
Efforts to target several other gene fusions are underway. The newly developed bromodomain inhibitors that have shown dramatic efficacy in hematological malignancies [140,141] are now being tested in multiple clinical trials for NUT midline carcinoma characterized by BRD3/4-NUT gene fusions, which represent a rare but highly aggressive class of tumors with no effective treatment currently available [104]. Also, the R-spondin fusions observed in colorectal and prostate cancer may be sensitive to Wnt pathway antagonist porcupine inhibitors [142].
Gene fusions involving ETS transcription factors have been utilized in diagnostic applications. A non-invasive assay system has been developed based on the detection of TMPRSS2-ERG fusion transcripts in urine samples from patients, which in combination with the detection of urine PCA3 improved the performance of the multivariate Prostate Cancer Prevention Trial risk calculator in predicting cancer on biopsy [143]. Detection of TMPRSS2-ERG in circulating tumor cells in therapynaive patients and in castration-resistant prostate cancer patients following treatment suggests potential applications in non-invasive monitoring of the therapeutic response [144]. While therapeutic targeting of transcription factor oncogenes is intrinsically challenging, on the basis of the interaction of ERG with the DNA repair enzyme PARP1 and DNA protein kinase DNA-PKc, use of PARP inhibitors was shown to inhibit growth of TMPRSS2-ERG-positive prostate cancer xenografts [145]. Additionally, PARP inhibition was associated with radiosensitization of TMPRSS2-ERG-positive prostate cancer cells [146,147]. These experimental leads point to possible therapeutic avenues targeting a prevalent gene fusion in a common carcinoma.

Perspectives and discussion
Genomic or transcriptomic sequencing has virtually supplanted molecular and cytogenetic techniques as the primary modality for discovery of gene fusions, and detection of gene fusions is increasingly incorporated into the standard workflow for genomic characterization of tumors in both research and clinical settings. Transcriptome sequencing has been useful in helping to identify expressed gene fusions based on evidence of the fusion of exon boundaries, but putative promoter fusions that do not generate chimeric transcripts are likely to go undetected. Furthermore, typically recurrent gene fusions characterized in cancers represent gain-of-function events arising from the juxtaposition of cell-type-or lineage-specific regulatory elements and proto-oncogenes, or novel combinations of functional domains derived from two proteins that provide combinatorial or additive functionalities to normal genes. However, NGS data also reveal less frequently described loss-of-function chimeras involving tumor suppressor genes such as TP53, PTEN, and others. A systematic analysis of loss-of-function gene fusions could identify additional cancer samples with loss of tumor suppressors that might be currently going unreported, and could help broaden our understanding of the role of gene fusions in cancer.
The rapid increase in detection of gene fusions across cancers has spawned multiple discovery and prioritization pipelines to help distinguish bona fide functional gene fusions from random chimeras (and experimental artifacts). However, the development of diverse pipelines following different analysis parameters underscores a need for standardization of the vocabulary and information content in recording and reporting gene fusions, along the lines of the Minimum Information About a Microarray Experiment [148,149]. Furthermore, even as bioinformatics analyses help prioritize fusion candidates, the "recurrence" of fusion genes and/or retention of functional domains provide the most compelling rationale for functional characterization.
The detection of distinct gene fusions across subtypes of common carcinoma also provides a basis for molecular subclassification of these cancers. Recurrent gene fusions that characterize distinct subtypes of cancers include BRD4-NUT in NUT midline carcinoma, ETV6-NTRK3 in secretory breast carcinoma, CRTC-MAML2 fusions in mucoepidermoid carcinoma, and RAF family fusions in pilocytic astrocytomas. It is expected that as more and more carcinomas are analyzed by sequencing, additional subclasses may be recognized on the basis of whether the detected molecular aberrations are driver fusions. Importantly, the emerging landscape of gene fusions in solid cancers also reveals many gene fusions involving oncogene families or isoforms that are seen across multiple tumor types or subtypes, for example, fusions involving RAF and FGFR family genes. This supports the notion that a molecular classification of tumors in terms of driver fusions (or SNVs) may complement histopathological descriptions.
Many oncogenes involved in gene fusions (for example, RET, BRAF, ALK, NOTCH or PIK3CA/B) are also known to harbor activating mutations. However, fusions and mutations tend to be mutually exclusive. This indicates that either fusions or activating mutations can independently provide oncogenic function, and that either of these aberrations may render the tumors sensitive to therapeutic targeting. Thus, for example, MEK inhibitors that have been found to be useful for tumors with a BRAF activating mutation may also benefit tumors with the BRAF fusion.
The development of technologies that enable the systematic detection of molecular aberrations in cancer has profound clinical implications, as high-throughput sequencing of individual tumor samples is expected to become available as a routine diagnostic modality (as for whole-body PET scans or MRI) in the not-too-distant future. Considering the important diagnostic and therapeutic implications, the integration of approaches for the detection of driver gene fusions into cancer genomics pipelines is crucial for precision cancer medicine.

Additional files
Additional file 1: Recurrent gene fusions in epithelial cancers. Summary of recurrent gene fusions in epithelial carcinoma across different tissues. a Gene fusions with common 5′ and 3′ genes. b Multiple 5′ partners with common 3′ genes. c Common 5′ gene partners with multiple 3′ genes.

Competing interests
The University of Michigan has filed for a patent on recurrent gene fusions in prostate cancer and AMC is named as a co-inventor. The technology has been licensed to Hologic Inc. to develop a molecular diagnostic.