The prognostic role of intragenic copy number breakpoints and identification of novel fusion genes in paediatric high grade glioma

DNA copy number and gene expression studies have highlighted key distinctions between high grade gliomas (HGG) arising in childhood and far more commonly, much later in adult life [1‐4]. Indeed, recent exome-level sequencing initiatives have conclusively shown the existence of subgroups of HGG marked by distinct driver mutations [5], which are significantly enriched in young children (H3F3A K27M), teenagers and young adults (H3F3A G34R/V), and middle-aged adults (IDH1/2) [6]. Specific driving events for infants and elderly patients remain to be elucidated, however they too represent biological sub-entities, with infants having few genomic alterations [4], and elderly patients harbouring frequent amplification of EGFR and other genomic events [2, 3].

The identification of driving genetic alterations at the DNA copy level are necessarily most commonly focussed on assessing the amplification/deletion of genes in their entirety, and approaches to ascribe significance to genomic events make use of overlapping regions across multiple samples to find genes consistently within regions of gain/loss [7]. This approach has the result of ignoring genes for whom the breakpoint, i.e. the specific location of copy number change, is found within the coding regions. Such events may be more than mere bystanders of the “driving” aberration, and may themselves play significant roles in tumour initiation and maintenance.

One key implication of copy number breakpoints occurring within genes is the possibility of generating novel fusions. Gene fusions can occur through both intra- and inter-chromosomal translocations, bringing together coding regions from two or more genes within a single reading frame allowing expression of a novel protein. Such gene fusions are common in cancer, but have historically been thought to be largely restricted to haematological malignancies and selected solid tumours such as sarcomas. Recent evidence has overturned this, with numerous novel gene fusions being discovered in a wide range of cancer types, exemplified by the identification of common TMPRSS2:ERG fusions in prostate cancer [8] and the EML4:ALK fusion in non-small cell lung cancer [9].

The first fusion gene found in glioblastoma was the rearrangement located at an amplified region at chromosome 4q12, resulting in the fusing of the kinase domain of PDGFRA with the regulatory domains of KDR (VEGFR2) [10]. This KDR:PDGFRA was found to be activating and tumorigenic, however to date only a single additional case has been found, in a paediatric high grade glioma (pHGG) [11], and thus these fusions do not represent a common event. Another low frequency fusion has more recently been identified in approximately 3% of adult HGG, involving FGFR1 or FGFR3 partnering with TACC1 or TACC3[12]. These FGFR:TACC fusions have been shown to localize to mitotic spindle poles, have constitutive kinase activity and induce mitotic and chromosomal segregation defects and aneuploidy [12]. The types of integrated analysis that identified these mutations have also begun to identify more common rearrangements, such as numerous fusions involving EGFR, the most frequently seen partner producing the EGFR-SEPT14 fusion demonstrated to activate STAT3 signaling and confer mitogen independence and sensitivity to EGFR inhibition [13].

Such analyses are clearly proving extremely valuable in furthering our understanding of HGG biology and generating novel targets for therapeutic intervention. As similar approaches are yet to be undertaken in the paediatric setting, we have applied an algorithm designed to identify intragenic copy number breakpoints in our previously published study of DNA copy number [4]. We identify numerous potentially functional gene disruptions and a novel validated complex fusion, DHX57:TMEM178:MAP4K3.

Comprehensive copy number profiling of adult and paediatric high grade gliomas was among the first data to demonstrate the biological differences between these similar-looking histological malignancies [18]. In this context, the focus has been on large-scale genomic copy number changes. A more refined analysis of copy number and exon-level expression data has identified new insights into genomic architecture and novel fusion proteins in adult glioblastoma [12, 13]. Here we leverage a large dataset we have previously generated [4] in the paediatric disease to carry out a scan of intragenic breakpoints, leading to the identification of novel gene disruptions and candidate gene fusions.

The presence of intragenic copy number aberrations was confirmed in the vast majority of pHGG cases, and was itself prognostic, with an absence of iCNAs conferring a longer overall survival in paediatric patients. This was associated with the infant age group, known to have a better clinical outcome than older children [19], and further highlights the biological distinctiveness of this age group. By contrast, the presence of large numbers of intragenic breaks conferred a shorter survival time, but was not a result of the grade of the tumour, nor associated with a second malignancy due to radiation treatment for an earlier cancer. We had previously reported an association of post-irradiated HGG with PDGFRA amplification and chromosome 1q gain [4], so it appears these are relatively selective radiation-induced changes, rather than reflecting a generalised genomic instability in secondary tumours from these patients. Importantly, we identified an increased number of iCNA in tumours harbouring an H3F3A K27M mutation, regardless of anatomical location. This is a group of thalamic and pontine HGG associated with a particularly dismal prognosis [18], for whom understanding the mechanisms of genomic instability and the identification of novel gene disruptions is of considerable interest.

The majority of intragenic breakpoints we identified were associated with gene disruption. This includes deletions of known tumour suppressors such as RB1 and NF1, but also more novel glioblastoma associated genes. FAF1 and MTAP were both recurrently targeted by intragenic deletion events in pHGG. These genes are localised close to known cyclin-dependent kinase inhibitors and tumour suppressors CDKN1C and CDKN2A/B, respectively, but both FAF1 and MTAP have recently been proposed to harbour tumour suppressor activity in their own right. FAF1 is associated with a FAS-mediated apoptosis response and restoration of the FAF1 protein in adult glioma cell lines significantly increases cell death [20], whilst in MTAP-deficient cells, methylthioadenosine, generated during polyamine biosynthesis, is not cleaved and the salvage pathway for adenine and methionine is absent [21]. It seems that such mechanisms are also likely in a subset of paediatric tumours.

Of note we identified novel deletions in the protein phosphatase epsilon, PTPRE. This has not been reported previously, although there are several reports of the tumour suppressive capacity of the related PTPRD[22, 23]. This gene also appears targeted by intragenic deletions, and human astrocytes lacking PTPRD exhibited increased growth, as it is thought the protein usually functions to dephosphorylate the oncoprotein STAT3 [23]. The wholly intragenic microdeletions observed in CSMD3 in four cases may represent another novel mechanism of gene disruption. CSMD3 encodes a gene with multiple CUB and Sushi domains whose function is poorly understood. Recently, CSMD3 was identified as the second most frequently mutated gene (next to TP53) in lung cancer, where it was demonstrated that loss of CSMD3 results in increased proliferation of airway epithelial cells [24].

Gene disruption may also play a significant functional role when known gain-of-function oncogenes are amplified. We report numerous intragenic breakpoints which may have been overlooked in the context of identifying the ‘driver’ event within a common amplicon, but which may themselves be tumorigenic. These include disruptions of KIDINS220, a functional mediator of multiple receptor signalling pathways and essential for cortical development [25, 26]; CHIC2, frequently deleted/rearranged in myeloid malignancies [27]; and KCND2, encoding a potassium voltage-gated channel, which is expressed in both neuronal and glial cells and has been shown to regulate ERK signalling in ganglioglioma [28]. All of these gene disruptions represent novel avenues for understanding the underlying biology of pHGG.

Of most interest was the use of the iCNA algorithm to identify potential novel fusion genes, as was demonstrated in adult glioblastoma with the identification of the KDR:PDGFRA fusion [10], which we also found in a case of pHGG [11]. Our analysis nominated two potential candidates – the first we were unable to conclusively validate, CSGALNACT2:RET. Such a putative fusion would retain the kinase domain of the RET oncoprotein, but would lose the autoregulatory portion of the protein, instead fusing it to the N terminal of chondroitin sulfate N-acetyl-galactosaminyltransferase 2. Although a precise cancer-related function has not been ascribed to the latter enzyme, it is though to play an important role in morphogenesis in zebrafish models [29, 30]. Whilst not validated, oncogenic RET rearrangements and fusions are common in thyroid and lung cancer [31, 32], and the presence of infrequent activating fusions in HGG do not seem unlikely.

We were able to validate a novel complex fusion involving three genes with intragenic breakpoints and amplification/rearrangement on chromosome 2p22.1. The resulting fusion gene, DHX57:TMEM178:MAP4K3 encompasses key regulatory domains from all three proteins, though a specific function is hard to predict. The helicase properties of the DHX57 component may be a candidate for oncogenicity, with numerous other DEAD-box helicases appearing to play a role in regulation of DNA repair, apoptosis and drug sensitivity [33]. MAP4K3 has been associated with several malignancies in both an oncogenic and tumour suppressor capacity [34, 35]. In particular, one function that has been ascribed includes activation of mTOR signalling via the TORC1 complex [36], a pathway commonly activated by diverse mechanisms in pHGG [18].

In summary these data represent a key addition to our understanding of the genomic alterations driving pHGG and provide novel avenues for developing sorely-needed novel therapeutic strategies for children with these otherwise incurable tumours.