Mutations in SETD2 and genes affecting histone H3K36 methylation target hemispheric high-grade gliomas

To identify genetic drivers in samples not carrying mutations in IDH1 and H3F3A, we analyzed 60 pediatric HGG tumors [grades III (n = 9) and IV (n = 51)] using WES (44 previously reported; Table S1) [32]. As matched normal DNA was unavailable for the majority of tumors, we identified private mutations that were present in tumors but were absent from public databases (1000 genomes project [24], NHLBI exomes) and from our set of 543 control exomes, and considered these as candidate somatic mutations. We compared the frequency of private mutations in each gene between the 60 tumors and 543 controls using Fisher’s exact test and used a false discovery rate threshold (FDR) of 0.05 to correct for multiple tests. Our case–control approach effectively corrects for the background rate of mutations in each gene (which implicitly includes the length of the gene and the mutability). We filtered out variants that were predicted to be tolerated/benign/unknown by both SIFT and PolyPhen-2 [1], and identified private mutations that we considered as candidate somatic mutations. The top genes by mutation frequency are shown in Table S2. As expected, four genes previously associated with pediatric HGG showed a highly significantly number of mutations (TP53, H3F3A, ATRX, NF1) [32] (Table S2). In addition, two genes not previously reported in HGG, SETD2 and CSMD3, achieved genome-wide significance (FDR = 0.029 and 0.031, respectively). For SETD2, this significance was more striking when only truncating mutations were considered (FDR = 0.0017), as no truncating mutations were seen in 543 controls, but tumor samples had frameshift (3), nonsense (1), and splicing (1) variants. In addition, the three missense variants in tumor samples occurred at highly conserved residues and were computationally predicted as damaging by both SIFT and Polyphen scores (Fig. 1a, b). In contrast, of the seven private variants in the control samples (all missense), only one is predicted damaging by both SIFT and Polyphen. H3.3 mutations occur at two positions within the histone tail involved in key regulatory post-translational modifications, K27 (directly) and K36 (indirectly). Driver loss-of-function SETD2 mutations have recently been identified in two high-grade cancers, renal cell carcinoma [6, 12] and early T-cell precursor acute lymphoblastic leukemia [37]. The other candidate gene, CSMD3 is expressed in adult and fetal brains; however, its functions are yet unclear [33]. Hence, we focused our next efforts on SETD2 as the top candidate gene.

We expanded our sequencing analysis and next sequenced SETD2 in 123 additional gliomas of various ages and grades (Table 1; Table S1). Combining the discovery and the validation datasets, SETD2 mutations were identified in a total of 15 % of pediatric HGG (11/73) and 8 % of adult HGG (5/65), and were not seen in low-grade diffuse gliomas (0/45) (P = 0.0133; Table 1; Table S1). Except for one sample, all mutations occurred in children above the age of 12, in adolescents and in younger adults, mirroring the age range of H3.3 G34R/V and IDH1 mutations in HGG (Fig. 1a; Figure S1) [18‐21, 32, 35]. Notably, all tumors carrying SETD2 mutations were localized in the cerebral hemispheres (P = 0.0055). SETD2 mutations were mutually exclusive with H3F3A mutations (P = 0.049) in HGGs (0/70), but showed partial overlap with IDH1 R132 mutations (4/14), TP53 (4/8) and ATRX (3/9) mutations (Table S1).

SETD2 encodes the only H3K36 trimethyltransferase in humans [11, 41]. To support computational predictions of the damaging nature of SETD2 mutations, we assessed H3K36 trimethyltransferase activity in histone acidic extractions of patient tissue samples through Western blotting for H3K36me3 levels, an indicator of SETD2 activity [11]. Immunoblot analysis revealed a significant decrease in total H3K36me3 levels in SETD2-mutant gliomas (Fig. 2a), as well as a significantly decreased normalized ratio of H3K36me3 to total H3 levels in SETD2-mutant tumors (P < 0.001; Fig. 2b) showing loss-of-function as a result of SETD2 missense/truncating mutations.

GBMs with epigenetic driver mutations such as H3.3 K27M or G34R/V, as well as those with IDH1 mutations, display distinct DNA methylation profiles and clinical characteristics. They also arise in distinct anatomic compartments, with IDH1- and H3.3 G34R/V-mutant tumors being restricted to areas of the cerebral hemispheres [18‐21, 32, 35]. We and others have previously described the distinct heterogeneity of epigenetic profiles underlying HGGs including GBM [2, 35, 36]. We thus sought to characterize the DNA methylation profiles of 36 pediatric HGG tumors with mutations likely to affect K36 methylation status, using the Illumina 450K array platform as previously described [35]. SETD2 mutations yielded global DNA methylation patterns distinct from tumors with H3.3 G34R/V mutations, but which partly overlapped with IDH1-mutant methylation patterns (Fig. 3a–f). Notably, promoters at OLIG1/2 loci, characteristically hypermethylated in G34R/V-mutated samples, were not hypermethylated in SETD2 mutants (Figure S2) [2, 35].

We further investigated our dataset for mutations in other genes affecting PTM of H3K27 or H3K36 but which did not reach the statistically significant mutation levels. Eight distinct mammalian enzymes methylate H3K36 and share the catalytic SET domain, but have varying preferences for K36 residues in different methylation states (reviewed in [41]). SETD2 is the only enzyme in humans to catalyze H3K36 tri-methylation [11], while its mono- and/or di-methylation is catalyzed by NSD1, NSD2, NSD3, SETMAR, ASH1L, SMYD2 or SETD3 (reviewed in [41]). We identified one missense and one nonsense mutation in ASH1L (concurrently with SETD2 mutation) and SETD3 (1 missense mutation concurrently with SETD2 mutation). One PGBM mutant for SETD2 also had a missense mutation in UTX/KDM6A (H3K27 demethylase). This same PGBM had a missense mutation in PBRM1, a gene frequently mutated in renal cell carcinoma in association with SETD2 [39]. We also identified two mutations in KDM5C (H3K4 demethylase) (Table S1). Interestingly, no mutations in the cancer-implicated histone methyltransferase EZH2 were identified. Mutations in these genes were not prevalent enough to be statistically associated with HGGs in our sample set; however, it remains possible that they contribute to pathogenesis in a small fraction of HGG cases.

Further investigation of the exome dataset revealed previously described mutations in BRAF (V600E [30, 31], 5/60 pediatric HGG), which did not overlap with the epigenetic driver mutations we identify (Table S1; Fig. 2c). Other alterations also previously described in GBM, which may provide pathways alternative or complementary to epigenomic dysregulation, included PTEN mutations (two samples) which overlapped with H3.3 K27M while EGFR mutation or amplification (three samples) and CDKN2A mutation/loss (five samples) partially overlapped with SETD2 mutations (Table S1). Truncating mutations in the mismatch repair genes [10] MSH6 (three samples) and MSH2 (one sample) were identified and were concurrent with IDH1 (two samples) and SETD2 (three samples) mutations. Of note, SETD2 mutations were absent in a large cohort of 125 cases of medulloblastoma [15] another major group of pediatric brain tumors.

Post-translational modification of resident histones modulates the properties of chromatin, impacting cell state and differentiation and determining the outcome of virtually all DNA processes in eukaryotes. Methylation of H3K36 is a key histone mark and has been widely associated with active chromatin but also with transcriptional repression, alternative splicing, DNA replication and repair, DNA methylation and the transmission of memory of gene expression from parents to offspring during development (reviewed in [41]). We identify loss-of-function mutations in SETD2, in 15 % of pediatric and 8 % of adult high-grade gliomas in a cohort of 183 samples from all ages and grades II–IV of glioma (Fig. 1a; Table S1). We further show SETD2 mutations to be specific to high-grade tumors (P = 0.013), to HGGs located within the cerebral hemispheres (P = 0.005), and to be mutually exclusive with H3.3 mutations we [18, 32] and others [42] previously identified in pediatric high-grade astrocytomas (P = 0.049). SETD2 alterations overlapped with IDH1 mutations in 4 of 14 tumors (Table S1). Strikingly, the oncometabolite produced by IDH1 mutations inhibits a plethora of histone demethylases (KDMs) causing aberrant histone methylation at defined residues including K27 and K36 and a block to cell differentiation [4, 22, 29, 43]. We [21] and others [14] have previously shown the association of ATRX and TP53 mutations in IDH1-mutant diffuse astrocytic gliomas, and others have pointed to mutations in CIC and 1p19q loss in IDH1-mutant oligodendroglial tumors. Thus, IDH1 mutations may require other key genetic events in a specific context for full-blown tumorigenesis, which may include SETD2 mutations as suggested by our cohort. H3.3K36 methylation can be thus disrupted by H3.3 G34R/V mutation, IDH mutations and the SETD2 mutations we report herein (Fig. 2a, b). Furthermore, our current analysis suggests that this histone mark is specifically altered in hemispheric adolescent and younger adult HGG (Fig. 2c, d) [2, 18, 19, 25, 32, 34, 35], and that the functional effect differs between SETD2 and H3.3 mutations (Fig. 3; Figure S2). Future studies directed towards elucidating the importance of H3K36 methylation in cortical astrocytes and neural progenitor cells, and its dysregulation in tumorigenesis may lend insight into the regional specificity of these defects, while improved understanding of the consequences of altered chromatin remodeling induced by these mutations will help guide alternative therapeutic avenues for these deadly cancers.