Main

Brain tumours are the most common cause of childhood oncological death, and medulloblastoma is the most common malignant paediatric brain tumour. Current medulloblastoma therapy including surgical resection, whole-brain and spinal cord radiation, and aggressive chemotherapy supplemented by bone marrow transplant yields five-year survival rates of 60–70%1. Survivors are often left with significant neurological, intellectual and physical disabilities secondary to the effects of these nonspecific cytotoxic therapies on the developing brain2.

Recent evidence suggests that medulloblastoma actually comprises multiple molecularly distinct entities whose clinical and genetic differences may require separate therapeutic strategies3,4,5,6. Four principal subgroups of medulloblastoma have been identified: WNT, SHH, Group 3 and Group 4 (ref. 7), and there is preliminary evidence for clinically significant subdivisions of the subgroups3,7,8. Rational, targeted therapies based on genetics are not currently in use for medulloblastoma, although inhibitors of the Sonic Hedgehog pathway protein Smoothened have shown early promise9. Actionable targets for WNT, Group 3 and Group 4 tumours have not been identified4,10. Sanger sequencing of 22 medulloblastoma exomes revealed on average only 8 single nucleotide variants (SNVs) per tumour11. Some SNVs were subgroup-restricted (PTCH1, CTNNB1), whereas others occurred across subgroups (TP53, MLL2). We proposed that the observed intertumoural heterogeneity might have underpowered prior attempts to discover targets for rational therapy.

The Medulloblastoma Advanced Genomics International Consortium (MAGIC) consisting of scientists and physicians from 46 cities across the globe gathered more than 1,200 medulloblastomas which were studied by SNP arrays (n = 1,239; Fig. 1a, Supplementary Fig. 1 and Supplementary Tables 1–3). Medulloblastoma subgroup affiliation of 827 cases was determined using a custom nanoString-based RNA assay (Supplementary Fig. 2)12. Disparate patterns of broad cytogenetic gain and loss were observed across the subgroups (Fig. 1b and Supplementary Figs 3, 7, 8, 10 and 11). Analysis of the entire cohort using GISTIC2 (ref. 13) to discover significant ‘driver’ events delineated 62 regions of recurrent SCNA (Fig. 1c, Supplementary Fig. 4 and Supplementary Tables 4 and 5); analysis by subgroup increased sensitivity such that 110 candidate ‘driver’ SCNAs were identified, most of which are subgroup-enriched (Fig. 1c–e and Supplementary Table 6).

Figure 1: Genomic heterogeneity of medulloblastoma subgroups.
figure 1

a, The medulloblastoma genome classified by subgroup. b, Frequency and significance (Q value ≤ 0.1) of broad cytogenetic events across medulloblastoma subgroups. c, Significant regions of focal SCNA identified by GISTIC2 in either pan-cohort or subgroup-specific analyses. d, e, Recurrent high-level amplifications (d, segmented copy number (CN) ≥ 5) and homozygous deletions (e, segmented CN ≤ 0.7) in medulloblastoma. The number of genes mapping to the GISTIC2 peak region (where applicable) is listed in brackets after the suspected driver gene, as is the frequency of each event.

PowerPoint slide

Twenty-eight regions of recurrent high-level amplification (copy number ≥ 5) were identified (Fig. 1d and Supplementary Table 7). The most prevalent amplifications affected members of the MYC family with MYCN predominantly amplified in SHH and Group 4, MYC in Group 3, and MYCL1 in SHH medulloblastomas. Multiple genes/regions were exclusively amplified in SHH, including GLI2, MYCL1, PPM1D, YAP1 and MDM4 (Fig. 1d). Recurrent homozygous deletions were exceedingly rare, with only 15 detected across 1,087 tumours (Fig. 1e). Homozygous deletions targeting known tumour suppressors PTEN, PTCH1 and CDKN2A/B were the most common, all enriched in SHH cases (Fig. 1e and Supplementary Table 7). Novel homozygous deletions included KDM6A, a histone-lysine demethylase deleted in Group 4. A custom nanoString CodeSet was used to verify 24 significant regions of gain across 192 MAGIC cases, resulting in a verification rate of 90.9% (Supplementary Fig. 5). We conclude that SCNAs in medulloblastoma are common, and are predominantly subgroup-enriched.

Subgroup-specific SCNAs in medulloblastoma

WNT medulloblastoma genomes are impoverished of recurrent focal regions of SCNA, exhibiting no significant regions of deletion and only a small subset of focal gains found at comparable frequencies in non-WNT tumours (Supplementary Figs 4, 6 and Supplementary Table 8). CTNNB1 mutational screening confirmed canonical exon 3 mutations in 63 out of 71 (88.7%) WNT tumours, whereas monosomy 6 was detected in 58 out of 76 (76.3%) (Supplementary Fig. 6; Supplementary Table 9). Four WNT tumours (4/71; 5.6%) had neither CTNNB1 mutation nor monosomy 6, but maintained typical WNT expression signatures. Given the size of our cohort and the resolution of the platform, we conclude that there are no frequent, targetable SCNAs for WNT medulloblastoma.

SHH tumours exhibit multiple significant focal SCNAs (Fig. 2a, Supplementary Figs 12, 15, 16 and Supplementary Tables 10 and 11). SHH enriched/restricted SCNAs included amplification of GLI2 and deletion of PTCH1 (Fig. 2a, e, f)10. MYCN and CCND2 were among the most frequently amplified genes in SHH (Supplementary Table 6), but were also altered in non-SHH cases. Genes upregulated in SHH tumours (that is, SHH signature genes) are significantly over-represented among the genes focally amplified in SHH tumours (P = 0.001–0.02, permutation tests; Supplementary Fig. 9). Recurrent amplification of SHH signature genes has clinical implications, as amplification of downstream transcriptional targets could mediate resistance to upstream SHH pathway inhibitors14.

Figure 2: Genomic alterations affect core signalling pathways in SHH medulloblastoma.
figure 2

a, GISTIC2 significance plot of amplifications (red) and deletions (blue) observed in SHH. The number of genes mapping to each significant region are included in brackets and regions enriched in SHH are shaded red. b, c, Recurrent amplifications of PPM1D (b) and PIK3C2B/MDM4 (c) are restricted to SHH. d, Fluorescence in situ hybridization (FISH) validation of MDM4 amplification. e, SHH signalling, TP53 signalling and RTK/PI3K signalling represent the core pathways genomically targeted in SHH. P values indicate the prevalence with which the respective pathway is targeted in SHH versus non-SHH cases (Fisher’s exact test). Frequencies of focal and broad (parentheses) SCNAs are listed. f, Mutual exclusivity analysis of focal SCNAs in SHH. g, Clinical implications of SCNAs affecting MYCN, GLI2 or PTCH1 in SHH (log-rank tests).

PowerPoint slide

Novel, SHH-enriched SCNAs included components of TP53 signalling, including amplifications of MDM4 and PPM1D, and focal deletions of TP53 (Fig. 2a–e). Targetable events, including amplifications of IGF signalling genes IGF1R and IRS2, PI3K genes PIK3C2G and PIK3C2B, and deletion of PTEN were restricted to SHH tumours (Fig. 2a, c, e). Importantly, focal events affecting genes in the SHH pathway were largely mutually exclusive and prognostically significant (Fig. 2f, g). Many of the recurrent, targetable SCNAs identified in SHH medulloblastoma (IGF1R, KIT, MDM4, PDGFRA, PIK3C2G, PIK2C2B and PTEN) have already been targeted with small molecules for treatment of other malignancies, which might allow rapid translation for targeted therapy of subsets of SHH patients (Supplementary Table 16). Novel SHH targets identified here are excellent candidates for combinatorial therapy with Smoothened inhibitors, to avoid the resistance encountered in both humans and mice9,14,15.

Group 3 and Group 4 medulloblastomas have generic names as comparatively little is known about their genetic basis, and no targets for rational therapy have been identified7. MYC amplicons are largely restricted to Group 3, whereas MYCN amplicons are seen in Group 4 and SHH tumours (Fig. 1d)3,4. Indeed, MYC and MYCN loci comprise the most significant regions of amplification observed in Group 3 and Group 4, respectively (Fig. 3a, b, Supplementary Figs 13, 14, 17–20 and Supplementary Tables 12–15). Group 3 MYC amplicons were mutually exclusive from those affecting the known medulloblastoma oncogene OTX2 (ref. 16) and were highly prognostic (Supplementary Fig. 21)3,16. Type II activin receptors, ACVR2A and ACVR2B and family member TGFBR1 are highly amplified in Group 3 tumours, indicating deregulation of TGF-β signalling as a driver event in Group 3 (Fig. 3c–e and Supplementary Fig. 22). The Group 3-enriched medulloblastoma oncogene OTX2 is a prominent target of TGF-β signalling in the developing nervous system17 and TGF-β pathway inhibitors CD109 (ref. 18), FKBP1A (refs 19 and 20) and SNX6 (ref. 20) are recurrently deleted in Group 3 (Fig. 3a, d). SCNAs in TGF-β pathway genes were heavily enriched in Group 3 (P = 5.37 × 10−5, Fisher’s exact test) and found in at least 20.2% of cases, indicating that TGF-β signalling represents the first rational target for this poor prognosis subgroup (Fig. 3d). Similarly, novel deletions affecting regulators of the NF-κB pathway, including NFKBIA (ref. 21) and USP4 (ref. 22) were identified in Group 4 (Supplementary Fig. 23), proposing that NF-κB signalling may represent a rational Group 4 therapeutic target.

Figure 3: The genomic landscape of Group 3 and Group 4 medulloblastoma.
figure 3

a, b, GISTIC2 plots depicting significant SCNAs in Group 3 (a) and Group 4 (b) with subgroup-enriched regions shaded in yellow and green, respectively. c, Recurrent amplifications targeting type II (ACVR2A and ACVR2B) and type I (TGFBR1) activin receptors in Group 3. d, Recurrent SCNAs affecting the TGF-β pathway in Group 3 (P = 5.73 × 10−5, Fisher’s exact test). Frequencies of focal and broad (parentheses) SCNAs are listed. e, Enrichment map of gene sets affected by SCNAs in Group 3 versus Group 4.

PowerPoint slide

Network analysis of Group 3 and Group 4 SCNAs illustrates the different pathways over-represented in each subgroup. Only TGF-β signalling is unique to Group 3 (Fig. 3e). In contrast, cell-cycle control, chromatin modification and neuronal development are all Group 4-enriched. Cumulatively, the dismal prognosis of Group 3 patients, the lack of published targets for rational therapy, and the prior targeting of TGF-β signalling in other diseases suggest that TGF-β may represent an appealing target for Group 3 rational therapies (Supplementary Table 16).

SNCAIP tandem duplication is common in Group 4

Although Group 4 is the most prevalent medulloblastoma subgroup, its pathogenesis remains poorly understood. The most frequent SCNA observed in Group 4 (33/317; 10.4%) is a recurrent region of single copy gain on chr5q23.2 targeting a single gene, SNCAIP (synuclein, alpha interacting protein) (Fig. 4a and Supplementary Fig. 24). SNCAIP, encodes synphilin-1, which binds to α-synuclein to promote the formation of Lewy bodies in the brains of patients with Parkinson’s disease23,24. Additionally, rare germline mutations of SNCAIP have been described in Parkinson’s families25. Large insert, mate-pair, whole-genome sequencing (WGS) demonstrates that SNCAIP copy number gains arise from tandem duplication of a truncated SNCAIP (lacking non-coding exon 1), inserted telomeric to the germline SNCAIP allele (Fig. 4b, c and Supplementary Fig. 25). Affymetrix SNP6 array profiling of patient-matched germline material confirmed that SNCAIP duplications are somatic (Supplementary Fig. 26), and subsequent whole-transcriptome sequencing (RNA-Seq) of select Group 4 cases (n = 5) verified that SNCAIP is the only gene expressed in the duplicated region (Supplementary Fig. 27). Analysis of published copy number profiles for 3,131 primary tumours26 and 947 cancer cell lines27 (total of 4,078 cases) revealed only four cases with apparent duplication of SNCAIP, all of which were inferred as Group 4 medulloblastomas (data not shown). We conclude that SNCAIP duplication is a somatic event highly specific to Group 4 medulloblastoma.

Figure 4: Tandem duplication of SNCAIP defines a novel subtype of Group 4.
figure 4

a, Highly recurrent, focal, single-copy gain of SNCAIP in Group 4. b, Paired-end mapping verifies recurrent tandem duplication of SNCAIP in Group 4. c, Schematic representation of SNCAIP tandem duplication. d, SNCAIP is a Group 4 signature gene. Upper panel, SNCAIP expression across subgroups in a published series of 103 primary medulloblastomas. Error bars depict the minimum and maximum values, excluding outliers. Lower panel, SNCAIP ranks among the top 1% (rank, 39th out of 16,758) of highly expressed genes in Group 4. e, NMF consensus clustering of 188 expression-profiled Group 4 tumours supports two transcriptionally distinct subtypes designated 4α and 4β (cophenetic coefficient = 0.9956). 21 out of 22 SNCAIP duplicated cases belong to Group 4α (P = 3.12 × 10−8, Fisher’s exact test). f, SNCAIP expression is significantly elevated in Group 4α versus 4β (P = 9.31 × 10−14, Mann–Whitney test). g, Group 4α cases harbouring SNCAIP duplication exhibit a 1.5-fold increase in SNCAIP expression. f, g, Error bars depict the minimum and maximum values, excluding outliers.

PowerPoint slide

Re-analysis of 499 published medulloblastoma expression profiles confirmed that SNCAIP is one of the most highly upregulated Group 4 signature genes (Fig. 4d and Supplementary Fig. 28). Profiling of 188 Group 4 tumours on expression microarrays followed by consensus non-negative matrix factorization (NMF) clustering delineates two subtypes of Group 4 (4α and 4β; Fig. 4e and Supplementary Fig. 29). Strikingly, 21 out of 22 SNCAIP duplicated cases belonged to Group 4α (P = 3.12 × 10−8, Fisher’s exact test). SNCAIP is more highly expressed in Group 4α than 4β (Fig. 4f), and 4α samples with tandem duplication showed approximately 1.5-fold increased expression, consistent with gene dosage (Fig. 4g and Supplementary Figs 35 and 36). Group 4α exhibits a relatively balanced genome compared to 4β (Supplementary Figs 30–32), and several 4α cases harbour SNCAIP duplication in conjunction with i17q and no other SCNAs (Supplementary Fig. 33). Importantly, SNCAIP duplications are mutually exclusive from other prominent SCNAs in Group 4, including MYCN and CDK6 amplifications (Supplementary Fig. 34).

PVT1 fusions arise via chromothripsis in Group 3

Although recurrent gene fusions have recently been discovered in solid tumours, none have been reported in medulloblastoma. RNA-Seq of Group 3 tumours (n = 13) identified two independent gene fusions in two different tumours (MB-182 and MB-586), both involving the 5′ end of PVT1, a non-coding gene frequently co-amplified with MYC in Group 3 (Fig. 5a, b, Supplementary Fig. 37 and Supplementary Tables 17 and 18). Sanger sequencing confirmed a fusion transcript consisting of exons 1 and 3 of PVT1 fused to the coding sequence of MYC (exons 2 and 3) in MB-182, and a fusion involving PVT1 exon 1 fused to the 3′ end of NDRG1 in MB-586 (Fig. 5a, b).

Figure 5: Identification of frequent PVT1-MYC fusion genes in Group 3.
figure 5

a, b, RNA-Seq identifies multiple fusion transcripts driven by PVT1 in Group 3. Schematics depict the structures of verified PVT1-MYC (a) and PVT1-NDRG1 (b) fusion genes. c, Heat map of the MYC/PVT1 locus showing a subset of 13 MYC-amplified Group 3 cases subsequently verified to exhibit PVT1 gene fusions (shown in d). Yellow box highlights the common breakpoint affecting the first exon/intron of PVT1, including miR-1204. d, Summary of PVT1 fusion transcripts identified in Group 3. e, f, WGS confirms complex patterns of rearrangement on chr8q24 in PVT1 fusion (+) Group 3.

PowerPoint slide

Group 3 copy number data at the MYC/PVT1 locus indicated that additional samples might harbour PVT1 gene fusions (Fig. 5c). PCR with reverse transcription (RT–PCR) profiling of select Group 3 cases confirmed PVT1-MYC fusions in at least 60% (12/20) of MYC-amplified cases (Fig. 5d and Supplementary Table 19). Fusion transcripts included many other portions of chr8q, with up to four different genomic loci mapping to a single transcript, a pattern reminiscent of chromothripsis28,29 (Fig. 5d). WGS performed on four MYC-amplified Group 3 tumours harbouring PVT1 fusion transcripts identified a series of complex genomic rearrangements on chr8q (Fig. 5e, f, Supplementary Fig. 38 and Supplementary Tables 20 and 21). Chromosome 8 copy number profile for MB-586 (PVT1-NDRG1) derived from WGS showed that PVT1 and NDRG1 are structurally linked, as predicted by RNA-Seq, and several adjacent regions of 8q24 were extensively rearranged (Fig. 5e, f and Supplementary Table 21). Monte Carlo simulation suggests that this fragmented 8q amplicon arose through chromothripsis, a process of erroneous DNA repair following a single catastrophic event in which a chromosome is shattered into many pieces (Supplementary Fig. 39). Further examination of our copy number data set revealed rare examples of chromothripsis across subgroups (Supplementary Fig. 40), with only chr8 in Group 3 demonstrating statistically significant, region-specific chromothripsis (Q = 0.0004, false discovery rate (FDR)-corrected Fisher’s exact test). Among Group 3 tumours, the occurrence of chr8q chromothripsis is correlated with deletion of chr17p (location of TP53; data not shown), in keeping with the association of loss of TP53 and chromothripsis recently described in medulloblastoma (P = 0.0199, Fisher’s exact test)28. Whereas the PVT1 locus has been suggested to be a genomically fragile site, we observe that the majority of MYC-amplified Group 3 tumours harbour PVT1 fusions that arise through a process consistent with chromothripsis.

PVT1 is a non-coding host gene for four microRNAs, miR-1204miR-1207. Previous studies have implicated miR-1204 as a candidate oncogene that enhances oncogenesis in combination with MYC30,31. PVT1 fusions identified in this study involve only PVT1 exon 1 and miR-1204. Importantly, miR-1204, but not the adjacent miR-1205 and miR-1206, is expressed at a higher level in PVT1-MYC fusion (+) Group 3 tumours compared to fusion (−) cases (P = 0.0008, Mann–Whitney test; Fig. 6a). To evaluate whether aberrant expression of miR-1204 contributes to the malignant phenotype, we inhibited miR-1204 in MED8A cells, a Group 3 medulloblastoma cell line with a confirmed PVT1-MYC fusion (Fig. 5d). Antagomir-mediated RNA interference of miR-1204 had a pronounced effect on MED8A growth (Fig. 6b). A comparable reduction in proliferative capacity was achieved with knockdown of MYC. Conversely, the medulloblastoma cell line ONS76 exhibits neither MYC amplification nor a detectable PVT1-MYC fusion gene, and knockdown of miR-1204 had no effect in this line (Fig. 6c).

Figure 6: Functional synergy between miR-1204 and MYC secondary to PVT1-MYC fusion.
figure 6

a, Quantitative RT–PCR of PVT1-encoded microRNAs confirms upregulation of miR-1204 in PVT1-MYC fusion (+) Group 3 tumours. MYC-balanced/fusion (−), n = 4; MYC-amplified/fusion (−), n = 6; MYC-amplified/fusion (+), n = 8. Error bars represent standard error of the mean (s.e.m.) and reflect variability among samples. b, c, Knockdown of miR-1204 attenuates the proliferative capacity of PVT1-MYC fusion (+) MED8A medulloblastoma cells (b) but has no effect on fusion (−) ONS76 cells (c). Error bars represent the standard deviation (s.d.) of triplicate experiments. CTL, control.

PowerPoint slide

PVT1 has been reported previously in fusion transcripts with a number of partners30,32,33. The most prevalent form of the PVT1-MYC fusion in Group 3 tumours lacks the first, non-coding exon of MYC, similar to forms of MYC that have been described in Burkitt’s lymphoma34 (Fig. 5a, d). The PVT1 promoter contains two non-canonical E-boxes and can be activated by MYC31. This indicates a positive feedback model where MYC can reinforce its own expression from the PVT1 promoter in PVT1-MYC fusion (+) tumours. Indeed, knockdown of MYC alone in MED8A cells resulted in diminished expression of both MYC and miR-1204, suggesting MYC may positively regulate PVT1 (that is, miR-1204) expression in medulloblastoma cells (Supplementary Fig. 41).

Discussion

Medulloblastomas have few SNVs compared to many adult epithelial malignancies11, whereas SCNAs seem to be quite common. Medulloblastoma is a heterogeneous disease7, thereby requiring large cohorts to detect subgroup-specific events. Through the accumulation of >1,200 medulloblastomas in MAGIC, we have identified novel and significant SCNAs. Many of the significant SCNAs are subgroup-restricted, highly supporting their role as driver events in their respective subgroups.

Expression of synphilin-1 in neuronal cells results in decreased cell doubling time35, decreased caspase-3 activation36, decreased TP53 transcriptional activity and messenger RNA levels, and decreased apoptosis37. Synphilin-1 is ubiquitinated by parkin, which is encoded by the hereditary Parkinson’s disease gene PARK2 (ref. 24), a candidate tumour suppressor gene38. Whereas patients with Parkinson’s disease have an overall decreased risk of cancer, they may have an increased incidence of brain tumours39,40. As tandem duplications of SNCAIP are highly recurrent, stereotypical, subgroup-restricted, affect only a single gene, and as SNCAIP-duplicated tumours have few if any other SCNAs, SNCAIP is a probable driver gene, and merits investigation as a target for therapy of Group 4α. Similarly, PVT1 fusion genes are highly recurrent, restricted to Group 3, arise through a chromothripsis-like process, and are the first recurrent translocation reported in medulloblastoma.

We identify a number of highly targetable, recurrent, subgroup-specific SCNAs that could form the basis for future clinical trials (that is, PI3K signalling in SHH, TGF-β signalling in Group 3, and NF-κB signalling in Group 4). Activation of these pathways through alternative, currently unknown genetic and epigenetic events could increase the percentage of patients amenable to targeted therapy. We also identify a number of highly ‘druggable’ events that occur in a minority of cases. The cooperative, global approach of the MAGIC consortium has allowed us to overcome the barrier of intertumoural heterogeneity in an uncommon paediatric tumour, and to identify the relevant and targetable SCNAs for the affected children.

Methods Summary

All patient samples were obtained with consent as outlined by individual institutional review boards. Genomic DNA was prepared, processed and hybridized to Affymetrix SNP6 arrays according to manufacturer’s instructions. Raw copy number estimates were obtained in dChip, followed by CBS segmentation in R. SCNAs were identified using GISTIC2 (ref. 13). Driver genes within SCNAs were inferred by integrating matched expressions, literature evidence and other data sets. Pathway enrichment of SCNAs was analysed with g:Profiler and visualized in Cytoscape using Enrichment Map. Fluorescence in situ hybridization (FISH) was performed as described previously8,10. Medulloblastoma subgroup was assigned using a custom nanoString CodeSet as described previously12. Tandem duplication of SNCAIP was confirmed by paired-end mapping as previously reported28. RNA was extracted, processed and hybridized to Affymetrix Gene 1.1 ST Arrays as recommended by the manufacturer. Consensus NMF clustering was performed in GenePattern. Gene fusions were identified from RNA-Seq data using Trans-ABySS. Medulloblastoma cell lines were maintained as described10. Proliferation assays were performed with the Promega CellTiter 96 Assay. Additional methods are detailed in full in Supplementary Methods.