The transcriptional landscape of Shh medulloblastoma

ARTICLE The transcriptional landscape of Shh medulloblastoma Sonic hedgehog medulloblastoma encompasses a clinically and molecularly diverse group of cancers of the developing central nervous system. Here, we use unbiased sequencing of the transcriptome across a large cohort of 250 tumors to reveal differences among molecular subtypes of the disease, and demonstrate the previously unappreciated importance of noncoding RNA transcripts. We identify alterations within the cAMP dependent pathway (GNAS, PRKAR1A) which converge on GLI2 activity and show that 18% of tumors have a genetic event that directly targets the abundance and/or stability of MYCN. Furthermore, we discover an extensive network of fusions in focally amplified regions encompassing GLI2, and several lossof-function fusions in tumor suppressor genes PTCH1,SUFU and NCOR1. Molecular convergence on a subset of genes by nucleotide variants, copy number aberrations, and gene fusions highlight the key roles of specific pathways in the pathogenesis of Sonic hedgehog medulloblastoma and open up opportunities for therapeutic intervention. https://doi.org/10.1038/s41467-021-21883-0 OPEN A full list of authors and their affiliations appears at the end of the paper. NATURE COMMUNICATIONS | (2021) 12:1749 | https://doi.org/10.1038/s41467-021-21883-0 | www.nature.com/naturecommunications 1 1234567890():,; Medulloblastoma (MB) is the most common malignant pediatric brain tumor and a major cause of morbidity and mortality in the pediatric population1. Current therapy consists of maximal safe resection, radiotherapy in patients over 36 months, and cytotoxic chemotherapy. MB is thought to comprise a group of four molecularly distinct diseases: Wnt, Sonic Hedgehog (Shh), Group 3, and Group 42. Shh-MB is clinically heterogeneous with infants, teenagers and adults affected. Shh-MB likely comprises four molecular subtypes, Shh-α (adolescents), Shh-β(babies with a poor prognosis), Shh-γ (babies with a good prognosis), and Shh-δ(adults)3. The vast difference in the host (babies versus adolescents versus adults) dictates different treatment approaches for different molecular subtypes. Prior delineation of Shh-MB subtypes used expression microarrays4, and/or DNA methylation arrays3, and the biology underlying the differences among the subtypes is poorly understood. To further understand the biology of Shh-MB and its molecular subtypes, we studied 250 human Shh-MB using strand-specific RNA sequencing with the incorporation of DNA methylation, whole-genome sequencing, and SNP 6.0 copy number analysis. This non-biased approach to the Shh-MB transcriptome allows us to understand the transcriptional basis and underlying biology of Shh-MB and reveals a previously unsuspected role for many non-coding RNAs. We find disruption in the cAMP pathway converging on Shh signaling and also detect a cluster of mutations in MYCN which prevent degradation by FBXW7. Alterations in these genes are mutually exclusive of each other and found in 18% of Shh-MB tumors. We also identify a number of fusion transcripts in Shh-MB, many of which fall within focally amplified regions and known Shh-MB tumor suppressors. This analysis of a large cohort of similar tumors highlights previously unsuspected examples of molecular convergence where the same gene or pathway is activated through diverse molecular mechanisms, emphasizing the importance of those drivers in Shh-MB. Genetic events in Shh-MB do not assort randomly across the cohort, but rather show very restricted patterns of mutual exclusivity, suggesting specific biology, with implications for Shh-MB modeling, and perhaps for the design of synthetic lethal approaches to therapy. Results Importance of the non-coding transcriptome in Shh-MB. Our Shh-MB strand-specific RNA-seq samples (n=250) were additionally characterized with whole-genome sequencing (WGS) (n=26), Infinium Human Methylation 450 K BeadChip (n= 196), Affymetrix HuGene 1.1 expression arrays (n=173), and Affymetrix SNP 6.0 arrays (n=130) (Fig. 1a; Supplementary Data 1). Integrative analysis and unsupervised clustering of both RNA-seq and 450 K methylation data allowed us to assign ShhMB samples to their appropriate molecular subtype3. Subtype assignment based on RNA-seq and 450 K methylation data highly overlap with subtyping using Affymetrix expression and 450 K methylation arrays (Fig. 1b, c). While protein-coding genes make up only 35% of the transcriptome in GENCODE (v19), 95% of subtype-specific genes identified using expression arrays are protein-coding (Fig. 1d). However, Shh-MB subtype-specific transcripts identified with RNA-seq encompass many non-coding RNA species, including long non-coding RNAs, expressed pseudogenes, and microRNAs (Fig. 1d; Supplementary Data 2). Indeed, the majority of genes differentially expressed between subtypes using RNA-seq data are non-coding transcripts, which are not evaluated by expression arrays (Fig. 1e). While many of these non-protein-coding genes are poorly annotated, pathway analysis reveals divergent biological mechanisms among Shh-MB subtypes (Fig. 1f). We conclude that each Shh-MB subtype has a unique landscape of non-coding transcripts which may play an important role in the biology of Shh-MB. cAMP-dependent pathway alterations converge on GLI2 activity. We investigated the incidence and patterns of mutations in a subtype-specific manner (Fig. 2a; Supplementary Data 3). We detect mutations in GNAS, a heterotrimeric Gs protein αsubunit (Gαs), in 4.4% of Shh-MB. Most mutations cluster between the GTPase and helical domains which are predicted to reduce GTP binding (Fig. 2b). GNAS activates adenylyl cyclase which increases intracellular cAMP, there-by activating protein kinase A (PKA), a negative regulator of the Shh signaling pathway. This is in line with the phenotype of Gnas knockout mice which develop Shh-MBs5. Direct phosphorylation of GLI2 by the PKA complex leads to proteolytic conversion of GLI2 into its repressor form and abrogation of Shh target gene expression. Correspondingly, we also observe mutations mutually exclusive of GNAS in PRKAR1A, a critical component of the PKA complex (Fig. 2c, d). All PRKAR1A mutations localize to the binding pocket of the cAMP-binding domain impairing the activation of PKA6. Nearly all patients with alterations in GNAS or PRKAR1A do not have any alterations in the Shh signaling pathway (i.e., PTCH1,SMO, SUFU,GLI2)(P=3.80 × 10−5; two-sided Fisher’s exact test), suggesting that aberration of the cAMP-dependent pathway can lead to Shh pathway activation (Fig. 2e). Single nucleotide variants (SNVs) were also found in GLI2 within the activation domain7(Fig. 2f) which are largely exclusive of GLI2 amplification or fusions (Fig. 2g). Most recurrent is the p.P1028L mutation found within a partial PKA consensus sequence8, which may interfere with phosphorylation and prevent conversion into its repressor form. Other SNVs can disrupt binding to SUFU (p.G274R). Interestingly, nearly all patients with mutations in GLI2 had no other alterations in Shh pathway constituents (PTCH1,SMO,SUFU)(P=0.015; two-sided Fisher’s exact test) further suggesting an oncogenic role. In conclusion, we describe an alternative axis of control for the Shh-signaling pathway and open up more opportunities for therapy through activation of cAMP signaling. Alterations in cell cycle control genes. Several Shh-MB drivers important for cell cycle control which were previously identified as amplified, (i.e., MYCN and PPM1D) also harbor damaging mutations in a subset of patients. PPM1D, a negative regulator of the p53 DNA damage response pathway9undergoes nonsense and frameshift mutations at its C-terminus (Fig. 3a, b), all of which are predicted to leave its phosphatase activity intact while significantly increasing protein stability10–12. We also detect a cluster of SNVs in MYCN within the phospho-degron containing MBI domain (Fig. 3c). MYCN amplifications and SNVs are mutually exclusive (Fig. 3d). Phosphorylation of MYCN at S62 primes for second phosphorylation at T58 by glycogen synthase kinase-3 (GSK3). Subsequent dephosphorylation at S62 leads to recruitment of the FBXW7 E3 ubiquitin ligase complex to a phosphodegron motif that includes amino acids both N-terminal and C-terminal to pT5813, and the consequent ubiquitination of MYCN14,15. Mutations in this region of MYCN disrupt FBXW7 binding and/or ubiquitination, and are predicted to stabilize MYCN16 (Fig. 3e). Remarkably, we also identify missense mutations of FBXW7 within tryptophan-aspartic acid motif (WD40) (Fig. 3f, g)17–20 that binds MYCN, in >10% of Shh-MB, which are mutually exclusive of MYCN amplification or SNVs. Finally, we found a mutational hotspot (p.R60Q) in the MYC heterodimer partner MAX (1.6% of Shh-MB tumors) (Fig. 3h). These alterations lie within the bHLH-Zip domain involved in protein–protein interactions and DNA binding and may ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-21883-0 2NATURE COMMUNICATIONS | (2021) 12:1749 | https://doi.org/10.1038/s41467-021-21883-0 | www.nature.com/naturecommunications Fig. 1 Importance of the non-coding transcriptome in Shh-MB. a Overview of Shh-MB RNA-seq samples and overlapping data sources. bHeatmap of the sample-to-sample fused network (SNF) by cluster (k=4, n=250). Sample similarity is represented by red (less similar) to yellow (more similar) coloring inside the heatmap. cSubtype clusters obtained by Similar Network Fusion (k=4) using Affymetrix expression +450 K methylation and RNA-seq +450 K methylation (n=196). Relationships between clustering methods are indicated by gray bars between columns. dBiotype distribution amongst all genes (top) as compared to genes that differentiate subtypes (significant normalized mutual information (NMI) from SNF RNA-seq +450 K methylation), in both RNA-seq and microarray datasets (middle) or restricted to only the RNA-seq dataset (bottom). eDifferentially expressed genes per subtype (RNAseq). Genes found only with RNA-seq data are indicated. fEnrichment map of biological processes and pathways in Shh-MB subtypes. Each node represents a pathway or process and connecting lines represent common genes between them. Nodes with many shared genes are grouped together and labeled with a biological theme. The color of the nodes refers to the subtype(s) in which the process is enriched. The size of the node is proportional to the number of genes in the process. NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-21883-0 ARTICLE NATURE COMMUNICATIONS | (2021) 12:1749 | https://doi.org/10.1038/s41467-021-21883-0 | www.nature.com/naturecommunications 3 Fig. 2 cAMP-dependent pathway alterations converge on GLI2 activity. a Oncoprint summaries of all fusion, mutation, and copy number data (n=196). Subtypes are denoted above. NA, not available. bGene-level summary of GNAS events. cMutual exclusivity of GNAS and PRKAR1A LOF events. LOF events include mutations and homozygous deletions. dGene-level summary of PRKAR1A events. ecAMP dependant signaling pathway schematic. Red indicates activating alterations while blue indicates inactivating alterations. f,gGene-level summary of (f)GLI2 events and (g) their overlaps. Mutations in fare shown as lollipop diagrams above the gene schematic and fusion events are shown below. The 5 prime and 3 prime orientation of the fusion transcript is indicated by the color orientation. In cases where GLI2 is the 3 prime partners, the fusion lollipop is red on the right. ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-21883-0 4NATURE COMMUNICATIONS | (2021) 12:1749 | https://doi.org/10.1038/s41467-021-21883-0 | www.nature.com/naturecommunications upregulate MYC activity21. In conclusion, we find that 18% of Shh-MB patients have a genetic event that directly targets the abundance and/or stability of MYCN. Somatic copy number aberrations in Shh-MB.Regionsof recurrent genomic gain and loss identify both known Shh-MB driver genes (i.e., MYCN,GLI2,PPM1D,PTEN)22,aswellasthe putative drivers (i.e., PRMT2,HECTD1,SOX11,andLHX1) (Fig. 4a; Supplementary Data 4). Several recurrent somatic copy number aberrations (CNAs) that do not contain any genes when studied by expression arrays, do contain transcripts when studied by RNA-sequencing (Fig. 4b). Regions of focal amplification are much more likely to show concomitant changes in gene transcription as compared to larger, broad copy number changes (Fig. 4c). A number of putative Shh-MB driver genes encompassed by focal gains or deletions demonstrate copy number-driven expression, further supporting their role as drivers (Fig. 4d; Supplementary Data 5). Notably, only 15% (378/2,536) of genes identified within GISTIC regions show copy number-driven expression (Fig. 4e, Supplementary Fig. 1A–C). In many cases, the copy number responsive genes are poorly annotated non-coding RNAs that might first be overlooked (Fig. 4e−h, Supplementary Fig. 1D−F). We also observe significant deletions in 9q34.11 encompassing the copy number responsive gene GPR107 (Fig. 4f).Thisregionisusually lost in the context of chromosome 9q loss along with PTCH1 and IKBKAP (Supplementary Fig. 1G, H). A substantial minority (24%) of Shh-MB are aneuploid; their transcriptome Fig. 3 Alterations in cell cycle control genes. a,bGene-level summary of (a)PPM1D events and (b) their overlaps. cGene-level summary of MYCN events. Only mutations in the canonical isoform NM_005378 are shown. dOverlap of MYCN fusion, amplification, and SNV events. eStructural model of MYCN highlighting positions affected by hotspot mutations (blue) near the FBWX7 protein binding region (purple), and phospho-degron positions (red). fGene-level summary of FBXW7 events. gMutual exclusivity of MYCN gain-of-function (GOF) and FBXW7 loss-of-function (LOF) events. P-value calculated using the DISCOVER package. GOF and LOF events include both high-level CNA and mutation events. hGene-level summary of MAX events. NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-21883-0 ARTICLE NATURE COMMUNICATIONS | (2021) 12:1749 | https://doi.org/10.1038/s41467-021-21883-0 | www.nature.com/naturecommunications 5 differs from diploid tumors by over-expression of genes involved in RNA processing and translation (Supplementary Fig. 2A−D). We conclude that regions of focal CNAs in the Shh-MB genome contain both copy number responsive and non-responsive genes, that many events focus on poorly characterized non-coding transcripts, and that non-copy number responsive genes within CNAs are likely to a poor choice for the development of targeted therapy. Identification of Shh-MB fusion genes. We identified fusion transcripts in the Shh-MB transcriptome using three distinct ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-21883-0 6NATURE COMMUNICATIONS | (2021) 12:1749 | https://doi.org/10.1038/s41467-021-21883-0 | www.nature.com/naturecommunications assembly and alignment-based callers (STAR-fusion, InFusion, Trans-Abyss)23–25,filtering out any readthrough transcripts or fusion contigs that were also observed in libraries of noncancerous brain tissue (Supplementary Figs. 3 and 4; Supplementary Data 6). A subset of Shh-MB patients (12/126, 10%) harbor a high number (top 25th percentile) of both fusions and copy number events and are significantly associated with both aneuploidies (10/12; P=7.4 × 10−7, two-sided Fisher’s exact test) and with TP53 mutation (6/12; P=1.2 × 10−4, two-sided Fisher’s exact test) (Supplementary Fig. 5A). Only a subset of fusion transcripts demonstrates substantial evidence of an underlying structural variant (SV) in the genome due to the presence of breakpoints in matching WGS or SNP 6.0 data and/or the identification of multiple splice variants of the same fusion transcript. The number of SV-supported fusions per patient was significantly different among subtypes (P=4.7 × 10−8; KruskalWallis rank-sum test), with Shh-αshowing the highest number of fusions per tumor (Supplementary Data 6). A large number of SV-supported fusions coincide with focal amplification of GLI2 (2q14.2), MYCN (2p24.3), CCND2 (12p13.32), and PPM1D (17q23.2) (Fig. 5a, b; Supplementary Fig. 5B−G). Most recurrently, we observe GLI2 fusion transcripts (11/250 Shh-MB) fused in the 5 prime ends of the mRNA which houses the repressor domain of the encoded protein, suggesting that the fusion leads to an overactive protein (Fig. 2f). GLI2 fusions were largely exclusive of detected SNV events and were also found in patients without GLI2 amplifications (Fig. 2a). We additionally observe recurrent fusion transcripts at nearby genomic loci, such as EPB41L5,NBAS,BCAS3,andGLIS3 which are likely a result of chromothripsis, and/or the formation of extrachromosomal double minutes (Fig. 5c−f)26,27.Itisuncleartheextenttowhich amplification versus the formation of a fusion transcript contributes to clonal selection (Supplementary Fig. 5B−G), nor is it obvious whether fusion transcripts involving nearby genes are drivers or passengers. Conversely, we now identify fusions in ZBTB20 (14/250 patients), which are not usually found in the context of amplification (Fig. 6a, b). We also identify fusion transcripts involving known Shh-MB tumor suppressor genes such as PTCH1 and SUFU,(Fig.6c–h), both of which are accompanied by decreased expression of the gene immediately following the breakpoint. These are likely markers of chromosomal events that result in loss of gene function and are largely mutually exclusive of tumors with mutations or large chromosomal deletions, supporting their functional role (Fig. 6g, h). We identify N-terminal missense mutations of SUFU which are predicted to be damaging, occur in a highly conserved portion of the gene, and are mutually exclusive with mutations in other Shh signaling genes (Fig. 6e). NCOR1, a transcriptional regulator of neural stem cell differentiation28,29 harbors similar loss-of-function (LOF) fusion transcripts and damaging mutations (13/250, 5.2% of patients) (Fig. 6i, j). We conclude that >20% of Shh-MB patients exhibit fusion transcripts with structural support for an event in the genome. The landscape of oncogenic alterations across Shh-MB. Transcriptional profiling of this large cohort of a single molecular tumor type permits identification of both recurrent and rare Shh-MB driver genes, and their patterns of mutual exclusivity (Supplementary Data 7 and 8). Most Shh-MBs (86%) have an identifiable event activating the Sonic Hedgehog signaling pathway, including mutations of PTCH1 (42%), SMO (12%), SUFU (10%), or GLI2 (9%) (Fig. 2a). About 11% of patients have previously unappreciated inactivating (i.e., SUFU or PTCH1), or activating (i.e., GLI2) fusion transcripts affecting Shh pathway genes. Pathways discovered using copy number aberrations, mutations, or fusion transcripts were numerous in Shh-αand Shh-δbut limited for Shh-βor Shh-γdue to their low number of mutational events (Fig. 7a). There is strong mutational convergence on genes important for Shh signaling, neuronal development, cell cycle progression, and modification of the epigenome (Fig. 7a, b). Of Shh-MBs without detected events that canonically lead to excess Shh signaling (PTCH1, SMO, SUFU, TP53, GLI2, 9q, 10q, and 17p loss) (45/250 patients), the most recurrent mutational events involved DDX3X (n=12), KMT2D (n=6), PRKAR1A,GNAS,GSE1 and CREBBP (each n=5) (Fig. 2a); all of which have been previously shown to interact with or potentiate Shh signaling5,30,31.DDX3X and GSE1 are potent medulloblastoma tumor suppressors in Gorlin 1 NES cells, PRKAR1A with its upstream g-protein GNAS are both regulators of Shh activity through cAMP, and CREBBP has been shown to promote cell-cycle exit during postnatal development in coordination with Shh pathway upregulation. We used MethylMix32 to identify potential Shh-MB driver genes affected by promoter CpG hypomethylation or hypermethylation, for which there is a correlative change in gene expression (Supplementary Fig. 6). We obtained a curated list of 735 promoter probe-gene pairs (540 and 195 for two and three methylation clusters, respectively), involving 727 genes in total (Supplementary Fig. 6A, B; Supplementary Data 9). Among these, we identify a number of known cancer genes (i.e., FOXL2, RUNX1T1), transcription factors (i.e., MEIS2), as well as LHX1 and PAX6 (which are also recurrently affected by mutations) (Supplementary Data 9). Transcriptional silencing of PAX6 through promoter CpG methylation, versus somatic mutations of PAX6, appear to be largely mutually exclusive (P=7.3 × 10−4, multinomial exact test), suggesting convergence on PAX6 loss of function (Supplementary Fig. 6C−H). Lastly, DISCOVER33 wasusedtoidentifynetworksof significantly mutually exclusive genes and chromosome arms across the subgroup and in a subtype-aware manner. We observe extensive significant mutual exclusivity between driver gene pairs in Shh-MB (Fig. 7c; Supplementary Data 8). As expected, the most pronounced negative gene correlations are Fig. 4 Somatic copy number aberrations in Shh-MB. a GISTIC significant amplifications (red) and deletions (blue) observed in Shh-MB (n=126). bLog2 fold increase of known annotated gene in GISTIC regions using RNA-seq compared to expression arrays. GISTIC regions with genes only found in the RNAseq dataset have points on the outermost circle. cNormalized expression density across broad and focal CNAs. dExpression difference between copy number neutral and aberrant states in GISTIC region copy number responsive genes. Each gene was normalized by its neutral copy number state distribution. Numbers in square brackets denote the number of patients detected with the CNA. The lower and upper hinges in the boxplot correspond to the first and third quartiles while the center line represents the median. The upper and lower whisker extends from the nearest hinge to the smallest/ largest value at most 1.5 times the interquartile range. Points outside this range are outliers and are plotted individually. eGISTIC copy number responsive gene types. f–hExpression difference between copy number neutral and aberrant states in (f) 9q34.11, (g) 8q22.1, and (h) 10q23.31. Asterisks annotate significant copy number responsive genes (FDR < 0.05) calculated using a Kruskal-Wallis rank-sum test. Please refer to Supplementary Data 5 for exact P-values. The SNP 6.0 copy number segments are shown to the left of each graph. The expression of each gene was normalized by the expression median of the neutral copy number state. NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-21883-0 ARTICLE NATURE COMMUNICATIONS | (2021) 12:1749 | https://doi.org/10.1038/s41467-021-21883-0 | www.nature.com/naturecommunications 7 Fig. 5 Fusion networks within somatic recurrently amplified regions. a The network of gene fusions in focally amplified regions. Node color signifies the most common orientation of the fusion gene, 5 prime (blue), 3 prime (red), or both (gray). The arrow and base color show the proportion of chimeric reads compared to wildtype supporting the fusion. The arrow line color shows the difference in expression of the 3 prime fusion partners compared to patients without the detected fusion. bOncoprint of fusions depicted in focally amplified regions illustrated in a.c–fGene-level summary of (c)EPB41L5,(d)NBAS, (e)BCAS3, and (f)GLIS3 events. Refer to Fig. 2b for schema description. ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-21883-0 8NATURE COMMUNICATIONS | (2021) 12:1749 | https://doi.org/10.1038/s41467-021-21883-0 | www.nature.com/naturecommunications Fig. 6 Recurrent fusions in Shh-MB. a Oncoprint of fusions detected in focally amplified regions and known Shh-MB tumor suppressors. NA, not available. bGene-level summary of ZBTB20 events. Mutations are shown as lollipop diagrams above the gene schematic and fusion events are shown below. The 5 prime and 3 prime orientation of the fusion transcript is indicated by the color orientation. In cases where ZBTB20 is the 3 prime partners, the fusion lollipop is red on the right. cGene-level summary of PTCH1 events. dRead depth diagrams of representative PTCH1 fusion events. eGene-level summary SUFU events. fRead depth diagrams of representative SUFU fusion events. gOverlap of PTCH1 fusion, amplification, and mutation events. hOverlap of SUFU fusion, amplification, and mutation events. IGene-level summary of NCOR1 events. jRead depth diagrams of representative NCOR1 fusion events. NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-21883-0 ARTICLE NATURE COMMUNICATIONS | (2021) 12:1749 | https://doi.org/10.1038/s41467-021-21883-0 | www.nature.com/naturecommunications 9 Competing interests The authors declare no competing interests. Additional information Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41467-021-21883-0. Correspondence and requests for materials should be addressed to H.S. or M.D.T. Peer review information Nature Communications thanks Scott Pomeroy and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Reprints and permission information is available at http://www.nature.com/reprints Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’sCreativeCommons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/. © The Author(s) 2021 Patryk Skowron1,2,3,77, Hamza Farooq1,2,3,77, Florence M. G. Cavalli1,3,77, A. Sorana Morrissy4,5,6,77, Michelle Ly1,2,3, Liam D. Hendrikse 1,3,7, Evan Y. Wang 1,3,7, Haig Djambazian 8,9, Helen Zhu7,10, Karen L. Mungall11, Quang M. Trinh 10, Tina Zheng12, Shizhong Dai13, Ana S. Guerreiro Stucklin 1,3, Maria C. Vladoiu1,2,3, Vernon Fong 1,3, Borja L. Holgado1,3, Carolina Nor1,3, Xiaochong Wu1,3, Diala Abd-Rabbo10, Pierre Bérubé 8, Yu Chang Wang8, Betty Luu1,3, Raul A. Suarez1,3, Avesta Rastan 1,3,14, Aaron H. Gillmor4,5,6, John J. Y. Lee1,2,3, Xiao Yun Zhang1, Craig Daniels1,3, Peter Dirks 1,3,15,16, David Malkin 7,17, Eric Bouffet3,17, Uri Tabori3,14,17, James Loukides3, François P. Doz18, Franck Bourdeaut18, Olivier O. Delattre 19, Julien Masliah-Planchon20, Olivier Ayrault 21, Seung-Ki Kim22, David Meyronet23, Wieslawa A. Grajkowska24, Carlos G. Carlotti25, Carmen de Torres26, Jaume Mora26, Charles G. Eberhart27, Erwin G. Van Meir28, Toshihiro Kumabe29, Pim J. French 30, Johan M. Kros31, Nada Jabado 32, Boleslaw Lach33,34, Ian F. Pollack35, Ronald L. Hamilton36, Amulya A. Nageswara Rao37, Caterina Giannini 38, James M. Olson39, László Bognár40, Almos Klekner40, Karel Zitterbart41, Joanna J. Phillips 42,43, Reid C. Thompson44, Michael K. Cooper45, Joshua B. Rubin 46, Linda M. Liau 47, Miklós Garami 48, Peter Hauser48, Kay Ka Wai Li49, Ho-Keung Ng49, Wai Sang Poon50, G. Yancey Gillespie51, Jennifer A. Chan6, Shin Jung52, Roger E. McLendon53,54, Eric M. Thompson54, David Zagzag55, Rajeev Vibhakar56, Young Shin Ra57, Maria Luisa Garre58, Ulrich Schüller 59,60,61, Tomoko Shofuda62, Claudia C. Faria63,64, Enrique López-Aguilar65, Gelareh Zadeh66,67, Chi-Chung Hui1,16, Vijay Ramaswamy 1,3,7,17, Swneke D. Bailey68,69, Steven J. Jones 11,70,71, Andrew J. Mungall 11, Richard A. Moore11, John A. Calarco72, Lincoln D. Stein16,73, Gary D. Bader 16,74, Jüri Reimand 7,10,16, Jiannis Ragoussis 8,9, William A. Weiss12,42,75, Marco A. Marra 11,70, Hiromichi Suzuki 1,3,78✉& Michael D. Taylor 1,2,3,7,15,76,78✉ 1 Developmental & Stem Cell Biology Program, The Hospital for Sick Children, Toronto, ON, Canada. 2 Department of Laboratory Medicine and Pathobiology, University of Toronto, Toronto, ON, Canada. 3 The Arthur and Sonia Labatt Brain Tumour Research Centre, The Hospital for Sick Children, Toronto, ON, Canada. 4 Department of Biochemistry and Molecular Biology, Cumming School of Medicine, University of Calgary, Calgary, AB, Canada. 5 Alberta Children’s Hospital Research Institute, Calgary, AB, Canada. 6 Charbonneau Cancer Institute, University of Calgary, Calgary, AB, Canada. 7 Department of Medical Biophysics, University of Toronto, Toronto, ON, Canada. 8 McGill University Genome Centre, McGill University, Montreal, QC, Canada. 9 Department of Human Genetics, McGill University, Montreal, QC, Canada. 10 Computational Biology Program, Ontario Institute for Cancer Research, Toronto, ON, Canada. 11 Canada’s Michael Smith Genome Sciences Centre, BC Cancer Agency, Vancouver, BC, Canada. 12 Department of Neurology, University of California San Francisco, San Francisco, CA, United States. 13 Department of Cellular and Molecular Pharmacology, University of California San Francisco, San Francisco, CA, United States. 14 Institute of Medical Science, University of Toronto, Toronto, ON, Canada. 15 Division of Neurosurgery, The Hospital for Sick Children, Toronto, ON, Canada. 16 Department of Molecular Genetics, University of Toronto, Toronto, ON, Canada. 17 Division of Haematology/Oncology, Department of Pediatrics, The Hospital for Sick Children, Toronto, ON, Canada. 18 SIREDO Center (pediatric, adolescent and young adults oncology), Institut Curie, University of Paris, Paris, France . 19 INSERM U 830, Institut Curie, Paris, France. 20 Unit of Somatic Genetics, Institut Curie, Paris, France. 21 PSL Research University, Université Paris Sud, Université Paris-Saclay, CNRS UMR 3347, INSERM U1021, Institut Curie, Paris, France. 22 Department of Neurosurgery, Division of Pediatric Neurosurgery, Seoul National University Children’s Hospital, Seoul, South Korea. 23 Hospices Civils de Lyon, Institute of Pathology, University Lyon ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-21883-0 16 NATURE COMMUNICATIONS | (2021) 12:1749 | https://doi.org/10.1038/s41467-021-21883-0 | www.nature.com/naturecommunications 1, Department of Cancer Cell Plasticity–INSERM U1052 Cancer Research Center of Lyon, Lyon, France. 24 Department of Pathology, The Children’s Memorial Health Institute, Warsaw, Poland. 25 Department of Surgery and Anatomy, Faculty of Medicine of Ribeirão Preto, University of Sao Paulo, São Paulo, Brazil. 26 Developmental Tumor Biology Laboratory, Hospital Sant Joan de Déu, Esplugues de Llobregat, Barcelona, Spain. 27 Departments of Pathology, Ophthalmology and Oncology, John Hopkins University School of Medicine, Baltimore, MD, United States. 28 Department of Hematology & Medical Oncology, School of Medicine and Winship Cancer Institute, Emory University, Atlanta, GA, United States. 29 Department of Neurosurgery, Kitasato University School of Medicine, Sagamihara, Kanagawa, Japan. 30 Department of Neurology, Erasmus University Medical Center, Rotterdam, Netherlands. 31 Department of Pathology, Erasmus University Medical Center, Rotterdam, Netherlands. 32 Division of Experimental Medicine, McGill University, Montreal, QC, Canada. 33 Department of Pathology and Molecular Medicine, Division of Anatomical Pathology, McMaster University, Hamilton, ON, Canada. 34 Department of Pathology and Laboratory Medicine, Hamilton General Hospital, Hamilton, ON, Canada. 35 Department of Neurological Surgery, University of Pittsburgh School of Medicine, Pittsburgh, PA, United States. 36 Department of Pathology, University of Pittsburgh School of Medicine, Pittsburgh, PA, United States. 37 Division of Pediatric Hematology/ Oncology, Mayo Clinic, Rochester, MN, United States. 38 Department of Laboratory Medicine and Pathology, Mayo Clinic, Rochester, MN, United States. 39 Clinical Research Division, Fred Hutchinson Cancer Research Center, Seattle, WA, United States. 40 Department of Neurosurgery, University of Debrecen, Medical and Health Science Centre, Debrecen, Hungary. 41 Department of Pediatric Oncology, Masaryk University School of Medicine, Brno, Czech Republic. 42 Department of Neurological Surgery, University of California San Francisco, San Francisco, CA, United States. 43 Department of Pathology, University of California San Francisco, San Francisco, CA, United States. 44 Department of Neurological Surgery, Vanderbilt Medical Center, Nashville, TN, United States. 45 Department of Neurology, Vanderbilt Medical Center, Nashville, TN, United States. 46 Departments of Neuroscience, Washington University School of Medicine in St. Louis, St. Louis, MO, United States. 47 Department of Neurosurgery, David Geffen School of Medicine at UCLA, Los Angeles, California, United States. 48 2nd Department of Pediatrics, Semmelweis University, Budapest, Hungary. 49 Department of Anatomical and Cellular Pathology, The Chinese University of Hong Kong, Shatin, New Territories, Hong Kong. 50 Department of Surgery, The Chinese University of Hong Kong, Shatin, New Territories, Hong Kong. 51 Department of Neurosurgery, University of Alabama at Birmingham, Birmingham, AL, United States. 52 Department of Neurosurgery, Chonnam National University Research Institute of Medical Sciences, Chonnam National University Hwasun Hospital and Medical School, Hwasun-gun, Jeollanam-do, South Korea. 53 Department of Pathology, Duke University, Durham, NC, United States. 54 Department of Neurosurgery, Duke University, Durham, NC, United States. 55 Department of Pathology and Neurosurgery, NYU Grossman School of Medicine and NYU Langone Health, New York, NY, United States. 56 Department of Pediatrics, University of Colorado Denver, Aurora, CO, United States. 57 Department of Neurosurgery, University of Ulsan, Asan Medical Center, Seoul, South Korea. 58 U.O. Neurochirurgia, Istituto Giannina Gaslini, Genova, Italy. 59 Institute of Neuropathology, University Medical Center, Hamburg-Eppendorf, Germany. 60 Research Institute Children’s Cancer Center, Hamburg, Germany. 61 Pediatric Hematology and Oncology, University Medical Center, Hamburg-Eppendorf, Germany. 62 Division of Stem Cell Research, Institute for Clinical Research, Osaka National Hospital, Osaka, Japan. 63 Division of Neurosurgery, Centro Hospitalar Lisboa Norte (CHULN), Hospital de Santa Maria, Lisbon, Portugal. 64 Instituto de Medicina Molecular João Lobo Antunes, Faculdade de Medicina, Universidade de Lisboa, Lisbon, Portugal. 65 Division of Pediatric Hematology/Oncology, Hospital Pediatría Centro Médico Nacional century XXI, Mexico City, Mexico. 66 Division of Neurosurgery, Toronto Western Hospital, University Health Network, Toronto, ON, Canada. 67 MacFeeters-Hamilton Center for Neuro-Oncology Research, Princess Margaret Cancer Centre, University Health Network, Toronto, ON, Canada. 68 Department of Surgery, Division of Thoracic and Upper Gastrointestinal Surgery, Faculty of Medicine, McGill University, Montreal, QC, Canada. 69 Cancer Research Program, Research Institute of the McGill University Health Centre, Montreal, QC, Canada. 70 Department of Medical Genetics, University of British Columbia, Vancouver, BC, Canada. 71 Department of Molecular Biology & Biochemistry, Simon Fraser University, Burnaby, BC, Canada. 72 Department of Cell & Systems Biology, University of Toronto, Toronto, ON, Canada. 73 Adaptive Oncology, Ontario Institute for Cancer Research, Toronto, ON, Canada. 74 The Donnelly Centre, University of Toronto, Toronto, ON, Canada. 75 Department of Pediatrics, University of California San Francisco, San Francisco, CA, United States. 76 Department of Surgery, University of Toronto, Toronto, ON, Canada. 77 These authors contributed equally: Patryk Skowron, Hamza Farooq, Florence M. G. Cavalli, A. Sorana Morrissy. 78 These authors jointly supervised this work: Hiromichi Suzuki, Michael D. Taylor. ✉email: [email protected];[email protected] NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-21883-0 ARTICLE NATURE COMMUNICATIONS | (2021) 12:1749 | https://doi.org/10.1038/s41467-021-21883-0 | www.nature.com/naturecommunications 17