Epigenetic activation of a RAS/MYC axis in H3.3K27M-driven cancer

Histone H3 lysine 27 (H3K27M) mutations represent the canonical oncohistone, occurring frequently in midline gliomas but also identified in haematopoietic malignancies and carcinomas. H3K27M functions, at least in part, through widespread changes in H3K27 trimethylation but its role in tumour initiation remains obscure. To address this, we created a transgenic mouse expressing H3.3K27M in diverse progenitor cell populations. H3.3K27M expression drives tumorigenesis in multiple tissues, which is further enhanced by Trp53 deletion. We find that H3.3K27M epigenetically activates a transcriptome, enriched for PRC2 and SOX10 targets, that overrides developmental and tissue specificity and is conserved between H3.3K27M-mutant mouse and human tumours. A key feature of the H3K27M transcriptome is activation of a RAS/MYC axis, which we find can be targeted therapeutically in isogenic and primary DIPG cell lines with H3.3K27M mutations, providing an explanation for the common co-occurrence of alterations in these pathways in human H3.3K27M-driven cancer. Taken together, these results show how H3.3K27M-driven transcriptome remodelling promotes tumorigenesis and will be critical for targeting cancers with these mutations.

D iffuse Intrinsic Pontine Glioma (DIPG) is a devastating paediatric high-grade glioma (HGG) that diffusely penetrates the brainstem and is the leading cause of paediatric brain tumour death 1 . Surgery is not possible, radiation is largely palliative, and conventional chemotherapy and targeted agents have proven ineffective. Around 80% of DIPGs harbour lysine 27methionine (K27M) substitutions in histones H3.3 (H3F3A) and H3.1 (HIST1H3B/C) 2,3 . More recently H3K27 mutations have also been found in a range of other tumours including haematopoietic malignancies and carcinomas [4][5][6][7][8] . Thus, H3K27M is assumed to be an oncohistone, although the precise nature of its oncogenic function remains obscure.
H3K27 is a key hub for transcriptional regulation: acetylation (H3K27ac) is associated with active enhancers while trimethylation (H3K27me3) represses transcription [9][10][11] . H3K27M has increased affinity for, and inhibits the activity of, Enhancer of zeste homologue 2 (EZH2), the catalytic subunit of Polycomb repressive complex 2 (PRC2) that deposits H3K27me3, leading to widespread H3K27me3 loss 12,13 . However, H3K27M also leads to H3K27me3 gains in some areas, suggesting that aberrant gene expression and repression may both be important 14 .
Despite clonal analysis indicating that H3K27M is the tumourinitiating mutation in DIPG 15,16 , in vivo models of H3K27Mdriven cancer, which to date have focused on brain tumours, have failed to generate tumours without additional oncogenic drivers. The endogenous H3f3a locus was modified to allow H3.3K27M expression in Nestin-positive cells, where neonatal H3.3K27M expression decreased the latency of gliomas driven by Trp53 loss and constitutively active Pdgfra 17 . In experiments using RCAS-TVA retroviral delivery, neonatal targeting of Nestin-positive brainstem progenitors with H3.3K27M accelerated tumours driven by Pdgfb overexpression and Trp53 loss 18 Similarly, H3.3K27M decreased the latency of glioma formation in in utero electroporation experiments introducing H3.3K27M alongside Trp53 loss, Pdgfb overexpression, or WT or constitutively active (D842V) Pdgfra overexpression 19,20 . None of these models yielded tumours where the sole exogenous mutation was H3.3K27M. Further, to produce tumours, H3.3K27M was introduced alongside other strong oncogenic drivers making it difficult to parse out the direct role of the mutant histone in oncogenesis. Overall, these models suggest that H3K27M cooperates with p53 dysfunction and constitutive RAS/MAPK activation, but is not itself sufficient to initiate tumorigenesis, at least in the cells and developmental time points tested thus far. This cooperation with p53 and the RAS/MAPK pathway is supported by the frequent co-occurrence of PDGFRA and TP53 alterations with H3K27M in human DIPGs, although in human tumours H3K27M is clearly the initiating event 21,22 . Thus, how H3.3K27M alters the cell state to permit or initiate tumorigenesis and the basis of its cooperation with additional mutations remains elusive. A clearer understanding of these mechanisms is critical to developing appropriate therapies for this devastating disease.
Here we used a mouse model expressing H3.3K27M from the Fabp7 promoter in both neural and non-neural precursor cells to investigate H3K27M-driven cancer. H3.3K27M alone was sufficient to drive development of lymphomas and carcinomas, while Trp53 loss was required for high grade gliomas to form. H3.3K27M imposed a common transcriptome, mirroring early gliomagenesis, which was shared between H3K27M-driven mouse tumours from all sites, H3.3K27M mouse embryonic brainstem and human DIPGs. A main component of the H3.3K27M transcriptome was an activated RAS/MYC axis which could be targeted therapeutically. H3K27M-mutant tumours acquired secondary mutations to reinforce this activation, suggesting a model where early epigenetic pathway activation by H3K27M is later reinforced by pathway activating genetic mutations and explaining the co-occurrence of these mutations in human DIPG.

Results
H3.3K27M increases cancer incidence and leads to early death. CD-1 mice, expressing FLAG/HA-tagged H3.3K27M (H3f3a-K27M) under the control of the Fabp7 promoter, were engineered by microinjection, with colonies established from 3 founders ( Supplementary Fig. S1a-c). Fabp7 promoter-directed H3.3K27M expression was compatible with viable embryonic and postnatal development; adult mice had similar phenotype and litter size to CD1 wild-type mice, although E12.5 and E14.5 H3.3K27Mpositive embryos occasionally developed signs of central nervous system haemorrhage ( Supplementary Fig. S1d). At the gene expression level, Fabp7 expression is highest in brain during embryogenesis, but is expressed in adult brain as well as in other embryonic and adult tissues including liver, intestine, heart, lung and bone ( Supplementary Fig. S1e). Correspondingly, we saw that H3.3K27M expression was highest in brain, yet readily detectable in other tissues ( Supplementary Fig. S1f, g) [23][24][25][26] . Importantly, the H3.3K27M-FLAG/HA transgene was not overexpressed versus endogenous H3.3 in the mouse samples or when compared to H3K27M levels in human DIPG ( Supplementary Fig. S1h, i).
H3.3K27M-driven mouse and human tumours have similar genomic features. We performed whole exome sequencing (WES) on 10 tumours from 8 H3.3K27M, H3.3K27M/Trp53 +/− or H3.3K27M/Trp53 −/− mice to investigate the evolution of H3.3K27M-driven tumorigenesis. We compared the genetic features of the H3K27M mutant mouse tumours with those of human brain and non-brain tumours. The mutation frequency was similar between mouse and human tumours (Fig. 1e, Supplementary Data 3) 5,21 . The burden of copy number variations (CNV) and the fraction of the genome affected by CNVs was also similar between mouse tumours and human H3.3K27M DIPG (Fig. 1f, g Fig. 1h).
H3.3K27M rapidly disrupts transcriptional networks to establish a glioma-like, proliferative phenotype in middevelopment in cooperation with SOX10. In order to investigate the earliest events resulting from H3.3K27M expression we examined brainstem tissue from E14.5 mouse embryos,~5 days after H3.3K27M expression is induced from the Fabp7 promoter. We further generated isogenic cell lines expressing H3.3K27M or empty vector (EV) using the human oligodendrocyte precursor cell (OPC) line MO3.13, as OPCs are a proposed cell of origin for DIPG 34 . H3.3K27M has an inhibitory effect on PRC2, and its expression in E14.5 midbrain/hindbrain (MB/HB) or MO3.13 cells reduced H3K27me3 compared to controls (Fig. 2a, b and Supplementary  Fig. S3a). This was accompanied by an increase in the proliferative markers PCNA and phospho-H3S10 in H3.3K27M mice compared with WT (Fig. 2a). A similar increase in proliferation was seen in the MO3.13 cells compared with empty vector (EV) and H3.3WT-expressing control cells (Fig. 2c). In a scratch assay, while EV and H3.3WT cells failed to close a wound after 24 h, H3.3K27M-expressing MO3.13 cells had fully closed the wound, further suggesting that H3.3K27M confers protumorigenic behaviour on MO3.13 cells ( Supplementary  Fig. S3b).
While loss of H3K27me3 is permissive for gene transcription, this loss alone is not enough; transcriptional upregulation is also dependent on the presence of transcription factors. We combined differential gene expression, transcription factor binding sites and protein-protein interactions to identify potential master regulators of the H3K27M phenotype. We found that the transcription factor SOX10 directly regulates~30% of genes upregulated by H3.3K27M in E14.5 MB/HB, 3.4x more than are direct PRC2targets (Fig. 3a). SOX10 influences lineage commitment in the developing brain to promote oligodendrogenesis 45 , and we noted  46 . Overall, the majority of genes upregulated in the early response to H3.3K27M are not canonical PRC2 target genes. This data suggests that SOX10 may be at least partially responsible for mediating this early response in the MB/HB by promoting oligodendrogenic expansion at the expense of neuronal differentiation.
To better understand how H3.3K27M and SOX10 interact we depleted SOX10 expression in MO3.13 cells ( Supplementary  Fig. S4g), and repeated the 4-day differentiation assay (Fig. 2f). S100-β and MBP are both SOX10 target genes 42 . H3.3K27M did not induce expression of these genes after SOX10 knockdown, confirming the importance of SOX10 in executing the H3.3K27M transcriptome (Fig. 3g). As expected, all differentiation conditions decreased cell proliferation (p < 0.0001; Fig. 3h and Supplementary Fig. S4h). Strikingly, SOX10 depletion reversed the growth advantage conferred by H3.3K27M in regular media and, when combined with differentiation media, H3.3K27M cells grew substantially more slowly than EV and H3.3WT controls (p < 0.0001 when normalising by EV condition to control for the dominating media effect; Fig. 3h and Supplementary Fig. S4h). Moreover, SOX10 depletion had a synthetic lethal effect with H3.3K27M expression, particularly under differentiation conditions, while EV and H3.3WT cell viability was unaffected (p < 0.0001; Fig. 3i and Supplementary Fig. 4i). Collectively these data suggest that H3K27M co-operates with SOX10 in the regulation of transcription programmes involved in cell fate, proliferation, and survival of OL cells consistent with our findings in H3K27M brains, mouse HGGs, and human DIPGs. In keeping with this, the H3.3K27M/SOX10 signature was highly upregulated in both H3.3K27M-mutant mouse HGG and human DIPG (Fig. 3j). Furthermore, 20-30% of differentially expressed genes in the E14.5, mouse HGG and human DIPG datasets are SOX10 target genes, suggesting that SOX10 activity is maintained throughout tumorigenesis.
To further investigate the similarities between the early effects of H3.3K27M on the brainstem and gliomagenesis we compared the transcriptomes of H3.3K27M mutant E14.5 brainstem with H3.3K27M/Trp53 +/− mouse HGGs and human DIPGs. The differentially expressed genes and pathways in all 3 datasets were strikingly similar (all two-way overlaps p ≤ 10 −7 ; Fig. 4a, b). 167 commonly regulated genes were enriched in extracellular remodelling pathways, consistent with the highly invasive nature of human DIPG (Fig. 4c), and 52 were SOX10 target genes, underscoring the importance of SOX10 activity in establishing the early transcriptional response to H3.3K27M that is maintained throughout tumorigenesis (p = 10 −56 , Fig. 4d). The 53 pathways commonly regulated across all 3 datasets included proliferative (RAS signalling, cell cycle, MYC), extracellular remodelling (epithelial-mesenchymal transition, adhesion) and metabolic (glycolysis, oxidative phosphorylation) terms. All 3 datasets showed strong enrichment of the mesenchymal subtype and strong repression of neural/proneural glioblastoma subtype signatures (Fig. 4e) 47 . These data suggest that H3.3K27M rapidly establishes a transcriptome in the brainstem that recapitulates features of gliomagenesis.
H3.3K27M drives a core transcriptome in different cellular contexts. To understand the effect of H3.3K27M in different tissues we compared RNA-Seq of HGG, lymphoma and brain from H3.3K27M mice with adult wild-type mouse brain, spleen, thymus and bone marrow 48 , as well as human DIPG and normal brain. Single-sample GSEA (ssGSEA) was used to analyse relative pathway activity across different samples. T-stochastic neighbour embedding (t-SNE) models and hierarchical clustering revealed that, while brain samples clustered separately from bone marrow, spleen and thymus, reflecting expected inter-tissue transcriptional differences, H3.3K27M-driven tumours clustered together, distinctly from their tissues of origin ( Fig. 4f and Supplementary Fig.  S5a). Similarly, mouse H3.3K27M HGG and lymphomas grouped with human H3.3K27M DIPG, while control H3.3K27M adult mouse brain samples grouped with normal human brain (Fig. 4g). This suggests that H3.3K27M establishes a transcriptome that is conserved across tumour types and overrides tissue-specific expression patterns. Gene sets or transcriptional pathways commonly regulated between H3.3K27M mouse HGG, H3.3K27M mouse lymphoma and human DIPG included RAS, PI3K/AKT signalling and MYC targets (Figs. 4h, 4i, 5a and Supplementary  Fig. S5b). Finally, we asked whether the H3.3K27M-SOX10 axis (Fig. 3) has a role in regulation of these pathways. Except for MYC targets at E14.5, which were not upregulated at E14.5 and therefore have no overlap with SOX10-regulated genes at this Fig. 2 H3.3K27M rapidly induces a glioma-like proliferative phenotype in developing brainstem and subverts oligodendrocyte differentiation programmes. a Western blot of H3.3K27M and WT MB/HB from littermates harvested at E14.5. Densitometry quantifies H3.3K27M versus WT (normalised to actin) and is plotted as mean ± standard deviation (n = 3 independent embryos). b Western blot of empty vector (EV), H3.3WT or H3.3K27M MO3.13 cells (results are representative of 2 independent experiments from different litters). c Growth curve of MO3.13 cells transduced with empty vector (EV) or H3.3K27M. Bars show mean ± standard deviation (n = 3). d GSEA (gene set enrichment analysis) of BENPORATH_PRC2_TARGETS gene set in H3.3K27M E14.5 MB/HB. NES: Normalised Enrichment Score. FDR: false discovery rate. e Heatmap and GSEA of differentially expressed genes in H3.3K27M E14.5 MB/HB with functions in different lineages. Oligo: oligodendrocyte, Astro: astrocyte. f Schematic of experiment differentiating MO3.13 cells from oligodendrocyte precusor cells (OPC) into early and late oligodendrocytes (OL) experiment using serum starvation and (PMA) phorbol 12myristate 13-acetate treatment. g RT-PCR analysis of MO3.13 cells grown in regular media or differentiated for 4 days. <: specific band. Results are representative of 6 independent replicates. h Western blot of MO3.13 cells grown in regular media or differentiated for 4 days. <: specific band. Results are representative of 2 independent replicates. i RT-PCR analysis of MO3.13 cells grown in regular media or differentiated for 10 days. <: specific band. Results are representative of 2 independent replicates. j Bright-field microscopy images of MO3.13 cells grown in regular media or differentiated for 10 days. Results are representative of 2 independent replicates. Scale bar: 100 µm. Source data are provided as a Source Data file. Statistical tests: t (a, c). ****p < 0.0001. developmental stage, 30-40% of the RAS/MYC targets and ECM genes differentially expressed in each of H3.3K27M E14.5 MB/ HB, mouse HGG and human DIPG are SOX10 target genes ( Supplementary Fig. S5c).
Recent data has suggested repression of CDKN2A may be important in H3.3K27M-mediated tumorigenesis, as this gene retains its promoter H3K27me3 in the presence of the oncohistone, and the expression level of p16 regulates the proliferative capacity of H3.3K27M-mutant cells 18,46 . We examined Cdkn2a promoter H3K27me3 in E14.5 MB/HB, finding that H3.3K27M reduced Cdkn2a methylation at this developmental time point (Supplementary Fig. S5d). There was not an accompanying expression change in Cdkn2a in RNA-Seq data and, in our mouse HGG and human DIPG datasets, it was   Fig. S5e, f), suggesting that p16 repression is not a key feature of tumorigenesis in this mouse model.
RAS/MAPK activation is an early response to H3.3K27M. RAS signalling was one of the most strongly activated pathways in established H3.3K27M mouse HGG and lymphoma as well as human DIPG (Fig. 4h, i and Supplementary Fig. S5b). To ask if this was an early response to H3.3K27M, we examined E14.5 MB/ HB. At this time point H3.3K27M had been expressed for~5 days and a proliferative phenotype established (Fig. 2a). Western blotting and RNA-Seq analysis of E14.5 mouse brainstem revealed that RAS/MAPK signalling was indeed already activated in an H3.3K27M associated manner (Fig. 5a-c). Pdgfra, one of the main receptors activating RAS signalling, was upregulated in H3.3K27M E14.5 MB/HB, as well as in mouse HGG and human DIPG (Fig. 5d-f and Supplementary Fig. S5h), and the promoters of Pdgfra, Hras, Kras and Nras had reduced H3K27me3 at E14.5 ( Fig. 5g and Supplementary Fig. S5i). MO3.13 cells expressing H3.3K27M had increased MEK and ERK phosphorylation compared with their empty vector (EV) counterparts (Fig. 5h). H3.3K27M markedly reduced H3K27me3 in the promoters of RAS target genes in NSCs (Fig. 5i) 46 . This data suggested that H3.3K27M drives epigenetic activation of the RAS/MAPK pathway by targeting multiple pathway members and its downstream target genes. To further test the dependence of the RAS/MAPK pathway activation on H3.3K27M expression, we next examined H3.3K27M-mutant primary paediatric HGG (pHGG) BT245 cells in which the H3.3K27M mutation had been reversed using CRISPR 49 . BT245-M27K cells showed increased H3K27me3 in the promoters of RAS target genes and significantly downregulated their expression compared with parental H3.3K27M-mutant BT245 cells (Fig. 6a, b). Similarly, depletion of H3F3A by shRNA in (H3F3A-K27M) H3.3K27M-mutant primary DIPG cells restored RAS target gene promoter H3K27me3 (Supplementary Fig. S6a) 50 . In H3-WT G477 pHGG cells overexpressing H3.3K27M or H3.3K27R 49 , K27R had no effect on H3K27me3 while K27M upregulated RAS target genes and decreased their promoter H3K27me3 (Fig. 6c and Supplementary Fig. S6b). Finally, in a panel of primary pHGG cells, H3.3K27M-mutant cells had reduced RAS target gene promoter H3K27me3 compared with H3WT cells (Fig. 6d) 49 . Together, this suggests that H3.3K27Mmutant tumours exhibit K27M-mediated epigenetic activation of the RAS/MAPK pathway.
Early epigenetic activation of RAS/MAPK pathway is later reinforced by genetic activation. Based on sequencing data it is well established that mutations in the RAS/MAPK pathway are frequent events in human DIPG 51 suggesting activation of this pathway is an important step in DIPG tumorigenesis with clonal evolution modelling implicating these as later events 15,16 . Interestingly, our RNAseq data showed activation of the RAS/MAPK pathway in both mouse tumours and human DIPG regardless of whether the RAS/MAPK/PI3K pathway was mutated ( Fig. 6e-g, Supplementary Fig. S6c and S6d) suggesting that activation of this pathway may be universal with epigenetic/non-mutation-dependent mechanisms responsible for this activation in the nonmutant cases. Intriguingly, in another mouse model where H3.3K27M accelerated development of Trp53 KO /Pdgfra-driven HGG 17 , the addition of H3.3K27M epigenetically activated RAS target genes compared to the H3.3WT HGG despite the pathway already being activated by constitutively active PDGFRA (Supplementary Fig. S6e, f) 17 suggesting epigenetic mechanisms of pathway activation may play a role, even in mutant tumours.
The rapid induction of a RAS/MAPK signalling signature in both our mouse and cell line models following introduction of H3.3K27M as detailed above suggests that epigenetic activation of the pathway may be an early event, with mutation-dependent activation occurring later in tumour development or in a subclonal fashion. This is consistent with a model where the early epigenetic activation of RAS by H3.3K27M is later reinforced by genetic activation of the pathway.
To test if RAS/MAPK genetic alterations in the H3.3K27M mouse tumours are early or late events we investigated the clonal evolution of these tumours (Fig. 6h). For the KP29 HGG and lung carcinoma this also provided a unique opportunity to understand how H3.3K27M affects tumorigenesis of two different tissues within the same animal. These two tumours had little genetic similarity, indicating that both were primary ( Fig. 1h and Supplementary Fig. S2e). RAS was activated in both tumours, with Kras alterations in the second and first clone of the HGG and lung carcinoma, respectively (Fig. 6h). The HGG additionally had a single copy Nf1 deletion in clone 1. The lung carcinoma also acquired a focal Mycn amplification and single-copy chromosome 11 (Trp53) loss in the first clone, and a broad single-copy loss in chromosome 4, containing Cdkn2a, in the second (Fig. 6h). Overall, as with human DIPG, most H3.3K27Mdriven mouse tumours acquire additional, subclonal, oncogenic alterations beyond the truncal histone mutation and Trp53 loss. Intriguingly, these alterations converge on the RAS/MAPK/PI3K pathway and MYC/MYCN, which are key features of H3.3K27M DIPG 21 .
Epigenetic activation of a RAS/MYC axis in H3.3K27M-driven cancer. MYC target genes were among the most upregulated gene sets in H3.3K27M mouse HGG and lymphomas, and human DIPGs (Fig. 4h, i and Supplementary Fig. S5b). As with RAS signalling, H3.3K27M mouse HGG and human DIPG Fig. 3 SOX10 is important for the early response to H3.3K27M and activates a signature maintained in gliomas. a Percentage of upregulated genes in H3.3K27M E14.5 MB/HB that are directly regulated by PRC2 and/or SOX10. b Western blot of H3.3K27M and WT MB/HB from littermates harvested at E14.5. Note that the actin loading control is shared with Fig. 5e (results are representative of at least 3 independent experiments). c SOX10 expression in human H3.3K27M DIPG (n = 28) and normal brain (n = 20). d Network diagram showing SOX10 transcriptional targets from key pathways that are differentially expressed at E14.5. Red: upregulated genes; blue: downregulated genes; colour intensity reflects expression fold-change. DEG: differentially expressed gene. e ChIP quantifying H3K27me3 at the Sox10 promoter in E14.5 WT or H3.3K27M MB/HB. f H3K27me3 ChIP-Seq density was profiled ± 15 kb around the (transcription start site) TSS of SOX10 target genes in H3.3K27M or H3.3WT NSCs 46 . g RT-PCR analysis of MO3.13 cells transduced with control (shCTR) or SOX10-specific (shSOX10) shRNA and grown for 4 days in regular or differentiation media. Results are representative of 6 independent replicates (shSOX10 clone 1 n = 2, clone 2 n = 4). EV: empty vector. h Cell counts of MO3.13 cells grown as in g. Bars show mean ± standard deviation (n = 3). i Percent viability of MO3.13 cells grown as in g. Bars show mean ± standard deviation (n = 3). # significance for EV vs H3.3K27M in shSOX10 cells grown in differentiation media was tested by normalising each condition by EV to control for media effect. j SOX10 target genes upregulated in H3.3K27M E14.5 MB/HB were used for gene set enrichment analysis (GSEA) in H3.3K27M mouse HGG and human DIPG. Source data are provided as a Source Data file. Statistical tests: t (c, e), ANOVA (h, i).    upregulated MYC targets regardless of whether they were MYC amplified ( Fig. 7a and Supplementary Fig. S7a-c). Overexpression of H3.3K27M in G477 HGG cells or NSCs resulted in loss of H3K27me3 in MYC target genes compared with control cells and their upregulation in H3.3K27M-G477 cells (Fig. 7b and Supplementary Fig. S7d, e) 46,49 . Similarly, the presence of H3.3K27M in Trp53 KO /Pdgfra mouse HGG led to upregulation of MYC target genes associated with loss of promoter H3K27me3 (Supplementary Fig. S7f, g) 17 . Conversely, depletion of H3F3A from H3.3K27M-mutant DIPG cells increased MYC target gene promoter H3K27me3 (Supplementary Fig. S7h) 50 . Furthermore, in BT245 HGG cells, even though MYC is amplified in these cells, CRISPR-mediated removal of H3.3K27M resulted in restoration of promoter H3K27me3 and MYC target gene downregulation (Fig. 7c, d) 49 . In agreement with this, primary H3.3K27M-mutant pHGG cells had reduced H3K27me3 at MYC target genes compared with H3WT cells (Fig. 7e) 49 . Together, this data indicates that H3.3K27M drives epigenetic activation of the MYC pathway.  Fig. S7i). Importantly, H3.3K27M MO3.13 cells had significantly increased cell death in response to Omomyc expression compared with control cells (Fig. 7f). MYC is a known effector gene of RAS signalling, which targets MYC twice (Supplementary Fig. S7j): MYC-S62 phosphorylation by ERK stabilizes MYC and prolongs its activation, and AKT suppresses GSK-3-mediated MYC-T58 phosphorylation leading to MYC degradation 53 . To test if these mechanisms were at play in the context of H3K27M we used H3.3K27M-expressing MO3.13 cells. These had increased phospho-MYC-S62 as well as activated RAS/MAPK compared with their EV counterparts, suggesting that both H3.3K27M and RAS signalling have roles in MYC activation (Figs. 5h, 7g). In support of this, RAS and MYC activation are tightly coupled in H3.3K27M DIPG (Fig. 7h).
We then examined the effects of AKT inhibition on phospho-MYC-T58 in MO3.13 cells. As previously reported, A443654 increased phospho-AKT-S473 54 . Phospho-MYC-T58 was increased in both A443654-treated WT and K27M MO3.13 cells, accompanied by a 30-40% reduction in the overall amount of MYC and HSP90 (Fig. 7i). However, H3.3K27M-MO3.13 cells were significantly more sensitive to A443654 (IC50 29 nM vs EV IC50 72 nM, p < 0.0001) than WT MO3.13 cells, suggesting that K27M alone leads to increased dependency on MAPK-mediated MYC activation (Fig. 7j). Thus, as a clinically viable alternative to directly targeting MYC, targeting either the MAPK or PI3K/AKT arms of the RAS cascade ( Supplementary Fig. S7j) could be used. To test this as a potential DIPG therapy we investigated the effects of multiple MEK inhibitors and AKT inhibition on 5 H3K27M-mutant primary DIPG cell lines. Both MEK and AKT inhibition resulted in dose-dependent DIPG cell death with IC50s in the nM range (Fig. 7k, l, Supplementary Fig. S7k, l) supporting these as important therapeutic targets for H3K27M-driven cancer.

Discussion
H3K27M mutations were first identified as high-frequency drivers of DIPG, and more recently have been found in additional tumour types [2][3][4][5][6][7][8] . Previous pre-clinical mouse models of K27M-driven cancer focussed on generating brain tumours and revealed a cooperative effect on p53/PDGFRA-driven mouse HGG 17,19,55 . However, the oncogenic role of H3K27M itself was unclear. To investigate this we created a mouse where H3.3K27M is expressed from the Fabp7 promoter in radial glia and astrocyte progenitors from E9.5, as well as in developing and adult non-brain tissues [23][24][25][26]56 . H3.3K27M mice developed tumours in multiple tissues, with increased frequency and decreased latency, with and without additional Trp53-deficiency. Thus, this is the first demonstration that H3.3K27M is sufficient to drive tumorigenesis in the absence of other initiating mutations, establishing its role as an oncohistone. This is in keeping with recent work where removing the K27M mutation from DIPG and HGG cells by CRISPR/Cas9 inhibited their tumorigenicity 49 . HGG occurred exclusively in association with Trp53 loss in our model, while H3.3K27M alone increased lymphoma and carcinoma susceptibility. Notably, the other tumour types in which H3K27 mutations have been identified in humans are haematopoietic malignancies and carcinomas 6-8 , perhaps explaining the spectrum of H3K27M-related cancers we saw in our model. H3K27M-mediated disruption of PRC2 in these diseases is in keeping with frequent PRC2 deregulation in haematological neoplasms and lung carcinomas, either by mutation ( Supplementary Fig. S8) or differential expression [57][58][59][60][61] . Finally, previous studies showed an association of high levels of H3.3 and invasive phenotypes in already developed cancers 62,63 .
While our in vitro studies support a phenotypic effect related to the mutant histone, rather than to ectopic expression of H3, our study is limited by the lack of a mouse line expressing the wildtype histone under the Fabp7 promoter. However, transgene expression in our model was very modest compared with endogenous H3.3 and H3.3K27M in human DIPG ( Supplementary  Fig. S1h, i), and the phenotypes we observed were related to cancer initiation rather than to metastasis, suggesting that they are likely due to the K27M mutation rather than histone overexpression. Although H3.3K27M expression disrupted canonical PRC2 target genes in the E14.5 MB/HB, many other genes were also upregulated, indicating that although PRC2 is important, the mutant histone must also act through additional signalling and transcriptional pathways to drive transformation. SOX10 and its target genes were epigenetically activated by H3.3K27M and, consistent with the role of SOX10 in promoting OPC commitment 45,64,65 , we found increased expression of oligodendroglial markers at the expense of other lineages. Importantly, an OPC-like cell is hypothesised as a putative DIPG cell of origin 34,55 , suggesting that SOX10 may have a key role in mediating the effects of H3.3K27M. This was confirmed in our cell culture model of OPC differentiation where H3K27M and SOX10 cooperated in activation of a mid-differentiation oligodendroglial transcription programme. Inhibition of SOX10 in this model also led to decreased proliferation and survival specifically in H3K27M cells, suggesting that H3K27M may use SOX10 as a proxy for maintaining oligodendroglial-like cells in a proliferative state. Given the frequent tissue specificity of transcription factors, it is likely that H3.3K27M will partner with other transcription factors in non-glial tissues.
H3.3K27M drove tumorigenesis in multiple tissues and, unexpectedly, the transcriptomes of H3.3K27M-driven tumours were far more similar to one another than to other WT tissues, including their tissues of origin. Remarkably, when comparing H3.3K27M mouse tissues with human DIPGs and normal brain, the mouse HGGs and lymphomas grouped with DIPG. Thus, oncohistone-mediated epigenetic changes override tissue-specific programmes to impose a core, lineage-independent H3.3K27M oncotranscriptome.
Among the most enriched pathways between H3.3K27M-driven mouse cancers and human DIPGs were target genes of RAS/ MAPK and MYC. We found that, both in vitro and in vivo, these pathways were epigenetically activated in an H3.3K27M-dependent fashion, including through upregulation of PDGFRA and MYC. This is supported by recent work showing that H3.3K27M leads to in vitro activation of the RAS/MAPK cascade 66  Interestingly, human DIPG and H3.3K27M-driven mouse tumours both develop secondary RAS/MAPK/PI3K pathway and MYC mutations. This suggests a model in which H3.3K27M initiates tumorigenesis in part through epigenetic activation of RAS and MYC and later mutational events act to lock in this activation in a portion of tumours, providing an explanation for the frequent association of PDGFRA/PI3K and MYC alterations in H3.3K27M DIPG (Supplementary Fig. S9).
Western blot analysis. Whole cell lysates of organs and cells were prepared in 2x SDS lysis buffer (20 mM Tris [pH 7.4], 20 mM EDTA, 2% SDS, 20% glycerol) and concentration determined by DC Protein Assay (Bio-Rad). 30 µg of protein was resolved on 10-20% SDS-PAGE and transferred to PVDF membranes that were blocked and incubated with antibodies diluted in 3% BSA in TBS-T. Binding was detected with enhanced chemiluminescence (Pierce).
Immunohistochemistry. Organs and tissues were fixed in 4% PFA and embedded in paraffin. Antigen retrieval using heat and citrate buffer was included for all antibodies. Signal was detected with DAB peroxidase substrate (Vector Laboratories). Images were captured on a Nikon Eclipse E600 microscope.  Reads were trimmed with Trimmomatic-v0.32 69 and aligned to GRCm38-v68 (mouse) or GRCh37-v75 (human) using STAR v2.5.0 70 in two-pass mode; duplicate reads were marked with Picard-v2.5.0. Gene expression was counted with HTSeq 71 , and differential expression calculated with edgeR 72 . Genes with absolute fold-change > 1 and Benjamini-Hochberg adjusted p-value < 0.05 were classed differentially expressed. Pathways were analysed with DAVID 73 . Genes were ranked by multiplying their fold-change sign with the -log10(adjusted p-value) for pre-ranked GSEA 37 , using human homologues of mouse genes obtained from Ensembl. For ssGSEA, reads were aligned to the transcriptome using RSEM-v1.2 74 . Genes with mean FPKM < 1 were discarded, and genes with duplicated names were filtered to keep the most expressed gene. SOX10 target genes were from the Harmonizome 75 . Fabp7 RNA-Seq expression data was from the Gene Expression Database 76 .
Network analysis. Transcription factor (TF) networks were scored on three metrics, similar to previous methods 77 , with improvements to network weighting by gene differential expression. Differential gene expression scores were calculated as for pre-ranked GSEA. Networks of TF-DNA and TF-Protein interaction edges were constructed from MARA 78 and String-DB 79 , respectively, allowing a maximal edge distance of 3 from the root TF. The network score (N t,n ) was computed by N t;n ¼ X rϵV t S r Á S n D r;n Á L r;n where r∊V t is the set of all genes (r) in the local subnetwork of transcription factor t; S r is the differential expression gene score, S n is the MARA/String relationship score; D r,n and L r,n is the edge distance of gene r from t and the outdegree of the parent node in the network, respectively. Each TF network was scored as the aggregate rank of each sub-score in decreasing order, such that the higher the rank, the stronger the effect of the TF influence among differentially expressed genes.
End-point PCR and Q-PCR. Total RNA was reverse transcribed using Reverse Transcription kit (Applied Biosystems). End point PCR was performed for 28 cycles unless indicated otherwise. End-point PCR-validated primers were used for qPCR with iTaq Universal SYBR green supermix (Bio-Rad).
Mouse exome sequencing. DNA was extracted from frozen tissue samples with DNeasy kit (Qiagen). Exome libraries were generated with the SureSelect Mouse All Exon Kit (Agilent). Paired end 125 bp Illumina HiSeq 2500 sequencing was done at The Hospital for Sick Children. We sequenced a pool of normal CD1 mice because there is no available CD1 reference genome or genome-wide SNP analysis.
Somatic variants were called with VarScan-v2.3.8 82 and Mutect2 83 , retaining those identified by both. For unmatched tumours, variants were called individually against all normals including the CD1 pool, retaining those identified with both tools against every normal and additionally discarding known variants in other mouse strains, as we could not exclude the possibility that these are also present in a particular CD1 mouse. Variants with >10x coverage in both tumour and normal samples, a minimum tumour variant allele frequency (VAF) of 0.05 and maximum normal VAF of 0.01 were annotated using SnpEFF-v4.3k 84 . Variants were compared with known human cancer variants using the COSMIC-v81 85 . Clonal evolution of tumours with a matched normal was analysed with SuperFreq 86 . Copy number variants were identified using both on-and off-target reads with CNVkit v0.8.6 87 .
Human exome sequencing. DNA was extracted from frozen tissue samples with DNeasy kit (Qiagen). Libraries were created with Ion TargetSeq Exome 50 Mb library (ThermoFisher), sequenced on Ion Proton machines (ThermoFisher), and aligned to human genome build hg19 with Torrent Suite Software (Ther-moFisher) at The Hospital for Sick Children. Variants were called with VarScan and annotated with SnpEFF. RAS pathway and MYC mutant samples were determined by the presence of an alteration in a core RAS/MAPK/PI3K pathway gene (Supplementary Data 4) or MYC, respectively. These were restricted to SNVs/indels that introduced missense, frameshift, nonsense or splice site mutations at variant-allele frequency >0.2 and with coverage >20, a copy number gain with 5+ copies, or a homozygous deletion, had to be present in the sample for any pathway gene. Wild-type samples had all pathway genes free of SNV/ indels and copy number changes.
Statistical analysis. Unless otherwise stated, all p-values were calculated by twotailed t-tests, not assuming equal variance between samples.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Raw mouse WES and RNA-seq data is available from the Gene Expression Omnibus (GEO), accession GSE120884. Human WES and RNA-seq data is available from the European Genomics Archive, accession EGAD00001006450 (https://ega-archive.org). Publicly available data was from GEO (https://www.ncbi.nlm.nih.gov/geo accessions GSE115875, GSE85390, GSE108364) or https://datahub-jv6f4mbl.udes.genap.ca, and is referenced both within the article and in Supplementary Table S8. All other information supporting the findings of this study are available within the article, its Supplementary Information files, a Source Data file and from the corresponding author upon reasonable request. Source data are provided with this paper.