SUMMARY While molecular subgrouping has revolutionized medulloblastoma classification, the extent of heterogeneity within subgroups is unknown. Similarity network fusion (SNF) applied to genome-wide DNA methylation and gene expression data across 763 primary samples identifies very homogeneous clusters of patients, supporting the presence of medulloblastoma subtypes. After integration of somatic copy-number alterations, and clinical features specific to each cluster, we identify 12 different subtypes of medulloblastoma. Integrative analysis using SNF further delineates group 3 from group 4 medulloblastoma, which is not as readily apparent through analyses of individual data types. Two clear subtypes of infants with Sonic Hedgehog medulloblastoma with disparate outcomes and biology are identified. Medulloblastoma subtypes identified through integrative clustering have important implications for stratification of future clinical trials.

Midline pediatric high-grade astrocytomas (pHGAs) are incurable with few treatment targets identified. Most tumors harbor K27M mutations on histone 3 variants. In 40 treatment-naïve midline pHGAs, 39 analyzed by whole-exome sequencing, we find additional somatic mutations specific to tumor location. Gain-of-function mutations in ACVR1 occur in tumors of the pons in conjunction with H3.1 K27M, while FGFR1 mutations/fusions occur in thalamic tumors associated with H3.3 K27M. Hyper-activation of the bone morphogenetic protein (BMP)/ACVR1 developmental pathway in pHGAs harbouring ACVR1 mutations led to increased phospho-SMAD1/5/8 expression and up-regulation of BMP downstream early response genes in tumour cells. Global DNA methylation profiles were significantly associated with the K27M mutation regardless of the mutant H3 variant and irrespective of tumor location, supporting its role in driving the epigenetic phenotype. This significantly expands the potential treatment targets and further justifies pre-treatment biopsy in pHGA as a means to orient therapeutic efforts in this disease.

Understanding the biology that underlies histologically similar but molecularly distinct subgroups of cancer has proven difficult since their defining genetic alterations are often numerous, and the cellular origins of most cancers remain unknown1–3. We sought to decipher this heterogeneity by integrating matched genetic alterations and candidate cells of origin to generate accurate disease models. First, we identified subgroups of human ependymoma, a form of neural tumor that arises throughout the central nervous system (CNS). Subgroup specific alterations included amplifications and homozygous deletions of genes not yet implicated in ependymoma. To select cellular compartments most likely to give rise to subgroups of ependymoma, we matched the transcriptomes of human tumors to those of mouse neural stem cells (NSCs), isolated from different regions of the CNS at different developmental stages, with an intact or deleted Ink4a/Arf locus. The transcriptome of human cerebral ependymomas with amplified EPHB2 and deleted INK4A/ARF matched only that of embryonic cerebral Ink4a/Arf−/− NSCs. Remarkably, activation of Ephb2 signaling in these, but not other NSCs, generated the first mouse model of ependymoma, which is highly penetrant and accurately models the histology and transcriptome of one subgroup of human cerebral tumor. Further comparative analysis of matched mouse and human tumors revealed selective deregulation in the expression and copy number of genes that control synaptogenesis, pinpointing disruption of this pathway as a critical event in the production of this ependymoma subgroup. Our data demonstrate the power of cross-species genomics to meticulously match subgroup specific driver mutations with cellular compartments to model and interrogate cancer subgroups.