Abstract Medulloblastoma (MB) is a malignant cerebellar tumor composed of biologically distinct molecular subgroups. Single-cell transcriptomic studies have linked MB origins to the developing rhombic lip (RL). Among consensus molecular subgroups, Group 3/4-MBs exhibit the most extensive heterogeneity, with eight subtypes defined by bulk DNA methylation landscapes. While molecular features of MB subgroups are well defined, cellular and epigenetic heterogeneity within subgroups and their constituent subtypes is still unresolved. Here, we generated whole-genome single-cell DNA methylation (scDNAm) profiles from 31 primary MB tumors spanning all subgroups and subtypes, alongside profiles of the developing RL across four prenatal timepoints. By leveraging snATAC-seq data from the same RL populations, we annotated RL cell types using hypomethylated ATAC peaks and robustly projected tumor cells onto their corresponding developmental states in DNA methylation space. We developed a neural network classifier trained on ∼2,000 bulk MB methylation array profiles, enabling assignment of molecular subgroups and subtypes at single-cell resolution and revealed substantial intra-patient heterogeneity. To dissect this heterogeneity, we applied Non-Negative Matrix Factorization to the array data and identified five major methylation programs in Group 3/4 MBs. Quantifying these programs in single cells revealed continuous variation along each program within samples, reflecting a spectrum of methylation states across cells. Importantly, cell clusters stratified by these programs exhibited locus-specific methylation changes relative to the most similar RL cells, highlighting somatic epigenetic alterations that diverge from normal development and may contribute to tumorigenesis. Using an extensive scDNAm dataset of MB as a foundation, our study provides a novel framework for interpreting tumor heterogeneity at cellular resolution. Ongoing analyses aim to identify early epigenetic changes driving tumorigenesis and potential subtype-specific vulnerabilities. By comparing tumor cells to their normal developmental counterparts in the RL, this work establishes a reference for understanding how deviations from normal cerebellar development shape MB pathogenesis.
Martin et al. (Tue,) studied this question.