Abstract 2494: Elucidating developmental methylation programs and cell type compositions in glioblastoma using combinatorial optimization

Abstract Glioblastoma (GBM) is one of the most aggressive forms of brain cancer. Previous research has shown that GBM tumors may comprise one or more developed brain cell types, including astrocytes, oligodendrocytes, oligodendrocyte precursor cells, and neural stem cells. However, it is not yet known whether GBM tumors originate from mature or progenitor states of these brain cell types. To explore the interplay between the developmental program of the brain cells and GBM tumor progression, we leverage methylation, an epigenetic marker that has been established as a reliable signal for clinical subtyping of brain cancers and understanding tumor evolution and brain development. While methylation profiles for developed brain cell types are available, little is known about the methylation profiles of their progenitors. To simultaneously determine cell type composition in bulk GBM samples and infer the methylation profiles of progenitor brain cells, we developed a novel computational framework, Qombucha (Quadratic prOgraMming Based tUmor deConvolution with cell HierArchy). Qombucha uses a tree hierarchy depicting the developmental progression of mature brain cells, with leaves representing established mature brain cell types and internal nodes representing unobserved progenitors. Based on the premise that the value of a specific progenitor methylation feature equals those of its mature descendants, provided the descendants are concordant, Qombucha infers the values of the remaining unknown progenitor methylation features by using bulk methylation profiles of tumor samples from GBM patients. Qombucha implements a combinatorial optimization formulation that simultaneously infers the cell type composition of each tumor sample. The objective of Qombucha is to minimize the differences between the inferred and observed methylation profiles across the entire cohort of tumor samples. The inferred cell type compositions for tumors can then be used to investigate differences between GBM subtypes. We applied Qombucha to real GBM patient data we compiled, representing three most common methylation-based subtypes: receptor tyrosine kinase (RTK) I, RTK II, and mesenchymal typical (MES TYP). Qombucha successfully recapitulated the known finding that GBM MES TYP patients exhibit higher immune infiltration, as indicated by the increased Qombucha-imputed microglia fractions. We also identified GBM subtype-specific brain development processes. Interestingly, the Qombucha-imputed cell compositions demonstrate comparable or better performance in clustering GBM subtypes than state-of-the-art brain tumor classifying methylation sites used for clinical decision-making. Overall, Qombucha provides a novel framework for understanding the crosstalk between brain development and tumor progression and offers valuable insights into the heterogeneity among GBM subtypes. Citation Format: Xuan Cindy Li, Ananth Hari, Yuelin Liu, Omkar Singh, Zhichao Wu, Stephen M. Mount, Kenneth D. Aldape, Eytan Ruppin, S. Cenk Sahinalp, Alejandro A. Schäffer. Elucidating developmental methylation programs and cell type compositions in glioblastoma using combinatorial optimization [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2494.