LTBK-08. Inferring cell type and cell state composition in glioblastoma from bulk DNA methylation profiles using multi-omic single-cell analyses
Year: 2022
Type: conference-abstract
Abstract: OBJECTIVE Cellular heterogeneity determines tumor phenotype and response to therapy. This is particularly pronounced in glioblastoma (GBM), which is characterized by multiple malignant cell states with distinct proliferation potentials, and different cell types of the microenvironment. Ideally, cellular heterogeneity is characterized using single cell genomic profiling techniques. However, these techniques remain challenging to apply in a diagnostic setting and to large retrospective patient cohorts (TCGA, GLASS, DKFZ and clinical trials). Instead, clinicians routinely support their diagnosis with bulk DNA methylation profiling, which generates robust results from low quality material but does not inform on cellular heterogeneity. We have developed a powerful new computational method to deconvolute bulk DNA methylation data and infer cellular heterogeneity within individual tumors, to support prognostic accuracy and personalized treatment decisions. METHODS Using both bulk and single-cell multi-omic datasets, we created a DNA methylation-based reference of cell types (malignant, glial, neuronal, and immune) within GBM tumors, and the state of malignant cells therein (stem-like vs. differentiated-like). Using this reference, our computational approach accurately deconvolutes bulk DNA methylation profiles of individual query samples. RESULTS High deconvolution accuracy of GBM heterogeneity was achieved from frozen and FFPE tissue samples, including those of low quality or purity (Jensen Shannon divergence for composition similarity < 0.05). Our approach eliminates bias derived from the microenvironment, and results in patient stratification that harmonizes the DNA methylation- and RNA-based classifications of GBM. It also reveals the inter- and intra-tumoral links between the genetic, DNA methylation, and transcriptomic components of GBM pathology, and suggests their specific impacts on treatment efficacy. To facilitate clinical translation, we created a public website that allows clinicians to infer the relative abundance of different cell states within a tumor at the click of a button.
Source: Neuro-Oncology
Authors: Dana Silverbush, Mario Suva, Volker Hovestadt
Language: English
Subfield: Genetics
Field: Medicine
Domain: Health Sciences
Open Access status: bronze
OpenAlex ID: W4309407264