Genomic estimates of aneuploid content in glioblastoma multiforme and improved classification

Accurate classification of glioblastoma multiforme (GBM) is crucial for understanding its biologic diversity and informing diagnosis and treatment. The Cancer Genome Atlas (TCGA) project identified four GBM classes using gene expression data and separately identified three classes using methylation data. We sought to integrate multiple data types in GBM classification, understand biologic features of the newly defined subtypes, and reconcile with prior studies.

We used allele-specific copy number data to estimate the aneuploid content of each tumor and incorporated this measure of intratumor heterogeneity in class discovery. We estimated the potential cell of origin of individual subtypes and the euploid and aneuploid fractions using reference datasets of known neuronal cell types.

There exists an unexpected correlation between aneuploid content and the observed among-tumor diversity of expression patterns. Joint use of DNA and mRNA data in ab initio class discovery revealed a distinct group that resembles the Proneural subtype described in a separate study and the glioma-CpG island methylator phenotype (G-CIMP+) class based on methylation data. Three additional subtypes, Classical, Proliferative, and Mesenchymal, were also identified and revised the assignment for many samples. The revision showed stronger differences in patient outcome and clearer cell type-specific signatures. Mesenchymal GBMs had higher euploid content, potentially contributed by microglia/macrophage infiltration.

We clarified the confusion about the "Proneural" subtype that was defined differently in different prior studies. The ability to infer within-tumor heterogeneity improved class discovery, leading to new subtypes that are closer to the fundamental biology of GBM.