Integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA, IDH1, EGFR, and NF1

Abstract

The Cancer Genome Atlas Network recently cataloged recurrent genomic abnormalities in glioblastoma multiforme (GBM). We describe a robust gene expression-based molecular classification of GBM into Proneural, Neural, Classical, and Mesenchymal subtypes and integrate multidimensional genomic data to establish patterns of somatic mutations and DNA copy number. Aberrations and gene expression of EGFR, NF1, and PDGFRA/IDH1 each define the Classical, Mesenchymal, and Proneural subtypes, respectively. Gene signatures of normal brain cell types show a strong relationship between subtypes and different neural lineages. Additionally, response to aggressive therapy differs by subtype, with the greatest benefit in the Classical subtype and no benefit in the Proneural subtype. We provide a framework that unifies transcriptomic and genomic dimensions for GBM molecular stratification with important implications for future studies.

Figure 1

Identification of four GBM subtypes. (A) Consensus clustering matrix of 202 TCGA samples for k=2 to k=5. (B) Consensus clustering CDF for k=2 to k=10. (C) SigClust p-values for all pair wise comparisons of clusters. (D) Silhouette plot for identification of core samples. Also see Figure S1.

Figure 2

Gene expression data identify four gene expression subtypes. (A) Using the predictive 840 gene list, samples were ordered based on subtype predictions and genes were clustered using the core set of 173 TCGA GBM samples. (B) Gene order from the TCGA samples was maintained in the validation dataset (n=260), which is comprised of GBMs from four previously published datasets. (C) Ordered gene expression for 24 xenograft samples. Samples are ordered based on their predicted identity using the 840 gene list. Selected genes are displayed for each gene expression subtype. Also see FigureS3 and TableS3.

Figure 3

Integrated view of gene expression and genomic alterations across glioblastoma subtypes. Gene expression data (ge) was standardized (mean equal to zero, standard deviation equal to 1) across the 202 dataset, data are shown for the 116 samples with both mutation and copy number data. Mutations (mut) are indicated by a red cell, a white pipe indicates loss of heterozygosity, and a yellow cell indicates the presence of an EGFRvIII mutation. Copy number events (cn) are illustrated by bright green for homozygous deletions, green for hemizygous deletions, black for copy number neutral, red for low level amplification, and bright red for high level amplifications. A black cell indicates no detected alteration.

Figure 4

Single sample GSEA scores of GBM subtypes show a relation to specific cell types. Gene expression signatures of oligodendrocytes, astrocytes, neurons and cultured astroglial cells were generated from murine brain cell types (Cahoy et al., 2008). Single sample GSEA was used to project the four gene sets on samples on the Proneural, Classical, Neural and Mesenchymal subtypes. A positive enrichment score indicates a positive correlation between genes in the gene set and the tumor sample expression profile; a negative enrichment score indicates the reverse. Also see FigureS6.

Figure 5

Survival by treatment type and tumor subtype. Patients from TCGA and Murat (Murat et al., 2008) were classified by therapy regimen: red, more intensive therapy: concurrent chemotherapy and radiation or greater than four cycles of chemotherapy; black, less intensive therapy: non-concurrent chemotherapy and radiation or less than four cycles of chemotherapy. (A) Proneural, (B) Neural, (C) Classical, (D) Mesenchymal. Also see Figure S7 and Table S7.