BACKGROUND: Glioma sphere-forming cells (GSCs) derived from surgical specimens are a fundamental resource to study glioblastoma (GBM) biology. Mesenchymal-expressing GSCs have been proposed as a source of treatment resistance and mesenchymal tumors correlate with poorer survival. Recently, we found that the anti-angiogensis drug bevacizumab appeared to provide no benefit to patients with mesenchymal tumors, in contradiction to expectations that a mesenchymal microenvironment may benefit from anti-angiogenesis therapy. We have developed a collection of GSCs that have undergone comprehensive genomic characterization, similar to that performed by the Cancer Genome Atlas (TCGA) for whole tumor specimens. We hypothesized that the genomic landscape of GSCs would recapitulate what was observed by TCGA. METHODS: 47 GSCs were obtained from primary culture of fresh tumor specimens obtained at surgery and cultured as 3-dimensional spheres in the absence of serum. All lines were subjected to RNAseq (75bp paired-end, 100X coverage), copy number analysis (Affymetrix Oncoscan 2.0), whole methylome (Illumina Infinium 450k bead array), and targeted resequencing of known cancer-associated genes. Whole exome sequencing was performed for 22 GSCs. Gene expression was determined by reads per kilobase per million (RPKM) using an RNA sequencing data analysis pipeline (PRADA) and somatic mutations identified by a commonly used detection method (MuTech). Consensus clustering based on none-negative matrix factorization (CNMF) was performed on expression data and correlation to TCGA clusters determined by single-sample gene set enrichment analysis (ssGSEA). RESULTS: While global copy number alterations such as gain of chromosome 7 at the EGFR locus or loss of chromosome 10 at the PTEN locus were shared between tumor and matched GSC, the rate of somatic events was significantly higher in GSCs compared to tumors (range 47-570, median 124 vs range 2-255, median 65). Optimization of CNMF identified a total of 5 gene-expression clusters. GSCs in only one of these clusters showed enrichment for a unique TCGA class, mesenchymal. GSCs in other clusters were divided among multiple TCGA classes. CONCLUSIONS: Mesenchymal glioblastomas are derived from mesenchymal GSCs, suggesting that the tumor component is the largest contributor to the aggressive biology of this subtype. GSCs from other tumor subtypes correlate to multiple TCGA classes, suggesting that tumor stroma may contribute to the expression phenotype in those cases. Therapeutics targeting the microenvironment, such as anti-angiogenesis drugs, may have a greater role in non-mesenchymal tumors where the stromal contribution is more prominent. SECONDARY CATEGORY: Neuropathology & Tumor Biomarkers.