1Department of Neurosurgery, The Rausing Laboratory, Lund University, 22100, Lund, Sweden.2Department of Theoretical Physics, Lund University, Lund, Sweden.3Department of Oncology, Lund University, Lund, Sweden.4Department of Pathology, Lund University, Lund, Sweden.5Department of Clinical Genetics, Lund University, Lund, Sweden.6Division of Tumor Immunology, Lund University, Lund, Sweden.7Institute for Cancer Genetics, College of Physicians and Surgeons, Columbia University, New York, NY, USA.8Departments of Pathology and Medicine, Herbert Irving Comprehensive Cancer Center, College of Physicians and Surgeons, Columbia University, New York, NY, USA.
Checking for direct PDF access through Ovid
Gliomas are among the most aggressive malignant tumors and the most refractory to therapy, in part due to the propensity for malignant cells to disseminate diffusely throughout the brain. Here, we have used 27 K cDNA microarrays to investigate global gene expression changes between normal brain and high-grade glioma (glioblastoma multiforme) to try and better understand gliomagenesis and to identify new therapeutic targets. We have also included smaller groups of grade II and grade III tumors of mixed astrocytic and oligodendroglial origin as comparison. We found that the expression of hundreds of genes was significantly correlated to each group, and employed a naÏve Bayesian classifier with leave-one-out cross-validation to accurately classify the samples. We developed a novel algorithm to analyze the gene expression data from the perspective of chromosomal position, and identified distinct regions of the genome that displayed coordinated expression patterns that correlated significantly to tumor grade. The regions identified corresponded to previously known genetic copy number changes in glioma (e.g. 10q23, 10q25, 7q, 7p) as well as regions not previously associated significantly with glioma (e.g. 1p13, 6p22). Furthermore, to enrich for more suitable targets for therapy, we took a bioinformatics approach and annotated our signatures with two published datasets that identified membrane/secreted genes from cytosolic genes. The resulting focused list of 31 genes included interesting novel potential targets as well as several proteins already being investigated for immunotherapy (e.g. CD44 and tenascin-C). Software for the chromosome analysis was developed and is freely available at http://base.thep.lu.se.