Using gene expression profiling to identify a prognostic molecular spectrum in gliomas

Abstract

Histopathological classification of gliomas is often clinically inadequate due to the diversity of tumors that fall within the same class. The goal of the present study was to identify prognostic molecular features in diffusely infiltrating gliomas using gene expression profiling. We selected 3456 genes expressed in gliomas, including 3012 genes found in a gliomal expressed sequence tag collection. The expression levels of these genes in 152 gliomas (100 glioblastomas, 21 anaplastic astrocytomas, 19 diffuse astrocytomas, and 12 anaplastic oligodendrogliomas) were measured using adapter-tagged competitive polymerase chain reaction, a high-throughput reverse transcription-polymerase chain reaction technique. We applied unsupervised and supervised principal component analyses to elucidate the prognostic molecular features of the gliomas. The gene expression data matrix was significantly correlated with the histological grades, oligo-astro histology, and prognosis. Using 110 gliomas, we constructed a prediction model based on the expression profile of 58 genes, resulting in a scheme that reliably classified the glioblastomas into two distinct prognostic subgroups. The model was then tested with another 42 tissues. Multivariate Cox analysis of the glioblastoma patients using other clinical prognostic factors, including age and the extent of surgical resection, indicated that the gene expression profile was a strong and independent prognostic parameter. The gene expression profiling identified clinically informative prognostic molecular features in astrocytic and oligodendroglial tumors that were more reliable than the traditional histological classification scheme.