INTRODUCTION: We used an integrated molecular analysis strategy to perform class discovery on a population of low-grade gliomas (astrocytomas, oligodendrogliomas, and mixed gliomas) to improve our understanding of the molecular relationships among these tumors and to reconcile genotypic relationships with current histologic and molecular strategies for tumor classification. METHODS: Gene expression profiling was performed on a cross-section of World Health Organization (WHO) grades I-II gliomas. Unsupervised class discovery algorithms identified and validated tumor clusters with genotypic similarity, and these data were integrated with chromosomal copy number assays and RT-PCR data to define molecular tumor subclasses. Machine learning models allowed accurate, prospective classification of unknown tumors into these molecular subgroups. This molecular classification model was compared to current histologic (WHO) and molecular pathologic (chromosome 1p and 19q deletions, p53 alterations, and Ki-67 expression) methods for glioma classification. RESULTS: Molecular class discovery suggested a three-class model for low-grade gliomas. One discrete cluster of gliomas identified the pilocytic astrocytomas, a second grouped the 1p/19q codeleted oligodendrogliomas, and the mixture of remaining 1p/19q intact gliomas, including astrocytomas, oligodendrogliomas, and oligoastrocytomas, formed a third cluster with a discrete pattern of expression. CONCLUSIONS: Integration of genomic, transcriptomic, and morphologic data for class discovery suggests a three-class model for low-grade gliomas. Class I represents tumors with molecular similarity to pilocytic astrocytomas, class II tumors are similar to 1p/19q codeleted oligodendrogliomas, and class III represents infiltrative low-grade gliomas. This classification is similar to current clinical paradigms for low-grade gliomas; our work suggests a molecular basis for such models. This classification may supplement or may serve as the basis for a molecular pathologic alternative to current grading schemes for low-grade gliomas and may highlight potential targets for future biologically based treatments or strategies for future clinical trials.
INTRODUCTION: We used an integrated molecular analysis strategy to perform class discovery on a population of low-grade gliomas (astrocytomas, oligodendrogliomas, and mixed gliomas) to improve our understanding of the molecular relationships among these tumors and to reconcile genotypic relationships with current histologic and molecular strategies for tumor classification. METHODS: Gene expression profiling was performed on a cross-section of World Health Organization (WHO) grades I-II gliomas. Unsupervised class discovery algorithms identified and validated tumor clusters with genotypic similarity, and these data were integrated with chromosomal copy number assays and RT-PCR data to define molecular tumor subclasses. Machine learning models allowed accurate, prospective classification of unknown tumors into these molecular subgroups. This molecular classification model was compared to current histologic (WHO) and molecular pathologic (chromosome 1p and 19q deletions, p53 alterations, and Ki-67 expression) methods for glioma classification. RESULTS: Molecular class discovery suggested a three-class model for low-grade gliomas. One discrete cluster of gliomas identified the pilocytic astrocytomas, a second grouped the 1p/19q codeleted oligodendrogliomas, and the mixture of remaining 1p/19q intact gliomas, including astrocytomas, oligodendrogliomas, and oligoastrocytomas, formed a third cluster with a discrete pattern of expression. CONCLUSIONS: Integration of genomic, transcriptomic, and morphologic data for class discovery suggests a three-class model for low-grade gliomas. Class I represents tumors with molecular similarity to pilocytic astrocytomas, class II tumors are similar to 1p/19q codeleted oligodendrogliomas, and class III represents infiltrative low-grade gliomas. This classification is similar to current clinical paradigms for low-grade gliomas; our work suggests a molecular basis for such models. This classification may supplement or may serve as the basis for a molecular pathologic alternative to current grading schemes for low-grade gliomas and may highlight potential targets for future biologically based treatments or strategies for future clinical trials.