Outcome-based profiling of astrocytic tumours identifies prognostic gene expression signatures which link molecular and morphology-based pathology

Abstract

Astrocytomas are intracranial malignancies for which invasive growth and high motility of tumour cells preclude total resection; the tumours usually recur in a more aggressive and, eventually, lethal form. Clinical outcome is highly variable and the accuracy of morphology-based prognostic statements is limited. In order to identify novel molecular markers for prognosis we obtained expression profiles of: i) tumours associated with particularly long recurrence-free intervals, ii) tumours which led to rapid patient death, and iii) tumour-free control brain. Unsupervised data analysis completely separated the three sample entities indicating a strong impact of the selection criteria on general gene expression. Consequently, significant numbers of specifically expressed genes could be identified for each entity. An extended set of tumours was then investigated by RT-PCR targeting 12 selected genes. Data from these experiments were summarised into a sample-specific index which assigns tumours to high- and low-risk groups as successfully as does morphology-based grading. Moreover, this index directly correlates with definite survival suggesting that integrated gene expression data allow individualised prognostic statements. We also analysed localisation of selected marker transcripts by in situ hybridization. Our finding of cell-specificity for some of these outcome-determining genes relates global expression data to the presence of morphological correlates of tumour behaviour and, thus, provides a link between morphology-based and molecular pathology. Our identification of expression signatures that are associated individually with clinical outcome confirms the prognostic relevance of gene expression data and, thus, represents a step towards eventually implementing molecular diagnosis into clinical practice in neuro-oncology.