Molecular Classification of Gliomas to Guide Treatment Decisions.

Introduction: In glioma patients, the histological classification and grading of tumors, often combined with perceived clinical prognostic features, guides treatment decisions. However, histological classification of gliomas is troublesome and subject to a large degree of interobserver-variation. At least four independent studies (including one at the Erasmus MC) have indicated that approximately 20-30% of all gliomas are incorrectly classified (1-4). For a large proportion of these patients, discrepancies in classification result in different treatment strategies (4) whereas the treatment strategies have been optimized for the different glioma subtypes (see e.g. (5)). The classification of gliomas therefore urgently requires improvement. Preliminary results: We have recently demonstrated that classification of tumors based on similarities in gene expression (unsupervised molecular clustering) is a robust and objective alternative to histological classification of gliomas (6). Molecular clustering identified seven subgroups that are distinct from the histological subgroups and correlate better with survival. The power of molecular clustering was demonstrated by the ability to identify a subset (10-15%) of prognostically favorable tumors within an external dataset that contains only confirmed glioblastomas and to identify a subset (10-15%) of grade II gliomas that have a prognosis comparable to those of grade IV (figure 1). Molecular clustering is highly robust (data were validated on five independent datasets) and improves on histological classification. Conversely, histological classification does not improve our unsupervised molecular clustering (table 1). Although other molecular markers also aid classification of gliomas (7), the molecular clusters correlate better with survival than those markers (table 1). Molecular classification therefore may aid diagnosis and can be used to guide clinical decision making. This is a long path though, which starts with assessing prognostic and predictive molecular characteristics within prospective clinical trials Aim: In this project we will use material from two large randomized phase III clinical trials (European Organisation of Research and Treatment of Cancer (EORTC) trials 26951 (completed) and 22033 (ongoing)) to validate our prognostic clusters in gliomas. The data generated will also allow identification of predictive markers. Ultimately, our aim is to use molecular clustering to identify tumors that should not be treated as low grade glioma despite favorable histological characteristics. Conversely, molecular clustering can be used to identify gliomas that should be treated conservatively, despite unfavorable histological characteristics. Plan of Investigation: Clinical samples are formalin-fixed and embedded in paraffin (FFPE), resulting in partially degraded RNA. We and others have already demonstrated that FFPE samples retain the differences in gene expression that are required for molecular clustering (figure 2). Nevertheless, a cohort of paired snap frozen-archival samples will be used to determine informative probesets on FFPE samples (aim 1). We will then validate prognostic markers on samples of EORTC 26951 as this study has been completed and clinical data and molecular data are directly available (aim 2). If required, molecular profiles can be optimized for FFPE samples based on data obtained from aim 1 and aim 2. Q-PCR on a subset (30-40) of classifier genes can be used as a validated backup technique (figure 2). A final validation of the prognostic value of molecular clustering will be performed as part of EORTC 22033, a large, phase III clinical trial (aim 3). Independent of the validation of prognostic clustering, the data generated using EORTC trials 26951 and 22033 will also be used to identify predictive profiles (aim 4). Expression profiling will be performed using HuEx_1.0_st arrays in combination with NuGen WT-Ovation technology and in collaboration with Dr. S. Pepper, an expert in the field of FFPE profiling. Tissue microarrays will be used to perform immunocytochemistry on prognostic/predictive markers. Samples will be assigned to one of the seven molecular subtypes based on its nearest centroid. Clinical benefit: By improving glioma classification, this project will result in more patients receiving the optimal treatment regimen. A better classification will result in a reduced burden of treatment in prognostically favorable tumors (despite histologically poor characteristics) and in more aggressive treatment regimens in prognostically poor tumors (despite histologically favorable characteristics). Ultimately our results will lead to the development of a diagnostic tool that will help clinical decision making.