Prediction of clinical outcome and biological characterization of neuroblastoma by expression profiling

Although numerous prognostic factors have been identified, precise risk evaluation in individual neuroblastoma patients remains difficult. To define a reliable predictor for event-free survival after first-line therapy and to identify gene signatures characteristic for clinical and biological subgroups, we performed expression profiling of 68 primary neuroblastomas of all stages by Affymetrix U95A arrays. Expression data from subgroups were analyzed using machine learning methods. Recurrence of neuroblastoma within two years of diagnosis was correctly predicted in 83% of cases using support vector machines (SVM). Significance analyses of microarrays (SAM) was applied to search for genes differentially expressed between subgroups. Surprisingly, expression profiles of stage 4 and stage 4s neuroblastomas could not clearly be distinguished by any of the mathematical methods applied. In contrast, MYCN-amplification as well as expression of TrkA demonstrated a strong association with specific gene expression patterns. We conclude that SVM is a suitable tool for prediction of clinical outcome in neuroblastoma. Gene signatures associated with biological factors including TrkA and MYCN will prove useful to understand neuroblastoma etiology and progression.

Supported by NGFN/BMBF.