Purpose: We summarize and discuss our recent work exploring the use of gene expression analysis for the prediction of histopathological features of prostate cancer and patient outcome following radical prostatectomy.
Materials and methods: Using "gene chips" capable of analyzing expression of greater than 10,000 genes simultaneously, expression data were collected from prostate tumors and adjacent normal tissue taken from patients undergoing radical prostatectomy between 1995 and 1997 at our hospital. Differentially expressed genes were analyzed using a signal-to-noise metric and by Pearson correlation, while predictive models were built and tested using nearest-neighbor prediction and leave-one-out cross validation. Patient outcome data were analyzed using Kaplan-Meier plots.
Results: Robust expression differences were readily detected between tumor and normal tissue, and computer models were built that could predict the identity of tumor or normal differences with 92% accuracy. We were able to find gene expression correlates of Gleason score and, more importantly, a preliminary 5-gene model was found that could separate cases into recurrent and nonrecurrent groups based on the expression patterns found in the primary tumors.
Conclusions: These data suggest that it may be possible to predict prostate tumor behavior based on expression signatures present at the time of surgery. A critical next step will be to try and validate such findings in larger independent datasets.