There has been little progress in identifying prognostic gene expression signatures in colorectal cancer compared with other solid malignancies. We investigated the possibility of building classifiers for disease progression based on the gene expression profile from 101 crude tumor samples from stage II and III patients. Unsupervised hierarchical clustering separated the tumors into groups with distinct microsatellite instability status that were strongly related to outcome in Stage II patients. Stage II colon cancer patients classified as microsatellite unstable had significantly longer overall survival than microsatellite stable (P=0.0014). Using a maximum likelihood classification method we identified nine genes that could separate the 101 tumor samples into stable and unstable tumors, and which were in 97% concordance with microsatellite analysis. The sensitivity for diagnosis of microsatellite stable tumors exceeded 98% with a specificity exceeding 94%. The positive and negative predictive values exceeded 98% and 91%, respectively. The classifier worked equally well when using real-time PCR analysis as 95.7% of an independent set of 47 tumors were correctly classified. Thus, based on quantification of nine gene transcripts we identified a group of stage II patients with a very good prognosis and a group, which are MSS, that have 50% mortality. In addition, microsatellite unstable tumors could be classified into sporadic and hereditary tumors with an accuracy of 97.2% (36 of 37) using the two genes MLH1 and PIWIL1. The combination of such different classifiers based on gene expression profiles may contribute to the development of microarrays as diagnostic tools for colorectal cancer.

[Proc Amer Assoc Cancer Res, Volume 46, 2005]