Heterogeneity of colorectal cancers and extraction of discriminator gene signatures for personalized prediction of prognosis.

Dissected specimens of colorectal cancer (CRC) have been intensively studied using molecular sketches (gene signatures) to obtain a set of discriminator gene signatures for accurate prognosis prediction in individual patients. The discriminators obtained so far are not universally applicable, as the gene sets reflect the method and site of the study. In this study, we show that dissected stage II and III CRC samples are significantly heterogeneous in molecular sketches, and are not appropriate sources for discriminator extraction unless handled individually. To search for an accurate discriminator gene set for prediction of metastases, we need to start with less heterogeneous stage II CRC. We examined 198 (92 stage II and 106 stage III) CRC dissected samples for the predictability of discriminator gene signatures by analyzing stage II CRC alone, stage III alone, or in combination. The best predictive power of discriminator genes was obtained only when these genes were extracted and validated with stage II CRC samples. An accurate discriminator gene set for the prediction of CRC metastases can be obtained by focusing on stage II CRC samples.