In
colorectal cancer, to predict the response to chemo- and/or radio-
therapy or the existence of
lymph node metastasis preoperatively, a more competent diagnostic system is required, in addition to conventional diagnosis based on morphology and pathology. The application of gene expression profiling to preoperative
cancer diagnosis using endoscopic biopsies could enable the selection of a more appropriate
therapy for patients. In this study, we evaluated the feasibility of gene expression profiling using preoperative biopsies of
colorectal tumors in a clinical setting, by investigating the influence of intra-
tumor heterogeneity on the profiles and testing the prediction ability of
tumor malignancy. Under endoscopic examination, two biopsies were sampled from each of 10
colorectal cancers and 10
adenomas, and their gene expression data were obtained using
cDNA microarrays. The intra- and inter-
tumor heterogeneities of the profiles were compared with unsupervised clustering analysis. Molecular prediction of
tumor malignancy using biopsies was performed with the supervised classification algorithm. In clustering analysis, almost all paired biopsies from the same
tumors joined each other. Pearson's correlation coefficients of the profiles between biopsies from the same
tumors (mean, 0.83) were significantly greater than those of the profiles between biopsies from other
cancers (mean, 0.58) (p<0.0001). In the supervised classification method,
malignancy was correctly predicted in 39 out of 40 biopsies with 8-71 informative genes. Gene expression profiling using endoscopic biopsies of
colorectal tumors revealed that the intra-
tumor heterogeneity was smaller than the inter-
tumor heterogeneity and
tumor malignancy was correctly predicted. Our findings suggest that the technique of gene expression profiling accurately represents the biological properties of
colorectal cancer and could help the preoperative diagnosis of this disease.