Metabolites are the end products of cellular regulatory processes, and their levels can be regarded as the ultimate response of biological systems to genetic or environmental changes. We have used a metabolite profiling approach to test the hypothesis that quantitative signatures of primary metabolites can be used to characterize molecular changes in human colon cancer in comparison with normal colon mucosa.

27 colon carcinomas and 18 samples from normal colon mucosa were analyzed by gas chromatography-time of flight mass spectrometry with automated mass spectral deconvolution. A total of 206 metabolites were detected. Carcinoma samples were compared to normal tissue samples using Welch’s t-test to assess the deregulation of each measured metabolite. At Bonferroni threshold (p<0.00024) we detected n=27 deregulated metabolites. With a less stringent threshold (p<0.01) we generated a list of n=61 metabolites. By a permutation procedure the rates of false discoveries among the latter list was estimated as FDR = 2.8%. Supervised clustering with respect to the top 61 deregulated metabolites led to a nearly perfect split between carcinomas and normal tissues, with only a single misclassification observed. Overall, with respect to the detected metabolites, the group of normal tissues exhibited a stronger homogeneity than the group of carcinomas. 11 amino acids were contained in our list of the top 27 metabolites, and 16 in our list of the top 61 metabolites.

Predictive models (NMC, LDA, SVM) were fitted by a combination of feature selection and machine learning. Full Leave-one-out cross-validation gave 0-3 misclassifications out of 45 samples, e.g. prediction accuracies between 93% and 100%.

Our study shows that large scale metabolic profiling using GC-TOF mass spectrometry is suitable for analysis of human tumor tissue and that there is a consistent and significant change in primary metabolism of malignant colorectal tumors compared to normal tissue which can be detected using multivariate statistical approaches. Therefore metabolomics is a promising high-throughput, automated approach in addition to functional genomics and proteomics for analyses of molecular changes in malignant tumors.

98th AACR Annual Meeting-- Apr 14-18, 2007; Los Angeles, CA