Male F344 rats aged 5-week old were treated by daily oral gavage with one of 85 hepatocarcinogenic, non-hepatocarcinogenic or non-carcinogenic compounds at two dose levels for 28 days. The livers were sampled at the following 6 time points: Days 1, 3, 7, 14, 21 and 28. For monitoring gene expression levels, an in-house oligomicroarray (NEDO-ToxArray III) covering approximately 6700 genes was prepared. Genes for prediction of carcinogenicity were selected by data cleaning and the Welch's t-test based on classification of the training set of chemical carcinogens. Then, two prediction models were established by supportive vector machine analysis, one based on the expression pattern and the other on analysis of chemical compound classification according to Ames’ mutagenicity profiles.
The data demonstrated very good predictivity; the prediction rates were 100 % for rat hepatocarcinogens (23 chemicals), 78 % for mouse hepatocarcinogens (7/9) and 96 % for other carcinogens (23/24). Negative predictivity was 96 % (24/26). Blind testing (17 chemicals), inter-laboratory validation (4) and external validation (3) all provided good results.
Thus, the toxicogenomics approach was proven to be very valuable tool for prediction of carcinogenic potentials of chemicals.
The data obtained were from collaboration by the Chemicals Evaluation and Research Institute, Sumitomo Chemical Co. Ltd., and Mitsubishi Chemical Safety Institute Ltd.