Abstract
Development of alternative methods to rat carcinogenicity tests is highly demanded, but it remains very difficult. This study aimed to establish a method for predicting non-genotoxic carcinogenicity by read-across using molecular descriptors and in vitro assay data. From the results of 2-year rat carcinogenicity tests of agrochemicals published by the Food Safety Commission of Japan, we selected 80 carcinogens that cause benign or malignant tumors in liver, thyroid, testis, uterus, ovary, breast, nasal cavity, stomach, or bladder/urethra and 46 non-carcinogens that showed no carcinogenicity in these organs, and subjected these chemicals to various cytotoxicity tests in HepG2 cells and reporter assays for several nuclear receptors. Analyses using Fisher's exact test demonstrated significant associations between thyroid tumor and PXR activation, testicular tumor and AHR activation, stomach tumor and changes in GSH levels, and bladder/urethra tumor and AHR activation. Then, read-across prediction for these 4 types of tumors were performed using neighboring substances selected based on Euclidean distance between substances calculated using alvaDesc molecular descriptors, with or without stratification of the test substances by in vitro assay results. The results indicated that the stratification improved prediction accuracy (concordance rate: 0.571-0.921) compared to the accuracy based on descriptors alone (concordance rate: 0.421-0.556), suggesting that in vitro assay results are useful for read-across prediction of non-genotoxic carcinogenicity.
