Use of in silico models for prioritization of heat-induced food contaminants in mutagenicity and carcinogenicity testing

Numerous Maillard reaction and lipid oxidation products are present in processed foods such as heated cereals, roasted meat, refined oils, coffee, and juices. Due to the lack of experimental toxicological data, risk assessment is hardly possible for most of these compounds. In the present study, an in silico approach was employed for the prediction of the toxicological endpoints mutagenicity and carcinogenicity on the basis of the structure of the respective compound, to examine (quantitative) structure–activity relationships for more than 800 compounds. Five software tools for mutagenicity prediction (T.E.S.T., SARpy, CAESAR, Benigni-Bossa, and LAZAR) and three carcinogenicity prediction tools (CAESAR, Benigni-Bossa, and LAZAR) were combined to yield so-called mutagenic or carcinogenic scores for every single substance. Alcohols, ketones, acids, lactones, and esters were predicted to be mutagenic and carcinogenic with low probability, whereas the software tools tended to predict a considerable mutagenic and carcinogenic potential for thiazoles. To verify the in silico predictions for the endpoint mutagenicity experimentally, twelve selected compounds were examined for their mutagenic potential using two different validated in vitro test systems, the bacterial reverse mutation assay (Ames test) and the in vitro micronucleus assay. There was a good correlation between the results of the Ames test and the in silico predictions. However, in the case of the micronucleus assay, at least three substances, 2-amino-6-methylpyridine, 6-heptenoic acid, and 2-methylphenol, were clearly positive although they were predicted to be non-mutagenic. Thus, software tools for mutagenicity prediction are suitable for prioritization among large numbers of substances, but these predictions still need experimental verification.