shortcoming, we have developed an approach that does not reflect the restricted bias of the chemicals that make up the Department of Environmental and Occupational Health, Graduate School of database from which the SAR model is derived, but allows Public Health, University of Pittsburgh, Pittsburgh, PA 15261, USA extrapolation to the 'universe of chemicals' (23). Essentially, 1 To whom correspondence should be addressed.rather than looking for structural overlaps among SAR models, Email: rsnkranz@pitt.edu we chose 10 000 chemical representatives of the 'universe of The mechanistic relationship of the inhibition of gap juncchemicals' (24). The various biological/toxicological properties tional intercellular communication (GJIC) to other toxicoof these chemicals are predicted using validated SAR models. logical phenomena was explored using a recently developedThe prevalence of chemicals predicted to possess two toxicolomethod that models the properties of a large population of gical properties simultaneously is then determined and commolecules chosen to represent the 'universe of chemicals'.pared with that expected. The rationale of the approach as The analyses indicate that inhibition of GJIC is strongly well as the interpretation of the results are described below. linked to the carcinogenic process in rodents, to cellular We applied this approach to the phenomenon of inhibition but not systemic toxicity, to biological phenomena that may of GJIC to determine whether it demonstrated the expected involve inflammatory processes and to development effects.interactions and/or whether in addition it generated new testable The inhibition of GJIC appears not to be associated with mechanistic hypotheses. genotoxic mechanisms. With respect to cancer causation, integration of the analyses suggests that inhibition of GJIC Materials and methods is involved in non-genotoxic cancer induction or in the SAR methodology non-genotoxic phases of the carcinogenic process (such asFor these studies we used the CASE/MULTICASE SAR expert systems inflammation, cell toxicity, cell proliferation, inhibition of described previously (25,26). Application of this methodology results in the cell differentiation and apoptosis).development of four submodels, each of which are derived from different algorithms and useful for investigating different aspects of the biological phenomena under consideration. The projections of the four individual submodels were integrated into a single prediction based upon Bayes' theorem

The CASE/MULTICASE structure-activity relationship (SAR) system was used to assess a new procedure to investigate the mechanistic relatedness of various toxicological endpoints. The method consisted of predicting the activity of 10,000 randomly selected chemicals using validated and characterized SAR models from a variety of biological and toxicological endpoints. The prevalence of chemicals predicted to possess the ability to induce two or more toxicological effects simultaneously should provide a measure of the mechanistic relatedness of these phenomena. Eight toxicological endpoints were predicted and the results were compared to predictions based on an eye irritation SAR model. Allergic contact dermatitis demonstrated a 29.6% greater than expected overlap between expected and observed results (p < 0.001). Similar results were seen for respiratory hypersensitivity (33.1%), sensory irritation (28.9%), cell toxicity (25.9%), and Ah receptor binding (19.8%). A lesser degree of overlap was seen between eye irritation and Salmonella mutagenicity (11.5%) and the inhibition of gap junction intercellular communication (6.7%). Moreover, a negative overlap, suggesting possibly an antagonistic phenomena, was observed between eye irritation and alpha 2 mu-induced nephropathy. These results indicate that this method can provide a useful tool to investigate mechanistic relatedness between diverse toxicological endpoints.

No abstract