Molecular modeling for screening environmental chemicals for estrogenicity: use of the toxicant-target approach

There is a paucity of relevant experimental information available for the evaluation of the potential health and environmental effects of many man made chemicals. Knowledge of the potential pathways for activity provides a rational basis for the extrapolations inherent in the preliminary evaluation of risk and the establishment of priorities for obtaining missing data for environmental chemicals. The differential step in many mechanisms of toxicity may be generalized as the interaction between a small molecule (a potential toxicant) and one or more macromolecular targets. An approach based on computation of the interaction between a potential molecular toxicant and a library of macromolecular targets of toxicity has been proposed for preliminary chemical screening. In the current study, the interaction between a series of environmentally relevant chemicals and models of the rat estrogen receptors (ER) was computed and the results compared to an experimental data set of their relative binding affinities. The experimental data set consists of 281 chemicals, selected from the U.S. EPA's Toxic Substances Control Act (TSCA) inventory, that were initially screened using the rat uterine cytosolic ER-competitive binding assay. Secondary analysis, using Lineweaver-Burk plots and slope replots, was applied to confirm that only 15 of these test chemicals were true competitive inhibitors of ER binding with experimental inhibition constants (K(i)) less than 100 microM. Two different rapid computational docking methods have been applied. Each provides a score that is a surrogate for the strength of the interaction between each ligand-receptor pair. Using the score that indicates the strongest interaction for each pair, without consideration of the geometry of binding between the toxicant and the target, all of the active molecules were discovered in the first 16% of the chemicals. When a filter is applied on the basis of the geometry of a simplified pharmacophore for binding to the ER, the results are improved, and all of the active molecules were discovered in the first 8% of the chemicals. In order to obtain no false negatives in the model that includes the pharmacophore filter, only 8 molecules are false positives. These results indicate that molecular docking algorithms that were designed to find the chemicals that act most strongly at a receptor (and therefore are potential pharmaceuticals) can efficiently separate weakly active chemicals from a library of primarily inactive chemicals. The advantage of using a pharmacophore filter suggests that the development of filters of this type for other receptors will prove valuable.