QSAR models of the in vitro steroid hormone receptor activity of alkylphenol and bisphenol A analogs

Bisphenol A is a monomer constituent of epoxy and polycarbonate resins used in consumer products. Nonylphenol is an alkylphenol used in surfactants and polymer stabilizers. These compounds are weak estrogen receptor agonists with endocrine disrupting potential in exposed organisms. Presented here are a series of quantitative structure-activity relationship (QSAR) models describing the in vitro hormone activity (estrogen receptor binding, reporter gene induction, and cell proliferation) of 25 bisphenol A analogs and 20 nonylphenol analogs. Statistically significant QSAR models were generated using SYBYLRTM 6.9. For bisphenols, optimal models were: comparative molecular field analysis (CoMFARTM), mean correlation coefficient r 2 = 0.988 and mean cross-validated correlation coefficient q2 = 0.753; comparative molecular similarity indices analysis (CoMSIA), mean r2 = 0.987 and q2 = 0.774; and hologram quantitative structure activity relationship (HQSAR(TM)), mean r2 = 0.957 and q2 = 0.598. For alkylphenols, optimal models were: CoMFARTM, mean r2 = 0.994 and q2 = 0.595; CoMSIA, mean r 2 = 0.962 and q2 = 0.682; and HQSAR(TM), mean r2 = 0.900 and q 2 = 0.745. Bisphenols with strong estrogenic activity contained two unencumbered phenolic groups in the para orientation, and multiple alkyl substituents extending from the ring-linking carbon. Bisphenols with methyl group hydrogens replaced by halogens were also strongly estrogenic. Alkylphenols with strong estrogenic activity contained a single tertiary branched alkyl group composed of between 6 to 8 carbons in the para orientation on an otherwise unhindered phenolic ring. Optimal QSAR models were used to predict the estrogen receptor's relative binding affinity of other bisphenols and alkylphenols identified in the Toxic Substances Control Act inventory. Binding activities of compounds with molecular substructures comparable to those in their respective QSAR training sets were accurately predicted (residuals < 1 log unit relative binding affinity). These results suggest that QSAR modeling may be used to predict the relative binding affinity of chemicals with sufficient accuracy to identify probable endocrine disrupting compounds.