Effect-directed analysis combined with machine learning and QSAR prioritization identifies major estrogen receptor agonists in industrialized coastal sediments.

Estrogen receptor (ER)-mediated activity in sediments remains insufficiently characterized due to unidentified ER-active compounds that may not be captured by targeted analysis and complex agonistic/antagonistic interactions in environmental mixtures. This study applied an integrated framework combining effect-directed analysis, nontarget screening, machine learning, and quantitative structure-activity relationship (QSAR) modeling to identify major ER agonists in sediments from Lake Sihwa, a highly industrialized coastal system in South Korea. ER-mediated activity, measured using the T47D-Kbluc bioassay, was detected at all sites and was mainly associated with the polar fraction and its reversed-phase high-performance liquid chromatography subfraction, particularly in industrial inland creeks. Sediment-associated ER activity frequently exceeded effect-based trigger values (3.6 × 10-1 ngE2 g-1OC for freshwater sediments and 1.8 × 10-1 ngE2 g-1OC for marine sediments), indicating widespread estrogenic risk. However, four target ER agonists (estrone, 17β-estradiol, arenobufagin, and loratadine) accounted for only 10-53% of the observed activity. A stepwise prioritization workflow integrating activity-guided filtering, descriptor-based machine learning, and VEGA QSAR prediction reduced more than 15,000 detected features to a small number of candidate compounds. Diethylstilbestrol (DES) and heptyl paraben were newly identified as contributors to sediment-associated ER-mediated activity. DES exhibited high estrogenicity (relative potency = 1.5 with E2) and accounted for up to 28% of the observed activity, indicating that it is a major contributor to sediment-associated estrogenicity. These findings suggest that persistent synthetic estrogens may represent previously overlooked drivers of estrogenic effects in contaminated sediments.