Combination of Chromatographic and Machine Learning-Driven Virtual Fractionation Identifies Aryl Hydrocarbon Receptor...

Complex organic chemical mixtures in aquatic ecosystems may cause adverse effects on aquatic and sediment-dwelling organisms. Identified chemicals typically explain less than 10% of the observed in vitro bioactivities of such complex mixtures extracted from sediments. In a proof-of-concept study, we combined high-resolution fractionation with machine-learning-driven virtual fractionation to identify aryl hydrocarbon receptor (AhR) agonists in sediment from the Elbe River, Germany. Reporter gene assay showed that only apolar fractions activated the AhR but without specificity among them, necessitating additional virtual fractionation after analysis by gas chromatography coupled with high-resolution mass spectrometry (GC-HRMS). Activity and potency predictions by machine learning models for deconvoluted GC-HRMS features from mass spectral reference library matching allowed the identification of 145 AhR-active HRMS features, with 26 chemicals bioanalytically and chemically confirmed, most of which were polycyclic aromatic hydrocarbons (PAHs). With semiquantified concentrations and estimated potency for the tentatively identified chemicals, the mixture effects of the identified agonists accounted for 14% to 47% of AhR activation in sediments, doubling the contribution of known US EPA priority PAHs. This study provides an effective tool for early screening of AhR agonists, serving as a blueprint to identify causative chemicals for other environmentally relevant modes of toxic action.