Linking mutagenic activity to micropollutant concentrations in wastewater samples by partial least square regression and subsequent identification of variables

Abstract

We deployed multivariate regression to identify compounds co-varying with the mutagenic activity of complex environmental samples. Wastewater treatment plant (WWTP) effluents with a large share of industrial input of different sampling dates were evaluated for mutagenic activity by the Ames Fluctuation Test and chemically characterized by a screening for suspected pro-mutagens and non-targeted software-based peak detection in full scan data. Areas of automatically detected peaks were used as predictor matrix for partial least squares projections to latent structures (PLS) in combination with measured mutagenic activity. Detected peaks were successively reduced by the exclusion of all peaks with lowest variable importance until the best model (high R2 and Q2) was reached. Peaks in the best model co-varying with the observed mutagenicity showed increased chlorine, bromine, sulfur, and nitrogen abundance compared to original peak set indicating a preferential selection of anthropogenic compounds. The PLS regression revealed four tentatively identified compounds, newly identified 4-(dimethylamino)-pyridine, and three known micropollutants present in domestic wastewater as co-varying with the mutagenic activity. Co-variance between compounds stemming from industrial wastewater and mutagenic activity supported the application of “virtual” EDA as a statistical tool to separate toxicologically relevant from less relevant compounds.