Sorption of pharmaceutically active compounds to sandy soils: Effects of compound and soil properties

• Sorption of 23 pharmaceutically active compounds (PACs) onto sandy soils was examined. • The sorption rate coefficient declines as the equilibrium coefficient (K L ) increases. • Principal component analysis (PCA) was performed for each ionic state of PACs. • Based on PCA, sorption of PACs is more influenced by molecular than soil descriptors. • Existing regression models over-predict sorption of PACs to soils with low SOM and CEC. • Models developed for sorption of PACs to sandy soils explained > 90 % of the variations in K L . In this study, batch experiments were conducted to evaluate the sorption behavior of 23 pharmaceutically active compounds (PACs) onto four sandy soils that vary in their organic matter (SOM) and cation-exchange capacity (CEC). Sorption of the PACs generally follows pseudo-second-order kinetics. The Freundlich model best described the equilibrium behavior of the PACs for soils with high SOM or high CEC, while the linear equilibrium model was adequate for soils with low SOM and low CEC. Comparison between the linear/linearized equilibrium sorption coefficient (K L ) and the sorption rate coefficient shows an inverse relationship, implying slower chemical interaction or mass transfer to sorption sites with increased sorption tendency. Principal component analysis of ten independent variables related to PAC and soil properties was performed for each ionic state to study their influence on K L . The results showed that molecular descriptors played a more significant role than soil descriptors in predicting the sorption behavior of PACs, and their influence varies across ionic states. Existing regression models failed to predict K L of the PACs and generally resulted in over-predictions. New predictive models developed for different ionic states of the PACs used in this study could explain 92 % to 98 % of the variations in the K L values. When tested with sorption data for similar PACs and soils in other studies, the models’ ability to predict the K L values within ± 1-log unit decreased from 64 % for neutral PACs to 50 % for cationic PACs and 30 % for anionic PACs.