Development of Bioaccessibility-Based Soil Environmental Criteria for PFAS through the Establishment of In Vitro Methods and Machine Learning Models

Per- and polyfluoroalkyl substances (PFAS) are ubiquitously distributed in the environment, with soil increasingly serving as one of the major matrices. Soil environmental criteria (SEC) are primarily based on total concentrations of pollutants. However, actual biological effects are driven by bioavailable concentrations. There is no systematic and standardized in vitro method for measuring PFAS bioaccessibility in soil. In this study, we developed extraction methods for PFAS bioaccessibility in soil toward different soil organisms (i.e., plants and earthworms) using in vivo tests as the benchmark. It was shown that water extraction was suitable for predicting the accumulation of 15 PFAS in plant shoots and short-chain PFAS (C ≤ 7) in plant roots, while C18 membrane extraction was feasible for long-chain PFAS (C > 7) in plant roots and 15 PFAS in earthworms. Based on 3,474 experimental data points from 44 soil samples across 18 provinces in China, a machine learning (ML) model for predicting PFAS bioaccessibility was developed. We further derived bioaccessibility-based ecotoxicity data based on the ML model to construct species sensitivity distribution curves, resulting in HC5 values of 0.229, 0.330, and 0.313 mg/kg for PFOA, PFHxS, and PFOS, respectively. These values reduced the uncertainty associated with soil variability and more accurately reflected the actual ecological risk of PFAS in soil. Overall, this study established a novel framework for deriving PFAS SEC and facilitated the transition from total concentration-based to bioaccessibility-based ecological risk assessment.