Discovery of Comprehensive Sets of Chemical Constituents as Markers of PFAS Sources through a Nontarget Screening and Machine Learning Approach.

The objective of this study was to identify chemical constituents as markers of six per- and polyfluoroalkyl substance (PFAS) sources including aqueous film-forming foam-impacted groundwater, landfill leachate, biosolids leachate, municipal wastewater treatment plant effluent, and wastewater effluents from the pulp and paper and power generation industries. Previous chemical fingerprinting methods relying on target and suspect PFASs alone have been unable to differentiate PFAS sources containing complex target and suspect PFAS profiles. Here, we demonstrate that high-resolution mass spectral acquisitions from six distinct PFAS sources can be processed by means of an integrated nontarget analysis (NTA) and machine learning (ML) classification-based approach to improve source classification. NTA was conducted from negative- and positive-mode acquisitions, resulting in 21,815 chemical features from 92 samples and 114,660 chemical features from 88 samples, respectively. The inclusion of non-PFAS markers such as preservatives, pesticides, pharmaceuticals, manufacturing intermediates, spectroscopy materials, fatty acids, metabolites, and plant-derived chemicals substantially enhanced the classification performance compared to PFAS-only classifiers. This study significantly advances PFAS forensic capabilities by offering a practical framework for source differentiation and providing critical tools for environmental monitoring and remediation efforts.