Examining environmental matrix effects on quantitative non-targeted analysis estimates of per- and polyfluoroalkyl substances.

Abstract

Non-targeted analysis (NTA) is commonly used for the detection and identification of emerging pollutants, including many per- and polyfluoroalkyl substances (PFAS). While NTA outputs are often non-quantitative, concentration estimation is now possible using quantitative non-targeted analysis (qNTA) approaches. To date, few studies have examined matrix effects on qNTA performance, and little is therefore known about the implications of matrix effects on qNTA results and interpretations. Using a set of 19 PFAS, we examined the impacts of drinking water (DW) and waste-activated sludge matrices on qNTA performance across three qNTA approaches: one structure-independent approach based on "global" surrogates and two structure-dependent approaches based on "expert-selected" surrogates and predicted ionization efficiency (IE) regression. The performance of each qNTA approach was examined separately for the PFAS prepared in pure solvent, DW extract, and sludge extract using leave-one-out modeling. Performance was evaluated using previously defined qNTA metrics that describe predictive accuracy, uncertainty, and reliability. The studied sample matrices had minimal effects on qNTA accuracy and larger effects on qNTA uncertainty and reliability. Using solvent-based surrogate data to inform matrix-based estimations yielded lower uncertainty, but also lower reliability, emphasizing that uncertainty must be considered in context of reliability. No single qNTA approach uniformly performed best across all comparisons. Since the IE regression and global surrogates approaches proved most reliable, we recommended them for future qNTA applications. This study highlights the importance of examining multiple performance metrics and utilizing matrix-matched surrogate data in qNTA studies.