Assessment on data processing methods for nontarget identification of per- and polyfluoroalkyl substances using liquid chromatography-high-resolution mass spectrometry

The widespread use， persistence， bioaccumulation， and toxicity of per- and polyfluoroalkyl substances （PFAS） have raised global concern. The number of PFAS types continues to grow， driven by changing industrial demands and regulatory environments. Non-target analysis using high-resolution mass spectrometry （HRMS） is an effective methodology for identifying novel and unknown PFAS in environmental matrices. The efficacy of non-target analysis is critically influenced by the data acquisition mode， peak picking algorithm， and deconvolution strategy. Using ultra-high performance liquid chromatography coupled with an Orbitrap mass spectrometer （UHPLC-Orbitrap MS）， this study aims to systematically evaluate data processing methods for non-targeted PFAS identification under data-dependent acquisition （DDA） and data-independent acquisition （DIA） modes. A clean sludge sample was spiked with 34 PFAS standards at three levels to assess method performance， alongside the analysis of three electroplating sludge samples. To compare the identification performance between DDA and DIA modes， a multi-step evaluation process was employed. Firstly， we assessed the peak picking capabilities of two widely used data processing software packages， MS-DIAL and MZmine. The key parameters for peak picking process are MS1 mass tolerance of 0.002 5 Da， MS2 mass tolerance of 0.01 Da， minimum peak height of 1 000， and retention time alignment tolerance of 0.1 min. Secondly， a comparison was made regarding DIA data deconvolution， specifically between MS2Dec algorithm and IonDecon algorithm. Finally， FluoroMatch was utilized to compare the true positive rate （TPR） and positive predictive value （PPV） of PFAS identification in both DDA and DIA datasets. In the spiked samples， the ［M-H］⁻ precursor ions for 33 PFAS standards and the ［M-CO2-H］⁻ ion for HFPO-DA were successfully detected and manually verified across all three levels. For peak picking， MS-DIAL demonstrated superior performance， achieving a 100% detection rate in all spiked samples， outperforming MZmine. When comparing deconvolution performance for DIA data， MS2Dec algorithm and the IonDecon algorithm showed similar efficacy， although MS2Dec algorithm exhibited slightly better results for low-concentration samples. In DDA mode， the true positive rate for PFAS identification increased from 80% to 100% with rising analyte concentration， accompanied by a minimal decrease in positive predictive value. Conversely， in DIA mode， the true positive rate remained at 100% across all concentrations， but positive predictive value decreased as concentration increased， primarily due to interferences from in-source fragmentation and adduct ions. The degree of in-source fragmentation of perfluorocarboxylic acids （PFCAs） decreases with increasing carbon chain length. However， the proportion of adduct ions remains nearly constant across different PFAS， leading to false positive identification of hydrogen-substituted PFAS. Based on the evaluation results， the data processing methods for DDA and DIA modes were optimized. These methods were then applied to three electroplating sludge samples， leading to the identification of 36 PFAS species belonging to 10 classes， including eight perfluorocarboxylic acids （PFCAs）， eight perfluorosulfonic acids （PFSAs）， one hydrogen-substituted perfluorosulfonic acid （H-PFSA）， five unsaturated perfluorosulfonic acids （UPFSAs）， one carbonyl perfluorosulfonic acid （KPFSA）， one chlorine-substituted perfluorosulfonic acid （Cl-PFSA）， one n∶2 fluorotelomer sulfonic acid （n∶2 FTSA）， five chlorinated polyfluoroethersulfonic acids （Cl-PFESAs）， two hydrogen-substituted polyfluoroethersulfonic acids （H-PFESAs）， and four polyfluoroethersulfonic acids （PFESAs）. Their presence was largely attributed to the use of chrome mist suppressants in the electroplating process. Combining DDA and DIA data for FluoroMatch input captured more information on unknown PFAS， possibly because the inclusion of multiple samples improves peak extraction. Based on the performance of PFAS identification in spiked and real samples， we developed a processing method that couples DDA and DIA data. This method can generate a composite list of identified PFAS while keeping data files independent， increasing the true positive rate and efficiency of identification. This study systematically evaluated nontargeted PFAS data processing methods， clarifying the optimal combination of tools for key steps （acquisition mode， peak picking， and deconvolution）， and validating its application potential in complex environmental matrices.