Introduction

Per- and polyfluoroalkyl substances (PFAS) are synthesized compounds produced through electrochemical fluorination (ECF) or telomerization1,2,3. These substances enhance the surface activity, performance, efficiency, and durability of a wide range of products, including non-stick coatings4,5, cosmetics6, food packaging5,7,8, waterproof textiles9,10, stain-resistant carpets9,10, and Class B aqueous film-forming foams (AFFF)11,12. However, the same stability that makes PFAS effective in industrial and commercial applications also contributes to their persistence in the environment, resulting in widespread contamination and public health concerns13,14,15. Long-chain PFAS are particularly concerning due to their potential to bioaccumulate in the food chain16,17,18 and their ability to undergo long-range atmospheric and aquatic transport19,20,21,22,23, which leads to widespread environmental exposure.

Drinking water contaminated with PFAS is a major pathway of human exposure24,25,26,27,28. Perfluorinated compounds, such as perfluorooctanoic acid (PFOA) and perfluorooctane sulfonic acid (PFOS), have been frequently detected in drinking water resources worldwide13,29,30. These occurrences are especially prevalent in areas where PFAS-containing AFFF have been used, whether during routine training exercises, emergency fire responses, or accidental releases31,32,33,34,35,36,37,38. Elevated serum levels of PFOS and PFOA in impacted communities underscore the human health risks25,26,39, which include immune system disruption, hormonal imbalances, and possible cancer40,41.

Existing research13,42,43,44,45 demonstrates that conventional drinking-water treatment processes are generally ineffective at fully removing perfluorinated compounds, particularly short-chain compounds. Enhanced and advanced treatments, including activated carbon adsorption, ion exchange resins, and high-pressure membrane filtration, show greater promise42; however, the resulting spent media and waste streams containing concentrated PFAS require additional treatment42 through thermal46,47,48 or non-thermal49,50,51,52 means. Conventional disinfection (e.g., chlorination) and advanced oxidation processes are typically ineffective against perfluorinated compounds due to the strength of the carbon–fluorine (C–F) bond53,54. In some cases, these treatments may even transform polyfluoroalkyl substances into more persistent perfluorinated compounds containing only C‒F bonds55,56 such as PFOA and PFOS.

Polyfluoroalkyl substances, which contain at least one carbon–hydrogen (C–H) bond in addition to C–F bonds, represent a structurally diverse and increasingly studied29,33,57 subclass of PFAS. The ECF process produces a variety of these compounds, including fluorinated sulfonamide derivatives (F(CF2)n–S(O)2NH− and fluorinated amide derivatives (F(CF2)n–C(O)NH−)58,59,60.

Polyfluoroalkyl substances have been detected in both drinking water and human blood samples26,61,62,63,64. Toxicological studies have shown that exposure to certain polyfluoroalkyl substances can cause developmental abnormalities in zebrafish, including altered embryonic and larval photomotor responses, as well as disruptions across multiple morphological endpoints65. Epidemiological evidence also links the polyfluoroalkyl compound N-ethyl-perfluorooctane sulfonamido acetic acid, detected in human blood, to adverse outcomes such as impaired fetal growth66, developmental issues, and an increased risk of breast cancer67. Recent research has demonstrated that selected polyfluoroalkyl substances can transform into environmentally persistent perfluorinated compounds, such as PFOA and PFOS, during drinking water chlorination55,56. As a result, removing these compounds prior to disinfection is critical to preventing such transformations and limiting human exposure via drinking water. However, unlike their perfluorinated counterparts, the fate of polyfluoroalkyl substances in conventional drinking-water treatment processes remains largely unexamined.

Although coagulation/flocculation45,68 and disinfection55,56,69 are widely used conventional drinking-water treatment processes, their effectiveness in removing or transforming polyfluoroalkyl substances remains poorly understood. Likewise, the fate and transformation of PFAS mixtures, such as those present in AFFF33,36,70,71,72,73, during chlorination have not been thoroughly investigated. To address these knowledge gaps, our team investigated the fate of complex mixtures of polyfluoroalkyl substances during conventional coagulation, flocculation, and chlorine disinfection. We employed non-target analysis using Kendrick mass defect (KMD) analysis and developed quantitative structure-activity relationship (QSAR) models to gain mechanistic insights.

Results

Removal of polyfluoroalkyl substances during coagulation and flocculation

Fig. 1 and Supplementary Fig. S1 illustrate the removal efficiencies of polyfluoroalkyl substances in AFFF-contaminated surface water following coagulation and flocculation treatments. Note that floc formation can occur during the coagulation stage, and thus, PFAS removal via enmeshment to fine flocs may initiate at this point. In this study, the distinction between coagulation and flocculation is based on the conventional operational parameters (rapid mixing versus slow mixing) rather than the specific mechanisms (e.g., charge neutralization and enmeshment/sweep coagulation).

Fig. 1: Removal efficiency of polyfluoroalkyl substances during coagulation and flocculation in surface water.
figure 1

a Removal of cationic and zwitterionic polyfluoroalkyl substances. b Removal of anionic polyfluoroalkyl substances. The y-axis shows the UPLC–ESI-ToF MS peak intensity for each compound. These intensities are used as relative measures of concentration, as authentic standards for quantification were unavailable. Structural isomers are distinguished based on UPLC retention time and configuration. Compound names follow the nomenclature in ref. 58, except for HOEAmES-FASAs (identified in ref. 1) and N-DAmEAmES-FASAs (identified in refs. 1,2). The exact chemical structures of compounds with m/z 711.0721 and m/z 489.2376 remain unresolved.

Full size image

During coagulation, removal was relatively modest, with most compounds experiencing reductions ranging from 4% to 28%. For instance, C-6 N-HOEAmPFASAHOPS showed an 18% reduction, and C6-N-OxAmPFASA exhibited a 16% reduction (Fig. 1a and Supplementary Fig. S1). The quadrupole time-of-flight mass spectrometer (QToF-MS/MS) used in this study typically shows less than 5% variation when analyzing the same sample, indicating that the observed changes exceed the instrument’s measurement uncertainty. This low removal (e.g., <20%) can be attributed to the formation of positively charged hydrolyzed aluminum species, such as Al(OH)2+ and Al3+ 74,75, during alum coagulation, which likely repel cationic and zwitterionic polyfluoroalkyl substances via electrostatic repulsion.

Flocculation proved more effective than coagulation, as depicted in Fig. 1a, particularly for the most abundant C6 perfluorooctane sulfonamide (FOSA)-based polyfluoroalkyl substances, including C6-N-DAmEAmES-FASA, C6-N-HOEAmP-FASA, and C6-N-TAmP-FASA, where a notable decrease in UPLC–ESI-ToF MS peak intensity was observed post-flocculation. The enhanced removal during this slow-mixing phase likely results from the enmeshment of polyfluoroalkyl substances into sizable flocs, which mitigates some of the charge repulsion observed in coagulation. Considering the cumulative effect of both processes, the average removal of cationic and zwitterionic polyfluoroalkyl substances reached 34.7 ± 9.3% (Supplementary Fig. S1).

In contrast, anionic polyfluoroalkyl substances exhibited a different removal profile (Fig. 1b and Supplementary Fig. S1). During coagulation, several anionic polyfluoroalkyl substances experienced substantial reductions, such as C4-NEtFASAA that was reduced by 48%. This relatively high efficacy is likely due to favorable interactions between anionic polyfluoroalkyl substances with hydrolyzed aluminum species through charge neutralization, as well as their incorporation into flocs via sweep coagulation. Flocculation provided additional removal, though its effect was less pronounced than that in coagulation, contributing to a total removal of 44 ± 15% for anionic polyfluoroalkyl substances (Supplementary Fig. S1). Previous studies13,44 have demonstrated the inefficient removal of anionic perfluorinated compounds (e.g., PFOA) through coagulation and flocculation processes. The comparatively higher removal of anionic polyfluoroalkyl substances observed in this study is likely influenced by interactions involving non-fluorinated moieties, although the exact mechanisms remain to be elucidated.

The fate of co-occurring polyfluoroalkyl substances during conventional chlorination treatment

Having established the removal of polyfluoroalkyl substances via coagulation–flocculation, we next examined their fate under conventional chlorine disinfection. As shown in Fig. 3, chlorination led to measurable transformations in cationic/zwitterionic and anionic polyfluoroalkyl substances. Due to the lack of authentic standards for these polyfluoroalkyl substances, the transformation was assessed by tracking the decline in UPLC–QToF-MS peak areas over time, from which pseudo–first-order rate constants (kobs) were derived.

The majority of polyfluoroalkyl substances followed first-order kinetics, allowing for straightforward calculation of transformation rates. The kobs values, provided in Fig. 2, and Supplementary Figs. S2, S3, indicate the reactivity and persistence of diverse PFAS structures. For example, cationic and zwitterionic compounds such as C6-N-HOEAmES-FASA, C6-N-DAmEAmES-FASA, and C-6 N-HOEAmPFASAHOPS exhibited relatively high transformation rates, likely due to electron-rich nitrogen groups enhancing chlorination susceptibility. In contrast, polyfluoroalkyl substances containing electron-withdrawing groups, such as -SO3H or -COOH (e.g., C5-N-SHOPAmPFASAHOPS), displayed slower transformation, attributable to reduced electronic heterogeneity or steric hindrance that impedes oxidative attack. Dissolved organic matter and other chlorine-demanding constituents in pretreated surface water further reduced transformation efficiency compared to idealized conditions (i.e., organic-free water or buffered DI water).

Fig. 2: Transformation of cationic and zwitterionic polyfluoroalkyl substances during chlorination in pretreated surface water.
figure 2

Each point represents the mean ± standard deviation of three triplicate samples. A0 and At denote the peak areas of a polyfluoroalkyl substance in a sample, as determined by UPLC-QToF-MS, before and after the chlorination process, respectively. The values of kobs are provided. The corresponding transformation results, obtained in buffered DI water at pH 7.4, can be referenced in Supplementary Fig. S2. Polyfluoroalkyl substances: 1, C4-N-HOEAmP-FASA; 2, C4-N-HOEAmHOPFASA; 3, C4-N-HOEAmPFASE; 4, C6-N-DAmEAmES-FASA; 5, C6-N-TAmP-FASA; 6, C6-N-OxAmPFASA; 7, C3 N-HOEAmPFASAHOPS; 8, C3-N-HOEAmES-FASA; 9, C4-N-SHOPAmPFASA; 10, C6-N-HOEAmP-FASA; 11, C4 N-HOEAmPFASAHOPS; 12, C4-N-HOEAmES-FASA; 13, C6-N-SHOPAmPFASA; 14, C4-N-SHOPAmPFASAHOPS; 15, C-6 N-HOEAmPFASAHOPS; 16, C6-N-HOEAmES-FASA; 17, C5-N-SHOPAmPFASAHOPS; 18. C6-N-SHOPAmPFASAHOPS. Supplementary Fig. S3 illustrates the transformation of anionic polyfluoroalkyl substances during chlorination in pretreated surface water.

Full size image

In both matrices (pretreated surface water and buffered DI water), kobs values for cationic and zwitterionic polyfluoroalkyl substances were comparable to those of anionic species (Figs. 3 and 4). In pretreated surface water, these values were ~3–24.5% of those observed in organic-pure water (Fig. 3), reflecting competition from dissolved organic matter that diminishes PFAS transformation rates.

Fig. 3: First-order transformation rate constants (kobs) of polyfluoroalkyl substances obtained in buffered deionized (DI) water (pH 7.4) and pretreated surface water.
figure 3

a shows kobs values for cationic and zwitterionic polyfluoroalkyl substances, highlighting differences in degradation between water types. b presents kobs values for anionic polyfluoroalkyl substances under the same conditions.

Full size image

As further highlighted in Fig. 4, a positive linear correlation (R2 > 0.95) emerges between the kobs measured in DI water and those in pretreated surface water, demonstrating that the same underlying reaction pathways operate in both matrices, albeit at a reduced rate in the presence of naturally occurring interferences. The regression slopes were 1.75 for cationic and zwitterionic polyfluoroalkyl substances (R2 = 0.99) and 1.93 for anionic compounds (R2 = 0.97).

Fig. 4
figure 4

The linear regression plot of the lnkobs values of cationic, zwitterionic, and anionic polyfluoroalkyl substances in buffered DI water (pH 7.4) versus those in pretreated surface water.

Full size image

QSAR analysis of polyfluoroalkyl substance stability

As illustrated in Figs. 2, 3 and Supplementary Fig. S2, certain polyfluoroalkyl substances exhibit only moderate or weak reactivity toward chlorine, requiring extended chlorination times to achieve notable transformation. Such extended contact times reflect conditions that may occur in drinking water distribution systems, where residence times can span several days.

To investigate how molecular structure governs the chlorination reactivity of polyfluoroalkyl substances, we combined the transformation data from our recently studied compounds55,56. For each compound, 663 molecular descriptors were calculated, spanning constitutional, topological, electronic, geometric, and autocorrelation categories. A preliminary screening identified 26 descriptors with Pearson correlation coefficients r > 0.7 relative to the experimentally determined lnkobs values in buffered DI water (pH 7.4). After excluding highly intercorrelated variables, the pool was reduced to 15 candidate descriptors (Fig. 5a and Supplementary Table S3).

Fig. 5: Relationship between molecular descriptors and observed transformation rates of polyfluoroalkyl substances.
figure 5

a displays an intercorrelation matrix of molecular descriptors. The color of each circle indicates the direction of the correlation, with red circles representing positive correlations and blue circles indicating negative correlations. The size of each circle corresponds to the strength of the correlation, with larger circles denoting stronger relationships. b presents linear regression plots showing the relationship between the natural logarithm of the observed lnkobs and the molecular descriptors SssNH and AATS4p. c compares the predicted and measured values of lnkobs using a multilinear regression model.

Full size image

In QSAR efforts based on multiple linear regression, there is often a temptation to include numerous predictor variables simply to achieve a marginal increase in R2. However, this practice can waste degrees of freedom, result in a cumbersome model, and increase the risk of overfitting by incorporating many statistically insignificant variables. In the present study, the optimal model determined by multiple linear regression is Eq. 1, with a relatively high Rfitting2 and a small number of predictor variables (Fig. 5b, c).

$${\rm{l}}{\rm{n}}{k}_{{\rm{o}}{\rm{b}}{\rm{s}}}=2.1\times {\rm{S}}{\rm{s}}{\rm{s}}{\rm{N}}{\rm{H}}+32.5\times {\rm{A}}{\rm{A}}{\rm{T}}{\rm{S}}4{\rm{p}}+(-32.6)$$
(1)

[Rfitting2 = 0.79; CI95%: 2.1, (−0.7 to 4.9); 32.5, (−19.1 to 84.1); −32.6, (−76.4 to 11.2)]

The positive coefficients for SssNH and AATS4p (Eq. 1) indicate that these descriptors positively influence the reactivity of precursor compounds toward chlorine. SssNH describes the electrotopological state (E-state) of secondary amine groups (–NH–) in the molecule, as defined by Kier and Hall76,77. The E-state reflects both the intrinsic electronic state of the -NH- group, influenced by its valence state electronegativity, and its perturbation by the electronic and topological effects of neighboring atoms. A higher SssNH value suggests a greater susceptibility to electrophilic attack at nitrogen sites, consistent with their known reactivity under oxidative conditions56.

AATS4p represents the average Broto-Moreau autocorrelation with a lag 4, weighted by polarizabilities. This descriptor captures the degree to which polarizability is distributed across atoms separated by four bonds, effectively reflecting the spatial uniformity of electron-rich regions in the molecule. The Broto-Moreau autocorrelation provides a statistical approach for interpreting the spatial distribution of various atomic properties within a molecule, including attributes such as electron density and electronegativity, which are vital to comprehend the reactivity of a molecule. This autocorrelation method assesses the correlation between these properties in neighboring atoms throughout the molecular structure. A polyfluoroalkyl substance exhibiting a higher AATS4p implies a more uniform distribution of electron-rich atoms within the molecule than a precursor with a lower autocorrelation. An even distribution of electron-rich atoms (signified by high autocorrelation) indicates that multiple locations within a precursor are susceptible to electrophilic attack.

Consistent with this interpretation, compounds lacking –NH– moieties (e.g., 5:3 FTB, 5:1:2, N-TAmP-FHxSA) exhibit SssNH = 0 and lower observed lnkobs values. In contrast, polyfluoroalkyl substances that contain –NH– functionalities or exhibit a more uniform electron distribution degrade more readily under chlorination.

Nontarget analysis

Figures 6, and S4 provide a detailed representation of the KMD spectrum for AFFF samples in both positive and negative ionization modes before and after chlorine treatment across various time intervals. The KMD approach enables the recognition of homologous series of compounds differing by repeating units such as CF2, which is characteristic of PFAS structures. By normalizing masses to the CF2 unit, structurally related polyfluoroalkyl substances can be grouped and identified based on their regular spacing in KMD plots78,79.

Fig. 6: KMD analysis of polyfluoroalkyl substances before and after chlorination.
figure 6

KMD plots are shown for features with masses normalized to CF2 units, highlighting changes in molecular composition during chlorination of an AFFF sample (#3-398). a shows KMD profiles acquired in positive ionization mode. b shows KMD profiles acquired in negative ionization mode. Solid black circles labeled “CK” represent the Kendrick mass profiles of polyfluoroalkyl substances prior to chlorination and serve as baseline references. Open symbols represent the corresponding features after chlorination, with different time points ranging from 1 to 96 h. The shift and evolution of points over time indicate chemical transformations occurring during the chlorination process.

Full size image

Initial observations from the chlorination process reveal minor shifts in KMD values at shorter exposure times, ranging from 1 to 5 h, suggesting preliminary interactions between chlorine and the polyfluoroalkyl substances, potentially involving halogenation or minor bond cleavages. As chlorination progresses (10 and 20 h exposure), the KMD values exhibit increased dispersion, indicating more extensive chemical transformations. This phase likely involves additional bond breakage and/or the formation of new, lower-molecular-weight products, as evidenced by the broad distribution of points in the KMD spectrum.

Upon long-term exposure, specifically from 72 to 96 h, the KMD data stabilizes, and new points emerge on KMD plots (Figs. 6, S4), indicating the formation of stable transformation products following extended chlorination treatment. Notably, perfluorinated compounds such as perfluoroalkyl carboxylates and perfluoroalkyl sulfonates, which exhibit KMD values ranging from −0.015 to 0.157, appear in the KMD plots after 96 h of chlorination.

Lastly, KMD plots visually reveal differences in chlorination responses between cationic/zwitterionic and anionic polyfluoroalkyl substances. Cationic and zwitterionic polyfluoroalkyl substances, typically containing positively charged or neutral functional groups, exhibit transformation patterns that differ markedly from those of anionic polyfluoroalkyl substances. Upon chlorination, anionic polyfluoroalkyl substances display more pronounced shifts and a broader dispersion in KMD plots, indicating a more substantial transformation.

The KMD plots indicate the transformation of polyfluoroalkyl substances and the formation of new species during chlorination, consistent with recent studies involving model compounds55,56. Unlike chlorinated and brominated disinfection by-products, the exposure risks associated with fluorinated transformation products from polyfluoroalkyl substances following chlorination remain largely unknown. This gap in knowledge warrants further research, particularly in the context of AFFF-contaminated sites where such transformations are likely to occur.

Discussion

This study advances the understanding of the transformation of polyfluoroalkyl substances derived from AFFF in contaminated water, offering critical insights into their fate during conventional drinking-water treatment processes, namely coagulation, flocculation, and chlorination. By systematically examining these processes, we elucidate how structural features govern the removal and transformation behaviors of cationic, zwitterionic, and anionic polyfluoroalkyl substances.

Chlorination induces partial transformation by cleaving non-fluorinated carbon chains, with transformation kinetics revealing structure-dependent reactivity. Compounds with nitrogen-containing functional groups, such as perfluorooctane sulfonamides, showed enhanced susceptibility, while those with electron-withdrawing groups (e.g., -SO3H, -COOH) exhibited slower transformation due to steric hindrance or reduced electronic heterogeneity. QSAR modeling, supported by high-resolution mass spectrometry (HRMS), identified the electrotopological state of -NH- and Broto-Moreau autocorrelation as key predictors of reactivity, providing a predictive framework for assessing degradability.

Given the scarcity of such comprehensive studies, our results fill a critical gap in the literature and offer a foundation for optimizing water treatment strategies, aligning with global efforts to mitigate fluorochemical pollution. One possible strategy42 involves implementing granular activated carbon or ion-exchange treatment of AFFF-contaminated water before the disinfection process to minimize the transformation of polyfluoroalkyl substances into perfluorinated species.

Methods

Reagents and samples

This study included three AFFF samples (3% v/v) and two Fluorad brand surfactant concentrates (FC100 and FC129) (Supplementary Table S1a), in which >50 cationic, zwitterionic, and anionic polyfluoroalkyl substances in >10 classes have been identified1,80. Exact extracted m/z values, ion formulas, and structures for these polyfluoroalkyl substances in AFFF and surfactant concentrates are listed in Supplementary Tables S1b and S1c. The concentrations of PFOS varied from 800 to 9000 mg/L in surfactant concentrates and from 12,000 to 16,000 mg/L in AFFF. The AFFF stock solution of polyfluoroalkyl substances was prepared by adding 800 µL AFFF or surfactant concentrate solution into 200 mL of HPLC/MS-grade methanol (Thermo Fisher Scientific, Inc., Waltham, MA, USA).

All other solvents and chemicals were purchased from Sigma Aldrich or Fisher Scientific at the highest purity available.

Experiments

Natural surface water was collected from the Red River (Grand Forks, ND, USA), featuring a dissolved organic carbon of 6.1 mg/L and a pH of 7.9. The water was spiked with 300 µL of an AFFF stock solution, resulting in a final AFFF concentration of 0.00012% (v/v), and stabilized at 22 °C for 10 h prior to undergoing coagulation, flocculation, sedimentation, and disinfection treatments.

The coagulation and flocculation experiments were performed following our previously established methods45,55. Briefly, 1 L of PFAS-spiked surface water underwent jar tests to simulate coagulation and flocculation using a jar test apparatus (model JLT6, VELP Scientific) that was equipped with a flat paddle impeller and adjustable speed settings. After adding 40 mg/L alum as the coagulant, the solutions were vigorously stirred at 150 rpm for 1 min for rapid mixing or coagulation, then gently stirred at 40 rpm for 20 min for slow mixing or flocculation, and finally allowed to settle for 30 min. After coagulation and settling, samples were microfiltered to determine the residual levels of polyfluoroalkyl substances. Although the pH of the coagulated/flocculated water was not measured, it is estimated to be between 7.4 and 7.6, given that 1 mg/L of alum consumes 0.5–0.6 mg/L of alkalinity as CaCO381,82.

The settled water sample was filtered through Whatman Grade 5 filter paper (9 cm diameter, ~2.5 µm pore size) to remove any remaining suspended particles. The filtered water then underwent breakpoint chlorination using sodium hypochlorite (Acros Organics), following a procedure described in our previous study55 at a free chlorine dose of ~5.1 mg/L with contact times of 1, 2, 5, 10, 20, 72, 96, 216, and 456 h in buffered DI water, and 1, 3, 5, 7.5, 24, 30, 48, 49, 51, 53, 55, 72, 77, 97, and 216 h in pretreated surface water. The dose of free chlorine is within the typical range used for conventional drinking-water disinfection (e.g., 1–6 mg/L of free chlorine)83.

After chlorination, samples were quenched with sodium thiosulfate at a molar dose five times that of the dosed chlorine before being analyzed via HRMS. Residual free and total chlorine values were measured by the conventional DPD (N, N-diethyl-p-phenylenediamine) method84 using a colorimeter (Hach, Loveland, CO).

Additional chlorination experiments were conducted in spiked organic-pure water, prepared with DI buffered with NaHCO3 (1 × 10−3 mol/L) to a pH of 7.4, to eliminate the impact of background interference from the water matrix. These experiments in the buffered DI water were conducted on both AFFF and individual polyfluoroalkyl substances.

Two sets of controls were prepared. The first set consisted of unspiked surface and synthetic water samples that underwent the same treatment processes (coagulation, flocculation, separation, and disinfection) to ensure that no PFAS contamination originated from laboratory wares and devices. The second set included spiked samples that were subjected to the aforementioned treatment processes, but without the addition of coagulants and disinfectants. The concentrations of the samples in the second set of controls were used to evaluate the removal of polyfluoroalkyl substances in water treatment processes. Our previous studies found no to negligible hydrolysis loss of the selected polyfluoroalkyl substances over a period of up to ~250 h55,56. Given the short residence time of PFAS during coagulation, flocculation, and separation, degradation via hydrolysis is unlikely. Any minimal loss potentially due to hydrolysis is accounted for through the second set of spiked control samples.

Target analysis

The analysis of polyfluoroalkyl substances in coagulated, flocculated, and disinfected water samples was conducted with a HRMS system consisting of a Waters Acquity ultra-performance liquid chromatograph (UPLC) coupled to a Waters Synapt G2-S QToF-MS/MS (Waters Corporation, Milford, MA, USA). Chromatographic separation was achieved on a Waters Acquity UPLC BEH Shield RP18 column (100 × 2.1 mm; 130 Å; 1.7 μm) equipped with a matching VanGuard pre-column. The mobile phase was composed of eluent A (2 mM ammonium formate in LC/MS grade Optima™ water) and eluent B (2 mM ammonium formate in HPLC/MS grade Optima™ methanol). The gradient started at 20% B for 0.5 min, increased linearly to 85% B over 5 min, jumped to 98% B for 0.1 min, and maintained isocratically for 1.5 min. The system returned to the initial A/B ratio of 80/20 within 0.1 min and re-equilibrated for another 1.3 min. The flow rate was set at 0.45 mL/min, with the column temperature maintained at 55 °C, and samples were injected using a well-plate autosampler kept at 8 °C.

Target mass spectrometric analysis was performed on the m/z values of these polyfluoroalkyl substances1. The mass spectrometric analysis was conducted with the Synapt G2-S QToF-MS equipped with an ESI source, operating alternately in negative and positive ion modes. The settings included a cone voltage of 20 V, capillary voltage of 1.8 kV, source temperature at 110 °C, desolvation temperature at 350 °C, cone gas flow at 10 L/h, and desolvation gas flow at 1000 L/h. The analyzer operated at a resolution of 10,000 (fwhm at m/z 554) with an acquisition time of 0.1 s. Data were collected in ToF MSE mode, alternating the transfer T-wave element voltage between 10 and 25 V for low and high energy states. Leucine enkephalin (400 pg/μL) was continuously infused at 10 μL/min for mass accuracy correction. Data acquisition and analysis were conducted using MassLynx V4.1 software (Waters). The MSE function facilitated the simultaneous capture of MS and MS/MS fragmentation data in a single chromatographic run, providing detailed structural information of the transformation products1.

Because the authentic standards of these polyfluoroalkyl substances were not available, the transformation of these polyfluoroalkyl substances was evaluated by the UPLC peak area (A). The transformation profile of polyfluoroalkyl substances as a function of chlorination time was fit to a pseudo-first-order (kobs) kinetics model. This relies on the fact that chlorine concentration is much higher than that of polyfluoroalkyl substances. The natural logarithm of the concentration of a precursor (At) was plotted against the chlorination time (t), and the slope of the resulting linear trend line was taken as the negative of kobs.

The removal of polyfluoroalkyl substances during each of the treatment units (e.g., coagulation) was evaluated by Eq. 2:

$${\rm{Removal\; efficiency}}\left( \% \right)=\frac{{A}_{0}-{A}_{t}}{{A}_{0}}\times 100$$
(2)

where A0 and At, respectively, represent the UPLC peak areas of the second set of control samples and the chlorinated samples at a given retention time.

Non-target analysis

Non-target experiments were conducted based on QTOF-MS/MS plots, which involved data-filtering steps, including blank filtering, intensity filtering, and KMD filtering. The mass tolerance was set for less than 5 ppm. Briefly, features were extracted from the MS plots by Masslyss 12.0 software (Waters Corporation, Milford, MA, USA). Features with an intensity exceeding ten times that of the blank samples were selected. Subsequently, blank subtraction was then performed using six LC/MS grade water samples and six LC/MS grade methanol samples. This step was critical to remove background contamination/interference from the system. Then, the CF2-normalized KMD values for all features were computed using Eqs. 3 and 4.

$$\mathrm{Kendrick\; mass}=\frac{\mathrm{nominal\; mass}\,\mathrm{of}\,{\mathrm{CF}}_{2}}{\mathrm{exact\; mass}\,\mathrm{of}\,{\mathrm{CF}}_{2}}\times \mathrm{measured\; mass}$$
(3)
$${\rm{K}}{\rm{M}}{\rm{D}}={\rm{n}}{\rm{o}}{\rm{m}}{\rm{i}}{\rm{n}}{\rm{a}}{\rm{l}}\,{\rm{K}}{\rm{e}}{\rm{n}}{\rm{d}}{\rm{r}}{\rm{i}}{\rm{c}}{\rm{k}}\,{\rm{m}}{\rm{a}}{\rm{s}}{\rm{s}}-{\rm{K}}{\rm{e}}{\rm{n}}{\rm{d}}{\rm{r}}{\rm{i}}{\rm{c}}{\rm{k}}\,{\rm{m}}{\rm{a}}{\rm{s}}{\rm{s}}$$
(4)

Structural effects of polyfluoroalkyl substances

The use of a mixture of polyfluoroalkyl substances in AFFF offered a realistic perspective on the changes these compounds undergo during chlorination treatment. However, the lack of authentic standards made it challenging to quantify these compounds and interpret the effects of their structure. To overcome this challenge, we leveraged transformation data from nine model polyfluoroalkyl substances collected in our previous studies55,85 and established a QSAR using molecular descriptors. These nine model polyfluoroalkyl substances included 2-[(4,4,5,5,6,6,7,7,8,8,8-undecafluorooctyl)dimethylammonio]acetate (5:3 FTB), 2-[(3,4,4,5,5,6,6,7,7,8,8,8-dodecafluorooctyl)dimethylammonio]acetate (5:1:2 FTB), N-AP-FHxSA, N-[3-(perfluoro-1-hexanesulfonamido)propan-1-yl]-N,N,N-trimethylammonium (N-TAmP-FHxSA), and 6:2 fluorotelomer sulfonamide betaine (N-CMAmP-6:2 FOSA or commonly known as 6:2 FTAB86,87) (Supplementary Table S2). All those polyfluoroalkyl substances have been observed in AFFF2,58 and AFFF-contaminated soil and water samples26,57,88.

Molecular descriptors are advanced mathematical representations that capture the structured information embedded within a molecule. They are viewed as the end product of a logical and mathematical process that converts the information encoded in a molecule’s symbolic representation into a practical numeric value89. A total of 663 molecular descriptors of polyfluoroalkyl substances, including constitutional, topological, electronic, thermodynamic, geometric, and autocorrelation descriptors, were calculated using PaDEL-Descriptor freeware90. The correlation of the descriptors with the experimentally derived lnkobs was evaluated, and descriptors that exhibited significant association with lnkobs were selected to establish a QSAR.

Supplementary material

This supplementary material includes tables summarizing reagents, sample compositions, polyfluoroalkyl substances used for QSAR analysis, and key molecular descriptors influencing chlorination kinetics. The figures illustrate the transformation kinetics of different precursor classes (cationic, zwitterionic, anionic, and sulfonamide-based) under buffered conditions, as well as KMD plots capturing PFAS transformation products in both positive and negative ionization modes.