Practical application guide for the discovery of novel PFAS in environmental samples using high resolution mass spectrometry

Abstract

Background:

The intersection of the topics of high-resolution mass spectrometry (HRMS) and per- and polyfluoroalkyl substances (PFAS) bring together two disparate and complex subjects. Recently non-targeted analysis (NTA) for the discovery of novel PFAS in environmental and biological media has been shown to be valuable in multiple applications. Classical targeted analysis for PFAS using LC-MS/MS, though growing in compound coverage, is still unable to inform a holistic understanding of the PFAS burden in most samples. NTA fills at least a portion of this data gap.

Objectives:

Entrance into the study of novel PFAS discovery requires identification techniques such as HRMS (e.g., QTOF and Orbitrap) instrumentation. This requires practical knowledge of best approaches depending on the purpose of the analyses. The utility of HRMS applications for PFAS discovery is unquestioned and will likely play a significant role in many future environmental and human exposure studies.

Methods/results:

PFAS have some characteristics that make them standout from most other chemicals present in samples. Through a series of tell-tale PFAS characteristics (e.g., characteristic mass defect range, homologous series and characteristic fragmentation patterns), and case studies different approaches and remaining challenges are demonstrated.

Impact statement:

The identification of novel PFAS via non-targeted analysis using high resolution mass spectrometry is an important and difficult endeavor. This synopsis document will hopefully make current and future efforts on this topic easier to perform for novice and experienced alike. The typical time devoted to NTA PFAS investigations (weeks to months or more) may benefit from these practical steps employed.

Background

The past several decades have seen a surge in interest in per- and polyfluoroalkyl substances (PFAS) as cornerstone industrial and consumer chemicals and concerning environmental pollutants leading to human and ecological exposure. The chemical class is highly varied, with expansive definitions of PFAS [1] including large fractions of partially fluorinated pesticides and pharmaceuticals in addition to classic fluorinated polymers and fluorinated surfactants. As various industrial producers, end-users, and regulatory bodies grapple with managing PFAS, there is a need for analytical chemical measurements to support and inform decision making at all levels. Robust and reliable chemical measurements are needed for characterization of PFAS content and extent in consumer/industrial products, food and environmental media, and for source identification and/or biomonitoring of PFAS. Most analytical methods are developed as targeted MS/MS methods designed to quantify selected sub-groups of PFAS based on the classic methods development paradigm, where targeted chemicals and matrices are identified, and methods optimized for those specific measurements.

High certainty in targeted LC-MS/MS measurements for PFAS is facilitated by the availability of native and isotope-labeled standards (as the gold standard for quantitation) for both initial method optimization and ongoing method performance evaluation. Targeted methods development for PFAS is a rather narrow scope of common chemical measurements and is often in response to a need; this is true for both regulatory methods (e.g., EPA method 537.1, 533) and methods developed by contract labs and academia. For example, the common EPA 537 PFAS method for water analysis includes a select range of carboxylic and sulfonic perfluoroalkyl acids (PFAAs) within a narrow range of mass and physicochemical properties typically analyzed by LC-ESI-MS/MS. Analysis is conducted using unit resolution MS/MS instruments operated in negative electrospray ionization mode (ESI−), as this currently gives the highest ionization efficiency and MS transmission rates, and hence highest the sensitivity for these types of anionic PFAS analytes. A deficiency with a common approach such as this is that the analytical coverage is fundamentally static and does not adapt easily to new analytes of interest including large groups of nonionic, volatile, cationic or high-molecular weight PFAS. This introduces several limitations, including: the 1-dimensional (1D) LC gradient has limitations in the number of substances that can be fully separated (which limits the ability to reduce co-elution interfering on the quantification and identification of the substances; the electrospray source operated in negative mode is effective mostly for anionic PFAS; and unit resolution MS/MS quantification requires optimization for mass transitions and introduces a risk of misidentification of PFAS forming few fragments. While analytical methods grow with time, they are not designed to keep up with the pace of emergence of new PFAS analytes. The ionization, transmission and detection capabilities of MS instruments also limit the total number of ions that can be simultaneously detected without interference and without loss of sensitivity. This inherently leads analytical chemists to prioritize a limited number of substances to be included into the methods. Some have termed the classical suite of perfluorinated anionic PFAS (i.e. EPA 537) “legacy PFAS”. These “legacy” methods include the most common perfluorinated alkyl acids, such as PFOA and PFOS, which have been used or formed as degradation products from commercially used PFAS precursors. Modernized methods also include a wider range of PFAS (e.g., EPA 1633), however, they significantly lag discovery and usage of emerging PFAS [2]. As a result of the limited coverage of PFAS diversity by targeted methods, other approaches, such as total organic fluorine and total oxidizable precursor (TOP) assays, have been used to report incomplete fluorine mass balances compared to the targeted methods. While these methods may help to quantify the fraction of un-identified PFAS, these methods do not provide structural information. NTA has become widespread to address the identification gap, leading to rapid expansion of PFAS knowledge with 700+ structurally defined compounds listed in the latest review [2] and the more than 6 million PFAS compiled in various databases [3, 4] This paper provides an overview of the tools available for PFAS nontargeted HRMS analysis and offers guidance for their practical application by end-users. Identification of novel PFAS is the first step for subsequent targeted methods to be developed later.

Alternative PFAS chemistry

As the EPA’s PFOA stewardship program was put in place, eight major US manufacturers of PFAS entered into a voluntary agreement to phase out usage and production of PFAS with eight or more fluorinated carbons [5]. In Canada, an agreement was negotiated between Environment Climate Change Canada, Health Canada, and four fluorochemical companies to eliminate “long-chain” perfluorocarboxylates, including PFOA [6]. Similar production, use or commerce phase outs were implemented in the European Union [7]. China, with a large PFAS manufacturing sector, begun to restrict new installations of PFOS and PFOA production facilities in 2011 [8], and, as of 2019, to phase out PFOS except for “acceptable purposes” specified in the Stockholm Convention [9]. The driving assumption behind these phase-outs was that shorter chain PFAS should be developed for industrial applications and products, with the hope that this chemical change would lead to shorter environmental half-lives, and a consequential reduction in bioaccumulation potential and concomitant bioavailability and risks.

As it currently stands, organofluorine chemistry is highly proprietary, with manufacturers developing specific products and manufacturing processes that make use of narrow suites of PFAS precursors and intermediates. While PFOS/PFOA were once somewhat ubiquitous, the identity of the replacement chemistries and associated precursors, byproducts, and waste generated are highly varied and protected by claims of confidential business interest. A number of manufacturers have begun to produce replacement chemistries such as per- and polyfluoroether PFAS as alternatives (e.g., ammonium 4,8-dioxa-3H-perfluorononanoate (ADONA) for ammonium perfluorooctanoate (APFO)) in fluoropolymer manufacture and metal plating [10]. Narrowly defined, manufactured replacements could be classified as “not C8”-based PFAS due to perfluoro insertion of mono- or polyether functionality, the reduction in perfluorinated chain length (from C8 to C4 or C6) in fluorotelomer-based products (e.g., 6:2 fluorotelomer alcohol or sulfonic acids), and/or the addition of non-fluorine heteroatoms, such as chlorine [2]. Additional replacements have included the development of cyclical dioxygen-bridged polyfluorinated carboxylic acids for use as polymer processing aids (e.g., C6O4—CAS 1190931–41-9) [11]. It is very likely with the vast applications of PFAS in industrial processes and products that there are currently overlooked PFAS. As evidenced by the discovery of a suite of associated PFAS near a fluorinated compound production facility in North Carolina, USA [12, 13] NTA investigations of specific manufacturing facilities can yield significant numbers of new chemicals. It is almost certain that this is the case at other facilities globally, with thousands of known PFAS applications and chemistries, any attempt to produce a complete PFAS inventory is likely incomplete. Beyond the analytical complexities imposed by the proliferation of these alternative PFAS, little is known regarding their environmental stability, degradation products, fate and transport or the bioaccumulation and toxicity potential. Due to the intrinsic persistence of perfluorinated carbon and the unmanageable uncertainties regarding PFAS exposure and risks with alternative PFAS production, academia has called on the international community to limit the production and use of PFAS altogether [14, 15].

Any attempt to manage the expansive PFAS chemical space will require expanded analytical tools to detect and characterize the widest fraction of PFAS produced, intentionally or otherwise, by replacement PFAS processes. Combined analytical techniques may be useful for investigation purposes in many cases. For an initial survey, total fluorine methods may help select areas or samples for further investigation or serve as documentation of compliance with PFAS levels in the supply chain of products. For purposes of enforcement, further identification of the substances may be needed, and eventual (targeted) quantification of the pollutant levels to support risk management processes. In this article we focus on tools which can facilitate identification of novel and emerging PFAS in environmental media.

Methods

HRMS and NTA

High-resolution mass spectrometry (HRMS) has emerged as a valid and critical analytical approach for the discovery of unknown chemicals, including PFAS. The expansion of this technique has been driven by availability of benchtop instruments such as quadrupole time-of-flight (QTOF) and Orbitrap mass spectrometers. Other capable HRMS platforms (e.g., Fourier-transform ion cyclotron resonance) possess mitigating factors that prevent widespread use, such as additional cost and upkeep and dedicated facilities and staff. One should not assume that having HRMS equipment negates the need for robust low-resolution MS/MS equipment for targeted well-defined analytical methods, but the technique offers distinct advantages. In conducting HRMS analysis the characteristics of instrumentation mass accuracy and resolving power are both important and necessary considerations we will further discuss [16]. Resolving power (or resolution) is the ability of an instrument to distinguish two peaks of slightly different mass-to-charge $\left(m/z\right)$ ratios. The resolving power can be easily defined for a measured mass by its full peak width at half maximum (FWHM). The narrower the peak width $\left(\Delta m\right)$ for a measurement of any given mass, the higher the resolving power $\left(R\right)$ (i.e., $R=m/\Delta m$). Typical “unit-mass” resolution or “low-resolution” mass spectrometers (LRMS), such as quadrupoles and ion-traps will have a FWHM near 0.6 Da, whereas HRMS equipment will have a value of 0.02 Da or less. Consequently, mass resolution for LRMS may be 3000–5000 whereas HRMS instruments may be ≥25,000. HRMS instruments with ultrahigh resolution in the range above 250,000 can identify fine isotope structures of specific elements in isotope clusters, aiding in chemical formula elucidation [17]. The second key analytical performance metric for HRMS is the mass accuracy, which is defined as the difference between measured accurate mass and theoretical exact mass of an ion. Mass accuracy can be expressed in absolute units of Da or relative to the measured mass in parts-per-million (ppm,$\Delta m/m\times {10}^6$ and is typically below 5 ppm. Modern HRMS instruments reliably measure mass accuracy often in the <1 ppm (or ~4 mDa) range, which can be combined with computational approaches to predict molecular formula with a high degree of accuracy. Resolving power becomes increasingly important at higher $m/z$ values, where there are proportionally more combinations of elemental compositions and structures, and hence the presence of similar mass substances rises [17]. Highly accurate mass measurements further bound the combinations of elements that can yield a particular mass, reducing the number of potential formulae to consider when characterizing a molecule. However, increases in resolution alone do not permit unique identification of chemical formulae.

Rather than an individual ion, most instrument software identifies chemical features, a candidate molecular ion with its associated retention time, abundance, and identified isotope ions. Ideally, for small user-specified mass error, only a few candidate chemical compositions would be possible. However, some $m/z$ values can fall within small mass error range of large numbers of candidate formulae. These unknown molecular features can be assigned formulae by matching exact masses and isotope abundance ratios using mathematical combinatorics. Proprietary vendor software conducts this formula assignment using algorithms to assign likely formula with confidence ratings, although the origins of the score are somewhat opaque to the user. A major step in the assignment of organic molecular formula is estimating the molecular carbon number $\left(n\right)$ :

$$n=\frac{\left\{\frac{M=1}{M}\right\}}{{\left\{\frac{13\text{C}}{12\text{C}}\right\}}_{\text{SRC}}}$$ (1)

where $\left(M\right)$ is the dominant mass-chromatographic peak area of the molecular feature, $\left(M+1\right)$ is the peak area of the $M+1$ isotopologue, assumed to reflect the presence of a single ¹³C in the molecule, and (¹³C/¹²C)_SRC is carbon isotopic ratio of the carbon source reservoir (SRC). The value of (¹³C/¹²C)_SRC can be reasonably estimated as that of the present-day atmosphere, 0.01114 [18]. The estimation of the molecular carbon number can be enhanced by HRMS, particularly in the analysis of large molecules. Because of accurate masses, characteristic A + 1 can be made more specific into A + 1.003355, where 1.003355 is the difference in the exact masses of ¹³C and ¹²C. Similar counting can be conducted for elements with appreciable isotope abundance (e.g., sulfur, chlorine) based on the isotope mass shift and known abundances. A noteworthy advantage of the isotopic analysis of HRMS data over that of LRMS is in the separation of the A + 2 isotopologues of molecules that contain sulfur, chlorine and/or bromine atoms. A shortlist of molecular features usually can be culled from longer lists of candidate formulae based on predicted elemental compositions, (e.g., the carbon number ±2). Then candidate-formula lists can be further limited by other atom-count ranges to be consistent with the estimated carbon-number range, e.g., H-number range = (0, (2C + 1)), or anticipated substructural elements (e.g., CO₂, SO₃). Incorporating this isotope information gives significantly more accurate lists of potential formulae than exact mass predictions alone (e.g., those derived from qualitative vendor software or Chemcalc) [19]. These calculations rely on the assumption that isotope abundance measurements have the same precision. This is generally true for TOF instrumentation for compounds of sufficient abundance, whereas for ion-trap instruments, including the Orbitrap, spectral precision can be reduced for increasing masses or at large ion fluxes.

If mass accuracy and resolution are the foundation of HRMS measurements, scan range, scan speed and molecular fragmentation are the remaining differentiating factors that can impact NTA experiments. For small-molecule science and the discovery of PFAS, an HRMS scan range of 75–3000  $m/z$ is adequate and achievable with minimal tradeoffs across a range of instruments. In performing true discovery NTA, the tradeoffs are between scan speed and mass accuracy. Both QTOF-based and Orbitrap-based HRMS equipment can generate MS/MS data with the commensurate high resolution and high mass accuracy for intact molecules. However, QTOF-based instruments generally have faster scan speeds compared to Orbitraps but lower maximum resolution. Orbitrap instruments are capable of trading resolution for scan speed within an experiment, sacrificing their significantly higher resolution for near-QTOF scan speeds. Scan speed is most relevant for NTA due to the ability to collect molecular structural information from fragmentation spectra. HRMS instrument software permits the collection of full-scan HRMS data interspersed with MS/MS data with user-defined settings in the same analytical run. Higher scan speeds permit the insertion of more MS/MS scans between MS scans, meaning fewer “missed” compounds with no fragmentation data and/or fewer gaps in the full-scan chromatograms. Additionally, the use of vendor specific data acquisition tools (e.g., iterative exclusion, data-dependent acquisition, all ions..) may be important to gain full coverage in particular for MS/MS specifically for low abundance peaks.

Another significant benefit of HRMS data is that, if properly acquired and stored, it allows for retrospective screening for unidentified compounds, including PFAS, if the experiment was amenable to their detection. This capability is a marked deviation from LRMS targeted experiments, which collects only MS/MS transitions from predefined chemicals. HRMS also enables trend-analysis post hoc for chemicals identified, without defining a panel before sample collection or data. Adding quantification marker compounds to samples, may allow for semi-quantification at a later point [20, 21] to estimate risk or assess sources of and when pollution has occurred. With suspect screening analysis (SSA) (comparing acquired data to a list of known or suspected contaminants) or NTA (true de novo structure elucidation) of HRMS data, comparison data is crucial. Some examples could be monitoring from a set location over a long period of time at discrete intervals, monitoring of water above and below an effluent source, before/after a treatment technology, or comparing soils’ extracts proximate and distant from a suspected source.

Limitations of HRMS

While HRMS measurements have significant utility for chemical analysis, they are not a panacea. HRMS compound elucidation and sample analysis has limitations due to the limits of accurate mass measurement, the availability of accurate mass fragments from the analysis, suitable reference databases, and lack of authentic standards for comparison. The advantage of not pre-supposing the chemical contents of a sample can easily overwhelm an analyst, yielding hundreds if not thousands of features with only limited ability to unequivocally determine chemical identities. Therefore, communicating the confidence of a compound detection is of exceptional importance [22, 23]. Schymanski et al. [22] defined the most widely used schema for reporting the confidence of NTA compound assignments, while Charbonnet et al. [23] updated with specificity for PFAS. The base of confidence is an assignment only of an exact mass of interest (Level 5), and the pinnacle of confidence, true confirmation, is the comparison of MS, MS/MS, and chromatographic information with a reference standard (Level 1). Intermediate confidence in between these two extremes includes an unequivocal molecular formula (Level 4), tentative candidate(s) structures with MS and MS/MS data (Level 3), and probable structure with MS and MS/MS data based on library matching and/or expert characterization (Level 2). Level 2 confidence is usually the highest degree of certainty that can be expected in true NTA applications, including PFAS analysis, when there are no existing authentic standards to make a true confirmation. More recently Charbonnet et al. [23] further expanded this confidence classification to PFAS specifically. This work further explores diagnostic fragmentation, homologous series, isomers and other information in additional PFAS specific confidence levels [23]. However, the lack of standards does not prevent a strong case for an HRMS-based assignment using rigorous supporting data. This system builds on the process of collecting identification points being the standard guidance for enforcement controls of contaminants (e.g., EU-only) and residues in food developed for LRMS and other analytical methods [11, 24]. Work at the European Union (EU) level is ongoing to update this process for HRMS methods.

When reporting compound identifications, it is important to state the entire “case” for the assignment. Supporting data include the annotated mass spectrum with mass accuracy, molecular formula and structures for major fragments, confidence level for the identification, and when possible, an explanation of potential mass spectral fragmentation pathways. Supporting data that is not unique to individual new compounds, e.g., isotope distribution and molecular formula criteria, should also be included. Furthermore, external evidence of chemical identities, such as potential sources of contamination, knowledge of synthetic production routes (e.g., for homologous series), fingerprints of raw material stocks, transformation pathways, and presence of other confirmed chemical species (e.g., homologous series, mixture co-occurrence) can provide more confidence in the likely source of a chemical feature [25]. Combined with chemometrics, such methodologies are used to profile and trace drugs [26], as well as authenticate foods and beverages. In recent years, structure elucidation has been made easier by extension of mass spectral libraries/databases to a wider range of substances and by new database search tools [17, 27, 28]. However, there are significant data gaps for composition of products and publicly accessible maps of pollutant fingerprints from sources.

Combining targeted analysis of PFAS concentrations with NTA experiments can allow for semi-quantitative estimates of concentration for unknowns. Estimates can be based on single-point abundance ratios compared with a known PFAS concentration, or concentrations can be based on full surrogate calibration curves using a “close match” PFAS. It is worth noting that these surrogate estimates do not benefit from the matched stable isotope normalization used in traditional targeted measurements, and cross-chemical abundance ratios or calibration can exhibit up to an order of magnitude in relative error [21, 29]. Best semi-quantification is obtained if samples are spiked with a series of marker compounds of known concentrations, whereby accuracies of about a factor of three can be obtained for unknowns quantified against closely eluting marker compounds [20]. Well matched surrogates are capable of yielding estimates equivalent to matched standards [30], but there is limited-to-no ability to select good surrogates or provide uncertainty bounds for these estimates. Further, these estimates assume similar method amenability, and analysts would require orthogonal methods (e.g., LC + GC) to examine disparate chemical species such as anionic PFAS and larger neutrals. This can be exacerbated when performing multimedia site studies; concentrations in soil or vegetation may exhibit significantly different abundance than a water sample from the same location due to the influence of matrix background and different sample preparation schemes (e.g., limited water solubility). Even with these limitations, identification and relative quantification (e.g., inter sample fold-change comparison of a single chemical) data are often sufficiently useful to make decisions, (e.g., assessing treatment technology effectiveness) in the absence of absolute concentrations.

HRMS data processing is an additional obstacle to applications of this technique. Post-processing of raw HRMS data includes, but is not limited to, feature (peak) extraction, mass and retention time alignment of chemical features, background subtraction, and aforementioned assignment of chemical identity using library, suspect screening lists, and/or manual inspection. Spectra may require the labor-intensive manual annotation of MS and MS/MS data for proposed formula and structure. While data collection times for HRMS instrument runs are comparable to LRMS methods, it is not uncommon for data storage needs (e.g., multiple GB per study) and data processing times for analysis and data reduction to exceed a comparable LRMS experiment by orders of magnitude. Data processing is often a combination of original equipment manufacturer (OEM) software and user-specific scripts (e.g., R, python, SAS), which conduct the signal processing, statistical filtering, and reference a combination of publicly available and purchased databases. It is not uncommon to perform multiple analyses of data sets with different data extraction thresholds and match tolerances until an acceptably succinct output is reached. There are no widely applied best case practices for benchmarking a non-targeted data processing workflow, though there are some efforts to define them [31, 32]. A good test is that NTA workflows should be able to isolate, align and output already known analytes which are spiked into the samples or have been measured by a targeted analysis and replicate gross trends observed in those sample sets. User discretion is often needed to set threshold values in determining analytes that are or are not found (e.g., PFAS found to be <LOQ of a targeted method should not be expected to be detected via NTA).

Contrary to early beliefs and practices in HRMS, keeping the performance of the instrument in strict control is even more important in NTA than for targeted analyses since internal standards are not available in NTA to correct for variations in instrument performance. It is hence critical to keep a high degree of method/instrument consistency and cleanliness to ensure comparability between samples and batches by preventing contaminants in blanks, carry-over from previous runs or deteriorating ionization efficiency. Absolute sensitivity and prevention of drifts in retention times can be maintained throughout a run, by addition of regular cleansing sequences in the sample list and periodic cleaning of the ion source, along with regular instrument maintenance. The same stable isotope-labeled internal standards used for targeted PFAS analysis can provide some of the necessary QC features for data reduction. As known standards, they provide a check on instrument performance (e.g., resolution, mass accuracy, sensitivity, retention time, concentration estimation) and permit scalar sample normalization. Quality control samples (to check for sensitivity and drift in LC retention times) and calibration samples throughout the run (sequence list) can provide feedback on instrument performance. Pooled QC samples, consisting of aliquots of all the samples in an experimental batch, can also be used to monitor long-term performance of the system across batches [33].

Telltale characteristics of PFAS seen during HRMS-based NTA

PFAS are a unique chemical class that are readily separated from other chemicals commonly detected by HRMS analysis. Four characteristics of prominence with respect to PFAS are: ion suppression, characteristic mass defect range, homologous series, and characteristic fragmentation patterns. Each will be discussed individually following.

In initial investigations, PFAS peaks may be detected by scrutinizing the total ion chromatogram (TIC). Most often the TIC will be a uniform line (e.g., mobile phase, reference masses) which increase with increasing organic solvent percentage. Where PFAS elute a small dip in signal (negative peak—ion suppression) may occur in particular at higher concentrations (e.g., contaminated water) [25]. When signal suppression is noted investigation of the “peak” may give ideas for $m/z$ values to extract to further explore if they are part of a homologs series, have negative mass defects (NMDs), or have telltale fragments of PFAS as described below.

Mass defect

HRMS measurements can resolve individual isotopologues of chemicals and therefore chemical features are commonly assigned a monoisotopic mass. The theoretical monoisotopic mass of a chemical is based on the sum of the exact masses of the most prevalent, stable isotope of each constituent atom. Due to the high precision of HRMS, it is possible to measure the minor deviations between the exact mass of an atom and its nominal, integer mass number. For example, the monoisotopic mass of fluorine-19 is 18.998403, exhibiting a deviation of –0.001597 Da. The origin of this deviation is beyond the scope of this manuscript, but this measurable mass defect can be a useful signature when analyzing organic molecules. By happenstance, the most common backbone elements of organic chemistry (C, H, N) have either a positive or neutral mass defect, while all the remaining elements of interest (e.g., S, O, F, Cl, Br) have a negative mass defect [34]. In a practical sense, when chemicals are assembled, the combination contribution from each element imparts an overall mass defect to the chemical. For PFAS, the inclusion of significant fluorination results in an increased mass defect compared to non-PFAS molecules of similar mass. The addition of ³²S, ³¹P, and ¹⁶O in acidic moieties as well as the addition of other halogens (³⁵Cl, ⁷⁹Br) are all structural components of emerging PFAS that can contribute to more negative mass defects. Legacy PFAS exhibit a molecular negative mass defect due to their small size and high degree of fluorination, while polyfluorinated species (e.g., fluorotelomers, AFFF components) may exhibit a small, but positive defect (Table 1). Polyfluorinated species with low degree of fluorination can easily exhibit net positive mass defects from the contribution of ¹H and ¹⁴N, since the relative effects of each atom differ (Fig. 1). A large percentage (86%) of all PFAS in the curated OECD PFAS Database [35] fall within the defect range between −0.1 and +0.15 Da, and thus can be a good first cut for suspect data for distillation. Note that although many of these species exhibit positive mass defects, they are still relatively lower than the mass defect of an equivalent hydrocarbon of the same mass and are visually separated on mass defect plots (plot of nominal mass vs. mass defect) [36]. Kendrick mass defects (KMDs) can also be used, with PFAS falling typically within the defect range of 0.1 to −0.15 or −0.13 to 0.1 for CF₂-normalized KMD, as discussed later. Recently a mass over carbon (m/C) and mass defect over carbon (md/C) ratio was calculated and used and also systematically evaluated to discriminated PFAS when plotting m/C vs. md/C in complex food matrices [37, 38].

Table 1.

Some commonly detected PFAS with $m/z$, mass defect, Kendrick mass (CF₂ based) and Kendrick mass defect calculated.

Analyte	$m/z$of detected ion	Mass defect (Da)	Kendrick mass/CF2 (Da)	Kendrick mass defect/CF2 (Da)
PFOA	412.9664	−0.0336	412.99283	0.00717
PFHxA	312.9728	−0.0272	312.99283	0.00717
PFBA	212.9792	−0.0208	212.99283	0.00717
PFOS	498.9302	−0.0698	498.96213	0.03787
PFHxS	398.9366	−0.0634	398.96213	0.03787
PFBS	298.943	−0.057	298.96213	0.03787
HFPO-DA	328.9677	−0.0323	328.98876	0.01124
ADONA	376.9689	−0.0311	376.99303	0.00697
6:2 FTS	426.9679	−0.0321	426.99523	0.00477
8:2 FTS	526.9615	−0.0385	526.99523	0.00477
PFOSA	498.9535	−0.0465	498.98544	0.01456
PFBSA	298.9663	−0.0337	298.98544	0.01456
11Cl-PF3OUdS	630.8892	−0.1108	630.92958	0.07042
9Cl-PF3ONS	530.8956	−0.1044	530.92958	0.07042
6:2 mono PAP	442.9723	−0.0277	443.00065	−0.00065
6:2 diPAP	788.9751	−0.0249	789.02560	−0.02560
6:2 fluorotelomer sulfonamide betaine	570.0858	0.0858	570.12229	−0.12229
4:2 FTOH	264.0197	0.0197	264.03660	−0.03660
6:2 FTOH	364.0133	0.0133	364.03660	−0.03660
8:2 FTOH	464.0069	0.0069	464.03660	−0.03660

Fig. 1.

Mass defect of common elements.

Homologous series detection

Another useful characteristic for the identification of PFAS in environmental samples is that they often exist as homologous series. The common manufacturing processes of PFAS, electrochemical fluorination (ECF) or telomerization, yield a series of related PFAS with defined repeating units (Fig. 2). In many PFAS this is seen as a series of compounds related by the difference of CF₂ (exact mass 49.9968 Da) between related homologs at various industrial use, production, and contamination sites with increasingly lower abundance as the mass trends away from the predominant peak. Chromatographic separation can help confirm the relationship between homologous compounds, with clear retention time shifts between subsequent homologous compounds based on polarity and mass. For example, in reversed phase chromatography polar, lower molecular weight congeners elute first with increasingly larger homologs eluting later with RT shifts.

Fig. 2.

Common PFAS repeating units.

Repeating subunits

Various repeat units can also occur, allowing for intersecting homologous series. For instance, copolymers with n and m subunits, e.g., perfluoroalkoxy alkane (PFA), (C₂F₄)_n(C₃F₆O)_m (or even more subunits) could be observed with both C₂F₄ and C₃F₆O repeating units [25]. As molecules increase in size, they may be excluded from inside the pores of the stationary phase as was seen with a homolog series of polyethoxylated PFAS [25]. However, while the exact amount of individual shift strongly depends on chromatographic settings, a homologous series should exhibit clear systematic behavior. This trend can be used to verify or exclude data from a homologous series, depending on whether features match the expected pattern. Compared to the negative/positive mass defect methods, the advantage is also that the mass shift is ”normalized” to the repeat unit in the molecule, so you get a constant value for each ”repeat” elemental composition, see Eqs. (2) and (3).

Kendrick mass defect plots

A common transformation and plotting technique for homologous series was first described by data transformation to a CH₂ mass scale [39]. However, the Kendrick mass (KM) can be calculated for any repeating unit using the following Eq. (2):

$$\text{KM}\left(X\right)=\left(\text{observed}\text{mass}\right)\times \left(\frac{\text{nominal}\text{mass}\text{X}}{\text{exact}\text{mass}\text{X}}\right)$$ (2)

$$\text{KMD}=\text{nominal}\text{Kendrick}\text{mass}-\text{KM}\left(\text{X}\right)$$ (3)

For visualization KM (Eq. (2)) is plotted on the x-axis vs. KMD (Eq. (3)) on the y-axis. In doing so PFAS that differ simply by the repeating unit of CF₂ (but where the rest of the molecule is identical) would show up on a horizontal line. In Fig. 3, the plotting of the PFAS found in Table 1 were used to demonstrate this visualization. For ease of visualization the series of PFAS differing only by CF₂ repeating units were color coded by representative PFAS classes. A series of papers use this technique to explore PFAS in AFFF [40], in contaminated soil [41], in paper products [42] in polyfluoropolyether-based formulations [43] and in compound discovery in environmental and human samples [2]. Additionally, some instrument companies have in recent years provided the option to calculate the KMD in their software (e.g., Agilent MassHunter, Thermo Compound Discoverer) and there are publicly available R-scripts for the automated transformation of HRMS data for additional visualizations of any possible repeating units [44, 45]. When using automated software, data completeness is dependent on extracted chemical features, and care should be taken to ensure that lower-intensity compounds are not eliminated upstream in data processing. For true PFAS compound discovery, even without knowing the identity of suspect PFAS, KMD visualizations would at least inform the researcher of the existence of a related set of chemicals differing by a typical PFAS subunit. Limitations of the KM plots arise from the inaccuracy of mass measurements. For example, a commonly accepted mass error of 0.001 Da for accurate mass measurement also produces an error of 0.001 of the KMD. In practice, a tolerance of 0.001 Da of the KMD is applied to assign accurate masses to homolog series. To exclude false positive assignments, the systematic retention time shift within a homologous series can be used as further criteria [41].

Fig. 3.

Kendrick mass defect plots of PFAS from Table 1 for the visualization of six homologous series (“other” are single PFAS not belonging to any homolog series).

Diagnostic ions and fragments

Another characteristic that can be a telltale signature of PFAS are diagnostic fragment ions and neutral fragment losses [25]. Intentional fragmentation occurs in a collision cell, where collisions with inert gas imparts energy to the molecule and induces the breakage of chemical bonds, which can also happen in source. Fragment ions can be formed from molecular rearrangement and strict cleavage at weaker chemical bonds, typically hetero-atom linkages [25], which are often part of common repeating units (Fig. 2). Perfluorinated alkyl chains produce CF₃ and longer perfluoroalkyl ions, while ether oxygen linkages fragment at the C–O bond(s) (Table 2). Head-group losses (e.g., –CO₂, –SO₃) are common, but not diagnostic to PFAS, but the production of fluorosulfonate (FSO₃⁻) is a unique diagnostic fragment for sulfonated PFAS. Neutral loss of HF is common in polyfluorinated species, with this being particularly characteristic of fluorotelomers and polyvinylidines, with one HF elimination per backbone hydrogen. As discussed in Trier et al. [25] multiple (–HF) losses can also be indicative of the presence of fluorotelomer chains. Neutral loss of (–CF₂O) was also reported in PFAS literature, which may be indicative of alcohol functionality [46, 47]. For fluoroaliphatic betaines, such as FTSA-PrB (previously referred as “FTAB”), FTSAB and others, characteristic neutral losses of 103 Da ((CH3)₂NCH₂COOH) were observed [48]. Some zwitterionic PFAS had 45 Da neutral losses, characteristic of a tertiary amine (N,N-dimethylamino moiety) [40, 48]. Fluoroaliphatic betaines commonly show $m/z$ 104 (C₄H₁₀O₂N⁺) in positive mode MS/MS spectra, and the deprotonated betaine fragment ion at $m/z$ 102 was also observed in negative mode [48]. At high enough collision energies, zwitterionic PFAS of diverse functionalities (e.g., FTSA-PrB, thioamidoalkyl betaines, sulfonamidoalkyl amine oxides, thiohydroxyammonium sulfoxides), can produce an intense MS/MS fragment ion at $m/z$ 58 N(CH₃)₂⁺) [49].

Table 2.

Common fragments, losses and adducts of PFAS in the ESI⁻ mode.

Fragment	Mass (Da)	Losses	Mass (Da)
CF3−, C2F5−, C3F7−	68.9952, 118.9920, 168.9888	CO2	43.9898
CF3O−, C2F5O−, C3F7O−	84.9901, 134.9869,184.9837	SO3	79.9568
C2F3−, C3F5−, C4F7−	80.9952, 130.9920, 180.9888	H2CO	30.0106
C2F3O−, C3F5O−, C4F7O−	96.9901, 146.9869, 196.9837	CO	27.9949
CF2O−, C2F4O−, C3F6O−	65.9917, 115.9885, 165.9853	HF	20.0062
C2F4−, C2F4−, C3F6−,	99.9936, 149.9904, 199.9872	FSO3	98.9552
C3F6ClO−	200.9542	H3PO4	97.9769
SO2F−, SO3F−	82.9603, 98.9552	Adducts and n-mers	Δ Mass (Da) from M
C2F4SO3−	179.9504	[M − H + Na]−	21.981945
C2F5N−	132.9951	[2M + H]−	M + 1.007825
		[2M + Na]−	M + 21.981945

The accurate mass is a summation of all composition elements while neglecting the electron (0.00054858 Da). M refers to the negatively charged molecular ion, and Δ Mass refers to the mass deviation from M.

A further, novel approach for NTS of PFAS is the comprehensive data mining of MS2 mass fragment differences [50]. In brief, specific structural moieties such as fluorotelomer chains, CF3 groups or repetitive molecular parts like CF2O units can result in fragmentation spectra which show characteristic mass differences like ΔHF (20.00623), ΔCnH3F2n-3 (e.g. 346.00271 for n = 8) or ΔCF2O (65.99172). A wide range of PFAS classes can be tentatively detected and identified by the occurrence of distinct mass differences without previously predefined diagnostic fragments. This was already demonstrated by identifying different unknown PFAS homologs in a paper extract. The software “FindPFΔS” is available as both Python code and executable Windows application [50].

Fragmentation can also happen in the source of the mass spectrometer, producing diagnostic fragment ions, diagnostic neutral losses, as well as adducts and gas phase n-mers [25]. By performing alternating full scan and in-source fragmentation scan and extracting the chromatograms of diagnostic fragments, retention times of parent PFAS were flagged, following strategies like mass defect filtering which can be used to discover unknown PFAS [46]. In-source fragmentation flagging significantly narrows down the mass spectral range requiring further investigation for potential PFAS ions. However, caution should be taken using this approach to ensure in source fragments are novel PFAS and not simply artifacts of other well-known PFAS (e.g., CO₂ loss). Common fragments are terminal backbone breakages including C_nF_2n + 1⁻ (n is usually 2 and 3) for saturated PFAS, C_nF_2n − 1⁻ and C_nF_2n − 3⁻ for unsaturated or multiple H-substituted PFAS after a few HF losses, and C_nF_2n + 1O_x (x is usually 1 and 2, e.g., COF₃⁻, C₂F₅O₂⁻) for poly/perfluorinated (poly)ether compounds. Functional group fragments include SO₃F⁻ and SO₃⁻ for PFSAs, SO₃⁻ and SO₂H⁻ for α-H substituted PFSAs, PO₄⁻ for poly/perfluorinated phosphates, and SO₂N⁻ for poly/perfluorosulfonamides, etc. Diagnostic neutral loss, in theory, can also be helpful in discovering potential PFAS ions by comparing the full scan and in-source fragmentation scan spectra at each time point, this strategy however has not been seen in publications. Regarding n-mers, polyfluorinated carboxylic acids can also dimerize in the gas phase in the source of a mass spectrometer leading to telltale series of spectral peaks that differ by mass 21.9819 (Table 2) [25]. These represent the H⁺ (+1.0078 Da) and Na⁺ (+22.9898 Da) dimers of a PFCA and can be used to diagnose unknown PFAS carboxylic acids. In addition, some PFAS may form single or mixed adducts with LC mobile phase additives, such as has been observed for mono-alkylated polyfluorinated alkyl phosphate surfactants (monoPAPs) with formic acid (HCOO⁻) and Na⁺ [24, 25]. It should be noted that the [M-H]⁻ of a polyfluoroether carboxylic acid (PFECA) could be negligible, while fragments and diagnostic n-mers are easily visible (e.g., HFPO-DA). Additionally, n-mers can form through other ions such as K⁺ and NH₄⁺ and are not limited to dimers (e.g., trimers and tetramers) depending on compound concentrations.

These four characteristics of PFAS (ion suppression, characteristic mass defect range, existence of homologs and diagnostic ions) can be used in concert to glean through large quantities of HRMS data for peaks for further investigation. None of these taken alone are enough to conclude a PFAS is present, however together a strong probability exists.

Once an ion fragment is tentatively identified for an unknown suspect molecular feature, the fragment can be quite useful for detection of related PFAS, e.g., homologs, by extracting the fragment mass from the high collision-energy mass chromatogram. Commonly peaks will arise in the extracted chromatogram, reflecting the elution time of the related PFAS present at lower concentrations. Extraction of spectra at these elution times in low and high energy collision settings can illuminate repeat patterns serving both to assess the presence of these related molecules well as the functional interrelations amongst members of the series.

User guide (Fig. 4)

Fig. 4.

User guide for recommended steps in PFAS analysis.

Planning: a well-planned measurement is crucial and starts with correct sample treatment and preparation. Include blanks (a minimum of one; ideally up to three) to check for background contamination of labware, sample preparation and handling and sampling gear. Which analytes can be expected? Which extraction technique is suitable for a broad range of PFAS or for a specific group of PFAS? Which are the appropriate steps in sample preparation (mind the sample gear and labware to minimize sorption and contamination)? Which are well-suited LC solvents, buffers, ionization modes, MS source settings? Which mass range is useful?

Data acquisition: after your method is correctly set up, analyze your samples, controls and blanks. MS/MS or parallel MS/MS (All-Ion, SWATH, DDA, IE-DDA…) provides additional information on fragment ions and can be helpful in retrospective analysis.

Data processing/reduction: HRMS scans comprise large amounts of data that need to be reduced for successful identification. Useful techniques include filtering by sample comparison, blank filtering, mass defect, minimum peak abundances, and peak shapes as well as replicate filtering.

Data evaluation: once the data set has been reduced, various techniques discussed in this paper can be applied for successful PFAS determination. They include homologous series detection which can be coupled to retention time shifts, KMD analysis, characteristic fragment ions and TIC suppression. Based on this information assignment of confidence should be added with confirmation with targeted analysis as needed.

Data storage: a proper way and format of data storage is always recommended and can be useful for retrospective data analysis [51]. PFAS chemistry is highly dynamic and new compounds are put on market regularly. Suitable data storage formats allow the most flexible application of data mining.

Tentative identity confirmation and semi-quantification

Once a list of identified features has been filtered and grouped by class, visual inspection of the extracted chromatograms by the analyst is a necessary step to bolster identification confidence. The observation of ascending retention time patterns within homolog series is of importance, and determination of whether retention time spacing is consistent between consecutive homologs may further require manual inspection. As a rule of thumb for reversed phase LC-MS, retention time difference between consecutive PFAS homologs will typically decrease as the fluoroalkyl chain length of the series increases and is dependent on the elution gradient of the LC system. The chromatographic peak shapes can also be useful to infer whether the compound is derived from ECF or telomerization. Theoretically, both ECF and telomerization can yield branched or linear isomers, but the ECF process will produce mixtures and thus characteristic RPLC peak shapes of co-eluting chain isomers. Chromatographic peak shapes and their variation with mobile phases could also help clarify structures or distinguish between functional group isomers (e.g., ketone-PFAS presenting significant peak broadening and tailing with acidic mobile phases but not with ammonium acetate). MS/MS experiments can help increase the confidence level in PFAS identification, as discussed earlier (Table 2). Comparison with built-in software spectral libraries or published MS/MS spectra, such as those presented in D’Agostino and Mabury [48], Bugsel et al. [42], or Barzen-Hanson et al. [40], could help attain confidence level 2 of the Schymanski classification [22]. In absence of spectral comparison data, it is possible to input the compound structure in commercial software packages that generate theoretical fragmentation reactions (e.g., Mass Frontier, ACD/MS Fragmenter, CFM-ID). An additional software package called FluoroMatch is available for the visualization of acquired PFAS NTA data [52]. This two-step process consisting of (1) running full scan MS acquisitions for NTA and (2) subsequent sample reanalysis using targeted high-resolution MS/MS for confirmation, can be applied as-is or accelerated using a combined procedure. For instance, data-dependent acquisition (DDA) combines within a single injection full scan MS event with MS/MS scans of the top N most abundant parent ions automatically picked by the instrument. Setting the number of MS/MS events within a cycle (top N) to a high value could reduce the number of points per peak in Full Scan MS, and an elongated chromatographic gradient may thus be needed. As the DDA process may also miss low abundance ions, a recent study applied iterative exclusion DDA and introduced automated PFAS annotation using a vendor-neutral software FluoroMatch [44, 53]. For identity confirmation at the highest confidence level, samples should be rerun together with a certified standard, if available. Patents can also be consulted to see if the identified PFAS has been already reported in formulations or processes (e.g., patents.google.com), though the documents generally will not inform whether they have been applied for widespread production/usage.

Providing a sense of the relative intensity of newly-identified PFAS compared with legacy PFAS—or better yet, concentration estimates—can help partly address the question of environmental significance of the findings [54] and could inform whether efforts should be engaged in further acquiring ecotoxicological data. At least one certified standard for each PFAS class identified by nontargeted analysis should (ideally) be obtained with a view of providing concentration estimates. Depending on the structural similarity with the reference compound, quantitation confidence levels as introduced in Backe et al. [55] and Allred et al. [56] can be assigned to caveat estimates (Qn: quantitative, Sq: semi-quantitative, Ql: qualitative, Sc: screening). For instance, in-house synthesized [48] or Zonyl technical grade [49] 6:2 FTSA-PrB was initially used for the semi-quantification of FTSA-PrB (FTAB) and FTB in AFFF-impacted environments until the commercially available certified standards were issued. Though the confidence level for FTB was set at the screen data quality level at that time, concurrent analysis of 6:2 FTSA-PrB and 5:3 FTB certified standards later informed that the 5:3 FTB estimate had only been off by a factor of ~2 (unpublished data). McCord et al. [30] evaluated several reference standards as surrogates for HFPO-DA and PFECAs and discussed the implications for semi-quantitative estimates. Though estimation biases could be up to an order of magnitude, they were determined to be systematic rather than random [30]. Assuming suitable instrument signal stability throughout an MS batch sequence, one could use concentration estimates to inform time trends or environmental fate properties of emerging PFAS. If semi-quantified PFAS have high concentrations and are found widely across samples, researchers may consider sharing this data with PFAS synthesis contract laboratories to demonstrate demand for them to manufacture new standards.

Results

Case studies highlighting the importance of non-target analysis for PFAS

Cape Fear River—North Carolina

Legacy PFAS contamination in the Cape Fear River Basin was a known issue prior to the PFOA/PFOS phaseout, with PFAS manufacturing in the vicinity of Fayetteville, NC a major source [57]. Following the phaseout, changes in PFAS usage by manufacturing facilities was anticipated [58], but the specifics of the replacement chemistry, and its potential impact on the watershed, was unknown. In late 2011, a series of water samples were taken upstream and downstream of the suspected source to confirm continued emission of PFAS and identify novel and/or replacement PFAS species. Samples were analyzed using targeted analysis for legacy PFAS, which indicated significant PFAS discharge, most prominently perfluoropentanoic acid (PFPeA) [12]. Additional samples were analyzed via HRMS using a time-of-flight (TOF) mass spectrometer using the techniques previously discussed (negative mass defect, homologous series, and diagnostic in-source fragmentation) to identify PFAS associated with the source effluent. The first compound to be identified in the samples was hexafluoropropylene oxide dimer acid (HFPO-DA) as confirmed by an authentic standard comparison (e.g., accurate mass match, retention time matching, diagnostic fragment match) [59] also known by its ammonium salt form trade name “GenX” [2]. Binary comparison of upstream/downstream samples revealed many other chemicals with diagnostic PFAS negative mass defect that were introduced by the intervening source. Homologous series analysis identified a series of monoether PFAS (repeating unit CF₂, mass = 49.9968 Da) as well as a polyfluoroether series (repeating unit CF₂O, mass = 65.9917 Da). The novel PFAS molecules exhibited a CO₂ loss (−43.9898 Da), diagnostic in-source n-mers of Na⁺ and H⁺ and telltale polyfluoroether fragments (Table 2). These species were ultimately identified as polyfluoroether carboxylic acids (PFECAs). Additionally, a series of polyfluoroether sulfonic acids (PFESAs) likely stemming from the manufacture of the co-polymer Nafion^™ were identified, along with dozens of additional manufacturing byproducts [13]. Examination of both source and treated drinking water downstream demonstrated only negligible removal efficacy for the novel PFAS and that they were being delivered in community drinking water [60].

This work was followed by intense scrutiny by the North Carolina Department of Environmental Quality (NCDEQ) and the public to eliminate this PFAS source to the Cape Fear River. As a result, concentrations of PFAS in the industrial wastewater, the Cape Fear River, and the subsequent drinking water dramatically dropped. This increased visibility led the State of North Carolina legislature to fund the North Carolina PFAS Testing Network to study statewide PFAS concentrations in water and air (NC PFAST, 2020—https://ncpfastnetwork.com/about/). Additionally, this spurred the biomonitoring of these novel PFAS in serum of residents consuming the finished drinking water in Wilmington, NC ~70 miles downstream of the source. Three of the novel PFAS were measurable in >87% of the serum above the LOQ, and one PFAS was detected in >76% of the samples but was not quantitated due to a lack of authentic standards at the time, with none being found in control location samples elsewhere in NC [61].

West Deptford—New Jersey

The New Jersey Department of Environmental Protection (NJDEP) identified a hotspot of PFNA in drinking water near West Deptford, NJ through the US. EPA’s UCMR3 [62, 63]. NJDEP sought technical support via targeted analysis and NTA to identify the PFAS source and any additional PFAS. NJDEP suspected the PFAS was generated by the historical use of a polymer processing aid called Surflon S-111 [64] comprised of a mixture C₉, C11, and C₁₃ PFCAs (PFNA, PFUnA and PFTrA, respectively) from a nearby fluorochemical manufacturer. Targeted analyses of water and soil samples collected from nearby locations were able to detect appreciable quantities of PFNA and PFOA but were not able to unequivocally assign a single source based on GIS mapping contours, perhaps due to variability imparted from other local sources of these PFAS as well as possible loss with reported cessation of Surflon use some years earlier. Subsequent NTA HRMS investigations of soil samples [65] and surface/well water samples [66] were able to identify a homologous series of chlorinated-PFECAs. The identified Cl-PFECAs differed by a combination of repeating subunits ([C₂F₄O]_ethyl and [C₃F₆O]_propyl) terminating with a C₃F₆Cl tail, which yielded a diagnostic fragment ion. This structure had been suggested [10] based on existing patent literature, but had never been reported in environmental samples. The MS data acquired for NTA water samples collected in Europe downstream of a facility for the same manufacturer, but analyzed 5-years previous, were also revisited and found to contain the same suite of Cl-PFECAs. Washington et al. [65] was able to discern through contour mapping of soil samples taken in prevailing wind directions that the fluorochemical manufacturer was the source. McCord et al. [66] was able to discern the treatment efficacy of granular activated carbon for the novel compounds and semi-quantitatively estimate concentrations in comparison to PFNA/PFOA in well water.

Rastatt case—Baden-Württemberg, Germany

A routine control of a drinking water well in Rastatt, Southwest Germany, revealed high concentrations of PFAS in 2013. Further investigations proposed residues from paper production as sources of contamination. Sludge and paper fibers were continuously disposed on agricultural land between 2000 and 2008, obviously from production of impregnated paper containing PFAS products [67]. Targeted analysis detected PFCAs, PFSAs and two homologs of diPAPs (6:2 and 8:2). Sum parameters such as the TOP assay (total oxidizable precursors) and EOF (extractable organic fluorine) indicated further PFAS contaminants not yet covered by the target analytes. Nontargeted screening utilizing HRMS and Kendrick mass analysis was able to show further diPAP homologs (4:2/6:2 up to 12:2/14:2) and diSAmPAPs in the topsoil samples which represent the major source of contamination [41]. Furthermore, typical degradation products of diPAPs and diSAmPAPs and PFOS were detected in different soil horizons which are dominated by PFCAs. This agrees well with the findings that high-molecular weight PFAS and long-chain PFCAs were less mobile and mainly found in upper soil horizons [68], whereas short-chain PFCAs were more mobile and found in lower soil horizons and even in groundwater and drinking water or can be taken up by plants [69]. Fluorotelomer mercapto alkyl phosphates (FTMAPs; 6:2/6:2 to 10:2/10:2) and fluorotelomer sulfonic acids (FTSAs) have been identified on another site in Southwest Germany with a similar contamination due to the application of paper sludge [42]. Data on chemical structure, NMR and mass spectra of a synthesized standard of 6:2 FTMAP have been made publicly available on PubChem (CID 156620404).

Changshu, Jiangsu Province—China

Large fluorochemical industrial parks in China have been investigated as sources of PFAS release to the nearby environments. For instance, target analyses revealed PFOA concentrations up to 1.7 mg/l in surface water downstream of a fluoropolymer production center in the Xiaoqing River basin in Shandong Province [70]. Elevated PFAA levels were also reported in groundwater and tap water at this and other sites impacted by such industrial effluents [70,71,72]. In 2001, a High-Tech Fluorine Chemical Industrial Park was built in Changshu, Jiangsu, China. Liu et al. applied a nontargeted in-source fragmentation flagging method on influent sample collected in 2014 from the wastewater treatment plant (WWTP) in the park [46], and revealed 5 classes of new PFAS, including hydrogen substituted perfluorocarboxylates (H-PFCAs), chlorine substituted perfluorocarboxylates (Cl-PFCAs) and polyfluorosulfonates. By using influent wastewater samples from the same WWTP sampled in 2011, Wang et al. conducted a suspect and nontargeted analysis and detected 90 PFAS chemicals, including 37 newly reported [73]. Serially connected SPE cartridges (Oasis MAX, MCX, and HLB) were used to capture a large breadth of PFAS and the eluent was analyzed by QTOF-MS using an elongated reversed phase LC gradient. Peak picking rules included signal-to-noise ratios and absolute intensities above the set threshold, linking of likely homologs using CF₂ or CH₂CF₂ mass differences, and verification of retention time consistency among homolog series (ascending trends) [73]. Retrospective suspect screening was also performed using PFASTRIER, a PFAS compilation available on the U.S. EPA CompTox Chemicals Dashboard [74]. in wastewater samples in the industrial park such as H-PFCAs [46, 73], and polyfluorinated carboxylates [73], were in fact also discovered in surface water downstream of manufacturing facilities near Decatur, Alabama, USA [75]. Though the industrial wastewater was treated on site, fold-changes between influent and effluent samples indicated that some classes were predominately generated by the water treatment process, including increases up to 2000 times for perfluorinated dioic acids [73]. The authors recently conducted a follow-up nontarget screening study of Chinese municipal WWTPs at nationwide scale, and the semi-quantitation results suggested the presence of certain emerging PFAS at comparable or greater concentration estimates than legacy PFAS [76]. By also analyzing natural water downstream of the park (i.e., Yangtze River), the study confirmed that beyond the legacy PFAS commonly analyzed for, industrial effluents are introducing a broad range of unmonitored PFAS classes into the environment. Liu et al. collected fish from Yangtze River, ~10 km downstream the fluorochemical park. While the detection of Cl-PFCAs shows the effects of the industrial park on the surrounding environment and the possible bioaccumulative potential of these new PFAS the failure in detecting any H-PFCA molecules also suggest the possible environmental or biological degradation. Recently, with serum samples from workers in fluorochemical companies, Liu et al. again raised concerns over new PFAS’ health risks by showing high intensity of previously undetected PFAS including pentafluorosulfide perfluoroalkane sulfonates that were only reported in AFFFs (unpublished data).

Significance

As previously discussed, there are challenges associated with the use of NTA for analysis of samples for PFAS content. In trying to understand the totality of PFAS content in any sample, targeted analysis gives the most robust indication. All remaining approaches, including NTA, TOP, and EOF have deficiencies in understanding the remaining unknown PFAS content. Each of these methods have pros and cons with respect to the degree with which they help to understand the unknown content of PFAS.

Additionally, there are the deficiencies of NTA and PFAS in the finding of false negatives, or indications that PFAS are not present in a sample, when in fact they are. There are no guarantees that the methods above are both all-inclusive and failsafe with respect to the finding of unknown PFAS. There are several initial decisions that are made with any analysis that bias the data regardless of solid-phase extraction adsorbents chosen for sample preparation, separation by LC vs. GC, ionization modes (e.g., electrospray, atmospheric pressure chemical ionization), or mass spectrometer operation settings. Regardless of how thorough one is in data reduction, and even if a multiplatform approach was used, it may be impossible to overcome the preselected biases.

Though nontargeted analysis may proceed without quantitative standards, the methods employed for extraction and instrumental analysis should be designed to avoid measuring signals that were not in the original sample. False positives that may derive from contamination should be checked by means of procedural and field blanks; blank detections should be reported as it would inform other end-users of precaution in data interpretation for certain PFAS classes (e.g., H-PFCAs were determined to also be from mobile phase or the overall HPLC system) [46]. Thus far, nontargeted analysis has been most often applied in the context of highly contaminated samples, and there are many challenges ahead for application in ultra-trace level analysis. Some PFAS may also be lost during sample handling and storage [77], potentially resulting in a false negative in NTA, which has not been systematically studied. Conversely, some PFAS may be generated as artifacts during extraction procedures. It is well-known that fluorinated carboxylic acids can undergo methyl esterification in anhydrous acidic methanol, but there are other examples reported in recent literature [78, 79]. In the absence of available standards, spikes with dilute AFFF/surfactant solutions [79] or concentrated sample extracts [80] can be conducted to verify analytical recoveries. Ionization can “modify” PFAS structures sometimes. For example, PFCAs can fragment easily under the ESI mode and form unsaturated perfluorinated alcohols via structural rearrangement, which can cause falsely assigning the relevant ions into a new PFAS class [81]. Multiple hydrogen substituted PFAS are also found to easily lose CO₂ and/or several HF in the source in mass spectrometers, resulting in artifact “new” PFAS. It is thus recommended always to check for retention time overlap before reporting any PFAS compound as routine practice.

The presence of multiple isomers is another issue that should be paid special attention to in future NTA research efforts. In a recent study on New Jersey soils, Washington et al. [65] noted peak splitting in chromatograms of Cl-PFPECAs, and on the basis of fragmentation patterns surmised the presence of regioisomers. Chain isomers of PFCAs and PFSAs (and related precursors) resulting from manufacturing processes are currently the focus of intensive research interests for PFAS environmental forensics. In addition to the complexity of positional and chain isomers, there exist multiple PFAS that are functional group isomers (i.e., share the same molecular formula but with distinctly different functional groups). For instance, (C₁₃H₂₀O₄SN₂F₉)⁺ could correspond either to a fluorotelomer sulfonamidoalkyl betaine (4:2 FTSA-PrB) or to N-trimethylammoniopropyl perfluorobutane sulfonamide propanoic acid (N-TAmP-FBSAP) (Fig. 5). As the former is present in some fluorotelomer-based AFFFs [48] and the latter in ECF-based AFFFs [40], both compounds could potentially co-occur at impacted sites. Compared to chain isomers, functional isomers may be chromatographically resolved, which may facilitate identification efforts. A recent example can be found in Nickerson et al. [79], who noted the presence of an unknown isomer eluting 2 min from TAmPr-FHxSAPrA, and proposed the new structure based on MS/MS and chromatographic evidence. Cases where chromatographic coelution between two PFAS with similar monoisotopic mass have also been reported (Fig. 6), for instance 6:2 FTSA and HydroEVE (an ether-PFAS related to Nafion by-product 2) [61].

Fig. 5.

Two AFFF components sharing the same formula but derived through different synthetic pathways (ECF vs. telomerization).

Fig. 6.

Two PFAS (6:2 FTS and Hydro-EVE) closely related in monoisotopic mass (±5 ppm).

Other areas for future research include the development of multiplatform approaches for comprehensive nontarget PFAS profiling; in particular, GC-HRMS has been less frequently applied than LC-HRMS for NTA of PFAS and could be highly relevant for certain types of environmental samples (e.g., evaluate aerial emissions at landfill sites or during operations of fire-equipment testing sites). After decades of fruitful academic research, the composition of proprietary AFFF formulations is now well-known, with telltale PFAS signatures associated with certain formulations. Knowledge of the composition of the wide variety of surfactants and detergents used in industrial applications is not nearly as extensive and suggest need for further study. Enhanced oil recovery is an example where an unusual PFAS, p-perfluoro nonenoxybenzenesulfonate (OBS), was recently highlighted across several SSA and NTA studies in China [76, 82], but there may be hundreds of other ”exotic” PFAS and related degradation products that can contribute to the so-called “PFAS dark matter” [83]. PFAS nontargeted studies conducted both in negative ionization and positive ionization mode are also needed for a more comprehensive coverage of the current PFAS diversity.

Reproducibility of nontargeted analysis has also been questioned in the past, as quality assurance and nontargeted procedures are often not standardized [84]; however, this diversity is not necessarily an issue as it “can also lead to more discoveries” [85]. There are ongoing efforts in assessing reproducibility and performance of PFAS nontarget analysis, for instance through collaborative trials. For example, the performance of PFAS nontargeted analysis was evaluated in spiked and unadulterated house dust, as part of EPA’s Non-Targeted Analysis Collaborative Trial (ENTACT) [86]. Future interlaboratory studies could also be conducted with a view of evaluating and standardizing semi-quantitation practices.

Once a compound is isolated and identified to be a potential PFAS the main barrier in most investigations is the lack of an authentic standard for compound confirmation and subsequent quantitation. This is no different than when Martin et al., in 2004 wrote about how the analytical challenges that hinder PFAS research were “low concentration, lack of authentic standards, and unusual physical and chemical properties” [87]. This will not change, as an ever-broadening net of HRMS methods and techniques are used to probe environmental samples for the unknown. One hope, however, is that more comprehensive lists of PFAS, their $m/z\text{’s}$, structures and for which uses they are applied become available, as well as libraries with MS spectra. The latter is challenged by lack of access to pure standards. In any case, examples of inventory libraries include the NORMAN network, and recently also the lists provided in Glüge et al. [88] (Appendix 2 in Glüge). Ideally these should be made available via publicly and sustained platforms, (e.g., European Information Platform for Chemical Monitoring [IPCHEM], National Center for Biotechnology Information [PubChem] or via US EPA sites (e.g., Chemicals dashboard). Retention behavior is an identifying factor that has been neglected due to large retention time variability in LC analysis warranting more investigation. Beyond using numerous columns to obtain multi-dimensional retention indicators for each PFAS to partially overcome the large LC retention time shifts, ion mobility mass spectrometry derived collision cross section (CCS) [89,90,91] should be included whenever possible. These hopefully can serve as an important filtering parameter in data processing and facilitating correct PFAS annotation.

The last hurdle appears to be the access of the HRMS techniques for NTA to non-government and non-academic researchers and scientists. These methods used remain part science and part expert judgment in their application. Access to very expensive equipment, and the knowledge of how to probe these data efficiently and properly, though growing rapidly, is not available to all. Most of the work being done on the cutting edge of this type of research are in government and academic research laboratories specifically dedicated to this pursuit. The work is neither routine nor simple to explain/report for context. A certain professional interpretation requirement often accompanies the output.

Even with all the previously mentioned challenges, the use of HRMS and NTA for PFAS identification and understanding appears to be the most powerful approach in unknown contamination situations. With the use of proper techniques and data distillation it appears to be the most comprehensive and powerful analytical tool available to the analyst. Reliance on targeted analysis solely is both insufficient and shortsighted. We urge future research planning to include the use of HRMS/NTA approaches and investment in equipment to facilitate this capability.

Data availability

There is no specific data available this effort. Any data available would be through the referenced work.