Evaluation and validation of methodologies for the extraction of per- and polyfluoroalkyl substances (PFASs) in serum of birds and mammals

Advances in analytical techniques have allowed greater detection of environmental contaminants from small volumes of sample. Four methodologies were evaluated for the extraction of 53 per- and polyfluoroalkyl substances (PFASs) from eight classes in 200 µL of avian and mammal serum. Spiked serums at four concentrations (0, 0.5, 5.0 and 25 ng mL−1) were prepared by protein precipitation (PPT), enhanced matrix removal (EMR), weak anion exchange (WAX), and hydrophilic-lipophilic balance (HLB) solid-phase extraction cartridges. The extract from each methodology was analysed by high-performance liquid chromatography coupled with tandem mass spectrometry (HPLC–MS/MS), and concentrations were compared with known concentrations in the spiked media. EMR performed the best overall, with 40 of 53 compounds effectively recovered at 5 ng mL−1. Furthermore, EMR was effective overall at concentrations ranging from 0.5 to 25 ng mL−1 for 39 out of 53. Similarly, PPT was effective for 35 of 53 compounds at all spiked serum concentrations. There was a negative correlation between internal standard recovery for compounds with increasing octanol–water coefficients (Kow) for WAX (R =  − 0.65, p = 0.0043) and HLB (R =  − 0.62, p = 0.0077) extractions, indicating methanol may not be a suitable solvent for long-chain PFAS extraction from protein-rich tissues. EMR and PPT represent fast and effective methodologies for the extraction of PFASs from low volumes of serum which allows greater accuracy and precision that can be applied to future human and wildlife biomonitoring programmes. Supplementary Information The online version contains supplementary material available at 10.1007/s00216-022-03962-3.


Introduction
The development of analytical techniques to quantitatively measure per-and polyfluoroalkyl substances (PFASs) in serum is essential due to the strong bioaccumulation potential and associated negative impacts of PFASs in humans and wildlife. Currently, there are 4730 registered CAS numbers of substances that can be defined as PFASs [1]. PFASs of variable carbon chain lengths can be categorised into discrete classes based on their functional group, such as perfluoroalkanecarboxylic acids (PFCAs) and perfluoroalkanesulfonic acids (PFSAs), among others [2]. PFASs continue to be used for a variety of industrial and commercial purposes [3], such as aqueous film forming foam (AFFF), Scotchgard™ and Teflon™ despite evidence demonstrating their persistence, environmental accumulation and adverse health effects [4]. PFASs enter the environment by intentional [5] and accidental [6] discharge of PFAS-containing products and can then bioaccumulate in plants and animals exposed to contaminated soil and/or water [7,8]. Therefore, there is an increasing demand to improve the measurement of PFASs in biological matrices to include a wide range of chemistries more easily and affordably.
The distribution of PFASs throughout body tissue in wildlife is species specific and can be impacted by environmental and physiological changes [13]. Monitoring blood and blood products is commonly employed as a reliable and accurate indicator for total PFAS body burden [14]. PFASs are found in relatively high concentrations in blood and blood products from fish [15], birds [16] and humans [17]. Though, other perfused tissues, such as liver, lung [18] and brain [19], will also accumulate PFASs in significant concentrations. Nevertheless, blood is a preferable biological matrix to study as it can be collected without destroying the subject or without intrusive surgery and allows for longitudinal studies on specific individuals from a population over time [20]. Due to the strong protein binding affinity of PFASs for serum albumins [21], they are found in greater concentrations in serum or plasma compared to whole blood [22]. Therefore, serum is a more useful biomonitor for PFAS exposure in humans and wildlife.
Generally, the sample volume needed to detect PFASs in trace concentrations has decreased from 5 mL [23] to 0.1 mL [24], reflecting the need to reduce the cost and storage per sample and the impact to the subject, whilst maintaining relevant method reporting limits for human and wildlife biomonitoring (0.1 to 1 ng mL −1 ). The total number of PFASs monitored has increased from PFCAs and PFOS-related substances [25] to include PFASs with many chain lengths and functional groups due to their increased prevalence in the environment [26]. PFAS extraction is similar to those currently used for the extraction of other organic compounds from serum which include solid-phase extraction (SPE), protein precipitation (PPT) and ion pairing (Table S1). Keller et al. [27] previously validated six standardised methods for the extraction of 13 of the most commonly measured PFASs in an interlaboratory study. Kaiser et al. [26] then expanded the number of PFASs to include 61 compounds from six functional groups, by solid-phase extraction and ion pairing methodologies. Given the complexity of the serum matrix and potential for interferences in the confirmation of ions [28], research is needed to validate the extraction of a wider set of PFASs to reflect the growing list of compounds that are now routinely monitored in the environment.
In the last decade, exposure to PFASs has been investigated in several human and wildlife biomonitoring studies across multiple countries. Typical background concentrations of PFASs in serum and plasma from human and wildlife could be expected to range from 0.1 to 100 ng mL − [33]. Exposure to PFASs creates the potential for a suite of adverse health effects in humans [8,34,35], and wildlife [36].
This study aims to provide a fast, simple and robust extraction method for the trace quantification of 53 PFASs, from eight functional chemistry classes, in relatively small volumes (< 1 mL) of serum. This is achieved by comparing the extraction efficiency and instrument response for each of the compounds and classes from a range of methodologies. Avian and mammal serum (including human serum) was used to validate existing and novel extraction methods for a wide range of PFAS compounds and the application of the evaluated methodologies will allow clear and reproducible results from human and wildlife biomonitoring programmes.

Overview
Four methodologies for the extraction of 53 PFASs from eight classes were evaluated by comparing measured concentration of serum spiked with the known concentrations for each compound. The measured concentrations at four spiked levels were compared statistically to the known concentration to evaluate the efficiency of the method at a range of concentrations. (1) Protein precipitation (PPT) is conducted by the addition of acetonitrile to serum, which results in the formation of a white precipitate, which is then separated by centrifugation or filtration. (2) Enhanced matrix removal (EMR) is a proprietary sorbent developed by Agilent (Delaware, USA), designed to reduce matrix effects by trapping lipids by size exclusion and hydrophobic interactions in a slightly acidic solution (with the addition of 1% formic acid) and allowing aliphatic compounds, such as PFASs, to pass through. The use of acetonitrile also has the effect of precipitating proteins on the cartridge which are then filtered from the extract-further reducing matrix interferences. (3) Weak anion exchange (WAX) and (4) hydrophilic-lipophilic balance (HLB) are sorbents developed by Waters Corporation (New South Wales, Australia), each designed to retain compounds like PFASs by hydrophobic and electrostatic interactions in aqueous solutions. A strong solvent (such as methanol) is then used to elute PFASs from the cartridges for analysis.

Extraction evaluation
Chicken serum was used to evaluate and compare the four extraction methodologies (Fig. 1). The extraction of PFASs from chicken serum was conducted at four concentrations: 0 ng mL −1 (control), 0.5 ng mL −1 , 5 ng mL −1 and 25 ng mL −1 (treatment), which includes the range of commonly detected PFASs from humans and wildlife exposed to background levels and contaminated land and water. A mixed solution of native PFASs in methanol (0.5 mL) was added gravimetrically to reach target concentrations for treatment groups and the methanol was allowed to evaporate to dryness at room temperature. Then, frozen chicken serum was allowed to thaw at 4 °C; then, 10 mL was transferred to the four polypropylene centrifuge tubes each. The serum was vortexed for 30 min and allowed to equilibrate for 24 h at 4 °C.
Equal volumes of spiked serum were used for each extraction methodology and the final extract volume was kept consistent at approximately 0.2 mL serum to 0.3 mL acetonitrile to reliably compare the extraction efficiencies and matrix interferences. In methods requiring the use of methanol (WAX and HLB), the solvent was evaporated and reconstituted in the final extract solution of 60% acetonitrile.

Extraction validation
The methodology which performed the best for the extraction of PFASs in chicken serum was applied to samples of serum from human and horse samples. The extraction of PFASs from human and horse serums was conducted at four concentrations: 0 ng mL −1 (control), 0.5 ng mL −1 , 5 ng mL −1 and 25 ng mL −1 (treatment). A mixed solution of native PFASs in methanol (0.200 mL) was added gravimetrically to reach target concentrations for treatment groups and the methanol was allowed to evaporate to dryness at room temperature. Frozen human and horse serums were allowed to thaw at 4 °C; then, 2 mL was transferred to the four polypropylene centrifuge tubes each. The serum was vortexed for 30 min and allowed to equilibrate for 24 h at 4 °C. Triplicate serum samples of each spiked concentration (0.200 mL) were extracted using EMR cartridges on a positive-pressure displacement manifold, as detailed above.

LC-MS/MS analysis
Sample analysis was performed on an Agilent Technologies 1290 Infinity II liquid chromatography (LC) coupled with Agilent 6495C tandem mass spectrometer (MS/MS) with Agilent Jet Stream negative electrospray ionisation (AJS ESI-) [37]. Separation was achieved on a Zorbax eclipse plus RRHD C18 column (3.0 × 50 mm, 1.8 µm, Agilent Technologies, USA) with a C18 guard column attached. Gradient elution (14 min) using 2-mM ammonium acetate in ultrapure water (A) and MeOH (B) at 400 µL min −1 was used. Source conditions for the mass spectrometer are as follows: drying gas = 250 °C at 11 L min −1 , sheath gas = 375 °C at 11 L min −1 , nebuliser pressure = 25 psi, capillary and nozzle voltage = − 2500 V and − 1500 V, and high-and low-pressure RF (iFunnel) = − 90 V and − 60 V. The m/z transition with the highest intensity was used for quantitation (complete MS/MS parameters listed in Table S2). Linear calibration curves for all compounds to the average of the triplicate injections of ten individual levels from 0.05 to 50 ng mL −1 (R 2 > 0.99) and with an accuracy of ± 30% were achieved in MeOH. 4:2 FTSA, 6:2 FTSA, 8:2 FTSA and 10:2 FTSA were linear across eight calibration points (R 2 > 0.99) from 0.05 to 10 ng mL −1 .

Quality assurance and quality control
Internal standard recovery was determined by comparing the response of each sample to the average response of the calibration curve. The method reporting limit (MRL) was defined by the lowest calibration concentration with a signal-to-noise ratio (S/N) greater than 10:1 or three times the concentration in the blank calibration solution, whichever is greater. Samples with S/N between 3 and 10 were defined as less than the method reporting limit (< MRL) and responses with S/N less than 3:1 were considered nondetected (n.d.). The qualifier ratio (where two transitions were available) was set to ± 20% of the median response ratio of the calibration curve. Acceptable retention times (RT) were defined as 5% relative to the internal standard response of the calibration standards. If criteria for qualifier ratio or RT were not met, the sample was designated < MRL.
Several PFASs were detected in the blank chicken serum matrix above the method reporting limits. Background concentrations of PFASs detected in serum were relatively consistent within triplicate samples and between extraction methodologies. PFBA was detected in serum extracted by EMR, WAX and HLB methodologies ranging in concentration between 0.31 and 0.90 ng mL −1 . FOSAA was detected in chicken serum extracted by WAX (average: 0.24 ng mL −1 ) and HLB (average: 0.14 ng mL −1 ) methodologies. PFPeA (average: 0.62 ng mL −1 ) and PFHxA (average: 0.38 ng mL −1 ) were detected in chicken serum extracted by EMR. The average concentrations of PFAS detected triplicate blank samples were subtracted from subsequent treatment groups to obtain a blank-corrected recovery.

Statistical analyses
Concentration data were acquired and quantitated using Agilent MassHunter Workstation v10.1 and Quantitative Analysis v10.1 respectively. The log octanol-water partition coefficient (log K oc ) for compounds was predicted by the EPI Suite v4.11 KOWWIN™ package [38] where mass-labelled internal standards were assumed to have the same value as the native compound. The KOWWIN™ model performed well in predicting log K ow against measured values from the literature [39]. Descriptive statistics, statistical tests and data visualisation were performed with R v4.0.2 using RStudio v1.2.5019 and the following packages: tidyverse v1.3.0.9000 [40], rstatix v0.060 [41] and RColorBrewer v1.1-2 [42]. Internal standard-corrected recoveries were calculated as a percentage of measured concentration to spiked concentration and were assessed by deviation from the expected value. Furthermore, internal standard-corrected instrument responses were plotted against internal standard-corrected concentration for each compound and analysed by linear regression. The estimates for the coefficient were compared statistically to the calibration curve (target concentration) by ANOVA with post hoc Tukey's pairwise test to determine the accuracy of each method.
where B is the regression coefficient for the calibration curve (Cal), protein precipitation (PPT), enhanced matrix removal (EMR), weak anion exchange (WAX) and hydrophilic-lipophilic balance (HLB) respectively. PFASs were measured in blank serum; the application of the standard addition method is used to measure recovery without blank subtraction. The linearity (R 2 ) from 0.5 to 25 ng mL −1 for each compound is reported here but must be interpreted objectively as there are only four points each.

Method performance (mass recoveries)
The evaluation of four methods for the extraction of PFASs from serum is based on the internal standard-corrected concentrations of each compound with its mass-labelled analogue. The recoveries of 18 mass-labelled internal PFAS standards from six classes were compared to the average response from 10 calibration solutions (injected in triplicate), each spiked with 5 ng mL −1 for each compound. The internal standard recoveries for protein precipitation (PPT) and enhanced matrix removal (EMR) were performed within acceptable ranges for most compounds. Average recoveries for PPT ranged from 55 ± 3% to 123 ± 5%, except for 6:2 diPAP-13 C 4 which was 230 ± 8%, and EMR ranged from 45 ± 2% to 153 ± 13% for all compounds (Fig. 2). The weak anion exchange (WAX) and hydrophilic-lipophilic balance (HLB) extraction methodologies had greater variability in recovery, where PFTeDA-13 C 2 , EtFOSA-d 5 and EtFOSEd 9 each had recoveries far below 50%-indicating that the estimation for concentrations of long-chain PFCAs and some PASFs may be negatively impacted by these methodologies. There is a moderate negative correlation for the recovery of compounds with higher octanol-water partition coefficient for the WAX and HLB methodologies (Fig. 2). Chelcea et al. [39] noted that the strong affinity of PFASs to proteins can lead to a discrepancy between K ow and bioconcentration factors in biological matrices. The increased hydrophobicity of these compounds and their affinity for proteins in the serum [21] may be stronger than the methanol used to elute the compounds from the WAX and HLB cartridges and further investigation to the strength of solvent used for elution is required. Therefore, the EMR and PPT extraction methodologies for PFASs are preferred, as there is no clear trend for decreasing internal standard recovery with the octanol-water partition coefficient.
The use of mass-labelled internal standards for the quantification of any compound by mass spectrometry is the "gold standard" as you can more accurately estimate the efficiency of the extraction method by the addition of a known concentration of compounds with identical chemical properties [43]. Comparing the response from the mass-labelled internal standard from the calibration levels and samples is a useful tool to measure extraction efficiencies. However, due to the expense of isotopically labelled standards, many researchers use the response from another compound for quantification. This can result in biased recoveries and measurements of concentrations, due to the sorption coefficients of compounds with different carbon chain lengths. It may be impractical to purchase isotopically labelled standards for each native congener; however, the selection of standards should reflect the complete range of functional group chemistry and chain lengths to prevent reporting biased results. For example, if the response from PFOA-13 C 4 (C 8 PFCA) was applied to PFTrDA (C 14 PFCA), the internal standard-corrected concentration would appear much lower compared with using PFTrDA-13 C 2 as the internal standard. The internal standard recoveries should be reported as part of the quality assurance criteria (even if they are single-point calibration), along with the standard that is used to measure the concentrations for each compound.
EMR performed within an acceptable range for the most compounds (40/53) at 5 ng mL −1 (Fig. 3). All eleven PFCAs and nine PFSAs, the most commonly measured compounds, were all recovered effectively between 70 and 130%. Eleven PASFs were also effectively recovered, with the exception of 6:2 FTAB (37 ± 32%); all eight PFESs, three FTCAs and four FTSAs were also recovered within ranges of 50 to 150%. Of the diPAPs, only 6:2 diPAP was recovered effectively, whilst 6:2, 8:2 diPAP; 8:2 diPAP; and diSAmPAP were recovered < 28%. Similarly, all three PFPiAs were not effectively recovered by EMR, with recovery < 58%. Fig. 2 Relationship between the recovery of mass-labelled internal standards and the predicted KOWWIN™ octanol-water partition coefficient [38] for each extraction methodology EMR sorbent is designed to exclude biological matrix interferences, such as phospholipids, by size exclusion and hydrophobic interactions. diPAPs have a bis(polyfluoroalkane) phosphate ester and PFPiAs have bis(perfluoroalkane) phosphate functional groups, which are structurally similar to phospholipids which have a phosphate ester group between two aliphatic carbon chains. The chemistry of the phosphate functional group in diPAPs and PFPiAs, or the very long chain length, may be negatively affecting the extraction of these compounds by EMR extraction.

Effects of concentration on recovery
The average internal standard-corrected recovery of PFASs from three treatment levels (0.5, 5.0 and 25 ng mL −1 ) and seven classes was successful in at least one methodology ( Table 1). The perfluorophosphinic acids (PFPiAs) were not able to be recovered consistently by the tested methodologies. To the authors' knowledge, PFPiAs are not routinely monitored in abiotic or biotic matrices [12] and are typically not detected in drinking water [44], surface water [45], wastewater or soil [46]. Whilst PFPiAs are offered for a range of consumer and industrial applications including wetting agents in household cleaning products [47], there is little information on the toxicology of PFASs with this functional group to make an assessment about their associated exposure risks.
Perfluoroalkylcarboxylic acids (PFCAs) were recovered from PPT (84 ± 15% to 140 ± 8%), EMR (76 ± 10% to 122 ± 8%), WAX (39 ± 23% to 135 ± 11%) and HLB (32 ± 10% to 127 ± 9%) methodologies on average across three concentrations. EMR performed the best on average for PFCA extraction with all compounds recovered within ± 30% of the spiked concentration for each level. Furthermore, the regression coefficients for eight out of eleven PFCAs were not significantly different from the expected concentrations, indicating good recovery over a range of concentrations (0.5 to 25 ng mL −1 ). PPT also performed well for all compounds, except for PFHpA (140 ± 8%) and PFHxA (138 ± 8%) which had slight enhancement. The regression coefficients for compounds extracted by PPT indicate good agreement for C n>9 PFCAs with the expected concentration. Recovery of PFASs with WAX and HLB cartridges did not perform within 70-130% for compounds with increasing chain lengths. The recovery of PFTeDA by WAX extraction was 39 ± 23% and recoveries of PFDoDA, PFTrDA and PFTeDA by HLB  extraction were 52 ± 15%, 35 ± 20% and 32 ± 10% respectively. Analysis of the regression coefficients for recovery of PFASs by WAX and HLB indicates that few compounds were effectively extracted across the treatment range. Perfluoroalkylsulfonic acids (PFSAs) were recovered from PPT (78 ± 20% to 111 ± 9%), EMR (66 ± 19% to 96 ± 8%), WAX (10 ± 4% to 95 ± 7%) and HLB (12 ± 4% to 108 ± 8%) methodologies on average across three concentrations. PPT performed the best on average for PFSA extraction with all compounds recovered within ± 30% of the spiked concentration for each level and the regression coefficients for each compound did not differ significantly from the expected concentration. EMR performed well for all PFSAs, except for PFDoDS (66 ± 19%) which had its recovery slightly suppressed. Recovery of PFSAs was acceptable for WAX and HLB extractions, except for PFDS and PFDoDS. Recoveries of PFDS and PFDoDS for WAX were 49 ± 8% and 10 ± 4% respectively, and for HLB, recoveries were 22 ± 4% and 12 ± 4% respectively. Similar to the recovery of long-chain PFCAs, the recovery of long-chain PFSAs was significantly suppressed compared to the expected concentration.

Considerations for the quantification of PFASs
Serum is a complex matrix of biological material, including a range of proteins and lipids of different chemistries. Whilst each extraction methodology aims to reduce the concentrations of each of these materials to exclude them from analysis, the selection of PFASs often includes other organic molecules that can impact the quantification of PFASs by LC-MS/MS, for example taurodeoxycholic acid (TDCA, C 26 H 45 NO 6 S), a bile acid present in serum, that shares a similar retention time, precursor and product ion mass with PFOS, using a C18 column (m/z 498.9 → 80.0) [28]. This presents an analytical challenge to prevent the over-quantification of PFOS in biological matrices, as the sulphite fragment (m/z 80; SO 3 − ) is greater in abundance compared to the fluorosulfate fragment (m/z 98.9; FSO 3 − ) [37]. However, as TDCA does not contain any fluorine, the m/z 98.9 transition is not present (Fig. 4). Separation of TDCA and PFOS can be achieved on a column using either perfluorooctyl stationary phases [23] or an ion-exchange column [48]. Although TDCA was not removed by any of the extraction protocols used in this study, the slow elution gradient presented in this analytical methodology (t = 14.5 min) has achieved separation with a C18 column. Analysis of the m/z 80 and 98.9 transitions reveals that PFOS is eluted at t = 6.98 min and TDCA is eluted at t = 7.23 min where the response returns to baseline between each peak.
As many analytical methodologies strive for shorter run times and faster elution gradients, it is important to ensure that potential interfering compounds, such as TDCA, are separated from PFASs with equal precursor and product ion masses. To this end, the known qualifier ratio of two transitions from the calibration levels must fall within ± 20% of the ratio measured in the sample. Any deviation from the known qualifier ratio may be due to the presence of another compound with a similar retention time and precursor mass. In cases where PFASs have lower precursor mass, and therefore only one positive transition with m/z > 100, such as PFBA, PFMBA and PFPeA, the identification and quantification do inherit some risk. As the results for these compounds in this study indicate accurate recoveries, there does not seem to be any interfering matrix components that are suppressing or enhancing the quantification.

Validation of human and horse serums
EMR performed well as an extraction methodology for PFASs in human and horse serums. All PFCAs, PFSAs and PFESs were effectively extracted from serum at concentrations ranging from 0.5 ng mL −1 to 25 ng mL −1 , except for NFDHA at 0.5 ng mL −1 (58 ± 20%). At higher concentrations (5 and 25 ng mL −1 ), FTCAs and FTSAs had slightly elevated recoveries in both horse and human serums. The recoveries of PASFs were more variable, with most compounds effectively recovered at all concentrations and EtFOSA, EtFOSE, MeFOSA and MeFOSE exhibiting very low recovery ≤ 5 ng mL −1 . Only 6:2 diPAP and 6:6 PFPiA were recovered well from their respective classes. Overall, these results are consistent with results from extraction of PFASs from chicken serum ( Table 2). Extraction efficiencies from human serum by EMR are equal to or exceed WAX and HLB methodologies previously reported, especially for long-chain PFCAs and PFSAs [26], likely due to their increased affinity for serum protein. In an interlaboratory study, Keller et al. [27] found increased relative standard deviation (RSD) for long-chain PFCAs (PFDA, PFUDA) potentially due to the lower concentrations in the human serum samples, although the affinity of these compounds for serum protein may also explain the variability in results due to different extraction methodologies used (WAX, HLB, C18, PPT, ion pairing).

Recommendations and conclusion
The results from this study establish a robust, fast and simple methodology for the extraction of the widest range of PFASs from serum of multiple species compared with those previously published (Table S1). There is evidence that compounds with relatively high K ow are not successfully recovered by WAX and HLB cartridges. As these compounds are typically recovered well from other matrices [37], this negative effect is likely due to the strong interactions between the compounds and the serum protein. Elution with a stronger organic solvent may be applied to help the recovery; however, the use of acetonitrile has similar results with this study [26]. Overall, enhanced matrix removal (EMR) was the most successful extraction methodology for chicken, horse and human serums as the majority of compounds ranging in concentration from 0.5 to 25 ng mL −1 were recovered within acceptable thresholds. This methodology could be scaled to allow for quantification of higher concentrations (with a greater calibration range), although decreasing the limit of reporting is currently challenging with the volume of serum at 0.2 mL. This methodology is an important contribution to the quantification of PFASs, in small volumes of biological matrices for improved and effective biomonitoring application.