Combining the targeted and untargeted screening of environmental contaminants reveals associations between PFAS exposure and vitamin D metabolism in human plasma †

We have developed, validated, and applied a method for the targeted and untargeted screening of environmental contaminants in human plasma using liquid chromatography high-resolution mass spectrometry (LC-HRMS). The method was optimized for several classes of environmental contaminants, including PFASs, OH-PCBs, HBCDs, and bisphenols. One-hundred plasma samples from blood donors (19 – 75 years, men n = 50, women n = 50, from Uppsala, Sweden) were analyzed. Nineteen targeted compounds were detected across the samples, with 18 being PFASs and the 19th being OH-PCB (4-OH-PCB-187). Ten compounds were positively associated with age (in order of increasing p -values: PFNA, PFOS, PFDA, 4-OH-PCB-187, FOSA, PFUdA, L-PFHpS, PFTrDA, PFDoA, and PFHpA; p -values ranging from 2.5 × 10 − 5 to 4.67 × 10 − 2 ). Three compounds were associated with sex (in order of increasing p - values: L-PFHpS, PFOS, and PFNA; p -values ranging from 1.71 × 10 − 2 to 3.88 × 10 − 2 ), all with higher concentrations in male subjects compared with female subjects. Strong correlations (0.56 – 0.93) were observed between long-chain PFAS compounds (PFNA, PFOS, PFDA, PFUdA, PFDoA, and PFTrDA). In the non-targeted data analysis, fourteen unknown features correlating with known PFASs were found (correlation coe ﬃ cients 0.48 – 0.99). Five endogenous compounds were identi ﬁ ed from these features, all correlating strongly with PFHxS (correlation coe ﬃ cients 0.59 – 0.71). Three of the identi ﬁ ed compounds were vitamin D 3 metabolites, and two were diglyceride lipids (DG 24:6;O). The results demonstrate the potential of combining targeted and untargeted approaches to increase the coverage of compounds detected with a single method. This methodology is well suited for exposomics to detect previously unknown associations between environmental contaminants and endogenous compounds that may be important for human health.


Introduction
5][6] To measure the exposome, different techniques are needed, 7 and an important group of compounds is environmental contaminants.A subset of environmental contaminants has been described as environmental organic acids, dened as environmental organic compounds utilized in commerce with at least one ionizable proton. 8This group of compounds includes per-and polyuoroalkyl substances (PFASs), 9,10 the phenolic metabolites of polybrominated diphenylethers (OH-BDEs) 11 and polychlorinated biphenyls (OH-PCBs), 12 and bisphenols. 137][18][19][20] Because of the typically low abundances of environmental contaminants in blood, about 1000 times lower compared to endogenous compounds and drugs, 6 highly sensitive methods are required.This may be achieved by using targeted measurements with triple quadrupole mass spectrometry 21 or untargeted discovery with high-resolution mass spectrometry instruments coupled to liquid chromatography (LC-HRMS).The benet of using LC-HRMS in the untargeted screening mode is that the data may be used to discover previously unknown environmental contaminants and to nd previously unknown associations with both endogenous and exogenous compounds. 21,22e developed a method for high-throughput combined targeted and untargeted screening of environmental contaminants in human plasma using LC-HRMS.The methodology was inspired by our previous work in untargeted metabolomics and lipidomics, [23][24][25] applying fast and straightforward workows for data analysis not typically used for environmental contaminant screening.The result is a method applicable for high-throughput exposome proling of large cohorts, giving both quantitative measures of environmental contaminants and the possibility to reveal previously unknown associations between environmental exposures and endogenous compounds in the same experiments.

Study samples
The use of surplus sample volumes from routine samples was approved by the Uppsala Regional Ethics Committee, Uppsala, Sweden (01/367).The blood donor plasma samples (n = 100) were collected in 2021 and in agreement with the ethical permit, only age and sex were gathered for the subjects.The samples were stored at −80 °C until sample preparation.Prepared samples were stored at −80 °C until analysis.The Declaration of Helsinki and its subsequent revisions were followed.The demographic data of the subjects are summarized in Table 1.Newborn bovine serum (USA) was purchased from Sigma-Aldrich and stored at −20 °C until analysis.

Sample preparation
The samples were prepared in randomized order.Fiy mL plasma was added to 140 mL MeOH in 0.6 mL Eppendorf tubes, followed by adding 10 mL internal standards (IS) in MeOH (nal concentration 2.5 ng mL −1 of each IS).The samples were then vortexed for 15 s and centrifuged at 4 °C with 14 800 rpm.One hundred mL of the supernatants were transferred to plastic HPLC vials (Thermo Scientic) with polyimide-lined caps (Macherey-Nagel, Dueren, Germany) and stored at −80 °C until analysis.Samples were prepared in six batches with each batch containing up to 28 samples.Ten mL of each prepared sample were pooled and used for quality control throughout the experiments.For MS/MS (MS2) data collection experiments, a pooled prepared sample was concentrated about four times.The sample was evaporated under a gentle stream of nitrogen and reconstituted in 3 : 1 MeOH : H 2 O.

Calibration curves and quality controls
Internal standard calibration was performed for all compounds where both native and heavy isotope labeled standards were available.For compounds for which only heavy isotope labeled standards were available, one-point calibration compared with the corresponding internal standard was performed.Calibration curves, quality controls, and blanks were prepared using newborn bovine serum ltered through ENVI-Carb SPE tubes (250 mg, Merck).The used lters contain graphitized nonporous carbon and were used for the purpose of obtaining a calibration matrix with decreased background levels of, e.g., PFASs. 26Calibrators and QCs were prepared the same way as the authentic samples, but the total 150 mL MeOH used for precipitation was made up of 130 mL MeOH, 10 mL IS MeOH solution (the same as for samples described above), and 10 mL native reference MeOH solution (of varying concentrations).
The calibration curves were prepared at seven calibrator levels, spanning the plasma concentration range of 0.02-80 ng mL −1 .As most of the reference standards used were provided as mixtures, the same concentrations were used for all targeted compounds, i.e., not considering individual reference intervals for each compound.The approximate limit of detection (LOD) and lowest limit of quantication (LLOQ) for each specic compound were estimated by visual evaluation of the corresponding peaks in the calibrator samples, considering a peak with height >2 × 10 3 that matches the corresponding internal standard as a clearly detected peak.For estimations of LODs and matrix effects, background (blank) subtraction was performed for the compounds for which background levels were observed.The QCs were prepared at two concentration levels: 2 ng mL −1 (QCL) and 20 ng mL −1 (QCH).The calibration curve was prepared once with the rst batch of samples.Three of each QC (QCL and QCH) were prepared with all batches.Also prepared with each batch of samples was one blank sample without IS and one blank sample with IS using the same newborn bovine serum used for calibration curves and quality controls.
The calibration curve was also prepared in 3 : 1 MeOH : H 2 O (without matrix) to estimate matrix effects.Two corresponding calibration curves were also prepared for the heavy isotopelabeled standards to control their linearity.Solvent blanks (3 : 1 MeOH : H 2 O) with and without IS were prepared to control possible carry-over.The bovine serum calibration curves were used for estimating sample concentrations and 1/x 2 weighting was used for each curve to acknowledge the large range of concentrations covered.

Liquid chromatography-mass spectrometry analysis
Samples were injected in a randomized order.Aer every tenth sample, an injection of the pooled sample followed by a newborn bovine serum blank injection was done.Prior to injecting the study samples, all calibrator samples and QC samples were injected.Six repeated newborn bovine serum blank injections conditioned the column before any runs, and before the injection of samples, six repeated injections of the pooled sample were done.
Twenty mL of each sample was injected on a reversed-phase HPLC column (Accucore C18, 100 × 2.1 mm, 2.6 mm, Thermo Scientic) using an Ultimate 3000 HPLC system (Thermo Scientic) interfaced to a high-resolution hybrid quadrupole Q Exactive Orbitrap MS (Thermo Scientic).A 17.5 min long chromatographic program including a gradient was applied using the mobile phases H 2 O with 0.1% acetic acid and 10 mM NH 4 Ac (mobile phase A) and MeOH with 0.1% acetic acid and 10 mM NH 4 Ac (mobile phase B); 5% B for 1 min, 5-100% B for 9.5 min, hold at 100% B for 3 min, return to 5% B over 0.1 min followed by re-equilibration at 5% B for 3.9 min.The ow rate was 0.6 mL min −1 , and the column temperature was 55 °C.
HRMS was operated in negative ionization mode with the following source parameters: spray voltage: 2.5 kV, capillary temperature: 275 °C, sheath gas ow rate: 55, auxiliary gas ow rate: 15, sweep gas ow rate: 3, S-lens RF level: 50, auxiliary gas heater: 450 °C.The HRMS analysis was performed in the full scan mode, collecting data in the prole mode with a resolution of 70 000 in the range of m/z 212.5-750.
Following the full-scan experiments of all samples, MS2 experiments were performed using a solution of all standards and references (both native and heavy isotope-labeled) at a concentration of 50 ng mL −1 as well as a concentrated pooled sample (four-fold concentration).The MS2 experiments were performed in six m/z windows: (1) 212.5-300 m/z, (2) 290-390, (3) 380-480, (4) 470-570 m/z, (5) 560-660 m/z, (6) 650-750 m/z.For each separate m/z window, ten experiments (with separate injections) were done, both for the reference solution and the pooled sample: one full scan (MS1) followed by nine datadependent MS2 scans at nine different collision energies (normalized collision energy, NCE): 10, 15, 20, 22.5, 25, 27.5, 30, 35, and 40 (in total 120 experiments).Apart from varying NCEs the settings were the same for all data-dependent MS2 experiments: resolution: 35 000, loop count: 10 (TopN), maximum injection time: 200 ms, and isolation window: 1.9 m/z, with data collected in the prole mode.Chromatography was performed in the same way as the main MS1 experiments.Twenty mL was injected for each experiment.

Statistical analysis
The LC-HRMS data for targeted compounds were processed using TraceFinder 4.1 soware (Thermo Scientic) using ten ppm mass tolerance to extract ion chromatograms.The concentration of targeted compounds was estimated using 1/x 2 weighted calibration curves with internal standard calibration.For some of the compounds, a background was detected in the newborn calf serum used as the matrix for calibration curves and QCs (in order of increasing background level: PFDA, 6:2FTS, PFOS, PFTrDA, PFNA, and PFUdA).For these compounds, the concentration in the calf serum was calculated using the standard addition method, and the estimated concentrations were adjusted using this information.Injections of the precipitation solvent showed negligible background of these compounds, thus identifying the calf serum as the background source.
To investigate the associations of the targeted compounds with age and sex, linear models were tted on log-transformed concentration data from TraceFinder, including compounds present in $15% of the samples.The linear models used age, sex and the interaction age: sex as independent variables and compound concentrations as dependent variables, according to the following equation: The variables were evaluated individually with p < 0.05 as the limit for signicance.The correlations of the concentrations between compounds were estimated using Pearson correlation.The resulting correlation coefficients were used for distance calculation between the compounds and the obtained distance matrix was subjected to hierarchical clustering.This is plotted as heatmaps, where the compounds are clustered based on the correlation between them.

Identication of unknown compounds
In the untargeted evaluation, features present in >90% of the samples and a correlation coefficient of absolutely >0.6 were used to obtain unknown features having at least a moderate correlation with targeted compounds.The m/z of the resulting features was compared against the NORMAN database. 35To improve our identication of unknown compounds, we made a linear model to predict log K ow values for the unknown compounds using the retention times and listed log K ow values in the NORMAN database for the targeted compounds (estimated by EPISuite 36 ).When multiple matches were found, the candidates were evaluated based on the listed log K ow values compared to our predicted log K ow values.A similar log K ow was deemed to increase the likelihood of the matching compound.The NORMAN database focuses on exogenous environmental compounds.For unknowns deemed to correspond to endogenous substances not found in the NORMAN database, we used HMDB, 37 Metlin, 38 and Lipid Maps 39 to nd likely candidates within 5 ppm of the observed m/z.Furthermore, the available MS2 spectra were compared with molecular structure databases using CSI:FingerID. 40Finally, the matching MS2 spectra were manually evaluated to strengthen the identication.

Results
We developed a combined targeted and untargeted method for screening environmental contaminants using LC-HRMS.The method was optimized for a total of 37 environmental contaminants across different chemical classes: PFASs (24 compounds), OH-PBDEs (two compounds), OH-PCBs (seven compounds), HBCDs (three isomers), and bisphenol A. One hundred plasma samples from blood donors ranging in age from 19 to 75, with 50 females and 50 males, were screened with the method.Nineteen of the monitored compounds were detected across the samples.Their concentrations were estimated based on calibration curves.Full names and abbreviations for the 19 compounds are given in Table 2.The method was validated by evaluating repeated injections of pooled and spiked QC samples.The MS1 and MS2 data were further investigated for untargeted compounds associated with the detected targeted compounds.

Method validation
The developed method displayed desirable robustness and selectivity for the targeted analytes.The CV in concentration and retention time for the detected targeted analytes was calculated based on the repeated injections of the pooled sample (containing equal volumes of all 100 samples, total number of injections, n = 22).The CV in concentration ranged from 0.9% to 12.3%, with a mean CV of 5.9%.The CV in retention time ranged from 0.1% to 0.4%, with a mean CV of 0.2%.This demonstrates the high instrumental reproducibility of the method.A representative chromatogram from an injection of the pooled sample is presented in the ESI (Fig. S1 †).Validation results for spiked QC samples and calibrators are summarized in the ESI (Table S2 †).
The calibration curves for the targeted analytes, using 1/x 2 weighting, generally exhibited good linearity with a mean coefficient of determination (R 2 ) of 0.959 (range 0.594-0.995).Except for 6:2FTS (R 2 = 0.594), BPA (R 2 = 0.612), and PFUdA (R 2 = 0.881), the remaining compounds exhibited R 2 -values in the range 0.974-0.995.Carry-over was evaluated by investigating the presence of the targeted compounds in the solvent blank injected directly aer the highest calibrator (80 ng mL −1 ).For the majority of compounds, no carry-over was observed, and for the two compounds with observed carry-over (FOSA and PFHpA) the carry-over was very low considering the concentrations observed in actual samples (corresponding to <40% of the lowest calibrator, i.e., <0.01 ng mL −1 aer injection of 80 ng mL −1 ).
For 21 of the 37 targeted compounds, including 16 of the 19 compounds detected in the samples under study (see Section 3.2.below), the response was good with clearly detected peaks in the lowest calibrator sample (0.02 ng mL −1 ) with average peak heights 7 × 10 3 (ranging 2 × 10 3 -2 × 10 4 ; adjusted for background concentrations when relevant) and matching the corresponding IS (where IS was available).By using 10 ppm ltering all noise was virtually removed from the extracted ion chromatograms, resulting in easily interpretable data.For the 21 compounds with good response at the lowest calibrator, LODs were estimated to be approximately 0.01 ng mL −1 , i.e., half of the lowest calibrator with the corresponding LLOQ (lowest limit of quantication) at 0.02 ng mL −1 .For these compounds the calibration curves spanned the plasma concentration range of 0.02-80 ng mL −1 .For the remaining 16 compounds, the response was not as good with large variations in response between compounds.The majority of these compounds, with the exception of N-MeFOSAA, N-EtFOSAA, and 4-OH-PCB-187 (that had acceptably low LODs), were not detected in the samples under study.By visually evaluating the peak height at the different calibration levels we estimated LODs and LLOQs as follows: 4:2FTS, N-MeFOSAA, N-EtFOSAA, 6-OH-BDE47, 4-OH-PCB-187, and 4-OH-PCB-61, LOD at 0.04 ng mL −1 and LLOQ at 0.08 ng mL −1 (calibration curves spanning the plasma concentration range 0.08-80 ng mL

Results of targeted analysis
The developed method was applied to analyze 100 plasma samples from blood donors.Out of the 37 targeted compounds, 19 were detected across the samples including 18 PFAS compounds and one hydroxylated PCB metabolite (4-OH-PCB-187) (Table 2).
Eleven of the 19 compounds were detected in $93% of the studied samples, whereas some were only detected in a subset.The least detected compounds were N-EtFOSAA and 4-OH-PCB-187, which were detected in 10 and 15 samples, respectively.The estimated concentrations for all compounds were within the low end of the calibration curve, with the highest single concentrations observed across the samples for PFHxS at 31.8 ng mL −1 and PFOSs at 15.2 ng mL −1 (Table 2).Ten compounds (in order of increasing p-values: PFNA, PFOS, PFDA, 4-OH-PCB-187, FOSA, PFUdA, L-PFHpS, PFTrDA, PFDoA, and PFHpA) were positively associated with age (pvalues ranging from 2.5 × 10 −5 to 4.67 × 10 −2 ) (Table 2).Three compounds (in order of increasing p-values: L-PFHpS, PFOS, and PFNA) were associated with sex (p-values ranging from 1.71 × 10 −2 to 3.88 × 10 −2 ), all with higher concentrations in male subjects compared with female subjects (Table 2).No signicant age × sex interactions were found.Four illustrative examples (PFNA, PFOS, PFDoA, and 4-OH-PCB-187) of the observed differences regarding age and sex are presented in Fig. 1.

Results of untargeted screening
In the untargeted data analysis, 39 538 features were detected and quantied.Eleven out of the 18 compounds found in the targeted approach were also quantied in the untargeted data.The compounds not detected in the untargeted data analysis were compounds with low concentrations or low responses, e.g., 4-OH-PCB-187 and 6:2FTS.For the eleven compounds detected using both methods, there was a good correlation between the concentrations estimated using the targeted approach and the same data analyzed using the untargeted approach (ESI, Fig. S5 †).The quantitative results of the untargeted approach were similar to the targeted approach with strong correlations between long-chain PFAS compounds (Fig. 3).
Including all features detected in $90% of the samples (4939 features), we found fourteen features with statistically signicant associations (correlation coefficients 0.48-0.99) to seven of the conrmed environmental contaminants (Fig. 4).Based on the available MS1 and MS2 data, we identied four compounds at condence level 2 (putative identications, supported by MS2 spectra) and three at condence level 3 (tentative identications).The condence levels for identication used here were introduced and described by Schymanski et al. to communicate the condence of HRMS identications.

View Article Online
Three compounds correlated strongly (correlation coefficients 0.76-0.99)with six targeted PFASs (PFOA, PFNA, PFOS, PFDA, PFUdA, and L-PFHpS).By evaluating the mass spectra and retention times compared to the known targeted compounds, these three were identied at condence level 2 (Table S3 †) as branched PFOS and two in-source fragments of PFOA and PFNA.The branched PFOS was identied based on exhibiting an m/z within 2 ppm of the theoretical m/z and a retention time 0.1 minutes before linear PFOS.Various branched PFOS variants were included in the used reference mix, matching the untargeted compound(s) observed here, although it is not possible to determine which specic branched forms were observed here based on the data.The two in-source fragments can be formed from loss of carboxylic groups from the parent compounds (loss of CO 2 ).They also share the exact retention times and exhibited high correlations (0.98-0.99) with the parent compounds, making in-source fragmentation their most probable source.
Five endogenous compounds, all strongly correlating with PFHxS, were identied (Table 3).The identication process for these endogenous compounds started from the observed m/z of the respective molecular features.Using the databases HMDB, 37 Metlin, 38 and Lipid Maps 39 we searched for compounds within 5 ppm from the observed m/z to nd likely candidates with matching molecular formulae.When several possible candidates were present we evaluated and compared the different candidates to select the most likely compound to be observed in human plasma (e.g., excluding plant metabolites).When available we compared the log P values presented in the databases between the different candidates for matching with the observed retention times.Finally, the MS2 spectra (available for four of the ve compounds, identied at condence level 2) were carefully evaluated to identify characteristic fragments to further strengthen the proposed identities.Annotated MS2 spectra for these compounds are presented in the ESI.† Three metabolites of vitamin D 3 were strongly positively correlated with PFHxS (correlation coefficients 0.59-0.71)and with moderate correlation towards branched PFOS and L-PFHpS (correlation coefficients 0.24-0.).For both compounds, both adducts correlated with PFHxS with highly similar correlation coefficients (Table 3).The observation of two adducts exhibiting the same correlation for these compounds gives further condence to the proposed identities.For 1,25(OH) 2 D 3 -26,23-lactone and 24(O)-1,23,25(OH) 3 D 3 -glucuronide, the concentration in the pooled sample used for MS2 data collection was sufficiently high to collect qualitative MS2 spectra supporting the proposed structures (ESI †).For 24(O)-1,23,25(OH) 3 D 3 no useful MS2 spectrum was collected; therefore, this compound is labeled as tentatively identied (condence level 3).However, the observation of the glucuronide conjugate (24(O)-1,23,25(OH) 3 D 3 -glucuronide) of the same compound, also correlated with PFHxS, supports the assigned identity of the non-conjugated form. 42 Finally, two forms of the diglyceride lipid DG 24:6;O (iden-tied at condence level 3) also strongly correlated with PFHxS (correlation coefficients 0.66 and 0.69) and with moderate correlation towards branched PFOS and L-PFHpS (correlation coefficients 0.24-0.41)were identied.The observed m/z for these lipids deviated only 0.11 ppm from the theoretical m/z, making the assigned molecular formula, typical of diglyceride lipids, highly probable.Annotated MS2 spectra supporting the proposed identities are found in the ESI.Several structural isomers of these lipids are present in plasma, and here two separate peaks (corresponding to two features) were observed and identied.The annotated identity DG 24:6;O refers to the total number of carbons, double bonds and modications (six double bonds and an OH-group) of the two fatty acids.In the HMBD, 20 different structural isomers for these DG lipids are reported, and authentic references would be required to specically identify which isomers are found to correlate with PFHxS in the present study.

Discussion
As outlined by Vermeulen et al., several challenges lie ahead for exploiting the full potential of exposome research. 7One major point is to improve the available technologies to screen for View Article Online exogenous chemicals at higher-throughput rates and lower costs and to further develop the chemical and spectral data resources to identify these chemicals in samples.With that in mind, we developed a combined targeted and untargeted LC-HRMS method that is fast, simple, and highly applicable for high-throughput analysis of large cohorts.To our knowledge, the metabolomics-inspired workows used here have not been previously used for the screening of environmental contaminants.
High-resolution mass spectrometry has been described as a prime example of a technique suited for assessing the exposome. 7Using HRMS, thousands to tens of thousands of chemical features may be measured in a single analytical run.Data analysis tools developed for, e.g., metabolome proling may be used to map the exposome from untargeted HRMS data.A wide range of analytical methodologies are available for different compound classes.For example, several classes of common environmental contaminants, e.g., PCBs and BDEs, typically require gas chromatography-based methods, whereas their hydroxylated metabolites (OH-PCBs and OH-BDEs, respectively) can be measured by LC-based methods with electrospray ionization. 43,44Here we focused our methodology on compounds that form ions during negative electrospray ionization, a group of compounds that have been labeled environmental organic acids and include, e.g., PFASs, OH-PCBs, OH-BDEs, bisphenols, and HBCDs. 8The developed method was optimized with regard to the response and chromatographic performance of all targets from the different compound classes.The targeted compounds were chosen based on environmental relevance and commercial availability.We decided to limit the required sample volume to 50 mL, which is relatively low in environmental analysis 26 so the method should apply to the analysis of cohorts where sample volume is limited.
Across the 100 studied samples from blood donors, we detected 19 of the 37 targeted compounds, with 18 of those being PFASs.PFASs are a large group of synthetic compounds with more than 10 000 different PFAS chemicals listed by the US Environmental Protection Agency. 45PFASs were detected in all 100 studied samples, with similar concentrations to those reported recently in similar study populations. 46,47The unusually high concentration of PFHxS in a single sample may be related to PFAS contamination of the drinking water in certain areas of Uppsala, which was discovered and mitigated in 2012.The contamination has been sourced to PFHxS-based aqueous lmforming foam used for reghting during military training at a nearby airport until at least 2003, which keeps leaking to the environment today. 48,49uring the last 20 years, the industrial production of PFASs has changed signicantly internationally due to increased awareness of the associated risks and regulatory pressures. 50ome specic PFASs have been phased out, most notably PFOS and PFOA, while short-chain alternatives have seen increased use as replacements.Short-chain PFASs are known to be eliminated faster in animals and humans than the long-chain legacy PFASs that bioaccumulate and are therefore expected to be less toxic.However, there is growing evidence that the toxicity of some short-chain PFASs has been severely underestimated, which raises new concerns. 51,52In the present study, nine PFASs showed a positive association with increasing age.Most of these were long-chained, 9 except for short-chain PFHpA.The higher concentrations in older subjects could be related to the bioaccumulating properties of these PFAS, and similar associations have been reported previously. 53,54We also found higher concentrations of L-PFHpS, PFOS, and PFNA in male compared to female subjects, which agrees with previous observations for several PFASs. 53,55It has been proposed that this sex difference could be related to occupational exposure to PFASs in maledominated occupations. 53,56en investigating the correlation between targeted analytes, we found strong correlations of the concentrations between long-chain PFASs across the samples, potentially related to the bioaccumulating properties of these compounds and possibly indicating common sources of exposures. 57Structurally related PFASs such as FOSA and N-MeFOSAA, and short-chained L-PFBS and L-PFPeS were also found to correlate.Among the targeted compounds, short-chained 6:2FTS exhibited virtually no correlation among the targeted compounds.This compound has been increasingly used as a substitute for various industrial applications that formerly required the use of PFOS or PFOA, and is less toxic and bioaccumulative compared to those more uorinated compounds. 9,58,59Decreased accumulation together with differing manufacturing trends could explain why no correlations were observed for this compound.As could be expected, structurally related compounds with similar chemical properties were found to correlate in concentrations.
The nal targeted compound detected in the samples was 4-OH-PCB-187.This OH-PCB had low coverage at 15% and was mainly detected in subjects >50 years old, showing a positive association with increasing age.Hydroxylated PCBs are strongly retained in human blood and may cause endocrine-related toxicity. 60,61The 4-OH-PCB-187 is one of the predominant OH-PCB congeners in human plasma, and in several studies, the most abundant OH-PCB. 43,60That this specic OH-PCB is detected in the studied samples is therefore not surprising, and the association with increasing age is in line with its accumulating properties.PCBs were phased out in the 1970s but are still present in our environment, with the main exposure source being the diet, in Sweden, particularly from the consumption of fatty sh. 60,62,63aving evaluated the method's performance on targeted analytes, we then explored the untargeted data.An initial challenge was optimizing our peak-picking methods to detect the typically low-level environmental contaminants amongst the much more abundant endogenous compounds also detected with the untargeted method.With too liberal criteria for peak-picking, noise will be included as features, and with too strict criteria, low-level compounds will not be detected.

View Article Online
Considering this, we found a balanced set of parameters that could extract and integrate most of the targeted compounds from the untargeted data.We compared the output from the untargeted data with that from the targeted approach and found high agreement and the same general trends with strong correlations between several long-chain PFASs.The targeted approach was, however, better for detecting compounds of very low concentrations or with low responses.The best choice may be to use a combined approach, collecting untargeted data that is evaluated using a list of targets or suspects and with a separate evaluation of the untargeted data.
It is challenging to identify features when data is collected in an untargeted mode.To facilitate feature identication, we therefore used a detailed strategy to collect high-quality MS2 spectra.The complete m/z range of the primary full scan method was divided into six overlapping m/z windows, and nine data-dependent MS2 scans, with varying collision energies, were collected using a pooled sample of all individuals included in the study.This improved our success rate for identication.As an example of this, linear peruorocarboxylic acids are almost fully fragmented even at low collision energies, while peruoroalkylsulfonates and the vitamin D 3 metabolites require high collision energies for sufficient fragmentation (example MS2 spectra in ESI †).
We identied three metabolites of vitamin D 3 with strong positive association with PFHxS: 1a,25-dihydroxyvitamin D 3 -26,23-lactone (1,25(OH) 2 D 3 -26,23-lactone), 24-oxo-1a,23,25-trihydroxyvitamin D 3 (24(O)-1,23,25(OH) 3 D 3 ) and 24-oxo-1a,23,25trihydroxyvitamin D 3 -glucuronide (24(O)-1,23,25(OH) 3 D 3glucuronide).Vitamin D 3 is a type of vitamin D made in the human skin under UV light, as opposed to vitamin D 2 , which humans get from the diet. 64Associations between vitamin D 3 biomarkers and PFAS have previously been reported, 65 and recently, the vitamin D receptor has been identied as a potential target for the toxic effects of PFASs. 66Previously Etzel et al. observed PFHxS to be associated with higher vitamin D concentrations and lower odds of being vitamin D decient. 65his observed association was not modied by age and sex.In that study, PFAS concentrations were compared with concentrations of the main vitamin D metabolite 25-hydroxyvitamin D 3 . 65We also found 25-hydroxyvitamin D 3 and the additional active vitamin D 3 metabolite 1,25-hydroxyvitamin D 3 in our untargeted data but found no statistically signicant associations with PFHxS or other measured PFAS compounds.It has been reported that other endocrine-disrupting chemicals may affect circulating concentrations of vitamin D, 67 and vitamin D status has been associated with several diseases, including cancer, cardiovascular disease, multiple sclerosis, 68 and other autoimmune diseases. 64,65,69Two forms of the diglyceride lipid DG 24:6;O were also strongly associated with PFHxS.1][72][73] Based on this, the interaction between the lipidome and PFASs is important to investigate in human health.

Conclusions
We here present a combined targeted and untargeted method suited for high-throughput analysis of environmental contaminants in plasma.This application of metabolomics-inspired workows for the screening of environmental contaminants has a high potential for exposome proling.Nineteen targeted environmental contaminants were detected in the study population, with ten associated with age and three with sex.In the evaluation of the untargeted data, three vitamin D 3 metabolites and two diglyceride lipids were shown to associate with PFHxS strongly.The results demonstrate the potential of combining targeted and untargeted approaches that may be important in understanding the effects of environmental contaminants on human health.

Fig. 4
Fig. 4 Fourteen unknown features were found to be associated with seven confirmed environmental contaminants.The percentage values within parentheses next to compound names indicate compound coverage.The stars (*) below correlation coefficients indicate significance: ***: p # 0.001; **: p # 0.01; *: p # 0.05.A detailed heatmap with the specific p-values is found in the ESI (Fig. S4 †).

Table 2
Concentrations for the 19 environmental contaminants detected in the 100 analyzed plasma samples from blood donors a a Bold: p < 0.05.

Table 3
Identified endogenous compounds correlating with PFHxS observed in untargeted analysis p-values for correlation coefficients.All p-values are presented in the ESI.b Identication condence levels according to Schymanski et al. 2014.
a 41 Paper Environmental Science: Processes & Impacts Open Access Article.Published on 11 May 2023.Downloaded on 9/16/2023 6:10:05 AM.This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.