Abstract

Purpose of Review

   There is a growing interest in understanding the health effects of
   exposure to per- and polyfluoroalkyl substances (PFAS) through the
   study of the human metabolome. In this systematic review, we aimed to
   identify consistent findings between PFAS and metabolomic signatures.
   We conducted a search matching specific keywords that was independently
   reviewed by two authors on two databases (EMBASE and PubMed) from their
   inception through July 19, 2022 following PRISMA guidelines.

Recent Findings

   We identified a total of 28 eligible observational studies that
   evaluated the associations between 31 different PFAS exposures and
   metabolomics in humans. The most common exposure evaluated was legacy
   long-chain PFAS. Population sample sizes ranged from 40 to 1,105
   participants at different stages across the lifespan. A total of 19
   studies used a non-targeted metabolomics approach, 7 used targeted
   approaches, and 2 included both. The majority of studies were
   cross-sectional (n = 25), including four with prospective analyses of
   PFAS measured prior to metabolomics.

Summary

   Most frequently reported associations across studies were observed
   between PFAS and amino acids, fatty acids, glycerophospholipids,
   glycerolipids, phosphosphingolipids, bile acids, ceramides, purines,
   and acylcarnitines. Corresponding metabolic pathways were also altered,
   including lipid, amino acid, carbohydrate, nucleotide, energy
   metabolism, glycan biosynthesis and metabolism, and metabolism of
   cofactors and vitamins. We found consistent evidence across studies
   indicating PFAS-induced alterations in lipid and amino acid
   metabolites, which may be involved in energy and cell membrane
   disruption.

   Keywords: PFAS, Metabolomics, Human metabolome, Endocrine-disrupting
   chemicals, Systematic review

Introduction

   Perfluoroalkyl and polyfluoroalkyl substances (PFAS) are a group of
   widespread human-made fluorinated compounds that can contribute to
   deleterious health effects and chronic diseases in humans [[52]1,
   [53]2]. Estimates from recent data derived from the National Health And
   Nutrition Examination Survey (NHANES) show that PFAS can be detected in
   nearly every sample [[54]3, [55]4], with perfluorooctanoic acid (PFOA),
   perfluorooctane sulfonate (PFOS), and perfluorohexanesulfonic acid
   (PFHxS) being present in at least 99% of the U.S. population [[56]5].
   While certain long-chain, or so-called legacy PFAS, such as PFOS and
   PFOA are being phased out in many countries [[57]6], a new generation
   of short-chain PFAS chemicals are being introduced with limited
   knowledge on their short- and long-term health effects in humans. Early
   indications in observational studies point at the persistent nature of
   these chemicals through overall long half-life, bioaccumulation, and
   slow degradation rate in the environment and in the human body
   [[58]7–[59]9]. In prior research, PFAS exposures have been linked to a
   wide array of diseases including, but not limited to, increased risk of
   cancer [[60]10, [61]11], asthma [[62]12], altered immune [[63]13],
   thyroid [[64]14, [65]15], or liver function [[66]16], kidney damage
   [[67]17], altered pregnancy outcomes [[68]18], and cardiometabolic
   disease [[69]19–[70]21]. These effects are observed in occupational and
   non-occupational settings and in both adult and child cohorts.
   Particular attention has been placed on the metabolic effects of PFAS
   in metabolic syndrome [[71]1], a complex combination of metabolic
   abnormalities including insulin resistance, dyslipidemia, glucose
   intolerance, hypertension, and obesity [[72]22]. Mechanisms underlying
   PFAS toxicity in humans are not yet fully understood, but exposure to
   PFAS has been hypothesized to play a role in pathways modulating
   insulin resistance and lipid metabolism, such as oxidative stress,
   inflammation, peroxisome proliferator-activated receptor (PPAR)
   signaling, metabolic hormones (i.e., adipokines, insulin) or sex
   steroid hormones [[73]21, [74]23–[75]27].

   Newly available emerging high-throughput technologies have made
   possible the investigation of the human metabolome as a novel way to
   understand mechanisms of health and disease [[76]28]. Metabolomics is
   the study of metabolites (or small molecules) within the body, tissues,
   and cells, at a large-scale. Metabolomic approaches have the ability to
   characterize the human exposome and are considered a promising tool to
   potentially unravel the etiology of certain diseases [[77]29–[78]31].
   Only one scoping review has been conducted on PFAS and metabolomics
   previously, including untargeted metabolomics studies only [[79]32]. To
   our knowledge, we compile for the first time a systematic review of all
   epidemiological studies that have examined PFAS exposures and
   metabolomics using both targeted and untargeted approaches. In this
   review, we summarized the human metabolite pathways associated with
   PFAS exposure, highlighting the major and common pathways identified
   across studies to gain more insight into biological responses of PFAS
   exposures in humans. We also discuss current research gaps and possible
   avenues of future research.

What Are PFAS? Definition and Types

   PFAS are a group of ubiquitously manufactured chemicals that have been
   broadly used worldwide in consumer and industrial products for their
   properties as surfactants and their coating resistance to heat, oil,
   stains, or water. PFAS are aliphatic substances made of strong
   carbon-fluoride bonds that give these compounds their slow degradation
   rate and enduring properties in the environment and within the human
   body [[80]33]. PFAS half-lives in humans reported in published
   literature to date are summarized in [81]Table S1. PFAS compounds have
   a characteristic perfluoroalkyl moiety, where for one or more carbon
   atoms the hydrogens have been replaced by fluorine atoms, as well as a
   functional group [[82]34] conferring them both hydrophobic and
   lipophobic properties [[83]35, [84]36]. Depending on the number of
   carbons, a PFAS compound is referred to as either long-chain or
   short-chain ([85]Table 1). Long-chain PFAS usually refer to any
   perfluoroalkyl carboxylic acid containing 8 carbons or greater, or any
   perfluoroalkyl sulfonic acid containing 6 carbons or greater.
   Perfluoroalkyl carboxylic acids and sulfonic acids containing less than
   8 and 6 per-fluorinated carbons, respectively, are commonly referred to
   as short-chain PFAS [[86]34, [87]37]. The longer their fluorinated
   carbon chain the more PFAS are thought to bioaccumulate, as suggested
   in previous animal studies [[88]38–[89]40], and thus are more subjected
   to monitoring [[90]41]. PFOA and PFOS are two of the most common
   long-chain PFAS studied. While regulation tends to focus more on
   long-chain PFAS [[91]42], short-chain PFAS also present a wide array of
   concerns. Short-chain PFAS can be considered toxic [[92]43] and as
   persistent as long-chain PFAS [[93]44]. They have a potential
   long-range transport in both biotic and abiotic environments compared
   to their long-chain counterparts [[94]45, [95]46]. This high mobility
   [[96]47, [97]48] means that they may reach bodies of water from which
   they are harder to remove than long-chain PFAS due to their lower
   adsorption potential [[98]46, [99]49]. A limited number of PFAS have
   been considered in biomonitoring and health studies to date. However,
   it is estimated that more than 4700 PFAS exist [[100]50–[101]52] and at
   least 3000 are readily available in products currently on the market
   [[102]53, [103]54].

Table 1.

   Summary of PFAS and their characteristics included in metabolomic
   studies in humans
   PFAS name Studied in relation to metabolomics Available descriptive
   statistics (when included in analytic dataset) Chemical formula Number
   of carbons Functional group Chain length Half-life (T[1/2])
   Perfluorotetradecanoic acid (PFTeDA/PFTDA) McGlinchey et al. 2020
   Median (IQR): 0.32 (0.1–0.75) ng/mL C[14]HF[27]O[2] 14 Carboxylate Long
   –
   Perfluorotridecanoic acid (PFTrDA) Lee et al. 2021 Median: 0.38 ng/mL
   C[13]HF[25]O[2] 13 Carboxylate Long –
   McGlinchey et al. 2020 Median (IQR): 0.11 (0.06–0.20) ng/mL
   Perfluorododecanoic acid (PFDoA/PFDoDA) Li et al. 2020 Median (IQR):
   0.00 (0.00–0.00) ng/mL^[104]a C[12]HF[23]O[2] 12 Carboxylate Long –
   McGlinchey et al. 2020 Median (IQR): 0.08 (0.06–0.12) ng/mL
   Perfluorododecane sulfonate (PFDoDS) McGlinchey et al. 2020 Median
   (IQR): 0.12 (0.11–0.2) ng/mL C[13]F[25]O[3]S^− 12 Sulfonate Long –
   Perfluoroundecanoic acid (PFUnA/PFUnDA/PFUdA) Huang et al. 2019 Mean
   (SD): 0.5 (0.5) ug/L C[11]HF[21]O[2] 11 Carboxylate Long 4.5–12 years
   [[105]242]
   Lee et al. 2021 Median: 0.53 ng/mL
   Li et al. 2020 Median (IQR): 0.00 (0.00–0.00) ng/mL^[106]a
   Li et al. 2021 In mothers:
   Mean (95% CI): 0.97 (0.76–1.17) ng/mL
   In fetus:
   Mean (95% CI): 0.70 (0.31–1.09) ng/mL
   Matta et al. 2022 OMA:
   Median (IQR): 0.13 (0.10–0.19) ng/m^[107]b
   noOMA:
   Median (IQR): 0.15 (0.10–0.19) ng/mL^[108]b
   Control:
   Median (IQR): 0.13 (0.10–0.17) ng/mL
   McGlinchey et al. 2020 Median (IQR): 0.27 (0.23–0.34) ng/mL
   Salihovic et al. 2019 Median (IQR): 0.29 (0.26–0.40) ng/mL
   Schillemans et al. 2021 T2D cases:
   Median (IQR): 0.16 (<LOQ–0.23) ng/mL ^[109]c
   Controls:
   Median (IQR): 0.18 (<LOQ–0.26) ng/mL^[110]c
   Sinisalu et al. 2020 Cord blood:
   Median (min–max): <LLQ (<LLQ–0.48)
   ng/mL^[111]d
   3 months
   Median (min–max): <LLQ (<LLQ–0.53) ng/mL^[112]d
   Stratakis et al. 2020 Median (IQR): 0.20 (0.13, 0.30) ng/mL
   You et al. 2022 Mean (SD): 6.70 (9.60) ng/mL
   Perfluorodecanoic acid (PFDA/PFDeA) Huang et al. 2019 Mean (SD): 0.3
   (0.3) ug/L C[10]HF[19]O[2] 10 Carboxylate Long 4.5–12 years [[113]242]
   Lee et al. 2021 Median: 0.69 ng/mL
   Li et al. 2020 Median (IQR): 0.00 (0.00–0.00) ng/mL^[114]a
   Li et al. 2021 In mothers:
   Mean (95% CI): 0.87 (0.70–1.04) ng/mL
   In fetus:
   Mean (95% CI): 0.37 (0.29–0.44) ng/mL
   Matta et al. 2022 OMA:
   Median (IQR): 0.21 (0.18–0.32) ng/mL^[115]b
   noOMA:
   Median (IQR): 0.22 (0.18–0.33) ng/mL^[116]b
   Control:
   Median (IQR): 0.19 (0.17–0.25) ng/mL
   McGlinchey et al. 2020 Median (IQR): 0.19 (0.14–0.25) ng/mL
   Schillemans et al. 2021 T2D cases:
   Median (IQR): 0.21 (<LOQ–0.29) ng/mL^[117]c
   Controls:
   Median (IQR): 0.23 (0.17–0.30) ng/mL^[118]c
   You et al. 2022 Mean (SD): 1.70 (2.60) ng/mL
   Pefluorodecane sulfonic acid (PFDS) McGlinchey et al. 2020 Median
   (IQR): 0.06 (0.05–0.08) ng/mL C[10]HF[21]O[3]S 10 Sulfonate Long 6.6
   years [[119]243]
   Perfluorononanoic acid (PFNA) Chang et al. 2022 Median (min-max): 0.27
   (<LOD–2.27) ng/mL C[9]HF[17]O[2] 9 Carboxylate Long 2.5–4.3 years
   [[120]242]
   Huang et al. 2019 Mean (SD): 1.0 (0.5) ug/L
   Kingsley et al. 2019 Mean (SD): 0.90 (0.70) ng/mL
   Lee et al. 2021 Median: 1.33 ng/mL
   Li et al. 2020 Median (IQR): 0.00 (0.00–0.00) ng/mL^[121]a
   Li et al. 2021 In mothers:
   Mean (95% CI): 0.90 (0.71–1.08) ng/mL
   In fetus:
   Mean (95% CI): 0.73 (0.28–1.18) ng/mL
   Maitre et al. 2018 Primary cohort:
   Median (IQR): 0.77 (0.56–1.05) ng/g lipid^[122]e
   Replication cohort:
   Median (IQR): 0.60 (0.46–0.78) ng/g lipid^[123]e
   Matta et al. 2022 OMA:
   Median (IQR): 0.49 (0.41–0.65) ng/mL^[124]b
   noOMA:
   Median (IQR): 0.48 (0.37–0.67) ng/mL^[125]b
   Control:
   Median (IQR): 0.46 (0.31–0.52) ng/mL
   McGlinchey et al. 2020 Median (IQR): 0.39 (0.28–0.54) ng/mL
   Mitro et al. 2021 Median (IQR): 0.60 (0.40–0.80) ng/mL
   Salihovic et al. 2019 Median (IQR): 0.71 (0.53–0.97) ng/mL
   Schillemans et al. 2021 T2D cases:
   Median (IQR): 0.55 (0.40–0.76) ng/mL^[126]c
   Controls:
   Median (IQR): 0.53 (0.42–0.78) ng/mL^[127]c
   Sen et al. 2022 Median (min–max): 0.37 (0.09–1.08) ng/mL
   Sinisalu et al. 2021 ^[128]f Median (min–max): 0.84 (LOD–O.31) ng/mL
   Stratakis et al. 2020 Median (IQR): 0.72 (0.47–1.11) ng/mL
   You et al. 2022 Mean (SD): 2.1 (3.1) ng/mL
   Yu et al. 2022 Mean (SD): 1.02 (0.89) ng/mL
   Perfluorononane sulfonate (PFNS) McGlinchey et al. 2020 Median (IQR):
   0.03 (0.00–0.06) ng/mL C[9]F[19]O[3]S^− 9 Sulfonate Long –
   Perfluorooctanoic acid (PFOA)^[129]g Alderete et al. 2019 Geometric
   mean (SD): 2.78 (1.29) ng/mL C[8]HF[15]O[2] 8 Carboxylate Long 1.8–22
   years [130]9,[131]242–[132]251]
   Chang et al. 2022 Median (min–max): 0.72 (<LOD–4.42) ng/mL
   Chen et al. 2020 Geometric mean (95% CI): 2.26 (1.61–3.18) ug/L
   Huang et al. 2019 Mean (SD): 2.5 (2.6) ug/L
   Ji et al. 2021 Median (IQR): 39.6(27.5–48.9) ng/g creatinine^[133]h
   Jin et al. 2020 Median (IQR): 3.42 (1.65) ng/mL ^[134]i
   Kingsley et al. 2019 Mean (SD): 2.6 (1.0) ng/mL
   Lee et al. 2021 Median: 2.99 ng/mL
   Li et al. 2020 Median (IQR): 0.40 (0.25–0.60) ng/mL ^[135]a
   Li et al. 2021 In mothers:
   Mean (95% CI): 2.66 (2.03–3.29) ng/mL
   In fetus:
   Mean (95% CI): 3.84 (2.38–5.29) ng/mL
   Lu et al. 2019 Median (min–max): 164.6 (2.00–7214) ng/mL ^[136]j
   Maitre et al. 2018 Primary cohort:
   Median (IQR): 2.68 (1.69–3.67) ng/g lipid ^[137]e
   Replication cohort:
   Median (IQR): 1.66 (1.28–2.32) ng/g lipid ^[138]e
   Matta et al. 2022 OMA:
   Median (IQR): 1.22 (0.81–1.58) ng/mL ^[139]b
   noOMA:
   Median (IQR): 1.21(0.81–1.58) ng/mL ^[140]b
   Control:
   Median (IQR): 1.10 (0.77–1.69) ng/mL
   McGlinchey et al. 2020 Median (IQR): 1.02 (0.68–1.41) ng/mL
   Mitro et al. 2021 Median (IQR): 5.0 (3.6–6.8) ng/mL
   Salihovic et al. 2019 Median (IQR): 3.33 (2.55–4.39) ng/mL
   Schillemans et al. 2021 T2D cases:
   Median (IQR): 2.8 (2.15–3.6) ng/mL ^[141]c
   Controls:
   Median (IQR): 3.0 (2.3–4.2) ng/mL ^[142]c
   Sen et al. 2022 Median (min–max): 1.89 (0.49–6.36) ng/mL
   Sinisalu et al. 2020 Cord blood:
   Median (min–max): 2.32 (1.31–4.80) ng/
   mL ^[143]d
   3 months:
   Median (min–max): 4.34 (1.23–9.17) ng/mL ^[144]d
   Sinisalu et al. 2021 ^[145]f Median (min–max): 0.66 (0.36–3.6) ng/mL
   Stratakis et al. 2020 Median (IQR): 2.38 (1.45, 3.45) ng/mL
   Wang et al. 2017 Median (IQR): 7.56 (6.09–10.7) nM
   You et al. 2022 Mean (SD): 20.1 (24.0) ng/mL
   Yu et al. 2022 n-PFOA:
   Mean (SD): 2.07 (0.85) ng/mL
   Perfluorooctane sulfonic acid (PFOS)^[146]k Alderete et al. 2019
   Geometric mean (SD): 12.22 (1.91) ng/mL C[8]HF[17]O[3]S 8 Sulfonate
   Long 2.9–8.2 years [[147]9, [148]89, [149]242, [150]243, [151]246,
   [152]247, [153]249–[154]251]
   Chang et al. 2022 Median (min–max): 2.10 (<LOD–12.40) ng/mL
   Chen et al. 2020 Geometric mean (95% CI): 4.29 (1.61–11.5) ug/L
   Hu et al. 2019 Median (IQR): 33.9 (16.1–61.0) ng/mL
   Huang et al. 2019 Mean (SD): 6.5 (4.6) ug/L
   Ji et al. 2021 Median (IQR): 67.6 (41.0–96.5) ng/g creatinine ^[155]h
   Jin et al. 2020 Median (IQR): 5.59 (4.46) ng/mL^[156]i
   Kingsley et al. 2019 Mean (SD): 4.4 (3.2) ng/mL
   Lee et al. 2021 Median: 3.79 ng/mL
   Li et al. 2020 Median (IQR): 33.9 (16.1–61.0) ng/mL^[157]a
   Li et al. 2021 In mothers:
   Mean (95% CI): 5.36 (4.59–6.14) ng/mL
   In fetus:
   Mean (95% CI): 2.53 (2.21–2.85) ng/mL
   Lu et al. 2019 Median (min–max): 909.3 (9.60–43,299) ng/mL^[158]j
   Maitre et al. 2018 Primary cohort:
   Median (IQR): 6 (3.94–8.15) ng/g lipid^[159]e
   Replication cohort:
   Median (IQR): 5.3 (4.15–7.18) ng/g lipid^[160]e
   Matta et al. 2022 OMA:
   Median (IQR): 2.45 (1.65–3.44) ng/mL^[161]b
   noOMA:
   Median (IQR): 2.09 (1.56–3.38) ng/mL^[162]b
   Control:
   Median (IQR): 1.87 (1.24–2.32) ng/mL
   McGlinchey et al. 2020 Median (IQR): 1.38 (0.97–1.86) ng/mL
   Mitro et al. 2021 Median (IQR): 26.6 (17.3–40.3) ng/mL
   Salihovic et al. 2019 Median (IQR): 13.35 (10.13–17.79) ng/mL
   Schillemans et al. 2021 T2D cases:
   Median (IQR): 19 (15–25) ng/mL^[163]c
   Controls:
   Median (IQR): 20 (16–27) ng/mL^[164]c
   Sen et al. 2022 Br-PFOS:
   Median (min–max): 2.13 (0.63–9.71) ng/mL
   L-PFOS:
   Median (min–max): 2.50 (0.74–11.8) ng/mL
   Sinisalu et al. 2020 Cord blood:
   Median (min–max): 2.21 (0.27–8.17) ng/mL^[165]d
   3 months:
   Median (min–max): 2.93 (0.27–7.66) ng/mL^[166]d
   Sinisalu et al. 2021 ^[167]f Median (min–max): 0.45 (LOQ–1.80) ng/mL
   Stratakis et al. 2020 Median (IQR): 6.74 (4.43, 10.4) ng/mL
   Wang et al. 2017 Median (IQR): 12.8 (10.5–15.6) nM
   You et al. 2022 Mean (SD): 9.1 (9.0) ng/mL
   Yu et al. 2022 n-PFOS:
   Mean (SD): 2 (1.31) ng/mL
   Sm-PFOS:
   Mean (SD): 0.77 (0.41) ng/mL
   Perfluorooctanesulfonamide (PFOSA) Li et al. 2020 PFOSA
   Median (IQR): 0.00 (0.00–0.04) ng/mL^[168]a
   MePFOSAAcOH
   Median (IQR): 0.00 (0.00–0.00) ng/mL^[169]a
   EtPFOSAAcOH
   Median (IQR): 0.28 (0.12–0.53) ng/mL^[170]a C[8]H[2]F[17]NO[2]S 8
   Sulfonamide Long 1.7 years [[171]251]
   McGlinchey et al. 2020 Median (IQR): 0.01 (0.002–0.01) ng/mL
   Chlorinated polyfluorinated ether sulfonic acids (Cl-PFESAs)^[172]a Li
   et al. 2021 6:2 Cl-PFESA
   In mothers:
   Mean (95% CI): 2.58 (2.24–2.92) ng/mL
   In fetus:
   Mean (95% CI): 1.16 (1.00–1.31) ng/mL
   8:2 Cl-PFESA
   In mothers:
   Mean (95% CI): 0.25 (0.21–0.30) ng/mL
   In fetus:
   Mean (95% CI): 0.21 (0.17–0.25) ng/mL – 8 Sulfonate Long 15.3 years
   [[173]252]
   Lu et al. 2019 Median (min–max): 8.90 (<LOD–43.4) ng/mL ^[174]j
   Sinisalu et al. 2021 ^[175]f Median (min–max): 0.19 (0.07–1.82) ng/mL
   Perfluoroheptane sulfonic acid (PFHpS) Li et al. 2021 In mothers:
   Mean (95% CI): 0.22 (0.19–0.25) ng/mL
   In fetus:
   Mean (95% CI): 0.16 (0.14–0.18) ng/mL C[7]HF[15]O[3]S 7 Sulfonate Long
   1.54.7 years [[176]9, [177]247]
   Matta et al. 2022 OMA:
   Median (IQR): 0.05 (0.05–0.12) ng/mL^[178]b
   noOMA:
   Median (IQR): 0.05 (0.05–0.13) ng/mL^[179]b
   Control:
   Median (IQR): 0.05 (0.05–0.07) ng/mL
   McGlinchey et al. 2020 Median (IQR): 0.06 (0.05–0.07) ng/mL
   Perfluorooctane sulfonamide acetic acids (EtFOSAA, MeFOSAA) ^[180]m Hu
   et al. 2019 EtFOSAA-AcOH:
   Median (IQR): 0.28 (0.12–0.53) ng/mL – – Sulfonamide Long –
   Mitro et al. 2021 EtFOSAA:
   Median (IQR): 1.2 (0.6–2.1) ng/mL
   MeFOSAA:
   Median (IQR): 1.0 (0.5–1.7) ng/mL
   Perfluoroheptanoic acid (PFHpA) Lee et al. 2021 Median: 0.30 ng/mL
   C[7]HF[13]O[2] 7 Carboxylate Short 62 days–1.5 years [[181]9, [182]242]
   Li et. al 2020 Median (IQR): 0.00 (0.00–0.00) ng/mL^[183]a
   McGlinchey et al. 2020 Median (IQR): 0.12 (0.04–0.19) ng/mL
   Salihovic et al. 2019 Median (IQR): 0.07 (0.05–0.11) ng/mL
   Sinisalu et al. 2020 Cord blood:
   Median (min–max): <LLQ (<LLQ–0.5) ng/mL
   3 months
   Median (min–max): <LLQ (<LLQ–1.09) ng/mL
   Potassium perfluoro-4-ethyl-cyclohexanesulfonate (PFECHS) McGlinchey et
   al. 2020 Median (IQR): 0.06 (0.06–0.06) ng/mL – – Sulfonate – –
   Perfluorohexane sulfonic acid (PFHxS) Alderete et al. 2019 Geometric
   mean (SD): 1.65 (2) ng/mL C[6]HF[13]O[3]S 6 Sulfonate Long 2.9–35 years
   [[184]9, [185]242, [186]246, [187]247, [188]250, [189]251]
   Chang et al. 2022 Median (min–max): 1.09 (<LOD–4.80) ng/mL
   Chen et al. 2020 Geometric mean (95% CI): 1.37 (0.32–5.79) ug/L
   Huang et al. 2019 Mean (SD): 0.7 (0.9) ug/L
   Jin et al. 2020 Median (IQR): 1.53 (3.17) ng/mL ^[190]i
   Kingsley et al. 2019 Mean (SD): 2.1 (2.7) ng/mL
   Lee et al. 2021 Median: 0.93 ng/mL
   Li et al. 2020 Median (IQR): 2.29 (1.02–3.68) ng/mL^[191]a
   Li et al. 2021 In mothers:
   Mean (95% CI): 1.42(1.24–1.61) ng/mL
   In fetus:
   Mean (95% CI): 1.46 (1.19–1.73) ng/mL
   Lu et al. 2019 Median (min–max): 785.2 (<LOD–1,226) ng/mL^[192]j
   Maitre et al. 2018 Primary cohort:
   Median (IQR): 0.87 (0.69–1.14) ng/g lipid^[193]e
   Replication cohort:
   Median (IQR): 0.44 (0.34–0.57) ng/g lipid^[194]e
   McGlinchey et al. 2020 Median (IQR): 0.33 (0.27–0.43) ng/mL
   Mitro et al. 2021 Median (IQR): 2.3 (1.50–3.8) ng/mL
   Salihovic et al. 2019 Median (IQR): 2.08 (1.61–3.45) ng/mL
   Schillemans et al. 2021 T2D cases:
   Median (IQR): 0.99 (0.69–1.40) ng/mL^[195]c
   Controls:
   Median (IQR): 1.10 (0.76–1.40) ng/mL^[196]c
   Sen et al. 2022 Median (min–max): 0.60 (0.16–10.6) ng/mL
   Sinisalu et al. 2020 Cord blood:
   Median (min–max): 0.55 (0.31–1.03) ng/mL
   3 months:
   Median (min–max): 0.70 (0.31–1.62) ng/mL
   Sinisalu et al. 2021^[197]f Median (min–max): 0.09 (LOQ–2.1) ng/mL
   Stratakis et al. 2020 Median (IQR): 0.59 (0.34, 0.93) ng/mL
   Yu et al. 2022 Mean (SD): 0.91 (1.32) ng/mL
   Perfluoropentanoic acid (PFPeA) McGlinchey et al. 2020 Median (IQR):
   0.15 (0.13–0.21) ng/mL C[5]HF[9]O[2] 5 Carboxylate Short –
   Perfluoropentane sulfonic acid (PFPeS) McGlinchey et al. 2020 Median
   (IQR): 0.04 (0.04–0.06) ng/mL C[5]HF[11]O[3]S 5 Sulfonate Short 0.6–1
   year [[198]9, [199]247]
   Perfluorobutane sulfonic acid (PFBS) Lu et al. 2019 Median (min–max):
   76.4 (0.80–2449) ng/mL ^[200]j C[4]HF[9]O[3]S 4 Sulfonate Short 26–44
   days [[201]9, [202]253]
   McGlinchey et al. 2020 Median (IQR): 0.0059 (0.003–0.064) ng/mL
   Perfluorobutyric acid or perfluorobutanoic acid (PFBA) Lu et al. 2019
   McGlinchey et al. 2020 Median (min–max): 17.4 (<LOD–189.6) ng/mL^[203]j
   Median (IQR): 0.37 (0.29–0.43) ng/mL C[4]HF[7]O[2] 4 Carboxylate Short
   74 days [[204]254]
   [205]Open in a new tab
   ^a

   Descriptive data reported in Hu et al. in the Child Health and
   Development Studies (CHDS) cohort [[206]112]
   ^b

   In patients with endometriosis with endometrioma (OMA) and in patients
   with endometriosis and no endometrioma (noOMA)
   ^c

   Descriptive data reported in Donat-Vargas et al. in type 2 diabetes
   cases and controls from the Västerbotten Intervention Programme (VIP)
   cohort [[207]255]
   ^d

   In patients with celiac disease
   ^e

   In blood from pregnant women
   ^f

   Study reporting 14 chemicals evaluated but for only 5 chemicals
   descriptive information was provided
   ^g

   Includes n-PFOA and Sb-PFOA
   ^h

   In urine from COVID-19 patients
   ^i

   In children with NAFLD
   ^j

   In occupational workers
   ^k

   Includes n-PFOS, Br-PFOS, L-PFOS, and Sm-PFOS
   ^l

   Includes 6:2 Cl-PFESA and 8:2 Cl-PFESA
   ^m

   Includes EtPFOSAAcOH

Sources of Human Exposure to PFAS

   PFAS have been in use in industry and consumer products since their
   introduction in the 1940s [[208]55]. Common PFAS sources range from
   site-specific and occupational exposures (military bases, airport
   sites, fire-fighting foams) to everyday consumer products, such as
   cookware, food packaging, impermeable gear, furniture, carpeting,
   coatings and paint, or cosmetics, among others [[209]56–[210]58]. The
   most prevalent PFAS exposure routes in humans are via dietary sources
   or contaminated water (oral route), followed by inhalation of dust or
   air particles, and dermal absorption [[211]59, [212]60]. To date, PFAS
   have been detected in rivers, rain water, soil, and ambient air of
   major cities around the world, as well as in remote areas [[213]61].
   There are no current adequate safety limits to PFOS or other PFAS in
   drinking water in most countries, and there are no proven safe levels
   of PFAS at lower dosages. For instance, even low doses of PFAS have
   been linked to adverse health effects in humans, such as decreased
   antibody response [[214]62–[215]64]. Additional routes of exposure also
   include early life exposure routes via the placenta and breastfeeding
   [[216]65–[217]67]. Prenatal and early-life PFAS exposures are
   particularly important, yet understudied, as they may contribute
   significantly to programming of later chronic health outcomes in
   adulthood [[218]68].

PFAS Transport and Clearance in Human Body

   Given the persistent nature of PFAS, with an average half-life ranging
   from a few days or months to several decades ([219]Table S1),
   bioaccumulation in humans has raised concerns. PFAS are structurally
   similar to fatty acids, but unlike other persistent organic pollutants,
   such as organochlorine and brominated compounds, PFAS compounds do not
   tend to accumulate in lipid tissue. Instead, due to their polar
   hydrophobic fluorine content, PFAS have a higher affinity for proteins
   [[220]69–[221]71]. After ingestion, PFAS tend to concentrate in high
   protein density tissues, such as compounds in blood, binding to serum
   albumin [[222]72]. Protein-rich tissues, such as the liver and blood,
   are major repositories of perfluorinated acids [[223]73, [224]74].
   Interestingly, an intervention study showed a decrease in serum and
   plasma PFAS levels of Australian firefighters after blood donation
   [[225]75]. Not surprisingly, PFAS are also commonly detected in human
   breast milk [[226]76], which has high antimicrobial and digestive
   enzymatic activity [[227]77, [228]78]. In addition to primarily
   accumulating after adsorption in media with high-protein content, such
   as liver and serum, PFAS can also accumulate in other organs, including
   the lungs, bone, brain, and kidney [[229]79, [230]80]. While the
   primary route of PFAS clearance is through the kidney in animal studies
   [[231]72, [232]81], renal urine excretion in humans has been modeled to
   be low for these compounds [[233]82] highlighting the persistent nature
   and difficult elimination of these compounds, particularly long-chain
   PFAS. Despite PFAS secretion via urine, an albumin-based reabsorption
   mechanism in the kidneys contributes to low elimination rates
   [[234]83]. Sex steroid hormones have also been observed to take part in
   facilitating renal clearance in animals [[235]84]. Additionally, fecal
   excretion has been suggested as a potential route of PFAS elimination
   [[236]85, [237]86]. In women, lactation, parity, and menstruation are
   other alternative excretion routes that may provide an advantage,
   compared to males, in terms of more rapid PFAS secretion from the body
   [[238]87–[239]90]. This is consistent with the higher levels of some
   PFAS found in males compared to women [[240]3], yet this sex difference
   in clearance may narrow as a function of age and menopausal status
   [[241]91, [242]92].

Characterization of Exposure–Disease Mechanisms via Metabolomics

   The metabolome refers to the collection of all the molecules (organic
   or inorganic) playing a role in physiological processes in the body or
   present in cells and tissues. The human metabolome can include
   metabolites involved in endogenous cellular processes, as well as
   metabolites external to the body such as dietary factors, xenobiotics,
   and environmental contaminants. Analytical methods using
   high-throughput technologies have emerged as novel tools to examine
   comprehensively both exogenous chemicals as well as endogenous
   metabolites in the body at a granular level. Targeted methods have been
   developed for a priori preselected metabolites while untargeted methods
   holistically and systematically analyze a wide array of metabolites in
   an organism by maximizing the number of metabolite features detected.
   Numerous laboratory methods, such as nuclear magnetic resonance (NMR)
   and liquid chromatography-mass spectrometry (LC–MS), enable the
   detection of metabolites. Recent advances, particularly in
   high-resolution metabolomics (HRM), such as ultra-high-resolution mass
   spectrometry (UHRMS), have allowed for an even more comprehensive
   characterization of exogenous and endogenous metabolites present at
   lower concentrations in the body [[243]28, [244]93]. The use of
   metabolomics as a tool to better understand the mechanisms associated
   with disease is still at its early stages. So far, metabolomics has
   facilitated the assessment of the effects from PFAS exposure in humans
   in recent studies, indicating putative intermediary markers for the
   initiation and progression of various diseases, such as cardiometabolic
   disease [[245]29], diabetes [[246]30], and cancer [[247]31]. For
   instance, candidate metabolites like glucose or creatine have been
   hypothesized as biomarkers for diagnosis or prognosis of diseases
   [[248]94] as metabolomics data encompass the downstream phenotypic
   product after gene transcription and translation [[249]95], as well as
   suggest possible interaction effects or cross-talk between the exposome
   and endogenous biological systems.

Methods

   We conducted a systematic review following PRISMA guidelines [[250]96]
   and selected studies that investigated human PFAS exposures (either
   studied as single compounds or as a mixture) in relation to metabolic
   profiling using high-throughput metabolomics technologies. Inclusion
   criteria were detailed in advance and registered in PROSPERO (ID:
   327,196; access registration via
   [251]https://www.crd.york.ac.uk/PROSPERO/display_record.php?RecordID=32
   7196).

Search Strategy

   We searched research articles found on EMBASE and PubMed databases from
   their inception through July 19, 2022. Two reviewers independently
   performed a study selection of eligible epidemiological research
   articles. Our search was based on matching words contained in the
   title, abstract, or as keywords including “PFAS” or “perfluoroalkyl” or
   “polyfluoroalkyl,” and “metabolomics” or “metabolome” or “metabolic
   profiles,” among others. A full list of terms and search strategy is
   provided in the [252]Supplementary Material ([253]Supplementary
   Material, [254]Methods) as well as in PROSPERO.

Study Selection

   Eligible studies included original epidemiological research articles
   reported as full-text articles in English that predominantly focused on
   at least one PFAS exposure and with metabolomics data (either as a
   primary or secondary outcome, or either as targeted or untargeted).
   Lipidomics studies were also included under the targeted study
   category. All research studies made use of high-throughput approaches
   for metabolite profiling. Research articles that focused solely on
   metabolic biomarkers with no metabolomic analytic approach used were
   excluded from this review. Studies with metabolomics data were eligible
   from any human biofluid sample as long as the study was
   epidemiological; animal studies, as well as experimental studies with
   human samples conducted in vitro or in vivo were ineligible
   ([255]Supplementary Material, [256]Methods).

Data Extraction from Eligible Studies

   Data extraction of eligible studies included first-author, year of
   publication, study design, sample size, participant characteristics and
   location, years of follow-up (if applicable), number and nomenclature
   of PFAS examined, window of exposure (and year), number of metabolites
   measured (outcome), metabolomics analytical method (including reported
   level of confidence in metabolites [[257]97]), data pre-processing,
   statistical methods (including false discovery rate correction),
   chemical databases used for identification or classification of the
   metabolites and pathways, summary of findings (analyzed metabolites and
   pathways), biological significance, adjusted confounders, effect
   modifiers, limitations, and additional findings. When possible, we
   classified each PFAS-associated metabolite using the criteria denoted
   by the human metabolome database (HMDB) ([258]https://hmdb.ca/) and
   pathways using the Kyoto Encyclopedia of Genes and Genomes (KEGG)
   database ([259]https://www.genome.jp/kegg/pathway.html). When a
   particular metabolite identifier was not found in the HMDB database,
   the Chemical Entities of Biological Interest (ChEBI)
   ([260]https://www.ebi.ac.uk/chebi/) and LipidMaps
   ([261]https://www.lipidmaps.org/) were used instead.

Quality Assessment

   We evaluated the quality of each study based on their limitations and
   potential biases. We created a quality score following 5-point criteria
   based on epidemiological [[262]98] and metabolomics guidelines
   [[263]99, [264]100] where points were assigned for each of the
   following: sample size (≥ 100 vs < 100 participants), adjustment for
   confounders (adjusted vs. none), false discovery rate (FDR-adjustment
   vs. no correction), study design (prospective vs. not prospective), and
   reporting of confidence level 1 metabolites identified according to the
   Metabolomics Standards Initiative (MSI). A higher score indicated a
   higher quality of evidence.

Data Synthesis

   We summarized the evidence for reported PFAS associations with
   individual metabolites or dysregulated metabolic pathways to identify
   consistent findings and research gaps. In this review, we refer to
   consistent findings as those metabolites, metabolite classes, or
   pathways reported most frequently across studies (Figs. 2–6, [265]Table
   S5), but we also report consistent directionalities based on the total
   number of metabolite associations independent of study number
   ([266]Figs. S1–[267]S3, [268]Table S3–[269]S4); we either presented the
   number of studies reporting specific PFAS-metabolite associations or
   the total number of reported PFAS-metabolite associations in all
   studies. We classified each reported metabolite by HMDB superclass,
   class, and subclass categories. We also specified the directionality of
   the association reported, as well as the PFAS exposure (either a
   combination of several PFAS mixtures or the individual PFAS studied)
   and metabolomics approach followed (either targeted or untargeted).
   Pathway analyses across studies were also summarized by KEGG modules
   and submodules. In our summary, we included results from correlation
   analyses, analyses adjusting for covariates, and analyses implementing
   FDR-correction. If a study presented more than one type of analysis, we
   extracted the results that were considered more rigorous (FDR-corrected
   and/or covariate-adjusted). In order to complement the summary of
   pathways reported across studies, we extracted each reported metabolite
   by study and conducted an independent pathway analysis using a
   systematic method with Reactome ([270]https://reactome.org) [[271]101].
   We extracted each identified metabolite name reported in the articles
   and converted them to KEGG IDs. In the case of lipids, when unsure, we
   searched on LipidMaps, and if a m/z ratio was provided, we matched it
   with what was reported by the study. Studies were included in Reactome
   analyses if they reported any PFAS-metabolite association with an
   available metabolite KEGG ID recognizable by Reactome (n = 26). We then
   used the KEGG IDs for each compound and uploaded them in Reactome to
   evaluate the pathways inferred from positive associations, negative
   associations, and all associations for each study separately. Lastly,
   we calculated combined p-values for top pathways across studies
   [[272]240, [273]241]. Quantitative summaries for descriptive statistics
   and visualizations for main findings were implemented in R (version
   4.1.2).

Results

Description of Studies included in the Review

   Out of 358 records retrieved from EMBASE (n = 207) and PubMed (n =
   151), a total of 28 unique records for original studies evaluating PFAS
   exposures and human metabolomics that were observational in design and
   in English language were eligible for this systematic review ([274]Fig.
   1). Effects of 31 different PFAS exposures (including isomers) have
   been evaluated in relation to the human metabolome across studies, with
   the most common exposure being legacy long-chain PFAS, such as PFOA,
   PFOS, and PFHxS ([275]Table 1). We accounted for PFAS that are included
   in the final analytic dataset across studies ([276]Table S2) and
   described their reported distributions in [277]Table 1. While the
   majority of studies examined single and mixed exposures to PFAS, there
   are several studies that examined other pollutants as well [[278]31,
   [279]102–[280]110]. Only one study [[281]109] examined PFAS-metabolite
   associations adjusting for other pollutants in partial correlation
   analyses and another two studies (Li et al. [[282]31] and Matta et al.
   [[283]104]) implemented multi-pollutant models (PCBs in combination
   with PFAS). Most PFAS were measured in samples collected during the
   2000s and only four studies had samples measured prior to 1999
   [[284]31, [285]111–[286]113]. The majority of studies measured PFAS
   exposures during adulthood (n = 15), followed by prenatally or
   perinatally (n = 9), and/or in childhood or adolescence (n = 7)
   ([287]Tables 2 and [288]3, [289]Table S2). Sample sizes across eligible
   studies also ranged widely from 40 to 1,105 participants recruited
   across three different continents as follows: America (n = 11 in the
   USA), Asia (n = 9), and Europe (n = 2 in Sweden, n = 2 in Finland, n =
   1 in France, n = 1 in Spain, and n = 2 across Europe).

Fig. 1.

   Fig. 1
   [290]Open in a new tab

   PRISMA 2020 flow diagram for PFAS exposures and metabolomics in humans
   summarizing eligible studies through July 19, 2022

Table 2.

   Summary of studies examining PFAS and untargeted metabolomics in
   humans^[291]a
   Author/year Study population Study design Window of PFAS exposure
   (calendar year) PFAS included in analytic dataset Metabolomics
   analytical method Main findings (metabolites) Main findings (inferred
   pathways) Multiple testing correction Covariates
   Maitre et al. 2018 [[292]102] N = 750. Birth cohort (INMA, Spanish
   cohort, from two locations) Cross-sectional and longitudinal
   associations (prospective) Prenatal (2004–2008) Multiple exposures: 35
   chemicals quantified in first-trimester blood samples (organochlorine
   pesticides, PCBs, PFAS-PFHxS, PFNA, PFOA, PFOS), in cord blood
   (mercury), and twice in urine at 12 and 32 weeks of pregnancy (metals,
   phthalates, bisphenol A) 1H nuclear magnetic resonance (NMR) PFHxS was
   correlated with ↓pregnanolone-3G, ↓acetone, and ↑succinate; PFOA with
   ↑3-hydroxybutyrate/3-aminoisobutyrate, ↓alanine, ↓glycine; but
   significantly in the population of 1 location only (stratified
   analysis) and with respect to trimester 3. PFAS did not show consistent
   associations with the urine metabolome during pregnancy NA FDR-adjusted
   to 5%: (q-value Confounders: Time of the day of sampling (nonpersistent
   analysis only), gestational week, age, and BMI (endogenous and
   exogenous factors affecting metabolism including diet and smoking)
   McGlinchey et al. 2020 [[293]30] N = 264 (dyads). Mother-children pairs
   (EDIA cohort), in Finland Prospective Prenatal (2013–2014) Detected
   PFAS: PFBA, PFBS, PFDA, PFDoDA, PFDoDS, PFDS, PFECHS, PFHpA, PFHpS,
   PFHxS, PFNA, PFNS, PFOA, L-PFOS, PFOSA, PFPeA, PFPeS, PFTDA, PFTrDA,
   PFUnDA UHPLC-QTOFMS High prenatal PFAS exposure associates with
   decreased postnatal serum ↓phospholipids. In mothers: total PFAS, as
   well as several individual PFAS levels were positively associated with
   maternal polar metabolite cluster (amino acids, saturated free fatty
   acids, and cholesterol). In cord serum: PFOS, PFOA, and total PFAS
   exposure inversely associated with ↓lipids, particularly for clusters
   CLC2 (sphingomyelins (SMs), abundant phosphatidylcholines (PCs);
   SM(38:1), SM(42:2), PC(34:2), PC(36:4)), CLC3 (lysophosphatidylcholines
   (LPCs); LPC(18:0), LPC(18:1), LPC(20:4)), and CLC4 (PUFA-containing
   phosphatidylcholines PCs; PC(36:5), PC(40:7), PE(38:6)); PFOS and PFDA
   exposure was associated with ↑CMC4 cluster (amino acids; serine,
   methionine, aspartic acid, phenylalanine). In univariate analyses, 39
   molecular lipids, mainly decreased ↓LPCs, ↓SMs and ↓PCs and 10 polar
   metabolites (increased ↑amino acids), differed between higher and lower
   quartiles of total maternal PFAS exposure. Regression analysis showed
   that PFNA, PFOS, and PFDA were the top predictors of cord serum LPC
   (20:4) (from CLC3) while PFNA, PFPeA, and PFOA were the linear
   predictors of SM (d38:1). PFHxS, followed by PFDA, PFOS, and PFNA,
   identified as a top predictors of cord serum methionine. Interaction
   effect between HLA risk and PFAS exposure on lipids NA NA (not
   for-omics) Age and BMI (in linear regression analyses only). Effect
   modification by HLA gene
   Chang et al. 2022 [[294]119] N = 313. Pregnant African-American women
   at 8–14 weeks of gestation in Atlanta, GA Cross-sectional Prenatal
   (2014–18) PFHxS, PFOS, PFOA, PFNA in serum LC-HRMS. HILIC with (+) ESI
   and reverse phase (C18) chromatography with (−) ESI Unique significant
   features associated with PFAS: 5000. 10 overlapping metabolites
   associated with both PFAS and fetal growth endpoints: level 1
   confidence- PFNA with ↑glycine, ↑taurine, ↑uric acid; PFOA with ↑uric
   acid, and ↓ferulic acid; level 2 confidence-PFOS and PFHxS with
   ↑2-hexyl-3-phenyl-2-propenal; unsaturated fatty acid C18:1 (PFOA and
   PFNA with↓elaidic acid, oleic acid, or vaccenic acid); PFNA with
   ↑androgenic hormone conjugates (PFNA and ↑DHEA-S, ↑testosterone
   sulfate, or androsterone sulfate); ↑parent bile acid (PFNA and PFOA
   with ↑CDCA, DCA, HDCA, or isoursodeoxycholic acid); and ↓bile
   acid-glycine conjugate (PFOS and PFOA with ↓chenodeoxycholylglycine,
   deoxycholylglycine, or ursodeoxycholylglycine) 21 pathways identified
   related to amino acid, lipid and fatty acid, bile acid, uric acid, and
   androgenic hormone metabolism. ≥2 PFAS/birth weight/SGA birth:
   linoleate metabolism, arginine and proline metabolism, histidine
   metabolism, nitrogen metabolism, alanine and aspartate metabolism,
   pyrimidine metabolism, tryptophan metabolism, and vitamin B3
   metabolism. ≥2 PFAS and birth weight: de novo fatty acid biosynthesis,
   fatty acid activation, purine metabolism, and vitamin D3
   (cholecalciferol) metabolism. ≥2 PFAS and SGA: glutamate metabolism,
   lysine metabolism, methionine and cysteine metabolism, aspartate and
   asparagine metabolism, glycan pathways (keratan sulfate degradation,
   glycosphingolipid metabolism, and glycosphingolipid
   biosynthesis-ganglioseries), glycerophospholipid metabolism, and
   butanoate metabolism FDR-adjusted using Benjamini–Hochberg method
   Confounders in linear regression: maternal age, education, parity, BMI,
   tobacco use, marijuana use, and infan’s sex
   Sinisalu et al. 2021 [[295]124] N = 104. Chinese infants
   Cross-sectional Prenatal (2018) Total PFAS and individual PFAS; 16 PFAS
   (14 detected) in cord plasma including 6:2 Cl-PFESA, PFOS, PFHxS, PFNA,
   PFOA (detected in > 50% for multivariable analyses) UHPLC-qTOF/MS, dual
   ESI in negative mode In adjusted analyses, PFAS was associated with
   ↑bile acids (HCA, GHCA, GCDCA, TaMCA), ↑TGs, ↑↓PCs,
   ↑lysophosphatidylethanolamines (total PFAS and lysoPE 18:1),
   ↑ceramides, ↑↓sphingomyelins, ↑C18:2, ↑C16:1, ↑palmitic acid, ↑oleic
   acid, ↓lysophosphatidylcholines, and ↑phosphatidylserine. Overall,
   PFOS, PFOA, PFNA, and PFHxS associated with ↑TG with saturated fatty
   acids and ↓phospholipids (i.e., LPC, PC and lysoPE) NA NA maternal age,
   delivery type and birth weight
   Abrahamsson et al. 2021 [[296]110] N = 590. Matched maternal and cord
   blood samples (total 295 pairs) enrolled at UCSF, in California
   Cross-sectional Prenatal (2014–2017) Exogenous chemicals (including
   PFAS-detected PFHxS, PFOS, PFDA, PFUnA, PFNA) LC-QTOF/MS in ESI+and
   ESI−and in soft ionization (MS) and fragmentation (MS/MS) modes In
   mothers: PFAS (alkyl acids: PFUnA, PFNA, PFDA) showed to correlate with
   certain fatty acids (PFDA with↑stearic acid, ↑4-oxopentanoic acid,
   ↑beta-hydroxyisovaleric acid, ↑nonanoic acid, ↓oleic acid; PFNA with
   ↑beta-hydroxyisovaleric acid, ↑nonanoic acid; PFUnA with ↑nonanoic
   acid), PFOS with ↑chrysanthemic acid. In cord blood: PFOS and↑FA 18:2 +
   2O, PFDA with↑hexadecanedioic acid, ↑4-oxopentanoic acid, ↑stearic
   acid, ↑beta-hydroxyisovaleric acid, PFUnA with ↑stearic acid, PFNA with
   ↑stearic acid, ↑beta-hydroxyisovaleric acid NA FDR-adjusted using
   Benjamini–Hochberg method NA
   Hu et al. 2019 [[297]112] N = 397. Child Health and Development Studies
   (CHDS) in California: leveraged matched case–control (breast cancer)
   population based on birth year and trimester of maternal blood draw
   (white, non-obese,~25 y.o.) Cross-sectional within matched case–control
   Perinatal (1959–1967) PFOS, its precursor EtFOSAA, and EtFOSAA-to-PFOS
   ratio High-resolution C18 LC coupled with MS with positive ESI 34
   features had an association with PFOS, and 49 with EtFOSAA. 63 features
   were commonly associated with EtFOSAA, PFOS and the ratio. EtFOSAA/PFOS
   ratio associated with ↓β-alanine, ↓creatinine, ↓pipecolate, ↓lysine,
   ↓arginine, ↓creatine, ↓adrenochrome, ↑homocysteine, ↑betaine,
   ↑phosphoserine, ↑N6,N6-dimethyl-L-lysine, ↑citrulline. EtFOSAA
   associated with ↑homocysteine, ↑betaine, ↑phosphoserine, ↑citrulline,
   ↓creatinine, ↓arginine, ↓creatine, ↑5beta-Cholestan-3-one,
   ↑aminobutanal, ↑mercaptopurine, ↓phosphoglycolate, ↑trihydroxyvitamin
   D3, ↑tyramine, ↓ureidoisobutyrate. However, PFOS associated
   with↑lysine, ↑targinine, ↑creatine, ↑adrenochrome, ↓betaine,
   ↓phosphoserine, ↓N6,N6-dimethyl-L-lysine, ↑aminoisobutyric acid,
   ↑clupanodonyl carnitine, ↑glutamate, ↑carnitine, ↑hexadecenoyl
   carnitine, ↓N1-methyl-4-pyridone-5-carboxamide, ↑peptide
   2-(3-carboxy-3-aminopropyl)-L-histidine,↓linoelaidyl carnitine,
   ↓histidine, ↑tetracosapentaenoyl carnitine All PFOS, EtFOSAA, and their
   ratio were associated with enriched pathways for glycine,
   serine,alanine, and threonine metabolism, and urea cycle/amino group
   metabolism. PFOS was strongly associated with carnitine shuttle, lysine
   metabolism, and BCAA metabolism (valine, leucine, and isoleucine
   degradation) and moderately associated with β-alanine, vitamin B3, and
   butanoate metabolism. Other mild associations were found between
   EtFOSAA with enriched pathways for bile acid biosynthesis, alanine and
   aspartate metabolism, and between EtFOSAA/PFOS ratio and pyrimidine
   metabolism, lysine metabolism, alanine and aspartate metabolism, and
   aspartate and asparagine metabolism FDR-adjusted using
   Benjamini–Hochberg method Confounders: total cholesterol, age and
   p,p’-DDE level
   Alderete et al. 2019 [[298]29] N = 40. High-risk overweight/obese
   population, Hispanic children (Los Angeles area), 3 year follow up
   Cross-sectional within a prospective cohort Childhood (2001–2011) PFOA,
   PFOS, PFHxS in plasma HRM, LC–MS, HILIC with ESI source operated in
   positive mode PFOA and PFHxS concentrations and ↑glucose. Identified
   149, 298, and 17 metabolite features associated with plasma
   concentrations of PFOA, PFOS, and PFHxS, respectively. The integrated
   analysis identified a cluster of children with increased 2-h glucose
   levels over follow up, characterized by increased PFAS levels and
   altered metabolite patterns: ↑palmitic acid, ↑hydroperoxylinoleic acid,
   ↑tyrosine, ↑phenylalanine, ↑arginine, and decreased plasma levels of
   ↓sphingomyelin, ↓linoleic acid, and ↓aspartate Pathway enrichment
   analysis showed significant alterations (24 pathways) in association to
   PFASs exposure: lipids (glycosphingolipid metabolism, glycoshingolipid
   biosynthesis-ganglioseries, glycoshingolipid biosynthesis-globoseries,
   de novo fatty acid biosynthesis, fatty acid metabolism, linoleate
   metabolism), amino acids (aspartate and asparagine metabolism, tyrosine
   metabolism, arginine, and proline metabolism, alanine and aspartate
   metabolism, glycine, serine, alanine and threonine metabolism,
   histidine metabolism, selenoamino acid metabolism, beta-alanine
   metabolism, glutathione metabolism, glutamate metabolism), aminosugars
   metabolism, vitamins and cofactors pathways (vitamin B3 (nicotinate and
   nicotinamide) metabolism, vitamin B9 (folate) metabolism), nitrogen
   metabolism, urea cycle/amino group metabolism, pyrimidine metabolism,
   sialic acid metabolism, drug metabolism-cytochrome P450 FDR-adjusted
   <20%) using Benjamini–Hochberg method Confounders: age, sex, and social
   position
   Jin et al. 2020 [[299]116] N = 74. Patients with nonalcoholic fatty
   liver disease (NAFLD), ages 7–19 y.o., 71% male, 51% Hispanic, 85%
   obese, in Atlanta, GA Cross-sectional Childhood (2007–2015) PFOA, PFOS,
   PFHxS LC and high-resolution MS, in plasma Overall, PFOA associated
   with 348 metabolite features; PFOS with 349; and PFHxS with 662
   features. For higher risk population (NASH), increased PFAS levels
   altered plasma metabolite levels ↑phosphoethanolamine, ↑tyrosine,
   ↑phenylalanine, ↑aspartate and ↑creatine, and ↓betaine Overall, 21
   metabolic pathways including tyrosine metabolism, aspartate and
   asparagine metabolism, xenobiotics metabolism, glycine, serine, alanine
   and threonine metabolism, urea cycle/amino group metabolism,
   glycerophospholipid metabolism, arginine and proline metabolism, lysine
   metabolism, valine, leucine and isoleucine degradation, butanoate
   metabolism, phosphatidylinositol phosphate metabolism, drug
   metabolism-other enzymes, ascorbate and aldarate metabolism, citrate
   cycle (TCA cycle), nicotinate and nicotinamide metabolism, alanine and
   aspartate metabolism, caffeine metabolism, glutamate metabolism,
   vitamin B6 metabolism, glyoxylate and dicarboxylate metabolism, carbon
   fixation FDR-adjusted using Benjamini–Hochberg method Confounders: age,
   sex, ethnicity, z-BMI Effect modifiers: sex, ethnicity, liver
   histologic features
   Kingsley et al. 2019 [[300]121] N = 114 8-year-old children from
   Cincinnati, OH (HOME cohort) Cross-sectional Childhood (2011–2014)
   PFOA, PFOS, PFNA and PFHxS (in serum) LC and high-resolution MS, in
   serum: HILIC with ESI in positive mode and reversed-phase
   chromatography (RPC) with ESI operated in negative mode (C18-negative)
   PFAS concentrations were associated with certain enriched metabolic
   features primarily ↑lipids and ↑dietary factors. In the C18-negative
   mode, 17, 63, 47, and 29 m/z features were associated with serum PFOA,
   PFOS, PFNA, and PFHxS. In the HILIC-positive mode, 18, 253, 76, and 39
   m/z features were associated with serum PFOA, PFOS, PFNA, and PFHxS
   Pathways associated with all four PFAS included lipid metabolism, amino
   acids metabolism-arginine and proline metabolism, aspartate and
   asparagine metabolism, butanoate metabolism, beta-alanine metabolism,
   glutamate metabolism, glycine, serine, alanine and threonine
   metabolism, glycerophospholipid metabolism, glycosphingolipid
   metabolism, glyoxylate and dicarboxylate metabolism, histidine
   metabolism, linoleate metabolism, methionine and cysteine metabolism,
   tyrosine metabolism, urea cycle/amino group metabolism, vitamin B1
   (thiamin) metabolism, vitamin B3 (nicotinate and nicotinamide)
   metabolism FDR-adjusted using Benjamini–Hochberg method Confounders:
   child age, sex, and race
   Yu et al. 2022 [[301]106] N = 152. Adolescent girls in the Growing Up
   Healthy Study in New York City Cross-sectional Adolescence (2005–2017)
   n-PFOS, n-PFOA, Sm-PFOS, PFHxS, PFNA in plasma Reverse-phase and
   hydrophilic LC-HRMS in negative and positive modes In positive mode,
   ↑betaine (n-PFOS), ↓LPE(18:0) (n-PFOS), ↓LPC(16:0) (n-PFOS, n-PFOA),
   ↑LPG(18:1) (n-PFOS), ↓LPC(18:0) (n-PFOS), ↓SM(d18:2/14:0) (n-PFOS),
   ↓PE(20:4/P-18:0) (Sm-PFOS), ↓GPE(16:0/22:6) (n-PFOA), ↓SM(d18:1/24:1)
   (Sm-PFOS); in negative mode, ↑hippuric acid (n-PFOS), ↑gamma-glutamyl
   leucine (n-PFOS), ↑N(2)-phenylacetyl glutamine (n-PFOS),
   ↓dihomolinolenic acid (n-PFOS), ↓11,14-trans-eicosadienoic acid
   (n-PFOS), ↓dehydroepiandrosterone sulfate (n-PFOA), ↓androsterone
   sulfate (n-PFOA), ↑glycocholic acid (n-PFOS), ↑taurodeoxycholate
   (n-PFOS, n-PFOA), ↑LPE(20:4) (n-PFOS), ↑LPC(18:2) (n-PFOS), and
   ↑GPC(P-18:0/20:4) (n-PFOS) NA FDR-adjusted using Benjamini–Hochberg
   method NA
   Chen et al. 2020 [[302]118] N = 102. Young adults (17–22 y.o.), 82%
   overweight or obese, 60% Hispanic (Meta-AIR study), in California
   Cross-sectional Early adulthood (2017) PFOA, PFOS and PFHxS in plasma
   LC and high-resolution MS, HILIC with positive ESI and C18 hydrophobic
   RPC with negative ESI (plasma). MS/MS (serum) In fasting plasma, 231
   metabolomic (HILIC positive) and 239 metabolomic features (C18
   negative). Using 30-min postglucose challenge plasma samples, 372
   metabolomic (HILIC positive) features and 518 metabolomic (C18
   negative) features. All marginal significant associations with at least
   one PFAS. 19 metabolites identified linked to pathway analyses
   (proline, glutamine (PFOS, PFHxS), methionine, citrulline (PFOA),
   mannose/galactose (PFOA, PFOS, PFHxS), lysoPC(18:0) (PFOA, PFOS),
   sphingosine (PFOS), lactate (PFOS, PFHxS), oxovalerate/ketoisovalerate
   (PFOS, PFHxS), hydroxymethylglutarate (PFOA, PFOS, PFHxS), glucose
   (PFHxS), fatty acids including octanoate (PFOS, PFHxS), palmitate
   (PFOA, PFOS), linolenic acid (PFOA, PFHxS), linoleic acid (PFOA, PFOS),
   oleic acid (PFOA, PFOS), stearic acid (PFOA, PFOS), homolinoleic acid
   (PFOA, PFOS), arachidonic acid (PFOA, PFOS)) Pathways with dysregulated
   metabolism of lipids (glycerophospholipid, glycosphingolipid, fatty
   acids), amino acids such as arginine, proline, and tryptophan, as well
   as hexoses NA for -omics data Confounders: age, sex, parental
   education, race/ethnicity, cigarette and e-cigarette smoking status in
   the past week, physical activity levels and dietary covariates, (body
   fat) Effect modifiers: obesity
   Lu et al. 2019 [[303]17] N = 92 (40 occupational workers and 52
   controls in China) Cross-sectional within case–control Adulthood (2017)
   Σ6PFASs (PFBA, PFOA, PFBS, PFHxS, PFOS, and 6:2 Cl-PFESA) LC–MS, and
   GC–MS, in plasma 14 metabolites: ↑3-hydroxyoctanoic acid, ↓azelaic
   acid, ↑sebacic acid, ↑gamma-CEHC, ↑acylcarnitines (C18:1-CN, C18:2-CN),
   ↑pyroglutamic acid, ↑ornithine, ↓methionine sulfoxide,
   ↑DL-2-aminooctanoic acid, ↑hypoxanthine, ↓myo-inositol,
   ↓glycerophosphocholine, ↓piperine Pathways involved included lipid
   metabolism, amino acids metabolism, purine metabolism, inositol
   metabolism, retinol metabolism, metabolism of alkaloids and their
   derivatives NA Confounders: Age, BMI, gender, smoking, and drinking
   status
   Schillemans et al. 2021 [[304]113] N = 374. 187 matched pairs (based on
   age, sex and date of blood draw), nested within the Vasterbotten
   Intervention Programme cohort who donated blood samples (Swedish
   population) Nested case–control (cross-sectional metabolomics design)
   Adulthood (baseline: 1990–2003: follow-up: 2000–2013) PFOS, PFOA,
   PFHxS, PFDA, PFNA, PFUnA LC-qTOF-MS on reverse phase and HILIC columns
   in both positive and negative ionization modes PFAS levels
   (particularly long-chain) correlated with 171 metabolite features. In
   partial correlations (adjusted), PFAS was associated with ↑5
   glycerophospholipids (PFOS (1), PFDA (5), PFNA (1), PFUnDA (3),
   long-chain PFAS (5)), ↓2 glycerophospholipids (PFDA (2), PFUnDA (2),
   PFNA (1), long-chain PFAS (2)), ↑1 fatty acid (PFDA, PFNA, PFUnDA,
   long-chain PFAS), ↑3 diacylglycerols (PFOS (1), PFDA (3), PFNA (3),
   long-chain PFAS (3)), ↓1 diacylglycerol (PFDA, PFNA, long-chain PFAS),
   ↓gammabutyrobetaine (PFHxS, PFDA, long-chain PFAS), ↑
   3,4,5-trimethoxycinnamic acid (PFNA, PFUnDA, long-chain PFAS),
   ↑docosahexaenoic acid (PFHxS, PFDA, PFNA, PFUnDA, long-chain PFAS) NA
   FDR-adjusted Confounders: sex, age, sample year, marital status,
   education, smoking status, physical activity, and case–control status
   Li et al. 2020 [[305]31] N = 397. Leveraged matched case–control
   population (based on birth year and trimester of maternal blood draw)
   of women (white, non-obese, ~ 25 y.o.) from the Child Health and
   Development Studies (CHDS) cohort in California Cross-sectional within
   matched case–control Adulthood (1960s) Mixed exposures (39 chemicals)
   including PFAS (PFOA,PFHxS, PFOS, PFOSA, PFHpA, PFNA, PFUdA, PFDeA,
   PFDoA, MePFOS-AAcOH, EtP-FOSAAcOH), DDT, and PCBs (from serum)
   High-resolution LC–MS using C18 column in positive ESI mode Metabolite
   communities showed strong associations with DDT, PFAS communities (and
   lipids) Common metabolite communities associated with PCB, PFAS and
   lipids include: linoleate metabolism, fatty acid metabolism, omega-3
   fatty acid metabolism, de novo fatty acid biosynthesis, xenobiotics
   metabolism, tryptophan metabolism, purine metabolism, sialic acid
   metabolism, bile acid bionsynthesis, vitamin E metabolism, TCA cycle,
   lysine metabolism, methionine and cysteine metabolism, valine, leucine
   and isoleucine metabolism, electron transport chain,
   glycerophospholipid metabolism, glycosphingolipid metabolism,
   chondroitin sulfate degradation, N-glycan degradation, heparan sulfate
   degradation, pentose phosphate pathway, beta-alanine metabolism,
   alanine and aspartate metabolism, glyoxylate and dicarboxylate
   metabolism, arginine and proline metabolism, urea cycle/amino group
   metabolism, aspartate and asparagine metabolism, drug metabolism-
   cytochrome P450 NA NA
   Wang et al. 2017 [[306]123] N = 181. Chinese male adult population
   Cross-sectional Adulthood (NA) PFOA, PFOS, Total PFCs (PFOA, PFOS,
   PFDA, PFHxS, PFNA, PFUnA) LC/obitrap-MS in negative ion mode, in serum
   High exposure PFAS levels associated with ↓D-glucurono-6,3-lactone
   (PFOA, PFOS, PFAS), ↓α-carboxyethyl hydroxychromanol (PFOA),
   ↓arachidonic acid (PFOA), ↓hypoxanthine (PFOA, PFOS, PFAS),
   ↓oxoglutaric acid (PFOA, PFOS, PFAS), ↓pyroglutamic acid (PFOA, PFOS,
   PFAS), ↓tetrahydrobiopterin (PFOA, PFOS, PFAS), ↓xanthine (PFOA, PFOS,
   PFAS), ↑deoxyarabinohexonic acid (PFOA, PFOS, PFAS), and
   ↑hydroxybutyric acid (PFOS, PFAS) NA/Pathway analyses not conducted
   (metabolite findings implied that PFAS exposure may play a role in
   lipid metabolism, oxidative stress, α-tocopherol metabolism, xenobiotic
   detoxification, antioxidation, nitric oxide (NO) signal pathways,
   glutathione (GSH) cycle, Krebs cycle, and purine metabolism) NA
   Confounders: age, BMI, smoking and drinking status
   Salihovic et al. 2019 [[307]117] N = 965. Swedish population (aged 70
   years, 50% women) Cross-sectional Late adulthood (2001–2004) PFOS,
   PFHpA, PFOA, PFNA, PFUnDA, PFHxS UPLC-QTOF-MS operated in positive
   electrospray mode, in plasma 15 metabolites were found to be associated
   with levels of PFASs, except PFHxS. PFNA and PFUnDA were associated
   with multiple glycerophosphocholines (↑phosphatidylcholines (P-36:4,
   40:6, 38:5, 38:6, 36:5), ↑lysophosphatidylcholines (20:5/0:0,
   0:0/20:5)), ↑dicarboxylic acid (C12H14O5). PFUnDA was also related to
   fatty acids (↑docosapentaenoic acid (DPA), ↑docosahexaenoic acid
   (DHA)), ↓L-proline, ↑lysophosphatidylcholines (P-16:0/0:0, 0:0/18:0).
   PFOA was only related to ↑dicarboxylic acid (C12H14O5), PFHpA only to
   ↑uric acid, and PFOS were associated with ↑dicarboxylic acid (C12H14O5)
   and ↓monoacylglycerol (16:1) Human metabolic pathways were
   significantly enriched in lipids and fatty acids: glycerophospholipid
   metabolism (significant), linoleic acid metabolism (significant), and
   α-linoleic acid metabolism (not significant) Bonferroni FDR-adjusted
   Confounders: sex, smoking, exercise habits, education, energy, and
   alcohol intake
   You et al. 2022 [[308]105] N = 496. Cohort with 5 major chronic
   diseases (71 obese, 81 hyperuricemia 83 hypertensive, 74 diabetic, 104
   dyslipidemia patients, and 83 controls) in China Cross-sectional
   Adulthood (2018–2019) Multiple chemicals including individual PFAS and
   ΣPFAS (PFOA, PFNA, PFDA, PFUnDA, PFDoDA, PFHxS and PFOS), ΣPFCA (PFOA,
   PFNA, PFDA, PFUnDA and PFDoDA), and ΣPFSA (PFHxS and PFOS). Only PFOA,
   PFNA, PFDA, PFUnDA and PFOS used solely in the matabolomics analyses
   LC-HRMS method 240 endogenous metabolite markers were significantly
   associated with at least one PFAS. At least 2 or more PFAS groups
   (total PFAS, PFCA, or PFSA) correlated with↑↓amino acids (↑creatine,
   ↑creatinine, ↑pyroglutamic acid, ↑3-methyl-L-histidine, ↑benzoic acid,
   ↓glutamic acid, ↓pipecolic acid), ↑peptides (prolyl-isoleucine,
   Phe-Pro), ↑↓xenobiotics (↑1,7-dimethylxanthine, ↑caffeine,
   ↓3,7-dimethyluric acid), ↑uric acid, ↑↓sterol lipids (↑DHEA-S, ↓ST
   28:1;O;S), ↑↑↓bile acids, ↑cholesterol, ↑↑↓carnitines, ↑↑↓fatty acids,
   ↑glycerides (MGs, DGs, TGs), ↑↑↓phosphatidylcholines (PCs),
   ↑phosphatidylethanolamine (PEs), ↑↑↓lysophospholipids,
   ↑glycerophospholipids, ↑↑↓ceramides, ↓tocopherol, and ↑↑↓sphingolipids
   NA NA. FDR-adjusted only in analyses comparing hyperuricemia vs.
   controls Gender, age, BMI, sampling time, location, education level,
   cigarette smoking and alcohol drinking history
   Bessonneau et al. 2021 [[309]109] N = 143. 69 California women
   firefighters (FF) and 74 office workers (OW) enrolled in the Women
   Firefighters Biomonitoring Collaborative (WFBC) study Cross-sectional
   Adulthood (2014–2015) Exposome data matrix of 620 unique chemicals
   which matched to 300 chemical formulas, including PFHxS and PFOS
   LC-HRMS in negative ESI mode In firefighters only, PFHxS was associated
   with microbial-derived secondary bile acid ↑sulfolithocholylglycine;
   PFOS was correlated with one inflammatory signaling molecule ↓15d PGD2,
   and ↑calcitriol (vitamin D) NA FDR-adjusted, threshold of 0.1 NA for
   -omics data
   Huang et al. 2019 [[310]107] N = 57. Males in the CISTPFP study in
   China Cross-sectional Adulthood (2009–2010) Arsenic, PAE, and 11 PFC
   (PFOA, PFOS, PFBA, PFBS, PFDA, PFHpA, PFHxA, PFHxS, PFNA, PFUnA, and
   PFDoA) in blood. Only PFOA, PFOS, PFHxS, PFUnA, PFDA, PFNA were
   detected and included in analyses UPLC/MS/MS with heated ESI in
   positive and negative ion mode PFHxS, PFUnA, total PFCs associated with
   ↑pivaloylcarnitine and PFHxS also with ↑glycerophosphocholine.
   (Eicosatetraenoate, carnitines, and tocotrienol possible mediators of
   the positive association between PFHxS and sperm concentration) NA
   FDR-adjusted Age, body mass index (BMI), abstinence time, smoking, and
   alcohol drinking status
   Li et al. 2021 [[311]122] N = 84. Pregnant women in Beijing, China
   Cross-sectional Adulthood and prenatal (2015–2016) PFOA, PFNA, PFDA,
   PFUnDA, PFHxS, PFHpS, PFOS, 6:2 Cl-PFESA, 8:2 Cl-PFESA measured in
   maternal and cord serum HPLC-QTOF-MS 279 metabolites in mothers and 338
   meteabolites in fetus associated with PFAS. Shorter legacy PFAS (PFOA
   and PFHxS) associated with more metabolites than novel PFAS (6:2
   Cl-PFESA and 8:2 Cl-PFESA). PFAS-related metabolites in maternal serum
   included steroid hormones (6:2 Cl-PFESA/8:2 Cl-PFESA and ↑pregnenolone;
   PFOA and ↓cortisol 21-sulfate), ↓anahormones (PFNA and
   ↓5α-pregnane-3,20-dione), ↓fatty acids (PFHpS and ↓16-hydroxypalmitate;
   PFHxS and ↓linoleate; PFDA and ↓decanoic acid), and ↓terpenoids (PFUnDA
   and ↓canthaxanthin). PFAS-related metabolites in cord serum included
   ↑fatty acids (PFDA and ↑docosahexaenoic acid; PFOA, PFNA, PFUnDA and
   ↑octadecenoic acid; PFHxS and
   ↑8-[(1R,2R)-3-oxo-2-{(Z)-pent-2-enyl}cyclopentyl]octanoate; PFHpS and
   ↑arachidonate; PFOS and ↑glycolithocholate) Pathways in maternal serum
   and pathways in cord serum found to be associated with PFAS exposure:
   steroid hormone biosynthesis, arachidonic acid metabolism, α-linolenic
   acid metabolism, linoleic acid metabolism, and retinol metabolism
   FDR-adjusted to 20%: q-value (pFDR) using Bayesian posterior
   probability Confounders: age, weight, gravidity, parity, and residence
   of the pregnant women, as well as the length, weight, and gender of the
   fetuses
   Chen et al. 2021 [[312]103] N = 120. 9–80 y.o. in Wuxi City, China
   Cross-sectional Childhood/Adolescence/Adulthood (NA) Multiple
   contaminants including 66 PFCs (PFOA and PFOS in metabolomics analyses)
   UPLC-Q-Orbitrap HRMS In untargeted analyses, PFOA and PFOS associated
   with ↑ total lysophosphatidylcholine (LPC); PFHxS with ↑total
   lysophosphatidylcholine (LPC) and ↑total sphingomyelin (SM); PFDA with
   ↑total triglycerides (TG): and PFUnA with total ↑TG, ↑SM, and ↑LPC NA
   NA Age, BMI, and health status
   [313]Open in a new tab
   ^a

   See [314]Table S2 for a complete summary of all significant metabolites
   and pathways altered by PFAS across studies. Table ordered by age or
   developmental stages. If multiple age stages are studied, these studies
   are included at the end. ↑ (upregulated metabolite), ↓ (downregulated
   metabolite), ↑↑↓ (both upregulated and downregulated metabolite across
   PFAS subtypes but with overall more upregulation), and ↑↓↓ (both
   upregulated and downregulated metabolite across PFAS subtypes but with
   overall more downregulation)

Table 3.

   Summary of studies examining PFAS and targeted metabolomics in humans
   Author/year Study population Study design Window of PFAS exposure
   (calendar year) PFAS included in analytic dataset Metabolomics
   analytical method Main findings (metabolites) Main findings (inferred
   pathways) Multiple testing correction Covariates
   Stratakis et al. 2020 [[315]115] N = 1,105. Mother-children pairs
   (multicenter study, European HELIX cohort), 6–10 year follow-up
   Prospective Prenatal (1999–2010) PFOS, PFNA, PFOA, PFHxS, and PFUnDA
   (in plasma from pregnant mothers around mid-2000s) LC ESI–MS/MS, in
   child serum (circa 2014) Network of children at high vs low risk for
   liver injury involved particularly metabolites: 5 amino acids (PFNA,
   PFUnDA with ↑leucine, ↑valine, ↑isoleucine, ↑tryptophan,
   ↑phenylalanine), 1 ↑↓biogenic amine (acetylornithine (↓PFHxS, ↑PFNA,
   ↑PFUnDA)), 18 glycerophospholipids (PC ae C36:3 (↓PFHxS, ↑PFNA,
   ↑PFUnDA), PC aa C34:1 (↓PFHxS, ↑PFNA, ↑PFUnDA), PC aa C40:6 (↑PFHxS,
   ↑PFOS, ↑PFOA), PC aa C36:6 (↑PFOA, ↑PFHxS, ↑PFOS), PC aa C38:0 (↑PFOA,
   ↑PFHxS, ↑PFOS), PC aa C38:6 (↑PFOA, ↑PFHxS, ↑PFOS), PC ae C38:0
   (↑PFHxS, ↑PFOS), PC aa C36:5 (↑PFHxS), PC aa C36:3 (↓PFHxS, ↑PFNA,
   ↑PFUnDA), PC aa C32:0 (↓PFHxS, ↑PFNA, ↑PFUnDA), PC ae C34:1 (↓PFHxS,
   ↑PFNA, ↑PFUnDA), PC ae C36:5 (↑PFNA, ↑PFUnDA), PC ae C36:4 (↑PFNA,
   ↑PFUnDA), PC ae C38:4 (↑PFNA, ↑PFUnDA), PC ae C38:5 (↑PFNA, ↑PFUnDA),
   LysoPC a C20:3 (↓PFHxS, ↑PFNA, ↑PFUnDA), LysoPC a C18:1 (↓PFHxS, ↑PFNA,
   ↑PFUnDA), LysoPC a C20:4 (↓PFHxS, ↑PFNA, ↑PFUnDA)), 3 sphingomyelins
   (↑SM C18:0 (PFOS, PFNA, PFUnDA), ↑SM C18:1 (PFNA, PFUnDA), ↑↓SM (OH)
   C22:1 (↓PFHxS, ↑PFNA, ↑PFUnDA)), 1 ↑hexose (PFNA, PFUnDA) Protein and
   amino acid pathways (protein digestion and absorption; aminoacyl-tRNA
   biosynthesis; valine, leucine, and isoleucine biosynthesis; valine,
   leucine, and isoleucine degradation; phenylalanine, tyrosine and
   tryptophan biosynthesis) and lipid metabolism pathways
   (glycerophos-pholipid metabolism, ether lipid metabolism, and
   sphingolipid signaling pathway) NA NA for -omics data
   Lee et al. 2021 [[316]125] N = 290. 8–10 y.o. Taiwanese children from
   the Taiwan Birth Panel Study (TBPS, N = 214) and Taiwan Early-Life
   Cohort (TEC, N = 76 Cross-sectional Childhood (2013–2014) 13 PFAS from
   which 8 were detected > 10% (PFOS, PFTrDA, PFDA, PFOA, PFUnDA, PFHpA,
   PFNA, PFHxS) UPLC-MS/MS PFTrDA and PFDA exposures (high vs low level)
   were associated with serum lipid profiles (PLS-DA), and PFOS was
   marginally associated. In multiple linear regression (tertile 3 vs 1),
   PFOS associated with ↑lysoPC(18:2/0:0),↓PC(16:1/16:1), ↓PC(16:1/16:0),
   ↓PC(16:1/18:2), ↓PC(16:0/18:1), ↓PC(16:0/20:3), ↓PC(18:0/22:4),
   ↓PC(18:0/20:3), ↑PC(20:4/20:4), ↑P-PC (P-16:0/18:2, P-16:0/18:1,
   P-36:4, P-36:2, P-16:1/22:5), ↑SM (32:1), ↑SM (33:1), ↑SM (34:1), ↑SM
   (38:1), ↑SM (40:2), ↑SM (40:1), ↑SM (41:2), ↑SM (41:1), ↑SM (42:2):
   PFTrDA associated with ↓PC(16:0/20:3), ↓PC(16:1/16:0), ↓PC(16:1/18:2),
   ↓PC(38:4), ↓lysoPC(18:2/0:0), ↓lysoPC(0:0/18:2), ↑PC(18:2/18:2),
   ↑PC(16:0/22:6), ↓PC(18:0/20:3), ↑PC(20:4/20:4), ↓PC(18:0/22:4),
   ↑P-PC(P-16:1/22:5), ↑SM(38:1): and PFDA associated with ↓PC(16:0/18:1),
   ↓PC(16:1/16:1), ↓PC(16:1/18:2), ↑PC(20:4/20:4), ↑PC(18:2/18:2), ↓SM
   (32:1), ↑SM(34:2), ↑SM(36:2), ↑SM(38:1), ↑SM(40:1), ↑SM(42:3) NA NA
   Gender, age, BMI, household income (in multivariable linear regression
   analyses)
   Chen et al. 2020 [[317]118] N = 102. Young adults (17–22 y.o.), 82%
   overweight or obese, 60% Hispanic (Meta-AIR study), in California
   Cross-sectional Early adulthood (2017) PFOA, PFOS and PFHxS in plasma
   LC and high-resolution MS, HILIC with positive ESI and C18 hydrophobic
   RPC with negative ESI (plasma). MS/MS (serum) In serum, non-stratified
   analyses: PFOA, PFOS, and PFHxS associated with ↑glycerol, and PFOA
   also associated with grouped ↑short-chain non-OH/DC acylcarnitines.
   PFHxS associated with ↑acylcarnitine C16:1-OH/C14:1-DC. Interaction
   between PFHxS and obesity on 3-hydroxybutyrylcarnitine (C4-OH) NA NA
   for -omics data Confounders: age, sex, parental education,
   race/ethnicity, cigarette and e-cigarette smoking status in the past
   week, physical activity levels, and dietary covariates, (body fat)
   Effect modifiers: obesity
   Mitro et al. 2021 [[318]111] N = 691. Participants at higher risk for
   diabetes (BMI > 24, > 25 y.o., with impaired glucose tolerance and a
   fasting plasma glucose of 5.3–6.9 mmol/L) enrolled in the Diabetes
   Prevention Program (DPP cohort), in the US Cross-sectional Adulthood
   (1996–1999) Total PFOS, n-PFOS, Sm-PFOS, total PFOA, n-PFOA, Sb-PFOA,
   PFNA, PFHxS, EtFOSAA, MeFOSAA, in plasma HILIC coupled to MS in
   positive ion mode (for amino acids and amines); C8 chromatography
   coupled to MS in positive ion mode (lipids); amide chromatography
   coupled to MS using negative ion mode electrospray ionization (for
   organic acids) 38 metabolites significantly associated with any PFAS. 4
   DAGs and 18 TAGs were associated with at least one PFAS. 4 plasmalogens
   (3↑ and 1↓) and 4 sphingomyelins were significantly associated with at
   least one PFAS. All PFOS were associated with 3 ↑PC plasmalogens (C36:1
   PC plasmalogen, C36:4 PC plasmalogen, C34:1 PC plasmalogen-A), ↑C16:1
   sphingomyelin (SM), and ↑C18:2 SM). All PFOA were associated with 3
   ↑phosphatidylethanolamines (C36:3 PE, C38:4 PE, C36:4 PE) and 4
   ↑triacylglycerols (C54:2 TAG, C52:1 TAG, C50:1 TAG, C50:3 TAG). Sb-PFOA
   was inversely associated with ↓C34:4 PC plasmalogen. PFNA was
   significantly associated with ↑C54:6 triacylglycerol, EtFOSAA was
   significantly associated with ↑C18:2 SM. Sb-PFOA was associated with
   ↑leucine. Total PFOA, n-PFOA, Sb-PFOA associated with ↓glycine. Total
   PFOS, total PFOA, n-PFOA, Sb-PFOA, and EtFOSAA, associated with total
   ↑BCAAs. Total PFOS, n-PFOS, Sm-PFOS, total PFOA, n-PFOA, and EtFOSAA
   associated with total ↑sphingolipids (C16:1 SM, C18:1 SM, C18:2 SM, and
   C22:1 SM) and total ↑glycerophospholipids (including total
   phosphatidylethanolamines and total plasmalogens). Total PFOA and
   Sb-PFOA was associated with ↑glycerolipids (diacylglycerols and
   triacylglycerols) NA/Pathway analyses not conducted (metabolite
   findings implied that PFAS concentrations were associated with amino
   acid, glycerolipid, and glycerophospholipid pathways) FDR-adjusted
   using Benjamini–Hochberg method Confounders: Age, sex, race/ethnicity,
   use of anti-hyperlipidemic or triglyceride-lowering medication income,
   years of education, marital status, smoking, family history of diabetes
   (BMI, waist circumference, parity, menopausal status, other PFAS)
   Sen et al. 2022 [[319]108] N = 105. NAFLD (70% female) patients
   undergoing a laparoscopic bariatric surgery. Northen European
   population Cross-sectional Adulthood (NA) Environmental contaminants
   including PFAS (PFHxS PFNA, PFOA,2 PFOS isomers) Hepatic polar
   metabolites (GC × GC–TOFMS), hepatic molecular lipids (UHPLC-QTOFMS),
   hepatic acylcarnitines (UHPLC-QQQMS), hepatic BAs (UHPLC-QQQMS) In
   females, serum concentrations of PFAS associated with glyco- and
   tauro-conjugated primary, and secondary↑BAs (PFNA with TCA (taurocholic
   acid), GCDCA (glycochenodeoxycholic acid), TCDCA (taurochenodeoxycholic
   acid), GUDCA (glycoursodeoxycholic acid); PFOA with TCDCA, GHCA
   (glycohyocholic acid), GUDCA, CA (cholic acid), DCA (deoxycholic acid);
   PFOS with CA and DCA. Furthermore, PFAS alters and ↑lipid metabolism
   (↑Cers and ↑HexCers (PFNA, PFOA, and PFOS with ↑HexCer d18:1/18:0, PFOA
   an PFOS with ↑HexCer d18:1/24:0, PFOS with ↑Cer d18:0/16:0 and Cer
   d18:0/18:0), ↑ether phospholipids (O-PCs (PFOA and PFOS with↑O-PC 40:4,
   O-PC 40:5, O-PC 44:5; phosphatidylcholines (PFNA with↓LysoPC d20:4)),
   ↑TGs (PFNA with TG 52:0, TG 54:1, TG 55:2; PFOA with TG 55:3, TG 55:2,
   TG 54:1, TG 53:2, TG 52:2, TG d18:1/18:1/18:1, TG(d16:0/18:1/2 0:1) +
   TG(d18:0/18:1/18:1); PFOS and TG 55:3, TG 55:2, TG 54:1, TG 53:2, TG
   53:1, TG 52:3, TG 52:2, TG 52:0, TG 51:2, TG 51:1, TG 50:1, TG 50:0, TG
   d18:1/18:1/18:1, TG (d16:0/18:1/20:1) + TG(d18:0/18:1/18:1) and ↑DGs
   (PFOA with DG 36:3, DG 36:2; PFOS with DG 36:2, DG 34:1)) and polar
   metabolites (PFNA and ↑alanine, ↓cholesterol; PFOA with ↓malic acid;
   PFOS with ↓ aspartic acid, cholesterol, serine, threonine). Inverse
   associations were observed in males: PFNA was associated with ↓THCA,
   cholesterol, PFOA with ↓GUDCA, TG 55:5, TG 60:4, and PFOS associated
   with ↓DCA, TG 60:3, DG 32:1, phosphatidylethanolamine PE 42:0, LysoPC
   d18:0, HexCer d18:1/16:0, Cerd18:1/26:1 +Cer d18:2/26:0, Cer
   d18:0/24:1, citric acid In both sexes, primary bile acid biosynthesis,
   glycerophospholipid metabolism, along with alanine, aspartate and
   glutamate metabolism were over-represented comparing high vs low PFAS
   level. Sphingolipid metabolism pathways were over-represented in highly
   exposed females FDR-adjusted NA
   Matta et al. 2022 [[320]104] N = 87. 18–45 y.o. women in France with
   endometriosis and controls Cross-sectional Adulthood (2018–2019)
   Persistent organic pollutants, including 14 PFAS (PFAS retained for
   analyses: PFHpS, PFOS, PFOA, PFNA, PFDA,PFUnA) Small molecules measured
   by LC–MS/MS and lipids by flow injection analysis (FIA)-MS/MS with ESI
   source PFAS associated with↓ratio of diacylhosphatidylcholines and
   cyl-alkylphosphatidylcholines to choline, ↓ratio of hydroxylated
   sphingomyelins to non-hydroxylated sphingomyelins (Ratio.
   SMOHs.to.SMNonOHs). PFOS, PFOA associated with ↓HexCer d18:1/24:0. PFOA
   associated with tetradecenoylcarnitine (C14:1), PFOS with TG 16:0
   (40.6) and TG 16:0 (40.8), PFDA with TG 16:0 (40.8) and TG 18:1 (35.3),
   PFNA with TG 18:1 (35.3) and ↑HexCer d18:1/24:0 NA NA NA
   Ji et al. 2021 [[321]120] N = 160. 80 COVID-19 patients and 80
   symptom-free controls were recruited from Shanxi and Shandong provinces
   in China Cross-sectional Adulthood (2020) PFOA, PFOS, and total 12 PFAS
   (PFOS, PFOA, PFBS, PFHxA, PFHpA, PFHxS, PFNA, PFDA, PFUnA, PFDoA,
   PFTrDA, and PFTeDA) in urine and serum LC–MS/MS In COVID patients, 49
   metabolites associated with PFOA, PFOS, and ィ(12) PFASs, including
   mitochondrial metabolism metabolites. In adjusted analyses in COVID
   patients total PFAS, PFOA, PFOS were associated with mitochondrial
   metabolism metabolites (↑2-aminobutyric acid, ↑2-hydroxyglutarate,
   ↑3-hydroxyisobutyric acid, ↑acetoacetic acid, ↑aconitic acid,
   ↑butyrylcarnitine, ↑glycolic acid, ↓hydroxypropionic acid, ↑itaconic
   acid, ↑L-acetylcarnitine, ↑pyruvic acid, ↑succinic acid,
   ↑tiglylglycine), kynurenine metabolism metabolites
   (↑3-hydroxyanthranilic acid, ↑hydroxykynurenine, ↑L-kynurenine), and
   eicosanoids metabolism metabolites (↑8-isoprostane, ↑11-dTxB2,
   ↑PGF2alpha, ↑tetranor-PGEM, ↑TXB2). PFOA additionally associated with
   ↑3-methyladipic acid, ↑quinolinic acid. PFOA and total PFAS were also
   associated with ↑N-formylkynurenine PFAS associated with mitochondrial
   metabolism, kynurenine metabolism, eicosanoid metabolism, glycolysis
   and gluconeogenesis metabolism, niacin metabolism, purine and
   pyrimidine metabolism, microbiome metabolism, sphingolipid and
   phospholipid metabolism, amino acid metabolism, bioamines and
   neurotransmitter metabolism in COVID patients FDR-adjusted, threshold
   of 0.2 Adjusted for age, gender, body mass index (BMI), diabetes,
   cardiovascular diseases (CVDs), and urine albumin-to-creatinine ratio
   (UACR) in random effects model
   Sinisalu et al. 2020 [[322]114] N = 66. Newborns in Finland in the Type
   1 Diabetes Prediction and Prevention (DIPP) study (17 progressors to
   celiac disease and 16 healthy controls) with measured samples at both
   birth and 3-month follow-up Prospective Prenatal and childhood
   (1999–2005) 7 PFAS compounds detected but 5 detected in > 10% of
   samples and used for analyses (PFHpA, PFHxS, PFOA, PFOS, and PFUnDA)
   both at birth and at 3 months of age UPLC-MS/MS PFAS exposure may
   modulate lipid and BA metabolism, and the impact is different in the
   infants who develop CD later in life, in comparison to HCs.In
   multivariate correlation analysis, there were associations among PFAS,
   BAs and molecular lipids that were different in cord blood than at 3
   months (from negative to positive correlations) including as
   cholesterol esters (CEs), lysophosphatidylcholines (LPCs),
   phosphatidylcholines (PCs), phosphatidylethanolamines (PEs),
   sphingomyelin (SMs) and triacylglycerols (TGs). More specifically, at
   baseline in controls (PFHpA associated with ↓GCDCA, ↓TG; PFOS with
   ↑GCDCA; PFUnDA with ↓LPC,↓PC; total PFAS with ↓GCDCA), at baseline in
   CD progressors (PFOA, PFOS, total PFAS associated with ↓DCA), at 3
   months in controls (PFHpA associated with ↓GUDCA; PFUnDA, PFOA with
   ↓LPC) and at 3 months in CD progressors (PFOA associated with
   ↑GCDCA-PFOS with ↑DCA, ↑CDCA, ↑GCDCA; total PFAS with ↑DCA, ↑CDCA,
   ↑GCDCA; PFUnDA with ↑TG; PFHpA with ↑TG_mfa). In multiblock analyses
   for lipids: in cord blood, comparing the celiac risk group vs.
   controls, where PFOA and PFOS contributed the most, there were inverse
   correlations with ↓CEs and ↓ether-linked PCs (PC(O-34:3), PC(O-36:3,4)
   and PC(O-38:4,5,6)); and at 3 months, where PFHxS contributed the most
   to celiac risk group compared to controls, several ↑PCs, ↑SM(d36:2),
   ↑PC(O-38:4,5) were upregulated NA NA NA
   Chen et al. 2021 [[323]103] N = 120. 9–80 y.o. in Wuxi City, China
   Cross-sectional Childhood/Adolescence/Adulthood (NA) Multiple
   contaminants including 66 PFCs (PFOA, and PFOS in metabolomics
   analyses) UPLC-Q-Orbitrap HRMS In targeted adjusted analyses, PFOA,
   PFOS, PFHxS, PFUnA associated with 18:0 lysophosphoethanolamine
   (LPE(18:0)) NA NA Age, BMI, and health status
   [324]Open in a new tab
   ^a

   See [325]Table S2 for a complete summary of all significant metabolites
   and pathways altered by PFAS across studies. Table ordered by age or
   developmental stages. If multiple age stages are studied, these studies
   are included at the end.↑ (upregulated metabolite), ↓ (downregulated
   metabolite), ↑↑↓ (both upregulated and downregulated metabolite across
   PFAS subtypes but with overall more upregulation), and ↑↓↓ (both
   upregulated and downregulated metabolite across PFAS subtypes but with
   overall more downregulation)

   A total of 19 studies reported a non-targeted metabolomics approach, 7
   studies used a targeted metabolomics approach, and 2 studies included
   both approaches. The most common analytical method used in samples was
   LC–MS or LC-HRMS (n = 11), followed by UHPLC-, UPLC-, HPLC-, or
   LC-qTOF-MS (n = 7), HPLC-, UPLC-, or LC–MS/MS (n = 7), and
   LC/Orbitrap-MS or UPLC-Q-Orbitrap HRMS (n = 2). Other methods such as
   ^1H-NMR (n = 1) or GC–MS (n = 1) were less commonly used for
   metabolomics analyses.

   Metabolomics data were collected from several media across studies: in
   serum (n = 16), plasma (n = 10), urine (n = 2), or semen samples (n =
   1). Most studies measured metabolomics in a single sample collected
   from participants, with only two studies integrating repeated
   metabolomics measures in plasma [[326]114] or urine [[327]102].

   The majority of studies were cross-sectional (n = 25), including four
   [[328]30, [329]102, [330]114, [331]115] with additional prospective
   analyses of PFAS measured prior to metabolomics. Several studies
   focused as well on examining associations with a specific disease or
   biomarker for disease (n = 18). Cardiometabolic outcomes were the most
   studied diseases in relation to PFAS exposure, including overall
   diabetes [[332]105], type 2 diabetes [[333]111, [334]113] type 1
   diabetes [[335]30], obesity [[336]105], hypertension [[337]105],
   dyslipidemia [[338]105], hyperuricemia [[339]105], non-alcoholic fatty
   liver disease or steatohepatitis [[340]108, [341]116], in addition to
   cardiometabolic markers (insulin resistance, lipids, glucose level,
   adiposity, BMI, or liver injury) [[342]29, [343]106, [344]109,
   [345]115, [346]117, [347]118]. A few other studies also examined
   reproductive outcomes (semen quality [[348]107] or endometriosis
   [[349]104]), birth weight or small-for-gestational age [[350]119],
   breast cancer [[351]31], celiac disease [[352]114], and COVID-19
   [[353]120]. Overall, most articles had a medium to high-level quality
   score with Kingsley et al. [[354]121], Maitre et al. [[355]102],
   Salihovic et al. [[356]117], Hu et al. [[357]112], and Chang et al.
   [[358]119] studies scoring the highest based on the aforementioned
   criteria ([359]Table S2).

Summary of Individual Metabolites Linked to PFAS Exposures Across Studies

   A total of 546 unique metabolite features, either classifiable or
   identifiable under HMDB, ChEBI, or LipidMaps (with reported names),
   were found to be dysregulated by PFAS exposures across studies. PFAS
   were associated more frequently with lipid metabolites, followed by
   organic acids and derivatives ([360]Fig. 2, [361]Figures S1-[362]S3).
   More specifically, metabolites that appeared significant most
   frequently across all studies pertained to the class of chemicals fatty
   acyls (present in n = 26 studies), carboxylic acids and derivatives (n
   = 26 studies), glycerophospholipids (n = 23 studies), steroids and
   steroid derivatives (n = 18 studies), or sphingolipids (n = 16 studies)
   ([363]Fig. 2). Overall, a positive association between PFAS and
   metabolites was more frequently reported across studies, particularly
   for fatty acyls, imidazopyrimidines, and benzene and substituted
   derivatives, whereas a negative association was reported for
   sphingolipids in more studies. A sensitivity descriptive analysis was
   conducted restricting to studies that included a metabolite confidence
   level 1 in their results ([364]Figure S4), which yielded consistent
   findings for most metabolite classes; PFAS had an overall positive
   association with imidazopyrimidines, benzene and substituted
   derivatives, organooxygen and organonitrogen compounds, while a
   negative association was found for sphingolipids across studies. Other
   metabolite classes had approximately a similar positive-to-negative
   association number ratio ± 1 study, with the exception of fatty acyls
   which were observed to revert in directionality with respect to PFAS
   but the number of studies in our sensitivity analysis was reduced
   significantly (from n = 26 studies that initially reported any
   PFAS-metabolite associations to only n = 11). Among all untargeted
   metabolomics data, the most common metabolites reported were
   glycerophospholipids (40.3%), fatty acyls (23.2%), sphingolipids
   (9.3%), carboxylic acids and derivatives (8.1%), glycerolipids (6.1%),
   and steroids (4.7%) ([365]Figure S1). A similar distribution of
   metabolites was also observed in studies reporting a targeted approach,
   where PFAS were found to be related the most to the following
   metabolite classes: glycerophospholipids (33.6%), glycerolipids
   (15.6%), sphingolipids (13.0%), carboxylic acids and derivatives
   (10.0%), steroids (9.1%), and fatty acyls (6.2%) ([366]Figure S2).
   Overall, the most frequent PFAS-metabolite associations were observed
   with respect to glycerophospholipids, followed by fatty acyls
   ([367]Figure S3).

Fig. 2.

   Fig. 2
   [368]Open in a new tab

   Summary of the number of studies reporting targeted and untargeted
   PFAS-metabolite associations by directionality and metabolite class

   Long-chain legacy PFAS were the predominant PFAS compound forms that
   were most examined in relation to metabolomics and also the most
   frequently associated with different metabolic profiles ([369]Table
   S3). These include primarily PFOA, PFOS, and PFHxS, followed by PFDA,
   PFNA, and PFUnDA. We observed a fewer number of associations for other
   long-chain PFAS such as PFTrDA, PFHpS, and EtFOSAA, short-chain PFAS
   (PFHpA, PFPeA), or novel PFAS (Cl-PFESAs). In terms of the total number
   of metabolites, an overall consistency in the directionality of
   associations was particularly observed between PFAS homologues and
   increased number of fatty acids [[370]17, [371]29, [372]105, [373]110,
   [374]113, [375]117, [376]120, [377]122, [378]123], glycerophospholipids
   (glycerophosphocholines and glycerophosphoethanolamines)
   [[379]105–[380]108, [381]111, [382]115, [383]117, [384]124, [385]125],
   glycerolipids (triradylcglycerols, diradylglycerols,
   monoradylglycerols) [[386]106, [387]108, [388]111, [389]113, [390]114,
   [391]124], bile acids [[392]31, [393]105, [394]108, [395]109, [396]114,
   [397]119, [398]124], and to a lesser extent, for fatty acid esters
   (acylcarnitines) [[399]17, [400]105, [401]107, [402]112, [403]118,
   [404]120], amino acids [[405]17, [406]29, [407]102, [408]105, [409]106,
   [410]108, [411]111, [412]112, [413]115, [414]116, [415]119, [416]120],
   phosphosphingolipids (sphingomyelins) [[417]105, [418]111, [419]115,
   [420]124, [421]125], purines [[422]17, [423]105, [424]112, [425]117,
   [426]119], sulfated steroids [[427]105, [428]119],
   glycerophosphoinositols [[429]105], carnitine [[430]105, [431]112], and
   benzoic acids [[432]105, [433]106, [434]120]. On the other hand,
   cholestane steroids [[435]105, [436]108, [437]112], quinone, and
   hydroquinone lipids [[438]105] appeared to be consistently
   downregulated in relation to PFAS. Ceramides were frequently reported
   as being potentially altered by PFAS, though with unclear or
   inconsistent directionality [[439]105, [440]108, [441]124].

   At the metabolite-level, we found an overlap in the number of
   significant metabolites across studies ([442]Table S4). A particularly
   high number of metabolites that were reported in at least three studies
   in relation to PFAS exposures were an increase in docosahexaenoic acid
   (Schillemans et al. [[443]113], Li et al. [[444]122], Salihovic et al.
   [[445]117]), phosphatidylcholine (PC) 40:6 (Salihovic et al.
   [[446]117], Sinisalu et al. [[447]124], You et al. [[448]105]),
   creatine (Jin et al. [[449]116], Hu et al. [[450]112], You et al.
   [[451]105]), and uric acid (Salihovic et al. [[452]117], You et al.
   [[453]105], Chang et al. [[454]119]). Overall, an additional 15
   metabolites were consistently upregulated (PC 35:1, PC 36:5, PC 38:5,
   PC 38:6, PC 40:5, ether-linked phosphatidylcholines PC O-38:5, PC
   O-40:4, triacylglycerols TG 54:2, TG 54:5, TG 54:1,
   glycochenodeoxycholic acid, pyroglutamic acid, phenylalanine, succinate
   or succinic acid, carnitine), two were downregulated (glycine,
   betaine), and two were either upregulated and downregulated
   (deoxycholic acid) or for which directionality was not reported
   (methionine) across several studies.

Summary of Dysregulated Pathways Linked to PFAS Exposures Across Studies

   In pathway analyses, a total of 101 different pathways were reported to
   be significantly altered in relation to PFAS exposures. In [455]Table
   S5, we classified these PFAS-related pathways based on KEGG
   identifiers. There were 10 untargeted studies (Jin et al. [[456]116],
   Lu et al. [[457]17], Kingsley et al. [[458]121], Li et al. [[459]31],
   Alderete et al. [[460]29], Li et al. [[461]122], Salihovic et al.
   [[462]117], Hu et al. [[463]112], Chang et al. [[464]119], Chen et al.
   [[465]118]) and 3 targeted studies (Sen et al. [[466]108], Ji et al.
   [[467]120], Stratakis et al. [[468]115]) that examined metabolic
   pathways related to PFAS using Mummichog, pathway enrichment analysis
   in MetaboAnalyst, or the KEGG database. Out of these 13 studies, the
   majority (8 studies) conducted analyses adjusting for confounders and
   incorporating FDR-correction (Jin et al. [[469]116], Kingsley et al.
   [[470]121], Alderete et al. [[471]29], Li et al. [[472]122], Salihovic
   et al. [[473]117], Hu et al. [[474]112], Chang et al. [[475]119], Ji et
   al. [[476]120]).

   The most frequently reported alteration of metabolism in most studies
   was among amino acids followed by lipid, carbohydrate, and metabolism
   of cofactors and vitamins, accounting for 27%, 25%, 15%, and 9% of the
   total significant PFAS-induced pathway associations across studies,
   respectively ([477]Table S5, [478]Fig. 3). The predominant pathways of
   amino acid metabolism included alanine and aspartate (n = 7), aspartate
   and asparagine (n = 6), arginine and proline (n = 6), urea cycle (n =
   5), lysine (n = 5), and glutamate (n = 5). The predominant pathways of
   lipid metabolism involved glycerophospholipid (n = 8),
   linoleate/linoleic acid (n = 6), glycosphingolipid (n = 5),
   glycosphingolipid biosynthesis (n = 5), bile acid-related pathways (n =
   3), or fatty acid-related pathways (n = 15). The most prevalent
   pathways of carbohydrate metabolism involved butanoate (n = 4), TCA
   cycle (n = 3), sialic acid (n = 3), glycolysis/gluconeogenesis (n = 3),
   glyoxylate and dicarboxylate (n = 3). The most frequent pathways of
   vitamin and cofactor metabolism involved vitamin A (n = 3), vitamin B3
   (n = 4), and vitamin D3 (n = 3). In addition, PFAS exposure also
   contributed to the alteration of several additional pathways:
   nucleotide metabolism (5%, i.e., purine and pyrimidine metabolism),
   glycan biosynthesis and metabolism (5%, i.e., N-glycan
   degradation/biosynthesis, keratan/chondroitin/heparin sulfate
   degradation), energy metabolism (4%, i.e., nitrogen metabolism),
   metabolism of other amino acids (3%, i.e., beta-alanine metabolism), or
   xenobiotic metabolism (3%). Results for other pathways were reported
   less consistently across studies.

Fig. 3.

   Fig. 3
   [479]Open in a new tab

   Summary of PFAS-related targeted and untargeted pathway associations
   reported across studies^[480]a

   ^aClassifications based on the Kyoto Encyclopedia of Genes and Genomes
   (KEGG) database. Pathways that were associated with PFAS in less than 3
   studies where counted as “Other”. Other: includes lipid metabolism
   (ether lipid metabolism, n=1, 0.42%; NA, n=1, 0.42%; phytanic and
   peroximal oxidation, n=0.42%; prostaglandin formation from dihomo
   gama-linoleic acid, n=1, 0.42%; saturated fatty acids beta-oxidation,
   n=1, 0.43%; carnitine shuttle, n=2, 0.85%; squalene and cholesterol
   biosynthesis, n=1, 0.42%; steroid hormone biosynthesis, n=1, 0.42%;
   arachidonic acid metabolism, n-1, 0.42%; alpha-linolenic acid, n=1,
   0.42%; omega-3 fatty acid metabolism, n=2, 0.85%; omega-6 fatty acid
   metabolism, n=2, 0.85%), amino acid metabolism (NA, n=2, 0.85%; valine,
   leucine, and isoleucine biosynthesis, n=1, 0.42%; valine, leucine, and
   isoleucine metabolism, n=1, 0.42%; phenylalanine, tyrosine, and
   tryptophan biosynthesis, n=1, 0.42%; alanine, aspartate metabolism, and
   glutamate metabolism, n=1, 0.42%), metabolism of other amino acids
   (glutathione metabolism, n=2, 0.85%; selenoamino acid metabolism, n=1,
   0.42%), carbohydrate metabolism (ascorbate and aldarate metabolism,
   n=2, 0.85%; inositol metabolism, n=1, 0.42%; propanoate metabolism,
   n=2, 0.85%; pentose phosphate pathway, n=2, 0.85%; pyruvate metabolism,
   n=1, 0.42%; pentose and glucoronate interconversions, n=1, 0.42%;
   C5-branched dibasic acid metabolism, n=1, 0.42%; hexose
   phosphorylation, n=1, 0.42%; galactose metabolism, n=2, 0.85%; fructose
   and mannose metabolism, n=2, 0.85%; aminosugars metabolism, n=2, 0.85%;
   hyaluronan metabolism, n=1, 0.42%; starch and sucrose metabolism, n=1,
   0.42%), energy metabolism (electron transport chain, n=1, 0.42%;
   mitochondrial metabolism, n=1, 0.42%), metabolism of cofactors and
   vitamins (nicotinate and nicotinamide metabolism, n=2, 0.85%; vitamin
   B1 metabolism, n=1, 0.42%; vitamin B2 metabolism, n=1, 0.42%; vitamin
   B6 metabolism, n=1, 0.42%; vitamin B9 metabolism, n=2, 0.85%; vitamin E
   metabolism, n=1, 0.42%; porphyrin metabolism, n=1, 0.42%; lipoate
   metabolism , n=1, 0.42%; biopterin metabolism, n=1, 0.42%; ubiquinone
   biosynthesis, n=1, 0.42%), biosynthesis of other secondary metabolites
   (caffeine metabolism, n=1, 0.42%; metabolism of alkaloids and their
   derivatives, n=1, 0.42%), metabolism of terpenoids and polyketides
   (limonene an pinene degradation, n=1, 0.42%), translation
   (aminoacyl-tRNA biosynthesis, n=1, 0.42%), structure-based
   classification (eicosanoid metabolism, n=1, 0.42%), NA (dynorphin
   metabolism, n=1, 0.42%; kynurenine metabolism, n=1, 0.42%; microbiome
   metabolism, n=1, 0.42%; bioamines and neurotransmitter metabolism, n=1,
   0.42%)

   Similar pathway results were found in our analyses using Reactome where
   metabolites participating in membrane transport, lipid-related, or
   amino acid-related mechanisms were predominant across studies
   ([481]Fig. 4A–[482]C and [483]Tables S6-[484]S8). A complete list of
   Reactome pathways including those that appeared less frequently across
   studies were also shown in [485]Tables S9-[486]S11. Top pathways
   associated with metabolites involved in the potential deleterious
   effects of PFAS belonged to transmembrane transport, transport of bile
   salts and organic acids, metal ions and amine compounds, plasma
   lipoprotein remodeling pathways (including HDL), phospholipid and
   phospholipase-related pathways (PLC beta mediated events or
   phospho-PLA2), phagocytosis, amino acid transport across the plasma
   membrane, Golgi-to-ER transport, glucose-dependent insulinotropic
   polypeptide, or acyl chain remodeling of lipids, among others
   ([487]Fig. 4C, [488]Table S6). Additional pathways were observed when
   we conducted analyses separately with metabolites from positive and
   negative PFAS associations ([489]Fig. 4A, [490]B, [491]Tables
   S7-[492]S8). For instance, while pathways implicated in triglyceride
   metabolism were present in studies that reported positive
   PFAS-metabolite associations ([493]Table S7), immune system or ceramide
   signaling pathways, as well as amino acid transport across the plasma
   membrane, were predominant across studies that reported metabolites
   with inverse PFAS associations ([494]Table S8). In addition to the top
   pathways (present in 25% of the studies and with an FDR < 0.05),
   summaries for Reactome pathways specific to lipid and amino acid
   metabolism are shown in [495]Figs. 5 and [496]6. It is noteworthy that
   at least 3 studies showed a pathway enrichment for PPAR-αregulation of
   lipid metabolism, lipid particle organization, sphingolipid, and
   phospholipid metabolism ([497]Fig. 5), as well as catabolism of
   tryptophan, threonine, and choline, creatine metabolism, and carnitine
   synthesis ([498]Fig. 6).

Fig. 4.

   Fig. 4
   [499]Open in a new tab

   Top Reactome pathways involved in PFAS-metabolite associations across
   studies^[500]a

   ^aMain Reactome pathways for PFAS-metabolite associations with positive
   (A), negative (B), and all directionalities (C) including positive,
   negative, and unknown (not reported) directionality. A specific pathway
   was shown if it was found across more than 25% of the studies and with
   an FDR<0.05. Reactome entities ratios are denoted by the bubble sizes.
   Combined p-values were calculated [[501]240, [502]241] for top pathways
   across studies and shown in the x-axis. Studies were included if they
   reported any PFAS-metabolite association with an available metabolite
   KEGG ID recognizable by Reactome (n=26). Abbreviations:
   glucose-dependent insulinotropic polypeptide (GIP), amino acids (AAs),
   high-density lipoprotein (HDL), ATP-binding cassette (ABC),
   diacylglycerol (DAG), triacylglycerol (TAG), phospholipase C (PLC),
   Fcgamma receptor (FCGR), cardiolipin (CL), phospholipase A2 (PLA2),
   coat protein complex I (COPI), endoplasmic reticulum (ER),
   phosphatidylcholine (PC), and solute-carrier gene (SLC).

Fig. 5.

   Fig. 5
   [503]Open in a new tab

   Main Reactome pathways involved in lipid metabolism across
   studies^[504]a

   ^aHighlighted pathways represent more consistently reported lipid
   pathways across studies as follows: red asterisk represents
   statistically significant pathways in at least 3 studies (FDR<0.05),
   orange asterisk represents a pathway present in at least 5 studies with
   at least one study being statistically significant (FDR<0.05), and a
   yellow asterisk represents a pathway present in at least 5 studies but
   not reaching statistical significance (FDR>0.05). This visualization is
   based on results from [505]Table S9 indicating Reactome pathways
   related to metabolites with any directionality with respect to PFAS.
   Studies were included if they reported any PFAS-metabolite association
   with an available metabolite KEGG ID recognizable by Reactome.

Fig. 6.

   Fig. 6
   [506]Open in a new tab

   Main Reactome pathways involved in amino acid metabolism across
   studies^[507]a

   ^a Highlighted pathways represent more consistently reported amino acid
   pathways across studies as follows: red asterisk represents
   statistically significant pathways in at least 3 studies (FDR<0.05),
   orange asterisk represents a pathway present in at least 5 studies with
   at least one study being statistically significant (FDR<0.05), and a
   yellow asterisk represents a pathway present in at least 5 studies but
   not reaching statistical significance (FDR>0.05). This visualization is
   based on results from [508]Table S9 indicating Reactome pathways
   related to metabolites with any directionality with respect to PFAS.
   Studies were included if they reported any PFAS-metabolite association
   with an available metabolite KEGG ID recognizable by Reactome.

PFAS-Metabolomic Associations Reported Between Disease and Control
Populations

   Nearly half of the studies included in this review (n = 13) made
   reference to populations with a specific disease or populations at
   increased risk for disease for whom metabolomics data were analyzed.
   Three studies evaluated metabolomics in patient populations: Jin et al.
   [[509]116] in children with nonalcoholic fatty liver disease (NAFLD),
   Ji et al. in COVID-19 patients [[510]120], and Sen et al. [[511]108] in
   NAFLD patients undergoing a laparoscopic bariatric surgery. Another
   three metabolomics studies were conducted in population subgroups at
   high risk for disease: Chen et al. [[512]118] focused on children who
   had a history of being overweight or obese but did not have diabetes or
   other disease, Mitro et al. [[513]111] focused on adult participants at
   higher risk for diabetes (with BMIs of above 24 and with high levels of
   fasting plasma glucose), and Alderete et al. [[514]29] focused on
   children with high risk of T2D but without clinical diagnosis. Lastly,
   nearly seven studies reported PFAS-associated metabolite features
   separately in disease patients (or at increased risk for disease) and
   control populations: Stratakis et al. [[515]115] compared children with
   high vs. low liver injury risk, Schillemans et al. [[516]113] compared
   T2D and control pairs, Li et al. [[517]31] and Hu et al. [[518]112]
   examined breast cancer cases and controls but no metabolomics
   comparison was conducted, You et al. [[519]105] compared cases with
   hyperuricemia and controls, Matta et al. [[520]104] compared women with
   and without endometriosis, and Sinisalu et al. [[521]114] evaluated
   metabolomics on celiac disease patients and healthy controls. When
   comparing PFAS-metabolite associations in a cardiometabolic disease
   population with a healthy control group, Stratakis et al. [[522]115]
   found primarily increases in several branched-amino acids, and lipid
   alterations (glycerophospholipids and sphingomyelins), Schillemans et
   al. [[523]113] study observed primarily increases of
   glycerophospholipids and diacylglycerols linked to T2D odds ratios, and
   You et al. [[524]105] compared differential metabolites in
   hyperuricemia patients versus controls (but disregarding PFAS exposure
   status) showing alteration in several lipids and aminoacids. Sinisalu
   et al. [[525]114] conducted metabolomics analyses with respect to
   celiac patients and controls and showed alterations in lipid and bile
   acid metabolism triggered by PFAS exposure in infants who developed
   celiac disease later in life in comparison to healthy controls. Lastly,
   Matta et al. [[526]104] observed a differential metabolome (lipids and
   functional metabolite ratios) between cases of endometriosis and
   controls but disregarding PFAS exposure levels between groups.

PFAS Effects During Susceptible Windows of Exposure

   Given the different ages from populations included in reviewed studies,
   we evaluated potential differential patterns across life stages
   focusing on early life exposures or sensitive windows of exposure. We
   observed reports of alterations to amino acids during the prenatal
   period [[527]102, [528]119], as well as alterations in the metabolism
   of several vitamins (B3, D, and retinol) [[529]112, [530]119, [531]122]
   and dysregulation of imidazopyrimidines in pregnant women [[532]112,
   [533]119]. These findings (dysregulation of imidazopyrimidines, amino
   acids, and vitamins) were also consistent with findings in adults.
   Additionally, we observed that aromatic amino acids [[534]29, [535]116]
   and arginine or related pathways [[536]29, [537]116, [538]121], were
   particularly specific to associations found in children across studies
   in this review.

Discussion

   This systematic review highlights the potential for PFAS exposures to
   alter several metabolic pathways in humans as reported by recent
   investigations of the human metabolome. The studies summarized in this
   review were recently published, with the oldest publication being in
   2017, highlighting the relevance of this emerging field. We found
   consistent evidence across several observational studies suggesting
   that PFAS exposures are associated with dysregulations in lipid and
   amino acid metabolites and related pathways particularly relevant to
   metabolic disease in humans. Many of these pathways may be involved in
   energy and cell membrane disruption. We additionally observed both
   potentially similar and divergent effects of PFAS by age or
   developmental stages. Research gaps from the reviewed studies include
   the lack of prospective studies or longitudinal measures of PFAS and
   metabolomics over the life-course, as well as limited adjustment for
   relevant confounders. The majority of studies to date have evaluated
   long-chain legacy PFAS. Therefore, future research to study emerging
   and shorter chain PFAS, either as individual compounds or as exposure
   mixtures, is warranted.

Potential Dysregulated Pathways Linked with PFAS Exposures in Humans

   Growing evidence from previous epidemiological and experimental studies
   indicate that PFAS can alter health via increased oxidative stress,
   inflammation, peroxisome proliferator-activated receptor (PPAR)
   signaling, or sex steroid hormone mechanisms [[539]23–[540]26,
   [541]126]. PPARs are nuclear receptors that regulate fatty acids,
   lipids, and glucose metabolism [[542]127]. A variety of PFAS have been
   shown to activate PPARα, primarily expressed in the liver, in human
   cells in vitro [[543]128–[544]130]. Several mechanisms for PFAS
   toxicity in mammals also show that PFAS can incur damage via
   non-PPARα-dependent pathways, such as other nuclear receptors (i.e.,
   PPARγ, CAR, ERα) [[545]131–[546]135] inducing gene expression changes,
   altering mitochondrial function [[547]136–[548]138] modifying membrane
   fluidity [[549]139], or inhibiting gap junction intercellular
   communication [[550]131, [551]140, [552]141]. Similarly, findings from
   Reactome pathway analyses in this review also highlighted that PFAS may
   modify transmembrane, lipid, and amino acid metabolism. Of note,
   alterations in PPAR-mediated mechanisms were also among the most
   recurrent (present in at least 3 studies with FDR < 0.05) PFAS-induced
   pathways related to lipid metabolism in this review.

   At the molecular level, the potential for PFAS as endocrine disruptors
   is reflected in their capacity to bind to other proteins, their
   structure similar to that of fatty acids, and their putative
   involvement in the displacement of endogenous ligands, which could
   explain their high retention in human serum and the potential to alter
   lipid metabolism and the hormonal system. For instance, PFAS can
   interfere with binding to albumin [[553]142, [554]143], sex
   hormone-binding globulin [[555]144], corticosteroid-binding globulin
   [[556]71], and liver fatty acid-binding protein (L-FABP) [[557]145].
   PFAS also can compete with thyroxine in binding to thyroid hormone
   transport protein or receptors [[558]146, [559]147], which could lead
   to a potential decrease in the normal levels of circulating hormones
   and derivatives (i.e., thyroid hormone, SHBG) resulting in hormone
   dysregulation. In our review, we found that PFAS altered levels of
   steroid-related and bile acid-related pathways and metabolites, such as
   pregnane steroids [[560]102, [561]122], overall levels of sulfated
   steroids [[562]105, [563]119], and bile acids were increased [[564]105,
   [565]106, [566]108, [567]109, [568]114, [569]119, [570]122, [571]124],
   and levels of cholestane steroids were decreased [[572]105, [573]108,
   [574]112]. Similarly, previous metabolomic studies on PFAS toxicity in
   animals indicated that PFAS exposures modulated metabolic pathways
   related to sterols and bile acids in mice [[575]148]. This is
   consistent with findings from pathway analyses in Reactome indicating
   bile acid transport dysregulation and a moderate alteration of steroid
   metabolism. Overall, the findings from our systematic review parallel
   prior findings in animal studies, which together support a causal role
   for PFAS on disrupting endocrine pathways.

   Our review detected a disproportionately higher number of lipids and
   membrane function-related pathways across studies and across pathway
   analyses. Glycerophospholipids, the main type of lipid in the cell
   membrane, may be an important component in driving cellular
   accumulation of PFAS. More specifically, in our review an overall
   increase in glycerophosphocholines, and in particular of
   phosphatidylcholines, may indicate mitochondrial membrane disruption
   with a subsequent hydrolysis of phospholipids [[576]149]. Furthermore,
   glycosphingolipid and sphingolipid metabolism were recurrent in pathway
   analyses across the studies [[577]29, [578]31, [579]118, [580]119,
   [581]121]; related metabolites (altered ceramides and increased
   phosphosphingolipids, such as sphingomyelin) were also reported
   frequently across the studies [[582]29, [583]104, [584]105, [585]108,
   [586]111, [587]115, [588]124, [589]125] and corroborated in Reactome
   showing consistent pathways related to ceramide signaling,
   phospholipids, sphingolipids, and phospholipase activity which may
   exacerbate inflammatory response [[590]150]. Furthermore, given that
   sphingolipids constitute a major part of the fluidity, structure, and
   permeability of membranes and are involved in cell-to-cell signaling
   [[591]151–[592]154], a PFAS-induced membrane alteration mechanism may
   be a common pathway by which PFAS may exert cellular damage.

   We also found a relative consistent positive association with PFAS
   among glycerolipids, particularly with triacylglycerides (TGs). While
   the association between PFAS and increased levels of unhealthy lipids,
   including triglycerides or LDL-cholesterol, has not been consistent
   across animal [[593]155, [594]156] and epidemiological studies
   [[595]157–[596]161], the majority of overall PFAS-TG associations
   across human metabolomics studies included in this systematic review
   were positive in directionality [[597]105, [598]108, [599]111,
   [600]114, [601]124]. Furthermore, pathway analyses in Reactome
   confirmed that mechanisms related to TG metabolism were involved in the
   positive association between higher PFAS exposure levels and increased
   TGs. Similarly, lipoprotein pathways, including dysregulation of
   healthy cholesterol pathways such as high-density lipoprotein (HDL),
   were observed to be predominant across studies reporting PFAS-TG
   associations [[602]105, [603]108, [604]111, [605]114, [606]124]. In
   humans, PFAS associations with serum triglycerides rendered
   inconsistent effects across different PFAS in previous non-metabolomics
   epidemiological studies [[607]158, [608]160, [609]162, [610]163],
   though more positive associations were observed for PFOA [[611]157,
   [612]158, [613]164–[614]166], PFOS [[615]157, [616]165, [617]167,
   [618]168], PFHxS [[619]166, [620]169], or PFNA [[621]165, [622]166]
   exposures across various populations of healthy and unhealthy adults,
   children, and adolescents. Moreover, lipidomics studies in mice and
   rats have suggested more consistent associations with hepatic
   triglycerides for several long-chain PFAS (PFDoDA, PFOS, PFHxS, APFO,
   PFNA) [[623]131, [624]170–[625]173]. The mechanism underlying a
   potential PFAS-induced alteration in triglycerides points at initiation
   by PPARs. An in vitro study with PFAS-exposed human liver cells where
   PFOS, PFOA, and PFNA activated PPARα signaling, suggested that the
   subsequent observed increase in cellular triglyceride levels could be
   due to an induction of lipid droplet-associated proteins or
   glyceroneogenesis [[626]174]. PPARγ could be also implicated in the
   PFAS-induced lipid alteration as a regulator of lipids and triglyceride
   fat storage in adipose tissue [[627]175], though mechanisms for
   PFAS-induced damage via changes in triglyceride levels are poorly
   understood.

   PFAS may contribute to lipid dysregulation by alterations in energy
   metabolism. Across studies, we observed that PFAS were positively
   associated with acylcarnitines [[628]17, [629]105, [630]107, [631]112,
   [632]118, [633]120], which are intermediate metabolites involved in the
   transport of fatty acids and long-chain acyl-CoA from the cytosol into
   the mitochondria. Hence, the potential for acylcarnitines to be used as
   a marker of mitochondrial functioning and fatty acid oxidation
   [[634]176], given that elevated levels could reflect either
   mitochondrial dysfunction or an adaptive change to disturbed lipid
   metabolism. Our findings across the reviewed studies are consistent
   with the studies in mice, the latter suggesting an overall increase in
   acylcarnitines (particularly hepatic) in relation to PFAS [[635]148,
   [636]177]. Similarly, we observed an implication of PFAS exposures on
   fatty acid metabolism, carbohydrate, and amino acid metabolism. Binding
   of PFAS to fatty acid binding proteins may be implicated in the
   reduction of bioavailable binding sites for endogenous fatty acids
   resulting in higher concentrations of fatty acids. Fatty acids may
   interact with PPARs [[637]178] and liver X receptors and could be
   involved in the regulation of gene expression and inflammation.
   Furthermore, several branched-chain amino acids (leucine, isoleucine,
   and valine), relevant in fatty acid oxidation and prevalent in the
   metabolome of those suffering from obesity-related conditions
   [[638]179], were overall increased across reviewed studies [[639]111,
   [640]115]. Similarly, aromatic amino acids, namely tyrosine and
   phenylalanine, which have been shown to be related to insulin
   resistance or diabetes [[641]180, [642]181], were also increased across
   studies [[643]29, [644]115, [645]116]. Pathway analyses in Reactome for
   metabolites sharing a negative association with PFAS, implicated
   pathways related to acyl-chain remodeling and amino acid transport
   across the plasma membrane, which could explain abundant levels of
   amino acids in obese subjects via dysregulation of key metabolites
   helping in the processing of fatty acids and amino acids. It is
   hypothesized that impaired branched-chain amino acid metabolism could
   lead to accumulation of toxic branched-chain keto acids and acyl-CoA
   precursors facilitating conversions into acylcarnitines [[646]182,
   [647]183], which are increased in obese and T2D individuals [[648]180,
   [649]184], consistent with our findings across studies in this review.
   Increased levels of acylcarnitines may reflect incomplete long-chain
   fatty acid beta-oxidation and limited intermediates or tricarboxylic
   acid (TCA) cycle utilization [[650]184]. Furthermore, glycan-related
   pathways were also recurrent across studies in this review. Glycans are
   polysaccharides and alterations in glycosylation (a post-translational
   modification) have been suggested to be involved in diabetes etiology
   in prior epidemiological studies [[651]185–[652]187]. Perturbations of
   several metabolites associated with the TCA cycle, along with the
   presence of enriched carbohydrate metabolism across studies in our
   review, such as electron transport chain, glycolysis and
   gluconeogenesis, or beta-oxidation, is also consistent with the notion
   of the toxicity exerted by PFAS. Thus, changes in energy metabolism may
   have widespread implications on development, growth, aging, protection
   against infectious and toxic exposures, and multiple disease processes.

PFAS and Increased Disease Risk in Human Metabolomic Studies

   In humans, a myriad of studies have linked environmental
   endocrine-disrupting chemicals to a variety of diseases and metabolic
   dysregulations. In vitro studies with animal and human cells, have
   shown that PFAS can exert immunotoxicity [[653]188], hepatic toxicity
   [[654]172], developmental [[655]189], and endocrine toxicity
   [[656]147]. Research indicates that PFAS affect cardiometabolic markers
   of disease, can increase cancer risk, alter immune response, and impair
   reproductive health, thyroid, liver, and kidney function, among others
   [[657]4, [658]10, [659]12, [660]14, [661]18, [662]190–[663]192]. For
   instance, various PFAS, such as PFOA, PFBS, or PFNA have been linked to
   diabetes across different populations and epidemiological study designs
   [[664]193–[665]195]. Interestingly, Reactome pathways related to
   disease, the immune system, and glucose-dependent insulinotropic
   polypeptide (GIP), which is a hormone regulating insulin secretion,
   were shared across reviewed studies.

   Several of the studies included in our review indicated that PFAS
   increased the risk of multiple cardiometabolic conditions. Given that
   prior epidemiological studies not including metabolomics reported
   inconsistent findings on PFAS and cardiometabolic outcomes [[666]1,
   [667]196–[668]200], either due to their cross-sectional design or
   heterogeneous populations, epidemiological studies including
   metabolomics are warranted to elucidate potential mechanisms at the
   metabolite-level in the plausible link between PFAS exposure and
   cardiometabolic outcomes. In the liver, PFAS increased liver enzymes
   characterizing injury risk in Stratakis et al.’s study [[669]115]. In
   Jin et al.’s study [[670]116], PFOS and PFHxS were associated with
   increased odds for liver fibrosis, lobular inflammation, or
   nonalcoholic steatohepatitis (NASH) [[671]116]. PFAS exposure mixtures
   together with other environmental chemicals increased NAFLD risk in Sen
   et al.’s study [[672]108]. A potential mechanism for liver injury could
   be via increased oxidative stress, as reflected in increased liver
   function biomarkers (i.e., ALT, AST) [[673]201, [674]202]. Furthermore,
   PFAS was associated as well with increased glucose in several studies
   in our review [[675]29, [676]118]. Mitro et al. 2021 [[677]111]
   identified particularly several sphingomyelins,
   phosphatidylethanolamines, and DGs and TGs related to legacy PFAS (PFOA
   and PFOS) in a population at high risk of developing T2D. Similarly,
   increased T2D risk was also reported for PFNA-associated
   diacylglycerols in Schillemans et al. 2021 [[678]113]. Lipids are
   mainly stored in mature white adipocytes in the form of TGs. Increased
   circulating free fatty acids and accumulation of TGs and derivatives,
   such as diacylglycerols, are deemed contributing factors to insulin
   resistance [[679]203]. Along with TGs, diacylglycerols were overall
   increased across the studies in this review [[680]105, [681]108,
   [682]113]. Furthermore, prenatal PFAS exposure contributed to increased
   postnatal risk of type 1 diabetes in neonates in one other study
   (McGlinchey et al. [[683]30]), and PFHxS contributed to metabolic
   syndrome and lower levels of healthy cholesterol (HDL) in Bessonneau et
   al.’s study in occupationally-exposed adult women [[684]109]. Elevated
   levels of acylcarnitines, which were increased across studies in our
   review (“Potential Dysregulated Pathways Linked with PFAS Exposures in
   Humans”), have been linked to risk of cardiovascular disease in prior
   cohort studies [[685]204, [686]205]. Other metabolic disorders such as
   hyperuricemia were found to increase in risk upon PFAS exposures
   [[687]105]. This is consistent with findings from several large cohorts
   from US and Chinese populations where PFOA, PFNA, PFOS, and PFHxS were
   reported to increase risk of hyperuricemia [[688]2, [689]206,
   [690]207].

   There was evidence across the included studies from this review that
   PFAS exposures also altered reproductive outcomes and immune-related
   diseases. PFNA concentration was significantly associated with higher
   odds of small-for-gestational age (SGA) birth in a population of
   African-American women in Chang et al.’s study [[691]119], and PFHxS
   was linked to lower sperm concentration in Chinese males in Huang et
   al.’s study [[692]107]. Chemical mixtures including PFAS exposures also
   exacerbated conditions like endometriosis in Matta et al.’s study
   [[693]104], suggesting that an endometriosis metabolic pattern could be
   characterized by dysregulation of bile acid homeostasis. PFAS was also
   found to increase risk of COVID-19 in one metabolomics study via
   impaired kynurenine metabolism (involved in immune responses), and
   eicosanoids (involved in inflammatory responses) [[694]120], consistent
   with previous research linking PFAS and impaired immune system.

   Overall, we observed consistent findings for reported PFAS-metabolite
   associations between studies conducted in population subgroups with a
   specific disease compared to population-based studies, particularly in
   regard to alterations in lipid metabolites (glycerolipids [[695]104,
   [696]105, [697]108, [698]111, [699]113, [700]114, [701]117, [702]124],
   sphingolipids [[703]29, [704]30, [705]104–[706]106, [707]108, [708]111,
   [709]115, [710]124, [711]125], and fatty acyls [[712]17, [713]29,
   [714]104–[715]107, [716]109, [717]110, [718]112, [719]113,
   [720]117–[721]120, [722]122–[723]124]), bile acids [[724]105, [725]106,
   [726]108, [727]109, [728]114, [729]119, [730]122, [731]124], and amino
   acids [[732]17, [733]29, [734]30, [735]102, [736]105, [737]106,
   [738]108, [739]111–[740]113, [741]115–[742]120, [743]123].
   Additionally, the field of metabolomics is at its early stages and
   could have the potential to become a key tool for disease biomarker
   detection and help elucidate pre-diagnostic stages of disorders that
   could be amenable to intervention. The lack of multi-omics studies
   integrating metabolomics with genomics, epigenomics, toxicogenomics,
   transcriptomics, proteomics, the microbiome, and other fields,
   highlights the need to advance this area of research as multi-omics may
   be a promising avenue of exposomics research where combined
   applications can provide mechanistic explanations for differential
   levels of environmental contaminants in the body and corresponding
   phenotypic states across individuals.

Effects of PFAS Exposures in the Developing Human Fetal Metabolome

   Different populations may undergo distinct effects from PFAS at
   different stages throughout their life course. Of particular concern
   are the effects of PFAS in the womb due to the critical windows of
   exposure increasing the risk of disease onset in the offspring. Based
   on experimental research in mice, it is hypothesized that PFAS
   exposures lead to a reduction in transport of amino acid analogues from
   mothers to the fetus [[744]208]. Though across studies we observed a
   perturbation of amino acids in mothers, we observed inconsistent
   results throughout pregnancy; for instance, amino acids, such as
   glycine, were increased in mothers at 8–14 weeks in the Chang et al.
   study [[745]119]; in the Maitre et al. study, glycine decreased at
   trimester 3 [[746]102]. Consequences of amino acid perturbation in the
   developing fetus could indicate improper nutrition, affect fetal
   growth, and contribute to small gestational size [[747]119, [748]209,
   [749]210].

   An association between PFAS and disrupted metabolism related to
   cofactors and vitamins B and D was also observed across studies and are
   key contributors to proper fetus development. Metabolism for vitamin B3
   (nicotinate and nicotinamide) [[750]29, [751]112, [752]119, [753]121],
   retinol [[754]17, [755]121, [756]122], and vitamin D [[757]109,
   [758]112, [759]118, [760]119, [761]121] were reported consistently
   across studies, three of them being conducted in pregnant women
   [[762]112, [763]119, [764]122]. In pregnant women, vitamin A (retinol)
   and its analogs can regulate gene transcription impacting embryonic
   development [[765]211] and the immune system [[766]212–[767]214].
   Moreover, vitamin B is a well-known antioxidant for fetal growth (i.e.,
   folate), and Vitamin D can be implicated also in oxidative stress and
   inflammatory response mechanisms, as well as in the metabolism of
   glucose and fetal growth skeletal development or placental function
   [[768]215–[769]217].

   An overall increase of imidazopyrimidines [[770]17, [771]105, [772]112,
   [773]117, [774]119], including increased uric acid [[775]117,
   [776]119], was observed across studies in this review, with two studies
   being conducted in pregnant women [[777]112, [778]119]. In pregnant
   women, it is hypothesized that PFAS may lead to decreases in uric acid
   secretion, resulting in elevated serum uric acid concentrations, which
   in turn, may trigger placental inflammation and oxidative stress,
   inhibit amino acid transport to placentas, alter the development of
   endothelial and trophoblast cell development in the fetus, or lead to
   higher risk for pre-eclampsia [[779]119, [780]218–[781]221].

Effects of PFAS Exposures on the Child’s Metabolome

   Seven studies in total have focused on PFAS and metabolomics and were
   conducted in a population of children or adolescents [[782]29,
   [783]106, [784]114, [785]116, [786]118, [787]121, [788]125].
   Interestingly, out of these studies, three [[789]116, [790]121,
   [791]29] performed pathway analyses with PFAS-associated metabolites
   and numerous amino acid pathways were reported in all three studies:
   tyrosine metabolism, aspartate and asparagine metabolism, glycine,
   serine, alanine and threonine metabolism, urea cycle/amino group
   metabolism, arginine and proline metabolism, alanine and aspartate
   metabolism, and glutamate metabolism. We did not see consistent
   associations between PFAS-induced alterations on branched-amino acids
   (valine, isoleucine, and leucine) and related pathways across reviewed
   studies in children, contrary to previous findings from adult studies
   [[792]105, [793]111]. However, an alteration in aromatic amino acids
   was evident in children. Tyrosine and phenylalanine are aromatic
   metabolites that were primarily increased in children populations
   across reviewed studies [[794]29, [795]116]. Aromatic amino acids, as
   well as associated pathways, have been previously linked to increased
   risk of developing insulin resistance or obesity in children
   [[796]222–[797]226], or increased liver injury in adolescents
   [[798]227]. We also observed that arginine or arginine-related
   pathways, previously linked to diabetes [[799]228], were recurrent
   across the reviewed studies in children [[800]29, [801]116, [802]121].
   These findings are consistent with the hypothesis that environmental
   contaminants, namely PFAS, could alter the susceptible metabolism of
   children in a more remarkable way than at other developmental stages
   across the lifespan suggesting a potential window of susceptibility on
   certain amino acid types, particularly aromatic. A possible avenue of
   future research could also focus on infant PFAS exposures and other
   environmental contaminants via breast milk.

Differential Effects Across PFAS Groups

   This review also intended to discern any potential patterns of effects
   across PFAS subtypes. Sulfonic acids are considered more potent than
   carboxylic acids [[803]229], and it is hypothesized that longer
   carbon-chain PFAS may exert more deleterious effects [[804]230],
   however, this premise has been contested [[805]231] as novel PFAS
   emerge. We observed similar effects across the major studied PFAS
   (PFOS, PFOA, PFHxS), though a more clear-cut positive association seems
   to be apparent between the long-chain sulfonate PFOS and amino acids,
   fatty acid esters, or glycerophosphocholines. Overall, the findings
   observed across studies indicated that PFAS with longer half-lives were
   associated with relatively more metabolites. Similarly, we found that
   more legacy or long-chain PFAS were associated with more metabolites
   compared to short-chain or novel PFAS [[806]113, [807]122], yet this
   could also be due to the frequency of the chemical being studied across
   studies and/or their detectability, where novel shorter-chain PFAS have
   lower exposure levels across samples, and their exposure assessment
   could be more prone to measurement error due to shorter half-lives.
   Animal studies have shown that PPARα receptor activation was increased
   the longer the carbon chain of the PFAS chemical [[808]128, [809]232].
   Furthermore, inhibition of gap junction intercellular communication is
   also considered more prominent for longer chain PFAS than in shorter
   chain PFAS [[810]140]. Although animal studies can provide some
   validity as to the reason why we may be encountering different potency
   across PFAS groups, we would need more evidence for emerging or
   short-chain PFAS, which may be similarly toxic but are understudied, to
   further assess their deleterious effects on the human metabolome and
   involvement in disease.

Common Limitations of Previous Studies, Risk of Bias, and Future Directions

   Several common limitations were found across studies rendering a lower
   quality score. These included confounding factors not accounted for,
   lack of false discovery rate (FDR) adjustment, a cross-sectional design
   introducing potential reverse causation bias, a small sample size, or
   not reporting a high confidence level for metabolites. Out of all 28
   studies included in the review, no study reached a maximum quality
   score of 5, meaning that no study accounted altogether for a
   longitudinal study design, a large sample size, multiple testing
   correction, MSI metabolite confidence level 1, and covariate adjustment
   in the study design. Future PFAS-metabolomics studies may consider
   addressing these limitations in their design. Longitudinal measures of
   PFAS and metabolomics with longer follow-up from birth through
   adulthood are needed to better capture metabolite variability and to
   elucidate persistent effects over the life-course.

   Metabolomics approaches handling repeated exposures and longitudinal
   -omics data are also needed to corroborate potential windows of
   enhanced vulnerability. Variability encountered in the laboratory
   methods for metabolomics profiling as well as the intra-individual
   variability in the samples could have also reduced the power to detect
   associations [[811]233], so it is likely that the observed associations
   may be an underestimate of the true associations. We also expect that
   technical methods are more reliable than intra-individual correlation,
   which could have low reliability over time, as metabolites may be
   dependent on sex, age, fasting status, and diet of the individual.

   On the other hand, there could be potential false positives (type I
   error) due to lack of multiple testing correction or lack of
   covariate-adjustment in metabolomics analyses. Overall, in metabolomics
   analyses, 16 studies included in the review reported FDR-correction
   ([812]Tables 2 and [813]3, [814]Table S2). Similarly, only 8 studies
   reporting pathway analyses both applied FDR-correction and adjusted for
   any covariates ([815]Table S5). Moreover, nearly 40% of the studies in
   this review (n = 11) report to have included metabolites with a
   confidence level 1. Our sensitivity analyses including only these
   studies that reported the highest confidence level indicated a similar
   pattern of PFAS-associated metabolites for most metabolite groups
   examined ([816]Figure S4). Given that more than half of the studies
   either did not include MSI confidence level 1, or did not mention at
   all confidence levels, caution when interpreting both findings should
   be given. Compliance with the Metabolomics Standards Initiative when
   reporting metabolomics methods is needed in future studies [[817]99,
   [818]100].

   Confounding bias could be present in reviewed studies that examined an
   association without controlling for sociodemographic and lifestyle
   variables, such as diet or exposure to other correlated chemicals. Diet
   was included as a confounder in the metabolomic analyses for only 2
   studies [[819]102, [820]117]. Interestingly, high level of
   docosahexaenoic acid (DHA) was observed across several studies
   [[821]113, [822]117, [823]122]. Evidence from animal studies indicated
   that undergoing fish oil supplements prevented the PFOA-induced
   increase in hepatic triglyceride content by depressing the formation of
   triglycerides by DHA [[824]234]. This suggests that potential
   deleterious effects exerted by PFAS exposures may be attenuated if fish
   consumption, a potential negative confounder of PFAS and omega-3 fatty
   acids, is not adjusted for in the analyses. Given that fish is a
   DHA-rich food source but also has high PFAS content
   [[825]235–[826]237], findings cannot be attributed to PFAS exposures as
   the only causative factor. The same would apply to a high correlation
   between packaged foods with high caloric density content and PFAS
   contamination [[827]238], which could instead skew results
   over-representing lipid metabolism. Similarly, the presence of several
   benzenoids and xenobiotic metabolism across studies may indicate that
   PFAS may act in conjunction with other exogenous chemicals to alter
   metabolite levels in humans. Therefore, it is not possible to parse out
   whether findings are due to the effects from PFAS exposures solely, or
   instead, stem in part from potential confounders, a combination of both
   (mixture), or effect modification. Of note, there is a research gap
   regarding co-exposures since most of the reviewed articles on PFAS and
   metabolomics do not adjust for other pollutants in their analyses and
   correlated exposures are not systematically taken into account.

   The vast majority of studies presented in this review were conducted in
   White and Asian populations with only a few studies including Hispanic
   and African-American populations in the US [[828]29, [829]119].
   Increased representation of minority ethnic groups with potentially
   different socioeconomic backgrounds, lifestyle, and dietary behaviors
   is needed in future studies to reassure generalizability of findings to
   minority populations that are disproportionally affected by metabolic
   and other chronic diseases. Moreover, most of the evidence compiled in
   this review is drawn from blood samples and additional studies
   comparing PFAS effects via the metabolome across tissues and other
   biofluid matrices can be informative. Additionally, genetic
   susceptibility to PFAS exposure in relation to metabolomics was taken
   into account in only one study [[830]30]. Emerging evidence suggest
   gene-PFAS interactions in disease risk in humans [[831]239]. To that
   end, an emerging field of “multi-omics” data incorporating both
   metabolomics and genomics has just recently started to be applied in
   the investigation of environmental exposures and disease. Together with
   novel PFAS exposures, effect modification by genetic variations or
   incorporating “multi-omics” approaches could be a focus of future
   research. Improvements in the sensitivity of analytic tools (i.e., HRM)
   and harmonization in chemical annotation and standardization of
   pre-processing and data preparation (i.e., imputation, transformations,
   CV %) can help address these limitations in this field of research and
   enable a better reproducibility of metabolomic studies.

Limitations and Strengths of this Systematic Review

   This review included metabolite classifications from the HMDB and ChEBI
   databases and KEGG pathway classifications, which are not fully
   comprehensive, challenging the summary and classification of findings
   across specific metabolite groups. About a third of unique metabolite
   features reported in our review, were either unmatched to a KEGG ID
   (not found) or their KEGG ID was not recognized by Reactome.
   Conversions were not possible particularly for several lipid
   metabolites and vitamins and were underrepresented when performing
   pathway analyses compared to other molecules, in spite of appearing
   frequently in results across studies. Thus, we can expect that lipid
   metabolism, yet present across studies in a significant manner, is
   underestimated in Reactome pathway analyses, although this also seems
   to be the case for individual studies conducting other pathway
   enrichment analyses. We also recognize that differentiation of targeted
   and untargeted studies is arguable in the field of metabolomics, so we
   attempted to refer to this classification as reported by the studies.
   Additionally, laboratory methods likely influencing quality assessment
   are not reported systematically throughout each study, and therefore,
   we were not able to account for other factors into the quality score.
   One important strength is that despite the high heterogeneity noted
   across epidemiology and laboratory methods, study design, and
   populations (i.e., country of study, sex, race, age, occupational,
   unhealthy), we found consistent findings for several metabolites and
   pathways across studies, and thus, results are more likely to reflect
   true associations. Furthermore, our study provides a comprehensive
   systematic review of all human studies published on PFAS and
   metabolomics, including both targeted and untargeted metabolomics, as
   well as lipidomics studies. Lastly, by using Reactome we provided a
   pathway analysis, which is considered a more systematic method to
   summarize and visualize findings.

Conclusion

   PFAS are ubiquitous chemicals that can alter health via disruption of
   key metabolites and pathways in the human body. In this review, we
   summarized and identified alterations in several metabolites (amino
   acids, fatty acids, glycerophospholipids, glycerolipids,
   phosphosphingolipids, bile acids, ceramides, purines, and
   acylcarnitines) and related metabolic pathways that could underlie
   PFAS-associated diseases in humans, including lipid, amino acid,
   carbohydrate, nucleotide, glycan, or energy metabolism, and metabolism
   of cofactors and vitamins. Future studies should consider prospective
   designs optimizing methods for exposure-metabolomics analyses with
   longitudinal measures, additional confounder adjustment, or assessment
   of emerging PFAS and mixture effects to address existing limitations in
   this field.

Supplementary Material

   Tables S3-S5
   [832]NIHMS1918423-supplement-Tables_S3-S5.xlsx^ (72.6KB, xlsx)
   Table S2
   [833]NIHMS1918423-supplement-Table_S2.xlsx^ (54.8KB, xlsx)
   Supplementary Material
   [834]NIHMS1918423-supplement-Supplementary_Material.docx^ (526.2KB,
   docx)
   Tables S6-S8
   [835]NIHMS1918423-supplement-Tables_S6-S8.xlsx^ (44.2KB, xlsx)
   Tables S9-S11
   [836]NIHMS1918423-supplement-Tables_S9-S11.xlsx^ (434.3KB, xlsx)

Acknowledgements