Associations of per-and polyfluoroalkyl substances (PFAS) and their mixture with oxidative stress biomarkers during pregnancy

Background: Oxidative stress from excess reactive oxygen species (ROS) is a hypothesized contributor to preterm birth. Per-and polyfluoroalkyl substances (PFAS


Introduction
Per-and polyfluoroalkyl substances (PFAS) are environmental chemicals of public health concern and clinical significance because of their ubiquity, toxicity, and persistence in humans (Sunderland et al., 2019).Over 9,000 PFAS are in existence, and they are used in the manufacturing of consumer goods, where they are released into the environment and slow to break down (Sunderland et al., 2019).The most common exposure route is ingestion of water and food contaminated with PFAS, which poses a health risk for the mother and fetus during pregnancy.(Sunderland et al., 2019).An estimated 99 % of pregnant people are exposed to perfluorooctanoic acid (PFOA), perfluorooctane sulfonic acid (PFOS), perfluorononanoic acid (PFNA), and perfluorohexane sulfonic acid (PFHxS) (Woodruff et al., 2011).These PFAS have also been found in follicular fluid, amniotic fluid, cord blood, fetal tissues, placenta, and breastmilk (Petro et al., 2014;Zhang et al., 2013;Tao et al., 2008;Panagopoulos Abrahamsson et al., 2021).Studies show that these toxicants increase risk of reproductive disorders, including preeclampsia, miscarriage, preterm birth, and reduced fetal growth, which are the leading morbidities among mothers and infants worldwide (Chambers et al., 2021;Bommarito et al., Oct 2021;Gao et al., 2021).The cause of such adverse pregnancy outcomes is multifactorial, and therefore the underlying molecular mechanisms have been difficult to elucidate, but mounting evidence suggests prenatal PFAS exposure is a contributing factor (Gao et al., 2021).
Oxidative stress in utero impacts the integrity and function of biomolecules and cells, thereby altering the normal course of pregnancy (Go and Jones, 2017;Marseglia et al., 2014).Environmental toxicants are a source of excess reactive oxygen species (ROS) in the perinatal period, as evidenced by recent population-based studies that demonstrate metals, phthalates, organophosphate esters, polycyclic aromatic hydrocarbons, parabens, and phenols are associated with oxidative stress and its sequelae, oxidative damage and inflammation (Aung et al., 2021;Ingle et al., 2020;Ferguson et al., Dec 2016;Watkins et al., 2015).Several adverse pregnancy and birth outcomes that are characterized by elevated levels of oxidative stress have also been linked to a high PFAS body burden (Aouache et al., 2018;Lembo et al., 2021;Eick et al., 2020).Hence, there is growing interest in determining if oxidative stress arises from PFAS exposure during pregnancy.
Experimental research and preliminary epidemiologic studies suggest a relationship between PFAS exposure and oxidative stress based on systemic responses and cellular reactions.For example, blood PFAS levels perturbed redox-related pathways and biomarkers in proteomic, metabolomic, and lipidomic studies conducted in adults, including pregnant people (Salihovic et al., Dec 2020;Lu et al., 2019;Chang et al., 2022).Cross-sectional studies have found that PFAS exposure is positively correlated with circulating levels of inflammatory and prooxidant markers in humans, albeit none have used isoprostanes, well-studied biomarkers of lipid peroxidation (Lu et al., 2019;Omoike et al., 2021;Lin et al., 2020).Other studies have reported higher immune cell counts and altered inflammatory responses among pregnant people, postpartum mothers, and children environmentally exposed to PFAS (Omoike et al., 2021;Zota et al., 2018).In laboratory settings, PFAS have been shown to generate ROS and elicit inflammation in vivo and in vitro (Jiao et al., 2021;Kleszczynski et al., 2009;Ojo et al., 2021;Xu et al., 2022).Taken altogether, the growing body of literature indicates that PFAS exposure promotes oxidative stress, which may be particularly detrimental to maternal and fetal health.
To our knowledge, no study has investigated PFAS as promoters of oxidative stress during pregnancy, an important developmental period that necessitates improved risk assessment.We sought to address this gap by leveraging two integrated, prospective birth cohorts, the Illinois Kids Development Study (IKIDS) and Chemicals In Our Bodies (CIOB), to test the hypothesis that prenatal PFAS exposure is associated with elevated levels of oxidative stress biomarkers during pregnancy.Oxidative stress was assessed through the quantification of urinary F 2 -Isoprostanes (F 2 -IsoPs), which are often considered the "gold standard" biomarkers of oxidative stress since they indicate lipid peroxidation.The NIEHS-sponsored Biomarkers of Oxidative Stress Study (BOSS) compared commonly used biomarkers of oxidative stress and F 2 -IsoPs were shown to reflect endogenous lipid peroxidationa hallmark of oxidative stressin a time-and dose-dependent manner.Importantly, urinary F 2 -IsoPs are stable and specific molecules that are not affected by dietary lipids in humans, allowing for direct comparison across different populations, which is particularly advantageous for epidemiological work (Roberts and Morrow, 2000).Herein, we included 8-isoprostaglandin-F 2α (8-iso-PGF 2α ), an abundant and the most well-studied F 2 -IsoP isomer (Roberts and Morrow, 2000).We additionally included two of the major 8-iso-PGF 2α metabolites, as they may be more sensitive than the parent compound when measured in urine (Dorjgochoo et al., 2012).Finally, we included prostaglandin F-2α (PGF 2α ) as the fourth biomarker since it is isomeric to 8-iso-PGF 2α and enzymatically derived, thereby reflecting inflammation pathways.

Study design and population
Participants included in our study were a subset of participants enrolled in CIOB and IKIDS, which together make up the ECHO.CA.IL cohort (Eick et al., 2021).This subset includes those for whom serum PFAS and urinary levels of oxidative stress biomarkers were available (Total N = 428 with 225 from IKIDS and 203 from CIOB; Figure S1).
A detailed description of the ECHO.CA.IL participants is provided elsewhere (Eick et al., 2021).Briefly, our study population is comprised of pregnant people with diverse sociodemographic backgrounds and geographic locations.Recruitment for IKIDS took place in Urbana-Champaign, Illinois at Christie Clinic and Carle Physicians Group while recruitment for CIOB took place in San Francisco, California at Zuckerberg San Francisco General, Moffitt Long, and Mission Bay Hospitals.Inclusion criteria for IKIDS were as follows: less than 15 weeks gestation, between 18 and 40 years of age, not pregnant with multiples, spoke English as a primary language, low-risk pregnancy, not already enrolled in the study with another child, and resided within a 30-minute drive of Champaign, IL.Pregnant people were eligible for inclusion in CIOB if they were in their second trimester, at least 18 years of age, English or Spanish speaking, and not pregnant with multiples.The Institutional Review Boards at the University of Illinois at Urbana-Champaign approved IKIDS and the University of California, San Francisco, and Berkeley approved CIOB (approval numbers 09498, 10-00861, 2010-05-04, respectively).
Upon enrollment in both cohorts, participants completed selfreported interview questionnaires that were used to ascertain information on maternal age at delivery (years), maternal education (<college degree, college degree, or graduate degree), maternal race and ethnicity (white, Black, Asian/Pacific Islander, Latina, other/multi-racial), current smoking status (yes, no), marital status (married or cohabitating with partner, single).Information regarding parity (no prior births, at least one prior birth) and pre-pregnancy body mass index (BMI, kg/m 2 ) was obtained via medical record abstraction in CIOB and calculated from pre-pregnancy weight and height obtained via questionnaire in IKIDS.

Measurement of PFAS exposure
Serum samples were obtained from participants during the second trimester (median: 18 weeks gestation) and frozen at − 80 • C until analysis.Twelve PFAS were quantified at the Environmental Chemical Laboratory at the California Department of Toxic Substances Control (DTSC) using methods previously described in detail (Morello-Frosch et al., 2016).Briefly, the serum was thawed, then spiked with internal standards, and denatured.Extractions were subsequently obtained from an automated Biotage rapid trace SPE Work Station and injected onto an automated on-line solid phase extraction method coupled to liquid chromatography and tandem mass spectrometry.
Prior to LC/MS analysis, the samples were purified by extraction on a Waters HLB 96-well microelution plate (Waters Corporation, Milford, MA USA).Sample wells were first washed methanol (200uL × 2) followed by 25 % methanol in water (200uL × 2).The sample was then loaded into the well and washed with 400uL of a solution of 0.1 % formic acid/methanol (95/5, v/v).The wells were then washed with 200uL hexanes.Isoprostanes and metabolites were eluted from the plate with 30uL 2-propanol/acetonitrile (50/50, v/v) into a 96-well collection plate containing 30uL water in each well.LC/MS was performed on a Waters Xevo TQ-XS triple quadrupole mass spectrometer connected to a Waters Acquity I-Class UPLC (Waters Corp., Milford, MA USA).Separation of analytes was obtained using a Waters BEH C18 UPLC column (1.0 × 100 mm, 1.7um) with mobile phase A being 0.01 % formic acid in water and mobile phase B acetonitrile.The gradient elution began with 30 % B and progressed linearly to 95 % B over 8 min at a flow rate of 0.300 mL/min.
Because 8-iso-PGF 2α is synthesized from both non-enzymatic lipid peroxidation and prostaglandin-endoperoxide synthases pathways, we conducted a sensitivity analysis where we quantified the proportions of 8-iso-PGF 2α derived from each of the respective pathways using a ratio formula of 8-iso-PGF 2α to PGF 2α (van 't Erve et al., 2015).First, we calculated the ratio of chemical lipid peroxidation (CLP) to prostaglandin-endoperoxide synthases (PGHS) with the equation CLP PGHS = − 0.996*(8− iso− PGF2α/PGF2α) +0.004 0.507*(8− iso− PGF2α/PGF2α) − 0.493 (van 't Erve et al., 2015).Second, we calculated the absolute contribution of PGHS and CLP using a custom interface for the R package "Constrained Linear Mixed Effects" (CLME) that has been described in detail by (van 't Erve et al., 2015).The "chemical fraction" reflects non-enzymatic lipid peroxidation resulting from oxidative stress involving free radicals, whereas the "enzymatic fraction" reflects prostaglandin-endoperoxide synthases and is more reflective of inflammation (van 't Erve et al., 2017).Measured oxidative stress biomarkers with values below the LOD were imputed with LOD/ ̅̅̅ 2 √ (Hornung and Reed, 1990).We accounted for urinary dilution by correcting for specific gravity with the equation ; where SpG Median is the SpG population median (1.012 and 1.015 for CIOB and IKIDS, respectively).Ox is the uncorrected oxidative stress biomarker level and Ox c is the SpG-corrected oxidative stress biomarker level.For participants with oxidative stress biomarkers available at both timepoints (N = 428 unique participants, of which 225 had a measurement available at both timepoints; Figure S1), we used the geometric mean (GM) of the biomarker concentrations (Eick et al., 2020) as the primary outcome in our mixture analyses.If oxidative stress biomarkers were measured once, we used only that measure.

Statistical analysis
We analyzed the distribution of demographic characteristics using descriptive statistics and reported them as mean (±standard deviation [SD]) or frequency (%).The distribution of serum PFAS and urinary oxidative stress biomarkers were examined with detection frequency percentiles, GMs, and geometric standard deviations (GSDs).We natural log-transformed PFAS for downstream analyses to address positive skewness and improve model fits.Pearson correlation coefficients were estimated for PFAS concentrations and averaged oxidative stress biomarkers.
We first modeled the unadjusted and adjusted associations between serum PFAS concentrations and urinary oxidative stress biomarkers levels using linear mixed-effects models with a random intercept for participant ID.This analytical approach allowed us to account for the repeated measures of oxidative stress biomarkers and maximize the usage of all available data.Covariates retained in adjusted models were selected based on the construction of a directed acyclic graph (DAG; Figure S2) and included maternal age, education, parity, pre-pregnancy BMI, gestational week of sample collection, and cohort.In linear mixedeffects models, PFAS were standardized to the population's interquartile range (IQR).All statistical analyses were performed in R (Boston, MA, USA, Version 4.1.0).

Mixture analysis
We examined the joint effects of seven PFAS in relation to oxidative stress biomarkers using two approaches for modeling and parametrizing exposure mixtures, quantile g-computation and Bayesian kernel machine regression (BKMR).Together, these models inform the partial and cumulative dose-response relationships between a mixture of environmental chemicals and health outcomes, which are often obscured by single pollutant models.Mixture models can also account for highly correlated co-exposures.All mixture models were adjusted for maternal age, education, parity, pre-pregnancy BMI, and cohort, and run on complete cases (N = 428; Figure S1).
Our first approach, quantile g-computation, uses a parametric, generalized linear model-based implementation of g-computation (Keil et al., 2020).Quantile g-computation estimates the overall change in an oxidative stress biomarker corresponding to a simultaneous one-quartile increase in all PFAS included in the mixture.Quantile g-computation also allows each exposure in the mixture to have a positive or negative effect on the outcome, which is reflected with relative weights for each exposure that sum to one (Keil et al., 2020).
In our second approach, we implemented BKMR to identify potential linear and non-linear associations and interactions between exposures.BKMR estimates a non-parametric high-dimensional exposure-response function using kernel machine regression (Bobb et al., 2015).BKMR was performed with 10,000 iterations and checked for convergence with the Markov chain Monte Carlo procedure.Linearity was assessed by examining univariate exposure-response functions, which reflect the relationship between levels of a single PFAS and levels of oxidative stress biomarkers, holding the remaining PFAS in the mixture constant at the median value.We assessed interaction using bivariate exposure-response functions, which indicate evidence of interaction when lines are not parallel and/or overlapping.We calculated posterior inclusion probabilities (PIPs) to determine the relative importance of a given PFAS exposure on oxidative stress biomarkers.A traditional threshold of 0.5 was used to identify "important" predictors (Cathey et al., 2021).We then estimated the overall effect of the PFAS mixture by comparing the expected change in individual oxidative stress biomarkers when all PFAS exposures were set at the first and third quartiles versus fixed at the median value.
To determine if our results were robust against the influence of one of the two cohorts in ECHO.CA.IL, we conducted a series of sensitivity analyses in which linear mixed-effects, quantile g-computation, and BKMR models were stratified by cohort.Cohort-specific models were adjusted for maternal age, education, parity, and pre-pregnancy BMI.We also analyzed the single pollutant associations with 8-iso-PGF 2α using the LOD/ ̅̅̅ 2 √ imputation procedure for PFAS detected below the LOD.

Results
The sociodemographic and clinical characteristics of the pregnant people included in this analysis were representative of all pregnant people enrolled in the ECHO.CA.IL cohort (Table 1).Of the 428 participants included in our analytic sample, 63.6 % self-identified as white and 16.1 % self-identified as Latina.Most participants were married or cohabitating with their partner (93.2 %) and had attained a college or graduate degree (75.3 %).On average, the participants were 32 years of age and delivered at 39 weeks gestation (SD = 2.1) (Table 1).Approximately half of the participants had at least one prior pregnancy (53.5 %) and delivered a female infant (54.4 %).The percentage of participants with gestational diabetes and gestational hypertension in the analytic sample was 15.7 % and 7.0 %, respectively.The sociodemographic characteristics were similar across the cohorts, although CIOB was more racially and ethnically diverse and a larger percentage of IKIDS participants had a graduate degree (Table S1).
Based on the magnitude of the weights obtained from quantile gcomputation, the chemical fraction contributed the most to the overall mixture effect on 8-iso-PGF 2α , which reflects the contribution of oxidative stress (Table 3).Both of the 8-iso-PGF 2α metabolites were inversely associated with the PFAS mixture, but 95 % confidence intervals included the null.Across oxidative stress biomarkers, PFOS was consistently identified as having the largest positive weight while PFNA was only assigned a negative weight (Fig. 1B).PFUdA had a positive partial effect, as suggested by its quantile g-computation weight, on the enzymatic fraction, which contrasted with the linear mixed-effect model results.This difference may be due to the consideration of co-exposures  Abbreviations: SD, standard deviation; LOD, limit of detection.Note: There were 425 samples from visit 1 and 228 samples from visit 2. The geometric mean, standard deviation, and percentiles urine concentrations were corrected for specific gravity.Geometric mean, geometric SD, and percentile values use the machine read value if it was available.If there was no machine read value, missing values were replaced with LOD/square root of 2. in the PFAS mixture analysis.In models stratified by cohort, increasing all PFAS in the mixture by one quartile was associated with slightly elevated levels of oxidative stress biomarkers among IKIDS only (Table S6).We observed PFOS to be the strongest driver of the overall mixture effect in both ECHO.CA.IL cohorts.Similarly, PFOA was assigned a positive weight across all oxidative stress biomarkers among IKIDS participants (Figure S4A).Using BKMR, we observed a non-significant, inverse association between the PFAS mixture and levels of 2,3-dinor-8-iso-PGF 2α and 2,3dinor-5,6-dihydro-8-iso-PGF 2α (Fig. 1C).This trend was pronounced in CIOB relative to IKIDS (Figure S4B).The univariate exposure-response functions showed that for all oxidative stress biomarkers, PFOS had a positive, linear relationship whereas PFNA had a negative, linear relationship (Fig. 2).We did not observe interaction, as all PFAS had parallel bivariate exposure-response functions (Figures S5 -S10).Finally, BKMR did not identify any PFAS as being "important" exposures since all PIPs were < 0.5 and ranged from 0.01 to 0.49 in the overall sample (Table S7).

Table 3
Quantile g-computation estimates and 95 % confidence intervals for the change in urinary oxidative stress biomarkers corrected with specific gravity for a one quantile increase in the per-and polyfluoroalkyl substances mixture in the ECHO.CA.IL cohort, 2014 -2019 (N = 428).

Fig. 2.
Univariate exposure-response functions and 95 % credible intervals for the change in urinary oxidative stress biomarker levels resulting from individual PFAS while holding all the remaining exposures in the mixture fixed at their median concentration, estimated using Bayesian kernel machine regression (BKMR) (N = 428).Note: Models are adjusted for maternal age, education, parity, pre-pregnancy body mass index, and cohort.

Discussion
The objective of our study was to examine the associations between prenatal PFAS exposure and oxidative stress biomarkers among pregnant people enrolled in the ECHO.CA.IL cohort.We found the PFAS exposure mixture was not strongly related to urinary levels of isoprostanes during pregnancy.However, there was a consistent and positive trend between PFOS and biomarkers of oxidative stress throughout our single pollutant and mixture analyses, which warrants further investigation.
In the present analysis, we estimated a modest, yet non-significant association, between the PFAS mixture and 8-iso-PGF 2α .Our findings point to the difficulty of studying oxidative stress from environmental chemical exposure in humans.It is well-documented that PFAS induce ROS and propagate oxidative stress events, including lipid peroxidation, DNA damage, mitochondrial destabilization, reduced antioxidant capacity, and cytotoxicity in vivo and in vitro.For example, a mouse study found exposure to PFOA elevated pancreatic and hepatic levels of 8-iso-PGF 2α (Kamendulis et al., 2014).In primary tilapia hepatocytes, PFOA also induced maleic dialdehyde, a biomarker of lipid peroxidation, in association with increased cytotoxicity, ROS, and activities of key antioxidant enzymes (e.g., superoxide dismutase, catalase and glutathione reductase) (Liu et al., 2007).The two most commonly detected PFAS in humans, PFOA and PFOS, generate ROS and impact antioxidant defenses in zebrafish embryos and human cells from the liver, microvascular endothelium, and umbilical vein (Shi and Zhou, 2010;Wielsoe et al., 2015;Qian et al., 2010;Liao et al., 2012).Exposure to PFAS is hypothesized to overwhelm and destabilize the mitochondria, thereby limiting the effective management of ROS (Jiao et al., 2021;Souders et al., 2021).Prolonged dysregulation of free radicals and antioxidant defenses in the mitochondria may prompt a cascade of cell death signals and pathways, ultimately resulting in cytotoxic effects (Kleszczynski et al., 2009).Select studies have suggested PFAS toxicity occurs independently of oxidative stress but there are multiple lines of evidence to support these toxicants induce oxidative stress (Ojo et al., 2021;Crebelli et al., 2019).Given the inconsistencies in experimental literature and paucity of epidemiologic studies, additional research is needed to improve our understanding of the redox mechanisms impacted by PFAS exposure.
A limited body of research has investigated the relationships between PFAS exposure and oxidative stress in humans, with no studies conducted among pregnant people or with 8-iso-PGF 2α , a highly sensitive and specific biomarker of oxidative stress.Prior work using National Health and Nutrition Examination Survey data from 2005 and 2012 found that circulating levels of PFOS, PFNA, and PFHxS were positively correlated with iron, albumin, bilirubin, which suggests that PFAS exposure leads to an increase in circulating antioxidants, presumably to counteract oxidative stress (Omoike et al., 2021).Another cohort of adults from the general population in Taiwan demonstrated that serum PFAS concentrations increase levels of 8-hydroxy-2-deoxyguanosine (8-OHdG) and 8-nitroguanine, two biomarkers of oxidative and nitrative stress (Lin et al., 2020).Similarly, a randomized controlled trial of 141 senior Koreans with elevated serum PFOS and PFDoDA concentrations also had higher urinary levels of malondialdehyde and 8-OHdG (Kim et al., 2016).Epidemiologic studies conducted in occupational settings have found comparable results.For example, forty adult workers at a Chinese fluorochemical manufacturing plant had depleted methionine sulfoxide activity, a potent scavenger of ROS (Lu et al., 2019).A separate investigation performed in China identified several relationships between environmental exposure to PFAS and perturbed metabolomic pathways involved in oxidative and nitrative stress among 181 adult males (Wang et al., 2017).While our study used different biomarkers of oxidative stress, our results build upon previous findings that exposure to certain PFAS may be associated with a rise in oxidative stress.
Prior work, including work in this study population, has shown that prenatal PFAS exposure is linked to an increased risk of preterm birth and restricted fetal growth (Eick et al., 2022;Eick et al., 2020).Systematic reviews and meta-analyses of pregnancy cohorts have additionally found mothers exposed to higher levels of PFAS are more likely to experience reproductive tract dysfunction and give birth to children with acute and chronic health issues (Chambers et al., 2021;Rickard et al., 2022;Starnes et al., 2022;Ding et al., 2020;Liew et al., 2018).Induction of oxidative stress may provide a mechanistic explanation for these associations.Urinary levels of isoprostanes in our study population are comparable to other studies which have found increased risk of adverse birth outcomes, including preterm birth (Eick et al., 2020;Arogbokun et al., 20212021;Morales et al., 2022;Eick et al., 2022).For example, prospective cohort studies of pregnant people in the mainland United States and Puerto Rico have demonstrated that elevated levels of 8-iso-PGF 2α and 2,3-dinor-5,6-dihydro-8-iso-PGF 2α are associated with any preterm birth as well as preterm birth of spontaneous origin (Eick et al., 2020;Rosen et al., 2019).Second trimester 8-iso-PGF 2α levels are also higher in preeclamptic pregnancies compared to normotensive pregnancies.(Ferguson et al., 2017) Pregnant people with increased levels of 8-iso-PGF 2α , PGF 2α , and 2,3-dinor-5,6-dihydro-8-iso-PGF 2α have a greater risk of developing intra-amniotic infection and gestational diabetes and delivering small-for-gestational age and lower birthweight infants (Arogbokun et al., 20212021;Hsieh et al., 2012;Park et al., 2016;Kapustin et al., 2020).Collectively, these adverse pregnancy and birth outcomes constitute a substantial portion of the disease burden in pregnant populations and many have been linked to prenatal PFAS exposure.Further, infants may be uniquely sensitive to changes in levels of isoprostanes, which have bioactive roles in vascularization and respiratory development in the postnatal period (Matthews et al., 2016;Chen et al., 2012).Hence, it remains an important public health initiative to identify risk factors for elevated oxidative stress during pregnancy, as this information may aid in the development of preventive, diagnostic, and therapeutic targets for improved maternal and child health.
The present analysis has several strengths.First, we leveraged single pollutant models to account for the repeated measures of oxidative stress biomarkers in the second and third trimesters among almost half of the pregnant participants.Evidence obtained from the linear mixed-effects models was further enriched by the two mixture modeling approaches, which enabled us to characterize partial and cumulative dose-response relationships.Single pollutant models may not accurately portray reallife exposures, as in reality, we are exposed to complex mixtures of environmental chemicals, the components of which may interact or have a joint effect on health outcomes.Second, urinary oxidative stress biomarkers were assessed using mass spectrometry.This method is preferred over blood samples analyzed by enzyme-linked immunoassays, which are subject to auto-oxidation during storage (Morrow et al., 1990;Klawitter et al., 2011).Third, we included two metabolites of 8iso-PGF 2α as they are hypothesized to be more sensitive than the parent compound when measured in urine (Dorjgochoo et al., 2012).Lastly, we examined the chemical and enzymatic fractions of 8-iso-PGF 2α , which allowed us to quantify the proportion of 8-iso-PGF 2α derived from oxidative stress and inflammation pathways.
We also acknowledge several limitations in our study.Our sample size is relatively small, which limited our statistical power, and this imprecision is reflected in our confidence intervals.Short chain PFAS, including PFDoA, PFOSA, PFBS, Et-PFOSA-AcOH, and PFHpA, were also detected in less than 50 % of our participants, which hindered our ability to examine their effects on oxidative stress biomarkers.Further, we did not correct for multiple comparisons, but rather focused on identifying patterns, as opposed to an overreliance on statistical significance (Rothman, Jan 1990).The mixture methods applied in our analysis do not account for toxicological effects of each PFAS congener.We also do not have information on participants' physical activity, which may have an effect on oxidative stress levels.Lastly, as with all observational studies, our results may not be generalizable to other populations and may be subject to residual confounding.

Conclusion
To our knowledge, this is the first study to examine associations between prenatal PFAS exposure mixtures and oxidative stress biomarker levels.Among participants enrolled in the ECHO.CA.IL birth cohort, we observed that prenatal PFOS exposure was associated with modestly elevated levels of oxidative stress biomarkers.The six additional PFAS included in our analysis had null and/or weak associations with isoprostanes.Our findings advance the current state of science on how PFAS, in particular PFOS, may affect maternal redox homeostasis during pregnancy.Quantification of additional PFAS not measured in our study may reveal greater insight into cumulative effects on oxidative stress.Further research is needed to confirm our findings in other populations with larger sample sizes and should investigate oxidative stress as a mediating pathway between PFOS exposure and adverse reproductive and perinatal health outcomes.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Fig. 1 .
Fig. 1.Associations between serum levels of per-and polyfluoroalkyl substances (PFAS; ng/mL) and urinary levels of specific gravity corrected oxidative stress biomarkers (ng/mL) during pregnancy in the ECHO.CA.IL cohort, 2014 -2019 (N = 428).(A) Beta and 95 % confidence interval indicating the change in oxidative stress biomarkers in association with an interquartile range increase in individual PFAS, estimated using linear mixed-effect models; (B) Positive and negative weights representing the partial effects of PFAS in the mixture on individual oxidative stress biomarkers, estimated using quantile g-computation; (C) Cumulative effect and 95 % credible intervals of the PFAS mixture on individual oxidative stress biomarkers, estimated using Bayesian kernel machine regression (BKMR).Note: Linear mixed-effects models adjusted for maternal age, education, parity, pre-pregnancy body mass index, cohort, gestational weeks at sample collection, and including a random intercept included for participant ID.Quantile g-computation and BKMR models are adjusted for maternal age, education, parity, pre-pregnancy body mass index, and cohort.

Table 1
Characteristics of the pregnant people in this analysis and all pregnant people in the ECHO.CA.IL cohorts, 2014 -2019.

Table 2
Distributions of second trimester serum levels of per-and polyfluoroalkyl substances (PFAS; ng/mL) and second and third trimester urinary oxidative stress biomarkers corrected with specific gravity (ng/mL) in the ECHO.CA.IL cohort, 2014 -2019.