Association of per- and polyfluoroalkyl substances with the antioxidant bilirubin across pregnancy

Abstract

Background:

In mechanistic and preliminary human studies, prenatal exposure to per- and polyfluoroalkyl substances (PFAS) is associated with oxidative stress, a potential contributor to maternal liver disease. Bilirubin is an endogenous antioxidant abundant in the liver that may serve as a physiological modulator of oxidative stress in pregnant people. Hence, our objective was to estimate the association between repeated measures of PFAS and bilirubin during pregnancy.

Methods:

The study population included 332 participants in the Atlanta African American Maternal-Child Cohort between 2014-2020. Serum samples were collected up to two times (early pregnancy: 6-18 gestational weeks; late pregnancy: 21-36 gestational weeks) for the measurement of perfluorohexane sulfonate (PFHxS), perfluorooctane sulfonate (PFOS), perfluorooctanoic acid (PFOA), perfluorononanoic acid (PFNA), and total bilirubin. We analyzed single PFAS with linear mixed effect regression and a mixture of the four PFAS with quantile g-computation. Models were repeated with a multiplicative interaction term to explore effect modification by study visit.

Results:

Overall, PFHxS was positively associated with bilirubin (β=0.08, 95% CI=0.01, 0.15). We also found during late pregnancy, there was a positive association of PFHxS and the PFAS mixture with bilirubin (β=0.12, 95% CI=0.02, 0.22; ψ=0.19, 95% CI=0.03, 0.34, respectively). Finally, study visit modified the PFOA-bilirubin association (interaction p-value=0.09), which was greater during early pregnancy (β=0.08, 95% CI=0.01, 0.15).

Conclusion:

In a prospective cohort of pregnant African Americans, an increase in PFOA, PFHxS, and the PFAS mixture was associated with an increase in bilirubin. Our results suggest that, depending on pregnancy stage, prenatal PFAS exposure disrupts the maternal liver antioxidant capacity.

I. INTRODUCTION

Per- and polyfluoroalkyl substances (PFAS) is a family that contains more than 12,000 different chemicals with strong carbon-fluorine bonds.¹ Since the invention of PFAS in the 1930s, many commercial goods and consumer products have been produced with these chemicals to resist heat, water, and oil.^{2, 3} The widespread use of PFAS has led to environmental pollution, which is exacerbated by their long half-lives in the environment.^{4, 5} Likewise, adult human exposure most often occurs through ingestion of food and drinking water, which has prompted intense investigation into PFAS removal strategies.^{6, 7} Organs in the digestive tract are the primary biological endpoints of PFAS toxicity, such as the liver, even before birth.^{8, 9} Population-based investigations have found an association between exposure to PFAS and liver dysfunction, liver disease, and liver cancer.^10–12 Given the upward trend in adult liver disease cases that require hospitalization, especially among pregnant people (2002-2004: 5.97 per 1,000; 2008-2010: 8.91 per 1,000), a major public health goal is to understand the role PFAS play in their pathogenesis and severity.^13–15

Every year in the United States (US), more than 3% of pregnant people are diagnosed with a pregnancy-related liver disorder, including intrahepatic cholestasis, acute fatty liver, HELLP syndrome (hemolysis, elevated liver enzymes, and low platelets), hyperemesis gravidarum, and pre-eclampsia.^{16, 17} The latter involves multiple systems, but most often presents as right upper quadrant abdominal pain (i.e., liver location) or abnormal liver function tests (i.e., elevated transaminase levels), particularly among patients with HELLP syndrome, which is considered a variant of severe pre-eclampsia cases. Non-alcoholic fatty liver disease (NAFLD) is the most common cause of pregnancy-related liver disorders and tripled among pregnant people from 2007 to 2015.¹⁸ Exposure to PFAS has been linked to NAFLD and pre-eclampsia in experimental and epidemiological research.^19–21 There is also a link between PFAS and bilirubin, a prognostic biomarker of pregnancy-related liver disorders and NAFLD, in non-pregnant adults.^22–29

Bilirubin is formed by the catabolism of heme, an iron-containing porphyrin that binds oxygen to various proteins, including cytochrome, catalase, hemoglobin, myoglobin, peroxidase, and pyrrolase.³⁰ Total bilirubin is a clinical measurement of indirect bilirubin, which exists in the bloodstream as a hydrophobic complex with albumin, plus direct bilirubin, which exists in the liver as a hydrophilic conjugate with glucuronic acid, before eventual excretion in bile acids.³¹ This endogenous antioxidant inhibits NADPH oxidase (NOX) complexes and neutralizes reactive oxygen species (ROS), particularly those which damage lipids via peroxidation, in the liver.³² Hence, bilirubin is critical to hepatic redox homeostasis during pregnancy, when mitochondrial respiration is high to meet the maternal-placental-fetal bioenergetic demands. ^{33, 34}

Similar to bilirubin, PFAS are attracted toward albumin and soluble in water, plus various chemicals in this family are linked to bile acid dysregulation.^35–37 There is also biological plausibility of an association between PFAS and bilirubin during pregnancy; prenatal exposure is associated with non-enzymatic oxidative stress pathways (e.g., lipid peroxidation), non-enzymatic antioxidants (e.g., albumin), non-enzymatic prooxidants (e.g., 8-isoprostane-prostaglandin F_2α), and disorders of lipids produced by the liver (e.g., cholesterol, triglycerides).^38–44 However, little is known about how repeated exposure to single and multiple PFAS affects bilirubin among pregnant people. Oxidative stress during late pregnancy is an important risk factor for preterm birth and other adverse birth outcomes associated with PFAS exposure, suggesting that a critical window of susceptibility may exist.^{38, 45–48} This information is needed for public health programs and clinical interventions, which would be easy to scale with bilirubin because the antioxidant biomarker is measured in the comprehensive metabolic panel, a lab test routinely ordered at prenatal visits.⁴⁹

The objective of this study was to estimate the association of four PFAS and their mixture with bilirubin during pregnancy in the Atlanta African American Maternal-Child Cohort. Our hypothesis was that higher serum concentrations of single and multiple PFAS would be associated with higher total serum bilirubin levels. We also explored effect modification by study visit, hypothesizing a stronger association at the second visit (range: 21-36 gestational weeks) than the first visit (range: 6-18 gestational weeks), based on previous reports of greater changes in oxidative stress biomarkers from prenatal PFAS exposure during late pregnancy.

II. MATERIALS AND METHODS

Study Population

Our study population included 332 participants from the Atlanta African American Maternal-Child Cohort, which is an ongoing, prospective birth cohort that has been previously described in detail.^{50, 51} Participants in the present analysis were recruited at Emory Midtown Hospital or Grady Memorial Hospital between 2014-2020. People were eligible for inclusion if presented with a singleton pregnancy, self-identified as an African American woman, aged 18-40 years, and had the ability to communicate in English. Another eligibility criterion was no diagnosis of chronic medical conditions, no chronic use of prescription medications, and no evidence of viral hepatitides on standard prenatal screening. All participants provided written informed consent prior to enrollment. The Emory University Institutional Review Board approved this study (approval reference #: 68441).

Data and blood samples were collected at enrollment (mean 11 weeks gestation, range=6–18) and at follow-up (mean 27 weeks gestation, range=21–36) from 332 participants. A total of 327 participants visited once during early pregnancy and a subset of 256 participants visited again during late pregnancy. The flowchart of participants included in this analysis is presented in Figure S1.

Collection of Covariate Data and Serum Samples

Participant data were collected by validated questionnaires and medical record abstraction. Sociodemographic characteristics included maternal age at enrollment (years), marital status (single, married or cohabitating), medical insurance type (private, Medicaid), highest level of education (<high school, high school, some college, college or graduate school), and self-reported any use of alcohol, tobacco, or marijuana in the month prior to enrollment (yes, no). We calculated an income-poverty ratio with household income and household size relative to the Federal Poverty Level. Clinical characteristics included body mass index (BMI) according to the clinical measurement of weight (kg) and height (m²) at enrollment, parity, and infant sex assigned at birth. None of the sociodemographic characteristics were time-varying. Venous blood samples were also collected at both study visits and centrifuged for serum before storage at −80 ◦C.

Measurement of PFAS

Maternal serum PFAS concentrations were measured by the Wadsworth Center Human Health Exposure Analysis Resource (HHEAR) Laboratory Hub and the Laboratory of Exposure Assessment and Development for Environmental Research (LEADER) at Emory University. Both laboratories participated in the Children’s Health Exposure Analysis Resource (CHEAR), which is supported by the National Institute of Environmental Health Sciences, and followed the same protocol for quality control and data harmonization.^{44, 52, 53}

The measurement protocol was as follows: (1) spike serum samples with internal standards; (2) perform solid phase extraction; (3) analyze with liquid chromatography interfaced with tandem mass spectrometry (LC-MS/MS); (4) quantify PFAS with isotope dilution calibration. Detailed methods are available elsewhere ^{51, 54} We retained PFNA, PFOA, PFOS, and PFHxS for primary and supplemental analyses based on >50% detection rates at both study visits. We imputed the serum PFAS concentrations detected below the limit of detection (LOD) with the LOD/$\sqrt{2}$.⁵⁵ All measurements were natural log-transformed to address right skewness.

Measurement of Bilirubin

Maternal total serum bilirubin concentrations were measured by the reference standardization of metabolomic signal intensities.^{56, 57} An established protocol of liquid chromatography coupled with high-resolution mass spectrometry (LC-HRMS) (Thermo Scientific^™ Q-Exactive^™ HF) was performed by the Clinical Biomarker Laboratory at Emory University to conduct untargeted high-resolution metabolomics.^{58, 59} First, serum samples were analyzed in triplicate with the hydrophilic interaction liquid chromatography (HILIC) using a positive electrospray ionization (ESI) analytical column. Second, the bilirubin intensity for each participant (HILIC mass to charge ratio (m/z)=114.0662, retention time (RT)=29.4) was normalized to the averaged intensity of the National Institute of Standards and Technology (NIST) sample. Third, the known bilirubin concentration in the NIST reference material was used to convert the normalized bilirubin intensities into molar concentrations (μmol/L). All measurements were natural log-transformed to address the observed right skewness in data prior to analysis.

Statistical Analyses

We performed all statistical tests in R (Boston, MA, USA, Version 4.1.0). The statistical significance threshold was set to p-value<0.05 for main effects and p-value<0.10 for interaction effects a priori.

Our study population included participants with up to two measurements of the exposure and outcome, and complete observations for the modifier and potential confounders. The visit-specific distributions of maternal serum PFNA, PFOA, PFOS, PFHxS, and bilirubin were examined with detection frequencies, select percentiles, and violin plots. We also estimated the Spearman correlation coefficient for each pair of PFAS by prenatal study visit. To estimate the individual variability of PFAS and bilirubin between the study visits, we calculated intraclass correlations (ICC). The concentration changes for all analytes were visualized with generalized additive mixed models throughout pregnancy (mgcv package).

Primary Analysis

We used linear mixed effect regression (lme4 package) as the single chemical model, which included one of the four PFAS analyzed (i.e., PFNA, PFOA, PFOS, or PFHxS), and quantile g-computation (qgcomp package) as the chemical mixture model, which included all four PFAS.⁶⁰ Both models allow for analysis of repeated measures and effect modification, plus their results are easily compared to each other given similar interpretations. Specifically, linear mixed effect regression estimates (β) are interpreted as the average change in one natural log-unit of bilirubin for an increase by one natural log-unit of PFNA, PFOA, PFOS, or PFHxS, while quantile g-computation estimates (ψ) are interpreted as the average change in one natural log-unit bilirubin for a simultaneous increase by one-quartile natural log-unit of PFNA, PFOA, PFOS, and PFHxS. This modeling approach allowed us to explore critical windows of vulnerability and preserve the longitudinal nature of our data.

Quantile g-computation models were parameterized with 4,000 Monte Carlo simulations and 400 bootstraps. All models were fitted with a random intercept for participant ID to consider intraindividual correlation of repeated measures and adjusted for gestational age at sample collection to consider interindividual concentration differences between timepoints. In the fully adjusted models, we included maternal age, parity, education, and use of alcohol, tobacco, or marijuana in the month prior to enrollment as potential confounders based on the construction of a directed acyclic graph (DAG; Figure S2). In a separate, previous study, we demonstrated that hemodynamic factors do not confound the association between serum PFAS concentrations during late pregnancy and adverse pregnancy outcomes.⁵⁶ Hence, we did retain prenatal PFAS exposures at the second visit and did not adjust for pregnancy-related hemodynamics in any of the models.

For the effect modification analysis, we used the lme4 package and qgcompint package.⁶⁰ In a similar procedure performed by others, a multiplicative interaction term between PFAS (i.e., single chemical or chemical mixture) and study visit (i.e., early pregnancy or late pregnancy) was included in the fully adjusted models.⁶¹ To obtain an estimate specific to early pregnancy, we report the main effect when the modifier referent was set to the first visit, then repeated the model with the second visit as the referent for the estimate specific to late pregnancy.

Supplemental Analysis

We examined study visit as an effect modifier for the association between prenatal PFAS exposure and total bilirubin in two ways: using an interaction term in the primary analysis and stratification in a supplemental analysis. The fully adjusted models were stratified by study visit and the reported estimates were the main effects only. For another supplemental analysis, we excluded participants with pre-eclampsia (N=26) from the analytic sample since bilirubin rises in the setting of this pregnancy complication. The third supplemental analysis included maternal BMI at enrollment as a potential confounder, in addition to the other covariates in the fully adjusted models. There is conflicting evidence about whether adiposity is a ‘cause’ or ‘effect’ of PFAS and their metabolism.⁶² Chemicals that belong to this family disrupt endocrine and metabolic processes in adipose tissue, so exposure may change how energy is created, used, and stored.^{63, 64} An alternative hypothesis is a person’s BMI, the non-invasive surrogate measure for adipose tissue, may alter how PFAS are metabolized. Next, we repeated the primary analysis among participants with analyte measurement data available at both study visits to evaluate any changes in the results (N=251). The final supplemental analysis was conducted with serum PFAS concentrations at the first visit and total bilirubin at the second visit. These single measure models minimize the potential for reverse causality from any changes that occur between the two timepoints.

III. RESULTS

Study Population

Characteristics of the study population are presented in Table 1. Among the 332 pregnant people in this analysis, 52 (15.7%) attained at least a college degree, 127 (38.3%) had a normal weight BMI between 18.5-24.9 kg/m², 149 (44.9%) were primiparous, and 179 (53.9%) self-reported no recent use of alcohol, tobacco, or marijuana. At enrollment, the average age of pregnant people was 25 years (Standard Deviation [SD]=4.8) and the average age of the fetus was 11 gestational weeks (SD=2.2). Among the 256 who completed a follow-up visit during late pregnancy, the average age of the fetus was 27 gestational weeks (SD=2.5).

Table 1.

Characteristics of study population from the Atlanta African American Maternal-Child Cohort, 2014 – 2020 (N=332).

Characteristic	N (%) or Mean (SD)
Gestational Weeks at Early Pregnancy Study Visit	11 (2.2)
Gestational Weeks at Late Pregnancy Study Visit	27 (2.5)
Gestational Weeks at Delivery	38 (3.1)
Age at Enrollment (years)	25 (4.8)
Marital Status
Married or Cohabitating	158 (47.6%)
Single	174 (52.4%)
Education Level
<High School	54 (16.3%)
High School	125 (37.7%)
Some College	101 (30.4%)
College or Graduate School	52 (15.7%)
Alcohol, Tobacco, and Marijuana Use
None	179 (53.9%)
Any	153 (46.1%)
Prenatal Body Mass Index
Underweight (<18.5 kg/m2)	10 (3.0%)
Normal Weight (18.5-24.9 kg/m2)	127 (38.3%)
Overweight (25.0-29.9 kg/m2)	74 (22.3%)
Obese (≥30 kg/m2)	121 (36.4%)
Parity
0	149 (44.9%)
1	92 (27.7%)
≥2	91 (27.4%)
Hospital
Emory University Hospital – Midtown	131 (39.5%)
Grady Memorial Hospital	201 (60.5%)
Medical Insurance
Medicaid	264 (79.5%)
Private	68 (20.5%)
Infant Sex
Male	167 (50.3%)
Female	165 (49.7%)
Pre-eclampsia
No	306 (92.2%)
Yes	26 (7.8%)

Abbreviations: N, number; SD, standard deviation.

Analyte Distributions

The distributions of maternal serum concentrations for PFAS and bilirubin are presented in Table 2 and Figure 1. We detected PFNA, PFOA, PFOS, and PFHxS in ≥94% of early pregnancy samples and ≥72% of late pregnancy samples. PFHxS was the most abundant (Geometric Mean [GM]=3.73 ng/mL; GSD=1.89) and most variable (ICC=0.11, p=0.17) among the 332 participants. This pattern was also evident in the generalized additive mixed models and violin plots. The Spearman correlations between PFHxS and the three other PFAS were strong during early pregnancy but weak during late pregnancy (Figure S3). Finally, the geometric mean of total serum bilirubin was not clinically elevated (i.e., ≥18.9 μmol/L) at either study visit.⁶⁵

Table 2.

Distributions of serum biomarkers in study population from the Atlanta African American Maternal-Child Cohort, 2014 – 2020.

	Early Pregnancy Study Visit (N=327)				Late Pregnancy Study Visit (N=256)				Overall (N=332)
Biomarker	% Detected	Min	Max	GM (GSD)	% Detected	Min	Max	GM (GSD)	ICC	p-Value
PFOS	97.29	<LOD	12.42	1.96 (2.13)	75.30	<LOD	19.20	2.75 (2.93)	0.43	<0.001
PFOA	96.08	0.03	4.42	0.62 (2.35)	72.59	<LOD	6.76	0.51 (3.35)	0.61	<0.001
PFHxS	94.88	0.08	4.80	0.99 (1.92)	76.51	<LOD	15.64	3.73 (1.89)	0.11	0.17
PFNA	94.58	<LOD	2.27	0.23 (2.39)	75.90	0.02	3.25	0.41 (2.05)	0.50	<0.001
Bilirubin	-	1.01	49.20	12.01 (1.80)	-	0.39	35.22	9.73 (1.79)	0.45	<0.001

Abbreviations: PFHxS, perfluorohexane sulfonate; PFOS, perfluorooctane sulfonate; PFOA, perfluorooctanoic acid; PFNA, perfluorononanoic acid

Notes: All PFAS reported in ng/mL. Bilirubin reported in μmol/L. Bold font denotes p<0.05.

Figure 1.

Serum concentrations of PFAS and bilirubin during pregnancy in the Atlanta African American Maternal-Child Cohort, 2014 – 2020.

Abbreviations: PFAS, per- and polyfluoroalkyl substances; PFHxS, perfluorohexane sulfonate; PFOS, perfluorooctane sulfonate; PFOA, perfluorooctanoic acid; PFNA, perfluorononanoic acid; w, weeks

Note: Generalized additive mixed models and violin plots were constructed to show the distributions of (A) maternal serum PFAS and (B) maternal total serum bilirubin. Visit 1 occurred during early pregnancy and Visit 2 occurred during late pregnancy.

Overall Association between PFAS and Bilirubin during Pregnancy

There were consistent patterns of association between the minimally adjusted models (Table S1) and fully adjusted models (Table S2). A one natural log-unit increase in PFHxS was associated with a 0.08 natural log-unit increase in bilirubin (95% Confidence Interval (CI)=0.01, 0.15) across pregnancy (Figure 2). A simultaneous increase by one-quartile natural log-unit in the PFAS exposure mixture yielded a comparable change in bilirubin (β=0.09; 95% CI=−0.004, 0.18). Null associations were found between repeated measures of PFOS, PFOA, and PFNA with bilirubin.

Figure 2.

Adjusted associations between PFAS and bilirubin during pregnancy in the Atlanta African American Maternal-Child Cohort, 2014 – 2020.

Abbreviations: PFAS, per- and polyfluoroalkyl substances; PFHxS, perfluorohexane sulfonate; PFOS, perfluorooctane sulfonate; PFOA, perfluorooctanoic acid; PFNA, perfluorononanoic acid.

Note: Linear mixed effect regression models fitted with single PFAS. Quantile g-computation models fitted with PFAS mixture. PFAS mixture included PFNA, PFOA, PFOS, and PFHxS. Visit 1 occurred during early pregnancy and Visit 2 occurred during late pregnancy.

Effect Modification by Study Visit for the Association between PFAS and Bilirubin during Pregnancy

In the single chemical models, study visit modified the association between prenatal PFOA exposure and total serum bilirubin (interaction p-value=0.09), which was greater in magnitude at enrollment (β=0.08; 95% CI=0.01, 0.15) than at follow-up (β=0.01; 95% CI=−0.05, 0.07) (Table S2, Figure 2). In comparison, we found the PFHxS-bilirubin association (β=0.12, 95% CI=0.02, 0.22) and the PFAS mixture-bilirubin association (ψ=0.19, 95% CI=0.03, 0.34) were greater in magnitude during late pregnancy.

Supplemental Analyses

We obtained similar estimates from models stratified by study visit relative to those that included a multiplicative interaction term between PFAS and study visit (Table S3). In the supplemental analysis that included the primary set of covariates plus maternal BMI, estimates for the main effects and interaction effects were comparable to those in the primary analysis (Table S4). Results from the analytic sample restricted to participants without pre-eclampsia were like those observed in the primary analysis; we found bilirubin was associated with PFOA during early pregnancy (β=0.09; 95% CI=0.02, 0.16), PFHxS during late pregnancy (β=0.10; 95% CI=0, 0.20), and the PFAS mixture across pregnancy (β=0.09; 95% CI=0, 0.18) and during late pregnancy (β=0.17; 95% CI=0.01, 0.33) (Table S5). Also, in the supplemental analysis conducted among participants who visited twice during pregnancy (N=251; Table S6), we observed a pattern of association like the primary analysis, such as with the PFAS mixture and bilirubin at any time during pregnancy (overall ψ=0.10, 95% CI=−0.002, 0.21; early pregnancy visit ψ=0.07, 95% CI=−0.04, 0.19; late pregnancy visit ψ=0.15, 95% CI=−0.01, 0.30). Finally, single measure estimates from the supplemental analysis of serum PFAS concentrations during early pregnancy and total bilirubin during late pregnancy are reported in Table S7 (N=268). The associations were attenuated and for some, in the opposite direction, relative to those obtained in the primary analysis.

IV. CONCLUSIONS

In a prospective cohort of pregnant African Americans, we found maternal serum PFHxS concentrations were positively associated with maternal total serum bilirubin. There was also evidence of effect modification by study visit for the association between PFOA and bilirubin, with stronger effects at the first visit during early pregnancy. Alternatively, at the second visit during late pregnancy, exposure to PFHxS and the PFAS mixture was positively associated with bilirubin. We did not observe meaningful associations for PFNA or PFOS.

Our findings extend mechanistic information about how PFAS exposure perturbs the delicate balance between ROS and antioxidants in pregnant people. There is already a predisposition toward oxidative stress during pregnancy because organs rich in mitochondria, like the liver and placenta, produce high amounts of energy for maternal and fetal growth.⁶⁶ A byproduct of mitochondrial metabolism, namely fatty acid oxidation, is ROS, which is counteracted by antioxidant defenses under normal conditions.⁶⁶ However, PFAS exposure may lead to harmful redox responses during pregnancy. A separate study performed by us revealed an association between a PFAS mixture of the four chemicals analyzed herein and pathways related to fatty acid oxidation in the Atlanta African American Maternal-Child Cohort.⁶⁷ We also found a mixture of PFNA, PFOA, PFOS, PFHxS, perfluoroundecanoic acid (PFUNDA), and perfluorodecanoic acid (PFDA) was associated with oxidative stress biomarkers indicative of lipid peroxidation in the Illinois Kids In Development Study (IKIDS) and Chemicals In Our Bodies (CIOB) pregnancy cohorts.³⁸ In the context of the present work, prenatal PFAS exposure may increase ROS, thereby triggering an increase in bilirubin to maintain redox homeostasis, as illustrated in Figure 3. Additional research is needed to validate the protective role of bilirubin for the relationship between prenatal PFAS exposure and adverse health outcomes characterized by oxidative stress.

Figure 3.

Bilirubin is an inhibitor of ROS generated by dysfunctional mitochondria from prenatal PFAS exposure.

The present analysis does build on several other reports of a direct relationship between PFAS and bilirubin in representative samples of regional and national populations. There was a correlation between serum PFOS concentrations and direct bilirubin in a cross-sectional analysis of adults part of the 2005-2006 C8 Health Project.⁶⁸ In the 2007-2010 US National Health and Nutrition Examination Survey (NHANES), higher PFOS and PFOA concentrations were correlated with total bilirubin among adolescents and adults.²⁴ A separate investigation of only adults in the 2003-2016 US NHANES identified a correlation between high PFOS (>18.40 ng/mL) and high total bilirubin (>95^th percentile).²⁶ Another group found adults with higher PFNA, PFOA, PFOS, and PFHxS had higher total bilirubin in the 2005-2012 US NHANES.²⁷ Similarly, higher PFOA, PFOS, and PFHxS was modestly correlated with total bilirubin among adults that participated in the Canadian Health Measures Survey between 2007-2011 or 2016-2017.²⁸ In general, study populations only comprised of males, seniors, or PFAS production workers do not exhibit a positive association between exposure and bilirubin, as observed herein among pregnant people with background exposure.^{25, 69, 70}

To the best of our knowledge, we only found two studies that have analyzed the association between a PFAS mixture and bilirubin in females of reproductive age. For example, an exposure mixture of PFNA, PFOA, PFOS, and PFHxS was correlated with total bilirubin among premenopausal women in the 1999-2018 US NHANES.²⁹ The cumulative mixture effect was driven by PFOA, which had the greatest quantile g-computation weight of the four PFAS, and comparable in magnitude to the estimate obtained by us. Furthermore, 420 Chinese pregnant people with higher exposure to a mixture of nine PFAS [i.e., PFNA, PFOA, PFOS, PFHxS, PFUNDA, PFDA, perfluorododecanoic acid (PFDoA), perfluoroheptanoic acid (PFHpA), perfluorobutanesulfonic acid (PFBS)] had higher total, direct, and indirect bilirubin.⁷¹ Mixture analysis with weighted quantile sum regression (positive and negative) also revealed PFHxS was the main contributor to the cumulative mixture effect, as observed in this study. Taken altogether, our results align with those previously reported that PFAS mixtures are positively associated with bilirubin and such associations are driven by PFOA and PFHxS in adult female populations.

The findings of this study suggest prenatal PFAS exposure between 6-18 gestational weeks has a greater adverse effect on the maternal liver antioxidant capacity than 21-36 gestational weeks. A pregnant woman’s liver must meet the demands of the growing maternal-placental-fetal unit, including bile production, hormone regulation, energy metabolism, and detoxification, which underscore the biological plausibility of late pregnancy as a critical window of vulnerability.⁷² For example, detoxification of endogenous substances like hormones and exogenous substances like PFAS may overwhelm the hepatic redox system as pregnancy progresses, thereby triggering an antioxidant response like bilirubin. Further investigation is needed to understand if such an association leads to adverse health outcomes in the mother or fetus.

Our analytical approach was a key strength. We used two robust models to estimate the association between repeated measures of PFAS and bilirubin during pregnancy. The combination of linear mixed effect regression and quantile g-computation enabled us to compare chemical effects by grouping (i.e., single PFAS versus PFAS mixture) and timing (i.e., early pregnancy versus late pregnancy). The analysis of study visit as an effect modifier by inclusion of a multiplicative interaction term with the exposure also enabled us to calculate a p-value for the interaction effect and retain statistical power with the maximum sample size, which are two advantages over stratification and may explain any modest differences in the reported estimates. Another strength is the study design. The Atlanta African American Maternal-Child Cohort is exclusive to a systematically marginalized racial group, who experience disproportionate rates of adverse pregnancy and birth outcomes, despite residing in a major metropolitan region of the US. Pregnant people with chronic medical conditions or chronic prescription medications, which are important sources of hepatocellular injury, are ineligible for participation. None of the participants had viral hepatitides on standard prenatal screenings either, which are abstracted from medical records. Our study was also strengthened by the analysis of total bilirubin, which is routinely measured in liver function blood panels during pregnancy. Our work may be scaled to other study populations and clinical settings with total bilirubin measurements available.

Unfortunately, we did not have data available for indirect and direct bilirubin, which are needed to explore where PFAS affect the hepatobiliary tract. Furthermore, the <50% detection rate for 11 of the 15 PFAS measured in the Atlanta African American Maternal-Child Cohort limited our analysis to PFNA, PFOA, PFOS, and PFHxS. Another limitation is reduced statistical power after excluding participants diagnosed with pre-eclampsia. Results from this sensitivity analysis should be interpreted cautiously as the smaller sample size may have contributed to the estimates.

In conclusion, our study in the Atlanta African American Maternal-Child Cohort provides evidence of an association between prenatal exposure to PFAS, a family of chemicals that generate ROS, and total bilirubin, an antioxidant that neutralizes ROS. A critical window of vulnerability may exist during pregnancy for certain PFAS, which we encourage to be explored in follow-up investigations on maternal liver antioxidant capacity.

Data Availability Statement

The datasets for this manuscript are not publicly available because, per the National Institutes of Health (NIH), approved Environmental influences on Child Health Outcomes (ECHO) Data Sharing Policy, ECHO-wide data have not yet been made available to the public for review/analysis. Requests to access the datasets should be directed to the ECHO Data Analysis Center, ECHO-DAC@rti.org.