Serum Per- and Polyfluoroalkyl Substance (PFAS) Concentrations and Predictors of Exposure among Pregnant African American Women in the Atlanta Area, Georgia.

Abstract

Exposure to per- and polyfluoroalkyl substances (PFAS) has been associated with adverse health outcomes, especially when exposure occurs within sensitive time windows such as the pre- and postnatal periods and early childhood. However, few studies have focused on PFAS exposure distribution and predictors in pregnant women, especially among African American women. We quantified serum concentrations of the four most common PFAS collected in all 453 participants and an additional 10 PFAS in 356 participants who were pregnant African American women enrolled from 2014 to 2018 in Atlanta, Georgia, and investigated the sociodemographic predictors of exposure. Additional home environment and behavior predictors were also examined in 130 participants. Perfluorohexane sulfonic acid (PFHxS), perfluorooctane sulfonic acid (PFOS), perfluorooctanoic acid (PFOA), and perfluorononanoic acid (PFNA) were detected in > 95% of the samples with PFOS having the highest concentrations (geometric mean (GM) 2.03 ng/mL). N-Methyl perfluorooctane sulfonamido acetic acid (NMeFOSAA), perfluoropentanoic acid (PFPeA), perfluorodecanoic acid (PFDA), and perfluoroundecanoic acid (PFUnDA) were found in 40–50% of the samples, whereas the detection frequencies for the other six PFAS were below 15%. When compared to National Health and Nutrition Examination Survey (NHANES) participants matching sex, race, and age with this study, our results showed similar concentrations of most PFAS, but higher concentrations of PFHxS (GM 0.99 ng/mL in this study; 0.63 and 0.4 ng/mL in NHANES 2014–2015 and 2016–2017 cycles). A decline in concentrations over the study period was found for most PFAS but not PFPeA. In adjusted models, education, sampling year, parity, BMI, tobacco and marijuana use, age of house, drinking water source, and cosmetic use were significantly associated with serum PFAS concentrations. Our study reports the first PFAS exposure data among pregnant African American women in the Atlanta area, Georgia. The identified predictors will facilitate the setting of research priorities and enable development of exposure mitigation strategies.

Introduction

Per- and polyfluoroalkyl substances (PFAS) are a group of synthetic compounds manufactured since the 1950s for use in a range of consumer products because of their ability to repel both water and lipids. Consumer products containing PFAS include food packaging, cookware, fabrics, carpet, upholstery, and personal care products (The Agency for Toxic Substances and Disease Registry, 2018; Paul et al., 2009). Because the carbon-fluorine covalent bond is one of the strongest, PFAS are persistent in the environment and some of them accumulate in wildlife and humans (Lau et al., 2007). Despite their phase-out starting in the early 2000s, PFAS have been widely detected, both in the environment and in human biospecimens over the past two decades (Harris et al., 2017; Kim et al., 2020; Sinclair et al., 2020).

A summary from the Agency for Toxic Substances and Disease Registry (2018) indicated that exposure to PFAS could cause hepatic, immune, reproductive, and developmental effects in animal oral exposure studies, and in a few inhalation and dermal exposure studies. Perfluorooctanoic acid (PFOA) and perfluorooctane sulfonic acid (PFOS) exposures were linked to adverse pregnancy and birth outcomes, and neurodevelopmental effects, such as preterm birth, growth restriction, and developmental delays in human studies (Chen et al., 2012; Fei et al., 2007; Hoffman et al., 2010; Jensen et al., 2015; Lau et al., 2007; Lenters et al., 2016; Maisonet et al., 2012; Savitz et al., 2012; Stein et al., 2009, Stein et al, 2014; Souza et al., 2020). More importantly, PFAS have been shown to cross the placenta. Concentrations in maternal peripheral blood highly correlate with infant cord blood concentrations suggesting that placental transfer could contribute to fetal exposure (Beesoon et al., 2011; Chen et al., 2017; Hanssen et al., 2010; Zhang et al., 2013). The positive correlation between duration of breastfeeding and PFAS concentrations in children suggests transfer can occur from mothers to their children during lactation (Mogensen et al., 2015; Tao et al., 2008). PFAS exposure during pregnancy, infancy, and early childhood is of interest because these are susceptible windows of exposure which may result in later health risk (The Agency for Toxic Substances and Disease Registry, 2018).

Racial differences in PFAS exposure have been reported in the general U.S. population: higher serum PFOS and perfluorononanoic acid (PFNA) levels were found among non-Hispanic blacks than whites and other races/ethnicities in multiple National Health and Nutrition Examination Survey (NHANES) cycles when adjusted for other variables (Calafat, et al., 2007a; Calafat, et al., 2007b; Jain, 2014; Nelson et al., 2012; Park et al., 2019). These differences in serum PFAS could potentially contribute to existing racial health disparities such as the disproportionately high risk of low birth weight, preterm delivery, and mortality in African Americans as compared to whites (Dunlop et al., 2011; Gee & Payne-Sturges, 2004; Giscombé & Lobel, 2005; Kramer & Hogue, 2009). While PFAS exposure has been documented and well-characterized in several studies, few studies have focused on exposure distribution percentile and exposure predictors in African Americans (Boronow et al., 2019; Park et al., 2019), much less pregnant African American women. The limited knowledge among this minority population can restrict our ability to identify critical risk factors for exposure and resulting health outcomes. Furthermore, little information is known about typical chemical exposures in the Southeastern United States. The purpose of this study is to characterize the chemical distribution percentiles and predictors of PFAS exposure in a cohort of pregnant African American women in the Atlanta area, Georgia.

Materials and methods

Study population

Participants for this study were drawn from the women participating in the Emory University African American Vaginal, Oral, and Gut Microbiome in Pregnancy Study (Corwin et al., 2017). The study recruited pregnant women from prenatal clinics affiliated with hospitals in Atlanta, Georgia: Emory University Hospital Midtown, a private hospital, which provides service to people with a diverse socioeconomic status (SES), and Grady Memorial Hospital, a county-support public hospital, which provides service mainly to low-income or underserved populations. Inclusion criteria for enrollment into the cohort included: (1) U.S.-born Black woman by self-report; (2) between 8 and 14 weeks gestation with a singleton pregnancy; (3) able to understand and speak English; (4) between 18 and 40 years old; and, (5) experiencing no chronic medical condition or not taking prescribed chronic medications. The subjects who met the criteria above were asked to provide written informed consent at enrollment. This study was reviewed and approved by Emory’s Institutional Review Board.

A total of 453 pregnant African American women were included in this study; these 453 women represented the first consecutively enrolled women in the cohort (enrolled between March 2014 and May 2018) whose pregnancy ended in a birth for whom a blood sample (for PFAS measurement) was available. Women enrolled in the study between January 2016 and May 2018 (n = 232) had the option of enrolling in a sub-study that involved a home visit between 20 and 24 weeks of gestation as part of the Center for Children’s Health, the Environment, the Microbiome, and Metabolomics (C-CHEM²), which seeks to understand the interaction between the prenatal and postnatal environmental toxicant exposures, the microbiome, the metabolome, and their impacts on birth outcomes and infant health and development.

Clinical and questionnaire data collection

The questionnaires were administered to the participants by trained research coordinators, who directly recorded the answers on either paper or tablet computers. Two questionnaires were included in this study: (1) the Sociodemographic Questionnaire, which contains self-report of household income, education, marital or cohabitating status, insurance status during pregnancy, and substance use, which was collected at enrollment at the hospitals; and, (2) the Home Environment and Behavior Questionnaire, which was administrated between 20 and 24 gestation at the participants’ home.

Of the 232 participants enrolled from January 2016 to May 2018, 130 participants completed the Home Environment and Behavior Questionnaire during the home visit. The participants were asked about housing characteristics, including age of their home, distance between their home and the nearest industry, dump or waste site, and behavioral factors such as home cleaning behaviors, primary drinking water source (tap/well/bottled water consumption), water consumption in the last 48 hours (tap/well/bottled water), takeout food or deliver pizza consumption last month, microwave popcorn consumption last month, frequency of personal care product use (hair products/shampoo/lotion/sunscreen), and cosmetic product use (foundation/rouge or blush/lipstick/mascara/nail polish/eye shadow). Additionally, some data were ascertained from the prenatal clinical record, such as maternal age, parity, and BMI (calculated by height and weight measured at prenatal visit).

Quantification of PFAS

A 30-mL venous blood sample was drawn from each participant at enrollment between March 2014 and May 2018. The sample tubes were transported to the laboratory, centrifuged to separate the serum from whole blood, and then stored at −80°C for subsequent analyses. The samples were analyzed at two laboratories in the Children’s Health Exposure Analysis Resource (CHEAR): Wadsworth Center (New York State Department of Health) and the Laboratory of Exposure Assessment and Development for Environmental Health Research (LEADER) (Emory University). CHEAR is an exposure assessment resource supported by the National Institute of Environmental Health Sciences to expand the ability of including environmental exposure analyses in children’s health research. Laboratories in CHEAR have implemented the same rigorous internal quality control procedures to provide harmonized data (Balshaw et al., 2018). All 453 samples were analyzed for perfluorohexane sulfonic acid (PFHxS), PFOS, PFOA, and PFNA from both laboratories, whereas 356 samples were analyzed for perfluorobutane sulfonic acid (PFBS), perfluorooctane sulfonamide (PFOSA), N-methyl perfluorooctane sulfonamido acetic acid (NMeFOSAA), N-ethyl perfluorooctane sulfonamido acetic acid (NEtFOSAA), perfluoropentanoic acid (PFPeA), perfluorohexanoic acid (PFHxA), perfluoroheptanoic acid (PFHpA), perfluorodecanoic acid (PFDA), perfluoroundecanoic acid (PFUnDA) and perfluorododecanoic acid (PFDoA) from Wadsworth.

Further details of the Wadsworth method are described elsewhere (Honda et al., 2018). For the LEADER method, each serum sample was spiked with an isotopically labeled standard solution, and deproteinated with formic acid. The supernatant was loaded to the Isolute C18 cartridge (Biotage, Sweden). The cartridge was washed with 1:1(v/v) formic acid and methanol solution, then eluted with methanol. The eluate was evaporated to dryness under a nitrogen stream. The sample was reconstituted with methanol for the analysis using high-performance liquid chromatography interfaced with tandem mass spectrometry (6460 Triple Quadrupole LC/MS, Agilent Technologies; Santa Clara, CA). In each analytical batch, a matrix-based calibration curve, a blank sample, and quality control samples were prepared and injected alongside 37 unknown serum samples. Quantification was performed using isotope dilution calibration with the limits of detection (LODs) ranging from 0.05 to 0.2 ng/mL. The LODs were defined as the lowest concentrations in the calibration curve, where the signal to noise ratio of the observed signal was ≥3 and the accuracy was within 100 ± 20% (Supplement Table 1) (FDA, 1994). The method precision, calculated as the relative standard deviation (RSD), was less than 10% for all analytes. Standard reference materials from the National Institute of Standards and Technology (NIST) were analyzed concurrently and the recoveries of target analytes were within 100 ± 20%. Good agreement was observed between the results from the two laboratories using 11 overlapped samples; the Pearson correlation coefficients ranged from 0.88 to 0.93, and the relative percent differences (RPD) ranged from 0.12 to 20.2% with a median of 4.8% (Supplement Table 2). Additionally, both laboratories successfully participate in and are certified semi-annually for PFAS measurements by the German External Quality Assessment Scheme (g-equas.de).

Statistical analyses

Descriptive analyses were performed for the serum concentrations, including detection frequencies, distribution percentiles, ranges, geometric means (GMs), and geometric standard deviation (GSD). PFAS levels below the LODs were imputed with LOD/√2 in the descriptive analyses (Hornumg & Reed, 1990). To conform to normality, we used natural log-transformed values of PFAS concentrations for statistical analyses. The PFAS with low detection frequencies (0–14%), including PFBS, PFOSA, NEtFOSAA, PFHxA, PFHpA, and PFDoA, were excluded from further statistical analyses due to the limited number of quantified measurements.

Pearson correlation coefficients were first calculated to investigate the correlations between the concentrations of individual PFAS. One-way analysis of variance (ANOVA) was performed to examine the concentration differences between different categorical predictors.

We fitted the multivariable linear regression models to assess the associations between serum PFAS levels and potential predictors. In the first stage, the associations between sociodemographic factors and serum PFAS levels were investigated in adjusted models mutually including age (continuous), sampling year (2014/2015/2016/2017–2018), parity (0/1–2/3–4), BMI (continuous), education (less than high school/high school/some college/college and above), poverty income ratio (the ratio of a family’s income to the US Census Bureau’s poverty threshold; <100/100–150/≥150%), married/cohabiting (yes/no), insurance status during pregnancy (private/Medicaid, a federal-state public health insurance program for low income and disabled people), hospital (Emory/Grady), and tobacco (ever/never), alcohol (during pregnancy/not during pregnancy), and marijuana use (during pregnancy/not during pregnancy). The covariates were selected based on previous studies (Kingsley et al., 2018; Pitter et al., 2020; Shu et al., 2018; Vuong et al., 2018). In the second stage, the home environment and behavior characteristics were included one at a time with the significant covariates (at a 0.05 α level) in the first stage analysis because the covariates could contribute to the main effects of serum PFAS concentrations. Sampling year was excluded from the models in the second stage because the participants with home environment and behavioral information were mostly enrolled in 2017. PFAS concentrations as outcome variables were natural log-transformed; therefore, we back-transformed coefficient estimates and confidence intervals using (e^β−1)×100, and reported effect estimates as a percent change in GM of PFAS concentrations relative to the reference group. Collinearity for the covariates in the models was also assessed by calculating variance inflation factors (James et al., 2013).

Due to missing values on poverty income ratio (missing n=62; 13.7%), tobacco use (missing n=6; 1.3%), alcohol use (missing n=6; 1.3%), marijuana use (missing n=7; 1.5%), and some home environment and behavior predictors (missing n=1–92; 0.8–70.8%; Supplement Table 3), multiple imputations by chained equations with regression imputation approaches were conducted (Buuren & Groothuis-Oudshoorn, 2010). Serum PFAS concentrations below the LODs were multiply imputed using maximum likelihood estimation with a lognormal distribution (Lubin et al., 2004). We imputed 10 datasets up to the full sample size of 453 and 356 for the first stage analyses (based on the availability of PFAS measurements), and 130 for the second stage analyses. The sensitivity analyses included: 1) the comparison between multiple imputations and complete case analysis for multivariate linear regression models to evaluate the impact of the missing values; 2) the comparison between simple imputation with LOD/√2 and multiple imputations using maximum likelihood estimation for PFAS values below LOD. Overall results did not change substantially; thus, multiple imputations are presented because the other methods tended to bias effect estimates away from the null hypothesis in our analyses.

The measured PFAS concentrations in this population were compared with those of the general U.S. population among the same demographic group (non-Hispanic black females between the age of 18 and 40 years in NHANES). Serum PFAS data from four NHANES (2009–2016) cycles (n=333) were extracted and evaluated accounting for the sampling weights, primary sampling units, and strata due to the complex survey design. The PFAS with more than 50% detection frequencies in this study were compared to the NHANES data. All statistical analyses were conducted in R version 3.6.1, and multiple imputations were done through Multivariate Imputation by Chained Equations (MICE) package.

Results

The demographic characteristics are shown in Table 1. Among the participants, the average age was 24.8 years. The majority had a high school or below education (54%), < 150% poverty income ratio (58%), Medicaid as medial insurance (78%), given birth one or more times (52%), and an overweight (21%) or obese (36%) status. PFAS concentrations and summary statistics are presented in Table 2. PFHxS, PFOS, PFOA, and PFNA were detected in > 95% samples with PFOS having the highest GM (2.03 ng/mL). NMeFOSAA, PFPeA, PFDA, and PFUnDA were detected in almost 50% of the samples, whereas only a small percentage of samples had detectable levels of PFBS (14%), PFOSA (0%), NEtFOSAA (5%), PFHxA (3%), PFHpA (14%), and PFDoA (2%).

Table 1.

Demographic characteristics of the participants in this study.

Characteristics	Total (n=453)
Maternal age at enrollment (years)
Mean (SD)	24.8 (4.67)
Median [Min, Max]	24.0 [18.0, 40.0]
Education
Less than high school	72 (15.9%)
High school	172 (38.0%)
Some college	137 (30.2%)
College and above	72 (15.9%)
Poverty income ratio (%)
< 100	189 (41.7%)
100–150	74 (16.3%)
≥ 150	128 (28.3%)
Missing	62 (13.7%)
Married/Cohabiting
Yes	215 (47.5%)
No	238 (52.5%)
Insurance
Private	98 (21.6%)
Medicaid	355 (78.4%)
Hospital
Emory	179 (39.5%)
Grady	274 (60.5%)
Sampling year
2014	93 (20.5%)
2015	124 (27.4%)
2016	115 (25.4%)
2017–2018	121 (26.7%)
Parity (#)
0	216 (47.7%)
1–2	192 (42.4%)
3–4	45 (9.9%)
BMI (kg/m2)
< 18.5	16 (3.5%)
18.5–24.9	176 (38.9%)
25–29.9	96 (21.2%)
≥ 30	165 (36.4%)
Tobacco use
Ever	63 (13.9%)
Never	384 (84.8%)
Missing	6 (1.3%)
Alcohol use
During pregnancy	37 (8.2%)
Not during pregnancy	410 (90.5%)
Missing	6 (1.3%)
Marijuana use
During pregnancy	99 (21.9%)
Not during pregnancy	347 (76.6%)
Missing	7 (1.5%)

Table 2.

Summary statistics for PFAS concentrations (ng/mL) measured in this study.

	# C	n	% > LOD	GMa	GSDa	P25	P50	P75	P95	Max
Perfluoroalkane sulfonic acids (PFSAs)
PFBS	4	356	14	--	--	<LOD	<LOD	<LOD	0.12	0.59
PFHxS	6	453	97	0.99	1.93	0.73	1.10	1.52	2.32	4.80
PFOS	8	453	99	2.03	2.07	1.43	2.17	3.22	5.31	12.4
Perfluoroalkane sulfonamides (FASAs)
PFOSA	8	356	0	--	--	<LOD	<LOD	<LOD	<LOD	<LOD
NMeFOSAA	11	356	49	--	--	<LOD	<LOD	0.07	0.26	1.46
NEtFOSAA	12	356	5	--	--	<LOD	<LOD	<LOD	0.02	0.11
Perfluoroalkyl carboxylic acids (PFCAs)
PFPeA	5	356	48	--	--	<LOD	<LOD	0.11	0.22	0.66
PFHxA	6	356	3	--	--	<LOD	<LOD	<LOD	<LOD	0.26
PFHpA	7	356	14	--	--	<LOD	<LOD	<LOD	0.10	0.31
PFOA	8	453	97	0.63	2.35	0.45	0.71	1.07	1.69	4.42
PFNA	9	453	97	0.24	2.35	0.15	0.27	0.42	0.74	2.27
PFDA	10	356	48	--	--	<LOD	<LOD	0.13	0.27	1.06
PFUnDA	11	356	43	--	--	<LOD	<LOD	0.06	0.17	0.53
PFDoA	12	356	2	--	--	<LOD	<LOD	<LOD	<LOD	0.11

#C = number of carbon atoms; n = sample number; LOD = limit of detection; GM= geometric mean; GSD= geometric standard deviation; P25 = the 25^th percentile.

^aGeometric means and standfard deviations were not calculated for congeners with detection frequencies < 50%; the values below LODs were replaced by LOD/√2.

Table 3 shows the Pearson correlation coefficients between the natural log-transformed concentrations of PFAS. The four most frequently detected PFAS, i.e., PFHxS, PFOS, PFOA, and PFNA, were moderately to strongly correlated with each other with the coefficients ranging from 0.39 to 0.71 (Schober et al., 2018). Although positive correlations were mostly found between any two PFAS, PFPeA concentrations were negatively associated with most of PFAS with the correlation coefficients ranging from −0.33 to −0.06.

Table 3.

Pearson correlation coefficients of natural log-transformed serum PFAS concentrations^a.

	PFHxS	PFOS	NMeFOSAA	PFPeA	PFOA	PFNA	PFDA	PFUnDA
PFHxS	1.00	0.66**	0.08	−0.25**	0.44**	0.39**	0.44**	0.10
PFOS		1.00	0.15*	−0.31**	0.65**	0.70**	0.55**	0.22*
NMeFOSAA			1.00	−0.06	0.07	0.13*	0.14*	−0.16*
PFPeA				1.00	−0.14*	−0.22**	−0.33**	0.04
PFOA					1.00	0.71**	0.35**	0.17*
PFNA						1.00	0.50**	0.29**
PFDA							1.00	0.11
PFUnDA								1.00

^*P-value < 0.05

^**p-value < 0.01

^aThe values below LODs were imputed by a lognormal probability distribution and maximum likelihood estimation.

We compared serum PFAS levels in this study with similarly aged, non-Hispanic black women in NHANES. Figure 1 shows PFHxS, PFOS, PFOA and PFNA concentrations decrease by sampling year among both NHANES and our study participants. Similar serum PFOS, PFOA, and PFNA concentrations were found among these two populations, whereas PFHxS levels were higher in this study than those in NHANES. The GM of serum PFHxS was 0.63 ng/mL (95% CI 0.07–0.50 ng/mL) and 0.40 ng/mL (95% CI 0.06–0.28 ng/mL) in the 2013–2014 and 2015–2016 NHANES cycles, respectively, which are about 0.4–0.6 times the GM of this study (GM 0.99; 95% CI 0.94–1.06 ng/mL).

Figure 1.

PFAS geometric means and 95% confidence interval (ng/mL) among non-Hispanic black females aged 18–40 years in 2009–2016 NHANES and in this study.

The unadjusted GMs and GSDs by different predictors are presented in Supplement Table 3. PFHxS, PFOS, PFOA, and PFNA concentrations were significantly different by education, poverty income ratio, insurance, sampling year, parity, BMI, tobacco use, marijuana use, age of house, bottle water as primary drinking water source, number of cosmetic product usually worn, usually worn foundation, and usually worn lipsticks in ANOVA analyses. The GM and GSD of NMeFOSAA, PFPEA, PFDA, and PFUnDA were not presented by groups because these PFAS were less frequently detected. The adjusted results using multivariable regression models are presented in Table 4. No collinearity was found in the models, and the values of variance inflation factors were between 1.02 and 2.36. Higher education levels were associated with higher PFNA concentrations; the participants with college education and above had a 57.8% (95% CI 14.4–118%) increase in serum PFNA levels compared to those with less than high school education. The concentrations of PFHxS, PFOS, PFOA, PFNA, NMeFOSAA, and PFDA in the samples collected during 2017 and 2018 were 36–90% lower than those collected in 2014. PFPeA and PFUnDA concentrations were higher when the samples were collected in the later years. Lower PFAS concentrations were found in the women with higher parity, with PFOA having the largest percentage difference in concentrations, yet PFPeA has a different direction of relationship from the other PFAS. Additionally, BMI was negatively associated with PFAS concentrations; a unit changed in BMI was significantly associated with 0.9% (95 CI −1.6, −0.3%), 1.2% (95% CI −2, −0.5%), and 2.4% (95% CI −4.5, −0.3%) decreases in PFHxS, PFOS, and PFUnDA concentrations, respectively. Ever tobacco users had lower PFHxS and PFOS concentrations than never tobacco users, whereas the people using marijuana during pregnancy had higher PFOS, PFOA, and PFNA levels than non-users.

Table 4.

Adjusted^a percent change in natural log-transformed serum PFAS^b by sociodemographic predictors^c.

	PFHxSd	PFOSd	PFOAd	PFNAd	NMeFOSAAe	PFPeAe	PFDAe	PFUnDAe
	% difference in PFAS concentrationf (95% CI)
Age (years)	−0.3 (−1.6, 1.1)	−0.2 (−1.8, 1.3)	−0.4 (−2.4, 1.7)	−0.3 (−2.2, 1.6)	−1.0 (−6.1, 4.3)	0.8 (−2.1, 3.8)	−1.0 (−4.5, 2.6)	3.1 (−1.5, 7.8)
Education (Ref = Less than high school)
High school	4.9 (−10.2, 22.6)	−0.2 (−16.4, 19.2)	6.5 (−15.2, 33.8)	1.2 (−18.4, 25.4)	−26.0 (−58.0, 30.4)	5.8 (−24.0, 47.3)	13.0 (−23.3, 66.6)	0.5 (−40.2, 69.0)
Some college	5.0 (−11.6, 24.7)	2.1 (−16.1, 24.4)	5.9 (−17.8, 36.4)	5.7 (−16.8, 34.1)	−44.4 (−69.4, 0.9)	−2.4 (−31.2, 38.4)	19.5 (−23.7, 87.2)	−2.5 (−46.8, 78.6)
College and above	9.5 (−13.2, 38.2)	25.8 (−3.5, 64.0)	35.7 (−3.5, 90.9)	57.8 (14.4, 118)**	−17.7 (−63.1, 83.5)	−20.6 (−51.2, 29.1)	47.9 (−19.5, 172)	78.1 (−15.0, 273)
Poverty income ratio (%) (Ref < 100)
100–150	8.4 (−6.2, 25.3)	17 (−1.1, 38.3)	19.2 (−3.9, 47.9)	15 (−5.8, 40.5)	2.1 (−38.7, 70.0)	4.8 (−24.0, 44.6)	−0.9 (−33.9, 48.6)	42.2 (−13.9, 135)
≥ 150	14.4 (−2.0, 33.6)	13.8 (−4.2, 35.2)	15.4 (−7.6, 44.2)	13.3 (−8.4, 40.0)	−21.0 (−54.4, 37.0)	−7.7 (−32.4, 26.0)	5.7 (−26.5, 52.1)	12.3 (−30.4, 81.2)
Married or cohabitating (Ref = No)
Yes	2.5 (−7.8, 14.0)	2.3 (−9.4, 15.6)	6.5 (−8.9, 24.5)	10.7 (−4.4, 28.3)	−14.0 (−40.2, 23.7)	−0.1 (−22.3, 28.5)	5 (−20.3, 38.3)	14.9 (−18.7, 62.2)
Insurance (Ref = Private)
Medicaid	4.6 (−12.8, 25.4)	4.4 (−15.1, 28.4)	−2.6 (−25.4, 27.2)	4.6 (−18.7, 34.5)	14.3 (−39.8, 117)	−9.5 (−40.1, 36.8)	−5.9 (−41.6, 51.4)	−19.1 (−54.2, 42.9)
Hospital (Ref = Emory)
Grady	4.2 (−9.8, 20.3)	−0.7 (−15.8, 17.1)	16.4 (−5.7, 43.8)	−0.8 (−18.7, 21.1)	38.0 (−17.5, 131)	3.3 (−23.9, 40.0)	5.0 (−26.1, 49.3)	−25.2 (−53.8, 21.3)
Sampling year (Ref =2014)
2015	−9.4 (−22.1, 5.4)	−12.8 (−26.7, 3.6)	12.4 (−10.0, 40.4)	-20.6 (−35.6, −2.1)*	-62.8 (−76.9, −40.1)**	43 (2.3, 100)*	-42.9 (−59.5, −19.5)**	64.6 (5.4, 157)*
2016	-57.6 (−63.7, −50.4)**	-50.2 (−58.3, −40.5)**	−4.4 (−23.9, 20.1)	−15.1 (−31.5, 5.3)	-63.6 (−84.0, −17.3)*	59 (−3.0, 161)	-87.2 (−93.2, −75.9)**	110.2 (2.0, 333)*
2017–2018	-35.5 (−44.7, −24.9)**	-54.4 (−61.7, −45.7)**	-37.4 (−50.0, −21.7)**	-60.3 (−67.8, −51.0)**	-59.4 (−74.4, −35.6)**	208.8 (119, 337)**	-89.7 (−92.8, −85.4)**	43.8 (−6.9, 122)
Parity (#) (Ref = 0)
1–2	−10.5 (−20.3, 0.6)	−6.7 (−18.3, 6.5)	-19.6 (−32.2, −4.6)*	−10.2 (−23.5, 5.5)	−3.8 (−35.9, 44.5)	1.1 (−21.2, 29.8)	−14.1 (−34.9, 13.4)	−12.5 (−42.1, 32.1)
3–4	-19.1 (−33.2, −2.0)*	−6.8 (−25.1, 16.1)	-25.8 (−44.0, −1.7)*	−11.4 (−32.1, 15.5)	−9.6 (−54.0, 77.5)	13.9 (−26.3, 76.2)	−22.8 (−53.5, 28.0)	−20 (−47.5, 82.8)
BMI (kg/m2)	-0.9 (−1.6, −0.3)**	-1.2 (−2.0, −0.5)**	−0.3 (−1.3, 0.6)	−0.4 (−1.3, 0.5)	0.1 (−2.1, 2.4)	−0.9 (−2.3, 0.4)	−0.6 (−2.2, 1.1)	-2.4 (−4.5, −0.3)*
Tobacco use (Ref = Never)
Ever	−22.9 (−34.9, −8.7)**	-25.2 (−38.4, −9.2)**	−18.9 (−36.6, 3.8)	−18.1 (−35.1, 3.3)	20.6 (−34.1, 121)	34.9 (−8.5, 98.9)	−18.5 (−50.6, 34.5)	−26.1 (−58.0, 29.9)
Alcohol use (Ref = Not during pregnancy)
During pregnancy	5.9 (−12.9, 28.6)	20.8 (−3.3, 50.8)	25.8 (−5.4, 67.5)	19.4 (−8.8, 56.3)	28.4 (−34.4, 151)	−32.8 (−57.8, 7.1)	1.6 (−37.9, 66.1)	79.1 (−5.8, 240)
Marijuana use (Ref = Not during pregnancy)
During pregnancy	9.8 (−4.4, 26.1)	21.8 (3.9, 42.7)*	38.8 (13.3, 70.0)**	42.4 (17.6, 72.4)**	41.9 (−11.4, 127)	−8.6 (−33.4, 25.3)	20.9 (−17.8, 77.9)	23.4 (−21.5, 93.9)

CI = confidence interval;

^*p-value < 0.05;

^**p-value < 0.01.

^aThe models were mutually adjusted for all the variables listed above.

^bThe values below LODs were imputed by a lognormal probability distribution and maximum likelihood estimation.

^cThe missing values were multiply imputed by chained equations with logistic and polytomous logistic regression imputation approaches.

^dSample size = 453.

^eSample size = 356.

^fPercent change in PFAS concentrations associated with each predictor by exponentiating regression coefficients, subtracting 1, and multiplying by 100%.

For the home environment and behavioral characteristics (Table 5), the participants with ≥ 20 years residential house had 32% and 39% decreases in PFOA and PFNA concentrations than those with <10 years house. The participants who reported bottled water as their primary drinking water source had 65% and 88% increases in PFOS and PFOA concentrations. Additionally, a larger number of cosmetic products usually worn (a 50% increase in PFNA concentrations when comparing 4–6 and 0 products used), and usually worn foundation (52% and 58 % increases in PFOS and PFNA concentrations) were associated with higher serum PFAS. The results of well water consumption and frequency of sunscreen use were not shown since most women reported no exposure to either of these. The results of NMeFOSAA, PFPeA, PFDA, and PFUnDA were not shown due to the small sample size.

Table 5.

Adjusted^a percent change in natural log-transformed serum PFAS^b by home environment and behavioral predictors^c (n=130).

	PFHxS	PFOS	PFOA	PFNA
	% difference in PFAS concentrationd (95% CI)
Age of house (years) (Ref < 10)
10–20	−1.4 (−31.4, 41.6)	−13.5 (−40.9, 26.5)	−14.3 (−45.4, 34.5)	0.8 (−34.3, 54.4)
≥ 20	9.1 (−20.2, 49.1)	a−25.1 (−45.0, 2.0)	−31.9 (−53.4, −0.3)*	−38.5 (−57.1, −11.8)**
Distance to nearest industrial plant, dump or waste site (meters) (Ref < 400)
≥ 400	3.4 (−29.7, 52.1)	2.2 (−26.2, 41.6)	1.1 (−33.4, 53.6)	0.6 (−41.6, 73.2)
Frequency of floor cleaning (Ref = Daily)
A few times a week	5.3 (−19.2, 37.1)	−2.7 (−25.5, 27.0)	1.0 (−27.9, 41.4)	−8.5 (−34.0, 27.0)
Once every couple weeks or less	−0.5 (−31.5, 44.6)	4.7 (−28.2, 52.6)	-6.6 (−42, 50.1)	11.5 (−29.8, 77.1)
Primary drinking water source - tap water (Ref = No)
Yes	1.9 (−20.3, 30.3)	5.1 (−18.0, 34.8)	11.8 (−18.2, 52.9)	−1.9 (−27.8, 33.2)
Primary drinking water source - bottled water (Ref = No)
Yes	34.5 (−8.0, 96.8)	64.6 (12.9, 140)*	88.3 (17.1, 203)*	57.2 (−1.8, 152)
Tap water consumption in the last 48 hours (8 oz cup) (Ref = None)
1–5	−9.8 (−31.2, 18.3)	0.7 (−23.4, 32.3)	22.2 (−13.6, 72.6)	28.0 (−8.4, 78.8)
≥ 6	15.4 (−18.7, 63.8)	28.2 (−10.0, 82.5)	19.3 (−23.8, 86.6)	35.7 (−12.1, 109)
Bottled water consumption in the last 48 hours (8 oz cup) (Ref = None)
1–5	15.4 (−17.6, 61.5)	31.6 (−6.0, 84.4)	12.1 (−27.0, 72.1)	9.2 (−28.2, 66.1)
≥ 6	13.4 (−23.9, 69.1)	30.5 (−12.4, 94.4)	23.7 (−25.5, 105)	2.3 (−37.7, 67.9)
Takeout consumption in the last month (times) (Ref = None)
1–2	14.0 (−16.5, 55.5)	−1.1 (−27.6, 35.1)	−19.6 (−45.8, 19.3)	−12.6 (−40.4, 28.3)
≥ 3	11.6 (−19.0, 53.6)	−16.1 (−39.2, 15.7)	−13.2 (−42.2, 30.3)	−26.3 (−50.3, 9.4)
Microwave popcorn consumption in the last month (times) (Ref = None)
1–2	4.2 (−22.5, 40.1)	2.1 (−24.3, 37.8)	−11.7 (−39.5, 28.8)	−11.4 (−38.6, 27.6)
≥ 3	−14.7 (−44.6, 31.4)	3.6 (−33.1, 60.5)	1.6 (−41.4, 76.1)	40.7 (−17.4, 140)
Frequency of cosmetic product use (Ref = Never or Occasionally)
Daily	−0.3 (−22.8, 28.7)	2.8 (−20.8, 33.5)	−2.0 (−29.2, 35.5)	−6.2 (−31.6, 28.7)
Number of cosmetic product usually worn (#) (Ref = 0)
1–3	5.3 (−22.4, 43.0)	22.5 (−10.0, 66.7)	14 (−22.4, 67.5)	9.5 (−24.7, 59.3)
4–6	−8.6 (−34.4, 27.4)	18.2 (−15.3, 65.0)	49.1 (−1.8, 126)	50.4 (0.2, 126)*
Usually worn foundation (Ref = No)
Yes	−19.8 (−40.3, 7.5)	−3.0 (−28.1, 30.8)	52.2 (5.2, 120)*	58.3 (10.5, 127)*
Usually worn rouge or blush (Ref = No)
Yes	26 (−13.4, 83.2)	14.2 (−21.8, 67.0)	18.7 (−26.4, 91.6)	31.6 (−17.4, 110)
Usually worn lipstick (Ref = No)
Yes	−1.2 (−24.0, 28.5)	8.0 (−17.2, 40.7)	36.5 (−1.9, 89.8)	36.8 (−0.9, 88.7)
Usually worn mascara (Ref = No)
Yes	−22.3 (−39.6, 0.0)	7.6 (−17.0, 39.3)	24.3 (−10.1, 71.9)	24.9 (−9.0, 71.3)
Usually worn nail polish (Ref = No)
Yes	15.2 (−10.7, 48.5)	21.4 (−6.0, 56.8)	27.5 (−7.6, 76.0)	17.9 (−14.1, 61.8)
Usually worn eye shadow (Ref = No)
Yes	−13.8 (−33.3, 11.5)	−3.3 (−25.5, 25.4)	10.8 (−20.2, 53.7)	19.6 (−13.0, 64.6)
Frequency of lotion use (Ref = Never or Occasionally)
Daily	−11 (−31.7, 16.0)	−17.1 (−36.5, 8.2)	−15.1 (−39.9, 19.8)	−7.3 (−33.4, 29.1)
Frequency of hair products use (Ref = Never)
Occasionally or Monthly	18.8 (−20.2, 76.9)	9.9 (−26.5, 64.3)	44.3 (−12.8, 139)	35.3 (−17.4, 122)
Weekly or Daily	12.5 (−24.0, 66.7)	13.1 (−24.0, 68.2)	28.6 (−21.8, 112)	25.3 (−23.0, 104)
Frequency of shampoo (Ref = Monthly)
Biweekly or more often	8.1 (−15.2, 37.9)	21.1 (−5.1, 54.4)	22.3 (−10.1, 66.5)	25.7 (−6.8, 69.7)

CI = confidence interval;

^*p-value < 0.05;

^**p-value < 0.01.

^aThe models were adjusted for education, parity, BMI, tobacco use, and marijuana use.

^bThe values below LODs were imputed by a lognormal probability distribution and maximum likelihood estimation.

^cThe missing values were multiply imputed by chained equations with logistic and polytomous logistic regression imputation approaches.

^dPercent change in PFAS concentrations associated with each predictor by exponentiating regression coefficients, subtracting 1, and multiplying by 100%.

Discussion

This is the first study focusing on PFAS exposure and their predictors among pregnant African American women in Atlanta, Georgia. Serum PFHxS, PFOS, PFOA, and PFNA levels in this population were generally lower than the measurements in the other birth cohorts in the United States in which samples were collected in earlier years (Boronow et al., 2019; Kingsley et al., 2018; Lyall et al., 2018; Romano et al., 2016; Sagiv et al., 2015). Serum PFAS levels were more comparable to the concentrations measured in the corresponding NHANES sampling years within a subset of similarly aged, non-Hispanic black females. It is noteworthy that serum PFAS levels among pregnant women, especially women in the late pregnancy, could be lower than non-pregnant women of reproductive age because of physiologic changes during pregnancy, such as increased blood volume, increased renal plasma flow and glomerular filtration rate, and decreased plasma protein levels (Pan et al., 2017; Soma-Pillay et al., 2016). Although the serum PFAS concentrations among pregnant women could underestimate the concentrations of the women in this population, we expect the effect to be less influential because the serum samples were collected early in pregnancy in this study.

An overall downward trend of serum PFAS concentrations over time was observed in both this population and the matched NHANES population. In addition, sampling year is a significant predictor in the adjusted regression analyses, showing a decrease in serum PFAS levels over the study period (2014–2018). This decline is likely due to the voluntary phase-out by the major manufacturer starting from the year 2000, and followed by EPA PFOA Stewardship Program, which reduced emission and product content of PFOA, and Significant New Use Rule (SNUR), which required a notice and a review before manufacturing, selling, importing, and using long-chain PFAS (EPA, 2020; Land et al., 2018). Previous studies indicated that serum PFAS concentrations among the U.S. population have been declining since the early 2000s, coinciding with the timing of the phase-out (Calafat et al., 2007b; Kato et al., 2011; Spliethoff et al., 2008).

Higher PFHxS concentrations were found in this population than in NHANES. A study proposed that PFHxS could be a tracer of exposure for PFAS in consumer products rather than the other exposure sources such as seafood consumption (Hu et al., 2018), suggesting consumer products could be important PFAS exposure sources in this population.

The positive correlations observed among most of PFAS suggested similar exposure sources or pathways. However, PFPeA, a short-chain PFAS (≤7 carbons for perfluoroalkyl carboxylic acids (PFCA); ≤5 carbons for perfluoroalkane sulfonic acids (PFSA)) with five carbon atoms (ITRC, 2020), was negatively correlated with the other PFAS such as PFHxS, PFOS, PFOA, PFNA, and PFDA, suggesting the possibility of long-chain PFAS substituted by short-chain alternatives. Moreover, an upward trend of PFPeA concentrations by sampling year was found (Table 4), whereas downward trends were shown for most of the long-chain PFAS. Previous studies showed that manufacturers have adopted short-chain PFAS and the other fluorinated alternatives for the consumer products due to the phase-out of long-chain PFAS (Ateia et al., 2019; Valsecchi et al., 2017). Despite limited environmental and biomonitoring data on PFPeA, several studies have detected PFPeA in precipitation, surface water (Gewurtz et al., 2019), drinking water source (Sun et al., 2016), indoor dust and air (Fraser et al., 2013; Haug et al., 2011; I. Ericson et al., 2012), consumer products (Ministry of Environment and Food of Denmark, 2018), and PFPeA has been the dominant PFAS in influent and effluent water of wastewater treatment plants in some areas during the 2010s (Arvaniti et al., 2012; Zhang et al., 2015). Despite the shorter biological persistence, several concerns about the application of short-chain PFAS were reported by the previous studies. First, short-chain PFAS are lower in technical performance than long-chain PFAS; thus, larger quantities of short-chain PFAS are required in products to reach a similar performance as long-chain PFAS (Lindstrom et al., 2011). Second, the toxicity and the effects on human health remain largely unknown. Third, shorter-chain PFAS could cross the placenta more efficiently than longer-chain PFAS, which has raised more concerns on the exposure among women in pregnancy (Chen et al., 2017; Needham et al., 2011; Zhang et al., 2013). These concerns underscore the need for future research on exposure distribution of PFPeA or its precursor and their potential health effects.

Higher SES has been linked to higher PFAS exposure in previous studies (Buekers et al., 2018; Nelson et al., 2012). A possible explanation is that the people who are wealthier or more highly educated have different lifestyles (e.g. purchasing more PFAS-containing products) and dietary habits (Herzke et al., 2012; Kato et al., 2011). In agreement with these results, we found higher PFAS concentrations in the higher education groups (Supplement Table 3), and PFNA levels were significantly higher among the participants with college education and above than those with less than a high school education in the adjusted regression models (Table 4). Although some studies indicated that family income was a significant or a stronger predictor among the SES-related factors (Buekers et al., 2018; Kato et al., 2014; Nelson et al., 2012; Sagiv et al., 2015; Tyrrell et al., 2013), we found mostly positive but non-significant associations between poverty income ratio and serum PFAS levels. The different results could be due to the different ranges of income levels among these populations, or the relationship between income and the behaviors linking to PFAS exposure.

Parous women had significantly lower PFHxS and PFOA concentrations in the adjusted models, and decreased GMs of PFAS were mostly shown among the women with more parity, which is consistent with previous findings (Berg et al., 2014; Brantsæter et al., 2013; Kato et al., 2014; Kingsley et al., 2018; Lauritzen et al., 2016; Manzano-Salgado et al., 2016; Sagiv et al., 2015). These results provide evidence of fetal transfer from the mother to her offspring through the placenta or breastfeeding (Kim et al., 2011). BMI was negatively associated with serum PFHxS, PFOS, and PFUnDA, in agreement with some previous findings (Berg et al., 2014; Kato et al., 2014). However, mixed results were found in the literature (Brantsæter et al., 2013; Hölzer et al., 2008; Ji et al., 2012; Jürgen et al., 2008; Rylander et al., 2009; Rylander et al. 2010; Sagiv et al., 2015). In contrast with lipophilic persistent organic pollutants, which are mainly stored in lipid-rich tissues, PFAS are distributed to protein-rich compartments such as the liver, kidneys, and blood; thus, dilution effects due to the amount of adipose tissue might be less significant for PFAS than more lipophilic compounds (The Agency for Toxic Substances and Disease Registry, 2018).

Higher PFAS concentrations were found among never tobacco users and marijuana users during pregnancy in this study. Previous studies have inconsistently reported the association between smoking status and PFAS (Brantsæter et al., 2013; Fei et al., 2007; Kato et al., 2014; Park et al., 2019; Sagiv et al., 2015), but only limited studies have reported the association between marijuana use and PFAS. In the Health Outcomes and Measures of the Environment (HOME) study, women who reported using marijuana during pregnancy had lower PFOS and PFHxS concentrations than non-users (Vuong et al., 2018). The reasons for these different findings remain unknown, but might be explained by different lifestyles, behaviors, or metabolism between tobacco/marijuana users and non-users (Eriksen et al., 2011; Lauritzen et al., 2016).

Serum PFAS concentrations were not associated with age in this study. Unlike the serum concentrations of lipophilic persistent pollutants which often increase when people aged, serum PFAS concentrations have shown inconsistent associations with age (Kato et al., 2011; Kato et al., 2014; Manzano-Salgado et al., 2016; Olsen et al., 2003; Rylander et al., 2009). Additionally, the associations between serum PFAS and age were often driven by the greater serum levels in older or younger individuals (Olsen et al., 2017); thus, little changes in PFAS by age were found in this study. The mixed results can also be explained by chemical persistence, and time elapsed after the peak emission (Quinn & Wania, 2012).

We observed that age of house is a significant predictor of serum PFOA and PFNA concentrations; for the other PFAS, although no significant result was found, the GMs of PFHxS and PFOS decreased when the age of house increased. Similar results were presented in previous studies, which showed that PFAS concentrations in indoor air or dust were negatively correlated with age of the residence. This finding can be explained by the difference in building construction and materials, and also the how long the materials have been used (Haug et al., 2011; Kubwabo et al., 2005).

Bottled water as a primary drinking water source was also associated with higher serum PFOS and PFOA concentrations. The result is somewhat unexpected since most studies have shown that the concentrations of PFAS in tap water were higher than in bottled water (Domingo & Nadal, 2019; Kaboré et al., 2018; Ünlü Endirlik et al., 2019), expect for contamination events (Department of Public Health, Bureau of Environmental Health, 2019). An increased PFAS level was shown among reproductive-aged women who mainly consumed tap water than those who mainly consumed bottled water in a highly polluted area in China (Zhou et al., 2019). It is possible that the effect was attributable to residual SES confounding rather than bottled water consumption since no dose-response relationship was found between PFAS levels and the amount of bottled water consumption.

Although positive associations were mostly found between serum PFAS levels and uses of the cosmetics and personal care products, only foundation use was significantly associated with higher PFOA and PFNA. The occurrence of PFAS in cosmetic and personal care products is not well characterized; however, some studies have detected PFAS in various products. For example, PFCAs were found in >86% of the cosmetic products, and >88% of the sunscreen samples in a Japanese study (Fujii et al., 2013). Samples taken from individuals who use of foundations displayed the highest detection frequencies and concentrations of PFAS among the other personal care products such as hair sprays, body lotions, cream, and powders in Denmark and Sweden (Ministry of Environment and Food of Denmark, 2018; Schultes et al., 2018). The sources of PFCAs in consumer products are postulated to be degradates or the results of biotransformation of polyfluorinated phosphate ester, which could either be an active ingredient or an impurity in cosmetic and personal care products (Fujii et al., 2013; Schultes et al., 2018). Due to higher PFAS concentrations among foundation users in this study and highly detected PFAS in the products, future study is needed to understand whether the use of cosmetic and personal care products, especially foundation, could be an important exposure source to PFAS among pregnant women.

Our study is limited in several ways. First, the questionnaires were not specially designed to evaluate predictors of PFAS exposure, which could limit our ability to identify the associations. For example, no information on the amount or frequency of cosmetic use was collected in the questionnaire; thus, we could not verify if serum PFAS concentrations increase as dose or frequency of foundation use increases. Second, most information was collected at the same time point in the early pregnancy in a cross-sectional study design, with home environment and behavior predictors collected in the mid-pregnancy, which suggests a potential of reverse causality in this study. Third, due to a wide list of predictors being examined, some of the findings might be due to chance. Fourth, diets were shown to be important predictors for serum PFAS (Berg et al., 2014; Domingo & Nadal, 2017; Jain, 2014; Wu & Kannan, 2019); however, we did not include the diet-related predictors for PFAS in this study. Fifth, the relative small sample size (n=130) for home environment and behavior predictors could reduce the statistical power, and generalizability in this study. Finally, the relatively low detection frequencies for some PFAS have limited our ability to investigate their association with the predictors.

In summary, our results indicate that PFAS are ubiquitous in the serum of pregnant African American women in Atlanta, Georgia. Characterizing PFAS exposure is important because the health burden and disparity found in this underserved and underrepresented population may be partly attributable to PFAS exposure, particularly among pregnant women and their fetuses when exposures occur at a relevant time window. We found higher serum PFHxS concentrations among the studied population than NHANES. Serum levels of a short-chain PFAS, PFPeA, increased by year in this population suggesting that PFPeA or a precursor might be a substitute to the long-chain PFAS which have been phased out. Future studies should consider measuring levels of shorter-chain PFAS and their precursors and investigating the associated health effects. Furthermore, we identified several important predictors for serum PFAS concentrations in this population, including parity, BMI, education, tobacco, and marijuana use, age of house, drinking water source, and cosmetic product use. This information can provide data for public health sectors to lower exposure or improve individual and population health by mitigating relevant exposure pathways.

Highlights.

• The four most common PFAS plus 10 additional PFAS were measured in 453 and 356 pregnant African American women, respectively.

• PFHxS, PFOS, PFOA, and PFNA were detected in > 95% samples.

• Higher PFHxS concentrations were found in the study sample as compared to NHANES data within the same demographic groups.

• Concentrations of PFPeA increased during 2014–2018.

• Education, parity, BMI, tobacco use, marijuana use, and year of sampling were significantly associated with some serum PFAS levels.