PFAS concentration during pregnancy in relation to cardiometabolic health and birth outcomes

Abstract

Introduction:

Poly- and perfluoroalkyl substances (PFAS) are persistent organic pollutants with pervasive exposure and suspected associations with metabolic abnormalities and adverse pregnancy outcomes. The goal of the present study was to examine the relationship between serum-PFAS concentrations measured in late pregnancy with relevant outcomes.

Methods:

The study sample included 433 pregnant women enrolled in the Vanguard Pilot Study of the National Children’s Study. Six PFAS were measured in primarily third trimester serum, as well as fasting insulin, total cholesterol, and triglycerides. The PFAS were examined in quartiles in relation to serum biomarkers, gestational age at birth and birth weight standardized for gestational age using multivariable-adjusted regression models.

Results:

Over 98% of the study population had detectable concentrations of four of the PFAS, and concentrations varied by race/ethnicity. Total cholesterol was positively associated with PFDA, PFNA, and PFOS, and triglycerides with PFDA, PFNA, PFOS, and PFOA, but PFAS were not associated with fasting insulin in adjusted models. Only PFNA was associated with an increased odds of birth at <37 weeks gestation. PFAS were generally not associated with birth weight, though PFHxS was associated with the first quartile of birth weight among males only.

Conclusions:

This study of pregnant U.S. women supports the ubiquitous exposure to PFAS and positive associations between PFAS exposure with serum-lipid concentrations. PFAS were largely unassociated with gestational age at birth and birth weight, though PFNA was associated with preterm birth. The results support the vulnerability to PFAS exposure of pregnancy.

1. Introduction

Per- and polyfluoroalkyl substances (PFAS) represent a class of persistent organic pollutants that have been used widely in consumer products for over 70 years as coatings to achieve nonstick and stain-resistant properties, in food packaging, and in firefighting foams (Agency for Toxic Substances and Disease Registry, 2018). The home environment, including house dust, and contaminated foods (e.g., seafood) are primary sources of human exposure, though drinking water in areas of the US has also been shown to be contaminated with PFAS due to nearby industrial and military use (Agency for Toxic Substances and Disease Registry, 2018). The major PFAS are resistant to environmental degradation and are bioaccumulative in tissues, all with elimination half-lives in humans of several years (Agency for Toxic Substances and Disease Registry, 2018).

PFAS are recognized endocrine disrupting chemicals, and animal studies have suggested multiple pathways of impact that include disruption of reproductive hormones and impaired signaling of thyroid hormones (Rappazzo et al., 2017). Higher concentrations of some PFAS, including PFOS, PFOA, and PFDA, have been associated with low birth weight in many studies, though only some reached statistical significance (Ashley-Martin et al., 2017; Bach et al., 2015; Manzano-Salgado et al., 2017; Marks et al., 2019; Sagiv et al., 2018; Valvi et al., 2017; Workman et al., 2019). There is also evidence that PFAS concentrations, particularly PFOS, PFOA, PFDA, and PFNA, may be associated with an increased risk of preterm birth (Meng et al., 2018; Sagiv et al., 2018). However, sample sizes have been small in some studies, and there were possible inconsistencies in the PFAS examined and their associations with birth outcomes. Relevant differences across studies include the gestational age at serum collection, adjustment for blood hemodynamics, inclusion of analyses stratified by offspring sex, the years during which PFAS data were collected in relation to population-level changes in exposure, the standardization of birth weight for gestational age at birth, the covariates included (e.g., diet), as well as statistical power and which PFAS were measured.

Data on the relationship between PFAS and cardiovascular risk factors in adults suggest associations between PFAS exposure and hypercholesterolemia and diabetes (Christensen et al., 2018; He et al., 2018; Sun et al., 2018; Seo et al., 2018). In particular, pollutant-associated gestational diabetes may impact the health of both mother and child (Valvi et al., 2017).

The goal of the present study was to examine the associations between serum-PFAS concentrations with total cholesterol, triglycerides, and fasting insulin among the pregnant mothers in the Vanguard Pilot study of the National Children’s Study, as well as the relationships of these parameters with offspring gestational age and size at birth.

2. Materials and Methods

2.1. Study Population

The study population included pregnant women enrolled in the Vanguard Pilot Study of the National Children’s Study (NCS) (Mortensen et al, 2014). The NCS was designed to be a nationally representative prospective cohort study of approximately 100,000 children from conception to age 21 to identify the associations of environmental exposures with health and developmental outcomes from pregnancy to adulthood. The NCS Vanguard Pilot study collected prenatal, birth, and early childhood data across 40 sites in the U.S. and was not nationally representative. Its goal was to determine the feasibility of the recruitment and data collection strategies and methodology proposed for the NCS (Mortensen and Hirschfeld, 2012). The current study comes from a pilot project to determine the feasibility of blood collection. All participants completed IRB-approved consent forms.

Recruitment for the Vanguard Pilot study was conducted in distinct phases with unique approaches beginning in 2009. The first involved neighborhood canvassing to locate and enroll women who were pregnant or likely to become pregnant, and the second phase involved healthcare provider-based recruitment, direct public outreach campaigns, and enhanced household-based sampling.

Women were enrolled in the Vanguard Pilot Study during preconception and pregnancy if they were between ages 18–49 years with a probability of becoming pregnant or pregnant women. Excluded were women who self-reported as infertile or prisoners, along with those who were unable to provide informed consent. In total, 89,886 were identified for contact in the Vanguard study, 74,004 completed the pilot screener, 11,165 were eligible for enrollment, 7,921 were enrolled, and of the 6,229 pregnant women, 5,420 remained in the study at child birth, for which birth weight was available for 4,653. Investigators collected blood samples primarily during the third trimester from approximately a quarter of the Initial Vanguard Study participants (N=433 out of approximately 1600), selected randomly, and PFAS was measured in all of these serum samples and were included in the current study.

2.2. Data Collection

Data collection included basic information on medical history, health behaviors and demographics assessed by questionnaire and telephone interview, physical assessments, biospecimens, and abstraction of prenatal, perinatal, and neonatal medical records. Dietary habits, including seafood consumption, were assessed by food frequency questionnaire during the third trimester of pregnancy. Race/ethnicity was based on self-report and categorized for the current study as Hispanic, non-Hispanic white, and non-Hispanic black. Educational attainment was examined as a proxy for socioeconomic status and was dichotomized in this analysis as any college vs. high school or less. Birth weight, recorded in grams, and gestational age in weeks, were collected primarily from neonatal medical record abstraction as well as participant self-report in study interviews.

Serum was analyzed from a random sample of fasting third-trimester mothers and analyzed at the CDC’s National Center for Environmental Health to quantify the concentration of six major PFAS using solid-phase extraction high-performance liquid chromatography isotope-dilution tandem mass spectrophotometry (Kato et al., 2011). The PFAS (level of detection, LOD in ng/mL) measured were 2-(N-Methyl-perfluorooctane sulfonamido) acetic acid (Me-PFOSA-AcOH, LOD=0.1), perfluorodecanoic acid (PFDA, LOD=0.1), perfluorononanoic acid (PFNA, LOD=0.08), perfluorooctane sulfonic acid (PFOS, LOD=0.2), perfluorooctanoic acid (PFOA, LOD=0.1), and perfluorohexane sulfonic acid (PFHxS, LOD=0.1). For samples that were below the LOD, the value of LOD/ √2 was imputed.

Triglycerides and total cholesterol were measured in the same serum samples at the CDC laboratory using the Roche Modular P Chemistry Analyzer (CDC, 2015). Both the triglycerides and total cholesterol assays use an enzymatic method, and the triglycerides one with glycerol blanking. Fasting insulin was measured in the same fasting serum, and analyzed using a solid-phase, two-site sequential chemiluminescent immunometric assay (Siemens IMMULITE® 2000 Immunoassay System).

The data used in this study are publicly available.

2.3. Statistical Analysis

The study population for the present study was restricted to those with any data collected on serum-PFAS concentrations (N=433). The primary independent variables of interest were the six measured PFAS concentrations, examined as continuous variables and in quartiles.

The dependent variables of interest measured in the pregnant women were total cholesterol, triglycerides, and fasting insulin, which were all examined as continuous measures. The dependent variables of interest measured in the offspring at birth were weight (grams) and gestational age (weeks). Birth weight was examined continuously as a z-score standardized for gestational age, and categorically as the first quartile of standardized birth weight z-score vs. quartiles 2-4. Gestational age was also examined continuously and dichotomized as <37 weeks (preterm) vs ≥37 weeks.

First, the distributions of the PFAS and dependent variables were examined in the study population. We calculated the distributions of PFAS by race/ethnicity, and determined differences using linear regression models, with natural log-transformed PFAS concentrations as the dependent variables to achieve near-normal distributions, while adjusting for maternal age. The survey procedure was used in SAS to account for the cluster sampling design (Montaquila et al., 2010). Second, we examined the associations between the log-transformed PFAS concentrations with total cholesterol, triglycerides, and fasting insulin first using Pearson’s correlation coefficient.

Multivariable-adjusted linear regression models were constructed to examine the associations between PFAS quartiles with total cholesterol, triglycerides, natural log-transformed insulin, gestational age at birth, and standardized birth weight z-score, adjusting for maternal age, college education, race/ethnicity, pre-pregnancy BMI, parity, prenatal smoking, and gestational age at serum collection. The means of each of these outcomes across PFAS quartiles were determined with adjustment for all covariates, and the trend p-values across the quartiles were calculated.

Multivariable adjusted logistic regression models were also constructed for the categorical outcomes gestational age <37 weeks and standardized birth weight in the first quartile. The PFAS quartiles were examined in relation to these dichotomous outcomes with the first PFAS quartile as the reference, and a p-value for trend across the quartiles was calculated. These analyses were adjusted for maternal age, college education, race/ethnicity, pre-pregnancy BMI, parity, prenatal smoking, and gestational age at serum collection. Analyses of birth weight were restricted to full-term births defined as 37 weeks gestation or later.

An exploratory analysis stratified by offspring sex was also conducted for PFAS quartiles in relation to birth weight z-score, assessed as a continuous variable in linear regression models and categorically as the first quartile vs quartiles 2-4 in logistic regression models, adjusted for all covariates. Lastly, we explored the possibility that seafood consumption might explain an association between PFAS and birth weight. Of note, exploratory spline curve analyses of PFAS in relation to gestational age and birth weight did not suggest any clear trends for the curve shapes.

The associations between fasting insulin, total cholesterol, and triglycerides with standardized birth weight z-scores were examined in separate linear regression models among term births ≥37 weeks gestation, adjusting for all covariates.

All analyses were conducted using SAS version 9.4 (SAS Institute, Cary, NC). We defined statistical significance as a two-sided alpha<0.05.

3. Results

Among the 433 NCS Vanguard mothers with PFAS analyses, complete data were available for total cholesterol and triglycerides, while fasting insulin was available for N=184 only, gestational age for N=390, and birth weight for N=354. Among all NCS participants, the mean±SD birth weight was 3353±535g, gestational age at birth was 38.7±1.9 weeks, 12% were Black, 19% Hispanic, and 60% white. Distributions of the PFAS concentrations, outcomes, and demographics in the study population with PFAS data available are shown in Table 1. Over 98% of the study population had detectable concentrations of PFNA, PFOS, PFOA, and PFHxS in serum. The mean total cholesterol was 6.15 mmol/L (standard deviation, SD=1.19), and triglycerides was 2.43 mmol/L (SD=0.97), the median fasting insulin was 55.56 (interquartile range, IQR 34.73-97.23) pmol/L, the mean birth weight was 3,370 grams (SD=530), the mean gestational age was 38.8 weeks (SD=1.7), and 7% were born before 37 weeks of gestation. A total of 52% of the children were male.

Table 1.

Distribution of variables in the study population.

PFAS	N detected	% detected	25%	Median	75%	Max
Me-PFOSA-AcOH (ng/mL)	266	61%	Non-detect	Non-detect	0.2	2.9
PFDA (ng/mL)	367	85%	0.1	0.2	0.3	2.6
PFNA (ng/mL)	433	100%	0.5	0.7	1.1	6.8
PFOS (ng/mL)	430	>99%	2.6	3.9	5.9	208.0
PFOA (ng/mL)	433	100%	0.9	1.4	2	7.2
PFHxS (ng/mL)	425	98%	0.3	0.5	0.9	12.6
Dependent variables	N	Mean (SD)	25%	Median	75%
Total cholesterol mmol/L	433	6.15 (1.19)	5.35	6.05	6.85
Triglycerides mmol/L	433	2.43 (0.97)	1.78	2.21	2.82
Fasting Insulin pmol/L	184	97.23 (131.96)	34.73	55.56	97.23
Birth weight (g)	354	3370 (530)	3080	3360	3675
Gestational age (weeks)	390	38.8 (1.7)	38	39	40
Demographics	N	Mean (SD)/%
Maternal age, mean (SD)	420	29.1 (5.7)
Race/ethnicity, %	420
Non-Hispanic White		79
Non-Hispanic Black		7
Hispanic		15
Education,%	416
<Highschool		27
Graduated high school/GED		30
Some college/associate		38
Bachelor’s		3
Post-graduate		1
Previous live births, %	424
0		32
1		33
2		23
3		8
≥4		5

Among the 4,653 NCS participants with birth weight data available, there was no statistically significant difference in birth weight between those with and without data on serum-PFAS concentrations. Also, among the 433 with PFAS data, there was no statistically significant difference in most PFAS concentrations between those with and without available data on birth weight and fasting insulin, with two exceptions: PFNA was lower among those with missing insulin (p=0.03), and PFHxS was lower among those missing birth weight information (p=0.01).

Table 2 shows the distribution of serum-PFAS concentrations by race/ethnicity. Although the trends varied across the PFAS, the disparity for Me-PFOSA-AcOH was just shy of statistical significance. PFDA and PFNA were highest in Black participants, while PFHxS was highest in whites. The only PFAS associated with education level was Me-PFOSA-AcOH, which was higher among those with any college education (p=0.03). Adjusting for educational attainment did not impact the relationship between race/ethnicity and Me-PFOSA-AcOH (data not shown).

Table 2.

Serum-PFAS concentrations (ng/mL) by race/ethnicity

	White (N=271)		Black (N=23)		Hispanic (N=51)		P-value*
	Median	IQR	Median	IQR	Median	IQR
Me-PFOSA-AcOH	ND	ND-0.2	ND	ND-0.3	ND	ND-0.1	0.07
PFDA	0.2	0.1-0.3	0.3	0.1-0.5	0.2	0.1-0.4	0.03
PFNA	0.7	0.5-1.0	1.0	0.7-1.5	0.8	0.6-1.1	0.02
PFOS	4.0	2.8-5.8	4.5	3.0-10.6	3.6	2.1-5.2	0.04
PFOA	1.4	1.0-2.2	1.4	1.0-2.1	1.1	0.9-1.7	0.02
PFHxS	0.6	0.4-1.0	0.4	0.3-1.1	0.3	0.2-0.5	0.01

^*Global p-values generated from linear regression models after log transformation of PFAS, adjusted for maternal age

The covariate-adjusted distributions of insulin, total cholesterol, triglycerides, gestational age at birth, and standardized birth weight z-scores across quartiles of PFAS concentrations are shown in Figures 1–5. Total cholesterol was positively associated with PFDA, PFNA, and PFOS (Figure 1, Table 3). Quartiles of PFDA, PFNA, PFOS, and PFOA were positively associated with triglycerides (Figure 2). For fasting insulin there were positive correlations with PFOS and PFNA (Table 3), but these associations were not apparent in multivariable-adjusted analyses of PFAS quartiles (Figure 3).

Figure 1.

Trend p-values across serum-PFAS quartiles generated from linear regression models, adjusted for maternal age, education, race/ethnicity, pre-pregnancy BMI, prenatal smoking, parity, and gestational age at serum collection.

Figure 5.

Trend p-values across serum-PFAS quartiles generated from linear regression models, adjusted for maternal age, education, race/ethnicity, pre-pregnancy BMI, prenatal smoking, parity, and gestational age at serum collection. This analysis was restricted to births at ≥37 weeks gestation.

Table 3.

Pearson correlation coefficient, and P-value, for associations between PFAS (natural log transformed) and fasting insulin (natural log transformed) and lipid concentrations.

	Insulin (N=184)	Total cholesterol (N=433)	Triglycerides (N=433)
Me-PFOSA-AcOH	−0.07, 0.34	0.06, 0.24	0.07, 0.14
PFDA	0.12, 0.11	0.11, 0.03	0.08, 0.10
PFNA	0.19, 0.01	0.11, 0.02	0.07, 0.17
PFOS	0.15, 0.04	0.13, 0.01	0.04, 0.35
PFOA	0.07, 0.36	0.07, 0.14	0.03, 0.55
PFHxS	0.06, 0.42	0.07, 0.15	0.01, 0.91

Figure 2.

Trend p-values across serum-PFAS quartiles generated from linear regression models, adjusted for maternal age, education, race/ethnicity, pre-pregnancy BMI, prenatal smoking, parity, and gestational age at serum collection.

Figure 3.

Trend p-values across serum-PFAS quartiles generated from linear regression models with log-transformed insulin as the dependent value, adjusted for maternal age, education, race/ethnicity, pre-pregnancy BMI, prenatal smoking, parity, and gestational age at serum collection.

There were no statistically significant associations between PFAS with gestational age at birth (Figure 4) or birth weight (Figure 5) when these outcomes were examined continuously. There was no association between PFAS and odds of having an offspring in the first quartile of standardized birth weight (Table 4). However, increasing PFNA quartiles were associated with gestational age at birth <37 weeks (trend across quartiles OR=1.44, 95% CI: 1.01-2.06), while Me-PFOSA-AcOH showed borderline association with a decreased odds of gestational age at birth <37 weeks (OR=0.58, 95% CI: 0.33-1.01).

Figure 4.

Trend p-values across serum-PFAS quartiles generated from linear regression models, adjusted for maternal age, education, race/ethnicity, pre-pregnancy BMI, prenatal smoking, parity, and gestational age at serum collection.

Table 4.

Associations between maternal serum-PFAS concentrations with gestational age at birth and birth weight

PFAS	Gestational age <37 weeks (vs ≥37 weeks)	Standardized birth weight z-score in the first quartile (vs quartiles 2-4)
	OR, 95% CI*	OR, 95% CI*¥
Me-PFOSA-AcOH
Quartile 3 vs 1-2	0.93 (0.39-2.19)	1.06 (0.56-2.01)
Quartile 4 vs 1-2	0.20 (0.05-0.94)	0.73 (0.36-1.44)
Quartile Trend p-value	0.05	0.42
PFDA
Quartile 2 vs 1	0.60 (0.14-2.59)	0.86 (0.34-2.19)
Quartile 3 vs 1	1.13 (0.31-4.11)	0.98 (0.40-2.40)
Quartile 4 vs 1	1.82 (0.54-6.19)	1.27 (0.50-3.21)
Quartile Trend p-value	0.12	0.43
PFNA
Quartile 2 vs 1	0.48 (0.12-1.94)	1.39 (0.65-2.94)
Quartile 3 vs 1	1.16 (0.38-3.61)	1.68 (0.78-3.62)
Quartile 4 vs 1	2.38 (0.87-6.49)	1.04 (0.46-2.34)
Quartile Trend p-value	0.04	0.79
PFOS
Quartile 2 vs 1	1.94 (0.66-5.68)	0.93 (0.43-2.04)
Quartile 3 vs 1	1.13 (0.34-3.73)	0.81 (0.36-1.82)
Quartile 4 vs 1	1.41 (0.46-4.33)	1.41 (0.66-3.03)
Quartile Trend p-value	0.82	0.40
PFOA
Quartile 2 vs 1	3.17 (0.94-10.70)	1.20 (0.56-2.59)
Quartile 3 vs 1	3.14 (0.95-10.31)	0.84 (0.40-1.80)
Quartile 4 vs 1	1.38 (0.32-5.97)	0.91 (0.41-2.02)
Quartile Trend p-value	0.53	0.62
PFHxS
Quartile 2 vs 1	2.11 (0.76-5.81)	0.97 (0.42-2.24)
Quartile 3 vs 1	1.21 (0.37-3.92)	1.74 (0.81-3.74)
Quartile 4 vs 1	1.27 (0.39-4.17)	1.20 (0.55-2.62)
Quartile Trend p-value	0.85	0.44

^*Adjusting for maternal age, education, race/ethnicity, pre-pregnancy BMI, prenatal smoking, parity, and gestational age at serum collection

^¥Restricted to births at ≥37 weeks gestation

Information on seafood consumption was available for 271 participants with PFAS data. In this sample there was no statistically significant association between seafood consumption (in daily oz) and standardized birth weight (linear regression Beta=−0.50, p=0.75, adjusting for maternal age, education, race/ethnicity, pre-pregnancy BMI, parity, and smoking). Among the natural log-transformed serum-PFAS concentrations, PFDA was associated with reported seafood consumption (linear regression adjusted Beta=0.18, p=0.004), as was PFNA (adjusted Beta=0.13, p=0.01). The associations between PFAS and birth weight remained virtually unchanged in analyses that controlled for seafood consumption.

There were no statistically significant interactions between PFAS and offspring sex in relation to birth weight (p>0.10), adjusting for all covariates, and stratified analyses suggested no differences between male and female births in relation to birth weight assessed continuously (not shown). However, PFHxS quartiles were associated with an increased odds of having a standardized birth weight in the first quartile (vs the top three quartiles of birth weight) among male births (fully adjusted OR=1.57, 95% CI: 1.05-2.36), but not among female birth (fully adjusted OR=0.87, 95% CI: 0.61-1.25). In offspring sex-stratified adjusted analyses, all other PFAS were unrelated to being in the first quartile of birth weight.

We did not observe statistically significant associations between maternal insulin levels, triglycerides, or total cholesterol with offspring standardized birth weight Z-scores, adjusting for all covariates (results not shown).

Only 6 study participants had babies classified as being small for gestational age (Villar et al, 2014) and only 18 had babies classified as low birth weight at <2,500 grams. Therefore, we were not able to examine the detailed relationship between PFAS and these cut-points for fetal growth in the present study.

4. Discussion

In the present study of pregnant women and their offspring in the U.S., maternal serum concentrations of multiple PFAS were positively associated with serum concentrations of total cholesterol and triglycerides, particularly PFDA, PFNA, PFOS, and PFOA. PFNA and PFOS were associated with fasting insulin only in unadjusted analyses. PFAS were largely not associated with gestational age at birth and birth weight, though PFHxS was associated with being in the first quartile of birth weight among male offspring, and PFNA was associated with birth at <37 weeks gestation in the full study sample.

These results are consistent with NHANES data demonstrating that PFAS exposure is ubiquitous in the U.S. (Calafat et al., 2019), and our results on limited racial and educational differences are also consistent with findings from NHANES (Calafat et al., 2007; Nelson et al., 2012).

The present findings are consistent with previous studies suggesting an association between increased PFAS exposure and elevated lipid levels (Christensen et al., 2018), though the literature is not entirely unequivocal (Agency for Toxic Substances and Disease Registry, 2018). PFAS, particularly PFOA, have been positively associated with total cholesterol in some studies. Evidence from both cross-sectional and prospective studies of PFOA-exposed workers suggests that increased PFOA exposure is associated with higher serum-cholesterol concentrations, often exceeding the reference range (Costa et al., 2009; Sakr et al., 2007a; Sakr et al., 2007b). Positive associations were also found in regard to PFOS exposure (Frisbee et al., 2010; Nelson et al., 2010; Steenland et al., 2009), However, data on long-chain PFAS more recently introduced (e.g., PFDA, PFNA) remain limited (Sunderland et al., 2018; Winquist et al., 2014). A study conducted in Korean adults indicated positive associations between PFAS with total cholesterol, LDL, and triglycerides and a negative correlation with HDL (Seo et al., 2018). NHANES data from 2003-2012 suggested a positive association between PFOA and total cholesterol in both men and women, though associations were not observed for other types of PFAS (He et al., 2018). In the prospective Nurses’ Health Study II, an increased risk of diabetes was observed among women with higher plasma concentrations of PFOS and PFOA before the beginning of the follow-up (Sun et al., 2018). Another prospective study in the US, the Diabetes Prevention Program Trial, demonstrated cross-sectional associations between plasma PFAS concentrations and markers of insulin secretion and beta-cell function, but no longitudinal associations with diabetes incidence (Cardenas et al., 2017).

Epidemiological data regarding the relationships between PFAS with insulin resistance and glucose metabolism have been inconsistent, perhaps in part due to differences in study designs and methods. Cross-sectional NHANES data suggest an association between PFOA with β-cell function, and PFOS with higher insulin, HOMA-IR, and β-cell function (Lin et al., 2009). It has been suggested that confounding by diet and also changes in exposures over time may help explain some inconsistencies in the current literature, and therefore data from other countries may not be generalizable to US populations with different or variable food production and consumption practices (Donat-Vargas et al, 2019). In the current study, the correlations for PFNA and PFOS with fasting insulin did not remain robust in the fully adjusted models, but our sample size with available fasting insulin was small.

Obesity may be an important mediating factor for the associations between PFAS and lipids and insulin resistance, though there was no observed association between PFAS and recorded pre-pregnancy BMI in this population. In a recent diet-induced weight loss trial, participants with higher baseline plasma PFAS concentrations experienced greater weight regain and altered patterns of resting metabolic rate changes through weight loss and regain (Liu et al., 2018). Also, a prospective European study demonstrated an association between childhood PFOS concentration with adiposity during adolescence (Domazet et al., 2016).

Several studies have suggested an association with preterm birth for PFDA, PFNA, PFOS, and PFOA (Apelberg et al., 2007; Fei et al., 2007; Meng et al., 2018; Sagiv et al., 2018). The results of the present study are consistent with a possible association between increased concentrations of PFNA with preterm birth, though the statistical power was limited. The relationships between PFAS and gestational size has received more attention in the literature with many studies suggesting that increased exposure to PFAS was associated with low birth weight, particularly for PFDA, PFOS, and PFOA (Ashley-Martin et al., 2017; Bach et al., 2015; Manzano-Salgado et al., 2017; Marks et al., 2019; Sagiv et al., 2018; Valvi et al., 2017; Workman et al., 2019). However, with small sample sizes common in the existing literature, statistical significance was often not reached in previous studies (Agency for Toxic Substances and Disease Registry, 2018). Overall, the current data do not support an association between elevated PFAS exposure with decreased birth weight but would be in accordance with a small decrease. The associations between PFAS and gestational growth parameters have varied across PFAS chemicals and between boys and girls (Cao et al., 2018), and previous data have suggested associations between PFAS and birth weight in boys but not in girls (Valvi et al., 2017; Bach et al., 2015). Though our tests for effect modification by offspring sex were underpowered and not statistically significant, we found that PFHxS was associated with being in the first quartile of birth weight among boys only. Interestingly, in the Avon Longitudinal Study of Parents and Children cohort, first trimester PFHxS was associated with decreased birth weight among girls, and though the direction of the effect estimate was consistent for boys it did not reach statistical significance (Marks et al., 2019). Although data have suggested associations between prenatal PFAS exposure with decreased birth weight, prenatal PFAS exposure has also been associated non-linearly with increased weight and adiposity during childhood, with data suggesting that PFAS exposure may accelerate the accumulation of fat mass during infancy. However, these trends have also varied by PFAS chemical and by sex (Gyllenhammar et al., 2018; Karlsen et al., 2017; Starling et al., 2019).

Mechanisms underlying an association between PFAS exposure and fetal growth remain poorly understood, though PFAS exposure has been associated with low weight at birth in animal models (Koustas et al, 2014). The possible occurrence of gestational diabetes apparently does not affect PFAS-associated decreases in birth weight (Valvi et al., 2017). Effects of PFAS on lipid metabolism have been hypothesized to play a mediating role in the relationship with fetal growth restriction, though we did not observe an association between cholesterol levels or triglycerides and birth weight in the current study.

Strengths of the study include the measurement of six PFAS in a sample size larger than in many previous studies. The analysis of PFAS in relation to lipids and insulin is cross-sectional, limiting inferences related to causality. However, this one-time assessment can be considered to represent long-term exposure due to the long half-lives of PFAS (Agency for Toxic Substances and Disease Registry, 2018). An additional limitation is the possibility of residual confounding by factors associated with PFAS exposure as well as metabolic and fetal health, such as dietary habits, income, and breastfeeding history. Because serum concentrations of lipid and insulin are impacted by pregnancy and gestational age, the observed associations with PFAS may not be generalizable to non-pregnancy. Measurement of serum-PFAS during the third trimester of pregnancy also represents a possible limitation, and several previous studies have measured prenatal PFAS exposure during the beginning of pregnancy, which may represent a more valid approach due to the impact of plasma volume expansion and glomerular filtration rate changes that occur during pregnancy. Plasma volume and glomerular filtration rate both increase in the early stages of pregnancy, may lead to enhanced PFAS dilution and excretion, respectively, and are believed also to impact fetal growth. Therefore, observed associations between prenatal PFAS concentrations and fetal growth outcomes may be biased without accounting for these hemodynamic changes (Sagiv et al., 2018), and small adverse effects may be difficult to identify.

It is important to note that the study population, although cluster-sampled, does not represent a population-based study, that serum-PFAS concentrations were obtained only in 433 pregnant mothers, that there was missingness for birth weight and fasting insulin for many participants with PFAS data available, and that the possibility of selection bias could not be excluded. However, the lack of association between PFAS data availability with offspring birth weight in the full study population, and the lack of association between birth weight and fasting insulin availability with PFAS concentration in the study population suggest that selection bias is unlikely. Calculation of gestational age based on reliable recording of last menstrual period or ultrasound data was not possible for all participants, and therefore some small and misclassification of gestational age is probable, though bias due to this random misclassification is unlikely to impact the conclusions of this study. Lastly, we did not have a sufficient sample size to examine PFAS in relation to clinical outcomes including gestational diabetes and intrauterine growth restriction.

PFAS exposure represents a research and public health priority due to their ubiquitous presence in blood samples from populations worldwide, and their persistent and bioaccumulative nature. Despite increasing regulations and proposed restrictions, their health and environmental effects are expected to persist for decades, highlighting the need to better elucidate the strength and range of their health impacts.

5. Conclusions

The present study utilized the available data on PFAS exposure and clinical parameters obtained in the Vanguard Pilot study to explore the public health concerns about PFAS exposure during pregnancy. Given the sample size of the study, the patterns of association are more relevant than the p-values. The results support a growing body of literature suggesting that increased PFAS exposure is associated with elevated serum-lipid concentrations and insulin, and overall do not support any strong association between increased PFAS exposure with lower birthweight or gestational age at birth.

Highlights.

• PFAS exposure was ubiquitous in a sample of third-trimester pregnant women in the U.S.

• Serum concentrations of PFDA, PFNA, and PFOS were associated with total cholesterol and triglycerides during pregnancy.

• PFAS exposure was largely not associated with gestational age at birth or birth weight.

Abbreviations:

BMI

Body mass index

CI

Confidence interval

CV

Coefficient of variation

GA

Gestational age

HOMA-IR

Homeostatic model assessment for insulin resistance

IQR

interquartile range

LH

Luteinizing hormone

LOD

Limit of Detection

Me-PFOSA-AcOH

2-(N-Methyl-perfluorooctane sulfonamido) acetic acid

NCS

National Children’s Study

PFAS

Poly- and perfluoroalkyl substances

PFDA

Perfluorodecanoic acid

PFHxS

Perfluorohexane sulfonic acid

PFOA

Perfluorooctanoic acid

PFOS

Perfluorooctane sulfonic acid

PFNA

Perfluorononanoic acid

SD

standard deviation