Prenatal Exposure to PFAS, Associations with Preterm Birth and Modification by Maternal Estrogen Levels: The Maoming Birth Study

Abstract

Background:

Estrogens play a critical role in parturition, and poly- and perfluoroalkyl substances (PFAS), which have estrogenic effects, have been associated with preterm birth. However, the impact of estrogens on the association between PFAS and preterm birth is unknown.

Objective:

The objective of this study is to investigate if estrogens modified the association between PFAS and preterm birth, using a nested case-control study design.

Methods:

A total of 371 preterm births and 508 controls were selected from a birth cohort study in China between 2016 and 2018. Perfluorobutanoic acid (PFBA), perfluorohexanesulfonic acid (PFHxS) and its branched isomer, perfluorooctanoic acid (PFOA), perfluorooctanesulfonic acid (PFOS) and its branched isomer, and perfluorononanoic acid (PFNA) were quantified in maternal serum (mean gestational age of 32 wk). Estradiol and estriol were quantified in cord serum. Preterm birth was defined as live delivery at $<37$ gestational weeks. Causal mediation analysis was used to estimate the mediation and interaction effects of estrogen on the association between PFAS and preterm birth. Latent profile analysis was used to identify important estrogen profiles. Multiple linear regression was used to estimate associations between PFAS and preterm birth and interactions between PFAS and estrogens on preterm birth.

Results:

Overall, higher odds ratios (ORs) of preterm birth were associated with each 1 ln-unit PFAS increase: PFBA [1.20, 95% confidence interval (CI): 1.14, 1.26], PFNA (1.30, 95% CI: 1.21, 1.39), PFOA (1.98, 95% CI: 1.54, 2.55), and PFOS (1.91, 95% CI: 1.76, 2.07) and its branched isomer (1.91, 95% CI: 1.90, 1.92). We detected statistically significant interactions between cord estradiol and PFAS on preterm birth, while no mediation effects of cord estrogen were observed. The ORs of PFOS (4.29, 95% CI: 1.31, 8.25), its branched isomer (6.71, 95% CI: 1.06, 11.91), and preterm birth were greater for participants with high cord estrogen levels than for participants with low cord estrogen levels.

Discussion:

Our findings suggest that estrogen modified the association between maternal PFAS exposure and preterm birth. Further studies on maternal PFAS exposure and preterm birth, taking interaction effects of cord estrogens into account, are warranted. https://doi.org/10.1289/EHP11377

Introduction

Poly- and perfluoroalkyl substances (PFAS) comprise more than 4,000 unique chemicals, some of which are ubiquitous in the environment.¹ Exposure to PFAS is a growing public health concern. Importantly, PFAS cross the placental barrier² and consequently, both mothers and their fetuses are at risk from prenatal exposure.³ While human PFAS exposure has been recognized for more than 20 years,⁴ questions remain regarding their reproductive health impacts, including how they affect birth outcomes, including preterm birth. This is especially important in the context of the increasing levels of human exposure to PFAS resulting from rapid industrialization in China.⁵

Preterm birth (i.e., delivery at $<37\mathrm{wk}$ gestational age) is not only the leading cause of death among children younger than 5 years old but also is associated with a wide array of adverse health effects across the lifespan.⁶ Despite growing concerns about the reproductive toxicity of PFAS, current evidence of associations between PFAS exposure and preterm birth is inconsistent. For example, some studies report a positive association,^7,8 but others have reported null effects of PFAS exposure on preterm birth.^9,10

One reason for the conflicting results from studies of PFAS exposure and preterm birth may be an underestimation of the endocrine disrupting effects of PFAS during pregnancy, if the potential effects of estrogen were not considered.¹¹ Estrogen is a sex-steroid hormone, largely synthesized by the placenta from precursor androgens produced by the fetal adrenal glands.¹² Increasing fetal and placental estrogen biosynthesis results in a progressive cascade of biological processes that lead to parturition, and these biological processes might be disturbed by exposure to endocrine disrupting environmental pollutants.^13,14 For example, Wang et al. found that cord serum estriol levels mediated the association between perfluorooctanesulfonic acid (PFOS) and birth weight, suggesting that estrogen may be an important mediator of associations of PFAS with preterm birth.¹⁵ Bjerregaard-Olesen et al. observed that PFAS-induced xenoestrogenic activity was associated with lower birth weight and length.¹⁶ An in vitro study observed that co-exposure to PFOS or perfluorooctanoic acid (PFOA) and estradiol enhanced the effects of estradiol on an estrogen-responsive pathway,¹⁷ further indicating that PFAS exposure may interact with estrogen on preterm birth. However, to our knowledge, there is no prior study of estrogen’s potential mediation or modification effects on the association between PFAS exposure and preterm birth.

Another issue that should be considered in the context of estrogen’s potential effects of PFAS toxicity is the complex interplay among estrogen types. Estrogen is comprised of three major types as follows: estrone, estradiol, and estriol. Different estrogen profiles may modify the effects of PFAS exposure on pregnancy differently.¹⁸ However, previous studies evaluated each estrogen type in isolation and did not account for the different biological functions of and transformations between different estrogen types.^19,20 Assessing modification of PFAS-preterm birth associations by the different estrogen types may shed light on the reproductive toxicity of PFAS exposure and inform public health strategies for maternal and child health.

In this context, we examined associations between maternal PFAS exposure and preterm birth in 879 Chinese women and then further explored if these associations were modified by estrogen profile.

Methods

Study Population

We conducted this nested case-control in the Maoming birth cohort study, a longitudinal investigation of pregnant women and their offspring, who delivered at the Maternal and Child Health Hospital in Maoming, China, between 2016 and 2018. The Maternal and Child Health Hospital is the sole “grade A” tertiary delivery hospital in Maoming city, which offers specialist and high-risk obstetrical care and services for close to 25% of the newborns in this region. Women were approached for study enrollment during their first prenatal care visit. A total of 15,635 pregnant women who lived in Maoming city over 2 years were enrolled. Participants completed a face-to-face study questionnaire administered by research assistants before delivery. Of these, 5,202 participants completed the study questionnaire data and provided both maternal and cord blood samples. We enrolled all preterm babies (407 [7.8%]) as the case group. We initially selected one control participant for each case matched on the date of delivery and later expanded the control sample by 50% because of a lower return visit rate in the cohort study, and so 619 controls were selected. After excluding mothers with liver, cardiovascular disease ($n=67$) and multiple gestation ($n=80$), a total of 371 cases and 508 controls were included in the final analysis (Figure S1). All research protocols were approved by the Human Studies Committee of Sun Yat-sen University, and informed consent was provided by each participant.

Outcome Assessment

Gestational age was estimated by the attending obstetrician according to Naegele’s rule, using each participant’s self-reported last menstrual period (LMP). The controls were defined as delivery at $\ge 37$ gestational weeks including term/later births, while preterm birth was defined as delivery before 37 gestational weeks. We further classified preterm birth as early/late preterm birth (delivery between 32 and 37 wk), very early preterm birth (delivery before 32 wk), spontaneous preterm birth (presentation of premature rupture of the membranes, spontaneous preterm labor, or both), and medically indicated preterm birth (preterm births with preeclampsia, gestational diabetes mellitus, or with both artificial membrane rupture and induced labor).

Biospecimen Collection and PFAS Assessment

We drew $5\mathrm{mL}$ of venous blood from women during pregnancy (mean gestational age: 32 wk, range: 7–40 wk), and the serum was separated and stored in $-80°\mathrm{C}$ freezers. Detailed methods for the determination of PFAS are described elsewhere.²¹ Briefly, all PFAS were determined by ultra-performance liquid chromatography–tandem mass spectrometry (UPLC-MS/MS). Each batch of 22 maternal serum samples included quality control (QC) materials and reagent blanks. Values less than the limits of detection (LODs) were replaced by $\text{LOD}/\surd 2$.²² The LODs, abbreviations, and nomenclature are provided for all PFAS in Table S1. Seven PFAS with high environmental and dietary concentrations in China and the United States^23,24 were selected for statistical analysis a priori, including perfluorobutanoic acid (PFBA), perfluorohexanesulfonic acid (PFHxS) and its branched isomer (br-PFHxS), PFOA, PFOS and its branched isomer (br-PFOS), and perfluorononanoic acid (PFNA).

Estrogen Measurements

We also collected $5\mathrm{mL}$ of venous umbilical cord blood at delivery. Cord serum was stored in $-80°\mathrm{C}$ freezers after separation. Estradiol and estriol were measured by immunoluminometric assay using an Architect random access assay system (Abbott Diagnostics, Abbott Park, IL, USA). The detailed measurement method was reported previously.²⁵ Briefly, we used $50\mathrm{\mu }\mathrm{L}$ of cord blood serum for estradiol and estriol measurements. We analyzed each sample in duplicate and then averaged the two measurements and used this value as the concentration of each sample. The LODs of estradiol and estriol were $5\text{\hspace{0.17em}pg}/\text{ml}$ and $0.5\text{\hspace{0.17em}ng}/\mathrm{mL}$, respectively.

Covariates

Potential covariates were selected a priori based on important covariates from previous literature^26,27 and using a directed acyclic graph (DAG)²⁸ (Figure S2), including household annual income ($<\mathrm{30,000}$ Yuan, 30,000–100,000 Yuan, and $>\mathrm{100,000}$ Yuan), maternal highest educational level (no school, elementary school, middle school, high school, and higher education), maternal alcohol consumption during pregnancy (more than one time per week or not), biospecimen collection time during pregnancy (days gestation), date of delivery, and environmental tobacco exposure during pregnancy (yes, no). Demographic information (e.g., household annual income and maternal highest educational level) and behavioral information (e.g., maternal alcohol consumption during pregnancy and environmental tobacco exposure during pregnancy) was self-reported on questionnaires. All of the above information was collected before delivery during the third trimester, except for cases of extremely preterm birth ($<28\mathrm{wk}$), in which information was collected postnatally. We did not include the maternal smoking variable, as only 18 participants (2%) reported previous or current smoking. Data on maternal age, maternal prepregnancy body mass index (BMI), number of pregnancies, infant sex, subtype of preterm birth, delivery information (e.g., household address, date of delivery, gestational age), and participants’ medical history information during pregnancy, including coronary heart disease, cerebrovascular disease, hypertension, and liver and kidney disease, were collected from the medical record corresponding to the most recent hospital visit.

Statistical Analysis

The distributions of demographic and clinical characteristics were expressed as mean [standard deviation (SD)], number (percentage), or median [interquartile range, (IQR)]. A comprehensive set of statistical tests, including $t$-tests for continuous variables, Kruskal-Wallis tests for nonparametric analysis, ${\mathrm{\chi }}^2$ tests for categorical variables, and analysis of variance (ANOVA) for comparing multiple groups, was employed to evaluate differences between the cases and control study groups. Five maternal prepregnancy weight values were missing and imputed using stochastic regression methods.²⁹ Inverse probability of sampling weights were applied to the subsequent analysis³⁰ to ensure that the study population was representative of the 5,202 participants in the cohort. Skewed variables, including PFAS, were normalized using a natural logarithmic transformation. Statistical significance was defined as $p<0.05$ for a two-tailed hypothesis test. All statistical analyses were performed using STATA version 17.0 (Stata Corp LP, College Station, TX, USA).

Main analysis.

The associations between PFAS exposure and preterm birth were estimated using logistic regression models, accounting for the region of residence, defined as city or rural area according to the Chinese household registration system (“Hukou”), as a random effect, and adjusted for the aforementioned covariates. Considering that sex-specific effects have been identified in previous studies,^15,21 we also repeated the main analysis stratified by infant sex and estimated the interactions between PFAS and infant sex on preterm birth by including a PFAS*sex cross-product term. Stratified analyses were also used to explore the relationships between PFAS and different subtypes of preterm birth, including spontaneous, medically indicated, very early ($<32\mathrm{wk}$) and early/late (32–37 wk).

Causal mediation analysis.

Previous evidence suggests that the maternal estrogen level is likely to intervene in the causal pathway between maternal PFAS exposure and preterm birth.¹⁸ However, cord estrogen level, which has been used to represent late gestational estrogen concentrations,³¹ may play a more important role in preterm birth than maternal estrogen level.³² Thus, we used cord estrogen levels in the following analysis. We first examined the associations between estrogen and PFAS or preterm birth using linear regression or logistic regression, respectively. Then, we used the “med4way” package in STATA software to perform a causal mediation analysis and estimated 95% confidence intervals (CIs) by bootstrapping with 1,000 resamples.³³ This approach estimated cord estrogens as potential mediators of PFAS-preterm birth associations, allowing for an interaction between PFAS and cord estrogens, adjusting for maternal age, BMI, education, parity, environmental tobacco smoke exposure (ETS), alcohol intake, household income, date of delivery, and biospecimen collection time during pregnancy. We assumed the following: a) no unmeasured confounders of the association between PFAS and preterm birth, b) no unmeasured confounders of the association between cord estrogens and preterm birth, c) no unmeasured confounders of the association between PFAS and cord estrogens, and d) that PFAS did not cause any confounder of the association between cord estrogens and preterm birth. We decomposed the total PFAS-preterm birth association into four components to describe the interactive, mediating, and direct effects of PFAS with preterm birth as shown in Figure S3, including:

1. the controlled direct effect, which is the difference in the log odds of a preterm birth for a 1-unit increase in PFAS when estrogen is fixed at the median level in the controls of $\text{estradiol}=2.42\text{\hspace{0.17em}ng}/\mathrm{mL}$ or $\text{estriol}=135.60\text{\hspace{0.17em}ng}/\mathrm{mL}$;

2. the natural direct or “interactive” effect, which corresponds to the multiplicative interaction effect in the absence of mediation and is the difference in the log odds of a preterm birth for co-exposure to a 1-unit increase in PFAS with a 1-unit increase in estrogen, compared to the sum of the individual differences of a 1-unit increase in PFAS without estrogen and a 1 unit increase in cord estrogen without PFAS;

3. the mediated interaction, which corresponds to the multiplicative interaction effect in the presence of mediation, and is the difference in the log odds of a preterm birth for co-exposure to a 1-unit increase in PFAS with a 1-unit increase in estrogen, at different measured values of estrogen; and

4. the pure indirect effect, which corresponds to the mediating effect of estrogen without a multiplicative interaction and is the difference in the log odds of preterm birth with a 1-unit increase in PFAS multiplied by a 1-unit increase in estrogen.

The proportion of interactive effects was defined as the sum of the proportions of the overall effect due to the natural direct (“interactive”) effect (i.e., “2” above) and the mediated interaction (i.e., “3” above). The proportion of mediating effects was defined as the sum of the proportions of the overall effect that were due to the mediated interaction (i.e., “3” above) and the pure indirect effect (i.e., “4” above).

Exploratory analysis.

We used a latent profile analysis (LPA), with the “gsem” procedure in STATA software, to identify estrogen profiles that accommodated a correlation between cord serum estradiol and estriol as moderators of PFAS-preterm birth associations, adjusting for maternal age and infant sex. Specifically, the gsem procedure maximizes between-class variance and minimizes within-class variance to categorize participants into mutually exclusive subgroups, or “profiles,” of an unobserved latent categorical variable based on the relative similarity of their exposure indicator value patterns. Bayes’ theorem was utilized to assign each study participant to the estrogen profile for which they had the highest probability of membership. We estimated models using 1–5 profiles and selected the best fit model using Akaike’s information criterion (AIC), Bayesian information criterion (BIC), and Lo-Mendell-Rubin test, as shown in Table S2.³⁴ Estrogen profiles were characterized by the z-standardized estimated levels of estradiol and estriol. We identified three estrogen profiles as follows: “least estrogen,” “high estradiol,” and “high estriol” as shown in Figure S4. Subsequently, Wald tests were conducted to assess the interaction between PFAS and estrogen profiles by including a PFAS*estrogen profiles cross-product term. We then explored the effect of PFAS on preterm birth across different estrogen profiles.

Sensitivity analyses.

We repeated the main analysis in 263 preterm and to 263 controls matched on the date of biospecimen collection ($\text{within}\pm 1\mathrm{wk}$) to estimate the impact of systematic differences in the time of biospecimen collection. The average of biospecimen collection time was 233 d gestation and 238 d gestation for the preterm and term group, respectively. Furthermore, we stratified participants into estradiol and estriol tertiles to estimate the PFAS-preterm birth association using logistic regression models, and interactions with PFAS on preterm birth by including a PFAS*estrogen profiles cross-product term.

Results

The demographic and clinical characteristics of mothers and infants are presented in Table 1. There were 371 preterm and 508 term births. Mothers of preterm cases had less education and were older (mean: 29.17 y old) than mothers of term controls (mean: 28.03 y old). Control mothers were more likely to be primiparous (64.57%) and samples were collected later than for cases. Preterm infants were more likely to be assigned male sex than female, but we did not observe any difference in the mean level of cord serum ln-estrogen between females and males (Table S3). Additionally, the median cord estriol concentration was significantly lower in preterm than in term births. Participants’ characteristics were similar to the 15,653 participants with questionnaire data and 5,202 participants who also had available blood samples in the Maoming birth cohort study, with the exceptions of household annual income and environmental tobacco smoke exposure (Table S4).

Table 1.

Sociodemographic, clinical, and PFAS exposure characteristics among participants in the Maoming birth cohort study ($n=879$).

Characteristic	Preterm ($n=371$)	Control ($n=508$)	$p$-Value
Maternal
Age (y)	$29.17\pm 7.85$	$28.03\pm 6.05$	0.008
BMI (${\mathrm{kg}/\mathrm{m}}^2$)	$20.98\pm 3.58$	$20.79\pm 3.16$	0.574
Education	—	—	$<0.001$
$<\text{High school}$	187 (50.40%)	156 (30.71%)	—
High school	84 (22.64%)	140 (27.65%)	—
$>\text{High school}$	100 (26.95%)	212 (41.73%)	—
Infant sex
Male	221 (59.57%)	262 (51.57%)	0.015
Female	150 (40.43%)	246 (48.43%)	—
Parity
First	168 (45.28%)	327 (64.57%)	$<0.001$
Second	168 (45.28%)	158 (31.1%)	—
$>\text{Second}$	35 (9.44%)	22 (4.33%)	—
Annual family income (Yuan)
$<\mathrm{30,000}$	88 (23.72%)	121 (23.82%)	0.749
30,000–100,000	243 (65.50%)	324 (63.78%)	—
$\ge \mathrm{100,000}$	40 (10.78%)	63 (12.40%)	—
ETS
No	205 (55.26%)	273 (53.74%)	0.656
Yes	166 (44.74%)	235 (46.26%)	—
Drink
No	359 (96.77%)	495 (97.44%)	0.552
Yes	12 (3.23%)	13 (2.56%)	—
Biospecimen collection time (days gestation)	$228.65\pm 29.58$	$254.58\pm 36.79$	$<0.001$
PFAS ($\text{ng}/\mathrm{mL}$)
PFBA	1.259 (0.341, 2.253)	0.679 (0.057, 1.428)	$<0.001$
PFHxS	0.151 (0.087, 0.273)	0.157 (0.104, 0.237)	0.481
br-PFHxS	0.001 (0.001, 0.011)	0.002 (0.001, 0.005)	0.002
PFOA	1.271 (0.826, 2.046)	0.978 (0.724, 1.367)	$<0.001$
PFOS	4.307 (2.622, 7.262)	2.753 (1.541, 4.339)	$<0.001$
br-PFOS	0.967 (0.603, 1.497)	0.815 (0.481, 1.189)	$<0.001$
PFNA	0.434 (0.308, 0.629)	0.415 (0.314, 0.552)	0.133
Cord estrogen ($\text{ng}/\mathrm{mL}$)
Estradiol	2.39 (1.59, 3.28)	2.42 (1.85, 3.32)	0.061
Estriol	113.04 (71.84, 145.35)	135.60 (107.29, 185.87)	$<0.001$

Note: Difference test using $t$-test, chi-square, or Wilcoxon rank-sum test; $n=5$ maternal prepregnancy weight values were missing and imputed with stochastic regression methods to calculate BMI. Values are presented as mean plus or minus SD, number (%) or median (quantile 1, quantile 3). —, no data; BMI, body mass index; br-PFHxS, branched perfluorohexanesulfonic acid; br-PFOS, branched perfluorooctanesulfonic acid; ETS, environment tobacco smoke exposure; PFAS, poly- and perfluoroalkyl substances; PFBA, perfluorobutanoic acid; PFHxS, perfluorohexanesulfonic acid; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctanesulfonic acid.

Concentrations of PFBA, PFOA, PFOS, and br-PFOS were significantly higher in preterm than in full-term births. However, br-PFHxS was higher in controls, and the concentrations were much lower than those of the other PFAS (Table 1). All PFAS were measured above the LODs, with the exceptions of 54.0% of br-PFHxS and 80.54% of PFBA (Table 2).

Table 2.

Distribution of maternal serum perfluoroalkyl substance concentrations ($\text{ng}/\mathrm{mL}$) among participants from the Maoming birth cohort study ($n=879$).

PFAS	LOD ($\text{ng}/\text{mL}$)	$n$ (%) $>\text{\hspace{0.17em}LOD}$	Minimum	25th percentile	Median	75th percentile	Maximum
PFBA	0.0081	708 (80.55%)	0.006	0.164	0.855	1.839	7.168
PFHxS	0.0011	877 (99.77%)	0.001	0.098	0.154	0.25	6.577
br-PFHxS	0.0011	475 (54.04%)	0.001	0.001	0.002	0.005	1.099
PFOA	0.0025	879 (100.0%)	0.002	0.752	1.094	1.61	18.578
PFOS	0.00025	879 (100.0%)	0.001	1.96	3.224	5.467	258.204
br-PFOS	0.00027	879 (100.0%)	0.029	0.509	0.867	1.354	52.458
PFNA	0.0024	879 (100.0%)	0.09	0.309	0.42	0.589	2.679

Note: br-PFHxS, branched perfluorohexanesulfonic acid; br-PFOS, branched perfluorooctanesulfonic acid; LOD, limit of detection; PFAS, poly- and perfluoroalkyl substances; PFBA, perfluorobutanoic acid; PFHxS, perfluorohexanesulfonic acid; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctanesulfonic acid.

Association between Maternal PFAS Exposure and Preterm Birth

On average, each 1-ln-unit PFAS increase was associated with a greater risk of preterm birth, especially for PFBA [odds ratio (OR) = 1.20, 95% CI: 1.14, 1.26], PFNA (OR = 1.30, 95% CI: 1.21, 1.39), PFOA (OR = 1.98, 95% CI: 1.54, 2.55), and PFOS (OR = 1.91, 95% CI: 1.76, 2.07) and its branched isomer (OR = 1.91, 95% CI: 1.90, 1.92), adjusted for covariates (Table 3).

Table 3.

Logistic regression models for associations between preterm birth and maternal serum PFAS among mother-infant pairs from the Maoming birth cohort study ($n=879$).

ln-PFAS	Crude OR (95% CI)	Adjusted ORa (95% CI)
PFBA	1.18 (1.11, 1.26)	1.20 (1.14, 1.26)
PFHxS	0.87 (0.76, 1.00)	0.96 (0.77, 1.20)
br-PFHxS	0.99 (0.89, 1.09)	0.91 (0.79, 1.06)
PFOA	1.44 (1.22, 1.68)	1.98 (1.54, 2.55)
PFOS	1.66 (1.42, 1.95)	1.91 (1.76, 2.07)
br-PFOS	1.46 (1.21, 1.76)	1.91 (1.90, 1.92)
PFNA	1.26 (0.96, 1.64)	1.30 (1.21, 1.39)

Note: br-PFHxS, branched perfluorohexanesulfonic acid; br-PFOS, branched perfluorooctanesulfonic acid; CI, confidence interval; ln-, per natural logarithm-transformed-unit increase; OR, odds ratio; PFAS, poly- and perfluoroalkyl substances; PFBA, perfluorobutanoic acid; PFHxS, perfluorohexanesulfonic acid; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctanesulfonic acid.

^a Model adjusted for age, maternal BMI, maternal education, infant sex, parity, environmental tobacco smoke, alcohol intake, household income, date of delivery, and biospecimen collection time during pregnancy.

In a stratified analysis, the ORs for different preterm birth subtypes were similar to the main analysis (Table S5). For example, a 1-ln-unit increase in maternal PFOS was associated with a 1.78 times (95% CI: 1.70, 1.87) greater risk of spontaneous preterm birth and 2.04 times (95% CI: 1.86, 2.44) greater risk for medically indicated preterm birth, 1.50 times (95% CI: 1.26, 1.77) greater risk for very early preterm birth, and 1.97 times (95% CI: 1.88, 2.07) greater risk for early/late preterm birth. The ORs for preterm birth were higher in females than males when stratified by infant sex (Table S6).

Causal Mediation Effects of Estradiol and Estriol

We found associations with weak strength between maternal PFAS and cord estradiol level (e.g., $\mathrm{\beta }$ for PFOS: $-0.04$, 95% CI: $-0.08$, $-0.01$) (Table S7), and cord estradiol was positively associated with preterm birth (OR = 1.07, 95% CI: 1.05, 1.08) (Table S8). While no statistically significant mediation effects were observed for cord estradiol (Table S9), the interaction effects for PFAS with cord estradiol accounted for 16.61% (95% CI: 11.28%, 21.94%) of the overall association for PFOS and preterm birth, and 4.35% (95% CI: $-24.52\%$, 33.21%) for the overall association between br-PFOS and preterm birth in our causal mediation analysis (Table 4). We did not observe interaction or mediation effects for estriol in cord serum (Table 4; Table S9).

Table 4.

Proportions attributable to interaction of the associations between preterm birth and maternal serum PFAS levels mediated and interacted by cord estrogen levels in the Maoming birth cohort study ($n=879$).

ln-PFAS	Estradiol [% (95% CI)]	Estriol [% (95% CI)]
PFBA	$-9.66\%$ ($-20.20\%$, 0.89%)	0.03% ($-1.07\%$, 1.14%)
PFHxS	3.61% ($-13.38\%$, 20.60%)	1.04% ($-3.49\%$, 5.57%)
br-PFHxS	$-4.31\%$ ($-12.47\%$, 3.86%)	$-6.49\%$ ($-51.00\%$, 38.02%)
PFOA	14.08% (3.85%, 24.31%)	0.84% ($-2.87\%$, 4.56%)
PFOS	16.61% (11.28%, 21.94%)	$-1.05\%$ ($-5.66\%$, 3.55%)
br-PFOS	4.35% ($-24.52\%$, 33.21%)	3.04% ($-9.62\%$, 15.69%)
PFNA	8.33% (0.52%, 16.13%)	5.26% ($-29.70\%$, 40.22%)

Note: Causal mediation models include adjustment for age, maternal BMI, maternal education, parity, environmental tobacco smoke, alcohol intake, household income, date of delivery, and biospecimen collection time during pregnancy. BMI, body mass index; br-PFHxS, branched perfluorohexanesulfonic acid; br-PFOS, branched perfluorooctanesulfonic acid; CI, confidence interval; ln-, per natural logarithm-transformed-unit increase; PFAS, poly- and perfluoroalkyl substances; PFBA, perfluorobutanoic acid; PFHxS, perfluorohexanesulfonic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctanesulfonic acid; PFNA, perfluorononanoic acid.

Modification Effects of Estradiol and Estriol

In an exploratory analysis, we considered cord estradiol and estriol levels together to explore the interaction effects of cord estrogen on the association between PFAS and preterm birth. We stratified cord estrogen levels into three latent profiles according to the z-standardized means of estradiol and estriol (Figure S4). The “least estrogen” profile ($n=736$, 83.7%) was chosen as the reference group and was defined by low z-standardized mean levels for both estrogens. Profile 2 ($n=120$, 13.7%) was defined by the highest z-standardized mean level of cord estriol, which was labeled as “high estriol.” Profile 3 ($n=23$, 2.6%) was defined by the highest z-standardized mean level of cord estradiol, which was labeled as “high estradiol.” The numerical values of z-standardized means are provided in Table S10. The sociodemographic, clinical, and PFAS exposure characteristics of participants are stratified by the three latent estrogen classes in Table S11. Compared to the “least estrogen” profile, greater serum PFHxS, PFOS, and br-PFOS were associated with increased risks of preterm birth among participants in the “high estradiol” and “high estriol” profiles (Table S12). Specifically, participants with high estradiol levels had stronger associations between preterm birth and serum PFOS ($\text{OR}=3.29$, 95% CI: 1.31, 8.25) and its branched isomer ($\text{OR}=6.71$, 95% CI: 1.06, 11.91) than participants without high estradiol levels (Figure 1).

Figure 1.

Logistic regression models for associations between preterm birth and serum PFOS (A) and br-PFOS (B) stratified by cord serum estrogen profile using latent profile analysis; least estrogen ($n=736$), high estriol ($n=120$), and high estradiol ($n=23$). All models were adjusted for age, maternal BMI, maternal education, infant sex, parity, environmental tobacco smoke, alcohol intake, household income, date of delivery, and biospecimen collection time during pregnancy. Additional details about the association from the analysis are in Table S12. Blue boxes indicate odds ratios; black bars indicate 95% confidence intervals. P for interaction test of estrogen profile by PFAS cross-product term. Note: BMI, body mass index; br, branch-; PFOS, perfluorooctanesulfonic acid.

Sensitivity Analysis

To assess the robustness of the results, we repeated the analyses after matching a subsect of controls to cases on the gestational age of biospecimen collection, and the results were consistent with the main analysis (Table S13). We also repeated the analyses stratified by tertiles of cord estradiol and estriol. We observed significant interactions between PFOS, br-PFOS, and estradiol (Table S14) and between PFOS and estriol (Table S15) with greater odds for preterm birth associated serum PFAS primarily among the low (first tertile) and high (third tertile) cord estrogen tertiles.

Discussion

Growing concern regarding the potential adverse reproductive health consequences of exposure to PFAS, specifically on birth outcomes, has fueled numerous studies that have yielded incongruent results. In the present investigation, we observed that higher maternal exposure to PFAS was associated with a greater risk of preterm birth. These associations were not mediated by cord estrogen levels. However, we detected significant interactions between the association of estrogen and maternal serum PFAS on preterm birth. Specifically, participants with high cord estradiol and estriol levels had higher odds of preterm birth with greater exposure to PFAS, especially PFOS and its branched isomer.

Previous studies have reported associations between preterm birth and maternal PFAS exposure, but the results were inconsistent. Compared to previously reported positive PFAS-preterm birth associations, the median PFAS concentrations in our study (e.g., ${\text{PFOS}}_{\text{median}}$: $3.22\text{\hspace{0.17em}ng}/\mathrm{mL}$) (Table 2) were lower than studies conducted from 1996 to 2002 in Denmark ($30.1\text{\hspace{0.17em}ng}/\text{mL}$),⁷ 1999 to 2002 in the United States ($25.7\text{\hspace{0.17em}ng}/\text{mL}$),³⁵ and 2013 in China ($7.15\text{\hspace{0.17em}ng}/\text{mL}$).²¹ Our PFOS concentration were also lower than those from studies conducted from 2010 to 2011 and 2013 to 2015 in Greenland ($8.99\text{\hspace{0.17em}ng}/\mathrm{mL}$)³⁶ and 2003 to 2008 in Spain ($6.05\text{\hspace{0.17em}ng}/\text{mL}$),³⁷ which reported no associations of maternal PFAS exposure with preterm birth. However, the studies that reported a null PFAS-preterm birth association measured maternal PFAS in the first or second trimesters, while we measured maternal PFAS in the third trimester. Maternal PFAS concentrations attenuate with greater gestational age,³⁸ which is also confirmed in our study, so the conflicting results may be influenced by biospecimen collection time. Another potential explanation for the discrepant results is that some sources of PFAS exposure, such as seafood and meat consumption,³⁹ may also promote maternal reproductive health via high polyunsaturated fatty acid content.⁴⁰ Since the consumption of red meat and marine polyunsaturated fatty acids is lower in Chinese than European populations,⁴¹ that may alter the association of PFAS with preterm birth. A future investigation with more detailed dietary information will be necessary to more clearly estimate the impact. Also, the inconsistent study results might be related in part to different subtypes of preterm birth as study outcomes, heterogeneity of the study populations, inconsistent adjustment for potential confounders, and inattention to important effect modifiers, such as estrogens.⁴²

As with other endocrine disrupting chemicals, previous studies have reported associations between PFAS exposure and cord blood estrogens levels.^43,44 Furthermore, several studies suggested that PFAS exposure may disrupt estrogenic activity by stimulating estrogen synthesis⁴⁵ and disturbing estrogen metabolism,⁴⁶ which may in turn impact birth outcomes.¹⁴ For example, Wang et al. reported that cord estriol levels mediated 30.43% (95% CI: 3.31%, 66.77%) of the association between cord PFOS levels and birth weight in 424 pregnant Chinese women.¹⁵ We, therefore, hypothesized that estrogen would mediate the association between PFAS exposure and preterm birth. Although the estrogen levels of our study were similar to previous studies,^15,43,44 the results of our causal mediation analysis did not support estrogens as causal mediators of the PFAS-preterm birth associations.

We observed significant interactions between PFAS and estradiol on preterm birth. To our knowledge, there are no directly comparable studies that investigated the interaction effects of cord estrogens on PFAS and preterm birth. Several potential mechanisms may help to explain our interaction findings. Greater estrogen levels may stimulate expression of the estrogen receptor (ER),⁴⁷ potentiating susceptibility to PFAS toxicity⁴⁸ in the absence of a mediating effect as we observed. Experimental studies reported that increased ER expression may enhance PFOS toxicity,^49,50 which may also drive our results, although we were unable to test this hypothesis directly. Alternately, Sonthithai et al. found that co-exposure to PFOS or PFOA and estradiol enhanced the effects of estradiol on an estrogen-responsive pathway,¹⁷ which may in turn disturb the parturition process and lead to preterm birth.¹⁴ Furthermore, other unobserved factors, such as meat intake,^51,52 maternal psychosocial stress,⁵³ or co-exposure to other endocrine disrupting chemicals⁵⁴ that influence birth outcomes could affect PFAS and cord estrogen levels and bias the causal mediation analysis results. Unfortunately, we did not collect this information, and, therefore, our results cannot rule out a chance effect and need to be interpreted cautiously.

Our results also suggested stronger associations between maternal levels of some PFAS and preterm birth in female infants, similar to a previous cohort study that also found a stronger negative association between maternal PFAS exposure and gestational age at delivery in female infants.²¹ A sex-difference may result from higher cord serum estrogen levels in female than in male fetuses,³² enhancing susceptibility to gestational PFAS toxicity. However, our results did not support this hypothesis, we did not observe a difference in levels of cord serum estrogens between females and males. Our results provide evidence that females may be more susceptible to PFAS toxicity than males, which may result from the fact that ER receptors are more abundantly expressed in female infants than in males.⁵⁵ Additional research with a larger sample size is needed to clarify the sex-specific effects between PFAS exposure and preterm birth.

Considering the complex biological roles of different estrogen types, we simultaneously considered the most bioactive and abundant estrogen, estradiol, and the predominant estrogen of pregnancy, estriol, using an LPA method.¹² Participants exhibiting both high estradiol and estriol levels had greater odds of preterm birth in association with PFAS exposure. Importantly, the odds of preterm birth in association with PFAS were higher among participants in the high estrogen LPA profiles (Table S12) than the odds when stratified by tertiles of each estrogen type individually (Table S14 and S15). We also found significant interaction effects between PFOS and estrogen on preterm birth, after considering estradiol and estriol together as estrogen profiles using LPA. These results incorporated the complex interplay of the estrogenic environment of pregnancy, rather than as isolated contributions from each estrogen type.

This study has several strengths. To our knowledge, we are among the first to examine interactions between maternal PFAS exposure and estrogen on preterm birth. The large sample size and representative population allowed for investigation into effect modification by estrogens. Additionally, we identified estrogen profiles by considering two estrogen types (estradiol and estriol) together, to more comprehensively capture effect modification by estrogens than possible using single estrogen types in isolation. Furthermore, our findings are consistent with recently reported human reproductive toxicity information for br-PFOS isomers and short chain PFHxS.⁵⁶

Despite these strengths, this study also has limitations. First, maternal PFAS concentrations are influenced by gestational age.³⁸ We measured maternal PFAS concentrations once in the third trimester of pregnancy, which may introduce exposure misclassification leading to imprecise effect estimates. Additionally, this may introduce immortal time bias, especially for very early preterm birth, in which more susceptible fetuses delivered earlier were less likely to have maternal blood collection than less-susceptible fetuses delivered later.⁵⁷ Second, given our nested case-control design, it is important to acknowledge the potential for selection bias. Although we used inverse probability of sampling weights to mitigate the potential impact of a selection bias, differences in income and ETS between our cohort and the baseline population may have influenced our findings. Third, we considered estradiol and estriol as potential hormonal mediators of the PFAS-preterm birth associations, while other reproductive hormones associated with preterm birth, such as progesterone or testosterone, may also be important.^44,58 Fourth, other factors that may confound the interaction between PFAS and estrogen should also be considered, including meat intake, maternal psychosocial stress, and co-exposure to other endocrine disrupting chemicals. Finally, the analyses may have been underpowered for high estrogen groups because of limited numbers, leading to imprecise estimates of the interactions, and the finding should thus be interpreted as preliminary.

In summary, our study provides new evidence that significant interactions exist between estrogen and exposure to PFAS on preterm birth. Specifically, participants with high estradiol and estriol levels had higher odds of preterm birth after exposure to PFHxS and PFOS and its branched isomer. A larger investigation in pregnant women with estrogen measurement will be required for more definitive results in the future.