Prenatal per- and polyfluoroalkyl substances (PFAS) exposure in relation to preterm birth subtypes and size-for-gestational age in the LIFECODES cohort 2006–2008

Abstract

Background:

Per- and polyfluoroalkyl substances (PFAS) are a group of synthetic chemicals widely used in consumer and industrial products. Numerous studies have linked prenatal PFAS exposures to increased risks of adverse pregnancy outcomes such as preterm birth (PTB) and small-for-gestational age (SGA).However, limited evidence is available for the effects of PFAS on PTB subtypes and large-for-gestational age (LGA).

Objective:

To examine the associations of PFAS with PTB [overall, placental (pPTB), spontaneous (sPTB)], BW Z-score, and size-for-gestational age (SGA, LGA).

Methods:

Our nested case-control study included 128 preterm cases and 373 term controls from the LIFECODES cohort between 2006 and 2008 (n = 501). Plasma concentrations of nine PFAS were measured in early pregnancy samples. Logistic regression was used to assess individual PFAS-birth outcome associations, while Bayesian Kernel Machine Regression (BKMR) was used to evaluate the joint effects of all PFAS. Effect modification by fetal sex was examined, and stratified analyses were conducted to obtain fetal sex-specific estimates.

Results:

Compared to term births, the odds of pPTB were higher from an interquartile range increase in perfluorodecanoic acid (PFDA) (OR = 1.60, 95% CI: 1.00–2.56), perfluorononanoic acid (PFNA) (OR = 1.67, 95% CI: 1.06–2.61), and perfluoroundecanoic acid (PFUA) (OR = 1.77, 95% CI: 1.00–3.12), with stronger associations observed in women who delivered males. BKMR analysis identified PFNA as the most important PFAS responsible for pPTB (conditional PIP = 0.78), with increasing ORs at higher percentiles of PFAS mixture. For LGA, positive associations were observed with PFDA and perfluorooctanoic acid in females only, and with PFUA in males only. BKMR analysis showed increasing, but null effects of PFAS mixture on LGA.

Conclusions:

The effect of prenatal exposure to single and multiple PFAS on PTB and LGA depended on fetal sex. Future studies should strongly consider examining PTB subtypes and sex-specific effects of PFAS on pregnancy outcomes.

1. Introduction

Per- and polyfluoroalkyl substances (PFAS) are a group of synthetic chemicals widely used in various consumer and industrial products due to their unique properties, such as resistance to grease, oil, water, and heat (Buck et al., 2011). Long-chain PFAS (≥6 carbons) tend to persist in the environment, bioaccumulate in wildlife and humans, and have prolonged half-lives (Dewitt, 2015). Despite the phase-out of certain legacy PFAS, including perfluorooctanoic acid (PFOA) and perfluorooctane sulfonic acid (PFOS) in the United States (US), their production persists in many parts of the world, and numerous PFAS that ultimately degrade into PFOA and PFOS are still manufactured and utilized globally (Wang et al., 2017). The main sources of PFAS exposure in the general population include drinking water, food, and household dust (D’Hollander et al., 2010). Some legacy PFAS have been detected in the blood of more than 90 percent of Americans (Kato et al., 2011), as well as in the cord blood and placenta of newborns (Hall et al., 2022; Apelberg et al., 2007), highlighting concerns about their potential toxic effects.

Animal and epidemiological studies have consistently linked PFAS with multiple health outcomes, including immunotoxicity, dyslipidemia, kidney and testicular cancer, metabolic syndrome, and notably adverse birth outcomes such as preterm birth (PTB) and reduced birthweight (BW) (Gui et al., 2022; Sunderland et al., 2019; Rappazzo et al., 2017). PTB, defined as delivery before 37 weeks of gestation, is a leading cause of global childhood mortality, and is responsible for approximately 28 percent of all neonatal deaths (Simmons et al., 2010; Lawn et al., 2005). Infants born preterm or with low BW are more likely to experience childhood morbidities and long-term neurodevelopmental disorders (Casey, 2008; Robert L Goldenberg and Culhane, 2007), while high BW infants have an increased risk of developing metabolic syndrome (Derraik et al., 2020). The exact causes of PTB and altered fetal growth are unclear, although several risk factors, including maternal sociodemographic characteristics, health status, pregnancy history, and genetic predisposition may play a role (Robert L. Goldenberg et al., 2008). The potential impact of environmental toxicants such as PFAS on PTB and altered fetal growth is a growing concern and a top public health priority that requires further investigation (Ferguson and Chin, 2017; Behrman and Butler, 2007).

In the past decade, several studies have explored the associations between prenatal PFAS levels and PTB and altered fetal growth, with mixed results. For instance, some studies have shown associations between certain PFAS and increased risk of PTB (Padula et al., 2023; Yang et al., 2022; Liao et al., 2022; Gardener et al., 2021; Chu et al., 2020; Meng et al., 2018; Sagiv et al., 2017; Chen et al., 2012). However, others have shown null (Eick et al. 2020, 2023; Liu et al., 2020; Huo et al., 2020; Manzano-Salgado et al., 2017; Darrow et al., 2013; Hamm et al., 2010), or in some cases, negative associations (Liao et al., 2022; Whitworth et al., 2012). The studies on PFAS and BW have reported mostly negative associations (Padula et al., 2023; Wang et al., 2023; Gardener et al., 2021; Hjermitslev et al., 2020; Chu et al., 2020; Marks et al., 2019; Sagiv et al., 2017; Bach et al., 2016; Maisonet et al., 2012; Chen et al., 2012; Wikstrom et al., 2020; Steenland et al., 2018; Washino et al., 2009), or null relationships (Eick et al. 2020, 2023; Hu et al., 2021; Huo et al., 2020; Chu et al., 2020; Manzano-Salgado et al., 2017; Hamm et al., 2010). These studies, however, have varied greatly in study design, sample size, cohort demographics, and exposure levels. For instance, among five studies that have explored associations between prenatal PFAS exposures and PTB in the general population in North America, the majority of them had less than 50 PTB cases and most focused on PFOA, PFOS, perfluorononanoic acid (PFNA), and perfluorohexanesulfonic acid (PFHxS). In addition, just a few of them have investigated the joint or combined effects of PFAS mixture, which more accurately reflects the exposure scenarios experienced by the general population (Rogers et al., 2021; Kwiatkowski et al., 2020). Evidence is also limited on the associations between PFAS and extreme measures of fetal growth such as small-for-gestational age (SGA) and large-for-gestational-age (LGA) that are especially detrimental to the long-term health of the newborns (Bommarito et al., 2021a).

Our understanding of the effects of prenatal PFAS exposure on PTB is also further limited in that most previous studies have primarily focused on overall PTB. This is a limitation given that PTB is a multifactorial disease with heterogeneous risk factors, clinical presentations, and biological mechanisms and studying it as a single entity may mask the specific effects of PFAS on different PTB subtypes. Although there are no official guidelines for defining PTB subtypes, they can be broadly categorized into one of the two types based their clinical presentations: spontaneous PTB (sPTB) and placental PTB (pPTB) (McElrath et al., 2008).

To address some of the aforementioned knowledge gaps, the current study aimed to expand on previous findings and examine the associations between prenatal exposure to eight PFAS and birth outcomes using data from a racially diverse cohort of pregnant women enrolled in the LIFECODES study. The birth outcomes assessed in this study include PTB and the subtypes (sPTB and pPTB), BW Z-score, SGA, and LGA. We also aimed to examine the joint effects of multiple PFAS on the aforementioned outcomes, assessing any nonlinear dose-response associations or interactions. We hypothesized that elevated prenatal PFAS levels will be associated with an increased risk of adverse birth outcomes, specifically PTB and its subtypes, reduced BW, and altered fetal growth (SGA or LGA).

2. Materials and methods

2.1. Study population

Participants in the current study were a subset of the larger LIFECODES prospective birth cohort recruited between 2006 and 2008 (n = 1,648) at Brigham and Women’s Hospital (BWH) in Boston, Massachusetts. The detailed descriptions of LIFECODES recruitment methods and study design have been documented elsewhere (Cantonwine et al., 2015; Ferguson et al., 2014). In brief, women were eligible for the study if they received prenatal care before 15 weeks of pregnancy, were over 18 years old, and planned to deliver at BWH. The only exclusion criterion was women with higher-order multiple gestations (triplets or greater). Information on sociodemographic status and personal medical history were collected via detailed questionnaires at the first study visit (median = 9.9 week gestation). Plasma samples were collected at the initial visit and again at three subsequent visits at median 18, 26, and 35 weeks gestation. Gestational age (GA) was estimated using the last menstrual period along with ultrasound measurements, in accordance with the guidelines established by the American College of Obstetrics and Gynecologists (ACOG, 2017).

This study combined 470 participants from a nested case-control study (Cantonwine et al., 2015) of singleton preterm births (n = 127 PTB cases; n = 343 non-PTB controls) with 31 participants (n = 1 PTB case; n = 30 non-PTB controls) from a later pilot study within LIFECODES that also measured PFAS concentrations. The original nested case-control study included 130 women who delivered preterm and 352 randomly selected women who delivered at term (after 37 weeks gestation). Twelve participants from the original study did not have enough plasma samples for PFAS analysis. The final analytical sample of the current study included 128 women who delivered preterm and 373 who delivered at term. All relevant protocols related to this research were approved by Institutional Review Boards at both BWH and University of Michigan. All study participants provided written informed consent prior to participation.

2.2. Outcomes classification

As stated earlier, overall PTB was defined as a live delivery before 37 weeks gestation, and further classification into sPTB (n = 72) and pPTB (n = 39) were based on their clinical presentations (McElrath et al., 2008; Aung et al., 2019). sPTB was characterized by preterm premature rupture of membranes, spontaneous preterm labor, placental abruption, and cervical insufficiency, while pPTB was characterized by preeclampsia or intrauterine growth restriction (IUGR). In addition to sPTB and pPTB, n = 17 newborns were electively delivered preterm due to obstetric protocol, including the management of prior cesarean section or other pregnancy complications. These cases were included in the overall PTB analysis but were not analyzed separately as a subtype.

Newborn BWs were obtained from medical records post-delivery, and BW Z-scores were calculated using a previously established internal growth standard (Cantonwine et al., 2016). SGA was defined as having a BW Z-score below the 10th percentile, while LGA was defined as having a BW Z-score above the 90th percentile.

2.3. Exposure measurements

Maternal plasma samples used to measure PFAS were collected at the first or second study visit. The majority of samples were collected during the first visit, with n = 45 (8.9 percent) samples being collected at the second (15.6–21.6 weeks gestation) when a sample from the first visit was not available. The samples were stored at −80 °C until they were analyzed by NSF International using the Centers for Disease Control and Prevention’s Polyfluoroalkyl Chemicals laboratory method No. 6304.1. We measured concentrations of nine PFAS congeners, including N-methylperfluoro-1-octanesulfonamidoacetic acid (MPAH), perfluorodecanoic acid (PFDA), perfluoroheptanoic acid (PFHP), PFHxS, PFNA, PFOA, PFOS, perfluorooctane sulfonamide (PFOSA), and perfluoroundecanoic acid (PFUA). The detailed description of the laboratory analytical method for PFAS analysis is available elsewhere (Bommarito et al., 2021b). Briefly, the frozen plasma samples were thawed to room temperature; and pre-concentrated through on-line solid phase extraction (SPE) interfaced with a Thermo Scientific Transend TXII system utilizing a Cyclone-P extraction column. The analytes were then separated and focused using a Dionex UltiMate 3000 ultra-high-performance liquid chromatography (UHPLC) system with reversed-phase chromatography using a Waters XBridge C18 analytical column. Lastly, the analytes were detected using a Thermo Scientific Transcend TXII Turbulent Flow system interfaced with a Thermo Scientific Quantiva triple quadrupole mass spectrometer (MS). The NSF method was validated and was found to perform within an acceptable criterion established by the CDC method (validated analyte calibration curve correlation coefficient (R²) > 0.998). The limit of detection (LOD) and the detection rate for each PFAS are presented in Table 2. In this analysis, machine-read values (MRVs) of PFAS concentrations were used for participants whose exposure levels were below the LOD. When MRVs were not available, the estimated levels were imputed by dividing the LOD by the square root of $2(\mathrm{L}\mathrm{O}\mathrm{D}/\sqrt{2})$. All PFAS, except PFOSA and PFHP, were detected in more than 75 percent of the samples. PFOSA was detected below the LOD in more than 99 percent of samples and was excluded from further analysis; PFHP was analyzed as a dichotomous variable (below/above the LOD) since 57 percent of samples were detected below the LOD.

Table 2.

Distributions of plasma PFAS concentrations (ng/mL) among study participants in LIFECODES, and in women of reproductive age (18–40 years) in the NHANES 2007–2008.

PFAS	LIFECODES (n = 501)									NHANES (n = 343)b
	LOD (ng/mL)	Detection ratea (% > LOD)	GM	25th	50th	75th	90th	95th	Max	GM
MPAH	0.1	77.49	0.17	0.11	0.16	0.27	0.44	0.60	6.16	0.24
PFDA	0.1	98.41	0.29	0.22	0.29	0.40	0.53	0.68	1.66	0.28
PFHxS	0.1	99.40	0.90	0.54	0.88	1.50	2.45	3.71	11.90	1.07
PFNA	0.1	99.40	0.85	0.65	0.86	1.13	1.62	1.94	4.00	1.04
PFOA	0.5	99.20	2.34	1.77	2.53	3.28	4.21	5.00	8.47	3.26
PFOS	0.1	100.00	6.86	4.81	7.14	9.96	13.25	16.89	39.50	8.35
PFUA	0.1	88.65	0.22	0.15	0.23	0.38	0.52	0.65	1.27	<LOD
PFHP	0.1	42.83	0.10	0.04	0.07	0.23	0.71	0.84	2.88	<LOD
PFOSA	0.1	1.00							0.18	<LOD

Note: NHANES, National Health and Nutrition Examination Survey; GM, Geometric mean.

^aWhen available, the percentiles of PFAS concentrations were computed using the machine read values.

^bNHANES estimates were calculated for women aged 18–40 years to facilitate a comparison with the current study. We applied the NHANES survey design and weights. The National Report on Human Exposure to Environmental Chemicals from CDC did not report GM for PFUA, PFHP, and PFOSA for the relevant cycles due to their low detection rates. Consequently, we did not calculate GM estimates for these PFAS when making comparisons. Similarly, percentiles for PFOSA were not determined, as its detection rate was a mere 1%.

2.4. Covariates

Information on sociodemographic characteristics, including maternal age, maternal race (Whites/non-Whites), insurance status (public/private), and maternal educational status (high school or less, some college, college graduate or more), and prepregnancy body mass index (BMI) was collected via questionnaires at the initial visit. Participants identifying as African American (n = 87), Asian (n = 37), Hispanic (n = 66), mixed (n = 10), and others (n = 14) were reclassified as non-Whites to simplify the number of variables used in the model. BMI was estimated from self-reported maternal prepregnancy weight and height. Information about smoking status (yes/no) during the pregnancy, as well as personal health information, including parity (nulliparous/multiparous) were obtained at the initial visit. The gestational age at the time of sample collection was recorded, and the biological sex of a newborn was collected post-delivery. Potential confounders were identified a priori based on published literature (Deji et al., 2021; Bach et al., 2015) and all models were adjusted for maternal age, race, insurance status, educational status, prepregnancy BMI, and parity. Additionally, the models for BW Z-scores, SGA, and LGA were adjusted for fetal sex (Bommarito et al., 2021a).

2.5. Statistical analysis

All statistical analysis was conducted using RStudio 2022.07.0 Build 548 (RStudio, 2022). The distributions of PFAS in the study population were compared with the data from the National Health and Nutrition Examination Survey (NHANES) cycles 2007–2008 for women of reproductive age (CDC, 2019). Spearman correlation was used to assess pairwise relationships between different PFAS (Myers and Sirois, 2004). Additionally, the distributions of maternal sociodemographic and birth characteristics in the study population were characterized and tabulated.

2.5.1. Single pollutant models

The associations between each PFAS and binomial outcomes (overall PTB and PTB subtypes, SGA, and LGA) were examined using logistic regression. Meanwhile, linear regression was used to model the associations between individual PFAS and BW Z-scores. Beta estimates and their 95 percent confidence intervals (CIs) were calculated for a change in outcome per interquartile range (IQR) increase in each PFAS concentration, except for PFHP. For binomial outcomes, beta estimates were converted to odds ratios (ORs) for ease of interpretation. For PFHP, beta estimates and ORs were calculated by comparing individuals with exposure levels below and above the LOD. Placental PTBs were excluded i.e., not recoded as controls, from the analysis of spontaneous PTBs, and vice versa. Similarly, the models for SGA excluded all LGA cases, and vice versa. PFAS concentrations were natural log-transformed to minimize the influence of outliers.

The primary single pollutant analysis was performed on a complete case dataset (n = 478 with 122 preterm cases and 356 controls) because the proportion of participants with at least one missing covariate was low (less than 5 percent). BW Z-score was missing from one participant and was excluded from analysis related to BW measurements. To generalize the associations observed in this study to the broader LIFECODES population, regression models for BW Z-scores, SGA, and LGA were adjusted using sampling weights for PTB cases and controls. The original nested case-control study weights were readjusted to account for 31 additional participants. The revised weights were 1.1 for PTB cases and 2.78 for controls.

2.5.2. Multi-pollutant models

The joint effects of PFAS and birth outcomes were examined using Bayesian kernel machine regression (BKMR), a statistical method that can account for flexible non-linear exposure-response relationships and interactions among different exposures (Bobb et al., 2018). Consistent with the single pollutant models, the exposure variables were log-transformed, and the same set of covariates were used. All PFAS except PFHP, which had detection rate below 50 percent, were included in the BKMR models.

The BKMR model for each outcome was run with default priors using 20,000 iterations to ensure convergence. Hierarchical variable selection was implemented to address high correlation among components of the PFAS mixture and identify the most important PFAS associated with a birth outcome (Bobb et al., 2018). Group specification for each PFAS was assigned based on hierarchical clustering that used Spearman correlation coefficients. as the distance metric. This resulted in the following groups: PFOA, PFOS, PFHxS in group 1 (average pairwise ρ = 0.63); PFNA, PFDA, and PFUA in group 2 (average pairwise ρ = 0.59); and MPAH in group 3 (Fig. S1). The posterior inclusion probability (PIP) for each group and conditional-PIPs for individual PFAS within a group were estimated as a measure of relative variable importance. A PIP threshold of 0.5 was used to identify the relative importance of each group or PFAS within a group. The dose-response relationship between each PFAS and outcome was examined using exposure-response plots with all other PFAS fixed at their medians. The pairwise interactions between different PFAS were evaluated using bivariate response plots. The overall effect of the PFAS mixture was assessed by comparing the ORs when all PFAS were set at their medians in relation to the 25th and 75th percentiles. The ORs for the overall effect plots were obtained by first multiplying the probit estimates from BKMR by 1.6 and then exponentiating them. The multiplication factor of 1.6 was used to approximately convert a probit estimate to a logit estimate (Amemiya, 1981).

Given the known endocrine disrupting effects of PFAS that may cause sex-dimorphic health effects (Lau et al., 2007; Andersen et al., 2010), we tested for the interaction by fetal sex. This was done by including a cross-product term between PFAS and fetal sex in full models. In addition, stratified models were also run to obtain fetal sex-specific estimates. All hypothesis testings were conducted at α = 0.05, and p-values below that were considered statistically significant.

2.5.3. Sensitivity analysis

To ensure the robustness of our findings, multiple sensitivity analyses were conducted. First, to assess the impact of additional data, all primary models from single pollutant analysis were rerun after excluding the 31 participants who were not part of the original case-control study. Second, all models were rerun with PFAS concentrations in their natural unit (ng/ml) to ensure the validity of conclusions from the primary analysis with log-transformed exposures, and to make sure that the observed findings were not merely due to data transformation (Choi et al., 2022). Third, given that the data for all study participants were not collected at the same time, and that the plasma PFAS concentrations may fluctuate over the course of pregnancy (Savitz, 2014), the effect of gestational age at the time of sample collection was also examined. Fourth, since a small proportion of participants reported smoking during pregnancy, models for the primary analysis were not adjusted for smoking behavior. Instead, the effect of smoking was tested by restricting analysis to non-smokers. Fifth, the analyses for BW measurements were restricted to term births to address potential heterogeneity in etiologies. Sixth, the associations between PFAS and GA at delivery were examined to capture more subtle effects of PFAS on timing of delivery. Seventh, potential biases arising from missing covariates data were assessed by rerunning the primary analysis using 20 multiply imputed datasets. Finally, biases resulting from the use of MRVs were addressed by rerunning the primary analysis with PFAS concentrations below the LOD replaced by $\mathrm{L}\mathrm{O}\mathrm{D}/\sqrt{2}$.

3. Results

3.1. Descriptive statistics

Table 1 (PTB) and Table S1 (BW measurements) present demographic information related to the study participants. The median age of participants was 32.7 years old, the median prepregnancy BMI was 24.4 kg/m², and 57.4 percent were White. Most participants had some college education, private health insurance, and refrained from smoking during pregnancy. About 54.4 percent of participants were nulliparous and 56 percent of all newborns were males.

Table 1.

Baseline demographics of the studied population in LIFECODES stratified by PTB case status.^a

Characteristics	Overall	Controls	Preterm cases
	(n = 501)	(n = 373)	(n = 128)
Maternal age (y)
Median [Min, Max]	32.7 [18.3, 50.2]	32.7 [18.3, 48.7]	32.7 [20.9, 50.2]
Maternal prepregnancy BMI (kg/m2)
Median [Min, Max]	24.3 [16.6, 52.8]	24.3 [16.9, 52.8]	24.5 [16.6, 46.3]
Missing	10 (2.0%)	7 (1.9%)	3 (2.3%)
GA at sampling (wk)
Median [Min, Max]	9.86 [5.14, 21.6]	9.86 [5.4, 20.3]	9.86 [5.1, 21.6]
Missing	2 (0.4%)	2 (0.5%)	0 (0%)
Maternal race/ethnicityb
Whites	288 (57.5%)	217 (58.2%)	71 (55.5%)
Non-whites	213 (42.5%)	156 (41.8%)	57 (44.5%)
Maternal education
High school or less	73 (14.6%)	52 (13.9%)	21 (16.4%)
Some college/technical school	84 (16.8%)	60 (16.1%)	24 (18.8%)
College or greater	333 (66.5%)	251 (67.3%)	82 (64.1%)
Missing	11 (2.2%)	10 (2.7%)	1 (0.8%)
Health insurance
Private insurance/HMO	389 (77.6%)	284 (76.1%)	105 (82.0%)
Self-pay or Medicaid/Mass	98 (19.6%)	77 (20.6%)	21 (16.4%)
Health
Missing	14 (2.8%)	12 (3.2%)	2 (1.6%)
Parity
Multiparous	228 (45.5%)	172 (46.1%)	56 (43.8%)
Nulliparous	273 (54.5%)	201 (53.9%)	72 (56.3%)
Smoking during pregnancy
No	471 (94.0%)	352 (94.4%)	119 (93.0%)
Yes	30 (6.0%)	21 (5.6%)	9 (7.0%)
Fetal sex
Female	220 (43.8%)	162 (43.4%)	58 (45.3%)
Male	281 (56.0%)	211 (56.6%)	70 (54.7%)

^aCovariate distributions were estimated for all participants in the study (n = 501). 23 participants were missing information for at least one covariate.

^bParticipants identifying as non-White included African American (17.3%) and a combined group of Asian, Hispanic, and others (25.3%).

Table 2 presents detection rates and percentile estimates of PFAS in the study participants. Most PFAS were detected above their LOD in over 75 percent of participants. The levels of PFAS in the study participants were comparable to or slightly lower than those observed in women aged 18–40 in the NHANES 2007–2008 cycle. As shown in Fig. S2, several PFAS congeners were highly correlated with each other, with the highest correlation observed between PFDA and PFNA (Spearman ρ = 0.74). Table S2 shows quartiles of PFAS concentrations categorized by levels of each categorical outcome and covariate. In general, levels of some PFAS tended to differ across different races, maternal education levels, health insurance status, and parity. For instance, nulliparous women tended to have higher median concentrations of PFHxS, PFOA, PFOS, and PFUA as compared to multiparous women.

3.2. Single pollutant models

Logistic and linear regression models were used to examine the associations between each PFAS and birth outcomes (Fig. 1; Table S3a). Elevated levels of some PFAS were associated with increased odds of pPTB. Specifically, an IQR increase in PFNA (OR: 1.67; 95% CI: 1.06–2.61), PFDA (OR: 1.60; 95% CI: 1.00–2.56) and PFUA (OR: 1.77; 95% CI: 1.00–3.12) were associated with increased OR for pPTB. The associations between other PFAS and pPTB were null. The associations between most PFAS and sPTB were also null, with only PFHP (OR: 1.77; 95% CI: 1.00–3.12) showing a positive association (PFHP detection rate among participants with sPTB (n = 69) = 29%). No significant association was observed between any PFAS and overall PTB, although most OR point estimates were above 1.0. The associations between PFAS and BW Z-score and SGA were all null, while a weak positive association was found between PFDA and LGA (OR: 1.26; 95% CI: 0.98–1.62) (Fig. 2; Table S4a).

Fig. 1.

Forest plots of ORs from overall and fetal sex stratified models. All models were adjusted for maternal age, race, prepregnancy BMI, education level, insurance status, and parity. All ORs, except for those related to PFHP, were calculated for IQR increase in PFAS concentrations. PFHP concentrations were dichotomized based on its LOD. Hence, the OR estimates were obtained by comparing exposure levels below and above the LOD. Associations with significant effect modification by fetal sex are represented by filled cyan markers (p-value for the interaction term PFAS*fetal sex <0.05). Fetal sex was entered as a covariate while testing for interaction effects.

Note: PTB, preterm birth; ORs, odds ratios; IQR, interquartile range; LOD, limit of detection.

Fig. 2.

Forest plots of ORs and betas from overall and fetal sex stratified models for BW measurements. All models were adjusted for maternal age, race, prepregnancy BMI, education level, insurance status, parity, and fetal sex (for overall model only). All models for BW measurements were weighted using inverse probability weights to account for uneven selection of preterm cases in the case-control study. All ORs, except for those related to PFHP, were calculated for IQR increase in PFAS concentrations. PFHP concentrations were dichotomized based on its LOD. Hence, the OR estimates were obtained by comparing exposure levels below and above the LOD. Associations with significant effect modification by fetal sex are represented by filled cyan markers (p-value for the interaction term PFAS*fetal sex <0.05). Fetal sex was entered as a covariate while testing for interaction effects.

Note: BW Z-score, birthweight for gestational age Z-score; SGA, small-for-gestational-age; LGA, large-for-gestational-age; ORs, odds ratios; IQR, interquartile range; LOD, limit of detection.

3.3. Stratification by fetal sex

Sex-stratified analysis showed that the associations between certain PFAS and PTB outcomes varied by fetal sex (Fig. 1; Table S3a). Specifically, the interaction by fetal sex was found to be statistically significant for the associations of MPAH and PFOA with overall PTB (Table S3a). There was a positive association between MPAH and overall PTB (OR: 1.45; 95% CI: 0.98–2.14) in women who delivered female offspring, while the association was negative in women who delivered male offspring (Table S3b–c). Conversely, the association between PFOA and overall PTB was positive in women who delivered male offspring and negative in women who delivered female offspring.

Additionally, the associations of PFNA and PFOA with pPTB were significantly modified by fetal sex, with positive effects observed in women who delivered male offspring, and negative or null effects in women who delivered female offspring (Tables S3a–c). The ORs for PFNA and PFOA in women who delivered male offspring were 4.01 (95% CI: 1.8–8.97) and 2.69 (95% CI: 1.2–6.01), respectively. On the other hand, the OR for PFOA-pPTB association in women who delivered female offspring was 0.53 (95% CI: 0.29–0.98). Sex-stratified analysis also showed that the association between PFDA and pPTB was positive (OR: 2.53; 95% CI: 1.25–5.13) in women who delivered male offspring.

For the associations between PFAS and BW Z-score or SGA, no statistically significant interaction by fetal sex was observed (Fig. 2; Table S4a). However, a significant effect modification was observed in the associations of PFDA (marginal significance), PFHxS, and PFOA with LGA. All three PFAS were positively associated with LGA in female offspring, with elevated odds ratios for PFDA (OR: 1.73; 95% CI: 1.08–2.77), PFOA (OR: 1.88; 95% CI: 1.17–3.01), and PFHxS (OR: 1.63; 95% CI: 0.98–2.72) in female offspring (Tables S4b–c). Conversely, a reverse trend was observed for the association between PFUA and LGA with an elevated odds ratio in male offspring (OR: 1.62; 95% CI: 1.01–2.59).

3.4. Multi-pollutant models

BKMR analysis was used to investigate potential non-linear dose-response relationships and interactions between different PFAS in relation to pPTB and LGA. All PFAS congeners except PFHP were entered in BKMR models, and all analyses were stratified by fetal sex.

PFNA was ranked as the most important PFAS linked to increased odds of pPTB in all women and women who delivered male offspring. This was determined by its conditional-PIPs, which were the highest among all PFAS (Tables S5a–b). In both instances, the joint PIPs for group 2 (PFNA, PFDA, and PFUA) were also the highest. Exposure response plots showed a positive and linear relationship between PFNA and pPTB in women who delivered male offspring, while the relationships were U-shaped and mostly null in all women or women who delivered female offspring (Fig. 3). Bivariate exposure-response plots showed no evidence of pairwise interactions between PFAS (Figs. S3–S5). The overall effect of joint exposures to all PFAS on pPTB was apparent in all women or women who delivered male offspring (Fig. 4). The ORs increased above 1 as the concentrations of all PFAS rose from median to higher quantiles. In contrast, the overall effect was mostly null in women who delivered female offspring.

Fig. 3.

BKMR exposure-response plots with 95 percent credible intervals for associations between PFNA and pPTB. The response plots were obtained by holding all other PFAS in the mixture at their medians. The response functions represent the log-odds of pPTB at log-PFAS concentrations. The log-odds and their 95 percent credible intervals were calculated as (1.6*BKMR estimate) to convert a probit estimate to a logit estimate. All BKMR models were adjusted for maternal age, race, prepregnancy BMI, education level, insurance status, and parity.

Note: BKMR, Bayesian Kernel Machine Regression; pPTB, placental preterm birth; ORs, odds ratios.

Fig. 4.

BKMR overall effect of PFAS mixture on pPTB stratified by fetal sex. The ORs and their 95 percent credible intervals were calculated by comparing the values of PFAS when all of them were held at different percentiles (0.25, 0.35, 0.45, 0.55, 0.65, 0.75, 0.85) as compared to when all of them were held at their medians. The ORs and their 95 percent credible intervals were calculated as exp(1.6*BKMR estimate) to convert a probit estimate to a logit estimate. The solid black line at OR = 1 represents the null association. All BKMR models were adjusted for maternal age, race, prepregnancy BMI, education level, insurance status, and parity.

Note: BKMR, Bayesian Kernel Machine Regression; pPTB, placental preterm birth; ORs, odds ratios.

With respect to LGA in female offspring, PFOA in group 1 (PFOA, PFOS, and PFHxS), which had the highest joint PIP, was identified as the most important PFAS. The conditional PIPs for all other PFAS were less than 0.5. Bivariate exposure response plots showed mostly linear relationships between PFAS and LGA, with no evidence of pairwise interactions (Figs. S6–S8). The exposure-response relationship for PFHxS was marginally U-shaped in all or male offspring, but positive and mostly linear in female offspring. The overall effect of joint exposures to all PFAS on LGA was mostly null in all cases (Fig. 5).

Fig. 5.

BKMR overall effect of PFAS mixture on LGA stratified by fetal sex. The ORs and their credible intervals were calculated by comparing the values of PFAS when all of them were held at different percentiles (0.25, 0.35, 0.45, 0.55, 0.65, 0.75, 0.85) as compared to when all of them were held at their medians. The ORs and their 95 percent credible intervals were calculated as exp(1.6*BKMR estimate) to convert a probit estimate to a logit estimate. The solid black line at OR = 1 represents the null association. The BKMR model for all women was adjusted for maternal age, race, prepregnancy BMI, education level, insurance status, parity, and fetal sex (for overall model only).

Note: BKMR, Bayesian Kernel Machine Regression; LGA, large-for-gestational-age; ORs, odds ratios.

Other birth outcomes were not the primary focus of the BKMR analysis since their associations with PFAS were largely non-significant in single pollutant models. However, we have provided PIPS (Tables S5a–b) and overall effect plots (Figs. S9–S12) corresponding to these outcomes as supplementary files. The ORs associated with PFAS mixture, as shown in these plots, were mostly null.

3.5. Sensitivity analysis

Multiple sensitivity analyses were conducted to assess the robustness of our findings. First, the analysis was restricted to participants selected for the original nested case-control study, which reduced the sample size by around six percent. Despite this, the main conclusions from the primary analysis remained largely unchanged (Tables S6a–b). Notably, the ORs for the associations of MPAH and PFOA with overall PTB in women who delivered female and male offspring respectively, increased above 1. Similarly, the ORs for PFDA, PFNA, and PFOA in relation to pPTB in all women or women who delivered male offspring also increased. For LGA, the ORs associated with PFDA and PFHxS in female offspring were attenuated, while the OR associated with PFOA increased. The OR for the association between PFUA and LGA in male offspring was still positive but attenuated.

Second, the findings were fairly robust to data transformation, as the direction of significant associations remained consistent when analysis was performed with non-log-transformed PFAS concentrations (Tables S7a–b). Third, adjusting for GA at the time of sample collection (Tables S8a–b) and restricting analysis to non-smokers (Tables S9a–b) did not meaningfully alter the main conclusions from the primary analysis for most PFAS and birth outcomes. When the analysis was restricted to term births, consistent results were found for BW Z-score and LGA (Table S10). However, the effect estimates for the associations between most PFAS, particularly PFNA, and SGA increased. The associations between prenatal PFAS levels and GA at delivery were mostly negative, but the 95 percent CIs included zero (Table S11). However, significant negative associations were observed between PFHxS and PFOS and GA at delivery when the analysis was restricted to term births.

The results based on multiply imputed datasets were generally consistent with those from the complete case data, although we observed some attenuation in the estimates for the associations between certain PFAS and pPTB in women who delivered male offspring (Tables S12a–b). Finally, the use of MRVs instead of $\mathrm{L}\mathrm{O}\mathrm{D}/\sqrt{2}$ to replace PFAS concentrations below the LOD had a negligible impact on the overall findings (Tables S13a–b).

4. Discussion

This study investigated the associations between prenatal exposure to PFAS and PTB, its subtypes, and fetal growth measurements, including BW Z-score, SGA, and LGA in a case-control study nested within the diverse LIFECODES cohort. In the primary analysis, positive associations were found between plasma levels of PFNA, PFDA, and PFUA and pPTB, while the associations between most PFAS and overall PTB were positive, but the confidence intervals included null. Similarly, the association between PFDA and LGA was weakly positive, while the associations between most PFAS and sPTB, BW Z-score, and SGA were null. Fetal sex-stratified analysis indicated some effect modification, with stronger positive associations of MPAH and PFOA with overall PTB in women who delivered female and male offspring, respectively. The effects of PFNA and PFOA on pPTB, and PFDA, PFHxS, and PFOA on LGA were also modified by fetal sex. It should be noted that the smaller sample size in the fetal sex-stratified models led to less precise effect estimates with wider confidence intervals in each fetal sex group. BKMR analysis identified PFNA, which was highly correlated with PFDA and PFUA, as the most important PFAS in relation to pPTB in all women or women who delivered male offspring. No meaningful pairwise interactions were observed between any PFAS in the studied associations. The findings were stable when accounting for data transformations, restrictions, or any potential confounding factors such as smoking or gestational age at the time of the sample collection.

Plasma PFAS levels observed in the study participants were close to those reported for women of comparable age in the NHANES 2007–2008 cycle. However, the detection rates of PFUA and PFHP were much higher in LIFECODES participants. Likewise, the observed PFAS levels were consistently higher than the current background level exposures in the US (CDC, 2019). Compared to other pregnancy cohorts in the US, the median PFAS levels in LIFECODES were generally higher, except for those reported by (Sagiv et al., 2017) in Project VIva in Boston, MA. The variations within the US might have resulted from differences in sample collection periods, exposure sources, demographic characteristics, or risk profiles of participants in each cohort.

The mostly positive yet null findings for overall PTB in this study align with previous studies that have reported primarily null associations between prenatal PFAS exposures and overall PTB (Manzano-Salgado et al., 2017; Hamm et al., 2010; Eick et al., 2020). They, however, deviate slightly from other studies that found significantly increased odds of overall PTB in relation to certain PFAS (Yang et al., 2022; Sagiv et al., 2017; Meng et al., 2018; Liao et al., 2022; Gardener et al., 2021; Chu et al., 2020; Chen et al., 2012). A recent meta-analysis by Deji et al. (2021) pooled results from most of the aforementioned studies and reported increased odds of PTB in relation to PFOS, but not PFHxS, PFNA, and PFOA. Among the studies that have specifically investigated PTB subtypes, Huo et al. (2020) and Liu et al. (2020) did not find any associations between PFAS and sPTB or indicated PTB. Similarly, the current study did not find evidence of reduced fetal growth in relation to elevated PFAS levels, which is consistent with a few other studies (Manzano-Salgado et al., 2017; Huo et al., 2020; Hu et al., 2021; Hamm et al., 2010; Eick et al. 2020, 2023; Chu et al., 2020), but different from others that have reported negative associations (Padula et al., 2023; Gardener et al., 2021; Hjermitslev et al., 2020; Chu et al., 2020; Marks et al., 2019; Sagiv et al., 2017; Bach et al., 2016; Maisonet et al., 2012; Chen et al., 2012; Wikstrom et al., 2020; Steenland et al., 2018; Washino et al., 2009).

The reasons for the observed heterogeneity among studies are not fully understood, but may involve differences in population characteristics, exposure levels, exposure matrix, timing of maternal sample collection, outcomes assessments, or the statistical method used for data analysis (Steenland et al., 2018; Bach et al., 2015). For example, half of the studies reporting positive associations between PFOS and overall PTB measured PFAS in samples collected later in pregnancy, after delivery, or in cord blood (Yang et al., 2022; Gardener et al., 2021; Chu et al., 2020; Chen et al., 2012). Likewise, the study of PTB subtypes by Huo et al. (2020) included mostly nulliparous women from Shanghai who may have substantially different exposure or risk profiles than the LIFECODES participants. Women in Shanghai Birth Cohort had higher levels of PFOA than PFOS, which is contrary to what has been observed in most pregnancy cohorts or the US general population. These differences in exposure profiles or population characteristics may affect the PFAS-birth outcome associations in either direction, complicating the granular comparisons of findings across different cohorts.

Very few studies have examined the association between prenatal PFAS exposures and LGA, which is an extreme measure of birthweight associated with chronic metabolic complications. The positive associations between several PFAS and LGA observed in this study contrast with Whitworth et al. (2012) that reported null associations for PFOA and PFOS in the Norwegian Mother and Child Cohort Study. The findings also contrast with recent results from a multi-cohort analysis from the ECHO study Padula et al. (2023), who observed decreased odds of LGA for PFDA, PFOA, and PFNA. Notably, these authors also reported positive, but null findings for several PFAS in relation to PTB, with the strongest association for PFNA which is consistent with our results. However, their study focused on overall PTB, defined using the 37 gestation weeks threshold, and did not distinguish between PTB subtypes, which might have masked potential associations between different PFAS and pPTB observed in this study.

The present study is one of the first to distinguish between sPTB and pPTB, which may correspond to distinct sets of pregnancy complications with common biological features. The etiologies leading to disorders preceding each PTB subtype are not well understood, but may involve inflammation, infection, oxidative stress, or placental dysfunction. A previous study showed that disorders leading to pPTB were commonly characterized by placental infarcts and increased syncytial knots (McElrath et al., 2008). These symptoms are potential indicators of uteroplacental insufficiency, which often results from abnormal implantation (Szilagyi et al., 2020) and may be affected by elevated maternal PFAS levels. Notably, the current findings align with preliminary results from a smaller, overlapping study in LIFECODES that investigated PFAS in relation to preeclampsia (Bommarito et al., 2021b). In that study, PFOS and PFDA were associated with increased odds of late-onset preeclampsia (occurring after 34 weeks gestation), but not with overall or early-onset preeclampsia.

Collectively, this study provides evidence for the associations of several PFAS with increased odds of pPTB, particularly in women who delivered male offspring, but an increased odds of LGA in female offspring. A few previously mentioned studies have reported sex-dimorphic effects of PFAS on birth outcomes, although the patterns of associations have been inconsistent (Zhang et al., 2022; Gao et al., 2022; Hall et al., 2022; Lind et al., 2017). The observed sex-specific effects could be attributed to multiple factors, including differences in hormonal regulation (Rivera-Nunez et al., 2023), gene expression (Petroff et al., 2023), or placental function and development that may interfere with fetal growth (Kalisch-Smith et al., 2017).

The current study has several strengths. It uses data from a racially diverse cohort with samples collected in 2006–2008 when the exposure levels of legacy PFAS in the US were just beginning to decrease. Moreover, this is one of the first US studies to evaluate the associations between prenatal exposures to MPAH, PFUA, and PFHP and birth outcomes. The number of PTB cases included in this study was among the most to date in relation to PFAS exposures, which enhances the robustness and accuracy of our findings. Additionally, it is among the first to differentiate between PTB subtypes or examine the effects of PFAS on LGA. Furthermore, the use of BKMR analysis in this study provided insights into possible non-linear dose-response relationships (for example, between PFNA and pPTB) and interactions or lack thereof among PFAS and their combined effects on birth outcomes.

The present study also has several limitations that should be considered when interpreting its findings. First, the sample sizes for the fetal-sex specific analysis were smaller, resulting in less precise estimates. The reduced statistical power might have limited the detection of subtle effects in both single pollutant and mixture analysis. Second, this study lacked data on maternal diet, physical activity levels, or psychosocial factors, making it difficult to identify potential exposure sources or assess whether these factors might confound or modify the effects of PFAS on birth outcomes. While a limited number of studies have explored the impact of diet or psychosocial stress on PFAS-birth outcomes relationships with mixed results, further research is needed to fully understand their influence. Lastly, plasma volume expansion and variability of renal function during pregnancy (Savitz, 2014) were not addressed in this study. However, since the samples in the current study were collected early in pregnancy (at a median gestation of approximately 10 weeks), it is less likely that these effects would meaningfully impact the observed associations between PFAS and birth outcomes (Verner et al., 2015; Sagiv et al., 2017).

5. Conclusions

In summary, the current study found associations between prenatal PFAS levels and placental PTB and LGA among pregnant women participating in the Boston-based LIFECODES study. Specifically, significant associations were observed between elevated levels of PFNA, PFDA, PFUA, and PFOA with pPTB in women who delivered male offspring, as well as increased levels of PFDA and PFOA and LGA in female offspring. The BKMR analysis largely supported the findings for pPTB from the single pollutant analysis, identifying PFNA as the most influential PFAS, and revealed no considerable interactions between PFAS congeners. Future research should focus on examining the effects of both legacy and emerging PFAS on birth outcomes, including PTB subtypes, within larger cohorts characterized by diverse exposure ranges and population demographics. Furthermore, additional studies are needed to better elucidate the underlying biological mechanisms responsible for overall and fetal sex-specific effects, explore potential modifying factors, and assess the long-term health outcomes for children exposed to higher levels of PFAS in utero.

Data availability

The authors do not have permission to share data.