Prenatal exposure to per- and polyfluoroalkyl substances and maternal thyroid function during pregnancy in a cohort with high familial likelihood of autism spectrum disorder

Abstract

Thyroid hormones supplied by the mother are essential for fetal brain development but could be disrupted by per- and polyfluoroalkyl substances (PFAS). We explored how prenatal PFAS exposures relate to maternal thyroid function in pregnant participants from the MARBLES cohort. We analyzed 212 serum samples from 151 pregnant women who later had a child with diagnosis of autism spectrum disorder (ASD), non-typical development (non-TD), or typical development (TD) by age 3. We quantified nine PFAS, total triiodothyronine (TT3), total thyroxine (TT4), and free thyroxine (FT4), and thyroid-stimulating hormone. We used a linear mixed effects model for individual and co-exposure effects, and Bayesian kernel machine regression (BKMR) for mixture effects. We conducted a mixed graphical model with child neurodevelopmental classification as an exploratory analysis. Perfluorooctane sulfonate (PFOS) and perfluorohexane sulfonate (PFHxS) were positively associated with TT4 (per 1-unit increase in ln-transformed concentrations β [95% confidence interval, CI]: 1.005 [0.108, 1.903] for PFOS; −0.581 [−1.160, −0.002] for PFHxS) and FT4-to-TT4 ratio (-0.153 [-0.270, −0.036] for PFOS; 0.090 [0.013, 0.166) for PFHxS). In the network map, PFAS were directly and indirectly associated with non-TD diagnosis, while TT3 was conditionally associated with non-TD. These findings indicate that prenatal PFAS exposure could interfere maternal thyroid function, potentially impacting fetal neurodevelopment.

Graphical Abstract

1. Introduction

Per- and polyfluoroalkyl substances (PFAS) comprise a large class of man-made substances used for decades in industrial and consumer applications.^1,2 Today, greater than 10,000 substances are classified as PFAS.^3,4 Due to their wide-ranging applications, PFAS are widespread in both environmental media and in human populations, referred to as “forever chemicals.” People are primarily exposed to PFAS through contaminated food, drinking water, air, and contact with PFAS-containing products.^5–7 Recognizing the various adverse health effects of PFAS, the U.S. Environmental Protection Agency (EPA) established legally enforceable limits for six PFAS compounds, including perfluorooctane sulfonate (PFOS), perfluorooctanoate (PFOA), perfluorohexane sulfonate (PFHxS), and perfluorononanoate (PFNA), in drinking water on April 10, 2024.⁸ However, despite regulatory efforts since the 2010s, long-chain legacy PFAS continue to be detected in human bodies due to their widespread use and long biological half-lives in human (e.g., PFOA: 2–10 years, PFOS: 3–27 years, PFHxS: 4–35 years, and PFNA: 2–4 years), with most geometric mean estimates across studies ranging from 3 to 8 years.^9–11 Notably, their presence in placenta, amniotic fluid, embryos, cord blood, and fetal organs^12–18 highlights their capacity to cross placental tissues and reach the developing fetus during pregnancy.

Maternal THs are essential for fetal brain development, as the fetus relies on the maternal TH supply before the second trimester.^19–21 Disruption of maternal TH levels during key stages of fetal brain development is linked to child neurodevelopmental effects, including reduced intelligence quotient (IQ) scores, autism spectrum disorder (ASD), and attention deficit hyperactivity disorder (ADHD).^3,22,23 Various factors, including iodine availability, thyrotropin (TSH) stimulation, and genetic background, can influence maternal thyroid function.^24,25 PFAS may also act as an additional influencing factor, creating a critical window of vulnerability by interfering with thyroid hormone homeostasis.^3,19

Several cohort studies have examined the associations between maternal PFAS concentrations and maternal TH levels measured in the same maternal blood during pregnancy. For example, a U.S. study (1999–2002) reported an inverse association between PFOA and free thyroxine (FT4).²⁶ A Taiwanese cohort (2000–2001) observed that PFHxS had a positive association with thyroid-stimulating hormone (TSH), and perfluorodecanoate (PFDA) was positively associated with total triiodothyronine (TT3), while PFNA, perfluoroundecanoate (PFUnDA), and perfluorododecanoate (PFDoDA) were negatively associated with total thyroxine (TT4) and FT4.²⁷ Similarly, a Swedish study (2007–2010) reported a negative association between PFUnDA and both TT4 and total triiodothyronine (TT3).²⁸ Additionally, a Danish study (2010–2012) observed positive associations of PFOA, PFOS, and PFNA levels with FT4.²⁹ Collectively, findings from these studies remain inconsistent, which may reflect differences in PFAS exposure levels, timing of TH measurement during gestation, and underlying population characteristics. Notably, no studies to date have specifically focused on sensitive populations who may be at greater risk of thyroid dysfunction.

We leveraged data from MARBLES (Markers of Autism Risk in Babies – Learning Early Signs), a high-risk cohort enriched for ASD due to elevated familial likelihood. We observed a higher prevalence of any type of thyroid dysfunction among MARBLES pregnant women compared with the U.S. general population (Table S1). Using this unique MARBLES data, we first examined the effects of (1) individual prenatal PFAS exposure and (2) PFAS mixtures on maternal thyroid function during pregnancy.

2. Materials and Methods

2.1. Study population

This study used the data of the MARBLES study, a prospective cohort study that began in 2006. The MARBLES cohort enrolls pregnant individuals with an older child previously diagnosed with ASD, placing them at an elevated likelihood (~20%) of having another affected child. Most participants were recruited from a list of families receiving state-funded services for children with ASD. Eligible participants were adults residing within driving distance of Davis/Sacramento who had an older child diagnosed with ASD and could complete study procedures in English. The study protocol was received and approved by the University of California Davis (UC Davis) Institutional Review Board (IRBNet Project ID: 225645–79) and by the California Health and Human Services Agency Committee for the Protection of Human Subjects (Project No. 12-11-0911). All participants provided informed consent prior to data and biospecimen collection. Comprehensive descriptions of the study design, recruitment procedures, and data collection methods have been published previously.³⁰

For this study, we included 151 mother-child pairs where maternal blood samples were collected at least once during pregnancy between 2009 and 2015, and their child was clinically classified at 3 years old. In addition to the mothers who delivered twins, participants were excluded if they (1) reported a history of thyroid conditions such as Graves’ disease or Hashimoto’s thyroiditis or (2) indicated current use of thyroid medication during pregnancy, based on self-reported questionnaire responses and medication records, or (3) had missing TH levels, chemical measurement or covariate data. A detailed flowchart of this study from the full cohort is shown in Figure 1.

Figure 1.

Flowchart for the study sample selection from the MARBLES full cohort. This study included mothers who provided at least one blood sample during pregnancy.

2.2. Serum sample collection and PFAS quantification

A total of 212 maternal blood samples were collected from 151 mother-child pairs across all trimesters of pregnancy. Each mother provided up to three samples, with 105 mother-child pairs providing one sample, 31 pairs providing two samples, and 15 pairs providing three samples (Figure 1). Most samples were collected during the second (n = 98, 46%) and third trimesters (n = 73, 34%), followed by the first trimester (n = 41, 19%). All blood samples were stored at −80 °C until analysis.³⁰ We measured nine PFAS in maternal serum: PFOA, PFOS, PFHxS, PFNA, perfluorodecanoic acid (PFDA), PFUnDA, PFDoDA, 2-(N-methyl-perfluorooctane sulfonamido) acetate (MeFOSAA), and N-ethyl perfluorooctane sulfonamido acetate (EtFOSAA). PFAS quantification was conducted at the Centers for Disease Control and Prevention (CDC) using automated solid-phase extraction followed by reversed-phase high-performance liquid chromatography with isotope dilution-tandem mass spectrometry. Additional details on sample preparation and instrumental analysis have been reported elsewhere.³¹ For quality control, each batch included blank samples and two quality control materials at low and high concentrations. For quality assurance, 25 blind duplicate samples were analyzed alongside study samples. The median coefficient of variation for duplicate pairs ranged from 0% to 11%, depending on the PFAS. The limit of detection (LOD) for all PFAS was 0.1 ng/mL, and values falling below this threshold were imputed as LOD √2.³²

2.3. Maternal serum collection and measurement of thyroid hormones and autoantibodies

We measured thyroid function tests (TFTs), including THs (TT3, TT4, and FT4), TSH, and thyroid antibodies (TAs) in the same serum samples analyzed for PFAS. All analyses were performed by the clinical laboratory at the University of Washington using automated immunoassays (Beckman DxI 800, Beckman Coulter Diagnostics, Brea, CA).³³ The limits of quantification (LOQ) were 40 ng/dL for TT3, 0.5 μg/dL for TT4, 0.2 ng/dL for FT4, 0.003 μIU/mL for TSH, and 1 IU/mL for both anti-thyroid peroxidase antibody (TPOAb) and anti-thyroglobulin antibody (TgAb). Following laboratory standards, TAs were classified as positive for TPOAb or TgAb. TPOAb positivity is defined as TPOAb levels ≥8.9 IU/mL, and TgAb positivity is defined as TgAb levels ≥1.9 IU/mL. Values below the LOQ were replaced with LOQ/√2. As additional thyroid function outcome metrics, we used the FT4/TT4 ratio to assess potential displacement of FT4 and its binding to transporter proteins, and the TT4/TT3 ratio to evaluate potential effects on T4 metabolism (via deiodinase enzymes).

2.4. Covariates

To determine potential confounders, we created a directed acyclic graph (DAG, Figure S1), selecting variables a priori based on prior literature.^34–39 The selected eight covariates include maternal age at delivery (<35 and ≥35 years), maternal education (less than bachelor’s and Bachelor’s or higher), homeownership (yes and no) as a proxy for socioeconomic status, child sex (male and female), smoking habits (non-smoker and regular smoker at any point), maternal pre-pregnancy body mass index (BMI) (<25 and ≥25 kg/m²), sampling year (2009–2010, 2011–2013, and 2014–2015) to account for temporal trends in PFAS concentrations, and trimester (first, second, and third trimester) to account for the variation in TFT levels (Table S2). As it is considered a potential mediator, we excluded antibody status (negative and positive) to avoid overadjustment but included it in the sensitivity analysis to assess the robustness of our findings.

2.5. Statistical analyses

We compared maternal characteristics across the three groups of child’s later neurodevelopmental classification. Descriptive statistics of mothers were summarized using the number of cases (n) and proportions (%). Differences in characteristics were compared using the chi-squared test. For further analysis, we included six PFAS with a detection rate of ≥60% (i.e., PFOA, PFOS, PFHxS, PFNA, PFDA, and PFUnDA) (Table 2). We computed intraclass correlation coefficients (ICCs) with ICCest() function in R to quantify within-person variability in PFAS concentrations and TFT levels. Due to their skewed distributions, we used ln-transformed concentrations for TFTs (except TT4) and PFAS (except PFDA and PFUnDA). PFDA and PFUnDA were analyzed as dichotomous variables (detected and non-detected) due to low variability, with median concentrations near or at the LOD and narrow interquartile ranges. We displayed the correlation heatmap using Pearson’s correlation coefficients for ln-transformed PFAS concentrations. We performed all statistical analyses in R (version 4.3.1, R Core Team, https://www.R-project.org/), and the results were considered statistically significant at the 0.05 level.

Table 2.

Summary statistics of maternal PFAS concentrations and thyroid function test (TFT) levels

Biomarker	LOD	Detection Rate	GM (95% CI)	Percentile					ICCs (95% CI)
				5th	25th	50th	75th	95th
PFAS (ng/mL)
PFOA	0.1	100.0	0.87 (0.80, 0.94)	0.30	0.60	0.90	1.28	2.13	0.83 (0.75, 0.89)
PFOS	0.1	100.0	2.92 (2.69, 3.17)	1.27	2.10	3.00	4.20	7.10	0.92 (0.87, 0.94)
PFHxS	0.1	100.0	0.45 (0.41, 0.50)	0.10	0.30	0.50	0.70	1.60	0.24 (0.001, 0.54)
PFNA	0.1	100.0	0.48 (0.44, 0.51)	0.20	0.33	0.50	0.70	1.03	0.82 (0.73, 0.88)
PFDA	0.1	84.6	0.15 (0.13, 0.16)	<LOD	0.10	0.10	0.20	0.40	0.96 (0.94, 0.97)
PFUnDA	0.1	62.6	0.11 (0.10, 0.12)	<LOD	<LOD	0.10	0.10	0.30	0.96 (0.94, 0.97)
PFDoDA	0.1	25.7	0.08 (0.08, 0.08)	<LOD	<LOD	<LOD	0.10	0.10	0.71 (0.57, 0.8)
MeFOSAA	0.1	59.3	0.12 (0.11, 0.13)	<LOD	<LOD	0.10	0.10	0.80	0.76 (0.64, 0.83)
EtFOSAA	0.1	2.3	0.07 (0.07, 0.07)	<LOD	<LOD	<LOD	<LOD	<LOD	0.94 (0.9, 0.96)
TFTs
TT3 (ng/dL)	40		139 (135, 144)	94.8	119.2	142.0	166.46	191.1	0.68 (0.52, 0.78)
TT4 (μg/dL)	0.5		11.1 (10.7, 11.4)	8.34	10.06	11.39	12.61	14.87	0.53 (0.33, 0.67)
FT4 (ng/dL)	0.2		0.64 (0.63, 0.66)	0.47	0.57	0.65	0.72	0.88	0.10 (0.0003, 0.47)
TSH (μIU/mL)	0.003		1.27 (1.15, 1.41)	0.35	1.02	1.43	1.90	2.81	0.75 (0.62, 0.83)

Abbreviations: GM, geometric mean; CI, confidence interval, ICCs, intraclass correlation coefficients; LOD, limit of detection; PFAS, per- and polyfluoroalkyl substances; PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate; PFHxS, perfluorohexane sulfonate; PFNA, perfluorononanoate; PFDA, perfluorodecanoate; PFUnDA, perfluoroundecanoate; PFDoDA, perfluorododecanoate; MeFOSAA, 2-(N-methyl-perfluorooctane sulfonamido) acetate; EtFOSAA, N-ethyl perfluorooctane sulfonamido acetate; TT3, total triiodothyronine; TT4, total thyroxine; FT4, free thyroxine; TSH, thyroid-stimulating hormone.

We examined the overall association between exposure to individual PFAS compounds and thyroid function using multiple regression. We investigated these associations under both single and co-exposure scenarios: (1) single-exposure models, in which each PFAS was modeled separately without adjustment for other PFAS compounds, and (2) a co-exposure model, which included multiple PFAS in the same model to estimate associations conditional on co-exposures. Both approaches assess the independent effect of each PFAS, but the co-exposure model accounts for potential confounding.

To examine mixture effects of PFAS on thyroid function, we applied the Bayesian Kernel Machine Regression (BKMR) using multiple parallel chains. This Bayesian approach is well suited for small samples and allows more flexible modeling than traditional frequentist methods.⁴⁰ Using BKMR, we obtained posterior inclusion probabilities (PIPs) to characterize the relative importance of individual PFAS. We assessed the overall mixture effect by estimating changes in thyroid hormone levels when all PFAS were simultaneously set at various percentiles (10th–95th, in 5th-percentile increments) compared with the 50th percentile. We further characterized the contribution of individual exposures to the overall effect by estimating differences when a single exposure was set at the 25th or 75th percentile while the other exposures were fixed at the 20th, 50th, and 80th percentiles.⁴¹ BKMR models were run using five parallel chains of 5,000 iterations each, following a 500-iteration burn-in, with default parameter settings. All models were adjusted for covariates, and subject ID was included as a random effect to account for individual variability and differing numbers of repeated measures across participants.⁴²

We performed sensitivity analysis by including TAs in the model to assess the robustness of our findings. As the potential mediator, TAs might have indirect effects of autoimmune-related thyroid damage. Autoimmune damage induced by PFAS may increase susceptibility to thyroid-disrupting effects, potentially reducing TH synthesis.^43,44 Thus, we conducted single- and co-exposure using the same methodology.

Interestingly, we found distinct associations between prenatal PFAS exposure and maternal thyroid function during pregnancy, along with changes in TH metabolism. To extend these findings, we conducted an exploratory analysis to examine the potential mediating role of thyroid dysfunction in child neurodevelopment using a mixed graphical model (MGM). The MGM estimates conditional associations among mixed data types (including both continuous and categorical variables) under the assumption of no inherent directionality.⁴⁵ Because of this assumption, the MGM can reveal potential mediation-like patterns by identifying conditional dependence among variables but does not allow causal inference or formal mediation testing.⁴⁶ In this framework, edges denote conditional relationships between pairs of variables, accounting for the influence of all remaining variables, and edge weight (line width) reflects the strength of the association, derived from standardized regression parameters in nodewise regressions.⁴⁵ Our model included PFAS, thyroid hormones (THs), TSH, covariates, and child neurodevelopmental outcomes (autism spectrum disorder [ASD], typical development [TD], and non-typical development [non-TD]). The child’s neurodevelopment status is classified according to MARBLES guidelines (Figure S2).^47–49

3. Results

3.1. Participant characteristics

Among the 151 mother-child pairs, most (83%) mothers were not regular smokers at any point during pregnancy. Only 12 mothers (8%) tested antibody-positive at least once during pregnancy (Table 1). We observed variation in participant characteristics across PFAS concentrations (Table S3). Overall, PFAS concentrations were higher among women aged ≥35 years at delivery and lower among those with two or more previous pregnancies. While concentrations significantly declined over the sampling years, no differences were observed by trimester or the number of sample collections (Table S3).

Table 1.

Participant characteristics included in the present study (n [%])

	151 Mother-child pairs
Maternal age at delivery (years)
<35	82 (54.3)
≥35	69 (45.7)
Maternal race/ethnicity
Non-Hispanic White	76 (50.3)
All other races and ethnicitiesa	75 (49.7)
Parity
≤1	66 (43.7)
>1	85 (56.3)
Child sex
Male	64 (42.4)
Female	87 (57.6)
Maternal education
Less than bachelor’s	76 (50.3)
Bachelor’s or higher	75 (49.7)
Homeownership
No	62 (41.1)
Yes	89 (58.9)
Smoking habits
No	125 (82.8)
Regular smoker at any point	26 (17.2)
Maternal pre-pregnancy body mass index (BMI) (kg/m2)
<25.0	74 (49.0)
≥25.0	77 (51.0)
Antibody status b
Negative	139 (92.1)
Positive	12 (7.9)

^aIncludes Hispanic persons, Black persons, Asian persons, and persons with other races.

^bDefined whether thyroid-peroxidase (TPO) or thyroglobulin (TG) antibody positive: TPO positivity is defined as TPO levels ≥ 8.9 IU/dL, and TG positivity is defined as TG levels ≥ 1.9 IU/dL.

3.2. Prenatal maternal PFAS concentrations and TFT levels

PFOS concentrations were predominant with a median of 3.0 ng/mL, followed by PFOA (0.90 ng/mL), PFHxS and PFNA (both 0.50) ng/mL, and PFDA and PFUnDA (both 0.10 ng/mL) (Table 2). Except for PFHxS, the ICCs of the other five ln-transformed PFAS ranged from 0.71 to 0.96, indicating that maternal PFAS serum concentrations varied only modestly within participants (Table 2). PFAS were weakly to moderately correlated with each other (ρ range: 0.14–0.70): PFNA with PFOA and PFOS had the highest correlation (ρ = 0.70, p < 0.05) (Figure S3). Maternal TT3, TT4, and FT4 were detected in full samples, and maternal TSH levels were above the LOQ in all but one sample (99.5%). The distributions of ln-transformed PFAS and TFT levels are shown in Figure S3. We compared PFAS concentrations in the MARBLES cohort to those reported among pregnant women in the National Health and Nutrition Examination Survey (NHANES) during a similar time period. Compared with NHANES, MARBLES participants exhibited marginally lower PFAS concentrations but a higher frequency of thyroid dysfunction, as determined by FT4 and TSH reference ranges (Table S1).

3.3. Associations between individual PFAS compounds and TFTs

We estimated the individual effects of prenatal exposure to PFAS on TFTs under the co-exposure scenario. Higher concentrations of PFOA were positively associated with TT3 (β = 0.098, 95% confidence interval [CI]: 0.014, 0.182), whereas PFHxS was inversely associated with TT3 (β = −0.105, 95% CI: −0.158, −0.051), a finding that was also significant in the single-exposure model. Moreover, higher concentration of PFOS was associated with an elevated level of TT4 (β = 1.005, 95% CI: 0.108, 1.903), and a lower FT4-to-TT4 ratio (β = −0.153, 95% CI: −0.270, −0.036). In contrast, PFHxS was associated with a lower level of TT4 (β = −0.581, 95% CI: −1.160, −0.002), and a lower FT4-to-TT4 ratio (β = −0.153, 95% CI: −0.270, −0.036). Lastly, PFNA showed a negative association with TSH (β = 0.090, 95% CI: 0.013, 0.166). (Figure 2 and Table S4). The most consistent associations were observed when TA status was included in the model as a covariate (Figure S5).

Figure 2.

Associations between individual ln-transformed PFAS concentration and ln-transformed TFT levels (except for TT4) under a co-exposure scenario.

Note: All models were adjusted for maternal age at delivery, child sex, maternal education, maternal homeownership, smoking habits, maternal pre-pregnancy body mass index (BMI), sampling year, and blood collection time (trimester). Red and blue shaded areas represent significant associations. Asterisks (*) indicate significant associations under a single exposure scenario.

Abbreviations: PFAS, per- and polyfluoroalkyl substances; TFTs, thyroid function tests; TT3, total triiodothyronine; TT4, total thyroxine; FT4, free thyroxine; TSH, thyroid-stimulating hormone; PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate; PFHxS, perfluorohexane sulfonate; PFNA, and perfluorononanoate; and PFDA, perfluorodecanoate; PFUnDA, perfluoroundecanoate; CI, confidence interval.

3.4. Associations between PFAS mixture and TFTs

From the BKMR, PFOA (PIP = 0.291) and PFHxS (PIP = 0.228) showed relatively high PIPs for TT3, whereas all PFAS had similar PIPs for TT4 (0.110–0.180), FT4 (0.216–0.281), and TSH (0.443–0.528). We observed slight decreasing trends in TT3 and TT4, and an increasing trend in TSH with higher PFAS mixture concentrations, but not significant. No significant single-exposure effects within the mixture were detected (Figure 3).

Figure 3.

Associations between PFAS mixtures and thyroid function tests (TFTs) estimated by Bayesian kernel machine regression. The left panel shows the posterior inclusion probabilities (PIPs) for individual PFAS, the middle panel displays the overall mixture effects on each TFT with estimates and 95% confidence intervals, and the right panel presents the single-chemical effects within the mixture across selected exposure quantiles.

Abbreviations: PC, principal component; PFAS, per- and polyfluoroalkyl substances; PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate; PFHxS, perfluorohexane sulfonate; PFNA, and perfluorononanoate; PFDA, perfluorodecanoate; PFUnDA, perfluoroundecanoate; TD, typically developing children; ASD, autism spectrum disorder; non-TD, children with other neurodevelopmental concerns.

3.5. Exploratory analysis

From the network map, PFAS compounds showed positive conditional associations with each other after controlling for covariates. Among thyroid hormones, TT3 and TT4 were positively associated, whereas FT4 and TSH were negatively associated. FT4 demonstrated a relatively strong association with trimester. Notably, PFDA was directly associated with non-TD diagnosis, while ASD was directly associated with child sex and maternal education. In addition, PFHxS appeared to be indirectly linked to non-TD through its association with TT3 levels (Figure 4).

Figure 4.

Conditional dependency-based network structure of maternal PFAS exposure, maternal thyroid function tests (TFTs), and covariates. Edges represent pairwise conditional associations adjusted for covariates. Line width and accompanying values indicate edge weights, with thicker lines corresponding to stronger associations. Value colors denote the direction of association (positive = red, negative = blue, inconsistent = grey). Node colors distinguish PFAS (purple), TFTs (sky blue), child neurodevelopmental status (pink), and covariates (yellow).

Abbreviations: BMI, body mass index; ASD, autism spectrum disorder; non-TD, children with other neurodevelopmental concerns; PFAS, per- and polyfluoroalkyl substances; PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate; PFHxS, perfluorohexane sulfonate; PFNA, and perfluorononanoate; and PFDA, perfluorodecanoate; PFUnDA, perfluoroundecanoate; TFTs, thyroid function tests; T3, triiodothyronine; T4, thyroxine; and TSH, thyroid-stimulating hormone.

4. Discussion

This study explored how prenatal PFAS exposures relate to maternal thyroid function in a high familial likelihood ASD cohort using repeatedly collected serum samples. Our findings revealed links between prenatal PFAS exposure and markers of maternal TH homeostasis. Specifically, PFOA was associated with TT3, PFOS with TT4, PFNA with TSH, and PFHxS with both TT3 and TT4. We also noted patterns suggesting potential thyroidal perturbations linked to PFOS and PFHxS exposure, consistent with established pathways of TH metabolism.

Our findings were consistent with previous studies. For example, in the Wuxi Birth Cohort (2019–2021), a study of 562 pregnant participants reported that higher PFNA concentrations were linked to lower TSH levels.⁵⁰ Similarly, a study of 726 TPOAb-positive mothers in the Boston area (USA) showed negative associations between PFOS, PFNA and TSH.²⁶ A recent meta-analysis reported significant pooled estimates supporting positive associations of PFOS with TT3 and TT4, consistent with our findings.⁵¹ Only five studies have examined the relationship between prenatal PFAS mixture exposure and maternal TH homeostasis.^50,52–55 Although our mixture exposure–response associations were not significant, due to limited sample size, we observed patterns similar to those reported previously. For instance, a U.S. fertility clinic study reported decreasing trends in TT3 and TT4 with higher PFAS mixtures.⁵⁴ The Wuxi Birth Cohort and the Health Outcomes and Measures of the Environment (HOME) Study both found decreasing trends in TSH associated with PFAS mixtures.^50,55 However, inconsistent findings in humans have also been reported, suggesting potential variations due to heterogeneity in design, participant characteristics, and PFAS exposure distributions.

Toxicological research have suggested several potential mechanisms by which PFAS might alter TH homeostasis, including disruptions in TH signaling pathways, inhibition of deiodinase enzyme activity, and alterations in TH-binding proteins.^3,51 Specifically, PFAS may promote increased T4-to-T3 conversion through stimulation of type 1 deiodinase,^28,56 leading to increased T3 and lower T4.⁵⁷ Additionally, PFAS may influence thyroid physiology by competing with free THs for binding to T4-binding globulin or transthyretin and by stimulating enzymes involved in hormone clearance, which can lead to higher levels of unbound THs.^58,59

Notably, we newly identified that PFOS and PFHxS was associated with TT4 and the FT4-to-TT4 ratio in opposite directions. These divergent associations might reflect differences in their physicochemical properties. Although both are perfluoroalkane sulfonates (PFSAs), PFHxS has a shorter carbon chain (C6) compared to PFOS (C8), influencing their tissue distribution and protein-binding affinity.^3,60 PFOS tends to accumulate more in the liver, where it could disrupt TH metabolism and interfere with the conversion of T4 to T3, contributing to elevated TT4 levels.⁶¹ In contrast, PFHxS exhibits lower hepatic retention and higher affinity for albumin, a major TH transport protein.^61,62 By competitively binding to albumin, PFHxS may displace T4, potentially reducing circulating TT4 levels.²⁶

Based on a network analysis, our results point to a possible mediating role of maternal thyroid function in the relationship between prenatal PFAS exposure and subsequent child neurodevelopment. Maternal THs are essential for fetal brain development until mid-gestation,⁶³ and disruption may increase the risk of neurodevelopmental disorders such as ASD and ADHD.^23,64–66 Recent studies indicate that exposure to endocrine-disrupting chemicals (EDCs) during pregnancy, including PFAS, phthalates, and PCBs, may contribute to adverse outcomes by disrupting maternal thyroid homeostasis.^67–69 While no study has directly examined this potential mediation pathway (PFAS exposure-thyroid dysfunction-child neurodevelopment), several prior studies support the individual links in this hypothesized pathway. For instance, findings from both previous and our own MARBLES studies consistently report associations between prenatal PFOA exposure and adverse neurodevelopmental outcomes. A prospective cohort in Norway found prenatal PFOA exposure increased ASD risk,⁷⁰ and a pooled analysis of seven studies linked higher PFOA exposure to greater ADHD prevalence.⁷¹ In our earlier MARBLES study, prenatal PFOA exposure was also associated with reduced MSEL scores at 24 and 36 months.⁷² Together, these findings highlight the plausibility of maternal thyroid disruption as a pathway linking prenatal PFAS exposure to child neurodevelopment, though further research is needed to confirm this mechanism.

This study has several notable strengths. First, it focused on a highly susceptible population—mothers with a prior child diagnosed with ASD—who also had a higher prevalence of thyroid dysfunction based on the reference range of FT4 and TSH levels, thereby increasing the relevance of our findings to at-risk groups. Second, the availability of repeated measurements of PFAS and thyroid hormones during pregnancy for a subset of participants strengthened both exposure and outcome characterization. Third, to our knowledge, this is one of the few studies to incorporate child neurodevelopmental outcomes into an investigation of prenatal PFAS exposure and maternal thyroid function, offering novel insights into potential pathways relevant to ASD etiology. Finally, by applying advanced co-exposure, mixture modeling and network approaches, we were able to account for the complexity of real-world exposures and capture interrelationships that might be overlooked in traditional single-exposure analyses.

There are also important limitations to consider. First, the generalizability of our findings to broader populations may be limited since our study used the data from mothers of children with a high familial likelihood of ASD. Second, the modest sample size (152 mother-child pairs) likely reduced the statistical power to detect certain associations. Third, residual confounding from factors not captured in our data such as iodine status, lifestyle, or behaviors may still be present. Iodine is an essential element in TH synthesis, acting as an intermediate factor in the relationship between PFAS exposure and thyroid function.⁷³ Lifestyle and behaviors influence PFAS exposure levels, given that major exposure pathways include drinking water, diet, and ingestion of household dust.⁷⁴ Fourth, PFAS exposure and TH levels were measured in the same maternal serum samples, making the exposure–outcome analysis cross-sectional in nature. However, given long elimination half-lives, typically ranging from 3 to 8 years, these serum measurements likely reflect cumulative exposure rather than recent exposure.^11,75 Fifth, PFDA and PFUnDA were assessed as dichotomous variables due to their limited variability—median concentrations were at or near the LOD, with an interquartile range of just 0.1 ng/mL. This may limit the ability to detect subtle associations and should be considered when interpreting results for these compounds. Sixth, biospecimen samples were primarily collected during the mid- to late pregnancies. Participants were enrolled at varying gestational ages. As a result, consistent sample collection during the first trimester was challenging, largely due to difficulties in recruiting participants early in pregnancy. Lastly, this study did not support a formal mediation analysis due to the relatively small sample size. Nonetheless, this study provides valuable insight into the potential differential effects of PFAS on thyroid function in relation to child neurodevelopment.

Our findings suggest that prenatal PFOS and PFHxS exposures may influence maternal thyroid function in pregnancies among women already at elevated familial risk for ASD. However, due to the relatively small sample size and the specific characteristics of the MARBLES cohort, additional studies are required to validate and further investigate these associations.

Supplementary Material

Descriptive statistics of PFAS concentrations, thyroid function test (TFT) levels, and prevalence of thyroid dysfunction among our study participants and U.S. pregnant women, results of associations between prenatal PFAS exposure and maternal TFT levels, directed acyclic graph (DAG) for covariate selection, classification criteria for children’s neurodevelopmental status in the MARBLES study, correlation matrices among ln-transformed PFAS concentrations, additional table and figure showing data distributions, and sensitivity analyses.

Synopsis:

This study highlights the importance of examining how prenatal PFAS exposure may differentially affect maternal thyroid function based on child neurodevelopmental outcomes, offering new insights into the role of maternal thyroid hormones in fetal brain development.

Abbreviations

ADHD

attention deficit hyperactivity disorder

ASD

autism spectrum disorder

BKMR

Bayesian kernel machine regression

BMI

body mass index

CDC

Centers for Disease Control and Prevention

DAG

directed acyclic graph

DSM-5

diagnostic and statistical manual of mental disorders, 5th edition

EDCs

endocrine disrupting chemicals

EtFOSAA

N-ethyl perfluorooctane sulfonamido acetate

FT3

free triiodothyronine

FT4

free thyroxine

ICCs

intraclass correlation coefficients

IQ

intelligence quotient

LOD

limit of detection

LOQ

limits of quantification

MARBLES

markers of autism risk in babies – learning early signs

MeFOSAA

2-(N-methyl-perfluorooctane sulfonamido) acetate

MSEL

Mullen scales of early learning

MGM

mixed graphical model

Non-TD

non-typical development

NIS

sodium iodide symporter

NHANES

the National Health and Nutrition Examination Survey

PCBs

polychlorinated biphenyls

PFAS

per- and polyfluoroalkyl substances

PFDA

perfluorodecanoic acid

PFDoDA

perfluorododecanoate

PFHxS

perfluorohexane sulfonate

PFNA

perfluorononanoate

PFOA

perfluorooctanoate

PFOS

perfluorooctane sulfonate

PFUnDA

perfluoroundecanoate

PIPs

posterior inclusion probabilities

SD

standard deviation

T4

thyroxine

TAs

thyroid antibodies

TD

typical development

TFTs

thyroid function tests

THs

thyroid hormones

TG

thyroglobulin

TgAb

thyroglobulin antibody

TPO

thyroid peroxidase

TPOAb

thyroid peroxidase antibody

TSH

thyroid-stimulating hormone

TT3

total triiodothyronine

TT4

total thyroxine

U.S. EPA

U.S. Environmental Protection Agency

Data availability

Data will be made available on request.