Maternal serum and placental metabolomes in association with prenatal exposure to per- and polyfluoroalkyl substances and their relevance to child neurodevelopment in an ASD-enriched cohort

Abstract

Prenatal exposure to per- and polyfluoroalkyl substances (PFAS) has been linked to altered neurodevelopment in children, but the contribution of maternal metabolic disruption to this relationship remains unclear. We investigated associations between prenatal PFAS exposure, maternal metabolism, and child neurodevelopment. We analyzed 172 mother-child pairs from the MARBLES (Markers of Autism Risk in Babies–Learning Early Signs) cohort. Nine PFAS were measured in maternal serum collected during pregnancy. Metabolites were quantified in third-trimester serum and placental tissue using proton nuclear magnetic resonance (¹H-NMR) spectroscopy. At age three, children were clinically classified as having autism spectrum disorder (ASD), typical development (TD), or non-typical development (non-TD), the latter including children with atypical developmental features who do not meet the criteria for ASD. Multiple linear regression assessed associations between individual PFAS and metabolites, and quantile-based g-computation evaluated PFAS mixture effects. Principal component analysis (PCA) summarized metabolomic profiles. One-way analysis of covariance (ANCOVA) and multinomial logistic regression examined associations between metabolites and child neurodevelopment. Correlation network analysis explored relationships among PFAS, serum, and placental metabolites. After multiple comparison correction, perfluorooctane sulfonate (PFOS) was significantly associated with serum 2-hydroxybutyrate (q < 0.10). Higher perfluorooctanoate (PFOA), PFOS, and PFAS mixture levels were associated with lower serum PC-2 scores. Higher serum PC-3 score, reflecting mitochondrial dysfunction, was associated with increased non-TD risk. Network analysis identified 2-hydroxybutyrate as a key serum metabolite potentially linked to PFAS and placental amino acids. Prenatal PFAS exposure was associated with maternal metabolic alterations; however, no clear linkage to child neurodevelopment were observed. These findings suggest the need to consider gene-environment interactions in studies of neurodevelopmental outcomes.

1. Introduction

Per- and polyfluoroalkyl substances (PFAS) have been widely manufactured and used for decades in various consumer products, including non-stick cookware, food packaging, textiles, furniture, and building materials (De Silva et al., 2021; Sunderland et al., 2019). Due to their widespread use, PFAS are commonly detected in the general population, including among pregnant women (Monroy et al., 2008). Importantly, PFAS have been detected in amniotic fluid, placenta, cord blood, embryos, and fetal organs, indicating the potential for maternal-fetal transfer during pregnancy (Bangma et al., 2020; Hall et al., 2022; Jensen et al., 2012; Mamsen et al., 2019; Mamsen et al., 2017; Monroy et al., 2008; Zhang et al., 2013). A growing body of epidemiologic evidence suggests that prenatal PFAS exposure may be associated with altered child neurodevelopment, such as autism spectrum disorder (ASD) (Harris et al., 2018; Oh et al., 2021a; Oh et al., 2021b; Shin et al., 2020; Skogheim et al., 2021; Spratlen et al., 2020; Strøm et al., 2014; Zhang et al., 2023; Zhou et al., 2023).

The mechanisms by which prenatal PFAS exposure could affect child neurodevelopment remain unclear. Maternal metabolic perturbation has been suggested as a possible pathway by some prior studies. During pregnancy, maternal physiology and metabolism undergo extensive adaptations to support the healthy growth and development of the fetus (Hadden and McLaughlin, 2009; Kalhan, 2000; Napso et al., 2018; Weissgerber and Wolfe, 2006). However, evidence suggests that PFAS exposure may disrupt these normal metabolic adaptations, potentially leading to abnormal fetal development (Chang et al., 2022; Hu et al., 2019; Li et al., 2021; Liang et al., 2023; Prince et al., 2023; Rabotnick et al., 2024; Sinisalu et al., 2021; Taibl et al., 2023). Notably, two studies using data from the same U.S. birth cohort reported that prenatal PFAS exposure was associated with metabolic perturbations in both maternal serum and newborn dried blood spots (DBS), which were in turn linked to birth outcomes such as fetal growth and gestational length (Chang et al., 2022; Taibl et al., 2023). Additionally, several cohort studies have found that mothers of children with neurodevelopmental concerns exhibit distinct metabolic patterns during pregnancy compared to mothers of typically developing children (da Silva Rosa Freire et al., 2024; Kim et al., 2021; Schmidt et al., 2021). For example, one U.S. cohort study that used non-targeted metabolomic analysis reported that alterations in the prostaglandin pathway, indicative of neuroinflammation, were associated with atypical development in children who are at elevated likelihood of developing ASD (Schmidt et al., 2021).

Previous studies have reported significant associations between maternal PFAS exposure and either metabolic alterations or neurodevelopmental outcomes in offspring, suggesting potential molecular pathways linking PFAS exposure to child’s neurodevelopment. However, to our knowledge, no studies have simultaneously examined the relationships among maternal PFAS exposure, metabolomic changes, and child neurodevelopmental outcomes. Moreover, most existing studies have focused on metabolites measured in blood-based specimens such as serum, plasma, DBS, or cord blood. While informative, these bio-specimens may not fully capture the biological processes most relevant to fetal development. The placenta, a transient yet essential organ, plays a critical role in fetal nutrient delivery, gas exchange, waste removal, and the synthesis of hormones, cytokines, and growth factors. Optimal placental function is essential for supporting healthy fetal development, and its dysfunction, such as with preeclampsia, has been linked to increased risk for adverse neurodevelopmental outcomes in offspring (Cai et al., 2016; Chen et al., 2023; Chen et al., 2021b; Krakowiak et al., 2012; Rodolaki et al., 2023; Saros et al., 2023). Given the placenta’s central role in influencing intrauterine exposures and supporting fetal development, characterizing the placental metabolome could provide critical insights into the biological mechanisms by which prenatal PFAS exposure influences neurodevelopmental trajectories. While maternal serum reflects systemic maternal physiology, placental metabolomics complements this by capturing localized biochemical alterations at the maternal-fetal interface. Therefore, integrating the data from both maternal serum and placenta enhances our ability to uncover shared and distinct metabolic signatures associated with prenatal chemical exposure. More studies utilizing multiple biological matrices are needed to comprehensively capture the metabolomic changes.

This study aimed to address key gaps in the literature by examining: (i) the associations between prenatal PFAS exposure and metabolomic profiles in both maternal serum and placenta; (ii) whether these PFAS-associated metabolomic changes are linked to child neurodevelopmental outcomes; and (iii) the overall structure and patterns of metabolic disruption associated with prenatal PFAS exposure which may help elucidate potential biological pathways through which PFAS exposure affects fetal brain neurodevelopment. To our knowledge, this is one of the first studies to integrate both maternal serum and placental metabolome data in the context of PFAS exposure. We also employed network-based correlation analysis to investigate complex PFAS-metabolite relationships and identify coordinated disruptions in metabolic pathways. Moreover, by leveraging an ASD-enriched cohort, this study provides a unique opportunity to explore gene-environment interactions in a population with elevated neurodevelopmental susceptibility.

2. Materials and methods

2.1. Study population

The data for this study were obtained from the MARBLES (Markers of Autism Risk in Babies – Learning Early Signs) cohort (Hertz-Picciotto et al., 2018). Established in 2006, MARBLES recruited pregnant women with an elevated likelihood (~19 %) of having another child who might develop ASD due to having an older child diagnosed with ASD (Hertz-Picciotto et al., 2018; Ozonoff et al., 2011). Eligible participants were identified from lists of families receiving state-funded services for children diagnosed with ASD and met the following criteria: they or a first-degree relative (i) had at least one child with ASD; (ii) were 18 years of age or older; (iii) spoke, read, and understood English; and (iv) resided within 2.5 h of the Davis/Sacramento region. The study received ethical approval from the Institution Review Boards for the University of California, Davis (UC Davis) and the State of California. Informed consent was obtained from all participants before data collection. More information on the study design, recruitment, and data collection processes can be found elsewhere (Hertz-Picciotto et al., 2018).

2.2. Child neurodevelopmental assessment

At 3 years of age, children underwent assessment with the Autism Diagnostic Observation Schedule (ADOS), which is a widely used standardized tool for diagnosing ASD (Lord et al., 2000; Ozonoff et al., 2005). Children were also assessed for cognitive development using the Mullen Scales of Early Learning (MSEL), a validated instrument designed for children from birth to 36 months (Mullen, 1995). Neurodevelopmental outcomes were categorized into three groups: ASD, non-typical development (non-TD), or typical development (TD). An algorithm based on the Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (DSM-5) criteria, combined with ADOS and MSEL scores, was used for the classification. Children identified as having ASD fulfilled DSM-5 diagnostic criteria for ASD and had ADOS calibrated severity scores (CSS) of ≥4. The non-TD group included children who did not meet DSM-5 criteria for ASD but had an ADOS CSS ≥3 and/or had two or more MSEL subscale scores ≥1.5 standard deviations (SD) below the mean and/or at least one MSEL subtest score ≥2 SD below the mean. Children classified as TD did not meet these criteria (Ozonoff et al., 2014).

2.3. PFAS quantification in maternal serum

Prenatal PFAS exposure was assessed by measuring serum PFAS concentrations in maternal blood samples collected during pregnancy. In the MARBLES study, mothers were asked to provide blood samples during each trimester for up to three samples per participant. However, due to the variation in maternal enrollment timing, the number of samples collected per participant varied. A total of 267 blood samples were collected from 172 mother-child pairs including two twin pregnancies. Most mothers provided samples during the second trimester (n = 123, 46 %), followed by the third (n = 88, 33 %) and first trimesters (n = 56, 21 %). PFAS concentrations in this population remained relatively stable and showed low within-subject variability throughout pregnancy (Oh et al., 2022). Thus, average PFAS concentrations were calculated for mothers with multiple samples and used in all statistical analyses. Whole blood samples were centrifuged to separate serum and stored at −80 °C until analysis. Additional details on blood collection, transport, and storage protocols can be found elsewhere (Hertz-Picciotto et al., 2018).

Serum concentrations of nine PFAS were quantified at the Centers for Disease Control and Prevention (CDC), including perfluorohexane sulfonate (PFHxS), perfluorooctane sulfonate (PFOS), perfluorooctanoate (PFOA), perfluorononanoate (PFNA), perfluorodecanoate (PFDA), perfluoroundecanoate (PFUnDA), perfluorododecanoate (PFDoDA), 2-(N-methyl-perfluorooctane sulfonamido) acetate (Me-FOSAA), and 2-(N-ethyl-perfluorooctane sulfonamido) (Et-FOSAA). Quantification was conducted using an online solid-phase extraction coupled to reversed-phase high-performance liquid chromatography-isotope dilution tandem mass spectrometry. The limit of detection (LOD) for all PFAS was 0.1 ng/mL. Detailed methods of sample preparation and instrumental analysis can be found elsewhere (Kato et al., 2011). For quality assurance, 25 duplicate samples were analyzed along with the study samples. The median coefficient of variation (CV) for these duplicates ranged from 0 % to 11 %, depending on the specific PFAS.

Given the long biological half-lives of PFAS and their relatively stable concentrations throughout pregnancy in this cohort, we considered the averaged serum PFAS concentrations as proxies for cumulative prenatal exposure, minimizing the potential for exposure misclassification due to sample timing variability.

2.4. ¹H NMR metabolomics analysis in maternal serum and placenta

For the metabolomic analysis, we used maternal serum collected during the third trimester and placental tissue obtained at delivery. While maternal PFAS exposure was assessed using samples collected throughout pregnancy, metabolomics analyses were performed on third-trimester serum and placental tissue collected at delivery. These later time points were selected because they may reflect cumulative physiological and metabolic responses to sustained prenatal PFAS exposure, particularly given the established stability of PFAS levels during pregnancy in this cohort. Maternal blood was collected into serum-separator tubes, processed within a few hours to isolate serum, and aliquoted into sterile cryovials stored at −80 °C. Samples were stored in a closed container and transported to the UC Davis specimen processing laboratory under controlled temperature conditions. Full-thickness placental tissue (~1–2 cm³) from the fetal side was collected within a similar time frame, rinsed, cooled, and frozen at −80 °C. Full collection and storage protocols are described elsewhere (Hertz-Picciotto et al., 2018).

For maternal serum, detailed sample preparation methods have been described elsewhere (Parenti et al., 2022). Briefly, samples were thawed on ice and filtered to isolate water-soluble (polar) metabolites using Amicon Ultra-0.5 mL 3000 MW centrifugal filters (Millipore, Burlington, MA, USA), which had been pre-washed three times with ultra-pure water to remove glycerol. An internal standard (Chenomx, Edmonton, AB, USA) containing 5.0 mM 3-(trimethylsilyl)-1-propanesulfonic acid-d₆ (DSS-d₆), 0.2 % NaN₃, and 99.8 % D₂O was added to the filtrate, with pH adjusted to 6.8 ± 0.1. Prepared samples were kept at 4 °C and analyzed on the same day using proton nuclear magnetic resonance (¹H-NMR) spectroscopy with a Bruker Avance 600 MHz spectrometer (Bruker, Billerica, MA, USA) using the noesypr1d pulse sequence. Spectra were manually corrected for phase and baseline shifts using Chenomx NMR Processor (v8.1, Chenomx, Edmonton, AB, Canada). Metabolite concentrations were quantified using Chenomx Profiler (v8.1), based on the internal standard (DSS-d₆) and a spectral library. Metabolite concentrations were adjusted for dilution and reported in μmol/L.

For placenta samples, preparation methods have also been described elsewhere (Parenti et al., 2022). In brief, partially thawed samples were aliquoted using a 6 mm biopsy punch, and aliquots were returned to −80 °C until processing. Then, samples were cryoground to a uniform powder using mortars, pestles, and liquid nitrogen. Polar metabolites were extracted from approximately 80 mg of tissue using a two-step CHCl₃: MeOH: H₂O extraction protocol (Hasegawa et al., 2020). The upper aqueous phase, containing polar metabolites, was isolated, measured, and frozen before being dried using a miVac concentrator (Genevac, Warminster, PA, USA). Dried extracts were stored at −80 °C and later reconstituted in 10 mM potassium phosphate buffer containing DSS-d₆. Spectral acquisition and metabolite identification were conducted using the same protocol as for serum samples. Placental metabolite concentrations were adjusted for dilution and reported in nmol/g of tissue.

2.5. Statistical analysis

We first computed descriptive statistics of participant characteristics and maternal serum PFAS concentrations. For values below the LOD (0.1 ng/mL), we imputed a proxy value (LOD/√2) (Hornung and Reed, 1990). Four PFAS compounds (i.e., PFOA, PFOS, PFHxS, and PFNA) were detected in over 80 % of the samples and included in the subsequent analyses. PFDA, though above the 80 % detection threshold, was excluded due to limited variability, as its median concentration was at the LOD and its interquartile range was narrow (0.1 ng/mL) (Table S1). Unadjusted group differences in participant characteristics by ASD diagnostic classification were evaluated using Pearson’s chi-squared test and the Wilcoxon rank sum test.

We quantified 59 serum and 62 placental metabolites. Metabolites were excluded if they were (i) potential contaminants from sample preparation (ethanol, glycerol, and isopropanol for serum; ethanol, isopropanol, and methanol for placenta) and (ii) detected in less than 80 % of the samples (acetaminophen, ethyl-β-D-glucuronide, fructose, inosine, N-phenylacetyl glycine, propylene glycol, trimethylamine N-oxide, and valproate in serum; acetone, ascorbate, propylene glycol, and urea in placenta). After exclusions, 48 serum and 54 placental metabolites were retained for analysis.

To guide covariate selection, we constructed directed acyclic graphs (DAG) a priori to identify potential confounders, mediators, and risk factors for PFAS–maternal metabolome and maternal metabolome–child neurodevelopment relationships, respectively, based on previous literature (Liang et al., 2023; Parenti et al., 2022; Schmidt et al., 2021) (Fig. S1). Considered covariates included: child sex (male, female), child’s birth year (2009–2010, 2011–2013, 2014–2015), gestational age at delivery (continuous, weeks), parity (≤1, >1, missing), maternal pre-pregnancy body mass index (BMI) (<25, ≥25 kg/m²), maternal age (continuous, years), maternal race/ethnicity (White and non-Hispanic, others including Hispanic, Black, Asian persons, and person with other races), maternal education (less than bachelor’s, bachelor’s or higher), home ownership (no, yes, missing), breastfeeding duration (<12, ≥12 months, missing), gestational age at serum sample collection (continuous; weeks), and fasting time since the last meal or snack reported (continuous; hours). Minimal sufficient adjustment sets were identified using Dagitty (Textor et al., 2016; Van der Zander et al., 2014), excluding potential mediators.

For the prenatal PFAS–maternal metabolome analyses, selected covariates included parity, maternal race/ethnicity, education, age, home ownership, and birth year. Birth year was included to consider temporal trends of PFAS exposure. Gestational age at delivery and maternal pre-pregnancy BMI were excluded as potential mediators (Deji et al., 2021; Qi et al., 2020). However, maternal pre-pregnancy BMI was additionally included in a sensitivity analysis to evaluate the robustness of results, given its potential role as both an indicator and a mediator of maternal metabolic status. For maternal metabolome–child neurodevelopment analyses, selected covariates were child sex, maternal race/ethnicity, maternal education, maternal age at delivery, maternal pre-pregnancy BMI, home ownership, and parity. Gestational age at delivery was excluded due to its potential mediating role (Taibl et al., 2023). For the serum metabolome models, we additionally adjusted for gestational age at serum sample collection to consider maternal metabolic variation during the third trimester. In addition, given that the blood samples were collected under non-fasting conditions, we included fasting time to assess the potential impact of fasting duration. However, due to missing information on fasting time (n = 14, 11 %), we conducted multiple imputation by chained equation with 20 imputations based on exposures, outcomes, and covariates (Zhang, 2016). Additionally, for models involving the placental metabolome, we adjusted for delivery method (vaginal delivery, cesarean section) to account for potential metabolic differences associated with delivery type.

Due to the right-skewed distributions, PFAS and metabolite concentrations were log₁₀-transformed prior to analysis. Associations between serum PFAS concentrations and metabolite levels were examined using multiple linear regression, adjusting for selected covariates. We reported regression coefficients (β) and 95 % confidence intervals (CI), and used the Benjamini-Hochberg false discovery rate (FDR) procedure to adjust for multiple comparisons (Benjamini and Hochberg, 1995). To evaluate the joint effect of PFAS mixtures, we applied quantile-based g-computation (QGC), estimating the overall effect of increasing PFAS exposure by one quantile.

Principal component analysis (PCA) was used to reduce dimensionality and identify dominant patterns of correlated metabolites in each matrix. PCs explaining approximately 50 % of cumulative variance were selected, and their scores were used in regression models to examine associations with PFAS exposure and child neurodevelopment. To examine associations between PFAS-associated serum/placental metabolites and child neurodevelopment, we conducted one-way analysis of covariance (ANCOVA), restricting the analysis to metabolites significantly associated with PFAS. Multinominal logistic regression was also used to estimate the relative risk ratios (RRRs) of ASD and non-TD compared to TD, per unit change in PC scores.

To explore potential metabolic pathways linking maternal PFAS exposure, maternal metabolism, and placental metabolism, we constructed an integrated correlation-based network comprising PFAS, serum metabolites, and placental metabolites. A total of 103 mothers with both serum and placental metabolome data were included in the analysis. Pairwise Pearson correlations were calculated for: (i) PFAS-serum metabolites, (ii) PFAS-placental metabolites, and (iii) serum-placental metabolites. When computing partial correlations, PFAS–metabolite partial correlations were adjusted using the same covariates from linear models, while serum–placenta partial correlations were adjusted for birth year, gestational age at blood sample collection, fasting time, and gestational age at placental sample collection (at delivery), following our previous study (Parenti et al., 2024a). Only correlations with absolute values |r| ≥ 0.25 and p-value <0.05 were retained in the network. FDR correction was performed to adjust for multiple comparisons using the Benjamini-Hochberg procedure.

All statistical analyses were performed suing R version 4.4.1 (R Core Team, https://www.R-project.org/). We used “mice” and “qgcomp” R packages for multiple imputation and QGC, respectively. Statistical significance was set at α = 0.05, except for non-parametric tests and FDR-corrected results, for which α = 0.10 was used.

3. Results

3.1. Participant characteristics

A total of 179 mother-child pairs were included in this study (Fig. S2), consisting of 100 TD (56 %), 52 ASD (29 %), and 27 non-TD (15 %) children (Table 1). Among these, 57 % of children were male 94 % of children were born after 37 weeks of gestation, and 61 % of children were born vaginally. Children who were diagnosed with ASD were more likely to be male (69 %) and born in later years compared to TD children. Fewer children with ASD had mothers who owned their homes (42 % vs. 61 %), and mothers of children with ASD or non-TD were more likely to have less than a bachelor’s degree (63 % in both groups vs. 46 % in TD). For serum metabolomics, data were available for 129 mother-child pairs: 76 TD (59 %), 34 ASD (26 %), and 19 non-TD (15 %) (Table S1, Fig. S3). Placental metabolomics data were available for 152 pairs: 84 TD (55 %), 47 ASD (31 %), and 21 non-TD (14 %). Participant characteristics were generally consistent across child diagnostic groups in the serum or placental metabolomics subsamples (Table S2).

Table 1.

Participant characteristics of the 179 mother-child pairs with maternal serum and/or placental metabolomic data (n [%] or Mean ± SD).

Characteristic	All (n = 179)	TD (n = 100)	ASD (n = 52)	Non-TD (n = 27)	TD vs. ASDa	TD vs. Non-TDa
Child sex					0.080	1.000
Male	103 (57.5)	53 (53.0)	36 (69.2)	14 (51.9)
Female	76 (42.5)	47 (47.0)	16 (30.8)	13 (48.1)
Birth year					0.087	0.605
2009–2010	67 (37.4)	40 (40.0)	18 (34.6)	9 (33.3)
2011–2013	67 (37.4)	34 (34.0)	21 (40.4)	12 (44.4)
2014–2015	45 (25.1)	26 (26.0)	13 (25.0)	6 (22.2)
Parity					0.102	0.150
≤1	75 (41.9)	45 (45.0)	19 (36.5)	11 (40.7)
>1	101 (56.4)	55 (55.0)	31 (59.6)	15 (55.6)
Missing	3 (1.7)	–	2 (3.8)	1 (3.7)
Maternal pre-pregnancy BMI					0.124	0.449
<25 kg/m2	88 (49.2)	55 (55.0)	21 (40.4)	12 (44.4)
≥25 kg/m2	91 (50.8)	45 (45.0)	31 (59.6)	15 (55.6)
Maternal race/ethnicity					0.935	0.233
White, non-Hispanic	95 (53.1)	56 (56.0)	28 (53.8)	11 (40.7)
Othersb	84 (46.9)	44 (44.0)	24 (46.2)	16 (59.3)
Maternal education					0.061	0.178
Less than bachelor’s	96 (53.6)	46 (46.0)	33 (63.5)	17 (63.0)
Bachelor’s or higher	83 (46.4)	54 (54.0)	19 (36.5)	10 (37.0)
Home ownership					0.021	0.062
No	81 (45.3)	39 (39.0)	28 (53.8)	14 (51.9)
Yes	95 (53.1)	61 (61.0)	22 (42.3)	12 (44.4)
Missing	3 (1.7)	–	2 (3.8)	1 (3.7)
Breastfeeding duration					0.917	0.513
<12 months	90 (50.3)	49 (49.0)	27 (51.9)	14 (51.9)
≥12 months	71 (39.7)	40 (40.0)	19 (36.5)	12 (44.4)
Missing	18 (10.1)	11 (11.0)	6 (11.5)	1 (3.7)
Delivery method c					0.090	0.620
Vaginal delivery	93 (61.2)	46 (54.8)	34 (72.3)	13 (61.9)
Cesarean section	56 (36.8)	35 (41.7)	13 (27.7)	8 (38.1)
Missing	3 (2.0)	3 (3.6)	–	–
Maternal age (years)	34.2 ± 5.1	34.6 ± 5.0	34.0 ± 5.3	33.3 ± 5.0	0.450	0.289
Gestational age at delivery (weeks)	39.1 ± 1.3	38.9 ± 1.4	39.4 ± 1.1	39.2 ± 1.0	0.153	0.754
Gestational age at serum sample collection (weeks) c	34.0 ± 2.8	33.9 ± 2.8	33.9 ± 2.9	34.7 ± 2.6	0.961	0.240
Fasting time (minutes) c	98.4 ± 69.5	101 ± 65.2	95.5 ± 67.4	98.1 ± 72.7	0.700	0.952

^aP-value from the Pearson’s chi-squared test for categorical variables and the Wilcoxon rank sum test for continuous variables. Bold value represents p-value <0.10.

^bIncludes Hispanic persons, Black persons, Asian persons, and person with other races.

^cVariables relevant to placental sample collection (n = 152).

^dVariables relevant to serum sample collection (n = 129). n = 14 (11 %) were missing for fasting time.

-: n = 0 (0 %)

Abbreviation: autism spectrum disorder (ASD), body mass index (BMI), non-typical development (non-TD), standard deviation (SD), typical development (TD).

3.2. Maternal serum PFAS concentrations

Detection frequencies of PFOA, PFOS, PFHxS, PFNA, and PFDA exceeded 80 %, while PFUnDA, PFDoDA, Me-PFOSAA, and Et-PFOSAA were detected in 4.1 %–60.3 % of the samples (Table S1). Among all PFAS measured, PFOS had the highest median (3.0 ng/mL), followed by PFOA (0.90 ng/mL), PFNA (0.50 ng/mL), and PFHxS (0.40 ng/mL). Of the 267 maternal serum samples analyzed, 189 were from mothers with serum metabolomics data and 227 from those with placental metabolomics data. Detection frequencies and concentration patterns were consistent across both groups (Table S1).

3.3. Associations between prenatal PFAS exposures and maternal serum and placental metabolomic profiles

Maternal serum PFAS concentrations were weakly correlated with both serum and placental metabolites (Fig. 1). Among serum metabolites, Pearson correlation coefficients (r) ranged from −0.17 (between PFOS and betaine) to 0.25 (between PFOS and 2-hydroxybutyrate). For placental metabolites, correlations ranged from −0.18 (between PFNA and creatine) to 0.29 (between PFOA and 4-hydroxybutyrate). None of these correlations remained statistically significant after FDR correction.

Fig. 1.

Pearson correlations between log₁₀-transformed concentrations of PFAS and serum (n = 129) and placental metabolites (n = 152). Positive correlations (red) indicate higher metabolite concentrations with increasing PFAS concentrations, while negative correlations (blue) indicate lower metabolite concentrations with increasing PFAS concentrations. None of the correlations remained statistically significant after FDR correction.

Overall, higher PFAS concentrations were associated with altered metabolite levels in serum or placental tissue (Table 2, Fig. S4). Some metabolites were associated with multiple PFAS compounds. For example, PFOA, PFOS, PFNA, and the PFAS mixture were all positively associated with serum 2-hydroxybutyrate. PFOA, PFOS, and PFAS mixture were associated with higher serum 3-hydroxybutyrate; PFOS, PFHxS and PFAS mixture with increased serum 3-methyl-2-oxo-butanoic acid; and PFOS and PFAS mixture with elevated serum 2-oxoglutarate and citrate. However, most associations did not remain statistically significant after FDR correction. Notably, only the association between PFOS and serum 2-hydroxybutyrate remained significant (β = 0.21, 95 % CI: [0.08, 0.34], q-value <0.10).

Table 2.

Associations between prenatal maternal PFAS and individual maternal serum (n = 129) and placental metabolites (n = 152).

Matrix	PFAS	Metabolites	β (95 % CI)	p	q
Serum	PFOA	2-Hydroxybutyrate	0.16 (0.02, 0.29)	0.028	0.420
		3-Hydroxybutyrate	0.33 (0.10, 0.57)	0.006	0.288
		Choline	−0.08 (−0.16, −0.003)	0.043	0.420
		Formate	−0.07 (−0.14, −0.002)	0.043	0.420
		Tyrosine	−0.09 (−0.17, 0.001)	0.048	0.420
	PFOS	2-Hydroxybutyrate	0.21 (0.08, 0.34)	0.002	0.096
		2-Oxoglutarate	0.09 (0.01, 0.17)	0.025	0.317
		3-Hydroxybutyrate	0.25 (0.02, 0.48)	0.033	0.317
		3-Methyl-2-oxo-butanoic acid	0.09 (0.02, 0.17)	0.012	0.288
		Betaine	−0.07 (−0.14, −0.01)	0.029	0.317
		Citrate	0.08 (0.001, 0.15)	0.047	0.376
	PFHxS	3-Methyl-2-oxo-butanoic acid	0.07 (0.002, 0.13)	0.042	0.874
	PFNA	2-Hydroxybutyrate	0.16 (0.001, 0.32)	0.049	0.679
		N, N-Dimethylglycine	0.07 (0.001, 0.15)	0.048	0.679
	Mixture	2-Hydroxybutyrate	0.05 (0.01, 0.09)	0.010	0.264
		2-Oxoglutarate	0.02 (0.002, 0.05)	0.032	0.307
		3-Hydroxybutyrate	0.07 (0.01, 0.13)	0.022	0.300
		3-Hydroxyisobutyrate	0.03 (0.002, 0.06)	0.039	0.312
		3-Methyl-2-oxo-butanoic acid	0.03 (0.006, 0.05)	0.011	0.264
		Citrate	0.02 (0.003, 0.04)	0.025	0.300
Placenta	PFOA	Betaine	0.15 (0.03, 0.26)	0.013	0.707
	PFOS	myo-Inositol	−0.10 (−0.17, −0.02)	0.012	0.652
	PFHxS	Fumarate	0.17 (0.04, 0.31)	0.011	0.575
	PFNA	N-Acetylneuraminate	−0.19 (−0.34, −0.03)	0.018	0.961
	Mixture	–

Note: Models were adjusted for parity, maternal race/ethnicity, maternal education, maternal age, home ownership, and birth year. Serum models were additionally adjusted for gestational age at serum sample collection and fasting time. Placental models were additionally adjusted for delivery method. The prenatal maternal PFAS and individual maternal serum and placental metabolites were log₁₀-transformed prior to analysis. Hence, regression coefficients approximately describe how much (in relative terms) the outcome mean changes as the exposure changes. Only results with p-value <0.05 are shown.

^aBold value indicates statistical significance at FDR corrected p-value (q) < 0.10.

-: No significant associations observed.

When considering metabolites as a mixture using PCA, four principal components (PCs) were identified for serum metabolites, collectively explaining approximately 50 % of the total variance. For placenta metabolites, the first two PCs explained a similar proportion of the variance (Table S3–S4, Fig. S5). Higher concentrations of PFOA, PFOS, and the PFAS mixture were associated with lower scores on serum PC-2 (PFOA: β = −2.18, 95 % CI: [−3.86, −0.49], PFOS: β = −2.59, 95 % CI: [−4.19, −0.99], PFAS mixture: β = −0.47, 95 % CI: [−0.92, −0.03]) (Fig. 2, Table S5). Serum PC-2 was characterized by relatively strong negative loadings for metabolites such as 2-hydroxybutyrate (−0.31), 3-hydroxybutyrate (−0.28), 2-oxoisocaproate (−0.27), 3-methyl-2-oxo-butanoic acid (−0.26), O-acetylcarnitine (−0.24), acetoacetate (−0.23), and 3-hydroxyisovalerate (−0.22), and positive loadings for glycine (0.22), aspartate (0.21), and myo-inositol (0.20) (Table S3), several of which were also individually associated with maternal PFOS exposure (Table 2). No significant associations were observed between PFAS concentrations and placental PCs.

Fig. 2.

Associations between prenatal maternal PFAS and maternal serum (n = 129) and placental metabolome (n = 152). Symbols represent estimates (β) and error bars represent 95 % confidence intervals (95 % CI). Asterisks indicate statistical significance at p-value <0.05. Models were adjusted for parity, maternal race/ethnicity, maternal education, maternal age, home ownership, and birth year. Serum models were additionally adjusted for gestational age at serum sample collection and fasting time. Placental models were additionally adjusted for delivery method.

In sensitivity analyses additionally adjusting for maternal pre-pregnancy BMI, the association between PFAS exposure and 2-hydroxybutyrate was no longer significant after FDR correction. However, inverse associations between PFOA and PFOS and serum PC-2 scores remained robust and statistically significant (PFOA: β = −1.81, 95 % CI: [−3.51, −0.11], PFOS: β = −2.30, 95 % CI: [−3.91, −0.70]) (Table S6–S7).

3.4. Associations between maternal metabolomic profiles and child neurodevelopmental outcomes

In one-way ANCOVA restricted to metabolites associated with maternal PFAS concentrations, no significant differences in metabolite levels were observed across child diagnostic groups (Table S8). A lower level of 3-hydroxybutyrate was observed in the non-TD group compared to other groups, but this association did not remain significant after FDR correction. When considering metabolites as a mixture using PCA, serum PC-3 was associated with increased risk of non-TD compared to TD (RRR = 1.45, 95 % CI: [1.02, 2.07]) (Table S9). Serum PC-3 was characterized by relatively strong negative loadings for metabolites such as 2-oxoglutarate (−0.30), lactate (−0.25), 2-hydroxyisobutyrate (−0.24), glutamate (−0.24), aspartate (−0.23), pyruvate (−0.22), and positive loadings for metabolites such as methionine (0.24), tyrosine (0.22), and asparagine (0.20) (Table S3).

3.5. Network-based insights into PFAS-associated maternal and placental metabolic changes

Based on the correlation network constructed in this study (Fig. 3), maternal serum PFAS concentrations were associated with several metabolites, including serum 2-hydroxybutyrate, 3-methyl-2-oxo-butanoic acid, 2-oxoisocaproate, choline, placental acetate, and N-acetylneuraminate overlapping with some of the associations observed in the linear regression analyses (Table 2). Some of these metabolites were further interconnected with other serum or placental metabolites within the network, suggesting that prenatal PFAS exposure could influence a broader web of metabolic changes across maternal and placental systems. For example, PFOS and PFHxS were positively correlated with serum 2-hydroxybutyrate (r = 0.31 for PFOS and r = 0.32 for PFHxS), which was in turn negatively correlated with several placental amino acids, including asparagine, histidine, and proline (r ranged from −0.36 to −0.28). However, only the correlations between PFOS and PFHxS and serum 2-hydroxybutyrate, as well as those between PFHxS and serum 2-oxoisocaproate and 3-methyl-2-oxobutanoic acid, remained significant after FDR correction (q-value <0.10). No correlations between PFAS and placental metabolites, nor between serum and placental metabolites, remained significant after FDR correction.

Fig. 3.

Integrated correlation-based network structure of maternal PFAS exposure, serum, and placental metabolites. Edges represent pairwise Pearson correlations adjusted for covariates. For PFAS-serum metabolites and PFAS-placental metabolites, models were adjusted for parity, maternal race/ethnicity, maternal education, maternal age, home ownership, and birth year. Gestational age at serum sample collection and fasting time were additionally adjusted to PFAS-serum metabolite correlation. Delivery method was additionally adjusted to PFAS-placental metabolite correlation. Correlations between serum and placental metabolites were adjusted for birth year, gestational age at serum sample collection, fasting time, and gestational age at delivery. Solid lines indicate correlations with |r| ≥ 0.25 and p < 0.05, with line thickness corresponds to the strength of the correlation. Line colors represent statistical significance after FDR correction: significant (FDR-corrected p < 0.10; red) and non-significant association (gray). Node colors distinguish PFAS (orange), serum (light blue), and placenta (green).

4. Discussion

In the present study, we examined associations between prenatal PFAS exposure and maternal metabolomic profiles using third-trimester serum and placental tissue collected at delivery. We further examined the relationships between these metabolomic profiles and child neurodevelopmental outcomes, including ASD, TD, and non-TD, to explore potential metabolic pathways linking prenatal PFAS exposure to early neurodevelopment. Our findings showed that prenatal exposure to PFOS was positively associated with serum 2-hydroxybutyrate. PCA further showed that exposure to PFOA, PFOS, and the PFAS mixture was associated with serum PC-2, a component characterized by high loadings of metabolites involved in branched-chain amino acid (BCAA) catabolism (i.e., 2-oxoisocaproate, 3-methyl-2-oxo-butanoic acid, 3-hydroxyisovalerate) and lipid catabolism (i.e., 3-hydroxybutyrate, acetoacetate, and O-acetylcarnitine). Additionally, serum PC-3, a component characterized by high loadings of metabolites related to potential mitochondrial dysfunction and altered energy metabolism (i.e., 2-oxoglutarate, lactate, glutamate, aspartate, and pyruvate), was significantly associated with increased risk of non-TD, defined as lower cognitive function without meeting criteria for ASD based on our child classification algorithm. However, these findings did not provide clear evidence of a linkage between prenatal PFAS exposure and neurodevelopment in offspring. To further explore the interconnected metabolic effects of PFAS exposure, we conducted an integrated correlation-based network analysis. We observed that PFAS exposure was associated with changes in some maternal metabolites and preliminary patterns of coordinated metabolic alterations, which further connected with other serum or placental metabolites within the network. These findings suggest that prenatal PFAS exposure could influence a broader web of metabolic disruption. However, the results should be interpreted with caution because none of these associations remained significant after FDR correction. We observed no significant associations between maternal serum and placental metabolites, consistent with our previous MARBLES study (Parenti et al., 2024a).

Among the quantified maternal serum and placental metabolites, serum 2-hydroxybutyrate (also known as 2-hydroxybutyric acid or α-hydroxybutyrate) was the metabolite most consistently associated with prenatal PFAS exposure. This metabolite is a downstream product of α-ketobutyrate, which is generated through the synthesis of glutathione via the cysteine synthesis pathway or through the catabolism of amino acid such as threonine and methionine (Adams, 2011; Sousa et al., 2021). Under conditions of metabolic stress, such as oxidative stress or xenobiotic detoxification in the liver, hepatic glutathione synthesis is upregulated, resulting in elevated production of α-ketobutyrate and 2-hydroxybutyrate (Gall et al., 2010; Sousa et al., 2021). Notably, 2-hydroxybutyrate has been identified as an early biomarker of insulin resistance and impaired glucose regulation, reflecting increased lipid oxidation and oxidative stress (Gall et al., 2010; Li et al., 2009; Sousa et al., 2021; Trico et al., 2017). During pregnancy, elevated levels of 2-hydroxybutyrate have also been implicated in early- and late-onset preeclampsia (Bahado-Singh et al., 2015; Bahado-Singh et al., 2017). Thus, our findings suggests that prenatal PFAS exposure might contribute to endocrine disruption through alterations in energy metabolism and oxidative stress pathways. One potential mechanism involves the interference of PFAS with peroxisome proliferator-activated receptors (PPARs), which are nuclear receptors crucial for regulating lipid metabolism, glucose homeostasis, and peroxisomal activity (Bonato et al., 2020; Han et al., 2017; Szilagyi et al., 2020; Takacs and Abbott, 2007). Disruption of PPAR signaling may impair maternal metabolic regulation during pregnancy, contributing to the changes we observed in serum and placental metabolites. Additionally, PFAS-induced oxidative stress may damage cellular components, including receptors involved in hormone synthesis and metabolic regulation (Bonato et al., 2020). These mechanistic pathways provide a biologically plausible explanation for the metabolic perturbations associated with PFAS exposure in our study. Our finding is consistent with previous epidemiological studies reporting associations between prenatal PFAS exposure and adverse metabolic outcomes, including impaired glucose tolerance and gestational diabetes mellitus (Jensen et al., 2018; Matilla-Santander et al., 2017; Ren et al., 2020; Yu et al., 2021; Zang et al., 2023). Moreover, maternal metabolic disturbance has been linked to poorer neurodevelopmental outcomes in offspring (Cai et al., 2016; Chen et al., 2023; Chen et al., 2021b; Krakowiak et al., 2012; Rodolaki et al., 2023; Saros et al., 2023).

In addition to the findings from multiple linear regression, our PCA results showed that PFAS exposure was associated with scores of serum PC-2, characterized by high contributions from 2-hydroxybutyrate and other metabolites involved in BCAA catabolism and lipid metabolism (i.e., 3-hydroxybutyrate, 2-oxoisocaproate, and 3-methyl-2-oxo-butanoic acid). The findings of several previous studies from pregnancy cohorts were similar to ours. For example, three U.S. pregnancy cohort studies using non-targeted metabolomics with high-resolution mass spectrometry reported associations between prenatal PFAS exposure and energy and amino acid metabolism. A study from California found that prenatal PFOS exposure was associated with metabolic features enriched in carnitine shuttle, lysine metabolism, and BCAA pathways using maternal serum collected across pregnancy and into the early postpartum period (Hu et al., 2019). Similarly, a study from Atlanta reported associations between PFAS mixtures and first-trimester maternal serum levels of amino acids (i.e., isoleucine, ketoleucine, leucine, and valine) and fatty acids (i.e, carnitine), highlighting enrichment of energy metabolism and BCAA metabolism (Liang et al., 2023). In addition, a Michigan study found that first-trimester PFOS exposure was associated with increased isovalerylcarnitine, a downstream product of L-leucine metabolism (Rabotnick et al., 2024). A Spanish cohort study also found that PFHxS exposure was associated with maternal acetone and succinate, while PFOA exposure was associated with maternal alanine, glycine, and 3-hydroxybutyrate/3-aminoisobutyrate levels in third-trimester maternal urine samples (Maitre et al., 2018). Taken together, these findings suggest that PFAS exposure might disrupt maternal energy metabolism and BCAA catabolism during pregnancy, with potential implications for maternal metabolic health. However, in the present study, neither 2-hydroxybutyrate nor the metabolic profile presented by serum PC-2 was associated with child neurodevelopmental outcomes. Lower serum 3-hydroxybutyrate levels were observed in mothers with non-TD children, consistent with our previous study (Parenti et al., 2024a), but the association did not survive after FDR correction in this present study. Our findings might also suggest potential implications for fetal metabolic health. However, we did not address short- or long-term child metabolic development in the present study, and future studies are needed.

When examining the associations between metabolic disruptions and child neurodevelopment, we observed that higher maternal serum PC-3 scores were associated with increased risk of non-TD. Serum PC-3 was characterized by a pattern of metabolites involved in mitochondrial energy production, including 2-oxoglutarate, lactate, glutamate, aspartate, and pyruvate. It also showed positive loadings of metabolites such as methionine, tyrosine, and asparagine, which are related to oxidative stress, methionine cycle, and neurotransmitter synthesis. These metabolic alterations could reflect mitochondrial dysfunction and/or impaired energy metabolism, given that mitochondria play a key role in energy production and cellular respiration through glycolysis, tricarboxylic acid (TCA) cycle, and oxidative phosphorylation. Similarly, a previous study reported elevated lactate levels in maternal plasma samples from mothers of children diagnosed with ASD (Parenti et al., 2024b). While this trend differs from our findings (i.e., negative loadings for lactate in PC-3), it might be due to differences in the timing of sample collection (prenatal vs. postnatal). However, both studies support the hypothesis that mitochondrial dysfunction might contribute to adverse neurodevelopmental outcomes in children. The MARBLES participants have a higher genetic susceptibility to neurodevelopmental conditions as the study recruited mothers who already have a child diagnosed with ASD. Given that mitochondrial DNA is inherited matrilineally, disruptions in maternal mitochondrial metabolism might influence a child’s susceptibility to adverse neurodevelopmental outcomes such as ASD. Supporting this, a case-control study found that children with ASD were more likely to have mitochondrial DNA over-replication and deletion compared to TD children (Giulivi et al., 2010). Mitochondrial energy metabolism plays a crucial role in early brain development (Oyarzábal et al., 2021) and several case-control studies have reported altered levels of mitochondrial biomarkers, including lactate, pyruvate, and carnitine, in children with ASD (Giulivi et al., 2010; Oliveira et al., 2005; Orozco et al., 2019; Rossignol and Frye, 2012). However, in our sub-study, we did not observe metabolic evidence linking maternal PFAS exposure and child neurodevelopmental outcomes, although a previous study using the larger MARBLES cohort reported associations between maternal PFAS exposure and ASD (Oh et al., 2021a). One possible explanation is that PFAS may influence neurodevelopment indirectly through disruptions in maternal or placental metabolism, as suggested by the observed associations between PFAS and specific metabolomic profiles, and between serum PC-3 and increased risk of non-typical development. This is consistent with a mediating role for metabolic pathways. In addition, the MARBLES cohort is enriched for ASD due to familial history, suggesting a genetic predisposition that may interact with environmental exposures. PFAS may act as a modifying factor in this genetically susceptible population, but only manifest effects above certain exposure thresholds or in combination with other stressors. Furthermore, our sample size may have limited the power to detect modest associations, particularly given the heterogeneity of neurodevelopmental outcomes. Lastly, the targeted metabolomic scope of our analysis may not have captured all relevant biochemical disruptions. Future studies with larger sample sizes, expanded metabolomic coverage, and formal mediation analysis will be critical to elucidating these complex relationships.

Overall, serum PC-2 and PC-3 represented distinct but partially overlapping metabolic patterns related to energy and amino acid metabolism. PC-2 primarily reflected BCAA catabolism, suggesting disruptions in maternal energy balance during pregnancy, whereas PC-3 was more directly related to mitochondrial dysfunction. Despite these differences, both components included metabolites related to mitochondrial-related metabolism. For example, 3-hydroxybutyrate and O-acetylcarnitine (in PC-2) are involved in fatty acid oxidation, and aspartate (in both PC-2 and PC-3) is involved in malate-aspartate shuttle. These overlapping features suggest that prenatal PFAS exposure may be associated with disruptions in mitochondrial-related metabolism, potentially affecting child neurodevelopment. However, further studies are needed to confirm these associations and elucidate underlying mechanisms.

We observed relatively weaker associations between prenatal PFAS exposure and placental metabolomic profiles compared to those seen in maternal serum. Several factors may contribute to this discrepancy. First, although PFAS can accumulate in the placenta, their concentrations are typically lower in placental tissue than in maternal blood (Chen et al., 2017; Mamsen et al., 2019), which may attenuate PFAS-induced metabolic perturbations detectable in the placenta. Second, the maternal serum metabolome reflects systemic maternal physiology, including nutrient status and overall metabolic health, whereas the placental metabolome reflects distinct physiological functions, such as nutrient transport, hormone production, and fetal-maternal signaling (Costa, 2016; Gude et al., 2004). These tissue-specific roles may result in different sensitivities to PFAS-related metabolic disruption. Finally, the targeted set of polar metabolites assessed in this study may have limited our ability to capture placenta-specific biochemical alterations. Future studies utilizing broader or non-targeted metabolomic approaches may provide more comprehensive insights into PFAS-associated changes in placental metabolism.

This study has several notable strengths. To our knowledge, it is the first study to investigate the integrated associations among prenatal PFAS exposure, maternal metabolomic profiles, and child neurodevelopmental outcomes. A key strength of the study design is the inclusion of metabolomic data from both maternal serum and placental tissue, providing a more comprehensive view of metabolic processes at the maternal–fetal interface. Metabolites were quantified using ¹H-NMR spectroscopy, a robust and reproducible analytical platform that enables absolute quantification of identified metabolites. The quantitative nature of NMR-based data also facilitates multivariate data analyses, including correlation-based network analysis (Heinzmann et al., 2022). In contrast to prior studies that relied primarily upon non-targeted HRM and focused exclusively on maternal blood, our study offers additional insights by incorporating the placenta, a critical yet often overlooked tissue in prenatal environmental health research.

However, several limitations should be considered. First, the generalizability of our findings may be limited, as the study population was drawn from an elevated-likelihood cohort—children with an older sibling diagnosed with ASD—who may have a higher genetic susceptibility to neurodevelopmental conditions (Hertz-Picciotto et al., 2018). Second, maternal serum samples were collected under non-fasting conditions, which may have influenced concentrations of certain diet-sensitive metabolites. To mitigate this limitation, we adjusted for fasting time (i.e., hours since last meal or snack) in all regression models involving maternal serum metabolomic data. However, detailed dietary intake data were not available, and residual confounding due to recent food consumption cannot be ruled out. As a result, observed associations involving metabolites such as 2-hydroxybutyrate and principal components PC-2 and PC-3 may be influenced by unmeasured dietary factors, which should be considered when interpreting our findings. Third, the relatively small sample size may have limited our statistical power to detect subtle associations, particularly after correction for multiple comparisons. For instance, among the 129 children whose mothers had serum metabolome data, only 34 were diagnosed with ASD and 19 were classified as non-TD. Consequently, most associations did not remain significant after FDR correction. These limitations constrain the interpretation and generalizability of our findings beyond the high-risk populations. Future studies with larger, population-based samples and greater ethnic and socioeconomic diversity are needed to validate our results and better characterize the broader implications of prenatal PFAS exposure on maternal metabolism and child neurodevelopment. In addition, we did not explore trimester-specific associations between PFAS exposure and maternal or placental metabolomic profiles due to limited statistical power. Although PFAS generally have long biological half-lives and showed relatively stable concentrations across trimesters in the MARBLES cohort (Choi et al., 2024; Oh et al., 2022), physiological changes during pregnancy, such as plasma volume expansion and altered placental permeability, may still influence PFAS distribution and bioavailability (Chen et al., 2021a). Future studies with larger sample sizes are warranted to investigate time-specific effects and better identify critical windows of fetal vulnerability.

5. Conclusions

This study provides evidence that prenatal PFAS exposure may influence maternal metabolic profiles, particularly through alterations in third-trimester serum metabolites such as 2-hydroxybutyrate and PC-2, which are primarily associated with pathways related to energy and BCAA metabolism. A maternal serum metabolic pattern suggestive of mitochondrial dysfunction (PC-3) was associated with increased risk of non-TD in children. However, we did not observe direct associations between prenatal PFAS exposure and child neurodevelopmental outcomes. While this study did not identify the coordinated metabolic alterations between maternal serum and placental tissue in this study, prior findings from the MARBLES cohort suggest metabolic continuity along the maternal-placental-fetal axis (Parenti et al., 2024a). Given the use of a targeted metabolomics approach, future research employing non-targeted approaches with broader metabolite coverage may better capture the complexity of these interactions. Larger and more diverse populations will be essential to further elucidate how environmental exposures and metabolic disruptions influence fetal development and neurodevelopmental risk.

Supplementary Material

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.envpol.2025.126811.

Data availability

Data will be made available on request.