Identifying critical windows of prenatal phenol, paraben, and pesticide exposure and child neurodevelopmental: Findings from a prospective cohort study

Abstract

Background:

This study aimed to investigate how exposure to a mixture of endocrine disrupting chemicals (EDCs) during two points in pregnancy affects early childhood neurodevelopment.

Methods:

We analyzed publicly-available data from a high-risk cohort of mothers and their children (2007-2014) that measured six EDCs including methyl-, ethyl- and propyl parabens (MEPB, ETPB, PRPB), Bisphenol-A (BPA), 3,5,6-trichloro-2-pyridinol (TCPy) ,3-phenoxybenzoic acid (3-PBA) in prenatal urine samples during the second and third trimesters. Neurodevelopmental scores were assessed using Mullen Scales of Early Learning (MSEL) at age 3. We used mean field variational Bayes for lagged kernel machine regression (MFVB-LKMR) to investigate the association between trimester-specific co-exposure to the six EDCs and MSEL scores at age 3, stratified by sex.

Results:

The analysis included 130 children. For females, the relationship between BPA and 3PBA with MSEL score varied between the two trimesters. In the second trimester, effect estimates for BPA were null but inversely correlated with MSEL score in the third trimester. 3PBA had a negative relationship with MSEL in the second trimester and positive correlation in the third trimester. For males, effect estimates for all EDCs were in opposing directions across trimesters. MFVB-LKMR analysis identified significant two-way interaction between EDCs for MSEL scores in both trimesters. For example, in females, the MSEL scores associated with increased exposure to TCPy were 1.75 units (95%credible interval −0.04, −3.47) lower in the 2^nd trimester and 4.61 (95%CI −3.39, −5.84) lower in the third trimester when PRPB was fixed at the 75^th percentile compared to when PRPB was fixed at the 25^th percentile.

Conclusion:

Our study provides evidence that timing of EDC exposure within the prenatal may impact neurodevelopmental outcomes in children. More of these varying effects were identified among females. Future research is needed to explore EDC mixtures and the timing of exposure during pregnancy to enhance our understanding of how these chemicals impact child health.

Graphical Abstract

1. Introduction

In the United States, approximately 18% of children are affected by neurodevelopmental disorders, including attention deficit hyperactivity disorder, autism spectrum disorder (ASD), and intellectual disabilities [1]. Exposure to environmental chemicals, including endocrine-disrupting chemicals (EDCs), during critical windows of human development may increase the risk of neurodevelopmental disorders in childhood [2, 3]. The process of neural development is intricate, spanning from the embryonic period through adolescence, with the prenatal period representing a crucial window of brain development and susceptibility to exposure to EDCs. [4–6]. Due to the sequential and cumulative nature of fetal development, the timing of exposure to EDCs during the prenatal window may result in diverse and adverse impacts on neurodevelopmental outcomes [4–6]. Despite an expanding body of epidemiological evidence supporting the association between prenatal exposure to specific EDCs and adverse neurodevelopmental outcomes in children, there remains a lack of data investigating the correlation between time-varying exposure to diverse EDCs throughout the prenatal period and neurodevelopmental outcomes in children. [2, 3].

EDCs, such as phenols, parabens, and organophosphate (OP) insecticides, are exogenous compounds that have the potential to mimic, alter, or attenuate the action of natural hormones in the body [7, 8]. EDCs can be classified into both natural and synthetic categories, arising from various sources and the widespread use of these compounds has led to widespread exposure in the population [8]. Biomonitoring data from the United States have shown detectable levels of phenols, parabens, and OP insecticides in pregnant women, which of is of major concern because these chemicals can cross the placenta, resulting in fetal exposure [9–16]. Current literature indicates a broad spectrum of potential targets and pathways through which EDC exposure during the prenatal period may influence child neurodevelopment, including disruptions in hormonal signaling, epigenetic modifications, and impacts on neuronal connectivity and synaptogenesis [3, 17, 18].

Phenols and parabens are present in many common household and personal care products. For example, bisphenol-A (BPA) is used in the manufacturing of polycarbonates and epoxy resins to produce items like food cans, plastic bottles, water supply lines, and some dental sealants and composite [19, 20]. Prenatal exposure to BPA can disrupt various hormonal pathways, including thyroid homeostasis and sex hormone signaling [18]. BPA can affect the hypothalamus-pituitary-thyroid axis, impacting thyroid hormone transporters, receptors, and metabolism, which, in turn, is associated with cognitive and psychomotor development impairment and structural brain changes due to thyroid hormone deficiency [21]. Prenatal exposure to BPA can interfere with sex hormone signaling, acting as an estrogen agonist and androgen antagonist [21], potentially leading to neurodevelopmental deterioration during brain development [18]. In addition, previous epidemiologic research has shown that prenatal maternal exposure to BPA can lead to alterations in the white matter microstructure of preschool-aged children [22]. These changes in white matter, particularly in the splenium and inferior longitudinal fasciculus, are linked to shifts in visual processing skills. Moreover, children exposed to prenatal BPA tend to experience deficits in executive functioning along with an increase in both internalizing (anxiety, depression) and externalizing (attention) behaviors [22].

Parabens, esters of p-hydroxybenzoic acid, are widely used as antimicrobial preservatives in pharmaceuticals, food products, and in cosmetic products like moisturizers and hair-care products [23, 24]. There are limited studies examining the impact of paraben exposure during pregnancy on child development, with the existing literature primarily focusing on interference in thyroid function as the driving biological mechanism [25]. Maternal thyroid hormones play an important in child neurodevelopment. There is growing evidence that even mild maternal thyroid hormone insufficiency is linked to a range of neurodevelopmental issues in children [26]. These include lower intelligence quotient (IQ) levels, an increased risk of autism, and a higher incidence of attention deficit hyperactivity disorder (ADHD) [26]. OP insecticides like 3,5,6-trichloro-2-pyridinol (TCPy), and 3-phenoxybenzoic acid (3PBA) are commonly found in agriculture settings and homes [27]. TCPy, the principal metabolite of chlorpyrifos (CPF), has been shown to affect key nuclear transcription factors essential for the replication and differentiation of brain cells, providing a basis for its noncholinergic effects that could potentially result in neurobehavioral abnormalities [28]. Previous data have found that prenatal exposure to CPF, even at low levels, has a significant impact on brain structure in children aged 5.9 to 11.2 years [29]. This exposure led to abnormal morphological changes in the cerebral surface, predominantly enlargements in specific brain regions associated with cognitive and behavioral functions such as attention, social cognition, and emotional control. These findings suggest that these changes may result from CPF’s cytotoxic effects on glial and neuronal cells, mirroring the damage seen in animal models, and also indicate a potential negative impact on general cognition related to CPF exposure [29]. The compound 3-PBA is a metabolite of pyrethroids; animal studies have shown that early exposure to pyrethroids can lead to inadequate brain development due to alterations in vascular development [30], decreased motor activity [31], and adverse effects on the developing nervous system, causing cholinergic dysfunction, which leads to deficits in learning and memory [32].

Evidence supporting the association between prenatal exposure to BPA [33–39], parabens [40–43], and OP [33, 44–46] with different measures of neurodevelopmental outcomes in children is inconclusive. The majority of studies evaluating the relationship between select EDC exposure and neurodevelopmental outcomes used a single-chemical analytical approach based on only a single prenatal measurement of exposure. Evaluating associations solely through a single-chemical and timepoint approach may have limitations in the scientific assessment of EDCs. While the individual biological effects of certain EDCs can be weak, these chemicals may collectively exert an influence on cognitive development through a synergistic mechanism, such as thyroid hormone disruption [47–49]. Moreover, due to the pervasive use of EDCs, it is highly that fetal exposure to multiple hormonally active agents occurs concurrently or sequentially. This underscores the need to consider the time-dependent biological effects of exposure to multiple EDCs. Previous studies have examined the potential combined effects of prenatal co-exposure to parabens and phenols on neurodevelopmental outcomes [40, 50] or the association between trimester-specific exposure to phenols on neurodevelopmental outcomes [39, 41]. There is a lack of studies investigating the association between time-varying co-exposure to phenols, parabens, and OP insecticides during the prenatal period and neurodevelopmental outcomes in children. Addressing the time-varying effects of mixtures is important, as it may provide insights into the underlying biological mechanisms and guide intervention strategies. We can gain a better understanding of their potential impact by determining whether co-exposure to these select EDCs influences neurodevelopment in children based on the timing of exposure within the prenatal window. Most statistical methods either assess the varying effect of a single pollutant across different time points [51, 52] or the joint exposure-response relationships of multiple exposures at a single time point but do not address mixture effects across time. Understanding exposure to real-world environmental mixtures during sensitive windows of development and the associated health effects is challenging given the statistical complexity of analyzing time-varying exposures to multiple environmental chemicals. These statistical challenges include multicollinearity among the exposures both within time points and across time points, and complex exposure–response relationships [53].

The goal of this study is to evaluate whether exposure to a mixture of EDCs during two prenatal time periods differentially affects measures of overall cognitive development and functioning in early childhood. To address this research question, we applied a statistical method developed to address the time-varying effects of mixtures, mean field variational Bayes for lagged kernel machine regression (MFVB-LKMR), to publicly available prospective data with biomarker measures of EDC exposure during pregnancy and child neurodevelopment outcomes at age 3 years. We hypothesized that the sex-specific joint exposure-response relationships of select phenols, parabens, OP insecticides, and measures of overall neurocognitive development would vary across the two prenatal time windows (second and third trimester).

2. Methods

2.1. Study data

To investigate the time-varying effects of EDC mixtures, we used data from the Markers of Autism Risk in Babies – Learning Early Signs (MARBLES) Study, contained in a publicly available dataset obtained through the Human Health Exposure Analysis Resource (HHEAR) Data Repository at the Icahn School of Medicine at Mount Sinai [54]. The MARBLES Study is a prospective observational study initiated in 2006 at UC Davis MIND Institute to study prenatal exposure to environmental chemicals and child neurodevelopmental outcomes [55]. MARBLES is an enriched cohort that enrolled pregnant women that had a previous child with ASD [55, 56]. Details of study design, recruitment, eligibility, and measured exposure and outcome data have been described previously [55]. Briefly, mothers were primarily recruited through the California Department of Developmental Services (DDS), the state agency that coordinates services for individuals with developmental disabilities. The study’s inclusion criteria required that at least one parent had a biological child with ASD and/or the gestating younger child had an older half-sibling. Additionally, the mother had to be 18 years or older, pregnant, able to speak and understand English, and live within 2.5 hours of the Davis/Sacramento area. The study included births up to December 31, 2015, and as of June 30, 2018, 463 pregnant mothers had enrolled in the study. The majority (over 75%) were above 30 years old, including 7.9% who were over 40 years old. The diverse sample comprised 22% Hispanic, 25% non-Hispanic Black, Asian, or multiracial, with 24% born outside the U.S. Mothers were followed through pregnancy and infants from birth to 3 years. Of the children evaluated at 36 months, 24% were diagnosed with ASD, and another 25% showed non-typical, non-ASD development [55].

Complete details of urine sample collection procedures are available elsewhere [57–59]. The data included children with prenatal measures of six environmental chemical exposures (bisphenol A (BPA), ethyl paraben (ETPB), methyl paraben (MEPB), propyl paraben (PRPB), 3,5,6-trichloro-2-pyridinol (TCPy) and 3-phenoxybenzoic acid (3PBA) measured in both the second and third trimester. The biomarker measures in the dataset, adjusted for urinary dilution of metabolites using specific gravity, were assessed by NSF International using an ATAGO handheld digital refractometer model PAL-10S. The concentrations of these analytes were calculated utilizing isotope dilution calibration [60, 61]. Biomarker exposure measurements in the dataset included pooled urine samples (66% of participants), spot urine samples (26%), and twenty-four-hour urine samples (8%) measured in the 2nd and 3rd trimester. Because measures below the limit of detection were included in the data as machine-read values, some observations had negative values. Given the implausible negative values, we replaced the machine-read negative biomarker concentration values with the minimum positive value for the corresponding chemical [62]. We then log-transformed the specific gravity-corrected biomarker concentrations to account for the right skewed distribution of the data.

At approximately 36 months of age children were assessed by a trained psychologist for cognitive development using MSEL [55]. MSEL is a comprehensive, standardized test that assesses early intellectual development and school readiness by evaluating cognitive and motor abilities based on five subscale scores representing developmental ages: gross motor, fine motor, visual reception, expressive language, and receptive language [63]. Scores from the subscales, excluding gross motor, are used to calculate an Early Learning Composite score, which serves as an indicator of overall cognitive development. Raw scores for five domains are transformed into T scores (mean = 50, standard deviation = 10), percentile ranks, age equivalents, developmental stages up to 33 months, and descriptive categories, while the Early Learning Composite is recalculated into a standard score (mean = 100, standard deviation = 15), percentile rank, and descriptive category [63]. The MSEL has shown utility in assessing children diagnosed with ASD [64, 65] and is commonly utilized to assess preschool children for eligibility for early intervention services, with T-scores below 30 indicating a need for such services and scores from 31 to 35 suggesting a child may be at risk [63]. The MSEL has been validated in children with and without autism spectrum disorder and other developmental disabilities, showing convergent validity against established measures of nonverbal and verbal IQ [66]. The data housed within the public repository included MSEL summary score as well as separate measures for fine motor, expressive language, receptive language, and visual reception.

2.2. Statistical analysis

We restricted the study sample to children with available data on the six prenatal urinary biomarker measures (BPA, ETPB, MEPB, PRPB, TCPy, and 3PBA), MSEL and covariate data, leaving a final sample of 130 for this analysis (52 females and 78 males). We compared study characteristics between mother-child pairs included in this analysis versus those not included using Chi-square tests (Supplemental Table S1).

We stratified analyses by the child’s sex, given data showing sex differences in brain maturation during childhood neurodevelopmental outcomes [67]. We selected potential covariates from the available dataset, based on previous literature, and constructed a directed acyclic graph to guide model adjustment. We hypothesized that the selected a priori set of covariates in relation to neurodevelopmental outcomes would be similar for parabens, insecticides, and BPA. These covariates included: mother’s nativity (born in or outside the U.S.), maternal body mass index (BMI) before pregnancy, maternal age at the time of the child’s birth, child’s birth year (continuous), mother’s number of metabolic conditions (gestational diabetes, diabetes type 1 and type 2, pre-eclampsia), maternal self-reported race (non-Hispanic White, non-Hispanic Black/African American, Asian, and Other) as a proxy for markers of differential stressors by race such as racism. We used proxy measures for socioeconomic status, which included the highest education level in the household and household home ownership (owner, non-owner) [68–73]. After identifying the set of covariates and considering the sample size of our final dataset, we then used the mean square error (MSE) approach to determine the minimally sufficient set variables to include in adjusted analyses to balance both bias and precision [74]. The final set of covariates for model adjustments included maternal BMI before pregnancy, maternal age at time of child’s birth, highest education in household, household home ownership, child’s birth year, and mother’s number of metabolic conditions.

We used MFVB-LKMR to investigate the relationship between prenatal time-varying co-exposure to the EDC mixture (BPA, ETPB, MEPB, PRPB, TCPy, and 3PBA) and MSEL measures at age 3 to identify differential effects on neurodevelopment outcomes based on exposures occurring in the second versus third trimester of pregnancy. MFVB-LKMR is a flexible statistical method that estimates the exposure–response relationships in a co-exposure setting across multiple time windows while accounting for complex non-linear and non-additive effects of the mixture at any given exposure window [53, 75]. Details on the MFVB-LKMR methods are available in previously published papers [53, 75]. Briefly, lagged kernel machine regression (LKMR) is a method that combines Bayesian group lasso and fused lasso penalization schemes together with kernel machine methods to investigate time-varying co-exposures in relation to a time-fixed outcome [53]. Bayesian group and fused lasso penalties account for dependencies of correlation among exposure mixtures within a given time window, as well as auto-correlation across time windows. The Bayesian grouped lasso penalty regularizes the exposure–response function at each time window and the fused lasso penalty shrinks the exposure–response functions of neighboring time windows toward each other [75]. LKMR, which uses Markov chain Monte Carlo (MCMC) methods to update model parameters, estimates the non-linear and non-additive effects of exposure mixtures at any given exposure window. Implementing the MCMC algorithm is computational burdensome due to the complexity of the LKMR model, particularly with increased sample sizes or time points studied [53, 75]. MFVB-LKMR is a computationally efficient approximation of LMKR that uses the mean field variational approximation method for Bayesian inference to approximate the LKMR model with reduced computation time [75].

We applied MFVB-LKMR to estimate the main effects of each EDC in the second and third trimester and to investigate two-way interaction between EDCs in each trimester. We standardized the six EDC concentrations and MSEL outcome measures and generated main effect and interaction analyses for components of the MSEL (fine motor, expressive language, receptive language, visual reception) and the final MSEL score. MFVB-LKMR quantifies the main effects of an individual EDC as the difference in the mean outcome predicted when the single EDC is high (fixed at the 75th percentile) versus when that single EDC is low (fixed at the 25th percentile), fixing all other EDCs constant at their median exposure levels and adjusting for covariates. MFVB-LKMR also estimates the 95% posterior credible intervals (CrI), a Bayesian metric analogous to the 95% confidence interval, for the main effect estimates. MFVB-LKMR assesses two-way interaction (EDC₁, EDC₂) within a mixture by estimating the difference between the mean predicted outcome at the 75^th percentile of EDC₁ versus the 25^th percentile of EDC₁ at the 75^th percentile of EDC₂, contrasted with the difference in the mean outcome predicted at the 75^th percentile of EDC₁ versus the 25^th percentile of EDC₁ at the 25^th percentile of EDC₂ (fixing all other EDCs at their median levels and adjusting for covariates) [75, 76]. Of note, MFVB-LKMR allows for interactions within a time point, but does not evaluate interactions across multiple time points [75]. We used a threshold of α = 0.10 for statistical significance to identify evidence of potential two-way interaction. We visually assessed significant two-way interactions for nonlinearity using plots of the cross-section of the estimated exposure–response surface. All statistical analyses were performed using R version 3.6 and adapting publicly-available code provided by Liu et al. [75].

3. Results

3.1. Characteristics of study data

Table 1 details characteristics of the publicly-available study data. Of the 130 mother-child pairs included within our analysis, more than half for the children were males (60%). Overall, the study sample included a higher proportion of mothers that were non-Hispanic white, over 35 years at the time of the child’s delivery, used prenatal vitamins in the first trimester of pregnancy and had no metabolic conditions at the time of pregnancy (e.g., diabetes or pre-eclampsia). The mean pre-pregnancy BMI in the sample was 25.5 kg/m². The majority of parents were homeowners and had a bachelor’s degree. MARBLES mother-child pairs included in this analysis versus those not included did not differ significantly based on the distribution of demographic and outcome factors with the exception of child’s birth year where no child from birth years 2015 to 2017 were included in this study (Supplemental Table S1). Spearman correlations between each EDC by trimester is shown in Supplemental Figure S1, where there were moderate to high correlation between EDCs within the same exposure class (e.g., high correlation between MEPB and PRPB). Table 2 summarizes the distributions of the six EDCs included in the analysis. The geometric mean of the EDC concentrations was slightly higher in the third trimester for four (MEPB, PRPB, TCPy, 3PBA) of the six EDCs. Supplemental Table S2 details the distribution of the neurodevelopmental outcomes overall and by male and females where the median MSEL score was 96 and 106, respectively.

Table 1.

Characteristics of the mother-child pairs included in the analysis of the MARBLES study data downloaded from the HHEAR Data Repository (n=130).

Characteristic	All (N=130)	Males (N=78)	Females (N=52)
Child’s race/ethnicity
White (non-Hispanic)	83 (64%)	46 (59%)	37 (71%)
Black (non-Hispanic)	4 (3%)	3 (4%)	1 (2%)
Asian	11 (8%)	9 (11%)	2 (4%)
Other	32 (25%)	20 (26%)	12 (23%)
Birth year
2007	2 (2%)	2 (3%)	0 (0%)
2008	21 (16%)	12 (15%)	9 (17%)
2009	29 (22%)	17 (22%)	12 (23%)
2010	21 (16%)	14 (18%)	7 (14%)
2011	11 (9%)	8 (10%)	3 (6%)
2012	10 (8%)	4 (5%)	6 (12%)
2013	20 (15%)	13 (17%)	7 (13%)
2014	16 (12%)	8 (10%)	8 (15%)
Maternal pre-pregnancy BMI, mean (SD)	26.5 (6.4)	25.3 (4.8)	28.3 (8.0)
Normal/underweight	65 (50%)	40 (51%)	25 (48%)
Overweight	38 (29%)	27 (35%)	11 (21%)
Obese	27 (21%)	11 (14%)	16 (31%)
Mother’s age at deliver, mean (SD)	34.4 (4.8)	34.3 (4.8)	34.5 (4.9)
≤ 30 years	26 (20%)	17 (22%)	9 (17%)
30 – 35 years	40 (31%)	20 (26%)	20 (38%)
>35 years	64 (49%)	41 (52%)	23 (44%)
Prenatal vitamin use in the first trimester of pregnancy
Yes	108 (83%)	66 (15%)	42 (81%)
No	22 (17%)	12 (84%)	10 (19%)
Maternal metabolic conditions
No metabolic conditions	101 (78%)	62 (79%)	39 (75%)
Any metabolic conditions	29 (22%)	16 (21%)	13 (25%)
Prior parity
0	1 (1%)	0	1 (2%)
1	59 (45%)	35 (45%)	24 (47%)
>1	70 (54%)	43 (55%)	27 (52%)
Homeowner
Yes	82 (63%)	51 (65%)	31 (60%)
No	48 (37%)	27 (35%)	21 (40%)
Maximum parental education
Less than a college degree	20 (15%)	13 (17%)	7 (13%)
Bachelor’s or associate degree	78 (60%)	47 (60%)	31 (60%)
Graduate or professional degree	32 (25%)	18 (23%)	14 (27%)

Abbreviation: BMI, body mass index; SD, standard deviation.

Table 2.

Descriptive statistics of specific-gravity adjusted urinary concentrations of the six chemicals measured during the second and third trimester of pregnancy (n=130).

		Second trimester					Third trimester
	LOD	% >LOD	10th	50th	90th	GM (95% CI)	% >LOD	10th	50th	90th	GM (95% CI)
BPA	0.8	61%	0.29	0.98	3.54	0.93 (0.75, 1.15)	64%	0.27	1.06	2.62	0.80 (0.60, 1.05)
ETPB	0.5	42%	0.01	0.68	18.20	0.68 (0.45, 1.01)	50%	0.01	0.79	20.70	0.63 (0.40, 1.00)
MEPB	0.5	96%	4.67	32.95	288.50	33.38 (24.91, 44.72)	92%	4.13	40.30	304.50	36.62 (26.62, 50.36)
PRPB	1.0	77%	0.46	7.42	62.20	6.26 (4.51, 8.68)	74%	0.22	9.52	57.65	6.88 (4.84, 9.75)
TCPy	0.5	89%	0.06	2.52	7.08	1.88 (1.42, 2.47)	88%	0.12	2.53	11.17	2.29 (1.81, 2.89)
3PBA	0.5	82%	0.21	1.67	4.80	1.44 (1.19, 1.75)	87%	0.24	1.58	6.14	1.50 (1.21, 1.84)

Abbreviations: 3PBA, 3-phenoxybenzoic acid, BPA, bisphenol A, ETPB, ethyl paraben, MEPB, methyl paraben, PRPB, propyl paraben, TCPy, 3,5,6-trichloro-2-pyridinol, LOD, limit of detection; CI, confidence interval; GM, geometric mean.

3.2. Main effects of prenatal concentrations of phenols, parabens, and OP insecticides on MSEL

Our results focus on the relationship between the prenatal EDC mixtures measured in the second and third trimester and MSEL composite scores. Results for the four MSEL subscales are detailed in the Supplementary Materials (Supplemental Tables S3–S5). For females, the strongest relationship was observed between prenatal urinary concentrations of MEPB and PRPB in the second trimester and standardized MSEL scores (Figure 1a). Specifically, mean standardized MSEL scores were 2.17 units higher (95%Crl: 0.09, 4.25) for MEPB levels at the 75^th compared with the 25^th percentiles and mean standardized MSEL scores for PRPB levels at the 75^th compared with the 25^th percentiles were lower by 2.57 units (95%Crl: −4.97, −0.17) while fixing all other EDCs at their median levels (see Supplemental Table S4 for numeric effect estimates). In the third trimester for females, prenatal levels of MEPB and PRPB showed the same direction of effect for MSEL scores, however effect estimates were stronger in magnitude with narrower credible intervals (estimate = 3.26, 95%Crl: 2.03, 4.48 and estimate = −2.12, 95%Crl: −3.76, −0.48, respectively). The relationship between BPA and 3PBA with MSEL composite score exhibited opposing direction of effects between the two trimesters. In the second trimester, BPA showed close to null effect estimates and 3PBA showed a slightly negative relationship with MSEL. In contrast, in the third trimester for females, BPA had a negative relationship with MSEL score (effect estimate = −0.63, 95%Crl: −1.11, −0.15) whereas 3PBA was positively associated with MSEL score (effect estimate = 1.73, 95%Crl: 0.94, 2.52) (Figure 1a and Supplemental Table S4).

Figure 1a.

Standardized difference and 95% credible interval for MSEL composite score for 75^th versus 25^th percentile exposure contrast by trimester for females. Mean field variational approximation method for Bayesian inference procedure for lagged kernel machine regression estimated relative importance of each EDC exposure biomarker at the two critical prenatal windows for females (n=52) on early learning composite score, while adjusting for confounders. Plot of the estimated relative importance of each EDC, as quantified by the difference in the mean score at the 75^th percentile versus the 25^th percentile of a given EDC exposure biomarker, while fixing all other EDC exposure biomarkers constant at their median values. Abbreviations: 3PBA, 3-phenoxybenzoic acid, BPA, bisphenol A, ETPB, ethyl paraben, MEPB, methyl paraben, PRPB, propyl paraben, TCPy, 3,5,6-trichloro-2-pyridinol.

For males, the relationship between prenatal concentrations for the six EDCs in the second trimester and MSEL score were largely null (Figure 1b). The effect estimates for all six EDCs were more precise and generally in opposing directions when examining exposure in the third trimester. In the second trimester for males, there was a weak negative relationship between concentrations of MEPB and ETPB with MSEL score (effect estimate = −1.06, 95%Crl: −4.76, 2.64 and effect estimate = −0.38, 95%Crl: −2.18, 1.43, respectively). In contrast mean standardized MSEL scores based on third trimester metabolite measures were higher by 2.53 (95%Crl: 0.56, 4.50) for MEPB and by 1.13 (95%Crl: −0.01, 2.26) for ETPB at the 75^th compared with the 25^th percentiles, with all other EDCs fixed at their median levels.

Figure 1b.

Standardized difference and 95% credible interval for MSEL composite score for 75^th versus 25^th percentile exposure contrast by trimester for males. Mean field variational approximation method for Bayesian inference procedure for lagged kernel machine regression estimated relative importance of each EDC exposure biomarker at the two critical prenatal windows for males (n=78) on early learning composite score, while adjusting for confounders. Plot of the estimated relative importance of each EDC, as quantified by the difference in the mean score at the 75^th percentile versus the 25^th percentile of a given EDC exposure biomarker, while fixing all other EDC exposure biomarkers constant at their median values. Abbreviations: 3PBA, 3-phenoxybenzoic acid, BPA, bisphenol A, ETPB, ethyl paraben, MEPB, methyl paraben, PRPB, propyl paraben, TCPy, 3,5,6-trichloro-2-pyridinol.

3.3. Two-way interaction between prenatal concentrations of phenols, parabens, and insecticides and MSEL

The interaction effect represents the change in outcomes associated with an interquartile range (IQR) increase in EDC1 when EDC2 is at the 75th percentile compared to the 25th percentile, while all other EDCs are held constant at their median values. For females, MFVB-LKMR analysis identified significant two-way interaction for MSEL scores in both the second and third trimesters (Figure 2a). In the second trimester, there was significant interaction between prenatal levels of TCPy with PRPB and 3PBA, but in opposite directions. For example, the difference in the estimated effect difference when contrasting TCPy (EDC₁) at the 75th versus 25^th percentile when 3PBA (EDC₂) was fixed at the 75th percentile, compared to when 3PBA (EDC₂) was fixed at the 25^th percentile (while keeping all other EDCs at their median levels and adjusting for covariates) was positive, suggesting higher test scores when both chemicals were at the 75^th percentile. In contrast, a negative difference was estimated when contrasting the effect of TCPy at the 75th percentile vs 25^th percentile, when PRPB was fixed at the 75^th and 25^th percentile. This suggests lower MSEL scores when both chemicals were at the 75^th percentile.

Figure 2a.

Interaction and 90% credible interval for select EDCs and MSEL composite score at the second and third trimester for females. Mean field variational approximation method for Bayesian inference procedure for lagged kernel machine regression estimated for statistically significant EDC₁–EDC₂ interactions at second and third trimesters for females in the MARBLES study (n=52). Plot of the estimated interaction effect between two EDC (x-axis), fixing the remaining four EDCs at median exposure levels and adjusting for covariates. Estimate = [E(MSEL|EDC₁ = 75%, EDC₂ = 75%) − E(MSEL| EDC₁ = 25%, EDC₂= 75%)] − [E(MSEL|EDC₁ = 75%, EDC₂= 25%) − E(MSEL| EDC₁= 25%, EDC₂= 25%)]. Abbreviations: 3PBA, 3-phenoxybenzoic acid, BPA, bisphenol A, ETPB, ethyl paraben, MEPB, methyl paraben, PRPB, propyl paraben, TCPy, 3,5,6-trichloro-2-pyridinol.

The number of significant two-way interaction relationships in the third trimester was greater than in the second trimester for females. The interaction relationship between TCPy and PRPB, identified in the second trimester for females, was also found in the third trimester, exhibiting a stronger magnitude of effect (Figure 2a and Supplemental Table 6). There was significant two-way interaction between BPA and two parabens (MEPB and PRPB) in opposing directions. The estimated difference in MSEL test scores associated with BPA at the 75^th percentile was lower (effect estimate: −1.29, 95%Crl: −2.20 to −0.38) when MEPB was fixed at the 75th percentile, compared to when MEPB was fixed at the 25th percentile. In contrast, the two-way interaction between BPA and PRPB estimated higher mean predicted MSEL scores (effect estimate: 4.15, 95%Crl: 2.74 to 5.57) with all other chemicals within the mixture fixed at their medians. There was significant evidence of lower mean predicted MSEL scores for two-way interaction relationships between ETPB and 3PBA. Higher mean predicted MSEL scores were found when assessing the effect of TCPy when each of two parabens (MEPB and ETPB) were fixed at the 75th percentile compared to when they were held at the 25th percentile. Additionally, the effect of PRPB in estimating higher mean predicted MSEL scores was identified when BPA or 3PBA were at the 75th percentile, compared to when they were held at the 25th percentile (Figure 2a).

For males, evidence of a significant two-way interaction on MSEL scores was observed, but it was only present in the third trimester (Figure 2b). The estimated effect of MEPB on MSEL test scores (compring exposure at the 75^th and 25^th percentiles) was lower when either ETPB or PRPB was fixed at the 75th percentile, compared to when ETPB or PRPB was fixed at the 25th percentile. Significantly higher mean predicted MSEL scores were observed when assessing the effect of MEPB with interactions with BPA or TCPY. Significant two-way interaction between PRPB-ETPB and PRPB-3PBA was also found for estimating higher mean predicted MSEL scores for males.

Figure 2b.

Interaction and 90% credible interval for select EDCs and MSEL composite score at third trimester for males Mean field variational approximation method for Bayesian inference procedure for lagged kernel machine regression estimated for statistically significant EDC₁–EDC₂ interactions for the third trimester for males in the MARBLES study (n=78). Plot of the estimated interaction effect between two EDC (x-axis), fixing the remaining four EDCs at median exposure levels and adjusting for covariates. Note: there was no statistically significant two-way interaction in the second trimester for males. Estimate = [E(MSEL|EDC₁ = 75%, EDC₂ = 75%) − E(MSEL| EDC₁ = 25%, EDC₂= 75%)] − [E(MSEL|EDC₁ = 75%, EDC₂= 25%) − E(MSEL| EDC₁ 25%, EDC₂= 25%)]. Abbreviations: 3PBA, 3-phenoxybenzoic acid, BPA, bisphenol A, ETPB, ethyl paraben, MEPB, methyl paraben, PRPB, propyl paraben, TCPy, 3,5,6-trichloro-2-pyridinol.

3.4. Non-linear effects of prenatal concentrations of phenols, parabens, and insecticides on MSEL

Figure 3a shows the cross-section plots of the exposure-response surface for the significant two-way interactions for MSEL for females. These plots illustrate the exposure-response for EDC₂ when EDC₁ is low (at the 25^th percentile) and high (at the 75^th percentile) while also fixing the other four EDCs at their median values and adjusting for covariates. For females, in the second and third trimester there is a null association between TCPy and MSEL scores for lower levels of PRPB, however when PRPB is high there is distinct negative association between TCPy and MSEL scores (Figure 3a, panel 1). In the second trimester, TCPy appears to have a slight positive relationship when 3PBA is low and distinct negative relationship on MSEL when 3PBA is high (Figure 3a, panel 2). In the third trimester, similar patterns of effect measure modification are seen for MEBP-BPA, PRPB-BPA, and MEPB-ETPB, where there is a null relationship between EDC₂ and MSEL score when EDC₁ is low and a negative relationship between EDC₂ and MSEL scores when EDC₁ is high. We observed slight negative associations between EDC₂ and MSEL scores when EDC₁ is low and an even stronger negative associations between EDC₂ and MSEL scores when EDC₁ is high for TCPy-ETPB, 3PBA-ETPB, TCPy-MEPB, 3PBA-PRPB.

Figure 3a.

Interaction cross-section of the exposure-response surface plotted for select EDC relationships for females. Mean field variational approximation method for Bayesian inference procedure for lagged kernel machine regression estimated trimester-specific exposure–response functions for five select EDC₁-EDC₂ chemical exposures relationships for females in the MARBLES study (n=52). Plot of the cross-section of the estimated exposure–response surface for EDC₁ (x-axis) at the 25th (top panel of cluster) and 75th (bottom panel of cluster) percentile of EDC₂ exposure, fixing the remaining four chemicals at median exposures levels and adjusting for covariates. Abbreviations: 3PBA, 3-phenoxybenzoic acid, BPA, bisphenol A, ETPB, ethyl paraben, MEPB, methyl paraben, PRPB, propyl paraben, TCPy, 3,5,6-trichloro-2-pyridinol.

Figure 3b shows the cross-section plots of the exposure-response surface for the significant two-way interactions for MSEL for males. Overall, there was no strong evidence of effect measure modification between the six EDCs in either trimester when comparing the magnitude of the estimated effects.

Figure 3b.

Interaction cross-section of the exposure-response surface plotted for select EDC relationships for males. Mean field variational approximation method for Bayesian inference procedure for lagged kernel machine regression estimated trimester-specific exposure–response functions for four select EDC₁-EDC₂ chemical exposures relationships for males in the MARBLES study (n=78). Plot of the cross-section of the estimated exposure–response surface for EDC₁ (x-axis) at the 25th (top panel of cluster) and 75th (bottom panel of cluster) percentile of EDC₂ exposure, fixing the remaining four chemicals at median exposures levels and adjusting for covariates. Abbreviations: 3PBA, 3-phenoxybenzoic acid, BPA, bisphenol A, ETPB, ethyl paraben, MEPB, methyl paraben, PRPB, propyl paraben, TCPy, 3,5,6-trichloro-2-pyridinol.

4. Discussion

We applied MFVB-LKMR to data on a prospective cohort of mother-child pairs, enriched with ASD cases, to investigate critical windows of co-exposure to phenols, parabens, and OP in relation to neurocognitive development at age 3 in children, focusing on the second and third trimesters of maternal pregnancy. This study provides evidence that timing of exposure matters and warrants more consistent exploration in research focusing on prenatal exposures and neurodevelopmental outcomes in children. We found stronger evidence of varying effects from the EDC mixture based on timing of prenatal exposure in females than for males. This finding aligns with the growing evidence that the impact of prenatal exposure EDCs on neurodevelopmental outcomes may differ between sexes [77]. This theory stems from the understanding that the hippocampus, crucial for learning and memory, is influenced by sex and thyroid hormones. This makes it a prime target for endocrine disruption, particularly sensitive to estrogenic compounds like BPA [77]. In our study, we found that for females, concentrations of MEPB in the second trimester were positively associated with MSEL score and PRPB was negatively associated with MSEL score, and these associations were stronger in the third trimester. We observed a negative association with BPA and a positive association with 3PBA in females only in the third trimester. In the second trimester, there was evidence of departure from additivity between select EDCs. In females, there was a significant two-way interaction between TCPy and PRPB, as well as between TCPy and 3PBA, in the second trimester, and a significant interaction among all three classes of EDCs in the third trimester. In males, there were no main effects between any of the EDCs and MSEL score in the second trimester and only concentrations of MEPB were positively associated with MSEL score in the third trimester. There was no evidence of two-way interaction between the EDCs for effecting MSEL in males.

Concurrent or sequential exposure to a variety of EDCs is more reflective of the real-world experience. Since fetal development is a sequential and cumulative process, the overarching purpose of this study was to identify critical windows within the prenatal period that are more sensitive to the effects of co-exposure to various EDCs for affecting neurodevelopmental outcomes in children. Assessing the varying effect of the joint exposure-outcome relationship in the context of multiple exposures is statistically challenging and complex. Mixture methods like Bayesian kernel machine regression (BKMR) [78] and weighted quantile sum (WQS) [79] can assess a variety of exposure–response relationships in a multipollutant setting but at a single time point and not across time. The implementation of MFVB-LKMR mixture method allowed us to estimate how the effects of the prenatal exposure mixture of phenols, parabens, and OP on MSEL differed based on co-exposures occurring midway and at the end of the gestational period. Additionally, the MFVB-LKMR mixture method allowed us to assess this time-varying relationship while accounting for potential collinearity between EDCs and the non-linear and non-additive effects of the mixture at the two distinct exposure time windows [53, 75]. As a result, we observed interactions between EDCs in specific trimesters. For example, for females there was negative interaction between TCPy and PRPB in association with MSEL in both the second and third trimester, with the third trimester showing a stronger magnitude of effect. There was a greater number of significant two-way interaction relationships in the third trimester than in the second trimester across different classes of EDCs for females. For males, there was only evidence of two-way interaction in association with MSEL in the third trimester across different classes of EDCs. These findings suggest that the effect of prenatal exposure to select EDCs may be dependent on the exposure level of other EDCs during distinct time periods to impact MSEL outcomes in children.

There is limited data on the relationship between prenatal exposure to parabens, pesticides, and phenols and its impact on neurodevelopmental outcomes in children while considering the timing of multiple exposures. Therefore, we acknowledge that differences in analytical methods and measures of neurocognitive development may prohibit direct comparisons of our study findings. Few studies have assessed prenatal concentrations of parabens in relation to neurocognitive outcomes in children. Three studies used a single-chemical analytical approach [41–43], one study assessed the effects of prenatal exposure based on timing of exposures [41] and a prior analysis in the MARBLES cohort examined the association between chemical mixtures with ASD based on summary urinary measures across pregnancy [40]. A prospective study, part of the Environment and Childhood (INMA) Project in Spain, investigated the relationship between placental concentrations of phenolic EDCs, including BPA and parabens, and cognitive development in preschoolers [42]. Using the McCarthy Scales of Children’s Abilities (MSCA) to measure cognitive and motor function assessment at ages 4–5, the study analyzed a sub-sample of 191 mother-child pairs with samples collected from 2000 to 2008. It found negative associations between placental PRPB levels and motor function in preschool children based on a single-chemical analysis approach. In a previous prospective study, the impact of prenatal exposure to phenols and parabens on child development was evaluated in a longitudinal cohort of 478 mother-child pairs in China [41]. Maternal urine samples collected during each trimester of pregnancy between 2014 and 2015 were used to measure exposure to bisphenols, parabens, and triclosan. Child development was assessed at the age of 2 using the Bayley Scales of Infant Development, which includes the mental development index (MDI) to evaluate cognition, language, and social development, as well as the psychomotor development index (PDI) to assess fine motor skills. This longitudinal cohort used single-chemical analytical approach and evaluated trimester-specific exposure to phenols and parabens and its effect on MDI and PDI at age 2 years [41]. Some but not all of our results were consistent with their findings. For example, MEPB, ETPB, PRPB and Σparabens concentrations were negatively associated with MDI in the second trimester for females only. This is consistent with our findings of a negative relationship between PRPB and MSEL in the second trimester for females and null findings for males. In contrast to Jiang et al.’s findings, for females we observed the same negative relationship between PRPB and MSEL in the third trimester and positive relationship between MEPB and MSEL in both trimesters [41]. Our analysis found negative associations between prenatal BPA levels and MSEL scores in the third trimester for females. Other single-chemical analyses for prenatal BPA levels and neurocognitive outcomes have reported mixed results [43] and there is no epidemiologic data assessing trimester specific concentrations of BPA in relation to neurocognitive outcomes.

A systematic review of prenatal exposure to OP insecticides and various neurodevelopmental outcomes in children reported overall evidence of cognitive deficits (related to working memory) in children at age 7 years and behavioral deficits (related to attention) in toddlers [44]. The outcomes and timing of prenatal exposure in our study differed from what was included in the systematic review, making direct comparisons of results challenging. We found that the relationship between 3PBA and MSEL scores differed for females depending on the timing of exposure (negative relationship in the second trimester and positive effects in the third trimester). For males, we found a positive relationship for 3PBA and MSEL score in the second trimester and null relationship between 3PBA and MSEL score in the third trimester.

There are a number of notable strengths in this current study. We utilized publicly available prospective data including urinary biomarkers of three different classes of EDC exposure measured during the prenatal period. Prospective measures of neurodevelopmental outcomes ensure temporality for the observed relationships in our study. In addition, we used repeated measures of biomarker EDC concentrations measured at two distinct time periods during the prenatal period, allowing us to assess time-varying exposures to mixtures within the critical prenatal window of fetal development. Notably, application of the MFVB-LKMR allowed us to explore the potential nonlinear main effects of each EDC at both time points in relation to the neurodevelopmental outcomes while considering co-exposure to the other EDCs within the complex mixture. By using MFVB-LKMR, we observed that the relationship between some EDCs and neurodevelopmental outcomes varied across the two prenatal time windows. Additionally, MFVB-LKMR identified evidence of departure from additivity in some EDC two-way relationships.

Nevertheless, there are some study limitations. Using publicly available data limited our ability to investigate research questions based on the available data provided and within a population enriched for risk of ASD. We recognize that our findings may not be generalizable to average risk populations. Further, we did not have robust measures of socioeconomic status and acknowledge the potential for residual confounding associated with lifestyle and inequities. As previously reported, most women were enrolled in their second trimester of pregnancy limiting our ability to investigate the impact of the chemical mixture across all three trimesters [58]. Each trimester plays a vital role in various aspects of brain development and the first trimester is often considered the most crucial for forming the brain’s basic structures [80]. However, the second trimester marks a period of significant growth and is important for developing more advanced brain structures and the proliferation of neurons. Prospective epidemiological data indicate that the critical window for the development of cognitive, language, and visual skills at age two occurs between 20-25 weeks of gestation, a time period that our study captures [81]. We restricted the study sample to mother-child pairs with complete available data, which reduced the sample size to 130 participants with second and third trimester data, and increased the potential for selection bias. Child’s birth year was the only significant characteristic that differed between those included versus those not included, and our study sample may reflect less contemporary exposure profiles. ETPB and BPA, two EDCs with more than a third of the samples having values below the LOD, were identified to exhibit significant two-way interactions. Therefore, caution must be exercised when interpreting these interactions, as the results are based on relatively low concentrations of these analytes. Furthermore, the use of single imputation for analytes below the LOD and negative machine-read values may underestimate uncertainty in estimates. Given that population level exposure to OP insecticides and BPA has decreased over time [72, 82, 83], it is possible that this selection may bias some of our findings away from the null. The EDCs included in this study have a relatively short half-life (range 4 to 33 hours) [84–86]. It is possible that urinary measures may not be representative of long-term exposure during pregnancy. However, the use of repeated measures during the second and third trimesters allowed for a close approximation of the overall exposure profile during the mid to late gestational period. Moreover, although the MSEL is extensively used by practitioners to supplement clinical observations of early development and intervention responses, its major limitation is the binary scoring system [87]. In this system, a child is typically scored as either able or unable to complete a task, emphasizing skill acquisition rather than offering a scaled score across various motor abilities. Consequently, this binary approach may compromise the reliability of the MSEL and increase the likelihood of errors in assessment. We acknowledge that MFVB-LKMR methods do not provide a single parameter estimate for each timepoint, which is a limitation when the focus is on the ‘overall’ mixture effect rather than on individual components. Lastly, our analysis relied on the MFVB extension of LKMR, an approach that was developed for use in large sample sizes where the computational burden of MCMC is prohibitive. The small sample size (n = 130) is a limitation that could lead to poor sigma coverage, biased effect estimation, and compromise the overall robustness of our results. Furthermore, the small sample size may have contributed to our null findings and diminished statistical power, as larger samples might be necessary to effectively perform the non-parametric kernel function, adjust for covariates, and detect two-way interactions. Thus, we cannot rule out the potential for chance to explain our findings given the multiple comparisons made.

5. Conclusion

We applied MFVB-LKMR methods to investigate whether co-exposure to phenols, parabens, and OP insecticides during the second and third trimester of pregnancy differentially affects measures of overall neurodevelopment in higher risk males and females at age 3. Our study identified evidence of varying effects from the EDC mixture based on timing of prenatal exposure in both females and males, with stronger findings for females. Additionally, we found that select EDCs could act jointly at distinct time points during the prenatal period to affect neurodevelopmental outcomes. Current findings may contribute to informing early environmental health intervention programs focused on environmental exposures. Larger studies in other cohorts with prospective measures across all three trimesters of pregnancy are needed to understand the differential neurodevelopmental effects of time-varying co-exposure to ubiquitous EDCs. Further, comparing different statistical approaches to address the same research question can be useful for future studies.

Highlights.

• First study to assess the effect of prenatal mixtures of phenol, parabens, and pesticides in 2^nd and 3^rd trimester on child neurodevelopment

• Timing of exposure and two-way interaction between EDCs impacts MSEL score

• Impact of time-dependent co-exposures on neurodevelopment were sex-specific

Data availability

This manuscript was prepared using National Children’s Study Research Materials obtained from the NCS Vanguard Data and Sample Archive and Access System and the NICHD Data and Specimen Hub (DASH). We acknowledge NICHD DASH for providing the National Children’s Study data that was used for this research. We appreciate the contributions of the NCS NICHD Principal Investigator Jack Moye, and the NCS Initial Vanguard Center Principal Investigators: Children’s Hospital of Philadelphia (Jennifer Culhane); Mt. Sinai Medical School (Philip Landrigan); South Dakota State University (Bonny Specker); University of California at Irvine (James Swanson and Dean Baker); University of North Carolina at Chapel Hill (Barbara Entwisle and Nancy Dole); University of Utah School of Medicine (Ed Clark); and the University of Wisconsin (Maureen Durkin). Data are not publicly available, as they were obtained through Data Use Agreements.