Gestational exposure to environmental chemical mixtures and cognitive abilities in children: A pooled analysis of two North American birth cohorts

Abstract

Background:

Gestational exposures to single toxic chemicals have been associated with cognitive deficits in children, but few studies have explored chemical mixtures.

Objectives:

To evaluate the associations between gestational chemical biomarker mixtures and cognitive abilities in children from two prospective cohorts.

Methods:

This study includes 617 birthing parent–child pairs from the Health Outcomes and Measures of the Environment (HOME) and Maternal-Infant Research on Environmental Chemicals (MIREC) Studies. We measured 29 chemical biomarkers (metals, persistent organic pollutants, perfluoroalkyl substances, organophosphate esters, phenols, phthalates, organophosphate pesticides, and parabens) in pregnant individuals during early pregnancy and their children’s cognitive abilities at ages 3 to 5 years using Wechsler Intelligence Scales. We assessed linear associations using quantile g-computation and non-linear associations using Bayesian Kernel Machine Regression (BKMR) methods, adjusted for covariates.

Results:

Using quantile g-computation, we observed overall null associations between the chemical biomarker mixture and cognitive outcomes among preschool-age children. Although statistical significance was not attained for child sex as an effect modifier, our stratified analysis unveiled a moderate divergence in association trends. We noted a marginal inverse trend between the chemical biomarker mixture and cognitive scores [Full-Scale Intelligence Quotient (FSIQ) & Performance Intelligence Quotient (PIQ)] among males. Using quantile g-computation and BKMR methods, we observed that PBDE47, PFHxS, and di-ethyl organophosphates commonly contributed towards a decline in FSIQ scores in males. Among males, a quartile increase in the chemical biomarker mixture was associated with a 0.64-point decrease (95% CI: −2.59, 1.31) in the FSIQ score and a 1.59-point decrease (95% CI: −3.72, 0.54) in the PIQ score.

Conclusion:

In this study, we observed a weak negative trend between the gestational chemical biomarker mixture and cognitive scores (FSIQ/PIQ) among males. Further studies are needed to confirm the findings between the longitudinal chemical biomarkers and child cognitive scores at school ages.

1. Introduction

Gestational exposure to environmental chemicals, including toxic metals, persistent organic pollutants, and pesticides, has been associated with adverse neurodevelopmental outcomes in children (Werler, 2011; Lanphear, 2015; Vrijheid et al., 2016), including poorer cognitive abilities (Lanphear, 2015; Grandjean and Landrigan, 2014; Bellinger, 2013; Lam et al., 2017; Bauer et al., 2020; Percy et al., 2021; Yonkman et al., 2023; Jackson-Browne et al., 2018; Chen et al., 2014; Desrochers-Couture et al., 2018; Packull-McCormick et al., 2023; Goodman et al., 2023; Nkinsa et al., 2020; Nakiwala et al., 2018; Tanner et al., 2020; Rauh et al., 2011; Jusko et al., 2019; Bornehag et al., 2021; Thilakaratne et al., 2023; Kalloo et al., 2021). Most of these chemicals cross the placenta (Mathiesen et al., 2021; Morello-Frosch et al., 2016; Rager et al., 2020; Fisher et al., 2016; Li et al., 2013; Andersen et al., 2021; Bräuner et al., 2022; Mitro et al., 2015), and exposure to environmental chemicals during pregnancy can adversely influence fetal brain development via disruption of neurotransmission, synaptogenesis, synaptic trimming, and gonadal or thyroid hormonal functioning (Lanphear, 2015; Schug et al., 2015; Vandenberg et al., 2012).

Most epidemiological studies have examined the effect of a single chemical in isolation with one or more health outcomes. However, since humans are simultaneously exposed to many environmental chemicals, it is critical to study the cumulative impact of chemical mixtures on cognitive outcomes (Carlin et al., 2013; Rider et al., 2013). Additionally, only a limited number of studies have focused on gestational chemical mixtures and cognitive outcomes (Thilakaratne et al., 2023; Tanner et al., 2020; Guo et al., 2020; Lee et al., 2021; Brennan Kearns et al., 2024). These studies have predominantly involved biomarkers within a specific chemical class. Few studies have attempted to elucidate the combined associations between gestational chemical biomarkers and cognitive outcomes using latent variables (Yonkman et al., 2023; Kalloo et al., 2021). Previously, the two cohorts under investigation here studied associations between chemical mixtures and cognitive outcomes. Kalloo et al. assessed associations between gestational exposure to chemical mixtures using latent variables produced using principal component analysis and identified statistically significant associations with working memory in the HOME Study (Yonkman et al., 2023). Yonkman et al. stratified study participants into five exposure profiles using latent profile analysis and identified statistically non-significant trends with child cognitive scores in the MIREC Study (Kalloo et al., 2021).

Findings from these studies suggest heterogeneity in the association between chemical biomarkers and cognitive abilities. The lack of consistent findings across these studies highlights the need for validation and replication (Hubbard, 2016; NASEM, 2019). Moreover, studying chemical biomarker mixtures could shed more light on synergistic or additive associations, but this might also penalize the study power with a limited sample size (Quiros-Alcala et al., 2023; Eick and Hüls, 2022).

To address these gaps, we assessed the joint association between 29 environmental chemical biomarkers measured during pregnancy and cognitive abilities among children from two prospective North American cohorts. Leveraging the availability of diverse environmental chemical biomarkers and a larger sample size, we aimed to quantify the potential associations between exposure to chemical biomarker mixtures in pregnant individuals and cognitive outcomes in their children.

2. Methods

2.1. Study participants

This study comprises participants from two prospective pregnancy and birth cohorts in North America. The Health Outcomes and Measures of the Environment (HOME) Study includes 389 pregnant individuals who delivered a singleton newborn from 2003 to 2006 in the Cincinnati metropolitan area (Braun et al., 2016). The Maternal-Infant Research on Environmental Chemicals (MIREC) Study includes 1,910 pregnant individuals who delivered singleton newborns from 2008 to 2011 from 10 Canadian cities (Arbuckle et al., 2013). Further information on the recruitment and inclusion criteria are available in the appendix (SM1). The Institutional Review Boards of Cincinnati Children’s Hospital Medical Center and affiliated hospitals approved the HOME Study. The research ethics boards of Health Canada, CHU Sainte-Justine Research Centre, and all the participating recruitment centers granted ethical approval for the MIREC study. All the study participants provided written informed consent for themselves and their children; all participating children provided assent.

2.1.1. Biomarkers quantifying environmental chemical exposures

We measured concentrations of 44 chemical biomarkers in the urine, serum, plasma, or whole blood samples collected from pregnant people between the 13th and 19th week of gestation in the HOME Study and between the 6th and 13th week of gestation in the MIREC Study. These chemical biomarkers encompass eight groups, including metals [lead (Pb), mercury (Hg), arsenic (dimethylarsinic acid – DMA)], pesticides (organochlorine, organophosphate), polybrominated diphenyl ethers (PBDEs), polychlorinated biphenyls (PCBs), phenols, phthalates, perfluoroalkyl substances (PFAS), and organophosphate esters (OPEs) (Table S1).

In the HOME Study, the urinary concentrations of paraben, phthalate, and phenol biomarkers were measured using high-performance liquid chromatography coupled with tandem mass spectrometry. (Ye et al., 2008; Silva et al., 2007) Urinary dialkylphosphates were measured using gas chromatography coupled with tandem mass spectrometry. (Bravo et al., 2004) Metal (Pb and Hg) concentrations from whole blood samples were quantified using an inductively coupled plasma mass spectrometry. (Nixon et al., 1999; Jones et al., 2017) Furthermore, PBDE, dichlorodiphenyldichloroethene (DDE), and PCB biomarkers from serum were quantified using gas chromatography coupled with the tandem mass spectroscopy method. (Sjödin et al., 2004) We quantified serum cotinine and PFAS concentrations using high-performance liquid chromatography-tandem mass spectroscopy. (DeLorenze et al., 2002; Kato et al., 2011) All of these chemical biomarkers were measured at the Centers for Disease Control and Prevention (CDC). The DMA biomarker was measured at the Trace Element Analysis Core at Dartmouth College. (Jackson, 2015) In the HOME Study, we supplemented measurements of DMA from the 26th week of gestation (94 %) and 16th (6 %) weeks of gestation.

In the MIREC Study, the urinary phthalate biomarker concentrations were measured using Ultra Performance Liquid Chromatography coupled with a tandem mass spectrometer (Arbuckle et al., 2014). Urinary phenol biomarker concentrations were measured using gas chromatography coupled with a tandem mass spectrometer (Arbuckle et al., 2014). Metals concentrations in whole blood (Pb & Hg) and urine (DMA) were measured using inductively coupled plasma mass spectrometry (Arbuckle et al., 2016; Ettinger et al., 2017). PFAS in plasma was measured using Ultra Performance Liquid Chromatography coupled with a mass spectrometer (Fisher et al., 2016). Furthermore, PCB, PBDE, and DDE biomarkers from plasma samples were measured using gas chromatography coupled with a mass spectrometry detector (Fisher et al., 2016). Urinary organophosphate metabolites were measured using gas chromatography with a mass spectrometer. All of these were measured at the Institut National de Santé Publique du Québec (INSPQ, Québec, Canada). In the MIREC Study, to ascertain total lipids, we measured total cholesterol (TC), free cholesterol (FC), triglycerides (TG), and phospholipids (PL) measured in grams per liter (g/L), employing enzymatic techniques and colorimetry at the laboratory of Centre Hospitalier de l’Université Laval. Maternal urine-specific gravity was measured using a handheld refractometer (Atago, Tokyo, Japan) in the HOME and MIREC Studies.

Twenty-nine of 44 chemicals measured in the HOME and MIREC studies were detected in at least 50 % of the study participants and included for further analysis (Fig. S1; Table S1). Chemical biomarkers detected among at least 50 % of the study population and values below the limit of detection (LOD) were imputed by employing the left-truncated lognormal distribution for each chemical through the “fill-in” method (Lubin et al., 2004; Mersmann et al., 2023).

We aggregated phthalate and organophosphate pesticide biomarkers representing a common parent compound by estimating the molar sum (∑metabolite concentration/metabolite molecular weight) of the metabolite concentrations. Phthalate metabolites MEHHP (mono 2-ethyl-5-hydroxyhexyl phthalate), MEHP (mono-2-ethylhexyl phthalate), MECPP (mono-2-ethyl-5-carboxypentyl phthalate), and MEOHP (mono-2-ethyl-5-oxohexyl phthalate) were summed to represent ∑DEHP [di(2-ethylhexyl) phthalate]. Similarly, we calculated the molar sum of diethyl organophosphate (∑DEP) pesticides using diethylphosphate and diethylthiophosphate concentrations and di-methyl organophosphate (∑DMP) concentrations using dimethylphosphate and dimethylthiophosphate. Serum/plasma biomarker concentrations, except for PFAS, were standardized using serum lipid concentrations, and urinary biomarkers were standardized using urine-specific gravity (O’Brien et al., 2016; Hauser et al., 2004).

$${\mathit{\text{Concentration}}}_{\mathit{\text{urine–specific}}\mathit{\text{gravity}}\mathit{\text{standardized}}}={\mathit{\text{Biomarker}}\mathit{\text{concentration}}}_i\frac{\left(S{G}_m-1\right)}{\left(S{G}_i-1\right)}$$

Concentration_{urine-specific gravity standardized} represents the urinary biomarker concentrations standardized using urine specific gravity, concentration_i is the urinary concentration of a biomarker measured for i^th observation, SG_m represents the median urinary specific gravity of the participant pool, and SG_i is the urinary specific gravity of the i^th participant.

$${\mathit{\text{Concentration}}}_{\mathit{\text{lipid–standardized}}}=\frac{{\mathit{\text{Biomarker}}\mathit{\text{concentration}}}_i}{{\mathit{\text{Total}}\mathit{\text{lipid}}}_i}$$

Concentration_{lipid standardized} indicates the serum/plasma biomarker concentration standardized using total lipid. In the MIREC Study, the total lipid value was estimated using: 1.677*(TC-FC) + FC + TG + PL (Patterson et al., 1991). Whereas in the HOME Study, total lipids values were estimated using a short form of the Philips formula: (2.27*TC) + TG + 62.3. (Phillips et al., 1989).

2.1.2. Cognitive assessment in children

We obtained cognitive assessment scores for 292 children from the HOME Study, assessed at around age 5 years, and for 607 children from the MIREC Study, assessed at around three years of age. Only 607 children from six out of the ten study cities underwent neurobehavioral assessments due to financial constraints. (Fisher et al., 2023) Among the 607 children who underwent neurobehavioral assessments in the MIREC Study, five were excluded due to the unavailability of sub-scores. Cognitive assessments at age 5 years in the HOME Study and age 3 years in the MIREC were performed using the Wechsler Preschool and Primary Scale of Intelligence – 3rd edition (WPPSI-III in HOME & WPPSI-III^CDN in MIREC). At age 8, children in the HOME Study additionally underwent cognitive assessments using the Wechsler Intelligence Scale for Children – 4th edition (WISC-IV). To maximize the study sample size and considering the comparability between the WPPSI-III and WISC-IV assessments, children in the HOME Study who lacked measured WPPSI assessments at age 5 (n = 23) had their data supplemented with WISC scores at age 8 (Braun et al., 2017). A total of 917 [(HOME: 292-WPPSI + 23-WISC)+(MIREC: 602-WPPSI)] children with age-standardized cognitive assessments were included in this study (Fig. S1). However, owing to unobserved maternal chemical biomarkers, further analysis was limited to a maximum of 847 mother–child dyads (Fig. S2). This subset includes at least one of the 29 maternal chemical biomarkers among the study subjects. Detailed information on the outcome assessment tools for full-scale IQ (FSIQ), verbal IQ (VIQ), and performance IQ (PIQ) is available in the appendix (SM2).

2.1.3. Covariates

We created a directed acyclic graph based on prior literature (Yonkman et al., 2023; Thilakaratne et al., 2023) and considered ten covariates as potential confounders (Fig. S3). Among the potential confounders, maternal race was coded as Non-Hispanic White and other. Maternal education was coded into three levels: less than or equal to high-school education, greater than high school and less than a postgraduate education, and postgraduate or more. Maternal fish intake during the first trimester was coded into three levels: none, 1–3 times per month, and greater than 3 times per month. Covariates such as maternal age at delivery, Home Observation for Measurement of the Environment score (HOME score), parity, and cotinine concentrations were included as continuous variables. We obtained the HOME score when the child was 12 months old in the HOME Study and at age 3 for MIREC, which assessed the nurturing aspects of the home environment (Bradley and Caldwell, 1980). Additionally, we adjusted for cohort and study city to account for potential heterogeneity.

We used the fill-in left-truncated imputation method to impute cotinine concentrations (nicotine biomarker – a measure of tobacco smoke exposure) for subjects with values below the detection limit (29.5 % in HOME and 43.9 % in MIREC) (Lubin et al., 2004). In the MIREC Study, we used the gradient boosting method to impute unmeasured continuous variable HOME scores (2.1 % [n = 18]; mean squared error: 16.1) (Friedman, 2001). Categorical variables such as maternal education (1 % (Yonkman et al., 2023); accuracy: 22.6 %), parity (0.7 % (Lam et al., 2017); accuracy: 56.9 %), and child sex (0.7 % (Lam et al., 2017); accuracy: 24.4 %) were imputed using an ensemble of classifiers (random forest and neural network) available through SuperLearner R-package (Polley et al., 2023; Breiman, 2001).

2.2. Statistical analysis

We calculated the Spearman correlations between the 29 chemical biomarkers. Based on guidelines endorsed by the American Statistical Association and others, we have chosen to avoid null hypothesis significance testing along with reporting findings based on p-value thresholds in the study of chemical mixtures (Hubbard, 2016; Amrhein et al., 2019; Wasserstein et al., 2016; McShane et al., 2019; Poole, 2001; Lash, 2017). Associations were interpreted by their direction, magnitude, and confidence intervals.

2.2.1. Joint associations between chemical biomarkers and child cognitive abilities

2.2.1.1. Quantile g-computation method.

While considering 29 chemical biomarkers as a mixture, we included a subset of 617 mother–child dyads due to the availability of maternal biomarkers and assessed the joint associations using the quantile g-computation method (Keil et al., 2020). The model was fit, assuming a normal distribution of cognitive outcomes and converting chemical biomarkers from continuous to quartiles. This model also included adjustments for the covariates previously mentioned. The quantile g-computation model operates by constructing individual regression models for each of the 29 biomarkers on a quartile scale. It combines these model estimates to determine the joint association representing the combined impact of the chemical biomarker mixture. For the negative coefficients, the model sums their values and denotes this sum as the scaled effect size of negative coefficients. Conversely, it sums the beta coefficients with positive values to denote the scaled effect size of positive coefficients. To estimate the weights of chemical biomarkers, the model divides each biomarker’s beta value (whether it contributes positively or negatively) by its respective scaled effect size in the corresponding direction. The joint association was calculated by summing all the numeric beta coefficients irrespective of the direction of the association derived from the individual multiple linear regression models. We obtained the chemical biomarker weights that contributed to the joint association from the non-bootstrap quantile g-computation model and multiplied the negatively contributing weights with “−1” for presentation purposes. We emphasized the chemical biomarker weights that exceeded a specific threshold (threshold = 2/n), which varied based on the number of chemical biomarkers (n) present in the mixture. Additionally, to highlight the potential uncertainty of the joint associations from the non-bootstrap model, we supplemented our results with the effect estimates based on 400 bootstrap iterations.

We explored the role of child sex as a potential effect modifier and reported p-values for the interaction term using the quantile g-computation method. Along with the overall pooled analysis, we performed stratified analyses by child sex, due to the known sexual dimorphic mechanisms driven by certain chemical exposures (Schug et al., 2015). We performed the analysis using the qgcomp and qgcompint R-packages. (Keil, 2022; Keil, 2023).

2.2.1.2. Bayesian Kernel Machine regression (BKMR).

We implemented the BKMR model with hierarchical variable selection of chemical biomarkers using Markov Chain Monte Carlo methods, assuming Gaussian distribution of cognitive outcomes, to visualize potential non-linear associations. The exposure mixture and covariate cotinine concentrations were converted to log₁₀ scales. We included the same covariates as the quantile g-computation method to maintain consistency. We set the model parameter to generate 50,000 iterations with a single chain and a flat prior distribution. To obtain the overall estimates and credible intervals, we burned the first 25,000 iterations and sampled 500 iterations by selecting every 50th iteration between the 25,001–50,000 iterations. Overall estimates were then visualized for every five percent change of the exposure mixture over a range of 10–90th percentiles. Additionally, we obtained the group and conditional posterior inclusion probabilities (PIPs) to highlight the chemical biomarker contribution to the overall estimates. Along with the overall pooled analysis, we performed stratified analyses by child sex. We implemented this model using the bkmr R-package (Bobb et al., 2014; Bobb, 2023).

2.2.2. Single chemical analysis

We used multiple linear regression to explore the potential uncertainties in joint association models. This allowed us to compare the effect estimates from both joint and single chemical approaches. We adjusted for the same covariates as in the joint association models. The effect estimates (β) correspond to a change in IQ for every 10-fold increase in chemical biomarker concentrations. Additionally, we obtained confidence intervals assuming alpha = 0.05 and 0.2 to yield 95 % and 80 % confidence intervals.

2.2.3. Secondary analysis

In our secondary analysis, our objective was to explore the joint associations between a restricted set of chemical biomarkers and cognitive outcomes. We identified a subset of chemical biomarkers by outcome and/or child sex, using our results from the multiple linear regression assuming alpha = 0.2. Subsequently, we employed the quantile-based g-computation approach to investigate potential relationships between a subset of these chemical biomarkers and cognitive performance. The quantile g-computation step may include up to a subset of 9 chemical biomarkers depending on the child’s sex and/or outcomes.

2.2.4. Sensitivity analysis

To further test the robustness of our results, considering 29 gestational chemical biomarkers, we further assessed the associations by: 1) excluding maternal fish intake as a covariate and 2) excluding maternal fish intake as a covariate and excluding children for whom IQ was assessed using WISC.

3. Results

The primary analysis included 617 mother–child dyads from the HOME and MIREC Studies. The pregnant individuals had a median age of 32 years (interquartile range (IQR):29–35), and predominantly identified as White (83 %). Their educational attainment mostly ranged from above high school to below postgraduate levels (Table S2). The majority were nulliparous (44 %), and approximately half delivered a female infant (53 %). The median gestational age at delivery was 40 weeks (IQR: 39–41). Notably, approximately 60 % of these study participants had cotinine concentrations below the level of detection (Table S1).

The majority (24 of 29) of chemical biomarker concentrations were somewhat higher among the HOME Study participants than the MIREC Study participants (Table 1). These include DDE, 2,2′4,4′-tetrabromodiphenyl ether (PBDE47), PCBs, PFAS, parabens, OPEs, triclosan, BPA, and phthalates. We observed positive correlations between the chemical biomarkers within the chemical classes (Fig. S4).

Table 1.

Summary of chemical biomarkers and cognitive assessment scores among study participants.

Biomarker	Pooled (n = 617)	HOME (n = 139)	MIREC (n = 478)
Pb	6.22 (4.70, 8.49)	6.3 (5.00, 8.10)	6.22 (4.56, 8.49)
Hg	0.66 (0.36, 1.20)	0.65 (0.4, 0.99)	0.66 (0.34, 1.28)
DMA	4.85 (3.07, 7.46)	2.65 (1.37, 4.76)	5.4 (3.68, 8.84)
DDE	54.17 (37.93, 78.85)	66.9 (50.30, 91.10)	48.66 (35.62, 75.00)
PBDE47	8.71 (4.78, 17.20)	19.1 (11.10, 34.20)	7.18 (<LOD, 12.13)
PCB118	2.79 (1.90, 4.34)	4.5 (3.15, 6.30)	2.48 (1.74, 3.41)
PCB138	5 (3.20, 7.69)	7.3 (5.40, 9.95)	4.38 (2.90, 6.44)
PCB153	8.29 (5.43, 12.30)	9.8 (7.60, 13.45)	7.59 (5.00, 11.77)
PCB180	5.29 (3.41, 8.47)	5.6 (3.95, 8.75)	5.23 (3.23, 8.42)
PFHxS	1.1 (0.70, 1.80)	1.5 (0.90, 2.40)	0.99 (0.67, 1.60)
PFOA	2.1 (1.30, 3.50)	5.5 (3.70, 7.65)	1.7 (1.10, 2.50)
PFOS	5.1 (3.50, 8.80)	13.8 (9.20, 18.60)	4.45 (3.20, 6.10)
BCEP	0.31 (0.16, 0.68)	0.64 (0.35,1.19)	0.25 (0.15,0.52)
BDCIPP	0.39 (0.20, 0.80)	0.88 (0.54,1.67)	0.3 (0.16,0.55)
DNBP	0.1 (0.06, 0.17)	0.21 (0.15, 0.32)	0.08 (<LOD, 0.13)
DPHP	0.77 (0.38, 1.45)	1.75 (1.14, 3.24)	0.58 (0.32, 1.01)
Triclosan	11.77 (3.65, 93.83)	19.58 (7.42, 63.64)	9.29 (2.70, 105.13)
Bisphenol-A	1 (0.54, 1.86)	2.21 (1.29, 3.50)	0.82 (0.48, 1.41)
MBP	13.2 (8.57, 24.00)	30.6 (17.13, 44.50)	11.54 (7.74, 18.00)
MBzP	5.5 (3.20,11.20)	11.51 (6.25, 20.62)	4.58 (2.94, 8.55)
MCPP	1.13 (0.55, 2.50)	3.15 (2.03, 4.57)	0.87 (0.46,1.60)
Σ DEHP	28.1 (17.88, 55.82)	92.5 (48.26, 253.23)	23.69 (16.45, 34.19)
MEP	41.6 (13.50, 128.70)	160.69 (71.30, 321.50)	24.55 (11.49, 69.87)
MiBP	5.61 (3.25, 9.46)	6.46 (4.45, 11.15)	5.46 (3.01, 8.67)
Σ Di-ethyl organophosphate	3.36 (1.75, 5.61)	2.47 (0.93, 6.44)	3.44 (1.98, 5.41)
Σ Di-methyl organophosphate	7.76 (3.49, 15.94)	6.02 (2.05, 20.19)	7.86 (3.74, 15.18)
Butyl paraben	0.35 (0.06, 3.25)	0.6 (0.11, 5.30)	0.32 (0.05, 2.44)
Methyl paraben	83.78 (28.16, 214.50)	227.46 (95.05, 417.56)	64.03 (19.07, 157.56)
Propyl paraben	20.37 (4.78, 65.68)	52.44 (14.06, 123.55)	15.93 (3.42, 51.73)
FSIQ	108 (96, 115)	104 (91, 112)	108 (98, 117)
PIQ	103 (94, 112)	103 (93, 114)	103 (94, 112)
VIQ	108 (100, 116)	101 (88, 111)	111 (100, 119)

The above table summarizes using the median and the interquartile ranges of the biomarkers and outcomes included in this study. Above chemical biomarkers were measured as ng/mL, except POPs measured as ng/g of lipid.

3.1. Joint associations between chemical biomarkers and cognitive outcomes

Using the quantile g-computation method, we observed several positive associations that did not reach statistical significance between the gestational chemical biomarker mixture and cognitive outcomes among children ages 3 and 5 years (Table 2 & Fig. S5). We observed statistically non-significant trends while stratifying the results by cohort (Table S3). Estimates from the MIREC Study showed a negative trend for PIQ scores. Child sex did not modify these associations, but our findings from sex-specific associations using quantile g-computation uncovered interesting trends. Notably, we observed negative trends between the gestational chemical biomarker mixture and cognitive outcomes (FSIQ −0.64, 95 %CI: −2.59, 1.31 and PIQ −1.59, 95 % CI: −3.72, 0.54) among males. Conversely, females exhibited a positive trend, suggesting the potential for divergent trends by child sex. Additionally, employing the BKMR approach, we found overall null associations between chemical biomarker mixtures (Fig. S6 & Table S4). However, when we stratified our analyses by child sex, we observed a negative trend between chemical biomarkers and cognitive outcomes (FSIQ and VIQ) among males. Our findings from the quantile g-computation method align with the BKMR analysis, showing a negative trend between chemical biomarker mixture and FSIQ score among male children, and a positive trend between chemical biomarkers and cognitive outcomes among females. Additionally, we observed similar trends between qgcomp and BKMR approaches, where PBDE-47, Perfluorohexanesulfonate (PFHxS), and di-ethyl organophosphates contributed toward a decline in FSIQ scores (Fig. S7).

Table 2.

Association between 29 chemical biomarkers and child cognitive outcomes using the quantile g-computation method.

	FSIQ			PIQ			VIQ
	All	Female	Male	All	Female	Male	All	Female	Male
Overall bootstrap effect (Ψ, 95 % CI)	0.20 (−1.25, 1.66)	0.37 (−1.64, 2.38)	−0.64 (−2.59, 1.31)	−0.27 (−1.85, 1.31)	0.77 (−1.36, 2.90)	−1.59 (−3.72, 0.54)	0.42 (−0.89, 1.73)	0.11 (−1.82, 2.03)	−0.08 (−2.11, 1.95)
Child sex-effect modifier p-value for trend (bootstrap)	0.44			0.51			0.54
Overall effect (Ψ, 95 % CI) non-bootstrap	0.20 (−1.16, 1.56)	0.37 (−1.46, 2.20)	−0.64 (−2.81, 1.52)	−0.27 (−1.82, 1.28)	0.77 (−1.35, 2.89)	−1.59 (−3.99, 0.81)	0.42 (−0.89, 1.73)	0.11 (−1.65, 1.86)	−0.08 (−2.18, 2.02)
Child sex-effect modifier p-value for trend (non-bootstrap)	0.39			0.47			0.52
Scaled negative effect	−3.86	−4.66	−5.01	−4.61	−4.61	−6.67	−2.83	−3.97	−4.78
Scaled positive effect	4.07	5.03	4.37	4.34	5.37	5.08	3.25	4.07	4.71
Pb	0	−0.05	0.07	−0.02	−0.04	0.02	0	−0.06	0.07
Hg	0.06	−0.02	0.08	0.11	0	0.15	0	−0.03	0.01
DMA	0.03	0.06	0.01	0.04	0.08	−0.01	0.01	0.03	0.02
DDE	0.07	0.12	−0.02	0.07	0.07	0.03	0.03	0.15	−0.07
PBDE47	−0.12	−0.09	−0.15	−0.09	−0.08	−0.11	−0.1	−0.07	−0.1
PCB118	0.06	0.06	0.06	0.08	0.09	0.09	−0.02	0	−0.03
PCB138	−0.27	−0.18	−0.12	−0.25	−0.28	0.08	−0.26	−0.14	−0.27
PCB153	−0.14	−0.3	−0.1	−0.08	−0.17	−0.29	−0.16	−0.3	0.22
PCB180	0.26	0.29	0.18	0.19	0.2	0.22	0.3	0.33	0.03
PFHxS	−0.11	−0.04	−0.14	−0.11	−0.04	−0.12	−0.07	0	−0.08
PFOA	0.08	0.1	0.01	0.05	0.16	−0.07	0.09	−0.01	0.07
PFOS	0.11	0.12	0.07	0.08	0.12	−0.01	0.13	0.1	0.12
BCEP	0.03	0	0.02	0.07	0.05	0.04	−0.05	−0.07	−0.03
BDCIPP	0.01	0.01	0.02	−0.01	0.02	−0.02	0.02	−0.01	0.05
DNBP	0.06	−0.01	0.14	0.06	0.01	0.1	0.05	−0.01	0.1
DPHP	0.02	0.04	0.01	−0.02	0.01	−0.01	0.07	0.08	0.01
Triclosan	−0.01	0.01	−0.04	−0.03	−0.03	−0.02	0.03	0.06	−0.04
Bisphenol-A	−0.02	0	−0.02	−0.02	0	−0.02	−0.03	−0.02	−0.02
MBP	0	−0.05	−0.03	−0.04	−0.06	−0.03	0	−0.04	−0.05
MBzP	0.05	0.04	0.04	0.09	0.06	0.1	0	0.03	−0.03
MCPP	−0.15	−0.11	−0.1	−0.14	−0.12	−0.12	−0.11	−0.06	−0.02
Σ DEHP	−0.01	0	−0.06	0	0.01	−0.02	0	−0.02	−0.04
MEP	0.02	−0.05	0.07	0	−0.04	0.03	0.04	−0.05	0.09
MiBP	0.04	0.08	0.04	0.03	0.07	0.01	0.07	0.08	0.07
Di-ethyl organophosphate	−0.04	0.01	−0.08	−0.01	−0.01	−0.01	−0.1	0.01	−0.12
Di-methyl organophosphate	0.01	0.03	−0.01	−0.02	0.01	−0.03	0.08	0.08	0.03
Butyl paraben	0.06	0.02	0.07	0.03	−0.02	0.05	0.07	0.05	0.06
Methyl paraben	0.05	−0.01	0.11	0.1	0.06	0.08	−0.03	−0.07	0.05
Propyl paraben	−0.13	−0.07	−0.14	−0.14	−0.11	−0.1	−0.06	−0.04	−0.08

Results presented in this table are generated using non-bootstrap quantile g-computation. Numeric values for negative biomarker contributions were multiplied by −1 for presentation purposes. We considered 0.07 (2/29 = 0.069) as a threshold for highlighting the chemical biomarker weights. The interpretation of the regression coefficients is the change in cognitive score for every quartile increase in chemical biomarker mixture. Regression estimates were adjusted for child sex (while analyzing male and females combined), caregiving environment, maternal race, education, age at delivery, parity, prenatal cotinine concentrations, fish consumption, city, and cohort.

Findings from the quantile g-computation method suggest that each quartile increase in the chemical biomarker mixture was associated with a 1.59-point lower PIQ score among males (95 % CI: −3.72, 0.54). Among the 29 chemical biomarkers included in this study, the observed negative trend with PIQ among male children was influenced by persistent organic pollutants [such as PBDE47 and 2,2′,4,4′,5,5′-Hexachlorobiphenyl (PCB153)], perfluoroalkyl substances [PFHxS, Perfluorooctanoate (PFOA)], mono(3-carboxypropyl) phthalate (MCPP), and propyl paraben (Table 2). Whereas, Hg, 2,3′,4,4′,5-pentachlorobiphenyl (PCB118), 2,2′,3,4,4′,5′-Hexachlorobiphenyl (PCB138), 2,2′,3,4,4′,5,5′-heptachlorobiphenyl (PCB180), di-n-butyl phosphate (DNBP), mono-benzyl phthalate (MBzP), and methyl paraben positively contributed to the joint association, counterbalancing the negative impacts of other biomarkers.

Our findings from the sensitivity analysis analyzing the joint associations between 29 gestational chemical biomarkers and child cognitive outcomes, suggested non-significant associations (Table S5). Among the cognitive outcomes, we observed a relatively strong negative trend between 29 chemical biomarker mixture and PIQ (PIQ −1.03, 95 %CI: −4.95, 2.88) while excluding maternal fish intake as a covariate and excluding children with cognitive outcomes measured using WISC, compared to our primary findings (PIQ −0.27, 95 %CI: −1.85, 1.31) (Table 2).

3.2. Single chemical biomarker associations

Among males, we observed similar trends for PIQ while comparing the negative contributors (PBDE47, PFHxS, MBP, MCPP, and propyl paraben) from the quantile g-computation method to the effect estimates from the multiple linear regression approach (Table S6). Overall, in our single chemical analysis, we observed that chemical biomarkers were differentially associated with cognitive scores, with variations in direction across sexes. We observed notable differences among PFAS in their trends of association.

3.3. Joint associations using a subset of chemical biomarkers

We identified a distinct subset of chemical biomarkers by cognitive outcome and/or child sex, relying on the single chemical associations at an alpha level of 0.2. Upon analyzing associations between these subsets of chemical biomarkers and cognitive outcomes, we noted an overall (males and females combined) negative trend between chemical biomarkers and cognitive outcomes (FSIQ −0.46, 95 %CI: −1.28, 0.36 and PIQ −0.68, 95 %CI: −1.48, 0.13), as depicted in Table 3 and Fig. S7. Conversely, VIQ displayed a positive trend (VIQ 0.18, 95 %CI: −0.57, 0.92).

Table 3.

Joint association between select chemical biomarkers and cognitive outcomes using quantile g-computation method.

	FSIQ			PIQ			VIQ
	All (n = 718)	Female (n = 382)	Male (n = 346)	All (n = 728)	Female (n = 388)	Male (n = 333)	All (n = 738)	Female (n = 396)	Male (n = 370)
Biomarkers included	6	6	5	5	5	9	8	3	2
Overall effect (Ψ, 95 % CI) bootstrap	−0.46 (−1.28, 0.36)	0.15 (−0.88, 1.17)	−1.29 (−2.46, −0.12)	−0.68 (−1.48, 0.13)	0.78 (−0.26, 1.82)	−1.56 (−3.08, −0.05)	0.18 (−0.57, 0.92)	−0.23 (−1.11, 0.64)	−0.13 (−0.81, 0.54)
Overall effect (Ψ, 95 % CI) non-bootstrap	−0.46 (−1.27, 0.35)	0.15 (−0.84, 1.14)	−1.29 (−2.36, −0.22)	−0.68 (−1.49, 0.14)	0.78 (−0.35, 1.90)	−1.56 (−3.08, −0.04)	0.18 (−0.61, 0.96)	−0.24 (−1.05, 0.58)	−0.13 (−0.83, 0.56)
Scaled negative effect	−0.97	−1.86	−1.75	−1.16	−0.83	−2.3	−1.12	−0.25	−0.5
Scaled positive effect	0.5	2	0.46	0.48	1.61	0.74	1.29	0.01	0.37
Hg	– –	– –	– –	1	– –	0.97	– –	– –	– –
PBDE47	−0.36	– –	−0.29	−0.22	– –	−0.17	−0.25	– –	−1
PCB118	0.03	– –	– –	−0.14	– –	– –	−0.05	– –	– –
PCB138	−0.23	−0.72	– –	−0.08	−0.69	– –	−0.66	−0.19	– –
PCB153	– –	0.53	– –	– –	– –	– –	0.48	– –	– –
PFHxS	– –	−0.12	−0.23	– –	−0.31	−0.21	– –	– –	– –
PFOA	0.37	0.15	– –	– –	0.55	−0.11	−0.02	– –	– –
PFOS	– –	0.32	– –	– –	0.28	−0.14	0.3	– –	– –
BCEP	– –	– –	– –	– –	0.17	– –	– –	– –	– –
BDCIPP	– –	– –	– –	– –	– –	−0.01	– –	– –	– –
Bisphenol-A	– –	– –	– –	– –	– –	– –	−0.02	1	– –
MBP	– –	– –	– –	– –	– –	0.03	– –	– –	– –
MCPP	−0.41	– –	−0.25	−0.55	– –	−0.31	– –	– –	– –
MEP	– –	−0.16	– –	– –	– –	– –	– –	−0.81	– –
Σ Di-ethyl organophosphate	– –	– –	−0.22	– –	– –	– –	– –	– –	– –
Butyl paraben	0.59	– –	1	– –	– –	– –	0.22	– –	1
Propyl paraben	– –	– –	– –	– –	– –	−0.06	– –	– –	– –

This table contains numeric values for biomarker weights only if corresponding biomarkers are included in the joint association models. Regression estimates were adjusted for child sex (while analyzing male and females combined), caregiving environment, maternal race, education, age at delivery, parity, prenatal cotinine concentrations, fish consumption, city, and cohort.

Moreover, we noted negative associations between the chemical biomarker mixture and cognitive outcomes (FSIQ and PIQ) in males (Table 3 & Fig. S8). These findings indicate that a quartile increase in chemical biomarker mixture was associated with 1.29 points (95 % CI: −2.46, −0.12) lower FSIQ score in males. Similarly, a quartile increase in chemical biomarker mixture is associated with 1.56 points (95 % CI: −3.08, −0.05) lower PIQ scores in males. Notably, among these associations between chemical biomarkers and FSIQ/PIQ among male children, MCPP, PFHxS, and PBDE47 emerged as common contributors to the negative joint associations.

4. Discussion

In our pooled analysis of two North American prospective studies, we observed overall null associations between gestational chemical biomarkers and child cognitive outcomes at ages 3 and 5 years using joint association approaches (quantile g-computation and BKMR). While the consideration of child sex as an effect modifier did not achieve statistical significance, upon stratifying the analysis by child sex using g-computation, a negative trend emerged between the chemical biomarker mixture and cognitive abilities (FSIQ and PIQ scores) among males. Notably, chemical biomarkers such as PBDE47, PCB153, PFHxS, MCPP, and propyl paraben consistently contributed to this negative trend in FSIQ/PIQ scores among males. The BKMR approach revealed a negative trend between the chemical mixture and FSIQ/VIQ, among males. The chemical biomarkers that contributed the most included PBDE47, PCB118, PCB138, and MBP. Our findings from the chemical mixtures approach using quantile g-computation and BKMR aligned with the single chemical approach employing multiple linear regression. In contrast, we observed a positive trend between chemical biomarker mixture and cognitive outcomes among females using both the qgcomp and BKMR approaches.

Furthermore, our secondary analysis, considering a subset of chemical biomarkers, indicated negative associations between the chemical biomarker mixture and cognitive outcomes (FSIQ & PIQ) among males. Notably, PBDE47, PFHxS, and MCPP consistently within this subset contributed to the negative association observed between the biomarker mixture and FISQ/PIQ. Given that PIQ scores are predominantly derived from tasks such as block design, matrix reasoning, and picture concepts, we suspect that gestational exposure to chemical mixtures may influence specific brain regions, potentially leading to adverse cognitive outcomes involved in performing these tasks. In females, we observed diverging trends in the direction of associations, where we observed a positive trend for FSIQ and PIQ while using select chemical biomarkers.

Our findings agree with some of the existing literature that demonstrated negative associations between chemical biomarkers and cognitive abilities among males from the United States (CHAMACOS, Project Viva), Human Early Life Exposome study (multi-country Europe), Sweden (SELMA), Netherlands (Generation R), France (EDEN), South Korea (Environment and Development Cohort), and China (Sheyang Mini Birth Cohort Study) (Nakiwala et al., 2018; Tanner et al., 2020; Jusko et al., 2019; Thilakaratne et al., 2023; Guo et al., 2020; Lee et al., 2021; Brennan Kearns et al., 2024; Eskenazi et al., 2013). Most studies (except Tanner et al. & Guo et al.) limited prenatal exposure mixtures to a chemical class of endocrine-disrupting chemical (EDC) biomarkers (PFAS, organophosphates, phthalates, phenols, PCBs, metals) and suggested negative associations with child cognitive outcomes such as VIQ and PIQ among male children (Tanner et al., 2020). Tanner et al. investigated 26 EDC biomarkers as a prenatal exposure mixture, assessing their associations with cognitive outcomes through generalized weighted quantile sum regression. Assuming directional homogeneity of biomarkers, they noted negative contributions to cognitive outcomes among male children from phenols (bisphenol-F, bisphenol-A), phthalates (MEP, MBzP), PFAS (PFHxS, PFOA), DPHP, and 3,5,6-trichloro-2-pyridinol (Tanner et al., 2020). In contrast, Guo et al. focused on nine chemical biomarkers representing metals, pesticides, and phenols. Using the BKMR approach, their study reported statistically non-significant associations, highlighting that lead and BPA notably contributed to negative trends with FSIQ among males (Guo et al., 2020). Our investigation, focusing on 29 chemical biomarker mixtures and PIQ scores using the g-computation method, aligned with Tanner et al.’s findings, where PFHxS negatively contributed to cognitive outcomes among males. In contrast to the joint association method from Tanner et al., we relied on the quantile g-computation and BKMR approaches, allowing for the heterogeneous direction of the associations (Bobb et al., 2014; Keil et al., 2020). Among the studies that assessed associations between gestational exposure to chemical mixtures and cognitive outcomes, a few (Tanner et al., 2020; Thilakaratne et al., 2023; Guo et al., 2020) have presented these associations stratified by child sex and reported non-significant positive trends among girls that align with our findings.

Due to the lower detection frequencies of other PBDE congeners in the MIREC study, we limited the analysis by including only PBDE47 in the exposure mixture. We observed that PBDE47 consistently negatively contributed to cognitive outcomes (FSIQ, PIQ, and VIQ) in males and females. Correspondingly, the CHAMACOS Study indicated a negative association between prenatal (~26 weeks of gestation or delivery) exposure to PBDEs (sum of 10 congeners) and child cognitive abilities (VIQ) at age 7 years (Eskenazi et al., 2013).

Furthermore, we observed that bisphenol-A (except for FSIQ and PIQ in females), propyl paraben, MCPP, and MBP consistently exhibited weak contributions to cognitive outcomes (FISQ, PIQ, and VIQ) among males and females. Additionally, DEHP demonstrated a marginal negative trend with cognitive outcomes in males, while MEP exhibited a weak negative trend in females. Similar patterns were reported based on the participants included in the French EDEN cohort that included only males, where parabens, bisphenol-A, and phthalates [Monoethyl phthalate (MEP), Mono-n-butyl phthalate (MBP), and DEHP] showed negative trends with VIQ and PIQ scores (Nakiwala et al., 2018).

The chemical biomarkers identified as negative contributors in this study had been previously documented for their potential to cause endocrine-disrupting activities (Płotka-Wasylka et al., 2023). Previous studies have suggested potential underlying mechanisms where exposure to these disruptors might influence dopaminergic neuron development and modify hippocampal neuronal plasticity, potentially contributing to cognitive impairment (Masuo and Ishido, 2011; Nesan and Kurrasch, 2020). Notably, this study highlighted sex-related differences in outcomes, such as variations in locus coeruleus size, while exposed to BPA (Masuo and Ishido, 2011). Additionally, preclinical studies suggest gestational exposure to EDCs such as PCBs and BPA was associated with altered gene expression of steroid hormone receptors (Ar, Thra, Gper), neuropeptide (Gnrh1, Kiss1r), methyltransferase (Dnmt1), and clock genes (Arntl, Per1) in the arcuate nucleus of the hypothalamus among males and not in females (Rebuli and Patisaul, 2016). Endocrine disruptors were proposed to potentially interfere with hypothalamic neuroendocrine cells by modifying synaptic connectivity, altering neurotransmitter/neuropeptide expression, and affecting neuronal differentiation (Gore 2010; Patisaul and Polston, 2008; Parent et al., 2011). Furthermore, Ramirez et al. outlined several potential mechanisms involving endocrine-disrupting chemicals, including interactions with nuclear hormone receptors (estrogen, androgen, and thyroid) and alterations in neurotransmitters (dopamine, serotonin, glutamate, gamma-aminobutyric acid, acetylcholinesterase), that could result in adverse neurodevelopmental outcomes (Ramírez et al., 2022). These chemicals were highlighted to potentially alter nuclear hormone receptors even at low doses, which affected a range of outcomes, including brain development (Vandenberg et al., 2012).

Summarizing our findings, we observed negative trends between 29 chemical biomarkers and cognitive outcomes (FSIQ and PIQ) among males. Chemical biomarkers, including propyl paraben, MCPP, PFHxS, PCB153, and PBDE47, exhibited negative contributions to the joint associations with cognitive outcomes among males included in this study. This study has both strengths and limitations. Its strengths include incorporating a wide array of maternal chemical biomarkers measured during pregnancy, representing a range of endocrine disruptors. Pooling study participants from two longitudinal cohorts from the North American population with diverse exposure patterns and sociodemographic profiles will reduce the random error and improve the risk assessment for policy making. Patterns of concurrent environmental mixture exposure by pooling study participants across cohorts may represent real-world settings and improve the generalizability over a single-site study.

In light of the primary objective of this study, which is to evaluate the joint association between chemical biomarkers and cognitive outcomes, our analysis has been constrained to estimating controlled direct effects. Due to the pooled sampling design of this study, we were limited to including chemical biomarkers available in both the HOME and MIREC Studies. Moreover, biomarkers detected in less than half of the study participants were excluded. This could result in biased associations reported in this study due to the potential influence of unmeasured exposures and confounding factors. Furthermore, it is important to note that our estimates might be susceptible to bias due to exposure misclassification. This could be due to our focus on chemical biomarker exposures measured during early pregnancy instead of utilizing repeated exposure measurements. Compounding this, exposure assessments representing persistent and non-persistent chemical biomarkers as a mixture may introduce potential differential exposure misclassification bias. Furthermore, this study did not analyze nutritional factors [ex: folate, iodine, vitamin B12, iron, vitamin D, polyunsaturated fat (omega-3/-6), and choline] that may mediate certain associations between endocrine disruptors and child neurodevelopmental outcomes.

5. Conclusions

The joint association approach using 29 chemical biomarkers showed a negative trend with cognitive outcomes (FSIQ and PIQ scores) among males. These associations were notably driven by endocrine disruptors such as PBDE47, PCB153, PFHxS, MCPP, and propyl paraben. Further studies are needed to potentially explore the exposure-outcome associations using repeated exposure assessments to identify critical windows of susceptibility.

Abbreviations:

Pb

Lead

Hg

Mercury

DMA

Dimethylarsinic acid

DDE

Dichlorodiphenyldichloroethylene

PBDE47

2,2′4,4′-tetrabromodiphenyl ether

PCB118

2,3′,4,4′,5-Pentachlorobiphenyl

PCB138

2,2′,3,4,4′,5′-Hexachlorobiphenyl

PCB153

2,2′,4,4′,5,5′-Hexachlorobiphenyl

PCB180

2,2′,3,4,4′,5,5′-Heptachlorobiphenyl

PFHxS

Perfluorohexanesulfonate

PFOA

Perfluorooctanoate

PFOS

Perfluorooctanesulfonate

BCEP

bis2-chloroethyl phosphate

BDCIPP

bis1,3-dichloro-2-propyl phosphate

DNBP

di-n-butyl phosphate

DPHP

Diphenyl phosphate Triclosan Bisphenol-A

MBP

Mono-n-butyl phthalate

MBzP

Mono-benzyl phthalate

MCPP

Mono3-carboxypropyl phthalate

MEHHP

Mono 2-ethyl-5-hydroxyhexyl phthalate

MEHP

Mono-2-ethylhexyl phthalate

MEOHP

Mono-2-ethyl-5-oxohexyl phthalate

MCPP

Mono-2-ethyl-5-carboxypentyl phthalate

DEHP

di2-ethylhexyl phthalate

MEP

Monoethyl phthalate

MiBP

Monoisobutyl phthalate

DEP

Diethylphosphate

DMP

Dimethylphosphate

TC

total cholesterol

FC

free cholesterol

TG

triglycerides

PL

phospholipids

SG

specific gravity

WPPSI

Wechsler Preschool and Primary Scale of Intelligence

WISC

Wechsler Intelligence Scale for Children

FSIQ

Full scale intelligence quotient

PIQ

Performance Intelligence Quotient

VIQ

Visual Intelligence Quotient

HOME score

Home Observation for Measurement of the Environment score

HOME

Health Outcomes and Measures of the Environment

MIREC

Maternal-Infant Research on Environmental Chemicals

qgcomp

quantile g-computation

BKMR

Bayesian Kernel Machine Regression

PIPs

posterior inclusion probabilities

Data availability

The data that has been used is confidential.