Prenatal Exposure to Organophosphate Esters and Child Cognitive Development in the MARBLES Prospective Cohort

Abstract

Background:

Epidemiological studies on prenatal exposure to organophosphate esters (OPEs) and child cognitive development remain inconclusive. We examined associations between prenatal maternal urinary OPE biomarker concentrations and child cognitive development assessed at multiple time points.

Methods:

The study included 311 mother–child pairs from MARBLES, a prospective cohort of children with elevated familial likelihood of autism and at least one available cognitive assessment. We quantified nine OPE biomarkers in one maternal urine sample per participant, collected predominantly during the third trimester. Cognitive development was assessed at 6, 12, 24, and 36 months, with observations at each time point included in the primary analyses. We examined associations between specific gravity-adjusted urinary OPE biomarker concentrations, both individually and as a mixture, and MSEL Composite and subscale scores at each time point. We also evaluated child sex as an effect modifier and assessed longitudinal changes in scores across assessments.

Results:

Associations between individual prenatal OPE biomarkers and child cognitive outcomes were largely null after correction for multiple comparisons. Associations between bis(2-chloroethyl) phosphate (BCETP) concentrations and Composite and Visual Reception scores differed by age across assessments, with scores appearing higher at 36 months. Trajectory analyses identified no associations. Effect modification by child sex was not observed.

Conclusions:

We did not observe that prenatal OPE exposure was associated with cognitive development across infancy and toddlerhood; however, we observed that associations differed by age at assessment, highlighting the importance of repeated outcome measurements. Larger studies in general-population cohorts with repeated exposure measurements throughout pregnancy are warranted.

1. Introduction

Organophosphate esters (OPEs) are a group of chemical compounds conventionally used as flame retardants after the phaseout of brominated flame retardants but are also increasingly used as plasticizers in a wide range of consumer products, including electronics, food packaging, and children’s toys (Bi et al., 2023; Wei et al., 2015; Zhang et al., 2022). OPEs generally become less water-soluble as molecular mass increases, while their volatility varies considerably (Bekele et al., 2021). Higher-molecular-weight OPEs tend to accumulate in surface water, soil, and sediment, whereas lower-molecular-weight compounds are more likely to volatilize and partition into indoor dust. Because OPEs are commonly used as additive (i.e., non-covalently bound) chemicals, they can readily migrate from consumer products and building materials, contributing to their ubiquitous presence in various environmental media (Wei et al., 2015). Major human exposure pathways include inhalation of contaminated air, incidental dust ingestion, dermal contact, and dietary intake (Bekele et al., 2021; Wang & Kannan, 2018). In addition to their high detection frequencies (DFs) in urine samples of the general U.S. population (Ospina et al., 2018), OPEs have also been detected in maternal and fetal biospecimens, including amniotic fluid, placental tissue, and umbilical cord blood (Choi et al., 2024; Lian et al., 2025; Shahin et al., 2024).

Prenatal OPE exposure is concerning because experimental studies have reported developmental neurotoxicity following OPE exposure in animal models, including zebrafish embryos and rodents (Shi et al., 2018; Zhong et al., 2020). In vitro studies have also suggested that OPE exposure may induce cytotoxic effects, including increased apoptosis and disruption of neural network formation (Ta et al., 2014). Importantly, gestational exposure studies in rodents have also reported adverse neurodevelopmental effects in offspring, including neuroinflammation, oxidative stress, and disruption of dopaminergic pathways (Newell et al., 2023; Rock et al., 2020). Given that pregnancy represents a critical window of susceptibility to environmental exposures (Braun, 2017), these findings highlight the importance of examining whether prenatal OPE exposure influences child neurodevelopment.

Previous epidemiologic studies have examined associations between prenatal OPE exposure and cognitive (Castorina et al., 2017; Doherty et al., 2019; Ghassabian et al., 2025; Liu et al., 2021), neurobehavioral (Castorina et al., 2017; Oh et al., 2025), and broader neurodevelopmental outcomes (Cheng et al., 2024; Hall et al., 2023; Hernandez-Castro et al., 2024; Ramos et al., 2023; Vuong et al., 2024; Zhou et al., 2024) in children across infancy and toddlerhood. Several studies have reported significant adverse associations between maternal prenatal urinary OPE concentrations and child neurodevelopmental outcomes (Cheng et al., 2024; Doherty et al., 2019; Hernandez-Castro et al., 2024; Liu et al., 2021; Ramos et al., 2023). To our knowledge, the only previous study focused explicitly on child cognitive function as the outcome (Doherty et al., 2019) reported adverse effects on child cognitive function from maternal exposure to isopropyl-phenyl phenyl phosphate (ip-PPP), with little to no effects from other OPEs. Additionally, only one prior study has evaluated longitudinal changes in early neurodevelopmental outcomes across multiple time points in relation to maternal prenatal urinary OPE concentrations (Hernandez-Castro et al., 2024), and longitudinal evidence specifically focused on cognitive function remains limited. Furthermore, most previous studies have assessed cognitive or neurodevelopmental outcomes at a single time point, limiting understanding of whether prenatal OPE exposure influences developmental trajectories across infancy and early childhood. Given that cognitive function undergoes dynamic changes during early life and may be particularly sensitive to prenatal chemical exposures (Oh et al., 2021), further investigation of its longitudinal trajectories is warranted. Because OPEs are widespread environmental contaminants and human exposure is nearly ubiquitous, even subtle neurodevelopmental effects during sensitive developmental windows could have important environmental health implications at the population level.

Several epidemiologic studies have suggested that child sex may modify associations between prenatal OPE exposure and neurodevelopment, although findings remain inconsistent. Some studies reported no evidence of sex-specific effects (Castorina et al., 2017; Percy et al., 2021; Ramos et al., 2023), whereas others found that either males or females may be more susceptible depending on the specific OPE biomarker and neurodevelopmental outcome assessed (Hall et al., 2023; Hernandez-Castro et al., 2024; Liu et al., 2021; Oh et al., 2025; Vuong et al., 2024; Zhou et al., 2024). Among studies evaluating cognitive outcomes specifically, findings have also been mixed, with some reporting adverse associations in males, others in females, and one reporting no sex differences (Castorina et al., 2017; Doherty et al., 2019; Ghassabian et al., 2025; Liu et al., 2021). To our knowledge, only one prior study has evaluated sex-specific associations between prenatal OPE exposure and child neurodevelopment across multiple time points (Hernandez-Castro et al., 2024). Therefore, further investigation of potential effect modification by child sex is warranted.

The present study aims to investigate associations between prenatal OPE exposure and cognitive function in children assessed at four time points during infancy and early childhood (i.e., 6, 12, 24, 36 months). We leveraged data from the MARBLES (Markers of Autism Risk in Babies – Learning Early Signs) prospective pregnancy cohort to evaluate both individual OPE biomarker concentrations and OPE mixtures. We also examined potential effect modification by child sex at each time point and evaluated longitudinal changes and developmental trajectories in cognitive function across the four assessment time points in association with prenatal OPE exposure. Because the MARBLES cohort comprises children with an elevated familial likelihood of autism, this study investigated these associations in a unique population across a range of cognitive developmental trajectories.

2. Methods

2.1. Study population

Our study includes participants from the MARBLES (Markers of Autism Risk in Babies – Learning Early Signs) study (Hertz-Picciotto et al., 2018). Launched in 2006, the MARBLES study aims to identify environmental factors that contribute to autism risk, investigate potential biological mechanisms, and identify early biomarkers of elevated risk (Hertz-Picciotto et al., 2018). MARBLES recruited pregnant women who had previously given birth to a child diagnosed with autism; therefore, their subsequent children had an elevated familial likelihood of autism (~20%) and other neurodevelopmental concerns compared with children in the general population (Hertz-Picciotto et al., 2018; Ozonoff et al., 2014). Most participants were recruited from lists of families who received state-funded services for a child with autism. Eligibility criteria for MARBLES enrollment included: (a) having one or more children with autism; (b) being at least 18 years old; (c) ability to speak, read, and understand English; and (d) living within 2.5 h of the Davis/Sacramento region. Study protocols were approved by the Institutional Review Boards of the University of California Davis (UC Davis) and the State of California. All data were collected with the informed consent of participants. Details of the study design, recruitment methods, and data collection are available elsewhere (Hertz-Picciotto et al., 2018).

For this study, we excluded mother-child pairs without prenatal urinary OPE biomarker measurements (n = 73) and those without at least one cognitive assessment at any time point (n = 3). Therefore, 311 mother-child pairs were included in the baseline study population. Mothers in the baseline population provided urine samples during pregnancy between 2007 and 2019, and their children completed an autism evaluation and at least one cognitive assessment (Figure S1). Of these, 169 children who completed cognitive assessments at all four time points (6, 12, 24, and 36 months of age) were included as the restricted complete-case sample in the mixture and sensitivity analyses to ensure that estimates at each assessment reflected the same study population and to maintain consistency across our analyses. Mothers in the restricted complete-case sample provided urine samples during pregnancy between 2007 and 2016.

2.2. Child cognitive function assessment

Licensed clinical psychologists at the UC Davis Medical Investigations of Neurodevelopmental Disorders (MIND) Institute conducted clinical evaluations to determine autism diagnosis at 36 months of age for all children included in this study. Children were classified into one of three diagnostic groups: autism, non-typical development (Non-TD), and typical development (TD), using standardized algorithmic procedures described elsewhere (Hertz-Picciotto et al., 2018). Separately, MARBLES staff, who were trained by senior clinicians, administered the Mullen Scales of Early Learning (MSEL) during home visits at 6 and 12 months of age and during visits to the UC Davis MIND Institute at 24 and 36 months of age. Assessors were unaware of maternal OPE biomarker measurements at the time of the MSEL assessments because urinary OPE biomarker quantification was conducted subsequently. The MSEL is a performance-based measure designed to assess cognitive abilities in infants and children up to 68 months of age and was used to evaluate cognitive development in infancy and toddlerhood. The MSEL includes four subscales used to assess cognitive functions: Fine Motor, Visual Reception, Receptive Language, and Expressive Language. Raw scores of the four subscales were converted to age-standardized T-scores, with a mean of 50 and a standard deviation (SD) of 10. MSEL T-scores are standardized by child age but not by sex. An Early Learning Composite (Composite) score was calculated by combining the four subscales at the respective time points, with a mean of 100 and a SD of 15 (Swineford et al., 2015; Turner-Brown et al., 2013). Higher Composite and subscale scores indicate better cognitive functioning.

2.3. Urine sample collection and OPE biomarker quantification

Each pregnant mother participating in the MARBLES study was instructed to collect urine samples once per week for four consecutive weeks in each trimester (Hertz-Picciotto et al., 2018). OPE biomarkers were quantified in one maternal urine sample per participant as part of a larger OPE biomonitoring effort (Oh et al., 2025). Although MARBLES participants provided repeated urine samples during pregnancy, samples closest to the first day of the third trimester with sufficient remaining volume in the biorepository were prioritized for OPE analysis. This approach provided relatively consistent sample collection timing across participants and maximized the number of participants with available OPE measurements within the larger biomonitoring effort. Among the 311 mother-child pairs with available MSEL assessment data, 27 samples (8.7%) were collected during the second trimester and 284 samples (91.3%) were collected during the third trimester (mean gestational age at collection: 30.3 weeks; interquartile range [IQR]: 28.3–32.5 weeks). Participants collected samples and stored them in their home freezers until the staff came to retrieve them. Urine samples were then brought to the laboratory at UC Davis, aliquoted, and then stored at −80˚C until analysis. Details of urine sample collection, transport, and storage are described elsewhere (Hertz-Picciotto et al., 2018).

Urine samples were analyzed at the Wadsworth Center-Human Health Exposure Analysis Resources (HHEAR) Laboratory for nine OPE biomarkers: dipropyl phosphate (DPRP), the sum of di-n-butyl phosphate and di-iso-butyl phosphate (DBUP+DIBP), bis(butoxyethyl) phosphate (BBOEP), bis(2-ethylhexyl) phosphate (BEHP), bis(2-chloroethyl) phosphate (BCETP), bis(1-chloro-2-propyl) phosphate (BCPP), bis(1,3-dichloro-2-propyl) phosphate (BDCPP), diphenyl phosphate (DPHP), and bis(2-methylphenyl) phosphate (BMPP). DBUP and DIBP, major metabolites of tri-n-butyl phosphate (TNBP) and tris(isobutyl) phosphate (TIBP), respectively, were quantified as the sum of their concentrations due to poor chromatographic separation and co-elution of their peaks. Parent compounds for all biomarkers are provided in Table 2. The analytical methods for OPE biomarker quantification are described elsewhere (Oh et al., 2024). Quantification of the target chemicals was performed utilizing high-performance liquid chromatography (HPLC; ExionLC^™ system, SCIEX), coupled with an AB SCIEX QTRAP 5500+ triple quadruple mass spectrometry (Applied Biosystems). Every sample batch included reagent blanks, matrix blanks, and matrix-spiked samples. Replicates of HHEAR Urine Quality Control (QC) Pools Standard Reference Materials (SRM3672 and SRM3673; National Institute of Standards and Technology) were analyzed along with sample batches. The limits of detection (LODs) ranged from 0.012 to 0.044 ng/mL for target analytes.

Table 2.

Distributions of SG-corrected OPE biomarker concentrations (ng/mL) in 311 maternal urine samples collected 2007–2019.

Parent compound	Biomarker	LOD (ng/mL)	% detect	GM (ng/mL)	Percentiles (ng/mL)
					5th	25th	50th	75th	95th
TPP	DPRP	0.03	33.44	0.03	<LOD	<LOD	<LOD	0.04	0.23
TNBP + TIBP	DBUP + DIBP	0.04	98.07	0.16	0.06	0.10	0.14	0.23	0.76
TBOEP	BBOEP	0.02	69.13	0.04	<LOD	<LOD	0.04	0.07	0.25
TEHP	BEHP	0.02	37.94	0.03	<LOD	<LOD	<LOD	0.05	0.46
TCEP	BCETP	0.02	83.92	0.38	<LOD	0.18	0.57	1.27	6.33
TCIPP	BCPP	0.02	73.31	0.20	<LOD	<LOD	0.37	0.95	4.03
TDCIPP	BDCPP	0.02	95.50	0.72	0.04	0.35	0.86	1.75	5.57
TPHP	DPHP	0.03	100.00	0.92	0.30	0.54	0.79	1.43	3.82
TMPP	BMPP	0.01	56.91	0.02	<LOD	<LOD	0.01	0.03	0.13

Abbreviations: bis(butoxyethyl) phosphate (BBOEP), bis(2-chloroethyl) phosphate (BCETP), bis(1-chloro-2-propyl) phosphate (BCPP), bis(1,3-dichloro-2-propyl) phosphate (BDCPP), bis(2- ethylhexyl) phosphate (BEHP), bis(2- methylphenyl) phosphate (BMPP), sum of di-n-butyl phosphate and di-iso-butyl phosphate (DBUP+DIBP), diphenyl phosphate (DPHP), dipropyl phosphate (DPRP), geometric mean (GM), limit of detection (LOD), tris(2-butoxyethyl) phosphate (TBOEP), tri(2-chloroethyl) phosphate (TCEP), tris(1-chloro-2-propyl) phosphate (TCIPP), tris(1,3-dichloro-2-propyl) phosphate (TDCIPP), tris(2-ethylhexyl) phosphate (TEHP), trimethylphenyl phosphate (TMPP), sum of tri-n-butyl phosphate and tris(isobutyl) phosphate (TNBP+TIBP), triphenyl phosphate (TPHP), tripropyl phosphate (TPP)

Note: Some biomarkers may have multiple parent compounds.

Concentrations below the LOD were imputed as LOD/√2 (Hornung & Reed, 1990). To account for urinary dilution, specific gravity (SG) was measured using a digital handheld refractometer (ATAGO Co., Ltd., Tokyo, Japan) (Adibi et al., 2008; Hauser et al., 2004). OPE concentrations were corrected for urinary dilution using the equation: C_SG = C×[(1.012–1)/(SG–1)], where C_SG is the SG-corrected biomarker concentration (in ng/mL), C is the measured biomarker concentration in urine (in ng/mL), 1.012 represents the median SG of all analyzed samples, and SG is the specific gravity of each sample.

2.4. Covariates

A directed acyclic graph (DAG) was constructed a priori to identify potential confounders, risk factors for the outcome, and mediators, based on variables identified in previous literature (Doherty et al., 2019; Oh et al., 2021) (Figure S2). Covariates considered and included in the models were child sex (male, female), year of birth (2007–2010, 2011–2014, 2015–2017), maternal age at delivery (<35 years, ≥35 years), parity (≤1, >1), maternal pre-pregnancy body mass index (BMI; <25.0, ≥25.0 kg/m²), maternal race/ethnicity (non-Hispanic White, Hispanic/Asian/Black/multiracial), maternal education (less than bachelor’s degree, bachelor’s degree or higher), and homeownership (yes, no). A minimal sufficient adjustment set was identified using Dagitty (Textor et al., 2016; van der Zander et al., 2014): child year of birth, maternal age at delivery, parity, maternal pre-pregnancy BMI, maternal race/ethnicity, maternal education, and homeownership. We adjusted for the minimal sufficient adjustment set and child sex as a precision variable, associated with MSEL Composite scores at one or more time point (p < 0.05) (Table 1). We excluded child autism diagnostic classification because it could be on a causal pathway. Potential mediators, such as gestational age at delivery, were not included in the models.

Table 1.

Characteristics of the mother-child pair participants for all MARBLES participants in the present study.

N (%) or median [IQR]	Children with complete MSEL assessments	Baseline population	p-valuec
	(n = 169)	(n = 311)
Child’s sex			1.00
Male	96 (57)	177 (57)
Female	73 (43)	134 (43)
Year of birth			0.20
2007 – 2010a	62 (37)	104 (33)
2011 – 2014	56 (33)	88 (28)
2015 – 2019a	51 (30)	119 (38)
Maternal age at delivery			0.55
< 35 years	83 (49)	163 (52)
≥ 35 years	86 (51)	148 (48)
Parity			0.77
≤ 1	74 (44)	130 (42)
> 1	93 (55)	176 (57)
Missing	2 (1)	5 (2)
Maternal pre-pregnancy BMI (kg/m2)			0.96
< 25.0	71 (42)	128 (41)
≥ 25.0	98 (58)	182 (59)
Missing	0 (0)	1 (0)
Maternal race/ethnicity			0.52
Non-Hispanic white	75 (44)	149 (48)
Hispanic and othersb	94 (56)	162 (52)
Maternal education			0.20
Less than bachelor’s degree	75 (44)	158 (51)
More than or equal to bachelor’s degree	94 (56)	152 (49)
Missing	0 (0)	1 (0)
Homeownership			0.51
Yes	106 (63)	182 (59)
No	60 (36)	120 (39)
Missing	3 (2)	9 (3)
Child autism outcome group			0.11
TD	115 (68)	163 (52)
Non-TD	24 (14)	43 (14)
Autism	30 (18)	71 (23)
Missing	0 (0)	34 (11)
Maternal urinary OPE biomarker concentrations (ng/mL)
BBOEP	0.04 [0.02–0.07]	0.04 [0.02–0.07]	0.78
BCETP	0.67 [0.20–1.32]	0.57 [0.18–1.27]	0.35
BCPP	0.48 [0.02–0.98]	0.37 [0.02–0.95]	0.25
BDCPP	0.87 [0.30–1.60]	0.86 [0.35–1.75]	0.52
DBUP+DIBP	0.15 [0.10–0.22]	0.14 [0.10–0.23]	0.94
DPHP	0.90 [0.55–1.56]	0.79 [0.54–1.43]	0.60
BEHP	0.01 [0.01–0.04]	0.01 [0.01–0.05]	0.82
BMPP	0.01 [0.01–0.03]	0.01 [0.01–0.03]	0.43
DPRP	0.02 [0.02–0.04]	0.02 [0.01–0.05]	0.56
MSEL T-scores at the 6th month assessment point
Composite	98 [91–107]	98 [91–107]	0.99
Fine Motor	46 [46–53]	46 [46–53]	0.47
Visual Reception	51 [45–58]	51 [45–58]	0.89
Receptive Language	50 [44–57]	50 [44–58]	0.78
Expressive Language	42 [42–48]	42 [42–48]	0.51

^aChildren with complete MSEL assessments (n = 169) were born between 2008 and 2017.

^bIncludes Black/African American, American Indian/Alaska Native, Asian, Native Hawaiian/Other Pacific Islander, Other, and more than one race.

^cP-value from the chi-squared test or Wilcoxon rank-sum test.

Abbreviations: body mass index (BMI), Mullen Scales of Early Learning (MSEL), non-typical development (non-TD), typical development (TD)

2.5. Statistical analysis

We compared maternal sociodemographic characteristics, maternal urinary OPE biomarker concentrations, and MSEL T-scores at the earliest assessment (6 months) between the baseline population (n = 311) and the complete-case sample with MSEL assessments at all four time points (n = 169) to evaluate potential differential loss to follow-up using chi-squared or Wilcoxon rank-sum tests. Differences in mean MSEL Composite scores across demographic characteristics at each time point were assessed using Student’s t-test or ANOVA. We displayed distributions of MSEL Composite scores and T-scores of four subscales at 6, 12, 24, and 36 months of age using boxplots and assessed Composite score mean differences across time using a linear mixed-effects model. Pairwise correlations among six SG-corrected OPE biomarker concentrations detected in ≥60% of samples were presented by Spearman’s correlation coefficients.

SG-corrected OPE biomarker concentrations were categorized into two groups based on DFs and modeled separately. OPE biomarker concentrations with DFs ≥60% (BBOEP, BCETP, BCPP, BDCPP, DBUP+DIBP, DPHP) were modeled with continuous log2-transformed concentrations because their distributions were right-skewed. OPE biomarkers with DFs <60% were dichotomized and modeled as binary variables: detect (≥LOD) versus non-detect (<LOD) (Choi et al., 2024).

First, we used multiple linear regression models to examine associations between individual prenatal urinary OPE biomarker concentrations and MSEL Composite and subscale T-scores at each of the four assessment time points, adjusting for all covariates. Multiple imputation by chained equations with 20 imputations, based on exposure, outcomes, and covariates (van Buuren & Groothuis-Oudshoorn, 2011) was performed to impute two covariates with missing information: parity (1%) and homeownership (2%). Parameter estimates were pooled across imputed datasets using Rubin’s rules, which combine the calculated estimates while accounting for variance from the imputation (Barnard & Rubin, 1999; Lee et al., 2025; Rubin, 1996). To test the significance of the derived associations, false discovery rate (FDR)-adjusted p-values were calculated using the Benjamini-Hochberg method (Benjamini & Hochberg, 2018). Given previous evidence of sex-modified associations between prenatal OPE exposure and child cognitive development (Hernandez-Castro et al., 2024; Liu et al., 2021), we assessed effect modification by child sex on Composite scores by conducting a stratified analysis. Regression models were refitted and adjusted for all covariates except child sex, which served as the effect modifier. Biomarkers with DFs <60% were excluded from the effect modification analyses.

Second, we conducted mixture analyses at each assessment time point using quantile g-computation to estimate the combined effects of prenatal OPE exposures on MSEL Composite and subscale T-scores. We restricted the mixture analyses to the complete-case sample and biomarkers detected in ≥60% of the samples and categorized the concentrations into tertiles rather than quartiles to reduce potential misclassification caused by tied values among observations below the LOD, particularly for BBOEP (64.9%) and BCPP (68.0%). Quantile g-computation is a parametric method based on a generalized linear model-based implementation of g-computation that estimates the effects for one-quantile increases of multiple exposures on the outcome (Keil et al., 2020). The overall effect results of the quantile g-computation were obtained as pooled estimates across all imputed datasets, represented by psi (Ψ), and individual biomarker weights were estimated to quantify their relative contributions to the mixture effect. Because quantile g-computation assumes linear and additive relationships, we performed an additional sensitivity analysis using Bayesian Kernel Machine Regression (BKMR) to evaluate potential non-linear exposure-response relationships and interactions among OPE biomarkers. BKMR models were fitted using Markov Chain Monte Carlo with multiple parallel chains and 5,000 iterations for each imputed dataset. The resulting exposure-response functions were pooled at each exposure concentration using Rubin’s rules.

Third, we evaluated longitudinal changes in MSEL Composite and four subscale T-scores across the four assessment time points in relation to prenatal maternal urinary OPE biomarker concentrations using population-averaged generalized estimating equations (GEE) models with an autoregressive (AR-1) working correlation structure (Ballinger, 2004; Liang & Zeger, 1986). Age at assessment (6, 12, 24, and 36 months) was modeled as a categorical variable and included as a main effect, along with an interaction term between age and log2-transformed OPE biomarker concentrations to assess whether associations varied across assessment visits. Overall interaction p-values were obtained using the D1 statistic, a pooled multivariate Wald test comparing nested models with and without the exposure-by-age interaction terms across multiply imputed datasets (Li et al., 1991). The “waves” option was used to correctly specify the ordering of repeated assessments (Halekoh et al., 2006). We ran a sensitivity analysis using the restricted complete-case sample, which excluded 142 children that did not have complete MSEL assessments at 6, 12, 24, and 36 months of age. To evaluate the potential impact of unequal time intervals between assessments, we conducted a sensitivity analysis using an unstructured correlation matrix. Although GEE models provide population-averaged estimates across repeated assessments, such models may obscure heterogeneous developmental patterns across children. Therefore, as an exploratory analysis, we used latent class mixed models (LCMM) to identify distinct developmental trajectories of MSEL Composite scores across the four assessment time points (Nagin, 2005; Oh et al., 2021). Competing class solutions were evaluated using the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), sample-adjusted BIC (SABIC), entropy, and class size (Table S1). Although the three-class solution yielded more favorable information criteria and slightly higher entropy, one class contained a relatively small group of children (n = 28). Given the modest sample size and limited precision expected for subsequent exposure–trajectory analyses involving this small class, we selected the more parsimonious two-class solution. The resulting high-score and low-score groups represented children with relatively higher versus lower developmental trajectories, respectively. Poisson regression models with robust error variance, adjusted for all covariates, were then used to estimate relative risks (RRs) of belonging to the low-score versus high-score trajectory groups in relation to prenatal OPE exposure.

Finally, we performed several sensitivity analyses to evaluate the robustness of our findings. First, to evaluate the robustness of the findings, we refitted the models using the restricted complete-case sample of 169 children with MSEL assessments at all four time points. To evaluate the potential influence of gestational age at delivery on our findings, we performed a sensitivity analysis that additionally adjusted for gestational age at delivery in the multiple linear regression models. Next, we performed additional sensitivity analyses for both the multiple linear regression and quantile g-computation models restricted to children classified as TD, as our study population included children with neurodevelopmental concerns (i.e., autism and Non-TD). Models were refitted and adjusted for all covariates.

All statistical analyses were performed using R version 4.3.0 (R Core Team, https://www.R-project.org). The “mice” R package was used for multiple imputation, the “qgcomp” package for quantile g-computation, and the “bkmr” package for BKMR. The significance level was set at α = 0.05 for most statistical analyses, except for FDR-corrected p-values, which were set at α = 0.10.

3. Results

3.1. Participant characteristics and MSEL composite and subscale T-scores

The final analytic baseline population comprised 311 mother-child pairs, of whom 57% of children were male (Table 1). Furthermore, 59% of mothers had a pre-pregnancy BMI ≥25 kg/m², and 57% were multiparous at enrollment. Maternal sociodemographic characteristics, median OPE biomarker concentrations, and baseline MSEL T-scores (6 months) were not statistically different between the complete-case sample and the baseline population, providing limited evidence of differential selection based on these measured characteristics (Table 1 and Table S4). Mean MSEL Composite scores differed by child’s sex, maternal education, and child autism outcome group (Table S2). Similar patterns were observed among the 169 mother-child pairs in the complete-case sample (Table S3).

Analyzing the distribution of the Composite scores and the respective subscale T-scores, mean Composite scores were lower at 24 months of age compared to other assessment time points (Fig. 1). Receptive Language scores were lowest at 12 months. Fine Motor scores were highest at 12 months, while Visual Reception scores peaked at 36 months.

Figure 1.

Distributions of MSEL Composite scores and T-scores of four subscales at 6, 12, 24, and 36 months of age for the selected baseline population (n = 311 children). Red crosses denote mean values at each time point. Blue asterisks denote differences in mean scores with other time points identified by the estimated marginal means (EMM) test with Tukey’s Honestly Significant Difference (HSD) correction following the linear mixed model at the significance level of 0.05. Composite scores have a normative mean of 100 and a standard deviation (SD) of 15 and T-scores of each subscale have a normative mean of 50 and an SD of 10.

3.2. OPE biomarker concentrations in maternal urinary samples during pregnancy

DFs for DPHP, DBUP+DIBP, BDCPP, and BCETP exceeded 80%, while those for BBOEP, BCPP, DPRP, BEHP, and BMPP were 69%, 73%, 33%, 38%, and 57%, respectively (Table 2). Among OPE biomarkers with DFs ≥60%, BDCPP (0.86 ng/mL) had the highest median concentration, followed by DPHP (0.79 ng/mL) and BCETP (0.57 ng/mL), while BBOEP (0.04 ng/mL) had the lowest. Medians, IQRs, and DFs of OPE biomarkers were generally similar between the complete-case sample and the baseline population, providing limited evidence of differential selection based on these measured characteristics (Table 1 and Table S4). The OPE biomarkers detected in ≥60% of the samples showed generally low correlations with one another (Spearman’s rho = −0.01–0.35) (Figure S3).

3.3. Associations between maternal prenatal OPE concentrations and child MSEL scores

Overall, associations between maternal urinary concentrations of individual OPE biomarkers and child MSEL scores were largely null after FDR correction. Several nominal associations were observed prior to FDR correction and should be interpreted as exploratory. For example, higher prenatal maternal urinary BDCPP concentrations were positively associated with Fine Motor scores at 6 months (β = 0.49, 95% CI: 0.004, 0.98). DBUP+DIBP concentrations were positively associated with Visual Reception scores at 12 months (β = 1.42, 95% CI: 0.54, 2.30). Detectable DPRP was associated with higher Receptive Language scores at 36 months (β = 4.31, 95% CI: 1.64, 6.98). Conversely, detectable BMPP (versus non-detect) was associated with lower Visual Reception scores at 6 months (β = −2.78, 95% CI: −5.28, −0.28) (Figure 2 and Table S5). These associations were similar in direction in the complete-case sample, though associations were null after FDR correction (Table S6).

Figure 2.

Adjusted mean differences (β) in child MSEL Composite and T-scores at 6, 12, 24, and 36 months of age associated with a doubling in prenatal maternal urinary concentrations of six OPE biomarkers detected in ≥60% of samples and with detected versus non-detected concentrations of three OPE biomarkers detected in <60% of samples. Biomarkers detected in ≥60% of samples were modeled as continuous log2-transformed concentrations, whereas biomarkers detected in <60% of samples were modeled as dichotomous variables (detect versus non-detect). Models were adjusted for child sex, child year of birth, homeownership, maternal age at delivery, maternal education, parity, maternal pre-pregnancy body mass index (BMI), and maternal race/ethnicity. Red and blue shaded areas represent negative and positive associations with a p-value <0.05, respectively. No associations remained significant after false discovery rate (FDR) correction (FDR <0.10).

Abbreviations: bis(butoxyethyl) phosphate (BBOEP), bis(2-chloroethyl) phosphate (BCETP), bis(1-chloro-2-propyl) phosphate (BCPP), bis(1,3-dichloro-2-propyl) phosphate (BDCPP), bis(2-ethylhexyl) phosphate (BEHP), bis(2-methylphenyl) phosphate (BMPP), the sum of di-n-butyl phosphate and di-iso-butyl phosphate (DBUP+DIBP), detection frequency (DF), diphenyl phosphate (DPHP), dipropyl phosphate (DPRP)

When stratifying the analysis by child sex, no associations remained significant after FDR correction (Table S7). Results were generally similar when gestational age at delivery was included in the covariate set (Table S8) and when analyses were restricted to the TD children (Table S9).

When examining the overall mixture effects of the six OPE biomarkers detected in ≥60% of the samples in the complete-case sample, higher prenatal OPE mixture was inversely associated with Expressive Language scores at 36 months (ψ: −4.01, 95% CI: −7.80, −0.21) (Table 3). All biomarkers except BCPP exhibited negative weights, and BCETP showed the largest and most significant negative weight (Figure S4). No other statistically significant associations were observed for the Composite score or other subscales. When restricting the population to TD children, increasing Receptive Language scores were observed at 6 months (ψ: 4.80, 95% CI: 0.32, 9.28) (Table S10), while other trends remained similar. In sensitivity analyses using BKMR, we did not observe meaningful evidence of non-linear overall mixture effects, and the significant result from the quantile g-computation analysis was not reproduced (Figure S5).

Table 3.

Quantile g-computation estimates of adjusted mean differences (ψ) in MSEL Composite scores and T-scores of subscales of children at 6, 12, 24, and 36 months of age in association with a simultaneous one-tertile increase in prenatal maternal urinary concentrations of six OPE biomarkers detected in ≥60% of samples.

MSEL	6 months	12 months	24 months	36 months
	β (95% CI)	β (95% CI)	β (95% CI)	β (95% CI)
Composite	1.24 (−3.02, 5.50)	2.76 (−1.93, 7.44)	−0.50 (−6.60, 5.61)	−2.03 (−9.13, 5.07)
Fine Motor	0.25 (−2.92, 3.42)	1.61 (−1.90, 5.12)	−0.11 (−3.82, 3.59)	−0.29 (−5.12, 4.53)
Visual Reception	−1.22 (−5.17, 2.73)	2.62 (−0.34, 5.58)	−0.89 (−4.91, 3.13)	1.87 (−3.19, 6.93)
Receptive Language	3.51 (−0.30, 7.31)	1.52 (−1.86, 4.89)	−0.07 (−4.97, 4.83)	−1.51 (−5.28, 2.25)
Expressive Language	−0.22 (−2.72, 2.27)	−0.26 (−3.97, 3.44)	0.01 (−3.51, 3.52)	−4.01 (−7.80, −0.21)

Note: All models were adjusted for child sex, child year of birth, homeownership, maternal age at delivery, maternal education, parity, maternal pre-pregnancy body mass index (BMI), and maternal race/ethnicity. Bolded numbers indicate p-values < 0.05.

3.4. Longitudinal changes and trajectories of child MSEL scores and their associations with prenatal OPE biomarker concentrations

When analyzing longitudinal changes in MSEL Composite and subscale scores over the four assessment time points, the associations of BCETP, BBOEP, and DBUP+DIBP with Visual Reception scores varied by age (p-int ≤0.04). Specifically, age-specific estimates indicated positive associations between BCETP concentrations and Visual Reception scores at the 36-month time point (β = 1.36, 95% CI: 0.59, 2.12) compared with those at 6 month (Table 4). In contrast, associations between BBOEP and DBUP+DIBP with Visual Reception at the individual assessment ages were not significantly different compared with those at 6 months. The associations of BCETP with Composite scores also appeared to vary by age (p-int = 0.02), with positive individual associations observed at two age points compared with those at 6 months (β = 0.63, 95% CI: 0.02, 1.23 at 12 months; β = 0.98, 95% CI: 0.001, 1.96 at 36 months). When the GEE analysis was narrowed to the complete-case sample, trends were overall similar, though overall effect estimates became slightly less positive, especially for association with DPHP concentrations (Table S11). Results were generally similar in sensitivity analyses using an unstructured correlation matrix (Table S12), suggesting that the primary findings were robust to alternative correlation structure assumptions.

Table 4.

Associations between prenatal maternal urinary concentrations of six log2-transformed OPE biomarkers and longitudinal changes in child MSEL Composite and subscale T-scores across four assessment time points among the selected MARBLES baseline population (n = 311).

MSEL	BBOEP	BCETP	BCPP	BDCPP	DBUP+DIBP	DPHP
	β (95% CI)	β (95% CI)	β (95% CI)	β (95% CI)	β (95% CI)	β (95% CI)
Composite
6 months	Ref	Ref	Ref	Ref	Ref	Ref
12 months	0.05 (−0.98, 1.07)	0.63 (0.02, 1.23)	0.28 (−0.36, 0.93)	−0.21 (−1.27, 0.84)	0.45 (−1.13, 2.03)	1.11 (−0.49, 2.72)
24 months	−0.37 (−1.57, 0.84)	0.25 (−0.62, 1.13)	0.18 (−0.59, 0.95)	−0.22 (−1.46, 1.01)	0.06 (−2.26, 2.39)	2.16 (0.27, 4.05)
36 months	−0.22 (−1.81, 1.38)	0.98 (0.00, 1.96)	0.59 (−0.34, 1.53)	0.03 (−1.36, 1.43)	0.91 (−1.50, 3.33)	2.45 (0.21, 4.69)
pinta	0.53	0.02	0.48	0.89	0.73	0.13
Fine Motor
6 months	Ref	Ref	Ref	Ref	Ref	Ref
12 months	−0.51 (−1.61, 0.59)	0.39 (−0.18, 0.96)	−0.02 (−0.53, 0.49)	−0.34 (−1.07, 0.40)	0.08 (−1.41, 1.57)	0.11 (−1.23, 1.46)
24 months	0.28 (−0.66, 1.22)	0.06 (−0.49, 0.61)	0.18 (−0.35, 0.71)	−0.76 (−1.38, −0.14)	0.18 (−1.04, 1.40)	0.83 (−0.29, 1.95)
36 months	0.04 (−1.11, 1.20)	0.27 (−0.43, 0.96)	0.18 (−0.46, 0.82)	−0.46 (−1.33, 0.41)	0.21 (−1.36, 1.78)	1.72 (0.15, 3.28)
pinta	0.53	0.46	0.85	0.12	0.99	0.08
Visual Reception
6 months	Ref	Ref	Ref	Ref	Ref	Ref
12 months	0.35 (−0.60, 1.30)	0.37 (−0.08, 0.81)	0.47 (−0.02, 0.96)	0.28 (−0.40, 0.96)	1.08 (−0.16, 2.33)	0.27 (−0.89, 1.42)
24 months	−0.60 (−1.67, 0.48)	0.31 (−0.29, 0.92)	0.29 (−0.29, 0.88)	0.37 (−0.54, 1.28)	−0.72 (−2.23, 0.79)	0.37 (−1.01, 1.75)
36 months	0.93 (−0.53, 2.39)	1.36 (0.59, 2.12)	0.86 (0.12, 1.60)	0.54 (−0.48, 1.56)	0.83 (−1.14, 2.80)	0.55 (−1.15, 2.26)
pinta	0.04	0.00	0.06	0.77	0.03	0.93
Receptive Language
6 months	Ref	Ref	Ref	Ref	Ref	Ref
12 months	−0.20 (−1.24, 0.84)	0.37 (−0.13, 0.86)	0.09 (−0.46, 0.64)	−0.16 (−0.95, 0.63)	−0.11 (−1.54, 1.32)	1.05 (−0.17, 2.27)
24 months	0.06 (−1.29, 1.41)	0.00 (−0.74, 0.75)	−0.15 (−0.79, 0.50)	−0.18 (−1.18, 0.81)	0.03 (−2.01, 2.07)	1.58 (0.01, 3.15)
36 months	0.28 (−0.84, 1.39)	0.16 (−0.48, 0.80)	0.12 (−0.49, 0.73)	0.22 (−0.67, 1.11)	0.04 (−1.45, 1.54)	1.46 (−0.09, 3.00)
pinta	0.76	0.34	0.73	0.55	0.99	0.21
Expressive Language
6 months	Ref	Ref	Ref	Ref	Ref	Ref
12 months	−0.32 (−1.25, 0.62)	0.15 (−0.44, 0.74)	−0.04 (−0.57, 0.50)	−0.12 (−0.88, 0.64)	−0.60 (−1.81, 0.62)	0.85 (−0.31, 2.01)
24 months	−0.66 (−1.77, 0.45)	0.00 (−0.55, 0.54)	−0.11 (−0.63, 0.41)	0.02 (−0.79, 0.83)	0.22 (−1.32, 1.75)	1.53 (0.12, 2.94)
36 months	−0.15 (−1.28, 0.98)	0.25 (−0.31, 0.80)	0.11 (−0.42, 0.64)	−0.38 (−1.16, 0.41)	0.13 (−1.35, 1.61)	1.25 (−0.14, 2.65)
pinta	0.54	0.68	0.78	0.59	0.66	0.18

^ap-value for interaction with age.

Note: Bolded numbers indicate interaction p-values < 0.05. All models were adjusted for child sex child year of birth, homeownership, maternal age at delivery, maternal education, parity, maternal pre-pregnancy body mass index (BMI), and maternal race/ethnicity. Regression coefficients (β) represent interaction terms between age at assessment and log2-transformed OPE biomarker concentrations in GEE models that included both main effects and their interaction term. Positive coefficients indicate that associations became more positive over time, whereas negative coefficients indicate that associations became more negative over time.

An exploratory trajectory analysis identified two groups: a low-score trajectory group (n = 61, 36.1%) and a high-score trajectory group (n = 108, 63.9%) (Figure S6). The high-score group remained within 1 SD of the normative mean across all assessments, whereas the low-score group declined from within 1 SD at 6 and 12 months to approximately 2 SDs below the mean at 24 and 36 months. Compared with the high-score group, the low-score group included a higher proportion of mothers younger than 35 years at delivery (61% vs. 46%) and mothers without a bachelor’s degree (61% vs. 35%) (Table S13). Prenatal OPE biomarker concentrations were not significantly associated with membership in the low-score versus high-score trajectory group (Table S14).

4. Discussion

In this study, we investigated associations of prenatal OPE exposure, assessed both individually and as a mixture, with child MSEL scores assessed at multiple time points during infancy and toddlerhood in a cohort enriched for autism and other neurodevelopmental concerns. Overall, we found no clear evidence of associations between maternal OPE biomarker concentrations during pregnancy and child MSEL scores, especially after FDR correction. In the mixture analysis, one significant result was observed; however, it was not consistent between modeling approaches. Child sex did not appear to significantly modify the associations. In the longitudinal analysis, we observed nominal evidence that exposure-outcome associations differ with respect to age at assessment, particularly between prenatal BCETP concentrations and Composite and Visual Reception scores. Overall, we did not identify significant associations in the trajectory analyses.

To our knowledge, only one prior cohort study has evaluated associations between prenatal OPE exposure and child cognitive function using the same cognitive assessment tool as our study (Doherty et al., 2019). Among six OPE biomarkers analyzed, that study found that ip-PPP, an analyte not measured in our study, was inversely associated with Fine Motor scores at 36 months. Several additional studies have examined prenatal OPE exposure in relation to other neurodevelopmental outcomes in children (Castorina et al., 2017; Liu et al., 2021; Oh et al., 2025; Percy et al., 2021). For example, in a Chinese cohort (2014–2016), higher maternal prenatal BDCPP concentrations were associated with lower Mental Development Index (MDI) and Psychomotor Development Index (PDI) scores at 2 years of age, whereas no significant associations were observed for DPHP (Liu et al., 2021). Similarly, a U.S. study of 233 mother–child pairs reported no associations with DPHP but found that higher BCETP concentrations were associated with higher full-scale intelligence quotient (IQ) at 8 years of age measured using the Wechsler Intelligence Scale for Children-IV (Percy et al., 2021). Another U.S. study reported that prenatal DPHP concentrations were inversely associated with full-scale IQ and working memory in children assessed at 7 years of age (Castorina et al., 2017).

We observed a significant inverse association between the OPE mixture and Expressive Language scores at 36 months; however, this association was not reproduced in the sensitivity analysis using BKMR. Quantile g-computation estimates an overall additive mixture effect under linear assumptions, while BKMR is more flexible for detecting non-linear exposure-response relationships and interactions. These discrepancies suggest that the observed mixture finding should be interpreted cautiously and may not represent a robust association. Only four studies have examined prenatal OPE biomarker mixtures in relation to child neurodevelopmental outcomes (Hernandez-Castro et al., 2024; Percy et al., 2021; Vuong et al., 2024; Zhou et al., 2024). A Chinese cohort study reported several significant associations between a mixture of seven OPE parent compounds in maternal cord serum and decreased scores in gross-motor, fine-motor, and problem-solving domains at 12 months of age (Zhou et al., 2024). In contrast, a U.S. cohort of mostly low-income Hispanic/Latina women reported no significant associations, although slight inverse trends were observed at 6 months (Hernandez-Castro et al., 2024), with two other studies observing no significant relationships (Percy et al., 2021; Vuong et al., 2024). Although our findings provided limited evidence of age-specific associations, developmental timing may influence vulnerability to prenatal OPE exposure, as distinct cognitive domains emerge rapidly during infancy and early childhood (Neumann et al., 2021; Thompson, 2024). Future experimental studies are needed to better characterize how prenatal OPE exposure may differentially affect neurodevelopment across developmental stages.

We did not observe evidence of sex-specific associations after FDR correction. Although several nominal sex-specific associations were observed prior to multiple-comparison correction, these findings should be interpreted cautiously and considered hypothesis-generating. Several prior studies have investigated sex-modified effects of prenatal OPE exposure on child neurodevelopment, although findings remain inconclusive. For example, a Chinese study reported strong inverse associations for several OPE biomarkers among boys, with no significant associations among girls (Liu et al., 2021). Conversely, a California study reported significant inverse associations between BCPP and DPRP and motor development only among females (Hernandez-Castro et al., 2024). Another U.S. study reported no sex-specific associations (Percy et al., 2021), although a related study from the same cohort found opposite associations for executive function between males and females (Vuong et al., 2024). The mechanisms underlying the observed sex-specific differences remain incompletely understood. Experimental studies have reported sex-specific neurotoxicity following OPE exposure in animal models (Comfort & Re, 2017; González et al., 2021). OPE exposure has been associated with disruption of genes involved in steroidogenesis, including inhibition of the cyp19 genes, which plays an important role in estrogen synthesis (Zhang et al., 2024). In addition, male rodents have shown greater sensitivity to certain OPE mixtures, including greater disruption of genes involved in brain energy production and neuronal communication (Witchey et al., 2023). OPE-related disruptions in endocrine signaling, mitochondrial function, and neurodevelopmental pathways may contribute to sex-specific vulnerability (Comfort & Re, 2017; González et al., 2021; Witchey et al., 2023; Zhang et al., 2024), although these mechanisms require further investigation.

Overall, inconsistent findings across studies may reflect differences in study populations, prenatal exposure assessment, and timing of neurodevelopmental outcome assessment. First, unlike general-population cohorts, MARBLES is enriched for familial autism, and the underlying neurodevelopmental susceptibility of this population may contribute to differences in observed associations. Second, prenatal OPE exposure assessment has varied across studies. Our study, and several others (Castorina et al., 2017; Doherty et al., 2019; Hernandez-Castro et al., 2024), relied on a single spot urine sample collected predominantly during the third trimester, whereas three prior studies measured urinary OPE biomarkers repeatedly across pregnancy (Liu et al., 2021; Percy et al., 2021; Vuong et al., 2024), potentially providing more representative estimates of prenatal exposure. Third, studies have assessed neurodevelopment at different developmental stages and with varying durations of follow-up. Our study and several prior studies examined outcomes during infancy and toddlerhood (Doherty et al., 2019; Hernandez-Castro et al., 2024; Liu et al., 2021), while others assessed outcomes later in childhood, including full-scale IQ and related cognitive domains at school age (Castorina et al., 2017; Percy et al., 2021; Vuong et al., 2024). Additionally, most previous studies evaluated outcomes at a single time point (Castorina et al., 2017; Doherty et al., 2019; Liu et al., 2021; Percy et al., 2021; Vuong et al., 2024), which may not capture fluctuating developmental patterns over time (Zhou et al., 2024).

This study has several strengths arising from the MARBLES design and data collection approach (Hertz-Picciotto et al., 2018). The MSEL is a well-validated and reliable assessment tool for early childhood cognitive development and was administered by trained clinicians and psychologists. Repeated cognitive assessments enabled analyses of time-specific associations and longitudinal changes in scores over the four assessment points. Prospective data collection allowed for more comprehensive covariate adjustment in the models.

Some limitations were also noted. First, the MARBLES cohort is enriched for familial autism, and children in this study therefore had a higher likelihood of autism and other developmental concerns than children in the general population. This underlying neurodevelopmental susceptibility may contribute to differences between our findings and those from general-population cohorts. Although results were similar in the analyses restricted to the TD children (Table S9), our findings may not be directly generalizable to children without an elevated familial likelihood of autism and should be interpreted within the context of an autism-enriched population. Replication in larger, general-population prospective cohorts is warranted. Second, MSEL assessments were conducted in different settings across follow-up, with evaluations performed in participant’s homes at 6 and 12 months and at the MIND Institute at 24 and 36 months. Although the MSEL is a standardized and well-validated developmental assessment that was administered by trained staff using standardized protocols across all study visits, we cannot exclude the possibility that differences in testing environment may have influenced children's performance and, consequently, longitudinal comparisons. Third, although measured sociodemographic characteristics, OPE biomarker concentrations, and early MSEL T-scores were generally similar between the complete-case sample and the baseline population, we cannot exclude differences in unmeasured characteristics between participants with complete and incomplete follow-up. If loss to follow-up was related to both prenatal OPE exposure and child cognitive outcomes, differential attrition could have introduced selection bias and affected the estimated exposure-cognition associations. Fourth, breastfeeding is an important predictor of early cognitive development (Gibbs & Forste, 2014; Quigley et al., 2012). Although breastfeeding occurs after prenatal exposure assessment and is unlikely to function as a primary confounder, it may reflect postnatal factors associated with child cognitive development. Because breastfeeding data were not consistently collected in this study population, we were unable to evaluate its potential influence on child cognitive outcomes. Fifth, although reproducibility varies across OPE biomarkers, prior studies have documented substantial within-person variability during pregnancy, and a single measurement may therefore inadequately characterize average prenatal exposure (Hoffman et al., 2014; Kuiper et al., 2020; Percy et al., 2020; Romano et al., 2017). Because our exposure assessment relied on a single spot urine sample collected predominantly during the third trimester, our exposure estimates are susceptible to measurement error and exposure misclassification. If this error is non-differential with respect to child cognitive outcomes, it would generally be expected to attenuate exposure-outcome associations and may have reduced our ability to detect modest adverse neurodevelopmental effects. Sixth, our modest sample size may have limited our ability to detect small to modest associations, particularly after correction for multiple comparisons in subgroup, interaction, and trajectory analyses. Therefore, our findings should be interpreted with caution within the context of this high-risk population and replicated in larger prospective studies.

In conclusion, we found no clear evidence of associations between prenatal maternal urinary OPE biomarker concentrations and childhood cognitive function across infancy and toddlerhood among children with elevated familial likelihood of autism. Findings for the mixture analysis were not consistent across mixture modeling approaches and require replication in larger studies. We did not observe sex-specific associations after FDR correction. Associations between prenatal maternal BCETP concentrations and Composite and Visual Reception scores appeared to vary with respect to age at assessment, underscoring the importance of repeated outcome measurements for future studies. Because this study population is at elevated risk for developmental delays and cognitive deficits, further studies in lower-risk populations are warranted to confirm these findings.

Abbreviations

BBOEP

bis(butoxyethyl) phosphate

BCETP

bis(2-chloroethyl) phosphate

BCPP

bis(1-chloro-2-propyl) phosphate

BDCPP

bis(1,3-dichloro-2-propyl) phosphate

BEHP

bis(2-ethylhexyl) phosphate

BMI

body mass index

BMPP

bis(2-methylphenyl) phosphate

DAG

directed acyclic graph

DBUP+DIBP

sum of di-n-butyl phosphate and di-iso-butyl phosphate

DF

detection frequency

DPHP

diphenyl phosphate

DPRP

dipropyl phosphate

FDR

false discovery rate

GEE

generalized estimating equations

ip-PPP

isopropyl-phenyl phenyl phosphate

LOD

limit of detection

MARBLES

Markers of Autism Risk in Babies Learning Early Signs

MIND

Medical Investigations of Neurodevelopmental Disorders

MSEL

Mullen Scales of Early Learning

Non-TD

non-typical development

OPE

organophosphate ester

SD

standard deviation

SG

specific gravity

TBOEP

tris(2-butoxyethyl) phosphate

TCEP

tri(2-chloroethyl) phosphate

TCIPP

tris(1-chloro-2-propyl) phosphate

TD

typical development

TDCIPP

tris(1,3-dichloro-2-propyl) phosphate

TEHP

tris(2-ethylhexyl) phosphate

TMPP

trimethylphenyl phosphate

TNBP+TIBP

sum of tri-n-butyl phosphate and tris(isobutyl) phosphate

TPHP

triphenyl phosphate

TPP

tripropyl phosphate

Availability of data and material

Portions of the datasets generated and analyzed during this study are publicly available in the National Institute of Mental Health (NIMH) Data Archive. The entire non-identifiable dataset is available from the authors upon reasonable request and with permission from the IRB at UC Davis.