Prenatal exposure to persistent organic pollutants and associations with child behavior problems at 1–5 years

Abstract

Existing studies have found inconclusive associations between prenatal exposure to persistent organic pollutants (POPs), including per- and polyfluoroalkyl substances (PFAS) and polybrominated diphenyl ethers (PBDEs), and offspring neurodevelopment. However, there is a significant gap in research involving African American populations, who face higher levels of exposure to many POPs relative to other groups. In this study, we assessed the joint effects of PFAS and PBDEs on child behavior problems among African American mother-child pairs in Atlanta, Georgia. Our study population included a subset of mother-child pairs participating in a prospective birth cohort (N=159) for whom exposure and outcome data were available. Four PFAS and three PBDEs were measured in serum samples obtained during the first trimester of pregnancy. The Child Behavior Checklist was administered annually from ages 1-5 years and used to assess internalizing and externalizing behavior problems (averaged across all timepoints). We used quantile g-computation, Bayesian kernel machine regression (BKMR), and self-organizing maps (SOM) to assess associations between POPs mixtures and internalizing and externalizing behavior problems. Using quantile g-computation, we observed that increasing concentrations of prenatal PBDEs were associated with more internalizing and externalizing behavior problems (e.g., ψ= 0.20, 95% CI= 0.04, 0.36 for externalizing problems). The SOM cluster reflecting high PFAS and high PBDEs was similarly associated with an increase in internalizing and externalizing behavior problems compared to the reference cluster (e.g., β= 0.44 95% CI= 0.08, 0.81 for internalizing problems). The positive associations were attributable to PBDEs, while PFAS were negatively associated with both outcomes across all three methods. To conclude, among mother-child pairs in Atlanta, we observed that exposure to PFAS and PBDEs was associated with internalizing and externalizing behavior problems between 1-5 years of age.

Graphical Abstract

Introduction

Between 2018 and 2019, the United States (US) Centers for Disease Control and Prevention (CDC) estimated that 1 in 7 children aged 3-17 years had a current diagnosed mental or behavioral health condition.¹ Notably, many of these mental health conditions co-occur, with 37% of those with a mental health condition experiencing two or more conditions.¹ Some of the most common comorbid symptom presentations are internalizing and externalizing behaviors.² Examples of internalizing behaviors in children include worry, anxiety, and depression, and externalizing behaviors include inattention, impulsivity, and aggression. Longitudinal epidemiologic studies have reported substantial economic and societal costs associated with internalizing and externalizing behaviors during childhood.^3–5 For example, adults who exhibited high externalizing and high internalizing behavior problems during childhood were significantly more likely to be a welfare recipient and significantly less likely to have an intimate partnership compared to those who exhibited neither.⁶ As such, an urgent need exists to identify risk factors for externalizing and internalizing behavior problems, including maternal psychopathology. However, it remains unclear whether exposure to environmental toxicants is one such risk factor.

Systematic reviews summarizing evidence from animal and human studies suggests that prenatal exposure to environmental chemicals can adversely impact childhood psychopathology.^7,8 Historically, some of the most widely used environmental chemicals include persistent organic pollutants (POPs), such as per- and polyfluoroalkyl substances (PFAS) and polybrominated diphenyl ethers (PBDEs). Both of these chemicals are man-made and have been added to many consumer products to make them water and grease resistant and non-flammable.^9,10 However, evidence linking these specific chemicals to childhood psychopathology, which includes internalizing and externalizing behaviors as significant aspects, is mixed. For example, prenatal exposure to low levels of PBDEs was positively associated with an increased frequency of internalizing and externalizing behaviors at ages 2 and 4 years in the Shanghai Birth Cohort.¹¹ Subsequent work also finds that these associations extend through later childhood, as prenatal exposure to the sum of these same five PBDEs was associated with child self-reported functional impairment, inattention/hyperactivity, internalizing problems at 12 years in the Health Outcome and Measures in the Environment (HOME) study.¹² In contrast, a 2024 systematic review concluded that prenatal PFAS exposure was associated with performance intelligence, executive function, psychomotor, attention, and language during childhood, but found no association with internalizing and externalizing behaviors.⁸ It is possible that the inconsistent results observed with PFAS and childhood psychopathology may be attributed to failing to account for co-exposures, as the vast majority of studies focus on the effect of a single chemical at a time, or mixtures of chemicals within one class, on health outcomes. Despite these inconsistent associations, there is biologic plausibility that supports an association. For example, PFAS and PBDEs are positively associated with increased oxidative stress and inflammation, both of which have subsequently been linked to behavior problems in offspring.^13–15 Epigenetic age acceleration has similarly been implicated as a biologic mechanism linking environmental toxicant exposure to internalizing and externalizing behavior problems.^16–18 While PFAS and PBDE exposures are known to commonly co-occur, to our knowledge, only one study has examined a mixture of both PFAS and PBDEs in relation to child psychopathology,¹⁹ finding inverse association with PBDEs but inconsistent association with other POPs (including PFAS).

Concerns regarding the toxicity of PBDEs resulted in the regulation of these compounds in the US in the early 2000s, while the US Environmental Protection Agency only established legally enforceable regulations on six PFAS in drinking water in 2024. Despite recent achievements in regulation, these chemicals continue to pose a major public health threat, as exposure remains high among non-white and socioeconomically disadvantaged populations as a result of historical environmental injustices that are often compounded by social stressors, such as discrimination.^20–22

Therefore, to build on these knowledge gaps, we leveraged a socioeconomically diverse prospective cohort of mother-child pairs in metropolitan Atlanta, Georgia to assess whether PFAS and PBDEs have a joint effect on internalizing and externalizing behaviors at ages 1-5 years. We hypothesized that combined impact of exposure to both PFAS and PBDEs (i.e., joint effects) would be associated with higher levels of internalizing and externalizing behavior problems. We further hypothesized that these associations would be greatest for the total mixture of both chemical classes, as compared to what is observed for PFAS alone or PBDEs alone.

Methods

Overview of the study population

Participants in the present analysis included mother-child pairs (N=159) recruited from an ongoing, prospective longitudinal birth cohort designed to examine the root causes of child health disparities.^23,24 Study participants were recruited during the first trimester pregnancy from Grady Memorial Hospital and Emory University Hospital Midtown in Atlanta, Georgia. Eligibility inclusion for the cohort was as follows: US born African American female, not pregnant with multiples, able to communicate in English, and had no chronic medical conditions. Those who had a live born infant and whose infant was free of congenital abnormalities were invited to participate in the child follow up study, with annual follow-ups occurring at ages 1-5 years. The Institutional Review Board (IRB) at Emory University approved the protocols for this study, and all participants provided written, informed consent prior to enrollment. Separate IRB protocols were approved for the prenatal and child follow up study. The subset utilized in this analysis included those for whom a first trimester serum sample was previously analyzed for both PFAS and PBDEs, and who participated in the child follow up study. Participants in this study delivered between 2014-2017.

At the first study visit, a self-reported questionnaire was used to collect information on sociodemographic characteristics (e.g., maternal education, marital status). Information on substance use (i.e., alcohol, tobacco, and marijuana) within the last month was ascertained using a timeline follow back interview. Parity and early pregnancy body mass index (kg/m²) at the first study visit were extracted from the maternal medical record. Child sex and gestational age at delivery were extracted from the neonatal medical record.

PFAS and PBDEs exposure assessment

Serum samples were collected during the first trimester of pregnancy and frozen at −80°C prior to exposure assessment at Emory University’s Laboratory for Exposure Assessment and Development in Environmental Research (LEADER). Methods for PFAS^25,26 and PBDE^27(p1),28,29 exposure assessment have been previously described in detail. In brief, concentrations of seven PBDEs were analyzed using a gas chromatographic-tandem mass spectrometry (GC-MS/MS) instrument, while concentrations of four PFAS were analyzed using a liquid chromatographic-tandem mass spectrometry (LC-MS/MS) instrument.

We focused our analysis on those chemicals that were detected in at least 80% of samples, which included BDE-47, BDE-99, BDE-100, perfluorohexane sulfonic acid (PFHxS), perfluorooctane sulfonic acid (PFOS), perfluorooctanoic acid (PFOA), and perfluorononanoic acid (PFNA). Values below the limits of detection (LOD) were imputed using the commonly used equation, LOD divided by the square root of two,³⁰ and all chemicals were natural log transformed to reduce right skewness.

Measurement of internalizing and externalizing behavior

Internalizing and externalizing problems were measured using the pre-school version of the Child Behavior Checklist (CBCL), which has been validated for ages 1.5 years (18 months) through 5 years and provides data on behavioral and emotional symptoms in children.³¹ The CBCL is a widely used instrument with high validity and reliability.³¹ The CBCL for ages 1.5-5 consists of 100 items in which the mother indicates the option that best describes her child within the last two months on a scale of 0 (not true) to 2 (very true or often true). We calculated the internalizing problem raw score, a sum of 36 items with possible scores ranging from 0 to 72, and the externalizing problem raw score, a sum of 24 items with possible scores ranging from 0 to 48. We visualized the distribution of raw CBCL scores using histograms, observing that the raw scores were approximately normally distributed, and thus no transformations were applied. The CBCL was administered at annual visits from child ages 1-5 years and scores were averaged across all available assessments, an approach commonly applied in child psychology.^32–35 There were 30, 121, 108, 84, and 85 CBCL assessments occurring at ages 1, 2, 3, 4, and 5 years, respectively. This corresponded to 388 CBCL assessments across 159 children (mean two visits per child).

Statistical analysis

We first assessed descriptive statistics, which included examining the distributions of sociodemographic characteristics and CBCL scores in our analytic sample using frequencies for categorical variables and means and standard deviations (SD) for continuous variables. We then summarized the distributions of PFAS and PBDEs using geometric means (GMs), geometric SDs (GSDs), and percentiles (i.e., minimum, median, maximum). We calculated Pearson correlation coefficients to assess the correlations between exposures and outcomes.

We then assessed the associations between each chemical concentration and each CBCL outcome using single pollutant, multivariable linear regression models. Confounders and covariates were identified using a directed acyclic graph (DAG; Figure S1) that was informed through a literature review and univariate associations between exposures and outcomes in our study population. Final models were adjusted for child sex (male, female), mean child age at CBCL assessment, maternal age, maternal education (less than some college, college degree or higher), marital status (married or living together, never married), parity (no prior births, 1 or more prior births), early pregnancy body mass index (in kg/m²) and prenatal substance use (any, none). In all linear regression models, all exposures and outcomes were scaled to a mean of zero and a standard deviation for one.

Mixtures analysis

Next, we utilized a variety of different mixtures methods to assess the joint effects of PFAS and PBDEs on internalizing and externalizing behavior problems: quantile g-computation, Bayesian kernel machine regression (BKMR), and self-organizing maps (SOM). Each method was used to assess a specific aspect of this research question. Specifically, we used quantile g-computation and BKMR to assess the overall mixture effect. We additionally used BKMR to assess non-linearity and interactions between exposures. We then used SOM to assess patterns of exposures. For consistency with linear regression models, chemical exposures were natural log transformed, all exposures and outcomes were scaled, and all models were adjusted for the same set of covariates.

We first used quantile g-computation, which estimates the overall effect of increasing all exposures in the mixture by one quantile (4 quantiles in this analysis).³⁶ Quantile g-computation does not assume directional homogeneity, and all exposures are assigned a positive or negative weight which reflect the proportion of the partial effect in either direction. We considered three mixture groups for quantile g-computation models: the total mixture of both PFAS and PBDEs, a mixture of PFAS, and a mixture of PBDEs.

Our second approach was to assess the overall mixture effect using BKMR with component wide variable selection (10,000 iterations).^37,38 Univariate exposure-response functions were used to assess linearity, while bivariate exposure-response functions were used to assess interactions between exposures. In BKMR models, the relative importance of each exposure was estimated using posterior inclusion probabilities (PIPs), which reflect the posterior probability that a given exposure is included in the model.

Finally, we applied SOM to identify exposure profiles (i.e., clusters reflecting the most common exposure patterns). SOM is a type of artificial neural network that uses unsupervised machine learning to extract features from a dataset and reduce dimensionality.^39,40 First, weights are initialized randomly for a predefined number of units on a two-dimensional plane. A data point is mapped to the unit it is closest to, or the “winning unit”, calculated as the Euclidian distance between the data point and each weight. Then the weight of the winning unit, and the weights of the nodes closest to it are updated to be more similar to the data point mapped to it. This process happens iteratively until the trained model preserves topology of the underlying data. Exposure profiles are homogenous within clusters, and heterogeneous across clusters. The number of clusters chosen for analyses was based on multiple statistical measures identifying group structure, including adjusted R², as well as visual inspection of the clusters for interpretability and suitable number of participants in each cluster. Once each participant is assigned an exposure cluster, adjusted linear regression models were used to assess the impact of the exposure cluster on each behavior outcome.

Sensitivity analysis

In our first sensitivity analysis, we additionally adjusted for total lipids in linear regression models which included PBDEs as the exposure. Second, we re-ran linear regression models removing those that delivered a preterm infant. Third, we accounted for the longitudinal nature of our data by re-running single pollutant associations using linear mixed effect models which included a random intercept for participant ID. We additionally re-ran all mixture models for the joint effect of both PFAS and PBDEs together using models which included a random intercept for participant ID.⁴¹ For the longitudinal data analysis, we re-ran the SOM clustering algorithm on the longitudinal dataset.

All analysis was conducted in R version 4.4.2.

Results

There were 159 mother-child pairs included in this analysis (Table 1). The average maternal age was 25 years (SD=4.46), and half of mothers had at least a college degree (N=76, 48%) and were married or living with a partner (N=84, 53%). Most participants had an early pregnancy BMI that was underweight or obese at the early pregnancy clinic visit (N=64, 40% and N=66, 42%, respectively) and had an income to poverty ratio <100% (N=49, 45%). The mean scores for internalizing behavior problems were 7.10 (SD=5.29), while mean scores for externalizing behavior problems were slightly higher at 11.37 (SD=7.62) (Table 1). The demographics in our analytic sample were similar to the larger cohort, however a greater number of participants in our analytic sample were multiparous (59% versus 53%) and fewer reported substance use (43% versus 56%) compared to the full cohort (Table 1).

Table 1:

Distribution of sociodemographic characteristics and Child Behavior Checklist (CBCL) outcomes among mother-child pairs in metropolitan Atlanta, Georgia.

	Analytic Sample	Full Cohort
Demographic	N = 159	N=709
	N (%)	N (%)
Maternal Age (Mean [SD])	25 (5)	26 (5)
Maternal Education
Less than some college	83 (52%)	378 (53%)
College degree or higher	76 (48%)	331 (47%)
Marital Status
Never married	75 (47%)	363 (51%)
Married/Living together	84 (53%)	346 (48%)
Parity
No prior births	65 (41%)	332 (47%)
1 or more prior births	94 (59%)	377 (53%)
Early Pregnancy BMI
Underweight or normal	64 (40%)	270 (38%)
Overweight	29 (18%)	153 (22%)
Obese	66 (42%)	286 (40%)
Substance Use
No	90 (57%)	309 (44%)
Yes	69 (43%)	400 (56%)
Insurance Type
Low-income Medicaid	69 (43%)	224 (32%)
Pregnancy Medicaid	58 (36%)	335 (47%)
Private	32 (20%)	150 (21%)
Income to Poverty Ratio
<100%	76 (48%)	294 (41%)
100% - 199%	52 (33%)	271 (38%)
≥200%	31 (19%)	130 (18%)
Missing	0 (0%)	14 (2%)
Child Sex1
Male	81 (51%)	304 (43%)
Female	78 (49%)	324 (46%)
Missing	0 (0%)	81 (11%)
Child Behavior Checklist (CBCL) Outcomes1		N = 272
Mean Child Age	2.82 (0.64)	2.71 (0.61)
Internalizing Behavior Problems	7.10 (5.29)	6.80 (5.61)
Externalizing Behavior Problems	11.37 (7.62)	10.66 (7.79)

Abbreviations: SD, standard deviation.

Note: Percentages may not sum to 100% due to rounding.

¹CBCL outcomes are averaged across ages 1-5.

When examining the distribution of exposures within our analytic sample, we observed that the median concentration was highest for PFOS (2.53 ng/mL) and BDE-47 (92.70 pg/mL) among the PFAS and PBDEs, respectively (Table S1). Median values of the PFAS and PBDEs in our analytic sample were comparable to the larger cohort (Table S1). PFAS and PBDEs were strongly correlated within each chemical class (e.g., ρ=0.46 for PFOS and PFHxS, and ρ=0.69 for BDE-47 and BDE-99). PFAS and PBDEs were not strongly correlated across classes. PFAS were negatively correlated with CBCL scores, while PBDEs exhibited small positive correlations (e.g., ρ=0.11 for internalizing behavior problems and BDE-47) (Figure S2).

In adjusted single-pollutant models, PFHxS, PFOS, PFOA, and PFNA were associated with lower internalizing and externalizing behavior problems (e.g., β_PFHxS = −0.17, 95% confidence interval [CI]= −0.33, 0.00 and β_PFHxS = −0.21, 95% CI= −0.36, −0.05 for internalizing and externalizing behavior problems, respectively) (Figure 1, Table S2). In contrast, BDE-47, and BDE-99 were consistently associated with an increase in internalizing and externalizing behavior problems (e.g., β_BDE-99= 0.09, 95% CI=−0.07, 0.25 and β_BDE-99= 0.15, 95% CI=−0.01, 0.30 for internalizing and externalizing behavior problems, respectively). In linear regression models stratified by child sex, these associations were consistently stronger among males compared to females (Table S2). Associations were similar in models which additionally adjusted for total lipids, and in models which restricted to those who delivered full term (Table S3 and Table S4).

Figure 1:

Associations between natural log transformed per- and polyfluoroalkyl substances (PFAS) and polybrominated diphenyl ethers (PBDEs), measured in serum samples obtained during the first trimester, and internalizing and externalizing behavior problems averaged across ages 1-5 years among mother-child pairs in metropolitan Atlanta, Georgia.

Note: Models are adjusted for child sex, mean child age, maternal age, maternal education, marital status, early pregnancy BMI (categorical), parity, and substance use.

Using quantile g-computation, a one quantile increase in the total mixture was associated with a nonsignificant, small reduction in internalizing and externalizing behavior problems (ψ=−0.04, 95% CI =−0.28, 0.20 for internalizing and ψ=−0.06, 95% CI=−0.30, 0.18 for externalizing) (Figure 2, Table S5). In models which considered a mixture of PFAS alone and PBDEs alone, the effect estimates were greater in magnitude and were statistically significant (e.g., ψ_PBDE= 0.20, 95% CI= 0.04, 0.36 and ψ_PFAS= −0.28, 95% CI= −0.46, −0.10 for externalizing behavior problems). In the total mixture and the PFAS alone mixture, PFOA was assigned the largest positive weight when externalizing behavior problems was the outcome, while BDE-47 was assigned the largest negative weight (Figure 2).

Figure 2:

Quantile g-computation joint effects (95% confidence intervals) and weights reflecting the associations between a mixture of natural log transformed per- and polyfluoroalkyl substances (PFAS) and polybrominated diphenyl ethers (PBDEs), measured in serum samples obtained during the first trimester, and internalizing and externalizing behavior problems averaged across ages 1-5 years among mother-child pairs in metropolitan Atlanta, Georgia.

Note: Models are adjusted for infant sex, mean child age, maternal age, maternal education, marital status, early pregnancy BMI (categorical), parity, and substance use. By design, the positive and negative weights obtained using quantile g-computation sum to 1 in either direction and are reflective of the proportion of the effect in that direction. Positive and negative weights should not be directly compared to one another.

The total mixture was not associated with either internalizing or externalizing behaviors in BKMR models. In contrast, in chemical specific BKMR models, increasing quantiles of the PFAS alone mixture was associated with a small reduction in internalizing and externalizing behavior problems, while PBDEs were positively associated with both outcomes (Figure 3). PIPs obtained from the total mixture model indicate that PFHxS, BDE-47and BDE-99 have the largest PIPs when internalizing behavior problems are the outcome (Table S6), which supports the finding from quantile g-computation that these two chemicals appear to be strong drivers of the effect. PFHxS and PFNA were also negatively associated with internalizing and externalizing behaviors in the univariate exposure-response functions (Figure S3). When visualizing the bivariate exposure-response functions, we observed some evidence of interaction between PFOA and BDE-99, and BDE-47 and BDE-99 in the total mixture model which included internalizing behaviors as the outcome. The association between BDE-99 and internalizing behaviors was u-shaped when holding each additional PFAS and PBDE constant at various quantiles (Figure S4).

Figure 3:

Joint effect (95% credible intervals) of the BKMR mixture of natural log transformed per- and polyfluoroalkyl substances (PFAS) and polybrominated diphenyl ethers (PBDEs), measured in serum samples obtained during the first trimester, and internalizing and externalizing behavior problems averaged across ages 1-5 years among mother-child pairs in metropolitan Atlanta, Georgia.

Note: Models are adjusted for child sex, mean child age, maternal age, maternal education, marital status, early pregnancy BMI (continuous), parity, and substance use

Using SOM, we identified three exposure groups reflecting the following: high PFAS and high PBDEs (cluster 3, N=49), high PFAS and low PBDEs (cluster 1, N=78), and modest PBDEs and low PFAS (cluster 2, N=32) (Figure 4). We used cluster 1 high PFAS and low PBDEs, as the reference group in linear regression models which used the SOM cluster as the categorical exposure. We observed significantly higher externalizing and internalizing behavior problems for individuals in the low PFAS and modest PBDE cluster (cluster 2), as well as for individuals in the high PFAS and high PBDEs cluster (cluster 3) as compared to the reference group (e.g., β_{cluster 3}= 0.44, 95% CI= 0.08, 0.81 and β _{cluster 3}=0.54, 95% CI= 0.18, 0.89, for internalizing and externalizing behavior problems, respectively) (Figure 4, Table S7).

Figure 4:

Self-organizing maps (SOM) cluster star plot reflecting patterns of exposure to per- and polyfluoroalkyl substances (PFAS) and polybrominated diphenyl ethers (PBDEs), measured in serum samples obtained during the first trimester, and association between SOM clusters and internalizing and externalizing behavior problems averaged across ages 1-5 years among mother-child pairs in metropolitan Atlanta, Georgia.

Note: Models are adjusted for child sex, mean child age, maternal age, maternal education, marital status, early pregnancy BMI (categorical), parity, and substance use. Cluster 2 was treated as the reference group.

In our sensitivity analyses accounting for the longitudinal nature of our data, single pollutant effects in linear mixed models were comparable to what was observed using linear regression (Table S8). In quantile g-computation models which included a random intercept for participant ID, the only notable change was that the joint effect of PBDEs alone on externalizing behaviors was attenuated to non-signifance (Table S9). No notable differences between our primary analyses using average outcomes and our sensitivity analyses with repeated measures were observed for SOM (Table S10), including that we identified the same exposure patterns (Figure S5). The joint effect of the total mixture was also comparable using BKMR (Figure S6).

Discussion

Among mother-child pairs in Atlanta, we observed that increasing concentrations of PBDEs were associated with an increase in internalizing and externalizing behavior problems among pre-school age children. In contrast, we observed that increasing PFAS concentrations were associated with a decrease in these same behavior problems. When we considered the joint effects of PFAS and PBDEs together, PFAS appeared to attenuate the effect of PBDEs, as we largely observed fewer internalizing and externalizing behavior problems in mixture models which included both PFAS and PBDEs as the exposures. These conclusions persisted across our three different mixture methods, including quantile g-computation, BKMR, and SOM, and advance our understanding of how PFAS and PBDEs, alone and in combination, impact psychopathology among pre-school age children.

The findings reported here contribute to a growing body of research assessing neurodevelopmental impact of POPs exposure.^7,8,42,43 Our findings extend mechanistic studies linking PFAS exposure to behavior problems, which have found that PFOA, PFOA, PFNA, and PFHxS are associated with perturbations in biological pathways (e.g., amino acid and bile acid metabolism) and epigenetic activity.^16,44 Across all methods, we observed that PFHxS and PFNA contributed the most to the observed inverse association with internalizing and externalizing behavior problems. This finding is consistent with prior work in the HOME cohort observing that childhood exposure to PFNA and PFHxS was associated with a modest reduction in externalizing problems at 8 years of age.⁴² However, that study also found a non-significant increase in internalizing behaviors in association with childhood PFAS exposure,⁴² which contrasts with our findings. Among school age children ages 6-13 years in Italy, the highest compared to the lowest quartile of PFHxS, estimated using single pollutant Poisson regression, was associated with more externalizing problems.⁴⁵ Within a birth cohort in upstate New York, newborn dried blood spot levels of PFOS and PFOA were further associated with higher odds of behavioral difficulties at 7 years.⁴⁶ Discrepancies between these findings and the inverse associations observed here could be attributed to timing of measurement, as our study measured PFAS in prenatal serum samples and assessed outcome measurements solely up to 5 years of age.

In mixture models assessing the impact of solely PBDEs, we observed that higher PBDE exposure was associated with an increase in internalizing and externalizing behavior problems. Across all methods, BDE-47 was found to be the biggest contributor to the total effect. Our findings corroborate prior research within a Shanghai birth cohort, which observed that internalizing and externalizing behaviors were increased following prenatal exposure to multiple PBDEs.¹¹ Within the HOME cohort, child PBDE levels were associated with more frequent behavior problems at 8 years.⁴⁷ However, no associations were observed when behavior was assessed at earlier timepoints.⁴⁷ Additionally, within the World Trade Center cohort, cord blood PBDE levels were linked to worsening physical and mental development between 1-2 and 6 years of age.⁴⁸ Systematic reviews have further concluded that PBDEs are associated with neurodevelopmental disorders.^7,43

To date, most studies assessing associations between environmental-induced behavior changes have focused on individual classes of chemicals. While these studies are essential, particularly for identifying determinants of exposures, they do not truly reflect real-world exposure patterns, as exposures rarely occur in isolation. An important aspect of our study was that we included two classes of chemicals, PFAS and PBDEs. Exposures to both classes of chemicals occur simultaneously and may have additive or synergistic effects. While we hypothesized that both PFAS and PBDEs would be associated with an increase in behavior problems, this was only observed for PBDEs. Experimental studies suggest that endocrine disrupting chemicals act on many biological systems simultaneously, leading to opposing health effects resulting from complex toxicodynamic influences.⁴⁹ Another reason for our observed opposing effects could be the result of non-linearity resulting from a threshold response, which could suggest that an increase risk on the outcome does not occur after a certain level of exposure is reached. The opposing effects observed across chemical classes could also be the result of dose-response, indicating that there may be a dosage effect if the PFAS levels were generally lower than what is environmentally relevant.^50–52 PFAS levels in our study population are, on average, slightly lower than what is observed among reproductive aged Black women in the general US population,²⁵ while levels of PBDEs are higher.⁵³

Notably, we observed some evidence of non-monotonic dose-response relationships between BDE-47, BDE-99, and PFOA using BKMR. This has important implications for regulatory agencies that are tasked with identifying a ‘minimally safe level,⁵⁴ as our results suggest that effects are only non-linear in the presence of other chemicals. This underscores the importance of considering complex interactions between exposures, as cumulative exposure to multiple compounds potentially alters the dose-response curves which are used for risk assessment.

Our findings should be interpreted considering their strengths and limitations. First, our study relied solely on exposures measured in sera of mothers during pregnancy, which do not necessarily reflect early childhood exposures, particularly as children spend more time away from the home (e.g., at preschool). Additionally, accurate information regarding breastfeeding (a potential source of POPs exposure in infancy⁵⁵) was unavailable for many of our study participants; postnatal exposures should be examined in future work. Further, we assessed internalizing and externalizing behavior problems using the parent-reported CBCL, which does not provide a clinical diagnosis. While the CBCL may be subject to subjective parent reporting, we note that the CBCL has been validated for children (with the pre-school version valid beginning at 1.5 years of age) and is widely used across studies.^56,57 We also had average scores on the CBCL, an important strength which allowed us to get a more robust estimate of internalizing and externalizing behaviors in early childhood. Nonetheless, we also acknowledge that this approach may mask timepoint specific associations if prenatal exposure is associated with CBCL scores at certain timepoints or across development. To address this, we conducted a sensitivity analysis using the repeated measures of the CBCL outcomes and results were similar, providing more evidence that the findings observed here are robust. Lastly, prior work finds that prenatal PFAS exposure is linked to increased odds of miscarriage,⁵⁸ thus our findings may be subject to live birth bias, which produces downward bias in exposure-outcome associations; this may be an important consideration in interpreting the observed inverse association with PFAS.⁵⁸ Relatedly, we cannot rule out the possibility of residual confounding, as parenting practices and parenting stress were unavailable at the time of our analysis but have been related to early displays of internalizing and externalizing behavior problems in previous work.⁵⁹ Additionally, we had a relatively small sample size, and future studies in larger sample sizes are needed to confirm generalizability of our findings. Our study was comprised exclusively of African Americans. We believe racial homogeneity should be viewed as an important strength of our study, as prior research addressing similar research questions in the US has been limited in terms of the racial makeup of the study population. Lastly, the most notable strength of this analysis was our application of three types of mixture methods, allowing us to assess many different aspects, including interaction, non-linearity, exposure patterns, and joint effects.

Conclusions

Within a prospective birth cohort of mother-child pairs in Atlanta, we consistently observed positive relationships between prenatal exposure to PBDEs and child behavior problems and negative relationships between prenatal exposure to PFAS and child behavior problems. When considering joint exposure to both classes of chemicals, the deleterious effects of PBDEs were attenuated, suggesting evidence of interaction between these exposures. Future research, particularly which considers childhood exposure to different types of POPs, is needed to understand if the association observed in this study persists at later timepoints in child development.

Highlights.

• Joint exposure to PFAS and PBDEs was associated with more child behavior problems

• These associations were primarily driven by PBDEs

• Effects of PFAS on child behavior problems were more inconsistent

• These associations were robust and persisted across four different mixture methods