Sex-specific associations between urinary bisphenols concentrations during pregnancy and problematic child behaviors at age 2 years

Abstract

Effects of prenatal bisphenol A (BPA) exposure on child behavior are mixed with some reports suggesting increased problematic behaviors in girls (e.g., aggression and emotional reactivity) and in boys (i.e., externalizing behaviors), while other reports suggest decreased problematic behaviors in girls. Little is known about the potential impact of pregnancy bisphenol S (BPS) exposure on child behavior. In a prospective cohort study (n=68), five maternal spot urine samples collected across pregnancy were pooled and analyzed for BPA and BPS. Child behavior at 2 years was assessed using the Child Behavior Checklist (CBCL). Linear regression models were used to assess associations between bisphenols concentrations and both composite and syndrome CBCL scales. Exposure x child sex interactions were included in addition to their main effects and sex-stratified analyses were conducted. Models were adjusted for maternal age, number of siblings, and child age at CBCL intake. Mean maternal age was 29.7 years. Most women were White (88%), had an annual household income ≥$50,000 (66%), and at least a college degree (81%). Median concentrations were 1.3 ng/mL (range 0.4–7.2) for BPA and 0.3 ng/mL (range 0.1–3.5) for BPS. Sex modified the relationship between BPA and scores on several syndrome scales—anxious-depressed, aggressive, and sleep problems—where the association was consistently inverse in males in lower BPA concentrations, and positive (more reported behavior problems) among girls in the higher BPA group. Higher BPS was associated with more problematic internalizing behaviors among girls but not boys, and sex modified the relationship between BPS and emotionally reactive behaviors (P_interaction = 0.128), with sex-specific estimates revealing more emotionally reactive behaviors among girls (expβ=3.92 95% CI 1.16, 13.27; P=0.028) but not boys. Findings were mixed overall, but one notable finding was that BPS, a replacement for BPA, was associated with increased problematic behaviors. There is a need for replication of findings due to our small sample size.

1. INTRODUCTION

Bisphenol A (BPA) is a well-studied, man-made endocrine-disrupting chemical (EDC). It is used in the manufacture of many consumer products, including polycarbonate plastics used in some food and drink packaging as well as receipt paper and the lining of aluminum cans.¹ The most common source of BPA exposure among humans is diet due to leaching from polycarbonate products and can liners into liquids and food intended for human consumption.¹ BPA crosses the placenta during in utero development.² After birth, infants can experience continued exposure to BPA through breastmilk.³ With a growing body of scientific literature on the negative health effects associated with BPA exposure, another phenolic compound, bisphenol S (BPS), has been replacing BPA in an increasing number of consumer products. A 2018 study showed that 89.4% of non-institutionalized US adults and children had detectable (above 0.1 μg/L) urinary concentrations of BPS.⁴ Although still understudied, existing research suggests that health effects of BPS may be analogous to, and in some cases more pronounced than those of BPA, including Type 2 diabetes, and hypertension.^5–10 At the time of this study, there are no known epidemiological studies on BPS and behavior at any age.

Studying childhood behaviors at an early age is important because problematic behaviors in early childhood are predictive of problematic behaviors in later childhood and even adulthood.^11–14 Findings from the limited existing literature on phenols exposure and problematic early childhood behaviors are mixed.¹⁵ Overall, BPA exposure in pregnancy has been shown to be associated with increased anxiety, depression, aggression, and hyperactivity among children up to age 15 years.^16–21 When those findings are limited only to young children, however, the relationship becomes less clear, with studies using different instruments to measure child behavior and reporting positive,^17,22,23 negative,²¹ or no association^24,25. Currently, little is known about pregnancy or childhood BPS exposure and problematic childhood behaviors. In this context, we aimed to: 1) assess the association between pregnancy BPA and BPS urinary concentrations and behavior problems among children at age 2 years, and 2) assess the potential for child sex to modify associations between pregnancy bisphenols concentrations and problematic behaviors.

2. MATERIAL AND METHOD

2.1. Study population

Pregnant women (n=157) were recruited to be part of the pilot phase of the Illinois Kids Development Study (IKIDS) at 16–19 weeks of pregnancy from a local obstetrical clinic in Champaign-Urbana, IL. The pilot study took place between 2010–2012, and aimed to assess putative associations between biomarkers of prenatal phthalate and phenol exposure and infant cognition. Inclusion criteria for pregnant women were: no significant chronic medical conditions or medication use at the time of enrollment, 18–40 years of age, fluent in English, singleton pregnancy, living within a 45-mile radius of the University of Illinois campus, and not planning to relocate out of the area for at least one year after childbirth. Women were considered to have a a chronic health condition if they had been taking any prescription medications for 8 or more weeks. Some examples of conditions that women were excluded from participation for were: a thyroid condition, diabetes, cardiovascular disease, high blood pressure, and genital herpes. Women who enrolled but subsequently had medically complex pregnancies or pre-term birth (pre-37 weeks gestation) were not retained in the study. Additional details regarding cohort recruitment and retention can be found elsewhere (Yazdy et al., 2018). Our study sample began with 157 pregnant women, eight were removed due to the child being sent to the neonatal intensive care unit after birth or the researcher missing delivery, and 81 more were removed due to missing or incomplete Child Behavior Checklist (CBCL) data, or the participant withdrawing from the study, resulting in an analytical sample of 68 mother-child pairs (Supplemental Figure 1). The study protocol was reviewed and approved by the Human Subjects Committees of the University of Illinois at Urbana-Champaign and Carle Foundation Hospital, Urbana, Illinois, and written informed consent was obtained from all participants.

2.2. Exposure assessment

We quantified BPA and BPS in a pooled sample of urine collected across 5 time points during pregnancy: 16–19 weeks (early pregnancy) and 34–38 weeks (late pregnancy), which were first-morning urines collected at the participants’ homes, and three more occurring at 19–34 weeks, approximately 4 weeks apart, which were spot urines collected during regular prenatal obstetrical visits. For the three spot urine collections, which took place at the participants’ obstetrics office, there was a designated refrigerator at the clinic where the urine sample was placed. Research staff would then retrieve the sample from the clinic typically the same or the following day. For first-morning urines taken at home, participants would place the sample in their own refrigerator until research staff picked it up from the home, again typically the same day but at the latest the following day. All urine samples were collected with a polypropylene urine collection cup (Thermo Scientific Nalgene). Back in the lab, all urine samples (whether from the home or the healthcare providers’ office) were kept refrigerated until processing (within approximately 24 hours of collection). The next step was to aliquot urine using disposable polyethylene transfer pipets (Fisherbrand) into polypropylene vials with HDPE lids (Thermo Scientific Nalgene). At each of the five gestational timepoints when a urine sample was collected, an Eppendorf pipet was used to aliquot 0.9 ml of the sample volume into a “pooled” 5 ml cryovial. The pooled sample was then placed in a −20° C freezer. When the next urine sample was collected, the process was repeated: 0.9 ml of urine from that gestational timepoint’s sample was added to the cryovial on top of the frozen portion, and placed back into the freezer. When the pooled sample was complete, it was thawed to room temperature, mixed on a vortex mixer, and specific gravity was measured with a refractometer (TS400; Reichert). Periodically, field blanks were collected, transported, processed and stored using the same protocols as used for the urine. When ready for analysis, samples (urine and field blanks alike) were shipped on dry ice via overnight courier to the Centers for Disease Control and Prevention (CDC) laboratory in Atlanta, GA where they were analyzed using online solid phase extraction with high-performance isotope - dilution liquid chromatography - tandem mass spectrometry during the years 2010–2012.^26,27 CDC laboratory methods have excellent reproducibility, with coefficients of variation of 2.7–15%.^26,28,29 BPA (LOD 0.4 ng/mL) was detected in the urine of 100% of participants. BPS was found in 99.3% of participants (LOD, 0.1 ng/mL). We used the instrumental-reading values, even those that were below the LOD, for our analyses rather than imputing values below the LOD.^19,30–34 Urinary phenols concentrations were normalized for specific gravity.^35–40

2.3. Main outcomes of interest

The Child Behavior Checklist (CBCL) is one of several standardized measures commonly used to assess child behavior, and is valid and reliable.^41–44 We used the preschool version of the CBCL, which was completed by the child’s mother at 26–28 months of age. It consists of 100 items, each describing a problem behavior, divided into 2 composite scales, called “Internalizing” behaviors (36 items) and “Externalizing” behaviors (24 items), a Sleep problems syndrome scale, and “Other Problems.” This project focuses on internalizing and externalizing behaviors, and sleep problems. Internalizing behaviors are further broken down into four syndrome scales: emotionally reactive (9 items), anxious/depressed (8 items), somatic complaints (11 items), and withdrawn (8 items). The Externalizing behaviors composite scale is broken down into two syndrome scales: attention problems (5 items) and aggressive behavior (19 items). Sleep problems (7 items) is a stand-alone syndrome scale. The parent rates how well each item describes the child’s behavior in the past 2 months, using a 3-point scale, where higher scores correspond with more problematic behaviors, as such: “not true” = 0 points, “somewhat true” = 1 point, and “often true” = 2 points. Point values are summed to provide syndrome scores as well as broader scores for the Internalizing and Externalizing behavior composite scales. We modelled regression analyses on each of the two composite scales (Internalizing and Externalizing), as well as the seven syndrome scales. The sum of points for all of the items in any given scale denotes the severity of problems reported for a child on that scale, with higher scores corresponding to more problems in that domain.

2.4. Statistical analysis

The distributions of continuous variables were examined using measures of central tendency and normality plots. Descriptive statistics and Cronbach’s alpha were calculated for all scales. Comparisons of categorical data were performed using the Chi square test or Fisher’s exact test. Continuous data were compared across groups using the t-test if they were normally distributed, and the Wilcoxon 2-sample test with t-approximation if not. Variables used to adjust the models were objectively determined using directed acyclic graphs (DAGs) and prior published literature on the associations of interest (Supplemental Figure 2).^15,45–47 Variables considered in the DAG were maternal age, child sex, gestational age, income, number of siblings, marital status, age of child at CBCL, education level, paternal education, smoking, drinking, prenatal vitamin use, delivery type, OB risk score, Peabody Picture Vocabulary Test, breastfeeding, and HOME score. The relationships between and among the variables were depicted as paths in a causal diagram, keeping in mind that variables could be related to 2 or more variables, and that paths could be bi-directional. Two investigators (SDG and SM) independently examined the DAGs and the distribution and correlations of the variables in the study data to reach a conclusion about which variables to include in the models (Supplemental Table 1). Among the key determining factors were that the selected variable must precede the outcome, and that the variable was not a mediator in the pathway between exposure and outcome. Differences in variables selected by both investigators were discussed and resolved. The final set of variables used to adjust the models were: maternal age during pregnancy, number of siblings, child sex, and child age at CBCL intake.

Non-linear associations of the exposure with the outcome were tested using Generalized Additive Models (GAMs). GAMs test a non-linear relationship between exposure and outcome using spline (smoothing) functions, in addition to the linear relationship. A non-linear relationship was determined by examining the analysis of deviance Chi-square and plots of the outcome as a function of the spline exposure (fit and 95% confidence interval (CI)). The point of curvature observed on the plot (inflexion point) indicated where the relationship changed between the exposure and the outcome. A non-linear relationship indicated the need to account for this change in direction in order to improve model fit. In addition to the main effects for exposure (linear and spline components), GAMs included the same covariates planned for the final models (maternal age, number of siblings, child sex, child age at CBCL intake, exposure x child sex interaction).⁴⁸ An interaction term [(bpa_p − 2)*dummy;] was also included, where dummy is a binary variable with a value of 1 if BPA>2 and a value of 0 when BPA<2. Analysis by GAMs indicated that there was evidence of a non-linear relationship for BPA but not BPS with the outcomes. Thus, associations of BPA concentrations with all composite and syndrome scale outcomes were tested using piecewise generalized linear regression. As is the convention when using multivariable-adjusted piecewise linear regression, models were presented by BPA segment below the inflexion point of 2 ng/mL (BPA ≤ 2 ng/mL) and BPA > 2 ng/mL. The associations of the BPS concentrations with the outcomes were tested using generalized linear models. For both BPS and BPA exposures, we modelled the composite outcomes (Internalizing and Externalizing) as a Normal distribution because they were standardized. However, the syndrome scale outcomes were modelled as a Poisson distribution. Parameter estimates (beta coefficients) in the models testing the syndrome scale outcomes were exponentiated (base e) and interpreted as a percent change in the outcome scale for each unit increase in exposure concentrations.^21,49,50 when other variables are held constant in the model.⁵¹

Each concentration/outcome pair was tested in a separate model. The exposure x child sex interaction was also included. The concentration x child sex interaction was included in addition to the main effects. Parameter estimates from the exposure x child sex are presented, as well as child sex specific estimates from this exposure x child sex interaction using post-estimation procedures.

Model adequacy and fit was examined using the Akaike Information Criterion (AIC),⁵² deviance divided by its degrees of freedom (optimal fit when close to 1),⁵³ the variance of the Pearson residual,⁵⁴ and residual plots. Influential observations were identified using Cook’s D.⁵⁵ As a sensitivity analysis, the models were tested with and without the influential observations to determine whether the model parameter estimates changed. However, the models presented use all of the observations. A 2-tailed P value < 0.05 was considered statistically significant. No adjustment for multiple testing was performed because this is a preliminary investigation and results interpretation should focus on magnitude and confidence intervals of estimates. The adjusted relationship between the exposure and outcome were plotted using penalized B-spline curves with the PBSPLINE statement in the SGPANEL procedure in SAS. All analyses were conducted using Statistical Analysis Software (SAS) version 9.4 (SAS Institute, Cary, NC, USA).

3. RESULTS

Table 1 shows descriptive characteristics of the study sample. Over 80% of the mothers in our sample had a bachelor’s degree or higher, and a majority had an annual household income of over $50,000. Almost all mothers were married and non-smokers during pregnancy. Average maternal age was 29.7 years. Sex of the child was nearly equally split, with 48.5% being female. Standardized Cronbach Alpha values for the Internalizing and Externalizing behavior scales were 0.59 and 0.75 respectively, indicating that the scores in the CBCL scales were moderately internally valid in this particular sample. We also compared exposure and outcome scores by child sex (Supplemental Table 2) as well as those included in versus those excluded from the study to assess selection bias (Supplemental Table 3).

Table 1.

Sample characteristics (N=68).

	n (%) or mean±SD	Range
Demographic data
Maternal education
Some college	13 (19.1)
College graduate or higher	55 (80.9)
Household income
Not reported	1 (1.5)
Under $50,000	11 (16.2)
$50,000-<$100,000	45 (66.2)
≥$100,000	11 (16.2)
Marital status
Single	1 (1.5)
Married/living together	67 (98.5)
Maternal race
Non-White	8 (11.8)
White	60 (88.2)
Child sex
Female	33 (48.5)
Number of siblings
0	28 (41.2)
1	26 (38.2)
2–3	14 (20.6)
Smoking during pregnancy
None	67 (98.5)
Maternal age (years)	29.7±3.0	23–38
Child age at time of CBCL (months)	28.5±1.5	26–32
Biomarkers concentrations
BPA (ng/mL)	1.6±1.1	0.4–7.2
BPS (ng/mL)	0.4±0.5	0.1–3.5
CBCL scales
Internalizing behaviors	42.0±7.6	29–67
Emotionally reactive	1.2±1.7	0–10
Anxious-depressed	1.1±1.6	0–9
Somatic complaints	1.1±1.2	0–4
Withdrawn	0.6±0.8	0–3
Externalizing behaviors	44.4±9.2	28–71
Attention problems	1.5±1.4	0–8
Aggressive	7.2±5.6	0–27
Sleep problems	2.0±2.4	0–14

SD, standard deviation; BPA, bisphenol A; BPS, bisphenol S

We first examined the relationship between BPA and BPS concentrations and the composite CBCL scales—internalizing behaviors and externalizing behaviors. BPA was significantly associated with fewer internalizing behaviors among 2-year olds (β=−6.01; 95% CI −11.33, −0.69; p=0.028) but not with externalizing behaviors (Supplemental Table 4). When presenting by BPA segment (≤2 ng/mL vs >2 ng/mL) and child sex, the only significant association was among males in the lower BPA segment, where each unit increase in BPA concentration (ng/mL) was associated with a 10.48 point decrease in the internalizing behaviors score, on average (Table 2). BPS was not significantly associated with either of the composite scales (Supplemental Table 4). Our next step was to examine associations with the syndrome scale components of each composite scale, along with the sleep problems syndrome scale, which stands alone.

Table 2.

Multivariable piecewise linear regression testing the association of BPA with the CBCL composite scores.

	BPA≤2 ng/mL (n=50; 23 girls, 27 boys)			BPA>2 ng/mL (n=18; 10 girls, 8 boys)
	β	95% CI	P	β	95% CI	P
Internalizing behavior
Child sex: male	−10.48	−18.28, −2.68	0.009	3.45	−1.94, 8.84	0.184
Child sex: female	−1.22	−7.66, 5.23	0.705	6.91	−6.29, 20.11	0.271
BPA x child sex (female vs. male)	9.27	−0.71, 19.24	0.068	3.46	−10.41, 17.33	0.591
Externalizing behavior
Child sex: male	109.11	−38.26, 256.48	0.143	36.55	−146.35, 219.45	0.669
Child sex: female	99.17	−35.29, 233.62	0.144	33.88	−133.64, 201.39	0.665
BPA x child sex (female vs. male)	−9.95	−22.87, 2.97	0.128	−2.67	−18.06, 12.72	0.709

Table 3 shows the results of multivariable-adjusted piecewise generalized linear regression assessing associations between BPA by segment, and each syndrome scale outcome, with sex-specific estimates. Interaction terms were significant for three of the seven syndrome scale outcomes: anxious-depressed (p=0.006 for BPA≤2 ng/mL, and p=0.037 for BPA>2 ng/mL), aggressive behavior (p=0.002 for BPA≤2), and sleep problems (p=0.013 for BPA>2). We also conducted post-estimation analyses by sex and found that overall, in the lower BPA segment associations tended to be inverse (fewer reported problems) and among males, but in the higher BPA segment, associations were positive (more reported behavior problems) and tended to be in females. For example, for every unit increase in BPA concentration, scores on the anxious-depressed syndrome scale for boys in the lower BPA segment decreased (exp(β)=0.19; 95% CI 0.06, 0.58; P=0.004) whereas anxious-depressed scores for girls in the higher BPA segment increased (exp(β)=5.72; 95% CI 1.33, 24.53; P=0.019). Similarly, we identified sex-specific associations between BPA and aggressive behavior, and this time the inverse association persisted in males in the lower BPA segment (exp(β)=0.39; 95% CI 0.26, 0.62; P<0.0001), whereas in the higher BPA segment, BPA was associated with more aggressive behavior problems in boys (P=0.001) and girls (P=0.002) alike. The trend for the sleep problems syndrome scale was consistent with those already discussed: in the lower BPA segment, chemical exposure was statistically associated with better scores (fewer reported sleep problems) among males, and worse scores (more sleep problems) among females in the higher BPA segment (exp(β)=2.73; 95% CI 1.39, 5.34; P=0.003).

Table 3.

Multivariable piecewise linear regression testing the association of BPA with the outcome subscales, by sex and segments below and above the inflexion point.

	BPA≤2 ng/mL (n=50; 23 girls, 27 boys)			BPA>2 ng/mL (n=18; 10 girls, 8 boys)
	Exp(β)	95% CI	P	Exp(β)	95% CI	P
Emotionally reactive
Child sex: male	0.32	0.11, 0.90	0.031	1.42	0.98, 2.07	0.064
Child sex: female	0.61	0.28, 1.33	0.214	1.21	0.36, 4.07	0.753
BPA x child sex (female vs. male)	1.92	0.53, 6.99	0.325	0.85	0.26, 2.80	0.794
Anxious-depressed
Child sex: male	0.19	0.06, 0.58	0.004	1.33	0.67, 2.62	0.410
Child sex: female	1.33	0.56, 3.14	0.520	5.72	1.33, 24.53	0.019
BPA x child sex (female vs. male)	7.03	1.75, 28.19	0.006	4.30	1.09, 16.91	0.037
Somatic complaints
Child sex: male	0.68	0.20, 2.25	0.525	1.25	0.86, 1.81	0.234
Child sex: female	1.17	0.45, 3.05	0.755	1.73	0.64, 4.69	0.279
BPA x child sex (female vs. male)	1.72	0.37, 7.97	0.488	1.385	0.52, 3.71	0.518
Withdrawn
Child sex: male	0.09	0.01, 0.76	0.026	1.17	0.75, 1.81	0.486
Child sex: female	0.49	0.12, 1.96	0.311	2.89	0.80, 10.45	0.105
BPA x child sex (female vs. male)	4.93	0.43, 56.47	0.199	2.48	0.72, 8.55	0.151
Attention problems
Child sex: male	0.63	0.24, 1.62	0.333	1.35	0.96, 1.90	0.089
Child sex: female	1.00	0.43, 2.31	0.998	1.34	0.35, 5.17	0.672
BPA x child sex (female vs. male)	1.60	0.46, 5.52	0.456	0.99	0.26, 3.81	0.992
Aggressive behavior
Child sex: male	0.39	0.26, 0.62	<0.0001	1.42	1.16, 1.72	0.001
Child sex: female	0.94	0.67, 1.33	0.733	2.24	1.36, 3.68	0.002
BPA x child sex (female vs. male)	2.37	1.36, 4.12	0.002	1.58	0.99, 2.53	0.057
Sleep problems
Child sex: male	0.32	0.14, 0.73	0.007	1.17	0.86, 1.58	0.312
Child sex: female	0.86	0.38, 1.93	0.707	2.73	1.39, 5.34	0.003
BPA x child sex	2.70	0.87, 8.44	0.087	2.33	1.19, 4.56	0.013

β, exponentiated parameter estimate

^*Models were adjusted for maternal age at registration, number of siblings in the home, and child age at time of CBCL intake.

Figure 1 shows the significant sex-specific associations in three panels, anxious-depressed (panel A), aggressive behavior (Panel B), and sleep problems (Panel C).

Figure 1.

Adjusted sex-specific associations between BPA urinary concentration (in ng/mL) and CBCL syndrome scales

When examining the relationship between BPS concentrations and problematic behaviors, we again first looked at the broader composite scales of the CBCL—internalizing and externalizing behaviors (Supplemental Table 4). We identified no main effects, but a marginally significant interaction effect between sex and BPS concentration for internalizing behaviors among 2-year olds (P=0.058). Figure 2 shows analyses by sex, with girls having an average increase of 13.96 points on the internalizing behaviors scale (indicating more behavior problems) for every unit increase in BPS (95% CI 2.16, 25.76; P=0.021). Significance did not persist among boys (β=1.63; 95% CI −2.50, 5.76; P=0.433). Although neither main nor sex effects were identified for externalizing behaviors, it should be noted that the overall F statistic was not significant (P=0.497), indicating poor model fit.

Figure 2.

Adjusted sex-specific association between BPS concentration (in ng/mL) and internalizing behaviors

Table 4 shows results of multivariable-adjusted generalized linear regression assessing pregnancy BPS concentration as well as BPS by sex interaction, and CBCL syndrome scale outcomes. The BPS x Sex interaction term was significant for only one syndrome scale--anxious-depressed (P=0.048). When examining sex-specific estimates, however, significance did not persist in either boys or girls. Sex-specific estimates across all syndrome scales did show a significant association between BPS and emotional reactivity among girls even in the absence of a significant BPS x Sex interaction term (Figure 3). In this case, scores on the emotionally reactive syndrome scale (indicating more problematic behavior) increased almost four-fold per unit increase in BPS among girls (exp(β)=3.92; 95% CI 1.16, 13.27; P=0.028).

Table 4.

Multivariable- generalized linear regression testing the association of BPS concentration with syndrome scales by sex.

	exp(β)	95% CI	P	Deviance/df	Variance of Pearson residual	AIC
Internalizing Syndrome Scales
Emotionally reactive				1.62	1.26	227.81
Boys	1.36	0.86, 2.15	0.190
Girls	3.92	1.16, 13.27	0.028
BPS x child sex (girls vs. boys)	2.88	0.74, 11.27	0.128
Anxious-depressed				1.67	1.43	208.30
Boys	0.52	0.18, 1.47	0.217
Girls	3.51	0.77, 15.99	0.105
BPS x child sex (girls vs. boys)	6.74	1.02, 44.49	0.048
Somatic complaints				1.21	1.01	191.51
Boys	1.67	0.96, 2.89	0.069
Girls	3.67	0.95, 14.16	0.059
BPS x child sex (girls vs. boys)	2.20	0.48, 10.05	0.308
Withdrawn
Boys	0.84	0.38, 1.88	0.677	1.23	1.02	153.60
Girls	1.93	0.22, 17.19	0.555
BPS x child sex (girls vs. boys)	2.29	0.21, 24.79	0.495
Externalizing Syndrome Scales
Attention problems				1.33	1.01	218.54
Boys	0.78	0.47, 1.31	0.349
Girls	0.77	0.16, 3.73	0.741
BPS x child sex (girls vs. boys)	0.98	0.18, 5.31	0.979
Aggressive behavior				4.37	3.64	508.50
Boys	0.85	0.67, 1.08	0.183
Girls	1.62	0.89, 2.94	0.112
BPS x child sex (girls vs. boys)	1.91	0.99, 3.69	0.055
Independent Syndrome Scale
Sleep problems				2.64	2.59	299.57
Boys	0.68	0.41, 1.13	0.134
Girls	1.56	0.43, 5.60	0.497
BPS x child sex	2.30	0.57, 9.37	0.244

exp(β), exponentiated parameter estimate; CI, confidence interval; AIC, Akaike Information Criterion; df, degrees of freedom

^*Models were adjusted for maternal age at registration, number of siblings in the home, and child age at time of CBCL intake.

Figure 3.

Adjusted sex-specific association between BPS concentration (in ng/mL) and emotionally reactive behavior.

4. DISCUSSION

In this analysis, we found that pregnancy BPA concentration was associated with fewer internalizing behaviors among 2-year olds, but there was no association of BPA with externalizing behaviors. We also found that sex modified the relationship between BPA and the ratings on several of the CBCL syndrome scales--anxious-depressed, aggressive, and sleep problems—where the association of higher BPA concentrations with problematic behavior was strong in girls, but in the lower BPA range the association was inverse and only among males. We identified no main effects of BPS on internalizing or externalizing behavior composite scales, but BPS was associated with an increase in parent-reported internalizing behaviors among girls only. Sex modified the relationship between BPS and emotionally reactive behaviors, where the association of higher BPS concentrations with higher behavior problem scores was significant for girls, but not boys.

To the best of our knowledge, this is the first study to examine the relationship between pregnancy BPA urinary concentrations and CBCL outcomes at age 2 years, and the first study to evaluate pregnancy BPS concentration and problematic child behaviors using any instrument. There is a small body of existing literature on prenatal BPA exposure and neurobehavior, but most studies are with children older than those in our sample. ^{16,17,19–22,56,57} In addition, the findings from published studies on children aged 4–15 years are mixed, with some identifying positive relationships where higher concentrations of BPA during pregnancy were associated with more behavior problems among children, consistent with our findings in the higher BPA concentrations and syndrome scale outcomes, and others reporting no association between the two. Only one study focused on pregnancy BPA concentrations and problematic behavior in children at age 2 years,¹⁷ using data from a US pregnancy cohort and assessing problematic behavior via the Behavior Assessment System for Children 2 (BASC-2) and the Behavior Rating Inventory of Executive Function-Preschool (BRIEF-P), not the CBCL. The BASC-2, similar to the CBCL, categorizes behavior problems as “internalizing” and “externalizing.” Findings indicated a positive multivariable-adjusted associations between pregnancy BPA and externalizing scores for females (β=2.9, 95% CI 0.2, 5.7), but no association among males (β=0.0, 95% CI −4.1, 4.2) at age 2. Although we did not identify associations with the internalizing and externalizing scales, we did see similar associations with syndrome scale outcomes, which are more specific than the broader composite scales, in the higher BPA segment among females.¹⁷

A few other studies have focused on young children, ages 1–4.^22,23,25,58 Among 3-year-olds, Braun et al. focused on subscales of the BASC-2 and clinical scales from the BRIEF-P, identifying a positive multivariable-adjusted association between pregnancy BPA urinary concentrations and anxiety and depression scale scores and a negative association between BPA and emotional control and inhibition (meaning that higher levels of BPA were associated with poorer emotional control and inhibition), all of which were stronger among girls than boys.²² In addition, BPA was associated with increased hyperactivity among girls (β=9.1; 95% CI 3.1, 15.0), but negatively with hyperactivity among boys (β=−6.3; 95% CI −12, −0.6), assessed by the BASC-2. Our findings are not directly comparable because the CBCL does not assess hyperactivity, but the direction and sex-specific nature of associations are consistent with the study described. Another pertinent study also focused on prenatal BPA exposure among 198 African-American and Dominican children ages 3–4 years, and preschool CBCL outcomes.²¹ Perera et al. treated BPA urinary concentration measured during pregnancy as increasing quartiles and identified significant associations between BPA and variable CBCL outcomes among boys versus girls. Boys in the highest prenatal BPA concentration quartile had 1.62 times greater (95% CI 1.13, 2.32) scores on the emotionally reactive syndrome scale, and 1.29 times greater (95% CI 1.09, 1.53) aggressive behavior syndrome scale scores, than girls. In girls, on the other hand, BPA was negatively associated with the anxious/depressed syndrome scale (0.75 times as high as boys, 95% CI 0.57, 0.99), and aggressive behavior (0.82 times as high, 95% CI 0.70, 0.97). Finally, consistent with the current study, Li et al. (2020) found the relationship between prenatal BPA exposure and CBCL outcomes at ages 2 and 4 to be non-linear. Key findings in this longitudinal study were a positive association between BPA and both composite scales (internalizing and externalizing), as well as the emotionally reactive, anxious-depressed, and somatic complaints syndrome scales.²³ Associations were stronger among boys than girls.

In summary, the literature on prenatal BPA exposure and behavior problems in early childhood is mixed. Our study is consistent with aspects of certain studies, but inconsistent with others. Again, our study was the only one to assess pregnancy BPA concentration and child behavior using the preschool CBCL at age 2 alone. Differences in findings, then, may be explained at least in part by methodological differences in study design and characteristics of the study sample. Child age is one difference across studies that may have impacted results; even in the studies that are most comparable to ours, the children in other studies tended to be older (3 years,²² 3–5 years,^21,58 2–4 years,²³ and 1.5–5 years²⁵) than IKIDS pilot study 2-year-olds. The preschool version of the CBCL is intended for use in children aged 1.5–5 years, but brain development is rapid and expansive during this neurodevelopmental period; even one year difference can mark many developmental changes.⁵⁹ Differences in outcome measure used (e.g. BASC vs CBCL) assess child behavior also introduce variability across studies. Although several instruments can be used to assess problematic early childhood behaviors validly and reliably, they may not be directly comparable. Studies on convergent validity have tended to be in older children who are known to have a specific disorder or have been referred for behavior problems,⁶⁰ and even though highly correlated, these instruments may have varying degrees of classification consistency.^61,62 England-Mason et al. (2020) conducted a study on phthalate exposure and CBCL outcomes compared to BASC outcomes; the findings were not equivalent across outcomes.⁶³ Another difference across studies is number of phenol measurement time points during pregnancy. As described, urine was collected and pooled from 5 gestational time points, a method which maximizes stability of the measure, given the short half-lives of environmental phenols. Other comparable studies relied on a single gestational measurement,^21,23,25 and one had three.²² Other differences across studies are analytical in nature, for example quantile categorization of BPA concentrations and the resulting focus on children with the most elevated prenatal BPA concentrations, as well as variability in potential confounders included in statistical models.

Biological mechanisms underpinning the observed associations are not yet clearly understood. Because of the endocrine-disrupting properties of phenols, they disrupt sex hormones including estrogen, which are known to be important for masculinization of the brain, although the mechanisms through which this occurs have been studied most extensively in rodents models.^64,65 The process may be less estrogen-dependent and more androgen-dependent in humans, although there is evidence bisphenols can also disrupt androgens.^66,67 Differences in chemical structure may account for observed differences in effects of BPA compared with BPS, though their analogous structures may explain some shared characteristics such as estrogenic, anti-estrogenic, androgenic, and anti-androgenic properties.^5,68,69 Given their structural similarity and evidence that they disrupt the same hormonal and neurotransmitter systems it is difficult to speculate about possible biological mechanisms for these differences. However, there is also a possibility that BPS has a non-estrogenic mechanism of action.⁷⁰

In contrast to BPA, currently little is known about pregnancy exposure to BPS, a BPA substitute, and early childhood behavior problems in humans. Several recent animal studies, however, have indicated a possible link between the two in zebrafish, nematode, and rodent models.^6,7,71,72 One C. elegans experiment showed that low-dose gestational BPS exposure had similar effects as that of BPA on behavior as adults.⁶ Kinch et al. found low doses of gestational BPS to have comparable effects (180–240% increases) on hypothalamic neurogenesis, which is tied to hyperactivity.⁷ As summarized in a review paper, several experimental animal studies have identified neurobehavioral effects of BPA analogs including gestational BPS. Effects included anxiety-like and depressive behaviors, increased velocity, and weight loss among male offspring.^71,72 Our findings are therefore somewhat consistent with the existing animal studies—the BPS x Sex interaction term was significant for the anxious-depressed syndrome scale (Table 4), but the association was not significant in either sex. We also saw that sex modified the association between BPS and emotional reactivity. BPS, which is typically a replacement for BPA, had negative impacts on child behavior in terms of the emotionally reactive syndrome scale, whereas BPA did not. Thus, the replacement may be more toxic in this way than the original compound it is replacing as has been demonstrated for other health outcomes including reproductive and cardiovascular.^73–75

The current study has a number of strengths. First, the IKIDS pilot study has a rich pool of covariate data, which allowed for a detailed DAG and covariate selection process, ensuring the most parsimonious multivariable-adjusted modeling. Second, we had direct measures of urinary concentrations of BPA and BPS during pregnancy. Third, urinary chemical biomarker concentrations were pooled across five gestational time points, which provides a more stable estimate of average exposure across pregnancy than studies that have only a single time point or even two or three pooled samples, due to the short half-lives and day-to-day variability in exposure to these phenols. Finally, the cohort study design is robust, enabling us to establish temporality of the association. We also experienced several limitations. First, our study was under-powered with a sample size of 68 children, and especially for those with BPA concentrations greater than 2 ng/mL, which resulted in several wide 95% CIs. However, the risk of a small sample size is mainly under-estimation of effect size rather than spurious findings.⁷⁶ Second, CBCL data were parent-reported, not directly observed by a child psychologist, but the extent to which this may bias our findings, if at all, is not well-established.³⁶ Children as young as those in our sample (aged 2 years), who are not yet as verbal as older children, may be more accurately observed by their parents over time than by a clinician in an unfamiliar setting.^36,77 Third, standardized reliability as indicated by Cronbach’s alpha, is adequate for the externalizing composite scale (0.75), while only moderately reliable for the internalizing scale (0.53), which indicates the scale may require further validation in this particular sample. Hence, the results should be interpreted with caution. Fourth, no adjustment for multiple comparisons was performed due to the exploratory nature of this investigation, so interpretation of findings should focus on magnitude and confidence intervals of estimates rather than solely on p-values. Fifth, selection bias was assessed by comparing the study sample with those excluded based on missing data (Table S3). Women excluded from this study had lower incomes, lower education level, and higher BPA concentrations compared to those included, but were similar on other variables compared. Sixth, BPA and BPS were weakly correlated and we examined only the effect of a single chemical at a time, which does not account for joint effects or mixtures of real-world exposure. Finally, we are unable to account for BPA or BPS exposure during infancy or early childhood, which may further affect the associations being examined.

5. CONCLUSIONS

Findings from the current study were mixed, indicating both positive and negative associations with CBCL outcomes. We identified inverse associations between BPA during pregnancy and internalizing behaviors among 2-year olds, and no association with externalizing behaviors. Child sex modified the relationship between BPA and anxious-depressed, aggressive, and sleep problems syndrome scales: increasing BPA was associated with fewer reported problems in the lower BPA segment among boys, and more reported problems in the higher BPA segment among girls. BPS was positively associated with internalizing behaviors and emotionally reactive behaviors, among girls but not boys. Study findings should be interpreted with caution given the small sample size. Future research should focus on elucidating biological mechanisms underlying the identified associations, as well as on epidemiological studies with more robust sample sizes, in order to confirm or refute our findings. Additionally, a longitudinal analysis within the same cohort would be beneficial in assessing changes across development.

Highlights.

• We evaluated pregnancy BPA and BPS, and problematic child behavior at age 2.

• BPA associated with fewer internalizing but not externalizing behavior scores.

• Sex modified BPA and anxious-depressed, aggressive, and sleep problems links.

• BPS was associated with increased internalizing behaviors among girls only.

• BPS was linked with emotionally reactive behaviors, among girls but not boys.