Prenatal Bisphenol A Exposure and Early Childhood Behavior

1Bisphenol A (BPA) is a weak estrogenic monomer used to produce plastic polymers such as polycarbonate plastics and epoxy resins that can be found in toys, water supply pipes, medical tubing, and food container linings. More than 6 billion pounds of BPA were manufactured worldwide in 2003, making it one of the highest production-volume chemicals. Polymers made from BPA can be hydrolyzed in high temperature, acidic, and basic conditions, leading to BPA leaching into water and food containers and baby feeding bottles.

2The ubiquity of BPA-containing plastics makes human exposure nearly universal in developed countries. More than 90% of the U.S. population has detectable levels of urinary BPA. BPA has also been detected in the urine and serum of pregnant women and serum, plasma, and placenta of newborn infants.

3Given the role that estrogen and other aromatizable androgens play in neuronal connectivity, distribution of serotinergic fibers, synaptic function, and dendritic length, exposure to endocrine-disrupting compounds such as BPA may alter the course of normal neurodevelopment. Prenatal BPA exposure in rodent studies is associated with increased aggression, morphine- and methamphetamine-induced hyperlocomotion, and memory impairment. Some rodent studies suggest that sex modifies the effect of prenatal BPA exposure, whereas others do not. The direction of the interaction between sex and BPA exposure is not clear because most of these studies have used different BPA doses, routes of exposure, and end points.

4To date, no studies have examined the association between prenatal BPA exposure and neurodevelopment in children. Given the near universal exposure to BPA among the general U.S. population and the association between BPA exposure and neurodevelopment and behavior in animal studies, even small effects of BPA on childhood neurodevelopment could have profound public health impacts.

5We used data from the Health Outcomes and Measures of the Environment Study to examine the association between prenatal BPA exposure and behavior in 2-year-old children. We also investigated whether sex modified the association between prenatal BPA exposure and childhood behavior.

6Data were collected from mothers and their children participating in the Health Outcomes and Measures of the Environment Study, an ongoing prospective birth cohort in the Cincinnati, Ohio (USA), metropolitan area designed to examine low-level environmental toxin exposure and the efficacy of injury and lead hazard controls in the home. Beginning in March 2003, women were identified from seven prenatal clinics associated with three hospitals. Eligibility criteria for the study included ≤ 19 weeks of gestation, age ≥ 18 years, living in a house built before 1978, negative HIV status, and not taking medications for seizure or thyroid disorders. We mailed letters to 5,184 women who were ≥ 18 years of age and living in a house built before 1978. Of the 1,263 eligible women, 468 provided informed consent and enrolled in the study.

7Women provided three spot urine samples collected at approximately 16 and 26 weeks of gestation and at birth. Sixteen-week urine samples were collected when women were met by study staff at their prenatal care appointments; 26-week urine samples were collected when women received their glucose tolerance test for gestational diabetes screening. Women could have had their urine samples taken anytime during the screening test. Birth urine samples were typically collected before or within 24 hr of parturition.

8Urine was collected in polyethylene containers and stored at −20°C until shipped to the Centers for Disease Control and Prevention (CDC) Environmental Health Laboratories, where they were stored at or below −20°C. Some urine samples were stored for 4–5 years before analysis. Recent work from the CDC laboratories suggests that urinary BPA species are stable for at least 6 months and even as long as 30 months when the urine sample is stored at subfreezing temperatures.

9The concentration of total (free plus conjugated) species of urinary BPA was quantified using modified analytical methods previously described. Urine samples were treated with β‐glucuronidase/sulfatase to hydrolyze conjugated BPA species. After hydrolysis, urine samples were acidified and then BPA was preconcentrated using online solid-phase extraction. BPA concentrations were quantified using high-performance liquid chromatography–isotope-dilution tandem mass spectrometry. The limit of detection (LOD) was 0.4 μg/L. Concentrations below the LOD were given a value of – for statistical analyses. Four quality control (QC) samples were analyzed in each analytic run. The low-concentration QC samples (~ 2.8 μg/L) had a coefficient of variation (CV) of 6.9–9.2%, whereas the high-concentration QC samples (~ 9.7 μg/L) had a CV of 3.4–7.6%.

10We assessed children’s behavior at 2 years of age using the second edition of the Behavioral Assessment System for Children (BASC-2) Parent Rating Scale for preschoolers.The BASC-2 is a 134-item parent-reported assessment of a child’s adaptive and problem behaviors in community and home settings. The preschool version is appropriate for children 2–5 years of age and has excellent reliability and validity for assessing adaptive and maladaptive behaviors. Composite scores are derived by the publisher’s computerized scoring program for externalizing, internalizing, and Behavior Symptom Index (BSI) scales, each with clinical subscales. Scores were normalized to a mean (± SD) of 50 ± 10 during the national standardization of the instrument. On the clinical behavioral subscales, scores in the range of 60–69 are considered “at risk,” and scores ≥ 70 are considered clinically significant. Composite scores for externalizing behaviors consist of hyperactivity and aggression scales, and internalizing composite scores consist of depression, anxiety, and somatization scales. The BSI is a composite score that reflects the overall level of problem behaviors and consists of scales for aggression, hyperactivity, depression, and attention. We used the general combined sex-normalized scores for our analyses. Creatinine-adjusted prenatal BPA concentrations were not qualitatively or statistically different between children with and without BASC-2 scores.

11The BASC-2 uses the F-index as a measure of a parent’s tendency to be overly negative when describing their child. BASC-2 assessments with F-index scores > 3.0 are to be interpreted cautiously; however, in this study no BASC-2 assessments at 2 years of age had F-index ratings > 3.0.

12We used directed acyclic graphs (DAGs) to examine the role of potential confounders in our exposure–disease relation. Simulation studies have shown that DAGs are a better method to assess the role of confounding variables than are change in estimate and significance testing approaches. Because little is known about factors associated with BPA exposures, we adjusted for many of the same confounders that are thought to confound other environmental toxin–neurodevelopment relations, including maternal and child demographic factors such as maternal age (< 25, 25–34, and ≥ 35 years), race (white and nonwhite), education (< 12, 12, and > 12 years), marital status (married and nonmarried), annual household income (< $40,000, $40,000 to < $80,000, ≥ $80,000), and child sex. Maternal depression during pregnancy was measured at 20 weeks of gestation with the second edition of the Beck Depression Inventory (BDI-II). Women were classified as minimally or mildly depressed (BDI-II score < 20) and moderately or severely depressed (BDI-II score ≥ 20). Caregiving environment was measured by administering the Home Observation for Measurement of the Environment (HOME) Inventory during a 12-month home visit. HOME scores were categorized as < 35, 35–39, and ≥ 40 based on the distribution of the scores.

13Demographic characteristics of the 389 mothers and their singleton children were compared with the 249 dyads with complete urinary BPA and BASC-2 data. Descriptive statistics of BPA concentrations at each time point were calculated along with the proportion of measurements below the LOD. We also calculated the Pearson correlation coefficient between the nonadjusted and creatinine-adjusted BPA concentrations at each time point.

14We examined the shape of the association between maternal BPA concentration and children’s behavior using graphical and locally weighted scatterplot smoothing (LOESS) analysis. LOESS uses locally weighted polynomials to fit subsets of data using ordinary least squares regression, giving more weight to points closer to the point being estimated, and less weight to points farther away.

15We started our analyses by examining the association between the mean prenatal BPA concentration (i.e., the mean of 16- and 26-week and birth BPA concentrations) and externalizing, internalizing, and BSI scores using multivariable linear regression. We used continuous log10-transformed mean BPA concentrations and quartiles of the mean BPA concentration in separate models. Coefficients from the continuous models represent the mean change in BASC-2 scores for a 10-fold increase in BPA concentrations. Adjusted mean scores for each quartile were calculated, as well as the difference between the first quartile mean score and the other three quartiles’ mean score.

16Next, we analyzed the association between each individual BPA measurement and externalizing, internalizing, and BSI scores to examine whether the association was dependent on the timing of exposure. We constructed models using each log10-transformed BPA concentration and quartiles of BPA concentration as the independent variable.

17Finally, we examined the association between early prenatal BPA exposure and BASC-2 scores, by looking for differences in the association between BPA concentrations and BASC-2 scores based on the timing of the 16-week urine measurement. Each individual woman had only one urine sample taken at approximately 16 weeks of gestation. We estimated the gestational week of the 16-week measurement using women’s reported last menstrual period (LMP). We dichotomized the timing of the 16-week urine measurement at the median value of 16 weeks. We included a product interaction term between the log10-transformed BPA concentration and the dichotomous measurement time variable. This allowed us to estimate the association between BPA concentrations and BASC-2 scores separately for women with measurements ≤ 16 weeks of gestation and > 16 weeks of gestation.

18Based on previous animal literature suggesting that BPA may be associated with disruptions in sexually dimorphic behaviors, we estimated sex-specific associations between BPA concentrations and BASC-2 scores by including a product interaction term between the BPA concentration and child sex variables. Male- and female-specific slope coefficients were computed from regression models using continuous BPA concentration, whereas sex-specific means and differences were calculated from models using creatinine-adjusted BPA concentration quartiles. We assessed for the presence of sex modification by examining the p-value for the interaction term between BPA concentration and sex. We considered p-values < 0.10 to be indicative of effect measure modification.

19We used urinary creatinine concentrations to control for urine dilution in all of our analyses. In models using continuous mean prenatal BPA concentrations, each individual BPA concentration was divided by the respective urinary creatinine concentration before taking the mean of the measurements. In models using individual BPA concentrations, we included the log10-transformed urinary creatinine value from the respective measurement in our models. In analyses involving percentiles, we created percentiles from the distribution of creatinine-adjusted BPA concentrations. We excluded 10, 20, and 15 women with urinary creatinine values < 20 mg/dL at the 16-week, 26-week, and birth measurement, respectively, to reduce the influence of values from very dilute urine.

20Several sets of secondary analyses were conducted. First, we sought to determine if mean and time-specific prenatal (16 and 26 weeks of gestation and birth) serum cotinine or blood lead levels confounded the association between BPA concentrations and BASC-2 scores. Concurrent log10-transformed serum cotinine or blood lead levels were entered into respective log10-transformed continuous BPA models (i.e., 16-week cotinine included in 16-week BPA model). We also ran models that included both serum cotinine and blood lead levels. Second, we conducted an additional analysis that included Full Scale maternal IQ (Wechsler Abbreviated Scales of Intelligence) as a continuous variable in our model. We also conducted additional analyses that included women with urinary creatinine concentrations < 20 mg/dL to see whether this restriction changed our estimates. All analyses were conducted with SAS version 9.2 (SAS Institute Inc., Cary, NC, USA).

21The institutional review boards (IRBs) of the University of North Carolina–Chapel Hill, Cincinnati Children’s Hospital Medical Center, and CDC approved this study. The University of Cincinnati College of Medicine IRB was involved in the oversight of this study throughout all stages of planning and implementation. All mothers provided written informed consent for themselves and their children before enrollment in the study.

22Of the 468 women who enrolled in our study, 67 dropped out before delivery. We excluded nine sets of twins, and three stillborn children. Among the 389 women with singleton live births, all three urine samples were collected from 332 (85.3%) women. BASC-2 scores were available for 249 (64.0%) of the children whose mothers had all three urinary BPA measurements. Women from dyads with complete BPA and BASC-2 data were more likely to be white, better educated, wealthier, married, and 25–34 years of age than women in the original sample (Table 1). Among the 249 children born to women with all three prenatal BPA concentrations, the mean (± SD) externalizing, internalizing, and BSI scores were 47.6 ± 7.8 (range, 30–84), 44.8 ± 7.0 (range, 31–68), and 50.5 ± 6.8 (range, 34–74), respectively.

23Almost all (n = 247, 99.2%) women had at least one urine sample with a detectable urinary BPA concentration (Table 2; 16 weeks, 89.2%; 26 weeks, 89.6%; birth, 87.1%). One woman’s 26-week urine BPA concentration was 3 orders of magnitude higher than the median 26-week concentration. Excluding this woman’s results did not substantially alter our effect estimates. Creatinine-adjusted BPA concentration distributions were very similar across pregnancy and at birth. Log10-transformed BPA concentrations were modestly correlated at 16 and 26 weeks (Pearson r = 0.30), 16 weeks and birth (Pearson r = 0.26), and 26 weeks and birth (Pearson r = 0.28). We found little correlation between log10-transformed creatinine-adjusted BPA concentrations at 16 and 26 weeks (Pearson r = 0.11), 16 weeks and birth (Pearson r = 0.11), and 26 weeks and birth (Pearson r = 0.08)

24LOESS analysis revealed linear associations between mean and individual urinary BPA concentrations and BASC-2 scores. We found no association between mean prenatal BPA concentrations and externalizing, internalizing, or BSI scores for the entire sample. In sex-stratified analyses, we observed a positive association among females between log10-transformed mean prenatal BPA concentrations and externalizing [β = 6.0; 95% confidence interval (CI), 0.1–12.0] and BSI scores (β = 5.5; 95% CI, 0.3–10.7), but not among males (p-value for interaction < 0.05; Table 3).

25Covariate and creatinine-adjusted log10-transformed 16-week BPA concentrations were positively associated with externalizing scores (Table 4; β = 2.9; 95% CI, 0.2–5.7). The association between continuous 16-week BPA concentrations and externalizing scores was stronger among females (β = 4.8; 95% CI, 1.3–8.3) than for males (β = 1.1; 95% CI, −2.4 to 4.6). Sex also modified the association between 16-week BPA maternal concentrations and children’s BSI scores. Maternal BPA concentrations at 26 weeks were not associated with children’s externalizing behaviors. We found a modest inverse association between maternal 26-week BPA concentration and children’s internalizing and BSI scores (β = −2.2; 95% CI, −4.6 to 0.7).

26Children born from women in the top three quartiles of birth BPA concentrations had mean externalizing scores approximately 2 points higher than those of children from women in the lowest quartile. Males born from women in the second and third quartiles of BPA concentration had higher mean internalizing and BSI scores compared with males born to women in the first quartile.

27Mean externalizing scores increased across quartiles of maternal creatinine-adjusted 16-week BPA quartile among female children, whereas externalizing scores did not differ among males (Figure 1). Externalizing scores among females in the third and fourth quartiles were comparable to mean externalizing scores for males.

28The timing of the 16-week urine measurement modified the association between BPA concentrations and externalizing scores. A total of 242 women had LMP-based estimates of the 16-week measurement. Among these women, 134 (55.4%) had their urine measurements taken at ≤ 16 weeks of gestation (minimum, 11.2 weeks) and 108 (44.6%) had their measurements taken at > 16 weeks (maximum, 21.4 weeks). Confounder-adjusted log10-transformed BPA concentrations taken at ≤ 16 weeks of gestation were positively associated (β = 5.1; 95% CI, 1.5–8.6) with externalizing scores, whereas those taken at > 16 weeks of gestation were not (β = 0.6; 95% CI, −2.9 to 4.1). The positive association between BPA concentrations and externalizing behaviors was stronger in females (β = 7.0; 95% CI, 3.0–11.0; n = 70) than in males (β = 3.0; 95% CI, –1.0 to 7.0; n = 64) among mothers whose urine samples were taken at ≤ 16 weeks of gestation (p-value for BPA × sex interaction term < 0.10). We found no qualitative or statistical differences in the mean log10-transformed creatinine-adjusted BPA concentrations taken at ≤ 16 weeks and > 16 weeks. The proportion of males and females was not qualitatively or statistically different by the timing of the 16-week measurement.

29In secondary analyses, the inclusion of maternal blood lead levels, maternal serum cotinine levels, or both did not substantially alter our results. The inclusion of maternal IQ did not substantially alter any of our findings and was not associated with BASC-2 scores. Including women with urinary creatinine values < 20 mg/dL did not substantially change the estimates in any of our analyses. Our results were unchanged when we conducted analyses that included women with one or two missing BPA measurements.

30This is the first study to examine the association between prenatal BPA exposure and childhood behavior. We did not observe an association between mean urinary BPA concentrations and behavior scores among all children. However, we report an association between mean prenatal BPA concentrations and externalizing scores in females. Further examination revealed that the association between prenatal BPA concentrations and externalizing behaviors in 2-year-old girls was driven by maternal urine BPA concentrations measured at 16 weeks of gestation. The association between early BPA concentrations and externalizing scores was strongest among women with urine measurements taken at ≤ 16 weeks of gestation. The interaction between sex and 16-week BPA concentrations persisted in samples taken at ≤ 16 weeks of gestation.

31Consistent with our findings, some animal studies have shown that prenatal BPA exposure is associated with increased aggression, changes in the dopaminergic system in the limbic forebrain, and other neurobehavioral changes. To our knowledge, only one rodent study has examined for time-sensitive periods of BPA exposure. reported that BPA exposure during organogenesis and breast-feeding caused morphine-induced hyperlocomotion, preference for drug-paired place, and up-regulation of dopamine receptor function, whereas exposure at other gestational time periods did not.

32We observed an inverted U-shaped association between maternal urinary BPA concentrations and internalizing behaviors among male children in our sample. Others have written extensively about the possibility of a nonlinear dose–response curve between environmental estrogens and various end points.

33Our results suggest that child sex may modify the association between prenatal BPA exposure and externalizing behaviors. Interactions between child sex and prenatal exposure to neurodevelopmental toxicants have been observed in some epidemiologic studies; however, sex-specific effect measure modification is usually not reported in epidemiologic studies. Results from rodent studies have reported that sex modifies the effect of prenatal BPA exposure and that prenatal BPA exposure may abolish or reverse differences in sexually dimorphic behaviors. However, it is difficult to label effects as feminizing or masculinizing without knowing whether these end points are sexually dimorphic.

34Two issues must be considered when comparing the results of our study to those of rodent studies. First, the route of exposure in animal studies has included oral, subcutaneous, and direct injection at target organs, whereas humans are exposed predominantly through the oral route. Second, animal studies have used doses spanning several orders of magnitude, many of which are not relevant to human exposure.

35Prenatal urinary BPA concentrations among women in this study were similar to those from women of childbearing age from a nationally representative U.S. sample, from a previous study examining the association between BPA exposure and infant birth outcomes among minority pregnant women in New York City, and from pregnant women in Rotterdam, the Netherlands.

36This study has several limitations. First, the children in our sample were 2 years of age at the time of behavioral assessment. Patterns of behavior are variable during early childhood. However, longitudinal analyses of children reveal that inattentive/hyperactive behaviors tend to be stable over childhood, whereas aggressive behaviors decrease. Second, unmeasured confounding may be responsible for some or all of our observed associations. have shown that unmeasured confounding can be responsible for much of the observed association between environmental neurotoxicants and neurobehavioral function. Although we included many of the same confounders used in studies of other neurotoxicants, it is possible that we did not adequately assess parental psychopathology (e.g., parental attention deficit/hyperactivity disorder). Third, there is evidence that prenatal BPA exposure disrupts normal mother–child interactions in rodent models. Female mice exposed to BPA spend less time in their nest and nursing their offspring and more time grooming and resting alone. Therefore, associations between childhood behavior and prenatal BPA exposure may be mediated through maternal behavior toward the child. Finally, maternal BPA measurements taken at 26 weeks and birth may be influenced by the glucose tolerance test and birthing process, respectively. However, this is unlikely because creatinine-adjusted BPA concentration distributions were almost identical at each time point. We would have expected BPA concentrations to be higher at birth because women would be more likely to be exposed to BPA containing plastics, such as intravenous tubing, in the hospital setting. The lack of a difference could be attributable to increased plasma volume from intravenous fluids, hospital diet or fasting near delivery, metabolic changes from the birthing process, hospital policies that limit BPA-containing plastics, or a combination of all of these factors.

37Exposure misspecification may be responsible for our observed results. This could bias our results toward or away from the null, depending on the nature of the misspecification and characterization of the exposure (continuous vs. categorical). BPA has a short half-life in biological tissues, although a recent analysis of the 2003–2004 National Health and Nutrition Examination Survey suggests that BPA may have a longer than expected half-life. Urinary BPA concentrations have a moderate degree of intraindividual variability, making it difficult to accurately characterize exposure from a single measurement. The lack of prior literature on the appropriate specification of BPA exposure during pregnancy makes it difficult to determine if a single measurement or summary measurement of BPA concentration more accurately quantifies exposure. Because women did not have repeated urine measurements around 16 weeks of gestation, we assumed that timing of the 16-week urine measurement was random. The results of these analyses should be interpreted cautiously because this assumption may not be true. As recommended previously by, additional research into characterizing BPA exposure is necessary. An additional source of exposure misspecification may have arisen from BPA degradation in samples that were stored for > 3 years. However, prior work indicates that storing urine samples at subfreezing temperatures for as long as 30 months does not result in substantial loss of BPA.

38Another limitation is that we were not able to examine whether postnatal BPA exposure was associated with childhood behavior. We are planning to examine this association in future studies of this cohort.

39Selection bias is another potential limitation of our study because women were required to be living in pre-1978 housing. It is possible that children living in older housing stock come from lower socioeconomic status, making them more likely to have behavior problems. However, this is unlikely because families in our sample have higher levels of education and income than the general population.

40Finally, there were two statistical limitations to this study. First, it could be argued that our observed associations are spurious because of the large number of associations we examined. However, multiple comparison arguments assume that randomness underlies the variability of all the observed associations. It is unlikely that our results are attributable to random variation because we observed a consistent association between prenatal urinary BPA concentrations and externalizing behaviors in females that became stronger in measurements taken earlier in pregnancy. Second, our power to detect associations between maternal BPA exposure and childhood behavior was diminished in our analyses examining sex- and time-specific associations, resulting in imprecise estimates.

41This study has several strengths. First, we had three urinary BPA measurements in the latter two-thirds of pregnancy to estimate gestational exposure. This allowed us to examine whether some periods of gestation are sensitive to BPA exposure. Second, we used a valid and reliable measure of adaptive and problem behaviors in children. Finally, we are continuing to follow these children through 5 years of age, which will allow us to examine whether our observed associations persist throughout early childhood.

42Mean prenatal BPA concentrations were not associated with childhood behavior. However, we did observe a positive association between these concentrations and externalizing behaviors among females. Early second-trimester maternal urinary BPA concentrations were positively associated with externalizing behaviors and global behavior scores among all children, but especially among 2-year-old female children. Our results suggest that BPA concentrations in samples taken earlier in pregnancy may be more strongly associated with externalizing scores than those in later samples. The reported associations and interactions between child sex and timing of BPA exposure should be viewed cautiously because these results could be biased by exposure misspecification or residual confounding. Future research should aim to develop methods to accurately characterize BPA exposure, especially during the first trimester of pregnancy; to validate the associations observed in this cohort; and to examine the association between pre- and postnatal BPA exposure and behavior in later childhood.