Prenatal Phthalate Exposure Is Associated with Childhood Behavior and Executive Functioning

Background Experimental and observational studies have reported biological consequences of phthalate exposure relevant to neurodevelopment. Objective Our goal was to examine the association of prenatal phthalate exposure with behavior and executive functioning at 4–9 years of age. Methods The Mount Sinai Children’s Environmental Health Study enrolled a multiethnic prenatal population in New York City between 1998 and 2002 (n = 404). Third-trimester maternal urines were collected and analyzed for phthalate metabolites. Children (n = 188, n = 365 visits) were assessed for cognitive and behavioral development between the ages of 4 and 9 years. Results In multivariate adjusted models, increased loge concentrations of low molecular weight (LMW) phthalate metabolites were associated with poorer scores on the aggression [β = 1.24; 95% confidence interval (CI), 0.15– 2.34], conduct problems (β = 2.40; 95% CI, 1.34–3.46), attention problems (β = 1.29; 95% CI, 0.16– 2.41), and depression (β = 1.18; 95% CI, 0.11–2.24) clinical scales; and externalizing problems (β = 1.75; 95% CI, 0.61–2.88) and behavioral symptom index (β = 1.55; 95% CI, 0.39–2.71) composite scales. Increased loge concentrations of LMW phthalates were also associated with poorer scores on the global executive composite index (β = 1.23; 95% CI, 0.09–2.36) and the emotional control scale (β = 1.33; 95% CI, 0.18– 2.49). Conclusion Behavioral domains adversely associated with prenatal exposure to LMW phthalates in our study are commonly found to be affected in children clinically diagnosed with conduct or attention deficit hyperactivity disorders.

Biomonitoring of phthalate metabolites has identified virtually ubiquitous human expo sure internationally and for all age groups (Centers for Disease Control and Prevention 2005). Phthalate metabolites have been detected in many body tissues including urine, blood, semen, amniotic fluid, and breast milk (National Research Council 2008). Despite the relatively rapid clearance of phthalate metabolites, urine phthalate biomarker meas urements are relatively stable over periods of days to months (Adibi et al. 2008;Hauser et al. 2004;Teitelbaum et al. 2008), probably because exposure sources and patterns of usage of phthalatecontaining products are com mon and fairly consistent (Hauser et al. 2004;Teitelbaum et al. 2008).
At least 10 different phthalates are used commercially as plasticizers, solvents, anti foam agents, or alcohol denaturants. High molecularweight phthalates (HMWP) [e.g., di(2ethylhexyl) phthalate (DEHP)] can be found in tubing, vinyl flooring, and wall covering. Lowmolecularweight phthalates (LMWP) (e.g., diethyl phthalate) more com monly can be present in personal care prod ucts (fragrances, shampoo, cosmetics, and nail polish). Phthalates are also found as both inert and active ingredients in some pesticide formulations. Human exposure to phthalates can occur through inhalation, ingestion, and dermal contact. Once absorbed, they are rap idly metabolized to monoesters, and the high molecularweight monoesters can undergo further oxidation to form oxidative metabo lites (National Research Council 2008).
Some phthalates have been regulated in consumer products in Europe and the United States (California Safe Cosmetics Act 2005; Consumer Product Safety Improvement Act 2008; Directive 2005/84/EC of the European Parliament and of the Council 2005). The regulations in the United States are aimed at reducing childhood exposure to DEHP, dibutyl phthalate (DBP), and benzyl butyl phthalate by regulating their presence in bath toys or other small plastic toys that can be placed in the mouth easily. However, in some circumstances, these phthalates have been replaced with others (e.g., diisononyl or diisodecylphthalates for DEHP) that remain unregulated. Regulatory action was based mainly on studies of male reproductive toxicity in both animals and humans, which is related to testicularbased androgen insuf ficiency (National Research Council 2008).
Antagonistic effects of phthalates on the thyroid gland in vivo and thyroid tissue in vitro have been reported (Hinton et al. 1986;Pereira et al. 2007; Poon et al. 1997;Price et al. 1988; Sugiyama et al. 2005). DBP has been associ ated with a dosedependent decrease in circu lating triiodothyronine and thyroxine (T 4 ) in rats (O'Connor et al. 2002). In humans, low serum free T 4 was associated with high uri nary concentrations of monobutyl phthalate (a metabolite of DBP) (Huang et al. 2007) and of mono(2ethylhexyl) phthalate (a metabolite of DEHP) during pregnancy (Meeker et al. 2007). Recently, phthalate exposure in childhood was associated with attention deficit hyperactivity disorder (ADHD) in a cross sectional study of Korean school children between the ages of 8 and 11 years (Kim et al. 2009).

Objective
The objective of the Mount Sinai Children's Environmental Health Center is to investi gate the role of prenatal toxicant exposures on childhood growth and neurodevelopment. We recently reported a relationship between prenatal maternal concentrations of phthalate metabolites and altered neonatal behavior, specifically in the orientation and motor scales and their overall quality of alertness, exam ined within 5 days of delivery (Engel et al. 2009). The consequences of prenatal expo sure on neurobehavioral development during childhood have not previously been reported. Therefore, we examined these relationships in a subset of our cohort who returned for followup visits between 4 and 9 years of age.
Background: Experimental and observational studies have reported biological consequences of phthalate exposure relevant to neurodevelopment. oBjective: Our goal was to examine the association of prenatal phthalate exposure with behavior and executive functioning at 4-9 years of age. Methods: The Mount Sinai Children's Environmental Health Study enrolled a multiethnic prenatal population in New York City between 1998 and 2002 (n = 404). Third-trimester maternal urines were collected and analyzed for phthalate metabolites. Children (n = 188, n = 365 visits) were assessed for cognitive and behavioral development between the ages of 4 and 9 years. results: In multivariate adjusted models, increased log e concentrations of low molecular weight (LMW) phthalate metabolites were associated with poorer scores on the aggression [β = 1.24; 95% confidence interval (CI), 0.

Enrollment of birth cohort and child follow-up.
The Mount Sinai Children's Environmental Health study enrolled a prospective multiethnic cohort of primiparous women with singleton pregnancies. Women presented for prenatal care either at the Mount Sinai Diagnostic and Treatment Center, which serves the predomi nantly minority East Harlem population, or at one of two private practices on the Upper East Side of Manhattan. Four hundred seventy nine mother-infant pairs were successfully recruited. Women delivered at the Mount Sinai Medical Center between May 1998 and July 2001 (Berkowitz et al. 2003(Berkowitz et al. , 2004. Seventyfive women were excluded for reasons detailed elsewhere (Engel et al. 2007), includ ing extreme prematurity; the final cohort was 404 women for whom birth data were avail able. Questionnaires were administered to participants during the third trimester of preg nancy to obtain information on sociodemo graphic characteristics, medical history, and lifestyle factors. A maternal spot urine sample was obtained between 25 and 40 weeks' ges tation (mean, 31.2 weeks). Delivery charac teristics and birth outcomes were obtained from a perinatal database maintained within the Mount Sinai Department of Obstetrics, Gynecology and Reproductive Science.
Women were invited to return for three followup visits when their children were between 4 and 9 years of age. We attempted to complete interviews in each of these three periods: 4.5-5.5 years; 6-6.5 years; and 7-9 years. The number of visits per child ranged from one to three (approximately 40% came once, 26% came twice, 34% came three times), totaling 365 visits completed by 188 children (Table 1). Among these were five children who were not included in the original birth cohort analysis (n = 404) because they were not delivered at Mount Sinai (and therefore birth outcome informa tion was unavailable), although they returned for followup study visits. The study was approved by the Institutional Review Board of the Mount Sinai School of Medicine; par ticipants provided written informed consent before the study, and children ≥ 7 years of age provided assent.
Phthalate metabolite measurements. Sufficient maternal urine remained for 177 women to measure phthalate metabolite concentrations. Maternal urine samples were analyzed at the Centers for Disease Control and Prevention for 10 phthalate metabolites. Methods and quality control procedures have been described previously (Kato et al. 2005;Silva et al. 2008). To limit the number of sta tistical tests performed, phthalate metabolites were grouped into two categories defined by the molecular weight of the monoesters [high (> 250 Da) and low (< 250 Da)], as they have similar biologic activity as their parent diesters and come from similar environmental sources (Wolff et al. 2008). The concentrations, by molar sum and individual metabolite, have been previously reported (Wolff et al. 2008). To account for urine dilution, we included log creatinine in models where metabolites were considered continuous, logtransformed vari ables. When metabolites were considered in tertiles, creatininebased concentrations (micro moles per gram creatinine) were used to derive tertile cut points.
Behavior and executive functioning outcomes. At each visit, mothers completed the parentreport forms of the Behavior Rating Inventory of Executive Function (BRIEF), and the Behavior Assessment System for Children Parent Rating Scales (BASCPRS). These are standardized instruments commonly used in both research and clinical environments that have been validated in diverse samples and have known psychometric properties (Gioia et al. 2000;Reynolds and Kamphaus 1998). Mothers also provided current sociodemo graphic and other information on their chil dren. Questionnaires were available in both Spanish and English.
The BRIEF is an 86item questionnaire designed to assess executive cognitive func tion in children 5-18 years of age. Executive functions are used to achieve goals that require planning and holding in memory a multi step sequence of thoughts or actions, to moni tor and control attention and emotion, to inhibit inappropriate behaviors, and to formu late mental models based on life experiences (Pennington et al. 1997). The BRIEF is com posed of eight clinical scales: • Inhibit-the ability to control impulses • Shift-the ability to transition between situations • Emotional control-the ability to modulate emotional responses • Initiate-the ability to begin a task • Working memory-the ability to retain information for task completion • Plan/organize-the ability to anticipate future events, set goals, and develop a system atic plan of action • Organization of materials-the ability keep workspace orderly • Monitor-the ability to asses personal per formance and to register the effect of one's own behavior on others. These eight scales generate two broad indexes: Behavioral Regulation Index (BRI) and Metacognition Index (MI). An over all score, the Global Executive Composite (GEC), is obtained from the raw scores for the MI and BRI. For all scales, higher scores indicate worse executive functioning.
The BASC (Reynolds and Kamphaus 1998) is designed to evaluate problematic behaviors in children and adolescents 2.5-18 years of age. The BASCPRS includes nine clinical scales to assess a child's adaptive and problem behaviors in home and community settings. Parents respond to 130 items on a 4point scale that ranges from never to almost always. Externalizing problems is a composite scale derived from the hyperactivity (includ ing both hyperactivity and impulsivity items), aggression, and conduct problems scale items. Internalizing problems is a composite of the anxiety, depression, and somatization scale items. Three additional scales are attention problems, atypicality, and withdrawal. The adaptive behavior skills composite com bines information from the adaptability, social skills, and leadership scale items. The Behavioral Symptoms Index (BSI) is the api cal summary score that assesses the overall level of behavioral functioning. For the clini cal and composite scales, higher scores indi cate more problem behaviors. For the adaptive scales, lower scores indicate more problem behaviors. Thresholds delineating an atrisk or clinically significant scaled score have been established. For the clinical scales, the atrisk threshold is a Tscore of 60-69, and the clini cally significant threshold is a Tscore of ≥ 70. For the adaptive scales, the atrisk threshold is a Tscore of 31-40, and the clinically signifi cant threshold is a Tscore of ≤ 30.
The BRIEF and BASC are standardized for age and sex and provide Tscores with a mean (± SD) of 50 ± 10. Both instruments have good internal consistency, test-retest reliability, content and construct validity, and convergent discriminate validity (Pizzitola 2002;Reynolds and Kamphaus 1998).
Statistical analyses. We analyzed data using SAS version 9.2 (SAS Institute Inc., Cary, NC). Both the BASC and BRIEF include standard validity scales to identify problematic observations. For the BASC, the Infrequency Index (F) includes items that are rarely endorsed when the assessment is valid and can indicate an excessively negative evaluation of the child, a failure to follow instructions, ran dom responding, or difficulty reading. Raw Fscores on the BASC Parent Rating Scales (PRS) can vary from 0 to 6, with higher scores warranting closer review. All surveys with Fscores greater than 3 (n = 2 surveys) were excluded a priori, because the validity of these surveys is questionable. In the case of Fscores of 2 (n = 15 surveys) or 3 (n = 10 surveys), a committee of three investigators reviewed the child's records for parental reports of clinically diagnosed or treated psychiatric or neurologic problems and current or past usage of psycho tropic medications. The reviewers also looked for evidence of invalid, inconsistent, or unusual response patterns. After review, we excluded 12 surveys for the following reasons: language difficulty (n = 2), evidence of random respond ing (n = 7), and overly negative or unrealistic evaluation of child's behavior (n = 3). We did not exclude surveys with Fscores of 2 or 3 if the parent reported negative behaviors and the child was currently receiving treatment for or had previously been diagnosed with psychiatric or neuro logic problems.
Children were invited to return for three visits, but compliance varied. For children who returned for more than one visit, we took the average of scalespecific Tscores. Because Tscores are agestandardized, no additional adjustment for child age in the model was included. We conducted multivariate analy ses using PROC GLM (SAS Institute Inc.). Independent variables were chosen based on their relationships with the outcomes of inter est, as well as their correlation with phthalate metabolite concentrations. Given that phtha lates are reported to be reproductive toxicants and exhibit hormonal antagonism, we a priori hypothesized that there may be sex interac tions. Therefore, we first examined the signifi cance of sex-phthalate interactions (α = 0.10), before considering sex as a possible confounder. We used a backward elimination method to obtain the most parsimonious model, elimi nating covariates that did not change the esti mate of the main effect by at least 10%. All models were adjusted for urinary creatinine concentration. As a first step, effects of phtha late metabolite concentration were examined using creatininebased tertiles. Because the trend in the leastsquare means across the ter tiles supported a linear effect, the final models were based on the log e linear term of phtha late metabolite concentrations (Figure 1). Alternative analytic approaches (e.g., repeated measures analyses) were examined to evaluate the consistency of our findings. In these mod els, the coefficient for visit did not significantly differ from zero, suggesting that scores did not systematically vary across visits. Additionally, the phthalate effects were very similar compar ing the average Tscores to the repeated meas ures models; thus, our final models were based on the average score.

Results
There were slight differences in the character istics of women who returned for followup versus the original birth cohort (Table 1). Compared with the original birth cohort, the women who returned for followup tended to be those who were slightly older at the time of enrollment. However, the racial/eth nic mix, educational attainment, marital sta tus, and median urinary concentrations of the metabolites of low and high molecular phthalates were similar. Among the children followed up, 54% were boys, 75% of the chil dren spoke only English at home, and 83% of mothers were the child's primary caretaker. Those with other caretakers generally reported either father or grandmother. HMWP were not associated with most of the BASC or BRIEF domains, except that increased logHMWP was associated with poorer scores on the adaptability scale of the BASC [β = -1.33; 95% confidence interval (CI), -2.53 to -0.14].
Low molecular weight phthalate (LMWP) concentrations were strongly related to a number of clinical and composite scales ( Table 2). In multivariate adjusted models, each logunit increase in LMWP metabo lite concentrations was associated with a 1.24point increase on the aggression scale [95% confidence interval (CI), 0.15-2.34], a 1.29point increase on the attention prob lems scale (95% CI, 0.16-2.41), a 2.4point increase on the conduct problems scale (95% CI, 1.34-3.46), and a 1.18point increase on the depression scale (95% CI, 0.11-2.24). Although the associations between urinary concentrations of phthalate metabolites and the adaptive scales did not reach conventional levels of statistical significance, there was a consistent pattern of poorer adaptive profiles with increasing metabolite concentrations. This pattern was also evident in the composite Adaptive Skills Index (β = -0.98; 95% CI, -2.05 to 0.09). However, the associations were stronger between phthalate metabolite concentrations and externalizing problems (β = 1.75; 95% CI, 0.61-2.88) and the BSI (β = 1.55; 95% CI, 0.39-2.71). These effects were not modified by the child's sex. Using creatininecorrected tertiles of phtha late exposure, we plotted the tertilespecific adjusted mean Tscore at the median micro molar concentration of metabolites of LMWP to examine the shape of the dose-response relationship (Figure 1). Consistent with the loglinear models, the strongest linear trends were demonstrated for conduct problems and externalizing problems; however, all scales demonstrated monotonically increas ing LSMEANS, with the exception of the   a Includes subjects with F-scores of 0-1 and adjusted for race, educational level of the primary caretaker, marital status of the primary caretaker, urinary creatinine, and including sex-LMWP interaction term. All scales except conduct problems and leadership included 161 subjects. Conduct problems and leadership items are queried only on the 6-to 11-year BASC-PRS, so the number of subjects with these scales was 149.
Few surveys met the threshold scores for at risk or clinically significant (Tscores ≥ 60 for the clinical and composite scales or ≤ 40 for the adaptive scales) in this population. However, in a sensitivity analysis, we examined whether greater than the median metabolite concentration was associated with increased risk of scoring in the atrisk or clinically signif icant range on the scales identified in Table 2. In general, we found that the effects were in the same direction, although only attention problems (n = 27 atrisk or clinically signifi cant surveys) reached statistical significance (relative risk = 2.66; 95% CI, 1.06-6.66).
We also examined the metabolite specific effects on the BASC domains (Table 4). In general, we found consistency between the LMWP sum effects and the metabolite specific models.
Except for aggression and conduct prob lems, the LMWP molar sum significance was replicated in at least two of the individual metabolite models for that scale. Although for monobutyl phthalate (MBP) only aggression and externalizing problems were statistically significant, the magnitude of the MBP asso ciations were very similar to the LMWP sum effects for attention problems, adaptability, and the BSI.
There were fewer consistent associations between LMWP metabolite concentra tions and executive functioning as measured with the BRIEF. However, similar to what was observed for the BASC, increased log e LMWP metabolite concentrations was asso ciated with poorer scores on the emotional control scale (β = 1.33; 95% CI, 0.18-2.49) and on the GEC index (β = 1.23; 95% CI, 0.09, 2.36) ( Table 5). There were no inter actions between phthalate exposure and the child's sex for the BRIEF scales. We also com pared the metabolitespecific models with the total LMWP molar sum (Table 6). There were a number of significant associations for monomethyl phthalate (MMP), however, this metabolite has the lowest concentration among all metabolites measured. MBP was associated with poorer scores on working memory (β = 1.53; 95% CI, -0.01 to 3.07), but in general, the molar sum effects were consistent in direction and magnitude with the metabolite specific effects.

Discussion
We report an association between prenatal exposure to LMWP and poorer parentrated behavioral and executive functioning profiles for children between the ages of 4 and 9 years. Specifically, higher LMWP metabolite con centrations were associated with poorer scores on the aggression, attention problems, con duct problems, depression, and externalizing problems scales, and for the overall BSI on the BASC. Similarly, poorer executive func tioning was indicated by elevated scores on the emotional control scale and on the GEC index of the BRIEF. These effects remained statistically significant among boys, even after restricting the eligible surveys to those with Fscores of 0 or 1. All of these effects were consistent with a dose-response gradient.
In psychometric validity studies com paring the BRIEF and BASC (Gioia et al. 2000), the BRIEF emotional control scale is strongly correlated (r > 0.6; p < 0.01) with the BASCPRS aggression, anxiety, and depres sion scales (Gioia et al. 2000). Also in these validity studies, the GEC index is strongly correlated with the BASCPRS aggression and attention problems scales. We find com parable associations among these measures, supporting the validity of our application of these instruments.
Taken as a whole, the profile of parent reported behaviors we find associated with prenatal LMWP metabolite concentrations is suggestive of the behavior profiles of children clinically diagnosed with disruptive behav ior disorders, for example, oppositional defi ant disorder, conduct disorder, or ADHD (Loeber et al. 2009;Reynolds and Kamphaus 1998, Table 13.25) The strongest associa tions we find are for conduct problems, exter nalizing problems, and the BSI, and ratings on these parentreport scales typically serve as one element in a clinical diagnosis. The strong associations with the composite scales (externalizing and BSI) reflect the fact that on every scale for which a higher score reflects more problem behaviors, the direction of the association suggests an adverse effect of pre natal LMWP exposure. The association for the GEC on the BRIEF also reflects the con sistency in the direction of association on that instrument.
ADHD and disruptive behavior disorders are highly comorbid conditions, and no clini cal diagnosis can be based solely on screening instruments reliant on the parent's percep tions and recollections of their child's behav iors. The instruments we used include specific  items and patterns of responding to iden tify some types of parent reporting bias (e.g., excessive negativity), but other biases might go undetected. Parents' perceptions are also based on behaviors occurring in family situa tions, whereas the child might show different behaviors at school, and parents might be unaware of the child's subjective experiences of anxiety and other forms of internalization.
Although few children in this study met the standard at risk or clinically significant criteria on the BASCPRS, the patterns across scales and the consistency in findings across instru ments warrant additional study on the role of prenatal exposure to LMWP in the emergence of disruptive behavior problems in children.
The mechanism underlying a possible association between phthalates and neurode velopment has not been established, but may include prenatal disruption of the mater nal thyroid hormone system (Breous et al. 2005;Hinton et al. 1986;Huang et al. 2007;Meeker et al. 2007;O'Connor et al. 2002;Pereira et al. 2007;Poon et al. 1997;Price et al. 1988;Sugiyama et al. 2005;Vermiglio et al. 2004) or activation of peroxisome pro liferatoractivated receptors (Braissant and Wahli 1998;Corton and Lapinskas 2005;Latini et al. 2008;Peters et al. 2000) (Xu et al. 2007). Early pregnancy mild hypothyroxine mia (normal thyroidstimulating hormone, abnormal free T 4 ) was found to be associated with abnormal motor and socialization quo tients in children at 18 months of age, which was preventable with early iodine supplemen tation (Berbel et al. 2009). Additionally, pre natal iodinedeficient hypothyroxinemia has been associated with ADHD in childhood (Vermiglio et al. 2004). However, additional research is needed to investigate these mecha nisms and determine their applicability to fetal and child neurodevelopment. We previ ously reported a relationship between prena tal urinary concentrations of some phthalate metabolites and neonatal behavior as meas ured by the Brazelton Neonatal Behavioral Assessment Scale (Engel et al. 2009). Here we extend those findings to behavior and execu tive functioning in childhood.
Direct child exposure to phthalates in the postnatal period may be independently associated with behavioral symptoms or may act cumulatively with prenatal exposure to increase risk. Recently, phthalate exposure during childhood was associated with ADHD in a crosssectional study of Korean children between the ages of 8 and 11 years (Kim et al. 2009). However, no previous study has exam ined the influence of prenatal exposure.
Although substantial attrition occurred during the 10 years since this cohort was recruited, we do not believe selection bias can account for our findings. For loss to followup to represent a bias in our study, phthalates would have to have a different relationship with behavior and executive functioning in the lost population. We carefully examined the characteristics of the full cohort (n = 404) and the followup cohort (n = 188) and found that the populations were quite similar, except for maternal age at enrollment (Table 1). This was expected, given that the youngest mothers tended to be more difficult to track using the contact information they provided and were also somewhat more likely to move out of the area shortly after delivery. However, there were no differences with respect to phthalate metabolite concentrations. Moreover, none of the mothers in our study were provided with their prenatal phthalate metabolite concen trations; therefore, their evaluation of their child's behavior was not influenced by their own prenatal exposure level.
Exposure to phthalates is ubiquitous, and although phthalate biological activity may be less severe than that of other endocrine disruptors, the potential public health impact of exposure to phthalates might be greater. We report a pattern of associations between prenatal exposure to some LMWP and child hood behavior and executive functioning that is consistent with domains typically affected in childhood oppositional defiant disorder, con duct disorder, and ADHD. Additionally, most associations were consistent with a monotonic dose response. The reported relations between phthalates and thyroid hormone make such findings biologically plausible; however, addi tional research is urgently needed to replicate these findings and determine the underlying etiologic mechanism. Preventive measures to reduce exposure during pregnancy may be warranted should these findings be verified.