The associations between prenatal phthalate exposure measured in child meconium and cognitive functioning of 12-month-old children in two cohorts at elevated risk for adverse neurodevelopment

Abstract

Background

Phthalate metabolites in gestational-maternal urine represents short-term maternal exposure, but meconium, the newborn’s first stool may better capture cumulative fetal exposure. We quantified phthalate metabolites in meconium from two cohorts of children at higher risk of adverse neurodevelopment and evaluated associations with their cognitive function at 12 months.

Methods

Meconium phthalate metabolites were quantified in the Safe Passage Study (SPS), N=720, a pregnancy cohort with high community-levels of prenatal alcohol use, and the Early Autism Risk Longitudinal Investigation (EARLI), N=236, a high familial autism risk pregnancy cohort. EARLI also had second and third trimester (T2/T3) maternal urine for exposure assessment. Molar sum of di(2-ethylhexyl) ($\sum \mathrm{DEHP}$) metabolites and an anti-androgenic score (AAS) using mono-isobutyl, mono-n-butyl, monobenzyl (MBZP), and DEHP metabolites were computed. Cognitive function was assessed at 12 months using the Mullen Scales of Early Learning-Composite (ELC). Multivariable linear regression assessed associations between log_e-transformed metabolites and ELC. Quadratic terms explored nonlinearity and interaction terms of metabolite by child’s sex examined effect modification.

Results

In SPS, MBzP (β_Linear= −6.73; 95% CI: −12.04, −1.42; β_quadratic=1.95; 0.27, 3.62) and mono(2-ethyl-5-carboxypentyl), (β_Linear= −3.81; −7.53, −0.27; β_quadratic=0.93; 0.09, 1.77) had U-shaped associations with ELC. In EARLI, T2 urine mono-carboxyisononyl was associated with linear decrease in ELC, indicating lower cognitive function. Interaction with sex was suggested (P<0.2) for several urine metabolites, mostly indicating negative association between phthalates and ELC among girls but reversed among boys. Only mono-isononyl phthalate and $\sum \mathrm{DEHP}$ had consistent main effect associations across matrixes and cohorts, but similar interaction with sex was observed for meconium-measured $\sum \mathrm{DEHP}$, AAS, MBzP, and mono(2-ethylhexyl) in both cohorts.

Conclusions

Few phthalate metabolites were consistently associated with children’s cognitive function, but a similar set of meconium metabolites from both cohorts displayed sex-specific associations. Gestational phthalate exposure may have sexually-dimorphic associations with early cognitive function in children at higher risk for adverse neurodevelopment.

¹Background

Phthalates are a multifunctional class of chemicals primarily used to improve the flexibility of plastic products. They have industrial applications as plasticizers in manufacture of commercial and household polyvinyl chloride products, food packaging, and have also been used as solvents and adhesives in cosmetics and personal care products (Fisher et al., 2019; Schettler, 2006). Due to the lack of a strong chemical bond, phthalates tend to leach from plastic products (Guart et al., 2011) and contaminate the built environment. Several commonly used phthalates have been detected in population biomonitoring programs such as the National Health and Nutrition Examination Survey (NHANES), where levels in women, especially those of childbearing age, were higher than those in men(Silva et al., 2004). Once absorbed, the parent phthalate is rapidly metabolized into primary metabolites(Frederiksen et al., 2007). Among phthalates with long molecular chains like di(2-ethylhexyl) phthalate (DEHP), primary metabolites are further oxidized into several secondary metabolites(Frederiksen et al., 2007). The molar sums of these metabolites can be useful to assess exposure to the parent compound or those compounds with similar biological activity or from the similar environmental sources.(Wolff et al., 2008) Both primary and secondary metabolites cross the placenta(Li et al., 2018; Mose et al., 2007) and therefore, maternal exposure in pregnancy could directly affect the fetus. Several recent studies have reported that prenatal phthalate exposure is associated with changes in aspects of fetal neurodevelopment(Ejaredar et al., 2015; Miodovnik et al., 2014; O’Shaughnessy et al., 2021).

Phthalates are thought to affect fetal neurodevelopment mainly through disruption of thyroid hormone homeostasis(Braun, 2017) but other theories involving endocrine-disruption have also been proposed (O’Shaughnessy et al., 2021). Prenatal phthalates and their corresponding biologically created metabolites (mono-n-butyl phthalate (MBP), mono-carboxyisooctyl phthalate (MCOP), mono-ethyl phthalate (MEP), mono(2-ethylhexyl) phthalate (MEHP), and mono(2-ethyl-5-hydroxyhexyl) phthalate (MEHHP) have been reported to reduce maternal thyroid hormones(Huang et al., 2007; Johns, Ferguson, et al., 2015; Romano et al., 2018; Yao et al., 2016). Since maternal thyroid hormones are critical for human fetal brain development, thyroid hormone deficiencies may impair aspects of child neurodevelopment, including cognition(Moog et al., 2017). Phthalates are also reported to be associated with reduction in sex-steroids (testosterone, estradiol, progesterone) necessary for the sexually dimorphic organization of the fetal brain(Araki et al., 2014), and could interfere with fetal neurodevelopmental (Braun, 2017; Gore et al., 2014) through this pathway.

Epidemiologic studies have reported inverse associations of specific phthalate metabolites (MBP, mono-isobutyl phthalate (MiBP), mono benzyl phthalate (MBZP), MEP, $\sum \mathrm{DEHP}$ – the molar sum of 4 DEHP metabolites) with cognitive development(Factor-Litvak et al., 2014; Ipapo et al., 2017; Li et al., 2019; Merced-Nieves et al., 2021; Tanner et al., 2020; Torres-Olascoaga et al., 2020). However, other studies report no associations (Gascon et al., 2015; Huang et al., 2015; Kim et al., 2017; Nakiwala et al., 2018; Polanska et al., 2014), and a few have reported protective associations with certain metabolites (mono(2-ethyl-5-oxohexyl) phthalate (MEOHP), a secondary metabolite of DEHP ) (Jankowska et al., 2019). Several studies have observed that the child’s sex modifies the effect of phthalate metabolites on cognitive development(Hyland et al., 2019; Ipapo et al., 2017; Merced-Nieves et al., 2021; Torres-Olascoaga et al., 2020), but results have varied by direction of association and across studies, preventing firm conclusions(Radke et al., 2020). All currently published reports have focused on general population samples. There has yet to be an investigation in children at higher risk for neurodevelopmental problems based on family history or increased likelihood of exposure to other developmental neurotoxicants. The investigation of high-risk populations may be advantageous as the potential for detection of associations may be enhanced in such samples due to expected higher outcome frequency.

In almost all published epidemiological studies of prenatal phthalate exposure and fetal neurodevelopment, metabolites were quantified in maternal urine, since urine is easy to collect and is the major route of metabolite elimination(Frederiksen et al., 2007). However, phthalate metabolites are eliminated quite rapidly from the body and thus measurement in maternal urine captures only short-term maternal and fetal exposure. Meconium, the newborn’s first stool, is a biosample from the fetal compartment consisting of water, lipids, sterols and end products of swallowed amniotic fluid and shed skin. Since meconium starts accumulating at about 12 weeks gestation, meconium may better reflect cumulative fetal exposure (Bearer, 2003; Shwachman, 1981). Although easily collectible, meconium has not been widely used to assess prenatal exposure to environmental chemicals, but it has been used to quantify other endogenous and exogenous exposures like hormones (androgens) and cocaine(Frey et al., 2017; Ostrea et al., 2009). The analytical chemistry methods to detect phthalate metabolites in meconium were developed(Kato et al., 2006) over a decade ago. However, the Plastics and Personal Care Products use in Pregnancy (P4) (Arbuckle et al., 2016) study was the first large epidemiological study to utilize this method and we replicated it using meconium from children in a high familial autism risk birth cohort(Mathew et al., 2021). Currently no epidemiological investigation has quantified phthalate metabolites in meconium and used the levels to evaluate associations of prenatal phthalate exposure with adverse neurodevelopment.

Therefore, our objective was to further explore the association between prenatal phthalate exposure and child cognitive function by capitalizing on the availability of higher risk population samples (a cohort of children with an older sibling diagnosed with an autism spectrum disorder (ASD), and a cohort drawn from communities with higher prevalence of prenatal alcohol exposure), and measurement in meconium.

Methods

Study populations

The Safe Passage study (SPS) was a large prospective multicenter study designed by the Prenatal Alcohol in SIDS and Stillbirth (PASS) network to evaluate the association between prenatal alcohol use on Sudden Infant Death Syndrome (SIDS) and Fetal Alcohol Spectrum Disorders (FASD) among surviving infants. The study planned to enroll 12,000 pregnant women between August 2007 and January 2015 at four sites in the Northern Plains (3 Native American reservations, 1 urban clinic in North and South Dakota), United States; and in Cape Town, South Africa. In these communities, approximately 50% of women consumed alcohol in the first trimester, and 23% and 15% continued to drink in the second and third trimesters(Dukes et al., 2017). It has been previously reported that children who had fetal alcohol exposure are more likely to have deficits in many aspects of cognitive function (Mattson et al., 2019). Eligibility criteria for the Safe Passage study included ability to speak English or Afrikaans, pregnant with one or two fetuses between 6 weeks and up to delivery, and over 16 years of age. Additional details of study design are reported elsewhere(Dukes et al., 2014). A random sample of 3750 women among those who were enrolled before 24 weeks gestation were selected for an embedded study focusing on maternal, environmental, and genetic risk factors that interact with prenatal alcohol exposure to affect risk of FASD. The study provided meconium collection kits to hospital staff who collected the meconium within 48 hours of delivery. The polypropylene container with the sample was kept frozen in the hospital until recovered by the study staff. The retrieved samples were stored at −80° C at the local study labs and batch shipped to the study biorepository and continued to be kept at −80° C. The current study sample consists of 720 singleton children born to mothers enrolled in the embedded study at the Northern Plains sites, who were not born preterm, did not have FASD and have both meconium sample and the cognitive development outcome (Supplementary Figure 1).

The Early Autism Risk Longitudinal Investigation (EARLI) is a prospective cohort study of pregnant mothers who had an older child diagnosed with Autism Spectrum Disorder (ASD). On average, children with ASD are at higher risk for cognitive impairment with the latest CDC ASD surveillance data indicating that about 56% of eight-year-old children with ASD had borderline or lower IQ scores(Baio et al., 2018). EARLI enrolled families in Philadelphia, Baltimore, and two regions of Northern California from 2009 to 2012. Additional eligibility criteria included ability to speak English or Spanish, older than 18 years, living within 2 hours of the study site, and less than 29 weeks gestation. Further details of the EARLI study design are reported elsewhere(Newschaffer et al., 2012). In EARLI, meconium was collected by parents of the child at the hospital or at home using a study-provided sampling kit for meconium collection. Parents were instructed to scrape meconium off the diaper using the study provided wooden tongue depressor into the polypropylene collection container. If passed at the hospital, study staff recovered the collected sample prior to discharge. If the child was discharged without passing meconium, parents collected the sample and it was kept frozen at home until recovered later by study staff. Once with study staff, samples were kept at −20°Celcius at the local study labs and batch shipped to the study biorepository where it was homogenized, aliquoted, and frozen at −190° C. In EARLI, women were also provided urine collection kits prior to the second and third trimester (T2, T3) home visits and instructed to collect the entire first morning void of urine. These urine samples were kept refrigerated at home and retrieved by the study staff within 24 hours. Urine samples were aliquoted and frozen at −20° C at the local lab until batch shipped to the study biorepository and kept in long-term storage at −190° C. Although EARLI enrolled 236 women, the analysis samples differ based on the type of biosample and availability of the cognitive development outcome (Supplementary Figure 2).

Outcome Assessment

The Mullen Scales of Early Learning (MSEL) AGS edition was administered at 12-months of age to assess cognitive function (Mullen, 1995) in both studies. MSEL is a widely used cognitive function measure for children up to 68 months of age that has good convergent(Bishop et al., 2011) and concurrent validity(Farmer et al., 2016) with other measures of cognitive status. The MSEL includes both verbal and non-verbal measures of cognition and is therefore able to assess cognitive function among children with and without neurodevelopmental disabilities. Researchers have previously used the MSEL to evaluate cognition among children with ASD(Wiggins et al., 2020), siblings of children with ASD(Charman et al., 2017), Native American children(Mitchell et al., 2011), and children with FASD(Fuglestad et al., 2015). The MSEL has four cognitive scales measuring visual reception, expressive language, receptive language, and fine motor that are aggregated to create the Early Learning Composite (ELC) scale normed to a mean (SD) of 100(15). The ELC has been shown to have good construct, convergent, and divergent validity in both ASD and typically developing children(Swineford et al., 2015). Specifically, in this study the domain scores of the MSEL strongly relate to each other and combine well into a construct measuring cognitive development. Additionally, among children with and without ASD, the MSEL domain scores conformed well with a developmental functioning construct (convergent validity) and not with autism symptoms or emotional/behavioral problems (divergent validity).

Phthalate metabolite measurement

Phthalate metabolites were quantified identically in SPS and EARLI meconium by the Exposure Science Laboratory at the A.J. Drexel Autism Institute based on methods developed by Kato et al.(Kato et al., 2006) and as used in the P4 study(Arbuckle et al., 2016). Briefly, meconium samples were pre-treated with phosphoric acid to denature active enzymes and then treated with β-glucuronidase for enzymatic deconjugation. Analytes were extracted using liquid-liquid extraction and then analyzed by high performance liquid chromatography-electrospray ionization-tandem mass spectrometry (HPLC-ESI-MS/High resolution MS) (Arbuckle et al., 2016). The values were calculated normalized to wet weight of meconium, and limits of detection (LOD) was 0.0005 ng/g. The detailed analytic methods focused specifically on the EARLI cohort are reported elsewhere(Mathew et al., 2021).

Urine phthalate metabolites in EARLI were quantified by the Personal Care Products Laboratory of the CDC. After enzymatic deconjugation to separate glucuronidated conjugates, total urinary concentrations of 14 phthalate metabolites were quantified using online solid phase extraction coupled (HPLC-ESI-MS/MS), based on modification of a published method(Silva et al., 2013). The LODs ranged from 0.2 (mono-isononyl phthalate (MiNP), to 1.2 ng/ml (MEP). Creatinine was also measured by the Personal Care Products Laboratory. Prior to analysis, the quantified metabolite values were divided by creatinine to adjust for urine dilution.

Covariates and Potential Confounders

Continuous demographic variables, such as maternal age at child’s birth and gestational age at birth were summarized using means and standard deviation, and with frequency distributions for categorical variables of race/ethnicity, maternal education, annual household income, parity, pre-pregnancy BMI, study site, birth year, and child’s sex. Tobacco and alcohol use were assessed at each clinic visit in the SPS study. Because meconium is expected to capture cumulative fetal environmental exposure from the second trimester, we created indicator variables for any alcohol, and any tobacco use, during the second or third trimester.

Statistical Analysis

Machine generated values of individual metabolites less than the LOD but higher than zero were left unchanged, but zero values were set to $\raisebox{1ex}{$\mathit{LOD}$}\!\left/ \!\raisebox{-1ex}{$\sqrt{2}$}\right.$ (Hornung & Reed, 1990). Geometric means (GM) of individual metabolite distributions with 95% Confidence Intervals (CI) were compared to GM (95% CI) of metabolite distributions from the P4 study (meconium). Similarly, maternal urine measured metabolite distributions were compared to reported results from the Health Outcomes and Measures of the Environment study (HOME), a Cincinnati based pregnancy cohort. Additionally, we also compared the EARLI urine measured phthalate distributions to the female, 2011-12 NHANES biomonitoring study sample.

We calculated the molar sum of the DEHP metabolites ($\sum \mathrm{DEHP}$) using MEHP, MEHHP, MEOHP and mono(2-ethyl-5-carboxypentyl) phthalate (MECPP). We also considered a previously used (Messerlian et al., 2017; Varshavsky et al., 2016) anti-androgenic (AA) phthalate summary score calculated as the anti-androgenic potency weighted sum of DEHP metabolites plus MBP, MiBP, and MBzP. Since MiBP was not detected in SPS meconium, the metabolite was omitted for the AA score calculated in SPS. The AA score was calculated similarly in both meconium and urine samples from EARLI as all necessary metabolites were available in both biosamples. Prior to use in statistical analysis, all individual metabolites and summary scores were log_e-transformed to reduce the influence of extremely high data points and skewness of distribution.

The associations between log-transformed metabolites and ELC were assessed separately in SPS and EARLI using phthalate-specific linear regression models adjusted for a priori selected covariates. Based on current literature documenting associations between phthalates and neurodevelopment, we considered maternal age, maternal education, maternal race, household income, parity, pre-pregnancy BMI, alcohol use, and tobacco use as confounders. However, only eight women reported cigarette use during pregnancy in the EARLI study, and over 20% of EARLI women had missing data on alcohol use. A sensitivity analysis was conducted by evaluating the change in the metabolite-ELC associations when a missing category augmented alcohol use variable was added to the full model. Additionally, considering the limited available sample size for EARLI and consequent constraints on the number of variables in regression models, we excluded tobacco use and alcohol use from EARLI study models. All models using phthalate metabolites measured in meconium included gestational age at birth as a covariate since gestational age could affect both phthalate levels and child neurodevelopment. Finally, we adjusted for study site and birth year of the study child in all models due to the potential for differences in exposure and outcome distribution by site location or by changes in formulation of phthalate containing products over the study years.

Since several studies on phthalates and neurodevelopment suggest effect modification of the phthalate-ELC association by child’s sex, we adopted a modelling approach to explore potential effect modification. Prior to fitting regression models, we created Locally Weighted Scatterplot Smoothing (LOWESS) curves for the phthalate-ELC association, stratified by child’s sex. We used the shape of the plots to determine the need for quadratic terms of metabolites in the stratified models. These curves also allowed us to assess observations with potential to disproportionately affect the shape and magnitude of associations. When nonlinearity was detected, we tested for the presence of quadratic interaction of the phthalate metabolite by child’s sex in the full sample by jointly testing the coefficients of the linear and quadratic product terms. If nonlinearity was absent, but stratified plots suggested effect estimates in opposite directions, a linear interaction term was tested in the full sample model. If a potential outlier with disproportionate effect on the curve was detected, the model was rerun without the observation to evaluate the impact on the model. If the exclusion of two datapoints or less resulted in a loss of statistically significant interaction effect in the model, we excluded those datapoints in the analysis for those particular metabolites and chose the non-statistically significant, but stable results. This approach led to the exclusion of one observation from the analysis of MBP and mono-3-hydroxy-n-butyl phthalate (MHBP), and two observations from MiBP and mono-hydroxy-isobutyl phthalate (MHiBP) measured in maternal urine. Given study sample size, interaction tests were anticipated to have lower power, so we used a p-value of 0.2 to flag potential sex by phthalate interaction. Consequently, we recognize the increased potential for false positive findings of effect modification, and results need to be interpreted appropriately. When interaction p-values were below this threshold, sex-specific regression models with linear or quadratic term for the metabolite as appropriate, were computed, and effect estimates calculated. When effect modification was suggested in the full model, we generated fitted model plots with sex specific curves, with continuous variables set to their mean and categorical variables set to reference values used in the regression. We used p<0.05 as a threshold for gauging statistical significance of main effects. Although we log_e-transformed the metabolite distributions, there were still some phthalate metabolites with extreme values. Therefore, we winsorized all metabolite distributions at the 2^nd and 98^th percentile values and recomputed all regression models to evaluate the effect of extreme values on estimates. Additionally, since MEHP measured in meconium had high proportion of observations below the LOD and since these datapoints seemed to have a substantial effect on the regression model, we recomputed the MEHP models without the observations below the LOD. All analysis was conducted using SAS V 9.4.

Results

As reported in Table 1, the EARLI study mothers were on average older than the SPS study mothers and had slightly higher proportion of boys. Most EARLI women identified as Non-Hispanic White, Hispanic, and Non-Hispanic Asian, whereas majority of SPS women identified as Non-Hispanic White and Native American. A lower proportion of EARLI women reported as high school educated compared to SPS and a higher proportion reported having a bachelor’s degree or higher Since EARLI recruited women who had an older child with autism, 44.6% had at least one previous child and all others had 2 or more children. SPS had no such eligibility criteria and therefore 36.8% of women were nulliparous and a similar proportion of women had 1, and 2+ children. The gestational age distribution in EARLI was more heterogenous than SPS since we had only term births in our SPS sample. The EARLI sample consisted of 236 children, and the SPS sample contained 720 children, but the effective sample size for various analyses differed based on the number of women who completed study specific visits and the number of children with completed cognitive assessment. (SPS meconium N = 606, EARLI meconium N = 160, EARLI T2 urine N = 146, and EARLI T3 urine N = 113). Detailed comparison of characteristics between analysis samples is provided in the Supplementary Tables 1–3.

Table 1:

Characteristics of EARLI & SPS cohorts in comparison to the general US population

EARLI, N = 236		SPS, N = 720
Variable	N (%)	Variable	N (%)
Site		Site
Drexel	61 (25.8)	Rapid City	180 (25)
Johns Hopkins	53 (22.5)	Sanford health	540 (75)
Kaiser Permanente	72 (30.5)
UC Davis	50 (21.2)
Birth Year		Birth Year
2009	6 (2.5)	2008	94 (13.1)
2010	53 (22.5)	2009	118 (16.4)
2011	89 (37.7)	2010	99 (13.8)
2012	84 (35.6)	2011	97 (13.5)
2013	2 (0.9)	2012	111 (15.4)
		2013	82 (11.4)
		2014	75 (10.4)
		2015	44 (6.1)
Child Sex- Male	121 (51.1)	Child Sex- Male	353 (49.0)
Maternal Age (mean, SD)	33.8 (4.6)	Maternal Age (mean, SD)	28.1 (4.9)
Race/ethnicity		Race/ethnicity
NH White	128 (54.2)	NH White	576 (80.0)
NH Black	27 (11.4)	NH American Indian	107 (14.9)
NH Asian	33 (13.9)
NH Other	8 (3.4)	NH Other	9 (1.3)
Hispanic	38 (16.1)	Hispanic	28 (3.9)
Missing	2 (0.9)	Missing	0(0)
Pre-Pregnancy BMI		Pre-Pregnancy BMI
Underweight	3 (1.3)	Underweight	13 (1.9)
Normal	97 (40.9)	Normal	331 (47.0)
Overweight	58 (24.5)	Overweight	188 (26.7)
Obese	70 (29.5)	Obese	172 (24.4)
Missing	8 (2.4)	Missing	16 (2.2)
Maternal Education		Maternal Education
High School or less	29 (12.3)	High School or less	160 (22.2)
Some college/ Tech School	67 (28.4)	Some college	196 (27.2)
Bachelors or more	137 (58.1)	Bachelors or more	364 (50.6)
Missing	3 (1.3)	Missing	0(0)
Parity:		Parity:
Nulliparous	0 (0)	Nulliparous	265 (36.8)
1	107 (44.6)	1	234 (32.5)
2 +	130 (55.1)	2 +	221 (30.7)
Gestational Age (mean, SD)	39.0 (3.9)	Gestational Age (mean, SD)	39.5 (1.1)
Annual Household Income		Monthly Household Income
(US$) <49,000	61 (25.9)	(US $) <= 2000	161 (22.8)
50,000- < 75,000	34 (14.4)	2001 – 3000	137 (19.4)
75,000- <100,000	37 (15.7)	3001 – 4000	148 (20.9)
100,000- <200,000	78 (33.1)	4001 – 5000	123 (17.5)
> 200,000	17 (7.2)	> 5000	136 (19.3)
Missing	9 (3.8)	Missing	15 (2.1)
		Trimester 2/3 Smoking	92 (12.8)
		Missing	10 (1.4)
		Trimester 2/3 Alcohol	75 (10.4)
		Missing	1 (0.1)
MSEL- ELC	101 (15)	MSEL- ELC	98.5 (12.8)

Note: EARLI, Early Autism Risk Longitudinal Investigation; SPS, Safe Passage Study; NHANES, National Health and Nutrition Examination Survey, survey weight adjusted proportions; UC, University of California; SD, Standard Deviation; NH, Non-Hispanic; MSEL-ELC, Mullen Scales of Early Learning – Early Learning Composite

We detected 8 and 9 of 14 assessed metabolites in over 90% of SPS and EARLI meconium samples, respectively (Supplementary Table 4). Thirteen metabolites were quantified in EARLI meconium and twelve in SPS meconium, with MiBP the one metabolite reliably detected in EARLI but not SPS meconium. As reported in Table 2, meconium measured metabolites (MBP, MBzP, MECPP, MEP, and mono-isopentyl phthalate (MiPP)) were higher in EARLI than SPS, but levels of mono-cyclohexyl phthalate (MCHP), MiNP, mono-n-octyl phthalate (MOP), mono-n-pentyl phthalate (MPP) were lower in EARLI. Over 50% of samples were below detectable level for MiPP in SPS, and so the metabolite was excluded from all statistical analysis in SPS. Both EARLI and SPS measured metabolite levels in meconium were higher than in the only other large epidemiological study to quantify phthalates in meconium, the P4 study(Arbuckle et al., 2016). The only exception was in the levels for MBP where P4 levels were slightly higher than SPS (Supplementary Figure 3). The secondary metabolites available in common between all three studies had similar variance and were closer in magnitude to the P4 levels. All quantified metabolites except MEHP, MHBP and MiNP were detected in over 90% of urine samples. Compared to the 2011-2012 NHANES data (from urine samples), the geometric means and 95% Confidence Intervals (CI) of common metabolites overlapped substantially (Supplementary Figures 4, 5). Several 2^nd trimester metabolite GM’s (MiBP, MEHP, MEOHP, MECPP, MBP, mono-carboxynonyl phthalate (MCNP) were higher in EARLI than NHANES, but some of these (MEHP, mono-3-carboxypropyl (MCPP), MCNP) were lower in the 3^rd trimester in EARLI than in NHANES. The HOME study estimates for all metabolites except MiBP were higher than EARLI or NHANES (Supplementary Figures 4, 5).

Table 2:

Median (25%, 75%) of phthalate metabolites measured in urine (ng/mg creatinine) and meconium (ng/g)

			EARLI		SPS
Parent Phthalate	Phthalate   Metabolite	Urine Tri 2, N = 183	Urine Tri 3, N = 140	Meconium N = 190	Meconium N = 718
DBP	mono-n-butyl (MBP)	15.3 (9.6, 24)	15.4(10.1, 24.0)	21.1 (12.9, 31.5)	1.9 (0.9, 4.4)
DBP	mono-3-hydroxy-n-butyl (MHBP)	1.2 (0.8, 2.1)	1.2 (0.8, 2.3)	NA	NA
DiBP	mono-isobutyl (MiBP)	9.9 (6.7, 14.9)	10.4 (6.6, 15.9)	39.3 (23.3, 78.4)	ND
DiBP	mono-hydroxy-isobutyl (MHiBP)	3.6 (2.3, 5.3)	3.4 (2.3, 5.4)	NA	NA
DEP	mono ethyl (MEP)	33.5 (18.5, 70.5)	34.1 (17.4, 68.9)	33.8 (2.7, 233.9)	5.2 (1.9, 9.5)
BBzP	mono benzyl (MBzP)	6.7 (4.2, 14.4)	6.9 (3.4, 15.8)	10.9 (7.5, 17.7)	2.9 (2.2, 4.1)
DiDP	mono-carboxynonyl (MCNP)	3.8 (2.3, 7.8)	3.5 (2.1, 6.7)	NA	NA
DiNP	mono-isononyl (MiNP)	1.3 (0.7, 3.0)	1.1 (0.6, 3.0)	2.0 (0.3, 7.9)	6.5 (2.0, 17.8)
DiNP	mono-carboxy-isooctyl (MCOP)	26.8 (11.3, 71.1)	18.2 (9.9, 56.1)	NA	NA
DOP/DBP	mono-3-carboxypropyl (MCPP)	2.5 (1.4, 5.2)	2.4 (1.2, 6.0)	ND	ND
DEHP	mono(2-ethylhexyl) (MEHP)	2.9 (1.4, 4.8)	2.3 (1.2, 4.8)	10.9 (5.7, 21.0)	13.6 (4.3, 33.6)
DEHP	mono(2-ethyl-carboxypentyl) (MECPP)	18.8 (11.9, 31.4)	16.5 (10.2, 30.8)	16.5 (10.9, 23.5)	6.7 (4.2, 10.1)
DEHP	mono(2-ethyl-5-hydroxyhexyl) (MEHHP)	11.8 (7.2, 19.1)	11.1 (6.4, 17.9)	2.5 (1.5, 3.8)	4.1 (2.3, 6.5)
DEHP	mono(2-ethyl-5-oxohexyl) (MEOHP)	8.9 (6.2, 14.2)	9.0 (5.5, 14.7)	1.3 (0.8, 2.3)	2.4 (1.7, 3.7)
DCHP	mono-cyclohexyl (MCHP)	NA	NA	0.5 (0.004, 2.6)	1.1 (0.8, 1.5)
DOP	mono-n-octyl (MOP)	NA	NA	< LD (< LD, 2.6)	8.3 (0.8, 28.5)
DPP	mono-n-pentyl (MPP)	NA	NA	1.2 (0.4, 3.4)	14.2 (4.2, 31.8)
DiPP	mono-isopentyl (MiPP)	NA	NA	128.7 (55.8, 322.5)	< LD (< LD, 3.1)

Note: EARLI, Early Autism Risk Longitudinal Investigation; SPS, Safe Passage Study; Tri, Trimester; DBP, di-n-butyl phthalate; DiBP, di-isobutyl phthalate; DEP, di-ethyl phthalate; BBzP, Butyl-benzyl phthalate; DiDP, di-isodecyl phthalate; DiNP, di-isononyl phthalate; DOP, di-n-octyl phthalate; DEHP, di(2-ethylhexyl) phthalate; DCHP, di-cyclohexyl phthalate; DPP, di-n-pentyl phthalate; DiPP, di-isopentyl phthalate; Limit of Detection (LD) = 0.0005 ng/g in meconium. Not Assessed (NA), Not Detected (ND)

Association of meconium phthalate metabolites in SPS with ELC

Table 3 summarizes results from adjusted regression models for meconium measured phthalate metabolites in SPS and EARLI cohorts. In SPS data models without the metabolite-child sex interaction term, statistically significant associations in a quadratic U-shaped form were observed between phthalate metabolite and ELC for MBzP (β_Linear= −6.73, 95% CI: −12.04, −1.42; β_quadratic = 1.95, 95% CI: 0.17, 3.62) and MECPP (β_Linear= −3.81, 95% CI: −7.53, −0.27; β_quadratic = 0.93, 95% CI: 0.09, 1.77). Both other secondary metabolites of DEHP were observed to have a similar functional relationship with the ELC where a unit increase in the metabolite resulted in a sharp decrease in the ELC, but an eventual increase at higher levels of the metabolite.

Table 3:

Associations between meconium assessed phthalate metabolites and children’s Mullen scales of early learning-composite score at 12-months of age in SPS (N = 606) and EARLI (N = 160) cohorts and estimates from sex-stratified models when effect modification (EM p < 0.2) was detected

				Girls	Boys
Phthalate	β (95% CI)	P value	EM P-value	β (95 % CI)	β (95 % CI)
SPS
MBP	0.10 (−0.31, 0.53)	0.610
MEP	0.19 (−0.11, 0.50)	0.214
MBZP	−6.73 (−12.04, −1.42)	0.013	0.048	−11.27 (−18.54, −4.01)	1.20 (−1.73, 4.14)
MBzP2	1.95 (0.27, 3.62)	0.023		2.99 (0.80, 5.17)
MCHP	−1.15 (−3.25, 0.93)	0.277
MEHP	0.20 (−0.14, 0.55)	0.246		−0.40 (−0.93, 0.13)	0.63 (0.23, 1.05)
MEHP2	−0.03 (−0.10, 0.04)	0.351	0.050	−0.10 (−0.21, 0.00)
MECPP	−3.81 (−7.53, −0.27)	0.035
MECPP2	0.93 (0.09, 1.77)	0.030
MEHHP	−3.51 (−7.18 ,0.15)	0.060
MEHHP2	1.16 (−0.02, 2.36)	0.055
MEOHP	−1.57 (−6.05, 2.90)	0.489
MEOHP2	1.02 (−1.11, 3.16)	0.345
MINP	0.64 (−0.11,1.41)	0.095	0.082	0.24 (−0.94, 1.42)	1.15 (0.13, 2.18)
MOP	0.28 (−0.08, 0.64)	0.129
MPP	0.05 (−0.36, 0.47)	0.802
$\sum \mathrm{DEHP}$	0.26 (−0.71, 1.22)	0.599	0.039	−1.32 (−2.84, 0.20)	1.31 (0.06, 2.57)
AASCORE	0.07 (−0.96, 1.10)	0.889	0.036	−1.57 (−3.19, 0.04)	1.27 (−0.09, 2.64)
EARLI
MBP	−0.26 (−1.41, 0.88)	0.650
MiBP	0.71 (−0.91, 2.34)	0.385	0.013	−3.03 (−6.05, −0.01)	2.21 (−0.03, 4.46)
MEP	−0.10 (−0.54, 0.34)	0.650
MBzP	1.05 (−1.57, 3.69)	0.427	0.018	−3.19 (−7.63, 1.24)	2.95 (−0.78, 6.68)
MCHP	0.10 (−0.68 ,0.89)	0.800
MEHP	0.47 (−0.78, 1.71)	0.461		−0.49 (−1.45, 0.46)	1.92 (0.06, 3.77)
MEHP2	0.14 (−0.09, 0.37)	0.231	0.106		0.41 (0.06, 0.75)
MECPP	1.17 (−1.78, 4.12)	0.433
MEHHP	0.35 (−0.85, 1.56)	0.560
MEOHP	0.84 (−1.37, 3.07)	0.451
MiNP	0.27 (−0.32, 0.87)	0.357
MPP	−0.44 (−1.19, 0.31)	0.248
MiPP	0.83 (−0.82, 2.48)	0.320	0.113	−1.25 (−3.86, 1.36)	1.28 (0.18, 5.28)
$\sum \mathrm{DEHP}$	2.34 (−0.81, 5.49)	0.144	0.150	−1.90 (−7.40, 3.60)	4.66 (0.31, 9.00)
AASCORE	6.24 (−0.14, 12.63)	0.055		−1.13 (−11.66, 9.39)	9.89 (0.30, 19.47)
AASCORE2	3.03 (0.17, 5.89)	0.038	0.036	2.14 (−2.72, 6.99)	2.91 (−1.16, 6.98)

Note: EARLI, Early Autism Risk Longitudinal Investigation; SPS, Safe Passage Study; EARLI models adjusted for child sex, study site, birth year, maternal age, maternal race/ethnicity, household income, maternal education, parity, gestational age at birth, and pre-pregnancy BMI. SPS models additionally adjusted for cigarette smoking and alcohol use. EM- Interaction of phthalate metabolite and child sex in adjusted model; MBP, mono-n-butyl phthalate; MiBP, mono-isobutyl phthalate; MEP, mono-ethyl phthalate; MBzP, mono-benzyl phthalate; MCHP, mono-cyclohexyl phthalate; MiNP, mono-isononyl phthalate; MEHP, mono(2-ethylhexyl) phthalate; MECPP, mono(2-ethyl-5-carboxypentyl) phthalate; MEHHP, mono(2-ethyl-5-hydroxyhexyl) phthalate; MEOHP, mono(2-ethyl-5-oxohexyl) phthalate; MOP, mono-n-octyl phthalate; MPP, mono-n-pentyl phthalate; $\sum \mathrm{DEHP}$, molar sum of di(2-ethylhexyl) phthalate metabolites; AASCORE, anti-androgenic score.

Statistically significant interactions of the phthalate metabolite and child sex on ELC were observed for MBzP, MEHP, $\sum \mathrm{DEHP}$, and the AA score, and suggested for MiNP. MiPP and MiBP were not reliably detected in SPS and so were excluded from this analysis. In sex-stratified adjusted models for MBzP and MEHP, associations were nonlinear among girls, with MBzP (β_Linear= −11.27, 95% CI: −18.54, −4.01; β_quadratic = 2.99, 95% CI: 0.80, 5.17) strongly associated with ELC similar to the model without interaction. Both metabolites were associated with an increase in ELC among boys. The interaction remained statistically significant in the model for MEHP without the observations below the LOD (sensitivity analysis), but the effect of MEHP on ELC among boys was more variable (β = 0.92, 95% CI: −0.05, 1.89), Supplementary Figure 6. In sex-stratified adjusted models, increasing levels of $\sum \mathrm{DEHP}$ and the AA score were associated with a linear decrease in ELC among girls, but a linear increase in ELC among boys. The point estimate of MiNP indicated a linear increase in ELC among both sexes, but the 95% CI was wide and the lower limit negative among girls. Fitted model plots from adjusted models with interaction terms are shown in figures 1 and 2, and a summary of all results reported in Table 5.

Figure 1:

Fitted model plot of ELC and 95% confidence bands from interaction model for log_e-transformed meconium phthalate metabolites (in ng/g) in SPS. P-values from test for linear or quadratic interactions in adjusted regression model. Note: SPS, Prenatal Alcohol in SIDS and Stillbirth; ELC, Mullen Scales of Early Learning-early learning composite; M-Male, F-Female; MBzP, mono-benzyl phthalate; MEHP, mono(2-ethylhexyl) phthalate; $\sum \mathrm{DEHP}$, molar sum of di(2-ethylhexyl) phthalate metabolites.

Figure 2:

Fitted model plot of ELC and 95% confidence bands from interaction model for log_e-transformed mono-isononyl phthalate (MiNP) (ng/g) in meconium from SPS. P-values from test for linear interaction in adjusted regression model. Note: SPS, Safe Passage Study; ELC, Mullen Scales of Early Learning-early learning composite; M-Male, F-Female

Table 5:

Summary of associations between maternal urine and child meconium measured phthalate metabolites in the EARLI and SPS cohorts with children’s Mullen scales of early learning-composite score at 12-months of age

	EARLI			SPS
Metabolite	Urine, Trimester 2	Urine, Trimester 3	Meconium	Meconium
MBP	Linear ↓, Interaction	Linear ↑	Linear ↓	Linear ↑
MHBP	Linear ↑, Interaction	Linear ↑	NA	NA
MiBP	Linear ↓	Linear ↑	Linear ↑, Interaction	NA
MHiBP	Linear ↑	Linear ↑	NA	NA
MEP	Quadratic ∪	Quadratic ∪, Interaction	Linear ↓	Linear ↑
MBzP	Linear ↑, Interaction	Quadratic ∪*	Linear ↑*, Interaction	Quadratic ∪*, Interaction
MCNP	Linear ↓*	Linear ↑	NA	NA
MCHP	NA	NA	Linear ↑	Linear ↓
MEHP	Linear ↓	Quadratic ∪	Quadratic ∪, Interaction	Quadratic ∩, Interaction
MECPP	Linear ↓	Linear ↑	Linear ↑	Quadratic ∪*
MEHHP	Linear ↓	Linear ↑	Linear ↑	Quadratic ∪*
MEOHP	Quadratic ∩	Linear ↑	Linear ↑	Quadratic ∪
MiNP	Linear ↑	Linear ↑	Linear ↑	Linear ↑, Interaction
MCOP	Linear ↓	Linear ↑	NA	NA
MCPP	Linear ↓	Linear ↑	NA	NA
MOP	NA	NA	NA	Linear ↑
MPP	NA	NA	Linear ↓	Linear ↑
MiPP	NA	NA	Linear ↑, Interaction	NA
$\sum \mathrm{DEHP}$	Quadratic ∩, Interaction	Linear ↑	Linear ↑, Interaction	Linear ↑, Interaction
AASCORE	Quadratic ∩, Interaction	Linear ↓	Quadratic ∪*, Interaction	Linear ↑, Interaction

^*Statistically significant at P < 0.05; Interaction, statistically significant at P <0.2; NA, Not Assessed. EARLI, Early Autism Risk Longitudinal Investigation; SPS, Safe Passage Study; MBP, mono-n-butyl phthalate; MHBP, mono-3-hydroxy-n-butyl phthalate; MiBP, mono-isobutyl phthalate; MHiBP, mono-hydroxy-isobutyl phthalate; MEP, mono-ethyl phthalate; MBzP, mono-benzyl phthalate; MCNP, mono-carboxynonyl phthalate; MCHP, mono-cyclohexyl phthalate; MEHP, mono(2-ethylhexyl) phthalate; MECPP, mono(2-ethyl-5-carboxypentyl) phthalate; MEHHP, mono(2-ethyl-5-hydroxyhexyl) phthalate; MEOHP, mono(2-ethyl-5-oxohexyl) phthalate; MiNP, mono-isononyl phthalate; MCOP, mono-carboxyoctyl phthalate; MCPP, mono-3-carboxypropyl phthalate; MOP, mono-n-octyl phthalate; MPP, mono-n-pentyl phthalate; MiPP, mono-isopentyl phthalate; $\sum \mathrm{DEHP}$, molar sum of di(2-ethylhexyl) phthalate metabolites; AASCORE, anti-androgenic score.

Association of meconium phthalate metabolites in EARLi with ELC

In EARLI, in adjusted models without any metabolite-child’s sex interaction terms, most phthalates except for $\sum \mathrm{DEHP}$, and AA score were not strongly associated with ELC. Similar to the SPS analysis, we included a quadratic term to better model the relationships observed in meconium data, and the AA score was positively associated with ELC. We observed statistically significant effect modification of this association by child sex. In sex-stratified adjusted models, the estimates decreased initially but increased at higher metabolite levels for girls (β_Linear= −1.13, 95% CI: −11.66, 9.39; β_quadratic = 2.14, 95% CI: −2.72, 6.99). The estimates of both linear and quadratic terms were positive among boys (β_Linear = 9.89, 95% CI: 0.30, 19.47; β_quadratic = 2.91, 95% CI: −1.16, 6.98), but the wide confidence intervals suggest that this result should be interpreted cautiously. The interaction term was statistically significant at for MBzP and MiBP, and in both cases the metabolites were associated with a decrease in ELC among girls. Effect modification of the metabolite-ELC association by child sex was suggested for MEHP, MiPP, and $\sum \mathrm{DEHP}$, and in adjusted sex-stratified models, higher levels of these metabolites were associated with a linear decrease in ELC among girls, but the 95% CI was wide and the upper limit was positive. Among boys, $\sum \mathrm{DEHP}$ and MiPP were associated with a linear increase in ELC, and MEHP had a U-shaped association with ELC. In the sensitivity analysis model run without the MEHP observations below the LOD, the interaction term remained statistically significant and although the associations were in the same direction, the sex- stratified model estimates and 95% CI overlapped substantially (Supplementary figure 7). Figure 3 reports the fitted model plots from interaction term models, and Tables 3 and 5 contrasts the results with those from the SPS cohort.

Figure 3:

Fitted model plot of ELC and 95% confidence bands from interaction model for log_e-transformed meconium phthalate metabolites (in ng/g) in EARLI. P-values from test for linear or quadratic interactions in adjusted regression model. Note: ELC, Mullen Scales of Early Learning-early learning composite; M-Male, F-Female; MBzP, mono-benzyl phthalate; MEHP, mono(2-ethylhexyl) phthalate; MiBP, mono-isobutyl phthalate, MiPP, mono-isopentyl phthalate, and $\sum \mathrm{DEHP}$, molar sum of di(2-ethylhexyl) phthalate metabolites; EARLI, Early Autism Longitudinal Risk Investigation.

Association of urine phthalate metabolites with ELC in EARLI

Table 4 summarizes results from the linear regression models for 2^nd, 3^rd, and trimester averaged urine samples. In adjusted models for 2^nd trimester urine measured metabolites, MCNP was statistically significantly associated with a linear decrease in ELC (β = −3.13 (−6.02, −0.24). As with the meconium measured metabolites, effect modification of the phthalate metabolite-ELC association by child sex, was suggested for MBzP, $\sum \mathrm{DEHP}$, and the AA score, and additionally for MBP and MHBP (Figures 4, 5). The effect modification was captured using linear terms for MBP, MHBP, and MBzP, but due to substantial nonlinearity in the metabolite-ELC relationship among boys, a quadratic term best captured the association for $\sum \mathrm{DEHP}$ and the AA score. In sex-stratified adjusted models, metabolites were linearly associated with decrease in ELC (lower cognitive function) among girls and the association was statistically significant for MBP, MBzP, and MHBP. In similar models among boys, MBzP was statistically significantly associated with linear increase in ELC (higher cognitive function, analogous to results in meconium). Nonlinear associations observed between $\sum \mathrm{DEHP}$, and the AA score with ELC among boys had high point estimates but the confidence intervals were imprecise.

Table 4:

Associations between 2^nd, 3^rd, and trimester averaged maternal urine assessed phthalate metabolites and children’s Mullen scales of early learning-composite score at 12-months of age (EARLI) and estimates from sex-stratified models when effect modification (EM p < 0.2) was detected

Second trimester, N = 146
				Girls	Boys
Phthalate	β (95% CI)	P value	EM P-value	β (95 % CI)	β (95 % CI)
MBP	−0.34 (−4.07, 3.39)	0.857	0.058	−5.30 (−10.32, −0.28)	3.73 (−2.30, 9.76)
MHBP	0.12 (−3.14, 3.39)	0.941	0.034	−5.18 (−9.61, −0.74)	4.45 (−0.91, 9.8)
MiBP	−0.34 (−4.58, 3.90)	0.873
MHiBP	0.19 (−4.62, 5.02)	0.935
MEP	−6.44(−21.13,8.24)	0.387
MEP2	0.91 (−0.98, 2.80)	0.342
MBzP	1.47 (−1.34, 4.29)	0.303	0.067	−4.79 (−9.41, −0.18)	5.09 (1.01, 9.18)
MCNP	−3.13 (−6.02, −0.24)	0.034
MiNP	0.29 (−1.66, 2.25)	0.764
MCOP	−0.76 (−2.86,1.34)	0.473
MCPP	−0.39 (−2.64,1.84)	0.726
MEHP	−0.73 (−3.06,1.59)	0.532
MECPP	−0.01 (−2.86, 2.84)	0.994
MEHHP	−0.53 (−3.30, 2.24)	0.705
MEOHP	3.41 (−4.79,11.61)	0.412
MEOHP2	−0.61 (−1.88, 0.65)	0.341
$\sum \mathrm{DEHP}$	−2.10 (−6.73, 2.52)	0.369	0.139	−2.94 (−9.09, 1.20)	−3.21 (−8.78, 2.36)
$\sum \mathrm{DEHP}$ 2	−0.58 (−1.89, 0.71)	0.371			−1.73 (−3.52, 0.04)
AASCORE	−2.75 (−8.49, 2.97)	0.342	0.063	−4.21 (−8.45, 0.02)	−3.75 (−10.77, 3.26)
AASCORE2	−0.69 (−2.33, 0.95)	0.407			−2.23 (−4.50, 0.03)
Third trimester, N = 113
MBP	2.31 (−1.36, 5.97)	0.214
MHBP	0.48 (−3.27, 4.24)	0.798
MiBP	2.09 (−2.13, 6.32)	0.327
MHiBP	2.94 (−1.64, 7.51)	0.205
MEP	−5.33 (−17.53, 6.85)	0.387		−38.30 (−64.11, −12.49)	1.60 (−2.48, 5.67)
MEP2	0.55 (−0.98, 2.08)	0.475	0.093	4.50 (1.39, 7.60)
MBzP	11.78 (0.90, 22.66)	0.034
MBzP2	−2.91 (−5.86 ,0.03)	0.052
MCNP	2.68 (−0.31, 5.69)	0.079
MiNP	1.93 (−0.51, 4.37)	0.119
MCOP	2.21 (−0.62, 5.05)	0.124
MCPP	0.21 (−2.03, 2.46)	0.851
MEHP	−1.71 (−5.68, 2.25)	0.392
MEHP2	1.32 (−0.31, 2.96)	0.110
MECPP	2.04 (−1.54, 5.63)	0.260
MEHHP	1.58 (−1.57, 4.74)	0.320
MEOHP	1.79 (−1.63, 5.22)	0.300
$\sum \mathrm{DEHP}$	1.21 (−1.88, 4.31)	0.437
AASCORE	−5.21 (−13.29, 2.86)	0.202
AASCORE2	−1.79 (−3.88, 0.28)	0.089
2nd and 3rd trimester averaged, N = 183
Phthalate	β (95% CI)	P value	EM P-value	β (95 % CI)	β (95 % CI)
MBP	0.26 (−3.09, 3.61)	0.877
MHBP	0.44 (−3.88, 4.76)	0.840	0.125	−1.89 (−6.21, 2.43)	7.60 (−0.79, 16.00)
MHBP2	−0.68 (−3.49, 2.14)	0.635			−4.58 (−10.41, 1.25)
MiBP	0.90 (−2.68, 4.48)	0.621	0.199	−2.96 (−9.17, 3.26)	3.87 (−1.19, 8.94)
MHiBP	1.88 (−2.08, 5.84)	0.349
MEP	−0.07 (−2.35, 2.21)	0.951
MBzP	1.19 (−1.47, 3.85)	0.377
MCNP	−0.27 (−2.78, 2.22)	0.828
MiNP	1.14 (−0.67, 2.96)	0.215
MCOP	0.39 (−1.58, 2.37)	0.695
MCPP	0.01 (−1.81, 1.84)	0.987
MEHP	−0.51 (−2.65, 1.61)	0.632
MECPP	0.45 (−2.14, 3.06)	0.729
MEHHP	−0.41 (−2.87, 2.04)	0.739
MEOHP	−0.29 (−2.90, 2.32)	0.827
$\sum \mathrm{DEHP}$	−0.42 (−2.75, 1.89)	0.718
AASCORE	−0.77 (−3.37, 1.82)	0.554

Note: All models adjusted for child sex, study site, birth year, maternal age, maternal race/ethnicity, household income, maternal education, parity, and pre-pregnancy BMI. EARLI, Early Autism Risk Longitudinal Investigation; EM- Interaction of phthalate metabolite and child sex in adjusted model; MBP, mono-n-butyl phthalate; MHBP, mono-3-hydroxy-n-butyl phthalate; MiBP, mono-isobutyl phthalate; MHiBP, mono-hydroxy-isobutyl phthalate; MEP, mono-ethyl phthalate; MBzP, mono-benzyl phthalate; MCNP, mono-carboxynonyl phthalate; MiNP, mono-isononyl phthalate; MCOP, mono-carboxyoctyl phthalate; MCPP, mono-3-carboxypropyl phthalate; MEHP, mono(2-ethylhexyl) phthalate; MECPP, mono(2-ethyl-5-carboxypentyl) phthalate; MEHHP, mono(2-ethyl-5-hydroxyhexyl) phthalate; MEOHP, mono(2-ethyl-5-oxohexyl) phthalate; $\sum \mathrm{DEHP}$, molar sum of di(2-ethylhexyl) phthalate metabolites; AASCORE, anti-androgenic score.

Figure 4:

Fitted model plots of ELC and 95% confidence bands from interaction models for log_e-transformed phthalate metabolites (in ng/mg creatinine) MBP, MHBP, $\sum \mathrm{DEHP}$, and anti-androgenic score in 2^nd trimester maternal urine in EARLI. P-values from test for linear or quadratic interactions in adjusted regression model. Note: ELC, Mullen Scales of Early Learning-early learning composite; M-Male, F-Female; MBP, mono-n-butyl phthalate; MHBP, mono-3-hydroxy-n-butyl phthalate; S-DEHP, molar sum of di(2ethylhexyl) phthalate metabolites; EARLI, Early Autism Longitudinal Risk Investigation.

Figure 5:

Fitted model plot of ELC and 95% confidence bands from interaction model for log_e-transformed monobenzyl phthalate (MBzP) in ng/mg creatinine, in 2^nd trimester maternal urine in EARLI. P-values from test for linear interaction in adjusted regression model. Note: ELC, Mullen Scales of Early Learning-early learning composite; M-Male, F-Female; EARLI, Early Autism Longitudinal Risk Investigation.

In 3^rd trimester adjusted models without interactions, MBzP (β_Linear = 11.78, 95% CI: 0.90, 22.66; β_quadratic = 2.91, 95% CI: −5.86, 0.03) was associated with ELC in an inverted U-shaped manner, with an initial increase in ELC for a unit increase in the metabolite, but a decrease at higher levels. There was also a suggestion of effect modification of the metabolite-ELC association by child sex for MEP (Figure 6). A quadratic term best captured the exposure outcome relationship among girls. The U-shaped association between MEP and ELC was statistically significant for girls but the magnitude and shape of association should be interpreted with caution due to the wide confidence intervals. Table 5 reports a qualitative summary of the meconium and urine results from both SPS and EARLI cohorts.

Figure 6:

Fitted model plot of ELC and 95% confidence bands from interaction model for log_e-transformed mono-ethyl phthalate (MEP) in ng/mg creatinine in 3^rd trimester maternal urine in EARLI. P-values from test for linear interaction in adjusted regression model. Note: ELC, Mullen Scales of Early Learning-early learning composite; M-Male, F-Female; EARLI, Early Autism Longitudinal Risk Investigation.

We also averaged levels of phthalate metabolites from 2^nd and 3^rd trimester and evaluated associations between this measure and ELC. None of the trimester averaged phthalate metabolites were strongly associated with ELC in adjusted models, but similar to T2 results, we observed a suggestion of effect modification for MHBP and MiBP. Interaction effects noted for MHBP in T2 urine retained directionality of association by sex in trimester averaged data, but nonlinearity required a quadratic term in the model. In sex-stratified adjusted models the association among boys had an inverted U-shape (Figure 7) where ELC increased for lower values of MHBP but decreased at higher levels of the metabolite.

Figure 7:

Fitted model plot of ELC and 95% confidence bands from interaction model for log_e-transformed 2^nd and 3^rd trimester averaged phthalate metabolites (in ng/mg creatinine) maternal urine in EARLI. P-values from test for linear or quadratic interactions in adjusted regression model. Note: ELC, Mullen Scales of Early Learning-early learning composite; M-Male, F-Female; MHBP, mono-3-hydroxy-n-butyl phthalate; MiBP, mono-isobutyl phthalate; EARLI, Early Autism Longitudinal Risk Investigation.

Discussion

We used phthalate metabolite levels quantified in meconium from two cohorts of children at higher risk for adverse neurodevelopmental outcomes, to evaluate the risk of prenatal phthalate exposure on cognitive function of 12-month-old children. A select number of meconium-measured phthalate metabolites were negatively associated with cognitive function, but others were positively associated with cognitive function in SPS. The AA score was positively associated with cognitive function in EARLI. Notably, in both SPS and EARLI, we observed effect modification of the association of meconium measured MBzP, MEHP, $\sum \mathrm{DEHP}$, and the AA score with cognitive function, by child’s sex. The direction of association of effect modification was identical in both cohorts for the metabolites in common between them. In both cohorts, metabolite levels in meconium were negatively associated with cognitive function among girls, but positively associated among boys. We detected decrease in cognitive function associated with increasing levels of T2 urine MCNP and suggestive positive associations between some metabolites in T3 urine and cognitive function in analyses on the one cohort for which maternal urine was available. We also observed effect modification of the association between a subset of phthalate metabolites in urine from 2^nd trimester, 3^rd trimester and levels averaged over both trimesters, mostly indicating decrease in cognitive function among girls but increase among boys. In EARLI, where we had metabolites in both urine and meconium, the results were somewhat inconsistent between the matrixes and across trimesters. We previously reported that the correlations between metabolites in urine at T2 and T3 varied by the type of metabolite (MiNP: 0.2 to MEP: 0.7); and the correlation between the same metabolites measured in urine and meconium were low(Mathew et al., 2021). Therefore, the differences in strength and direction of association of the metabolites with ELC is perhaps not surprising, but the differences in effect could also be due to the timing of exposure measurement.

To date, several epidemiological investigations have assessed the relationship between prenatal phthalate exposure and cognition/intelligence over a wide age range of children, from approximately 16 weeks to 16 years, but none have used the MSEL to assess cognitive function. Two studies based on different samples from the same prospective cohort study of pregnant women, the Polish Mother and Child Cohort, evaluated the association between phthalate metabolites measured in T3 maternal urine and cognitive development based on the Bayley Scales of Early Development-III (BSID-III) [N = 165] (Polanska et al., 2014), and the Intelligence and Development Scales (IDS) [N = 134] (Jankowska et al., 2019) at 2 years of age. In Polanska et al., although minimally adjusted regression models reported negative associations between several prenatal phthalate (MiBP, $\sum \mathrm{DnBP}$, MBzP, and DEHP) metabolites and BSID-III, indicating association with lower cognitive function, once adjusted for covariates these were no longer statistically significant. In comparison, the latter reported positive associations between one secondary metabolite of DEHP (MEOHP) and IDS measures of fluid intelligence and cognition. Researchers used the General Cognitive Index (GCI) of the McCarthy Scales of Children’s Abilities in a Mexico based prospective cohort [N = 218] to evaluate the association between high molecular weight phthalates in maternal urine and cognitive ability of 48-month-old children, and observed a decrease in cognitive level for higher levels of T1 and T2 urine measured MBzP. (Torres-Olascoaga et al., 2020). Although this report did not demonstrate statistically significant interaction effect of the metabolites with child’s sex, the decrease in cognition was stronger among boys than girls. In a prospective cohort study [N = 168] among infants, researchers used the Fagan Test of Infant Intelligence, an instrument that measures the infant’s visual recognition to a familiar photograph and capacity to distinguish between different visual stimuli to evaluate the association between phthalate metabolites measured in T3 maternal urine samples and cognitive function(Ipapo et al., 2017). This report did not observe any associations between T3 maternal urine-measured phthalates and novelty preference in the full sample but noted significant effect modification by child sex. Increase in levels of MBzP, MEP, and $\sum \mathrm{DEHP}$ were associated with lower cognitive function among girls but the association was null or in the positive direction among boys, similar to the effect modification observed in our study. Similarly, another prospective cohort study among infants [N = 159] that used a visual measure of physical reasoning to evaluate cognitive function did not detect any main effect associations between phthalate metabolites in maternal urine pooled across pregnancy, with physical reasoning, but observed significant interaction effects.(Merced-Nieves et al., 2021). An inter-quartile range increase in levels of MEP, $\sum \mathrm{DEHP}$, and $\sum \mathrm{DiNP}$ were associated with worse cognitive function among boys but no association among girls, contrary to our observations. Several studies have used the Weschler Intelligence Scale to evaluate the association between phthalate exposure and IQ among both young and adolescent children. Among studies evaluating the effects of prenatal phthalate exposure on young children, researchers have observed negative association of T3 urine assessed MBP and MiBP with Full Scale IQ(FSIQ) among 7 year old’s (prospective cohort [N = 328]) (Factor-Litvak et al., 2014), and negative association of T2 urine-measured MBzP with 5, and 8 year old children’s FSIQ (prospective cohort [N = 253]) (Li et al., 2019). A recent prospective cohort [N = 334] study evaluating these associations among 16-year-old children observed negative association of sum of metabolites (MBzP, MCPP, MCNP, MCOP) and $\sum \mathrm{DEHP}$ with FSIQ among boys but positive associations among girls(Hyland et al., 2019) – the direction of effect modification here opposite to what we observe. Other investigations among older children have not detected any associations between prenatal phthalate exposure and cognitive outcomes(Kim et al., 2017; Nakiwala et al., 2018). Two recent studies used chemical mixture analysis methods to characterize the association between prenatal exposure to neurotoxic chemicals including phthalates and cognitive function, but did not observe any strong associations of phthalates with the outcome(Kalloo et al., 2021; Loftus et al., 2021). The lack of uniformity in the type of, and number of metabolites of phthalates assessed across studies limits our ability to make strong conclusions on the relative importance of specific metabolites. However, considering the weight of the current evidence of association of phthalates on cognitive development(Radke et al., 2020) along with our results, MiNP, the DEHP metabolites, and MBzP, in this order of importance, are of most concern.

The mechanisms underlying sex-differential associations between prenatal phthalate exposure and cognitive function are not fully understood. However, since phthalates have been shown to affect sex-steroids such as testosterone(Araki et al., 2014), and since stable testosterone levels are critical in the sex-differential organization of the fetal brain(Gore et al., 2014), the perturbation of fetal sex-steroids might underlie the effect modifications that we observe in our study. . Data on maternal thyroid hormones were not available to us but future studies from cohorts with data available on maternal thyroid hormone and fetal sex-steroids could potentially evaluate the effect of these factors in etiological pathway between prenatal phthalate exposure and fetal neurodevelopment. Studies developed within the framework of the Environmental influences on Child Health Outcomes (ECHO) program, for example, may be able to conduct such analyses with adequate statistical power. We observed nonlinear, quadratic U-shaped or inverted U-shaped relationships between several metabolites and the ELC. Similar nonlinear dose-response relationships have been previously shown between DEHP and testosterone, and brain aromatase (critical for estrogen metabolism) in animal models(Andrade et al., 2006; Do et al., 2012). There are several potential biological reasons for these nonlinear relationships(Lagarde et al., 2015). For example, if the substance/chemical has multiple molecular targets such as receptors and these have differential affinity towards the substance or at different concentration levels, the receptors would be activated separately and could induce opposite effects. Another possibility is the desensitization of receptors at high dose or due to the effect over time resulting in weaker response to the substance. Additionally, saturation of the metabolic system at higher doses of the substance resulting in an opposing effect than at low concentrations has also been suggested as a potential explanation for nonlinear effects(Lagarde et al., 2015). Examples of these biologically plausible explanations have been demonstrated for other endocrine disrupting chemicals. Since all these potential explanations do not preclude the harmful effects of the chemicals, the observation of a nonlinear relationship should not be considered as beneficial.

In almost all current epidemiological studies of prenatal phthalate exposure and neurocognitive outcomes, exposure has been quantified in maternal urine. As a matrix that could capture cumulative gestational exposure, and since it contains lipids, meconium could possibly attract more of the lipophilic High Molecular Weight (HMW) metabolites like those of DEHP. Therefore, we expected the effect estimates of association between meconium measured phthalates and cognitive function to be stronger in magnitude compared to urine measured metabolites from a discrete point in pregnancy. We did not observe differential associations by high and low molecular weight phthalate on ELC between the two matrixes in EARLI where we had metabolites from both, but in SPS we note that the effect estimates were generally stronger among HMWP (DEHP, MBzP, MiNP). It is possible that phthalate exposure measured in meconium has better comparability to metabolites measured in a pooled gestational urine sample from an exposure period equivalent to that of meconium accumulation in pregnancy.

In our meconium phthalate analysis from both cohorts, we observed only a few metabolites that displayed statistically significant associations in adjusted models. However, several metabolites and summary scores (MBzP, MEHP, $\sum \mathrm{DEHP}$ and the AA score) were noted to have significant interaction effects by sex, with similar results in both cohorts. MiPP and MiBP also showed sex-modified effects in EARLI but did not have reliable measurement in SPS. Although we observed similar effects for the AA score in both cohorts, MiBP was not included in the AA score calculation for SPS (because of measurement limitations). Since the parent phthalates of MiBP and MBP are structurally similar and used in similar applications, and since the potency weight for MiBP is the lowest (equal to MBP) among the metabolites used in calculating the score, we do not expect the omission of MiBP to have substantially affected the results.

The levels of phthalate metabolites measured in meconium were somewhat similar between the two cohorts only for DEHP, and we observed that the levels of those phthalates used more in personal care product applications (DBP, DiBP, DEP, BBzP) were higher in EARLI meconium whereas those phthalates used in industrial applications (DiNP, DOP, DPP) were detectable at higher levels in SPS meconium. This could be due to the differences in the urban (EARLI: Philadelphia, Baltimore, Northern California) and rural locations (SPS: Sanford Health, Rapid City, South Dakota) of the studies. Regardless of these differences in the exposure levels across cohorts, the results are consistent across them.

Limitations and Strengths

There are limitations to consider when interpreting the results from our study. Although we observed similar sexually dimorphic associations in meconium in both cohorts, the magnitude and width of the confidence intervals in EARLI reflects the high variability in phthalate metabolite levels. Therefore, in addition to the log-transformation we also re-computed all models with data winsorized at the 2^nd and 98^th percentile to reduce the disproportionate effects of potential outliers. We also recomputed the models for MEHP measured in meconium without the observations below the detectable level and did not detect any results that could affect our conclusions (Supplementary tables 5–7, Supplementary Figures 6, 7). We are also reassured by the consistency of results with those from the larger SPS cohort. In EARLI, although we collected the entire first morning void urine at 2^nd and 3^rd trimester visits, these are still spot urine samples. Since phthalates have short biological half-lives of < 24 hours(Koch et al., 2012; Koch et al., 2006), they reflect short-term exposures and thus our metabolite measurements in urine may not reflect the longer-term exposure. However, exposure to some of the phthalates used in personal care products have been shown to be fairly stable over pregnancy(Braun et al., 2012) and, since EARLI was a prospective cohort, any misclassification of longer-term exposure from spot samples would likely be non-differential with respect to our 12 month outcomes (and therefore bias associations toward the null). Since there was no maternal urine sample collection in the SPS cohort, we were unable to conduct urine phthalate analysis similar to EARLI. Since the interaction effects detected in meconium were similar across cohorts, we speculate that urine results would have been similar to the EARLI results, but with increased precision due to the larger sample size. Contamination of biosamples with external phthalates can be an issue in phthalate exposure assessment (Johns, Cooper, et al., 2015) and is relevant for meconium since meconium contains enzymes that can convert exogenous parent phthalates to their primary metabolites, which would appear identical in assays to those created within the body and thus artificially inflate the exposure level(Kato et al., 2006). The EARLI study was not specifically designed to eliminate external phthalate contamination and except for the use of polypropylene collection containers, we did not have any contamination mitigation or evaluation methods like phthalate-free diaper inserts or field blanks in collection kits. The levels of meconium-measured metabolites were higher in SPS and EARLI as compared to the P4 study, the only other North American epidemiologic cohort with meconium measured phthalates. Since the P4 study had specific sample collection protocols designed to eliminate external contamination, we used the magnitude of the ratio of the meconium metabolites to those in urine as a marker of contamination-controlled sample collection. We observed that the ratio was higher than 1 for all primary metabolites in EARLI, as opposed to those in P4, whereas it was lower than 1 (and similar to P4) for the secondary metabolites(Mathew et al., 2021). While these results suggest that there could be potential contamination issues in EARLI for some primary metabolites, since the direction and strength of associations in meconium samples were similar across SPS and EARLI, it seems unlikely that the contamination is dramatically influencing conclusions.

Another limitation of our study is that we did not have data on other chemicals (Bisphenol-A, lead etc.) available for analysis and therefore we did not adjust for co-pollutant confounding. Recent articles on endocrine disrupting chemicals note that humans are simultaneously exposed to a multitude of chemicals and thus the exposure should ideally be modelled as a chemical mixture. While we agree with this observation, we believe that studies of single metabolites or their molar sums in novel matrixes like meconium, and in large samples is necessary to strengthen the evidence of the relationship between prenatal phthalate exposure and adverse fetal neurodevelopment.

Since both SPS and EARLI cohort children were at a higher risk for adverse cognitive function than the general population, the results from our study may not be generalizable to the general population, but the similarity of sexually dimorphic results with other studies suggests that prenatal exposure to phthalates may affect early child cognition in the same manner in populations with a genetic propensity for ASD and otherwise. Although we adjusted for potential confounders in the analysis, we cannot exclude the possibility of residual confounding due to unmeasured factors. Notably, an often-used confounder is the caregiving environment of the child, which could influence both exposure and outcome. Unfortunately, no measures were available to us. Nevertheless, since our results, while not identical, are in a number of key ways (e.g., significant sexually dimorphic effects) similar to previous reports of the effect of prenatal phthalates on neurodevelopment, we expect to have adjusted for salient factors. Finally, we are mindful of the increase in Type-I error due to the many regression models fit so, rather than emphasize the effect of individual metabolites, we highlight the general similarity of the results across the two cohorts. This provides modest additional evidence suggesting that prenatal phthalate exposure may affect cognitive functioning in a sexually dimorphic manner.

However, this investigation also had considerable strengths. We conducted parallel analyses of prenatal phthalate exposure and child neurodevelopment, including the empirical examination of effect modification by child sex, which has been suggested in past studies, in two distinct cohorts, one of which has one of the largest samples used to date to evaluate such associations. Both cohorts were at higher risk a priori of adverse neurodevelopmental outcomes, helpful in increasing power to detect effects. The cohorts were high risk for different reasons, so the consistency across findings seen for some key exposure metrics supports the generalizability of these findings. Prior studies relied on maternal prenatal urine, which captures point-in-time exposure for measurement of phthalate metabolites. Our study is the first large epidemiologic study to utilize meconium samples which capture cumulative prenatal phthalate exposure beginning at gestational week twelve. We were also able replicate analyses with exposure data from prenatal maternal urine in one of the cohorts that had both biosamples available. The fact that we saw similar results in these analyses in key metabolites provides added support for the robustness of those findings.

Conclusions

In conclusion, we observed that only few prenatal phthalate metabolites were consistently associated with early child cognitive function across metabolites and cohorts. The direction and strength of association varied somewhat by metabolite, time of urine assessment, and biosample. However, several key individual metabolites and summary scores in meconium showed similar sexually dimorphic effects across two cohorts and, in all cases, the effect of prenatal exposure was detrimental to the cognitive development of girls, but was null or associated with better cognition among boys. As this is the first study to evaluate these associations in meconium, future epidemiological studies of prenatal phthalate exposure with follow up through delivery could include meconium collection in their protocols (although careful collection procedures and assessment of external contamination from the diaper is important). At this point, it seems advisable that future studies should be designed to explicitly assess sexually dimorphic effects of phthalates on neurodevelopment with further exploration of the interplay between antiandrogenic and other mechanisms in order to further clarify the mechanism driving such effects.

Gestational phthalate exposure in high-risk cohorts similar to general population

Gestational phthalate exposure negatively associated with child cognitive function

Child sex-differential associations between phthalates and child cognitive function

Similar associations detected across two different cohorts