Prenatal Exposure to Polycyclic Aromatic Hydrocarbons and Executive Functions at School Age: Results from a Combined Cohort Study

Abstract

Background:

Executive functions develop rapidly in childhood, enabling problem-solving, focused attention, and planning. Exposures to environmental toxicants in pregnancy may impair healthy executive function development in children. There is increasing concern regarding polycyclic aromatic hydrocarbons (PAHs) given their ability to transfer across the placenta and the fetal blood-brain barrier, yet evidence from epidemiological studies is limited.

Methods:

We examined associations between prenatal PAH exposure and executive functions in 814 children of non-smoking mothers from two U.S. cohorts in the ECHO-PATHWAYS Consortium. Seven mono-hydroxylated PAH metabolites were measured in mid-pregnancy urine and analyzed individually and as mixtures. Three executive function domains were measured at age 8-9: cognitive flexibility, working memory, and inhibitory control. A composite score quantifying overall performance was further calculated. We fitted linear regressions adjusted for socio-demographics, maternal health behaviors, and psychological measures, and examined modification by child sex and stressful life events in pregnancy. Bayesian kernel machine regression was performed to estimate the interactive and overall effects of the PAH mixture.

Results:

The results from primary analysis of linear regressions were generally null, and no modification by child sex or maternal stress was indicated. Mixture analyses suggested several pairwise interactions between individual PAH metabolites in varied directions on working memory, particularly interactions between 2/3/9-FLUO and other PAH metabolites, but no overall or individual effects were evident.

Conclusion:

We conducted a novel exploration of PAH-executive functions association in a large, combined sample from two cohorts. Although findings were predominantly null, the study carries important implications for future research and contributes to evolving science regarding developmental origins of diseases.

Introduction

Polycyclic aromatic hydrocarbons (PAHs) are chemical compounds formed from incomplete combustion of carbon-based fuels, and are found in tobacco smoke, chargrilled or broiled foods, and particulate air pollution¹. Exposure occurs via inhalation, ingestion, and skin contact, and is ubiquitous in the general population. In a nationally representative sample, PAH compounds were detected in 98-100% of U.S. pregnant women ². Of particular concern, PAHs can readily transfer across the placenta and the fetal blood-brain barrier ^3-5. The fetus is vulnerable to PAH exposure due to the rapid and dynamic brain development that occurs shortly after conception, as well as a limited ability to efficiently detoxify and clear chemicals. These chemical insults can induce a cascade of effects including DNA damage ^6-8, epigenetic programming ⁹, infant growth restriction ¹⁰, and immune and metabolic disorder ^11,12, ultimately causing irreversible neurodevelopment impairment ^13,14.

Prenatal exposure to PAHs has been linked to a variety of neurodevelopmental outcomes among children in the U.S., China, and Poland ^15-42. Data from the Columbia Center for Children’s Environmental Health New York City (CCCEH NYC) cohort have showed the associations of prenatal PAH exposures measured by either air monitoring or PAH-DNA adducts in cord blood with cognitive and mental developmental delay at age 3 (n=183) ²², reduced cognitive performance at age 5-7 (n=249 to 326) ^18,28,31, and increased behavioral problems at age 7-9 (n=252 and 253) ^29,37. Studies based on different source populations in China – two successive cohorts before and after the shutdown of a coal-fired power plant in Tongliang (Cohort one n=150; Cohort two n=158) ^15,16,23,27, the Qingdao birth cohort (n=306 and 348) ^19,30, and the Taiyuan cohort (n=158 and 283) ^39,42 – have generated similar conclusions in children aged 0-3. Consistent findings have also been reported for children in Kraków, Poland (n= 170 to 381)^20,32,38. However, in much larger samples from the ECHO-PATHWAYS consortium (n=835 and 1118), we previously found no association of PAH metabolites in maternal urine with children’s language development, intellectual function, or behavioral problems in early childhood; with the only significant association detected being specifically between 1-hydroxypyrene and neurodevelopmental delay ^35,40.

We identified two key knowledge gaps in prior research on maternal PAH exposures during pregnancy and child neurodevelopment. First, we are not aware of any existing study specifically clarifying the association between maternal PAH exposures and child executive functions, which encompasses higher-order cognitive processes, including working memory, inhibitory control, and cognitive flexibility, to formulate goals and stay focused ⁴³. Deficits in executive functioning can hinder an individual's ability to succeed in school and life ^44-48. The documented associations between child executive functions and several environmental exposures, such as maternal tobacco smoking ⁴⁹, air pollution exposures ⁵⁰, and perfluoroalkyl substance exposure ^51,52, are worrying. There is, however, limited research on the impact of PAH exposure, which may function on shared biological pathways with well-studied chemical exposures. Second, many previous studies used a single PAH-DNA adduct to reflect the overall exposure ^{15-18,23,27,29,34,36,37,41}, which missed the opportunity to examine the potential interplay among PAH components. Cohorts with available data for PAH metabolites/compounds generally evaluated single metabolites in isolation ^{21,24,38,39,42}, despite the fact that pregnant women are exposed to these chemical compounds simultaneously.

Herein, we investigated the associations of individual prenatal PAH exposures measured as mono-hydroxylated metabolites in urine with child overall executive functions and three domains at age 8-9, using data from two U.S. cohorts in the ECHO-PATHWAYS Consortium. In addition, previous studies have shown associations between maternal psychosocial stress (e.g., from material hardship, violence, or traumatic life events) experienced in pregnancy with a broad suite of offspring developmental and health outcomes ^53-57. The co-exposure to psychosocial stressors and environmental contaminants (termed “double jeopardy” ⁵⁸ ) can potentiate each other through common immune, endocrine, metabolic, and epigenetic pathways, increased allostatic load, and further impair individual resilience and ability to recover ⁵⁹. We thus evaluated maternal stressful life events during pregnancy as a potential effect modifier for these associations. Moreover, research has identified several sex differences in neurodevelopment, including morphological, physiological, and chemical differences ⁶⁰. Considering the endocrine disrupting potential of PAHs, we further determined whether these associations differ by child sex. Finally, in exploratory analyses, we examined potential mixture effects of PAH metabolites on executive functions, including non-linear dose response associations of each PAH metabolite in the context of co-exposure to other metabolites, and quantified the overall effects and potential interactions among PAH compounds. We hypothesized that higher prenatal exposures to PAHs would be associated with poorer executive functions in children at school age, and these associations would be more pronounced in males and in children of mothers with greater psychosocial stress during pregnancy.

Methods

Study population

Our analytic sample consisted of mother-child dyads from two cohorts in the ECHO PATHWAYS consortium ⁶¹ -- the Conditions Affecting Neurocognitive Development and Learning in Early childhood (CANDLE) study and The Infant Development and the Environment Study (TIDES). From 2006 to 2011, the CANDLE study recruited 1,503 pregnant women in their second trimester from either the general community or affiliated medical group clinics, and generally reflects the sociodemographic characteristics of the source population in Shelby County, TN. Women were identified and invited to the study after meeting the following criteria: (1) aged 16-40 years old; (2) had a low-risk singleton pregnancy; (3) intended to deliver at one of four Memphis hospitals. From 2010 to 2012, the TIDES study enrolled women in the first trimester of pregnancy from obstetrical clinics affiliated with participating academic medical centers in four cities: San Francisco, California; Rochester, New York; Minneapolis, Minnesota; and Seattle, Washington. Similarly, eligible women included those who were at least 18 years old, had a low-risk pregnancy, and planned to deliver at one of the study hospitals. A total of 803 women were retained throughout the pregnancy and delivered a live birth. Study design, enrollment process, and data collection for these two cohorts have been described elsewhere ^62,63. Several in-person visits at regular intervals and phone-based assessments were conducted across childhood, and data for each participating family member have been collected from questionnaires, clinical and biospecimen measurements, and medical record abstraction. Pregnant women provided informed consent upon enrollment, and children further gave their assents at the age 8-9 visit at individual study sites. The current analysis was approved by the Human Subjects Division at the University of Washington. We included 814 non-smoking mothers and their children who had available urinary PAH measures during pregnancy, at least one valid measure of child executive functions at age 8–9, and complete data for major covariates.

PAH exposure assessments

Monohydroxy-PAH (OH-PAH) metabolites were measured in spot urine samples collected in mid-pregnancy (median gestational week: 22). We determined specific gravity (SG) via a handheld refractometer at the time of collection. Samples were stored at −80 °C in local study biorepositories before analysis at the Wadsworth Laboratory, New York State Department of Health. Laboratory methods for the OH-PAH extraction and analysis have been previously described ^35,40,64. In brief, we performed liquid–liquid extraction followed by liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis. After fortification with an isotopically labeled internal standard mixture, urinary samples were mixed with an 0.5 M ammonium acetate buffer containing 200 units/mL of β-glucuronidase/sulfatase enzyme. Samples were incubated overnight at 37 °C, and then extracted with 7 mL of 80 % pentane: 20 % toluene (v:v) by shaking on a reciprocating shaker for one hour, and centrifuged at 3600g for 20 min. We used a Waters Acquity I-Class UPLC system connected with an Acquity UPLC BEH C18 column to chromatographically separate OH-PAHs. PAH metabolites were identified and quantified using an ABSCIEX 5500 triple quadrupole mass spectrometer. Robust quality assurance protocols were implemented, including analysis of two Standard Reference Materials (SRM^® 3672, SRM^® 3673; National Institute of Standards and Technology, Gaithersburg, MD, USA) containing certified values for several OH-PAHs, recoveries of analytes in reference materials ranging from 79 to 109% ⁶⁵. Calibration standards were injected repeatedly throughout the sample run to monitor for instrument stability.

We quantified 12 OH-PAH metabolites in total that were commonly studied in current environmental health literature. Seven metabolites that were detected in greater than 70% of the combined analytic sample were included: two metabolites of naphthalene (1-hydroxynaphthalene [1-NAP], 2-hydroxynaphthalene [2-NAP]), three metabolites of phenanthrene (2-hydroxyphenanthrene [2-PHEN], 3-hydroxyphenanthrene [3-PHEN], combined 1/9-hydroxyphenanthrene [1/9-PHEN]), the pyrene metabolite 1-hydroxypyrene (1-PYR), and a combination of the fluorene metabolites 2-, 3-, and 9-hydroxyfluorene (2/3/9-FLUO). The limits of detection (LOD) ranged from 0.02 to 0.12 ng/mL. Concentrations below the LOD were assigned a value of LOD/$\sqrt{2}$.

Child executive function measurements

Under standardized protocols, three core skills of executive functions in children – cognitive flexibility, working memory, and inhibitory control – were assessed by a cognitive battery via electronic tablet at age 8-9. Examiners underwent regular training and recertification on the administration of these tests. Cognitive flexibility, the ability to switch between different tasks and maintain multiple activities simultaneously, was quantified as percent accuracy on the Hearts and Flowers Task mixed block condition ⁶⁶. Children were presented with a series of visual stimuli (i.e., hearts and flowers) and were instructed to press the button on the same or opposite side of the stimulus. Eight responses faster than 200ms were characterized as anticipatory and excluded from the analysis ⁶⁶. Further exclusion (n=18) was made based on being correct less than 60% of the time in the simplest trial (the Heart only block), given that children may not understand the instructions appropriately. Working memory, the ability to store and manipulate information temporarily, was quantified using raw scores from the Digit Span subtest of the Wechsler Intelligence Scale for Children, 5^th Edition ^67,68. Children were read a sequence of numbers and recalled the numbers in the same order (forward task), reverse order (backward task), and ascending order (sequencing task). Inhibitory control, the ability to suppress irrelevant information, resist distractions, and overcome habitual behaviors, was assessed using the NIH Toolbox^® Flanker Inhibitory Control and Attention test ^69,70. It requires children to focus on a particular stimulus (an arrow in the middle of the screen) while inhibiting attention to the stimuli flanking it. During the test, children were instructed to press one of the two buttons that corresponded to the direction to which the middle arrow was pointing. To remain consistent with the unscaled measures from the Hearts and Flowers Task, we elected to use raw scores for the Digit Span subtest and uncorrected standard scores for the Flanker test, as is common practice in neuropsychological research ⁷¹.

We additionally computed a composite score of overall executive functions by integrating measures from all three core domains ⁷². The measures in each domain were standardized based on their respective mean and standard deviation. Subsequently, the average of standardized measures across the three domains was re-standardized and transformed by multiplying by 10 and adding 100, resulting in a final composite score with a mean of 100 and a standard deviation of 10. Higher composite and subdomain scores indicated better performance.

Covariates

A number of covariates related to maternal, child, and family characteristics were included, derived from questionnaires, medical/birth records, and clinical measurements. To account for variations in urinary PAH metabolites due to hydration, specific gravity was treated as a covariate. We identified the following child characteristics from the literature: sex (male vs. female), age at executive function assessment, race (White vs. Black vs. Asian vs. American Indian or Alaska Native vs. other vs. multiple races), ethnicity (Hispanic vs. non-Hispanic), birth order (firstborn vs. non-firstborn), secondhand smoking exposure (never vs. ever), and receptive language measured by the NIH Toolbox^® Picture Vocabulary test at age 8-9 ⁷⁰. Parent-reported child race and ethnicity were conceptualized as important social constructs that reflect membership in marginalized groups who have experienced higher exposures to environmental toxicants but fewer returns on educational investments with respect to health and financial gains ^73,74.

Several maternal characteristics were assessed at baseline and updated during follow-up, including age at delivery, maternal education at child visit (less than high school vs. high school/general equivalency diploma vs. college/technical school vs. graduate or professional degree), region and inflation-adjusted household annual income ⁷⁵, number of household members (2-3 vs. 4 vs. 5 vs. ≥6), marital status (married/living as married vs. single/living as single), breastfeeding duration (never vs. <2 months vs. 2-4 months vs. 4-6 months vs. ≥6 months), IQ measured by the Wechsler Abbreviated Scale of Intelligence ⁷⁶, and depression measured by the Patient-Reported Outcomes Measurement Information System ⁷⁷. Stressful life events (SLE) that occurred during pregnancy were assessed using an inventory of 14 events adapted from the Centers for Disease Control and Prevention Pregnancy Risk Assessment Monitoring System (PRAMS) survey ⁷⁸. Women were asked to identify prenatal experiences with illness, death, relationship problems, housing difficulties, legal issues, and financial problems; and all affirmative responses were summed. Neighborhood socioeconomic status was represented by indices in two of the three domains of the Childhood Opportunity Index (COI) -- the social and economic subscale and the educational subscale – geospatially linked to the pregnancy residential address history ^79,80. A higher index suggests better socioeconomic or educational resources in the community.

Statistical analysis

Primary analysis

We reported descriptive statistics on participant characteristics, child executive functions, and the eight PAH metabolites without specific gravity adjustment. Raw PAH metabolites were log₂-transformed to mitigate the influence of extreme values and impose the assumption that risk increases linearly per multiplicative increase in PAH exposure levels. The associations of individual PAH metabolites with the executive function composite score and each of the three core domains were estimated using linear regressions with Huber-White robust standard errors ⁸¹. The selection of adjustment variables was grounded in the existing literature and further informed by a conceptual model (Appendix Figure 1). We developed a staged modeling approach to evaluate the impact of increasing adjustment for confounders and precision variables on effect estimates: The minimal model (Model 1) controlled for batch, an interaction between specific gravity and cohort, child sex, child age, and study site. The primary model (Model 2) was additionally adjusted for child race, child ethnicity, birth order, secondhand smoke exposure, maternal age at delivery, maternal education, marital status, an interaction between region-and inflation-adjusted household income and number of household members, breastfeeding duration, maternal IQ, maternal depression, counts of SLE during pregnancy, and COI in economic and educational domains. The extended model (Model 3) further included an adjustment for child verbal skills in addition to the variables controlled in the primary model, aiming to address the influence of the child’s v receptive language on reading and interpreting test instructions. Since we assigned a slightly different response window to the initial wave of CANDLE participants (n=37) in the Hearts and Flowers Task (either 750ms or 2000ms instead of 1500ms), an indicator for response window was adjusted in all analyses with cognitive flexibility and executive function composite scores.

Secondary analysis

Potential effect modification by child sex or SLEs during pregnancy was examined by interaction models. We introduced product terms for each PAH metabolite with either child sex or continuous SLE counts in separate primary models and estimated interaction p-values. Sex-specific associations were further reported. In addition, Bayesian kernel machine regression (BKMR), a semi-parametric approach that flexibly models complex relationships among multiple pollutants, was employed to explore possible nonlinear associations with individual PAH metabolites, potential pairwise interactions among PAH metabolites, and overall mixture effects on child executive functions ⁸². The exposure mixture included all seven PAH metabolites. We constructed BKMR with a Gaussian kernel function utilizing a component-wise variable selection and fit the models with 200,000 iterations (including 100,000 burn-in iterations) of a Markov chain Monte Carlo algorithm. Posterior Inclusion Probabilities (PIPs) were used to quantify the relative importance of individual PAHs to the overall mixture effect in predicting child executive function. The model converge was diagnosed using trace plots.

Sensitivity analysis

Several sensitivity analyses were conducted to test the robustness of our primary findings. First, to quantify heterogeneity in associations by cohort and study site, we ran the primary analysis leaving out one cohort and one site at a time. Second, we implemented the Levine-Fahey equation to generate specific gravity adjusted PAH metabolites as an alternative method to account for hydration status ⁸³. The equation is specified as: ${\mathsf{P}}_{\mathsf{c}}=\mathsf{P}\ast [\frac{S{G}_{median}-1}{SG-1}]$,

where $\mathsf{P}$ is the raw PAH concentration in urine, SG is the measured specific gravity, and ${\mathsf{SG}}_{\text{median}}$ is the median of specific gravity in each bath. Furthermore, we excluded the initial wave of CANDLE participants with a response window of either 750ms or 2000ms in the Hearts and Flowers Task from the analysis for cognitive flexibility (measured by percent accuracy on the mixed block) and overall executive function (measured by composite score). Lastly, we repeated the analyses of the minimal and primary models in children with complete data for verbal skills to compare results across three models based on an identical analytic sample. All statistical analyses were performed in R (version 4.2.2; R Development Core Team).

Results

A flowchart describing participants is shown in Appendix Figure 2. Descriptive statistics for the 814 mother-child dyads are presented in Table 1. Children (416 females and 398 males) had an average age of 8.8 (SD: 0.7) years when executive functions were measured. Half (52%) were identified by their parents as Black and 39% as White. About one fifth (19%) of children were exposed to secondhand smoke in the home environment. Nearly two thirds of mothers (66%) were married or living with a partner. The mean maternal IQ was 101 (SD: 18), and more than half of mothers (56%) finished college. There were 62% of mothers reporting zero or one major stressful life event during pregnancy. The CANDLE study contributed 75% of the analytic sample and had more Black participants and low-income families than the TIDES study. The current analytic sample had a greater proportion of Black participants than the overall population at enrollment, but other baseline characteristics were similar (Appendix Table 1).

Table 1.

Characteristics of included participants in the CANDLE and TIDES studies from the ECHO PATHWAYS consortium

	Total	Cohort
		CANDLE	TIDES
	(N=814)	(N=613)	(N=201)
Child characteristics
Child age at age 8-9 years visit (year)	8.8 (± 0.7)	8.8 (± 0.7)	8.8 (± 0.5)
Child sex
Female	416 (51 %)	312 (51 %)	104 (52 %)
Male	398 (49 %)	301 (49 %)	97 (48 %)
Child race
White	319 (39 %)	171 (28 %)	148 (74 %)
Black	420 (52 %)	402 (66 %)	18 (9 %)
Asian	16 (2 %)	5 (1 %)	11 (5 %)
American Indian or Alaska Native	2 (0 %)	1 (0 %)	1 (0 %)
Other	44 (5 %)	28 (5 %)	16 (8 %)
Multiple races	13 (2 %)	6 (1 %)	7 (3 %)
Child ethnicity
Hispanic or Latino	30 (4 %)	19 (3 %)	11 (5 %)
Not Hispanic or Latino	784 (96 %)	594 (97 %)	190 (95 %)
Birth order
Not first born	456 (56 %)	370 (60 %)	86 (43 %)
First born	358 (44 %)	243 (40 %)	115 (57 %)
Secondhand smoke exposure
Never	660 (81 %)	461 (75 %)	199 (99 %)
Ever	154 (19 %)	152 (25 %)	2 (1 %)
Gestational week	39.0 (± 1.9)	38.9 (± 1.9)	39.3 (± 1.9)
Birth weight (kg)	3.3 (± 0.6)	3.3 (± 0.6)	3.4 (± 0.6)
Child receptive language	78.9 (± 8.2)	77.5 (± 8.1)	83.2 (± 7.1)
Maternal characteristics
Maternal age at delivery (year)	28.3 (± 5.9)	27.1 (± 5.6)	32.1 (± 5.2)
Region-, inflation-adjusted household income (dollar)	47421 [63543]	30989 [53436]	110813 [94942]
Household counts
2-3	189 (23 %)	140 (23 %)	49 (24 %)
4	322 (40 %)	225 (37 %)	97 (48 %)
5	180 (22 %)	141 (23 %)	39 (19 %)
≥6	123 (15 %)	107 (17 %)	16 (8 %)
Maternal education
Less than high school	30 (4 %)	27 (4 %)	3 (1 %)
High school/GED	237 (29 %)	229 (37 %)	8 (4 %)
Vocational or Technical school or associate degree	97 (12 %)	76 (12 %)	21 (10 %)
College degree	242 (30 %)	165 (27 %)	77 (38 %)
Graduate or Professional degree	208 (26 %)	116 (19 %)	92 (46 %)
Marital status
Married/living as married	537 (66 %)	364 (59 %)	173 (86 %)
Single/living as single	277 (34 %)	249 (41 %)	28 (14 %)
Breastfeeding duration
Did not breastfeed	203 (25 %)	192 (31 %)	11 (5 %)
<2 Months	81 (10 %)	74 (12 %)	7 (3 %)
2-4 Months	115 (14 %)	101 (16 %)	14 (7 %)
5-6 Months	78 (10 %)	67 (11 %)	11 (5 %)
>6 Months	337 (41 %)	179 (29 %)	158 (79 %)
Maternal IQ	101 (± 18.0)	96.2 (± 16.3)	115 (± 15.2)
Maternal depression (PROMIS)	51.8 (± 9.1)	48.2 (± 6.9)	62.8 (± 5.3)
Stressful life events	1 [2]	1 [3]	1 [2]
Urinary specific gravity	1.02 (± 0.01)	1.02 (± 0.01)	1.01 (± 0.01)
Child Opportunity Educational Index (Pregnancy)	−0.03 (± 0.1)	−0.05 (± 0.1)	0.04 (± 0.1)
Child Opportunity Economics Index (Pregnancy)	−0.05 (± 0.3)	−0.11 (± 0.3)	0.12 (± 0.2)
Recruitment site
Memphis SafetyNet, TN	106 (13 %)	106 (17 %)	0 (0 %)
Memphis Community, TN	507 (62 %)	507 (83 %)	0 (0 %)
San Francisco, CA	54 (7 %)	0 (0 %)	54 (27 %)
Minneapolis, MN	60 (7 %)	0 (0 %)	60 (30 %)
Rochester, NY	34 (4 %)	0 (0 %)	34 (17 %)
Seattle, WA	53 (7 %)	0 (0 %)	53 (26 %)

Shown in the table are mean (± SD), counts (percentage), and median [interquartile range]

In the combined sample, the percentage of samples with OH-PAH concentrations <LOD ranged from 0-23%. The concentration was highest in 2-NAP (Geomean [SD]: 3.45 [3.24] ng/mL) and lowest in 3-PHEN (Geomean [SD]: 0.07 [2.64] ng/mL) (Table 2). There was a medium to high pairwise correlation across PAH metabolites (Spearman correlation: 0.49-0.92) (Appendix Figure 3). Exposure levels were generally higher in CANDLE participants than in TIDES participants (Appendix Table 2). Percent accuracy for cognitive flexibility exhibited a slight left-skewed distribution (median [Interquartile range]: 85% [21%]), while the other executive function measurements were distributed normally (mean [SD]: 21.8 [5.0], 93.0 [12.0], and 100.5 [9.9] for the digit span raw score, the Flanker test standard score, and the composite score, respectively) (Appendix Figure 4).

Table 2.

Distributions of mid-pregnancy urinary raw PAH metabolites in included participants in the CANDLE and TIDES studies from the ECHO PATHWAYS consortium

PAHs	Counts of <LOD	Percentage of <LOD	Min.	1st.Qu.	Median	Mean	3rd.Qu.	Max.	Geomean (SD)
1-NAP	54	7%	0.03	0.30	0.65	2.94	1.70	329.02	0.68 (4.49)
2-NAP	3	0%	0.01	1.67	3.89	6.44	7.65	181.50	3.45 (3.24)
2-PHEN	86	11%	0.00	0.04	0.07	0.11	0.13	6.61	0.07 (2.59)
3-PHEN	88	11%	0.00	0.04	0.07	0.22	0.13	94.00	0.07 (2.64)
1/9-PHEN	125	15%	0.00	0.08	0.23	0.38	0.49	6.17	0.18 (4.16)
1-PYR	185	23%	0.01	0.04	0.12	0.20	0.25	4.36	0.09 (4.26)
2/3/9-FLUO	164	20%	0.06	0.34	0.67	1.44	1.38	177.50	0.72 (2.63)

LOD: limit of detection. LODs vary by cohort; 1-NAP: CANDLE 0.020 ng/mL, TIDES 0.040 ng/mL; 2-NAP: CANDLE 0.025 ng/mL, TIDES 0.017 ng/mL; 1/9-PHEN: CANDLE 0.080 ng/mL, TIDES 0.007 ng/mL; 2-PHEN: CANDLE 0.030 ng/mL, TIDES 0.003 ng/mL; 3-OH-PHEN: CANDLE 0.030 ng/mL, TIDES 0.003 ng/mL; 1-PYR: CANDLE 0.030 ng/mL, TIDES 0.009 ng/mL; 2/3/9-FLUO: CANDLE 0.120 ng/mL.

The unit for each PAH metabolite is ng/mL. Analytes below the LOD were assigned a value of LOD/$\sqrt{2}$.

We detected a few significant inverse associations between individual PAH metabolites and executive functions from the minimal model, including 2-NAP and all four outcomes; however, these associations were attenuated to null after additional adjustments for covariates in the primary and the extended model (Figure 1 and Appendix Table 3). We did not find evidence of sex-specific associations (Appendix Table 4) nor effect modification by maternal SLEs during pregnancy (Appendix Table 5), with all P _interaction greater than 0.05. The trace plots from BKMR displayed smooth and stationary traces for each parameter, suggesting successful convergence of the Markov chains. When estimating the univariate dose-response association using BKMR, we found 2/3/9-FLUO to be the most influential predictor (PIP: 0.61) of a non-linear association with working memory (Appendix Figure 5). An inverse association was observed when 2/3/9-FLUO concentrations were restricted to 8-65.9 ng/mL, although the results were highly imprecise. All other univariate responses were negligible, and no evidence was found for a non-linear relationship. Using the 10^th percentile as the reference, each 5-percentile increase in the overall mixture was not associated with any change in executive functions (Appendix Figure 6). Several pairwise interactions between individual PAH metabolites in the associations with working memory were suggested: for example, the adverse associations of working memory with 1/9-PHEN, 1-NAP, and 2-NAP were stronger in those with higher levels of 2/3/9-FLUO; by contrast, the positive associations with 1-PYR and 3-PHEN were more pronounced with higher levels of 2/3/9-FLUO (Appendix Figure 7).

Figure 1.

Associations of maternal urinary PAH metabolites with overall executive function and the three core skills – cognitive flexibility, working memory, and inhibitory control – estimated from multivariable linear regressions

PAH metabolites were log2 transformed.

The minimal model (Model 1) controlled for an indicator for batch, an interaction between specific gravity and cohort, child sex, child age, and study site. The primary model (Model 2) was additionally adjusted for child race, child ethnicity, birth order, secondhand smoking exposure, maternal age at delivery, maternal education, marital status, an interaction between region-and inflation-adjusted household income and number of household members, breastfeeding duration, maternal IQ, maternal depression, counts of stressful life events during pregnancy, and Child Opportunity Indices in economics and educational domains. The extended model (Model 3) further included an adjustment of child receptive language. An indicator for response window was adjusted in all analyses with percent accuracy in the Hearts & Flowers Task and the composite score. Numeric data has been shown in Appendix Table 3.

When restricting data to the CANDLE participants, we found that each 2-fold increase in 1-NAP was associated with a 0.27-point lower digit span raw score for working memory (95%CI: −0.5, −0.04); the other associations were null (Appendix Table 6). Other sensitivity analyses confirmed the main findings, including when we left out one site in TIDES each time (Appendix Table 7), when raw PAH metabolites and specific gravity were replaced by specific gravity adjusted PAH concentrations generated from the Levine-Fahey equation (Appendix Table 8), when the initial wave of CANDLE participants with a response window of either 750ms or 2000ms in the Hearts and Flowers Task were excluded from the analysis (Appendix Table 9), and when the analytic sample was further restricted to those with available child verbal IQ (Appendix Table 10).

Discussion

Based on two socio-demographically diverse pregnancy cohorts in the U.S., we report herein the first investigation of associations of prenatal PAH exposure with executive functions in school aged children. The associations were predominately null, and we found no modification by child sex or maternal SLEs during pregnancy. Our exploratory mixture analysis suggested that 2/3/9-FLUO exposures may be curvilinearly related to working memory, and there was also evidence of several pair-wised interactions, particularly between 2/3/9-FLUO and other PAH metabolites, in adverse associations with working memory.

We did not find evidence of an association between individual PAH metabolites in pregnancy and child executive functions in the primary analysis, which is consistent with our previous studies in the ECHO-PATHWAYS Consortium suggesting no relationship between prenatal PAH exposure and other neurodevelopmental outcomes in children, including IQ and behavioral problems at age 4-6. ³⁵ Two studies in other pediatric populations -- one based on the World Trade Center cohort ⁴¹, another based on a combined cohort in Karviná and České Budějovice, Czech Republic ²¹ – also reported null findings in child cognitive development at age 3 and age 5, respectively. However, the majority of existing studies have shown significantly inverse associations between prenatal PAH exposures and child neurodevelopment in different domains, including the CCCEH cohort in New York City ^{18,22,28,29,31,37}, the cohorts in Tongliang ^15,16,23,27, Qingdao ^19,30, and Taiyuan ^39,42, China, and the cohorts in Kraków, Poland ^20,32,38.

The discrepancy across studies may reflect the following considerations: First, we were specifically interested in executive functions, while many other studies included metrics of overall cognitive performance or developmental delay. Second, in our five study sites and other urban settings, traffic, environmental smoking exposures, residential heating, and food processing are major sources of PAH. Populations living in industrial or low-income areas are more likely to be exposed through heavily trafficked roadways, power plants, industrial boilers, and cooking using solid fuel ⁸⁴. For example, coal-burning and truck depots constituted important sources of PAH in Tongliang ^15,16,23,27 and Taiyuan ^39,42, China, as well as in Kraków, Poland ^20,32,38, where the exposure concentration was significantly higher than in many other study locations such as New York City ⁸⁵. The null results in our primary analysis could be partially explained by relatively low PAH concentrations if a threshold effect exists. Third, several previous studies quantified the cumulative exposures to high molecular weight PAHs by measuring either benzo[a]pyrene-specific DNA adducts or a broader spectrum of adducts from maternal or cord blood, which was resource intensive ^{15-18,23,27,29,34,36,37,41}. In some other studies, PAH concentrations were ascertained by air monitoring ^{20-22,28,31-33,36,37}or from house dust ²⁶, which may precisely capture exposure from one source but ignore the others. We chose to measure urinary PAH metabolites, as these biomarkers are used as proxies for internal doses to assess maternal PAH exposure from inhalation, ingestion, and dermal absorption ⁸⁶. However, there are limitations associated with this approach: the seven urinary metabolites measured in our study only reflect exposure to low molecular weight PAHs ^87,88, and a single measurement of PAH metabolites in urine can only represent a relatively short-term exposure with a half-life of 6-35 hours ⁸⁹. Although short lived pollutants are correlated with many stable environmental features and human activities, the potential misclassification in our exposure assessment should be considered. Fourth, compared to the current analysis, previous studies generally consisted of relatively small samples ranging from 100 to 300 with limited sociodemographic variability ^{15-29,31,33,34,36,39,41,42}. The CCCEH cohort recruited African American and Dominican pregnant women in New York City ^{18,22,28,29,31,37}, the Kraków cohort in Poland was restricted to Caucasian mothers and children ^20,32,38, and the three Chinese cohorts were located in single cities ^{15,16,19,23,30,39,42} Therefore, these results lack transferability to other study settings. Their small samples also posed a challenge to control for important confounders thoroughly, particularly the absence of socioeconomic status in both individual and neighborhood level, as well as maternal psychosocial stress and cognitive function. The robustness of the associations detected from many previous studies may be hampered by residual confounding.

Mixture analysis using BMKR showed a potential non-linear association between prenatal 2/3/9-FLUO exposures and working memory and indicates that 2/3/9-FLUO may also interact with other PAH metabolites in the association with working memory. This result was in accordance with our previous finding in CANDLE from weighted quantile sum regression that PAH metabolite mixture was inversely associated with age 2 Bayley Language scores, and the highest weight was found in 2/3/9-FLUO ⁴⁰. As metabolites of fluorene, hydroxyfluorenes are frequently detected in human urine samples ^90-92. Fluorene is widely used as an intermediate chemical in various industrial applications ⁹³. Animal studies showed the ability of fluorene to be metabolized and reach the cerebral compartment, with the potential to induce behavioral disturbances, including motor activity, responsiveness to sensory stimuli, and physiological and autonomic responses ^94-96. The underlying biological mechanisms are not yet known and may be multifactorial, possibly including tyrosine hydroxylase inhibition and direct effects on dopaminergic and noradrenergic systems ^95,97. We are not aware of any current data showing interactions within PAH mixture in associations with child neurodevelopment. The synergistic interactions of 2/3/9-FLUO with 1/9-PHEN, 1-NAP, and 2-NAP could be explained by the fact that they may converge on common biological pathways and amplify the toxic effect of each other. But caution should be taken when interpreting the mixture interaction effects from BKMR, as the results may be imprecise. Future studies are warranted to re-examine this research question and to elucidate the mechanistic pathways.

In agreement with our previous ECHO-PATHWAYS study ³⁵ and another study in the CCCEH cohort about prenatal PAHs and child behaviors ²⁹, the current study did not detect a sex-specific difference in the association between prenatal PAHs and child executive functions. Nonetheless, a separate ECHO-PATHWAYS analysis found the inverse association between prenatal 2-NAP and child IQ more pronounced in boys.⁹⁸ Another cross-sectional study in the U.S. using data from the National Health and Nutrition Examination Survey reported that boys with high levels of urinary fluorene metabolites were more likely to receive special education services than girls ⁹⁹. Additionally, we did not identify the modified associations by maternal stress during pregnancy. This conclusion conflicted with four other studies on a similar topic: Based on the CCCEH cohort, Perera et al. (2018) found significant differences in the number of Attention-deficit/hyperactivity disorder (ADHD) symptoms in children with both elevated prenatal PAH exposures and material hardship in the prenatal period and early childhood, compared to their counterparts with a single or no exposure ¹⁷; Vishnevetsky et al. (2015) presented significant interaction between high PAH-DNA adducts in cord blood and prenatal material hardship on child IQ and working memory ¹⁸; Pagliaccio et al. (2020) showed that relative to those with lower exposures, children with higher prenatal airborne PAH exposures exhibited stronger positive associations of early life stress with attention problems and ADHD symptoms ³⁶. Moreover, a study in Kraków, Poland also detected a synergetic effect between maternal demoralization and prenatal airborne PAH exposures on child behaviors ³². Others also documented that disadvantages in social conditions can amplify the neurotoxicity of environmental pollutants apart from PAH, such as lead, air pollution exposures, and passive smoking ^100-103. As most mothers in our study reported zero to one stressful life events during pregnancy, the low variability may restrict our statistical ability to detect the hypothesized modified effects. Mothers who suffer from high PAH exposures and a stressful life may also have experienced other risk factors that contribute to impaired executive functions in children, such as poor pregnancy diet and increased exposures to other environmental toxicants. Failure to account for these factors may mask the PAH-executive functions associations in those with more stressful life events. Future studies of large well-characterized samples are warranted to address residual confounding and better characterize vulnerable subgroups.

Our study has several notable strengths and provides unique contributions to the current literature. It encompassed a relatively large sample across five U.S. cities with high diversity in sociodemographic characteristics. Prenatal PAH exposure was estimated using urinary biomarkers, and child executive functions were measured by trained examiners using validated methods. We employed a combination of conventional and novel analytic approaches and comprehensively accounted for key confounders. It is also important to acknowledge that our study has limitations. One is the single metabolites measured from spot urine in mid-pregnancy, which only captured exposures to the low molecular weight PAHs in the past 1-2 days. We also lacked exposure data for other trimesters and the postnatal period. Future studies are needed to include repeated exposure assessments in different pre- and postnatal windows to reflect cumulative exposures as well as to identify critical windows of exposure and tease apart the effects of distinct biological mechanisms. Another limitation is that analyses of PAH metabolites does not allow for differentiating the sources of exposure. Further research is warranted to collect airborne PAH measurements and dietary data to explore how source-specific PAH exposure is associated with neurodevelopment. The reliance on self-report for ascertaining maternal stress during pregnancy is a limitation, but it is likely minor given the high agreement between self-reported psychosocial stress and stress-related biomarkers, such as cortisol, in pregnant women ¹⁰⁴. Lastly, the results may be biased by unmeasured confounders, other environmental risk factors in particular. Subsequent investigation may fill in this gap by collecting data on lead or mercury ¹⁰⁵.

This study marks the inaugural exploration of the association between individual and mixture of prenatal PAH exposures and child executive function. While findings based on our large sample from two U.S. cohorts were predominantly null, the exploratory analysis suggested that there may be a non-linear association between PAH metabolites and executive function, and individual PAH metabolites may interact with each other to adversely impact child executive function. We also investigated the potential for “double jeopardy”, an interaction between PAH exposure and maternal stress, but the data did not support our hypothesis. Our study contributes to the evolving science regarding developmental origins of diseases and underscores the need for continued exploration and refinement of methodologies in future research endeavors, aiming to elucidate the complex effects of chemical exposure mixtures on child neurodevelopment.

Highlights:

1. The PAH-executive functions associations were predominantly null.

2. Evidence suggested pairwise interactions between PAH metabolites on working memory.

3. Results did not support the “double jeopardy” of PAHs and maternal stress on executive functions.