Prenatal urinary biomarkers of metal and metalloid exposure and child neurodevelopmental outcomes in Puerto Rico

Abstract

Exposure to toxic metals, and insufficient or excess levels of essential metals, particularly during sensitive stages of development, can pose health risks. Previous research reports that prenatal exposure to metal mixtures is associated with poorer cognitive and behavioral outcomes. However, little research has studied the impact of prenatal exposure to complex metal mixtures on child behavioral and emotional development specifically in Puerto Rico, a population with exposure and health disparities. This study aims to understand how prenatal exposure to 14 metals affects child neurodevelopment using the Child Behavior Checklist (CBCL) at 1.5, 2, 3, 4, and 5 years. We examined associations between urinary prenatal metal concentrations, collected at 18 ± 2, 22 ± 2, and 26 ± 2 weeks of gestation, and repeated CBCL raw scores in 301 mother-child pairs from the PROTECT birth cohort in Northern Puerto Rico. The median raw CBCL total problem score was 19. In single-pollutant linear mixed-effects models, As, Co, and Sn were significantly associated with higher internalizing and total problem scores, indicating more parent-reported emotional and behavioral problems. An interquartile range increase in ln(Sn) was associated with a 0.17 (95% CI: 0.05, 0.29) point higher ln(Total Problems + 1) score. As exposure was associated with higher CBCL problem scores for male children across all domains, while Cu was associated with higher externalizing problems among females. In mixture analyses using adaptive elastic net and Bayesian kernel machine regression, Co, As, and Sn predominantly contributed to higher CBCL scores, specifically among male children. However, there were mostly only additive interactions between metals. These results elucidate the effects of multiple metals on behavioral and emotional development.

Introduction

Metals and metalloids are ubiquitous in the environment and widely detected in humans due to exposure through inhalation, dermal absorption, or ingestion (Ashrap et al., 2020a; Borowska and Brzóska, 2015; Gorman Ng et al., 2017; Martin and Grisworld, 2009; Rehman et al., 2018; Shim et al., 2017; Singh et al., 2011). Some metals, such as cobalt (Co), copper (Cu), manganese (Mn), molybdenum (Mo), and zinc (Zn) are essential for human health, but may pose health risks at insufficient or excess levels (Ashrap et al., 2020b; Caito and Aschner, 2015; Horning et al., 2015; Nordberg et al., 2000; O’Neal and Zheng, 2015). Other metals are nonessential and can be toxic even at low levels (Caito and Aschner, 2015; Kortenkamp et al., 2007). For example, lead (Pb) and mercury (Hg) are neurotoxicants and reproductive toxicants, and arsenic (As) and cadmium (Cd) are known human carcinogens (Agency for Toxic Substances and Disease Registry, 2022; Ashrap et al., 2020b; Caito and Aschner, 2015; Jaishankar et al., 2014; Jan et al., 2015; Singh et al., 2011). Pb, Hg, Cd, As, Mn, and Zn are also suspected endocrine disruptors at low or excess levels (Ashrap et al., 2020b; Bloom et al., 2010; De Coster and van Larebeke, 2012; Diamanti-Kandarakis et al., 2009; Mendiola et al., 2011). However, metals typically coexist in the environment and their toxic effects may depend on interactions with other metals (Claus Henn et al., 2014; Shah-Kulkarni et al., 2020). Exposure to a mixture of metals may lead to additive, synergistic, or antagonistic adverse health effects, especially during sensitive stages of development (Sanders et al., 2015a; Shah-Kulkarni et al., 2020). In particular, the prenatal period is one of the most sensitive stages for early-life neurodevelopment, and most metals can cross the placental barrier and accumulate in fetal tissue (Dórea, 2019; Iyengar and Rapp, 2001; Lozano et al., 2022; Shah-Kulkarni et al., 2020). Prenatal exposure to mixtures of metals can cause long-term health consequences to child neurodevelopment through various proposed mechanisms in both in vitro and in vivo studies, such as interference with neurotransmitters; impeded cell proliferation and migration; disruption of calcium, Zn, or other essential metal processes; oxidative stress; lipid, protein, or DNA damage to placental tissue; or endocrine disruption (Aflanie et al., 2015; Ashrap et al., 2020a; De Coster and van Larebeke, 2012; Diamanti-Kandarakis et al., 2009; Fitsanakis and Aschner, 2005; Lidsky and Schneider, 2003; Martin et al., 2018; Méndez-Armenta and Ríos, 2007; Mendiola et al., 2011; Strużyńska, 2000; Sultana et al., 2017; Valko et al., 2005; Wang and Du, 2013).

Previous epidemiological research has established the neurotoxic effects of prenatal exposure to Pb and Hg (Julvez et al., 2012; S. Kim et al., 2018; Sanders et al., 2015a, 2009; Schildroth et al., 2022). More recent evidence has also suggested that perinatal exposure to As, Cd, Mn, and mixtures of metals affects child cognitive and behavioral outcomes (Bauer et al., n.d.; Bobb et al., 2015; Claus Henn et al., 2012; Heng et al., 2022; Horton et al., 2018; S. Kim et al., 2018; Kim et al., 2009, 2013; Lee et al., 2021; Levin-Schwartz et al., 2021; C. Li et al., 2020; Y. Li et al., 2020; Lin et al., 2013; Liu et al., 2018; McDermott et al., 2011; Menezes-Filho et al., 2018; Nyanza et al., 2021; Renzetti et al., 2021; Rodríguez-Carrillo et al., 2022; Shah-Kulkarni et al., 2020; Stein et al., 2022; Tung et al., 2022; Valeri et al., 2017; Wei et al., 2022; Wright et al., 2006; Yorifuji et al., 2011). Other studies have also suggested protective effects of certain metals like iron (Fe) or selenium (Se) on child neurodevelopment (Jeong et al., 2015; Sanders et al., 2015a; Schildroth et al., 2022; Yang et al., 2014). However, most studies have looked at the effects of either individual or mixtures of metals specifically on cognition in children. The effect of metals is also likely dependent on child sex, due to sexual dimorphisms in neurotransmission, gray matter distribution, neuroendocrine activity, hormones, genetics, epigenetics, or metabolism (Gade et al., 2021; Mergler, 2012). Additionally, both timing of exposure during pregnancy and the child age at which subsequent health effects occur may depend on time-specific windows of susceptibility and developmental plasticity (Merced-Nieves et al., 2021; Rice and Barone, 2000). Therefore, there is a need to further understand the impact of prenatal exposure to complex mixtures of metals at low levels on emotional and behavioral outcomes in children across developmental stages, as well as potential sex-specific and temporal effects.

This is particularly important to study in Puerto Rico, as previous research has shown that pregnant women in Puerto Rico had higher biomarker levels of As, barium (Ba), Co, Hg, Mo, antimony (Sb), Mn, and Tin (Sn) compared to women in the National Health and Nutrition Examination Survey (NHANES) in the continental U.S. (Ashrap et al., 2020b). Additionally, there are over 200 hazardous waste sites and 19 active Superfund sites in Puerto Rico; many of which are in the Northern region of Puerto Rico near porous Karst aquifers that supply much of Puerto Rico’s drinking water (Lugo et al., 2001; Padilla et al., 2011). Puerto Rico has high rates of preterm birth, adverse birth outcomes, childhood obesity, and children receiving special education services compared to national averages (“Distribution of gestational age categories,” n.d.; Elías-Boneta et al., 2015; National Council on Disability, 2022). Environmental exposures to metals may be a significant contributing factor to various health disparities in Puerto Rico, in addition to the cumulative impacts of social inequities like high poverty rates and defunded public utilities, education, and healthcare (Lloréns and Stanchich, 2019; Khanam et al., 2021; Karakis et al., 2021; Wu et al., 2019; Vrijheid et al., 2020; Gorini et al., 2014). However, Puerto Rico is often excluded from many national environmental health assessments in the U.S. Thus, the aim of this study is to further understand the association of metals both individually and as a mixture and indicators of emotional and behavioral outcomes in the Child Behavior Checklist (CBCL/1.5–5), at the ages of 18 months, 2 years, 3 years, 4 years, and 5 years old in Northern Puerto Rico.

Methods

I. Study Population

This study is conducted among pregnant participants in PROTECT, an ongoing prospective birth cohort in the Northern Karst region of Puerto Rico. PROTECT is a Superfund Research Program and began in 2010 to assess environmental toxicant exposures in Northern Puerto Rico and risk factors for preterm birth and adverse birth outcomes. PROTECT participants were recruited at approximately 14 ± 2 weeks gestation at seven collaborating prenatal clinics and hospitals. All participants were between 18 and 40 years old, lived in the Northern Karst region of Puerto Rico, did not use oral contraception for at least three months before becoming pregnant, did not use in vitro fertilization to become pregnant, and did not have known medical or obstetrical complications. Participants participated in up to three study visits at 18 ± 2, 22 ± 2, and 26 ± 2 weeks of gestation. Information on medical and pregnancy history and questionnaires on sociodemographics were collected at the initial visit. Spot urine samples and questionnaires were collected at each study visit. Participants were followed until their delivery and information on birth outcomes was abstracted from medical records.

From 2016 to 2019, children up to 4 years old that were born to PROTECT mothers could enroll in the Center for Research on Early Childhood and Development in Puerto Rico (CRECE) study. CRECE was designed to understand how multiple environmental exposures and maternal psychosocial stress impact fetal and early childhood health and development. Follow-up of PROTECT children has continued through the ongoing ECHO-PRO study as a part of the National Institute of Health (NIH) Environmental Influences on Child Health Outcomes (ECHO) program. All participants provided informed consent before participating, and this study was approved by the Institutional Review Boards at the University of Michigan, the University of Puerto Rico, and Northeastern University. More information on PROTECT and CRECE/ECHO-PRO can be found in Ferguson et al. (2019), Meeker et al. (2013), and Manjourides et al. (2020) (Ferguson et al., 2019; Manjourides et al., 2020; Meeker et al., 2013).

II. Prenatal Metal Exposure Biomarkers

Concentrations of the following 14 metals and metalloids were measured in urine: antimony (Sb), total arsenic (As), barium (Ba), cadmium (Cd), cesium (Cs), cobalt (Co), copper (Cu), lead (Pb), manganese (Mn), mercury (Hg), molybdenum (Mo), nickel (Ni), tin (Sn), and zinc (Zn). Spot urine was collected using sterile polypropylene cups, and samples were aliquoted within one hour of collection. Urinary specific gravity was measured with a hand-held digital refractometer (Atago Co., Ltd., Tokyo, Japan) at the University of Puerto Rico Medical Sciences Campus to represent urine dilution. Samples were frozen and stored at −80°C and shipped on dry ice to the University of Michigan, Ann Arbor and then later brought to NSF International (Ann Arbor, MI, USA) for analysis. Concentrations were measured using inductively coupled plasma mass spectrometry (ICP-MS), which was calibrated with a blank and a minimum of four standards for each metal. Calibration curves were verified initially using three points within the curve and continuously after every 10 samples. Acceptable calibration curves had R² values above 0.995. The ICP-MS analysis is described in greater detail by Kim et al. 2018 and Ashrap et al. 2020 (Ashrap et al., 2020b; S. S. Kim et al., 2018).

Machine-read values were used in analyses, and samples with concentrations below the limit of detection (LOD) were replaced by the LOD divided by $\sqrt{2}$. If more than 50% of the samples for a given metal were below the LOD, the metal was omitted from the analysis. For urinary metal biomarker concentrations above the LOD, concentrations were corrected for specific gravity using the following equation:

$$P_c=P\left[\frac{{(SG}_p-1)}{({SG}_i-1)}\right]$$

In this equation, P_c is the specific gravity-corrected concentration, P is the initially measured concentration, SG_p is the median specific gravity of the study population, and SG_i is the specific gravity of the individual participant. The geometric mean urinary metal concentrations across pregnancy at up to three visits were used in subsequent analyses. Exposure biomarkers were right-skewed, and therefore, natural log-transformed.

III. Child Behavior Checklist (CBCL/1.5–5)

The Child Behavior Checklist (CBCL/1.5–5) questionnaire was conducted at 18 months, 2 years, 3 years, 4 years, and 5 years in PROTECT participants’ children to assess emotional and behavioral problems in children. Among our sample of 301 mother-child pairs, 167 individuals had one CBCL assessment, 89 had two, 35 had three, and 10 had four. We had CBCL assessments for 134 children at 18 months, 107 children at 24 months, 113 children at 36 months, 91 children at 48 months, and 45 children at 60 months. In the CBCL, parents’ or guardians’ responses are coded as 0 (not true), 1 (somewhat or sometimes true), or 2 (very true or often true) about the child’s behavior for 99 problem items (Achenbach and Rescorla, 2001). The relevant question scores are totaled to produce seven syndrome scales: emotionally reactive; anxious/depressed; somatic complaints; withdrawn; sleep problems; attention problems; and aggressive behavior. The syndrome scales are then summed into total problems (the sum of internalizing and externalizing problems), internalizing problems (summarizing emotionally reactive, anxious/depressed, withdrawn, and somatic complaints), and externalizing problems (summarizing attention problems and aggressive behaviors). CBCL scores were right-skewed and therefore natural log-transformed, after adding 1 due to the presence of 0 values; this transformation has been utilized in previous studies with the CBCL (Brennan et al., 2024; Huang et al., 2023).

IV. Statistical Analyses

a. Single-Pollutant Analyses

We included 301 mother-child pairs who had complete data on at least one exposure, at least one outcome, sex, maternal age at pregnancy, maternal education level, and child age at assessment in months. Descriptive statistics in our study population were calculated, and Spearman’s correlations between urinary metals and CBCL scores were assessed. Linear mixed-effect models were used to assess the association between the geometric mean of prenatal urinary metal concentrations and repeatedly measured, natural log-transformed CBCL scores + 1. In linear mixed-effects models, we adjusted for sex, maternal age at pregnancy, maternal education level, and child age at assessment in months. In addition, we included three sensitivity analyses, using the same linear mixed-effects models, where we either omitted participants who reported drinking during pregnancy, participants that reported smoking during pregnancy, or participants that reported any environmental tobacco smoke exposure during pregnancy. Participants with missing data were also omitted from the respective models. We also tested for effect modification by sex by including an interaction term between metal exposure and sex as a dummy variable in mixed-effect models. Effect estimates and 95% confidence intervals (CIs) are represented as the change in ln(CBCL score + 1) per an interquartile range (IQR) increase in the natural log-transformed exposure biomarker, by multiplying the effect estimate value by the IQR of each metal biomarker. Multiple linear regression was used to assess associations between prenatal metal concentrations and CBCL scores at each age, adjusting for sex, maternal age, and maternal education level.

b. Mixture Analyses

We utilized two approaches for mixture analyses: adaptive elastic net (adENET) and Bayesian kernel machine regression (BKMR). adENET is a regularized regression method which combines penalties of the least absolute shrinkage and selection operator (LASSO) and Ridge regression, while using collinear variables and conducting statistical inference (Park et al., 2017; Zou, 2006; Zou and Zhang, 2009). Using adENET, we can identify the most important predictor variables relating to CBCL outcomes and create an environmental risk score (ERS) that is a weighted summary of the effect of prenatal metal concentrations as a mixture on CBCL outcomes (Ashrap et al., 2020a; Park et al., 2014). In the adENET analysis, urinary metals and the median of natural log-transformed CBCL outcomes + 1 across ages were used, since adENET is a cross-sectional method and unable to utilize longitudinal, repeated measurements. For adENET, we adjusted for maternal age at pregnancy, maternal educational level, and sex, and we did not penalize the covariates in the model. In adENET, metals with weights not shrunk to zero were included in the calculation of the ERS. The ERS represents a composite summary measure of natural log-transformed, IQR-standardized metals for each participant, capturing the risk effect of the mixture specific to the outcome, weighting metals based on their association with median CBCL scores in adENET. The ERS from the median CBCL scores over time was then used as a predictor to examine the association between the metal mixture and median CBCL scores using multiple linear regression models, adjusting for the aforementioned covariates. Training and testing datasets were not used for the creation of adENET weights and the utilization of the ERS, due to limited sample size.

Additionally, BKMR was used to characterize the joint effect of various urinary metals, interactions, and potential non-linear relationships with repeated CBCL outcomes (Ashrap et al., 2020a; Bobb et al., 2015). Using the results from BKMR, the dose-response relationship of each metal, interactions between metals, and the overall mixture effect was assessed. We used BKMR with a hierarchical variable selection, using 10,000 iterations to ensure convergence, in a Markov chain Monte Carlo algorithm, which allows the user to group metals based on correlations or similar mechanisms of action (Ashrap et al., 2020a; Bobb et al., 2018, 2015). We grouped metals into toxic (Sb, As, Ba, Cd, Cs, Pb, Hg, Ni, Sn) and essential (Co, Cu, Mn, Mo, Zn) groups. More specifically, in our BKMR model, we incorporated a random intercept for participant identification number to account for correlated repeated CBCL outcomes (Bobb et al., 2018, 2015). These BKMR models were adjusted for maternal educational level, maternal age, exact child age in months at assessment, and sex. Subsequently, we performed sex-stratified analyses, adjusting for maternal educational attainment, maternal age, and child age.

Results

Demographic information related to both child and maternal health are included in Table 1 for our study population of 301 mother-child pairs with complete data on at least one exposure and outcome and relevant covariates. Most pregnant participants were of 25–29 years of age, married, had Bachelor’s or higher degrees, reported their household income as $10K to $29.9K per year, and reported paternal educational attainment of some college or less. Additionally, the majority of participants had never smoked, had never drank alcohol, were experiencing their first pregnancy, and reported that they were not prenatally exposed to tobacco smoke. 7.3% of the study participants experienced preterm birth.

Table 1.

Sociodemographics of individuals with at least one urine exposure measurement, one CBCL measurement, and complete cases of covariates included in the main model in our study population (N = 301).

	Mean (SD) or N (%)
Gestational age (weeks)	39.04 (1.72)
Birth weight (lbs)	6.98 (1.23)
Missing	113
Head circumference at birth (cm)	34.01 (2.37)
Missing	116
Length at birth (in)	19.78 (1.23)
Missing	115
Sex of baby
Female	154 (51.2%)
Male	147 (48.8%)
Maternal Age
18–24	68 (22.6%)
25–29	109 (36.2%)
30–34	75 (24.9%)
35–40	49 (16.3%)
Marital status
Single	45 (15.1%)
Married	169 (56.7%)
Cohabitating	84 (28.2%)
Missing	3
Maternal education
GED or less	35 (11.6%)
Some college	103 (34.2%)
Bachelor’s or higher	163 (54.2%)
Annual household income
< $10K	66 (24.0%)
$10K – $29.9K	95 (34.7%)
$30K – $49.9K	79 (28.8%)
>$50K	34 (12.4%)
Missing	27
Smoking
Never smoked	268 (89.0%)
Have smoked before	31 (10.3%)
Current	2 (0.7%)
Alcohol consumption
Never	155 (51.8%)
Before pregnancy	132 (44.1%)
During	12 (4.0%)
Missing	2
Number of previous pregnancies
0	131 (43.5%)
1	103 (34.2%)
2+	67 (22.3%)
Prenatal environmental tobacco smoke exposure
Never	245 (91.8%)
1 hour	9 (3.4%)
>1 hour	13 (4.9%)
Missing	34
Pre-pregnancy BMI
(<24]	151 (52.6%)
(25 to 30]	79 (27.5%)
>30	57 (19.9%)
Missing	14
Preterm birth	22 (7.3%)
Paternal education
GED or less	74 (25.3%)
Some college	135 (46.2%)
Bachelors or higher	83 (28.4%)
Missing	9

Among the 301 mother-child pairs, there were 551 observations for all urinary metals, across the three study visits. Distributions for the geometric mean urinary metals are reported in Table 2. Essential metals, like Zn and Mo, were detected at the highest geometric mean concentrations of 310.97 and 58.96 μg/L, while toxic metals, like Sb and Cd, were detected at lower geometric mean concentrations of 0.08 and 0.19 μg/L. Ni and Mn (−0.15) and Co and Mn (−0.22) were the most negatively correlated metals in Spearman correlations, while Co and Ni (0.54) and Ba and Ni (0.48) were the most positively correlated metals (Supplementary Figure S1).

Table 2.

Descriptive statistics of non-log transformed, geometric mean metal measurements (in μg/L) among individuals with at least one urine exposure measurement, one CBCL measurement, and complete cases of covariates in the main model (N = 301 individuals), showing the percentage of observations below the limit of detection among all repeated samples for the 301 individuals, the number of individuals, the minimum, 25th percentile, 50th percentile, 75th percentile, 90th percentile, maximum, geometric mean, geometric standard deviation, and interquartile range.

	% bLOD	N	Min	P25	P50	P75	P90	Max	GM	GSD	IQR
uAntimony	58.62	301	0.03	0.06	0.08	0.10	0.15	1.27	0.08	1.63	0.04
uArsenic	0.73	301	0.68	6.61	9.62	14.21	20.92	388.51	10.04	1.90	7.60
uBarium	2.00	301	0.26	1.67	2.92	5.17	8.60	31.88	2.82	2.35	3.50
uCadmium	38.29	301	0.03	0.11	0.17	0.30	0.46	1.86	0.19	1.96	0.18
uCesium	0	301	3.02	4.73	5.87	6.89	8.12	11.78	5.78	1.32	2.16
uCobalt	0.54	301	0.11	0.76	1.00	1.32	1.76	4.06	1.00	1.57	0.56
uCopper	6.53	301	5.52	12.06	14.48	18.21	23.10	84.17	14.87	1.46	6.14
uLead	43.01	301	0.06	0.17	0.26	0.38	0.54	4.62	0.27	1.82	0.20
uManganese	14.70	301	0.06	0.63	0.92	1.23	1.60	39.03	0.76	2.33	0.60
uMercury	2.75	301	0.06	0.43	0.76	1.27	2.23	7.69	0.73	2.32	0.85
uMolybdenum	0	301	15.12	45.95	58.18	73.68	95.87	567.67	58.96	1.52	27.73
uNickel	13.43	301	0.92	3.31	4.78	6.67	9.13	33.57	4.68	1.75	3.36
uTin	7.44	301	0.11	0.61	1.09	2.32	4.42	33.68	1.21	2.76	1.72
uZinc	0	301	49.98	212.22	333.09	484.13	623.65	1495.55	310.97	1.83	271.91

Our study population had 490 observations of CBCL assessments across the 5 study visits. The maximum scores were 27 for internalizing problems, 37 for externalizing problems, and 100 for total problems (Table 3). The median scores were 3 for internalizing problems, 8 for externalizing problems, and 19 for total problems. Spearman correlations, among metals and between metals and CBCL scores are reported in Supplementary Figure S2.

Table 3.

Descriptive statistics of raw CBCL summary scores, across all time points, of individuals with at least one urine exposure measurement, one CBCL measurement, and complete cases of covariates in the main model (N = 301 individuals), showing the number of observations, the minimum, 25th percentile, 50th percentile, 75th percentile, 90th percentile, maximum, mean, standard deviation, and interquartile range.

CBCL Domain	N	Min	P25	P50	P75	P90	Max	Mean	SD	IQR
Internalizing	490	0	1	3	6	10	27	4.34	4.50	5
Externalizing	490	0	4	8	13	18.1	37	9.18	6.78	9
Total	490	0	11	19	30	45	100	22.20	16.10	19

I. Single-Pollutant Analyses

In linear mixed-effects models, associations between geometric mean prenatal urinary metal concentrations and repeatedly measured CBCL scores were variable, but mostly null (Figure 1). Statistically significant associations were found among As, Co, and Sn and internalizing problems and total problems (p < 0.05). As was associated with a 0.11 (95% CI: 0.01, 0.21) increase in total problems and 0.15 (0.05, 0.25) higher internalizing problems per IQR increase in exposure (Supplementary Table S1). Co was also positively associated with total problems and internalizing problems, 0.11 (0.01, 0.22) and 0.11 (0.002, 0.21), respectively. Lastly, Sn was associated with higher total and internalizing problems, 0.14 (0.01, 0.26) and 0.17 (0.05, 0.29), respectively. We did not adjust for maternal smoking, environmental tobacco smoke, or maternal alcohol consumption in our main models because most participants reported that they did not smoke, did not drink, and were not exposed to tobacco smoke during pregnancy. However, using single pollutant linear mixed-effects models, we included three sensitivity analyses showing mostly consistent results, aside from larger confidence intervals, after omitting 14 participants that reported drinking during pregnancy or had missing alcohol consumption data (Supplementary Table S2), 2 participants that reported smoking (Supplementary Table S3), and 56 participants that reported any environmental tobacco smoke exposure during pregnancy or were missing data (Supplementary Table S4).

Figure 1.

Results from linear mixed-effect models of the effect of prenatal urinary metals on repeated ln(CBCL summary score + 1), adjusted for sex, child age in months at assessment, maternal education level, and maternal age (N = 301 individuals). The effect estimate is shown as the change in ln(CBCL score + 1) per interquartile range increase in ln(exposure). Red effect estimates and CIs indicate if p-values were below 0.05.

Sex-specific effects varied across metals and CBCL score domains (Figure 2). The effect of As on internalizing problems, externalizing problems, and total problems was significantly modified by sex. An IQR increase in As exposure was associated with a 0.24 (0.08, 0.41) higher total problems among males, compared to 0.03 (−0.09, 0.16) in females (p of interaction term < 0.1, Supplementary Table S5). For externalizing problems, As exposure was associated with a 0.21 (0.04, 0.37) change per IQR increase among males, compared to −0.02 (−0.15, 0.10) change in females (p of interaction term < 0.1). Lastly, for internalizing problems, As exposure was associated with a 0.26 (0.09, 0.42) change per IQR increase among males, compared to 0.09 (−0.04, 0.22) change in females (p of interaction term < 0.1). Contrastingly, Cu was modified by sex in the opposite direction. For externalizing problems, Cu exposure was associated with a 0.10 (−0.03, 0.23) higher score among females and a −0.07 (−0.20, 0.05) lower score among males (p of interaction term < 0.1). Separately, we also explored stratification by age in Supplementary Table S6 and Supplementary Figure S3, in which we found several positive associations between metals and CBCL scores at 24 and 36 months, similar to the linear mixed effects model results.

Figure 2.

Results from linear mixed-effect models of the effect of prenatal urinary metals on repeated ln(CBCL summary score + 1), including an interaction term between sex as a dummy variable and exposure, adjusted for child age in months at assessment, maternal education level, and maternal age (N = 301 individuals). The effect estimate is shown as the change in ln(CBCL score + 1) per interquartile range increase in ln(exposure), with the p-value of the interaction and number of observations. Starred effect estimates if the p-value of the interaction term between sex and exposure was below 0.05.

II. Mixture Analyses

When using adaptive elastic net to assess the effect of combined exposure to the mixture of metals, As, Pb, Hg, Sn, Co, Cu, Mn, Mo, and Zn had weights greater than zero across at least one of the three different CBCL domains (Figure 3, Supplementary Table S7). Specifically, Sn was selected as positively associated predictors of all three CBCL domains, while As, Co, Mn, and Zn were positively associated to at least one domain. Pb, Hg, Cu, Mn, and Mo were negatively weighted in relation to median internalizing or externalizing problems. In analyses using these weights from the adENET as an ERS, adjusting for maternal age, maternal educational level, and sex, the cumulative effect of the metals as a mixture was associated with higher reported CBCL problems (Figure 3, Supplementary Table S8). Specifically, an increase in ERS from the first to the second quartile was associated with a 0.27 (0.02, 0.52) higher total problems, and an increase in ERS from the first to the fourth quartile was associated with a 0.33 (0.07, 0.59) change (p < 0.05). Similarly, an increase in ERS from the first to the second quartile was associated with a 0.34 (0.11, 0.58) change in externalizing problems, and an increase in ERS from the first to the fourth quartile was associated with a 0.28 (0.01, 0.55) higher score (p < 0.05). For internalizing problems, only the fourth quartile showed a significant association of a 0.55 (0.28, 0.82) higher internalizing problems relative to the first quartile (p < 0.05).

Figure 3.

Left: Weights from adaptive elastic net with geometric mean urinary metals and median ln(CBCL score + 1), adjusted for sex, maternal educational level, and maternal age. Orange colors denote a higher positive weight, while blue colors indicate a lower negative weight. Right: Results from a multiple linear regression between the Environmental Risk Score (ERS), stratified by quartiles, and median ln(CBCL score + 1), adjusted for sex, maternal educational level, and maternal age. The effect estimate is shown as in ln(CBCL score + 1) per increase in ERS quartile, relative to the first quartile (N = 301 individuals).

Sex-stratified adENET and ERS results showed different associations between sexes (Figure 4). For total problems, Sb and Zn had non-zero weights among females, with Sb as the only metal negatively weighted. Yet, for males, As, Sn, Cu, and Mo were slightly positively weighted with median total problems, while Sb, Hg, and Co were slightly negatively weighted (Supplementary Table S9). For median externalizing problems, As, Sn, and Co were positively weighted, and Hg and Cu were negatively weighted in males. Only Zn was positively weighted in females for median externalizing problems. For internalizing problems, As and Co were positively weighted and Cu and Mo were negatively weighted for males. As, Sn, and Mn were positively weighted among females, while Pb and Hg were negatively weighted. Using the ERS method, the cumulative effect of the weighted metals was associated with greater median CBCL scores among males compared to females (Figure 4, Supplementary Table S10). For males, all ERS quartiles were associated with significantly higher median CBCL scores (p < 0.05). Most notably, the fourth quartile ERS was associated with a 0.41 (0.21, 0.62) higher internalizing problems compared to the first quartile (p < 0.05). Higher ERS quartiles were generally associated with higher CBCL scores among males. For females, only ERSs at the third or fourth quartile compared to the first quartile were associated with significantly higher internalizing problems (0.3 (0.13, 0.47) for the third quartile and 0.32 (0.15, 0.49) for the fourth quartile, p < 0.05).

Figure 4.

Left: Sex-specific weights from adaptive elastic net with geometric mean urinary metals and median ln(CBCL score + 1), stratified by sex and adjusted for maternal educational level and maternal age (N = 154 females, 147 males). Orange colors denote a higher positive weight, while blue colors indicate a lower negative weight. Right: Results from a multiple linear regression between the Environmental Risk Score (ERS) and median ln(CBCL score + 1), stratified by sex and quartiles, relative to the first quartile, and adjusted for maternal educational level and maternal age. The effect estimate is shown as change in ln(CBCL score + 1) per increase in ERS quartile.

In BKMR hierarchical variable selection models, most metals showed linear relationships to repeatedly measured total problems. The essential metal group had a posterior inclusion probability (PIP) of 0.61, deeming it as more important to the outcome, while the toxic metal group had a PIP of 0.50 (Supplementary Table S11). All conditional PIP values were less than 0.5, except for Co (PIP = 0.76). However, among the toxic metals, Sn (PIP = 0.21) had the largest conditional PIP value. Most metals had modest positive univariate associations with total problems at higher levels, while holding other exposures at their median values, with the exception of Cu, Pb, Hg, and Mo (Figure 5). Sb, Ba, Cs, and Zn also showed relatively null associations with median total problems. BKMR results for internalizing and externalizing problems were similar to total problems. Results for other summary scores and additional sex-specific BKMR results are included in Supplementary Tables S11–S12 and Supplementary Figures S4–S10.

Figure 5.

Univariate response plots, showing the shape of the association between each exposure and total problems, while holding other exposures at their median values, from a Bayesian Kernel Machine Regression (BKMR) with urinary, natural log-transformed metals and repeated ln(CBCL score + 1), adjusted for sex, age at assessment, maternal educational level, and maternal age (N = 301 individuals and 490 observations for total problems).

In bivariate response plots from BKMR hierarchical variable selection models, we assessed potential interactions, looking at the shape of the association between one exposure and outcome, while holding another exposure at varying quantiles and the remaining exposures at their median values (Figure 6). We specifically focused on metals that showed interesting bivariate patterns: As, Co, Pb, and Sn. Most metals showed additive effects of higher total problems in association with increases with another metal, in which the intercepts positively increased per quantile of another metal. However, Pb and Hg still showed unexpectedly negative associations and additive or minor synergistic effects of reduced total problems at increasing quantiles in relation to other metals in Figure 6. The slope of associations with As or Co showed slight changes per quantiles of the other metals, suggesting potential interacting effects.

Figure 6.

Bivariate response plots, showing the shape of the association between each exposure and total problems, holding the metal on the right at varying quantiles and the remaining exposures at their median values, from a Bayesian Kernel Machine Regression (BKMR) with urinary, natural log-transformed metals and repeated ln(CBCL score + 1), adjusted for sex, age at assessment, maternal educational level, and maternal age (N = 301 individuals and 490 observations for total problems).

Discussion

In this study, we examined the association between prenatal urinary concentrations of 14 metals, as both individual pollutants and a mixture, and emotional and behavioral development using the CBCL. In single-pollutant analyses, we found that As, Co, and Sn were significantly associated with higher internalizing and total problem scores, indicating poorer outcomes. In sex-specific, single-pollutant linear mixed-effects models, As was more positively associated with externalizing, internalizing, and total problems in males compared to females. Meanwhile, Cu was more positively associated with externalizing problems in females compared to males. In mixture analyses, As, Co, and Sn predominantly contributed to higher emotional and behavioral problems in CBCL scores, showing elevated cumulative effects as an ERS particularly for male children and mostly additive relationships between metals.

Our findings, particularly on the impacts of As on higher CBCL problems, is consistent with previous research looking at a variety of neurodevelopmental outcomes. The effects of As on cognitive function, Autism or intelligence measures, as well as in-vivo effects on neurotransmission, have been detected in previous research (Bjørklund et al., 2018; Hamadani et al., 2011; He et al., 2021; Obrenovich et al., 2011; Rodríguez-Barranco et al., 2013; Wasserman et al., 2011, 2004). The evidence is emerging in regards to emotional and behavioral outcomes; several previous studies using CBCL scores have found that urinary As was associated with higher internalizing or externalizing problems, especially when looking at the individual syndrome scales (such as emotionally reactive, withdrawn, somatic complaints, anxious/depressed, or thought problems for internalizing problems; attention problems or rule-breaking behavior for externalizing problems; and affective, anxiety, or pervasive developmental problems for other DSM-oriented scales) (Liao et al., 2024; Renzetti et al., 2021; Rodríguez-Carrillo et al., 2022). In a recent study by Liao et al. (2024), they also found As to be the main contributor affecting CBCL problems when using BKMR to look at the effects of a mixture of 6 metals (Liao et al., 2024). Although we found an elevated effect of arsenic on male children, other studies suggest that the effect on arsenic is elevated in female children (Dai et al., 2023; Hamadani et al., 2011; Liao et al., 2024).

In contrast, our findings on the impact of Co and Sn are more novel. Previous epidemiological research has found associations with Sn and lower IQ score, but less research has investigated the neurodevelopmental effects of Co, an essential metal and constituent of vitamin B12, that may be neurotoxic at high levels (Bora et al., 2019; Catalani et al., 2012; Leyssens et al., 2017; Shen et al., 2021). We also unexpectedly detected null or negative associations with Pb and Hg, potentially due to low levels of Pb exposure in our study population as well using urinary measures that likely do not capture exposure to neurotoxic methylmercury or the overall lead burden (Basu et al., 2018; Esteban-López et al., 2022; Gulson et al., 1998; Sallsten et al., 2022). Hg can also be complex to investigate due to co-exposures that may be beneficial, such as selenium, polyunsaturated fatty acids, and omega-3 fatty acids like docosahexaenoic acid, through fish or rice consumption (Bauer et al., n.d.). Many studies investigating metals and neurodevelopment also have found detrimental effects of Cd exposure; although we detected some positive, but not statistically significant associations with Cd, this may also be due to using urinary Cd in place of blood Cd (Kampouri et al., 2024; Kao et al., 2023; Liao et al., 2024; Liu et al., 2022; Rodríguez-Carrillo et al., 2022; Sanders et al., 2015b). We found predominantly linear relationships between metals and CBCL scores, which is what we anticipated to find for toxic metals. Yet, we expected to find non-linear or U-shaped relationships with essential metals, but since our population has elevated levels of several essential metals compared to NHANES averages, we may be representing only part of the dose-response curve in relation to metals like Co, Mo, or Mn (Ashrap et al., 2020b).

Furthermore, our study has several limitations. Although we utilize repeated outcome measures to improve our statistical power in the linear mixed-effects models and BKMR, our study utilizes a relatively small sample size of 301 individuals. Our results were not statistically significant when adjusting for multiple comparisons (q-values > 0.1). Due to our limited sample size, we are unable to split our data into training and testing data when performing the adENET and ERS analyses, making our ERS results potentially overfit. In the adENET and ERS analyses, we are also using median CBCL scores assessed at one to five different ages per participant. Therefore, there may be less accuracy in predicting the CBCL outcome when using the median, given that the median may be representative of only one CBCL measurement for some and up to five for others. We also use urinary measures from prenatal exposure, and although not likely correlated, we cannot account for the potential effects of early childhood exposures. This study also was not able to account for other potential confounders, such as nutritional status or environmental tobacco smoke exposure. The CBCL is also a parent-reported scale and a screening tool, rather than a confirmed diagnosis of neurodevelopmental disorder or direct measure of child neurological functioning. Urinary measures are also not the preferred biomarker media for several metals, depending on the half-life, target exposure period, and elimination pathway, such as Hg or Pb ((Esteban-López et al., 2022; Flora, 2014; Gulson et al., 1998). In a future study, we will investigate comparing associations using urinary metals, blood metals, or integrated multi-media biomarkers measurements.

The present study also has many strengths, as we assess prenatal exposure to 14 different metals and metalloids in relation to repeatedly measured emotional and behavioral development, a less often studied area of neurodevelopment among children. We also employed a variety of statistical methods to explore individual metals, sex- and age- specific effects, cumulative exposure, and non-linear, joint effects. Additionally, pregnant participants in Puerto Rico in previous studies had elevated levels of several metals, but low levels of lead exposure, compared to U.S. averages, which allows us to investigate the effect of different, lesser studied metals (Ashrap et al., 2020b). This study also examines the effect of environmental exposures on children’s developmental health in Puerto Rico, which has important environmental justice implications due to historical and ongoing environmental and health inequities on the island and can inform the development of future metal mixture studies.

Conclusions

In conclusion, we identified that As, Co, and Sn were related to higher emotional and behavioral problems across two CBCL summary scores in single-pollutant analyses. Our results also suggest potential sex-specific effects of As and Cu. As, Co, and Sn also predominantly contributed to higher emotional and behavioral problems, when looking at effects of the metal mixture. Our results underscore previous research on the effects of As on neurodevelopment and suggest potential effects of Co and Sn at elevated concentrations. These findings help further elucidate the effect of multiple metals on emotional and behavioral development in an underrepresented population with low levels of lead exposure. Future research should investigate mixtures of metals in a larger sample size and explore different biomarker media and neurodevelopment measures.

Highlights.

• As, Co, and Sn were associated with higher internalizing and total problem scores

• As and Cu showed sex-specific effects on CBCL scores

• Co, As, and Sn contributed to higher problem scores as a cumulative metal mixture

• Metal mixtures showed mostly additive relationships, with little to no interactions