Associations between Maternal Urinary Metabolite Concentrations of Selected Volatile Organic Compounds and Child Neurocognitive Development at Two Years of Age

Abstract

Background:

Neurotoxicity of exposure to volatile organic compounds (VOCs) has been documented in animal studies, but related epidemiological investigations are very limited; particularly, those based on biomonitoring data are not available yet.

Objectives:

This study aimed to evaluate the trimester-specific association between maternal urinary concentrations of multiple VOC metabolites and child neurodevelopment.

Methods:

Twenty VOC metabolites were measured in urine samples repeatedly collected at the first, second, and third trimesters from 1,023 pregnant women in Wuhan, China. The Bayley Scales of Infant and Toddler Development (Chinese Revision) was used to assess children’s neurocognitive development at 2 years old. General linear models and generalized estimating equations were used to analyze the associations of individual urinary VOC metabolite concentrations with children’s neurodevelopment. Weighted quantile sum (WQS) regression and Bayesian kernel machine regression (BKMR) were used to evaluate the effect of the VOC metabolite mixture on children’s neurodevelopment.

Results:

Risk assessment showed that the average hazard quotients of cyanide, 1,3-butadiene, and acrolein during pregnancy exceeded the recommended safety thresholds in more than 90% of the women. Higher urinary concentration of 2-aminothiazoline-4-carboxylic acid (ATCA) (a metabolite of cyanide) was associated with lower child mental development index (MDI) score, and the association was significant at the first trimester among males. Each interquartile ratio-fold increase in the urinary concentration of ATCA at the first trimester was associated with a decrease of 4.25 points (95% confidence interval: $–6.57$, $–1.94$) in males’ MDI scores. Additionally, WQS regression and BKMR analyses suggested that the VOC metabolite mixture was significantly associated with lower MDI in males, and the association was mainly driven by ATCA.

Conclusions:

Exposure to cyanide at environmentally relevant doses may impact neurodevelopment, particularly among males. Early pregnancy appeared to be the sensitive window of the exposure. Further studies are needed to confirm these findings. Corresponding measures need to be taken to reduce gestational exposure to cyanide. https://doi.org/10.1289/EHP15539

Introduction

Volatile organic compounds (VOCs) are ubiquitous in the environment, and people can be simultaneously exposed to multiple VOCs through inhalation,^1,2 ingestion,^3,4 and dermal contact^5,6 via air, food, water, dust, and soil.⁷ After absorption, VOCs are rapidly metabolized in the liver to various hydroxylated or open-ring products, which are excreted primarily through urine.⁷ Adverse health effects of long-term exposure to VOCs have raised significant public concern, including carcinogenicity,^2,8,9 neurotoxicity,^10–12 reproductive toxicity,^13,14 and genotoxicity.^15,16

Some VOCs such as benzene, toluene, and xylene can be transferred from mother to fetus after exposure during pregnancy.^17,18 The fetal period is a critical window of development during which exposure to environmental pollutants like VOCs could affect a child’s neurodevelopment and lead to lifelong health problems. Animal studies have suggested that prenatal exposure to certain VOCs could impair the neurodevelopment of offspring.^19–21 A few epidemiological studies have reported harmful effects of prenatal VOC exposure on children’s mental^22,23 or psychomotor development^24,25 or their behavior.²⁶ However, these studies only focused on a specific VOC (e.g., benzene)²² or used self-administered questionnaires to assess occupational²⁴ or environmental exposure^25,26 to organic solvents. People are often co-exposed to multiple VOCs at the same time, and it is imperative to explore the effect of the VOC mixture on neurodevelopment. Furthermore, since VOCs exist extensively in the environment and humans might have multiple routes of exposure to them, it could be difficult for the participants to comprehensively recognize various VOC exposures in their daily life.²³ In addition, the data about the associations of prenatal exposure to VOCs with child neurodevelopment are scarce in populations outside the developed countries, such as the Chinese population.

In recent years, measurement of urinary VOC metabolites is considered an appropriate approach to quantify human exposure to VOCs,^27–32 which possesses several advantages including the noninvasiveness of urine collection, the integration of exposures via multiple pathways, the relatively longer physiological half-lives compared with those of their parent compounds [e.g., N-acetyl-S-(N-methylcarbamoyl)-L-cysteine (AMCC) has a longer half-life (23 h) of urinary excretion vs. that of its parent compound N,N-dimethylformamide (2 h)],³³ as well as the specificity of most mercapturic acid metabolites corresponding to their parent VOCs.^7,28,34,35 Some urinary VOC metabolites were also included in the National Health and Nutrition Examination Survey (NHANES) 2011–2012 biomonitoring program of the United States Centers for Disease Control and Prevention (USCDC).³⁶

Because of temporal variability in VOC exposure during pregnancy, a single measurement may not fully reflect exposure levels throughout pregnancy.³⁷ Repeated measurements during pregnancy may help to recognize critical windows of susceptibility to the exposure associated with neurodevelopment.^38,39 In this study, we selected 20 urinary VOC metabolites based on the biomonitoring program of the USCDC³⁶ and previous related studies^7,28,32,40 and determined the concentrations of the VOC metabolites in repeated spot urine samples collected at three different trimesters from 1,023 Chinese women to examine individual associations between the VOC metabolites and 2-year-old children’s neurocognitive development. Further, the association between VOC metabolite mixture and neurodevelopment as well as the contribution of each VOC metabolite to the association were examined by weighted quantile sum (WQS)⁴¹ regression and Bayesian kernel machine regression (BKMR)⁴² analyses.

Methods

Study Population

This study included mother–child pairs of a cohort that was established at Wuhan Healthcare Center for Women and Children (WHCWC) since October 2013. In brief, women were included in this study if they had $<16$ gestational weeks at enrollment, had a singleton gestation, were residents in Wuhan, China, donated urine samples during pregnancy, gave birth at the study hospital, and had their child complete neurocognitive development assessment at 2 years old. The ethics committees of Tongji Medical College and WHCWC approved this study. All of the participants provided written informed consent.

From January 2014 to June 2017, a total of 5,112 pregnant women participated in the study and provided at least one urine sample. Among them, 2,782 mothers had their children complete the evaluation of neurocognitive development at approximately 2 years old until May 2020. Among these mother–child pairs, 1,041 mothers who provided a spot urine sample at each trimester during pregnancy were included in the analysis. Active smokers (assessed using urinary cotinine concentrations in any trimester of $\ge 100\text{\hspace{0.17em}ng}/\mathrm{mL}$)⁴³ ($N=15$) and drinkers ($N=2$) during pregnancy and women with missing covariates ($N=1$) were excluded, and 1,023 mother–infant pairs were included in the final study.

Sample Collection and Measurement

Maternal random spot urine samples were collected at the first, second, and third trimester ($13.0\pm 1.0$, $24.5\pm 3.2$, and $33.9\pm 2.9$ weeks, respectively) during pregnancy. Midstream urine samples were collected using $50\text{-mL}$ polypropylene tubes and stored at $–20°\mathrm{C}$ until analysis. Concentrations of the 20 VOC metabolites (selected based on the biomonitoring program of the USCDC³⁶ and previous related studies^7,28,32,40) were measured, and information about the VOC metabolites and their parent compounds is displayed in Table 1. Urinary cotinine concentration was also measured to assess the smoking status of the participants. Formic acid, acetonitrile, methanol, and water were liquid chromatography–mass spectrometry (LC-MS) grade and purchased from Fisher Chemical, Inc.; $\mathrm{\beta }\text{-glucuronidase}$ was purchased from Sigma-Aldrich, Inc.⁴⁴

Table 1.

Volatile organic compound metabolites and the corresponding parent compounds.

Metabolite	Full name of the metabolite	Parent compound
AAMA	N-Acetyl-S-(2-carbamoylethyl)-L-cysteine/Acrylamide mercapturic acid	Acrylamide
AMCC	N-Acetyl-S-(N-methylcarbamoyl)-L-cysteine	N,N-Dimethylformamide, methyl isocyanate
ATCA	2-Aminothiazoline-4-carboxylic acid	Cyanide
BMA	N-Acetyl-S-(benzyl)-L-cysteine/Benzyl mercapturic acid	Toluene, benzyl chloride, benzyl alcohol
BPMA	N-Acetyl-S-(n-propyl)-L-cysteine/n-Propyl mercapturic acid	1-Bromopropane
CEMA	N-Acetyl-S-(2-carboxyethyl)-L-cysteine/2-Carboxyethyl mercapturic acid	Acrolein
CYMA	N-Acetyl-S-(2-cyanoethyl)-L-cysteine/2-Cyanoethyl mercapturic acid	Acrylonitrile
DHBMA	N-Acetyl-S-(3,4-dihydroxybutyl)-L-cysteine/3,4-Dihydroxybutyl mercapturic acid	1,3-Butadiene
GAMA	N-Acetyl-S-(2-carbamoyl-2-hydroxyethyl)-L-cysteine/Glycidamide mercapturic acid	Acrylamide
HEMA	N-Acetyl-S-(2-hydroxyethyl)-L-cysteine/2-Hydroxyethyl mercapturic acid	Acrylonitrile, ethylene oxide, vinyl chloride
2HPMA	N-Acetyl-S-(2-hydroxypropyl)-L-cysteine/2-Hydroxypropyl mercapturic acid	Propylene oxide
3HPMA	N-Acetyl-S-(3-hydroxypropyl)-L-cysteine/3-Hydroxypropyl mercapturic acid	Acrolein
HPMMA	N-Acetyl-S-(3-hydroxypropyl-1-methyl)-L-cysteine/3-Hydroxypropyl-1-methyl mercapturic acid	Crotonaldehyde
2MHA	2-Methylhippuric acid	Xylene
$3\text{MHA\hspace{0.17em}}+4\text{MHA}$	3-Methylhippuric acid plus 4-Methylhippuric acid	Xylene
MHBMA3	N-Acetyl-S-(4-hydroxy-2-buten-1-yl)-L-cysteine/Monohydroxybutenyl mercapturic acid	1,3-Butadiene
MU	trans,trans-Muconic acid	Benzene
PMA	N-Acetyl-S-(phenyl)-L-cysteine/S-Phenyl mercapturic acid	Benzene
TGA	Thiodiglycolic acid	1,2-Dichloroethane, acrylonitrile, vinyl chloride
TTCA	2-Thioxothiazolidine-4-carboxylic acid	Carbon-disulfide

Details about the analytical procedure for the measurement of the VOC metabolites and cotinine have been reported previously.^27,45 Briefly, each sample ($0.1\mathrm{mL}$ urine) was spiked with $10\text{\hspace{0.17em}ng}$ of each internal standard and 85 units of $\mathrm{\beta }\text{-glucuronidase}$ ($10\mathrm{\mu }\mathrm{L}$ of $\mathrm{8,500}\text{\hspace{0.17em}units}/\mathrm{mL}$ in $1\mathrm{M}$ ammonium acetate with $\text{pH}=5.0$), incubated overnight at 37°C, diluted five times with 0.05% formic acid water solution, vortexed, transferred into an ultrafiltration unit, and ultrafiltered at $\mathrm{12,000}\times g$ for 30 min. The filtrate was used for further analysis. The analytes were separated using an ultraperformance liquid chromatography (ExionLC; SCIEX) system with an ACQUITY HPLC HSS T3 column ($1.8\text{-}\mathrm{\mu }\mathrm{m}$, $2.1\text{-mm}\times 150\text{-mm}$ column; Waters) maintained at 40°C. The detection was achieved using an electrospray ionization triple quadrupole mass spectrometer (SCIEX $6500+$) with multiple reaction monitoring modes. The injection volume was $10\mathrm{\mu }\mathrm{L}$.

Urine samples collected from the same mother at the three trimesters were allocated into the same batch, and samples from different mothers were randomly distributed to the batches. Quality control procedures including calibration curves, monitoring of carryover, instrumental sensitivity drift check, field & procedural blanks, duplicates, and matrix spikes were performed. Intraday coefficient of variation (CV) and interday CV were below 7.6% (Table S1). Mean recovery rates (RRs) were determined by the matrix spike samples, ranged between 72.0% and 111%, with standard deviations (SDs) ranging from 4.96% to 18.1%. Since the relatively low RR ($<80\%$), urinary concentration of 2-methylhippuric acid (2MHA) was adjusted using the equation adjusted concentration = detected concentration $\times 89.0\%/72.0\%$ [89.0% was the RR of 3-methylhippuric acid plus 4-methylhippuric acid ($3\text{MHA\hspace{0.17em}}+4\text{MHA}$); 72.0% was the RR of 2MHA], to avoid the underestimation of the concentration of 2MHA.⁴⁶ The instrumental limits of quantification (LOQs), background concentrations of blank samples, method detection limits (MDLs) for all of the analytes are listed in Table S1. The background value of the sample blank was higher than the LOQ for 2-aminothiazoline-4-carboxylic acid (ATCA), thus the MDL of ATCA was calculated by the average background value of the sample blanks plus three times the standard deviation (SD) of the blanks, and the average background value was deducted from concentration of each urine sample.⁴⁷ Urinary specific gravity (SG) was determined to correct the concentrations of the analytes.⁴⁸

Maternal urinary concentrations of zinc, iron, and lead (ng/mL) at the first, second, and third trimesters, which may affect child neurodevelopment, were also measured in this study and included as additional covariates in sensitivity analyses. The measurement of the metal concentrations was conducted from 2016 to 2018, which aimed to assess the impact of prenatal exposure to metals on mother and child health. Details about the measurement of the metal concentrations were described in our previous study.⁴⁹ Briefly, each urine sample ($0.5\mathrm{mL}$) was nitrated overnight with 3% ${\text{HNO}}_3$ and sonicated by ultrasound at 40°C for 1 h. The supernatant was collected after centrifugation and analyzed using inductively coupled plasma mass spectrometry (ICP-MS) (Agilent 7700; Agilent Technologies) in helium mode.

Neurocognitive Development Assessment

At approximately 2 years old (23–26 months), childhood neurocognitive development was evaluated using the Chinese revision⁵⁰ of the Bayley Scales of Infant Development (BSID-CR), which has been applied to 2,409 normal infants and toddlers at 2–30 months of age in 12 cities in China, demonstrating high reliability, validity, and suitability for assessment of neurocognitive development in Chinese children.^50–52 The BSID-CR generates two indices including the mental development and psychomotor development index (MDI and PDI) based on age-adjusted raw scores. Higher score of MDI or PDI represents better neurodevelopment of the child. In the study hospital, the test was conducted by certified psychologists, and quality control was performed through video reviewing of the evaluations. Four to five subjects were selected (each month) and tested by different psychologists to ensure the standardization, reliability, and accuracy of the test results.

Covariate Data Collection

Sociodemographic data about the participants (maternal age, height, prepregnancy weight, education level, annual household income, and paternal education level), as well as information on prenatal care and lifestyle (nutritional supplementation and medication use during pregnancy and drinking habits) were obtained through face-to-face questionnaire surveys conducted by trained investigators. Maternal socioeconomic status (high/low level) was generated by combining maternal and paternal education level and annual household income using a latent class model,^53,54 which identifies groups sharing common characteristics. Pregnancy-related information (history of gestation, maternal anemia, maternal hemoglobin levels in late pregnancy, and mode of delivery) and newborn birth outcome (sex, gestation age at birth, and birth weight) were obtained from the hospital’s electronic medical record system. The breastfeeding status was obtained through telephone follow-up interview at approximately 6 months old. The season of child neurodevelopment assessment was calculated according to the test date (warm season: May–October; cold season: November–April). Passive smoking during pregnancy was assessed using urinary cotinine concentrations (cotinine levels of $5–99\text{\hspace{0.17em}ng}/\mathrm{mL}$ were categorized as passive smoking, and levels of $<5\text{\hspace{0.17em}ng}/\mathrm{mL}$ were categorized as nonsignificant exposure).⁴³

Data Analysis

Concentrations of the VOC metabolites lower than the MDLs were substituted with the $\text{MDL}/\surd 2$.⁵⁵ SG-corrected concentration (ng/mL) of the target analyte was calculated using the following equation: $P_c=P\left[\left({\text{SG}}_{\mathrm{m}}-1\right)/\left(\text{SG}-1\right)\right]$, where $P_c$ means the SG-corrected concentration, P is the measured concentration, and ${\text{SG}}_{\mathrm{m}}$ represents the median SG of urine samples at each trimester. The concentrations of the urinary VOC metabolites were right skewed and then ${\log }_2\text{-transformed}$ for further analyses. The temporal variability of the urinary VOC metabolites across the three trimesters was assessed by intraclass correlation coefficient (ICC) using an iterative Gibbs sampler based left-censored missing value imputation approach (GSimp).^56,57

We calculated estimated daily intake (EDI) and hazard quotient (HQ) based on the urinary concentrations of the VOC metabolites to conduct health risk assessment of the VOC exposures of each participant during pregnancy utilizing Equations 1 and 2, as follows^58,59:

$$\mathit{\text{EDI}}(\mathrm{\mu }\mathrm{g}/\mathrm{kg}/\text{day})=\frac{C_m(\mathrm{\mu }\mathrm{g}/\mathrm{L})\times \mathrm{V}(\mathrm{L}/\text{day})\times {\text{MW}}_p(\mathrm{g}/\text{mol})}{{\mathrm{F}}_{\text{UE}}\times \mathit{\text{BW}}\left(\mathrm{kg}\right)\times {\text{MW}}_m(\mathrm{g}/\text{mol})},$$ (1)

$$\text{HQ}=\frac{\mathit{\text{EDI}}(\mathrm{\mu }\mathrm{g}/\mathrm{kg}/\text{day})}{\text{RfD}(\mathrm{\mu }\mathrm{g}/\mathrm{kg}/\text{day})},$$ (2)

where $C_m$ is the SG-corrected urinary concentration of the VOC metabolite, and V is the urinary excretion volume with 1.2 L/day for women.⁶⁰ ${\text{MW}}_p$ and ${\text{MW}}_m$ is the molecular weight of the parent VOC and relevant metabolite, respectively; BW is the average body weight of each pregnant women before pregnancy and at delivery; and ${\mathrm{F}}_{\text{UE}}$ is the urinary excretion fraction after intake of the VOC (Table S2).²⁷ The reference dose (RfD) of each VOC is obtained from the United States Environmental Protection Agency (USEPA) document or calculated based on its reference concentration for inhalation exposure (Table S2). We calculated EDI and HQ only for VOCs with specific metabolite(s) and established ${\mathrm{F}}_{\text{UE}}$ and RfD values, including acrolein, 1,3-butadiene, cyanide, acrylamide, acrylonitrile, and xylene (Table S2). The details of the calculation method have been described elsewhere.²⁷

General linear models (GLM) were performed to evaluate the associations of urinary average VOC metabolite concentrations across three trimesters with child MDI or PDI scores. Then, generalized estimating equation (GEE) models³⁸ were used to examine the trimester-specific association between the VOC metabolite concentrations and child neurodevelopment and to indentify the critical exposure windows. GEE models included a multiplicative interaction term of the exposure levels and timing of exposure, allowing for the autocorrelation among repeated measurements of the VOC metabolite concentrations; meanwhile, the models could jointly examine the exposure–outcome associations at each trimester and identify whether the association was dependent on the timing of exposure.³⁸ Since acrylamide, acrolein, and xylene exposure levels were reflected through multiple metabolites’ concentrations, we calculated $\mathrm{\Sigma }\text{Acrylamide}$, $\mathrm{\Sigma }\text{Acrolein}$, and $\mathrm{\Sigma }\text{Xylene}$ using the sum of molar concentrations (nmol/L) of acrylamide mercapturic acid/N-acetyl-S-(2-carbamoylethyl)-L-cysteine (AAMA) and glycidamide mercapturic acid/N-acetyl-S-(2-carbamoyl-2-hydroxyethyl)-L-cysteine (GAMA), 2-carboxyethyl mercapturic acid/N-acetyl-S-(2-carboxyethyl)-L-cysteine (CEMA) and 3-hydroxypropyl mercapturic acid/N-acetyl-S-(3-hydroxypropyl)-L-cysteine (3HPMA), and 2MHA and $3\text{MHA\hspace{0.17em}}+4\text{MHA}$, respectively. The above models were also used to evaluate the associations of $\mathrm{\Sigma }\text{Acrylamide}$, $\mathrm{\Sigma }\text{Acrolein}$, and $\mathrm{\Sigma }\text{Xylene}$ with child neurodevelopment. Furthermore, we also evaluated these associations stratified by sex and introduced the interaction terms (child sex × VOC metabolites) into the models to test for the modification effect by child sex. Changes in MDI and PDI for per interquartile ratio (IQR)-fold increase in the urinary VOC metabolite concentrations were calculated using estimates ($\mathrm{\beta }$) and 95% confidence intervals (CIs) from GLM and GEE models with the following equation: $\mathrm{\Delta }=\mathrm{\beta }\times {\log }_2\left(\mathrm{Q}3/\mathrm{Q}1\right)$,⁶¹ where Q3 and Q1 are the 75th and 25th percentile concentrations of the VOC metabolites, respectively. The false discovery rate (FDR) corrections on $p$-values ($p_{\text{FDR}}$) were conducted to control false positive rates in multiple tests.⁶²

We used a directed acyclic graph (DAG)⁶³ (Figure S1) to inspect possible confounding factors and further used the “change-in-estimate” (CIE) strategy⁶⁴ to select confounders. The CIE strategy compares the estimate of the association in the model with adjustment for a covariate with the estimate in the model that excludes the covariate by a backward elimination approach,⁶⁴ and the covariates were selected if they changed the estimate of the association more than 10% (Excel Table S1). Considering that birth weight and gestational age may be mediators of the association between the VOC exposure and child neurodevelopment, we did not include them as covariates. Folic acid supplementation during pregnancy was included as a covariate considering that it had been reported to be significantly associated with child neurodevelopment.⁶⁵ The final models were adjusted for maternal prepregnancy body mass index (BMI) ($<18.5$, 18.5–23.9, $\ge 24{\mathrm{kg}/\mathrm{m}}^2$), age (continuous), socioeconomic status (high/low level), passive smoking during pregnancy (yes/no), folic acid supplementation during pregnancy (yes/no), parity (nulliparous/multiparous), delivery mode (vaginal/cesarean), infant sex (males/females), breastfeeding status ($<6$ months, $\ge 6$ months, or missing), and neurodevelopment assessment season (warm/cold season). In the present analyses, breastfeeding status had missing data (2.9% of values), which was included as a separate missing category in the models.⁶⁶

Additionally, to explore the overall effect of the VOC metabolite mixture on children’s neurodevelopment and estimate the contribution of each VOC metabolite to the mixture effect, we performed WQS regression analyses,⁶⁷ which combined the VOC metabolites into an index (represents quartiles of the VOC metabolites) with adjustment of all the covariates described above and constrained the association to the negative direction, based on the results about the vulnerable group and critical periods from GLM and GEE models. Repeated holdout validation (40%/60% training/testing splits) for 100 times was used in WQS analyses to improve the stability of the results, with the mean (95% confidence interval) as the final estimate.⁶⁸ We also used BKMR to explore the joint effect of the VOC metabolite mixture on children’s neurodevelopment, capture the potential nonlinear effects, and identify the interactions within multiple VOC metabolites.⁴² BKMR utilizes a kernel function to flexibly model the exposure–response relationship while allowing for high collinearity among chemicals and nonlinear associations as well as potential interactions among chemicals.⁶⁹ We used 25,000 iterations to ensure convergence of the models and calculated posterior inclusion probabilities (PIP) to measure the importance of each exposure variable. Prior to the BKMR analyses, all of the VOC metabolite concentrations were centered and scaled after ${\log }_2\text{-transformation}$ to reduce the influence of the variability in different VOC metabolite concentrations.

Considering the potentially residual confounding factors that may affect the reliability of the results, we conducted five sensitivity analyses to test the robustness of the associations as follows:

1. Additional adjustment for more covariates which may affect child neurodevelopment, including multivitamin supplementation during pregnancy (yes/no), iron supplementation during pregnancy (yes/no), medication use during pregnancy (yes/no), maternal hemoglobin levels in late pregnancy (g/L), and gestational weight gain (inadequate, adequate, excessive).

2. Excluding women with anemia during pregnancy.

3. Excluding women with passive smoking during pregnancy (assessed using urinary cotinine concentrations in any trimester of $\ge 5\text{\hspace{0.17em}ng}/\mathrm{mL}$).

4. Excluding children with preterm birth or low birth weight.

5. Additional adjustment for maternal urinary average concentrations of zinc, iron, and lead (ng/mL) across the three trimesters, which may affect child neurodevelopment.

Among the first sensitivity analysis, 11 women were excluded because they had missing data about hemoglobin levels in late pregnancy, and 1,012 participants were included in the analysis. The fifth sensitivity analysis was based on the combination of this study and another study of ours which measured maternal urinary metal (including zinc, iron, and lead) concentrations (ng/mL) at the first, second, and third trimesters.⁴⁹ A total of 568 mother–child pairs who had available data of the maternal urinary concentrations of zinc, iron, lead, and the VOC metabolites as well as child neurodevelopment assessment were included in the fifth sensitivity analysis.

WQS regression and BKMR was fitted using the “gWQS” package (version 3.0.5) and “bkmr” package (version 0.2.2) of R (version 4.4.1; R Development Core Team), respectively. The R source code used for the WQS and BKMR analyses is available in the supplemental data (“Supplementary_SourceCode_WQS_BKMR.R”). All figures were created using the ggplot package in R or GraphPad PRISM (version 7.0; GraphPad PRISM, Inc.). All other analyses were conducted with SAS (version 9.4; SAS Institute Inc.). The CIE strategy used to select confounders was based on SAS macro provided by Atashili and Ta.⁶⁴ Statistical significance was set at $p<0.05$ (two-tailed) or $p_{\text{FDR}}<0.05$.

Results

Participant Characteristics

Among the 1,023 mothers (Table 2), the majority aged 26–34 years (79.2%), were nulliparous (77.2%), had a normal prepregnancy BMI (66.4%), had at least a college-level education (79.8%) and high level of socioeconomic status (73.9%), reported folic acid (87.7%) and multivitamin (88.8%) supplementation during pregnancy, and had no anemia (95.0%) or medication use (86.2%) during pregnancy. Approximately 32.6% of the women were exposed to secondhand smoking during pregnancy. The mean gestational age and birth weight were 39.0 wk and $\mathrm{3,356}\mathrm{g}$, respectively. About half of the women delivered males (52.9%) and breastfed children for at least 6 months (54.9%). Approximately half of the children (52.5%) completed neurodevelopment assessment in the warm season (from May to October). The $\text{mean}\pm \text{SD}$ scores of MDI and PDI were $110\pm 21.7$ and $111\pm 17.1$, respectively, for the 2-year-old children.

Table 2.

Demographic characteristics of the study participants from Wuhan Healthy Baby Cohort in Wuhan, China, 2014–2017 ($N=\mathrm{1,023}$).

Characteristic	$\mathrm{Mean}\pm \mathrm{S}\mathrm{D}$ or $N$ (%) or median (${\mathrm{P}}_{25}–{\mathrm{P}}_{75}$)
Maternal age (years)	$29.4\pm 3.7$
$\le 25$	112 (10.9)
26–34	810 (79.2)
$\ge 35$	101 (9.9)
Maternal prepregnancy BMI ($\mathrm{kg}/{\mathrm{m}}^2$)	$21.1\pm 2.9$
$<18.5$	193 (18.9)
18.5–23.9	679 (66.4)
$\ge 24$	151 (14.8)
Maternal education level
$\le \text{High\hspace{0.17em}school}$	207 (20.2)
$\ge \text{College}$	816 (79.8)
Paternal education level
$\le \text{High\hspace{0.17em}school}$	206 (20.1)
$\ge \text{College}$	817 (79.9)
Annual household income (CNY)
$<\mathrm{50,000}$	117 (11.4)
50,000–100,000	378 (37.0)
$\ge \mathrm{100,000}$	528 (51.6)
Socioeconomic statusa
Lower level	267 (26.1)
Higher level	756 (73.9)
Folic acid supplementation during pregnancy
Yes	897 (87.7)
No	126 (12.3)
Multivitamin supplementation during pregnancy
Yes	908 (88.8)
No	115 (11.2)
Iron supplementation during pregnancy
Yes	439 (42.9)
No	584 (57.1)
Passive smoking during pregnancyb
First trimester
Yes	346 (33.8)
No	677 (66.2)
Second trimester
Yes	338 (33.0)
No	685 (67.0)
Third trimester
Yes	244 (23.9)
No	779 (76.1)
Average over pregnancy
Yes	334 (32.6)
No	689 (67.4)
Medication use during pregnancyc
Yes	141 (13.8)
No	882 (86.2)
Maternal anemia
Yes	51 (5.0)
No	972 (95.0)
Maternal hemoglobin levels in late pregnancy (g/L)d	$121\pm 9.7$
Average maternal urinary zinc levels during pregnancy (ng/mL)e	203 (148–274)
Average maternal urinary iron levels during pregnancy (ng/mL)e	25.7 (15.2–44.4)
Average maternal urinary lead levels during pregnancy (ng/mL)e	1.80 (1.33–2.50)
Pregnancy induced hypertension
Yes	26 (2.5)
No	997 (97.5)
Gestational diabetes mellitus
Yes	97 (9.5)
No	926 (90.5)
GWG categories according to NHC
Inadequate total GWG	44 (4.3)
Adequate total GWG	349 (34.1)
Excessive total GWG	630 (61.6)
Parity
Nulliparous	790 (77.2)
Multiparous	233 (22.8)
Delivery mode
Vaginal delivery	495 (48.4)
Cesarean delivery	528 (51.6)
Child’s sex
Male	541 (52.9)
Female	482 (47.1)
Gestational age at delivery (wk)	$39.0\pm 1.1$
$<37$ (preterm birth)	27 (2.6)
$\ge 37$	996 (97.4)
Child’s birth weight (g)	$\mathrm{3,356}\pm 424$
$<\mathrm{2,500}$ (low birth weight)	21 (2.1)
2,500–3,999	935 (91.4)
$\ge \mathrm{4,000}$	67 (6.5)
Breastfeeding status (months)
$<6$	431 (43.4)
$\ge 6$	562 (56.6)
Missing	30
Neurodevelopment assessment seasonf
Warm season	537 (52.5)
Cold season	486 (47.5)
MDI [$\mathrm{mean}\pm \mathrm{S}\mathrm{D}$ (range, ${\mathrm{P}}_{25}/{\mathrm{P}}_{75}$)]	$110\pm 21.7$ (50–150, 98/126)
PDI [$\mathrm{mean}\pm \mathrm{S}\mathrm{D}$ (range, ${\mathrm{P}}_{25}/{\mathrm{P}}_{75}$)]	$111\pm 17.1$ (51–150, 99/123)

Note: BMI, body mass index; CNY, Chinese Yuan; GWG, gestational weight gain; MDI, mental development index; NHC, National Health Commission of the People’s Republic of China; ${\mathrm{P}}_{25}$, 25th percentile; ${\mathrm{P}}_{75}$, 75th percentile; PDI, psychomotor development index; SD, standard deviation.

^a Socioeconomic status was generated from maternal and paternal education level and annual household income using latent class analysis.

^b Passive smoking during pregnancy was assessed using urinary cotinine concentrations (cotinine levels ranging from $5\text{\hspace{0.17em}to\hspace{0.17em}}99\text{\hspace{0.17em}ng}/\mathrm{mL}$ were categorized as passive smoking, and levels of $<5\text{\hspace{0.17em}ng}/\mathrm{mL}$ were categorized as nonsignificant exposure).

^c Medication use including western medicines and Chinese patent medicines for cold, fever, or other diseases during pregnancy.

^d A total of 1,012 women had the data of hemoglobin levels in late pregnancy (g/L), and women who had missing data ($N=11$) were excluded in the sensitivity analysis.

^e Urinary levels of zinc, iron, and lead (ng/mL) were specific gravity-adjusted. A total of 568 women had this data available; women with missing data ($N=455$) were excluded in the sensitivity analysis.

^f Warm season: May–October; cold season: November–April.

Compared with the population who completed neurodevelopment assessment ($N=\mathrm{2,771}$, excluding participants who had missing covariates), the study population ($N=\mathrm{1,023}$) had a higher proportion of women with low socioeconomic status (26.1% vs. 21.8%), who were multiparous (22.8% vs. 20.2%), and who breastfed their children for less than 6 months (42.1% vs. 39.8%) (Table S3). No other statistically significant differences were observed for the demographic characteristics among the total population ($N=\mathrm{5,077}$, excluding participants who had missing covariates), the population who completed neurodevelopment assessment, and the study population (Table S3).

Urinary Concentrations of VOC Metabolites and Risk Assessment

The urinary VOC metabolite concentrations in different trimesters during pregnancy are shown in Table 3. Most metabolites had a detection rate of $\ge 72.2\%$, except for S-phenyl mercapturic acid/N-acetyl-S-(phenyl)-L-cysteine (PMA) and monohydroxybutenyl mercapturic acid/N-acetyl-S-(4-hydroxy-2-buten-1-yl)-L-cysteine (MHBMA3) (both nearly not detected). Among them, 3-hydroxypropyl-1-methyl mercapturic acid/N-acetyl-S-(3-hydroxypropyl-1-methyl)-L-cysteine (HPMMA) had the highest concentration [geometric mean (GM): $877\text{\hspace{0.17em}ng}/\mathrm{mL}$] throughout the pregnancy, followed by 3HPMA ($203\text{\hspace{0.17em}ng}/\mathrm{mL}$), thiodiglycolic acid (TGA) ($189\text{\hspace{0.17em}ng}/\mathrm{mL}$), ATCA ($126\text{\hspace{0.17em}ng}/\mathrm{mL}$), and 3,4-dihydroxybutyl mercapturic acid/N-acetyl-S-(3,4-dihydroxybutyl)-L-cysteine (DHBMA) ($118\text{\hspace{0.17em}ng}/\mathrm{mL}$). The Spearman correlation coefficients among the VOC metabolites ranged from 0.14 to 0.87 (Figure S2) based on the average concentrations over the three trimesters and from 0.18 to 0.89 (Figure S3), 0.23 to 0.90 (Figure S4), and 0.27 to 0.90 (Figure S5) at the first, second, and third trimester, respectively. Most VOC metabolite concentrations showed a decreasing trend from the first to the second and third trimesters. The ICCs of the VOC metabolite concentrations across the trimesters ranged from 0.04 to 0.19, indicating the variability of the VOC metabolite concentrations during pregnancy.

Table 3.

Distributions of the urinary SG-corrected concentrations of volatile organic compound metabolites during pregnancy and the trimester-specific concentrations (ng/mL) in participants from Wuhan Healthy Baby Cohort in Wuhan, China, 2014–2017 ($N=\mathrm{1,023}$ participants; $n=\mathrm{1,023}$ samples for each trimester).

Analytes	Detection frequency (%)			Average concentrations during pregnancy		Trimester-specific median			ICC
	First	Second	Third	Median	GM	First	Second	Third
AAMA	100	99.9	99.8	25.1	25.1	28.4	20.0	18.3	0.14
GAMA	99.9	99.9	99.6	3.79	3.76	4.20	3.31	2.77	0.19
$\mathrm{\Sigma }\text{Acrylamide}$ (nmol/L)	—	—	—	122	123	139	100	89.8	0.14
AMCC	97.8	95.3	92.1	24.8	24.5	31.8	20.9	16.0	0.04
ATCA	100	100	100	123	126	129	112	100	0.17
BMA	99.2	99.3	98.4	2.35	2.47	2.35	1.78	1.66	0.15
BPMA	91.7	85.2	83.0	2.74	2.87	2.67	1.46	1.09	0.11
CEMA	100	99.9	99.7	46.9	45.9	46.6	40.2	36.1	0.14
3HPMA	100	99.9	99.7	209	203	205	155	161	0.10
$\mathrm{\Sigma }\text{Acrolein}\left(\text{nmol}/\mathrm{L}\right)$	—	—	—	1,141	1,121	1,143	896	891	0.11
CYMA	100	99.7	99.8	0.89	0.88	0.95	0.74	0.68	0.17
DHBMA	100	99.8	99.7	124	118	137	104	92.6	0.11
HEMA	100	99.8	99.1	0.93	1.00	1.10	0.78	0.63	0.16
2HPMA	100	99.7	99.6	13.5	13.9	14.5	12.3	10.8	0.13
HPMMA	100	99.9	100	904	877	939	709	683	0.09
2MHA	100	99.9	99.9	24.9	25.3	27.8	20.5	18.7	0.10
$3\text{MHA\hspace{0.17em}}+4\text{MHA}$	100	100	99.9	63.7	65.8	67.2	55.2	49.4	0.10
$\mathrm{\Sigma }\text{Xylene}\left(\text{nmol}/\mathrm{L}\right)$	—	—	—	463	474	491	396	366	0.10
MHBMA3	0.10	0.00	0.00	<MDL	<MDL	<MDL	<MDL	<MDL	—
MU	100	100	100	38.1	40.4	36.3	25.3	29.6	0.10
PMA	0.00	0.00	0.00	<MDL	<MDL	<MDL	<MDL	<MDL	—
TGA	100	100	100	190	189	197	157	159	0.13
TTCA	80.7	72.2	73.5	11.6	11.6	7.06	4.05	4.45	0.09
Cotinine	100	100	100	6.15	3.78	3.50	3.45	2.66	0.49

Note: —, no data; $\mathrm{\Sigma }\text{Acrolein}$, the sum of molar concentrations (nmol/L) of CEMA and 3HPMA; $\mathrm{\Sigma }\text{Acrylamide}$, the sum of molar concentrations (nmol/L) of AAMA and GAMA; $\mathrm{\Sigma }\text{Xylene}$, the sum of molar concentrations (nmol/L) of 2MHA and $3\text{MHA\hspace{0.17em}}+4\text{MHA}$; 2HPMA, N-acetyl-S-(2-hydroxypropyl)-L-cysteine; 3HPMA, N-acetyl-S-(3-hydroxypropyl)-L-cysteine; 2MHA, 2-methylhippuric acid; $3\text{MHA\hspace{0.17em}}+4\text{MHA}$, 3-methylhippuric acid plus 4-methylhippuric acid; AAMA, N-acetyl-S-(2-carbamoylethyl)-L-cysteine; AMCC, N-acetyl-S-(N-methylcarbamoyl)-L-cysteine; ATCA, 2-aminothiazoline-4-carboxylic acid; BMA, N-acetyl-S-(benzyl)-L-cysteine; BPMA, N-acetyl-S-(n-propyl)-L-cysteine; CEMA, N-acetyl-S-(2-carboxyethyl)-L-cysteine; CYMA, N-acetyl-S-(2-cyanoethyl)-L-cysteine; DHBMA, N-acetyl-S-(3,4-dihydroxybutyl)-L-cysteine; GAMA, N-acetyl-S-(2-carbamoyl-2-hydroxyethyl)-L-cysteine; GM, geometric mean; HEMA, N-acetyl-S-(2-hydroxyethyl)-L-cysteine; HPMMA, N-acetyl-S-(3-hydroxypropyl-1-methyl)-L-cysteine; ICC, intraclass correlation coefficient; MDL, method detection limit; MHBMA3, N-acetyl-S-(4-hydroxy-2-buten-1-yl)-L-cysteine; MU, trans,trans-muconic acid; PMA, N-acetyl-S-(phenyl)-L-cysteine; SG, specific gravity; TGA, thiodiglycolic acid; TTCA, 2-thioxothiazolidine-4-carboxylic acid.

Table 4 presents the HQs of the selected parent VOCs during pregnancy. Generally, the average exposure levels of cyanide (parent compound of ATCA), 1,3-butadiene (parent compound of DHBMA), and acrolein (parent compound of CEMA and 3HPMA) in 1,023 (100%), 967 (94.5%), and 948 (92.7%) participants exceeded the RfD values recommended by the USEPA, respectively.

Table 4.

HQs of the selected parental volatile organic compounds during pregnancy based on USEPA RfD values for participants from Wuhan Healthy Baby Cohort in Wuhan, China, 2014–2017 ($N=\mathrm{1,023}$ participants; $n=\mathrm{3,069}$ samples).

Parent compounds	Minimum	25th	50th	75th	Maximum	$>1$ [$n$ (%)]
						Averaged	Overall
Acrolein	0.26	1.66	2.38	3.55	39.36	948 (92.7)	2,455 (78.8)
1,3-Butadiene	0.26	1.81	2.55	3.52	13.34	967 (94.5)	2,555 (82.0)
Cyanide	1.00	2.64	3.58	4.99	22.78	1,023 (100)	3,000 (96.3)
Acrylamide	0.01	0.03	0.05	0.06	0.27	0 (0.00)	0 (0.00)
Acrylonitrile	0.004	0.01	0.02	0.03	0.17	0 (0.00)	1 (0.03)
Xylene	0.001	0.002	0.003	0.004	0.06	0 (0.00)	0 (0.00)

Note: HQ, hazard quotient; USEPA, the United States Environmental Protection Agency; RfD, reference dose.

Associations between Child Neurodevelopment and Individual VOC Metabolites

The GLM models showed significant associations between higher (average) maternal urinary concentrations of several VOC metabolites and lower MDI scores among the males (Figure 1; Table S4). Specifically, each IQR-fold increase in the averaged urinary concentrations of ATCA, CEMA, HPMMA, and TGA were significantly associated with $–4.33$ (95% CI: $–6.87$, $–1.80$), $–2.77$ (95% CI: $–5.23$, $–0.31$), $–2.68$ (95% CI: $–5.00$, $–0.37$), and $–2.96$ (95% CI: $–5.42$, $–0.50$) changes in the males’ MDI scores, respectively. After the FDR correction, the association between elevated ATCA and lower males’ MDI score remained significant ($p_{\text{FDR}}=0.02$) (Table S4). The interactions of child sex with both ATCA ($p_{\text{sex-int}}<0.01$) and TGA ($p_{\text{sex-int}}=0.01$) on the MDI scores were significant (Table S4).

Figure 1.

The changes and 95% CIs in MDI and PDI scores associated with each IQR-fold increase in averaged SG-adjusted maternal urinary analyte levels (ng/mL) during pregnancy using general linear models for participants from Wuhan Healthy Baby Cohort in Wuhan, China, 2014–2017 ($N=\mathrm{1,023}$ participants; $n=\mathrm{3,069}$ samples). $\mathrm{\Sigma }\text{Acrylamide}$, $\mathrm{\Sigma }\text{Acrolein}$, and $\mathrm{\Sigma }\text{Xylene}$ was the sum of molar concentrations (nmol/L) of AAMA and GAMA, CEMA and 3HPMA, and 2MHA and $3\text{MHA\hspace{0.17em}}+4\text{MHA}$, respectively. Details of the chemical abbreviations are provided in Table 1. The models were adjusted for maternal prepregnancy body mass index, age, socioeconomic status, passive smoking during pregnancy, folic acid supplementation during pregnancy, parity, delivery mode, infant sex, breastfeeding status, and neurodevelopment assessment season. Sex-stratified analyses were adjusted for the abovementioned covariates except for child sex. *$p<0.05$ after false discovery rate correction for multiple comparison. The numerical data are listed in Table S4. Note: 2HPMA, N-acetyl-S-(2-hydroxypropyl)-L-cysteine; 3HPMA, N-acetyl-S-(3-hydroxypropyl)-L-cysteine; 2MHA, 2-methylhippuric acid; $3\text{MHA\hspace{0.17em}}+4\text{MHA}$, 3-methylhippuric acid plus 4-methylhippuric acid; AAMA, N-acetyl-S-(2-carbamoylethyl)-L-cysteine; AMCC, N-acetyl-S-(N-methylcarbamoyl)-L-cysteine; ATCA, 2-aminothiazoline-4-carboxylic acid; BMA, N-acetyl-S-(benzyl)-L-cysteine; BPMA, N-acetyl-S-(n-propyl)-L-cysteine; CEMA, N-acetyl-S-(2-carboxyethyl)-L-cysteine; CI, confidence interval; CYMA, N-acetyl-S-(2-cyanoethyl)-L-cysteine; DHBMA, N-acetyl-S-(3,4-dihydroxybutyl)-L-cysteine; GAMA, N-acetyl-S-(2-carbamoyl-2-hydroxyethyl)-L-cysteine; HEMA, N-acetyl-S-(2-hydroxyethyl)-L-cysteine; HPMMA, N-acetyl-S-(3-hydroxypropyl-1-methyl)-L-cysteine; IQR, interquartile ratio; MDI, mental development index; MU, trans,trans-muconic acid; PDI, psychomotor development index; SG, specific gravity; TGA, thiodiglycolic acid; TTCA, 2-thioxothiazolidine-4-carboxylic acid.

The trimester-specific associations between the individual urinary VOC metabolites during pregnancy and children’s neurodevelopment are shown in Figure 2 and Table S5. Based on the first-trimester urinary concentrations, higher levels of ATCA and HPMMA were associated with lower child MDI score. After being stratified by child sex, significant associations between higher levels of the VOC metabolites or calculated parent compound and lower Bayley scores were found only among males. Specifically, each IQR-fold increase in urinary concentrations of ATCA, CEMA, 3HPMA, $\mathrm{\Sigma }\text{Acrolein}$, and HPMMA was significantly associated with $–4.25$ (95% CI: $–6.57$, $–1.94$), $–3.01$ (95% CI: $–5.42$, $–0.60$), $–2.44$ (95% CI: $–4.80$, $–0.07$), $–2.63$ (95% CI: $–5.02$, $–0.23$), and $–2.37$ (95% CI: $–4.38$, $–0.35$) change in the males’ MDI scores, respectively. Higher urinary concentrations of benzyl mercapturic acid/N-acetyl-S-(benzyl)-L-cysteine (BMA) ($\mathrm{\beta }=–2.09$; 95% CI: $–3.81$, $–0.37$), DHBMA ($\mathrm{\beta }=–1.65$; 95% CI: $–3.24$, $–0.07$), and HPMMA ($\mathrm{\beta }=–1.55$; 95% CI: $–3.07$, $–0.03$) were significantly associated with lower PDI scores among males. Based on the second-trimester urinary concentrations, ATCA ($\mathrm{\beta }=–3.30$; 95% CI: $–5.71$, $–0.89$), n-propyl mercapturic acid/N-acetyl-S-(n-propyl)-L-cysteine (BPMA) ($\mathrm{\beta }=–2.93$; 95% CI: $–5.59$, $–0.27$), and 2-thioxothiazolidine-4-carboxylic acid (TTCA) ($\mathrm{\beta }=-3.07$; 95% CI: $–5.74$, $–0.40$) were negatively associated with the males’ MDI scores. Based on the third-trimester urinary concentrations, HPMMA ($\mathrm{\beta }=–1.42$, 95% CI: $–2.83$, $–0.003$) was significantly associated with lower MDI scores in the total children. After the FDR correction, the negative association between ATCA at the first trimester and the males’ MDI scores remained significant ($p_{\text{FDR}}=0.01$) (Table S5).

Figure 2.

The changes and 95% CIs in MDI and PDI scores associated with each IQR-fold increase in SG-adjusted maternal urinary analyte levels (ng/mL) at the first, second, and third trimester during pregnancy using generalized estimating equation models for participants from Wuhan Healthy Baby Cohort in Wuhan, China, 2014–2017 ($N=\mathrm{1,023}$ participants; $n=\mathrm{1,023}$ samples in each trimester). $\mathrm{\Sigma }\text{Acrylamide}$, $\mathrm{\Sigma }\text{Acrolein}$, and $\mathrm{\Sigma }\text{Xylene}$ was the sum of molar concentrations (nmol/L) of AAMA and GAMA, CEMA and 3HPMA, and 2MHA and $3\text{MHA\hspace{0.17em}}+4\text{MHA}$, respectively. Details of the chemical abbreviations are provided in Table 1. The models were adjusted for maternal prepregnancy body mass index, age, socioeconomic status, passive smoking during pregnancy, folic acid supplementation during pregnancy, parity, delivery mode, infant sex, breastfeeding status, and neurodevelopment assessment season. Sex-stratified analyses were adjusted for the abovementioned covariates except for child sex. *$p<0.05$ after false discovery rate correction for multiple comparison. The numerical data are listed in Table S5. Note: 2HPMA, N-acetyl-S-(2-hydroxypropyl)-L-cysteine; 3HPMA, N-acetyl-S-(3-hydroxypropyl)-L-cysteine; 2MHA, 2-methylhippuric acid; $3\text{MHA\hspace{0.17em}}+4\text{MHA}$, 3-methylhippuric acid plus 4-methylhippuric acid; AAMA, N-acetyl-S-(2-carbamoylethyl)-L-cysteine; AMCC, N-acetyl-S-(N-methylcarbamoyl)-L-cysteine; ATCA, 2-aminothiazoline-4-carboxylic acid; BMA, N-acetyl-S-(benzyl)-L-cysteine; BPMA, N-acetyl-S-(n-propyl)-L-cysteine; CEMA, N-acetyl-S-(2-carboxyethyl)-L-cysteine; CI, confidence interval; CYMA, N-acetyl-S-(2-cyanoethyl)-L-cysteine; DHBMA, N-acetyl-S-(3,4-dihydroxybutyl)-L-cysteine; GAMA, N-acetyl-S-(2-carbamoyl-2-hydroxyethyl)-L-cysteine; HEMA, N-acetyl-S-(2-hydroxyethyl)-L-cysteine; HPMMA, N-acetyl-S-(3-hydroxypropyl-1-methyl)-L-cysteine; IQR, interquartile ratio; MDI, mental development index; MU, trans,trans-muconic acid; PDI, psychomotor development index; SG, specific gravity; TGA, thiodiglycolic acid; TTCA, 2-thioxothiazolidine-4-carboxylic acid.

The interactions between child sex and ATCA ($p_{\text{sex-int}}<0.01$)/CEMA ($p_{\text{sex-int}}<0.05$)/TGA ($p_{\text{sex-int}}<0.05$) during all of the three trimesters and that between child sex and TTCA ($p_{\text{sex-int}}<0.05$) at the second trimester on MDI score were significant (Table S5). The interactions between the timing of exposure and ATCA ($p_{\text{time-int}}<0.01$)/CEMA ($p_{\text{time-int}}<0.05$)/TTCA ($p_{\text{time-int}}<0.05$) among the males and those between the timing of exposure and HPMMA among all the children ($p_{\text{time-int}}<0.05$) and among females ($p_{\text{time-int}}<0.05$) on the MDI score were significant (Table S5).

In the sensitivity analyses, most significant associations between higher urinary concentrations of the VOC metabolites and lower Bayley scores remained, especially that between ATCA and males’ MDI score (Excel Tables S2–S6). Additionally, some significant associations were found between the urinary concentrations of the VOC metabolites or the corresponding parent compounds (e.g., AAMA, $\mathrm{\Sigma }\text{Acrylamide}$, BPMA, and CEMA) and higher females’ Bayley scores in the main analyses and sensitivity analyses. After FDR correction, no significantly positive associations between the urinary concentrations of the VOC metabolites or the parent compounds and the females’ Bayley scores remained (Tables S4 and S5).

Associations between Child Neurodevelopment and VOC Metabolite Mixture

According to the results of the GLM and GEE models, males might be the vulnerable population to VOC exposures, and the first trimester appeared to be a sensitive window of VOC exposures. The WQS regression and BKMR analyses were further performed to assess the overall effect of the VOC metabolite mixture based on the concentrations at the first trimester and the average concentrations during pregnancy on the males’ MDI scores.

For the first trimester, each quartile increase in the VOC metabolite mixture was significantly associated with a decrease of 2.72 points (95% CI: $–4.99$, $–0.45$) in the males’ MDI score, and ATCA contributed the most [$\text{weight\hspace{0.17em}}\left(\mathrm{w}\right)=42\%$] in the overall association, followed by TGA (8%), TTCA (8%), and HPMMA (7%) (Figure 3; Table S6). For the average concentrations during pregnancy, each quartile increase in the VOC metabolite mixture was significantly associated with a decrease of 2.75 points (95% CI: $–4.85$, $–0.66$) in the males’ MDI score, and ATCA also contributed the most [$\text{weight\hspace{0.17em}}\left(\mathrm{w}\right)=20\%$] in the overall association, followed by TGA (12%), BPMA (11%), trans,trans-muconic acid (MU) (8%), and HPMMA (7%) (Figure 3; Table S6).

Figure 3.

WQS regression models with repeated holdout validation evaluating associations (estimates and 95% CIs) of males’ MDI scores and multiple SG-adjusted maternal urinary metabolites of volatile organic compounds based on concentrations (ng/mL) at the first trimester ($N=541$ participants; $n=541$ samples) and averaged concentrations (ng/mL) across three trimesters during pregnancy ($N=541$ participants; $n=\mathrm{1,623}$ samples) for participants from the Wuhan Healthy Baby Cohort in Wuhan, China, 2014–2017. Details of the chemical abbreviations are provided in Table 1. Two separate WQS indices were generated and constrained in the negative direction. The models were adjusted for maternal prepregnancy body mass index, age, socioeconomic status, passive smoking during pregnancy, folic acid supplementation during pregnancy, parity, delivery mode, breastfeeding status, and neurodevelopment assessment season. The numerical data are listed in Table S6. Note: 2HPMA, N-acetyl-S-(2-hydroxypropyl)-L-cysteine; 3HPMA, N-acetyl-S-(3-hydroxypropyl)-L-cysteine; 2MHA, 2-methylhippuric acid; $3\text{MHA\hspace{0.17em}}+4\text{MHA}$, 3-methylhippuric acid plus 4-methylhippuric acid; AAMA, N-acetyl-S-(2-carbamoylethyl)-L-cysteine; AMCC, N-acetyl-S-(N-methylcarbamoyl)-L-cysteine; ATCA, 2-aminothiazoline-4-carboxylic acid; BMA, N-acetyl-S-(benzyl)-L-cysteine; BPMA, N-acetyl-S-(n-propyl)-L-cysteine; CEMA, N-acetyl-S-(2-carboxyethyl)-L-cysteine; CI, confidence interval; CYMA, N-acetyl-S-(2-cyanoethyl)-L-cysteine; DHBMA, N-acetyl-S-(3,4-dihydroxybutyl)-L-cysteine; GAMA, N-acetyl-S-(2-carbamoyl-2-hydroxyethyl)-L-cysteine; HEMA, N-acetyl-S-(2-hydroxyethyl)-L-cysteine; HPMMA, N-acetyl-S-(3-hydroxypropyl-1-methyl)-L-cysteine; MDI, mental development index; MU, trans,trans-muconic acid; SG, specific gravity; TGA, thiodiglycolic acid; TTCA, 2-thioxothiazolidine-4-carboxylic acid; VOC, volatile organic compound; WQS, weighted quantile sum.

Distribution of the scaled concentrations of ${\log }_2\text{-transformed}$ VOC metabolites included in BKMR models is presented in Excel Table S7. For the first trimester, BKMR analyses showed significantly negative overall effect of the VOC metabolite mixture on the males’ MDI scores (Figure 4A; Table S7), with the PIP of ATCA higher than 0.5 ($\text{PIP}=0.86$) (Excel Table S8). Higher urinary ATCA concentration (75th percentile vs. 25th percentile) was significantly associated with lower males’ MDI score when all of the remaining VOC metabolites were fixed at their 25th, 50th, or 75th percentiles (Figure S6A; Excel Table S9). Figure S7 (Excel Table S10) shows a negative and approximately linear relationship between ATCA and the males’ MDI score when the other VOC metabolites were fixed at their median value. No interactions were observed among the VOC metabolites (Figure S8; Excel Table S11). For the average concentrations during pregnancy, results were generally similar to those at the first trimester (Figure 4B; Figures S6B, S9, and S10; Table S8; Excel Tables S8, S9, S12, and S13).

Figure 4.

BKMR models evaluating overall associations of males’ MDI score and maternal urinary concentrations of multiple VOC metabolites at the first trimester ($N=541$ participants; $n=541$ samples) (A) and averaged concentrations across three trimesters during pregnancy ($N=541$ participants; $n=\mathrm{1,623}$ samples) (B) for participants from Wuhan Healthy Baby Cohort in Wuhan, China, 2014–2017. The plot depicted the estimated changes and 95% CIs in males’ MDI in relation to all of the VOC metabolites at different percentiles compared with holding all the VOC metabolites at their 50th percentiles. The models were adjusted for maternal prepregnancy body mass index, age, socioeconomic status, passive smoking during pregnancy, folic acid supplementation during pregnancy, parity, delivery mode, breastfeeding status, and neurodevelopment assessment season. The numerical data are listed in Table S7. Note: BKMR, Bayesian kernel machine regression; CI, confidence interval; MDI, mental development index; VOC, volatile organic compound.

Discussion

This study first examined the associations between child neurocognitive development and repeatedly measured maternal urinary concentrations of multiple VOC metabolites during pregnancy. The average HQs of cyanide, 1,3-butadiene, and acrolein during pregnancy in more than 90% women exceeded the recommended safety thresholds. Certain VOC metabolites and the VOC metabolite mixture were significantly associated with lower child MDI score, respectively. The strongest VOC metabolite that contributed to the association was ATCA among the VOC metabolite mixture. Males were more vulnerable to VOC exposures compared with females, and the significant associations appeared to be more pronounced at the first trimester.

We observed relatively high urinary concentrations of HPMMA, 3HPMA, TGA, DHBMA, and ATCA (a specific metabolite of cyanide) during pregnancy, which were similar to those observed by Boyle et al.²⁹ in urine samples (collected at the third trimester) of 488 pregnant women from seven locations of the United States. However, the concentrations of most VOC metabolites in this study were approximately one- to five-fold lower than those in the study of Boyle et al.,²⁹ while HPMMA was approximately two-fold higher in this study. The higher emissions of vehicular exhaust in regions with higher economic levels may lead to higher concentrations of certain VOCs in the ambient air,^70–72 which may partially explain the higher concentrations of most urinary VOC metabolites among pregnant women in the United States compared with those in China. The higher concentrations of HPMMA in pregnant women in this study may be related to the Chinese cooking style with oils heated at high temperature, which is a significant source of human exposure to crotonaldehyde (parent compound of HPMMA) in food.^4,73 Additionally, the three isomers of HPMMA [N-acetyl-S-(3-hydroxypropyl-1-methyl)-L-cysteine, N-acetyl-S-(3-hydroxypropyl-2-methyl)-L-cysteine, and N-acetyl-S-(3-hydroxypropyl-3-methyl)-L-cysteine, which might derive from different parent compounds]⁷⁴ were not well separated in the present study, which may result in the higher concentrations of HPMMA among the participants in this study compared with those among the population in the United States.

Several other studies from China also reported similar profiles of the top urinary VOC metabolites in adults from Wuhan,²⁷ children aged 0–7 years from Wuhan and Shenzhen,⁴⁵ and children aged 6–12 years from Guangzhou.³² Overall, urinary concentrations of the VOC metabolites among the pregnant women in this study were similar to those among children in Wuhan, Shenzhen, and Guangzhou, but one to four times lower than those among adults (including males and smokers) in Wuhan. Different lifestyles and exposure sources among different populations may explain these discrepancies. Compared with the adults in the study reported by Qian et al.,²⁷ pregnant women and children may be less exposed to tobacco smoke (geometric mean concentration of urinary cotinine in the pregnant women vs. children⁴⁵ vs. adults²⁷ in Wuhan: 3.78 vs. 1.11 vs. $8.62\text{\hspace{0.17em}ng}/\mathrm{mL}$), which contains higher concentrations of many VOCs, and previous studies have reported associations between elevated urinary concentrations of the VOC metabolites and tobacco smoke exposure.⁷

Health risk assessment in this study implied high risk of cyanide, 1,3-butadiene, and acrolein exposure among the pregnant women, which were similar to the results reported in our previous study among the general population in Wuhan.²⁷ However, the variability of the VOC exposure across the three trimesters, e.g., the ICCs of the VOC metabolites during pregnancy ranged from 0.04 to 0.19, may cause uncertainty to the estimation of the HQs. The increase in urinary volume⁷⁵ and glomerular filtration rate throughout the pregnancy because of the increase in plasma volume⁷⁶ also added uncertainty to the estimation of the HQs based on urinary levels of the VOC metabolites. Future studies about additional health risk assessments based on external exposure of the VOCs could help us to realize how to reduce the exposure of the high-priority hazardous VOCs.

As far as we know, previous human data on maternal urinary VOC metabolites and childhood neurocognitive development are limited. A study from Korea²³ measured expectant mother’s personal airborne total VOC (TVOC) exposure for 3 days during the second trimester using passive samplers and investigated its relationship with children’s neurobehavioral development at 6, 12, 24, and 36 months ($N=383$). They observed lower mean PDI and MDI scores in the high TVOC exposure group compared with those in the low-exposure group.²³ Their findings and ours both suggested the potential impact of in utero simultaneous exposure to multiple VOCs on child neurodevelopment.

Higher maternal urinary concentrations of ATCA were found to be significantly associated with lower MDI scores of two-year-old children in this study after the FDR correction. We found ATCA concentrations contributed most to the negative association between the maternal VOC metabolite mixture and child MDI score based on both the first trimester and averaged concentration during pregnancy, suggesting potential neurodevelopmental toxicity from exposure to cyanide, the parent compound of ATCA. Cyanide can be found in cigarette smoke, vehicle exhaust, and released into the environment during the course of industrial usage⁷⁷ or converted from cyanogenic glycosides contained in almonds, linseed, lima beans, cassava, grains, bamboo shoots, soybeans, cherries, biscuits, macaroons, juice, nectar, pastries, and cakes (after grinding or chewing).⁷⁸ We found that the urinary ATCA concentrations (averaged) during pregnancy among the pregnant women with passive smoking exposure were significantly higher than those among the pregnant women without passive smoking exposure (median concentration: 130 vs. $118\text{\hspace{0.17em}ng}/\mathrm{mL}$), implying that cigarette smoke could be one of the exposure sources of cyanide in the study population. Other natural or processed foodstuffs containing cyanogenic glycosides are possible exposure sources of the study population and future work with detailed dietary surveys during pregnancy can help to identify major sources of cyanide exposure.

Laboratory studies have shown that exposure to cyanide can cause neurotoxicity, the potential mechanisms of which include inhibiting cellular oxygen consumption,^79,80 interfering with thyroid hormone levels,⁸¹ disrupting neurotransmitter release,^82–84 and inducing neuron apoptosis.^85,86 A few cross-sectional studies from the Democratic Republic of Congo revealed that dietary cyanide exposure⁸⁷ or urinary thiocyanate concentration⁸⁸ was adversely associated with early childhood neurodevelopment, such as cognition and motor development. Our findings add epidemiological evidence supporting the potential neurotoxicity of prenatal exposure to cyanide at environmentally relevant doses.

Furthermore, we found relatively high effect estimates for the associations of average urinary concentrations of TGA, CEMA, HPMMA, and TTCA during pregnancy with males’ MDI scores, suggesting the potential neurodevelopmental toxicity of exposure to the corresponding parent compounds of these metabolites. TGA is a shared metabolite of 1,2-dichloroethane, acrylonitrile, vinyl chloride,²⁷ dichloroacetic acid,⁸⁹ and trichloroacetic acid.⁸⁹ Previous experimental studies reported neurotoxicity of exposure to 1,2-dichloroethane⁹⁰ and acrylonitrile.⁹¹ Long-term exposure to 1,2-dichloroethane has been shown to reduce learning and memory abilities in mice and impair the structure and morphology of neurons in the hippocampal region.⁹⁰ The neurotoxicity induced by acrylonitrile may be explained by the redox imbalance in astrocytes,⁹² and the decreased content of gamma-aminobutyric acid and the inhibition of the glutamate decarboxylase activity in brain tissue.⁹³ Vinyl chloride is considered to be possibly neurotoxic to humans based on a previous literature review.⁹⁴ Dichloroacetic acid and trichloroacetic acid are disinfection by-products formed in drinking water, which have also been associated with neurotoxicity in experimental studies.^95,96 CEMA, HPMMA, and TTCA are the metabolites of acrolein, crotonaldehyde, and carbon-disulfide, respectively. Chronic exposure to acrolein, crotonaldehyde, and carbon-disulfide had been reported to cause neurotoxicity in animal studies^97–99; and the potential mechanisms may involve interference with neurotransmitter release,^99,100 induction of oxidative damage and inflammation in brain tissue,⁹⁸ disruption of neurofilament homeostasis, and activation of calpains.⁹⁷

In addition to most of the associations (corresponding to individual VOC metabolite analyses) observed at the first trimester being more pronounced compared with those in other trimesters, we also found relatively high effect estimates of the association between urinary BPMA (a metabolite of 1-bromopropane) concentrations at the second trimester and males’ MDI scores. In utero exposure to 1-bromopropane has been reported to cause adverse effects on the hippocampus neuronal excitability of rat offspring.¹⁰¹ The mid fetal period of gestation is the key developmental stage for maturation of excitatory neurons in the hippocampus,¹⁰² which may explain the more pronounced effect of exposure to 1-bromopropane in the second trimester observed in this study. To the best of our knowledge, this is the first epidemiological study to document the developmental neurotoxic effects of prenatal exposure to the aforementioned VOCs.

We found that the associations between maternal urinary concentrations of the VOC metabolites and impaired child neurodevelopment were more pronounced among males compared with females. A previous study from Japan examined the associations between indoor air pollutant exposure (including VOCs by analyzing indoor air samples) during childhood and neurodevelopment and also found significant associations between higher levels of air pollutant mixture (VOCs contributed the most) and poorer 3-year-old child neurodevelopment only among males,¹² suggested that males might be more vulnerable to VOC exposures compared with females. Previous epidemiological studies have found that the male brain is more vulnerable to many toxic exposures than the female brain, which may be explained by the greater glutathione availability and sulfate-based detoxification capacity in females, stronger protective effect of female fetus placenta (in response to maternal perturbations, through expressing more genes and proteins involved in transport, immune regulation, growth, and development than the male fetus placenta), and better neuroprotective/neuroreparative properties of female hormones (such as estrogen and progesterone, which may help to enhance cell proliferation and reduce inflammation and oxidative stress).^103,104

Neurodevelopment spans from the embryonic period to adolescence.¹⁰⁵ Compared to adults, the developing central nervous system is much more vulnerable to adverse environmental exposure.^106,107 Identifying critical windows of the exposure is crucial for children’s health.³⁸ In this study, the first trimester appeared to be the sensitive window of exposure to the VOCs (cyanide contributed the most), which may be related to the toxic mechanisms of cyanide and the temporal emergence of critical neruodevelopmental processes.¹⁰⁵ The neurogenesis, proliferation, migration, and differentiation of brain cells begin in early gestation,¹⁰⁵ during which the immature brain is highly dependent on energy.¹⁰⁸ Thus, exposure to cyanide can inhibit cell oxygen use⁸⁰ at the first trimester, which may result in more pronounced neurotoxicity. Furthermore, exposure to cyanide may be likely to produce neurotoxicity at the first trimester through interfering with the thyroid hormone levels,⁸¹ which are crucial for neurodevelopment and largely depend on maternal transformation and cannot be produced by the fetus until mid-pregnancy.¹⁰⁹ Future studies are needed to verify our findings and clarify the specific mechanisms.

The present study was based on a prospective birth cohort and repeated measurements of the VOC metabolite concentrations. These allowed us to explore the potential causal associations between the VOC metabolite concentrations and child neurocognitive development and to identify vulnerable windows of the exposures. Additionally, the use of WQS regression and BKMR analyses helped to assess the joint effect of the VOC metabolites on neurodevelopment and characterize each VOC metabolite’s contribution. Finally, we conducted risk assessment based on biomonitoring data, which provided evidence to recognize high-priority toxicants for management in the studied pregnant women.

The major limitation of this study was that we were unable to identify the major exposure sources of the VOCs (only based on the measurement of urinary VOC metabolites), which is important for susceptible populations to reduce their exposure. Cyanides mentioned in the discussion should be interpreted cautiously since they could be originated from cyanogenic glycosides contained in certain foodstuffs rather than the free form of cyanides in air or smoke. Further studies are needed to identify dietary preferences linked to cyanide exposure in pregnant women. Additionally, future studies combining biomonitoring data and exogenous exposure data may assist in figuring out the major exposure sources of the neurotoxic parent compounds.²⁹ Although we adjusted for the selected covariates in the statistical models and conducted sensitivity analyses to increase the robustness of the results, potential residual confounding factors (e.g., other neurotoxicants, such as mercury, and maternal intelligence quotient, information on maternal psychosocial factors such as maternal social support during pregnancy, maternal social assistance and depression postdelivery, and Home Observation for Measurement of the Environment in childhood) still existed. Future studies considering more comprehensive confounding factors are needed to validate the findings in this study. In addition, we only investigated the associations between prenatal urinary VOC metabolite concentrations and neurodevelopment of children aged two, and follow-up studies are warranted to address whether such associations persist in later life.

Conclusions

To our knowledge, this is the first study to observe associations between repeatedly measured maternal urinary concentrations of multiple VOC metabolites and child neurodevelopment. Risk assessment based on the RfD values suggested that cyanide, 1,3-butadiene, and acrolein should be considered as high-priority hazardous VOCs for risk management. We found significantly negative association of the VOC metabolite mixture with mental development, and ATCA (a metabolite of cyanide) contributed the most. In our study, males were more vulnerable to the VOC exposures compared with females, and the first trimester appears to be a critical window of susceptibility to the VOC exposures on neurodevelopment. Our findings suggested that prenatal exposure to certain VOCs, such as cyanide, may have adverse impacts on child neurodevelopment. Further studies are needed to confirm the association of real-world exposures with neurodevelopment and clarify the underlying mechanisms. Until then, it might be prudent to take measures to reduce the corresponding exposures. Integration of epidemiological investigations and major exposure source identification could assist to make favorable policies or interventions in the future.

Conclusions and opinions are those of the individual authors and do not necessarily reflect the policies or views of EHP Publishing or the National Institute of Environmental Health Sciences.