Abstract
Prenatal exposure to metals has been associated with impaired
neurodevelopment in children, but the detailed molecular mechanisms
remain largely unknown. Based on the Wuhan Healthy Baby Cohort, China
(N = 1088), eleven metals were measured in maternal urine during early
pregnancy (13.1 ± 1.1 weeks) and metabolomics profiling was conducted
in cord blood. Neurodevelopment was evaluated using the Bayley Scales
of Infant Development in 2-year-old children to obtain the mental
development index (MDI) and psychomotor development index (PDI). After
false discovery rate correction, higher maternal urinary levels of
manganese, nickel, aluminum, rubidium, gallium, and the summary score
of metals were only significantly associated with lower MDI scores. The
weighted quantile sum index of the metal mixture showed a significant
inverse association with MDI and PDI scores, with aluminum contributing
the most to the associations. Histidine, beta-alanine, purine, and
pyrimidine metabolism significantly mediated the above associations,
suggesting that disturbances in amino acids, neurotransmitter and
neuroendocrine metabolism may be important mediators in contributing to
impaired neurodevelopment of children.
Subject terms: Preventive medicine, Risk factors, Epidemiology
__________________________________________________________________
Prenatal exposure to metals has been linked to impaired
neurodevelopment in children, but the underlying mechanisms are
unclear. Here, authors show that disturbances in amino acids,
neurotransmitters, and neuroendocrine may mediate the association.
Introduction
The developing brain is particularly vulnerable to environmental
pollutants, and even low exposure levels can induce adverse
neurodevelopmental outcomes^[38]1–[39]4. Maternal exposure to metals,
such as lead (Pb) and cadmium (Cd), has been linked to adverse
neurodevelopmental outcomes in children, including impaired cognitive
development and behavior development^[40]2,[41]5,[42]6, as well as an
elevated risk of autism spectrum disorder (ASD) and attention deficit
hyperactivity disorder (ADHD)^[43]7,[44]8. However, previous studies
have primarily focused on single toxic metals (e.g. Pb, Cd) or mixtures
of several metals, and the effects of mixed metal exposure remain
inconclusive. Moreover, the biological mechanisms linking prenatal
exposure to mixed metals and children’s neurodevelopment have not yet
been fully established. Previous studies have suggested that
disruptions of homeostatic processes and bioenergetic disturbances
caused by environmental exposures in utero, such as increased oxidative
stress and lipid oxidation by high maternal levels of aluminum (Al),
Cd, Pb, and manganese (Mn)^[45]9,[46]10, disturbances of
neurotransmission by Al and Mn^[47]11,[48]12, mitochondrial dysfunction
induced by Pb and Mn^[49]13,[50]14, which may be underlying mechanisms
leading to adverse neurodevelopmental outcomes^[51]15,[52]16. However,
the majority of these findings are based on several traditional
biomarkers, including 8-hydroxy-2’-deoxyguanosine, 8-hydroxyguanosine,
malondialdehyde, and 4-hydroxy-2-nonenal-mercapturic
acid^[53]17,[54]18, which provide limited insight into complex
biological processes.
Metabolomics, as a high-throughput analytical approach to characterize
biological perturbations, has been extensively employed to investigate
the influence of environmental exposure on health to enhance our
comprehension of pathogenesis^[55]19–[56]21. Two previous studies with
a small sample size examined the effects of metal exposure during
pregnancy on maternal metabolism and then analyzed the biomarkers or
metabolic pathways associated with impaired neurodevelopment in
children^[57]22,[58]23. However, the mediating roles of metabolites or
metabolic pathways have not been quantified to explore the causal
pathways for the exposure-outcome associations. Metabolomics combined
with a meet-in-the-middle (MITM) method and mediation analysis with
metabolites and metabolic pathways may facilitate the characterization
of the molecular signatures perturbed by metal exposure in utero and
understand the pathogenesis of adverse outcomes in early
life^[59]24–[60]26. In particular, the aforementioned methods can
elucidate not only the impact of environmental exposure on individual
metabolites that are associated with the outcome but also the role of
the metabolic pathway, thereby providing a more comprehensive
understanding of the perturbed biological processes. To the best of our
knowledge, no studies have employed this approach to elucidate the
potential mechanisms of how prenatal metal exposure affects the cord
blood metabolome, and how these responses are linked to children’s
neurodevelopment.
Based on a prospective birth cohort in Wuhan, China, we measured the
levels of eleven metals in maternal urine from the first trimester,
analyzed metabolomics in cord blood, and evaluated the neurodevelopment
of 2-year-old children (N = 1088 mother-child pairs). The Bayley Scales
of Infant Development was used to evaluate the children’s
neurodevelopment because it is an extensive formal assessment tool that
evaluates cognitive and psychomotor development in infants and young
children to obtain the motor development index (MDI) and psychomotor
development index (PDI)^[61]27,[62]28. This study aimed to investigate
the associations between prenatal exposure to single and mixed metals
and neurocognitive development, and to identify the molecular
perturbation and underlying mechanism by applying a comprehensive
metabolomics workflow.
Results
Participant information
As illustrated in Table [63]1, the median age of mothers at delivery
was 28 years. Among the 1088 mothers, 89.0% had folic acid
supplementation during pregnancy, 82.0% had a bachelor’s degree or
above, 81.0% were primiparous, and 52.4% had an annual household income
<100,000 yuan. The mean birth weight of the children was 3345 g, of
which 54.1% were boys. The prevalence of low birth weight and preterm
birth was 2.4% and 2.5%, respectively. At 2 years of age, the median
(25th, 75th percentile) of MDI and PDI scores of the children were 114
(95, 126) and 108 (96, 122), respectively. About 8.3% and 2.9% had
moderate to severe cognitive development delay and psychomotor
development delay, respectively.
Table 1.
Basic characteristics of the study population from a birth cohort study
in Wuhan, China (N = 1088 mother-child pairs)
Total
Maternal characteristics
Age at delivery, years 28 (27, 31)
Pre-pregnancy BMI, kg/m² 20.7 (19.1, 22.8)
Folic acid supplementation during pregnancy
No 120 (11.0%)
Yes 968 (89.0%)
Maternal education level
High school or equivalent and below (≤12 years) 196 (18.0%)
Bachelor’s degree and above (>12 years) 892 (82.0%)
Annual household income (yuan)
<100,000 570 (52.4%)
≥100,000 518 (47.6%)
Parity
Nulliparous 881 (81.0%)
Multiparous 207 (19.0%)
Neonatal characteristics
Infant sex
Female 499 (45.9%)
Male 589 (54.1%)
Gestational age, weeks 39.4 (38.7, 40.1)
Birth weight, g 3345 (3070, 3650)
Low birth weight
No 1062 (97.6%)
Yes 26 (2.4%)
Preterm delivery
No 1061 (97.5%)
Yes 27 (2.5%)
MDI 114 (95, 126)
PDI 108 (96, 122)
Cognitive development delay (MDI < 70)
No 998 (91.7%)
Yes 90 (8.3%)
Psychomotor development delay (PDI < 70)
No 1056 (97.1%)
Yes 32 (2.9%)
[64]Open in a new tab
Continuous variables are median (25th, 75th); categorical variables are
n (%).
MDI mental development index, PDI psychomotor development index.
Almost all metals were detected in over 91% of maternal urine samples,
but the detection rate of gallium (Ga) was 72% (Table [65]S1). Rubidium
(Rb) had the highest median concentration (1653.04 μg/g creatinine),
followed by arsenic (As) and Al (20.38 and 20.34 μg/g creatinine,
respectively). Thallium (Tl), Cd, vanadium (V), chromium (Cr), Mn,
nickel (Ni), and Pb had median creatinine-corrected concentrations
ranging from 0.37 to 2.49 μg/g creatinine, while Ga had the lowest
median concentration at 0.03 μg/g creatinine. Spearman correlation
coefficients between the concentrations of these eleven metals in
maternal urine ranged from 0.25 to 0.77 (all P < 0.05) (Fig. [66]S1).
Associations between maternal urinary concentrations of metals and children’s
MDI/PDI scores
Linear regression models showed significant inverse associations
between multiple metals and MDI scores (Fig. [67]1A). Specifically,
after adjusting for covariates, each interquartile range (IQR) increase
in maternal urinary concentration of Cr, Mn, Pb, Ni, Tl, Al, Rb and Ga
during early pregnancy was significantly linked to a decrease of 1.38
points (95% CI: −2.73, −0.04), 2.48 points (95% CI: −3.87, −1.08), 1.57
points (95% CI: −3.01, −0.12), 1.27 points (95% CI: −2.26, −0.27), 1.71
points (95% CI: −3.38, −0.04), 1.97 points (95% CI: −3.23, −0.71), 1.77
points (95% CI: −3.26, −0.27), and 2.42 points (95% CI: −4.11, −0.73)
in MDI scores, respectively. After multiple testing corrections, the
associations of Mn, Ni, Al, Rb, and Ga with lower MDI scores remained
significant (P[FDR] < 0.05). Each IQR increase in the summary score of
metals was significantly associated with a decrease of 2.68 points (95%
CI: −4.27, −1.10, P[FDR] = 0.0056) in MDI score (Fig. [68]1A). By using
the weighted quantile sum (WQS) analysis, each quartile increase in the
WQS index of the metal mixture was associated with a decrease of 3.47
points in MDI score (95% CI: −5.00, −1.95, P < 0.001), and Al
contributed the most to the mixture effect (49.2%), followed by Ga
(21.6%), Mn (11.8%), Cd (7.5%), and Cr (3.7%) (Fig. [69]2A).
Fig. 1. Associations between maternal urinary metals and neurodevelopment of
2-year-old children (N = 1088).
[70]Fig. 1
[71]Open in a new tab
A Linear regression analysis of the associations between maternal
urinary metals and MDI scores of children (Per interquartile range
increase in the creatinine-adjusted urinary metal concentration (μg/g
creatinine)). B Linear regression analysis of the associations between
maternal urinary metals and PDI scores of children. The x-axis shows
the estimated coefficients of the associations between maternal urinary
metals and MDI or PDI scores. Data are presented as point estimates
(linear regression beta coefficients) and 95% confidence intervals
(beta coefficient * ± 1.96 (standard error[beta coefficient])).
Statistical tests were performed with two-sided linear regressions.
Adjustments for multiple comparisons were accounted for by calculating
P[FDR]. Estimated coefficients were adjusted for pre-pregnancy BMI,
maternal age, parity, folic acid supplementation, maternal education
level, infant sex, and cotinine levels in cord blood. Bold indicates
statistical significance (P < 0.05), and “*” indicates statistical
significance (P[FDR] < 0.05). Source data are provided as a Source Data
file. MDI mental development index, PDI psychomotor development index,
CI confidence interval, Cr chromium, Mn manganese, Pb lead, Ni nickel,
Cd cadmium, Tl thallium, V vanadium, Al aluminum, Rb rubidium, Ga
gallium, As arsenic.
Fig. 2. WQS regression analysis of the associations between maternal urinary
metals and neurodevelopment of 2-year-old children (N = 1088).
[72]Fig. 2
[73]Open in a new tab
A WQS regression analysis of the associations between maternal urinary
metals and MDI scores (Per quantile range increase in WQS index). B WQS
regression analysis of the associations between maternal urinary metals
and PDI scores. Data are presented as point estimates (linear
regression beta coefficients) and 95% confidence intervals (beta
coefficient * ± 1.96 (standard error[beta coefficient])). Statistical
tests were performed with two-sided WQS regressions. Estimated
coefficients were adjusted for pre-pregnancy BMI, maternal age, parity,
folic acid supplementation, maternal education level, infant sex, and
cotinine levels in cord blood. Source data are provided as a Source
Data file. WQS weighted quantile sum, MDI mental development index, PDI
psychomotor development index, CI confidence interval, Cr chromium, Mn
manganese, Pb lead, Ni nickel, Cd cadmium, Tl thallium, V vanadium, Al
aluminum, Rb rubidium, Ga gallium, As arsenic.
For PDI, each IQR increase in maternal urinary concentration of Pb, Rb,
and As was significantly associated with a decrease of 1.26 points (95%
CI: −2.50, −0.01), 1.34 points (95% CI: −2.63, −0.05), and 1.33 points
(95% CI: −2.60, −0.06) in PDI scores (Fig. [74]1B). However, none of
the associations were significant after false discovery rate (FDR)
correction (P[FDR] > 0.05). Each IQR increase in the summary score of
metals was significantly associated with a decrease of 1.38 points (95%
CI: −2.74, −0.01, P = 0.049, P[FDR] = 0.15) in PDI score (Fig. [75]1B).
A significant association was also observed between the WQS index of
metal mixture and PDI scores, with one quartile increase in the WQS
index being associated with a decrease of 1.88 points in PDI score (95%
CI: −3.32, −0.44, P = 0.011) (Fig. [76]2B). Al (30.9%) made the largest
contribution to the association, followed by Pb (22.6%), As (19.3%), Cd
(10.0%), Rb (9.1%).
In the sensitivity analyses that (1) excluding participants with
preterm birth and low birth weight, (2) excluding pregnant women who
did not take folic acid supplements, (3) excluding tobacco exposure
during pregnancy, or (4) adjusting for more covariates (gestation age
and birth weight), the results were not materially changed
(Table [77]S2). Sex-stratified analyses revealed associations between
some metals and Bayley scores were stronger among boys than girls, but
the interaction term between metal exposure and sex was not significant
(all P[sex-int] > 0.05, Tables [78]S3 and [79]S4), indicating that
child sex did not significantly modify the neurotoxic effect of metals.
Metabolome-wide association study (MWAS) on maternal urinary metals and
children’s MDI/PDI scores
Of the 438 metabolites detected, 11 metals (Cr, Mn, Pb, Ni, Cd, Tl, V,
Al, Rb, Ga, and As) were significantly associated with changes of 145,
113, 70, 28, 15, 26, 91, 104, 89, 79, and 16 metabolites in cord blood,
respectively, after adjusting for covariates (P < 0.05 and
P[FDR] < 0.2, Fig. [80]3A, Supplementary Data [81]1). Higher exposure
levels of Mn, Pb, Ni, V, Al, and Ga were mainly linked to the
upregulation of amino acids and their metabolites such as carnosine,
valine, hydroxylysine, asymmetric dimethylarginine, and targinine, as
well as the downregulation of glutamine. Elevated levels of Cr, Mn, Ni,
Tl, Al, and Rb were linked with the upregulation of riboflavin levels.
The increase in summary score of metal concentrations and WQS index of
metal mixture for MDI and PDI were significantly associated with 138,
136, and 115 metabolites, respectively, mainly related to the
upregulation of amino acids and their metabolites, phosphatidylcholines
(PCs), and phosphatidylethanolamines (PEs) (P < 0.05 and P[FDR] < 0.2,
Fig. [82]3A, Fig. [83]S2, Supplementary Data [84]1).
Fig. 3. Metabolome-wide association study (MWAS) on maternal urinary metals
and children’s MDI/PDI scores (N = 1088).
[85]Fig. 3
[86]Open in a new tab
A Volcano plot of the associations between maternal urinary metals and
metabolites in cord plasma samples, analyzed by linear regression
models. Statistical tests were performed with two-sided linear
regressions. The x-axis shows the estimated coefficient of the
association between maternal urinary metals and metabolites in cord
plasma samples, and the y-axis shows its value of −log[10] (P-value).
The significance P-value threshold of 0.05 is shown as a dashed line.
Adjustments for multiple comparisons were accounted for by calculating
P[FDR]. Exact P-values and P[FDR] are provided in Supplementary
Data [87]1. The adjusted covariates for analyzing the association
between maternal urinary metal exposure and metabolites included
pre-pregnancy BMI, maternal age, parity, folic acid supplementation,
maternal education level, infant sex, and cotinine levels in cord
blood. B Volcano plot of the association between metabolites in cord
plasma samples and MDI scores. C Volcano plot of the association
between metabolites in cord plasma samples and PDI scores. Statistical
tests were performed with two-sided linear regressions. The x-axis
shows the estimated coefficient of the association between metabolites
in cord plasma samples and MDI/PDI scores, and the y-axis shows its
value of −log[10] (P-value). The significance P-value threshold of 0.05
is shown as a dashed line. Adjustments for multiple comparisons were
accounted for by calculating P[FDR]. Exact P-values and P[FDR] are
provided in Supplemental Table [88]6. Different-colored dots represent
different types of metabolites. The adjusted covariates for analyzing
the association between metabolites and children’s MDI/PDI scores were
consistent with the above. D Overlapping metabolites in cord plasma
samples that were significantly associated with maternal urinary metal
and MDI. E Overlapping metabolites in cord plasma samples that were
significantly associated with maternal urinary metal and PDI. Source
data are provided as a Source Data file. MDI mental development index,
PDI psychomotor development index, Cer ceramide, PA phosphatidic acid,
PC phosphatidylcholine, PE phosphatidylethanolamine, PG
phosphatidylglycerol, PI phosphatidylinositol, SM sphingomyelin, Cr
chromium, Mn manganese, Pb lead, Ni nickel, Cd cadmium, Tl thallium, V
vanadium, Al Aluminum, Rb rubidium, Ga gallium, As arsenic.
In the MWAS on children’s MDI/PDI scores, there were 71 and 24
metabolites in cord blood significantly associated with MDI and PDI,
respectively (P < 0.05 and P[FDR] < 0.2, Fig. [89]3B, C, Table [90]S5).
Higher levels of carnosine, targinine, bilirubin,
lysophosphatidylethanolamine (LPE), lysophosphatidylcholine (LPC), and
PEs had significant associations with lower MDI scores. On the
contrary, the increase in levels of 10 metabolites, including lysine,
glutamine, leucine, alpha-aminobutyric acid, deoxycholic acid glycine
conjugate, riboflavin, 4-ethylphenylsulfate, fatty acid (FA) 12:0,
sphingomyelin (SM) 38:0, and SM 40:1, showed significant associations
with higher MDI scores. The increase in levels of 11 metabolites,
including 8 lysophospholipids, 1 amino acid, 1 FA, and 1
phosphatidylinositol (PI) was significantly linked to lower PDI scores.
A total of 13 metabolites exhibited a statistically significant
correlation with higher PDI scores.
Based on the MITM approach, 56 overlapping metabolites were found in
the MWAS on maternal urinary metals and the MWAS on children’s MDI
scores (P < 0.05 and P[FDR] < 0.2, Fig. [91]3D). The overlapping
metabolites profile consisted of 5 amino acids, 2 bile acids, 4
ceramides, 1 FA, 38 phospholipids, 2 SMs, 1 vitamin and 3 other
metabolites. There were 11 overlapping metabolites in the MWAS on both
maternal urinary metals and the MWAS on children’s PDI scores,
including 2 amino acids, 1 phosphatidic acid (PA), 3 PCs, 4 PIs, and 1
phosphatidylglycerol (PG) (Fig. [92]3E).
Overlapping enriched pathways associated with maternal urinary metals and
children’s MDI/PDI scores
Among the 45 mapped Kyoto Encyclopedia of Genes and Genomes (KEGG)
pathways, maternal urinary concentrations of Cr, Mn, Pb, Ni, Cd, Tl, V,
Al, Rb, Ga, and As were found to be significantly associated with 12,
26, 20, 21, 3, 12, 19, 23, 16, 15, and 14 metabolic pathways,
respectively (P < 0.05, Fig. [93]4A, Supplementary Data [94]2). A
hierarchical clustering analysis of the metabolic pathways revealed
that Mn, Ni, Pb, and Al clustered together, and the remaining metals
(V, Rb, Tl, Ga, As, Cr, and Cd) were grouped. Compared to most of the
single metals (except Mn, Ni, and Al), the summary score of metals, the
WQS index of metal mixture for MDI and PDI disturbed more metabolic
pathways, which were related to significant changes in 21, 26, and 22
pathways, respectively, including amino acid, lipid, and vitamin
metabolic pathways (P < 0.05, Supplementary Data [95]2).
Fig. 4. Metabolic pathways related to maternal urinary metals and
neurodevelopment of 2-year-old children (N = 1088).
[96]Fig. 4
[97]Open in a new tab
A Cluster heatmap of metabolic pathways for each maternal urinary
metal. Statistical tests were performed with two-sided tests, with
statistical significance set at 0.05. The color of the heatmap
represents −log[10](P-value) of metabolic pathways. B Venn diagrams of
significantly enriched KEGG pathway from one or more of the maternal
urinary metals, MDI, and PDI scores (P < 0.05). Statistical tests were
performed with two-sided tests, with statistical significance set at
0.05. Source data are provided as a Source Data file. MDI mental
development index, PDI psychomotor development index, Cr chromium, Mn
manganese, Pb lead, Ni nickel, Cd cadmium, Tl thallium, V vanadium, Al
aluminum, Rb rubidium, Ga gallium, As arsenic.
The MITM results indicated that 8 metabolic pathways (including lysine
degradation; valine, leucine, and isoleucine degradation; valine,
leucine, and isoleucine biosynthesis; histidine metabolism; riboflavin
metabolism; beta-alanine metabolism; purine metabolism, and pyrimidine
metabolism) in cord blood were associated with maternal urinary metals
and MDI scores (P < 0.05, Fig. [98]4B). Three pathways (including
lysine degradation; valine, leucine, and isoleucine degradation;
valine, leucine, and isoleucine biosynthesis) were linked to maternal
urinary metals and PDI scores (P < 0.05, Fig. [99]4B).
Mediating role of overlapping pathways and metabolites in the association
between maternal urinary metals and children’s MDI/PDI scores
We evaluated the potential mediation effect of the overlapping enriched
pathways and metabolites on the associations between maternal urinary
metals and children’s MDI/PDI scores. There were three pathway
mediators significantly associated with higher levels of metals and
lower children’s MDI scores, with the most frequently identified being
beta-alanine metabolism, followed by histidine metabolism, and
riboflavin metabolism. Beta-alanine and histidine metabolic pathways
had significant mediation effects on the association of seven metals
(Mn, Pb, Ni, V, Al, and Ga), with the mediation effects ranging from
10% to 26% (Table [100]2, Fig. [101]5). Among the metabolites detected
in the two metabolic pathways, only carnosine significantly mediated a
decrease in MDI associated with the six metals, with the mediation
effect of 7%, 15%, 15%, 11%, 10%, and 10% for Mn, Pb, V, Al, Rb and Ga,
respectively (Table [102]S6). Consistently, the pathways of
beta-alanine and histidine metabolism were found to significantly
mediate a 13% and 11% change in the association between the summary
score of metals and lower MDI, respectively. Additionally, these
pathways mediated a 9% and 8% change in the association between the WQS
index and lower MDI, respectively. In addition, glutamine, as a common
metabolite in purine and pyrimidine metabolic pathways, had significant
mediation effects on the association between metals (including Cr, Mn,
Pb, V, Al, Rb, Ga, the summary score and WQS index) and lower MDI
scores, with mediation effects ranging from 4% to 15% (Table [103]S6).
Table 2.
Mediation analyses of maternal urinary metal-related metabolic pathways
and metabolites on the association between maternal urinary metals and
MDI scores of children (N = 1088)
Exposure Pathway Indirect effects^a Proportion mediated Metabolites
Indirect effects^b Proportion mediated
Estimate (95% CI) P P[FDR] Estimate (95% CI) P P[FDR]
Cr Riboflavin metabolism 0.09 (0.01, 0.20) 0.005 0.030 −12% Riboflavin
0.07 (0.01, 0.17) 0.019 0.093 −9%
Mn beta-Alanine metabolism −0.12 (−0.22, −0.04) <0.0001 <0.0001 13%
Carnosine −0.07 (−0.14, −0.01) 0.005 0.130 7%
Histidine metabolism −0.11 (−0.20, −0.04) 0.001 0.003 11%
Riboflavin metabolism 0.05 (0.00, 0.12) 0.032 0.064 −5% Riboflavin 0.05
(0.00, 0.11) 0.036 0.130 −5%
Pb beta-Alanine metabolism −0.24 (−0.42, −0.09) <0.0001 <0.0001 23%
Carnosine −0.16 (−0.31, −0.04) 0.004 0.136 15%
Histidine metabolism −0.23 (−0.40, −0.08) <0.0001 <0.0001 21%
Ni beta-Alanine metabolism −0.09 (−0.17, −0.02) 0.011 0.033 12%
Carnosine −0.07 (−0.14, −0.01) 0.023 0.204 8%
Histidine metabolism −0.07 (−0.16, 0.00) 0.033 0.066 10%
Riboflavin metabolism 0.07 (0.01, 0.15) 0.006 0.033 −9% Riboflavin 0.06
(0.01, 0.14) 0.013 0.204 −8%
V beta-Alanine metabolism −0.24 (−0.44, −0.07) 0.002 0.012 26%
Carnosine −0.15 (−0.31, −0.02) 0.017 0.086 15%
Histidine metabolism −0.16 (−0.34, −0.01) 0.034 0.102 17%
Al beta-Alanine metabolism −0.10 (−0.18, −0.04) <0.0001 <0.0001 15%
Carnosine −0.07 (−0.14, −0.02) 0.005 0.114 11%
Histidine metabolism −0.09 (−0.16, −0.03) 0.001 0.003 14%
Riboflavin metabolism 0.06 (0.01, 0.12) 0.005 0.010 −8% Riboflavin 0.05
(0.00, 0.10) 0.014 0.118 −7%
Rb beta-Alanine metabolism −0.26 (−0.54, −0.04) 0.023 0.069 12%
Carnosine −0.21 (−0.46, −0.03) 0.019 0.157 10%
Riboflavin metabolism 0.25 (0.04, 0.52) 0.007 0.042 −11% Riboflavin
0.22 (0.02, 0.47) 0.012 0.157 −10%
Ga beta-Alanine metabolism −0.15 (−0.30, −0.03) 0.009 0.054 11%
Carnosine −0.14 (−0.28, −0.03) 0.006 0.090 10%
Histidine metabolism −0.13 (−0.28, −0.02) 0.021 0.063 10%
As beta-Alanine metabolism −0.19 (−0.41, −0.01) 0.037 0.189 14%
Carnosine −0.15 (−0.34, −0.01) 0.039 0.435 11%
Summary scores beta-Alanine metabolism −1.49 (−2.66, −0.53) <0.0001
<0.0001 13% Carnosine −1.05 (−2.10, −0.24) 0.007 0.090 9%
Histidine metabolism −1.26 (−2.36, −0.38) 0.001 0.003 11%
Riboflavin metabolism 0.83 (0.13, 1.79) 0.006 0.012 −7% Riboflavin 0.76
(0.09, 1.64) 0.013 0.090 −6%
WQS index beta-Alanine metabolism −0.33 (−0.59, −0.11) <0.0001 <0.0001
9% Carnosine −0.26 (−0.50, −0.05) 0.008 0.104 7%
Histidine metabolism −0.28 (−0.52, −0.08) 0.001 0.003 8%
Riboflavin metabolism 0.24 (0.06, 0.48) 0.002 0.004 −7% Riboflavin 0.20
(0.03, 0.42) 0.006 0.104 −6%
[104]Open in a new tab
Statistical tests were performed with two-sided tests, with statistical
significance set at 0.05. Adjustments for multiple comparisons were
accounted for by calculating PFDR. Adjusted covariates included
pre-pregnancy BMI, maternal age, parity, folic acid supplementation,
maternal education level, infant sex, and cotinine levels in cord
blood.
MDI mental development index, Cr chromium, Mn manganese, Pb lead, Ni
nickel, V vanadium, Al aluminum, Rb rubidium, Ga gallium, As arsenic.
^aIndirect effects of the pathway between maternal urinary metals and
neurodevelopment of 2-year-old children.
^bIndirect effects of metabolites between maternal urinary metals and
neurodevelopment of 2-year-old children.
Fig. 5. Interrelationship among maternal urinary metals, metabolic pathways
and metabolites in cord plasma samples, and neurodevelopment of 2-year-old
children (N = 1088).
[105]Fig. 5
[106]Open in a new tab
A Sankey plot was utilized to examine and visualize the correlation
between maternal urinary metal and neurocognitive development of
2-year-old children in the mediation of key cord plasma metabolic
pathways and metabolites. Source data are provided as a Source Data
file. MDI mental development index, PDI psychomotor development index,
Cr chromium, Mn manganese, Pb lead, Ni nickel, Cd cadmium, Tl thallium,
V vanadium, Al aluminum, Rb rubidium, Ga gallium, As arsenic.
Additionally, the riboflavin metabolic pathway mediated a −5% to −12%
change in the association between metals (including Cr, Mn, Ni, Al, Rb,
the summary score, and WQS index) and lower MDI scores. Among the three
metabolites detected in this pathway, only riboflavin was found to
significantly mediate the association between metals (including Cr, Mn,
Al, Rb, the summary score, and WQS index) exposure and lower MDI
scores, with a mediation effect of −5% to −10%.
In the mediation analysis of metals with PDI, we found a 10%–26%
mediation effect of the lysine degradation pathway on the association
between metals (Cr, Mn, Pb, Ni, Al, Ga, As; the summary score and WQS
index) and lower PDI scores (Table [107]S7, Fig. [108]5). Among the
three metabolites detected in this pathway, 5-hydroxylysine was the
only one found to significantly mediate the association between
exposure to these metals and lower PDI scores, with a mediation effect
of 17% to 42% (Table [109]S7).
In addition to the aforementioned metabolic metabolites, multiple
lysophospholipids (e.g., LPA 20:4, LPC O-16:0, LPC O-18:0, LPC O-18:1,
LPC O-24:2, LPE 22:4, and LPG 20:4) were significantly mediated the
associations between several metals (Cr, Mn, Pb, Ga, the summary score
and WQS index) and lower MDI scores, with the mediation effects ranging
from 4% to 19% (Table [110]S6). Furthermore, LPA 20:4, LPC O-18:1, and
LPG 20:4 also significantly mediated associations between those metals
and lower PDI scores, with the mediation effect values ranging from 8%
to 25% (Table [111]S8).
The results of the mediator analysis in the sensitivity analyses
mentioned above (excluding preterm birth, low birth weight, without
folic acid supplements, and tobacco exposure; adjusting for more
covariates) and sex-stratified analyses showed no material changes
(Tables [112]S9–[113]S12).
Discussion
Based on a prospective birth cohort, we conducted metabolomics analysis
of cord blood to explore the biologic pathways and metabolites that may
be involved in the relationship between maternal metal exposure during
early pregnancy and children’s neurodevelopment at 2 years of age. We
found higher levels of single and mixed metals were significantly
linked with reduced MDI scores in children. Beta-alanine and histidine
metabolic pathways played intermediate roles in the association between
these metals and lower MDI scores. Glutamine, a common metabolite of
purine and pyrimidine metabolic pathways, also had significant
mediation effects on these associations. Additionally, enhanced
riboflavin metabolism may partially mitigate the adverse effects on
cognitive development associated with metals.
Compared to the metal concentrations detected in pregnant women
worldwide, the median urinary metal concentrations (except V) observed
in our study were similar to those pregnant women from other regions in
China, such as Nanjing^[114]29,[115]30 and Taiwan^[116]31, and were
generally comparable to the level reported in Bangladesh^[117]32,
Mexico^[118]33 and Spain^[119]34 (Table [120]S13). However, the median
concentrations were approximately 1–2 times higher than those in
developed countries such as Greece^[121]35, United
States^[122]36–[123]40 and Australia^[124]41,[125]42. The median
urinary concentration of V was similar to the pregnant women from our
birth cohort reported previously^[126]43, but was about five times
higher than that reported in pregnant women from Nanjing,
China^[127]29, the United States^[128]38, and Australia^[129]42.
Consistent with previous epidemiological studies, we found that
maternal exposure to higher levels of Mn, Ni, Pb, and Al, which are
regarded as neurotoxic metals, were linked to lower MDI scores of
2-year-old children, indicating that prenatal exposure to these metals
can affect children’s cognitive development^[130]5,[131]44–[132]46. We
also found significant links between maternal exposure to Cr, Tl, Rb,
and Ga and lower MDI scores in children, which had been rarely
reported. However, we didn’t observe significant associations between
urinary metals and the categorical outcomes (PDI < 70 or MDI < 70)
after multiple corrections. The possible reasons may be that there were
not enough cases of neurodevelopmental delay in the study
population^[133]47,[134]48, and the metal exposure levels observed in
the present study could impair neurodevelopment but may not be
sufficient to significantly increase the risk of neurodevelopmental
delay. Another prospective cohort study conducted in Anhui, China,
which measured metals in maternal serum during pregnancy, also found
that higher levels of Tl was associated with poorer cognitive
development in children^[135]49. Nevertheless, it was also worth noting
that several researches found no associations between Cr, Tl, and Ga
exposure during pregnancy and impaired children’s cognitive
development^[136]50–[137]52. The inconsistent results may be
attributable to discrepancies in the study area, study population,
sample type, exposure levels, and unmeasured confounding factors.
Consistent with the existing epidemiological studies, the impact of
prenatal metal exposure on MDI was more pronounced compared to PDI,
indicating that the fetal nervous system may be more susceptible to
cognitive impairments caused by toxic exposures^[138]1,[139]53. This
differential impact may be attributable to the disruption of
neurotransmitter synthesis, brain cell development, oxidative stress
and lipid oxidation caused by metals (such as Mn and
Al)^[140]9,[141]10,[142]54, which are critical for cognitive
functions^[143]15,[144]55. Furthermore, a greater number of cord blood
metabolites and metabolic pathways were associated with MDI than PDI in
our study, further confirming that the impact of metabolic disturbances
in utero on cognitive development may be greater than on psychomotor
development.
In addition, to explore the cumulative effects of metal exposure on
neurodevelopment and verify the possible common metabolic pathways of
metal effects on neurodevelopment, we used a summary score to assess
the co-exposure of metals^[145]56, and found the estimate of a decrease
in Bayley score associated with the summary score was greater than that
of a single metal. The previous study used a summary score to assess
the mixture exposure levels of herbicides in adolescents, and also
found consistently worse performance on all five neurobehavioral
domains assessed^[146]56, suggesting that exposure to multiple
pollutants during pregnancy may exhibit synthetical effects^[147]1.
Furthermore, to minimize the potential for collinearity of the metals,
we employed a WQS model and found elevated metal mixture levels were
significantly associated with reduced Bayley scores. Al was identified
as the primary neurotoxic biomarker of concern. Similarly, in a
previous study that measured prenatal exposure levels of 20 metals in
703 mother-child pairs from Guangxi, China, maternal serum Al was
identified as the largest contributor to reduced fine motor and
adaption developmental quotients in children aged 2–3 years^[148]5. In
a prospective birth cohort from Bangladesh, Wei et al.^[149]52 also
observed that among 52 trace elements analyzed, cord serum Al was one
of the most significant contributors to decreased cognitive composite
scores in children aged 20–40 months. Furthermore, our study identified
a greater number of cord blood metabolites and metabolic pathways
associated with Al, providing further evidence that Al exposure during
pregnancy may lead to a range of metabolic disturbances.
Prenatal exposure to environmental pollutants can disrupt metabolic
pathways, rather than just affecting individual metabolites^[150]20.
Therefore, the metabolic pathway may be more helpful in understanding
the potential mechanism mediating the association between maternal
exposure levels of metals and children’s neurodevelopment^[151]57. We
found that all the mediation effects of pathways were higher than or
equal to those mediated solely by metabolites in the association
between maternal urinary metals and children’s MDI scores. This finding
may be attributed to the high correlation between metabolites involved
in the same pathway, as well as the similar strength and direction of
association between these metabolites and outcomes^[152]58. Mediator
dimension reduction allows all metabolites detected in a pathway to be
evaluated as a whole to assess the potential mediation pathway for
individual chemical^[153]57.
In the current research, we observed that the associations between
several metals (Mn, Pb, Ni, V, Al, Ga, and the mixed exposure) and
impaired cognitive development (decreased MDI scores) in children
shared the same two mediating metabolic pathways, including
beta-alanine and histidine metabolism, which are two important amino
acid metabolic pathways. They play a crucial role in fetal
neurodevelopment, particularly in terms of neurotransmitter synthesis,
myelin formation, and neuroprotection^[154]59. Insufficient histidine
intake can lower brain histamine levels (an important neurotransmitter)
of C57BL/6J male mice, leading to anxiety-like
behaviors^[155]60,[156]61. Previous case-control studies also showed
significant changes in urinary histidine metabolism in children with
ASD^[157]62, with significantly lower levels of urinary beta-alanine
and histidine compared to healthy controls^[158]63,[159]64. The
metabolism of beta-alanine and histidine are also important for the
synthesis of carnosine^[160]65, which serves as the main metabolite
mediating the decrease in MDI scores associated with Mn, Pb, V, Al, Rb,
Ga, and the mixed metals. Carnosine is a dipeptide that can act as a
neuroprotective factor due to its antioxidative properties and
chelating ability to divalent metal ions^[161]65–[162]69. Carnosine
metabolism disorder in childhood has been linked to mental deficiency
in children^[163]70,[164]71, while carnosine supplementation in
childhood may improve communication skills and behavior in children
with ASD^[165]72. Our results indicated that beta-alanine and histidine
metabolism as well as carnosine, which were perturbed by maternal
exposure to multiple metals during pregnancy, may be an important
metabolic molecular mechanism affecting children’s cognitive
development. This suggests that these pathways and metabolite may be a
common target for the wide-ranging effects of environmental factors.
Future studies are needed to elucidate the specific mechanisms and
explore their potential as a target for the control and prevention of
pollutants affecting cognitive development.
In addition, glutamine was observed to act as a co-mediator in the
association of multiple metals (including Cr, Mn, Pb, V, Al, Rb, Ga,
and mixed exposure) and cognitive developmental damage. Glutamine, as a
precursor of neurotransmitters, is directly involved in several
critical brain processes, such as energy metabolism, ammonia
homeostasis, and neurotransmitter cycling^[166]73,[167]74. The results
of our study indicated a positive association between glutamine levels
and MDI scores. Furthermore, the findings of several epidemiological
studies were in agreement with our results, which demonstrated that
plasma glutamine levels in children with ASD were significantly lower
than in controls^[168]75,[169]76. Glutamine is primarily synthesized
and released from astrocytes^[170]73. The target metals can transfer
across both the placental and blood–brain barriers, and they can
accumulate in astrocytes^[171]77, and impair their homeostatic
capabilities, posing a potential threat to fetal neurodevelopment
during the most sensitive early stages of life^[172]78. In vitro cell
experiments showed that Mn overexposure could dysregulate astrocytic
cycling of glutamine and glutamate by aberrant phosphate-activated
glutaminase pathway^[173]10,[174]79, and Al could decrease astrocytic
glutamate intake^[175]80, thereby inducing excitotoxicity and secondary
neuronal death. Perinatal low-dose Pb exposure in rats could cause
glutamate and glutamine dyshomoeostasis in microglia and astrocytes by
downregulating the protein level and activity of glutamine synthase,
leading to neuroinflammation and further neuropathological
changes^[176]81. For infant neurodevelopment, glutamine dyshomoeostasis
may play an important role in the mechanisms that underlie the
association between metal toxicity and decreased MDI scores in
children.
The negative mediation effects of the riboflavin metabolic pathway
suggested that Cr, Mn, Ni, Al, and Rb exposure may upregulate the
riboflavin metabolic pathway, partially offsetting the effects of
metals on reducing MDI scores, which was its mediating role between the
mixed metal exposure and impaired cognitive development. Riboflavin
metabolism is an important vitamin metabolic pathway, providing
important cofactors for energy metabolism, fatty acid oxidation, amino
acid and purine metabolism^[177]82,[178]83. A cohort study from Japan
(N = 1199 mother-child pairs) demonstrated that maternal intake of
riboflavin during pregnancy may serve as a protective factor against
childhood emotional problems in children aged five years^[179]84.
Metals such as Mn, Ni, and Al have been reported to induce oxidative
stress in the brain^[180]9,[181]11,[182]85, while riboflavin may
counteract this stress by enhancing antioxidant enzyme activities and
the glutathione redox cycle^[183]86. Mn has been demonstrated to cause
neuronal apoptosis through inducing pro-inflammatory response^[184]87,
but riboflavin has anti-inflammatory effects that can moderate
cognitive impairment in mouse models of inflammation and Alzheimer’s
disease^[185]88. Our observed findings may be attributable to
self-regulation mechanisms in individuals, which resemble negative
feedback processes occurring before the onset of disease.
Prenatal exposure to metals may reduce MDI and PDI scores by disrupting
the metabolism of lysophospholipids. A recent study from Puerto Rico
reported associations between metal exposure during pregnancy and
maternal lipids (N = 83 pregnant women) with significant alterations in
lysophospholipids^[186]89, which was aligned with the findings of the
present study. Metals such as Cr, Mn, and Pb, induce oxidative stress
through the production of reactive oxygen species, which in turn
increase lipid peroxidation of cell membranes^[187]11,[188]90,[189]91.
Consequently, elevated levels of lysophospholipids, which are products
of lipid peroxidation, can contribute to neuronal sheath demyelination
and neuronal apoptosis, ultimately impacting brain
development^[190]92,[191]93. The process of lipid peroxidation has been
demonstrated to damage the structure of cell membranes and to reduce
the permeability of the blood–brain barrier^[192]94, which may further
exacerbate adverse neurodevelopmental effects. Previous studies have
reported elevated levels of lipid peroxidation markers in two-year-old
children with ASD and ADHD^[193]18,[194]95, and identified increased
cord blood lysophospholipid levels as risk factors for ADHD and ASD
symptoms^[195]96. Consistent with these findings, our study also
observed that higher levels of lipid peroxidation products were
associated with lower Bayley scores. These findings suggest that lipid
peroxidation during fetal neurodevelopment may be one of the mechanisms
linking prenatal metal exposure and impaired neurodevelopment.
This study has several strengths, including its prospective birth
cohort design and comprehensive data collection from pregnancy. The
comprehensive metabolomics framework was used to discover the
overlapping pathways and metabolites in the association between
maternal urinary metals and children’s MDI/PDI scores, and allow all
metabolites detected in a pathway to be evaluated as a whole to assess
the potential mediation pathway. Moreover, our study had a relatively
larger sample size in comparison to other investigations on prenatal
exposure and metabolomics^[196]23,[197]89,[198]97. The present study is
subject to several limitations. First, residual confounding may persist
in any observational study and may affect the observed associations.
Nevertheless, after adjusting for various social and lifestyle
variables, our findings remained robust and consistent with previous
experimental and human evidence regarding the neurotoxic effects of
metals^[199]1,[200]45. Second, a single-time measurement of urinary
metals may not fully capture true exposure levels during pregnancy.
Urinary Cd is a reliable biomarker to reflect body burden over a
relatively long period of time^[201]98, but urinary levels of other
metals have moderate to high variability throughout pregnancy^[202]99,
due to the changes in individual physiological characteristics, dietary
habits, and environmental exposures. To enhance the reliability of
exposure assessments, future studies of repeated measurements of metal
concentrations during pregnancy are needed. Third, in this
observational exploratory study, the use of looser thresholds for FDR
correction in metabolomics analyses may lead to a higher incidence of
type I errors, but this was a necessary compromise to balance the
likelihood of missed discoveries and the risk of false
discoveries^[203]57,[204]100. Fourth, although we used a highly
sensitive ultra-performance liquid chromatography-tandem mass
spectrometry (UPLC-MS/MS) detection method to accurately quantify
metabolites reported in the literature as important for fetal growth
and development, it cannot fully cover all metabolites^[205]101.
Complementing the current study with an untargeted metabolomics
approach in the future will help to expand metabolite coverage and
capture more metabolic pathways that may be relevant to metal
neurotoxicity. Fifth, some of the covariates were self-reported, such
as annual household income, may be subject to recall bias. Finally, the
metabolic pathway of cord blood can serve as a mediator to reveal the
mechanism between prenatal metal exposure and neurodevelopmental
toxicity in children based on a longitudinal study, and the molecular
mechanisms responsible still require further elucidation and
validation.
Our study suggests that maternal metal exposure during early pregnancy
may affect children’s neurodevelopment. Metabolomics analysis indicates
that perturbed amino acids, lipids, neurotransmitters, and
neuroendocrine metabolism by metal exposure in utero may play important
roles in contributing to impaired neurodevelopment in childhood. The
upregulation of riboflavin metabolism may help alleviate some adverse
impacts on neurodevelopment associated with metals. Further
investigation is warranted to validate our findings, explore the
underlying metabolic mechanisms, and assess the potential of biomarkers
for prediction, therapy, and prevention approaches to mitigate the
adverse effects of environmental pollution exposure during pregnancy on
neurodevelopmental outcomes in children.
Methods
Study participants
This study was conducted using the Wuhan Health Baby Cohort (WHBC),
which was established at the Wuhan Women and Children Medical Care
Center in Wuhan, Hubei, China. Details about this cohort have been
previously reported^[206]43. Pregnant women were recruited if they met
the following criteria: (1) Chinese residents of Wuhan city; (2) had a
singleton pregnancy and completed a prenatal care examination in early
pregnancy (<16 weeks); (3) gave birth at the study hospital; and (4)
completed the questionnaires. For this study, 2120 pregnant women who
provided urine samples during the first trimester and cord blood
samples were included between March 2014 and March 2016. Subsequently,
the newborns with congenital malformations (N = 15) and fetal metabolic
diseases (N = 30) were excluded. Of the 2075 mother-child pairs, 1100
mothers whose children completed the neurodevelopment assessment at 2
years of age ultimately remained in this study. Participants with
missing values for pre-pregnancy body mass index (BMI) (N = 1) or
urinary metal concentration outliers (>five times the standard
deviation, N = 11) were excluded from the study. No statistically
significant differences were found in demographic characteristics about
the population whether or not they completed the Bayley scales at 2
years of age (Table [207]S14), except that the mothers of children who
completed the Bayley scales had a higher pre-pregnancy BMI (20.7 kg/m²
vs 20.4 kg/m², P = 0.048). Finally, a total of 1088 mother-child pairs
were included in the final analysis. All participants provided informed
consent before participating in the study. The research protocol was
approved by the ethics committee of Tongji Medical College, Huazhong
University of Science and Technology and Wuhan Medical & Healthcare
Center for Women and Children.
Neurodevelopment assessment
When the children were around 2 years old (range: 23–26 months), they
were invited back to the hospital to evaluate neurodevelopment using
the standardized Chinese revision of the Bayley Scales of Infant
Development (BSID-CR). BSID is one of the most widely used intelligence
scales for infants and young children, and BSID-CR is the Chinese
version which has been validated and used in Chinese children to assess
cognitive and psychomotor development^[208]27,[209]28. Cognition,
language and social development are assessed by the mental scale to
generate the MDI. Fine and gross motor development is assessed by the
psychomotor scale to generate the PDI. Then, MDI and PDI scores are
derived by converting the raw scores into standardized scores using
age-specific norms. Higher MDI and PDI scores indicate better
neurodevelopment. Children with MDI or PDI scores below 70 were
classified as moderate to severe cognitive or psychomotor development
delay, respectively^[210]47,[211]48. All BSID-CR tests were conducted
in a quiet room at the hospital by certified psychologists according to
standardized guidelines. Video reviews and grading were used for
quality control purposes.
Urine collection and metal exposure assessment
During early pregnancy (13.1 ± 1.1 weeks), pregnant women provided spot
urine samples in trace element-free containers, which were then stored
at −20 °C. Urine samples were thawed at room temperature prior to trace
metal determination. To prepare the samples for analysis, 2.0 mL of a
1.2% HNO[3] solution was added to a 0.5 mL sample of urine, and the
mixture was left to undergo nitrification overnight. The concentrations
of Cr, Mn, Pb, Ni, Cd, Tl, V, Al, Rb, Ga, and As were analyzed using
Inductively Coupled Plasma Mass Spectrometry (ICP-MS) (Agilent 7900,
Agilent Technologies, Santa Clara, CA, USA) in helium mode^[212]102.
To verify the accuracy of the measurements, each batch included the
addition of human urine standard reference material as an external
quality control sample (SRM2670a, the National Institute of Standards
and Technology, Gaithersburg, MD, USA). Each batch included a blank
sample containing 1.2% HNO[3], which was included solely to minimize
the potential for contamination. The spike recovery values for urinary
metals exhibited a range between 98.81% and 106.44%. The intra-assay
coefficients of variation (CV) ranged from 0.29% to 3.22%, while the
inter-assay CVs ranged from 0.05% to 7.26%. The limits of detection
(LODs) of these metals ranged from 0.001 to 0.106 µg/L. The
concentration of creatinine in the urine samples was quantified using a
creatinine kit (Mindray BS-200 CREA Kit, Mindray Bio-medical
Electronics Co., Ltd., Shenzhen, China), and the adjusted metal
concentrations were ultimately expressed as µg/g creatinine.
Cord blood collection and metabolomic analysis
After the babies were delivered, umbilical cord blood samples were
collected in tubes containing ethylenediaminetetraacetic acid (an
anticoagulant) by trained nurses. Subsequently, the cord plasma was
separated by centrifugation and stored at −80 °C in order to facilitate
subsequent research. The same sample pretreatment method was used for
the analysis of both metabolites and polar lipids^[213]103. After the
plasma was thawed at room temperature, 50 μL of cord plasma and 200 μL
of the extract (acetonitrile: methanol = 1:4, v/v), which contained the
61 labeled internal standards (Tables [214]S15 and [215]S16), were
added to a 2 mL centrifuge tube. The samples were subjected to a 15-min
vortexing process and then centrifuged at a rate of 19,000 × g for
10 min at 4 °C. Two hundred μL of the supernatant was collected for
analysis. The order for sample preparation and data collection of
biological samples were randomized. In this study, cord plasma was
collected from 1088 newborns.
We applied the stable isotope dilution internal standard method to
quantify cord plasma metabolites using a UPLC-QTRAP-MS system (SCIEX,
ExionLC AD-QTRAP 7500) (Table [216]S17). Two different chromatographic
columns [Waters ACQUITY UPLC HSS T3 (1.8 µm, 2.1 mm × 150 mm) and BEH
Amide (1.7 µm, 100 mm × 2.1 mm)] were used in the UPLC-QTRAP-MS
analysis. For Waters ACQUITY UPLC HSS T3, the mobile phases A and B
were 0.1% formic acid in water and 0.1% formic acid in acetonitrile,
respectively. The column temperature was 40 °C, with a flow rate of
0.4 mL/min. The injected volume was 10 μL. For Waters ACQUITY UPLC BEH
Amide, the mobile phases A and B were 0.1% formic acid in water
(containing 5 mM ammonium formate) and 0.02% formic acid in
acetonitrile, respectively. The column temperature was 40 °C, with a
flow rate of 0.3 mL/min. The injected volume was 1 μL. The scheduled
multiple reaction monitoring (sMRM) mode was used for detection, and
positive and negative ionization modes were used simultaneously. The
following general MS parameters were utilized: ion spray voltage of
+5500/−4500 V, source temperature of 350 °C, curtain gas at 40 psi, and
ion source gases 1 and 2 at 50 psi each. To optimize the MRM
transitions, standard solutions dissolved in a mobile phase (A: B ratio
of 1:1 v/v) were injected into the mass spectrometer. Optimized MRM
transitions, retention times, chromatographic columns, and the
deuterated internal standards of metabolites are shown in
Table [217]S17. Data acquisition and processing were performed using
the Sciex OS-Q software (Sciex, USA). In every 15 analytical samples, a
quality control (QC) sample was added to track the instrumental
analysis procedure. The target metabolites (Table [218]S17) selected
for detection in this study are those identified in previous literature
as critical for fetal growth and development. Alongside the 20 protein
amino acids, the assay included some neurotransmitters, tryptophan
metabolites, and dipeptides, such as γ-aminobutyric acid, indoleacetic
acid, and carnosine, which play important roles in fetal growth,
neurodevelopment, immune function, and energy
metabolism^[219]104–[220]108. After excluding metabolites with poor
reproducibility (relative standard deviation of QC > 20%), a total of
45 amino acids and their metabolites were retained. Furthermore, the
analysis includes 11 hormones (e.g. thyroxine,
cortisol)^[221]109,[222]110, 9 carnitines (e.g. acetylcarnitine,
palmitoylcarnitine)^[223]96, 7 bile acids (taurocholic acid,
glycocholic acid, etc.)^[224]111, and 7 vitamins (riboflavin,
thiamine)^[225]112,[226]113, which are crucial for maintaining fetal
hormone balance, energy metabolism, bile acid synthesis, and nutrient
supply. Furthermore, five additional metabolites (e.g. betaine) that
play a crucial role in fetal neurological and growth development were
also considered and analyzed^[227]114. All metabolites were detected in
over 96% of the cord plasma samples, and concentrations of metabolites
below the LOD were imputed using LOD/
[MATH: 2 :MATH]
. The coefficients of variation from QC samples ranged from 2 to 20%
(Table [228]S17).
We applied the relative quantitative analysis method to quantify cord
plasma polar lipids employing a UPLC-QTOF-MS/MS system (SCIEX, ExionLC
AD-QTOF X500R). A Phenomenex Kinetex C18 column (100 mm × 2.1 mm,
2.6 μm) was used to analyze polar lipids in the UPLC-QTOF-MS/MS system.
The column temperature was 55 °C, and the flow rate was 0.4 mL/min.
Mobile phase A consisted of a mixture of water, methanol, and
acetonitrile in a ratio of 1:1:1 (v:v:v) with 5 mM ammonium acetate,
while mobile phase B comprised a mixture of isopropanol and
acetonitrile in a ratio of 5:1 (v:v) with 5 mM ammonium acetate.
Information-dependent acquisition (IDA) was employed for the
acquisition of MS/MS data in negative electrospray ionization modes.
The following general MS parameters were utilized: a mass range of
50–1200 m/z, ion spray voltage of −4500 V, declustering potential of
−80 V, and collision energy at −30 V. Polar lipids identification was
carried out using MS-DIAL (version 4.80,
[229]https://systemsomicslab.github.io/compms/msdial/main.html). In the
identification process, mass tolerances with an accuracy of 0.01 Da for
MS and 0.05 Da for MS/MS were employed. The peaks were integrated
automatically using Sciex OS software (version 1.6, Sciex), and manual
checks and adjustments were made in case of peak misalignments. The K
nearest-neighbor method was used to impute missing lipid values. A
specific class of lipid molecular species was quantified by normalizing
the peak area of individual molecular ions to that of a selected
internal standard (Table [230]S16). The CV was calculated for each QC
sample across the batches, and metabolites with a CV greater than 30%
were excluded. A total of 354 polar lipids were selected and utilized
for further analyses.
Covariates
Information on maternal demographic information, including
pre-pregnancy weight, age, annual household income, maternal education,
and data on folic acid supplementation, was collected by trained nurses
through interviews prior to delivery. The pre-pregnancy BMI was
calculated by self-reported pre-pregnancy weight and height, which was
measured at admission to the hospital. Information regarding parity,
newborn sex, birth weight, and fetal metabolic disease was retrieved
from medical records. Cord plasma cotinine, a biomarker of tobacco
exposure, was analyzed by UPLC-QTRAP-MS as described above for the
metabolites, and the optimized MRM transitions, retention times,
chromatographic columns, and the deuterated internal standards of
cotinine are shown in Table [231]S17.
Statistical analysis
Maternal urinary concentrations of metals below the LODs were imputed
with LOD/
[MATH: 2 :MATH]
. All urinary concentrations of metals corrected by creatinine were
log2-transformed prior to statistical analysis. Furthermore, the
correlations between the urinary levels of metals were calculated using
Spearman analysis and presented in the form of a heatmap. To
demonstrate the calculation of the sample size, the significance level
was set at 0.05, the power level at 0.95, and the effect size at 0.15
based on the mean and standard deviation of the estimates reported in
the literature^[232]45,[233]53. The two-sample t-test for power
analysis was employed to calculate the sample size using the R software
(pwr package), resulting in a sample size of 178. Our sample size
(N = 1088) is much larger than the required, giving a test power of
over 0.99.
We investigated the associations of individual metals in maternal urine
and the summary score of metals with children’s MDI/PDI scores by
fitting linear regressions. Multiple testing was corrected using the
Benjamini–Hochberg false discovery rate (FDR) procedure^[234]115. Given
that metals were log2-transformed, we multiplied the β estimates by
log(Q3/Q1, 2) to determine the change in children’s MDI/PDI scores per
IQR increase in urinary concentrations, where Q1 and Q3 were the first
and third quartile of the metal concentrations, respectively^[235]116.
We used the multivariable logistic regression model to estimate the
associations of individual metals in maternal urine and the summary
score of metals with neurodevelopmental delay (PDI < 70 or MDI < 70).
Given the lack of statistical significance for the associations with
neurodevelopmental delay after multiple corrections (Table [236]S18),
subsequent analyses were conducted using the Bayley scores.
To analyze the association between the sum metals and neurodevelopment,
the summary score of metal concentrations was calculated by an approach
that standardizes and normalizes metal concentrations^[237]56. The
concentrations of each metal were added by one and natural
log-transformed. The log-transformed concentrations were divided by the
standard deviation of all metals, and then the average was calculated
as the summary score. We multiplied the β estimates by the difference
between Q3 and Q1 to determine the change in children’s MDI/PDI scores
per IQR increase in the summary score of metal concentrations.
Considering the potential interaction effects between metals, the WQS
regression, which can reduce dimensionality and avoid
multicollinearity, was conducted to assess the associations between the
potentially highly relevant co-exposures and health outcomes^[238]117.
Briefly, WQS combined the metals into an index by transforming the
metals into quartiles and constrained the association to a negative
direction, and the result was interpreted as the change in children’s
MDI/PDI scores associated with one quartile increase in WQS index of
the metal mixture. In addition, the weight for each metal expressed as
the percentage of that sum to one, was estimated from 1000 bootstraps
to identify their contribution to the mixture effect on the outcome.
Based on the biological plausibility and previous
studies^[239]22,[240]23, a directed acyclic graph was used to determine
the covariates (Fig. [241]S3). Finally, the models were adjusted for a
set of covariates, including maternal pre-pregnancy BMI (continuous),
maternal age at recruitment (continuous), parity (categorical:
nulliparous or multiparous), folic acid supplements (categorical: yes
or no), infant sex (categorical: male or female), maternal education
(categorical: high school or equivalent and below, bachelor’s degree
and above), and cotinine levels in cord blood (continuous).
We applied a comprehensive metabolomics workflow including a
metabolome-wide association study (MWAS), pathway enrichment analysis,
and MITM approach, to identify the molecular perturbation^[242]118.
Specifically, the MWAS was used to explore the associations of cord
blood metabolomics with maternal urinary metal concentrations and
children’s MDI/PDI scores. As the concentrations of metabolites
exhibited a right-skewed distribution, a log2 transformation was
employed to normalize the data. The covariates in the MWAS analysis
were consistent with the above. The KEGG database for Homo sapiens was
employed in Metaboanalyst 6.0 ([243]https://www.metaboanalyst.ca/) to
identify pathways associated with urinary metal concentrations and
children’s MDI/PDI scores, respectively. Furthermore, we used the MITM
approach to discover the overlapping pathways and metabolites.
Next, we used pairwise mediation analysis (R version 4.1.1; “mediation”
package^[244]57) and mediator dimension reduction (R version 4.1.1;
“hdmed” and “mediation” package^[245]57,[246]119) to explore the
mediating effect of overlapping metabolites and pathways, respectively,
in the association between maternal urinary metals and children’s
MDI/PDI scores in order to identify the underlying mechanism
(Fig. [247]S4). We hypothesized that the cord plasma metabolites and
pathways mediate the relationship between maternal urinary metals and
children’s MDI/PDI scores. In the analysis of the mediating effect of
pathways, the metabolites detected in an overlapping metabolic pathway
constitute a mediator group. We employed a Penalized Dimensional
Mediation Model to estimate the weights for each metabolite within
these mediator groups in a high-dimensional mediation framework. This
model facilitated the combination of mediator variables based on their
estimated weights, and then the mediating effects of this combined
mediating variable in the association between maternal urinary metal
exposure and children’s MDI/PDI scores were calculated.
Furthermore, in order to ascertain whether adverse perinatal factors or
other confounding factors influenced the reliability of the results,
the following sensitivity analyses were conducted: (1) excluding
children with preterm birth (gestation age <37 weeks) or low birth
weight (<2500 g); (2) excluding pregnant women who did not take folic
acid supplements during pregnancy; (3) excluding those with tobacco
exposure during pregnancy (cord plasma cotinine concentrations
>1.78 ng/mL)^[248]120,[249]121; (4) additional adjustment for more
covariates (gestation age and birth weight). Sex stratification was
also performed to explore the potential sex-specific neurotoxic effect.
All statistical analyses were performed using SAS software (version
9.4; SAS Institute Inc., Cary, NC, USA) or R software (version 4.1.1,
“pwr”, “gWQS”, “corrplot”, “ggplot2”, “ggforestplot”, “tidyverse”,
“ComplexHeatmap”, “circlize”, “hdmed” and “mediation” package).
Statistical significance was defined as a two-sided P-value less than
0.05. The condition of FDR < 0.2 or larger has previously been employed
to evaluate multivariate omics data^[250]57,[251]100, with the
objective of identifying more meaningful biomarkers. Accordingly, we
used an FDR threshold of 0.2 in metabolomic analyses.
Reporting summary
Further information on research design is available in the [252]Nature
Portfolio Reporting Summary linked to this article.
Supplementary information
[253]Supplementary Information^ (5.8MB, pdf)
[254]Supplementary Data 1^ (270.8KB, xlsx)
[255]Supplementary Data 2^ (21KB, xlsx)
[256]41467_2025_57253_MOESM4_ESM.pdf^ (4.7KB, pdf)
Description of Additional Supplementary Files
[257]Reporting Summary^ (293.8KB, pdf)
[258]Transparent Peer Review file^ (763.8KB, pdf)
Source data
[259]Source Data^ (285.2KB, xlsx)
Acknowledgements