The study was established at the affiliated hospitals of Nanjing Medical University and details about this population have been previously reported (22). Pregnant women who met the following criteria were recruited: (1) Residents of Nanjing City; (2) Age ≥ 18 years old; (3) Delivering at the study hospital; and (4) No serious disease such as cancer, hepatitis B or psychiatric illness. Initially, a total of 698 pregnant women were recruited in this study and maternal blood samples were collected. Follow-up was conducted when children were aged 5–8 years. Participants were excluded if they (1) had no birth information (n = 169); (2) had twins (n = 5); (3) were conceived through assisted reproduction technology (n = 25); (4) had no anthropometric measures information of children (n = 316); (5) had insufficient serum samples (n = 53). Finally, a total of 130 mother-child pairs were retained for the final analysis. Ethical approval was obtained from the Institutional Review Board of Nanjing Medical University, and all participants provided written informed consent prior to participation.

At enrollment, peripheral blood samples were collected from pregnant participants. Serum was isolated by centrifugation at 3,000 revolutions per minute for 5 min at 4 °C, and subsequently aliquoted and preserved at −80 °C until analysis. Metal levels were quantitatively measured in the collected samples. Quantification of metal concentrations in serum was conducted, with 15 elements initially detected. Among all measured metals, those with acceptable quality control performance and detection rates above 20% were retained for subsequent analyses, consistent with common practice in environmental epidemiology to exclude analytes with low detection frequencies to ensure statistical reliability (23). Of these, ten metals, namely iron (Fe), Cu, molybdenum (Mo), V, zinc (Zn), cobalt (Co), arsenic (As), Hg, Pb and cadmium (Cd), were selected for inclusion in the current study. Analytical procedures adhered to a previously validated protocol (24). Briefly, each serum sample underwent acid digestion using a mixture comprising 0.05% (v/v) Triton X-100, 0.1% (v/v) nitric acid, and internal standards (10 mg/L each of terbium [Tb], indium [In], bismuth [Bi], yttrium [Y], and scandium [Sc]). Metal quantification was performed using an inductively coupled plasma mass spectrometer (iCAP Q, Thermo Scientific, Bremen, Germany). Quality assurance was maintained by incorporating spiked quality control samples—constituting 10% of the sample batch—and procedural blanks, with one blank processed per ten samples. The relative standard deviations (RSDs) for both pooled and quality control samples ranged from 2.1% to 11.1%. Limits of detection (LODs) were estimated as three times the standard deviation derived from ten replicate blank digests. Detection rates of metals were calculated as the percentage of samples in which the measured concentration of a given metal exceeded its respective LOD.

Metabolomic profiling was conducted following a validated protocol described before (author?) (25). In brief, 20 μl of serum was aliquoted into a 1.5 ml Eppendorf tube and extracted with methanol at a 1:3 volume ratio. After vortex mixing for 10 s, and the samples were centrifuged at 15,000g for 15 min at 4 °C to precipitate proteins. The collected supernatant was dried under a gentle nitrogen stream and reconstituted in 10 μl of solvent prior to instrumental determination. Metabolite detection was carried out using an Ultimate 3,000 UPLC system (Dionex) coupled with a Q-Exactive mass spectrometer (Thermo Fisher Scientific, Bremen, Germany). Data acquisition was conducted in full-scan mode (resolution 70,000; m/z 70–1,050). Chromatographic separation used a gradient elution with mobile phases of 0.1% formic acid in water (A) and in acetonitrile (B). The flow rate was 0.4 ml/min, and the total run time was 15 min. To minimize analytical bias, sample injections were performed in random order. Metabolites were identified by comparing accurate mass, retention time, and tandem MS (MS/MS) spectra with those in a validated in-house reference library of authentic standards.

At the school-aged assessment (5–8 years), anthropometric measurements, including height and weight, were obtained using standardized and routinely calibrated equipment. Children's age-and-sex-adjusted BMI z-scores were derived based on the international World Health Organization (WHO) reference curves (26, 27). According to the WHO growth reference, overweight and obesity in children were defined as age- and sex-adjusted BMI z-scores ≥1 and ≥2, respectively (27). For the primary analyses, child OWO was defined as having a BMI z-score ≥1, encompassing both overweight and obesity.

Demographic information for both mothers and children were extracted from hospital maternity records by qualified medical personnel. At enrollment, trained interviewers administered standardized questionnaires to obtain sociodemographic and health-related data. Maternal variables included pre-pregnancy BMI, educational attainment, and passive smoking history. Child-related data included age, sex, birth weight, duration of outdoor activity, and frequency of sugar-sweetened beverage intake.

Differences between children with OWO and healthy controls were evaluated using independent-sample t-tests for continuous variables and Chi-square tests for categorical variables. To normalize the right-skewed distribution of serum metal concentrations, a natural logarithmic transformation was applied before statistical analysis.

Associations between individual metal exposures and the risk of OWO and BMI z-scores were examined using multivariable logistic regression models and generalized linear regression models, separately. Potential confounders were selected based on prior evidence and empirical associations with the outcomes (p < 0.20). These variables included continuous variables (maternal age, pre-pregnancy BMI, children's birth weight) and categorial variables (history of passive-smoking, maternal education, parity, children's gender, outdoor activity time, and the frequency of sugar-sweetened beverage intake). Metals levels blow the LOD were imputed by LOD/2 (28). Dose-response relationships were further assessed using restricted cubic spline (RCS) functions.

To explore associations between prenatal exposure to metal mixtures and OWO as well as BMI z-scores, we applied elastic net (ENET) regression, with model optimization achieved through 10-fold cross-validation (29). Furthermore, the Deletion-Substitution-Addition (DSA) model (30) was also employed, with 50 iterations performed, to identify metals jointly associated with childhood health outcomes. For both models, co-exposure models were then constructed by incorporating all measured metals into a logistic regression framework for OWO and a generalized linear regression model for BMI z-scores, adjusting for relevant covariates.

Non-linear and interactive associations among metal mixtures were investigated using Bayesian kernel machine regression (BKMR) model (31). The joint effect of all ten metals was estimated by comparing the OWO risk at the 25th−75th percentile relative to median exposure levels. Each metal's independent effects were evaluated by varying the concentration of one metal while holding all others fixed at their medians. BKMR models were fit using a binomial distribution for OWO and a Gaussian distribution for BMI z-scores with 100 knots and 5,000 iterations of Markov Chain Monte Carlo (MCMC) sampling. Metals with correlation coefficients greater than 0.3 were grouped into the same category. Based on Spearman's correlation coefficients, Fe, Zn, and Mo were classified into Group 1; Cu and Co into Group 2; V, Cd, and Pb into Group 4; and Hg into Group 5. Group posterior inclusion probabilities (groupPIP) derived from the BKMR model were used to characterize the overall importance of each exposure group. Conditional posterior inclusion probabilities (condPIP) were calculated to evaluate the relative contribution of individual metals to the OWO risk and BMI z-score variation. PIP values greater than 0.5 were considered indicative of a substantial contribution to OWO risk and BMI z-score variation. Bivariate exposure–response functions were additionally estimated within the BKMR framework to assess potential interactions between metals, with all remaining metals fixed at their median concentrations.

Associations of metabolites and metal exposures with OWO and BMI z-scores were analyzed using generalized linear regression models. To explore potential biological mediation pathways, the meet-in-the-middle (MITM) framework was applied to assess the extent to which metabolites mediated the relationship between prenatal metals exposure and OWO and BMI z-scores (32).

In addition, to explore potential sex-specific and age-related differences in the associations between maternal metal exposure and child growth, we conducted stratified analyses by child sex (boys vs. girls) and by age group (≤6 years vs. > 6 years) using BMI z-scores as the outcome. In the stratified models, all covariates included in the main analysis were adjusted for, except for sex or age, respectively.

Statistical analyses were performed using R software (version 4.5.0). Statistical significance was determined based on a two-tailed p-value threshold of < 0.05.

The characteristics of study participants are presented in Table 1. The median age of the included pregnant women was 27.5 years. Among the 130 mothers, 66.7% held a bachelor's degree or higher, 78.7% received folic supplementation during pregnancy, 97.5% were primiparous, and 61.4% had a history of passive smoking. The mean age of the included children was 6.49 years, with an average birth weight of 3,423 g. Of the 130 mother-child pairs, 44 children (33.9%) were classified as OWO, with 19 (14.6%) were diagnosed as obese. Mothers of OWO children exhibit significantly higher BMI values compared to those with children of normal weight (p = 0.001). Additionally, a higher proportion of boys were observed in the OWO category, and these boys also had higher birth weights. Other demographic characteristics, such as maternal age, parity, outdoor activity and daily consumption of sugary drinks, were comparable between the two groups. The maternal and infant characteristics included in the analysis and those excluded from the analysis are comparable (Table S1).

Figure 1 illustrates the distribution of serum metal concentrations. Detection rates were generally high (>90%) for most metals, except for vanadium (V, 56.15%), cadmium (Cd, 48.46%), and mercury (Hg, 47.96%). Among the analyzed elements, copper (Cu) demonstrated the highest mean concentration (1,771.14 μg/L), followed by iron (Fe, 1,539.13 μg/L), while cadmium (Cd) had the lowest (0.09 μg/L; Table S2). Pairwise correlation analysis revealed predominantly positive correlations among metals, with coefficients ranging from −0.20 to 0.44.

Logistic regression models revealed significant associations between prenatal exposure to multiple metals and the risk of OWO in children (Figure 2A). Specifically, after adjusting for potential covariates, prenatal Cu and V exposure were significantly associated with higher odds of OWO in school-aged children (Cu: OR = 24.171, 95% CI: 2.351–403.256, p = 0.014; V: OR = 2.534, 95% CI: 1.273–5.623, p = 0.014). Additionally, prenatal V exposure showed a positive association with BMI z-scores in this population (β: 0.293, 95% CI: 0.015–0.572, p = 0.042), while prenatal Cu exposure showed a borderline positive association (β: 0.758, 95% CI: −0.001–1.517, p = 0.054; Figure 2B). Sex-stratified analyses indicated that maternal V exposure was positively associated with BMI z-scores in boys, whereas maternal Cu exposure was positively associated with BMI z-scores in girls (Figure S1). Further stratification by children's age revealed that maternal V and Cu exposure were positively associated with BMI z-scores in children older than 6 years (Figure S2), whereas no associations were observed in children younger than 6 years.

RCS model indicated a non-linear relationship between V levels and OWO risk (Figure 2C, p for nonlinear = 0.004), wherein low concentrations of V were associated with an increased risk of OWO, while higher concentrations appeared to exert a protective effect. No other significant non-linear associations were identified between either two metals and children's OWO or BMI z-scores (Figures 2C, D).

Variable selection method identified Cu and V as being positively associated an increased risk of childhood OWO, whereas Cd exposure demonstrated a negative association with OWO risk (Figure 3A). A comparable trend was observed for BMI z-scores (Figure 3B), whereby higher BMI z-scores were noted when all metals were above the 50th percentile relative to their median exposure levels (Figure S3). With respect to PIP values, Cu, As, and Hg exhibit the highest conPIP values (conPIP > 0.8) for their association with OWO. Among metals with groupPIP > 0.5 for BMI z-scores, Cu demonstrated the highest conPIP values (0.79, Table S3), underscoring its prominent role in influencing child growth. Furthermore, univariate exposure-response analysis revealed approximately linear associations of key metals with both OWO and BMI z-scores. No interactive effect between metals on OWO risk was identified in bivariate exposure response analysis (Figures S4, S5). Specifically, Cu and V were found to be positively associated with OWO risk (Figure 3C), and their concentrations were likewise positively correlated with elevated BMI z-scores in school-aged children (Figure 3D).

Of the 170 metabolites detected, prenatal Cu and V were significantly associated with 79 and 17 metabolites, respectively (p < 0.05, Figure 4A). Higher levels of Cu were primarily associated with lipid metabolism. Notable associations included positive correlations with glycerophosphocholine, 17-hydroxyprogesterone, and 2-hydroxycaproic acid, alongside negative associations with 5-hydroxylysine, capric acid, and L-serine. Similarly, V exposure was predominantly associated with metabolites involved in lipid metabolic processes, such as 2-hydroxycaproic acid, 2-methoxyestradiol, cortisol, dodecanoic acid, sphingosine, progesterone, and testosterone.

In the Metabolite-Wide Association Study (MWAS) of childhood OWO and BMI z-scores, elevated levels of D-Glucurono-6,3-lactone, glycerophosphocholine, glycocholic acid, and L-Phenylalanine were significantly associated with an increased risk of OWO. In addition, glycerophosphocholine was positively associated with children's BMI z-scores, whereas glycine exhibited a negative association (Figure 4A).

Based on the MITM approach, glycerophosphocholine was identified in the MWAS of maternal Cu exposure, as well as in the MWAS for both children's OWO risk and BMI z-scores (Figure 4B). Moreover, glycine showed a significant association with maternal Cu levels and a negative association with children's BMI z-scores, indicating its potential mediating role in the relationship between prenatal Cu exposure and children's growth outcomes.