Assessment of individual and mixture effects of element exposure measured in umbilical cord blood on birth weight in Bangladesh

We conducted a prospective cohort study to identify associations between elements exposure and birth outcomes from 2008 to 2011 at Sirajdikhan and Pahna Sadar Upazilas in Bangladesh. The following study population inclusion criteria were used: (1) an ultrasound confirming a singleton pregnancy of ≤16 weeks gestation, (2) tube well water as the primary drinking water source for at least 6 months before enrollment, and (3) prenatal care provided by Dhaka Community Hospital Trust (DCH) throughout pregnancy and birth. Cohort recruitment and enrollment processes and other study details have been described [16]. The study was described in detail to all participants, and informed consent was obtained before study enrollment. All protocols performed in studies involving human participants were reviewed and approved by the Human Research Committees at Harvard T.H. Chan School of Public Health, Nanjing Medical University, and DCH.

In all, 1,613 eligible individuals were recruited at enrollment, of which 244 (15.2%) experienced fetal loss or neonatal death, and 302 families (18.8%) were lost to follow-up because of family migration or study withdrawal. At delivery, there were 745 mother-infant pairs with umbilical cord serum samples available.

Birth weight was measured within 120 min of delivery on a pediatric scale calibrated and rounded to the nearest 10 g before each measurement by trained healthcare workers. Gestational age was determined through ultrasonography by a licensed general practitioner using either (1) the gestational sac mean diameter if the pregnancy was between 4 and 7 weeks or (2) the crown–rump length if the pregnancy was between 7 and 16 weeks.

Demographic and anthropometric information of mothers and children were collected at less than 16 weeks gestation and birth. Other covariates in this analysis were collected at the time of enrollment using a structured questionnaire, including maternal age, body mass index (BMI), age of marriage, maternal and spouse education level, family income level, and second-hand smoking exposure.

Umbilical cord serum samples were collected immediately at delivery. Cord serum samples from newborns were obtained through venous puncture of the umbilical cord. All samples were collected and kept at −80 °C, shipped to the Trace Metals Laboratory at the Harvard Chan School, received in perfect condition, and correctly identified.

We analyzed 56 elements in umbilical cord serum (Figure S1 (available online at stacks.iop.org/ERC/3/105001/mmedia)) using an ICAP QC inductively coupled plasma mass spectrometer (ICP-MS) (Agilent 7700x ICP-MS, USA) at Shanghai Biotree Biotech Co., Ltd, including 9 alkali or alkaline earth metals [beryllium (⁹Be), sodium (²³Na), magnesium (²⁴Mg), potassium (³⁹K), calcium (⁴⁴Ca), rubidium (⁸⁵Rb), strontium (⁸⁶Sr), caesium (¹³³Cs), and barium (¹³⁷Ba)], 10 transition metals [thallium (⁴⁷Ti), vanadium (⁵¹V), chromium (⁵²Cr), manganum (⁵⁵Mn), iron (⁵⁶Fe), cobalt (⁵⁹Co), nickel (⁶⁰Ni), copper (⁶⁵Cu), zinc (⁶⁶Zn), and antimony (¹²¹Sb)], 16 rare earth elements (REEs) [scandium (⁴⁵Sc), yttrium (⁸⁹Y), lanthanum (¹³⁹La), cerium (¹⁴⁰Ce), praseodymium (¹⁴¹Pr), neodymium (¹⁴⁶Nd), samarium (¹⁴⁷Sm), europium (¹⁵¹Eu), gadolinium (¹⁵⁷Gd), terbium (¹⁵⁹Tb), dysprosium (¹⁶³Dy), holmium (¹⁶⁵Ho), erbium (¹⁶⁶Er), thulium (¹⁶⁹Tm), ytterbium (¹⁷²Yb), and lutetium (¹⁷⁵Lu)], and 21 other elements [lithium (⁷Li), boron (¹¹B), aluminum (²⁷Al), gallium (⁷¹Ga), arsenic (⁷⁵As), selenium (⁷⁸Se), zirconium (⁹⁰Zr), niobium (⁹³Nb), molybdenum (⁹⁵Mo), silver (¹⁰⁷Ag), cadmium (¹¹¹Cd), stannum (¹¹⁸Sn), hafnium (¹⁷⁸Hf), tantalum (¹⁸¹Ta), wolfram (¹⁸²W), hydrargyrum (²⁰²Hg), thallium (²⁰⁵Tl), lead (²⁰⁸Pb), bismuth (²⁰⁹Bi), thorium (²³²Th), and uranium (²³⁸U)].

For elements analyses, we followed a validated procedure for human blood with certified reference materials. Briefly, 60 μl of cord blood were diluted using 1800 μl of ammonia solution (1% of NH₄OH), and internal standards (ISTD) were added. The ISTD solution comprised Li (40 μg l⁻¹) and rhodium (Rh), indium (In), and rhenium (Re) (20 μg l⁻¹ each). A blank sample was also introduced to each of the 10 samples to ensure that there was no memory effect for any element. The limits of detection (LOD) and limits of quality (LOQ) were established as the signals that were three and ten times higher, respectively, than the signal of the average of ten consecutive blank measurements. Element concentrations below the LOD were imputed by LOD/2.

Descriptive statistics [relative standard deviation (RSD), median] of umbilical cord blood concentrations were calculated to describe the distributions of element concentrations among study participants. All concentrations of elements displayed skewed distributions (Figure S1) and thus were natural log transformed before statistical analysis.

Multivariate linear regression was used to evaluate the associations between individual element exposure measured in cord blood and birth weight adjusted for covariates, including maternal age, gestational age, infant sex, BMI, second-hand smoking, marriage age, education level, and family income levels, which were significantly associated with the outcome (P ≤ 0.05). To evaluate potential non-linear associations, linear models were run with each element exposure converted as a categorical variable based on interquartile range (IQR). We also explored associations between elements and birth weight stratified by population characteristics. Statistical significance was evaluated with the false discovery rate (FDR) to control for multiple comparisons, and an FDR-q ≤ 0.05 was considered statistically significant.

To explore interaction effects between multiple metals on birth weight, we used Bayesian kernel machine regression (BKMR), a non-parametric Bayesian variable selection framework for conducting mixture analysis without the assumption of linearity of the associations [17]. BKMR combines Bayesian and statistical learning methods to regress an exposure–response function iteratively by a Gaussian kernel function. BKMR can identify interactions between element mixtures. Here, BKMR modeled flexible function of element concentrations while adjusting for the same covariates described above. We assessed bivariate exposure-response effects of two metals if the second metal was fixed at the 10th, 50th, or 90th quantile while the other element was at the median.

We further verified the interaction effects among identified trace elements by BKMR using linear regression models with cross-product terms and assessed the significance of the coefficients using the Wald statistic. Further, categorical variables and restricted cubic spline were applied to explore metal-metal non-linearity interaction effects. For the categorical approach, levels of identified elements were re-categorized as 'low' (1st and 2nd tertiles) and 'high' (the 3rd tertile) [8]. For example, we analyzed effects as low Li with high Er, low Li with low Er, high Li with high Er, and high Li with low Er. Also, we added interaction effects in the restricted cubic spline model to explore the potential non-linear interaction effects between elements by setting the target elements at different quantiles [18]. And, FDR was used to evaluate statistical significance to control for multiple comparisons.

Weighted quantile sum (WQS) regression analyses were applied to construct an element risk score (ERS) adjusting for the same covariates described above. WQS regression was developed to assess the combined and discrete effects of multiple predictors in the context of correlated high-dimensional mixtures, which estimates an index that identifies the influential exposure variables with non-negligible weights and tests for associations between the exposure index and an outcome in a traditional linear framework as: g(μ) = β₀ + β₁WQS + z'ϕ. Here, g(μ) reflects a nonlinear link function allowing generalization to continuous, binary, and other distributions. As in typical regression approaches, β₀ represents the model intercept while β₁ represents the parameter estimate for the co-exposure index, represented here as WQS; the significance of this parameter reflects a straightforward test of associations between the co-exposure index and the outcome. The WQS index is constructed such that $WQS=\sum _{i}{w}_{i}{q}_{ij},$ where ${w}_{i}$ indicates a vector of empirically estimated weights for each mixture component, and ${q}_{ij}$ indicates the values of the mixture variables to each subject per element, which have been standardized [19].

All statistical analyses were performed using R version 3.6.0.

Demographic characteristics of the 745 mother-infant pairs and associations between demographic characteristics and birth weight are presented in table 1. The mean marriage age of the study population was 17.58 ± 2.29 years, with 59.46% of women married before 18 years of age. Further, 62.55% of women had a normal pre-pregnancy BMI (18.5–23.9 kg m⁻²), with a mean weight of 46.73 ± 7.74 kg and height of 151.09 ± 5.66 cm. In addition, 312 (41.88%) women were exposed to second-hand smoking, 188 (25.23%) spouses had no formal education, 245 (32.89%) had secondary education, and 320 (42.95%) received an income between 4000 and 6000 taka. Among infants, the average birth weight was 2841.05 g (standard deviation: 424.09 g; range: 800–4500 g), 376 (50.47%) were boys, 159 (21%) infants were born preterm (< 37 weeks of gestation), and 137 (18.39%) were small for gestational age.

Descriptive statistics (LOD, LOQ, RSD, and quartiles) of cord serum element concentrations are presented in table S1. Most of the samples had detectable element concentrations; only three elements exhibited concentrations higher than LOD in less than 50% of samples and were excluded from the subsequent statistical analysis [Cd (56.387% > LOD), Lu (70.47% > LOD), and Tl (69.66% > LOD)]. All the elements followed right-skewed distributions (Figure S1).

Multivariate linear regression model for each element adjusted for common covariates identified seven elements related to birth weight (FDR-q ≤ 0.05) (figure 1(A) and table S2). The presence of Li [b = −33.5, 95% CI: (−46.26, −20.44), P = 5.21 × 10⁻⁷, FDR-q = 3.07 × 10⁻⁵)], Mn [b = −38.94, 95% CI: (−63.24, −14.64), P = 0.002, FDR-q = 0.021], Co [b = −49.68, 95% CI: (−73.01, −26.35), P = 3.37 × 10⁻⁵, FDR-q = 0.001], and Cu [b = −74.89, 95% CI: (−121.52, −28.27), P = 0.002, FDR-q = 0.021] in umbilical cord serum were associated with low birth weight respectively. Also, we found the concentration of Y [b = −45.57, 95% CI: (−76.52, −14.62), P = 0.004, FDR-q = 0.039] and Er [b = −41.33, 95% CI: (−69.82, −12.84), P = 0.005, FDR-q = 0.039], both REEs, were associated with decreased birth weight. Interestingly, per each unit increment of Se umbilical cord serum levels was associated with an average increase in birth weight of 161.77 g [95% CI: (83.98, 239.56), P = 5.10 × 10⁻⁵, FDR-q = 0.001]. The relationships between above elements and birth weight according to quartiles of exposure were shown in table S3. And the results of categorical analyses shown that exposure to higher concentrations of elements would have greater effects on birth weight. Significant correlations between the above identified elements are shown in figure 1(B).

Zoom In Zoom Out Reset image size

Also, the associations between these elements and birth weight stratified by population characteristics were generally consistent with our findings (Table S4). Interestingly, the negative effects of Mn and Cu seemed to be more obvious in preterm birth. Pregnant women exposed to second hand smoking or marrying before 18 years may be affected easily by elements exposure.

BKMR analyses were used to explore potential interactions among candidate elements, indicating potential interactive associations between Li and other elements (Mn, Co, Cu, Se, Y, and Er) (figure 2). We also further explored the protective effect of Se against toxicity of the other elements on birth weight.

Zoom In Zoom Out Reset image size

Multivariate linear regression showed that there was no significant interaction on the multiplicative scales after multiple testing correction (FDR-q > 0.05) (Table S5). Metal-metal non-linearity interaction analyses showed that high Li exposure weakened the protectiveness of Se and aggravated the toxicity of other elements exposure on birth weight (Table S6). For example, there was a significant negative association between element exposure and birth weight only among those exposed to high levels of Er and Li compared with the other three groups. By contrast, the negative effect of element exposure on birth weight could be weakened in the population with high Se concentration compared with those with low Se concentration (Table S7).

Restricted cubic spline showed that associations between the identified elements (except Li) and birth weight differed by varying the series quantiles of Li and exposure to high Li may increase the risk of lower birth weight (figure 3). For example, Er had a strengthened negative association on birth weight according to the increased percentile of Li concentration (figure 3(F)), while the toxicity of Er on birth weight was considerably weakened with the increased percentile of Se concentration (figure 4(F)). High concentration of Se could weaken negative effects of toxicity elements exposure and similar opposite patterns of other non-linear interactions to Li between Se and the other risk elements are shown in figure 4.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Using the WQS model, the six risk elements (Li, Mn, Co, Cu, Y, and Er) were spontaneously evaluated for their joint effect. The ERS was constructed as a weighted sum of the elements using the mean weight of elements from the WQS regression attributed to the mean contributions of Li (48%), Er (21%), Co (16%), Cu (7%), Mn (7%), and Y (1%) (figure 1(C)). The ERS ranged from −3.10 to 2.21 with a mean (SD) of −0.3 (1.20). Note that the higher ERS index meant higher risk element exposure because WQS regression focuses inference in the negative direction in this study. Higher ERS values indicated more susceptibility to decreased birth weight in relation to cumulative exposure to element mixtures; per each unit increment of ERS, birth weight decreased 64.73 g [95% CI: (−86.75, −42.71), P = 1.23 × 10⁻⁸] after adjustment for the same covariates above. Further, the quantile analyses illustrated that the negative effect of element mixture on birth weight was more obvious among those with a high ERS [b = −21.73, 95% CI: (−31.16, −12.31), P_trend  = 7.21 × 10⁻⁶] (figure 1(D) and table S8).