Umbilical Cord Blood Metal Mixtures and Birth Size in Bangladeshi Children

Abstract

Background: Studies have evaluated environmental exposure to toxic metals such as arsenic (As), cadmium (Cd), manganese (Mn), or lead (Pb) on birth size; however, information on potential effects of exposures to metal mixtures is limited.

Objectives: We assessed the association between metal mixtures (As, Cd, Mn, Pb) in umbilical cord blood and neonate size in Bangladeshi children.

Methods: In this birth cohort study, pregnant women who were ≥18 years of age with an ultrasound-confirmed singleton pregnancy of ≤16wk gestation were recruited from two Bangladesh clinics between 2008 and 2011. Neonate size metrics were measured at the time of delivery. Metals in cord blood were measured using inductively coupled plasma mass spectrometry. We employed multivariable linear regression and Bayesian kernel machine regression (BKMR) to estimate associations of individual metals and metal mixtures with birth size parameters.

Results: Data from 1,088 participants was assessed. We found a significant negative association between metal mixture and birth length and head circumference when all metal concentrations were above the 60th and 55th percentiles, respectively, compared with the median. An interquartile range (IQR) increase in log Cd concentration {log[Cd (in micrograms per deciliter)] IQR=2.51 } was associated with a 0.13-standard deviation (SD) decrease in mean birth length (95% CI: −0.25 , −0.02 ) and a 0.17-SD decrease in mean head circumference (95% CI: −0.28 , −0.05 ), based on linear regression models adjusted for covariates and the other metals. An IQR increase in log Mn concentration {log[Mn (in micrograms per deciliter)] IQR=0.69 } was associated with a 0.07-SD decrease in mean birth weight (95% CI: −0.15 , 0.002).

Discussion: Metal mixtures in cord blood were associated with reduced birth size in Bangladeshi children. Results from linear regression models adjusted and the BKMR mixtures analyses suggest adverse effects of Cd and Mn, as individual metal exposures, on birth size outcomes.

Introduction

Prenatal exposure to environmental chemicals can disrupt fetal growth and developmental programming of organ function, thereby increasing the risk of diseases later in life (Grandjean et al. 2008). Environmental exposures to toxic metals have been associated with adverse birth outcomes, such as preterm birth and reduced birth size (Khoshhali et al. 2020; Hu et al. 2017; Bloom et al. 2014), which are closely associated with neonatal mortality and morbidity (Katz et al. 2013) and increased disease risks in adulthood (Gluckman et al. 2008).

Environmental metals such as arsenic (As), manganese (Mn), and lead (Pb) may cross the placenta and harm fetal development (Caserta et al. 2013). Although the placenta serves as a partial barrier for cadmium (Cd), Cd concentrations in cord blood have been associated with decreased birth length (Zhang et al. 2004), head circumference (Lin et al. 2011), and ponderal index (Guo et al. 2017), suggesting direct effects of exposure via transplacental circulation. However, it has also been proposed that maternal Cd exposure may affect fetal growth through effects on placental function (Geng and Wang 2019). Several epidemiologic studies have examined associations between birth outcomes and metal concentrations measured in maternal blood or urine, cord blood, or placenta (Ashrap et al. 2020; Kalloo et al. 2020; Kim et al. 2020; Signes-Pastor et al. 2019; Ashley-Martin et al. 2018; Neda et al. 2017; Arbuckle et al. 2016; Claus Henn et al. 2016; Thomas et al. 2015; Al-Saleh et al. 2014; García-Esquinas et al. 2013; Zota et al. 2009). However, the majority of these studies estimated associations with individual metal exposures, and there is limited information on the potential effects of metal mixtures on birth size. Environmental exposure to toxic metals (As, Cd, Mn, and Pb) is prevalent in Bangladesh (Kile et al. 2009), where rapid industrialization, urbanization, and geological and anthropogenic contamination have caused recent and widespread metal pollution (Islam et al. 2018).

In this birth cohort study, we measured metal concentrations in umbilical cord blood as a noninvasive biomarker of fetal exposure during the second and third trimester (Grandjean et al. 2005; Kaufmann and Scheffen 1998), and estimated associations with birth weight, birth length, and head circumference in a population-based sample from rural Bangladesh that experiences metal exposures from multiple sources. We estimated associations with individual metals and used Bayesian kernel machine regression (BKMR) to estimate the effects of co-exposure to metal mixtures (Bobb et al. 2018; Hamra and Buckley 2018).

Methods

Study Population

The study population has been described previously (Gao et al. 2019; Rahman et al. 2017). During 2008–2011, 1,613 pregnant women with an ultrasound-confirmed singleton pregnancy of ≤16wk gestation were recruited for a longitudinal prospective birth cohort study from two rural health clinics operated by the Dhaka Community Hospital Trust (DCH Trust) in the Sirajdikhan and Pabna Sadar Upazilas of Bangladesh. Of the 1,097 participants who had live births and umbilical cord blood samples, 9 were excluded because of incomplete information for covariates (household income, n=7 ; infant sex, n=1 ; gestational age, n=1 ; secondhand smoke, n=1 ) leaving 1,088 for the present analysis. All protocols were reviewed and approved by the human research committees at the Harvard T.H. Chan School of Public Health (HSPH) and the DCH Trust. All women provided written informed consent prior to participation.

Sample Collection and Analysis

Umbilical cord blood samples were collected by trained birth attendants using the same protocol. Specifically, the trained birth attendant double-clamped the umbilical cord after the baby was delivered. A vacutainer needle was used to puncture the umbilical vein to collect the blood sample into a trace element–free vacutainer tube (BD Vacutainer Royal Blue Cap; Becton Dickinson) with dipotassium ethylenediaminetetraacetic acid. If the umbilical cord blood had started to clot, the blood was manually massaged out of the umbilical cord vein into the vacutainer tube. Cord blood samples were immediately stored at 4°C and shipped on dry ice to the Trace Metals Laboratory at the HSPH. Detailed methods for the analysis of cord blood metal concentrations have been described previously (Gao et al. 2019). Briefly, metal concentrations in cord blood were analyzed by inductively coupled plasma mass spectrometry following an acid digestion method (Rodrigues et al. 2015). The limit of detection (LOD) varied by analytical batch. The mean LOD was 0.03μg/dL for As (range: 0.0006–0.2263μg/dL ), 0.009μg/dL for Cd (range: 0.00027–0.06393μg/dL ), 0.021μg/dL for Mn (range: 0.008–0.2115μg/dL ), and 0.013μg/dL for Pb (range: 0.00048–0.04626μg/dL ). Metal measurements in cord blood below a given LOD were imputed as the LOD divided by 2 (Claus Henn et al. 2016).

Measurement of Newborn Size at Birth

Newborn birth size indicators were assessed at the time of delivery by well-trained field staff from DCH Trust clinics. Birth weight was recorded to the nearest 10g using calibrated digital infant scales (Zota et al. 2009). Birth length was measured to the nearest 0.1cm on an infantometer in the supine position with the baby’s knees fully extended and soles of the baby’s feet held firmly against the measuring board. Head circumference was measured to the closest 0.1cm at the maximal occipitofrontal circumference (OFC) using a standard measuring tape. The gestational age of the fetus (in weeks) was determined by ultrasound measurement at the time of enrollment (≤16wk gestation). Gestational age at birth was rounded to the nearest completed week. We calculated z -scores for the birth size measures that were standardized to the study population mean and standard deviation (SD) for each week of gestational age as follows:

Z-score=Individual Value−MeanSD

Therefore, we did not adjust for gestational age when modeling associations with birth outcome z -scores. The z -scores could not be estimated for birth weight and head circumference for two births (1 birth was at 22 wk of gestation and 1 at 28 wk of gestation), or for birth length for five births (1 birth at 22 wk of gestation, 1 at 28 wk of gestation, and 3 at 29 wk of gestation) because the numbers of births during the gestational weeks (1 birth was at 22 wk of gestation and 1 at 28 wk of gestation) were too small to derive gestational-week–specific mean values and SDs. Three births at 29 wk of gestational age had the same birth length, so the z -score could not be estimated for birth length for three births at 29 wk of gestation.

Covariates

At enrollment at ≤16wk of pregnancy (first visit), trained DCH health care workers who lived in the local area administered questionnaires to collect the participants’ sociodemographic, lifestyle, and environmental information. Maternal weight and height were measured at scheduled clinic visits during pregnancy. Covariates were selected a priori based on previous literature and plausible associations with birth outcomes (Gao et al. 2019; Rahman et al. 2017), including maternal age (in years), maternal body mass index (BMI) at enrollment (in kilograms per meter squared), infant sex, household income (in taka), secondhand smoke exposure during pregnancy (yes or no), daily tea intake during pregnancy (yes or no), and study site (Pabna vs. Sirajdikhan). We included daily tea intake because of the potential for heavy metal contamination (Schwalfenberg et al. 2013; Jin et al. 2014; Falahi and Hedaiati 2013) and previously reported associations with adverse birth outcomes (Chen et al. 2018; Okubo et al. 2015; Bakker et al. 2010). None of the participants smoked during pregnancy (Gao et al. 2019), but 42% reported secondhand smoke exposure. For analyses of unstandardized (raw) birth size measures [birth weight (in grams), birth length (in centimeters), and head circumference (in centimeters)], we also adjusted for gestational age at birth (in weeks).

Statistical Analyses

Descriptive statistics were examined for all variables. Cord blood metal concentrations were right-skewed. Therefore, to reduce the influence of outliers, concentrations were natural log-transformed and centered at the median. Statistical analyses were performed using SAS (version 9.4; SAS Institute, Inc.) and R (version 4.0.3; R Development Core Team).

Multivariable Linear Regression

Associations between metals in cord blood and birth size z -scores and raw measures were estimated using the following multivariable linear regression:

Yi=β0+β1Asi+β2Cdi +β3Mni +β4Pbi +βZi+ei

where the Y denotes the birth size outcome (z -score or measured value) for individual i; As , Cd , Mn , and Pb are the centered log-transformed concentrations; and Z and β represent the covariates and corresponding regression coefficients. Results are presented as mean differences in birth size [with 95% confidence intervals (CIs)] per interquartile range (IQR) increase in log-transformed cord blood metal concentrations.

Bayesian Kernel Machine Regression

To estimate the effect of metal mixtures and potential nonlinear effects, we implemented BKMR, a statistical approach for multipollutant mixtures that flexibly models the joint effect of a mixture using a kernel function (Bobb et al. 2018). The BKMR model is specified as follows:

Yi=h(Asi, Cdi,Mni,Pbi)+βZi+ei

where the Y represents the outcome (birth size z -score) for individual i; and h() is an exposure–response function that can accommodate nonlinear and nonadditive exposure–outcome relationships and interactions among the mixture components. We used the Gaussian Kernel with the assumption of cor(hi,hj)=exp{−(1/ρ)∑4m=1(zim−zjm)2} to flexibly address a wide range of underlying functional forms for h(·), which effectively assumes that two individuals (zim, zjm) with similar exposure profiles [similar values of zim(Asi, Cdi, Mni, Pbi) ] will have similar health effects (Bobb et al. 2015, 2018; Valeri et al. 2017). The parameter ρ regulates the smoothness of the exposure–response function, and our primary models assumed ρ∼Unif (a,b) where a=0 and b=500 , and used quartiles of the As, Mn, and Pb concentrations (25th, 50th, and 75th percentiles) as knots. BKMR models were fit using a Markov Chain Monte Carlo algorithm with 10,000 iterations using the Gaussian kernel. Credible intervals obtained from BKMR incorporate additional uncertainty due to estimation of a high-dimensional set of exposures and, thus, account for multiple-testing (Valeri et al. 2017). BKMR results are displayed as estimates of the a) overall effect of the metal mixture; b) single metal associations; c) exposure–response relationships; and d) bivariate exposure–response functions for pairs of individual metals.

Sensitivity Analyses

We performed several sensitivity analyses. We repeated linear regression models of the raw (unstandardized) birth outcomes without adjusting for gestational age, given that it is a potential intermediate variable, and additionally adjusted linear regression models of z -scores and raw outcomes for type of delivery (caesarean vs. vaginal), another potential intermediate that also may influence precision. To assess the potential influence of Cd concentrations below the LOD (48%), we ran linear regression models with all metals coded as binary variables (<50th percentile vs. ≥50th percentile). In addition, we repeated linear regression analyses after excluding observations with any metal concentration ≥99th percentile to assess the possible influence of outliers.

To address concerns about the effect of multicollinearity on BKMR results, we ran BKMR using the group option (Bobb et al. 2018) for Cd and Mn (Yi=h[group(Cdi, Mni), Asi, Pbi]+βZi+ei) , which were moderately correlated (r=0.49 , p<0.001 , compared with r=−0.31 to 0.37 for other metal pairs) (Figure S1). In addition, we applied the group option to metals with common patterns of exposures based on a principal component analysis (PCA) of standardized metal concentrations (means=0 and SDs=1 ) (Gibson et al. 2019). For PCA, we used the varimax rotation to identify the principal components (PCs). We identified two PCs with eigenvalues >1 by PCA (Yi=h[group(Asi, Cdi, Mni),Pbi]+βZi+ei ). The first PC (PC1) alone accounted for 48% and the first two PCs accounted for about 74% of the total variance in metal exposure concentrations, suggesting two PCs were identified to explain the majority of the variation in the data. As, Cd, Mn were loaded on the PC1 with the loading values ranging from 0.70 to 0.83, and Pb was loaded on the PC2 with the loading value of 0.99 (Table S5). Finally, we assessed the sensitivity of the BKMR model to lower and higher smoothing parameters for the exposure–response function ρ (b=400 and b=600 , respectively).

Results

Study Population Characteristics

The overall study population included 1,088 participants [mean (SD) age at delivery, 23.0 y (4.2); mean (SD) BMI at enrollment, 20.5 (3.2); 550 (50.5%) male baby] (Table 1). Head circumference was missing for one birth (n=1,087 ). The z -scores could not be estimated for two births for birth weight and head circumference (1 birth at 22 wk of gestation and 1 at 28 wk of gestation) and for five births for birth length (1 birth at 22 wk of gestation, 1 at 28 wk of gestation, and 3 at 29 wk of gestation with having the same birth length), as described in the “Methods” section. Average infant sizes at birth (SD) were 2,838 (414) g for birth weight, 47 (2.6) cm for birth length, and 33 (1.2) cm for head circumference. Geometric means±geometric standard eviations [GMs±GSDs (ranges)] for cord blood metal concentrations were 0.57±2.37 (0.0002–1.63) μg/dL for As, 0.02±4.56 (0.0002–1.63) μg/dL for Cd, 6.44±2.13 (1.24–303.19) μg/dL for Mn, and 3.18±2.35 (0.36–83.50) μg/dL for Pb (Table 2). Concentrations were <LOD for Cd in 520 samples (47.8%) and were <LOD for As in 12 samples (1.1%).

Table 1 Study population characteristics, n=1,088 .

Table 1 has four columns, namely, Characteristic, Mean plus or minus standard deviation or lowercase italic n percentage, Median, and Range.

Characteristic	Mean±SD or n (%)	Median	Range
Maternal characteristics
Age at delivery (y)	23.0±4.2	22	18–41
BMI at enrollment (kg/m2 )	20.5±3.2	20.5	13.5–36.0
Monthly household income (taka)
≤3,000	164 (15.1)	—	—
3,001–4,000	291 (26.8)	—	—
4,001–5,000	330 (30.3)	—	—
5,001–6,000	168 (15.4)	—	—
>6,000	135 (12.4)	—	—
Secondhand smoke exposure during pregnancy
Yes	452 (41.5)	—	—
No	636 (58.5)	—	—
Daily tea intake during pregnancy
Yes	174 (16.0)	—	—
No	914 (84.0)	—	—
Study site
Pabna	520 (47.8)	—	—
Sirajdikhan	568 (52.2)	—	—
Neonatal characteristics
Gestational age at birth (wk)	38.0±2.0	38	22–42
Birth weight (g)	2,838±414	2,860	1,020–4,800
Birth length (cm)	46.6±2.6	47	28–74
Head circumference (cm)a	32.7±1.2	33	24–37
Sex
Male	550 (50.6)	—	—
Female	538 (49.5)	—	—

Note: Data are complete for all variables unless otherwise indicated. —, not applicable; BMI, body mass index; SD, standard deviation.

^an=1,087 due to missing data.

Table 2 Cord blood metal concentrations, n=1,088 .

Table 2, in six main columns, lists Metals (micrograms per deciliter), Less than the Limit Of Detection lowercase italic n (percent), Geometric Mean plus or minus Geometric Standard Deviation, Mean plus or minus standard deviation, Percentile, and Interquartile Range. The column Percentile has five subcolumns, namely, Twenty fifth, Fiftieth, Seventy fifth, Minimum, and Maximum. The column Interquartile Range has two subcolumns, namely, Untransformed and log-transformed.

Metals (μg/dL )	<LOD [n (%)]	GM±GSD	Mean±SD	Percentile					IQR
				25th	50th	75th	Min	Max	Untransformed	Log-transformed
As	12 (1.1)	0.57±2.37	0.88±1.30	0.32	0.55	0.93	0.06	23.44	0.61	1.05
Cd	520 (47.8)	0.02±4.56	0.06±0.12	0.005	0.01	0.06	0.0002	1.63	0.06	2.51
Mn	0 (0.0)	6.44±2.13	9.89±17.10	4.00	5.19	7.97	1.24	303.19	3.97	0.69
Pb	0 (0.0)	3.18±2.35	4.63±5.38	1.59	3.07	6.04	0.36	83.50	4.45	1.33

Note: The distributions were derived after substituting concentrations <LOD with the LOD divided by 2. The mean (range) of LOD was 0.032μg/dL (0.0006–0.2263μg/dL ) for As, 0.009μg/dL (0.00027–0.06393μg/dL ) for Cd, 0.021μg/dL (0.008–0.2115μg/dL ) for Mn, and 0.013μg/dL (0.00048–0.04626μg/dL ) for Pb. As, arsenic; Cd, cadmium; GM, geometric mean; GSD, geometric standard deviation; IQR, interquartile range; LOD, limit of detection; max, maximum; min, minimum; Mn, manganese; Pb, lead; SD, standard deviation.

Multivariable Regression Analyses

When adjusted for covariates and co-exposures to other metals, an IQR increase in log Mn concentrations {log [Mn (in micrograms per deciliter)] IQR=0.69 } was associated with a 0.07-SD decrease in mean birth weight (95% CI: −0.15 , 0.002) (Table 3). An IQR increase in log Cd concentration{[log [Cd (in microgram per deciliter)] IQR=2.51 } was associated with a 0.13-SD decrease in mean birth length (95% CI: −0.25 , −0.02 ) and a 0.17-SD decrease in head circumference (95% CI: −0.28 , −0.05 ). No significant associations were estimated for cord blood As or Pb and birth size outcomes. Associations with raw birth size measures were consistent with the results for z -scores: An IQR increase in log Mn concentration was associated with a 29-g decrease in mean birth weight (95% CI: −58 , −0.29 ), and an IQR increase in log Cd concentration was associated with a 0.31-cm decrease in birth length (95% CI: −0.6 , −0.01 ) and a 0.17-cm decrease in mean head circumference (95% CI: −0.23 , −0.03 ).

Table 3 Adjusted effect estimates (β and 95% CIs) from multivariable linear regression models for birth size outcomes in association with IQR increases in log-transformed cord blood metal concentrations.

Table 3 has three main columns, namely, Metals, Birth size lowercase z scores begin superscript lowercase a end superscript, and Untransformed birth size measures begin superscript lowercase b end superscript. The column Birth size lowercase z scores begin superscript lowercase a end superscript has three subcolumns, namely, Birth weight (lowercase italic n equals 1,086), lowercase italic p-Value, Birth length (lowercase italic n equals 1,083, lowercase italic p-Value, and Head circumference (lowercase italic n equals 1,085) lowercase italic p-Value. The column Untransformed birth size measures begin superscript lowercase b end superscript has three subcolumns, namely, Birth weight (lowercase italic n equals 1,088), lowercase italic p-Value, Birth length (lowercase italic n equals 1,088, lowercase italic p-Value, and Head circumference (lowercase italic n equals 1,087) lowercase italic p-Value.

Metals	Birth size z -scoresa						Untransformed birth size measuresb
	Birth weight [(n=1,086 )]	p -Value	Birth length [(n=1,083 )]	p -Value	Head circumference [(n=1,085 )]	p -Value	Birth weight [g (n=1,088 )]	p -Value	Birth length [cm (n=1,088 )]	p -Value	Head circumference [cm (n=1,087 )]	p -Value
As	0.03 (−0.06 , 0.12)	0.50	0.04 (−0.05 , 0.13)	0.43	−0.07 (−0.16 , 0.02)	0.11	14.71 (−19.05 , 51.36)	0.41	0.11 (−0.12 , 0.34)	0.34	−0.08 (−0.19 , 0.03)	0.14
Cd	0.02 (−0.09 , 0.14)	0.71	−0.13 (−0.25 , −0.02 )	0.03	−0.17 (−0.28 , −0.05 )	0.004	2.15 (−42.43 , 46.73)	0.92	−0.31 (−0.60 , −0.01 )	0.04	−0.17 (−0.23 , −0.03 )	0.02
Mn	−0.07 (−0.15 , 0.002)	0.06	−0.03 (−0.10 , 0.05)	0.47	0.03 (−0.05 , 0.09)	0.60	−29.32 (−58.36 , −0.29 )	0.05	−0.12 (−0.31 , 0.07)	0.21	−0.01 (−0.09 , 0.08)	0.89
Pb	−0.04 (−0.19 , 0.11)	0.61	−0.06 (−0.20 , 0.09)	0.44	0.08 (−0.06 , 0.23)	0.27	−20.68 (−78.43 , 37.08)	0.48	−0.23 (−0.61 , 0.15)	0.24	0.08 (−0.10 , 0.25)	0.39

Note: Numbers of observations differ because of missing head circumference data (n=1 ) and missing z -scores for birth weight (n=2 ), head circumference (n=2 ), and birth length (n=5 ). Concentrations <LOD (n=12 for As and n=520 for Cd) were replaced with the LOD divided by 2. As, arsenic; BMI, body mass index; Cd, cadmium; CI, confidence interval; IQR, interquartile range; LOD, limit of detection; Mn, manganese; Pb, lead.

^aModels adjusted for all metals, maternal age, maternal enrollment BMI, infant sex, income, daily tea intake, study site, and secondhand smoke.

^bModels adjusted for all metals, maternal age, maternal enrollment BMI, infant sex, income, daily tea intake, study site, secondhand smoke, and gestational age.

BKMR Analyses

We first estimated differences in birth size associated with a concurrent increase or decrease in all four metals from their median values (Figure 1A–C). We found significant negative associations of the metal mixture with birth length when all four metal concentrations were ≥60th percentile (As >0.65μg/dL , Cd>0.03μg/dL , Mn>6.83μg/dL , Pb>4.22μg/dL ) (Figure 1B), and with head circumference when all four metals were ≥55th percentile (As>0.59μg/dL , Cd>0.01μg/dL , Mn>5.5μg/dL , Pb>3.66μg/dL ) (Figure 1C) compared with their median values (As median=0.55μg/dL , Cd median=0.01μg/dL , Mn median=5.19μg/dL , Pb median=3.07μg/dL ), with stronger associations as the concentrations of the four metals increased.

[Image omitted - see PDF]

We then used the same models to estimate the difference in birth size with an IQR increase in each individual metal when the other three metals were fixed at their 25th, 50th, or 75th percentiles (Figure 2A–C). Consistent with results from the multivariable linear regression models (Table 3), an increase in log Mn concentrations from the 25th to 75th percentile was associated with nonsignificant reductions in mean birth weight of SD=0.05 (95% CI: −0.13 , 0.01), SD=0.05 (95% CI: −0.12 , 0.01), and SD=0.04 (95% CI: −0.11 , 0.01) when As, Cd, and Pb were set at their 25th, 50th, and 75th percentiles, respectively (Figure 2A). An increase in log Cd concentrations from the 25th to 75th percentile was associated with SD reductions in mean birth length of 0.50 (95% CI: −0.65 , −0.05 ), 0.42 (95% CI: −0.52 , −0.08 ), and 0.37 (95% CI: −0.60 , 0.09) (Figure 2B) and SD reductions in mean head circumference of 0.22 (95% CI: −0.44 , 0.13), 0.35 (95% CI: −0.43 , −0.06 ), and 0.49 (95% CI: −0.52 , −0.17 ) (Figure 2C), when As, Mn, and Pb were set at their 25th, 50th, and 75th percentiles, respectively. The negative association between Cd and head circumference increased as the concentrations of the other metals increased from their 25th to their 75th percentiles, suggesting a potential interaction with the metal mixture.

[Image omitted - see PDF]

To explore potential nonlinearity, we estimated single metal exposure–response functions when all other metals were fixed at their median values (Figure 3A–C). The metal–response curves suggest the linear inverse association between Mn and birth weight (Figure 3A); the inverse association between Cd levels above the LOD and birth length, although with considerable variation of association with the wide CI for high concentration; and the weak positive association between Mn and birth length (Figure 3B). The curves also suggest the linear inverse association between Cd and head circumference and the linear positive association between Mn and head circumference (Figure 3C). Results for As suggested weak positive associations with birth weight and length (U-shaped and linear, respectively) and a weak inverse association with head circumference (Figure 3A–C). Pb had weak linear inverse associations with birth weight and length, and a U-shaped association with head circumference (Figure 3A–C).

[Image omitted - see PDF]

We visualized the bivariate exposure–response function for two metals while the other metals are fixed at their median values (Figure 4A–C). Based on a visual inspection of the graphs, the association between the mixture and head circumferences becomes inverse and increases in magnitude as the concentrations of As and Cd increase when Mn and Pb concentrations are held at their median values (Figure 4C). To investigate further the interaction between two metals in association with neonate size, we fitted the BKMR models with a bivariate pairwise exposure–response function (Figure S2) (Valeri et al. 2017). The figure suggests the more-than-additive interaction between Cd and As in their association with head circumference, whereby the negative slope of Cd becomes steeper as As concentration increases from its 25th to 75th percentiles, and vice versa, when Mn and Pb are fixed at their median (Figure S2C). The bivariate exposure–response function also showed that the birth weight decreases when the concentrations of Mn and Pb increase or when the concentration of Mn increases and the concentration of As decreases (Figure 4A). The bivariate pairwise exposure–response function showed that slightly less-than-additive interaction on birth weight as the negative slope of Mn becomes slightly less steep as Pb concentration increases from its 25th to 75th percentiles, and vice versa, when Cd and As are fixed at their median (Figure S2A). The bivariate pairwise exposure–response function does not show more- or less-than-additive interaction on birth weight given that the slope of As was not influenced by Mn concentration, and vice versa, when Cd and Pb are fixed at their medians (Figure S2A).

[Image omitted - see PDF]

Sensitivity Analyses

When associations between cord blood metals and raw birth outcomes were estimated using multivariable regression without adjustment for gestational age at birth (a potential causal intermediate), estimates for birth length and head circumference were very similar to the corresponding estimates adjusted for gestational age (Table 3), including associations between IQR increases in log Cd concentrations and birth length [–0.33cm (95% CI: −0.63 , −0.03 )] and head circumference [–0.19cm (95% CI: −0.33 , −0.04 )] (Table S1). Associations with birth weight showed more variation between the two models than the other outcomes but were generally consistent with the primary model estimates, for example, –27g (95% CI: −58 , 4.4). Associations between cord blood metals and birth outcome z -scores were very similar to estimates from the primary model when additionally adjusted for type of delivery (cesarean vs. vaginal delivery) (Table S2).

Positive, negative, or null associations based on linear regressions of birth weight z -scores and cord blood metals as modeled as binary variables (<median vs. ≥median ) (Table S3) were generally consistent with the corresponding results based on models of log-transformed concentrations as continuous variables (Table 3), with the possible exceptions of associations between cord blood Cd and head circumference z -score [–0.04 (95% CI: –0.17 , 0.08) for Cd≥ vs. <0.01μg/dL vs. –0.17 (95% CI: –0.28 , –0.05 ) for an IQR increase in log-transformed Cd] and cord blood Pb and birth length z -score [–0.17 (95% CI: –0.35 , 0.01) for Pb≥ vs. <3.07μg/dL vs. –0.06 (95% CI: –0.20 , 0.09) for an IQR increase in log-transformed Pb]. Estimates for linear regression models of IQR increases of log-transformed cord blood metals and birth outcome z -scores were very similar to the primary model estimates (Table 3) when 11 observations with the concentration of any cord blood metal >99th percentile for the study population was excluded (Table S4).

Sensitivity analyses of BKMR with the group option applied to Cd and Mn (Figures S3A–C and S4A-C) were similar to results from the primary model (Figures 2A–C and 3A–C). In addition, when we repeated BKMR grouped according to the PCA (Figures S5A–C and S6A–C), the results were similar to the primary model (Figures 2A–C and 3A–C). When we repeated BKMR using two different smoothing parameters (b)=400 (Figure S7A–C) and 600 (Figure S8A–C), results were similar to the primary model (Figure 3A–C).

Discussion

Using linear regression adjusted for multiple metals and BKMR analysis of the metal mixture, we found significant negative associations of higher metal mixture concentrations with birth length and head circumference. Cord blood Cd was associated with significantly lower birth length and head circumference based on linear regression adjusted for covariates and the other metals. Metal exposure–outcome curves from BKMR were linear for head circumference but not birth length when other metals were held at their median concentrations, and almost half of the cord blood samples were <LOD for Cd. In addition, cord blood Mn was inversely associated with birth weight based on linear regression (p=0.05 ) and BKMR.

Few studies have estimated the effects of metal mixtures in cord blood on birth size. Although direct comparison is limited, studies have reported a significant inverse association of cord blood Cd with birth length (Zhang et al. 2004) and head circumference (Lin et al. 2011), as found in the present study. In a study of 44 newborns from a Cd-polluted area in Da-Ye county of China (Zhang et al. 2004), higher levels of cord Cd (>0.4μg/L ) were associated with a 2.24-cm decrease in birth length compared with lower Cd levels, and no association was found with birth weight. A study by Lin et al. (2011) of 402 newborns in Taiwan found a negative association between cord Cd and decreased head circumference and no associations with birth weight and length. In a study of 1,566 newborns in Saudi Arabia (Al-Saleh et al. 2014), cord Cd was associated with an increased risk of newborns with <10th percentiles birth weight and birth length, but the associations were null when adjusted for gestational age. In Spain, cord Cd was not associated with birth weight and birth length, probably due to the small sample size (n=114 ) (García-Esquinas et al. 2013). In the present study, we found a negative association of cord Mn with birth weight. In contrast to the present study, associations were not observed in other populations of 470 mother–infant pairs living in the Tar Creek Superfund site in Ottawa County, Oklahoma, USA (Zota et al. 2009; cord Mn mean=42μg/L ), and 1,519 mother–infant pairs from 10 cities in Canada (Ashley-Martin et al. 2018; cord Mn median=32μg/L ), where the concentrations of cord blood Mn were lower than in the present study (mean=99μg/L , median=52μg/L ). We found no associations of cord As with birth size outcomes. In a birth cohort study of 622 infants in the United States (Claus Henn et al. 2016), where cord As was lower than in the present study, compared with lower quartile levels (<3.3μg/L ), upper quartile levels of cord As (>1.8μg/L ) were associated with lower birth weight [β=−34.6 (95% CI: −146.1 , 76.9)], but the association was not statistically significant. Consistent with our findings, no significant associations were estimated for cord Pb and birth size outcomes in prior studies of 147 newborns in Iran (Neda et al. 2017) and 114 newborn–mother–father triads in Madrid, Spain (García-Esquinas et al. 2013).

The magnitude of estimated associations is considerable as compared with the estimated effect of maternal age, which may be clinically relevant (Lean et al. 2017). For example, the estimated effect of Mn on birth weight [SD=−0.07 (95% CI: −0.14 , 0.004) per IQR] is similar in magnitude to the estimated effect of 5 y of maternal age as an untransformed continuous variable [SD=−0.08 (95% CI: −0.15 , −0.01 ) per 5 y] in linear regression model adjusted for all metals, maternal BMI, infant sex, income, daily tea intake during pregnancy, study site, and secondhand smoke exposure during pregnancy as adjusted in the primary model (Table 3). Especially, the estimated effect of Cd on birth length [SD=−0.13 (95% CI: −0.24 , −0.01 ) per IQR] and head circumference [SD=−0.14 (95% CI: −0.25 , −0.03 ) per IQR] is much greater than the estimated effect of maternal age on birth length [SD=−0.03 (95% CI: −0.11 , 0.04) per 5 y] and head circumstance [SD=−0.03 (95% CI: −0.09 , 0.04) per 5 y].

The level of Cd in cord blood found in the present study [median=0.01μg/dL , arithmetic mean=0.57μg/dL , range: 0.0002–1.63μg/dL ] was within the range of levels seen in prior studies, and almost half of the study samples (48%) were below the LOD. Our measured level was higher than cord Cd levels measured in an agricultural region in Jiangsu Province, China (mean=0.28μg/L ) (Sun et al. 2014), Busan Metropolitan city in South Korea (median=0.04μg/L ) (Kim et al. 2015), and 10 cities across Canada (median<LOD ) (Arbuckle et al. 2016), but lower than levels reported in the capital city of Riyadh, Saudi Arabia (median=0.7μg/L ) (Al-Saleh et al. 2014), Taipei City and County in Taiwan (median=0.3μg/L ) (Lin et al. 2011), and urban and metropolitan areas in Madrid, Spain (median=0.27μg/L ) (García-Esquinas et al. 2013). Thirty percent (n=326 ) of samples in the present study surpassed the reference level of 0.5μg/L blood Cd established by the German Human Biomonitoring Commission of the German Federal Environmental Agency for children (Wilhelm et al. 2006). The concentrations of other metals in cord blood were higher than those reported in most populations (Table S6), suggesting excess fetal exposure to metals in this population.

Potential mechanisms that might explain Cd-associated fetal growth impairment are unclear but might include oxidative stress, placental dysfunction, impaired nutrient transport, and altered DNA methylation. Cd is known to induce oxidative stress in the fetus. In animals, Cd exposure during pregnancy has elevated oxidative stress-related proteins, including nicotinamide adenine dinucleotide phosphate oxidase 2 (NOX2), NOX4, and heme oxygenase-1 (HO-1), in the fetal liver and decreased body weight and liver weight in fetuses (Yi et al. 2021). Another animal study found that fetal weight, crown–rump length, and placental growth (weight and diameter) were decreased in Cd-treated mice (Wang et al. 2016). In humans, maternal oxidative stress during pregnancy has been associated with lower birth weight (Arogbokun et al. 2021). During pregnancy, Cd accumulates in the placenta which may impair placental function to transfer oxygen, nutrients, and waste (Esteban-Vasallo et al. 2012; Geng and Wang 2019). Excessive reactive oxygen species (ROS) in the placenta can impair placental function that may cause adverse pregnancy outcomes (Wu et al. 2016). Cd also interferes with maternal–fetal micronutrient transfer across the placenta (Wang et al. 2016; Mikolić et al. 2015). Cd in the human placenta has been inversely associated with zinc (Zn) in cord blood, suggesting Cd may inhibit Zn transfer to the fetus (Kippler et al. 2010). At the cellular and molecular level, an epigenetic mechanism, such as DNA methylation, may be involved in Cd-associated fetal development. Prenatal Cd exposure has been associated with variations in DNA methylation in maternal and fetal blood (Vilahur et al. 2015; Kippler et al. 2013) and the placenta (Everson et al. 2018). A study of maternal blood Cd concentrations during pregnancy and cord blood methylation in 127 mother–child pairs in Bangladesh reported sex-specific associations between Cd and methylation at individual CpG sites, some of which also showed sex-specific inverse correlations with birth weight (Kippler et al. 2013). The epigenome-wide association study (EWAS) of placental Cd and placental DNA methylation in two independent cohorts of mother–infant pairs in the United States reported Cd-associated variations in methylation at loci involved in inflammatory signaling and cell death as well as the association between gene expression in inflammatory signaling [i.e., the genes for tumor necrosis factor alpha-induced protein 2 (TNFAIP2), acyl-CoA thioesterase 7 (ACOT7), and retinoic acid-related orphan receptor (RORA)] and birth weight (Everson et al. 2018).

Mn is a trace essential nutrient that plays a role in physiological processes, including bone formation and growth in fetal development; however, it is also reported that high-level Mn may have developmental toxicity (Williams et al. 2012; O’Neal and Zheng 2015; Zota et al. 2009). Maternal Mn exposure crosses the placenta via active transport (Krachler et al. 1999), and excess Mn exposure during gestation has been associated with a decreased fetal size and weight in pregnant mice (Colomina et al. 1996; Sánchez et al. 1993), consistent with the negative association between cord blood Mn and birth weight in the present study population (Table 3). Potential mechanisms for these associations are uncertain, but they may include oxidative stress caused by high Mn exposures, leading to the impairment of cellular function and growth and fetal growth restriction (Erikson et al. 2006; Duhig et al. 2016). Blood Mn has been suggested as a biomarker of Mn exposure (Laohaudomchok et al. 2011; O’Neal and Zheng 2015), and the half-life of Mn in blood has been reported to be 10–42 d (Nelson et al. 1993), suggesting that the cord Mn represents fetal exposure during the third trimester. However, further studies with sequential measurements of maternal Mn are needed to confirm our findings and determine whether associations between Mn and birth weight depend on the timing of exposure.

Although the sources of metals were not assessed in the present study population, previous studies in Bangladesh suggests various potential routes of environmental exposures (Ghosh et al. 2020; Hossain et al. 2019; Forsyth et al. 2019; Rashid et al. 2016; Rodrigues et al. 2015; Kile et al. 2009), including As and Mn in drinking water (Ghosh et al. 2020; Rodrigues et al. 2015; Kile et al. 2009), Cd in food and sediment (Hossain et al. 2019), and Pb in turmeric (Forsyth et al. 2019) and tea (Rashid et al. 2016). Further studies are needed to identify the contribution of sources of metals exposure to give an insight into policy development.

Limitations

Our study has several limitations. First, anthropometric measurements of infants and exposure measurements in cord blood were performed only at birth, which may not have captured the most relevant time period of exposure or usual levels of exposure. Cord blood metal biomarkers in this study have different half-lives in blood. Unlike the Cd, Mn, and Pb (Colomina et al. 1996), the half-life of As in blood is very short, approximately a few hours (Awata et al. 2017), so cord blood As measurements may not represent long-term exposure. Inconsistent association between cord blood As and birth size metrics may be due in part to the short half-life of As. Because the anthropometric measurements and the exposure measurements in cord blood were conducted independently, we expected that measurement errors were independent; therefore, misclassification would be nondifferential misclassification and dilute the true association. Second, despite the inclusion of important covariates in our models, the potential for uncontrolled confounding cannot be ruled out in the present study or any observational studies. Third, umbilical cord blood was collected in trace element–free tubes as in other studies (Claus Henn et al. 2016). We cannot rule out the likelihood of contamination by the influence of blood collection devices although no contamination with the metals for blood collection materials and travel blanks was reported (Nakayama et al. 2019). Fourth, although ultrasound estimation of gestational age is more accurate in the first trimester (before 13 wk and 6–7 d) than later in pregnancy (Morgan and Cooper 2020), the timing ranged from 4–16 wk in the present study population, and we cannot rule out the possibility of some misclassification of gestational age for women who enrolled later in the study. Finally, we acknowledge the large proportion of samples <LOD for cord blood Cd in this population. We recruited participants from only two health clinics. These clinics are representative of the region but may not represent the general population in Bangladesh.

Conclusions

In conclusion, metal mixtures in cord blood were associated with reduced newborn birth size in a cohort of Bangladeshi children. In addition, results from linear regression models adjusted for metals co-exposures and from the BKMR mixtures analyses support negative associations of birth outcomes and Cd and Mn specifically. Future studies are warranted to confirm our findings.

Acknowledgments

M.-S.L. and D.C.C. had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. M.-S.L., K.-D.E., and D.C.C. were responsible for concept and design. All authors were responsible for acquisition, analysis, or interpretation of data. M.-S.L., K.-D.E., M.M., M.L.K., and D.C.C. drafted the manuscript. All authors assisted in the critical revision of the manuscript for important intellectual content. M.-S.L., K.-D.E., M.M., M.L.K., and D.C.C. performed statistical analysis. D.C.C. obtained funding. M.G., Q.Q., and D.C.C. provided administrative, technical, or material support. K.-D.E., M.M., M.L.K., and D.C.C. provided supervision.

This work was supported by the National Institute of Environmental Health Sciences (NIEHS) of the National Institutes of Health (NIH) (grants R01ES015533 and P30ES000002). The funding organizations had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.