Prenatal exposure to metals and metal mixtures influences birth weight in affected populations in the Republic of Suriname, South America

Abstract

This study examined the relationship between infant birth weight and exposure during pregnancy to three non-essential trace metals, cadmium, lead, and mercury, and one essential trace metal, manganese. Small-scale gold mining has contaminated freshwater fish with mercury, a primary food resource, in interior regions of Suriname, South America. Whole blood samples from pregnant women collected in early pregnancy (late first trimester or early second trimester) were used to determine trace metal concentrations. Inductively-coupled plasma mass spectrometry was used to measure trace metal concentrations. Bayesian kernel machine regression was used to evaluate the association of trace metal exposures on infant birth weight and possible interactions among trace elements. Lead was negatively associated with birth weight. Mercury appeared to affect birth weight in a non-linear fashion. Low birth weight was observed at low concentrations of blood mercury. Birth weight increased with increasing concentrations of mercury up to the higher concentrations of mercury where birth weight appeared to trend downward. Lead and mercury interacted such that high concentrations of lead and low concentrations of mercury were associated with low birth weight. Cadmium and manganese were not associated with birth weight in this study. Lead has been observed in previous studies to negatively influence birth weight as this study found. The association of mercury with birth weight in this study might reflect diet quality. With fish being a principal dietary source of mercury in this population and knowing that consumption of nutrient packed fish is beneficial for children’s physical and neurological development, low concentrations of mercury may reflect the absence or low level of fish consumption in the maternal diet. The mechanistic and clinical health implications of these associations with birth weight require further investigation.

1.1. Background

Elements and metals have been examined for their effects on numerous health outcomes. Increasingly non-essential and essential trace metals have been the focus of research examining their association with or effects on maternal-child health and specifically on birth outcomes. Adverse birth outcomes include pre-term birth, small for gestational age, intrauterine growth restriction, large for gestational age, low Apgar score, low birth weight, and macrosomia (Al-Saleh et al., 2014; Ashley-Martin et al., 2018; Ashrap et al., 2020; Baldewsingh et al., 2021; Baldewsingh et al., 2020a; Berky et al., 2023; Chen et al., 2014b; Gokoel et al., 2020; Hu et al., 2024; Hu et al., 2015; Issah et al., 2024; Kort et al., 2024a; Okubo and Nakayama, 2023; Rabito et al., 2014; Sabra et al., 2017; Takatani et al., 2022; Zilversmit Pao et al., 2019a; Zinia et al., 2023). Traditionally, studies have focused on how single trace elements affect birth outcomes. However, recent research is shifting to explore the combined effects of metal mixtures and the interactions between non-essential and essential trace elements on birth outcomes (Ashrap et al., 2020; Berky et al., 2023; Issah et al., 2024; Michael et al., 2022; Takatani et al., 2022).

Birth outcomes are important indicators of pediatric health and development and influence both mental and physical health as individuals age. For example, low birth weight is associated with substantially increased risks for current and future morbidity and mortality, growth inhibition, cognitive impairment, and a range of non-communicable diseases (Belbasis et al., 2016; Bianchi and Restrepo, 2021; Mathewson et al., 2017; Nakano, 2020; Zhang et al., 2023). Some non-essential trace elements, such as lead, have been consistently associated with lower birth weight (Berky et al., 2023; Dettwiler et al., 2023; Hu et al., 2024; Okubo and Nakayama, 2023; Rabito et al., 2014; Takatani et al., 2022; Zinia et al., 2023). The extensive body of independent studies linking higher maternal blood lead concentrations to reduced birth weight provides strong evidence that this relationship is not due to chance. Studies have also found that increasing maternal blood cadmium concentrations are negatively associated with infant birth weight and growth (Al-Saleh et al., 2014; Dettwiler et al., 2023; Okubo and Nakayama, 2023; Takatani et al., 2024; Zinia et al., 2023). Blood or hair mercury concentrations in pregnant women have shown mixed effects. Many studies report no association of mercury with birth weight, a few find a negative association, and emerging research suggests a positive association (Baldewsingh et al., 2020b; Berky et al., 2023; Chen et al., 2014b; Dack et al., 2023; Hu et al., 2024; Takatani et al., 2022). In some studies, lower concentrations of manganese are associated with adverse birth outcomes including low birth weight (Ashley-Martin et al., 2018; Chen et al., 2014a; Wood, 2009; Zota et al., 2009). Other studies find a positive benefit of manganese with birth weight (Takatani et al., 2022). These mixed findings may result from variations in exposure levels, elemental doses, and other study characteristics, all of which are important considerations.

The Republic of Suriname, located in northern South America, has a population of approximately 600,000 people. The country is racially and ethnically diverse, with approximately 60% of the population residing in the densely populated, urbanized coastal capital of Paramaribo. Another significant portion lives along the rest of the coastal plain, including the highly agricultural region of Nickerie. The remaining communities are situated in the largely undeveloped Amazonian rainforest interior. Recent and on-going artisanal small-scale goldmining activities, primarily in the interior regions, have contaminated freshwater systems and fish populations with mercury (Abdoel Wahid et al., 2023; Basu et al., 2015; Ouboter et al., 2018; Ouboter et al., 2012; Wickliffe et al., 2021). Most interior residents rely heavily on locally harvested freshwater fish as a primary source of dietary protein and essential nutrients, contributing to elevated mercury exposure in these regions. In addition, lead exposure is notably higher in the interior communities, likely due to other dietary practices and environmental sources (Rimbaud et al., 2017).

Previous research from our research group has found limited associations, either positive or negative, between non-essential trace elements and birth weight in a Surinamese cohort (Baldewsingh et al., 2020b; Gokoel et al., 2021; Gokoel et al., 2020; Kort et al., 2024b). In one study, focusing primarily on participants from the interior, we observed that pregnant women with low hair mercury concentrations had higher odds of delivering low-birth-weight babies. However, high mercury concentrations in hair were not associated with birth weight (Baldewsingh et al., 2020a). In another study with a larger cohort of pregnant participants, we found no association between mercury levels and birth weight or low birth weight (Gokoel et al., 2020). Importantly, these earlier studies only evaluated one metal at a time in relation to birth weight, and other metals or elements were not included.

Investigating the effects of complex metal mixtures, rather than individual elements, provides a deeper understanding of how multiple exposures interact to influence health outcomes by capturing synergistic or antagonistic effects that single-exposure analyses may miss (Billionnet et al., 2012). Metals can impact overlapping metabolic or cellular pathways, creating complex, nonlinear relationships between simultaneous exposures and developmental outcomes, such as birth weight (Issah et al., 2024). These relationships are challenging to model using traditional approaches like multiple linear regression, which requires specifying a large number of interaction terms a priori, often resulting in unstable estimates (Dominici et al., 2010). Bayesian Kernel Machine Regression (BKMR) offers a robust alternative by using a Gaussian kernel to flexibly model high-dimensional, non-additive dose-response relationships (Bobb et al., 2015). BKMR can be used to identify important interaction effects between metals in the mixture without the need to pre-specify these interactions in advance.

In this manuscript, we use BKMR to examine the complex relationships between a metal mixture and birth weight. We include a large sample of participants (n=757) from our Surinamese cohort of pregnant women and expand our scope to assess the influence of four metals on birth weight: three non-essential elements lead, mercury, and cadmium and one essential element, manganese. We also examined interactions among these four elements for evidence of more complex associations with birth weight.

1.2. Methods

1.2.1. Population

Mother/child dyads (n=757) were selected from the Caribbean Consortium in Environmental and Occupational Health (CCREOH) cohort, an environmental epidemiological study conducted in the Republic of Suriname, South America, which included 1,194 pregnant women in the total cohort (Zijlmans et al., 2020). To arrive at the final sample size of n=757, mother/child dyads were required to have complete data on blood levels of cadmium, manganese, mercury and lead, as well the child’s birth weight. Inclusion also required that we had complete data on important covariates used in the mixture models. Concentrations of trace elements were measured in blood samples collected in early pregnancy, either late in the first trimester or early in the second trimester. Two laboratories participated in analyzing blood samples for these elements. More detail on these two laboratories will be presented below. Interlaboratory comparisons indicated there were no differences across the analytical results. For cases where data were available from both labs, records from the Wadsworth lab were used to ensure consistency.

This study was approved by the Institutional Review Board (IRB) of Tulane University (839093) and the Medical Ethical Commission of Suriname’s Ministry of Health (VG 023-14). All included women (18+) provided written informed consent, and assent was obtained from participants who were 16 or 17 years old. Clinical trial number: not applicable.

1.2.2. Analytical Methods in Maternal Blood Samples

Wisconsin State Laboratory of Hygiene

Samples were stored at <−20°C prior to preparation. Whole blood metal analysis was conducted at the Wisconsin State Laboratory of Hygiene (WSLH) in a Trace Element Clean Lab (TECL). This facility, covering approximately 3,000 square feet, is designed for ultra-low-level elemental analysis and operates as an ISO 6 clean room or better. The TECL suite minimizes metal contamination using HEPA-filtered air, routine monitoring of air-borne particles, and acid precleaning of materials that contact samples. Acid handling is conducted in a separate room to avoid instrument corrosion from acid fumes.

The samples underwent digestion in acid-washed polypropylene containers. The blood was pipetted into tared vials and the masses collected to determine volume. Ultrapure nitric acid and hydrogen peroxide were added, and the samples were allowed to pre-digest at room temperature overnight. The following day, ultrapure hydrochloric acid was added, and the samples were heated to 85°C in a convection oven for 2 hours. The samples were cooled and diluted with ASTM Type 1 reagent water.

Elemental analysis was performed using a Thermo Element XR magnetic sector Inductively Coupled Plasma Mass Spectrometer (ICP-MS), based on EPA 200.8 analytical method. Various isotopes and resolution settings were employed to measure target elements and confirm the absence of isobaric interferences. Counts were normalized to internal standards (gallium, indium, and bismuth), and results were blank-corrected using batch-specific method blanks. Propagated uncertainty was calculated using the root-sum-square of three components: (a) SF-ICP-MS analysis (standard deviation of triplicate analyses), (b) blank subtraction (standard deviation of four method blanks), and (c) digestion recovery (long-term standard deviation of standard reference materials and spike recoveries).

Wadsworth Center New York State Department of Health

Blood specimens were shipped on dry ice to the Laboratory of Inorganic and Nuclear Chemistry at the Wadsworth Center, New York State Department of Health (Albany, NY). Specimens were prepared for analysis in a SterilGARD^® e3 Class II, Type A2 Biological Safety Cabinet (The Baker Company, Sanford, MA) which meets ISO 5 (Class 100) standards. All other preparation work was performed under ISO 5 (Class 100) clean room conditions or better (Terra Universal, Fullerton, CA). Trace element analysis was carried out on a Thermo Scientific iCAPTM TQ triple quadrupole inductively coupled plasma mass spectrometer (ICP-MS/MS), using an analytical method validated for lead, cadmium, mercury, and manganese, and optimized for human biomonitoring studies. Manganese in blood by quadrupole ICP-MS can be affected by the presence of high levels of ⁵⁶Fe (in addition to various polyatomic interferences which include ³⁹K¹⁶O and ⁴⁰Ar¹⁵N) (Praamsma et al., 2011). In this study, the interference from high levels of Fe on Mn was minimized by operating the ICP-MS/MS in the standard (non-CCT) mode with a 1 amu Q1 setting, and a higher mass analyzer (Q3) resolution of 0.3 amu at m/z 55, rather than the conventional 0.7 amu, which was used for other target analytes. Multielement calibration standards were prepared from a NIST-traceable (i.e., National Institute of Standards and Technology) stock solution (High Purity Standards, Charleston, SC, USA) and an eight-point, base-blood matrix-matched calibration curve used for each element (1+49 dilution of blood sample to diluent) consistent with the laboratory’s well-established biomonitoring method for trace elements in whole blood (Palmer et al., 2006). Whole blood specimens were thawed to room temperature and each one carefully homogenized using an indirect sonication system (UP200st ultrasonic transducer and Vial-Tweeter Sonotrode, Hielscher Ultrasonics GmbH). The ultrasonic system was optimized using an amplitude of 85% and a continuous pulse. Quality control (QC) materials and archived proficiency testing (PT) samples were homogenized by ultrasonication prior to sampling for analysis. Method limits of detection (LODs) based on a blood matrix were calculated per IUPAC guidelines, and were 0.07 μg/dL for lead, 0.07 μg/L for cadmium, 0.20 μg/L for mercury, and 2.0 μg/L for manganese. Four concentration levels of internal QC blood materials were analyzed at the start, end, and throughout each analytical run. Method repeatability (i.e., between-run precision or CV) for the low QC level was 5.3% for manganese at 7.3 μg/L; 2.4% for cadmium at 0.86 μg/L; 7.2% for mercury at 0.90 μg/L; and 2.3% for lead at 1.06 μg/dL.

The Wadsworth laboratory participates successfully in numerous PT programs and external quality assessment (EQA) schemes for trace elements including those operated by: the New York State Department of Health, Albany, NY (NYSDOH); L’Institut National de Santé Publique du Québec, Le Centre de Toxicologie du Québec (CTQ); the Centers for Disease Control and Prevention, Lead, Cadmium, Manganese and Mercury PT program (LAMP); the Trace Elements External Quality Assessment Scheme, at the University of Surrey, UK; and the German External Quality Assessment Scheme, operated by the Institute and Outpatient Clinic for Occupational, Social and Environmental Medicine of the Friedrich-Alexander University, Erlangen-Nuremberg, Germany. Each analytical run was monitored using four levels of internal QC materials, plus two additional levels of blood-based reference materials validated for the Children’s Health and Exposure Analysis Resource (CHEAR), which contains trace element concentrations that cover the clinical range for each target analyte. Method accuracy is periodically assessed using NIST SRM 955c (Toxic Metals in Frozen Caprine Blood), and NIST 955d (Toxic Elements and Metabolites in Frozen Human Blood). Complete details of the internal QC data, PT performance data from NYSDOH, and CTQ are provided in supplementary Tables S1 and S2.

1.2.3. Analysis of Participant Characteristics

Demographic characteristics, metal exposures, and birth weight of children in the study were summarized using medians, interquartile ranges, and ranges for continuous variables, as well as counts and percentages for categorical variables. Distributions of continuous variables were graphically examined using histograms and kernel density plots. Metal concentrations were natural log-transformed and then standardized for all analyses (Gibson et al., 2019). Pairwise Spearman correlations were computed for all standardized, natural-log-transformed metal concentrations, and variance inflation factors (VIFs) were used to evaluate multicollinearity among these standardized log-transformed metal exposures.

1.2.4. BKMR Mixture Modeling

We conducted mixture analyses to assess the effects of cadmium, manganese, mercury, and lead exposures on birth weight. These metals were selected based on prior evidence suggesting associations between their exposure and variations in birth weight (Issah et al., 2024; Michael et al., 2022; Zinia et al., 2023). BKMR modeling was chosen for its capability to flexibly model complex, high-dimensional, nonlinear, and non-additive exposure-response relationships. BKMR employs a Gaussian kernel, shrinking the estimated mixture effects among participants with comparable exposure patterns towards one another.

The BKMR model is specified as: $Y_i=h({\mathit{z}}_i)+{\mathit{x}}_i^{\prime }\beta +{ϵ}_i$, where the error terms, ${ϵ}_i$, are independent and normally distributed with mean 0 and a constant variance. In our analysis, ${\mathit{z}}_i$ denoted the vector of log-transformed and standardized metal concentrations for participant $i$. The function $h({\mathit{z}}_i)$ represents the complex exposure-response function that related the mixture of metals to each child’s birth weight, $Y_i$. This function was modeled using a Gaussian kernel. The vector ${\mathit{x}}_i^{\prime }$ denoted the set of baseline covariates that were adjusted for in the model using reference-cell coding. The included baseline covariates were household income, maternal education level, maternal age, ethnicity, preterm birth status, and the laboratory where samples were analyzed. Weakly informative inverse-Gamma priors were specified for the variance components of the model, and a uniform prior distribution with a range of 0 to 100 was specified for the Gaussian kernel weights that controlled the smoothness of $h(\cdot )$.

The BKMR analysis was conducted using 120,000 posterior samples drawn from four separate MCMC chains with random initialization. After discarding the first 15,000 samples from each chain as burn-in, 60,000 posterior samples remained for inference. Convergence was evaluated visually using trace plots and numerically using Gelman-Rubin diagnostics. Given that the posterior sampling algorithm returns samples that contain some autocorrelation, effective sample sizes were calculated to ensure that the posterior samples drawn corresponded to a sufficient number of independent posterior samples. Component-wise variable selection was not applied so that the effects of all four components of the mixture were included in the joint effects.

1.3. Results

1.3.1. Population Characteristics

Demographics for the pregnant women and their children in this study are presented in Table 1. Preterm birth was higher, at 15.2%, in the total cohort compared with 12.0% in this subset of study participants.

Table 1.

General demographics of the sample of pregnant women and their children in the study. Included are other characteristics such as preterm birth and analytical laboratory performing blood element analyses.

	Overall (N=757)
Birth weight (grams)
Median (IQR)	3090 (2820-3370)
Age (years)
16–19	85 (11.2%)
20–34	544 (71.9%)
35+	128 (16.9%)
Ethnicity
African descent	292 (38.6%)
Asian descent	209 (27.6%)
Indigenous	110 (14.5%)
Other	146 (19.3%)
Education level
None or primary	190 (25.1%)
Secondary or tertiary	567 (74.9%)
Household income (SRD)
<3,000	525 (69.4%)
3,000+	232 (30.6%)
Place of Residence
Paramaribo	431 (56.9%)
Nickerie	164 (21.7%)
Interior	162 (21.4%)
Gestational age at birth < 37 weeks
Yes	91 (12.0%)
No	666 (88.0%)
Lab
Wadsworth	520 (68.7%)
WSLH	237 (31.3%)

Demographics for the pregnant women in this study categorized by low and normal birth weight are presented in Table 2. Low birth weight was higher, at 13.2%, in the total cohort compared with 10.2% in study participants.

Table 2.

Demographic and other study variables summarized for babies who are born at birthweights of < 2500 grams and ≥ 2500 grams.

	Birth weight < 2500 grams (N=77)	Birth weight 2500 ≥ grams (N=680)
Age (years)
16–19	8 (10.4%)	77 (11.3%)
20–34	58 (75.3%)	486 (71.5%)
35+	11 (14.3%)	117 (17.2%)
Ethnicity
African descent	30 (39.0%)	262 (38.5%)
Asian descent	22 (28.6%)	187 (27.5%)
Indigenous	10 (13.0%)	100 (14.7%)
Other	15 (19.5%)	131 (19.3%)
Education level
None or primary	20 (26.0%)	170 (25.0%)
Secondary or tertiary	57 (74.0%)	510 (75.0%)
Household income (SRD)
<3,000	56 (72.7%)	469 (69.0%)
3,000+	21 (27.3%)	211 (31.0%)
District
Paramaribo	49 (63.6%)	382 (56.2%)
Nickerie	17 (22.1%)	147 (21.6%)
Interior	11 (14.3%)	151 (22.2%)
Gestational age at birth < 37 weeks
Yes	39 (50.6%)	52 (7.6%)
No	38 (49.4%)	628 (92.4%)
Lab
Wadsworth	49 (63.6%)	471 (69.3%)
WSLH	28 (36.4%)	209 (30.7%)

1.3.2. Elemental Exposures

Metal blood concentrations for the 4 metals in this study are presented in Table 3. Additional information regarding the concentrations of these metals and relationships with other characteristics of the cohort can be found in previous publications (Abdoel Wahid et al., 2023; Wickliffe et al., 2021; Zijlmans et al., 2020).

Table 3.

Median blood metal concentrations in pregnant women in this study. Interquartile ranges and total ranges are also provided.

Element	Median concentrations (IQR)	Range
Mercury (μg/L)	3.06 (1.86-6.29)	(0.185-73.4)
Manganese (μg/L)	12.9 (10.3-16.7)	(5.10- 52.20)
Cadmium (μg/L)	0.18 (0.13-0.26)	(0.02-1.338)
Lead (μg/dL)	1.85 (1.21-3.59)	(0.390-28.6)

Concentrations of these metals with regards to low birth weight and normal birth weight are presented in Table 4.

Table 4.

Median (IQR) concentrations of metals in pregnant women as a function of birth weight categorized as low birth weight or normal birth weight.

Element	Birth weight < 2500 grams (N=77)	Birth weight ≥ 2500 grams (N=680)
Mercury (μg/L)	2.59 (1.73-5.20)	3.13 (1.87-6.50)
Manganese (μg/L)	14.4 (11.7-18.3)	12.8 (10.2-16.4)
Cadmium (μg/L)	0.21 (0.14-0.27)	0.18 (0.12-0.26)
Lead (μg/dL)	1.88 (1.23-3.82)	1.83 (1.21-3.55)

Spearman correlation coefficients among metals in blood samples from pregnant women in this study are presented in Table 5. Lead and mercury are modestly positively correlated in this study, while all other correlations are weak. Furthermore, VIFs from models including the main effect of each standardized and log-transformed metal concentration indicated no evidence of multicollinearity, as the highest VIF was approximately 2.

Table 5.

Pairwise Spearman correlations between standardized, natural log-transformed metal concentrations.

Metals	Lead	Mercury	Manganese	Cadmium
Lead	1.000	0.561*	0.181*	−0.153*
Mercury		1.000	0.078	−0.319*
Manganese			1.000	0.214*

^*significant after adjusting for multiple comparisons using a Bonferroni correction. Family-wise error rate of alpha = 0.05.

Neither cadmium nor manganese exhibited an association with birth weight in this study (Figures 1 and 2). In contrast, lead demonstrated an inverse relationship with birth weight, where higher concentrations were associated with lower birth weight when the remaining metals were fixed at their median concentrations. Conversely, increasing concentrations of mercury were associated with increasing birth weights when the remaining metals were fixed at their median concentrations (Figure 2). Mercury displayed a non-linear relationship with birth weight. At lower concentrations, mercury was associated with lower birth weight, but as concentrations increased, birth weight also increased. Thus, depending on the concentrations of the remaining metals, mercury exhibited either an upside-down U shape or an increasing relationship with birthweight (Figure 2). Additionally, the positive relationship between mercury and birth weight was strongest at higher lead concentrations (e.g., the 75^th percentile, or 3.59 μg/L) when the remaining metals were held at their median levels.

Figure 1.

Univariate exposure-response curves showing the relationship between the metal and birth weight while other metals are fixed at their median levels. The y-axis illustrates the impact on birth weight in grams, adjusted for household income, maternal education level, maternal age, ethnicity, preterm birth status, and laboratory effects.

Figure 2.

Pairwise effects of two metals on birth weight, adjusted for household income, maternal education level, maternal age, ethnicity, preterm birth status, and laboratory effects, while holding other metals constant at their median levels. Curves that are not parallel or that cross indicate potential interaction effects.

Cadmium and manganese also did not show notable interactions with other metals in this study (Figure 2). The association of higher concentrations of lead with lower birth weight was evident and lead and mercury appear to interact in relation to birth weight. Specifically, the negative effects of lead on birth weight were more pronounced at lower mercury concentrations (e.g., the 25th percentile, or 1.86 μg/L), where increasing lead levels led to the steepest declines in birth weight. However, at higher mercury concentrations (e.g., the 75th percentile, or 6.29 μg/L), the negative effects of lead were more moderate, suggesting that mercury may attenuate the impact of lead on birth weight. Additionally, the positive relationship between mercury and birth weight was strongest at higher lead concentrations (e.g., the 75^th percentile, or 3.59 μg/L) when the remaining metals were held at their median levels. This suggests that elevated mercury levels or aspects associated with elevated mercury levels may mitigate some of the adverse effects of lead on birth weight. Notably, the combination of high lead and low mercury concentrations was associated with the most significant reduction in birth weight.

Figure 2 illustrates interactive effects between each metal and the remaining metals on birth weight. For a metal of interest, the point estimates and 95% credible intervals correspond to the change in birth weight when changing the metal from its 25^th percentile to 75^th percentile while holding all the other metals constant at a given quantile (adjusted for covariates). When the remaining metals were at high concentrations (75^th percentile), mercury exhibited a notable positive effect on birth weight, with a point estimate of almost 100 grams. Table 6 quantifies these interaction patterns using a difference-of-differences approach. Specifically, the estimated effect of a metal at its 75th percentile minus its effect at its 25th percentile was calculated while fixing all remaining metals at either their 75th or 25th percentiles. The difference between these two estimates captures interaction patterns. These correspond to subtracting the estimates in red on Figure 3 from the estimates in blue.

Table 6.

The interaction effects were quantified by subtracting the difference in $h(\cdot )$ between the 75th and 25th percentiles of the metal, calculated while all other metals were fixed at their 25th percentile from their 75th percentile. CI: 95% credible interval based on the 2.5^th and 97.5^th percentiles of posterior distribution.

Metals	Estimate	Lower CI	Upper CI
Lead	74.32	−68.95	217.58
Mercury	96.98	−47.48	241.42
Manganese	1.58	−66.26	69.41
Cadmium	1.32	−62.03	64.67

Figure 3.

Contribution of individual metals to $h(\cdot )$ going from their 25^th percentile to their 75^th adjusted for household income, maternal education level, maternal age, ethnicity, preterm birth status, and laboratory effects, while holding other metals constant at their 25^th, 50^th, and 75^th percentiles. Point estimates further from zero indicate larger contribution to $h(\cdot )$ at the given percentiles.

The large positive point estimates for lead and mercury indicate that increases in their concentrations were linked to larger positive changes in birth weight when the remaining metals were fixed at their 75th percentile compared to their 25th percentile. This finding suggests there may be interactive effects between lead and the remaining metals, as well as between mercury and the remaining metals possibly driven by the lead-mercury interaction. Although the 95% credible intervals for these estimates include zero, the posterior probabilities provide additional insight. Specifically, the probability that the interaction effect of lead is greater than zero is 75.9% and the probability that the mercury interaction effect exceeds 0 is 83.6%. In contrast, the point estimates for the interaction parameters were close to zero for cadmium and manganese, with posterior probabilities approximately equal for positive and negative effects. This suggests minimal evidence of interaction effects involving cadmium or manganese in relation to birth weight given the data.

1.4. Discussion

Previous studies have explored the relationships between birth weight and adverse birth outcomes and exposure to several essential and non-essential elements (Ashley-Martin et al., 2018; Ashrap et al., 2020; Baldewsingh et al., 2020b; Berky et al., 2023; Chen et al., 2014b; Dettwiler et al., 2023; Gokoel et al., 2021; Gokoel et al., 2020; Hu et al., 2024; Hu et al., 2015; Issah et al., 2024; Kort et al., 2024b; Michael et al., 2022; Okubo and Nakayama, 2023; Perng et al., 2019; Rabito et al., 2014; Sabra et al., 2017; Takatani et al., 2022; Takatani et al., 2024; Zilversmit Pao et al., 2019b; Zinia et al., 2023). Lead, mercury, and cadmium have shown various effects on birth weight (Berky et al., 2023; Hu et al., 2024; Issah et al., 2024; Rabito et al., 2014; Zinia et al., 2023). Consistent with prior findings, our study observed that increasing maternal blood lead levels are associated with decreasing birth weight. The effects of mercury on birth weight have been mixed. Some studies have found a negative association between mercury and birth weight, but some studies, including this one find a more complex relationship (Berky et al., 2023; Hu et al., 2024).

We found no apparent relationship between blood concentrations of cadmium and manganese in these Surinamese pregnant women and birth weight. The concentrations of cadmium in these women are quite low and tends to be lower than in those studies finding a negative association with birth weight. This may explain the lack of an apparent impact or interaction with the other metals being evaluated. Manganese concentrations in these pregnant women were within a normal range (4-15 μg/L in whole blood as reported in Ross et al), neither deficient nor excessive, so it was expected that manganese alone would not significantly influence birth weight in this population (Ross et al., 2014). Additionally, no meaningful interactions between manganese and other metals were observed in this study.

Lead at increasing concentrations was associated with lower birth weight. This supports the findings of previous studies finding a negative association of lead with infant birth weight (Berky et al., 2023; Hu et al., 2024). Lead exposure in pregnant women in Suriname likely comes from food and the diet, but other sources may exist (e. g. personal care products). Identifying the source(s) of the exposure should allow for the development of possible interventions to reduce or mitigate these exposures. Lead was not included in previous analyses of birth weight in this cohort, so this study represents additional and confirmatory evidence of the detrimental and inverse relationship lead has with children’s birth weight.

Mercury at lower blood concentrations was associated with lower birth weight on average. This finding may seem counterintuitive given its classification as a non-essential, toxic metal. Further, we are not suggesting that mercury itself plays any direct positive role. However, mercury exposure primarily originates from dietary sources, particularly fish consumption, which is also the case in Suriname. Fish is a highly nutritious food, providing numerous essential elements, fatty acids, and proteins. Therefore, low blood concentrations of mercury may serve as a proxy for a less nutritious diet, which may be a factor in lower birth weight. Poor maternal diet and limited dietary diversity during pregnancy are, with a few exceptions, associated with lower birth weight outcomes (Abu-Saad and Fraser, 2010; Chia et al., 2019; da Mota Santana et al., 2021; Englund-Ögge et al., 2019; Imdad and Bhutta, 2012; Kheirouri and Alizadeh, 2021; Lechtig et al., 1975).

Fish consumption provides beneficial elements such as selenium and polyunsaturated fatty acids, which support fetal growth and development. Up to a certain threshold, mercury from fish may not be harmful, and low blood mercury levels may indicate a diet deficient in vital nutrients. In fact, studies have linked maternal diets lacking in fish and seafood consumption to lower birth weight (Brantsæter et al., 2012; Englund-Ögge et al., 2019; Leventakou et al., 2014; Nykjaer et al., 2019; Wei et al., 2023). Previous research on this cohort, which measured mercury in maternal hair, also found evidence of an association between low concentrations of hair mercury and lower birth weight (Baldewsingh et al., 2020a). The larger study reported here, which examined blood mercury concentrations, supports that initial finding with hair mercury. Since methylmercury comprises the majority of both hair and blood total mercury concentrations in this cohort, results may reflect effects associated with this organic form of mercury (Wickliffe et al., 2021).

Our findings do indicate that the positive influence of mercury on birth weight plateaus and may begin to decline at higher blood concentrations, suggesting a potential toxic effect at elevated levels, particularly when lead levels are low. At higher blood mercury concentrations, we observed a slight decline in birth weight which was also influenced by lead concentrations, reinforcing the notion that the benefits of a diet rich in seafood but also mercury may be counterbalanced by mercury toxicity at excessive levels.

We also found that lead and mercury interact to further impact birth weight. Our analyses indicate that birth weight is more impacted and declines markedly when blood lead concentrations are high and blood mercury concentrations are low. This may be a function of mercury serving as a proxy for a healthy vs. a less healthy diet; low concentrations of mercury indicate a diet low in important developmental nutrients and higher lead concentrations thus have a more pronounced, negative impact on infant birth weight.

These results highlight the importance of understanding not only single metal exposures but also the combined effects of multiple metal exposures and the broader nutritional and environmental context. Use of BKMR mixture modeling techniques shows promise for disentangling these complex interactions to inform effective interventions and public health strategies.

1.5. Conclusions

By using Bayesian Kernel Machine Regression, a method designed to capture the complex relationships between a mixture of environmental exposures and a health outcome, we uncovered a complex interactive effect between prenatal exposure to lead and mercury and birth weight among the babies born in this cohort. Traditionally, studies in the field of environmental epidemiology have examined the contribution of metals to birth weight and other health outcomes using a “one-at-a-time” approach. This approach involves examining the impact of a single metal (e.g. lead) on birth weight by including only one metal at a time in the respective regression model, rather than considering the mixture of metals a baby may be exposed to prenatally. The “one-at-a-time” approach fails to consider the correlated nature of the metals, typically treats the effects of the metal concentrations (or log metal concentrations) on birth weight as linear and cannot identify potential effect modification that one metal may have on another metal’s effect. While non-linear relationships and interactions can be modeled using traditional methods such as splines, generalized additive models (GAMs), or interaction terms in regression, these approaches become increasingly unwieldy in mixture settings due to the need to pre-specify functional forms (e.g., number and placement of knots for splines) and a large number of potential interaction terms, which can lead to model overfitting and reduced interpretability. The BKMR approach we have utilized has numerous advantages. First, the full set of metals is considered within the same model, allowing researchers to identify interactive effects that subsets of metals have with one another. It also allows researchers to uncover nonlinear dose-response relationships between metals and birth weight. Each of these advantages were highlighted in this research study.

Lead negatively impacts birth weight in this study, especially when paired with the lower concentrations of the remaining elements. Overall, increasing levels of mercury were associated with increasing birth weight, up to a plateau or slight downward trend, depending on the values of the remaining metal concentrations. These effects have been observed in previous studies. Lead at higher concentrations coupled with mercury at lower concentrations appear to interact to have a more substantial negative impact on birth weight. This interaction further suggests that mercury or perhaps a healthy diet rich in fish, may moderate the negative effects of lead on birth weight. There remains a need to identify the source(s) of lead exposure and reduce or eliminate those when and where possible.

Despite the interesting influence of mercury on birth weight we and others have observed, we stand firm behind current public health efforts to reduce or remove mercury and methylmercury from the human diet. Methylmercury is a known neurotoxicant and can impact neurological and mental development and health. Removing mercury from the environment in Suriname, as in many other areas, is an actionable goal that should be achieved to mitigate such negative, irreversible impacts. Suriname has signed on to the Minamata Convention so efforts should be implemented to reduce or eliminate the use of elemental mercury in the gold mining sector. Further, Suriname in signing on to the convention has agreed to develop a national action plan with a goal of reducing and eliminating mercury use in the ASGM sector which would markedly reduce the amount of elemental mercury entering the environment, ecosystems, and human food chain.

Though we observed negative impacts on birth weight with respect to lead and the combination of lead and mercury, we did not examine other physical and neurodevelopmental effects associated with lower birth weight in this study. It is possible that despite having a lower birth weight, such children may continue to develop normally. The clinical significance of these apparent effects on lower birth weight should be examined to determine what, if any, physical or mental impacts are influenced by this outcome. Furthermore, future studies may wish to collect information on gestational age at the time of blood collection. While not collected in this study, it may be an important covariate to consider in future work. Gestational age at the time of blood collection could also be useful in identifying critical exposure windows during pregnancy. Exploring differences in the effects of the metal mixture by child’s sex is also another future direction. Applying BKMR to subgroups while ensuring a meaningful comparison is not straightforward, as the quantiles across subgroups are not directly comparable and there is no clear way to test for differences in effects. This would also involve stratifying the sample, so studies with larger sample sizes may be best positioned to study potential effect modification by sex. Another important contribution to studies such as these would be the robust collection of dietary information which would serve to inform the nutritional aspects of the maternal diet as well as the source of potential environmental risks that may influence maternal and child health. The lack of dietary intake of fish among our pregnant women is a limitation of this study. Future work should include the robust collection of individual fish consumption data for pregnant mothers to better match this with the observed relationship with mercury and low birth weight in this study and perhaps studies in similar populations. In addition, physiological and toxicological mechanisms by which lead and mercury influence birth weight both alone and in combination remain to be determined.

Highlights.

• Exposure to lead and mercury was highest in the interior, Amazonian region.

• Lead exposure was associated with negative impacts on birth weight.

• Mercury was nonlinearly associated with negative impacts on birth weight.

• Lead and mercury interacted in an association with negative impacts on birth weight.