Learn more: PMC Disclaimer | PMC Copyright Notice
Cord blood DNA methylation of DNMT3A mediates the association between in utero arsenic exposure and birth outcomes: Results from a prospective birth cohort in Bangladesh
Associated Data
Abstract
Background:
Fetal epigenetic programming plays a critical role in development. DNA methyltransferase 3 alpha (DNMT3A), which is involved in de novo DNA methylation (DNAm), is a prime candidate gene as a mediator between prenatal exposures and birth outcomes. We evaluated the relationships between in utero arsenic (As) exposure, birth outcomes, and DNMT3A DNAm.
Methods:
In a prospective Bangladeshi birth cohort, cord blood DNAm of three DNMT3A CpGs was measured using bisulfite pyrosequencing. Maternal toenail As concentrations at birth were measured to estimate in utero exposure. Among vaginal births (N = 413), structural equation models (SEMs) were used to evaluate relationships between DNMT3A methylation, log2 (toenail As), birth weight, and gestational age.
Results:
In an adjusted SEM including birth weight and gestational age, maternal toenail As levels were associated with DNMT3A DNAm (B = 0.40; 95% CI: 0.15, 0.66) and gestational age (B = −0.19 weeks; 95% CI: −0.36, −0.03). DNMT3A DNAm was associated with gestational age (B = −0.10 weeks; 95% CI: −0.16, −0.04) and birth weight (B = −11.0 g; 95% CI: −21.5, 0.4). There was a significant indirect effect of As on gestational age mediated through DNMT3A DNAm (B = −0.04; 95% CI: −0.08, −0.01), and there were significant indirect effects of maternal toenail As levels on birth weight through pathways including gestational age (B = −14.4 g; 95% CI: −29.2, − 1.9), DNMT3A DNAm and gestational age (B = −3.1 g; 95% CI: −6.6, −0.8), and maternal weight gain and gestational age (B = −5.1 g; 95% CI: −9.6, −1.5). The total effect of a doubling in maternal toenail As concentration is a decrease in gestational age of 2.1 days (95% CI: 0.9, 3.3) and a decrease in birth weight of 29 grams (95% CI: 14, 46).
Conclusions:
DNMT3A plays a critical role in fetal epigenetic programming. In utero arsenic exposure was associated with greater methylation of CpGs in DNMT3A which partially mediated associations between prenatal As exposure and birth outcomes. Additional studies are needed to verify this finding.
INTRODUCTION
Chronic exposure to arsenic (As) persists in many regions of the world. In Bangladesh, approximately 40 million individuals rely on household drinking water with As concentrations exceeding the World Health Organization (WHO) guideline of 10 µg/L, half of which are also above the Bangladesh standard 50 µg/L (Bangladesh Bureau of Statistics and United Nation Children’s Fund, 2015). Inorganic As and As metabolites readily pass the placenta, resulting in a high correlation between As concentrations measured in maternal and cord blood with subsequent fetal exposure (Concha et al., 1998; Hall et al., 2007). Arsenic is an established human toxicant and group 1 carcinogen (World Health Organization, 2011), and maternal As exposure during fetal development has been linked to increased risk of adverse health outcomes later in life including cancers of the lung, bladder, liver, and larynx; cardiovascular disease; and reduced lung function (Farzan et al., 2013; Vahter, 2009). In utero and early life exposure has been associated with increased childhood morbidity (Farzan et al., 2013; Rahman et al., 2017).
The teratogenic effects of As have been well established in rodent models (Ferm and Hanlon, 1985; Hill et al., 2008; Hood, 1972; Hood and Bishop, 1972; Kozul-Horvath et al., 2012; Moore et al., 2019; Morrissey and Mottet, 1983; Nagymajtényi et al., 1985). However, findings from epidemiological studies of the association between in utero As exposure and birth outcomes in humans have been inconsistent, possibly due to differences in study design, exposure assessment, sample size, population studied, and level of exposure. Considering the cumulative evidence, a recent review by Milton et al. found an insufficient number of studies addressing neonatal death and preterm birth but “consistent and convincing evidence” of the association between high As exposure and increased risk of spontaneous abortion, stillbirth, and low birth weight (Milton et al., 2017). Additionally, a meta-analysis conducted by Zhong et al. found a negative association between in utero As exposure and birth weight (Zhong et al., 2019).
Numerous epidemiological studies have examined the association between in utero As exposure and birth weight, shortened gestation, and/or intrauterine grown restriction. These outcomes have been shown to have varying effects on disease risk later in life (Wardlaw et al., 2004). Shortened gestation is associated with increased risk of infant mortality, morbidity, and disability, whereas restricted intrauterine growth is associated with decreased growth in childhood and increased morbidity in later in life. Several studies have evaluated birth weight and gestational age as independent outcomes associated with in utero As exposure, finding no significant association with either outcome, particularly at low levels of exposure (Bloom et al., 2016; Freire et al., 2019), or a significant negative association with both outcomes (Xu et al., 2011). Mediation analyses including birth weight and gestational age may help to understand the health effects of in utero As exposure. For example, in this cohort our group found that the in utero As exposure was related to gestational age and maternal weight gain during pregnancy, and these factors mediated the effect between As and birth weight (Kile et al., 2015). Subsequently, there was no direct effect of As on birth weight in this cohort.
Changes in the epigenome, including DNA methylation (DNAm), may be one mechanism underlying the associations between As exposure and multiple health outcomes (Bailey et al., 2016; Bjørklund et al., 2018). As explained by the developmental origins of health and disease hypothesis, environmental exposures during embryogenesis, a time of cellular differentiation and epigenetic reprogramming, may result in epigenetic dysregulation and increased disease risk (Heindel and Vandenberg, 2015). The DNA methyltransferases 3 alpha (DNMT3A) and 3 beta (DNMT3B) are responsible for de novo DNAm in embryonic cells (Okano et al., 1999). DNMT3A and DNMT3B have differential spatial and temporal expression during early embryogenesis (Uysal et al., 2017). DNMT3A may have a more significant role in maintaining global DNAm, particularly at distal promotors and non-CpG sites (Gu et al., 2018), and in imprinting (Kaneda et al., 2004). DNMT3A is also involved in RNA polymerase binding and transcription, genetic imprinting, mitotic cell cycle, protein binding, aging, and regulation of cell death. Although the role of DNMT3A in birth outcomes has not fully been established, DNAm within the gene body of DNMT3A has previously been associated with reduced gestational age in a Norwegian birth cohort (pFDR = 0.03) (Bohlin et al., 2016).
Due to the critical role of DNMT3A in establishing de novo DNAm, we hypothesized that DNAm of this gene could mediate the association between As exposure occurring in the critical prenatal development phase and adverse birth outcomes, specifically birth weight and gestational age. We also hypothesized that gestational age would act as a mediator between in utero As exposure, epigenetic changes, and birth weight. We explored this hypothesis using structural equation models (SEMs) with our a priori hypothesis depicted in Figure 1.
METHODS
Study Population
Participants were recruited as part or prospective birth cohort of women exposed to As through drinking water in Bangladesh in the Sirajdikhan and Pabna Sadar upazilas. This cohort has been described in detail previously (Kile et al., 2014). Women were recruited at ≤16 weeks gestations age by Dhaka Community Hospital (DCH) Trust (N = 1,458). Eligibility criteria were having a single pregnancy, using a tube well as the main source of drinking water, and planning to remain in the current residence for the duration of the pregnancy. Women received free prenatal care provided by DCH, as well as prenatal vitamins, which were given to participants during monthly home visits. Women gave birth with DCH-trained medical personnel at a local clinic or at their home who collected all birth outcome data and cord blood was collected at birth. A sub-sample of 569 newborns from the cohort were selected for measurement of DNAm in whole cord blood using pyrosequencing. This sub-sample was randomly selected across a range of arsenic exposure that was measured in the household’s drinking water. We also restricted analyses to only vaginal births to minimize the possibility of confounding which limited the final sample size to 413 newborns.
Ethics
The study protocol was approved by the Human Research Committees at the Harvard School of Public Health, Oregon State University, and DCH Trust. Prior to enrollment in the study, informed consent was obtained from all participants.
Laboratory measures
Maternal toenail As:
Arsenic concentrations in maternal toenail samples were measured as a biomarker of exposure during pregnancy. Arsenic has an affinity for sulfhydryl groups and accumulates in scleroproteins, and therefore toenail As concentrations at the distal tip represents a time-integrated measure of ingested As during the previous 9 to 12 months of nail growth and has been shown in previous research to be highly correlated with water arsenic concentrations (Karagas et al., 1996; Rodrigues et al., 2015). Toenail clippings were collected from mothers at the time of enrollment (e.g. maternal toenails at enrollment) and within one month of delivery (e.g. postpartum maternal toenails). The methods used to measure arsenic exposure have been previously published (Rodrigues et al., 2015). Briefly, samples were sonicated in 1% Triton X-100 solution (Sigma-Aldrich, Inc., St. Louis, MO) and rinsed in Milli-Q water (Millipore Corporation, Billerica, MA) to remove external contamination. Then, nails were digested in Trace Select Ultra Pure nitric acid (HNO3; Sigma-Aldrich, Inc.) and diluted with Milli-Q water. Inductively coupled plasma mass spectrometry was used to measure total arsenic concentrations (Perkin-Elmer Model DRC-II 6100, Norwalk, CT). Arsenic concentrations of human hair references were used to correct for method error using black correction and normalization (CRM Hair; Shanghai Institute of Nuclear Research, Academia Sinica, China); human hair was selected as the reference due to lack of available toenail references. Eighteen postpartum toenail samples were missing. Samples were dropped from analyses if they had a mass < 5 mg (N = 1) or relative standard deviation > 25% (N = 3). One sample below the batch limit of detection (ranging 0.004 – 0.85 µg As/g) was also excluded.
Maternal drinking water As:
At the time of enrollment, participants were asked to identify their main source of drinking water (Kile et al., 2014). Water samples were collected, preserved with ultrapure nitric acid to a pH < 2, and stored at room temperature. Arsenic concentrations in water samples were measured using inductively coupled mass spectrometry (ICP-MS) with the US EPA method 200.8 (Environmental Laboratory Services, North Syracuse, NY) (Creed et al., 1994). Samples had an average percent recovery from plasmaCal multi-element QC standard #1 solution (SCP Science) of 102 ± 7%. Thirty samples below the LOD of 1 µg/L were replaced with LOD/2.
Bisulfite pyrosequencing:
DNAm was quantified for cg26544247 (GRCh37/hg19, chr2:25,473,782) located on the north shore of a CpG island (chr2:25474757–25475598) within the gene body of DNMT3A. These CpG sites were selected as the target site for bisulfite pyrosequencing because they were included on the Illumina Infinium HumanMethylation450 BeadChip. Our rationale for this choice was to help make the pyrosequencing data in this study relevant to other methylation array data. Coverage also included two downstream CpG sites (chr2:25,473,813 and chr2:25,473,843). Bisulfite pyrosequencing was performed at EpigenDx (Hopkinton, MA) using 20 ng/ul whole cord blood DNA. Bisulfite conversion was used to convert unmethylated cytosines to uracil. Following PCR amplification and direct pyrosequencing, the average percent DNAm was calculated. Control samples with low methylation, medium methylation, and high methylation were included on each plate.
Covariates:
Sociodemographic data were collected by trained interviewers during clinical visits. Infant birth weight was measured by trained study staff using a pediatric scale calibrated before each use. Gestational age in weeks was determined by ultrasound measurements taken at enrollment by a licensed general practitioner using either (1) gestational sac mean diameter if the pregnancy was between 4 and 7 weeks or, (2) crown-rump length if the pregnancy was between 7 and 16 weeks.
Statistical analysis
Descriptive statistics were calculated for all variables. Maternal toenail As concentrations were right skewed and therefore log2 transformed. We assessed relationships between loci with quantified DNAm (cg26544247, chr2:25473813, and chr2:25473843) using Spearman correlations. Statistical mediation was evaluated using SEMs that included DNAm of all measured loci. SEMs are a multivariate statistical technique that allow for confirmatory analysis of a given hypothesis (Gunzler et al., 2013). By simultaneously estimating multiple and related regression-like models, SEMs allow for a given variable to act as both an independent and dependent variable (the term “exogenous” is used to refer to variables that act only as independent variables, and the term “endogenous” is used to refer to variables that act as dependent variables in at least one modeled regressions). SEMs use underlying latent (or unmeasured) variables defined by observed variables representing same construct, therefore addressing some forms of multicollinearity. DNAm of cg26544247, chr2:25473813, and chr2:25473843 was used as indicator variables to construct a latent variable representing DNAm of DNMT3A. Confirmatory factor analysis was performed to test how well DNAm of the three CpG sites represent DNAm of DNMT3A region.
To understand the associations between As concentrations, DNAm of DNMT3A, and birth weight independent of gestational age, we first built an SEM including log2-transformed maternal toenail As concentrations as independent variable, birth weight as the dependent variable, and the latent variable representing DNAm of DNMT3A as a mediator. We similarly assessed associations between As concentrations, DNAm of DNMT3A, and gestational age independent of birth weight. A single SEM including both outcomes of interest (birthweight and gestational age) was then used to assess mediation, adjusting for the potential confounders of infant sex, and maternal education (> primary vs. ≤ primary education), and the potential mediator of maternal weight gain between enrollment and delivery. Covariates were selected based on a review of previous literature and observed univariate associations between As exposure and birth outcomes in the current study. Birth weight was modeled in kg to ensure variances were similarly scaled. The full information maximum likelihood approach (FIML) was used to estimate model parameters, and robust estimates of model fit were used due to skewness in DNAm variables (Brosseau-Liard et al., 2012; Brosseau-Liard and Savalei, 2014). For models including categorical covariates (i.e., infant sex and maternal education), standard errors were calculated from 10,000 bootstrap samples. In each SEM, model indices were examined to determine if including residual correlations would improve model fit.
Our primary measure of exposure was postpartum maternal toenail As concentrations. Toenail As concentrations provide a time-integrated biomarker of internal dose that reflects arsenic exposure from all routes of exposure that occurred several months to a year prior to their collection (Kile et al., 2005). Therefore, postpartum maternal toenail As concentrations were selected as the measure of internal As dose that occurred during pregnancy. We conducted sensitivity analyses using two other arsenic exposure metrics: log2(maternal toenail As concentrations at enrollment) and log2(maternal water As concentrations). In addition, we used linear models to test the associations between DNAm of the target CpG site (cg26544247), log2-transformed postpartum maternal toenail As concentrations, birth weight, and gestational age. Specifically, we evaluated associations between the exposure (maternal toenail As concentration) and outcome (gestational age and birthweight), the mediator (DNAm) and the exposure (maternal toenail As concentrations), and the mediator (DNAm) and the outcome (gestational age and birthweight).
All statistical analyses were performed using the R statistical package, version 3.5.0 (R Core Team, 2015). SEMs were conducted using the R lavaan package (Oberski, 2014).
RESULTS
Participant Characteristics
Measures of DNAm was available for 413 infants born by vaginal births. Maternal and infant characteristics are summarized in Table 1. Approximately half of the infants were male (52.7%). The median ± IQR maternal drinking water As concentrations at recruitment was 17.0 ± 73.5 µg/L (range: 0.5 – 545.0 µg/L), and the mean maternal toenail As concentrations at delivery was 1.7 ± 3.2 ng/µg (range: 0.04 – 34.8). The median gestational age was 38 ± 3 weeks (range: 22 – 42 weeks), and the median birth weight was 2,800 ± 600 g (range: 1,400 – 4,600 g). Consistent with other birth cohort studies in Bangladesh, there was a high rate of preterm birth (< 37 weeks gestation, 29.0%) (Shah et al., 2014) and low birth weight (< 2,500 g, 25.9%) (Monawar Hosain et al., 2005).
Table 1:
Participant characteristics (N = 413).
| Median ± IQR | Range | |
|---|---|---|
| Drinking water As at recruitment (µg/L) | 17.0 ± 73.5 | 0.5 – 545.0 |
| Maternal toenail As at enrollment (ng/µg) | 2.1 ± 3.7 | 0.1 – 51.3 |
| Postpartum maternal toenail As (ng/µg) a | 1.7 ± 3.2 | 0.04 – 34.8 |
| Gestational age at delivery (weeks) b | 38 ± 3 | 22–42 |
| Preterm birth (< 37 weeks gestation), N (%) | 120 (29.0%) | |
| Birth weight (g) c | 2,800 ± 600 | 1,400 – 4,600 |
| Low birth weight (< 2,500 g), N (%) | 107 (25.8%) | |
| Infant sex | ||
| Male, N (%) | 218 (52.7%) | |
| Female, N (%) | 196 (47.3%) | |
| Percent DNAm | ||
| cg26544247 (chr2: 25,473,782) c | 2.5 ± 6.1 | 0.0 – 25.0 |
| chr2:25,473,813 | 2.1 ± 5.3 | 0.0 – 19.7 |
| chr2:25,473,843 | 0.0 ± 3.6 | 0.0 – 25.4 |
As = arsenic; IQR = interquartile range; DNAm = DNA methylation.
Relatively, levels of DNAm at each of the CpG sites were low. The median ± IQR percent methylation for the CpG sites were 2.5 ± 6.1, 2.1 ± 5.3, 0.0 ± 3.6 for cg26544247 (chr2: 25,473,782), chr2:25,473,813, and chr2:25,473,843, respectively. DNAm of was significantly and positively correlated between all CpG sites measured (rSpearman > 0.8, p < 0.001) (Figure 2). There was a trend toward increasing DNAm with greater As exposure at each CpG site. Figure 3 shows the median and IQR of DNAm by quartiles of postpartum maternal toenail As concentrations.
Correlation matrix of DNAm of CpGs measured in DNMT3A. Spearman correlations are presented in the top right cells, scatter plots of % DNAm are presented in the bottom left cells, and density plots of % DNAm are presented in the diagonal cells. DNAm = DNA methylation.
Mediation analysis
A latent variable representing DNMT3A methylation was constructed using percent methylation of cg26544247, chr2:25,473,813, and chr2:25,473,843. Confirmatory factor analysis indicated that DNAm measured at cg26544247, chr2:25473813, and chr2:25473843 represented DNAm of DNMT3A well (χ2 P-value < 0.001; Root mean square error of approximation (RMSEA) = 0.00; Standardized root mean squared residual = 0.00; Comparative fit index (CFI) = 1.00; Tucker-Lewis/non-normed fit index = 1.00).
We first assessed mediation of the associations between log2-transformed postpartum maternal toenail As concentrations and birth outcomes by DNAm of DNMT3A in separate SEMs for birth weight and gestational age. We observed a significant positive association between maternal As levels and DNMT3A DNAm (B = 0.37; 95% CI: 0.12, 0.62) in an SEM for the mediated effect of log2-transformed postpartum maternal toenail As concentrations on birth weight. We also observed a significant negative association between DNMT3A DNAm and birth weight (B = −20.7 g; 95% CI: −32.3, −9.1) (Supplemental Table S1 and Supplemental Figure S1). The direct effect of maternal As levels on birth weight was not significant (B = 1.6 g; 95% CI: −26.5, 29.8); however, the indirect effect of maternal As levels on birth weight through DNMT3A DNAm was statistically significant (B = −7.6 g; 95% CI: −13.8, −1.5). Similarly, in an SEM for the mediated effect of log2-transformed postpartum maternal toenail As concentrations on gestational age through DNMT3A methylation, there was a significant positive association between maternal As levels and DNMT3A DNAm (B = 0.39; 95% CI: 0.14, 0.65), and a significant negative association between DNMT3A DNAm and gestational age (B = −0.12 weeks; 95% CI: −0.18, −0.05) (Supplemental Table S2 and Supplemental Figure S2). Both the direct and indirect effects of maternal As levels on gestational age were statistically significant (direct effect: B = −0.25 weeks; 95% CI: −0.43, −0.08; indirect effect: B = −0.05 weeks; 95% CI: −0.08, − 0.01).
Subsequently, we constructed a single SEM to test for the indirect effects of log2-transformed postpartum maternal toenail As concentrations on birth weight and gestational age through DNMT3A DNAm. This model accounted for an indirect effect of maternal As exposure on birth weight though gestational age, but we did not include the direct effect of As exposure on birth weight because of the prior results of our individual models (Tables 2 and 3,3, and Figure 4). There was a significant positive association between log2-transformed postpartum maternal toenail As concentrations and DNMT3A methylation (B = 0.39; 95% CI: 0.14, 0.65), and negative associations between maternal toenail As levels and gestational age (B = −0.25 weeks; 95% CI: −0.42, −0.08), and between DNMT3A DNAm and gestational age (B = −0.12 weeks; 95% CI: −0.18, −0.06). There were also significant indirect effects of maternal toenail As levels on gestational age through DNMT3A methylation (B = −0.05 weeks; 95% CI: −0.09, −0.01). The direct effect of DNMT3A DNAm on birth weight and the indirect effect of maternal toenail As levels on birth weight through DNMT3A DNAm did not achieve significance (B = −10.1 g; 95% CI: −21.2, 1.0 and B = −4.0 g; 95% CI: −8.6, 0.6, respectively). However, we observed a significant association between gestational age and birth weight (B = 77.4 g; 95% CI: 62.0, 92.7) and significant indirect effects of maternal toenail As levels on birth weight through gestational age (B = −19.5 g; 95% CI: −33.5, −5.5) and through DNMT3A DNAm and gestational age (B = − 3.6 g; 95% CI: −6.8, −0.4).
Unadjusted structural equation model for the mediated effect of postpartum maternal toenail As levels on gestational through DNMT3A DNAm, and the mediated effect of postpartum maternal toenail As levels on birth weight through gestational age and DNMT3A DNAm. As = arsenic; DNAm = DNA methylation.
Table 2:
Unadjusted structural equation model for the mediated effect of postpartum maternal toenail As levels on gestational through DNMT3A DNAm, and the mediated effect of postpartum maternal toenail As levels on birth weight through gestational age and DNMT3A DNAm.
| Pathway | Effect | B (95% CI) | P |
|---|---|---|---|
| DNMT3A | |||
| log2(Maternal toenail As) → DNMT3A DNAm | Direct | 0.39 (0.14, 0.65) | 0.002 |
| Gestational age | |||
| log2(Maternal toenail As) → Gestational age | Direct | −0.25 (−0.42, −0.08) | 0.004 |
| DNMT3A DNAm → Gestational age | Direct | −0.12 (−0.18, −0.06) | < 0.001 |
| log2(Maternal toenail As) → DNMT3A DNAm → Gestational age | Indirect | −0.05 (−0.09, −0.01) | 0.017 |
| Birth weight | |||
| DNMT3A DNAm → Birth weight | Direct | −10.12 (−21.23, 0.99) | 0.074 |
| Gestational age → Birth weight | Direct | 77.37 (62.01, 92.73) | < 0.001 |
| log2(Maternal toenail As) → DNMT3A DNAm → Birth weight | Indirect | −3.98 (−8.55, 0.59) | 0.088 |
| log2(Maternal toenail As) → Gestational age → Birth weight | Indirect | −19.52 (−33.50, −5.52) | 0.006 |
| log2(Maternal toenail As) → DNMT3A DNAm → Gestational age → Birth weight | Indirect | −3.60 (−6.77, −0.44) | 0.026 |
As = arsenic; DNAm = DNA methylation.
Table 3:
Fit indices for the unadjusted structural equation model for the mediated effect of postpartum maternal toenail As levels on gestational through DNMT3A DNAm, and the mediated effect of postpartum maternal toenail As levels on birth weight through gestational age and DNMT3A DNAm. As = arsenic; DNAm = DNA methylation.
| Index | Criterion for good fit | Model fit1 |
|---|---|---|
| χ2 P-value | > 0.05 | 0.245 |
| Root mean square error of approximation (RMSEA) | < 0.05 | 0.025 |
| Comparative fit index (CFI) | > 0.95 | 0.999 |
| Tucker-Lewis/non-normed fit index | > 0.90 | 0.998 |
| Standardized root mean squared residual | < 0.05 | 0.012 |
Similar results were observed in an SEM including infant sex, maternal weight gain, and maternal education (Tables 4, and Figure 5). Fit indices for this final model indicated good model fit (Table 5). Maternal toenail As levels were significantly associated with DNMT3A DNAm (B = 0.40; 95% CI: 0.15, 0.66) and gestational age (B = −0.19 weeks; 95% CI: −0.36, −0.03), and DNMT3A DNAm was significantly associated with gestational age (B = −0.10 weeks; 95% CI: −0.16, −0.04) and birth weight (B = −11.0 g; 95% CI: −21.5, −0.4). There was a significant indirect effect of maternal toenail As levels on gestational age through DNMT3A methylation (B = −0.04 weeks; 95% CI: −0.08, −0.01), and significant indirect effects of maternal toenail As levels on birth weight through gestational age (B = −14.4 g; 95% CI: −29.2, −1.9), through DNMT3A DNAm and gestational age (B = −3.1 g; 95% CI: −6.6, −0.8), and through maternal weight gain and gestational age (B = −5.1 g; 95% CI: −9.6, −1.5). Calculated from the adjusted SEM, the total effect of a doubling in maternal toenail As concentration is a decrease in gestational age of 2.1 days (95% CI: 0.9 – 3.3) and a decrease in birth weight of 29 g (95% CI: 14, 46).
Adjusted structural equation model for the mediated effect of postpartum maternal toenail As levels on gestational through DNMT3A DNAm, and the mediated effect of postpartum maternal toenail As levels on birth weight through gestational age and DNMT3A DNAm, adjusted for infant sex, maternal weight gain, and maternal education. As = arsenic; DNAm = DNA methylation.
Table 4:
Adjusted structural equation model for the mediated effect of postpartum maternal toenail As levels on gestational through DNMT3A DNAm, and the mediated effect of postpartum maternal toenail As levels on birth weight through gestational age and DNMT3A DNAm, adjusted for infant sex, maternal weight gain, and maternal education.
| Pathway | Effect | B (95% CI) | P |
|---|---|---|---|
| DNMT3A | |||
| log2(Maternal toenail As) → DNMT3A DNAm | Direct | 0.40 (0.15, 0.66) | 0.002 |
| Infant sex → DNMT3A DNAm | Direct | −0.75 (−1.47, −0.04) | 0.041 |
| Maternal weight gain | |||
| log2(Maternal toenail As) → Maternal weight gain | Direct | −0.24 (−0.41, −0.07) | 0.004 |
| Gestational age | |||
| log2(Maternal toenail As) → Gestational age | Direct | −0.19 (−0.36, −0.03) | 0.027 |
| DNMT3A DNAm → Gestational age | Direct | −0.10 (−0.16, −0.04) | 0.001 |
| Maternal weight gain → Gestational age | Direct | 0.28 (0.20, 0.36) | < 0.001 |
| Maternal education (> primary vs. ≤ primary) → Gestational age | Direct | 0.56 (0.17, 0.97) | 0.006 |
| log2(Maternal toenail As) → DNMT3A DNAm → Gestational age | Indirect | −0.04 (−0.08, −0.01) | 0.024 |
| log2(Maternal toenail As) → Maternal weight gain → Gestational age | Indirect | −0.07 (−0.12 −0.02) | 0.006 |
| Birth weight | |||
| DNMT3A DNAm → Birth weight | Direct | −11.03 (−21.54, 0.41) | 0.048 |
| Gestational age → Birth weight | Direct | 77.04 (58.08, 94.39) | < 0.001 |
| Infant sex → Birth weight | Direct | −118.96 (−196.45, −41.25) | 0.003 |
| Maternal weight gain → Birth weight | Direct | 7.46 (−10.60, 26.00) | 0.425 |
| Maternal education → Birth weight | Direct | −60.82 (−140.15, 20.83) | 0.139 |
| log2(Maternal toenail As) → DNMT3A DNAm → Birth weight | Indirect | −4.37 (−9.69, 0.17) | 0.081 |
| log2(Maternal toenail As) → Gestational age → Birth weight | Indirect | −14.44 (−29.24, −1.92) | 0.038 |
| log2(Maternal toenail As) → DNMT3A DNAm → Gestational age → Birth weight | Indirect | −3.14 (−6.59, −0.77) | 0.037 |
| log2(Maternal toenail As) → Maternal weight gain → Birth weight | Indirect | −1.80 (−6.73, 2.82) | 0.440 |
| log2(Maternal toenail As) → Maternal weight gain → Gestational age → Birth weight | Indirect | −5.14 (−9.60, −1.48) | 0.012 |
As = arsenic; DNAm = DNA methylation.
Table 5:
Fit indices for the adjusted structural equation model for the mediated effect of postpartum maternal toenail As levels on gestational through DNMT3A DNAm, and the mediated effect of postpartum maternal toenail As levels on birth weight through gestational age and DNMT3A DNAm, adjusted for infant sex, maternal weight gain, and maternal education. As = arsenic; DNAm = DNA methylation.
| Index | Criterion for good fit | Model fit1 |
|---|---|---|
| χ2 P-value | > 0.05 | 0.166 |
| Root mean square error of approximation (RMSEA) | < 0.05 | 0.028 |
| Comparative fit index (CFI) | > 0.95 | 0.997 |
| Tucker-Lewis/non-normed fit index | > 0.90 | 0.994 |
| Standardized root mean squared residual | < 0.05 | 0.028 |
Sensitivity analyses
Sensitivity analyses that were performed using two other exposure metrics (e.g., log2-transformed maternal toenail As concentrations at enrollment and log2-transformed maternal drinking water arsenic concentrations) yielded similar results. Postpartum maternal toenail As concentrations were significantly correlated with maternal toenail As concentrations at enrollment (rSpearman = 0.80, p < 0.001) and maternal drinking water As concentrations (rSpearman = 0.61, p < 0.001). Results from adjusted SEMs using maternal toenail As levels collected postpartum and at enrollment were consistent: maternal toenail As levels at enrollment were associated with DNMT3A DNAm (B = 0.40; 95% CI: 0.14, 0.67) and gestational age (B = −0.24 weeks; 95% CI: −0.42, −0.07), DNMT3A DNAm was associated with gestational age (B = −0.10 weeks; 95% CI: −0.16, −0.04) and birth weight (B = −11.0 g; 95% CI: −21.7, 0.69), and there were significant indirect effects of maternal toenail As levels on birth weight through gestational age (B = −18.3 g; 95% CI: −33.7, −5.3) and through DNMT3A DNAm and gestational age (B = −3.1 g; 95% CI: −6.6, −0.7) (Supplemental Table S3 and Figure S3). However, in an adjusted SEM using maternal drinking water As levels, we observed a smaller effect size of the association between As exposure and DNMT3 DNAm (B = 0.18; 95% CI: 0.06, 0.31). The association between maternal drinking water As levels and gestational age, and the indirect effect of maternal drinking water As levels on birth weight through gestational age were not significant, although the effect estimates were negative (B = −0.06 weeks; 95% CI: −0.13, 0.02 and B = −4.5 g; 95% CI: −10.4, 1.3, respectively) (Supplemental Table S4 and Figure S4).
We also assessed the associations between in utero As exposure, birth outcomes, and DNMT3 DNAm using linear models to confirm results obtained through SEMs. Specifically, we evaluated associations between the exposure (postpartum maternal toenail As levels) and outcomes (birth weight and gestational age), the mediator (represented by the target CpG site DNMT3, cg26544247) and the exposure (postpartum maternal toenail As levels), and the mediator (DNMT3 DNAm) and the outcomes (birthweight and gestational age). Results from unadjusted models and models adjusted for infant sex, maternal weight gain, and maternal education are shown in Supplemental Table S5. Results from crude and adjusted linear models were consistent with SEMs. In adjusted models, log2-transformed postpartum maternal toenail As concentrations were associated with decreased gestational age (B = −0.16 weeks, 95% CI: −0.27, −0.04) and increased methylation of cg26544247 (B = 0.44; 95% CI: 0.14, 0.74), DNAm of cg26544247 was associated with decreased gestational age (B = −0.09 weeks; 95% CI: −0.13, −0.04) and birth weight (B = −12.6 g; 95% CI: −24.8, −6.4), and gestational age was associated with birth weight (B = 79.1 g; 95% CI: 60.7, 97.5). Maternal As levels were not significantly associated with birth weight (B = 1.9 g; 95% CI: −25.6, 29.5).
DISCUSSION
We observed that DNAm of DNMT3A mediated the association between in utero As exposure and birth outcomes. We found that increased log2-transformed postpartum maternal toenail As concentrations were associated with increased DNMT3A methylation and decreased gestational age, and there was a significant negative indirect effect of As on gestational age through DNMT3A methylation. In addition, we observed significant negative indirect effects of maternal toenail As levels on birth weight through gestational age and though DNMT3A methylation and gestational age. The indirect effect of maternal As levels on birth weight through DNMT3A methylation alone did not achieve statistical significance. Overall, a doubling in maternal toenail As concentration was associated with a decrease in gestational age of 2.1 days and a decrease in birth weight of 28.9 g. We used an SEM approach. SEMs are advantageous for testing mediators that are located on a causal pathway (i.e., variables that act as a dependent variable in one model and as an independent variable in another model) (Gunzler et al., 2013). Furthermore, SEMs incorporate unmeasured latent variables representing multiple, related indicator variables (i.e., measured variables). This helps to address limitations posed by multicollinearity in traditional biostatistical approaches (Geoffrey Maruyama, 1998). Although commonly used in other fields of research, SEMs can be a valuable tool in environmental epidemiology for understanding complex relationships (Buncher et al., 1991). In this study, SEMs allowed us to create a latent variable representing DNAm of three correlated CpG sites located in the gene body of DNMT3A and simultaneously evaluate the direct and indirect effects of in utero As exposure on birth outcomes.
Our results are consistent with other studies finding a negative association between in utero As exposure and gestational age (Röllin et al., 2017; Xu et al., 2011), although some studies have reported a borderline significant negative association (Laine et al., 2015) or a null association (Bloom et al., 2016; Freire et al., 2019; Sun et al., 2019). Characteristics of study populations overall, including level of exposure, As methylation capacity, and maternal diet, may contribute these differences. It should also be noted that epidemiological studies have generally analyzed associations with preterm birth (< 37 weeks gestation), rather than gestational age (Bloom et al., 2014). However, health effects later in life may be associated with early term, in addition to preterm, birth (Boyle et al., 2012). Furthermore, this clinical categorization may not be sufficient to detect variation in gestational age, and may be not be appropriate for populations with high rates of preterm birth such as Bangladesh (Shah et al., 2014). In fact, 29% of births in our study population were preterm (Table 1). Although we did not observe a direct effect of in utero As exposure on birth weight, we did find a significant total effect of As exposure on birth weight fully mediated through pathways including gestational age; mediation through DNAm of DNMT3A alone did not achieve statistical significance. Multiple epidemiological studies have investigated the effects of in utero As exposure on birth weight (as reviewed by Bloom et al., Milton et al., and Zhong et al., 16,17,46). However, our results indicate the importance of including gestational age as a birth outcome of interest, and addressing relationships between birth outcomes. Furthermore, decreased birth weight may be due to shortened gestational period and/or intrauterine grown restriction (Wardlaw et al., 2004). Investigating the etiology of observed reductions in birth weight is important to understanding the health effects of in utero As exposure.
Consistent with our findings, DNAm of the target CpG site, cg26544247, has previously been negatively associated with gestational age in a Norwegian birth cohort (pFDR = 0.03) (Bohlin et al., 2016). This study also found differential methylation throughout the epigenome (44,359 probes at pFDR < 0.05 for ultrasound-estimated gestational age and 44,544 probes for last menstrual period-estimate gestational age at pFDR < 0.05). However, 18 additional CpG sites annotated to DNMT3A were identified as differentially methylated and a gene ontology (GO) analysis identified several pathways that included the DNMT3A gene. In addition, in birth cohort in Tennessee, one CpG annotated to DNMT3A was identified as associated with gestational age (pFDR = 0.04) (Parets et al., 2013). Methylation at additional CpG sites located in DNMT3A have been associated with the in utero environment. Specifically, increased methylation in cord blood of cg13344237, located within 200 base pairs of the transcription start site, has been associated with lower maternal pre-pregnancy BMI (underweight <18.5 kg/m2 vs. normal 18.5–24.9 kg/m2) (pFDR = 0.02) (Sharp et al., 2015). Increased methylation in cord blood of cg15843262, located in the gene body, was associated with antidepressant use during pregnancy (p = 8x10−4) (Non et al., 2014).
The role of DNMT3A in de novo DNAm is well-established (Okano et al., 1999), and DNMT3A is expressed during embryogenesis (Watanabe et al., 2002). However, how DNMT3A may affect birth outcomes is not well understood. DNMT3A is involved in multiple, diverse biological pathways; GO terms annotated to DNMT3A include roles in DNAm, RNA polymerase binding and transcription, genetic imprinting, mitotic cell cycle, protein binding, aging, and regulation of cell death. Further research is needed to understand the biological pathways involved in the observed association between DNMT3A methylation and birth outcomes.
Our study strengths included that gestational age was determined by ultrasound at the time of enrollment, which provides a more accurate estimation than reported last menstrual period. In addition, our study is strengthened by the prospective birth cohort design with multiple measures of As exposure during pregnancy. Toenail As concentrations reflect exposure during the prior several months to a year (Kile et al., 2005). Maternal toenail samples were collected within one moth of delivery to provide an estimate of internal dose during pregnancy and establish temporality. Drinking water samples and maternal toenail samples were also collected at the time of enrollment early in pregnancy. Overall, results from sensitivity analyses using maternal drinking water As levels and maternal toenail As concentration at enrollment as the measures of exposure were consistent. The slight differences in results could be a function of exposure misclassification from using drinking water arsenic concentrations instead of biomarkers of internal dose (e.g., toenails), or differences in As metabolism and excretion between individuals (Marchiset-Ferlay et al., 2012). There is evidence that the As metabolism may be associated with birth outcomes (Gelmann et al., 2013) and may also influence DNAm. Sex has also been shown to influence arsenic metabolism. Subsequently, future studies could examine As metabolism modifies DNAm of DNAMT3A and birth outcomes.
Several limitations of our study should be noted. We had missing data on maternal toenail As concentrations (N = 21 for maternal toenail concentrations at enrollment, N = 23 for postpartum maternal toenail concentrations). Gestational age and birth weight were lower among mother-infant pairs with missing maternal toenail As concentrations at enrollment (mean gestational age: with missing data = 34.3 weeks, without missing data = 37.6 weeks; Kruskal-Wallis p = 0.003; mean birth weight: with missing data = 2,481 g, without missing data = 2,764 g; Kruskal-Wallis p = 0.012) and postpartum (mean gestational age: with missing data = 34.5 weeks, without missing data = 37.6 weeks; Kruskal-Wallis p = 0.005; mean birth weight: with missing data = 2,510 g, without missing data = 2,763 g; Kruskal-Wallis p = 0.018). However, we had full data on maternal water As concentrations, and maternal drinking water concentrations were significantly correlated and toenail As concentrations at enrollment (rSpearman = 0.60, p < 0.001) and postpartum (rSpearman = 0.61, p < 0.001). Although in sensitivity analyses using maternal water As levels we observed reduced significance of the indirect effect of in utero As on birth weight through DNMT3 methylation and gestational age, the confidence intervals were wide and tended toward a negative association (log2-transformed maternal water As → DNMT3A DNAm → birth weight: B (95% CI) = −2.01 (−4.59, 0.16), p = 0.095; log2-transformed maternal water As → gestational age → birth weight: B (95% CI) = −4.52 (−10.45, 1.28), p = 0.125). Furthermore, SEMs allow for the inclusion of cases with missing data on some exogenous variables, and therefore participants with missing data on maternal toenail As concentration could be still be used to estimate parameters for other pathways in the model.
We did not have data on DNAm at additional DNMT3A CpG sites and did not measure gene expression. Therefore, we cannot determine if changes in DNAm at the interrogated CpGs of DNMT3A lead to changes in gene expression or chromatin conformation. Nor can we infer that methylation at these 3 CpG sites represent methylation across the entire gene. The three CpGs measured in the current study are located with the gene body, and DNAm within gene bodies has previously been associated with gene expression, (Lou et al., 2014), with possibly a positive association between DNAm and gene expression (Rauch et al., 2009). There is little previous research, however, on the role of DNAm in regulating DNMT3A expression. One study of epigenetic aging observed significant associations between aging and DNAm of cg26544247, but no association between DNAm and DNMT3A expression (Marttila et al., 2015). However, this study population included young adults and nonagenarians and results may not be generalizable to fetal gene expression. Further research is necessary to determine if DNAm of DNMT3A, both at the three interrogated CpGs and CpGs located in other gene regions, is associated with gene expression. In addition, while it is expected that changes in DNMT3A expression would be reflected in DNAm levels of other genes, we did not have epigenome-wide DNAm data on this subsample of the birth cohort. Finally, we restricted our study to vaginal births and subsequently our results may not be generalizable to births delivered by cesarean section. Restricting our analysis to vaginal births might also lead to biased estimates if arsenic exposure or DNMT3A DNAm is related to birth complications that would require Cesarean delivery in this population. Subsequently, future studies should attempt to replicate these findings in a birth cohort that includes both vaginal and cesarean delivery.
CONCLUSION
In this prospective birth cohort of mothers exposed to As primarily through drinking water in Bangladesh, we assessed the relationships between in utero As exposure, gestational age, birth weight, and methylation of DNMT3A, a gene involved in de novo DNAm. We show that in utero As exposure is negatively associated with gestational age and birth weight. The effect of maternal As exposure on gestational age were partially mediated by DNAm of DNMT3A, and the effect on birth weight is fully mediated by pathways including gestational age. These results provide evidence that in utero As exposure affects fetal development through epigenetic dysregulation. Further research is need to understand how DNAm of DNMT3A is associated with gene expression, and in turn if this affects the methylation status and expression of other genes related to infant health outcomes.
ACKNOWLEDGEMENTS
We thank our field staff and the study participants in Bangladesh, without whom this work would not have been possible.
FUNDING SOURCES
This work was supported by the US National Institute of Environmental Health Sciences (NIEHS) grants R01 ES015533, R01 ES023441, P42 ES010349, and F31ES029019, and the National Center for Advancing Translational Sciences (NCATS) grant TL1 TR001875. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
The study protocol was approved by the Human Research Committees at the Harvard School of Public Health, Oregon State University, and Dhaka Community Hospital Trust.
Abbreviations:
| As | arsenic |
| DCH | Dhaka Community Hospital |
| DNAm | DNA methylation |
| DNMT3A | DNA methyltransferase 3 alpha |
| DNMT3B | DNA methyltransferase 3 beta |
| GO | Gene Ontology |
| SEM | structural equation model |
| WHO | World Health Organization |
Footnotes
Competing financial interest: The authors declare no competing financial interests.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Declaration of interests
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
REFERENCES
- Bailey KA, Smith AH, Tokar EJ, Graziano JH, Kim K-W, Navasumrit P, Ruchirawat M, Thiantanawat A, Suk WA, Fry RC, 2016. Mechanisms underlying latent disease risk associated with early-life arsenic exposure: current research trends and scientific gaps. Environ. Health Perspect 124, 170–175. 10.1289/ehp.1409360 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Bangladesh Bureau of Statistics, United Nation Children’s Fund, 2015. Bangladesh multiple indicator cluster survey 2012–2013 final report Dhaka, Bangladesh. [Google Scholar]
- Bjørklund G, Aaseth J, Chirumbolo S, Urbina MA, Uddin R, 2018. Effects of arsenic toxicity beyond epigenetic modifications. Environ. Geochem. Health 40, 955–965. 10.1007/s10653-017-9967-9 [PubMed] [CrossRef] [Google Scholar]
- Bloom MS, Neamtiu IA, Surdu S, Pop C, Anastasiu D, Appleton AA, Fitzgerald EF, Gurzau ES, 2016. Low level arsenic contaminated water consumption and birth outcomes in Romania-An exploratory study. Reprod. Toxicol 59, 8–16. 10.1016/j.reprotox.2015.10.012 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Bloom MS, Surdu S, Neamtiu IA, Gurzau ES, 2014. Maternal arsenic exposure and birth outcomes: a comprehensive review of the epidemiologic literature focused on drinking water HHS Public Access. Int J Hyg Env. Heal 217, 709–719. 10.1016/j.ijheh.2014.03.004 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Bohlin J, Håberg SE, Magnus P, Reese SE, Gjessing HK, Magnus MC, Parr CL, Page CM, London SJ, Nystad W, 2016. Prediction of gestational age based on genome-wide differentially methylated regions. Genome Biol 17, 207 10.1186/s13059-016-1063-4 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Boyle EM, Poulsen G, Field DJ, Kurinczuk JJ, Wolke D, Alfirevic Z, Quigley MA, 2012. Effects of gestational age at birth on health outcomes at 3 and 5 years of age: population based cohort study. BMJ 344, e896 10.1136/BMJ.E896 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Brosseau-Liard PE, Savalei V, 2014. Adjusting incremental fit indices for nonnormality. Multivariate Behav. Res 49, 460–470. 10.1080/00273171.2014.933697 [PubMed] [CrossRef] [Google Scholar]
- Brosseau-Liard PE, Savalei V, Li L, 2012. An investigation of the sample performance of two nonnormality corrections for RMSEA. Multivariate Behav. Res 47, 904–930. 10.1080/00273171.2012.715252 [PubMed] [CrossRef] [Google Scholar]
- Buncher CR, Succop PA, Dietrich KN, 1991. Structural equation modeling in environmental risk assessment. Environ. Health Perspect 90, 209–13. 10.1289/ehp.90-1519490 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Concha G, Vogler G, Lezcano D, Nermell B, Vahter M, 1998. Exposure to inorganic arsenic metabolites during early human development. Toxicol. Sci 44, 185–190. [PubMed] [Google Scholar]
- Creed J, Brockhoff C, Martin T, 1994. Method 200.8, Revision 5.4: Determination of trace elements in waters and wastes by inductively coupled plasma – mass spectrometry Cincinnati, OH. [Google Scholar]
- Farzan SF, Karagas MR, Chen Y, 2013. In utero and early life arsenic exposure in relation to long-term health and disease. Toxicol. Appl. Pharmacol. 272, 384–390. 10.1016/j.taap.2013.06.030 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Ferm VH, Hanlon DP, 1985. Constant rate exposure of pregnant hamsters to arsenate during early gestation. Environ. Res 37, 425–32. [PubMed] [Google Scholar]
- Freire C, Amaya E, Gil F, Murcia M, LLop S, Casas M, Vrijheid M, Lertxundi A, Irizar A, Fernández-Tardón G, Castro-Delgado RV, Olea N, Fernández MF, INMA Project, 2019. Placental metal concentrations and birth outcomes: the Environment and Childhood (INMA) project. Int J Hyg Env. Heal 222, 468–478. 10.1016/j.ijheh.2018.12.014 [PubMed] [CrossRef] [Google Scholar]
- Gelmann ER, Gurzau E, Gurzau A, Goessler W, Kunrath J, Yeckel CW, McCarty KM, 2013. A pilot study: The importance of inter-individual differences in inorganic arsenic metabolism for birth weight outcome. Environ. Toxicol. Pharmacol 36, 1266–1275. 10.1016/J.ETAP.2013.10.006 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Maruyama Geoffrey, 1998. Basics of Structural Equation Modeling SAGE Publications, Thousand Oaks, CA. [Google Scholar]
- Gu T, Lin X, Cullen SM, Luo M, Jeong M, Estecio M, Shen J, Hardikar S, Sun D, Su J, Rux D, Guzman A, Lee M, Qi LS, Chen J-J, Kyba M, Huang Y, Chen T, Li W, Goodell MA, 2018. DNMT3A and TET1 cooperate to regulate promoter epigenetic landscapes in mouse embryonic stem cells. Genome Biol 19, 88 10.1186/s13059-018-1464-7 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Gunzler D, Chen T, Wu P, Zhang H, 2013. Introduction to mediation analysis with structural equation modeling. Shanghai Arch. Psychiatry 25, 390–4. 10.3969/j.issn.1002-0829.2013.06.009 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Hall M, Gamble M, Slavkovich V, Liu X, Levy D, Cheng Z, van Geen A, Yunus M, Rahman M, Pilsner JR, Graziano J, 2007. Determinants of arsenic metabolism: blood arsenic metabolites, plasma folate, cobalamin, and homocysteine concentrations in maternal–newborn pairs. Environ. Health Perspect 115, 1503–1509. 10.1289/ehp.9906 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Heindel JJ, Vandenberg LN, 2015. Developmental origins of health and disease: a paradigm for understanding disease cause and prevention. Curr. Opin. Pediatr 27, 248–53. 10.1097/MOP.0000000000000191 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Hill DS, Wlodarczyk BJ, Finnell RH, 2008. Reproductive consequences of oral arsenate exposure during pregnancy in a mouse model. Birth Defects Res. (Part B) 83, 40–7. 10.1002/bdrb.20142 [PubMed] [CrossRef] [Google Scholar]
- Hood RD, 1972. Effects of sodium arsenite on fetal development. Bull. Environ. Contam. Toxicol 7, 216–222. 10.1007/BF01684401 [PubMed] [CrossRef] [Google Scholar]
- Hood RD, Bishop SL, 1972. Teratogenic effects of sodium arsenate in mice. Arch. Environ. Health 24, 62–65. 10.1080/00039896.1972.10666051 [PubMed] [CrossRef] [Google Scholar]
- Kaneda M, Okano M, Hata K, Sado T, Tsujimoto N, Li E, Sasaki H, 2004. Essential role for de novo DNA methyltransferase Dnmt3a in paternal and maternal imprinting. Nature 429, 900–903. 10.1038/nature02633 [PubMed] [CrossRef] [Google Scholar]
- Karagas MR, Steven Morris J, Weiss JE, Spate V, Baskett C, Robert Greenberg E, K, lM R., 1996. Toenail samples as an indicator of drinking water arsenic. Biomarkers Prev 5, 849–852. [PubMed] [Google Scholar]
- Kile ML, Cardenas A, Rodrigues E, Mazumdar M, Dobson C, Golam M, Quamruzzaman Q, Rahman M, Christiani DC, 2015. Estimating effects of arsenic exposure during pregnancy on perinatal outcomes in a Bangladeshi cohort. Epidemiology 27, 1 10.1097/EDE.0000000000000416 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Kile ML, Houseman EA, Baccarelli AA, Quamruzzaman Q, Rahman M, Mostofa G, Cardenas A, Wright RO, Christiani DC, 2014. Effect of prenatal arsenic exposure on DNA methylation and leukocyte subpopulations in cord blood. Epigenetics 9, 774–82. 10.4161/epi.28153 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Kile ML, Houseman EA, Rodrigues E, Smith TJ, Quamruzzaman Q, Rahman M, Mahiuddin G, Su L, Christiani DC, 2005. Toenail arsenic concentrations, GSTT1 gene polymorphisms, and arsenic exposure from drinking water. Cancer Epidemiol. Biomarkers Prev 14, 2419–2426. 10.1158/1055-9965.EPI-05-0306 [PubMed] [CrossRef] [Google Scholar]
- Kozul-Horvath CD, Zandbergen F, Jackson BP, Enelow RI, Hamilton JW, 2012. Effects of low-dose drinking water arsenic on mouse fetal and postnatal growth and development. PLoS One 7, e38249 10.1371/journal.pone.0038249 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Laine JE, Bailey KA, Rubio-Andrade M, Olshan AF, Smeester L, Drobná Z, Herring AH, Stýblo M, García-Vargas GG, Fry RC, 2015. Maternal arsenic exposure, arsenic methylation efficiency, and birth outcomes in the Biomarkers of Exposure to ARsenic (BEAR) Pregnancy cohort in Mexico. Environ. Health Perspect 123, 186–192. 10.1289/ehp.1307476 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Lou S, Lee H-M, Qin H, Li J-W, Gao Z, Liu X, Chan LL, KL Lam V, So W-Y, Wang Y, Lok S, Wang J, Ma RC, Tsui SK-W, Chan JC, Chan T-F, Yip KY, 2014. Whole-genome bisulfite sequencing of multiple individuals reveals complementary roles of promoter and gene body methylation in transcriptional regulation. Genome Biol 15, 408 10.1186/s13059-014-0408-0 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Marchiset-Ferlay N, Savanovitch C, Sauvant-Rochat M-P, 2012. What is the best biomarker to assess arsenic exposure via drinking water? Environ. Int 39, 150–171. 10.1016/j.envint.2011.07.015 [PubMed] [CrossRef] [Google Scholar]
- Marttila S, Kananen L, Häyrynen S, Jylhävä J, Nevalainen T, Hervonen A, Jylhä M, Nykter M, Hurme M, 2015. Ageing-associated changes in the human DNA methylome: genomic locations and effects on gene expression. BMC Genomics 16, 179 10.1186/s12864-015-1381-z [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Milton AH, Hussain S, Akter S, Rahman M, Mouly TA, Mitchell K, 2017. A review of the effects of chronic arsenic exposure on adverse pregnancy outcomes. Int. J. Environ. Res. Public Health 14 10.3390/ijerph14060556 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Monawar Hosain GM, Chatterjee N, Begum A, Saha SC, 2005. Factors associated with low birthweight in rural Bangladesh. J. Trop. Pediatr 52, 87–91. 10.1093/tropej/fmi066 [PubMed] [CrossRef] [Google Scholar]
- Moore CL, Flanigan TJ, Law CD, Loukotková L, Woodling KA, da Costa GG, Fitzpatrick SC, Ferguson SA, 2019. Developmental neurotoxicity of inorganic arsenic exposure in Sprague-Dawley rats. Neurotoxicol. Teratol 72, 49–57. 10.1016/j.ntt.2019.01.007 [PubMed] [CrossRef] [Google Scholar]
- Morrissey RE, Mottet NK, 1983. Arsenic-induced exencephaly in the mouse and associated lesions occurring during neurulation. Teratology 28, 399–411. 10.1002/tera.1420280311 [PubMed] [CrossRef] [Google Scholar]
- Nagymajtényi L, Selypes A, Berencsi G, 1985. Chromosomal aberrations and fetotoxic effects of atmospheric arsenic exposure in mice. J. Appl. Toxicol 5, 61–3. [PubMed] [Google Scholar]
- Non AL, Binder AM, Kubzansky LD, Michels KB, 2014. Genome-wide DNA methylation in neonates exposed to maternal depression, anxiety, or SSRI medication during pregnancy. Epigenetics 9, 964–972. 10.4161/epi.28853 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Oberski D, 2014. lavaan.survey: an R package for complex survey analysis of structural equation models. J. Stat. Softw 57, 1–27. 10.18637/jss.v057.i01 [CrossRef] [Google Scholar]
- Okano M, Bell DW, Haber DA, Li E, 1999. DNA methyltransferases Dnmt3a and Dnmt3b are essential for de novo methylation and mammalian development. Cell 99, 247–57. [PubMed] [Google Scholar]
- Parets SE, Conneely KN, Kilaru V, Fortunato SJ, Syed TA, Saade G, Smith AK, Menon R, 2013. Fetal DNA methylation associates with early spontaneous preterm birth and gestational age. PLoS One 8, e67489 10.1371/journal.pone.0067489 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- R Core Team, 2015. R: A Language and Environment for Statistical Computing Vienna, Austria. [Google Scholar]
- Rahman A, Granberg C, Persson L-Å, 2017. Early life arsenic exposure, infant and child growth, and morbidity: a systematic review. Arch. Toxicol 91, 3459–3467. 10.1007/s00204-017-2061-3 [PubMed] [CrossRef] [Google Scholar]
- Rauch TA, Wu X, Zhong X, Riggs AD, Pfeifer GP, 2009. A human B cell methylome at 100-base pair resolution. Proc. Natl. Acad. Sci. U. S. A 106, 671–8. 10.1073/pnas.0812399106 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Rodrigues EG, Kile M, Dobson C, Amarasiriwardena C, Quamruzzaman Q, Rahman M, Golam M, Christiani DC, 2015. Maternal–infant biomarkers of prenatal exposure to arsenic and manganese. J. Expo. Sci. Environ. Epidemiol 25, 639–648. 10.1038/jes.2015.45 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Röllin HB, Channa K, Olutola BG, Odland JØ, 2017. Evaluation of in utero exposure to arsenic in South Africa. Sci. Total Environ 575, 338–346. 10.1016/j.scitotenv.2016.10.044 [PubMed] [CrossRef] [Google Scholar]
- Shah R, Mullany LC, Darmstadt GL, Mannan I, Rahman SM, Talukder RR, Applegate JA, Begum N, Mitra D, Arifeen S. El, Baqui AH, ProjAHNMo Study Group in Bangladesh, 2014. Incidence and risk factors of preterm birth in a rural Bangladeshi cohort. BMC Pediatr 14, 112 10.1186/1471-2431-14-112 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Sharp GC, Lawlor DA, Richmond RC, Fraser A, Simpkin A, Suderman M, Shihab HA, Lyttleton O, McArdle W, Ring SM, Gaunt TR, Davey Smith G, Relton CL, 2015. Maternal pre-pregnancy BMI and gestational weight gain, offspring DNA methylation and later offspring adiposity: findings from the Avon Longitudinal Study of Parents and Children. Int. J. Epidemiol 44, 1288–1304. 10.1093/ije/dyv042 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Sun X, Liu W, Zhang B, Shen X, Hu C, Chen X, Jin S, Jiang Y, Liu H, Cao Z, Xia W, Xu S, Li Y, 2019. Maternal heavy metal exposure, thyroid hormones, and birth outcomes: a prospective cohort study. J. Clin. Endocrinol. Metab Epub ahead of print 10.1210/jc.2018-02492 [PubMed] [CrossRef] [Google Scholar]
- Uysal F, Ozturk S, Akkoyunlu G, 2017. DNMT1, DNMT3A and DNMT3B proteins are differently expressed in mouse oocytes and early embryos. J Mol Histol 48, 417–426. 10.1007/s10735-017-9739-y [PubMed] [CrossRef] [Google Scholar]
- Vahter M, 2009. Effects of arsenic on maternal and fetal health. Annu. Rev. Nutr 29, 381–399. 10.1146/annurev-nutr-080508-141102 [PubMed] [CrossRef] [Google Scholar]
- Wardlaw T, Blanc A, Zupan J, Ahman E, Åhman E, 2004. Low birthweight: country, regional, and global estimates UNICEF, New York, NY. [Google Scholar]
- Watanabe D, Suetake I, Tada T, Tajima S, 2002. Stage- and cell-specific expression of Dnmt3a and Dnmt3b during embryogenesis. Mech. Dev 118, 187–190. 10.1016/S0925-4773(02)00242-3 [PubMed] [CrossRef] [Google Scholar]
- World Health Organization, 2011. Arsenic in drinking-water: background document for the development of WHO guidelines for drinking-water quality, IARC Monographs on the Evaluation of Carcinogenic Risks to Humans Geneva, Switzerland: 10.1016/j.kjms.2011.05.002 [CrossRef] [Google Scholar]
- Xu L, Yokoyama K, Tian Y, Piao F-Y, Kitamura F, Kida H, Wang P, 2011. Decrease in birth weight and gestational age by arsenic among the newborn in Shanghai, China. Nihon. Koshu Eisei Zasshi 58, 89–95. [PubMed] [Google Scholar]
- Zhong Q, Cui Y, Wu H, Niu Q, Lu X, Wang L, Huang F, 2019. Association of maternal arsenic exposure with birth size: a systematic review and meta-analysis. Environ. Toxicol. Pharmacol 69, 129–136. 10.1016/j.etap.2019.04.007 [PubMed] [CrossRef] [Google Scholar]