Concentrations of persistent organic pollutants in maternal plasma and epigenome-wide placental DNA methylation

Prenatal maternal plasma persistent organic pollutant (POP) concentrations have been associated with neonatal outcomes. However, the underlying mechanisms remain unknown. Placental epigenetic mechanisms may be involved, but no prior epigenome-wide studies have investigated the impact of maternal POPs on placental DNA methylation. We studied the association between maternal plasma POP concentration in early pregnancy and epigenome-wide placental DNA methylation among 260 pregnant women from the NICHD Fetal Growth Studies. Our analysis focused on POPs with more than 80% plasma concentrations above the limit of quantification, including 3 organochlorine pesticides (hexachlorobenzene, trans-nonachlor, p,p’-dichlorodiphenyldichloroethylene), 1 polybrominated diphenyl ether (PBDE 47), 3 polychlorinated biphenyls (138/158, 153, 180), and 6 poly- and perfluorinated alkyl substances (PFASs) (perfluorodecanoic acid, perfluorohexanesulfonic acid, perfluorononanoic acid, perfluorooctanesulfonic acid, perfluoroundecanoic acid (PFUnDA)). Using 5% false discovery rate, POPs were associated with a total of 214 differentially methylated CpG sites (nominal p values ranging from 2.61 × 10−21 to 2.11 × 10−7). Out of the 214 CpG sites, 24 (11%) were significantly correlated with placental expression of 21 genes. Notably, higher PFUnDA was associated with increased methylation at 3 CpG sites (cg13996963, cg12089439, cg18145877) annotated to TUSC3, and increased methylation at those 3 CpG sites was correlated with decreased expression of TUSC3 in the placenta. Increased methylation at cg18145877 (TUSC3) and decreased expression of TUSC3 were correlated with shorter birth length. Out of the 214 CpG sites, methylation at 44 CpG sites was correlated (p value < 0.10) with at least one neonatal anthropometry measure (i.e., birth weight, birth length, and head circumference). Seven CpG sites mediated (p value < 0.05) the association between PBDE 47 and neonatal anthropometry measures. Genes annotating the top differentially methylated CpG sites were enriched in pathways related to differentiation of embryonic cells (PBDE 47) and in pathways related to brain size and brain morphology (PFASs). DNA methylation changes in the placenta were significantly associated with maternal plasma POPs concentration. The findings suggest that placental DNA methylation and gene expression mechanism may be involved in the prenatal toxicity of POPs and their association with neonatal anthropometry measures.


Background
Persistent organic pollutants (POPs) have been used for decades in a large variety of products. Despite the international consensus to reduce or ban these chemicals, exposure persists, mainly through diet, with concentrations still detected in human serum, including in US pregnant women [1] and newborns [2][3][4]. POPs are ubiquitous endocrine-disrupting compounds (EDC) that interfere with maternal hormones and can impact fetal development and health in later life. Studies have reported that maternal levels of POPs during pregnancy are associated with decrements in fetal growth and birth weight [5][6][7][8][9], neurodevelopmental disorders [10], earlier age of menarche [11], and genitourinary conditions in offspring [12].
The mechanisms underlying the relationships between maternal exposure to these chemicals and fetal outcomes are not yet fully understood. Alterations in placental development have been reported in response to maternal EDC exposure [13], such as modification of the size of the placenta in mice [14], and degeneration of placental trophoblast in rats [15]. Therefore, changes in DNA methylation in the placenta may be one of the potential mechanisms that explain the impact of POPs on human fetal outcomes [16]. Existing studies of POPs and placental DNA methylation were based on candidate gene-based approaches [17][18][19][20]. Two studies among 109 pregnant women from the CHECK (Children's Health and Environmental Chemicals in Korea) cohort reported associations of β-hexachlorocyclohexane (β-HCH) with decreased methylation in LINE-1 (a surrogate marker of global methylation) and p,p′-dichlorodiphenyltrichloroethane (p,p′-DDT) with increased methylation of IGF2 (implicated in placental and fetal growth) [17] and MCT8 among boys [18]. P,p′-dichlorodiphenyldichloroethylene (p,p′-DDE) and polybrominated diphenyl ether-47 (PBDE 47) were significantly associated with increased methylation in DIO3 among female infants [18]. Higher PBDE 66 in cord blood was associated with decreased placental methylation in LINE-1 and higher PBDEs 153 and 209 with decreased placental methylation of IGF2 among Chinese [20] while others did not find any associations [19]. Most studies have investigated individual POPs, although pregnant women are exposed to a mixture of chemicals [1]. In this study, we assessed placental DNA methylation related to individual POPs and to chemical classes (i.e., sum of POPs in each chemical class).
We performed an epigenome-wide association study (EWAS) to identify placental DNA methylation associated with maternal plasma concentration of POPs in early gestation (10 weeks 0 days to 13 weeks 6 days) among 260 pregnant women participating in the Eunice Kennedy Shriver National Institute of Child Health and Human Development's (NICHD) Fetal Growth Studies-Singletons cohort (which comprised 2802 pregnant women from 12 clinic sites within the USA). Genes annotated to the differentially methylated CpG sites were tested for enrichment of molecular pathways. We assessed correlations between DNA methylation at the POP-associated CpG sites and placental expression of the annotated genes. Further, to evaluate the relevance of the differentially methylated CpG sites to fetal growth, we examined correlations of methylation and gene expression with birth weight, birth length, and head circumference. Lastly, we investigated the potential mediating pathway from POPs to neonatal anthropometry measures via placental DNA methylation at the POP-associated CpG sites that were correlated with neonatal anthropometry measures (i.e., birth weight, birth length, and head circumference).

Correlations between DNA methylation and gene expression
Out of the 214 differentially methylated CpG sites, 24 CpG sites were correlated with placental gene expression of 21 unique genes (p values < 0.05, Table 4, Supplementary Table S5). The strongest correlation was between cg12089439 and expression of TUSC3 (r = − 0.55, p value = 3.70 × 10 −6 ). Further examination of the correlations between individual POP concentrations and gene expression levels found that higher maternal transnonachlor concentration was marginally correlated (r = − 0.25, p value = 0.05) with decreased expression of SH3PXD2B (Supplementary Table S6). This is consistent with our DNA methylation analysis finding where higher maternal plasma concentration of trans-nonachlor was associated with decreased methylation at cg00718518 (SH3PXD2B) ( Table 2), and lower methylation at cg00718518 was correlated with decreased expression of SH3PXD2B (Table 4). SH3PXD2B displayed the highest expression in female reproductive tissues (Supplementary Figure S4).

Canonical pathways
The genes mapping the differentially methylated CpG sites were enriched in Ingenuity Pathway Analysis (IPA) canonical disease and function pathways (Supplementary Table S7). Genes annotating the CpG sites differentially methylated with increasing levels of PBDE 47 were enriched in disease and functional annotation pathways including differentiation of embryonic cells (p value = 5.52 × 10 −7 ). Genes annotating CpG sites associated with PFAS exposure were enriched in disease and functional annotation pathways, including size of the brain (p value = 3.39 × 10 −4 ) and morphologies of the central nervous system (p value = 2.41 × 10 −4 ), brain (p value = 9.75 × 10 −4 ), and head (p value = 1.70 × 10 −3 ). The top IPA canonical pathways included Estrogen Receptor Signaling  Table S9).

Functional annotation analyses
To understand whether single nucleotide polymorphisms (SNPs) influence DNA methylation of the top-associated CpG sites, we assessed methylation quantitative trait loci (cis-meQTLs) within 1 mega base distance from the differentially methylated CpG sites in published databases of meQTLs in the placenta [21,22] and blood (including cord blood and peripheral blood samples from whole blood, buffy coats, white blood cells, and blood spots) [23]. In the placenta, cg12599971 (GRAMD3) and cg18145877 (TUSC3) that were associated with PFUnDA concentration in our study have been reported to be cis-meQTL targets for SNPs in GRAMD3 and TUSC3 (Supplementary Table S10). Both genes displayed the highest expression in the placenta (Supplementary Figure S3). In the blood, 22 CpG sites that were associated with transnonachlor, PBDE 47, PFDA, and PUnDA in our study have been reported to be cis-meQTL targets for 3831 unique SNPs located in/near 104 genes (Supplementary  Table S11), including SH3PXD2B and TUSC3.

Correlations between top CpG sites and neonatal anthropometry measures
Previously, we found significant associations between maternal plasma concentrations of specific POPs and fetal growth and birth anthropometry measures [8,9].
To examine whether the change in methylation at the CpG sites significantly associated with POPs in the present analysis are related to neonatal anthropometry, we tested the correlations of methylation levels at each of the 214 CpG sites with neonatal anthropometry measures (birth weight, birth length, and head circumference). Among them, 44 CpG sites were correlated with at least one neonatal anthropometry measure (Supplementary  Table S12). We conducted mediation analysis to investigate whether placental DNA methylation at those 44 CpG sites was in the pathway between POPs and neonatal anthropometry. Five CpG sites (cg02584377 (BCL7A), cg14402591 (SEMA6B), cg18601261 (STK38L), cg07463167 (MAP4K5), and cg04789362 (KIAA0284)) significantly mediated the inverse association between PBDE 47 and birthweight (Table 5). In addition, two CpG sites (cg01327147 (KIAA1161) and cg04789362 (KIAA0284)) significantly mediated the inverse association between PBDE 47 and head circumference. PBDE 47 was negatively correlated with birth weight (r = − 0.16, p value = 0.01) and head circumference (r = − 0.16, p value = 0.01) (Fig. 3). Out of the 21 genes for which expression levels were correlated with DNA methylation at the top-significant CpG sites, higher expressions of TUSC3 were positively correlated with increased neonatal length (r = 0.26, p value = 0.04) (Supplementary Table S13).

Discussion
In this study, we assessed the relation between maternal plasma POP concentrations during early pregnancy and genome-wide DNA methylation in the placenta. Specific maternal plasma POP concentrations were associated with 214 differentially methylated CpG sites. Of the 214 CpG sites, 24 were correlated with placental expression of the annotated genes. We found strong evidence for association of maternal plasma PFUnDA concentration with TUSC3 based on consistent findings from DNA methylation, gene expression, and meQTL analyses. Similarly, we found consistent evidence from DNA methylation and gene expression data on the impact of maternal plasma concentration of trans-nonachlor on SH3PXD2B. The correlations between DNA methylation at the POPs-associated CpG sites and neonatal anthropometry suggest that placental epigenetic mechanisms may underlie the influence of specific maternal plasma POP concentrations on fetal growth.
Two differentially methylated CpG sites identified in the present study have been associated with exposure to chemicals in previous EWAS analyses: cg02343823 (ZNF300) associated with PBDE 47 in our study has been associated with polybrominated biphenyl in adult blood [24], and cg18145877 (TUSC3) associated with PFUnDA in our study has been associated with PCBs in peripheral blood leucocytes [25]. Furthermore, CpG sites   Table S14). For example, CUX2, FAM150A, and FTCD-annotated CpG sites were associated with specific PFASs in our study as was PFAS concentrations in cord blood [26]. Many genes identified as being associated with maternal plasma concentration of PBDE 47 in our study have been associated with polybrominated biphenyl in adult blood (i.e., HLA-DM, CIB4, ARHGEF10L, BCL11A, CDS2, ENO1, HOXB7, INHBA, PCBP2, SDK1, SPRED2, ZBTB38, ZEB2, ZHX2, ZNF300, and ZNF710) [24]. The SH3PXD2B gene found to have methylation at cg00718518 and gene expression signatures associated with trans-nonachlor in our study is a genome-wide association study (GWAS) locus for waist-hip ratio [27,28], body weight [29], and balding measurement [27,30]. Methylation at cg00718518 (SH3PXD2B) in liver tissue biopsies has been previously associated with obesity [31]. Mediation analysis revealed a potential explanation of the association between PBDE 47 and smaller birth weight through placental DNA methylation at MAP4K5.
These findings support earlier studies that reported associations between POPs and birth weight [32] and a study in mice that highlighted activation of mitogen-activated protein kinases (MAPK) in the placenta of mice treated with PBDE 47 [33]. We also observed positive association between DNA methylation at another mitogenactivated protein kinase gene (MAPK8IP1) and PFUnDA. Both MAP4K5 and MAPK8IP1 placental DNA methylation levels were negatively correlated with birth weight in our study. Furthermore, PBDE 47 was associated with increased DNA methylation at cg04789362 (KIAA0284), and higher methylation at cg04789362 was associated with smaller weight and head circumference at birth. PBDE 47 was also associated with increased DNA methylation at cg01327147 (KIAA1161), and higher DNA methylation at cg01327147 was associated with decreased head circumference. Differential methylation at cg01327147 has been previously associated with neurodevelopmental syndromes [34]. The KIAA1161 gene is known to play a role in brain calcification [35], abnormal cerebellum morphology, and functional neurological abnormalities related to dysfunction of the   [36]. Moreover, our mediation analysis suggested relations between PBDE 47 and head circumference via placental DNA methylation (cg01327147 (KIAA1161) and cg04789362 (KIAA0284)). Together, our study corroborates previous findings on the impact of maternal plasma POP concentrations and placental methylation, and adds to recent EWAS evidence for 15 novel placental methylation sites that could potentially impact placental function and fetal development [37]. We found strong corroborative evidence of exposure to PFUnDA on methylation and gene expression of the TUSC3 gene, which is highly expressed in the placenta. In addition, placental methylation at cg18145877 (TUSC3) and expression of TUSC3 gene was correlated with birth length. A published placental meQTL study [21] showed that sequence variants regulate methylation of cg18145877 of TUSC3 in the placenta. TUSC3 gene is a protein coding gene associated with several biological functions including cellular magnesium uptake, protein glycosylation, and embryonic development. TUSC3 is a GWAS locus for mental health disorders and general cognitive ability including educational attainment and mathematical ability [38], obsessive-compulsive disorder [39], and schizophrenia [40]. In previous EWAS analyses, methylation at cg18145877 (TUSC3) in cord blood has been associated with prenatal arsenic exposure [41] and PCB 156 exposure [25]. PFUnDA was observed to be associated with decreased methylation at cg24722365 (KIF1B), and lower methylation at cg24722365 was associated with smaller birth length. KIF1B has been associated with body height in a published GWAS. In the same cohort, maternal plasma PFUnDA and PFHxS concentrations have been significantly associated with decreased neonatal thigh length at birth [8]. Studies have shown associations between prenatal exposure to PFAS and bone development [42,43]. Our findings may point to potential Corr. Spearman correlation, ACME average causal mediation effects, ADE average direct effect, TE total effect P-values: * < 0.10; **p < 0.05; ***p < 0.01; ****p < 0.001 Adjusted for maternal self-reported race/ethnicity, maternal age in years, fetal sex, maternal pre-pregnancy BMI, cotinine level, total lipids (except PFASs), methylation sample plate, first three methylation principal component (PCs), and the first 10 genotype PCs pathway through placenta DNA methylation; however, the mediation analysis results were not significant. Our study had several potential limitations that need to be considered in weighing our results. We are not aware of data from similarly-designed cohort studies, limiting our ability to replicate our findings in other independent populations. However, we were able to enrich the interpretation of our results using previous EWAS analyses involving biospecimens collected from adults or cord blood. There may be spatial and cell populationbased differences in gene expression in the placenta, which was beyond the scope of our study. Our EWAS was able to identify modest methylation changes associated with maternal plasma POPs concentration, but we acknowledge that studies with larger sample sizes are needed to detect CpG sites with smaller methylation changes. The study was undertaken at 12 clinical sites that could potentially be related to differential methylation. However, in a sensitivity analysis evaluating models using the Akaike information criterion (AIC), 206 out of 214 differentially methylated CpG sites (96.3%) were better explained in models without than models with clinical sites, and all association β values were comparable (Supplementary Table S15). Finally, as chemicals mixture may be more complex than the sum of chemicals, we encourage future studies to further investigate the potential implication of interaction and non-linear relationships of POPs on placental epigenetic changes. Our study had several strengths. We adjusted our analysis for genotype-based principal components (PCs) in addition to methylation-based PCs, effectively minimizing spurious associations due to population stratification [44]. To date, there is no reference for cell type composition for placenta; therefore, we implemented a validated reference-free adjustment for cell type proportion variation implemented in SVA [45] and further corrected our analysis for genomic inflation using BACON, a method demonstrated to maximize study power while controlling for false discovery rate [46]. We also integrated our EWAS findings with placental gene expression and neonatal anthropometry measures to highlight potential mechanisms of prenatal POP toxicity through placental epigenetic changes.

Conclusions
Findings from the present study suggest that maternal plasma concentrations of specific POPs may influence placental DNA methylation resulting in differences in neonatal birth size. Furthermore, we observed strong evidence consistently supporting a role for PFUnDA concentration on TUSC3 including placental DNA methylation, gene expression, and placental meQTL analyses. Taken together, these findings shed light on potential placental epigenetic mechanisms that may explain associations between prenatal exposure to POPs and birth outcomes.

Study population
This analysis included 260 pregnant women from the NICHD Fetal Growth Studies-Singleton cohort-who provided placenta samples at delivery and had POPs concentration measures. Briefly, the NICHD Fetal Growth Studies-Singleton cohort included 2802 pregnant women enrolled between 8 weeks and 6 days and 13 weeks and 6 days between July 2009 and January 2013 from 12 clinic sites within the USA [47]. Eligible women could not have a past history of adverse pregnancy outcomes or self-reported behavioral risk factors such as use of cigarettes, illicit drugs, or alcohol in the months prior to pregnancy. The study was approved by institutional review boards at NICHD, all participating clinical and laboratory sites, and the data coordinating center.

Environmental exposure data
For this ancillary study, all maternal blood samples used were collected at enrollment (10 weeks 0 days to 13 weeks 6 days) in phlebotomy equipment determined to be free of the POP contaminants under study. A total of 76 persistent organic pollutants (POPs) were measured, as described in a previous published paper [8]. We excluded chemicals where more than 20% of the concentrations were below laboratory LOQ. To capture the effect of the total exposure of each chemical class (i.e., OCPs, PCBs, PFASs), we summed individual chemicals within a class assuming additivity.
Briefly, PFASs were quantified using 200 μl of plasma, and PBDEs, PCBs, and OCPs were quantified using 1 ml of plasma. All samples were shipped in dry ice to the Wadsworth Center, New York State Department of Health [8]. Total plasma lipids were calculated using the short formula: total lipids (in ng/mL) = 2.27 * total cholesterol + triglycerides + 62.3 [48,49], where total cholesterol and triglycerides (in nangrams per millimeters) were measured from non-fasting plasma stored plasma in − 80°C freezers using the Roche COBAS 6000 chemistry analyzer (Roche Diagnostics, Indianapolis, IN) [50].
Machine-measured POP concentrations were modeled without substituting concentrations below the LOQ with a constant to minimize bias introduced when assessing health outcomes [51]. For analysis, POP concentration were log (1 + chemical) transformed for all POPs except for p,p′-DDE, PCB #138/158, sum of OCPs, and sum of PCBs where a log (10 + chemical) transformation was used. All concentrations were then rescaled by their standard deviations to provide results in interpretable units, i.e., change per 1 SD for each and summed POPs.

Placental DNA methylation measurement and quality control
Placental samples (n = 312) were obtained within 1 h of delivery. Briefly, placental parenchymal biopsies measuring 0.5 cm × 0.5 cm × 0.5 cm were taken from the fetal side, placed in RNALater, and frozen for molecular analysis, as previously described [21]. Extracted DNA was assayed using Illumina's Infinium Human Methylation450 Beadchip (Illumina Inc., San Diego, CA). Standard Illumina protocols were followed for background correction, normalization to internal control probes, and quantile normalization. Quality control procedures were followed as previously described [52]. Of the 301 pregnant women with placental DNA methylation data that passed quality filters, 260 women had POP concentrations available for analysis and represent the study cohort (260/ 312, 83.3%). For analyses, beta values were converted to the M value scale by using the formula: M value = log2(Beta/(1-Beta)). Any resulting infinity or missing M values (6.0%) were imputed by the k-nearest neighbors method, setting k = 10 for inclusion in the analysis [53].

Placenta RNA quantification for gene expression
The placental biopsies used for DNA extraction were also used for extracting RNA from 80 placentas using TRIZOL reagent (Invitrogen, MA), and sequenced using the Illumina HiSeq2000 system. The expression of the transcripts were quantified using Salmon [54] which account for experimental attributes and biases such as fragment GC-content bias that is commonly observed in RNA-seq data. Participants with RNA-seq, DNA methylation, and POPs level at enrollment (n = 62) were used to test correlation between DNA methylation and gene expression levels.

Statistical analysis Epigenome-wide analyses
Epigenome-wide analyses were performed for each POP and each summed chemical class as the predictor and placenta DNA methylation at each CpG site as the outcome using the R/Bioconductor package "limma" [55]. Placental genome-wide SNP genotype data were used to estimate 10 genotype-based PCs representing the population structure. The R package "prcomp" was used to calculate PCs on the samples' percent methylation profiles [56]. The EWAS analysis included robust linear regression models that were adjusted for self-reported maternal race/ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, Asian), age (in years), offspring sex (male/female), pre-pregnancy BMI (kg/m 2 ), total plasma lipid concentration (ng/mL, except PFASs), logtransformed plasma cotinine level (ng/mL), methylation sample plate (n = 5), the first three methylation PCs and the first 10 genotype PCs to account for population structure [44], and putative cell-mixture estimated using surrogate variable analysis (SVA) components (n = 20) to account for latent source of noise such as batch effects and cell composition [57][58][59]. To account for the inflation of statistical test in EWAS, we implemented the Bayesian method to obtain BACON-corrected inflation estimates (Bλ) and BACON-corrected p values using the R/Bioconductor package BACON [46]. Corrected p values from BACON were then controlled for FDR [60] giving BACON-corrected FDR p values. Quantilequantile plots of p values and the corresponding inflation estimates before (λ) and after (Bλ) BACONcorrection are reported in Supplementary Figure S4. The inflation statistic was close to 1 after BACON correction (Supplementary Figure S4). Genome-wide Manhattan plots were used to report results from EWAS. The residual-vs-fitted plots for the first top hits of each EWAS are presented in Supplementary Figure S5.
To identify genomic regions that are differentially methylated with maternal plasma concentrations of POPs, we implemented DMR analysis using the R package dmrff [61]. Significant DMR was defined based on three criteria: (1) the DMR length can be at most 500 base pairs; (2) the region has EWAS FDR p values < 0.05, and (3) EWAS effect estimates for the CpGs have the same direction.

Gene expression analyses
We estimated the correlation between mRNA levels of the genes mapping to POP-associated CpG sites in our study (BACON-corrected FDR p values < 0.05) and placenta DNA methylation and corresponding maternal POPs blood concentration using Spearman correlations.

Canonical pathway analysis
Genes annotating to the epigenome-wide significant CpG sites (BACON-corrected FDR p values < 0.05) were further explored. We identified canonical pathway, networks, and diseases and biological function involved using the "Core Analysis" function in Ingenuity Pathway Analysis (IPA, QIAGEN Redwood City, CA, USA, www. qiagen.com/ingenuity). Diseases and function annotation with 3 molecules or more were considered. Statistical significance of overrepresented canonical pathways was determined using Fisher's exact test after adjustment for multiple testing using the Benjamini-Hochberg method [60].

Functional annotation and analysis
To identify SNPs that may influence DNA methylation at the CpG sites in cis (i.e., within 1 megabase on either side of a gene's transcript start site (TSS) [62]), we examined the significant CpG sites (BACON-corrected FDR p values < 0.05) in a list of SNPs that are known to be meQTL in the placenta from two previously published papers [21,22].
Additional file 5: Figure S5. Residual-vs-fitted plots for the top hits of each EWAS. For each figure, one color represents one CpG site.
Additional file 6: Supplementary Tables: Table S1. Characteristics of the study subsample and of the full NICHD fetal Growth Studies -Singletons. Table S2: Description of the maternal plasma persistent organic pollutants concentrations. Table S3: Top-significant adjusted difference (BACON-corrected FDR p-values < 0.05) in placenta DNA methylation associated with maternal POP. Table S4: Differentially methylated regions (DMR) analysis. Table S5: Correlations between DNA methylation at the top differentially methylated CpG sites and gene expression in placenta. Table S6: Significant associations between maternal blood levels of POP and expression of genes near the top significant DNA methylation CpG sites. Table S7: Top 10 pathways of diseases and biological function from IPA. Table S8: IPA Canonical Pathway. Table S9: Network identified by Ingenuity Pathway Analysis. Table S10: Placental cis-eQTL analysis of top-significant (Bacon-adjusted FDR P-value<0.05) CpGs. Table S11: Cis-meQTL analysis of top-significant (Bacon-adjusted FDR P-value<0.05) CpGs. Table S12: Spearman correlation between top differentially methylated CpG sites and neonatal anthropometry. Table S13: Spearman correlation between neonatal anthropometry and gene-expression that were significantly correlated with the methylation on the corresponding CpG sites. Table S14: Comparison with previous EWAS on chemicals. Table S15: Sensibility analysis further adjusted for clinical sites for the CpG sites differentially methylated in the original model.