Ambient air pollution during pregnancy and DNA methylation in umbilical cord blood, with potential mediation of associations with infant adiposity: The Healthy Start study

Abstract

Background:

Prenatal exposure to ambient air pollution has been associated with adverse offspring health outcomes. Childhood health effects of prenatal exposures may be mediated through changes to DNA methylation detectable at birth.

Methods:

Among 429 non-smoking women in a cohort study of mother-infant pairs in Colorado, USA, we estimated associations between prenatal exposure to ambient fine particulate matter (PM_2.5) and ozone (O₃), and epigenome-wide DNA methylation of umbilical cord blood cells at delivery (2010–2014). We calculated average PM_2.5 and O₃ in each trimester of pregnancy and the full pregnancy using inverse-distance-weighted interpolation. We fit linear regression models adjusted for potential confounders and cell proportions to estimate associations between air pollutants and methylation at each of 432,943 CpGs. Differentially methylated regions (DMRs) were identified using comb-p. Previously in this cohort, we reported positive associations between 3^rd trimester O₃ exposure and infant adiposity at 5 months of age. Here, we quantified the potential for mediation of that association by changes in DNA methylation in cord blood.

Results:

We identified several DMRs for each pollutant and period of pregnancy. The greatest number of significant DMRs were associated with third trimester PM_2.5 (21 DMRs). No single CpGs were associated with air pollutants at a false discovery rate <0.05. We found that up to 8% of the effect of 3^rd trimester O₃ on 5-month adiposity may be mediated by locus-specific methylation changes, but mediation estimates were not statistically significant.

Conclusions:

Differentially methylated regions in cord blood were identified in association with maternal exposure to PM_2.5 and O₃. Genes annotated to the significant sites played roles in cardiometabolic disease, immune function and inflammation, and neurologic disorders. We found limited evidence of mediation by DNA methylation of associations between third trimester O₃ exposure and 5-month infant adiposity.

Introduction

Exposure to ambient air pollution during pregnancy has been linked to adverse offspring health outcomes, including greater risk of infant low birth weight and preterm birth (1–8). The strength of such associations may differ by the timing of exposure during gestation (9–15). Air pollution exposures in utero have also been associated with childhood health outcomes, including altered growth and body composition (16–18) and increased blood pressure (19). Prenatal environmental exposures may have delayed or lasting effects on offspring phenotype by causing epigenetic modifications to offspring DNA at sensitive periods of development (20).

Short-term and long-term exposures to various air pollutants have been linked to differences in DNA methylation in peripheral blood of adults (21–26) and children (27). Furthermore, epidemiologic studies of air pollution exposures during pregnancy have demonstrated epigenetic changes to the placenta, which may be particularly relevant for impacts on fetal growth (28–36). However, relatively few studies have examined air pollution-associated changes to DNA methylation in umbilical cord blood cells collected at birth. DNA methylation differences in cord blood may reflect in utero programming of the child’s DNA by the intrauterine environment, with consequences for childhood health outcomes including body weight and excess adiposity (37–40), asthma (41), and cognitive and psychiatric disorders (42, 43).

A limited number of previous studies have reported air pollution-associated epigenetic differences in cord blood. Some have analyzed global methylation (44, 45) or methylation at long interspersed nuclear elements (LINE-1) (46–48), while others have examined methylation at target sites within selected genes (49–54). Still fewer studies have examined epigenome-wide associations in cord blood with air pollution measures or proxies (29, 55, 56), including two meta-analyses of ambient nitrogen dioxide (NO₂) and fine particulate matter (PM_2.5), respectively, in several US and European cohorts (57, 58). However, previous studies have not examined trimester-specific exposures to identify potential sensitive windows, nor have they explored the potential for locus-specific differences in cord blood DNA methylation to mediate associations between prenatal air pollution exposure and offspring obesity-related outcomes. Previous studies focused on other exposures have identified differentially methylated positions in cord blood DNA that predict adiposity in infancy (38), early childhood (37–40) and mid-childhood (37).

In our previous work in the Denver, Colorado-based Healthy Start cohort, we observed a positive association between maternal exposure to ozone (O₃), but not PM_2.5, in the 3^rd trimester and greater offspring adiposity at 5 months of age, but no exposure-related differences in infant weight or adiposity at birth (59). Notably, the Colorado Front Range region including Denver has been classified as a “serious nonattainment area” by the U.S. Environmental Protection Agency due to O₃ levels routinely exceeding federal air quality standards (60). We hypothesized that prenatal air pollution exposures may produce delayed effects on offspring growth and body composition, potentially via changes to DNA methylation detectable in cord blood at delivery. In this study, we aimed to estimate associations between trimester-specific and full-pregnancy ambient PM_2.5 and O₃, and umbilical cord blood DNA methylation in the Healthy Start cohort. Additionally, we estimated the potential for mediation by cord blood methylation at selected CpGs of the previously observed association between prenatal exposure to O₃ and infant adiposity at 5 months of age.

Methods

Study population and eligibility

Pregnant women at ≤24 weeks of gestation were recruited into the Healthy Start cohort study (2009–2014) from the University of Colorado Hospital clinics. Women were excluded from enrollment if they had multiple gestation, were younger than 16 years of age, had a history of stillbirth or preterm birth prior to 25 weeks of gestation, or were currently receiving treatment for any of the following conditions: diabetes, cancer, asthma, or psychiatric illness. Participants provided written informed consent. The study protocol was reviewed and approved by the Colorado Multiple Institutional Review Board.

A total of 1,410 pregnant women enrolled in the Healthy Start cohort. For this study, eligible participants were identified among 600 women who previously had cord blood analyzed for DNA methylation markers as described below. Participants were not eligible for this analysis if their residential address was located outside of the Denver, Colorado metropolitan area or if we were unable to geographically locate the address (n=39). Additionally, we excluded women who reported sustained maternal smoking during pregnancy (n=51) or had missing data on smoking during pregnancy (n=77), leading to a potentially eligible sample size of 433 mother-infant pairs, with 4 subsequent exclusions for methylation quality control and some additional missing period-specific exposure data, as described below.

Prenatal ambient air pollution exposure assessment

We obtained and geocoded participant residential addresses at study enrollment. We obtained daily or hourly monitoring data for the entire study period (2009–2014) from the US EPA AQS Data Mart and the Colorado Department of Public Health and Environment. As previously described in detail (59), we calculated average daily values from each monitor for the first trimester, second trimester, third trimester, and across the full pregnancy (estimated conception to delivery date). We averaged hourly O₃ values across each 8-hour interval to obtain the maximum daily 8-hour average for each 24-hour period. We applied the following weighting formula in averaging values from all monitors for a given residence: 1/distance-squared. All monitors with non-missing data within 50km of the participant’s address were included in the inverse-distance-weighted average (up to 10 monitors for PM_2.5 and 19 for O₃). For most periods of pregnancy, data from a median of 6 PM_2.5 monitors and 11 O₃ monitors was included in the weighted average (Supplemental Table S1).

Infant adiposity at 5 months of age

Infant body composition was evaluated at approximately 5 months of age using air displacement plethysmography (PEA POD, COSMED, Rome, Italy). Infants wore a dry diaper and a tight-fitting synthetic cap to cover the hair. The PEA POD uses measured body density to estimate fat mass and fat-free mass using an infant-specific algorithm. We defined adiposity as fat mass as a percentage of total body mass. Each infant was evaluated twice, and a third measurement was taken if adiposity estimates differed by >2%. The closest two measures were then averaged to obtain the infant adiposity variable used in this study.

Other variables

At study enrollment, participants reported their race, ethnicity, smoking history, and educational level attained via questionnaire. Participants repeatedly reported recent smoking behavior, twice during pregnancy and once shortly after delivery. Smoking was classified as “sustained” during pregnancy if women reported smoking at least 100 cigarettes during their lifetime in addition to any smoking after study enrollment; if women quit smoking early in pregnancy (prior to enrollment) then they were considered not sustained smokers and therefore eligible for this analysis. We obtained information on infant sex and gestational age at birth from medical records. Season of birth was based on the month of delivery and was classified as winter (December-February), spring (March-May), summer (June-August), or fall (September-November). We obtained hourly temperature data from up to 16 monitors within 50km of any participant residential address and calculated inverse-distance-weighted daily average temperature for each pregnancy period, using an approach parallel to that described above for the air pollutants.

Analysis of DNA Methylation in Cord Blood

We used the Illumina Infinium HumanMethylation450 array to evaluate genome-wide DNA methylation in umbilical cord blood (61). Methods and quality control procedures for DNA extraction and bisulfite conversion are described in our previous publication (62). Briefly, the QIAamp DNA Blood Mini Kit (Qiagen) was used for extraction of DNA from stored buffy coats. Quality control checks excluded samples that did not meet the following criteria: 260/280 ratio >1.8 indicating DNA purity, DNA integrity score >7, and at least 500 ng of DNA available. We estimated the relative proportions of seven cell types in umbilical cord blood based on a combined cord blood reference dataset (63) using the R package FlowSorted.CordBloodCombined.450k (R version 3.6.2, R Foundation for Statistical Computing). After initial quality control procedures to exclude probes with high detection p-value or low bead count (62), we then excluded 11,335 probes located on the X or Y chromosome, and 39,983 cross-reactive or polymorphic probes using the rmSNPandCH function in the R package DMRcate (version 2.2.3), based on Chen et al. (64). We logit-transformed raw methylation (beta) values to M-values to meet modeling assumptions. We performed stratified quantile normalization with the preprocessQuantile function in minfi (65).

We removed extreme methylation outliers, defined as M-values more than three times the interquartile range above the 75^th percentile or below the 25^th percentile (66, 67). This resulted in missing values (0.1% of all M-values), which we replaced using single imputation with the R package impute, version 1.62.0 (68) to facilitate surrogate variable analysis, described below. We excluded 4 participants for whom the reported sex of the infant did not match the predicted sex based on methylation data (n=4).

Statistical Analysis

We employed univariate analyses and visualizations to check for missing or implausible values of predictor variables. We used separate linear regression models to estimate associations between continuous PM_2.5 or O₃ averaged across trimester 1, trimester 2, trimester 3, or the full pregnancy, and methylation (M-values) at each of 432,943 CpG sites remaining after filtering. We addressed potential confounding by technical batch by adjusting for one batch variable and for 5 surrogate variables generated with the R package sva, version 3.36.0 (69). We constructed a directed acyclic graph based on previous literature (Supplemental Figure S1), and all models were additionally adjusted for the following potential confounders: maternal education (2 categories: completed high school or less, more than high school education), maternal race/ethnicity (3 categories: non-Hispanic white, Hispanic, and all others), season of birth (4 categories: winter, spring, summer, fall), and average temperature during the pregnancy period of interest. Models were additionally adjusted for the following precision variables: infant sex, gestational age at birth (days), and estimated proportions of seven cell types (CD4+ T cells, CD8+ T cells, B cells, granulocytes, monocytes, NK cells, nucleated red blood cells). Because cell type proportions were not associated with air pollution exposures (Supplemental Table S2) and therefore not traditional confounders, we also examined results from models without cell type adjustment. We adjusted p-values for multiple testing using the Benjamini-Hochberg procedure (70) to control the false discovery rate (FDR). We performed gene set analysis using the methylglm function from the R package methylgsa (71) to identify enrichment of pathways from the Gene Ontology (GO) and Reactome databases.

We additionally compared our PM_2.5 results to the findings of a previously published meta-analysis of full-pregnancy PM_2.5 exposure and epigenome-wide methylation in umbilical cord blood (57). For each of the 11 CpGs significantly associated with PM_2.5 in the previous study and not excluded from our analysis, we examined whether the association between PM_2.5 in each pregnancy period and cord blood DNA methylation at that CpG was nominally significant (p<0.05) in our results.

Differentially methylated regions (DMRs) were identified using comb-p (72), using as seeds the CpGs with raw p-values <0.1 from analysis of differentially methylated positions (DMPs). We merged peaks within 750 bps into a single DMR. We excluded DMRs containing fewer than 3 probes. We defined significance for DMRs as Šidák-adjusted p-value <0.05. We calculated the proportion of CpGs in each region with methylation positively associated with the exposure, and we confirmed that >80% of probes in each region had a consistent direction. We annotated DMRs via chromosome and location to associated genes using BioMART (73). We searched the NCBI Gene database to identify functions of genes annotated to each significant DMR, with a specific focus on health outcomes previously linked to air pollution exposure: adiposity, cardiometabolic disease, inflammation and neurologic function.

In a previous analysis in the same cohort using the same exposure data, we observed a positive association between 3^rd trimester O₃ concentrations (but not PM_2.5) and infant adiposity at 5 months of age (59). Here, we explored the potential for epigenetic markers to mediate the previously observed association, among the subset of participants with complete data on both cord blood DNA methylation and 5-month adiposity. We evaluated as potential mediators the top CpG (by p-value) from each of the 9 significant DMRs associated with 3^rd trimester O₃ concentrations. We first confirmed that 3^rd trimester O₃ remained associated with 5-month adiposity in the subset of participants with complete adiposity data, and using the same covariate set employed in the epigenome-wide analysis but excluding surrogate variable adjustment. We then included two additional variables as precision covariates for the outcome of 5-month adiposity: infant age at the time of body composition measurement, and maternal pre-pregnancy BMI.

For those CpGs with consistent direction of association with both the exposure (third trimester O₃) and the outcome (5-month adiposity), we estimated the proportion of the total effect mediated through changes in DNA methylation, using the model-based tools for causal mediation analysis as implemented in the R package mediation (74). The mediate function calculates the average causal mediation effect, or indirect effect, using a potential outcomes framework. Specifically, the indirect effect is the expected difference in the outcome when treatment (exposure) is held constant but the mediator is changed from its expected value in the absence of exposure, to its expected value in the presence of exposure. The total effect of third trimester O₃ on 5-month adiposity was decomposed into a direct effect and an average causal mediation effect, or indirect effect, by separately fitting a mediator model and an outcome model.

We constructed a separate directed acyclic graph to represent the hypothesized associations between the exposure, mediator, outcome, and potential confounders of the exposure-outcome association, exposure-mediator association, and mediator-outcome association (Supplemental Figure S2). The mediator model regressed methylation (M-value) at each CpG on average 3^rd trimester O₃, adjusted for maternal education, race/ethnicity, pre-pregnancy BMI, infant sex, gestational age at birth, season of birth, average temperature during the 3^rd trimester of pregnancy, age at infant adiposity measurement, one technical batch variable and seven estimated cell proportions. The outcome model regressed infant adiposity on 3^rd trimester O₃ adjusted for all of the covariates in the mediator model, and additionally adjusted for the M-value at the CpG (the hypothesized mediator). We tested for exposure-mediator interaction using the test.TMint function within the mediation R package.

Results

Following exclusions, there were 429 mother-infant pairs included in this analysis. Characteristics of these 429, and of the subset of 294 with complete data on infant adiposity at 5 months of age, are presented in Table 1. The participants included in this analysis were generally representative of the larger Healthy Start cohort (Supplemental Table S3), with the notable exception that we excluded women who reported sustained smoking during pregnancy from this analysis.

Table 1.

Characteristics of 429 eligible mother-infant pairs in epigenome-wide analysis, and a subset of 294 with infant adiposity assessed at 5 months of age.

	Mean ± SD or N (%)
	Main analysis (n=429)	With 5-month infant adiposity data (n=294)
Infant sex
Male	224 (52)	149 (51)
Female	205 (48)	145 (49)
Race/ethnicity
Non-Hispanic white	218 (51)	156 (53)
Hispanic	115 (27)	77 (26)
Non-Hispanic Black	69 (16)	45 (15)
All other race/ethnicity groups	27 (6)	16 (5)
Birth weight (g)	3279 ± 431	3267 ± 445
Gestational age at birth (days)	276 ± 9	276 ± 9
Season of birth
Winter	98 (23)	71 (24)
Spring	103 (24)	71 (24)
Summer	132 (31)	77 (26)
Fall	96 (22)	75 (26)
Maternal educational level attained
More than high school	289 (67)	210 (71)
High school or less	140 (33)	84 (29)
Maternal age at delivery (years)	27 ± 6	28 ± 6
Maternal BMI prior to pregnancy (kg/m2)	26 ± 7	26 ± 7
Infant age (days) at follow-up	--	152 ± 35
Infant adiposity (%) at follow-up	--	25 ± 6
Mean PM2.5 in trimester 1 (μg/m3)a, b	7.6 ± 1.0	7.5 ± 0.9
Mean PM2.5 in trimester 2 (μg/m3)a, b	7.5 ± 1.0	7.4 ± 1.0
Mean PM2.5 in trimester 3 (μg/m3)a, b	7.5 ± 1.1	7.6 ± 1.1
Mean PM2.5 across full pregnancy (μg/m3)	7.5 ± 0.6	7.5 ± 0.5
Mean O3 in trimester 1 (ppb)	43 ± 11	44 ± 11
Mean O3 in trimester 2 (ppb)	43 ± 10	43 ± 10
Mean O3 in trimester 3 (ppb)	45 ± 11	44 ± 11
Mean O3 across full pregnancy (ppb)	44 ± 4	44 ± 4

^aMissing data by variable among the 429 participants in the main analysis: PM_2.5 in trimester 1, 40; Mean PM_2.5 in trimester 2, 29; PM_2.5 in trimester 3, 8.

^bMissing data by variable among the 294 participants in the main analysis: PM_2.5 in trimester 1, 30; Mean PM_2.5 in trimester 2, 17; PM_2.5 in trimester 3, 6.

Average concentrations of PM_2.5 during the pregnancy periods were relatively low in the study area compared to many other urban areas in the US, with a median 24-hour average concentration of 7.5 μg/m³ (standard deviation 0.6 μg/m³) across the full pregnancy (Table 1). Average daily 8-hour maximum concentrations of O₃ had a mean ± standard deviation of 44 ± 4 ppb across the full pregnancy. Due to strong seasonal trends in air pollution, exposures tended to be negatively correlated between trimesters and PM_2.5 and O₃ measures were negatively correlated with each other within the same trimester, as expected (Supplemental Table S4).

Epigenome-wide analyses showed no notable evidence of genomic inflation, with all lambda values between 0.968 and 1.015 (Supplemental Table S5). Models without cell type adjustment had a slightly wider range of lambda values (0.951–1.057). In fully adjusted models, there were no differentially methylated positions associated with PM_2.5 or O₃ at a false discovery rate (FDR) of <0.05. All CpGs with p<0.05 for each pollutant and period are presented in Supplemental Tables S6–S13. Although we present the cell type-adjusted estimates as the primary results, we also provide results from models without cell type adjustment (Supplemental Tables S14–S21). Manhattan and volcano plots for all models are provided (Supplemental Figures S3–S34). Enrichment analyses identified only one significant pathway; the GO pathway for “organelle fusion” (GO:0048284) was associated with full pregnancy O3.

We identified several differentially methylated regions (DMRs) with FDR<0.05 for each pollutant and period of pregnancy exposure. For PM_2.5 exposure, there were 13 DMRs for trimester 1, 5 for trimester 2, 21 for trimester 3, and 4 for full pregnancy average PM_2.5 (Table 2). Genes annotated to these regions included some with previously reported associations with cardiometabolic disease (PON1 (75), NKAPL (76), KLB (77), HOOK2 (78), TYW3 (79)) and others associated with neurologic disorders (SLC6A3 (80), LOXHD1 (81), DLGAP2 (82, 83)), immunologic functions (CD6 (84), POU2AF1 (85)), and certain cancers (MGMT (86) ZFP57 (87), PAX8 (88)). While some genes were associated with both full pregnancy PM_2.5 exposure and one other period, no genes were associated with more than one trimester of PM_2.5 exposure. Differentially methylated regions for PM_2.5 from models without cell type adjustment are presented in Supplemental Tables S22–S25.

Table 2.

Differentially methylated regions (false discovery rate [FDR] <0.05) in cord blood associated with exposure to PM_2.5 during pregnancy among 429 eligible mother-infant pairs in the Healthy Start study (2010–2014).

Chromosome	Start position	End position	Raw p-value	Šidák-adjusted p-value	Annotated Gene(s)	Top CpG (lowest p-value) from region	Number of probes in region	Proportion of probes with positive association	Pregnancy period
18	44236847	44237852	1.41E-14	2.78E-10	LOXHD1	cg05264816	11	0	Trimester 1
7	94953653	94954253	2.85E-14	9.41E-10	PON1	cg04155289	9	1	Trimester 1
6	33282624	33283344	5.34E-11	1.47E-06	ZBTB22	cg04415168	25	0.11	Trimester 1
1	159825552	159825812	2.27E-11	1.73E-06	VSIG8	cg12217625	4	0	Trimester 1
2	72079276	72079660	3.17E-10	1.63E-05	DYSF	cg09467436	6	0	Trimester 1
15	89959984	89960794	3.40E-09	8.29E-05	LOC105371031	cg17326095	7	0	Trimester 1
17	48911983	48912815	2.36E-08	0.00056	WFIKKN2	cg14074284	11	0	Trimester 1
6	29404667	29404784	9.94E-09	0.00168	OR10C1	cg03554196	4	0	Trimester 1
4	46126066	46126424	5.64E-08	0.003109	GABRG1	cg22324939	6	0	Trimester 1
5	1394296	1394684	6.65E-08	0.003384	SLC6A3	cg26924408	5	0	Trimester 1
6	28226885	28227271	7.86E-08	0.004019	NKAPL; ZKSCAN4	cg09523275	9	0	Trimester 1
13	112860420	112860558	1.55E-07	0.02198	LINC01070	cg04038089	3	0	Trimester 1
3	23244051	23244176	1.87E-07	0.0292	UBE2E2-AS1	cg18257541	4	0	Trimester 1
2	113992762	113993364	1.85E-11	6.09E-07	PAX8; PAX8-AS1	cg07594247	8	0	Trimester 2
17	43222106	43222309	4.37E-09	0.0004253	ACBD4	cg00625783	4	0	Trimester 2
11	111250093	111250482	1.48E-08	0.0007502	POU2AF1	cg11362935	9	0	Trimester 2
16	1583810	1584169	1.73E-08	0.0009529	IFT140; TMEM204	cg00463982	6	1	Trimester 2
14	81421983	81422211	4.48E-08	0.003882	TSHR	cg19563130	5	0	Trimester 2
6	32063394	32065094	1.08E-17	1.25E-13	TNXB	cg01033871	55	1	Trimester 3
11	60775545	60776175	1.83E-12	5.75E-08	CD6	cg10981907	4	0	Trimester 3
16	85362963	85363260	6.60E-12	4.39E-07	MIR5093	cg08428188	4	0	Trimester 3
4	39448432	39449181	1.85E-11	4.88E-07	KLB	cg06235390	6	0	Trimester 3
6	31650735	31651413	4.45E-10	1.30E-05	LY6G5C	cg25417675	21	0	Trimester 3
6	29648161	29648952	1.62E-09	4.05E-05	ZFP57	cg08022281	20	0	Trimester 3
1	75198403	75199168	2.22E-09	5.73E-05	TYW3; CRYZ	cg26855724	10	1	Trimester 3
12	54070378	54070661	3.10E-09	0.0002166	ATP5MC2	cg22997177	6	0.17	Trimester 3
10	131193104	131193323	3.92E-09	0.0003536	MGMT	cg13420320	3	1	Trimester 3
22	50482925	50483596	3.31E-08	0.0009761	TTLL8	cg02879960	5	0	Trimester 3
16	57831745	57832360	3.07E-08	0.0009879	KIFC3	cg27312338	6	0	Trimester 3
2	468737	469265	2.67E-08	0.0009977	LINC01874	cg22490454	5	1	Trimester 3
7	155584080	155584470	3.48E-08	0.001765	SHH	cg09322003	4	1	Trimester 3
4	77723057	77723466	4.65E-08	0.002245	SHROOM3	cg14095959	4	0	Trimester 3
11	2890319	2890776	7.63E-08	0.003297	KCNQ1DN	cg13081704	23	1	Trimester 3
16	88590325	88590518	3.52E-08	0.003604	LOC100128882; ZFPM1	cg06913337	5	0	Trimester 3
19	12876846	12877239	9.30E-08	0.004668	HOOK2	cg23899408	4	1	Trimester 3
8	637909	638381	1.65E-07	0.006871	ERICH1	cg12641240	3	0	Trimester 3
5	50673033	50673859	4.68E-07	0.01115	LOC642366	cg10536898	5	0	Trimester 3
1	201979478	201979815	8.41E-07	0.04816	ELF3; ELF3-AS1	cg18966401	5	0	Trimester 3
8	1094575	1094955	9.75E-07	0.04948	DLGAP2	cg03636930	4	0	Trimester 3
11	89867385	89868027	2.78E-12	8.57E-08	NAALAD2	cg05500015	10	0	Full
15	89959984	89960794	4.16E-11	1.02E-06	LOC105371031	cg06059461	7	0	Full
6	29404667	29404784	1.27E-07	0.02116	OR10C1	cg03554196	4	0	Full
6	29648345	29648952	1.15E-06	0.03669	ZFP57	cg08041448	16	0	Full

In one previous meta-analysis of prenatal exposure to PM_2.5 and epigenome-wide cord blood DNA methylation (57), 14 CpGs were identified as differentially methylated. Of these, 11 were included in our analysis and we examined the coefficients and p-values for each of these CpGs in association with PM_2.5 in different periods of pregnancy. Three of the 11 CpGs were nominally associated (p<0.05) with PM_2.5 exposure in our data: one was associated with PM_2.5 exposure during trimester 2 only (cg05479174) and the other two (cg24709511, cg16811875) were associated with PM_2.5 exposure throughout the full pregnancy (Table 3). However, in each of these cases, the direction of association in our study was opposite to that reported in the previous meta-analysis.

Table 3.

Associations^a with PM_2.5 by pregnancy period for CpGs reported as significant in previous meta-analysis by Gruzieva et al.^b

		Trimester 1 (n=389)		Trimester 2 (n=400)		Trimester 3 (n=421)		Full pregnancy (n=429)
CpG	Coefficient from Gruzieva et al. (2019)	Coefficient	p-value	Coefficient	p-value	Coefficient	p-value	Coefficient	p-value
cg24709511	−0.001	0.0142	0.17	0.0171	0.07	0.0083	0.31	0.0349	0.02
cg02236896	−0.003	0.0012	0.92	−0.0100	0.37	0.0051	0.61	−0.0028	0.87
cg16253537	0.001	0.0038	0.89	−0.0188	0.44	0.0089	0.68	0.0160	0.68
cg06846669	−0.002	0.0149	0.34	−0.0110	0.42	−0.0134	0.27	−0.0171	0.43
cg16811875	−0.003	0.0109	0.39	0.0221	0.05	0.0191	0.06	0.0358	0.04
cg22038738	−0.004	−0.0036	0.91	0.0089	0.76	−0.0129	0.61	0.0024	0.96
cg12044654	−0.001	−0.0100	0.19	0.0030	0.64	0.0003	0.96	−0.0064	0.54
cg05479174	−0.001	−0.0067	0.60	0.0275	0.01	0.0016	0.87	0.0253	0.15
cg19120073	−0.001	−0.0045	0.62	−0.0045	0.58	−0.0018	0.80	−0.0055	0.66
cg11886880	−0.001	0.0305	0.20	0.0214	0.33	0.0131	0.49	0.0522	0.12
cg23270359	−0.001	−0.0137	0.40	−0.0106	0.45	−0.0036	0.78	−0.0232	0.29

^aCoefficients from linear regression models, adjusted for average temperature during the exposure period of interest, infant sex, maternal education, maternal race/ethnicity, season of birth, infant gestational age at birth, technical batch, estimated proportions of seven cell types (B cells, CD4+ T cells, CD8+ T cells, granulocytes, monocytes, NK cells, and nucleated red blood cells), and five surrogate variables.

^bGruzieva et al. (2019) Environmental Health Perspectives 127(5):057012.

For O₃ exposure, there were 12 DMRs associated with trimester 1 exposure, 12 associated with trimester 2, 9 associated with trimester 3, and 8 regions associated with full-pregnancy exposure (Table 4). Genes annotated to O₃-associated DMRs included some previously linked to lipid metabolism and atherosclerosis (ALOX12 (89), PON3 (90)), inflammation and immune response (MACROD1 (91), NCR1, RNF39, RAET1L), as well as neurologic disease (MCCC1 (92), DLGAP2 (82)) and one previously linked to preterm birth (VTRNA2-1 (93)). Some genes were linked to more than one trimester of exposure for O₃ (FBXO27, RUFY1). Differentially methylated regions for O₃ from models without cell type adjustment are presented in Supplemental Tables S26–S29. There were two genes in common between the PM_2.5-associated DMRs and the O₃-associated DMRs: DLGAP2, which plays a role in neuronal cell signaling and has been linked to cognitive decline (82) and schizophrenia (83), and OR10C1, which codes for an olfactory receptor protein.

Table 4.

Differentially methylated regions (false discovery rate [FDR] <0.05) in cord blood associated with exposure to O₃ during pregnancy among 429 eligible mother-infant pairs in the Healthy Start study (2010–2014).

Chromosome	Start position	End position	Raw p-value	Šidák-adjusted p-value	Annotated Gene(s)	Top CpG (lowest p-value) from region	Number of probes in region	Proportion of probes with positive association	Pregnancy period
5	178986131	178986957	3.16E-11	7.57E-07	RUFY1	cg00080972	9	0	Trimester 1
17	6899085	6899628	1.63E-10	5.94E-06	ALOX12-AS1; ALOX12	cg03407747	11	1	Trimester 1
6	30038859	30039651	5.28E-10	1.32E-05	RNF39	cg03293507	35	0.03	Trimester 1
6	29404667	29404784	1.33E-10	2.24E-05	OR10C1	cg02270518	4	1	Trimester 1
16	2848409	2848881	1.32E-09	5.53E-05	PRSS41	cg27561907	4	1	Trimester 1
11	368351	368763	4.61E-09	0.0002213	B4GALNT4	cg15971656	12	1	Trimester 1
4	1523074	1523377	1.11E-08	0.0007214	FAM53A	cg04300684	3	1	Trimester 1
2	241075915	241076652	3.01E-08	0.0008075	COPS9	cg22062537	21	0.90	Trimester 1
8	119086580	119086813	1.44E-08	0.001225	EXT1	cg20547777	3	0	Trimester 1
19	39523158	39523677	5.84E-08	0.002224	FBXO27	cg02197293	5	0	Trimester 1
11	1036677	1036989	6.30E-08	0.003984	MUC6	cg22685816	7	1	Trimester 1
7	95025829	95026299	4.78E-07	0.01993	PON3	cg04080282	17	0	Trimester 1
8	143859669	143860041	1.77E-11	9.42E-07	LYNX1-SLURP2; LYNX1	cg08090164	7	0	Trimester 2
5	178986131	178986957	1.34E-09	3.21E-05	RUFY1	cg22764044	9	0	Trimester 2
1	2885085	2885295	9.66E-09	0.0009094	ACTRT2	cg27537199	4	0	Trimester 2
19	55417361	55417698	2.38E-08	0.001396	NCR1	cg12952132	6	1	Trimester 2
1	1003126	1003580	6.82E-08	0.002966	LOC105378948	cg08709360	4	0	Trimester 2
19	39523158	39523487	2.45E-07	0.01464	FBXO27	cg08404221	4	0	Trimester 2
6	150346721	150347104	3.13E-07	0.01602	RAET1L	cg20226154	10	0	Trimester 2
11	34460107	34460608	4.22E-07	0.0165	CAT	cg01847719	10	0	Trimester 2
5	135416331	135416664	4.01E-07	0.02353	VTRNA2–1	cg18797653	9	0	Trimester 2
13	113540400	113540682	4.38E-07	0.03025	ATP11A	cg11520003	5	0	Trimester 2
11	63775071	63775487	8.37E-07	0.03903	MACROD1	cg14085017	4	0	Trimester 2
10	48416781	48417028	5.82E-07	0.04555	GDF2	cg06840491	7	1	Trimester 2
3	182817190	182817677	3.99E-23	1.62E-18	MCCC1	cg03344955	11	0	Trimester 3
5	102898223	102898784	6.32E-13	2.23E-08	NUDT12	cg02976617	6	0	Trimester 3
8	689326	690110	1.81E-09	4.56E-05	LOC401442; DLGAP2	cg05022059	5	0	Trimester 3
5	178986131	178986957	3.80E-09	9.10E-05	RUFY1	cg19626725	9	0	Trimester 3
1	1003126	1003580	2.74E-08	0.001192	LOC105378948	cg08709360	4	0	Trimester 3
5	177366867	177367064	1.92E-08	0.001921	LOC100128340	cg07372500	2	0	Trimester 3
20	47013687	47013892	8.77E-08	0.008422	LINC00494	cg25181043	3	0	Trimester 3
8	23162162	23162434	1.75E-07	0.01262	LOXL2	cg07396746	3	0	Trimester 3
13	36871754	36872217	1.13E-06	0.04724	CCDC169; CCDC169-SOHLH2	cg11678027	11	0.09	Trimester 3
5	178986131	178986957	3.24E-13	7.76E-09	RUFY1	cg00080972	9	0	Full pregnancy
17	6899085	6899628	1.03E-09	3.74E-05	ALOX12-AS1; ALOX12	cg03407747	11	1	Full pregnancy
19	39523158	39523677	1.79E-08	0.00068	FBXO27	cg02197293	5	0	Full pregnancy
6	29404667	29404784	7.62E-09	0.001287	OR10C1	cg02270518	4	1	Full pregnancy
2	241076131	241076652	6.92E-08	0.002623	COPS9	cg20221591	15	1	Full pregnancy
4	1523074	1523377	1.22E-07	0.007932	FAM53A	cg04300684	3	1	Full pregnancy
1	2885085	2885295	2.11E-07	0.0197	ACTRT2	cg11737757	4	0	Full pregnancy
8	143859734	143860041	6.56E-07	0.04139	LYNX1-SLURP2; LYNX1	cg10555383	5	0	Full pregnancy

Mediation by DNA methylation of associations between O₃ and 5-month infant adiposity

Of the nine DMRs associated with 3^rd trimester O₃, the top CpGs (by p-value) from six of these DMRs were identified as potential candidates for mediation analysis, based on a consistent direction of association with both the exposure (3^rd trimester O₃) and the outcome (5-month adiposity). The estimated percentage of the total effect mediated by changes in methylation ranged from approximately 0 to 8%, although confidence intervals were wide and all included the null (Table 5).

Table 5.

Mediation analysis^a for the top CpGs from DMRs significantly associated with 3^rd trimester O₃, and with consistent direction of association with the outcome of adiposity at 5 months of age among 294 infants.

CpG	Annotated gene(s)	Direction of association with 3rd trimester O3	Direction of association with 5-month adiposity	Indirect effect (95% CI)b	Direct effect (95% CI)b	Proportion mediated (95% CI)b
cg03344955	MCCC1	negative	negative	19.95 (−43.94, 89.47)	197.32 (−28.09, 421.17)	0.08 (−0.46, 0.89)
cg19626725	RUFY1	negative	negative	−4.2 (−47.76, 37.70)	223.38 (4.50, 437.28)	−0.01 (−0.43, 0.35)
cg07372500	LOC100128340	negative	negative	−1.90 (−48.47, 45.10)	221.39 (−0.61, 438.99)	−0.01 (−0.45, 0.38)
cg25181043	LINC00494	negative	negative	6.31 (−34.21, 49.29)	210.86 (−10.89, 427.14)	0.02 (−0.33, 0.45)
cg07396746	LOXL2	negative	negative	8.31 (−11.90, 40.50)	210.72 (−3.35, 427.21)	0.02 (−0.10, 0.32)
cg11678027	CCDC169; CCDC169-SOHLH2	negative	negative	4.43 (−22.90, 36.45)	214.81 (1.35, 428.05)	0.01 (−0.21, 0.31)

^aConducted using the R package mediation, as described in Tingley et al. (2014).

^bEstimate and 95% quasi-Bayesian confidence interval, from models adjusted for average temperature during the exposure period of interest, infant sex, maternal education, maternal race/ethnicity, season of birth, infant gestational age at birth, technical batch, and estimated proportions of seven cell types (B cells, CD4+ T cells, CD8+ T cells, granulocytes, monocytes, NK cells, and nucleated red blood cells).

Discussion

In a population of non-smoking women from a Colorado-based cohort study, we conducted an epigenome-wide analysis of associations between pregnancy-average and trimester-specific ambient concentrations of O₃ and PM_2.5, and DNA methylation in umbilical cord blood. We did not identify any single differentially methylated CpGs with epigenome-wide significance; however, we did identify several differentially methylated regions. Many of these regions were annotated to genes with roles in disease processes previously linked to air pollution exposure, including cardiometabolic disease, inflammation and immune function, and neurologic disorders.

The greatest number of DMRs in cord blood were associated with maternal exposure to PM_2.5 in the third trimester of pregnancy. Genes annotated to PM_2.5-associated regions included the following with previously reported cardiometabolic activities: PON1, involved in metabolism of oxidized lipids (94) and implicated in coronary artery disease (75);; KLB, associated with obesity and fatty liver disease (77); HOOK2, previously linked to type 2 diabetes (78); and TYW3, linked to circulating resistin levels (79). Other notable genes include SLC6A3 (also called DAT1), a dopamine transporter implicated in several neuropsychiatric disorders including schizophrenia (95), Parkinson’s disease (96) and attention-deficit hyperactivity disorder (97); DLGAP2, associated with cognitive decline (82) and schizophrenia (83); and CD6, linked to multiple sclerosis (98).

Our results were not consistent with those of a previous meta-analysis that examined associations between full-pregnancy PM_2.5 and epigenome-wide DNA methylation in umbilical cord blood (57). The Healthy Start cohort data was not included in the previous meta-analysis. Of the 11 differentially methylated CpGs reported in that study and also analyzed here, two had nominal significance (p<0.05). However, in all of these cases, the direction of association was opposite to that reported previously. Possible reasons why our results differ from that of the meta-analysis, which included 1,551 mother-infant pairs from nine cohorts in the USA and Europe, may include: geographic differences in composition of PM_2.5; additional confounder adjustment in our study including maternal education, season, and temperature; and heterogeneity between the populations in other unmeasured characteristics. Our analysis excluded women with sustained smoking during pregnancy, while women who smoked were included in the meta-analysis. The smaller sample size of our study may have also limited our ability to detect small effect sizes, although we evaluated replication using a generous threshold of p<0.05 without adjustment for multiple testing.

Compared to the PM_2.5 results, there were fewer genes with reported cardiometabolic functions annotated to O₃-associated DMRs; notable exceptions were PON3, involved in oxidation of LDL cholesterol and linked to coronary artery disease (90), and ALOX12, involved in lipid metabolism and platelet function and previously identified as an epigenetic marker of atherosclerosis (89). DMRs associated with O₃ were annotated to some genes involved in inflammatory and immune functions including MACROD1, reported to enhance inflammation in adipocytes (91), and RNF39, at which methylation has been previously linked to multiple sclerosis (99). Some O₃-associated DMRs were annotated to genes linked to neurologic disorders, including MCCC1, associated with susceptibility to Parkinson’s disease (92), and DLGAP2, involved in neuronal signaling and linked with cognitive decline (82). There were also suggestive associations with respiratory diseases among the genes annotated to ozone-associated DMRs. A previous study reported that methylation of nasal epithelial cells at a CpG in ATP11A, associated with 2^nd trimester O₃ in our study, predicted lung function and BMI in cystic fibrosis patients (100). The LYNX1 gene, also associated with 2^nd trimester O₃ exposure, is involved in regulation of nicotinic cholinergic signaling in the lung and has been proposed as a target for treatment of asthma or chronic obstructive pulmonary disease (101).

Our epigenome-wide results share some common findings with previous studies. One previous study including 175 cord blood samples from the Early Autism Risk Longitudinal Investigation (EARLI) study identified differentially methylated regions (DMRs) in cord blood cells associated with prenatal O₃ and NO₂ exposures (29). Of note, one of the four DMRs associated with prenatal O₃ in that study was located at the RNF39 gene and this gene was also annotated to one of the DMRs associated with first trimester O₃ exposure in our study. RNF39 is located in the major histocompatibility complex (MHC) I region and methylation at this locus has been previously linked with immunological outcomes including multiple sclerosis (99) and cutaneous manifestations of systemic lupus erythematosus (102).

We additionally examined the potential for differentially methylated CpGs in cord blood to act as mediators of the effect of prenatal air pollution exposure on infant and child health. In a previous publication using data from the Healthy Start cohort, we reported a positive association between 3^rd trimester average 8-hour maximum O₃ concentrations, and infant adiposity at a 5-month follow-up visit (59). There was no association in our previous study between average PM_2.5 concentrations during pregnancy and infant adiposity. We therefore examined the potential for top CpGs from differentially methylated regions identified in this study to mediate the effect of prenatal O₃ on 5-month infant adiposity. The top CpGs from six DMRs were identified as potential candidate mediators, and the percent of the total effect mediated ranged from 0–8%. However, the 95% quasi-Bayesian confidence intervals included the null. The CpG estimated to mediate 8% of the effect of third trimester O₃ on 5-month adiposity (cg03344955) was annotated to the gene MCCC1, variants of which have been associated with Parkinson’s disease risk (92, 103).

Previous literature suggests multiple pathways by which maternal air pollution exposure may influence offspring health outcomes, and particularly those related to obesity and cardiometabolic disease. Exposures to ozone and particulate matter during pregnancy have been associated with impaired maternal glucose metabolism and gestational diabetes risk (104, 105); however, this is not consistent across studies (106, 107). A large study in southern California found positive associations of O₃ and PM_2.5 with gestational diabetes risk in single-pollutant models, but observed that the associations were attenuated after adjusting for other pollutants, suggesting that a different component of the air pollution mixture (such as NO₂) may be responsible for observed associations (108). Offspring exposed in utero to maternal diabetes and hyperglycemia experience greater risk of obesity and metabolic disease later in life (109–112).

Other maternal cardiometabolic effects associated with ambient concentrations of O₃ and PM_2.5 during pregnancy include higher blood pressure (113) and risk of gestational hypertension (114–116), which could in turn influence fetal growth and epigenetic marks in cord blood. In a European study of 361 pregnant women, 2^nd trimester PM_2.5 exposure was associated with hypomethylation of the LEP gene promoter in placental tissue (36). The LEP gene codes for the protein leptin, which is an important regulator of energy intake and expenditure. Leptin dysregulation or resistance is associated with obesity (117). Finally, multiple studies of non-pregnant individuals have reported positive associations between air pollution exposure and blood lipids (118), and these effects may be more pronounced among participants with obesity (119, 120).

Extensive previous literature has examined epigenetic markers of childhood obesity, growth, and adiposity. A systematic review conducted in 2021 identified >100 studies reporting on DNA methylation in association with childhood obesity (121). Among 23 epigenome-wide association studies included in the review (most using child peripheral blood rather than cord blood), only a small number of genes were identified in more than one study as differentially methylated in association with obesity (HDAC4, PRLHR, TNXB, and PRDM16); none of these were annotated to differentially methylated regions associated with air pollutants in our study.

Our study has some limitations. Ambient air pollution exposure during pregnancy was estimated using inverse-distance-weighted averages of concentrations from stationary monitors, which may not accurately reflect concentrations inside the maternal residence. The available network of stationary monitors is not sufficiently dense to capture PM_2.5 variation on a fine scale, such as residences with locally high exposure due to a nearby high-traffic road (122). Furthermore, we were unable to estimate pregnant women’s exposure to air pollution at locations other than the residential address, such as employment, commuting, or recreation sites.

DNA methylation was evaluated in umbilical cord blood, which includes a combination of cell types. Although we adjusted for seven cell types, methylation profiles may differ in other relevant offspring cell types including those in adipose tissue. Estimated proportions of cell types were not associated with the exposure in our data, however we included cell type adjustment in our main results to minimize genomic inflation and for comparability with other studies. Umbilical cord blood was collected from a convenience sample of participants in the Healthy Start cohort, but the characteristics of this sample were similar to those of the larger study population.

Additionally, this study was restricted to women who reported no sustained smoking during pregnancy, representing >90% of the Healthy Start cohort, but this exclusion may limit the generalizability of the study. Although the sample size was fairly large compared to many previous studies, statistical power to detect epigenome-wide associations after correction for multiple comparisons was limited. Finally, we adjusted for several hypothesized confounding variables, but we cannot rule out residual confounding of the results by unmeasured variables. This is particularly important for the mediation analysis, which relies on a number of strong assumptions for causal interpretation, including no unmeasured confounders of the exposure-mediator relationship or the mediator-outcome relationship (123). The degree of bias introduced by unmeasured confounders depends on the plausible magnitude of associations between the confounder and exposure, mediator, and outcome variables (123).

Notable contributions of this study include the examination of both PM_2.5 and O₃ in each trimester of pregnancy in relation to cord blood DNA methylation, in a relatively understudied geographic location with low PM_2.5 but O₃ levels that regularly exceeds federal air quality standards. We compared our epigenome-wide results to those of previous studies to attempt replication, and we examined the potential for locus-specific methylation changes to mediate previously observed associations between prenatal exposure to O₃ and infant adiposity.

Conclusions

We identified several differentially methylated regions in cord blood associated with trimester-specific concentrations of ambient PM_2.5 and O₃ among mother-infant pairs in a Denver, Colorado-based cohort. Genes annotated to these regions are involved in health outcomes plausibly linked to air pollution exposure, including cardiometabolic disease, inflammation and neurologic disorders. We found limited evidence that differential methylation could partially mediate the previously observed association between 3^rd trimester O₃ concentrations and infant adiposity at 5 months of age (59); all confidence intervals included the null. Future studies will examine whether methylation differences detected at birth are associated with offspring health outcomes later in childhood and early adulthood, and therefore may serve as useful markers of adverse environmental programming received during gestation.

Highlights.

• Air pollution during pregnancy was associated with DNA methylation in cord blood.

• Associated genes were linked to cardiometabolic, immune, and neurologic diseases.

• Limited support for mediation by methylation of associations with infant adiposity.