Particulate matter, the newborn methylome, and cardio-respiratory health outcomes in childhood.

Ambient air pollution is associated with adverse health outcomes including cardio-respiratory diseases. Epigenetic mechanisms such as DNA methylation may play a role in driving such associations. We investigated the effects of prenatal particulate matter (PM) exposure on DNA methylation of 178,309 promoter regions in 240 newborns using the Infinium HumanMethylation450 BeadChip, using a generalized linear regression model with a quasi-binomial link family, adjusted for gender, plate, and cell types. PM-associated CpG loci were then investigated for their associations with childhood asthma, carotid intima-media thickness (CIMT), and blood pressure (BP) using logistic or linear regression. Thirty-one loci were associated with either PM10 or PM2.5 using FDR-corrected p-values of less than 0.15. Two loci were evaluated for replication in a separate population of 280 Children's Health Study (CHS) subjects using Pyrosequencing, of which one successfully replicated (COLEC11 cg03579365). Three of the 31 loci were also associated with physician-diagnosed asthma at 6 years old, two were associated with CIMT and one with systolic BP at 10 years old. A higher methylation level in TM9SF2 (cg02015529) and UBE2S (cg00035623), respectively, was associated with a 2SD increase in prenatal PM and was also associated with 36% and 98% increased odds of asthma; whereas methylation of TDRD6 (cg22329831) was negatively associated with PM and a 24% decreased odds of asthma. Prenatal PM exposure was associated with altered DNA methylation in newborn blood in a small number of gene promoters, some of which were also associated with cardio-respiratory health outcomes later in childhood. Keywords: methylation, particulate matter, air pollution, asthma, cardiovascular.

Introduction

Ambient air pollution is associated with numerous adverse health outcomes, the most notable of which are allergic, respiratory, and cardiovascular diseases [1-3]. Long-term exposures have been associated with measures of atherosclerosis, including carotid intima-media thickness (CIMT) and blood pressure, both of which predict future cardiovascular events [4-8]. Regional and near-roadway pollutants have also been associated with childhood asthma, but the strength of association may depend on timing and duration of exposures [9-11]. Air pollutant exposures early in life, particularly during the prenatal period, have been associated with asthma development by age 6 years in some but not all studies [12, 13]. Prenatal air pollutant exposures have not been evaluated for their contribution to cardiovascular disease (CVD) risk, although this hypothesis is supported by the developmental origins of adult disease theory (DOHaD) and increasing data from animal models [14-22].

The biological mechanisms driving these exposure-disease associations are thought to be largely through oxidative stress, inflammation, or endothelial or autonomic dysfunction [23]. Importantly, genetic variation in genes involved in these biological pathways can alter an individual’s susceptibility to air pollution health effects [24], providing further evidence for their involvement. In recent years, the hypothesis that epigenetics might play a role in driving exposure-disease associations has gained traction, in part, because epigenetic modifications are labile and may respond to environmental exposures in ways that directly or indirectly affect gene transcription and disease risk.

Several studies have now been conducted to evaluate the effects of air pollutants on epigenetics, with a primary focus on DNA methlyation [25]. Particulate matter (PM) for example, has been associated with hypomethylation of selected tandem repeats [26] as well as changes in DNA methylation in a handful of candidate genes involved in asthma, inflammation, and oxidative stress [27-31]. These studies have been conducted in adults and largely in occupational or controlled exposure settings over relatively short exposure time windows. Moreover, few studies have taken an epigenome-wide approach in evaluating exposure effects.

Given our current understanding of the important and dynamic role that DNA methylation plays in embryogenesis [32] and the likelihood that epigenetic mechanisms play a role in the DOHaD hypothesis [19], we sought to investigate the effects of prenatal particulate matter (PM10 and PM2.5) exposure on DNA methylation profiles in newborns using the Infinium HumanMethylation450 BeadChip (HM450) in a subset of the Children’s Health Study [33, 34]. PM-associated CpG loci were also investigated for their associations with childhood cardio-respiratory health outcomes, including asthma prevalence, carotid intima-media thickness, and systolic and diastolic blood pressure.

Results

Demographic characteristics of the 240 study subjects at study entry as well as the 280 subjects in the replication population are shown in Table 1. Generally the characteristics of both populations were similar. Participants averaged 11 years of age (range 10–13 years) at the time of CIMT assessment, although there were more females (58%) than males (42%) in the study population. Sixteen (7%) were exposed to maternal smoking during pregnancy, while 29 (12%) were exposed to paternal smoking. The replication population had no subjects exposed to maternal smoking. Prevalence of asthma was 10–12%. Distributions of prenatal air pollutant exposures and cardiovascular phenotypes are shown in Figure 1 and Table 1. The median levels of PM10 were 38.9, 41.1, 39.2 µg/m3 for the first, second, and third trimesters, respectively, with an interquartile range 19.3, 20.6, and 17.8 µg/m3 whereas median levels of PM2.5 were 26.3, 26.7, and 24.0 µg/m3 with an interquartile range of 8.1, 8.2, and 5.6 µg/m3, respectively. PM10 and PM2.5 were modestly correlated within each trimester, but levels were not correlated across trimesters (r2 ranged from 0.54 to 0.69) (see online Supplementary material, Table S1).

Table 1:

demographic characteristics and cardio-respiratory phenotypes of participants

	Primary study population (N = 240)						Replication population (N = 280)
	N	%	Median	Min	Max	IQR	N	%	Median	Min	Max	IQR
Male sex	100	41.7	–	–	–	–	140	50.0	–	–	–	–
Race/Ethnicity
Hispanic White	137	57.1	–	–	–	–	151	53.9	–	–	–	–
Non-Hispanic White	73	30.4					93	33.2
Asian	15	6.3					11	3.9
Black/Other	15	6.3	–	–	–	–	25	8.9	–	–	–	–
Maternal smoking during pregnancy	16	6.7	–	–	–	–	0	0.0	–	–	–	–
Paternal smoking during pregnancy	29	12.0	–	–	–	–	28	10.0	–	–	–	–
Asthma	23	9.6	–	–	–	–	33	11.8	–	–	–	–
Mother's Education
High school or less	76	31.7	–	–	–	–	80	28.6	–	–	–	–
Some college	79	32.9	–	–	–	–	96	34.3	–	–	–	–
College grad/some grad school	69	28.8	–	–	–	–	86	30.7	–	–	–	–
Age at CIMT assessment	–	–	11.2	9.9	12.8	1.0	–	–	11.3	10.0	12.8	1.0
CIMT, µm	–	–	564.5	398.8	674.2	48.9	–	–	565.3	434.5	692.5	55.0
BMI, kg/m2	–	–	19.2	13.6	34.2	5.5	–	–	18.6	13.6	34.9	5.4
SBP, mmHg	–	–	105.0	85.0	129.0	13.0	–	–	104.0	86.0	124.0	11.3
DBP, mmHg	–	–	56.0	46.0	80.0	7.5	–	–	56.7	40.3	83.7	7.3

CIMT = carotid intima-media thickness, BMI = body mass index, SBP = systolic blood pressure, DBP = diastolic blood pressure.

Numbers do not always add up to 100% due to missing data.

Figure 1:

(a–b) distribution of cumulative air pollutant exposures across three trimesters (n = 240). (a) PM2.5 and (b) PM10.

The effects of trimester specific PM10 and PM2.5 on methylation of 178 309 promoter CpG loci were evaluated. We identified one CpG (cg22506605) associated with second trimester PM10 (FDR-corrected P-value = 0.07) and nine CpGs associated with third trimester PM10 (FDR-corrected P-value < 0.15) (Table 2). We also identified 10 CpGs associated with second trimester PM2.5 and 11 CpGs associated with third trimester PM2.5 using FDR-corrected P-values of less than 0.15 (Table 3). One CpG was found to be associated with both PM2.5 and PM10. The locus cg17486097 in the UNC5D gene was associated with higher methylation for both third trimester PM10 and PM2.5. Additional analyses evaluating potential confounding of these associations by maternal education, in utero tobacco smoke exposure and ethnicity largely did not change the results (see online Supplementary material, Tables S2 and S3).

Table 2:

Association between a 2SD increase in prenatal PM10 exposure and DNA methylation (n = 240)**.

					Trimester 1		Trimester 2			Trimester 3
Probe	Chr	Position	Gene	Mean methylation	Coeff	P-value	Coeff	P-value		Coeff	P-value		Direction of effect across trimesters
cg22506605	20	44462172	SNX21	0.06	0.06	0.15	0.20	3.7 × 10−7	*	0.04	0.32		+++
cg17486097	8	35093411	UNC5D	0.02	−0.02	0.40	0.06	3.4 × 10−3		0.11	2.7 × 10−7	*	− ++
cg22329831	6	46655820	TDRD6	0.95	0.00	0.96	−0.10	2.4 × 10−2		−0.19	1.0 × 10−6	*	+–
cg02015529	13	100153203	TM9SF2	0.04	−0.03	0.67	0.18	9.9 × 10−4		0.24	5.4 × 10−6	*	− ++
cg03579365	2	3642211	COLEC11	0.89	0.02	0.50	0.08	0.01		0.14	4.5 × 10−6	*	+++
cg04048259	20	57875346	EDN3	0.02	0.05	0.24	0.09	0.03		0.17	4.1 × 10−6	*	+++
cg07575624	1	192778623	RGS2	0.05	0.01	0.82	0.06	0.23		0.20	4.2 × 10−6	*	+++
cg13394864	2	97563733	FAM178B	0.78	0.00	0.82	0.00	0.85		−0.06	5.9 × 10−6	*	++−
cg15155738	12	121454335	C12orf43	0.01	−0.04	0.19	−0.08	4.6 × 10−3		−0.13	5.4 × 10−6	*	—
cg25430696	17	2240014	TSR1;SGSM2	0.07	−0.03	0.32	0.10	3.9 × 10−4		0.15	2.1 × 10−8	*	− ++

Indicates loci that are statistically significant after FDR correction (<0.15).

Results are from a quasi-binomial regression model adjusted for gender, plate, and cell types. 2 SD for PM10 for trimesters 1,2, and 3 are 32.4, 32.6, and 31.6.

Table 3:

Association between a 2SD increase in prenatal PM2.5 exposure and DNA methylation (n = 185)**.

					Trimester 1		Trimester 2			Trimester 3
Probe	Chr	Position	Gene	Mean methylation	Coeff	P-value	Coeff	P-value		Coeff	P-value		Direction of effect
cg01439366	20	388553	RBCK1	0.04	−0.01	0.62	0.13	5.5 × 10−6	*	0.03	0.27		− ++
cg02733795	6	33422526	ZBTB9	0.04	0.06	0.07	0.14	2.8 × 10−6	*	0.08	0.02		+++
cg04335339	10	116853118	ATRNL1	0.05	0.02	0.43	0.13	2.9 × 10−7	*	−0.04	0.18		++−
cg04784263	16	57318624	PLLP	0.02	0.07	0.06	0.17	5.1 × 10−7	*	0.03	0.36		+++
cg05233674	6	34433773	PACSIN1	0.03	0.07	0.22	0.22	5.0 × 10−6	*	0.02	0.79		+++
cg06618764	7	2719090	AMZ1	0.96	−0.06	0.72	−0.59	5.4 × 10−6	*	−0.13	0.42		—
cg10109500	3	172165884	GHSR	0.03	0.00	0.95	0.12	7.5 × 10−6	*	−0.01	0.65		0 + −
cg16522462	20	1875565	SIRPA	0.03	0.04	0.17	0.12	6.5 × 10−6	*	−0.05	0.07		++−
cg20909686	11	65554042	OVOL1	0.04	−0.01	0.88	0.20	4.5 × 10−6	*	−0.05	0.32		− +−
cg27399414	1	9884593	CLSTN1	0.02	0.07	0.10	0.18	8.4 × 10−6	*	−0.01	0.74		++−
cg00035623	19	55919836	UBE2S	0.04	−0.01	0.73	0.02	0.68		0.18	6.5 × 10−7	*	− ++
cg02656441	13	111367915	ING1	0.05	−0.10	0.12	0.09	0.17		0.27	1.3 × 10−6	*	− ++
cg07246225	2	145277659	ZEB2	0.06	−0.01	0.87	−0.04	0.61		0.30	5.0 × 10−6	*	–+
cg10217503	12	65153209	GNS	0.02	−0.06	0.08	0.01	0.70		0.17	1.7 × 10−6	*	− ++
cg11625178	12	98909434	TMPO; LOC100128191	0.04	−0.03	0.31	−0.03	0.32		0.17	1.4 × 10−8	*	–+
cg12749048	13	45011471	TSC22D1	0.05	−0.06	0.04	−0.03	0.31		0.12	7.7 × 10−6	*	–+
cg15142938	11	59383312	OSBP	0.02	−0.09	0.03	0.05	0.17		0.17	4.6 × 10−6	*	− ++
cg16918263	1	229644277	NUP133	0.03	−0.02	0.59	−0.01	0.79		0.15	3.9 × 10−6	*	–+
cg17486097	8	35093411	UNC5D	0.02	−0.05	0.07	0.00	0.92		0.12	1.6 × 10−6	*	–+
cg25976563	3	179169592	GNB4	0.04	−0.01	0.75	−0.03	0.42		0.17	4.0 × 10−6	*	–+
cg27277978	16	30457458	SEPHS2	0.09	−0.18	0.02	0.02	0.81		0.32	6.5 × 10−7	*	− ++

Indicates loci that are statistically significant after FDR correction (<0.15).

Results are from a quasi-binomial regression model adjusted for gender, plate, and cell types. 2 SD for PM2.5 for trimesters 1, 2, and 3 are 13.8, 14.2, and 11.2.

We chose two loci with the largest magnitude of effects from each analysis for further replication: cg03579365 in the COLEC11 gene from the PM10 analysis and cg27277978 in the SEPHS2 gene from the PM2.5 analysis (Figure 2). Replication was conducted in newborn bloodspots from an additional population of 280 subjects selected from the Children’s Health Study [33, 34] using Pyrosequencing. Replication for cg03579365 was successful (Table 4), further supporting an association between methylation in this locus and PM10; however, the results were strongest for first trimester rather than third trimester exposures. Replication was not successful for cg27277978.

Figure 2:

Scatterplot of 3rd trimester PM10 exposure and methylation values in cg27277978 (SEPHS2) and cg03579365 (COLEC11).

Table 4:

Effects of per 2SD increase in prenatal PM10 or PM2.5 exposure on DNA methylation in a replication population*.

					Trimester 1			Trimester 2			Trimester 3
Pollutant	Probe	Gene	Mean methylation	N	β	P-value	% change in methylation	β	P-value	% change in methylation	β	P-value	% change in methylation
PM10	cg03579365	COLEC11	0.91	279**	0.08	0.02	0.62%	−0.008	0.81	−0.06%	−0.05	0.14	−0.42%
PM2.5	cg27277978	SEPHS2	0.02	149***	0.14	0.41	0.20%	−0.02	0.90	−0.06%	−0.25	0.20	−0.35%

Results are from a quasi-binomial regression model adjusted for gender, plate, ethnicity and maternal education.

One methylation outlier was excluded from analyses.

Sample size is reduced due to missing exposure information for PM2.5.

Finally, we related methylation level at birth in the list of PM-associated CpG loci with cardiovascular and respiratory health outcomes in early childhood. Of the 31 loci tested, three were associated with physician-diagnosed asthma at 6 years of age, two were associated with CIMT and one with systolic BP z-score at 11 years of age (Table 5). A higher methylation level in the promoters of TM9SF2 (cg02015529) and UBE2S (cg00035623), respectively, was associated with a 2SD increase in prenatal PM and was also associated with 36 and 98% increased odds of asthma at 6 years of age; whereas methylation of TDRD6 (cg22329831) was negatively associated with PM and a 24% decreased odds of asthma.

Table 5:

Association between DNA methylation at birth and cardio-respiratory health outcomes in childhood*.

				CIMT***		SBP Z-score***		DBP Z-score***		Asthma**
Probe	Chr	Position	Gene	β	P-value	β	P-value	β	P-value	OR	P-value
cg00035623	19	55919836	UBE2S	6.02	0.05	0.01	0.90	−0.06	0.12	1.98	0.004
cg01439366	20	388553	RBCK1	2.04	0.62	−0.04	0.59	0.06	0.26	1.13	0.74
cg02015529	13	100153203	TM9SF2	−1.18	0.44	−0.05	0.08	0.00	0.98	1.36	0.005
cg02656441	13	111367915	ING1	−0.03	0.98	−0.02	0.47	−0.01	0.74	1.07	0.64
cg02733795	6	33422526	ZBTB9	4.37	0.11	−0.03	0.50	−0.03	0.36	1.17	0.49
cg03579365	2	3642211	COLEC11	1.59	0.22	0.00	0.95	−0.02	0.25	1.18	0.13
cg04048259	20	57875346	EDN3	−8.42	0.02	−0.03	0.69	−0.04	0.37	0.65	0.31
cg04335339	10	116853118	ATRNL1	−0.86	0.78	0.03	0.63	0.06	0.16	1.00	0.99
cg04784263	16	57318624	PLLP	−2.71	0.62	0.00	0.99	0.05	0.51	1.16	0.75
cg05233674	6	34433773	PACSIN1	0.28	0.90	−0.02	0.58	−0.01	0.63	1.28	0.14
cg06618764	7	2719090	AMZ1	−0.15	0.87	0.00	0.84	0.02	0.12	1.14	0.29
cg07246225	2	145277659	ZEB2	0.71	0.44	0.00	0.89	0.00	0.83	0.91	0.37
cg07575624	1	192778623	RGS2	−0.59	0.76	0.02	0.54	−0.01	0.71	0.81	0.30
cg10109500	3	172165884	GHSR	−7.73	0.19	0.10	0.35	0.10	0.19	0.97	0.96
cg10217503	12	65153209	GNS	0.88	0.86	−0.14	0.12	−0.10	0.11	1.15	0.73
cg11625178	12	98909434	TMPO;LOC100128191	−3.11	0.35	−0.04	0.49	0.00	0.96	1.40	0.21
cg12749048	13	45011471	TSC22D1	−2.89	0.37	−0.02	0.67	0.03	0.47	1.13	0.69
cg13394864	2	97563733	FAM178B	1.15	0.43	−0.01	0.75	0.01	0.67	0.99	0.92
cg15142938	11	59383312	OSBP	−0.71	0.88	0.13	0.12	0.11	0.08	1.76	0.12
cg15155738	12	121454335	C12orf43	14.89	0.12	0.04	0.80	0.06	0.66	0.62	0.63
cg16522462	20	1875565	SIRPA	3.89	0.38	0.02	0.85	0.04	0.44	1.10	0.82
cg16918263	1	229644277	NUP133	0.06	0.99	0.03	0.73	−0.01	0.88	1.13	0.72
cg17486097	8	35093411	UNC5D	−12.91	0.06	−0.02	0.84	−0.03	0.73	0.51	0.39
cg20909686	11	65554042	OVOL1	1.79	0.32	0.04	0.24	0.03	0.22	1.00	0.99
cg22329831	6	46655820	TDRD6	1.12	0.53	0.00	0.89	0.02	0.40	0.76	0.04
cg22506605	20	44462172	SNX21	−0.17	0.90	−0.06	0.01	−0.01	0.68	1.06	0.64
cg25430696	17	2240014	TSR1;SGSM2	−1.01	0.64	0.00	0.94	−0.03	0.22	1.04	0.83
cg25976563	3	179169592	GNB4	−2.42	0.44	−0.02	0.66	−0.01	0.85	1.28	0.29
cg27277978	16	30457458	SEPHS2	0.45	0.51	0.01	0.61	0.00	0.65	0.99	0.91
cg27399414	1	9884593	CLSTN1	−3.53	0.40	0.06	0.44	0.04	0.41	0.70	0.44

Models adjusted for maternal education, sex, ethnicity, plate, age at clinic visit, in utero tobacco smoke exposure, and BMI Z-score.

Logistic regression model evaluating odds ratio for asthma risk per 1% increase in DNA methylation.

Linear regression model evaluating the change in CIMT (µm), SBP z-score or DBP z-score per 1% increase in DNA methylation. Bold values are statistically significant.

Discussion

Evidence demonstrating associations between air pollution and DNA methylation is sparse and few, if any, studies have evaluated epigenome-wide DNA methylation at birth in association with prenatal exposures. The epidemiologic studies that do exist have been conducted in adults and largely in occupational or controlled exposure settings over relatively short exposure time windows. In these studies, particulate matter has been associated with hypomethylation of selected tandem repeats [26] as well as changes in DNA methylation in a handful of candidate genes involved in asthma, inflammation, and oxidative stress [27–31, 35–37]. In vitro studies in murine macrophages have also provided evidence that PM10 alters methylation machinery, specifically by decreasing expression levels of DNMTs [38]. In one of the only studies to evaluate prenatal PM exposure, Janssen et al. found that prenatal PM2.5 was associated with mitochondrial DNA methylation in placental tissue [39, 40].

In this study, we found that prenatal exposure to PM10 and PM2.5 was associated with altered DNA methylation in newborn blood in a small number of gene promoters, some of which were also associated with cardio-respiratory health outcomes later in childhood. The loci associated with either PM10 or PM2.5were largely independent, with only one locus associated with both exposures. The association between PM exposure and one locus in the COLEC11 gene promoter was replicated in an independent population of subjects using a second laboratory technique. No loci were associated with first trimester exposures.

Only one of the two loci selected for replication was successful. Prenatal PM10 exposure was associated with a 1.3 higher DNA methylation level at cg03579365 in COLEC11 in the analysis using the HM450 platform, whereas the magnitude of the association was smaller using Pyrosequencing and was with 1st trimester PM10 exposure instead. Thus while a consistent direction of effect was observed with prenatal PM10 exposure, timing of exposure and sensitivity of exposure time windows, remains in question. COLEC11 is a gene important in embryonic development, as mutations in COLEC11 are one of the causes of the developmental defect syndrome 3MC, which has a wide spectrum of developmental features including facial dysmorphism, cognitive impairment, hearing loss, and vesicorenal anomalies [41, 42]. COLEC11encodes the serum protein collection kidney 1 (CL-K1) comprising part of the collectin family of innate immune proteins. The lectin pathway is one of three pathways by which the complement system can be activated and thus CL-K1 plays an important role in the immediate response against microorganisms [43]. Interestingly, in a study by Kolarova et al. that evaluated DNA methylation and intellectual disability, hypermethylation of COLEC11 was found to be associated with developmental delay [44]. Herein we report that higher methylation in cg03579365 was associated with higher first and third trimesters of PM10 exposure.

DNA methylation in the promoters of three genes (TM9F2, UBE2S, and TDRD6) at birth was associated with physician-diagnosed asthma at 6 years of age. TM9F2 is a member of the evolutionarily conserved nonaspanin family, whose function as a whole is largely unknown though some evidence suggests they play a role in cellular immunity and response to bacteria [45]. UBE2S, ubiquitin-conjugating enzyme E2S, plays an important role in the pluripotent state of embryonic stem cells. It also can degrade SOX2 [46], a transcription factor that regulates embryonic development, cell fate, and neuronal development [47] and a critical role in branching morphogenesis and epithelial cell differentiation of the lung [48, 49]. Specifically, continuous expression of SOX2 inhibits the branching process resulting in a severe reduction of the number of airways [48]. Thus, it is biologically plausible that the associations we observed between higher PM levels, higher DNA methylation and UBE2S and increased asthma risk are mediated through effects on SOX2 expression. TRDR6 on the other hand, is essential for spermiogenesis and for proper miRNA expression [50]. Deficiencies in TRDR6, a member of the tudor domain containing proteins, which constitutes a conserved class of chromatoid body components causes male sterility and is essential for haploid spermatid development [51]. Exposure to titanium dioxide nanoparticles in the testes has been shown to affect sperm formation and to alter TDRD6 expression [52]. In one study, TDRD6 was implicated in atherosclerosis [53]. However, none of these genes have been previously implicated in asthma pathogenesis.

One of the great strengths of this study is the temporal separation of PM exposure assessment, DNA methylation measurement, and childhood health outcomes, enabling the study to truly address a DOHaD hypothesis. Several limitations should also be noted. We chose to focus the analysis on loci in promoter regions, and therefore, we may have missed interesting associations between exposure and methylation in loci outside these regions. We processed HM450 data using methylumi [54] however, use of other processing and data cleaning techniques may produce different results. Although we made every effort to control for potential confounders, we cannot exclude the possibility of residual confounding by some unknown factor that is associated with DNA methylation levels, ambient air pollution, and cardio-respiratory phenotypes. We measured DNA methylation in newborn blood. Although we adjusted for six common cell types using the Houseman method [55], the small changes in methylation observed may still be the result of shifts in cell populations of smaller subtypes [56]. Moreover, the Houseman method is meant for adult blood, and thus cannot address nucleated red blood cells or immature forms present in newborn blood. Only one of two loci evaluated for replication was successful. There may be several reasons for why one of the loci failed to replicate, including the original result was a false positive, differences in the laboratory assay sensitivities and/or measurement error, the fact that we could not adjust for cell fraction in the replication population, and statistical differences between the original and replication analyses leading to differential results. Misclassification of exposure is another limitation since air pollution exposure based on residential address does not capture individual behaviors. Estimation of trimester exposures based on birth certificate reporting of gestational age is prone to error [57]. Despite this inherent measurement error, we observed multiple trimester-specific effects of exposures and future studies specifically designed to capture weekly exposures in pregnancy will help to further narrow the relevant biological window of susceptibility.

Materials and Methods

Study P opulation

This study was nested in the Children’s Health Study, a longitudinal cohort study of respiratory health [58]. A subset of 737 children was recruited to participate in a sub-study of air pollution and atherosclerosis. Within this sub-study, 273 children had a newborn bloodspot in which DNA methylation was assessed using the Infinium HumanMethylation450 BeadChip (HM450). An additional subset of 280 different children with newborn bloodspots was chosen for a replication population in which DNA methylation was assessed using Pyrosequencing. All subjects had systolic/diastolic blood pressure, supine heart rate, standing height, and weight measured during a classroom visit, and B-mode carotid artery ultrasound was performed. Personal, parental, and socio-demographic characteristics, including maternal smoking during pregnancy, were obtained by parent-completed questionnaire. Children were classified as having asthma if the adult completing the questionnaire reported that a doctor had “ever diagnosed the child as having asthma.” Participants’ parents were asked about previous occurrences of stroke, heart failure, and heart attack or angina. Affirmative responses were coded as having a prior family history of heart disease if an individual responded yes to any of these inquiries. DNA methylation was measured in newborn bloodspots that were obtained from the California Department of Public Health Genetic Disease Screening Program. Birth weight, gestational age, mode of delivery, and other reproductive data were obtained from California birth records. The estimated date of conception was assigned using the birth date and gestational age, corrected for the average 2-week difference between the last menstrual period and conception.

Air P ollution A ssessment

The CHS air quality monitoring data [33, 34, 59] and the US EPA air Quality System (AQS) were used to assign estimates of prenatal air pollution exposures for PM2.5 and PM10, based on residential address reported on the birth certificate and at the time of the baseline questionnaire. Addresses were geocoded using TeleAtlas Inc.’s Address Point Geocoding Services. Station-specific air quality data were spatially interpolated to each birth residence using inverse-distance-squared weighting. The data from up to four air quality measurement stations were included in each interpolation. Due to the regional nature of PM10 and PM2.5 concentrations, a maximum interpolation radius of 50 km was used for all pollutants. However, when a residence was located within 5 km of one or more stations with valid observations, the interpolation was based solely on the nearby values. When multiple addresses were reported, the average concentrations were time-weighted to account for portion of year spent at each address. Individuals with incomplete residential histories were excluded from analyses. Prenatal air pollution assignments were successfully made for 241 of the 273 primary participants, with the exception of PM2.5, for which we had only 186 subjects with assigned exposure due to lack of monitoring data in some communities. Subjects with and without exposure data were largely similar except for differences in prevalence of in utero smoke exposure and asthma (see online Supplementary material, Table S4). In the replication population, 280 subjects had PM10 and 149 had PM2.5 measurements available.

The study protocol was approved by the University of Southern California Institutional Review Board and informed, written consent and assent were provided by the parents and children respectively.

Health Measurements

CIMT, heart rate, and blood pressure were assessed by a single physician-imaging specialist from the USC Atherosclerosis Research Unit (ARU) Core Imaging and Reading Center (CIRC). As described previously (Patents 2005, 2006) [60-63], the jugular vein and carotid artery were imaged transversely with the jugular vein stacked above the carotid artery. All images contained internal anatomical landmarks for reproducing probe angulation and a single-lead electrocardiogram was recorded simultaneously with the B-mode image to ensure that CIMT was measured at the R-wave in the cardiac cycle. CIMT was measured along the far (deep) wall of the distal common carotid artery (0.25 cm from the carotid artery bulb) along a standard 1 cm length that was automatically determined by a computer-generated ruler. This method standardizes the timing, location, and distance over which CIMT is measured, ensuring comparability across participants [60-63]. Duplicate scans were conducted 2.5 days apart on average for CIMT (n = 44) and the intra-class correlation between replicate scans was 0.84.

DNA M ethylation

DNA methylation was measured in archived newborn bloodspots as the most proximal biomarker of DNA methylation reflecting the fetal experience. Newborn bloodspots were stored by the state of California at −20°C. Upon receipt, the bloodspots were stored in our lab at −80°C until DNA extraction. Laboratory personnel performing DNA methylation analysis were blinded to study subject information. DNA was extracted from one half of a newborn bloodspot using the QiaAmp DNA blood kit (Qiagen Inc, Valencia, CA) and stored at −80°C. Average yield of DNA from half a spot was 550ng (ranged from 17 ng to 2342 ng). Generally, 700 ng to 1 µg of genomic DNA from each sample was treated with bisulfite using the EZ-96 DNA Methylation Kit™ (Zymo Research, Irvine, CA, USA), according to the manufacturer’s recommended protocol and eluted in 18 µl. The results of the Infinium HumanMethylation450 BeadChip (HM450) were compiled for each locus as previously described and were reported as beta (β) values [64]. A normal exponential background correction was first applied to the raw intensities at the array level to reduce background noise followed by dye bias correction [54]. We then normalized each sample’s methylation values to have the same quantiles to address sample to sample variability [65].

CpG loci on the HM450 array were removed from analyses if they were on the X and Y chromosomes, or if they contained SNPs, deletions, repeats, or if they have more than 10% missing values, leaving 383 857 probes for analysis. Only probes mapped to the promoter region were included in the analysis yielding 178 309 loci for interrogation and multiple comparison correction. Samples from 273 participants were included for initial analysis of DNA methylation. Thirty-two samples were removed for missing PM10 air pollution exposure and an additional one was excluded since many probes for this sample were identified as outliers. This left 240 samples for the primary analysis.

For replication analyses, PCR primers targeting the selected loci were developed using MethPrimer software [66]. Primers were designed to cover the HM450 loci of interest and the specificity of the primer sequences were confirmed using insilico PCR. 250 ng DNA was treated with bisulfite using the EZ-96 DNA Methylation Kit™ as described above and Methylation analyses were performed using the Pyrosequencing (PSQ) HS 96 Pyrosequencing System (Biotage AB, Uppsala, Sweden) as described in previous work [67]. The output from Pyrosequencing is reported as a percent of DNA methylation at each CpG locus. As a quality control check to estimate the bisulfite conversion efficiency, we placed duplicate genomic DNA samples on each bisulfite conversion plate to estimate the internal plate variation of bisulfite conversion and the Pyrosequencing reaction. Conversion efficiency was greater than 95%. We also added universal PCR products amplified from cell line DNA on each Pyrosequencing plate to check the run-to-run and plate-to-plate variation in performing Pyrosequencing reactions. In addition, the Pyrogram peak pattern from every sample was checked to confirm the quality of reaction.

Statistical A nalyses

Descriptive analyses were performed to examine the distribution of subject characteristics. Density plots of DNA methylation values from the HM450 BeadArray were created and evaluated for quality control. Outlier DNA methylation values were identified as values that were either greater than the median+ 5*SD or less than the median − 5*SD and were removed from analyses.

The association between air pollution exposure and percent DNA methylation was analyzed by a generalized linear regression model using quasi-binomial link family, adjusted for gender, plate, and cell types. Additional sensitivity analyses were conducted to evaluate potential confounding of results by maternal education, ethnicity, and in utero tobacco smoke and were found to be minimal. The following cell types were estimated using the method of Houseman et al. [55]: B-lymphocytes, granulocytes, monocytes, natural killer cells, CD4+ T-lymphocytes, and CD8+ T-lymphocytes. The generalized linear regression model was used to address the non-normal distribution of DNA methylation values, which are bounded by 0 and 1 and in many cases heavily skewed toward one end or the other. All regression analyses were adjusted for multiple testing at a false discovery rate (FDR) of 0.15, using the method of Benjamini and Hochberg [68]. A less conservative FDR threshold was chosen because we were employing a two-step screening and replication process and did not want to be too restrictive in the first step.

Finally, we used logistic or linear regression models to evaluate the associations between methylation and childhood health outcomes including asthma, CIMT, and z-score transformed blood pressure (normalized for age, height, and sex [69]), adjusting for maternal education, sex, ethnicity, DNA methylation plate, age at clinic assessment, BMI z-score, and in utero tobacco smoke exposure. These covariates were chosen for adjustment based on a priori knowledge of their associations with health outcomes, with the exception of methylation plate, which was included to adjust for potential batch differences. A similar generalized linear regression model using quasi-binomial link family was used in the replication population to evaluate the association between air pollutants and DNA methylation measured by Pyrosequencing, with adjustment for gender, plate, maternal education, and ethnicity. In utero tobacco smoke was not adjusted because no subjects had exposure. Blood cell fractions could not be estimated for this population because 450K data were not available. All tests assumed a two-sided alternative hypothesis, a 0.15 false-discovery rate, and were conducted using the R programming language, version R3.2.2.

Supplementary Data

Click here for additional data file.