Association of cigarette smoking and CRP levels with DNA methylation in α-1 antitrypsin deficiency

Abstract

Alpha-1 antitrypsin (AAT) deficiency and tobacco smoking are confirmed risk factors for Chronic Obstructive Pulmonary Disease. We hypothesized that variable DNA methylation would be associated with smoking and inflammation, as reflected by the level of C-Reactive Protein (CRP) in AAT-deficient subjects. Methylation levels of 1,411 autosomal CpG sites from the Illumina GoldenGate Methylation Cancer Panel I were analyzed in 316 subjects. Associations of five smoking behaviors and CRP levels with individual CpG sites and average methylation levels were assessed using non-parametric testing, linear regression and linear mixed effect models, with and without adjustment for age and gender. Univariate linear regression analysis revealed that methylation levels of 16 CpG sites significantly associated with ever-smoking status. A CpG site in the TGFBI gene was the only site associated with ever-smoking after adjustment for age and gender. No highly significant associations existed between age at smoking initiation, pack-years smoked, duration of smoking, and time since quitting smoking as predictors of individual CpG site methylation levels. However, ever-smoking and younger age at smoking initiation associated with lower methylation level averaged across all sites. DNA methylation at CpG sites in the RUNX3, JAK3 and KRT1 genes associated with CRP levels. The most significantly associated CpG sites with gender and age mapped to the CASP6 and FZD9 genes, respectively. In summary, this study identified multiple potential candidate CpG sites associated with ever-smoking and CRP level in AAT-deficient subjects. Phenotypic variability in Mendelian diseases may be due to epigenetic factors.

Keywords: :

• 68kDa (TGFBI)
• C-Reactive Protein (CRP)
• Chronic Obstructive Pulmonary Disease (COPD)
• Illumina GoldenGate Methylation Cancer Panel I
• alpha-1 antitrypsin (AAT) deficiency
• beta-induced
• methylation
• smoking behaviors
• transforming growth factor

Introduction

Chronic Obstructive Pulmonary Disease (COPD) is a common disorder with complex genetic etiology; the main environmental cause of COPD is tobacco smoking. Deficiency in alpha-1 antitrypsin (AAT), caused by rare polymorphisms in the serpin peptidase inhibitor, clade A (α-1 antiproteinase, antitrypsin), member 1 (SERPINA1) gene on chromosome 14q32.1, was the first established genetic determinant of COPD.¹ Subsequent studies, including whole genome association studies, identified additional, yet more common, genetic variants associated with risk to develop COPD^2-4 and smoking behaviors.^5-7 Since tobacco smoking associates with severe airflow limitation in AAT-deficient subjects,⁸ it is of importance to study biological factors responsible for the harmful effects of smoking in these subjects. However, the development of COPD in subjects with severe homozygous AAT deficiency is unpredictable regardless of smoking status⁸ and the impact of smoking is variable; some of this variability may be due to epigenetic alterations.

Cigarette smoking, as well as the smoking-related disease COPD, has systemic impact. For example, systemic inflammation, as reflected by the elevated levels of C-Reactive Protein (CRP), observed in subjects who have COPD, predicts hospitalization and mortality.⁹ Although genetic factors influencing smoking-related phenotypes and levels of CRP have been widely studied in genome-wide association studies, it is less known how epigenetic variation, such as DNA methylation, correlates with these traits. Studies of global methylation level, as reflected by the methylation status of Long Interspersed Nucleotide Elements (LINE-1), in cancer suggested that smoking associates with global DNA hypomethylation,¹⁰ although this finding was not confirmed in a study of healthy subjects.¹¹ Recent developments in high throughput epigenetic techniques has facilitated assessing the level of DNA methylation of thousands of individual CpG sites throughout the genome.¹² It has been reported that the methylation levels of 138 individual CpG sites are significantly associated with ever-smoking status in 53 lung tissue samples using the Illumina GoldenGate Methylation Cancer Panel I.¹³ A recent study of blood-derived DNA samples found a hypomethylated CpG site in the coagulation factor II (thrombin) receptor-like 3 (F2RL3) that was associated with current smoking in a study encompassing 27,578 CpG sites.¹⁴

Mendelian diseases, once considered to have phenotypes that were the outcome of single-gene inheritance, are now considered to be impacted by modifier genes. However, given the phenotypic heterogeneity frequently observed in Mendelian diseases, it is likely that genetic and epigenetic factors, interacting with environmental exposures, together play a role. The aim of the current study was to investigate associations of DNA methylation of multiple individual CpG sites as well as their global average, with smoking and CRP levels in subjects with severe AAT-deficiency, a Mendelian disease caused by homozygosity for the non-synonymous “Z” (Glu342Lys; rs28929474) SERPINA1 allele.

Results

This study included data from 316 subjects from the AAT Genetic Modifiers Study (Table 1). The percent methylation (i.e., the Beta level, ranging from 0 to 1), of 1,505 CpG sites was assessed using Illumina GoldenGate Methylation Cancer Panel I in DNA from blood samples. The initial data set underwent strict quality control (QC) procedures (Table 2). The histogram of all individual methylation Beta levels, passing QC, showed a bimodal distribution (Fig. S1). A total of 1,411 autosomal CpG sites were carried forward for further analysis.

Table 1. Characteristics of 316 subjects from 162 families studied

Males	n (%)	143 (45.3)
Age in years	mean (SD)	52.6 (9.7)
Ever-smokers	n (%)	196 (62.0)
Current smokers	n (%)	13 (4.1)
Pack-years smoked**	mean (SD)	18.1 (14.4)
Age at smoking initiation**	mean (SD)	17.7 (4.7)
Duration of smoking in years**	mean (SD)	18.5 (8.4)
Time since quitting smoking in years***	mean (SD)	15.4 (10.3)
CRP level in mg/L	mean (SD)	4.8 (12.8)
ZZ AAT genotype at rs28929474	n(%)	316 (100%)

** Within 196 ever-smokers; ***within 183 former smokers; SD, standard deviation; CRP, C-Reactive Protein.

Table 2. Quality control criteria applied in the current study

Step	Number of subjects	Number of CpG sites
Starting data set	480	1,505
Remove positive/negative controls and samples that failed bisulfite conversion	398	1,505
Remove duplicates	364	1,505
Remove subjects with gender inconsistency	361	1,505
Remove subjects with who failed “First Hybridization” control	358	1,505
Remove sex chromosome markers	358	1,421
Remove CpG sites with mean detection p value ≥ 0.05	358	1,411
Normalize methylation Beta levels	358	1,411
Keep subjects with ZZ genotype at SERPINA1 locus – final data set	316	1,411

SERPINA1, serpin peptidase inhibitor; clade A (α-1 antiproteinase, antitrypsin), member 1.

Association of individual methylation β levels with smoking-related phenotypes

Using a conservative Bonferroni-corrected significance threshold (p < 3.54 × 10⁻⁵), 16 CpG sites significantly associated with ever-smoking status in univariate regression analysis; a CpG site (cg07852148) in the transforming growth factor, β-induced, 68kDa (TGFBI) was the most significantly associated (Table 3). All of those 16 CpG sites were relatively hypomethylated in the whole cohort as well as in ever- as compared with never-smokers (Table 3 and Figs. S2–8). All 16 sites were located in the CpG islands. Except for a site in the gene deleted in bladder cancer 1 (DBC1), all 16 sites remained significant after accounting for family structure using linear mixed effect (LME) models (Table 3). Nine and six CpG sites remained significant using Box-Cox transformation of the methylation Beta level or using Mann-Whitney nonparametric tests (Table 3), respectively.

Table 3. CpG sites associated with ever-smoking status with p < 10⁻⁴ in a univariate linear regression analysis and corresponding results from the linear mixed effect model and Mann-Whitney test

Gene	CG id	Methylation level in 316 subjects studied			Linear regression		LME model		Mann-Whitney test p value
		Median	IQR		B	p	B	p
TGFBI	cg07852148	0.038	0.027	0.055	-0.0148	1.55x10^−6 *	-0.0143	5.89x10^−6	1.54x10^−6
SCGB3A1	cg19421379	0.047	0.035	0.070	-0.0177	2.40 x10^−6 *	-0.0175	5.22x10^−6	1.81x10^−5
ISL1	cg16203270	0.052	0.037	0.081	-0.0180	3.35 x10^−6 *	-0.0187	2.30x10^−6	1.95x10^−5
FZD9	cg25707686	0.196	0.155	0.243	-0.0386	3.70x10^−6 *	-0.0384	7.01x10^−6	0.00013
SOX2	cg18036037	0.049	0.037	0.062	-0.0130	6.06x10^−6 *	-0.0131	8.59x10^−6	7.14x10^−5
SLIT2	cg13875122	0.075	0.051	0.108	-0.0266	6.58x10^−6 *	-0.0267	9.58X10^−6	8.91x10^−6
FAT	cg06830038	0.076	0.053	0.105	-0.0238	9.99x10^−6 *	-0.0254	4.49x10^−6	7.49x10^−5
EPHA3	cg10315231	0.046	0.034	0.072	-0.0160	1.08x10^−5	-0.0160	1.78x10^−5	4.18x10^−5
RAB32*	cg21792703	0.025	0.019	0.038	-0.0105	1.45x10^−5	-0.0108	1.19x10^−5	0.00019
DBC1	cg14445814	0.067	0.046	0.094	-0.0223	1.70x10^−5 *	-0.0216	4.39x10^−5	1.19x10^−5
NR2F6	cg24945194	0.060	0.046	0.075	-0.0113	1.79x10^−5	-0.0117	1.41x10^−5	0.00012
DLK1	cg20175079	0.067	0.048	0.096	-0.0198	1.85x10^−5	-0.0202	2.15x10^−5	0.00053
ISL1	cg14024101	0.041	0.030	0.058	-0.0130	2.90x10^−5	-0.0133	2.71x10^−5	0.001
TGFBI	cg00833799	0.236	0.179	0.300	-0.0437	3.09x10^−5	-0.0446	3.16x10^−5	6.10x10^−5
RAB32	cg01851450	0.040	0.028	0.057	-0.0136	3.10x10^−5	-0.0147	9.00x10^−6	9.35x10^−5
EYA4	cg23916819	0.087	0.062	0.118	-0.0244	3.52x10^−5	-0.0253	2.72x10^−5	4.44x10^−5
HTR1B	cg26034501	0.112	0.079	0.153	-0.0293	3.69x10^−5	-0.0299	2.78x10^−5	6.45x10^−5
FRZB	cg01872931	0.067	0.050	0.094	-0.0191	3.99x10^−5	-0.0193	4.77x10^−5	0.00014
DAPK1	cg01984172	0.029	0.021	0.040	-0.0079	5.25x10^−5 *	-0.0079	7.08x10^−5	1.40x10^−5
RARB	cg14265392	0.016	0.014	0.022	-0.0056	6.40x10^−5	-0.0056	8.11x10^−5	7.61x10^−5
DCC	cg22444505	0.083	0.061	0.110	-0.0195	7.48x10^−5	-0.0195	9.52x10^−5	0.00012
ADAMTS12	cg13887940	0.054	0.041	0.071	-0.0129	9.04x10^−5	-0.0131	9.21x10^−5	0.00018
GABRB3	cg20421990	0.041	0.031	0.053	-0.0100	9.57x10^−5	-0.0101	1.07x10^−4	1.612x10^−5

Association p values < 3.54 × 10⁻⁵ are depicted in bold. *p < 3.54 × 10⁻⁵ after Box-Cox transformation of the methylation Beta level. B, regression/LME model coefficient for the ever-smoking variable set as independent variable and coded as “1” for ever-smokers and as “0” for never-smokers (i.e., the reference group); IQR, interquartile range; LME, linear mixed effect.

Using univariate linear regression and a conservative Bonferroni-corrected significance threshold, no CpG sites significantly (p < 3.54 × 10⁻⁵) associated in this cohort with pack-years smoked, duration of smoking or time since quitting smoking (Table 4). Two CpG sites in the membrane metallo-endopeptidase (MME) and TGFBI genes significantly associated with age at smoking initiation in a univariate linear regression model (Table 4). However, these two associations were attenuated (p = 0.002 and p = 0.008, respectively) or not significant (p > 0.16) after Box-Cox transformation of methylation Beta level or according to non-parametric Spearman’s rank correlation test, respectively. CpG sites in the myeloid leukemia factor 1 (MLF1; cg11432768) and tumor suppressor candidate 3 (TUSC3; cg10249582) additionally associated with age at smoking initiation at p < 10⁻⁴ using normalized methylation Beta levels (B = 0.00075, p = 7.46x10⁻⁵ and B = 0.0011, p = 7.60x10⁻⁵, respectively). These associations were less significant after Box-Cox transformation of methylation Beta level (p = 0.002 and p = 0.001, respectively) and not significant using Spearman’s rank correlation test (p = 0.368 and p = 0.054, respectively). No additional associations at p < 10⁻⁴ were identified in the analysis using normalized methylation Beta level and smoking-related phenotypes.

Table 4. Five CpG sites most significantly associated with the age at smoking initiation, pack-years smoked, time since quitting smoking or duration of smoking in a univariate linear regression model and corresponding results from the linear mixed effect model

Phenotype	Gene	CG id	Methylation level*			Linear regression		LME model
			Median	IQR		B	p	B	p
Age at smoking initiation	MME	cg20228377	0.048	0.030	0.072	0.00378	7.62x10^−8	0.00380	4.34x10^−7
	TGFBI	cg07852148	0.034	0.025	0.049	0.00157	1.16x10^−6	0.00157	4.73x10^−6
	CSF1	cg06450700	0.026	0.020	0.035	0.00083	7.05x10^−5	0.00083	0.00016
	GABRB3	cg20421990	0.038	0.028	0.048	0.00121	7.19x10^−5	0.00121	0.00013
	TUSC3	cg10249582	0.050	0.037	0.070	0.00174	0.00012	0.00192	5.46x10^−5
Duration of smoking	RET	cg21859545	0.052	0.040	0.068	0.00071	0.00072	0.00065	0.0025
	CEBPA	cg23071874	0.035	0.027	0.047	0.00057	0.00374	0.00057	0.0046
	EGFR	cg24522859	0.037	0.030	0.046	-0.00040	0.00484	-0.00042	0.0033
	MYH11	cg15581089	0.039	0.026	0.058	-0.00068	0.00874	-0.00071	0.0075
	NOTCH1	cg26924342	0.055	0.041	0.075	0.00067	0.00932	0.00057	0.0284
Pack-years smoked	RET	cg21859545	0.052	0.040	0.068	0.00040	0.0007	0.00039	0.0014
	HLA-DOA	cg23794671	0.151	0.117	0.215	-0.00120	0.0013	-0.00113	0.0022
	GABRA5	cg17525249	0.734	0.693	0.777	0.00090	0.0018	0.00082	0.0048
	AXL	cg20384525	0.099	0.070	0.137	-0.00070	0.0078	-0.00070	0.006
	APC	cg24136097	0.206	0.166	0.248	-0.00090	0.0079	-0.00073	0.0254
Time since quitting smoking	PLSCR3	cg26114392	0.394	0.339	0.455	0.0019	0.0010	0.0018	0.0028
	ERN1	cg13884260	0.219	0.180	0.261	-0.0015	0.0010	-0.0015	0.0016
	CAV1	cg07838272	0.161	0.129	0.187	-0.0012	0.0013	-0.0012	0.0019
	EYA4	cg24842760	0.072	0.054	0.093	0.0006	0.0016	0.0006	0.0030
	BCR	cg01574123	0.797	0.751	0.840	-0.0017	0.0023	-0.0016	0.0054

Association p values < 3.54 × 10⁻⁵ are depicted in bold. *In subjects with non-missing, smoking-related phenotype studied. B, regression/LME model coefficient for the continuously defined smoking-related phenotype set as independent variable; IQR, interquartile range; LME, linear mixed effect.

Association of methylation levels with CRP level

No CpG sites associated at the Bonferroni-corrected significance threshold (p < 3.54x10⁻⁵) with the level of CRP set as dependent or independent variable (Table 5). However, CpG sites in the runt-related transcription factor 3 (RUNX3) and keratin 1 (KRT1) associated with log₂-transformed CRP level, set as dependent variable and adjusted for gender in a linear regression analysis, at p < 10⁻⁴ (Table 5). These associations were nominally significant according to Spearman’s rank correlation test that did not take gender into account (p = 0.0016 and p = 0.0006, respectively).

Table 5. Five CpG sites most significantly associated with CRP (as a dependent or independent variable)

Level of CRP set as:	Gene	CG id	Methylation level in 316 subjects studied			Linear regression		LME model
			Median	IQR		B	p	B	p
Independent variable*	RIPK4	cg11719297	0.087	0.068	0.106	0.00042	0.0009	0.00041	0.0012
	ST6GAL1	cg23385525	0.097	0.072	0.147	0.00082	0.0025	0.00074	0.0039
	TIE1	cg19492156	0.328	0.255	0.384	-0.00130	0.0034	-0.00131	0.0034
	RUNX3	cg21368948	0.873	0.837	0.901	0.00070	0.0044	0.00071	0.0040
	KRT1	cg06030058	0.823	0.798	0.841	0.00055	0.0047	0.00051	0.0071
Dependent variable (log2-transformed)**	RUNX3	cg21368948	0.873	0.837	0.901	7.1	6.57x10^−5	7.1	6.04x10^−5
	KRT1	cg06030058	0.823	0.798	0.841	8.8	8.18x10^−5	8.1	0.00028
	JAK3	cg05244380	0.746	0.692	0.792	-4.3	0.00059	-4.1	0.0010
	CD82	cg24700286	0.248	0.186	0.316	-3.4	0.00075	-3.6	0.00024
	MMP9	cg12243192	0.049	0.036	0.064	-11.1	0.0014	-11.0	0.0013

* Tested with univariate linear regression analysis. **Regression analysis was additionally adjusted for gender. CRP, C-Reactive Protein; B, regression/LME model coefficient; IQR, interquartile range; LME, linear mixed effect.

Analysis of normalized methylation Beta levels revealed one additional site in the gene T-cell lymphoma invasion and metastasis 1 (TIAM1, cg0652489, B = 0.0005, p = 5.0 × 10⁻⁵) gene associated with CRP level set as independent variable at p < 10⁻⁴. This association was not significant as assessed by Spearman’s rank correlation test (p = 0.533) or after Box-Cox transformation of methylation Beta level (p = 0.718).

Unadjusted percent methylation Beta levels in the Janus kinase 3 (JAK3, Table 6) associated with the level of log₂-transformed CRP below suggestive significance level using normalized methylation Beta levels (p = 4.29 × 10⁻⁵). This association was confirmed by Spearman’s rank correlation test that did not take gender into account (p = 8.3 × 10⁻⁵). Of interest, linear regression analysis of normalized methylation Beta levels yielded suggestive associations of KRT1 (p = 3.61 × 10⁻⁵) and RUNX3 (p = 4.39 × 10⁻⁵) with log₂-transformed CRP level set as dependent variable respectively.

Table 6. Association of mean methylation Beta level with smoking related phenotypes and CRP level in a univariate linear regression analysis

Phenotype	Linear regression		LME model
	B	p	B	p
Ever-smoking	-0.00452	0.007	-0.00467	0.006
Age at smoking initiation	0.000384	0.062	0.000452	0.031
Duration of smoking	0.00000877	0.939	-0.0000399	0.729
Pack-years smoked	-0.0000492	0.464	-0.0000425	0.525
Time since quitting smoking	-0.000153	0.118	-0.000137	0.171
CRP level	-0.000029	0.651*	-0.0000452	0.478**

P values < 0.05 are depicted in bold. *p = 0.84 for log₂-transformed level of CRP set as dependent variable and adjusted for gender. **p = 0.708 for log₂-transformed level of CRP set as dependent variable and adjusted for gender. CRP, C-Reactive Protein; B, regression/LME model coefficient for the continuously defined smoking-related or CRP level phenotype set as independent variable; IQR, interquartile range; LME, linear mixed effect.

Impact of the adjustment of age and gender on the top associated CpG sites

In a multivariate linear regression model, 18 and 26 CpG sites associated with gender and age below the suggestive significance threshold, respectively (Table S1). A CpG site (cg07942426) in the caspase 6, apoptosis-related cysteine peptidase (CASP6) gene was the most significantly associated CpG site with gender (p = 1.1 × 10⁻¹²), while CpG site cg25707686 in the frizzled homolog 9 (Drosophila) (FZD9) was the most significantly associated CpG site (p = 4.2 × 10⁻¹⁰) with age (Table S1). Four additional CpG sites associated with age or gender using normalized methylation Beta levels below the suggestive significance threshold (Table S1). Three CpG sites associated with ever-smoking status at p < 10⁻⁴ (Table S2), and one CpG (cg07852148) site in the TGFBI gene remained significant at p < 10⁻⁴ after Box-Cox transformation of methylation Beta level (p = 2.2 × 10⁻⁵) after adjusting for gender and age. CpG sites in the MME and TGFBI genes, already identified as significant in a univariate regression analysis, were the only sites associated with age at smoking initiation at p < 10⁻⁴ after adjusting for gender and age (Table S2), yet the associations were less significant after Box-Cox transformation of methylation Beta level (p = 0.01 and p = 0.02, respectively). No additional associations at p < 10⁻⁴ were identified for other smoking-related phenotypes using raw or normalized betas and adjusting for age and gender. CpG site cg0652489 in TIAM1 was the only site associated with CRP level set as independent variable (p = 5.87 × 10⁻⁵ for normalized methylation Beta level), yet not after Box-Cox transformation of the methylation Beta level (p = 0.74).

Pyrosequencing

We selected several of the top associated CpG sites from each analysis for validation using pyrosequencing. All of the CpG sites selected demonstrated correlation between the pyrosequencing percent methylation and the Illumina Beta values [all p for correlation (Pearson) < 0.05] although the degree of correlation varied (JAK3 r² = 0.50, RUNX3 r² = 0.50, KRT1 r² = 0.29, MME r² = 0.15, TGFBI r² = 0.12). When examined in the specified model that demonstrated association with the Illumina percent methylation, a CpG site in TGFBI demonstrated a trend for association of lower methylation with younger age at smoking initiation (p = 0.06), lower methylation with higher pack-years of smoking (p = 0.05) and lower methylation with more total years smoked (p = 0.005), without any significant impact of adjustment for age, gender or batch; there was no trend with ever-smoking (p = 0.17). For the CpG site in KRT1, we observed association of higher methylation with higher CRP (p = 0.04); for JAK3 we observed a trend for association of lower methylation with higher CRP (p = 0.06). The associations for KRT1 and JAK3 demonstrated the same trend when adjusted for gender (KRT1 p = 0.046, JAK3 p = 0.070). The trend for JAK3 remained the same after adjustment for pyrosequencing batch (JAK3 p = 0.06) but was further attenuated with adjustment for pyrosequencing batch for KRT1 (p = 0.09). There were no observable trends for methylation of a CpG in MME and age at smoking initiation or RUNX3 and CRP level (p > 0.10). We also performed pyrosequencing at two sites in the SERPINA1 α-1 antitrypsin gene (which have corresponding Illumina cg designations cg02181506 and cg2462104; probe details in methods). We observed lower percent methylation at cg02181506 in ever-smokers vs. never-smokers (11.1% vs. 13.4%, p = 0.04), and this finding remained robust after adjustment for age, gender and pyrosequencing batch (p = 0.04); there was no association with CRP (p = 0.29). We did not demonstrate association of cg2462104 with any smoking phenotype or CRP (all p > 0.1).

Analyses of average methylation levels across all 1,411 autosomal CpG sites

Ever-smoking status associated with lower mean methylation Beta level (p = 0.007, Table 6). This result was further extended by an observation that 34% (n = 485) of all individual CpG sites analyzed were associated with ever-smoking status below a nominal significance threshold of p < 0.05 (n = 380 after adjusting for age and gender), and 88% (n = 427) of those sites were relatively hypomethylated in ever- as compared with never-smokers [93% (n = 354) after adjusting for age and gender]. Lower age at smoking initiation associated with lower mean methylation Beta level at borderline significance using linear regression and nominal significance using LME model (Table 6). Subsequent analysis revealed that age at smoking initiation associated with individual methylation Beta level of 130 sites at p < 0.05 (n = 123 after adjusting for age and gender), and 98% (n = 128) of those sites were hypomethylated with younger age at smoking initiation [99% (n = 122) after adjusting for age and gender]. Less significant associations were found when analyzing normalized methylation Beta levels and age at smoking initiation with the adjustment for gender and age (Table S3). However, similar results were obtained for the ever-smoking phenotype (Table S3). In univariate and multivariate linear regression models, age and gender were not significantly associated with the mean methylation Beta level (p > 0.76 for the average of raw betas and p > 0.21 for the average of normalized betas). Pack-years of cigarettes smoked, duration of smoking and time since quitting were not significantly associated with mean methylation Beta level (Table 6).

Discussion

This study identifies DNA methylation at candidate CpG sites associated with ever-smoking status, age at smoking initiation and CRP level in AAT-deficient subjects. Importantly, this study showed that ever-smoking associates with relative global hypomethylation in subjects with AAT deficiency, as approximated by the mean of methylation level across all CpG sites tested, while earlier age at smoking initiation associates with global hypomethylation in AAT-deficient subjects. Therefore, DNA methylation may be an important mediator of the effects of smoking on disease heterogeneity in AAT deficiency, either at specific genes or more globally across the genome. Importantly, our study also highlights that model selection, including covariate inclusion, will impact the outcome of epigenetic studies.

Besides numerous significant associations between methylation levels and smoking-related phenotypes, the current study additionally illustrates potential difficulties in the analysis of DNA methylation data. A CpG site in the TGFBI gene, was among the most significantly associated CpG sites with ever-smoking status and age at smoking initiation using linear regression analysis, however the association with the latter phenotype was not confirmed by non-parametric testing. The attenuation of p value of the association between the TGFBI site, and another one in the MME gene, with age at smoking initiation after Box-Cox transformation of methylation Beta level, suggests initial inflation of statistics due to non-normal distribution of the methylation Beta values. Of importance, adjustments for age and gender attenuated the majority of p values observed in the analysis of ever-smoking as compared with univariate regression. This suggests a potential confounding effect of those covariates on the univariate associations observed. Taken together, non-normality of the distribution of methylation Beta levels as well as potential confounding by age or gender were the additional challenges addressed in the current data analysis.

We identified association of DNA methylation at CpG sites in the RUNX3, KRT1 and JAK3 genes with the level of CRP set as dependent variable, and all those associations achieved, at least, a nominal significance threshold using non-parametric testing. To our knowledge, there have been no studies reporting on the association between CRP and methylation level of individual CpG sites using high-throughput methylation techniques. It is therefore important to verify these associations in additional cohorts and using expanded platforms, as it may lead to new insights into the network of events that contribute to systemic inflammation.

Of interest, our study identified several CpG sites associated with age and gender below a conservative significance threshold. Sites in genes such as FZD9 or activin A receptor, type I (ACVR1) were also significantly associated with age in previous studies.^13,15 Although our study was cross-sectional in nature, these genes may be considered as plausible epigenetic candidate markers of the aging process. We also confirmed the previously observed (in Gene Expression Omnibus)¹⁶ association of gender with methylation status of a single CpG site in the CASP6 gene.¹⁵ There is an ongoing interest in the validity of assessing association of methylation marks with gender due to cross-reactivity of probes, as observed by Chen and colleagues for the Illumina Infinium 27k chip.¹⁷ To our knowledge, it is yet to be established whether this limitation applies to the Golden Gate platform used in the current study; thus, we interpret this replicated finding with caution.

We averaged the methylation level of 1,411 analyzed autosomal loci and used it as an approximation of global DNA methylation, similar to the approach of Bibikova and colleagues.¹⁸ We found that ever-smoking associates with lower level of global methylation. Moreover, the number of single CpG sites associated with ever-smoking was much larger than expected by random chance. Global hypomethylation of LINE has been associated with smoking status in head and neck squamous cell carcinoma patients;¹⁰ however, this was not the case in a study involving healthy subjects.¹¹ Our study additionally suggested that earlier age at smoking initiation may associate with global DNA hypomethylation in AAT-deficient subjects. In contrast, no significant association with age at smoking initiation and the level of global methylation, measured in LINE and Alu sequences, has been observed in healthy subjects,¹¹ and the hypermethylation of the tumor suppressor Ras association (RalGDS/AF-6) domain family member 1 isoform A (RASSF1A) associates with earlier age at smoking initiation in non-small cell lung cancer.^19,20 Although it is unclear how well average methylation level of the preselected CpG sites on the Illumina GoldenGate Methylation Cancer Panel I reflects the level of global DNA methylation and to what extent correlation of methylation status with certain traits differs between various groups of subjects, these findings may provide insights into the systemic impact of smoking in patients with AAT deficiency. Since nicotine was shown to downregulate expression of DNA (cytosine-5-)-methyltransferase 1, an enzyme responsible for DNA methylation in vivo,²¹ we believe that smoking may determine methylation status of numerous CpG sites in the human genome. However, this phenomenon may be dependent on the phenotypic and genetic susceptibility characteristics of studied subjects, as well as tissue used as a source for DNA analysis. In contrast to the analysis of the ever-smoking phenotype, we found no significant association between level of CRP and the surrogate marker for global DNA methylation. This observation supports a previous study that reported no significant association between global DNA methylation assessed with LINE-1 and CRP levels.²²

The major limitation of the current study is relatively small number of CpG sites tested with pre-selection of the CpG sites, based on their relevance to cancer. Several biologically interesting candidate CpG sites such as those in the F2RL3 and the α/β nicotinic cholinergic receptors on chromosome 15q25 (i.e., determinants of smoking intensity)^5-7 are not included on the Cancer Panel I platform and were not evaluated in this cohort. It is therefore of future interest to assess association of other CpG sites with smoking related phenotypes and CRP level in AAT-deficient subjects. The source of DNA may be considered as another limitation of the current study. Since DNA methylation varies across tissues, it is not clear how well the CpG methylation assessed in blood correlates with the level found in the lungs or liver of AAT-deficient subjects. This may partially explain the lack of overlap between CpG sites significantly associated with ever-smoking status in the current study and those found by Christensen and colleagues in lung tissue samples; however, our goal was to focus on these methylation marks as a biomarker of cigarette smoke exposure and systemic inflammation.¹³ We are also aware of the fact that we were unable to establish the causal pathways, since it is certainly possible that variability in DNA methylation may be actually a cause of the phenotype studied, and not an outcome. We attempted validation of the initial Golden Gate results using pyrosequencing. We were surprised to observe only moderate to low correlations between the percent methylation findings between these two platforms. We should make note that the pyrosequencing was performed on DNA from an aliquot that was collected at the same time as the original sample but that was thawed and bisulfite treated specifically for this validation step. This may be one possible explanation for the moderate degree of correlation between the pyrosequencing and the Golden Gate results. It is also possible that our findings for our initial array analysis represent false positive statistical observations as previously mentioned; however, it is encouraging that there is still a trend for association of KRT1 and JAK3 with CRP levels, and TGFBI with smoking phenotypes, and that we have replicated the association with gender for CASP6 already published in the literature. Repeat experimentation in a larger cohort of AAT deficient subjects on newer generation platforms and more extensive pyrosequencing on contemporaneously bisulfite-converted samples are needed. We observed that CpG sites in the KRT1 and RUNX3 were among most significantly associated sites with CRP level irrespective of whether methylation level was set as predicting or predictor variable. This important question concerning phenotypic causation needs to be addressed in longitudinal studies.

In summary, our current study suggests an association of ever-smoking and younger age at smoking initiation with relative hypomethylation of the methylome in AAT-deficient subjects. We also highlight the need for methods development to inform model specification and covariate adjustment in large-scale studies of DNA methylation. Future epigenetic studies, investigating smoking behaviors and systemic inflammation, are required to establish whether our findings may be translated to other cohorts, as well as to unravel molecular mechanisms driving epigenetic changes in AAT-deficiency that may contribute to the protean manifestations in this and other Mendelian disease.

Methods

Subjects, measurement of DNA methylation and level of CRP

Subjects from the AAT Genetic Modifiers Study,⁸ homozygous for the “Z” allele of the Glu342Lys (rs28929474) polymorphism in SERPINA1, were included (Table 1). This cohort has been described previously⁸; our current analysis includes 316 siblings from 162 families. DNA was extracted from blood samples and underwent bisulfite conversion using the 96 well EZ DNA Methylation Gold Deep-well kits (Zymo Research).1 µg of input DNA was used per sample. The methylation levels of 1505 CpG sites in the Illumina GoldenGate Methylation Cancer Panel I were assayed using the Sentrix Array Matrix (SAM) 96-well format, according to the manufacturer’s standard protocol (Illumina, Inc.). The SAMs were scanned on the BeadStation 500GX, and intensity values from the red and green channels were converted to methylation Beta values, using the Illumina BeadStudio software. As such, the methylation level of each CpG site was represented as Beta i.e., a ratio, ranging from 0 to 1; this ratio represents the fluorescent signal intensity of the methylated channel to the sum of methylated and unmethylated channels plus 100. Standard quality control (QC) procedures, with exclusion of CpG sites located on sex chromosomes, were performed (Table 2) and resulted in a final data set encompassing 316 subjects and 1,411 CpG sites across 767 autosomal genes. Serum C-reactive protein (CRP) levels were measured using a high sensitivity immunoturbidimetric assay on the Hitachi 917 analyzer (Roche Diagnostics), with reagents from Denka Seiken (Denka Seiken). This high sensitivity assay has a lower limit of detection for CRP of 0.03 mg/L. All participants provided written informed consent, and the study protocol was approved by individual Institutional Review Boards at each study site.

Pyrosequencing

For validation of the top associated CpGs we used DNA from the AAT cohort for pyrosequencing. Forward, reverse and sequencing primers were designed using the Pyromark Assay Design Software (version 2.0.1.15) (Human Genome build 19). A touchdown PCR program was used to amplify the bisulfite converted DNA using the HotStarTaq kit (Qiagen, Inc.). Pyrosequencing was performed using the PyromarkQ96MD per manufacturer’s instructions. Using the Pyromark CpG Software (version 1.0.11), CpG methylation percentages were calculated using the height of the T and C peaks at the methylation site and applying the formula (C/C+T)x100.

The primers and completion rates for each assay include:

CpG TGFBI (cg07852148)

Forward PCR primer AGGGTAAGGGTTGGGAAAATT; Reverse PCR primer AACCTACTATACTACAACACCAACTAAT; Sequencing Primer GTTAGTAGTTTATTTAATTTGG.27 samples failed quality control for a 92.5% completion rate; data from 289 subjects were carried forward for validation analysis.

CpG JAK3 (cg05244380)

Forward PCR primer: AAAGTTAGGGTGTTAGGATAGGTAT; Reverse PCR primer: ACCTCCTAACTACTCTTCAAACTACTA; Sequencing Primer: GGATAGGTATAGATTGGAAT. Nine samples failed quality control for a 97% completion rate; data from 307 samples were carried forward for validation analysis.

CpG KRT1 (cg06030058)

Forward PCR primer: AGTAATAGAAAATAGGTAGTAATGGATAGT; Reverse PCR primer: CCACCCCTATACTTTACAAAATATCCT; Sequencing Primer: TGAAAAAGTTTATAGAGTAGTGA. Eight samples failed quality control for a 97.5% completion rate; data from 308 samples were carried forward for validation analysis.

CpG RUNX3 (cg21368948)

Forward PCR primer: AAGGAAAGTAAGTTTTTGTTTGTATTTAAG; Reverse PCR primer: AACACTAAAAACCACCCACTCTACTC; Sequencing Primer: CCCACTCTACTCTATCA.5 samples failed quality control for a 98.4% completion rate; data from 311 samples were carried forward for validation analysis.

CpG MME (cg20228377)

Forward PCR primer: AGTGATTGGATTTGGGAGATT; Reverse PCR primer: CCAAAAATTAAACCACTACTCCAACCTACT; Sequencing Primer: TGTAAGTGGAGAAGTTTGAT. Six samples failed quality control for a 98% completion rate; data from 310 samples were carried forward for validation analysis.

CpG SERPINA1 (cg02181506)

Forward PCR primer: GTTTAGGATTTTGAGGGTTGTT; Reverse PCR primer: CCTCCACCCCAAATCTACTTCCTAAATAA; Sequencing Primer: ATTTTGAGGGTTGTTG. Nine samples failed quality control for a total 97% completion rate; data from 305 subjects carried forward for validation analysis.

CpG SERPINA1 (cg2462104)

Forward PCR primer: GTTTAGAGGATTATTAGAAATGATAGG; Reverse PCR primer: ACCTACCAATTATTAATACCAAATCTATAC; Sequencing Primer: GGATTATTAGAAATGATAGGTT. Eighteen samples failed quality control for a 94.5% completion rate; data from 304 samples were carried forward for validation analysis.

Statistics

We utilized several statistical models to identify the most robust associations with DNA methylation. Discovery analysis was performed using univariate linear regression to associate smoking-related phenotypes and CRP level with individual unadjusted methylation Beta levels, set as dependent variables. CpG sites associated with any smoking-related phenotype at p < 10⁻⁴ were processed for additional testing using linear mixed effect (LME) model analysis that accounted for the family structure of the AAT Genetic Modifiers study, non-parametric testing (i.e., Spearman’s rank correlation test for continuous independent variables or Mann-Whitney test for a binary ever-smoking status variable) and Box-Cox transformation of methylation Beta level. CRP level was also analyzed as a dependent variable using linear regression. This analysis used log₂-transformed CRP level and additional adjustments for gender, since log₂-transformed CRP level was lower in males as compared with females (p = 0.05). Furthermore, we checked whether any additional associations at p < 10⁻⁴ could be identified using normalized methylation Beta values. Additionally, we repeated all of the analyses for CpG sites set as dependent variables while adjusting for gender and age using a multivariate linear regression. Lastly, as performed in a study by Bibikova and colleagues, we calculated a mean from 1,411 individual methylation Beta values for every subject and used this variable as a proxy for global DNA methylation.¹⁸ We considered p < 3.54 × 10⁻⁵ as statistically significant according to Bonferroni method (i.e., assuming 1,411 tests per phenotype performed, 0.05/1,411). Since methylation levels within genes and CpG islands may be correlated, 1,411 is likely an overestimation of the number of independent tests performed. Therefore p = 6.52 × 10⁻⁵ was used as a suggestive significance threshold (i.e., assuming 767 tests, corresponding to the number of analyzed genes, performed, 0.05/767). p < 0.05 was considered as nominally significant.

Software

R was used for all statistical analyses.²³ QC procedures and normalization of methylation Beta levels were performed using the methylumi R package.²⁴ Box-Cox transformation of the methylation Beta levels was performed using the car R package for specific linear regression models.²⁵ LME models were assessed using the nlme R package.²⁶

Acknowledgments

The authors would like to thank all of the participants in the AAT Genetic Modifier Study for their enthusiastic support and participation. We also thank Ms. Toni M. Delorey for technical assistance. We would also like to thank Dr. Robert Crapo for his review and quality assurance of a subset of the spirometry data.

Disclosure of Potential Conflicts of Interest

No potential conflicts of interest were disclosed.

Funding

Alpha-1 Foundation Research Grant (DLD), Doris Duke Clinical Scientist Award (DLD), R01 HL 089438 (DLD), R01 HL68926 and HL105339; M. Siedlinski was supported during this work by a generous postdoctoral fellowship from the Niels Stensen Foundation.