Epigenome-wide association study on asthma and chronic obstructive pulmonary disease overlap reveals aberrant DNA methylations related to clinical phenotypes

We hypothesized that epigenetics is a link between smoking/allergen exposures and the development of Asthma and chronic obstructive pulmonary disease (ACO). A total of 75 of 228 COPD patients were identified as ACO, which was independently associated with increased exacerbations. Microarray analysis identified 404 differentially methylated loci (DML) in ACO patients, and 6575 DML in those with rapid lung function decline in a discovery cohort. In the validation cohort, ACO patients had hypermethylated PDE9A (+ 30,088)/ZNF323 (− 296), and hypomethylated SEPT8 (− 47) genes as compared with either pure COPD patients or healthy non-smokers. Hypermethylated TIGIT (− 173) gene and hypomethylated CYSLTR1 (+ 348)/CCDC88C (+ 125,722)/ADORA2B (+ 1339) were associated with severe airflow limitation, while hypomethylated IFRD1 (− 515) gene with frequent exacerbation in all the COPD patients. Hypermethylated ZNF323 (− 296) / MPV17L (+ 194) and hypomethylated PTPRN2 (+ 10,000) genes were associated with rapid lung function decline. In vitro cigarette smoke extract and ovalbumin concurrent exposure resulted in specific DNA methylation changes of the MPV17L / ZNF323 genes, while 5-aza-2′-deoxycytidine treatment reversed promoter hypermethylation-mediated MPV17L under-expression accompanied with reduced apoptosis and decreased generation of reactive oxygen species. Aberrant DNA methylations may constitute a determinant for ACO, and provide a biomarker of airflow limitation, exacerbation, and lung function decline.

A large proportion (15-30%) of patients with chronic airways disease has features of both asthma and chronic obstructive pulmonary disease (COPD) (Asthma and COPD Overlap, ACO). ACO patients experience more frequent exacerbations, have poorer quality of life, decline in lung function more rapidly, and consume a disproportionate amount of healthcare resources than asthma or COPD alone 1,2 . Up to date, no universal definition criteria exist [3][4][5][6][7] , and few studies have investigated the pathogenesis of this overlap syndrome 8,9 . Only 25% of life-long smokers develop COPD, and asthma exhibit a strong familial connection, suggesting genetic determinants in susceptibility to both COPD and asthma 10,11 . In the past three decades, extensive research to identify genetic determinants of COPD and asthma has shown that only a few single nucleotide polymorphisms (SNP) are independently and consistently associated with fixed airflow limitation 12,13 . Epigenetics, which refers to the process of influencing gene expression through other genetic mechanisms without affecting DNA sequences, may accounts for this discrepancy. DNA methylation occurring at position 5 of the pyrimidine ring of cytosines in the context of the cytosine followed by guanine dinucleotide sequence (CpG) form the basis of epigenetic mechanisms through inhibiting the binding of transcription factors at the promoter regions or influencing transcriptional elongation and alternative splicing at the intragenetic regions. Gene promoter methylation often leads to transcriptional repression of the gene, whereas gene body methylation is frequently associated with high gene expression levels 14,15 . DNA methylation patterns are not only inheritable but also susceptible to change in response to environmental stimuli, such as smoking and allergens 16 . Additionally, SNPs in non-coding regions may simultaneously alter both the consensus sequence and its DNA methylation, if they alter or generate CpG dinucleotides 17 . Recent candidate gene and epigenome-wide association studies (EWAS) have identified several CpG site-specific aberrant DNA methylation changes associated with COPD and asthma individually [18][19][20][21] , but none has been performed in this overlap group 22 . We hypothesized that gene-specific CpG methylation profiles of peripheral blood mononuclear cells (PBMCs) may contribute to disease susceptibility, severity, and clinical phenotypes in ACO patients, with the goal of identifying novel epigenetic changes related to frequent exacerbation, rapid lung function decline, or severe airflow limitation.
Whole-genome DNA methylation profiles and enrichment pathway analysis in the discovery cohort. Twelve ACO patients and 6 healthy non-smokers (HS) enrolled in the discovery cohort were matched in terms of age, BMI, and Charlson co-morbidity index (Supplementary Table S1). A total of 21 PBMC samples www.nature.com/scientificreports/ The top-ranking pathways enriched in comparison I included apoptosis and survival of APRIL and BAFF signaling (Fig. 1C), immune response of NF-AT signaling in leukocyte interactions ( Supplementary Fig. S1), and development role of HDAC and CaMK in control of skeletal myogenesis (Supplementary Table S4). The topranking pathways enriched in comparison II included NF-AT signaling, regulation of epithelial-to-mesenchymal transition, and TGF/WNT signaling for cytoskeletal remodeling (Supplementary Table S5). Differential PDE9A, SEPT8, and ZNF323 gene methylations with respect to the presence of ACO in the validation cohort. The 22 ACO patients, 48 pure COPD patients, and 10 HS enrolled in the validation cohort were matched in terms of age, BMI, and Charlson co-morbidity index (Supplementary Table S6).
PDE9A gene (+ 30,088, Fig. 2A) was hypermethylated in ACO patients versus pure COPD patients or HS, and negatively correlated with post-BD FEV1%predicted (Fig. 2B).  The significantly hypermethylated genes were highlighted with a red-colored barometric bar, while hypomethylated genes in a blue-colored bar. The changes represent the differences between the mean β-values of normal and ACO patients. For example, the mean β-value for normal and ACO patients is 1.44 and 1.94, respectively, for calcineurin A (CACNA1C), indicating a higher methylation level (+ 0.5) in ACO. The image was created by the Metacore software. www.nature.com/scientificreports/ were all positively correlated with post-BD FEV1%predicted, while TIGIT gene methylation (− 173, Fig. 3D) was negatively correlated with post-BD FEV1%predicted. IFRD1 gene methylation (− 515, Fig. 3E) was decreased in all COPD patients with frequent severe AE versus those without frequent severe AE, and negatively correlated with exacerbation frequency (Fig. 3F).
Differential ZNF323, MPV17L, and PTPRN2 gene methylations with respect to rapid lung function decline in the validation cohort. DNA methylation levels over 7 CpG sites of 7 selected genes from comparison II were measured in 5 ACO patients and 8 pure COPD patients with rapid lung function decline after 1-year follow-up (FEV1%predicted 69.78 ± 12.15 versus 60.72 ± 12.18%, mean difference 9.06 ± 7.54%, p = 0.002), as well as in 12 ACO patients and 9 pure COPD patients without rapid lung function decline. ZNF323 (− 296, Fig. 3G) and MPV17L gene methylations (+ 194, Fig. 3H) were both elevated after 1-year follow-up (visit 2) versus at enrollment (visit 1) in those with rapid lung function decline, but remained the same in those without rapid lung function decline, while MPV17L gene methylation at visit 1 and visit 2 (Fig. 3I) were both negatively correlated with the difference in FEV1% predicted values between visit 2 and visit 1. PTPRN2 gene methylation (+ 10,000, Fig. 3J) were reduced after 1-year follow-up in 16 patients with frequent moderate to severe AE, and remained the same in those without frequent moderate to severe exacerbation.
Effects of in vitro concurrent cigarette smoke extract (CSE) and ovalbumin (OVA) stimuli on DNA methylation levels or gene expressions of the 10 candidate genes. ZNF323 gene methylation (− 264) was increased in response to CSE plus OVA treatment (p < 0.05, Fig. 4A). PTPRN2 gene methylation (+ 10,000) was decreased with OVA stimuli (p < 0.05, Fig. 4B). MPV17L gene (+ 194) methylation was increased in response to CSE plus OVA treatment (p values < 0.05, Fig. 4C). Pre-treatment with de-methylation agent (5-aza-2ʹ-deoxycytidine; 5-aza) in the presence of CSE plus OVA stimuli resulted in decreased methylation over − 113 CpG site of the MPV17L gene, increased MPV17L gene expression, reduced reactive oxygen species production, reduced late apoptosis, and increased cell viability, as compared with CSE plus OVA treatment alone (all p values < 0.05, Fig. 4D-H), whereas DNA methylation levels of the other 8 candidate genes were not altered despite increased gene expressions (Supplementary Fig. S2).

Discussion
T helper type 2 immune gene signals associated with greater BD reversibility, eosinophilia, and better response to inhaled corticosteroids, have been identified in ACO patients 8,24 . Although one report is available on sputum DNA methylation changes of the PCDH20 and SULF2 genes in relation to ACO, and one genome-wide association study has identified SNPs in the CSMD1, SOX5 and GPR65 genes for ACO 25,26 , this is the first study to perform a whole genome DNA methylation analysis in ACO with replication of the principal findings, and identify a specific association of ACO and several clinical phenotypes with aberrant DNA methylation patterns.
Aberrant methylation patterns of the PDE9A, SEPT8, and ZNF323 genes showed the most significant associations with ACO. PDE9A is the most selective for cGMP degradation, and its inhibitor can enhance memory function through promoting synaptic plasticity and counter pathological remodeling of the heart [27][28][29][30] . Given that PDE-4 is responsible for metabolizing adenosine 3′,5′-cyclic monophosphate that reduces the activation of a wide range of inflammatory cells including eosinophils, targeting PDE9A signaling pathway may be a novel strategy for managing ACO 31 . SEPT8 contributes to kidney and liver fibrosis, and modulates the generation of toxic amyloid-beta peptides in Alzheimer's disease 32,33 . Given that SEPT8 functions in various biological processes of cell cytokinesis and migration, it may be another novel target for ACO 34 . Interestingly, using the BiosQTL database (https ://genen etwor k.nl/biosq tlbro wser/), there is evidence that the effect of the SEPT8 CpG site (cg13334727) methylation on whole blood may be genetically regulated by the cis-methylation quantitative trait loci (meQTL), rs39855. This SNP is associated with several respiratory phenotypes in the United Kingdom Biobank (UKBB) such as asthma, hay fever or allergic rhinitis, or it is in linkage disequilibrium with the lead variants associated with asthma and respiratory diseases based on the information from the OpenTarget Genetics database (https ://genet ics.opent arget s.org/varia nt). Our in vitro experiments showed increased gene expressions but no significant changes in the methylation levels for 8 of the 11 selected genes, suggesting that the altered Table 2. Top differentially methylated loci in the comparison between patients with asthma and COPD overlap (ACO) and healthy non-smokers (comparison I) in the discovery cohort. UTR un-translated region; TSS transcription start site.    www.nature.com/scientificreports/ methylation patterns in patients may be inheriting epigenotypes rather than changes after cigarette smoke exposures. Several SNPs of the ZNF323 (ZSCAN31) gene are associated with lung function in asthmatic patients 35,36 , while its hypermethylation was noted both in ACO patients and in response to CSE plus OVA stimuli, suggesting a role of acquired ZNF323 hypermethylation in the pathogenesis of ACO. Aberrant methylation patterns of the TIGIT, CYSLTR1, CCDC88C, and ADORA2B genes were associated with severe airflow limitation. TIGIT can enhance Th2 response 37 , and shows aberrant methylation in allergic asthma 38 . Interestingly, the probe (cg19421218) annotated to TIGIT is described as an expression quantitative trait methylation (eQTM) for the same gene in the BiosQTL database. CysLTR1 mediates bronchoconstriction and eosinophil migration in asthma 39,40 . CCDC88C genetic mutation is implicated in eosinophilia-associated myeloid/lymphoid neoplasms 41 . ADORA2B contributes to pulmonary fibrosis and pulmonary hypertension associated with COPD [42][43][44] . Based on the information from the BiosQTL and OpenTarget Genetics databases, DNA methylation of the ADORA2B CpG site (cg07563400) may be genetically regulated by two Cis-meQTL, rs3925260 and rs12452624, which have been linked to lung function parameters and asthma, respectively. Our results suggest that these epigenotypes related to allergic responses may be important determinates of lung function in early life and predispose individuals to COPD.

Column ID
Aberrant methylation patterns of the MPV17L, ZNF323, and PTPRN2 genes associated with rapid lung function decline were identified, while altered methylations of the ZNF323 and IFRD1 with frequent exacerbation. MPV17L protects against mitochondrial oxidative stress and apoptosis by activation of Omi/HtrA2 protease, and its methylation may act as a biomarker for the prognosis of lung adenocarcinoma [45][46][47][48] . Our in vitro experiments showed that de-methylation agent could reverse promoter hypermethylation-mediated under-expression of the MPV17L gene and oxidative stress-mediated cell apoptosis in response to concurrent CSE and OVA stimuli, supporting the use of this epigenetic mark as potential therapeutic targets of ACO. Aberrant PTPRN2 gene methylation has been identified in smoking-related COPD patients in two previous EWAS 49,50 . IFRD1 regulates the pathogenesis of asthma and cystic fibrosis through mediating neutrophil function 51,52 .
There are several limitations in the present study. First, the sample size of the discovery cohort is relatively small and the analyses of both the DMLs and enriched pathways would have not met a more stringent threshold for significance of false discovery rate < 0.5 or 0.1. Three fifths of the results in the comparison I would not remain significant after applying a more stringent threshold of q-value of 0.1 to control the rate of false positives, but all the results in the comparison II remain significant. There would be 162 DMLs with 118 hypomethylated and 44 hypermethylated in the comparison I, if q value is less than 0.1. However, we used a 2-tiered approach to verify some of the findings. Moreover, testing for differentially methylated regions where the cumulative effect of probes included in the same genomic region will be conducted in the future. Second, the cause and effect relationship could not be straightforward determined in this association study. CSE and OVA co-exposure resulted in under-expression of the MPV17L gene and hypermethylation of its promoter region, which were in accordance with the findings in the clinical samples and fit the scientific consensus of an anti-correlation between promoter methylation and gene expression. Third, peripheral blood mononuclear cells are comprised of a mixture of cell types, which may contribute to different methylation changes. Single-cell multi-omic strategies will enhance the specificity and sensitivity of the analysis of DNA methylation patterns over both CpG and non-CpG sites, and site-specific methylome editing will serve as a key technique for the study of 5-methylcytosine function, although both were still un-mature 5 years ago when the current study started. The percentage of monocytes, T cells and B cells in the PBMC samples was not determined, so we could not make corrections for cell-type composition in the best way 53 . However, the results open the possibility of using de-methylation in the treatment of ACO.

Conclusions
We reported a novel association of ACO in adults of Asian origin with aberrant DNA methylation in the promoter or body regions of the PDE9A, SEPT8, and ZNF323 genes. The findings extend reports linking hypomethylated CYSLTR1/ADORA2B/CCDC88C and hypermethylated TIGIT with more severe fixed airflow limitation in COPD patients, identifying hypomethylated IFRD1/hypermethylated ZNF323 as biomarkers of frequent exacerbation, and providing direct evidence that perturbation of MPV17L signaling through epigenetic programming may play a role in the mediation of both inflammatory and allergic responses in ACO. Our findings provide a new direction for this disease and might establish novel biological insights into the development and effective treatment of ACO.

Materials and methods
This study was approved by the Institutional Review Board of Chang Gung Memorial Hospital, Taiwan (certificate number: 103-3366B). The study participants were recruited from the pulmonary clinics of Kaohsiung Chang Gung Memorial Hospital from October 2014 to July 2017. All the participants were Taiwanese Han people in ancestry. Written informed consent was obtained from each subject participating in the study. The enrollment and exclusion criteria for COPD, and the definition of acute exacerbation (AE) were in accordance with GOLD guideline (Supplementary-Appendix 1 Text). ACO was defined by the presence of three elements: (I) COPD diagnosis, (II) positive bronchodilator (BD) test, and (III) blood eosinophil > 3%, or history of atopic diseases, including asthma, allergic rhinitis, or atopic dermatitis. A total of 364 subjects were screened, and 228 COPD patients were enrolled for final analysis. The discovery cohort used for the EWAS microarray experiment included 12 ACO patients and 6 healthy non-smokers (HS) with normal lung function. The non-overlapping validation cohort included 22 ACO patients, 48 patients with pure COPD, and 10 HS. The experiment, enrolment and exclusion criteria for COPD, and the definition of acute exacerbation (AE) were in accordance with GOLD guideline. www.nature.com/scientificreports/ out 485 CpG sites that have bead count smaller than 3 in 5% of total samples and filter out 901 CpG sites that have detection p-value greater than 0.01 in 5% of the total samples with R package wateRmelon 54 . We then transfer the methylation β value into M values 55 which has better statistical properties for latter non-parametric statistical analysis. For those differentially methylated CpG sites, their corresponding gene symbols were used for pathway analysis and interaction networks contruction by MetaCore software (Thomson Reuters Incorporation, Philadelphia, USA). The significance threshold was a p < 0.0005 and a false discovery rate (q) < 0.3. All methylation datasets have been deposited in the NCBI Gene Expression Omnibus with the accession number GSE118468. Significantly differentially methylated CpG sites with at least a 10% difference in their β value (large effect size) and known biological or functional relevance were selected for further verification and validation by pyrosequencing method using PyroMark Q24 1.010 (Qiagen) 56 (Supplementary Appendix 1 Text).

DNA methylation measurement and analysis.
In vitro human monocytic THP-1 cell culture under the stimuli with cigarette CSE and OVA allergen. THP-1 immortalized monocyte-like cell lines are derived from the peripheral blood of a childhood case of acute monocytic leukemia (M5 subtype) and represent valuable tools for investigating circulatory monocytes, which are one of the main sources of inflammatory cytokines in response to allergens and smoking exposures 57 . DNA Methyl-Transferase (DNMT) 3A and 3B mediate de novo deposition of C5-methylcytosine to establish methylation marks in CpG sites, while 5-aza is a chemical nucleoside analog of cytidine, which can incorporate into DNA, and trap DNMTs through an irreversible covalent interaction 58 . Thus, THP-1 and 5-aza were adopted in the in vitro experiments. THP-1 were treated with normal medium, 100 ng/ml lipopolysaccharide (LPS), 2.5% CSE, 25 ug OVA, or CSE (2.5%) plus OVA (25 ug) mix for 48 h, and also treated with 1uM 5-aza (Sigma-Aldrich Corp) in advance for 24 h. Gene expressions were measured by quantitative real-time reverse transcription-PCR method with Taqman probe and specific primers (Supplementary Table S7). Relative expression levels were calculated using the ∆∆Ct method. (Supplementary Appendix 1 Text).
Statistical analysis. Data were expressed as the mean ± standard deviation. One-way analysis of variance were be used for comparing mean values of more than two experimental groups. Categorical variables were analyzed using Chi-square test. Stepwise multivariate linear regression analysis was used to adjust for confounding factors and obtain adjusted p values. The differences of continuous variables between study enrollment and follow-up after one year were analyzed by paired t-test. Pearson's correlation was used to determine the relationship between selected variables. A p-value of less than 0.05 is considered statistically significant.

Ethics approval and consent to participate. This study was approved by the Institutional Review
Board of Chang Gung Memorial Hospital, Taiwan (certificate number: 103-3366B). Written informed consent was obtained from each subject participating in the study.

Data availability
All methylation datasets have been deposited in the NCBI Gene Expression Omnibus with the accession number GSE118468.