Abstract

Background

   The causes of poor respiratory function and COPD are incompletely
   understood, but it is clear that genes and the environment play a role.
   As DNA methylation is under both genetic and environmental control, we
   hypothesised that investigation of differential methylation associated
   with these phenotypes would permit mechanistic insights, and improve
   prediction of COPD. We investigated genome-wide differential DNA
   methylation patterns using the recently released 850 K Illumina EPIC
   array. This is the largest single population, whole-genome epigenetic
   study to date.

Methods

   Epigenome-wide association studies (EWASs) of respiratory function and
   COPD were performed in peripheral blood samples from the Generation
   Scotland: Scottish Family Health Study (GS:SFHS) cohort (n = 3781; 274
   COPD cases and 2919 controls). In independent COPD incidence data
   (n = 149), significantly differentially methylated sites (DMSs;
   p < 3.6 × 10^−8) were evaluated for their added predictive power when
   added to a model including clinical variables, age, sex, height and
   smoking history using receiver operating characteristic analysis. The
   Lothian Birth Cohort 1936 (LBC1936) was used to replicate association
   (n = 895) and prediction (n = 178) results.

Findings

   We identified 28 respiratory function and/or COPD associated DMSs,
   which mapped to genes involved in alternative splicing, JAK-STAT
   signalling, and axon guidance. In prediction analyses, we observed
   significant improvement in discrimination between COPD cases and
   controls (p < .05) in independent GS:SFHS (p = .016) and LBC1936
   (p = .010) datasets by adding DMSs to a clinical model.

Interpretation

   Identification of novel DMSs has provided insight into the molecular
   mechanisms regulating respiratory function and aided prediction of COPD
   risk. Further studies are needed to assess the causality and clinical
   utility of identified associations.

Fund

   Wellcome Trust Strategic Award 10436/Z/14/Z.
     __________________________________________________________________

Research in context.

Evidence before this study

   We searched for articles in PubMed published in English up to July 25,
   2018, with the search terms “DNA methylation” and “respiratory
   function”, or “COPD”. We found some evidence for association between
   differential DNA methylation and both respiratory function and COPD. Of
   the twelve previous studies identified, eight used peripheral blood
   samples (sample size [N] range = 100–1,085) and four used lung tissue
   samples (N range = 24–160). The number of CpG loci analysed range from
   27,578 to 485,512. These studies have not identified consistent changes
   in methylation, most likely due to a combination of factors including
   small sample sizes, technical issues, phenotypic definitions, and study
   design. In addition, no previous study has: analysed a sample from a
   large single cohort; used the recently released Illumina EPIC array
   (which assesses ~850,000 CpG loci); adjusted both methylation data and
   phenotype for smoking history, or used both prevalent and incident COPD
   electronic health record data.

Added value of this study

   To our knowledge, this is the largest single cohort epigenome-wide
   association study (EWAS) of respiratory function and COPD to date
   (n = 3,781). After applying stringent genome-wide significance criteria
   (p < 3.6 × 10^−8), we found that DNA methylation levels at 28 CpG sites
   in peripheral blood were associated with respiratory function or COPD.
   Of these 28, seven were testable in an independent population sample:
   all seven showed consistent direction of effect between the two samples
   and three showed replication (p < .007 [0.05/7 CpG sites tested]). Our
   results suggest that adjustment of both the phenotypic and the DNA
   methylation probe data for smoking history, which has not been carried
   out in previous studies, reduces the confounding effects of smoking,
   identifies larger numbers of associations, and reduces the
   heterogeneity of effects across smoking strata. We used gene set
   enrichment and pathway analyses, together with an approach that
   combines DNA methylation results with gene expression data to provide
   evidence for enrichment of differentially methylated sites in genes
   linked to alternative splicing, and JAK-STAT signalling and axon
   guidance. Finally, we demonstrated that the inclusion of DNA
   methylation data improves COPD risk prediction over established
   clinical variables alone in two independent datasets.

Implications of all the available evidence

   There is now accumulating evidence that DNA methylation in peripheral
   blood is associated with respiratory function and COPD.

   Our study has shown that DNA methylation levels at 28 CpG sites are
   robustly associated with respiratory function and COPD, provide
   mechanistic insights, and can improve prediction of COPD risk. Further
   studies are warranted to improve understanding of the aetiology of
   COPD, explore causality and to assess the utility of DNA methylation
   profiling in the clinical management of this condition.

   Alt-text: Unlabelled Box

1. Introduction

   Respiratory function is influenced by both environmental factors and
   genetic factors, with heritability estimates ranging from 39 to 66%
   [[51]1,[52]2]. Epigenetic modifications are at the interface of
   genetics and the environment. DNA methylation, the covalent binding of
   a methyl group to the 5′ carbon of cytosine-phosphate-guanine (CpG)
   dinucleotide sequences in the genome, is an epigenetic modification of
   DNA that is associated with gene expression. Epigenome-wide association
   studies (EWASs) have the potential to provide mechanistic insights into
   impaired respiratory function and COPD pathogenesis. Previous EWASs of
   spirometric measures of respiratory function and respiratory disease
   have however produced inconsistent results, with some identifying
   significant associations [[53][3], [54][4], [55][5], [56][6]], and
   others not [[57][7], [58][8], [59][9]]. Moreover, there has been little
   consistency between the positive findings reported [[60]9,[61]10].
   Studies of lung tissue [[62]5,[63]8] have been constrained by sample
   availability, with the largest study to date comprising 160 subjects
   [[64]5]. Inconsistency amongst the results of the peripheral
   blood-based studies [[65]3,[66]6,[67]7,[68]9,[69]11] is likely to be
   due to a number of factors, including small sample size (e.g., two
   studies had <200 samples) [[70]6,[71]11] and/or investigation of a
   relatively small number (~27,000) of CpG loci [[72]3,[73]11]. The study
   with the largest number of samples (n = 1085) analysed only 27,000 CpG
   loci, while the largest study using the 450 K array (the predecessor to
   the array used here) analysed 920 samples [[74]12]. Differences in
   spirometric measures, definitions of COPD, study population
   characteristics and study design, in particular in the method used to
   adjust for smoking history, are also likely to be important sources of
   variation [[75]9,[76]10]. Smoking is established as a major risk factor
   for COPD [[77]13], and previous genome-wide DNA methylation have
   focused on DNA methylation associated with smoking and COPD
   [[78]5,[79]14,[80]15]. However, not all smokers develop COPD and >25%
   of COPD cases occur in never smokers [[81]16]. Results from a growing
   number of studies suggested that impaired respiratory function and COPD
   are strongly associated with risk factors other than smoking [[82][17],
   [83][18], [84][19]], and have a strong genetic component [[85][20],
   [86][21], [87][22]] that generally acts independently of smoking
   [[88]23]. To understand the pathological mechanisms of impaired
   respiratory function and COPD other than smoking we sought to identify
   robust associations by assessing methylation in a large single cohort
   sample, applying a more rigorous correction for smoking history and by
   performing sensitivity analyses. In contrast to prior studies, we used
   the recently released Illumina EPIC array, which interrogates over
   850,000 methylation sites. All 3781 individuals in our sample were from
   a single cohort with extensive and consistent phenotyping comprising
   clinical investigation, questionnaire, and linkage to routine medical
   health records. The cross-sectional design of prior studies has limited
   their capacity to distinguish cause and effect [[89]10]. To identity
   predictive biomarkers of COPD, and to provide insights into the causal
   nature of our findings we tested our findings for their predictive
   power. We used an independent subpopulation of 150 participants with
   incident COPD who were disease free at the time of blood sampling.
   Finally, where data were available, we attempted to replicate our EWAS
   and prediction findings in an independent cohort, LBC1936, drawn from
   the same population.

2. Material and methods

   A flow chart showing the overall study design is outlined in [90]Fig.
   1, and full description of the methods is provided in the appendix.

Fig. 1.

   [91]Fig. 1
   [92]Open in a new tab

   Flow-chart showing the analysis pipeline. Direction of the arrows
   represents the workflow of the study design with performed analysis
   indicated. Lemon and blue boxes represent the in the discovery
   Generation Scotland: Scottish Family health study cohort and the
   replication Lothian Birth Cohort of 1936 (LBC1936) data sets
   respectively. The grey box indicates input data of COPD case-control
   differential expression in lung tissue. The green boxes indicate the
   analyses undertaken. The black arrows and gold boxes indicate output of
   significant results. (For interpretation of the references to colour in