Genetic Advances in Chronic Obstructive Pulmonary Disease. Insights from COPDGene

Abstract

Chronic obstructive pulmonary disease (COPD) is a common and progressive disease that is influenced by both genetic and environmental factors. For many years, knowledge of the genetic basis of COPD was limited to Mendelian syndromes, such as alpha-1 antitrypsin deficiency and cutis laxa, caused by rare genetic variants. Over the past decade, the proliferation of genome-wide association studies, the accessibility of whole-genome sequencing, and the development of novel methods for analyzing genetic variation data have led to a substantial increase in the understanding of genetic variants that play a role in COPD susceptibility and COPD-related phenotypes. COPDGene (Genetic Epidemiology of COPD), a multicenter, longitudinal study of over 10,000 current and former cigarette smokers, has been pivotal to these breakthroughs in understanding the genetic basis of COPD. To date, over 20 genetic loci have been convincingly associated with COPD affection status, with additional loci demonstrating association with COPD-related phenotypes such as emphysema, chronic bronchitis, and hypoxemia. In this review, we discuss the contributions of the COPDGene study to the discovery of these genetic associations as well as the ongoing genetic investigations of COPD subtypes, protein biomarkers, and post–genome-wide association study analysis.

Contents

COPDGene Study Overview

COPD GWASs

GWASs of COPD-related Phenotypes

 Genetic Association Studies of Lung Function Phenotypes

 Genetic Association Studies of Imaging Phenotypes

 Genetic Association Studies of Other COPD-related Phenotypes

Genetic Studies of COPD Subtypes

Applications of Genome Sequencing in COPDGene

Candidate Gene Studies: Alpha-1 Antitrypsin (SERPINA1)

Genetic Determinants of Protein Biomarkers

Additional Investigations after Genetic Association Analysis

Advances in Statistical Genetics

Conclusions and Future Directions

Chronic obstructive pulmonary disease (COPD) is a complex disease influenced by both environmental and genetic factors. Through research performed largely in the past decade, the genetic basis of COPD susceptibility is now beginning to be understood. Aside from rare genetic variants causing Mendelian syndromes, such as alpha-1 antitrypsin deficiency (caused by variants in SERPINA1 [serpin family A, member 1]) and cutis laxa (caused by mutations in ELN [elastin], FBLN5 [fibulin 5], and other genes), there were frustratingly few consistently replicated genetic variants associated with COPD reported during the era of candidate gene association studies from the 1970s to early 2000s (1). Soon after the COPDGene study (Genetic Epidemiology of COPD; www.COPDGene.org) was launched, two genome-wide association studies (GWASs) published in 2009 demonstrated significant associations with COPD. One study found associations with SNPs at the chromosome 15q25 CHRNA3 (cholinergic receptor nicotinic α3 subunit)/CHRNA5/IREB2 (iron-responsive element–binding protein 2) region in the GenKOLS study (Genetics of COPD, Norway) sample (a case–control study in Norway) that were also replicated using COPD cases in the NETT study (National Emphysema Treatment Trial) and control subjects from the NAS study (Normative Aging Study) (2). At the same time, another genome-wide significant locus near HHIP (hedgehog interacting protein) was shown to be associated with spirometric measures by Wilk and colleagues (3). In this review, we highlight genomic regions influencing COPD susceptibility and COPD-related phenotypes identified through collaborative studies involving COPDGene. We also discuss other study populations that have contributed to progress in COPD genetics (see Table E1 in the online supplement). Furthermore, we review contributions from the COPDGene study to the evolving understanding of COPD genetics based on studies of COPD subtypes, DNA sequencing, and proteomic data.

COPDGene Study Overview

The NHLBI-funded COPDGene study is one of the largest studies ever undertaken to investigate the genetic factors involved in COPD development. COPDGene participants were enrolled from 2008 to 2011; 5-year follow-up data collection was completed, and 10-year follow-up is now underway. The study includes 10,198 individuals with inclusion criteria of age 45 to 80 years old and at least a 10–pack-year smoking history. Individuals who self-identified as non-Hispanic white (NHW) or African American (AA) were enrolled. COPDGene has recruited one of the largest cohorts of AA current and former smoking subjects, enriched for COPD cases, comprising approximately one-third of the COPDGene study population. Significant findings within the AA population are highlighted throughout this review and summarized in Table E2. Additional details regarding the COPDGene study population are included in the online supplement.

COPD GWASs

GWASs detect associations between genetic variants and traits in study populations. To detect associations between variants and diseases such as COPD, a case–control design is often used. Genetic associations with greatest confidence are those that have met genome-wide statistical significance (typically with P < 5 × 10⁻⁸) in one study and then have been replicated in another study. Findings that meet genome-wide significance in one study but have not yet been replicated are intriguing, whereas findings below genome-wide statistical significance are of uncertain importance.

The first genetic association study using COPDGene subjects was based on GWAS of subjects from GenKOLS, NETT/NAS, and ECLIPSE (Evaluation of COPD Longitudinally to Identify Predictive Surrogate Endpoints), with the initial 1,006 subjects in COPDGene used to replicate the most statistically significant findings (4). This GWAS confirmed prior COPD-associated loci at 4q31 (HHIP) and 15q25 (CHRNA3/CHRNA5/IREB2) and found a new COPD locus at 4q22 near FAM13A (family with sequence similarity 13 member A), a locus contemporaneously identified as associated with quantitative measures of lung function (5, 6).

A follow-up case–control study used genome-wide SNP genotyping in the initial 1,006 COPDGene subjects for discovery, combining them with the previous cohorts from ECLIPSE, NAS, NETT, and GenKOLS. This study found and replicated a novel genome-wide significant association with COPD at 19q13 near RAB4B, EGLN2 (egl-9 family hypoxia-inducible factor 2), MIA, and CYP2A6 (cytochrome P450, family 2, subfamily A, member 6) (7) in a region previously identified as associated with smoking behavior (8, 9). Although this study and previous COPD GWASs had adjusted for pack-years of smoking, these results suggested that at least some of the genetic susceptibility to risk of COPD at this locus is mediated through genetic control of smoking behavior.

The contribution of genetic factors to COPD susceptibility had been demonstrated previously in family-based studies (10, 11). Genome-wide SNP genotyping data in COPDGene allowed the use of newer methods to estimate heritability. Zhou and colleagues applied these methods to demonstrate substantial heritability for risk to COPD (37.7% in NHW and 37.9% in AA), lung function (for FEV₁, 38.4% in NHW and 50.9% in AA), and emphysema (28.2% in NHW and 31.3% in AA) using SNP genotyping array data from COPDGene (12).

When genome-wide SNP genotyping with the Illumina OmniExpress chip was available in the entire COPDGene study population, Cho and colleagues combined the populations in ECLIPSE, NETT/NAS, GenKOLS, and COPDGene, achieving a larger sample (total of 6,633) of moderate-to-severe COPD cases and 5,704 current and former smoking control subjects (13). In addition to confirming the three previously identified COPD susceptibility loci (FAM13A on 4q22, CHRNA3/CHRNA5/IREB2 on 15q25, and HHIP on 4q31), this study also discovered one additional novel locus at 14q32 near RIN3 (Ras and Rab interactor 3). The effect estimates for the loci near HHIP and CHRNA3 became significantly stronger when the same set of control subjects was compared with cases with only severe COPD. Two new genome-wide significant loci in this analysis of patients with severe COPD compared with control subjects were found at 11q22 (near MMP3 [matrix metalloproteinase 3], MMP12, and MMP1) and at 1q41 (near TGFB2 [transforming growth factor-β2]).

Although genome-wide SNP genotyping arrays provided reasonable coverage across the genome, more thorough investigations of the coding exons throughout the genome became possible with the commercial development of exome genotyping arrays covering both common and rare nonsynonymous coding variants. Using a combined analysis of two family-based cohorts and three case–control studies (including COPDGene) totaling 12,165 subjects, investigators performed an exome-wide genetic association analysis (14) and found one novel exome-wide significant association with a SNP in IL27 in the chromosome 16p11.2 locus. Noncoding variants in linkage disequilibrium with the IL27 variant were previously associated with other diseases, including early-onset inflammatory bowel disease and type 1 diabetes mellitus (15, 16). Subsequent COPD association studies found noncoding variants with stronger association at the 16p11.2 locus, suggesting that a noncoding variant, and not the IL27 nonsynonymous variant, explains the 16p11.2 locus COPD association.

Recognizing that further advances in COPD genetics would require larger sample sizes and well-defined phenotypes, a group of international experts met to develop plans for future collaborative work (17). Their recommendations led to the development of the International COPD Genetics Consortium (ICGC; www.copdconsortium.org) to encourage use of collaborative study populations to identify COPD subtypes and search for their possible genetic basis.

COPDGene was a key member in the first genetic projects of the ICGC, including the ICGC GWAS (18) and a concurrently published UK Biobank/UK BiLEVE (UK Biobank Lung Exome Evaluation) study (19). The ICGC GWAS examined COPD cases (defined as prebronchodilator moderate-to-severe airflow limitation) and control subjects with normal spirometry (18). The ICGC investigators combined data across 26 cohorts totaling over 83,000 subjects in a two-stage analysis followed by a meta-analysis. Twenty-two genome-wide significant loci were identified (Figure 1), including nine loci reaching genome-wide significance in previous GWASs of COPD, an additional nine loci not previously described as associated with risk of COPD but identified as influencing spirometric measures of lung function (FEV₁ and/or FEV₁/FVC), and four novel loci not previously achieving genome-wide significance for risk of COPD or quantitative measures related to COPD (EEFSEC [eukaryotic elongation factor, selenocysteine-tRNA specific], DSP [desmoplakin], MTCL1 [microtubule crosslinking factor 1], and SFTPD [surfactant protein D]). This study also implicated variants overlapping with genetic determinants of pulmonary fibrosis (but with opposite directions of effect) and an overall risk of COPD shared with lung function and asthma. The UK Biobank/UK BiLEVE study used a staged analytical approach for FEV₁, FVC, and FEV₁/FVC in general population studies to ultimately identify 97 independent risk signals for lung function, including many that were also associated with COPD in COPDGene and other COPD cohorts (19). These loci included potential targets for therapeutics.

Figure 1.

Manhattan plot of chronic obstructive pulmonary disease genome-wide association study by the International COPD Genetics Consortium. The dashed line corresponds to the threshold for genome-wide statistical significance (P < 5 × 10⁻⁸). Reprinted by permission from Reference 18.

Thus, through multiple collaborative studies, more than 20 genetic loci have been convincingly associated with risk of COPD. These regions of the genome likely contain functional genetic variants related to COPD pathogenesis. Table 1 summarizes the genome-wide significant loci for COPD susceptibility identified through collaborations with COPDGene.

Table 1.

Genomic Regions Significantly Associated with Chronic Obstructive Pulmonary Disease Affection Status in Meta-analyses That Have Included COPDGene

Top SNPs	Odds Ratio (95% CI) or β-Coefficient (SE)*	P Value*	Reference
Chromosome 1q41: TGFB2
rs10429950	1.11 (1.07–1.14)	1.66 × 10−10	18
Chromosome 2q26: PID1
rs16825267	1.19 (1.12–1.25)	1.68 × 10−9	18
Chromosome 3p24: RARB
rs1529672	1.11 (1.07–1.15)	2.47 × 10−8	18
Chromosome 3q21: EEFSEC
rs2955083	1.18 (1.13–1.24)	4.16 × 10−13	18
Chromosome 4q22: FAM13A
rs1903003	0.78 (0.64–0.95)†	7.19 × 10−3†	4
rs7671167	0.77 (0.63–0.93)‡	3.39 × 10−3‡	4
rs7671167	0.73 (0.66–0.81)	2.22 × 10−9	7
rs1964516	0.73 (0.66–0.81)	1.88 × 10−9	7
rs4416442	1.28 (1.2–1.36)	1.12 × 10−14	13
rs6837671	1.12 (1.09–1.15)	7.48 × 10−15	18
Chromosome 4q24: GTSCD
rs11727735	1.26 (1.18–1.33)	3.84 × 10−14	18
Chromosome 4q24: TET2
rs2047409	1.12 (1.08–1.15)	2.46 × 10−13	18
Chromosome 4q31: HHIP
rs13118928	0.77 (0.70–0.85)	9.3 × 10−8	4
rs13118928	0.76 (0.65–0.89)	5.97 × 10−7	7
rs13141641	0.76 (0.65–0.88)	2.56 × 10−7	7
rs13141641	1.27 (1.19–1.36)	1.57 × 10−12	13
rs145506456	1.22 (1.16–1.27)	9.1 × 10−41	18
Chromosome 5q32: HTR4
rs7733088	1.18 (1.14–1.21)	5.33 × 10−26	18
Chromosome 5q33: ADAM19
rs113897301	1.16 (1.12–1.21)	1.58 × 10−13	18
Chromosome 6p24: DSP
rs2076295	1.09 (1.06–1.12)	3.97 × 10−9	18
Chromosome 6p21: AGER
rs2070600	1.24 (1.16–1.32)	5.94 × 10−10	18
Chromosome 6q21: ARMC2
rs2806356	1.12 (1.08–1.16)	8.34 × 10−10	18
Chromosome 6q24: ADGRG6
rs9399401	1.15 (1.12–1.19)	1.81 × 10−19	18
Chromosome 10q22: SFTPD
rs721917	1.08 (1.05–1.11)	2.49 × 10−8	18
Chromosome 14q32: RIN3
rs754388	1.28 (1.18–1.39)	5.25 × 10−9	13
rs754388	1.15 (1.11–1.20)	4.96 × 10−14	18
Chromosome 15q23: THSD4
rs1441358	1.13 (1.10–1.16)	8.22 × 10−16	18
Chromosome 15q25: CHRNA3/5 and IREB2
rs8034191	1.29 (1.16–1.41)	8.32 × 10−8	4
rs13180	0.77 (0.70–0.85)	2.01 × 10−8	4
rs12914385	1.28 (1.2–1.36)	6.38 × 10−14	13
rs16969968	0.27 (0.04)	1.7 × 10−14	14
rs754388	1.18 (1.15–1.22)	1.77 × 10−28	18
Chromosome 16p11: IL27, CCDC101
rs181206§	−0.16 (0.04)	4.7 × 10−6	14
rs17707300	1.10 (1.06–1.13)	6.75 × 10−10	18
Chromosome 16q23: CFDP1
rs7186831	1.12 (1.08–1.16)	1.12 × 10−11	18
Chromosome 18p11: MTCL1
rs647097	1.10 (1.06–1.13)	6.14 × 10−9	18
Chromosome 19q13: RAB4B/EGLN2/CYP2A6
rs7937	0.73 (0.63–0.83)	2.88 × 10−9	7
rs2604894	0.74 (0.65–0.84)	3.41 × 10−8	7
rs12459249	1.10 (1.06–1.14)	3.42 × 10−8	18

Definition of abbreviations: CI = confidence interval; COPDGene = Genetic Epidemiology of Chronic Obstructive Pulmonary Disease.

Genes located near top SNPs in each region are shown, but they may not be the genes influenced by functional variants in that region; odds ratios and confidence intervals were taken from the cited references.

^*For meta-analysis.

^†Odds ratio and confidence interval for COPDGene non-Hispanic whites only; genome-wide meta-analysis P value = 9.47 × 10⁻¹¹.

^‡Odds ratio and confidence interval for COPDGene non-Hispanic whites only; genome-wide meta-analysis P value = 1.22 × 10⁻¹¹.

^§This study was an exome-wide study, so this SNP was considered significant, given this study design.

GWASs of COPD-related Phenotypes

GWASs using COPDGene subjects and other cohorts have identified multiple genetic loci associated with COPD susceptibility and severity. However, owing in part to the complex and heterogeneous nature of COPD, effect sizes of individual genetic loci remain modest. Thus, COPDGene investigators have sought to determine the heritable factors leading to clinical manifestations of COPD by examining the genetic associations with lung function phenotypes, imaging phenotypes, and other COPD-related phenotypes (summarized in Table E3).

Genetic Association Studies of Lung Function Phenotypes

Because spirometric lung function tests measuring FEV₁ and FEV₁/FVC are used to diagnose and determine the severity of COPD, investigators have reasoned that genetic loci associated with these phenotypes could also be associated with COPD susceptibility. Population-based studies that have been conducted before and concurrently with COPDGene have to date discovered nearly 100 different loci associated with FEV₁ and FEV₁/FVC ratio, beginning with the discovery of HHIP by Wilk and colleagues (3) and followed by the CHARGE (Cohorts for Heart and Aging Research in Genomic Epidemiology) and SpiroMeta consortiums and the UK BiLEVE and UK Biobank studies (5, 6, 19–21). Among the first studies to demonstrate the relevance of these measures in COPD (and COPDGene) was by Castaldi and colleagues, who hypothesized that other loci associated with FEV₁ and FEV₁/FVC were likely to be associated with COPD susceptibility (22). They tested 32 genome-wide significant SNPs in 17 genes for association with COPD status in four COPD case–control study samples (NETT/NAS, the GenKOLS case–control study from Norway, ECLIPSE, and the first 1,000 subjects in COPDGene), totaling 3,456 cases and 1,906 control subjects. Three of these loci were confirmed to contain SNPs associated with risk for COPD: 4q24 (FLJ20184/INTS12 [integrator complex subunit 12]/GSTCD [glutathione S-transferase C-terminal domain containing]/NPNT [nephronectin]), 6p21 (AGER [advanced glycosylation endproduct specific receptor]/PPT2 [palmitoyl-protein thioesterase 2]), and 5q33 (ADAM19 [ADAM metallopeptidase domain 19]). Subsequent larger studies by Wain and colleagues (19) and Hobbs and colleagues (18) reinforced the overlap between these lung function loci and COPD.

Examining genetic determinants of lung function specifically within COPD cohorts, Lutz and colleagues performed a GWAS of post-bronchodilator pulmonary function (FEV₁ and FEV₁/FVC) (23). NHW and AA COPDGene subjects were analyzed separately. In the NHW individuals, these investigators found that FEV₁ and FEV₁/FVC ratio were significantly associated with SNPs in the 15q25 region, near CHRNA3, CHRNA5, AGPHD1 (aminoglycoside phosphotransferase domain–containing protein 1), IREB2, and CHRNB4 (cholinergic receptor nicotinic β4 subunit), and that FEV₁/FVC was associated with HHIP, replicating previously seen associations when analyzing COPD affection status as a categorical phenotype. Meta-analysis of COPDGene NHW subjects combined with ECLIPSE and GenKOLS subjects revealed that most of the genome-wide significant results for FEV₁ and FEV₁/FVC were at the 15q25 locus, chromosome 4q22 near FAM13A, and chromosome 4q31 near HHIP. In addition, genome-wide significant associations were identified for both FEV₁ and FEV₁/FVC with regions on chromosomes 1 (near TGFB2) and 19 (near CYP2A6). Genome-wide significant associations were found for FEV₁/FVC on chromosomes 11 (near MMP3, MMP12, and MMP1) and 14 (near RIN3), as well as for FEV₁ on chromosome 9 (near DBH). Analysis of COPDGene AA subjects for FEV₁ and FEV₁/FVC did not find any genome-wide significant associations using GWAS, but admixture mapping was subsequently applied in COPDGene AA subjects (24). African ancestry was associated with lower FEV₁/FVC values, and a region of suggestive association to FEV₁/FVC was identified in FAM19A2.

Genetic Association Studies of Imaging Phenotypes

Variable combinations of emphysema and small airway disease leading to chronic airflow obstruction contribute to the phenotypic heterogeneity of COPD. Several methods of computed tomography (CT) imaging quantification have been developed to assess both pathologic emphysema patterns and functional small airway disease, and they have been applied in genetic studies outside of COPDGene. For example, using these imaging parameters in a population-based GWAS, Manichaikul and colleagues identified several genetic determinants associated with percentage emphysema (SNRPF [small nuclear ribonucleoprotein polypeptide F], PPT2, MAN1C1 [mannosidase α class 1C member 1]) and (upper lobe emphysema)/(lower lobe emphysema) ratio (MAN2B1 [mannosidase α class 2B member 1], DHX15 [DEAH-box helicase 15], as well as MGAT5B [α-1,6-mannosylglycoprotein 6-β-N-acetylglucosaminyltransferase B]) (25). In individuals with COPD (from the ECLIPSE, NETT, and GenKOLS cohorts), Kong and colleagues discovered a SNP in BICD1 (BICD cargo adaptor 1) that was associated with the presence or absence of emphysema determined by radiologist visual assessments (26).

Densitometric measures of emphysema provide objective and quantitative assessments of emphysema severity. In a total sample size of 12,031 subjects from COPDGene, ECLIPSE, NETT, and GenKOLS, Cho and colleagues performed a GWAS on two quantitative emphysema phenotypes (percentage of lung density histogram below −950 Hounsfield units [HU] and 15th percentile point of the lung density histogram), two quantitative airway imaging phenotypes (wall area percentage and square root of airway wall area of 10 mm internal perimeter airway), and the combined emphysema/small airway phenotype of expiratory gas trapping in smokers with and without COPD (27). Five different loci were associated with emphysema phenotypes, including the previously reported HHIP, CHRNA3, and AGER loci, as well as novel associations near DLC1 (deleted in liver cancer 1) and SERPINA10 (serpin family A member 10). The SERPINA10 association is located near the SERPINA1 gene and appears to be driven largely by the low-frequency SERPINA1 Z allele, demonstrating that even relatively rare genetic variants such as alpha-1 antitrypsin deficiency can be confirmed using large samples of subjects with disease-specific phenotypes. Two loci were associated with gas trapping on expiratory CT scans, near AGER on chromosome 6p21 and LINC00310 (long intergenic non–protein-coding RNA 310)/KCNE2 (potassium voltage-gated channel subfamily E regulatory subunit 2) on chromosome 21q22.

Local histogram–based emphysema quantification is a texture-based method that captures emphysema patterns and is more strongly associated with COPD-related physiological and functional measures than densitometric measures (28). Castaldi and colleagues sought to identify genetic determinants associated with distinct patterns of emphysema on chest CT scans within the COPDGene study (29). They performed a GWAS using 9,614 total COPDGene subjects (6,456 NHW and 3,158 AA) for five emphysema patterns. Two novel genome-wide significant associations were identified for SNPs in MYO1D (myosin ID) and VMA8 with severe centrilobular and panlobular local histogram–based emphysema patterns, respectively. Five other genome-wide significant associations were found for loci previously associated with COPD susceptibility and quantitative measures of lung function (HHIP, IREB2/CHRNA3, CYP2A6/ADCK [aarF domain–containing kinase], TGFB2, and MMP12).

Boueiz and colleagues examined a total of 11,532 subjects from COPDGene, ECLIPSE, and GenKOLS in a GWAS to determine the genetic determinants of the anatomic distribution of emphysema (30). Two loci previously reported to be associated with COPD susceptibility (specifically, 4q31 near HHIP and 15q25 near CHRNA5) were associated with both measures of emphysema distribution used in this study: the difference and ratio of upper-third to lower-third percentage emphysema at −950 HU. The COPD-associated variants in these two regions were associated with greater amounts of upper lobe emphysema. For the emphysema ratio phenotype, three novel associations were reported, near SOWAHB (sosondowah ankyrin repeat domain family member B), TRAPPC9 (trafficking protein particle complex 9), and KIAA1462.

Halper-Stromberg and colleagues hypothesized that visual assessment of CT images may provide independent information for genetic association studies compared with the semiautomated CT imaging measures of emphysema and airway disease described above (31). They performed a GWAS for visual assessment of CT images in 1,540 NHW COPDGene subjects to determine the genetic determinants contributing to the presence and severity of radiographic features representing airway wall thickness and emphysema. This study identified a genome-wide significant association with visual emphysema at the 15q25 region (within IREB2) but found no significant associations for visual airway wall thickness. Furthermore, these authors identified several previously identified COPD susceptibility loci that were also nominally associated with either an emphysema or airway phenotype. To demonstrate how visual measures can provide independent information, these associations were adjusted for quantitative densitometric assessment of emphysema.

Genetic Association Studies of Other COPD-related Phenotypes

In addition to spirometry and CT measures of emphysema and airway disease, COPDGene investigators have sought to identify the genetic determinants of multiple other disease-related phenotypes contributing to the extensive heterogeneity of COPD. Before COPDGene, Pillai and colleagues demonstrated associations of known COPD GWAS loci with several COPD-related phenotypes using ECLIPSE study subjects, including spirometry (CHRNA3/CHRNA5, HHIP, FAM13A), smoking (CHRNA3/CHRNA5), body mass index (BMI) (FAM13A), fat-free mass index (HHIP), emphysema (CHRNA3/CHRNA5, HHIP), COPD exacerbations (HHIP), and BODE score (BMI, airflow obstruction, dyspnea, and exercise capacity; a metric of risk of death in patients with COPD) (CHRNA3/CHRNA5) (32). A meta-analysis of 2,950 individuals with COPD from ECLIPSE, GenKOLS, and NETT conducted by Wan and colleagues demonstrated an association of the well-known FTO (FTO α-ketoglutarate dependent dioxygenase) locus with measures of cachexia, BMI, and fat-free mass index (33). Replication of the most significantly associated SNP with BMI was achieved using 502 genotyped subjects from COPDGene with Global Initiative for Chronic Obstructive Lung Disease (GOLD) spirometry grade 2 or higher (moderate, severe, and very severe) COPD.

McDonald and colleagues performed a GWAS for hypoxemia, as measured by pulse oximetry, in subjects from COPDGene (2,810 NHW and 820 AA) and ECLIPSE (1,758 total) (34). Although no SNPs reached genome-wide significance in the GWAS of NHW subjects, several suggestive loci were identified. In the GWAS of AA COPDGene subjects, two regions were found to be significantly associated with resting oxygen saturation: an intergenic region on chromosome 14 bounded by FOXG1 (forkhead box G1) and LINC00645 and a region on chromosome 15 encompassing TICRR (DNA topoisomerase II binding protein 1 interacting checkpoint and replication regulator) and KIF7 (kinesin family member 7).

In previous studies, chronic bronchitis, defined as symptoms of chronic cough and chronic phlegm production, has been associated with accelerated lung function decline and an increase in risk of respiratory infections (see additional citations in Kim and colleagues [35, 36]). Using subjects from COPDGene, GenKOLS, and ECLIPSE, Lee and colleagues investigated the genetic variants associated with chronic bronchitis in COPD subjects relative to current and former smokers with normal spirometry or subjects with COPD without chronic bronchitis (37). Comparing COPD subjects with chronic bronchitis with current or former cigarette smokers without COPD (“smoking control subjects”), they identified a novel genome-wide significant locus on chromosome 11p15.5 near the genes EFCAB4A (EF-hand calcium-binding domain–containing protein 4A), CHID1 (chitinase domain–containing protein 1), and AP2A2 (adaptor-related protein complex 2 subunit α2). The four most significant SNPs were located in the FAM13A region. Comparing subjects with COPD who had chronic bronchitis with subjects with COPD who did not have chronic bronchitis, they identified a novel locus yielding suggestive evidence of association on chromosome 1q23 containing genes RPL31P11 (ribosomal protein L31 pseudogene 11) and ATF6 (activating transcription factor 6). Because chronic bronchitis was present in some of the smoking control subjects, further association analysis of chronic bronchitis in COPD was completed using smoking control subjects without chronic bronchitis, resulting in slightly reduced statistical significance of the top SNPs. Together, these analyses assessed the combined outcome of COPD with chronic bronchitis; genome-wide significant genetic determinants of chronic bronchitis independent of COPD status have not been reported.

Enlargement of the pulmonary arteries was found to be associated with increased COPD exacerbation frequency in COPDGene (38). To determine the genetic determinants of pulmonary arterial hypertension associated with advanced COPD, Lee and colleagues performed a GWAS of pulmonary artery enlargement (PAE) in subjects with COPD (39). This study included 1,266 subjects with COPD with PAE, 3,755 subjects with COPD without PAE, and 4,461 current and former smokers with normal spirometry from both the COPDGene and ECLIPSE cohorts. Relative to subjects with COPD without PAE, the GWAS of subjects with PAE showed significantly associated SNPs on chromosomes 15q25 (within IREB2) and 14q31 (near GALC [galactosylceramidase]). Relative to smokers with normal spirometry, the GWAS of subjects with PAE demonstrated genome-wide significant association for markers in the 15q25 region containing CHRNA5, PSMA4 (proteasome subunit α4), IREB2, AGPHD1, and CHRNA3, in which all SNPs were either identical to or in strong linkage disequilibrium with SNPs previously reported as associated with risk for COPD. These results emphasize the complexity of the chromosome 15q25 region in controlling risk for COPD, where multiple regional genes could influence COPD-related phenotypes.

History of childhood pneumonia was shown to be associated with increased risk of COPD in COPDGene subjects (40). In COPDGene, Hayden and colleagues hypothesized that genetic variants would be associated with development of pneumonia during childhood and over the course of the lifetime, which may contribute to reduced lung function and risk for COPD (41). Although there were no genome-wide significant findings, several loci with suggestive associations were identified for childhood pneumonia (in NGR1, PAK6 [p21 {RAC1} activated kinase 6], and near MATN1 [matrilin 1]) and for lifetime pneumonia (in LOC339862, RAPGEF2 [Rap guanine nucleotide exchange factor 2], PHACTR1 [phosphatase and actin regulator 1], and PRR27 [proline-rich 27] and near MCPH1 [microcephalin 1]).

Because cigarette smoking is the principal environmental risk factor for developing COPD, genetic factors associated with smoking behavior and nicotine dependence may contribute to COPD susceptibility and severity. Previous GWASs of smoking behavior have identified significant associations, including the 15q25 (CHRNA5–CHRNA3–CHRNB4) and 8p11 (CHRNB3–CHRNA6) loci, with nicotine dependence (8, 9, 42, 43). As noted above, the 15q25 locus was one of the first COPD GWAS loci identified (2). Kim and colleagues confirmed the association of 15q25 SNPs with nicotine dependence in COPDGene subjects and demonstrated association of CHRNA3/CHRNA5 genetic variants with severity of emphysema in former smokers (44). Hancock and colleagues conducted a GWAS meta-analysis including 17,074 subjects from COPDGene, deCODE, EAGLE (Environment and Genetics in Lung Cancer Etiology study), COGEND (Collaborative Genetic Study of Nicotine Dependence), and SAGE (Study of Addiction: Genetics and Environment) to identify novel loci associated with nicotine dependence in ever smokers (45). In addition to confirming the associations with 15q25 and 8p11, the authors identified a novel genome-wide significant association with a locus on chromosome 20q13, within the CHRNA4 gene.

In addition to COPD susceptibility, genetic determinants could also influence response to COPD treatment. To determine whether genetic variants contributed to the acute response to inhaled β₂-agonists, Hardin and colleagues performed a GWAS of quantitative bronchodilator responsiveness using 5,789 current or former smokers with COPD from the COPDGene (both NHW and AA), ECLIPSE, GenKOLS, and NETT studies (46). Among NHW COPDGene subjects, a variant located upstream of the gene EPHA7 (EPH receptor A7) was significantly associated with the absolute difference in pre- versus post-bronchodilator FEV₁ levels. Among AA COPDGene subjects, two low-frequency variants in the gene CDH13 were significantly associated with the absolute volume change with bronchodilator or the difference in pre- versus post-bronchodilator change expressed as a percentage of the predicted FEV₁. A rare SNP in GOLGA8B (golgin A8 family member B) was associated with bronchodilator response expressed as change from baseline FEV₁. The reliability of these associations to low-frequency variants in a relatively small study population is uncertain. In addition, a more common SNP in SGCD (sarcoglycan delta; minor allele frequency, 0.07) was significantly associated with the difference in pre- versus post-bronchodilator FEV₁ as a percentage of baseline FEV₁. In the meta-analysis of all five study populations, several SNPs upstream from the gene KCNJ2 (potassium voltage-gated channel subfamily J member 2) were found to be associated with several different bronchodilator responsiveness metrics, but none achieved genome-wide significance. Although most genetic association analyses of COPD-related phenotypes in COPDGene have focused on SNPs, Begum and colleagues assessed association of estimated larger structural variants (also known as copy number variants) with COPD-related phenotypes and found suggestive associations of a structural variant on chromosome 5q35 with TLC (from chest CT scans) among AA (47) and on chromosome 3p26 with smoking behavior among AA (48).

Thus, studies conducted by COPDGene investigators have identified numerous genetic associations with lung function, lung imaging phenotypes, and COPD-related clinical phenotypes. It is notable that a considerable number of genetic loci associated with COPD susceptibility are also associated with one or more of these additional COPD-related phenotypes. As reported by Hobbs and colleagues (18), nearly half of COPD susceptibility loci are also associated with measures of lung function in general population samples. Two COPD susceptibility loci (CYP2A6 and MMP1/3/12) exhibited overlap with local histogram imaging phenotypes and lung function. Furthermore, four COPD susceptibility loci (TGFB2, FAM13A, HHIP, and CHRNA3/CHRNA5/IREB2) overlapped with multiple imaging, lung function, and COPD-related phenotypes, demonstrating their likely importance to COPD etiology. These loci also represent the most frequently reported and replicated signals associated with COPD. However, other genetic associations, such as the DLC1 locus with quantitative emphysema and TRAPPC9 with emphysema distribution, have not been associated with other COPD-related phenotypes; these unique associations could also point to specific biological mechanisms relevant for COPD heterogeneity. A summary of the associations of COPD-related phenotypes with genome-wide significant COPD susceptibility loci is provided in Figure 2.

Figure 2.

Summary phenogram plot of genome-wide association studies for chronic obstructive pulmonary disease (COPD) susceptibility and COPD-related phenotypes reviewed in this article. Individual SNP associations have been collapsed into the nearest gene or locus where genes are densely clustered. Although multiple SNP associations may be reported in an individual locus for a given phenotype, only one association is depicted. COPD susceptibility associations (blue) include all 22 loci reported by Hobbs and colleagues (18). Lung function (FEV₁, FVC, and FEV₁/FVC) associations depicted here (red) include associations from the comprehensive study conducted by Wain and colleagues (19). BDR = bronchodilator response; BMI = body mass index; CB = chronic bronchitis; EMPH.DIST.RATIO = emphysema distribution ratio of upper divided by lower lung fields; FFMI = fat-free mass index; LAA.950 = percentage of lung density histogram below −950 Hounsfield units; LHE = local histogram–based emphysema; MOD = moderate centrilobular emphysema on local histogram–based emphysema analysis; NORM = normal/nonemphysematous on local histogram–based emphysema analysis; OS = resting oxygen saturation; PAE = pulmonary artery enlargement; PAN = panlobular emphysema on local histogram–based emphysema analysis; PB = post-bronchodilator; PCT.GAS = percent gas trapping at −856 Hounsfield units on expiratory computed tomographic scan; PERC15 = 15th percentile point of the lung density histogram; SEV = severe centrilobular emphysema on local histogram–based emphysema analysis; VIS.EMPH = visual emphysema (presence/absence).

Genetic Studies of COPD Subtypes

Because COPD is a heterogeneous clinical syndrome, COPDGene has aimed to define clinical subtypes of COPD to enable better elucidation of the pathophysiologic mechanisms of the disease. The clinical subtypes examined to date for genetic associations are preserved ratio impaired spirometry (PRISm), mild COPD (GOLD spirometry grade 1), and asthma–COPD overlap. In addition, genetic determinants of COPD subtypes determined by unsupervised clustering approaches have been assessed.

PRISm is defined as a reduced FEV₁ with a preserved FEV₁/FVC ratio. Subjects with PRISm have previously been shown to have increased respiratory symptoms, decreased exercise capacity, more difficulty with activities of daily living, and increased mortality (49, 50). The prevalence of PRISm among all COPDGene subjects was estimated to be 12.3%, and these patients showed increased dyspnea, reduced 6-minute walk distance, increased percentage emphysema, decreased TLC, and increased segmental bronchial wall area percentage. Cluster analysis suggested that a subset of PRISm subjects may have COPD (50). However, in a GWAS of PRISm, there were no common genetic variants significantly associated with this clinical subtype at a genome-wide level. Of interest, a nominally significant association of PRISm with Klinefelter’s syndrome (47XXY) was observed.

Because FEV₁/FVC normally declines with advancing age, the GOLD 1 spirometry group (FEV₁ ≥80% predicted with FEV₁/FVC <0.70) includes subjects with mild COPD as well as healthy subjects with reduced FEV₁/FVC due to advanced age. Lee and colleagues performed a GWAS of GOLD 1 subjects in COPDGene (51). Although no genome-wide significant associations were identified, the COPD GWAS region near FAM13A was nominally associated with GOLD 1 status. K-means cluster analysis was used to dissect GOLD 1 subjects into four groups, which showed nominally significant evidence for differential associations with previously identified COPD GWAS SNPs.

In the COPDGene cohort, 13% of the patients with COPD also reported a history of asthma. Compared with the population of patients with COPD alone, the subjects with COPD and asthma were younger, had lower lifetime smoking intensity, had worse disease-related quality of life, were more likely to have had a COPD exacerbation in the past year, and were more likely to report frequent exacerbations (52). A GWAS in the NHW group of patients with COPD–asthma overlap identified significant SNPs in CSMD1 (CUB and Sushi multiple domains 1) on chromosome 8p23 and SOX5 (SRY-box 5) on chromosome 12p12. A meta-analysis of the NHW and AA subjects identified a region containing GPR65 (G protein–coupled receptor 65) on chromosome 14q31 that was associated with the COPD–asthma overlap phenotype.

COPDGene has also assessed sex-specific genetic influences on COPD. Hardin and colleagues identified SNP-by-sex interactions for SNPs in the CELSR1 (cadherin EGF LAG seven-pass G-type receptor 1) gene (53). In a sex-stratified association analysis, a SNP in CELSR1 approached genome-wide significance in females but was not even nominally associated with COPD in males. Further research to identify sex-specific determinants of COPD and COPD-related phenotypes may provide insight into the variable susceptibility to COPD among men and women.

In addition to clinically defined subtypes, machine learning approaches have also been used to classify COPD into distinct subtypes. Castaldi and colleagues used k-means cluster analysis to group COPDGene subjects into four clusters that showed distinct characteristics: relatively resistant smokers, mild upper zone–predominant emphysema, airway-predominant disease, and severe emphysema (54). Among NHW from COPDGene, they found evidence of association (although not genome-wide significance) with a SNP near the HHIP gene for the mild upper zone–predominant emphysema group, and within the severe emphysema group, there was also evidence for association near HHIP and in the chromosome 15q25 locus containing CHRNA3, CHRNA5, and IREB2.

Applications of Genome Sequencing in COPDGene

Although genome-wide SNP panels can provide reasonable coverage of common genetic variants, whole-exome sequencing (WES) provides comprehensive coverage of both rare and common variants in the protein-coding portion of the genome (encompassing ∼1% of the total genome), whereas whole-genome sequencing provides similar coverage across essentially the entire genome. To focus on the extremes of COPD susceptibility and resistance, the WES studies in COPDGene have focused on highly resistant smokers and severe COPD cases. Qiao and colleagues used WES in COPDGene early-onset COPD cases and resistant smoking control subjects to attempt to replicate genes segregating with COPD in families from the Boston Early-Onset COPD study (55). Although nominally significant replication was observed for variants in DNAH8 (dynein axonemal heavy chain 8), ALCAM (activated leukocyte cell adhesion molecule), RARS (arginyl-tRNA synthetase), and GBF1 (Golgi brefeldin A–resistant guanine nucleotide exchange factor 1), rare coding genetic determinants of COPD other than SERPINA1 appear to be quite heterogeneous. Using WES from COPDGene, Stanley and colleagues identified rare functional variants in the TERT (telomerase reverse transcriptase) gene in severe COPD cases, demonstrating the importance of this gene previously reported to be associated with risk for pulmonary fibrosis (56). The entire COPDGene study population is undergoing whole-genome sequencing through the TOPMed (Trans-Omics for Precision Medicine) project, which will enhance opportunities for future identification of rare variants influencing COPD susceptibility.

Candidate Gene Studies: Alpha-1 Antitrypsin (SERPINA1)

Although earlier candidate gene association studies had focused on biologically plausible candidates, selection of candidate genes from Mendelian syndromes or large-scale GWAS gives a greater likelihood of finding valid associations. Alpha-1 antitrypsin deficiency, caused primarily by homozygosity for the Z allele of the SERPINA1 gene, is the most well-established genetic cause of COPD. In addition to candidate studies of variants identified through lung function GWAS noted above (22), COPDGene has also been important for studying Z allele heterozygotes (PI MZ). In COPDGene, NHW heterozygous Pi MZ patients had significantly lower FEV₁ and FEV₁/FVC, as well as significantly more radiographic emphysema, than subjects not carrying SERPINA1 Z alleles (57). Findings were similar in the substantially smaller AA PI MZ group of COPDGene. A larger collaborative study in severe COPD cases versus control subjects found a genome-wide significant association in SERPINA1, adding further support to the elevated risk for heterozygote carriers of this variant (58). These findings, combined with efforts identifying this locus in emphysema GWAS (27), highlight the contribution of COPDGene to the increasing evidence that PI MZ smokers are at significantly increased risk for COPD (59).

Genetic Determinants of Protein Biomarkers

One of the major strengths of COPDGene is its abundance of proteomic data. These data have allowed COPDGene investigators to explore the relationships between genetic variation and blood biomarker concentrations using quantitative trait locus (QTL) analysis. In COPDGene and SPIROMICS (Subpopulations and Intermediate Outcome Measures in COPD Study), genome-wide SNP data were combined with proteomic data to identify hundreds of novel SNPs that were associated with blood protein concentrations of quantitative trait loci (pQTLs) (60) (Figure 3). These pQTLs are distinct from expression quantitative trait loci (eQTLs) identified from the transcriptome of RNA in an appropriate tissue sample. SNPs shown to be associated at genome-wide levels with disease status become more biologically interesting if they are also associated with RNA and/or protein expression within cells of relevant tissues (because it suggests the SNP may be directly functional and not just a “tagging” marker correlated with some unobserved causal SNP). pQTLs are enriched for missense mutations, which alter the amino acid sequence of the resulting protein. Although many of the strongest pQTL SNPs were within or near the gene encoding their respective proteins, several genetic “hot spots” were found where pQTLs for multiple proteins were detected. For instance, the ABO locus on chromosome 9, which encodes a glycosyltransferase, was a significant pQTL for six different proteins, all of which were coded by genes on different chromosomes.

Figure 3.

Many novel genome-wide associations between SNPs and blood protein biomarkers were found. Combined Manhattan plots show SNPs associated with blood protein levels quantitative trait loci (pQTL) by chromosomal location for 38 biomarkers with at least one SNP significant at genome-wide significance after adjustment for multiple testing (red line). The −log₁₀ P values are shown using results from a meta-analysis of both SPIROMICS and COPDGene SNPs. More than 300 novel pQTLs were identified. COPD = chronic obstructive pulmonary disease; COPDGene = Genetic Epidemiology of COPD; SPIROMICS = Subpopulations and Intermediate Outcome Measures in COPD Study. Reprinted by permission from Reference 60.

Additional Investigations after Genetic Association Analysis

The discovery of genomic regions associated with COPD and COPD-related phenotypes needs to be followed by efforts to identify the functional variant(s) and gene(s) in these regions and mechanistic studies to determine how those variants and the genes that they influence impact COPD susceptibility and heterogeneity. A substantial amount of work has leveraged the genetic findings in COPDGene to help identify enriched cell and tissue types as well as potentially causal variants and genes (18). Genes identified with support from both COPD GWAS association evidence in COPDGene and murine emphysema models include FAM13A (61), HHIP (62), IREB2 (63), MMP12 (64), MMP1 (65), SFTPD (66), and AGER (67).

Although the impact of each common variant consistently and significantly associated with risk for COPD is modest, when combined into a genetic risk score reflecting the combined impact of multiple genetic risk factors, there is the potential for more clinically important findings. For example, Busch and colleagues showed that a genetic risk score based on seven COPD susceptibility SNPs conferred a 1.9% reduction in FEV₁ percentage predicted for each risk allele in an independent study population (International COPD Genetics Network) (58). An expanded risk score using 97 variants associated with quantitative measures of lung function from COPD case–control studies (including COPDGene) was highly significant, with a 1-SD difference in risk score resulting in an odds ratio of 1.36 for COPD (19). Although these cumulative risk scores could be useful in identifying high-risk groups for clinical trials, their relevance for individual subjects remains uncertain.

Because complex diseases such as COPD are likely influenced by a network of interacting genes and proteins, efforts to identify the biologically relevant networks for COPD susceptibility have been pursued. McDonald and colleagues built a series of disease network modules within the protein–protein interaction network based on genetic association evidence in COPDGene and GenKOLS using the dmGWAS software (https://bioinfo.uth.edu/dmGWAS/); they identified genes that were involved in several novel biological pathways, including the IL-7 pathway, as influencing risk for COPD (68).

Advances in Statistical Genetics

The statistical analysis of genetic data for COPD studies is challenging because of the complex genetic architecture of COPD (i.e., multiple genes contributing to multiple correlated phenotypes) and the inclusion of skewed, nonnormally distributed traits such as lung function and/or cigarette smoking intensity. To deal with these challenges, multiple statistical genetic analysis approaches were developed and applied to COPDGene datasets.

To incorporate information from genetic and phenotypic sources, several statistical approaches were developed to jointly use this information. The multimarker association test by Qiao and colleagues exploited the polygenic architecture of COPD by testing multiple SNPs jointly for association with the outcome of interest (69). Zhou and colleagues developed an approach combining multiple phenotypes into a single-variant association test (70). Lutz and colleagues developed an approach and corresponding R package to test for pleiotropy (i.e., a SNP that is jointly associated with multiple phenotypes) (71). In addition, new methods to identify genetic markers related to imaging features have been developed by Batmanghelich and colleagues (72, 73).

For COPDGene genetic analysis, methods have also been developed for the adjustment of genetic heterogeneity and population substructure in GWAS and whole-genome sequencing studies. Prokopenko and colleagues developed new approaches based on the Jaccard index that are particularly well suited to detecting local/recent substructure captured in rare variant data from whole genome sequencing (74). Fier and colleagues proposed clustering approaches for the region-based analysis of rare variant data (75). Llinares-López and colleagues developed a new method to discover genetic heterogeneity and applied this approach to the COPDGene study (76).

Statistical methods for genetic association studies were also developed to account for issues relating to challenging phenotypic distributions and ascertainment bias. Lutz and colleagues derived the nonparametric population-based association test approach for the association of a SNP with the outcome of interest that is robust against phenotypic misspecifications (77). This method is implemented as an interactive software package for genetic association analyses in population-based studies using generalized estimating equations (sites.google.com/site/genenpbat/home/npbat). Lutz and colleagues also developed an approach to adjust for ascertainment bias when testing for genetic associations of secondary phenotypes in case–control studies (78).

Mediation analysis is an important tool for discriminating between direct and indirect effects of genetic markers on risk for COPD. To examine how a SNP affects an outcome of interest through a mediator, Siedlinski and colleagues used mediation analysis to examine how SNPs on chromosome 15q25 act on FEV₁ through cigarette smoking (79). Because mediation analysis assumes no unmeasured confounding of the exposure–mediator–outcome relationship and most sensitivity analyses focus on measured confounding of the mediator–outcome relationship (80), Lutz and colleagues developed the R package Umediation to analyze how results of mediation analysis would change in the presence of unmeasured confounding of the exposure–mediator–outcome relationship (81).

The COPDGene study, as a large, well-phenotyped and genotyped resource, has been important for other studies requiring adult control subjects and for method development by external investigators. For example, the recent upgrade of the popular software package plink2 uses a highly efficient algorithm for logistic regression developed using COPDGene data (82).

Conclusions and Future Directions

Substantial progress in understanding of the underlying genes influencing risk for COPD has been made in the past decade, and COPDGene has played a major role in these advances. Although understanding of COPD genetics remains incomplete, collaborative studies involving COPDGene and other large, well-phenotyped study populations (e.g., ECLIPSE and SPIROMICS) will be essential to provide greater insight into the genetic architecture of COPD. For example, after this review article was drafted, a larger collaborative GWAS using the ICGC and UK Biobank identified 82 genome-wide significant loci for COPD (83). In addition, studies that include younger subjects than those in COPDGene will be necessary to differentiate the impact of genetic determinants on reduced maximally attained lung function versus accelerated lung function decline on the development of COPD. Furthermore, ongoing research using novel machine learning and deep learning methods to define clinically relevant subtypes of COPD may ultimately allow for targeted therapy and prognostication of clinical course. With the addition of longitudinal phenotypes and whole-genome sequencing, especially when combined with proteomic, transcriptomic, epigenetic, and metabolomic data, COPDGene will likely provide important new information regarding the biological pathways controlling COPD risk, phenotypic heterogeneity, and disease progression that may allow more accurate diagnosis and prognosis, as well as new targets for treatment intervention.