Exome Array Analysis Identifies a Common Variant in IL27 Associated with Chronic Obstructive Pulmonary Disease

Abstract

Rationale: Chronic obstructive pulmonary disease (COPD) susceptibility is in part related to genetic variants. Most genetic studies have been focused on genome-wide common variants without a specific focus on coding variants, but common and rare coding variants may also affect COPD susceptibility.

Objectives: To identify coding variants associated with COPD.

Methods: We tested nonsynonymous, splice, and stop variants derived from the Illumina HumanExome array for association with COPD in five study populations enriched for COPD. We evaluated single variants with a minor allele frequency greater than 0.5% using logistic regression. Results were combined using a fixed effects meta-analysis. We replicated novel single-variant associations in three additional COPD cohorts.

Measurements and Main Results: We included 6,004 control subjects and 6,161 COPD cases across five cohorts for analysis. Our top result was rs16969968 (P = 1.7 × 10⁻¹⁴) in CHRNA5, a locus previously associated with COPD susceptibility and nicotine dependence. Additional top results were found in AGER, MMP3, and SERPINA1. A nonsynonymous variant, rs181206, in IL27 (P = 4.7 × 10⁻⁶) was just below the level of exome-wide significance but attained exome-wide significance (P = 5.7 × 10⁻⁸) when combined with results from other cohorts. Gene expression datasets revealed an association of rs181206 and the surrounding locus with expression of multiple genes; several were differentially expressed in COPD lung tissue, including TUFM.

Conclusions: In an exome array analysis of COPD, we identified nonsynonymous variants at previously described loci and a novel exome-wide significant variant in IL27. This variant is at a locus previously described in genome-wide associations with diabetes, inflammatory bowel disease, and obesity and appears to affect genes potentially related to COPD pathogenesis.

At a Glance Commentary

Scientific Knowledge on the Subject

Chronic obstructive pulmonary disease (COPD) is a complex disease, with susceptibility modified in part by genetic variants. Genome-wide association studies have identified genetic variants that impact COPD susceptibility; however, the majority of COPD heritability remains unexplained. Genome-wide association studies have not been optimized to capture coding genetic variants that affect COPD susceptibility.

What This Study Adds to the Field

In our analysis of the protein-coding region of the genome (the exome), we identified an IL27 variant associated with COPD susceptibility. This variant is highly correlated with genetic variants associated with diabetes, inflammatory bowel disease, and obesity, and it also regulates expression of several genes in lung tissue and in blood, including TUFM.

Chronic obstructive pulmonary disease (COPD) is a highly prevalent condition that is projected to be the third leading cause of death worldwide by 2020 (1). COPD is a complex disease whose development depends on both environmental and genetic risk factors. The genetic component of COPD is demonstrated by studies showing strong familial aggregation. Furthermore, several genome-wide association studies (GWASs) have implicated COPD risk loci at chromosomal regions 15q25 and 19q13, and near genes HHIP, FAM13A, RIN3, MMP12, and TGFB2 (2–6). Although multiple COPD susceptibility loci have been identified through GWASs, known COPD risk loci explain only a small fraction of the observed variability in risk (7). Therefore, a large portion of the heritability of COPD has yet to be explained.

Analyzing coding variation may reveal novel pathobiology contributing to the development of COPD. While the majority of genome-wide association variants in complex diseases are likely regulatory, coding variants, accounting for only 1% of the genome by size, are overrepresented among these associations (8). In addition, uncommon (minor allele frequency [MAF], 1–5%) and rare (MAF < 1%) genetic variants, not well captured by GWAS arrays, are one of several possible causes of “missing heritability” (9). Rare coding variants are important in COPD susceptibility, as illustrated by alpha-1 antitrypsin deficiency (AATD), a genetic disorder in which rare variants in a serine protease inhibitor (SERPINA1) greatly impact COPD susceptibility (10). Traditional GWAS genotyping arrays have a large portion of genetic markers outside the coding genome, so many GWAS associations in complex disease have yet to be functionally classified. Restricting analysis to the coding regions of the genome (the exome) could enable more direct biological and functional interpretation of association study results. Exome genotyping arrays have been developed to specifically query genetic variation in the coding genome. These exome arrays contain approximately 250,000 putatively functional (nonsynonymous, stop, or splice) single-nucleotide polymorphism (SNP) probes (11). In complex disease phenotypes such as insulin secretion, type 2 diabetes risk, blood lipid levels, and fasting glucose levels, this exome array technology has already been employed to add to working knowledge of these phenotypes (12–15). We hypothesized that an exome genotyping analysis would identify coding variants showing an association with COPD susceptibility.

Some of the results of this study were previously reported in the form of an abstract (16).

Methods

Study Cohorts

The primary analysis included individuals from two family-based cohorts identified through a proband with COPD (the BECOPD [Boston Early-Onset COPD] study and the ICGN [International COPD Genetics Network] study), as well as three case–control studies (the TCGS [Transcontinental COPD Genetics Study] in Poland and in Korea, and the COPDGene [Genetic Epidemiology of COPD] study, which included non-Hispanic white [NHW] and African American [AA] individuals). Institutional review board approval and written informed consent were obtained for all of these cohorts. Persons were labeled unaffected with COPD if they had an FEV₁/FVC ratio greater than or equal to 0.7 and an FEV₁ greater than or equal to 80% of the predicted value. Moderate to severe COPD was defined by an FEV₁/FVC ratio less than 0.7 and an FEV₁ less than 80% of the predicted value, and severe COPD was defined by an FEV₁/FVC ratio less than 0.7 and an FEV₁ less than 50% of the predicted value.

Single-Variant Testing for Association with COPD

All individuals were genotyped using the Illumina HumanExome arrays (v1.1 and v1.2; Illumina, San Diego, CA). Single-variant testing was performed on nonsynonymous, stop, and splice-site exome array variants with cohort-based MAF greater than 0.5%. Analyses were adjusted for age, sex, pack-years of smoking, and ancestry-based principal components. Association analysis for the TCGS-Korea, TCGS-Poland, COPDGene AA, and COPDGene NHW cohorts were performed using an additive genetic model through logistic regression in PLINK v1.9 (17, 18). To match the analytical methods for gene-based association testing (see later), the BEOCOPD and ICGN family-based cohorts were analyzed using logistic regression via generalized estimating equations implemented in GWAF version 2.2 (19, 20). Results were combined in a fixed effects meta-analysis using METAL (version 2011-03-25) (21, 22). Top coding variants (P < 1 × 10⁻⁵) not previously implicated in COPD susceptibility were further assessed with previously reported genotyped and imputed GWAS data derived from the ECLIPSE (Evaluation of COPD Longitudinally to Identify Predictive Surrogate Endpoints), GenKOLS (Genetics of Chronic Obstructive Pulmonary Disease, Norway), and NETT/NAS (National Emphysema Treatment Trial and Normative Aging Study) studies using the same analysis and covariates (with the exception of NETT/NAS, in which all control subjects are male) (2, 4).

Gene-based Testing for Association with COPD

Statistical tests that combine multiple variants in a gene may be more powerful than tests of individual variants, particularly when the variants are individually rare. Thus, in each analysis cohort (BEOCOPD and ICGN, TCGS-Korea, TCGS-Poland, COPDGene AA, and COPDGene NHW) and for each gene, we combined nonsynonymous, splice, and stop variants with a cohort-specific MAF less than 5%. Our analysis was limited to genes with a minimum minor allele count of 5. We used SNP-set (Sequence) Kernel Association Test (SKAT)-O (23), for all gene-based COPD association testing. The BEOCOPD and ICGN family-based cohorts were combined, using a covariate to denote study, to meet minimum sample size requirements of the statistical model, and then analyzed using a logistic mixed model in GMMAT version 0.5 (24). We performed a meta-analysis of SKAT-O results using SKAT version 1.0.1 (25, 26) and MetaSKAT version 0.40 (27), assuming homogeneous effects of SNPs across study cohorts. For both single-variant and gene-based meta-analyses, we adjusted for multiple testing using the Bonferroni correction, requiring that a single variant or gene be present in at least three of the five meta-analysis groups.

Secondary Analyses

We performed meta-analysis of white cohorts (European ancestry) to assess possible differences in results due to racial heterogeneity among cohorts in the primary analysis. We additionally performed an analysis with cases defined by severe COPD (Global Initiative for Chronic Obstructive Lung Disease spirometry stages 3–4) to search for enrichment of any association signals, as has been done in prior COPD GWASs (6). Finally, we evaluated the association statistics for a list of 119 genes identified from previous GWASs of COPD and lung function as well as genes implicated in Mendelian diseases leading to syndromes that included COPD (subsequently referred to as candidate genes) (see Table E2 in the online supplement) (6, 28, 29).

Additional cohort information and details of the methods used in this study are available in the online supplement.

Results

Participant Details

After subject quality control measures, a total of 6,004 control subjects and 6,161 cases were available for analysis. Baseline characteristics are shown in Table 1. As the two family-based cohorts, BEOCOPD and ICGN, were combined (see Methods section), we considered five analysis groups: (1) BEOCOPD and ICGN, (2) TCGS-Korea, (3) TCGS-Poland, (4) COPDGene AA, and (5) COPDGene NHW.

Table 1.

Cohort Compositions and Demographics, Separated by COPD Status (Control Subjects, Moderate to Severe COPD, and Severe COPD)

	BEOCOPD (n = 201 Pedigrees)	ICGN (n = 1,103 Pedigrees)	TCGS-Korea	TCGS-Poland	COPDGene African Americans	COPDGene Non-Hispanic Whites
Control subjects
Number of subjects	560	696	219	307	1,715	2,507
Males, %	41.6	48.3	96.8	67.4	58.1	49.4
Age, yr	39.3 (26.3–51.5)	54.9 (48.1–60.7)	53 (46–59)	58.3 (54.3–62.9)	51.8 (48.2–55.6)	59.3 (52.4–65.9)
Pack-years of smoking	1.66 (0–15.9)	25.1 (15.7–38.8)	25.5 (16.5–34)	32.2 (22.9–41.1)	32.7 (21.2–43.8)	35 (23.4–47)
FEV1, % predicted	94.5 (86.9–103)	97.5 (88.6–108)	93.6 (88.2–100)	102 (92.5–111)	96.6 (88.9–106)	95.5 (88–104)
Moderate to severe cases
Number of subjects	366	1,769	149	304	796	2,777
Males, %	39.9	58.6	99.3	70.1	55.5	55.7
Age, yr	50.8 (46.4–59)	59.4 (54.9–63.6)	69 (65–74)	62.2 (57.8–67.7)	58.2 (52.4–64.8)	65.2 (59–71)
Pack-years of smoking	37.5 (25.4–54)	45 (32.1–64.5)	40 (31–52)	40.3 (30–52.8)	37.8 (25.2–51.6)	49.8 (38–70.5)
FEV1, % predicted	29.2 (18.5–51.8)	39.2 (26.8–52.7)	33.2 (27.4–40.1)	28.7 (22.3–35.4)	54 (39.3–67.1)	50 (34.6–65.5)
Severe cases
Number of subjects	251	1,099	149	304	343	1,385
Males, %	33.1	60.9	99.3	70.1	58.0	57.7
Age, yr	50 (46.3–52.4)	59.5 (55.4–63.4)	69 (65–74)	62.2 (57.8–67.7)	60.1 (54.2–66.2)	65.8 (59.9–71.1)
Pack-years of smoking	39.1 (26.4–54)	46.4 (34–68)	40 (31–52)	40.3 (30–52.8)	38.5 (26–54)	52 (40–73.5)
FEV1, % predicted	22.6 (15.6–29.8)	29.9 (22.1–37.5)	33.2 (27.4–40.1)	28.7 (22.3–35.4)	36.5 (26.6–43.3)	34.6 (26.2–42.2)

Definition of abbreviations: BEOCOPD = Boston Early-Onset COPD; COPD = chronic obstructive pulmonary disease; COPDGene = Genetic Epidemiology of COPD; ICGN = International COPD Genetics Network; TCGS = Transcontinental COPD Genetics Study.

Values represent number, percent, or the median and interquartile range (lower quartile–upper quartile).

Marker Filtering for Analysis and Calculation of Statistical Significance Level

The number of nonsynonymous, splice, and stop variants present in at least three of the five analysis groups in the meta-analysis was 21,947, for a Bonferroni-adjusted, exome-wide, single-variant significance threshold of 2.3 × 10⁻⁶. For gene-based SKAT-O meta-analysis, 12,133 genes in at least three cohorts required a significance threshold of 4.1 × 10⁻⁶. Table E3 illustrates the number of variants and gene sets considered in both single-variant testing and gene-based testing with SKAT-O.

Single-Variant Meta-analysis for Association with COPD Affection Status

In our COPD affection status single-variant meta-analysis, we identified a single exome-wide significant variant, rs16969968 in CHRNA5, yielding P = 1.7 × 10⁻¹⁴. The CHRNA5 locus is known to be associated with COPD susceptibility and nicotine addiction (2, 4, 6). No novel variants reached exome-wide significance in our meta-analysis; however, rs181206 in the interleukin 27 gene (IL27) on chromosome 16 was just below the exome-wide significance threshold with meta-analysis (P = 4.7 × 10⁻⁶). Top results from the meta-analysis across the five analysis cohorts are shown in Table 2. Of the top 10 results, the majority were common variants with overall effect allele frequency greater than 5%; however, variants in AGER, FAM208B, CRAMP1L, and SERPINA1 were uncommon, with effect allele frequency between 1% and 5%. Note that the 10th most significant variant association (P = 1.4 × 10⁻⁴) was rs28929474, the SERPINA1 Z allele, which causes AATD (10). Although persons with known AATD were excluded from all study cohorts, three individuals in the TCGS-Poland study cohort were found to be homozygous for the Z allele. Removal of these three subjects diminished, but did not eliminate, the association for this variant in the TCGS-Poland study population (P value increased from 0.009 to 0.04) and in the overall meta-analysis results (P value increased from 1.4 × 10⁻⁴ to 2.4 × 10⁻⁴).

Table 2.

Top 10 Results from Meta-analysis Across All Cohorts of Nonsynonymous, Stop, and Splice Variant Associations with Moderate to Severe COPD

Chr	SNP	Gene	Effect Allele	Effect Allele Overall Frequency	β Value	SE	Test for Heterogeneity		Direction*†	P Value
							I2 Statistic	Q Statistic
15	rs16969968	CHRNA5	A	0.36	0.27	0.035	68	13	+++++	1.7 × 10−14
16	rs181206	IL27	A	0.69	−0.16	0.036	1.2	4.0	−−−−−	4.7 × 10−6
11	rs8177374	TIRAP	T	0.16	0.19	0.045	31	5.8	+−−++	2.6 × 10−5
6	rs2070600	AGER	T	0.042	−0.33	0.079	28	5.6	−−++−	2.7 × 10−5
11	rs679620	MMP3	T	0.48	0.12	0.030	0	2.5	+++++	3.8 × 10−5
6	rs10499052	AKD1	A	0.27	0.15	0.036	0	1.8	+++++	4.0 × 10−5
6	rs59056467	MICAL1	T	0.32	0.13	0.033	0	1.5	+−+++	4.7 × 10−5
10	rs41290259	FAM208B	A	0.98	0.58	0.15	0	1.2	+??++	8.8 × 10−5
16	rs61746451	CRAMP1L	T	0.011	−0.57	0.14	0	2.4	−?+−−	8.9 × 10−5
14	rs28929474	SERPINA1	T	0.021	0.45	0.12	50	4.0	+?+?+	1.4 × 10−4

Definition of abbreviation: Chr = chromosome; COPD = chronic obstructive pulmonary disease; SNP = single-nucleotide polymorphism.

Only variants observed in at least three of the five analysis cohorts are reported.

^*The direction of effect is reported for each analysis cohort. The symbols from left to right represent (1) Boston Early-Onset COPD study and International COPD Genetics Network study, (2) Transcontinental COPD Genetics Study (TCGS) Korea cohort, (3) TCGS Poland cohort, (4) Genetic Epidemiology of COPD Study (COPDGene) African American cohort, and (5) COPDGene non-Hispanic white cohort.

^†The symbol for each cohort is “+” for effect allele association with increased COPD susceptibility, “−” for association with decreased COPD susceptibility, and “?” if the variant was not present at a sufficient allele count in the analysis cohort.

The severe COPD association analysis (Table E4) showed stronger association (lower P values and higher effect sizes) at several known COPD loci, including the regions around CHRNA5 and ADAMTS7, MMP3, and SERPINA1. No additional exome-wide significant single-variant associations were discovered in this analysis. To assess nonsynonymous, splice, and stop variant associations with COPD in a more racially homogeneous set of individuals, we performed a meta-analysis limited to the white (European ancestry) cohorts (BEOCOPD and ICGN combined, TCGS-Poland, and COPDGene NHW). The results were similar overall (Table E5).

Gene-based Meta-analysis for Gene Association with COPD Affection Status

In gene-based association analysis, 16,006 genes were evaluated; 12,133 genes were present in at least three of the five analysis groups. Our top association was in the gene KIAA0020, but it did not reach exome-wide significance (P = 1.2 × 10⁻⁴). Our second most significant association was in AGER. When the variant rs2070600 (a top association from single-variant analysis) was removed from the AGER gene set, the association signal was markedly attenuated (P = 0.46). The top 10 most significant P values from SKAT-O meta-analysis of moderate to severe COPD are provided in Table 3. Analyses of severe COPD and limiting the sample to white subjects did not result in any significant associations (Tables E6 and E7).

Table 3.

Top 10 Results from Meta-analysis Across All Cohorts of SKAT-O Gene-based Association with Moderate to Severe COPD

Gene Set	BEOCOPD and ICGN	TCGS-Korea	TCGS-Poland	COPDGene AA	COPDGene NHW	Meta-analysis
KIAA0020	0.054	0.34	0.51	0.47	1.3 × 10−3	1.2 × 10−4
AGER	0.34	0.35	0.48	0.42	1.7 × 10−5	1.3 × 10−4
RRP1B	0.33	0.69	0.49	4.9 × 10−5	0.25	1.6 × 10−4
SQSTM1	0.045	—	0.74	0.35	2.2 × 10−3	1.6 × 10−4
RUFY4	0.060	0.11	0.63	0.14	0.040	2.3 × 10−4
TYRO3	0.36	—	—	7.8 × 10−5	0.18	2.8 × 10−4
IKZF2	0.078	—	0.58	0.12	0.011	4.2 × 10−4
TAF12	6.6 × 10−3	—	0.43	—	0.025	4.4 × 10−4
GON4L	1	0.86	0.40	4.9 × 10−4	0.24	7.3 × 10−4
MYEOV2	0.023	—	0.39	9.1 × 10−3	0.13	9.1 × 10−4

Definition of abbreviations: AA = African Americans; BEOCOPD = Boston Early-Onset COPD; COPD = chronic obstructive pulmonary disease; COPDGene = Genetic Epidemiology of COPD; ICGN = International COPD Genetics Network; NHW = non-Hispanic whites; SKAT-O = SNP-set (Sequence) Kernel Association Test; TCGS = Transcontinental COPD Genetics Study.

Results are given as P values. Gene sets present in at least three of the five cohort-level analyses are reported.

Candidate Gene Evaluation

In the meta-analysis of single nonsynonymous, splice, and stop variants, 14 variants in candidate genes were nominally significant (P < 0.05). Because the majority of our candidate genes were selected on the basis of significance in prior GWASs of COPD and lung function, we examined linkage disequilibrium (LD) between our variants and the previously reported GWAS SNPs. Most of these variants had either a high correlation coefficient (r²) or normalized coefficient of LD (D′) with previously reported SNPs. The results of these candidate gene evaluations are reported in Tables E8 and E9.

Further Assessment of IL27 Variant rs181206 in Additional Populations

The nonsynonymous variant rs181206 in IL27 was our only novel finding close to exome-wide significance; therefore, we sought to determine whether this variant could be tested in additional available cohorts with genome-wide SNP genotyping data. As this variant was common, genotyped or imputed data for this variant were available (see Table E10) in all three additional COPD cohorts (NETT/NAS, ECLIPSE, and GenKOLS). When these results for rs181206 were combined with our exome array data in a meta-analysis, the association signal for rs181206 was P = 5.7 × 10⁻⁸, exceeding the threshold for exome-wide significance.

We sought evidence for a functional effect of rs181206 in computational prediction models. This variant is predicted to be “probably damaging” by PolyPhen2 (30) and has a combined annotation-dependent depletion (31) scaled score of 23.5, which suggests that this variant has a deleterious effect on the protein. However, we also considered whether this variant (or variants in LD with it) could have effects on gene expression. Our nonsynonymous IL27 variant, rs181206, is in LD (r² > 0.5 in HapMap3 Utah residents of northern and western European ancestry [CEU]) with previously reported noncoding GWAS variants of body mass index (32), early-onset inflammatory bowel disease (33), Crohn’s disease (34), type 1 diabetes mellitus (35), and type 1 diabetes autoantibodies (36). Table E11 shows the P values from publicly available GWAS results for association of 16p11.2 variants with body mass index, type 1 diabetes, and inflammatory bowel disease. We evaluated the overall COPD association signal (primary analysis plus replication cohort meta-analysis) for rs181206 and the previously reported noncoding GWAS variants, and we report these results in Figure E1. These variants lie in the chromosome 16p11.2 region, which is complex, with 26 genes in a ±500-kb window around our lead association (see Table E12). Gene expression effects of rs181206 and the related 16p11.2 noncoding GWAS variants were evaluated in public or previously published studies of expression quantitative trait loci (eQTL) in blood, sputum, and lung. Sixteen different eQTL genes (of which 11 are in the ±500-kb window around variant rs181206) were identified and are shown in Table E13. None of these eQTL sources specifically pointed to IL27, but at least one source identified an effect on Tu, translation elongation factor, mitochondrial (TUFM) in both the blood and the lung for each variant examined at the 16p11.2 locus.

In an attempt to further explore the activity at the 16p11.2 locus related to COPD and lung function, we assessed lung tissue gene expression obtained from a separate study of former smokers with severe COPD and with normal spirometry (37). Details of this analysis are provided in the online supplement. Several genes in the region, including CD19, LAT, SULT1A1, SULT1A2, and TUFM, had nominal statistical significance (unadjusted P value < 0.05), but did not meet the false discovery rate threshold of less than 5%, for differential expression between COPD cases and control subjects. CD19, SULT1A1, and TUFM were significantly associated with FEV₁ and FEV₁/FVC ratio with a false discovery rate less than 5%.

Discussion

Despite substantial progress in identifying genetic risk factors for COPD from GWASs, much of the genetic risk for COPD remains undefined. Coding variants are known to predispose individuals to COPD, but most studies to date have been focused on the entire genome, the majority of which is noncoding. We performed a comprehensive assessment of coding variation on the exome array, analyzing both single-variant and gene-based associations with COPD affection status as the primary outcome phenotype, and included five study populations with 6,004 unaffected and 6,161 affected persons. Our study did not identify a large number of coding variants responsible for COPD susceptibility. These results are consistent with studies of other complex diseases showing that while coding variation, accounting for less than 1% of the genome, has a disproportionate contribution to disease susceptibility, the majority of heritability for most complex diseases is likely due to regulatory variation.

Gene-based analyses may have more power to detect an association than single-variant associations. However, our gene-based SKAT-O meta-analysis did not reveal any exome-wide significant genes associated with COPD affection status. We were able to recover evidence for the previously described association of COPD with AGER, though this association was driven by a single coding variant (rs2070600) that has been reported to be functional (38, 39). Our top gene-based COPD association was the pumilio domain–containing protein KIAA0020 (KIAA0020) (P  =  1.2 × 10⁻⁴). The KIAA0020 protein product is also known as the minor histocompatibility antigen HA-8, where minor histocompatibility antigens are T-cell epitopes that are presented by major histocompatibility complexes to function in immune regulation (40). While KIAA0020 has not previously been studied in COPD, it is expressed in lung tissue (41, 42) and is present at moderate to high levels in bronchial respiratory epithelial cells, pulmonary macrophages, and pneumocytes (43, 44). Future studies with increased sample size are needed to corroborate an association of KIAA0020 with COPD affection status.

Our top single nonsynonymous variant was rs16969968 in CHRNA5, a gene known to be associated with COPD susceptibility. The SNP in CHRNA5 is in strong LD (r² = 0.89; D′ = 0.95) with rs8034191, one of the first-identified COPD associations through GWAS (2). The 15q25 locus encompassing CHRNA5 is known to be complex, with a prior association with smoking behavior (45), and multiple independent variants explaining the association signals for COPD and smoking behavior (46, 47). A previous mediation analysis found that some but not all of the COPD risk at this locus is attributable to smoking behavior (48). Supporting this data is evidence of association of the 15q25 locus with airflow obstruction, even in never-smokers (49). While we adjusted for pack-years of smoking, we cannot rule out that the observed increase in strength of association of rs16969968 in severe COPD is due to increased smoke exposure. Another top association was with rs679620 in MMP3, which is in the same region recently associated with severe COPD in a GWAS meta-analysis (6). The variant rs2070600 in AGER is associated with COPD and quantitative computed tomography–identified emphysema, and was first identified as genome-wide significant in GWASs of FEV₁/FVC ratio (50–53).

We also identified a new exome-wide significant association of COPD with the nonsynonymous variant rs181206 in IL27. The variant rs181206 is a missense variant in which the minor allele (and COPD risk allele) G results in a leucine-to-proline amino acid change in the IL-27 subunit α-protein (54). Of interest, variants at 16p11.2 have previously been identified in GWASs of early-onset inflammatory bowel disease (33), Crohn’s disease (34), type 1 diabetes mellitus (35), and type 1 diabetes autoantibodies (36), adding to existing evidence of pleiotropy in GWASs (55) and to a potential biological link between autoimmune disease and COPD (56–58). Interestingly, effects at this variant appear to be discordant between diabetes and inflammatory bowel disease; the direction of effect that we found at this locus in COPD seems to be concordant with inflammatory bowel disease but discordant with risk to diabetes (59). The former may be of particular interest, given that some studies suggest an increased risk of COPD in patients with inflammatory bowel disease (60, 61).

IL27 (along with EBI3) encodes a heterodimeric cytokine complex that has a complex role in immune regulation and has been found to exhibit both pro- and antiinflammatory properties (62). Regarding COPD, Huang and colleagues showed that two IL27 polymorphisms, c.-964A/G (rs153109, LD with rs181206, r² = 0.49 in HapMap3 CEU; r² = 0.32 in HapMap3 Japanese individuals from Tokyo, Japan, and Han Chinese individuals from Beijing, China [JPT+CHB]) and g.2905T/G (rs17855750, LD with rs181206, r² = 0.03 in HapMap3 CEU, r² = 0.016 in HapMap3 JPT+CHB), are associated with COPD susceptibility in a Chinese population (63). Cao and colleagues reported higher levels of IL27 in the sputum and plasma of patients with COPD than in those of healthy control subjects, with a negative correlation between IL27 levels and FEV₁ in patients with COPD (64). Angata and colleagues recently suggested serum IL27 levels are a promising biomarker for COPD exacerbation (65). Our in silico analysis suggests the amino acid change of rs181206 could impact IL27 function. Alternatively, this variant could be in LD with a regulatory variant. In a study of lymphoblastoid cell lines, IL27 expression was decreased in healthy individuals with two copies of the rs1968752 (also in LD with rs181206) risk allele (33).

In fact, despite our hypothesis that using the exome array would identify functional coding variants, our discovered variant may be a marker for broader regulatory effects at the complex 16p11.2 region. An integrative analysis of expression data from the ECLIPSE study with previously published GWAS results identified the 16p11.2 locus as harboring evidence for association with COPD and gene expression and also showed that CCDC101 in the 16p11.2 locus had evidence of colocalization (46). In previous work, CCDC101 was also identified in a consensus module in a network-guided approach to COPD (66). Our survey of eQTL data indicates that CDC37P1, SULT1A1, and TUFM are seen as blood and lung and/or sputum eQTL signals in each of the five examined SNPs at the 16p11.2 locus. Further, SULT1A1 and TUFM are nominally significant for differential expression in COPD (46). Adding to the complexity of the 16p11.2 region, a common, approximately 0.45-Mb 16p11.2 inversion has been associated with the joint susceptibility to asthma and obesity (67). This 16p11.2 inversion encompasses the nonsynonymous variant rs181206. Although the LD structure between rs181206 and the inversion haplotype is not reported, some of the SNPs in LD with rs181206 also appear to be in LD with the inversion haplotype. The inversion genotype was strongly associated with expression levels of TUFM (P = 3.0 × 10⁻⁴⁰) and SULT1A4, SPNS1, SULT1A1, and CCDC101 (67). Gonzalez and colleagues (67) proposed that the expression variation of genes in the 16p11.2 region was “most likely” driven by the inversion itself. In the same study, IL27 levels were somewhat correlated (P = 2 × 10⁻²) with the inversion genotype, though not to the same degree as TUFM, SULT1A4, SPNS1, SULT1A1, and CCDC101.

The TUFM gene is intriguing in relation to COPD pathogenesis. TUFM has been described as interacting with the mitochondria-localized, nucleotide-binding, leucine-rich repeats (NLRX1) gene to reduce cytokine response and augment virus-induced autophagy (68, 69). In a recent study, three independent cohorts of patients with COPD were demonstrated to have decreased levels of NLRX1 mRNA, which was associated with increased disease severity, decreased pulmonary function, poorer quality of life, and worse prognosis (70). The TUFM and NLRX1 relationship, together with the association between variants in the 16p11.2 region and TUFM expression levels, and evidence for differential TUFM expression in COPD tissue provide support for a potential role of TUFM and more generally lend biologic plausibility to the association between COPD susceptibility and the 16p11.2 locus containing rs181206.

Our study has some potential limitations that deserve comment. First, despite our analysis of over 12,000 subjects, lack of statistical power may be a limitation. The number of subjects required to identify association with rare variants likely requires large sample sizes, and our effective sample size was decreased because of imbalances in cases and control subjects within individual cohorts. Second, COPD is a heterogeneous disease (71–73). Even when degree of airflow obstruction is similar, the clinical presentation of individuals can vary markedly, including the severity of emphysema and the frequency of exacerbations. Refining and studying phenotypes related to COPD pathogenesis may give additional power to detect novel genetic associations in COPD. In particular, efforts to define biologically similar subtypes of COPD may lead to improved ability to detect genetic contributions to COPD. The racial and/or ethnic heterogeneity of the COPD cohorts analyzed may also limit our observations. This heterogeneity was most notable in the gene-based analysis, where only 4,519 of 16,006 gene sets were common across all cohorts (where 9,023 gene sets were common across the white cohorts). We performed an analysis across all cohorts with the hope of maximizing power, but we also attempted to assess the impact of racial heterogeneity on our results by analysis only among the white cohorts. The results were not significantly different from when we included all cohorts in the meta-analysis, suggesting that racial heterogeneity across cohorts did not significantly impair our ability to discover genetic associations. However, the lack of a difference in findings may have been affected by the reduced sample size with the white cohort–only meta-analysis. Third, by using an exome genotyping array (as opposed to exome or whole-genome sequencing), we were unable to detect very low–frequency and private variants, which were explicitly excluded from the exome arrays. Fourth, our functional investigation of this locus is limited, and additional studies are needed to confirm the gene(s) involved, identify the mechanisms for increased susceptibility, and elucidate their role in the context of other genes related to COPD.

In summary, in a large study of coding variation in COPD, we identified a novel, nonsynonymous variant, rs181206, in IL27 that is associated with COPD susceptibility. This variant is in LD with variants identified by GWASs in body mass index, type 1 diabetes, and inflammatory bowel disease. Our work and the work of others suggest that this locus is complex and that susceptibility may be due to an effect on IL27 or other genes at this locus, particularly the mitochondria- and autophagy-related gene TUFM.