In a cohort of well-characterized COPD subjects, we integrated genomewide SNP and gene expression data derived from induced sputum, a biologically-relevant tissue in COPD, to identify a set of eQTL SNPs affecting gene expression levels. The SNPs were then tested for association with the clinical phenotype of COPD; gene expression was not tested for association with disease status in this set of COPD cases only. Using the eQTL results, we implicated two distinct COPD susceptibility genes in a previously identified region of chromosome 15q25. Additionally, we provide evidence for a potential novel COPD susceptibility locus in the HLA region on chromosome 6.
The initial GWAS in COPD found significant associations on chromosome 15q25, with SNPs in the genes CHRNA3
, encoding two subunits of the nicotinic acetylcholine receptor 
. This region has also been associated with lung cancer, peripheral arterial disease, and smoking behavior 
, so it is not clear whether these genes have a direct effect on COPD susceptibility, or their effects are at least partially influenced through cigarette smoking, the major environmental risk factor for COPD 
. In terms of genetic regulation of expression of the chromosome 15q25 genes, we found similar eQTL associations with CHRNA5
expression in induced sputum as has been found in brain 
and lung tissue 
. We found additional sputum eQTL SNPs for CHRNA5
in moderate LD with previously defined eQTLs. The previous papers on brain and lung tissue gene expression did not report testing IREB2
, a gene previously associated with COPD 
. The specific IREB2
SNPs associated in GWAS (rs13180) 
and in a candidate gene analysis of differentially expressed genes (rs2656069) 
were in only moderate LD (r2
0.44) with each other, implying independent effects on IREB2
expression. The IREB2
eQTL SNPs were not in LD with each other, suggesting the presence of at least two COPD susceptibility genes on chromosome 15q25. Previous studies have similarly used eQTL analyses to add functional information about genes identified through GWAS, including studies of asthma 
, celiac disease 
, and Crohn's disease 
. However, these prior studies have examined gene expression in blood cells, and not primary disease tissues.
However, we did not finding significant cis-eQTL SNPs for two other known COPD loci, HHIP and FAM13A. The associated SNPs found through GWAS may exert their effects on phenotype via other mechanisms besides influencing gene expression. Alternatively, the GWAS SNPs may actually be eQTLs acting in other tissues besides sputum, such as alveolar or bronchial epithelial cells, which were not assessed in our study.
Besides improving understanding of the COPD susceptibility locus on chromosome 15q25, we identified a potential novel COPD locus on chromosome 6. The SNP maps to gene PSORS1C1
, but it is associated with expression levels of the neighboring gene PSORS1C3
. Variants in PSORS1C3
have been reported to be associated with psoriasis 
, an immune-mediated skin disease. PSORS1C3
is located in the major histocompatability (MHC) region, and subsequent papers have found that the associations with psoriasis may be due to variants in HLA-C
(MIM 142840) 
. Interestingly, one study has reported an epidemiologic association between psoriasis and COPD 
, and cigarette smoking is a risk factor for psoriasis as well 
. Although there are no reports of HLA-C
associations with COPD, alleles of other MHC class I genes, HLA-A
, have been associated with COPD 
. The locus encompassing PSORS1C1/3
will require additional replication studies and functional validation to confirm its role in COPD susceptibility.
Prior studies have also used eQTL analyses to identify novel genes for complex traits, including age related decline in kidney function 
and body mass index 
. In contrast to our study, these papers first found gene transcripts correlated with the phenotype, then tested SNPs in/near these genes for association with expression levels. We performed the cis-eQTL analysis as the initial step, then tested the eQTL SNPs for phenotype association. This limits multiple testing compared to a GWAS, enriching for eQTL SNPs which may be more likely to be associated with disease 
This study has several limitations. The sample size of 131 subjects, though adequate for gene expression analyses, may be underpowered to detect all potential eQTL associations. Therefore, we limited the cis-acting analysis to SNPs within 50 kb from the gene, to limit the multiple testing burden. Based on RNA sequencing data, Pickrell et al. estimate that 90% of eQTL SNPs are within 15 kb of a gene 
. Previous papers have used a 50 kb limit to define cis-acting eQTLs 
. Using this method, we were able to replicate published eQTL associations from other tissues and were able to identify a set of significant eQTL SNPs to carry forward for COPD association studies. However, our method would be unable to detect cis-eQTLs located >50 kb from the TSS, such as a SNP in an upstream enhancer or in the 3′ UTR of a large gene. Due to the sample size, we limited our investigation to cis-acting eQTL SNPs, as a full search for trans-acting regulatory SNPs greatly increases the number of tests performed. The literature suggests that sample sizes under 200 subjects may be inadequate to find true trans-eQTLs 
Several groups have compared eQTLs in different tissues from the same individual, finding both overlapping and tissue-specific eQTLs 
. Multiple tissues are known to be important in COPD biology, including large and small airways, lung parenchyma and immune cells. By only surveying sputum, we may have missed significant eQTLs for COPD genes that are expressed in other tissues. Multiple cell types may be present in sputum, yet neutrophils have been shown to be the predominant cell type in the sputum samples from COPD subjects in ECLIPSE 
. Despite these limitations, sputum is a clinically important tissue in COPD and is more accessible for genomic and biomarkers studies than lung tissue. Studying diseased individuals may be advantageous to identify eQTL SNPs for potential disease genes, which may only be expressed, or may be expressed at higher levels, in patients compared to healthy controls.
In conclusion, we combined genomewide SNP genotyping with genomewide expression profiling from a relevant tissue in well-characterized subjects with a common chronic disease. Using this strategy, we were able to gain insights into the functional role of SNPs previously associated through GWAS, as well as identify a potential novel disease susceptibility gene which would have been missed using standard GWAS analysis. Previous eQTL studies have provided important information about genetic control of human gene expression. Integrative genomics studies in relevant tissue from well-phenotyped individuals, as we have performed, will be required to apply this knowledge to human disease.