Abstract

INTRODUCTION

Chronic lower respiratory diseases are the third leading cause of death In the United States, resulting in 137,082 deaths in 2009¹. Lung function, as assessed by the spirometric measures of forced expiratory volume in one second (FEV₁), forced vital capacity (FVC) and the FEV₁ to FVC ratio (FEV₁/FVC), is an objective indicator of general respiratory health, as well as an important long-term predictor of morbidity and mortality^2–6. Family- and twin-based studies provide consistent evidence of genetic contributions to lung function, with estimates of heritability ranging as high as 85% for FEV₁, 91% for FVC, and 45% for the FEV₁/FVC ratio^7–20.

Recently, genome-wide association studies (GWAS) have begun to shed light on the complex genetic architecture of lung function measures. Two large meta-analyses of lung function GWAS in subjects of European ancestry who participated in the SpiroMeta²¹ or CHARGE²² consortium reported 11 loci associated with FEV₁/FVC or FEV₁. A subsequent combined meta-analysis of 48,201 individuals from both consortia reported 16 additional loci influencing lung function²³. However, variants at these highly significant loci in the SpiroMeta-CHARGE meta-analysis explained only 3.2% of the variance for FEV₁/FVC and 1.5% of the variance for FEV₁²³. Thus, similar to studies of other complex phenotypes, a significant proportion of the heritability remains unexplained by individual variants identified in GWAS^24–26.

This missing heritability following GWAS has been attributed to numerous potential causes^24–27, many or all of which likely contribute. In particular, the assumptions about the genetic model underlying complex phenotypes that are inherent in standard GWAS approaches may not reflect the true genetic architecture for many phenotypes. GWAS typically assess the effect of each (common) single nucleotide polymorphism (SNP) individually using stringent thresholds of significance. While this strategy has been effective in minimizing false positive associations and capturing the ‘low hanging fruit’, the inability to identify genetic variation that accounts for significant proportions of human phenotypic variation suggests that alternative analytic strategies are required to differentiate the true from false positive associations among the variants with more modest P-values. For example, considering 294,831 SNPs simultaneously in a linear model, Yang et al. found that common SNPs accounted for as much as 45% of the phenotypic variance and 50% of the heritability of height in 3,925 subjects²⁸, compared to only 5% of the variance of height explained by ~50 SNPs that reached genome-wide thresholds of significance in earlier studies^29–32.

Here, we conducted a GWAS of lung function phenotypes in members of a founder population, the Hutterites^{20, 33–34}. In addition to loci reported in previous GWAS, multimarker modeling identified a novel set of airway epithelial cell derived host defense genes.

METHODS

RESULTS

A total of 1,144 Hutterites (613 females, 53.6%) aged 6–89 years (mean ± SD, 30.6 ± 18.4 years) with both genome-wide genotyping and spirometry phenotypes were included in the GWAS (Table 1). These same data are shown for the non-asthmatic and asthmatic sample subsets in Table E2 in the Online Repository.

GWAS of lung function traits

We identified genome-wide significant associations between FEV₁/FVC and five SNPs at the THSD4-UACA-TLE3 locus on chromosome 15q23 (see Fig. E2a in the Online Repository), replicating results from previous GWAS^{21, 23}. Overall, there were 21 SNPs at this locus with P < 10⁻⁵ (see Table E3 in the Online Repository). The most significant SNP at this locus, rs12441227, explained 2.9% of the residual variance in FEV₁/FVC in the Hutterites. The evidence for association with SNPs at this locus remained when the individuals with asthma were excluded (Fig. E2d), and when the sample was stratified by age (Table E4).

Nine additional SNPs at four loci had P-values < 10⁻⁵ with FEV₁/FVC, including SNPs downstream of the C10orf11 gene on chromosome 10q22.3, which was associated with FEV₁ in a meta-analysis of lung function GWAS²³. When a sub-analysis was performed excluding the Hutterites with asthma, the evidence for association at this locus increased to genome-wide levels of significance (Table E4, Fig. E2f) The evidence for associations with SNPs at three of these loci with P-values < 10⁻⁵, CCL23-CCL18 on chromosome 17q12 (Fig. E2b and E2e), PITPNC1 locus on chromosome 17q24.2, and CHAF1B on chromosome 21q22.13, remained in sub-analyses excluding asthmatics. The evidence for association at all locus with P-values < 10⁻⁵ was present in subset analyses stratified by age (Table E4). Only two SNPs had P-values < 10⁻⁵ in the GWAS for the other two phenotypes: one SNP 7k downstream of the IL37 gene on chromosome 2q13 was associated with FEV₁ and one SNP in an intron of ASXL3 on chromosome 18q12.1 was associated with FVC.

The Manhattan and Q-Q plots of P-values for the GWAS of the three phenotypes are shown in Fig. 1; results for all SNPs with P < 10⁻⁵ are shown in Table E3 in the Online Repository. The GWAS P-values in the Hutterites for the 27 loci associated with lung function in previous meta-analyses^21–23 are shown in Table E5 in the Online Repository. Overall, we found nominal evidence (P < 0.05) of association with at least one of the three phenotypes for 14 SNPs at 9 of the 27 previously reported loci; and five SNPs at an additional five previously reported loci were associated with at least one of the three phenotypes at P < 0.01.

Open in a separate window

Figure 1

Manhattan and Q-Q plots of P-values from the GWAS of FEV₁/FVC (a, b), FEV₁ (c, d), and FVC (e, f). SNPs with P-values < 10⁻⁵ are shown in red. The horizontal red line shows the genome-wide significance threshold (P < 2.0 x 10⁻⁷).

Multimarker modeling

We assumed that there were additional true associations among the GWAS SNPs that did not reach genome-wide levels of significance because their effects are too small to detect in single SNP analyses, especially in a sample size of ~1,000 subjects. Therefore, to assess a multimarker model of risk including all SNPs with P < 10⁻³, we performed LASSO regression to identify minimum sets of SNPs that provided the smallest mean square error of FEV₁/FVC in the Hutterites. A set of 80 SNPs yielded the best predictive value and were used for further study (see Table E6 in the Online Repository).

First, we assessed the phenotypic effects of these 80 SNPs by binning individuals by the total number of alleles associated with reduced FEV₁/FVC that they carried (total possible = 160) and calculated the mean (SE) residual FEV₁/FVC for Hutterites in each bin. The mean residual FEV₁/FVC decreased with increasing number of ‘low FEV₁/FVC alleles’, consistent with an additive genetic architecture (Fig. 2).

Open in a separate window

Figure 2

The combined effects of genotypes for 80 SNPs on the residual FEV₁/FVC in the Hutterites. Hutterites were binned by their total number of alleles associated with reduced FEV₁/FVC (x-axis); the mean (SE) residual FEV₁/FVC for each bin is plotted on the right y-axis (blue dots and bars) and the number of subjects in each bin on the left y-axis. The linear regression line through these points is shown in red.

Next, we used the GRAIL algorithm⁴⁸ to mine the Gene Ontology database for evidence of functional connections between genes near the 80 predictive SNPs. We identified a subset of six SNPs with significantly related genes (GRAIL P < 0.05), including three clusters of β-defensin genes, two chemokine genes (CCL18 and CXCL12), and TNFRSF13B (Table 2 and Fig. 3). Notably, the associated GWAS SNPs at two replicated loci, THSD4-UACA-TLE3 and C10orf11, were not functionally connected to any other genes defined by the 80 SNPs. However, a SNP at the CCL23-CCL18 locus, the second most significant locus in the Hutterite GWAS (see Fig. E3b in the Online Repository), was significantly connected to the β-defensin genes, as well as to CXCL12 and TNFRSF13B in the GRAIL analysis. These six SNPs by themselves explained 5.8% of the residual variance in FEV₁/FVC in the Hutterites.

Open in a separate window

Figure 3

GRAIL functional connections between the 80 predictive SNPs. Six SNPs with no nearby genes defined by GRAIL are not shown. GRAIL identified six pairs of SNPs that implicated the same genes; only one SNP from each these 6 pairs is shown in the figure. The regions (SNPs; outer ring) and genes (inner ring) are optimally ordered to display connections with a minimal number of intersections. Only the genes with P_GRAIL-values < 0.05 have connections displayed. The thickness and redness of the connectors reflects the significance of the connections. Three clusters of β-defensin genes are the most connected sets.

Table 2

High scoring regions from the GRAIL analysis, sorted by the GRAIL P-value. P_GWAS is the P-value from the FEV₁/FVC GWAS in the Hutterites; Beta (SE) is that of the predictive SNP in the regression model for the 865 Hutterites; and P_GRAIL is the region’s P-value given by GRAIL. The last column shows the candidate gene identified by GRAIL.

SNP	Chr.	NCBI36 Position	PGWAS	Beta (SE)	PGRAIL	Implicated Gene
rs365548	20	185618	5.61E-04	0.003 (0.046)	6.4E-04	DEFB129, C20orf96
rs2921026	8	8384658	6.14E-04	−0.133 (0.046)	6.6E-04	DEFB107A
rs4815436	20	25521423	9.45E-04	0.181 (0.069)	9.7E-04	DEFB115, DEFB116, DEFB123, DEFB124
rs854679	17	31382952	3.40E-07	−0.049 (0.059)	0.020	CCL18
rs1570846	10	43776486	7.06E-04	−0.139 (0.055)	0.037	CXCL12
rs7216399	17	16797303	5.96E-04	−0.065 (0.054)	0.043	TNFRSF13B

Open in a separate window

DISCUSSION

The success of GWAS for unraveling the genetic architecture of complex phenotypes has been widely debated^{24–27, 49–51}. While there have been many robust associations discovered for a wide spectrum of diseases and phenotypes⁵², the associated variants typically explain relatively little of the phenotypic variation. Several recent studies have highlighted the importance of approaches that consider multiple variants simultaneously^{28, 46, 53–56}, a more suitable approach if the genetic architecture of common phenotypes is polygenic with many contributing loci with small effects. However, the best way to identify multiple contributing loci is at present unclear.

The GWAS of the FEV₁/FVC ratio in the Hutterites revealed two previously reported associations with measures of lung function. Associations with multiple SNPs at the highly replicated locus on 15q23^{21, 23} reached genome-wide significance in the combined sample, and SNPs at the C10orf11 on chromosome 10q22.3²³ reached genome-wide significance in the non-asthmatic subset of the Hutterite sample. These results were robust to age, with evidence for association present in both the child and adult subsets of the population. Moreover, we detected nominal levels of significance with SNPs at 9 previously reported loci associated with lung function measures. Together, these results indicate that genes influencing lung function in Europeans and European Americans from the general population also contribute to lung function phenotypes in the Hutterites.

To assess the combined effects of these and other SNPs with less significant evidence of association, we used LASSO regression to select the minimum set of SNPs from among the 312 with P-values < 10⁻³. The LASSO regression selected 80 independent SNPs as the best predictor of the FEV₁/FVC ratio. Consistent with an additive genetic model, the mean phenotypic value decreases with increasing number of “risk” alleles (Fig. 2). Moreover, this approach led to the discovery of additional genes, including three independent clusters of β-defensin genes, two chemokine genes, and a tumor necrosis factor (TNF) family receptor, suggesting an important link between host defense mechanisms and lung function. Defensins are antimicrobial peptides that recruit inflammatory cells and modulate innate and adaptive immune responses, participating in both the promotion and resolution of inflammatory responses⁵⁷. There are three classes of defensins, but only the β-defensins are specifically expressed in epithelial cells, including those lining the respiratory tract. Genetic studies have implicated the β-defensin genes on chromosome 8p23 in lung function in patients with asthma⁵⁸, chronic obstructive pulmonary disease (COPD)⁵⁹ and with cystic fibrosis (CF)⁶⁰. In particular, DEFB1 mRNA in bronchial epithelial cell biopsies was significantly elevated in COPD patients compared to controls and significantly associated with both reduced FEV₁ and the FEV₁/FVC ratio in COPD patients and controls⁵⁹. The results of our studies would further suggest that all three clusters of β-defensin genes on chromosomes 8p23, 20p13, and 20q11 contribute to lung function in healthy, unselected subjects. Chemokines are small proteins that bind to G-protein-coupled receptors and orchestrate the migration of circulating leukocytes to sites of inflammation. CCL18 (also named PARC, pulmonary and activated-regulated cytokine) is constitutively and highly expressed in the human lung⁶¹, and can generate regulatory T cells from CD4⁺CD25⁻ T cells in healthy individuals via direct induction of transforming growth factor β1 (TGF-β1)⁶². Functional polymorphisms in the promoter of the TGFB1 gene have been associated with airway responsiveness and asthma exacerbations, and haplotypes comprised of polymorphisms and specific coding variants in this gene have been associated with lung function in CF patients^63–64, although the exact variants and direction of effect are inconsistent across studies. Moreover, both β-defensin-2 and CCL18 were significantly elevated in peripheral blood from COPD patients compared to smoking and non-smoking controls⁶⁵, CXCL12 (also name SDF-1, stromal derived growth factor 1) is critical to bone marrow-derived stem cell production, and shows increased expression in bronchial alveolar lavage fluid after bleomycin-induced lung fibrosis in a murine model, and in airway tissues in patients with idiopathic pulmonary fibrosis compared to controls⁶⁶. The TNFRSF13B gene encodes the transmembrane activator and calcium modulator and cyclophilin ligand interactor (TACI), which binds two ligands, BAFF (B-cell activating factor) and APRIL (a proliferating-inducing ligand). It is thought that TACI plays a key role in B cell activation and differentiation into plasma cells. In a recent study, rare mutations in TNFRSF13B were associated with asthma symptoms in Swedish children⁶⁷. Moreover, BAFF expression in alveolar macrophages was inversely correlated with lung function in COPD patients⁶⁸. Our study extends the roles of these two chemokines and TNF-family receptor to inter-individual variability in normal lung function.

Despite conducting this study in a relatively small sample (~1,000 Hutterites) and the absence of a major locus influencing variation in lung function compared to other traits (for examples, see references^{39, 69}), we were successful in identifying both genome-wide significant associations with replicated loci on chromosome 15 in the combined sample and on chromosome 10 in the non-asthmatic subset, in addition to a set of novel variants that are highly predictive for lung function in the Hutterites. The power of our study was likely enhanced by the homogeneity of the Hutterite population compared to the larger population samples that have been included in previous studies of lung function^21–23. The advantages of this population for genetic studies of complex phenotypes are primarily two-fold. On the one hand, it is possible that there are fewer lung function-associated alleles segregating in the Hutterites due to the population bottleneck that occurred prior to their emigration to the United States^35–36. This would result in a simpler genetic architecture due to both overall reduced genetic variation and increased frequencies of some variants with potentially larger phenotypic effects that are rare in other European populations. On the other hand, their communal lifestyle and shared environmental exposures³³, which include the absence of exposure to cigarette smoke and air pollution, may have enhanced the effects of genetic variation in general, and on specific pathways in particular, on lung development and subsequent lung function. In this population, exposures are remarkably similar during critical periods of lung development both in utero and in early life. Hutterite women and young children are not directly involved in farming activities, and their homes are generally distant from the agricultural fields and animal barns. Meals are prepared in a communal kitchen, using traditional recipes that are shared among the colonies. There are no pets, televisions, radios, or computers in the homes, and, as a result, Hutterite children spend significant proportions of each day playing outside. Thus, the absence of important environmental exposures that impact lung development and lung function, combined with a shared environment throughout life, not only reduces non-genetic heterogeneity but also allows for the detection of lung function alleles that are not confounded with those related to socioeconomic factors or behavior, such as cigarette smoking, or to ecogenetic pathways that are important in metabolizing inhaled particles. These population characteristics possibly enabled the novel finding in this study of an enrichment of genes involved in anti-microbial immunity in the airways among those associated with lung function.

In summary, this study identifies genome-wide significant associations between lung function and SNPs at the THSD4-UACA-TLE3 locus on chromosome 15q23 and the C10orf11 on chromosome 10, and replicates many other previous GWAS results. Moreover, using LASSO regression, we identified 80 independent SNPs as the best predictor of the FEV₁/FVC ratio, with the mean phenotypic value decreases with increasing number of “risk” alleles, consistent with an additive genetic architecture. Of note is that multimarker modeling implicated for the first time common variation in three independent clusters of β-defensin genes, two chemokine genes, and a TNF family receptor that involved in anti-microbial immunity in airway mucosa influences lung function.

Supplementary Material

Acknowledgments

Abbreviations

Footnotes

References