|Home | About | Journals | Submit | Contact Us | Français|
We aimed to identify novel genetic variants affecting asthma risk, since these might provide novel insights into molecular mechanisms underlying asthma.
We performed a genome-wide association study (GWAS) in 2,669 physician-diagnosed asthmatics and 4,528 controls from Australia. Seven loci were prioritised for replication after combining our results with those from the GABRIEL consortium (n=26,475), and these were tested in an additional 25,358 independent samples from four in-silico cohorts. Quantitative multi-SNP scores of genetic load were constructed on the basis of results from the GABRIEL study and tested for association with asthma in our Australian GWAS dataset.
Two loci were confirmed to associate with asthma risk in the replication cohorts and reached genome-wide significance in the combined analysis of all available studies (n=57,800): rs4129267 (OR=1.09, combined P=2.4×10−8) in the interleukin-6 receptor gene (IL6R) and rs7130588 (OR=1.09, P=1.8×10−8) on chromosome 11q13.5 near the leucine-rich repeat containing 32 gene (LRRC32, also known as GARP). The 11q13.5 locus was significantly associated with atopic status among asthmatics (OR = 1.33, P = 7×10−4), suggesting that it is a risk factor for allergic but not non-allergic asthma. Multi-SNP association results are consistent with a highly polygenic contribution to asthma risk, including loci with weak effects that may be shared with other immune-related diseases, such as NDFIP1, HLA-B, LPP and BACH2.
The IL6R association further supports the hypothesis that cytokine signalling dysregulation affects asthma risk, and raises the possibility that an IL6R antagonist (tocilizumab) may be effective to treat the disease, perhaps in a genotype-dependent manner. Results for the 11q13.5 locus suggest that it directly increases the risk of allergic sensitisation which, in turn, increases the risk of subsequent development of asthma. Larger or more functionally focused studies are needed to characterise the many loci with modest effects that remain to be identified for asthma.
A full list of funding sources appears at the end of the paper.
Eight loci were reported to associate with asthma risk with genome-wide significance, namely the locus containing GSDMB, ORMDL3 and GSDMA locus on chromosome 17q21 (1), PDE4D (2), DENND1B (3), the locus containing IL1RL1 and IL18R1 on chromosome 2q12.1 (4), HLA-DQ, IL33, IL2RB and SMAD3 (5). Notably, these findings point to a genetically-linked dysregulation of cytokine signalling in asthma, and provide a number of specific targets for the development of novel biological therapies. They also implicate previously unsuspected risk loci, such as the 17q21 region; as our understanding of the biological mechanisms underlying these associations improves, novel insights into the pathophysiology of asthma are likely to emerge.
The risk variants identified to date explain only a small fraction of the disease heritability (< 1% each), indicating that many more loci remain to be identified. Because of the proven success of genome-wide association studies (GWAS) to identify common risk variants (6), further dissection of this uncharacterised component of disease risk through well powered genetic studies represents a unique opportunity to advance our knowledge of the mechanisms that trigger asthma.
In this paper, we describe a series of genetic association analyses carried out to identify novel risk loci for asthma, including (1) a GWAS of physician-diagnosed asthma using data for 7,197 individuals of European descent from Australia; (2) a meta-analysis of these results with findings from 26,475 individuals studied by the GABRIEL consortium (5); and (3) testing the most significant regions of association in a further 25,358 independent samples.
We first carried out a GWAS of 7,197 individuals of European ancestry from Australia; throughout this paper, we refer to this analysis as the “Australian GWAS”. Participants were drawn from three cohorts (webappendix pp 2-6): the Australian Asthma Genetics Consortium (AAGC) cohort (n=1,810); the Busselton Health Study cohort (n=1,230); and the Queensland Institute of Medical Research (QIMR) GWAS cohort (n=4,157). Patients were generally recruited between 1964 and 2010.
Of the 2,669 asthmatic patients, 759 (28%) were diagnosed through clinical examination and 1,910 (72%) reported a lifetime physician diagnosis of asthma in epidemiological questionnaires. With respect to disease onset, 1,438 (54%) subjects were classified as having childhood asthma (defined by an age-of-onset at or before age 16 years), 697 (26%) subjects with later onset asthma (age-of-onset after the age of 16 years) and 534 (20%) with unknown age-of-onset. 1,570 (59%) of asthmatics were atopic, as defined by a positive skin prick test (SPT) response to at least one common allergen; 1,444 (54%) had at least one first-degree relative with asthma; and 1,026 (38%) reported lifetime smoking (webappendix pp 2-3).
The 4,528 controls included 2,701 (60%) individuals who were classified as asthma-free based on clinical examination (109 [2%]) or epidemiological questionnaires (2,592 [57%]). The remaining 1,827 (40%) individuals provided no information about their asthma status. As we show in the webappendix (pp 7), including this group of asthma-unknown individuals in the analysis as controls improved power to detect a true genetic association. SPT information and lifetime smoking status was unavailable for most controls (3,903 [86%] and 3,822 [84%], respectively; webappendix pp 2-3).
Overall, the mean age of participants was 39 years (SD=18.5, range 2 to 92) and 3,986 (55%) were women. This dataset includes 4,259 samples that have not been previously included in any asthma GWAS, 1,708 that were included in Ferreira and colleagues, (7) and 1,230 samples from the Busselton cohort included in the GABRIEL study (5).
Next, to prioritise loci for replication, results from the Australian GWAS were combined with those published and made publicly available by the GABRIEL consortium (5). After excluding overlapping samples between the two studies, the meta-analysis was based on results from 12,475 physician-diagnosed asthmatic patients and 19,967 controls.
Lastly, the most significant regions of association were tested for replication in four additional cohorts (3,322 cases and 22,036 controls) that contributed in-silico results for analysis: the Western Australian Pregancy Cohort (Raine) study (654 asthmatic and 621 control patients); the QIMR follow-up cohort (602 asthmatic and 2,206 control patients), consisting of individuals unrelated to those included in the QIMR GWAS cohort; the Netherlands Twin Registry (NTR) study (350 asthmatic and 2,321 control patients); and the Analysis in Population-based Cohorts of Asthma Traits (APCAT) consortium (1,716 asthmatic and 16,888 control patients). The APCAT consortium included six population-based cohorts from Finland and the United States: the Helsinki Birth Cohort (123 asthmatic and 1,533 control patients), Health 2000 (153 asthmatic and 1,841 control patients), Finrisk (160 asthmatic and 1,705 control patients), the Northern Finland Birth Cohort 1966 (364 asthmatic and 3,502 control patients), the Young Finns Study (119 asthmatic and 1,844 control patients) and the Framingham Heart Study (797 asthmatic and 6,463 control patients). The webappendix (pp 8-10) contains a detailed description of every cohort. All participants gave written informed consent and the study protocols were reviewed and approved by the appropriate review committees.
The Australian GWAS included data from 7,197 individuals of whom 5,523 (77%) were genotyped with Illumina 610K (Illumina, San Diego, CA, USA) array 1,674 (23%) with Illumina 370K array as part of four genotyping projects that are summarised in the webappendix (pp 11): the AAGC (n=1,810), QIMR_610K (n=2,483), QIMR_370K (n=1,674) and Busselton (n=1,230) projects. The same single-nucleotide polymorphism (SNP) quality control filters were applied to each project individually, including the removal of SNPs with call rate <95%, minor allele frequency (MAF) < 0.01 and Hardy-Weinberg equilibrium test P < 10−6. Autosomal SNPs passing quality control were then used to impute 7.8 million variants available from the combined 1000 genomes (60 individuals with northern and western European ancestry from the Centre d’Etude du Polmorphisme Humain collection (CEU), March 2010 release) and HapMap 3 (955 individuals from 11 populations, February 2009 release) reference panels using Impute2 (8). The AAGC and QIMR_610K datasets were merged before imputation as both were genotyped with the same array, were available concurrently for analysis and had no evidence for systematic allele frequency differences between controls (genomic inflation factor λ = 1.014 for AAGC controls vs. QIMR_610K controls) nor between asthmatic patients (λ = 1.001 for AAGC cases vs. QIMR_610K cases). We nonetheless removed a small subset of 1,104 SNPs (0.2%) that had significant (P < 0.001) allele frequency differences between the two datasets, as these probably indicated technical artifacts. The genomic inflation factor from a case-control association analysis in the resulting AAGC+QIMR_610K dataset (n=4,293) was 1.014, further indicating that batch effects between the AAGC and QIMR_610K datasets did not have a systematic effect on the results.
Imputation and subsequent SNP quality control were carried out as three separate analyses, corresponding to the AAGC+QIMR_610K, QIMR_370K and Busselton datasets. The QIMR_370K and Busselton datasets were imputed separately because the former was genotyped using a different array and because the latter only became available for analysis at a later stage. After imputation, we excluded SNPs with low imputation accuracy (information < 0.3), MAF < 0.01 or Hardy-Weinberg equilibrium test P < 10−6. To minimise the impact of potential dataset-specific technical artifacts, we also excluded from analysis SNPs with significant (P < 0.001) allele frequency differences between the three imputation analysis groups (from case-case and control-control comparisons, as described above). After quality control, genotype data for 5.7 million common SNPs (webappendix pp 12) were merged across the analysis datasets of AAGC+QIMR_610K, QIMR_370K and Busselton. All subjects included were confirmed to be unrelated and of European ancestry (webappendix pp 6) through the analysis of genome-wide allele sharing. Comparable procedures were used for the replication cohorts (webappendix pp 8-10).
For the Australian GWAS, individual SNPs were tested for association with lifetime physician-diagnosed asthma using a Cochran-Mantel-Haenszel test as implemented in PLINK (9), with three strata corresponding to the three imputation analysis groups described in the previous section. This analysis had adequate power (80% for α = 5×10−8) to detect loci with a genotype relative risk ranging from 1.23 (MAF = 0.50) to 2.23 (MAF = 0.01) (webappendix pp 7). For imputed SNPs, we analysed best-guess (i.e. most likely) genotypes. The Breslow-Day test of homogeneity was applied to test for differences in SNP effects between the three groups.
To prioritise SNPs for follow-up amongst lower-ranked loci from the Australian GWAS, we performed a fixed-effects meta-analysis of our results (after excluding the 1,230 overlapping samples from Busselton) with those published and made available by the GABRIEL (5), which were based on 10,365 asthmatic patients and 16,110 controls genotyped with the Illumina 610K array. We restricted our analysis to 421,334 SNPs that were available in all 36 GABRIEL cohorts, had no strong evidence for significant heterogeneous effects between the 36 cohorts (P > 0.01) and which were tested in our study. We used the Cochran’s Q test to identify SNPs with significant heterogeneous effects between our study and the GABRIEL study (5). No correction for genomic inflation of test statistics was applied to either set of results before the meta-analysis.
In the replication study, SNPs were tested in each cohort separately (logistic regression with cohort-specific adjustments) and then combined by performing a fixed-effects meta-analysis with METAL (10).
Lastly, we used a recently described (11) approach for phenotype prediction from genome-wide SNP data to address the hypothesis that hundreds or thousands of loci with individual weak effects contribute to asthma risk. Briefly, we selected groups of SNPs based on their level of association with asthma in the GABRIEL (5) analysis (for example, SNPs with a P ≤ 5×10−7) and created a quantitative score of genetic load based on these SNPs for every individual included in the Australian GWAS, after restricting the sample to 2,082 patients with asthma and 2,211 controls genotyped with the 610K array and that did not overlap with those included in the GABRIEL. This genetic score was calculated as the weighted sum of the number of reference alleles for each genotyped SNP, with the weight corresponding to the effect size (log of the odds ratio [OR]) estimated for that marker in the GABRIEL study (5). We then used logistic regression to test the association between a specific genetic score and case-control asthma status in our study. The log of the odds ratio was selected as the weight (instead of, for example, simply counting the number of risk alleles) because it considers both the expected magnitude and direction of effect of an individual SNP. We used the odds ratio estimated in the GABRIEL study because currently that study represents the largest asthma GWAS done so far, hence providing the most accurate estimates of SNP effects available for asthma. We considered seven groups of SNPs based on arbitrary thresholds of significance in the GABRIEL study, from very strict to very liberal: P ≤ 5×10−7, 5×10−7 < P ≤ 10−4, 10−4 < P ≤ 0.01, 0.01 < P ≤ 0.1, 0.1 < P ≤ 0.2, 0.2 < P ≤ 0.5 and 0.5 < P ≤ 1. To facilitate interpretation, we restricted our analysis to directly genotyped SNPs that were in low linkage disequilibrium (r2 < 0.1) with each other, using the clump routine implemented in PLINK (9). Results were unchanged by including as covariates the first four components obtained from the multidimensional scaling analysis of identity-by-state allele sharing.
The sponsors of the study had no role in the study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.
To identify novel SNPs associated with asthma risk, we first performed a GWAS in 2,669 physician-diagnosed asthmatic patients and 4,528 controls from Australia. We recognise that with this sample size, power was only adequate (≥ 80%) at the genome-wide level (α = 5×10−8) to detect loci with strong effects (e.g., an OR ≥ 1.25 for an allele frequency of 0.40). The genomic inflation factor confirmed that population substructure or other potential technical artifacts did not have a systematic impact on the results (λ=1.014; Q-Q plot in the webappendix pp 13).
In the Australian GWAS, five independent SNPs (linkage disequilibrium r2 < 0.1) were associated with asthma risk at a pre-defined cut-off of P ≤ 5×10−6 (webappendix pp 14). Among these regions were two loci previously reported (1,4) to associate with asthma – chromosome 17q21 near GSDMA (rs8071050, OR = 1.27, P = 6.3×10−10) and 5q22.1 near WDR36 (rs1438673, OR = 0.84, P = 3.2×10−6) – and three novel loci: chromosomes 12q24.31 near CABP1 (chr12:119562269, OR = 1.85, P = 8.3×10−7), 16q24.1 in KCNG4 (rs7196274, OR = 0.81, P = 1.3×10−6) and 8q22.1 in CDH17 (rs11776675, OR = 1.21, P = 2.7×10−6).
The 12q24.31 locus corresponds to an uncommon variant (chr12:119562269: MAF = 0.02) from the 1000 Genomes Project imputed in our data with modest confidence (imputation information of 0.86 for the 610K array and 0.52 for the 370K array). However, there was no supporting evidence for association from neighbouring markers (webappendix pp 15) nor from haplotype analyses performed after exclusion of the chr12:119562269 variant (data not shown), suggesting that this association probably represents a technical artifact. The 16q24.1 and 8q22.1 associations were supported by multiple markers (webappendix pp 15) and so were selected for follow-up in an independent replication panel, as described below.
Next, to prioritize additional loci for follow-up, we performed a fixed-effects meta-analysis of our results with those published and made publicly available by the GABRIEL consortium (5). After exclusion of overlapping samples, the combined analysis was based on results from 12,475 asthmatic patients and 19,967 controls genotyped at 421,334 autosomal SNPs. The genomic inflation factor for this analysis was 1.068 (Q-Q plot in the webappendix pp 16); only ~1% of SNPs had a Cochran’s Q test P < 0.01, indicating that SNP effects were largely homogeneous between the two studies.
19 independent (r2 < 0.1) SNPs had a meta-analysis association P ≤ 5×10−6 (listed in the webappendix, pp 17). Of these, 14 were not considered for further analysis because they were not associated with asthma in the Australian GWAS (n=2) or because they were located in regions previously reported to associate with asthma risk (n=12). The remaining five SNPs were nominally significant (P ≤ 0.05) in our dataset and were located in novel regions for asthma; as such, they were also selected for follow-up (webappendix, pp 17).
Thus, in total, we identified seven putative novel asthma risk loci (P ≤ 5×10−6) that we sought to follow-up in an independent panel of 3,322 asthmatics and 22,036 controls previously genotyped as part of four studies (APCAT, RAINE, QIMR and NTR). Power to replicate each individual SNP association at a Bonferroni-corrected threshold (α ≤ 0.05/7=0.007) ranged between 64% and 100%.
Follow-up analyses replicated at P ≤ 0.007, the association with one SNP located in the interleukin-6 receptor (IL6R) gene on chromosome 1q21.3 (rs4129267, uncorrected P = 0.0033; Table 1). In the overall analysis of the discovery and follow-up panels of all the studies (15,797 asthmatic patients and 42,003 controls), this variant was highly associated with asthma risk (OR = 1.09, P = 2.3×10−8; Figure1A ). A second SNP, rs7130588 on chromosome 11q13.5, replicated less strongly in the follow-up panels (combined P = 0.0328) but reached genome-wide significance in the overall analysis of all samples (OR = 1.09, P = 1.8×10−8; Table 1 and Figure 1B). The remaining five regions of association (PCDH20, PRKG1, KCNG4, IGHMBP2, CDH17) were not significantly associated with asthma in the follow-up study (P > 0.05; webappendix pp 19-20). Thus, genome-wide association analyses followed by replication identified variants in IL6R and on chromosome 11q13.5 as novel risk loci for asthma.
To further characterise the association between the IL6R and 11q13.5 variants and asthma risk, we tested each locus for association with nine asthma subphenotypes measured in up to 2,669 asthmatic individuals (webappendix pp 21). The rs7130588:G predisposing variant on 11q13.5 was more common in atopic asthmatic patients (defined by a positive skin prick test to at least one common allergen) than non-atopic asthmatic patients (G allele frequency 38% vs. 32%, P = 0.0007; webappendix pp 21-22). Consistent with this result, there was no evidence for an increased risk of asthma associated with this allele when considering only non-atopic individuals (webappendix pp 22); however, given the modest sample size for this secondary analysis, these findings require confirmation by independent studies.
We also sought to validate the genome-wide significant associations reported by the GABRIEL study (5), Sleiman et al. (3) and Himes et al. (2). We replicated (same SNP, P ≤ 0.05 and same direction of effect) the association with GSDMB (and GSDMA), IL18R1, IL33 and IL2RB (webappendix pp 23). A consistent but non-significant effect was observed for HLA-DQ and SMAD3. There was no support for an association with PDE4D or DENND1B (webappendix pp 23-24).
Next, we investigated the hypothesis that hundreds or potentially thousands of common variants with weak effects influence asthma risk. A multi-SNP score computed based on data for the ten most associated loci reported in the GABRIEL study (5) was significantly associated (P = 8.2×10−15) with asthma status in our study (Figure 2). Multi-SNP scores based on loci with less remarkable levels of association with asthma in the GABRIEL study (5) were also associated with asthma status in our study. For example, a score computed based on 2,520 largely independent SNPs (r2 < 0.1) that had individual P-values between 10−4 and 0.01 in the GABRIEL study (5) significantly associated with asthma case-control status (P = 1.2×10−4), indicating that many of these SNPs probably represent genuine asthma risk loci with modest effects. We stress, however, that these results have little clinical relevance, as the multi-SNP scores tested had very poor discriminative ability, with values for the area under the receiver operator characteristic curve (AUC) not exceeding 0.576 (Figure 2).
Three confirmed asthma loci are shared with Crohn’s disease, namely ORMDL3, IL1RL1 and now chromosome 11q13.5. We postulated that loci with modest effects on asthma risk might also be shared with other inflammatory or immune diseases. To explore this possibility, we identified 356 SNPs listed in the catalog of published GWAS (12) that were previously reported to associate at P ≤ 5×10−8 with 54 such traits or diseases excluding asthma (webappendix pp 25). Importantly, this list did not include any SNP located near (<500 kb) the ten GABRIEL loci (5), PDE4D (2), DENND1B (3), IL6R or 11q13.5. After excluding redundant markers (r2 ≥ 0.1), results for 207 largely independent SNPs from the original list were available in the meta-analysis of the Australian GWAS and the GABRIEL (5), either directly or through a proxy SNP (r2 ≥ 0.8). Of these, 16 (8%) were associated with asthma risk with a P ≤ 0.01 (webappendix pp 26), when only about two were expected at this threshold under the null hypothesis of no association between the 207 SNPs and asthma (Fisher’s exact test P = 0.001). Four SNPs survived a Bonferroni correction for multiple testing (P ≤ 0.00024): rs11167764 near NDFIP1 (OR = 1.11, P = 4.6×10−6) and rs1847472 in BACH2 (OR = 1.07, P = 0.00023), two variants associated with Crohn’s disease (13); rs2596560 near HLA-B (OR = 0.92, P = 6.5×10−5), a proxy SNP (r2 = 1.00) for a variant (rs3134792) reported to associate with psoriasis (14); and rs13076312 in LPP (OR = 0.93, P = 0.00016), a proxy SNP (r2 = 1.00) for a variant (rs1464510) reported to associate with celiac disease (15). These variants represent putative novel associations for asthma that require further confirmation and draw attention to molecular pathways that are probably shared between asthma and other inflammatory or immune diseases.
We identified two new loci with genome-wide significant association with asthma risk: rs4129267 in IL6R and rs7130588 on chromosome 11q13.5 (panel). Multiple lines of evidence suggest that IL6R is indeed the causal gene underlying our observed association with rs4129267. First, rs4129267 is strongly associated with variation in serum concentration of the soluble form of the IL-6 receptor (sIL-6R) (18). Each copy of the rs4129267:T allele increases sIL-6R protein concentration by about 1.4-fold (18), while it increases the risk of asthma by 1.09-fold based on our analyses. Second, the concentration of sIL-6R is increased in both the serum (19) and airways (20) of patients with asthma, and it correlates with Th2 cytokine production in the lung (20). Lastly, selective blockade of sIL-6R in mice suppressed IL-4, IL-5, and IL-13 production and decreased eosinophil numbers in the lung; on the other hand, blockade of sIL6-R plus the membrane-bound form of the receptor (mIL-6R) caused not only the suppression of Th2 cytokine production but also expansion of CD4+CD25+ Tregs in the lung with increased IL-10 production and suppressive capacity (20). Thus, together with our results, these data suggest that rs4129267, or a causal variant in linkage disequilibrium with it, increases the risk of developing asthma by up-regulating protein concentrations of sIL-6R or mIL-6R, or both, which in turn contributes to the development and maintenance of a Th2 immune response in the lung. An IL-6R antagonist (tocilizumab) has been approved as an effective biologic drug to treat rheumatoid arthritis (21); further studies are warranted to test the hypothesis that tocilizumab might also be effective to treat asthma, particularly for patients with the rs4129267:T risk variant.
Hundreds of studies have attempted to map genes that contribute to the risk of developing asthma. Most linkage and candidate-gene association studies were underpowered to detect genetic loci with effect sizes now considered realistic for common diseases (36) and so often the identified regions or genes failed to replicate in independent studies. As a result, prior to 2007, over 100 loci had been implicated in asthma causation (37), but which, if any, represented genuine risk factors was largely unclear. Since 2007, genome-wide association studies (GWAS) have contributed significantly to change this landscape. In the largest GWAS of asthma published to date, the GABRIEL consortium (5) identified six loci associated with asthma risk at the genome-wide significance level: IL18R1, HLA-DQ, IL33, SMAD3, the ORMDL3 locus, and IL2RB. Given the size of that study, these represent the most convincing genetic risk factors for asthma reported until now. However, each explains only a small fraction of the disease heritability, indicating that many more risk loci remain to be identified. This can be achieved by expanding and combining existing asthma GWAS, as we have done in this study.
Our study identifies two additional loci with genome-wide significant association with asthma risk: IL6R and chromosome 11q13.5. The IL6R finding raises the possibility that an approved IL6R antagonist may be effective to treat asthma. We also provide independent support for four of the six GABRIEL loci: IL18R1, IL33, ORMDL3 and IL2RB. Taken together, these results support the hypothesis that a genetic dysregulation of cytokine signalling increases disease risk.
The second locus that our analyses implicate in the causation of asthma is represented by rs7130588, which is located on chromosome 11q13.5 near a SNP recently reported to associate with two immune-related diseases, Crohn’s disease (CD) (22) and atopic dermatitis (AD) (23). The variant reported to associate with Crohn’s disease and atopic dermatitis (rs7927894) is in complete linkage disequilibrium (LD) (r2 = 1) with rs7130588, and the two predisposing alleles (rs7927894:T and rs7130588:G) occur on the same haplotype. This indicates that the same underlying causal variant is likely to explain the association between this locus and the three diseases. Each copy of the rs7130588:G allele increases the risk of atopic dermatitis by 1.22-fold (23), Crohn’s disease by 1.16-fold (22) and asthma in our study by 1.09-fold. The association with atopic dermatitis and our finding that rs7130588:G does not seem to increase asthma risk in non-atopic individuals collectively suggests that this allele directly increases the risk of allergic sensitisation and, if this develops, it increases the risk of subsequently developing asthma. This mechanism is consistent with the epidemiological observations that sensitization, eczema and allergic rhinitis often precede the development of asthma symptoms (24-27). While this work was under review, Marenholz and colleagues (38) also reported an association between 11q13.5 and asthma, although not at the genome-wide significance level. Results from that study are also consistent with an effect for the 11q13.5 locus that is specific to allergic asthma. The leucine rich repeat containing 32 gene (LRRC32) is a plausible causal candidate in this region, as it is expressed in activated Treg cells (28), of which numbers and immune suppressive function appear to be impaired in asthma (29).
Our analyses also provided independent evidence for association with four of the eight loci previously reported to associate with asthma risk (5) at the genome-wide significance level, specifically with the ORMDL3 locus, IL18R1, IL33, and IL2RB (5). The reported SNPs for HLA-DQ and SMAD3 had consistent but non-significance evidence for association in our study; these results suggest that they might represent true risk factors for asthma in the population we studied, but have an associated risk that is lower than originally reported (5). On the other hand, we found no supporting evidence for an association with PDE4D (2) or DENND1B (3) variants, despite appropriate power. Larger meta-analyses of available GWAS are in progress and will be able to study these loci in greater detail.
We also investigated the hypothesis that the genetic component of asthma risk includes a highly polygenic contribution, with hundreds or potentially thousands of variants that individually explain only a very small fraction of the disease heritability, as suggested for other traits such as schizophrenia (30) and height (31). Consistent with this hypothesis, we observed that quantitative genetic scores that represented the combined effect of thousands of common SNPs – each individually influencing asthma risk only weakly (e.g., median OR = 1.05 for Group 3 in Figure 2) in the GABRIEL study (5) – were significantly associated with asthma case-control status in our study. These results thus suggest that many of these SNPs are either in high LD with true causal variants that are common in the population but influence disease risk weakly, or they are in low LD with causal variants that are rarer but increase disease risk more strongly. The presence of both types of variants is not mutually exclusive; very large sample sizes will be needed to identify these through GWAS with genome-wide significance. The multi-SNP scores tested here, despite being significantly associated with asthma risk, provided low discrimination in disease status (i.e., low sensitivity and specificity) and so currently have little or no diagnostic utility per se, consistent with recent findings (30, 32, 33). However, as larger asthma GWAS are conducted, SNP effects will be estimated more precisely, and this will improve the discrimination accuracy of genome-wide SNP scores.
Lastly, our analysis of SNPs previously reported to associate with immune or inflammatory diseases identified several loci that may represent genuine asthma risk factors with weak effects. These include variants near plausible functional candidates, such as NDFIP1, which causes severe inflammation of the skin and lung when knocked-out in mice (34) and BACH2, a B-cell-specific transcription repressor that is a key regulator of antibody response (35). Further studies are required to confirm these as genuine asthma risk loci.
When interpreting results from this study it is important to recognise that we used a broad definition of asthma status, which grouped together not only clinically diagnosed and self-reported physician-diagnosed cases, but also groups of cases that were probably exposed to different environmental risk factors. Similarly, asthma controls included both asthma-free and asthma-unknown individuals, with limited information on atopic status. As a result, our primary association analysis provided improved power to detect risk loci with homogeneous effects across different asthma subtypes but was likely underpowered to detect loci with subtype-specific effects. Furthermore, a large fraction of SNPs discovered by the 1000 Genomes Project were imputed with modest confidence. For these SNPs, power to detect an association with asthma will also have been reduced.
In conclusion, we identified novel variants in IL6R and chromosome 11q13.5 with genome-wide significant association with asthma. The IL6R findings further support the hypothesis that a genetic dysregulation of cytokine signalling increases disease risk and raise the possibility that tocilizumab may be effective to treat asthma, perhaps in a genotype-dependent manner; studies that address this possibility are warranted. At this stage, it is unclear which gene underlies the association with 11q13.5. Given that no specific gene in this region has been directly implicated in allergic disease previously, further characterisation of this region of association is likely to discover novel molecular mechanisms involved in the causality of eczema, atopy and asthma. Lastly, our results are consistent with the contribution of hundreds or potentially thousands of variants with weak effects on asthma risk, which can be identified through larger GWAS as already shown for other diseases (13).
The Australian Asthma Genetics Consortium was supported by the Australian National Health and Medical Research Council (613627). Detailed acknowledgements are provided in the webappendix (pp 27-29).
Collaborators of the Australian Asthma Genetics Consortium Désirée Mészáros1, Mary Roberts2, Melissa C. Southey3, Euan R. Tovey4, Nicole M. Warrington5, Mathijs Kattenberg6, Lyle J. Palmer7,8, Loren Price9, Margaret J. Wright10, Scott D. Gordon10, Li P. Chung9, Anjali K. Henders10, Graham Giles11, Jouke-Jan Hottenga6, Paul S. Thomas12, Suzanna Temple9, John B. Whitfield10, Ian Feather13, Anna-Liisa Hartikainen14, Tari Haahtela15, Tarja Laitinen16, Pekka Jousilahti17, Johan G. Eriksson18-21, Elisabeth Widen22, Olli T. Raitakari23,24, Terho Lehtimäki25,26, Mika Kähönen26,27, Jorma Viikari28,29, Aarno Palotie22,30-32, Zofia K. Gajdos33-35, Helen N. Lyon34,35, George T. O’Connor36,37, Stephen Morrison38, Peter D. Sly39, Chalermchai Mitrpant9, Warwick J. Britton40, David John1, Pat G. Holt5, Andrew S. Kemp41
1Menzies Research Institute, Hobart, Australia.
2Department of Respiratory Medicine, Royal Children’s Hospital, Parkville, Australia.
3Department of Pathology, The University of Melbourne, Melbourne, Australia.
4Woolcock Institute of Medical Research, University of Sydney, Sydney, Australia.
5Telethon Institute for Child Health Research, Centre for Child Health Research, University of WA, Perth, Australia.
6Netherlands Twin Register, EMGO & NCA Institute, Department of Biological Psychology, VU University, Amsterdam, The Netherlands.
7Genetic Epidemiology and Biostatistics Platform, Ontario Instiute for Cancer Research, Toronto, Canada.
8Samuel Lunenfeld Research Institute, Unviversity of Toronto, Toronto, Canada.
9Lung Institute of WA and Centre for Asthma, Allergy and Respiratory Research, University of WA, Perth, Australia.
10The Queensland Institute of Medical Research, Brisbane, Australia.
11Cancer Epidemiology Centre, The Cancer Council Victoria, Melbourne, Australia.
12Faculty of Medicine, University of New South Wales, Sydney, Australia.
13Gold Coast Hospital, Southport, Australia.
14Department of Clinical Sciences, Obstetrics and Gynecology, Institute of Clinical Medicine, University of Oulu, Oulu, Finland.
15Skin and Allergy Hospital, Helsinki University Hospital, Helsinki, Finland.
16Deptartment of Pulmonary Diseases and Clinical Allergology, Turku University Hospital and University of Turku, Finland.
17Department of Chronic Disease Prevention, National Institute for Health and Welfare, Finland.
18National Institute for Health and Welfare, Finland.
19Department of General Practice and Primary health Care, University of Helsinki, Finland.
20Helsinki University Central Hospital, Unit of General Practice, Helsinki, Finland.
21Folkhalsan Research Centre, Helsinki, Finland.
22Institute for Molecular Medicine Finland (FIMM), University of Helsinki, Helsinki, Finland.
23Research Centre of Applied and Preventive Medicine, University of Turku, Turku, Finland.
24Department of Clinical Physiology, Turku University Hospital, Turku, Finland.
25Department of Clinical Chemistry, University of Tampere, Tampere, Finland.
26Tampere University Hospital, Tampere, Finland.
27Department of Clinical Physiology, University of Tampere, Tampere, Finland.
28Turku University Hospital, Turku, Finland.
29Department of Medicine, University of Turku,Turku, Finland.
30Wellcome Trust Sanger Institute, Wellcome Trust Genome Campus, Hinxton, United Kingdom.
31Program in Medical and Population Genetics and Genetic Analysis Platform, The Broad Institute ofMIT and Harvard, Cambridge, United States.
32Department of Medical Genetics, University of Helsinki and University Central Hospital, Helsinki, Finland.
33Department of Genetics, Harvard Medical School, Boston, United States.
34Divisions of Genetics and Endocrinology, Children’s Hospital, Boston, United States.
35Broad Institute, Cambridge, United States.
36Pulmonary Center, Department of Medicine, Boston University School of Medicine, Boston, United States.
37The National Heart, Lung, and Blood Institute’s Framingham Heart Study, Framingham, United States.
38University of Queensland, Brisbane, Australia.
39Queensland Children’s Medical Research Institute, Brisbane, Australia.
40Centenary Institute and University of Sydney, Camperdown, Australia.
41The Children’s Hospital, Westmead, Sydney, Australia.
Contributors Writing group: M.A.R.F, M.C.M., D.L.D., G.B.M., P.LeS, S.D., M.J.A., P.J.T., G.W.M., M.A.B.
Analytic group: M.A.R.F., M.C.M., D.L.D., S.B., W.A., N.M.W., P.D., S.G., M.K., A.R., S.V., J.H., M.R., M.A.B., P.M.V.
Project management: M.A.R.F., M.C.M., D.L.D., G.B.M., J.H., P.LeS, S.D., M.J.A., M.A.B., G.W.M., P.J.T.
Phenotype and genotype collection for QIMR studies: D.L.D., M.A.R.F., D.R.N., S.D.G., M.J.W., A.K.H., J.B.W., P.A.M., A.C.H, G.W.M., N.G.M.
Phenotype and genotype collection for TAHS: D.M., M.C.S., G.G., P.S.T., I.F., S.M., M.J., M.J.A., J.L.H., H.W., M.C.M., S.C.D.
Phenotype and genotype collection for MESCA: C.H., M.R., P.LeS., C.R.
Phenotype and genotype collection for CAPS: G.M., A.J.K., E.R.T., W.J.B., G.J., L.R.S.
Phenotype and genotype collection for RAINE: C.P., P.D.S., P.G.H.
Phenotype and genotype collection for NTR:G.W., E.J.C.deGeus, M.K., JJ.H., D.I.B.
Phenotype and genotype collection for APCAT: V.S., M-R.J., O.T.R., J.V., T.L., M.K., J.E.
Phenotype and genotype collection for Busselton: J.H., J.B., L.J.P., A.J., B.M.,
Genotyping for AAGC: P.D., M.A.B.
Phenotype and genotype collection for LIWA: S.B., F.C., L.P.C., S.T., C.M., B.S., L.P., P.J.T.
Conflicts of interest MAF owns stock in Illumina, Roche and Complete Genomics. PST’s institution has received consulting fees from GSK as a result of activities outside this work. PST has received lecture fees from GSK and Novartis. GTO has received consulting fees from Novartis and Sunovion, Inc, as a result of work unrelated to this study. GBM was a board member of Novartis; GBM’s institution received fees for lectures and preparation of educational presentations from Boehringer Ingelheim and AstraZeneca. CFR received consulting fees from GSK and Merck as a result of activities outside this work. ASK’s institution received speaker’s fees from Nutricia. All other authors declare that they have no conflicts of interest.