Defining the contribution of SNPs identified in asthma GWAS to clinical variables in asthmatic children

Asthma is a complex respiratory disease with susceptibility involving the interplay between genetic and environmental factors [1]. The chronic inflammation of the airways is characterised by reversible airflow obstruction and increased hyperresponsiveness of the airways. Genome-wide association studies (GWAS) are currently the preferred method for the identification of genetic determinants in complex multifactorial diseases such as asthma. These studies have the advantage of being able to identify polymorphisms with small effect sizes and localise smaller susceptibility regions due to linkage disequilibrium (LD). The first GWAS for asthma was completed involving 994 childhood-onset asthma patients and 1,243 non-asthma controls and identified single nucleotide polymorphisms (SNPs) within chromosome 17q21 spanning the ORMDL3/GSDMB genes [2]. These SNPs showed highly significant association with childhood asthma and were also replicated in two independent cohorts [2]. A study in an Icelandic cohort showed significant association with asthma and a SNP in IL1RL1[3] and SNPs in PDE4D have also shown association with genome-wide significance with asthma [4]. IL-13 and RAD50 on 5q31.1 and the HLA-DR/DQ locus at 6p21.3 have been implicated by Li et al. [5]. Other polymorphisms spanning IL33, IL18R1, SMAD3, IL2RB and CRB1 have been proposed by other GWAS using large scale discovery and replication approaches [6, 7]. However, due to the phenotype used for case definition, which is often a physician diagnosis or self-report of asthma, there remains a need to determine the contribution of these polymorphisms to clinically relevant endpoints in asthma. This will provide a greater insight into the mechanisms altered in individuals carrying the relevant risk alleles.

One problem that arises from GWAS is that LD patterns can result in multiple polymorphisms at a locus being associated with disease phenotype, even if only one of them is causal. A problem therefore exists in the identification of potential causal variants. This can be simply analysed for polymorphisms within coding regions, as they could potentially affect the polypeptide structure and function and those SNPs in regulatory or intronic regions could affect splice site motifs, transcriptional regulation or RNA stability [8, 9]. However, many GWAS have implicated polymorphisms in non transcribed regions far away from annotated genes, where their role is likely to be regulatory [10]. The ENCODE project was established to identify all the functional elements such as protein binding sites, transcripts (coding and non coding) and chromatin marks in the human genome. This represents a valuable resource to provide additional meaning to association data provided from GWAS [11, 12]. Taken in consideration with LD data from the 1000 Genomes project [13], computational analysis will enable assignment of more meaningful potential function to SNPs identified from GWAS or those SNPs which are in LD with a GWAS associated SNP [10], to help better prioritise SNPs for functional analysis.

GWAS have been successful in identifying susceptibility genes for asthma through the use of large scale population cohorts and replication approaches [2–7]. However, there remains a need to determine the contribution of these polymorphisms to clinically relevant endpoints in asthma as opposed to disease diagnosis. We adopted a two step process where we selected SNPs from key genes in the Caucasian population which have met genome-wide significance and/or with replication, namely eleven SNPs in the following genes; ORMDL3/GSDMB, IL33, IL18R/IL1RL1, SMAD3, IL2RB, PDE4D, CRB1 and RAD50. These SNPs were genotyped in a UK asthma family cohort with at least two affected siblings, to test for association with clinically relevant endpoints in asthma. Our data provide support for the original GSDMB association; however SNPs in the other genes did not show association after correction for multiple testing. Secondly, as these SNPs have previously shown association at a genome-wide level, we also investigated the potential for these SNPs and those in LD to affect chromatin states and alter regulatory motifs/binding sites to try to assign potential function to these associations. This analysis used recently reported data from the ENCODE project. These data support the hypothesis that SNPs associated in GWAS may be tagging SNPs for the actual causative variant, but also provide tentative evidence that these genotyped SNPs may themselves play a functional role e.g. GSDMB rs2305480 (Ser311Pro).

GWAS have the advantage over linkage studies in that they can identify polymorphisms with small effect sizes and localise smaller susceptibility loci as LD only generally spans <500 kb [22]. Since the identification of childhood asthma associated variants on chromosome 17q21 [2], many other GWAS have been completed in different populations implicating variants spanning different genes (see above). One problem with these studies lies in defining the contribution of these associated variants to clinical endpoints in asthma. To this end we have completed a SNP for SNP analysis of key variants identified in the Caucasian population meeting genome-wide significance or where there was replication, focussing on GWAS from 2007–2011 (Table 1). We utilised a well characterised UK family based asthma cohort [15] to test for association with asthma diagnosis and the following clinical endpoints: FEV₁ (% predicted), BHR (to methacholine), BTS defined severity and positive SPT to one or more allergens, see [15]. These data show significant association for the two GSDMB SNPs; rs2305480 (Ser311Pro) and rs11078927 (intron) across more than one phenotype, with other phenotypes showing association which does not reach significance after correction for multiple testing. These data both replicate and extend previous findings [2, 6] and identify this chromosome 17 locus as containing genetic determinants underlying multiple clinical features of asthma, including FEV₁ (% Predicted) and disease severity defined by BTS. These two polymorphisms are in high LD suggesting these associations represent a single locus. The BHR association (rs2305480) is supported by different SNPs in this gene also showing association with this phenotype, albeit in different ethnic populations [23, 24]. Interestingly, a recent study has found association between rs2305480 and wheezing phenotypes in asthmatic children, but did not find association with intermediate phenotypes such as atopy or lung function [25].

When interpreting the results of GWAS, the associated variant could be tagging another variant in a larger region of LD [26]. This is particularly important as it is common that only a few variants in a haplotype block are present on genotyping platforms. Methods such as imputation and using genotyping data from the 1000 Genomes project can help to further characterise these patterns [27]. However, once these variants are identified it is still necessary to assign potential function and with many associated variants from GWAS residing in non-transcribed regions, they are likely to play a regulatory role in gene expression. We used HaploReg [14] to screen chromatin data, conservation data and regulatory motif databases which was obtained from the ENCODE project to see if any biological plausible role could be assigned to the SNPs genotyped in this study and those SNPs which were in LD with r² = 1. Our results show that these SNPs in LD can potentially lead to changes in regulatory element binding motifs affecting the transcription of genes positively or negatively and there is also the suggestion that alterations in enhancer histone marks or DNase sites could be affected. The non-synonymous SNP (Ser 311 Pro) is predicted to be deleterious to protein structure (Polyphen) in Gasdermin B suggesting a potential functional mechanism altering protein function. This may be occurring at the mRNA level (altering the efficiency of splicing) as this polymorphism is predicted to be located in an exonic splicing enchancer site (FuncPred). Both GSDMB SNPs are in high LD with several SNPs spanning ZPBPZ and GSDMB which result in alterations in enhancer and transcription factor binding sites. However the rs2305480 and rs11078927 SNPs themselves are not predicted to have any regulatory role (ENCODE data). Gasdermin-family proteins have been implicated in TGFβ1 signalling and epithelial cell apoptosis [28], with increased airway cell apoptosis reported in severe asthma [29]. The extended chromosome 17 locus has also been implicated previously as containing SNPs associated with asthma spanning three main genes; ZPBP2/GSDMB/ORMDL3 and it is still unclear which genes/polymorphisms underlie these associations. The finding that the GSDMB SNPs including a non-synonymous SNP are particularly relevant to clinical endpoints including lung function, BHR and severity in our childhood asthma cohort is interesting as few recent studies have genotyped this GWAS SNP or investigated clinical outcomes with the same criteria such as BTS used in this study [24, 30]. However, a recent study has investigated the effect of rs7216389 (which is in linkage disequilibrium with both GSDMB SNPs in this study) with respect to severe asthma subjects classified with poorly controlled disease whilst taking high doses of inhaled corticosteroids, long-acting bronchodilators and short acting β₂ agonists and showed association [31]. Similarly, while not meeting genome-wide association significance, we have previously provided evidence that rs2305480 is associated with severe asthma (p = 5.5×10^-5, [32]), although we do acknowledge that the criteria for BTS severity is defined in part by lung function and so this association with severe asthma may be related to a synergistic effect. These data support previous studies that identified the chromosome 17 locus as containing markers of childhood onset disease [2, 6, 24, 30], BHR [24] and elevated total serum IgE [24] in diverse ethnic populations.

Interestingly, no other SNP survived correction for multiple testing for any phenotype analysed, including asthma diagnosis and there was no indication of any other SNP showing a trend towards significance. CRB1 rs2786098 (p = 0.089 in the current study) has shown replicated association in multiple childhood cohorts, but did not reach statistical significance in our study. Perhaps this lack of association may be due to power (MAF 0.22, see earlier). One potential explanation for the lack of association for asthma and related phenotypes for the majority of SNPs is that many of the SNPs/genes were identified in adult asthma populations and the relative contribution to childhood asthma remains unclear at this time. However, in the results not meeting conventional correction for multiple testing, IL33 SNP rs1342326G (25.7 kb from gene transcription start site (TSS)) showed a level of association with atopic asthma defined by positive skin prick test and clinical diagnosis (p = 0.0097, Table 4). These data are in line with the accumulating evidence suggesting a role of IL33 in allergic mechanisms in asthma (Reviewed in [33]) although we cannot exclude that this is a false positive. Similarly, while not surviving correction, suggestive evidence for SNP associations for CRB1/BHR, IL18R/severity and IL33/SPT were apparent (p < 0.05), which may represent true associations.

Studies have shown the potential for GWAS SNPs to be tagging SNPs associated with eQTL [10]. Our analyses did provide some support for this particularly for the two GSDMB SNPs (rs2305480 and rs11078927) when eQTL analyses was completed across multiple cell types and tissues. These eQTL analyses identified regulation of multiple genes in the region, demonstrating the need for further work to define the biology underlying these genetic signals.

It is important to note the limitations of our study, while a SNP for SNP approach focussed to the pivotal SNPs from the GWAS has strengths to directly replicate and extend previous findings we acknowledge the assumption that the linkage disequilibrium pattern between the discovery Caucasian population and our UK population will be the same, which may at least in part explain the inability to detect associations with all variants. Similarly, we did not genotype all GWA significant SNPs identified in each of the regions. We acknowledge the reduced power of the current study using 370 families is a limitation potentially explaining while several SNPs showed nominal significance and did not meet stringent statistical criteria. A cohort of 602 families would be required to provide a power of 0.99 for the lowest MAF observed in the current study (relative risk of 1.5 and significance level of 0.05). We also acknowledge that the design of this study compared to discovery cohorts using case/control design also has reduced power, however we consider the use of an alternative association design i.e. family based to those previously published a strength. Furthermore the lack of assessing gene-environment interactions in GWAS and the current study may also account for not replicating previous observed effects due to differential environmental exposures between study populations. Finally, the potential for winners curse bias is also possible for direct replication of e.g. association with asthma diagnosis.

In summary, this study has provided a greater insight into the relative contribution of asthma GWAS SNPs to clinically relevant endpoints in asthmatic children and further defines the diverse role of GSDMB SNPs particularly in lung function and disease severity.