Asthma and genes encoding components of the vitamin D pathway

Subjects were from the Saguenay^_Lac-Saint-Jean (SLSJ) asthma study, which consists of French-Canadian families ascertained through asthmatic probands. Probands were included in the study if they fulfilled at least two of the following criteria: 1) a minimum of three clinic visits for acute asthma within one year; 2) two or more asthma-related hospital admissions within one year; or 3) steroid dependency, defined by either six month's use of oral, or one year's use of inhaled corticosteroids. A total of 1064 individuals from 388 nuclear families were included in the present analyses. Families were included in the study if at least one parent was available for phenotypic assessment, at least one parent was unaffected, and all four grandparents were of French-Canadian origin. Family members were considered asthmatics if both a self-reported history of asthma and a history of physician-diagnosed asthma were recorded, or by clinical evaluation following a methacholine provocation test. Skin-prick tests were performed for 26 inhalant allergens and subjects were considered atopic if they had at least one positive response (wheal diameter ≥ 3 mm at 10 min) [25]. Spirometry, methacholine challenge and IgE measurements are described in detail elsewhere [21]. Table 1 presents the characteristics of the subjects. The SLSJ local ethics committee approved the study, and all subjects gave informed consent.

Data from the Canadian Asthma Primary Prevention Study (CAPPS) study, the Study of Asthma Genes and the Environment (SAGE) birth cohort and the Busselton Health Study (BHS) were used to replicate the findings. The Childhood Asthma Management Program (CAMP) study was also used to replicate a specific gene-gene (VDR-IL10) interaction models. The CAPPS and SAGE studies have been described elsewhere [26]. Briefly, the CAPPS study was initiated in 1995 to assess the effectiveness of a multifaceted intervention program in the primary prevention of asthma in high-risk infants [27, 28]. High-risk infants were identified before birth as having at least one first-degree relative with asthma or two first-degree relatives with other IgE-mediated allergic diseases. A total of 549 children and their parents forming 545 families were enrolled in the study during the second and third trimester of pregnancy. The children were followed since birth and were assessed by a pediatric allergist for the presence of asthma and allergies. Atopy was defined by skin-prick test. A total of 16 allergens were tested and the diagnosis was positive if at least one wheal ≥ 3 mm than the negative control was observed. Children with 7 year follow-up data and DNA were included in the current study (380 children/families). The SAGE study is a population-based cohort of 16,320 children born in the province of Manitoba, Canada, between January 1, 1995 and December 31, 1995. Parents of these children were first survey by mail in 2002. A subset of children was then invited to join the study at age 8-10 years. This subset included children with parent-declared asthma and children without asthma. A total of 723 families were recruited into the study. All recruited children underwent clinical assessment of asthma by a pediatric allergist. Skin prick testing for 16 allergens was used to define atopy. In the two latter studies (SAGE and CAPPS), children affected with asthma/atopy and their parents were genotyped and analysed in trios. In contrast, the BHS was analysed using a case-control design. This study comprised a series of six cross-sectional health surveys that took place every three years from 1966 to 1981 in all adults and children residing in the Shire of Busselton, Western Australia and a follow-up study of all previous participants (residing within and outside this Shire) in 1994/1995. Busselton is a coastal town in the South West region of Western Australia with a population that is predominantly of European origin. In the present case-control study, all subjects (n = 1395, 751 controls and 644 cases) who attended both the 1981 and the 1994 survey and who had a diagnosis of asthma as well as available DNA were included. Subjects were considered to have asthma if they answered yes to the question "Has your doctor ever told you that you had asthma/bronchial asthma?" in a written questionnaire at either survey. Subjects were considered controls if they answered no at both surveys. Skin prick testing for 12 allergens was used to define atopy. Finally, CAMP is a multicentered North American clinical trial designed to investigate the long-term effects of inhaled anti-inflammatory medications in children with mild to moderate asthma [29, 30]. A total of 1625 individual members of 428 non-Hispanic white nuclear families were included in the present analyses. This represents the subset non-Hispanic with CAMP families with available SNP genotype data at both the VDR and IL10 loci. The diagnosis of asthma was based on a methacholine provocation test and one or more of the following criteria for at least 6 months in the year before recruitment: 1) asthma symptoms at least two times per week, 2) at least two uses per week of an inhaled bronchodilator, and 3) daily asthma medication. A local ethics committee approved the protocol independently in each study. Written informed consent was obtained from all study participants.

Eleven candidate genes were chosen for this study. Figure 1 is a cartoon of the vitamin D pathway that illustrates the implication of each gene selected. Briefly, genes encoding key components of the vitamin D pathway were chosen, which include: enzymes responsible for the activation and inactivation vitamin D (CYP27A1, CYP27B1, CYP2R1 and CYP24A1) [31, 32]; the vitamin D binding protein (GC) that binds to vitamin D and its plasma metabolites and transports them to target tissues; SKIIP, also known as NCoA62/SKIP, that serves as a coactivator a vitamin D-mediated transcription [33]; and five revevant genes for asthma that are known to be transcriptionally regulated by vitamin D (IL10, IL1RL1, CD28, CD86 and IL8) [14].

SNPs were selected using the CEPH genotype dataset from phase 1 of the International HapMap project [34]. The genotype data were downloaded from the genomic region covering ten kilobases up- and downstream of each gene. A maximally informative set of SNPs was selected using a pairwise tagging algorithm described by Carlson et al. [35]. A Perl program, called ldSelect http://​droog.​gs.​washington.​edu/​ldSelect.​html, was used to select the SNPs in each gene. Briefly, this program analyzes the pattern of LD between SNPs and forms bins of SNPs in LD based on an r² threshold. The algorithm ensures that all pairwise LD values between SNPs in the same bin exceed the r² threshold. Accordingly, any SNP in a bin can serve as a proxy (tagSNP) for all other SNPs in the same bin. Only one tagSNP needs to be typed per bin. At this level, nonsynonymous SNPs genotyped in the HapMap dataset were prioritized using the "-required" option. Similarly, some SNPs were prioritized based on the type of variation (A/T, C/T, etc) to meet the genotyping technology requirement. The minor allele frequency and the r² thresholds were set at 0.05 and 0.8, respectively, using the "-freq" and "-r2" options. Known nonsynonymous SNPs or functional variants not genotyped in the HapMap dataset were also selected for genotyping. Selected SNPs and their characteristics are shown in additional file (see Additional file 1). The location of SNPs relative to the gene structure is illustrated in Additional file 2.

In the SLSJ study, a total of 87 SNPs were genotyped using the SNP stream^® UHT technology [36]. Primers were designed using FastPCR version 3.8.78 for multiplex PCR [37]. Single base extension primers were designed using Autoprimer.com (Beckman Coulter). The protocol and reaction conditions were performed in accordance with the manufacturer [36]. SNPs were genotyped in different panels that were organized by grouping SNPs with the same type of variation (A/T, C/T, etc) and by respecting the 12-plex maximum capacity of the system. For the replication studies (CAPPS, SAGE and BHS), 52 SNPs located in five genes were genotyped using the Illumina GoldenGate assay [38] as part of a larger SNP genotyping panel http://​www.​genapha.​ca. SNP genotypes in CAMP available from prior analyses were generated for VDR [22] and IL10 [39] using the MassARRAY platform (Sequenom, San Diego, CA) and SNaPShot (Applied Biosystems, Forrest City, CA), respectively, as previously described.

Mendelian inheritance incompatibilities were inspected using Pedmanager version 0.9 and Hardy-Weinberg equilibrium was evaluated using a χ² test among parents. For the SLSJ, CAPPS and SAGE studies the Family Based Association Test (FBAT) program was used to test association with single SNPs [40]. All tests were performed with an additive model using the empirical variance-covariance estimator that adjusts for the correlation among sibling genotypes and for multiple nuclear families within a single pedigree. The FBAT test provides a Z-statistic with the corresponding p value. A positive Z-statistic is indicative of a high-risk allele and a negative Z-statistic is indicative of a protective allele. In the BHS, the single SNP associations were evaluated using the Cochran-Armitage test for trend with additive coding of alleles. Genes showing at least one SNP with a p < 0.05 in the SLSJ collection were considered for validation in the other populations. Our strategy to deal with multiple testing was to replicate the associations in independent populations instead of using an adjusted p value. LD values were evaluated using the r² metrics and calculated with Haploview 3.32 [41]. Power calculations for the four study populations were recently described [42].

Gene-gene interactions were evaluated for asthma and atopy using a multilocus analysis method following the framework described in Millstein et al. [23]. This strategy is based on likelihood-ratio tests that used a log-additive coding scheme, where genotypes aa, Aa, and AA are coded as 0, 1, and 2, respectively. Briefly, the analyses were performed in two stages. In stage one, single SNP tests of associations were evaluated by contrasting the null hypothesis of no association with the alternative hypothesis. The threshold for significance for stage one of the gene-gene interaction tests was then adjusted for multiple testing using the Bonferroni correction. In stage two, a full two-gene interaction model, including the two main effects and the interaction term, was tested against the reduced model that includes only the main effects that were declared significant in the first stage, if any. This strategy avoids retesting the same effects detected in stage one. While the framework of Millstein et al. [23] was described for a case-control dataset, it can easily be adapted to a case-parent design by following the case/pseudocontrol design described by Cordell et al. [43], where each case is matched with three pseudocontrols derived from the untransmitted parental alleles. Following this, conditional logistic regression is used to assess the significance of the main and interaction terms. Since transmission to multiple affected siblings cannot be assumed to be independent events, and since the families in our sample may contain more than one case, robust estimates for the variance and Wald tests were used instead of likelihood ratio tests for the SLSJ, CAPPS and SAGE studies. Considering the number of SNPs genotyped in genes involved in the vitamin D pathway, a total of 5003 two-gene interaction models were evaluated.

Post hoc analyses were performed with the combined dataset (SLSJ, CAPPS, SAGE, and BHS). Tests of association were performed using the likelihood method implemented in UNPHASE v3.0.10 [44], which allow data from family studies and case-control individuals to be analyzed together.

Additional file 1 presents the 87 genotyped SNPs and their characteristics in the SLSJ study. Two SNPs failed the assay design including one in the CD86 gene (rs1915087) and another in the GC gene (rs1491711). Both of these SNPs are singletons and are not tagging other SNPs in the genes. A third SNP (rs8176353) located in the CYP27B1 gene was monomorphic. Additional file 1 also showed the minor allele frequencies for a reference population (CEPH from HapMap) and for the SLSJ study. In most cases, the minor allele frequencies were very similar between the two populations with a mean difference of 2% and the largest difference was 16% for SNP rs4308217 located in the CD86 gene. After Bonferroni correction, only one SNP was out of Hardy-Weinberg equilibrium (see Additional file 1). This SNP (rs4809960) is a singleton in the CYP24A1 gene and was removed from further analyses. Accordingly, a total of 83 SNPs were tested for association with asthma and atopy. Additional file 2 shows the exon-intron structure of each gene and the location of genotyped SNPs.

The overall distribution of single marker FBAT association tests shows a greater number of small p values for asthma compared to what was expected by chance (see Additional file 3). Results for genes with at least one significant p value for asthma and atopy are illustrated in Figure 2 (results for all genes are illustrated in Additional file 2). The details of these tests are shown in Table 2 for SNPs having at least one p value < 0.1 for asthma or atopy (FBAT results for all SNPs can be found in Additional file 4). Five SNPs in the IL10 gene, three located in the promoter (rs1800872, rs1800871, rs1800896), one in intron 1 (rs3024490), and the other located in the 3' region (rs4844553) were significantly associated with asthma. Three of them (rs1800871, rs1800872, and rs3024490) were in tight LD (r² > 0.97), while the others were in low to modest LD (Figure 3). Haplotype analysis for the three tightly linked SNPs revealed the presence of only two haplotypes with an allele frequency above 1% in the SLSJ population. The TAA haplotype had a frequency of 0.268 and was overtransmitted to asthma patients (p = 0.024), while the CCC haplotype had a frequency of 0.726 and was undertransmitted (p = 0.013). Four out of the eight genotyped SNPs in the CYP24A1 gene were also modestly associated with asthma or atopy (range of p values = 0.051 to 0.015). The two intronic SNPs associated with atopy (rs912505 and rs927650) were in modest LD (r² = 0.36) and the two SNPs associated with asthma (rs2248359 and rs8124792) located in the promoter and the 3' region of the gene were in complete equilibrium (r² = 0) (Figure 3). Three intronic SNPs in the IL1RL1 gene were also associated with asthma. Two of them (rs1420089 and rs1861245) were in modest LD (r² = 0.28) and the third one (rs1946131) showed no LD with the other (Figure 3). Also worth mentioning is a SNP (rs2715267) in the promoter region of the CD86 gene that was significantly associated with atopy (p = 0.004). A trend for this SNP was also observed for asthma (p = 0.069). Finally, one SNP (rs11023374) in intron 2 of the CYP2R1 gene was associated with asthma (p = 0.017). LD plots for all genes are illustrated in Additional file 5.

Interaction among functionally related genes may not be surprising. Hence, all possible two-gene interactions were tested for asthma in the SLSJ study for the 11 genes under study plus the VDR gene (Figure 4). Two concentrated spots of significant two-gene models for asthma are observed in this figure. The one at the bottom represents two-gene models involving SNPs in the IL10 and VDR genes. The second spot located in the center of Figure 4 represents two-gene models involving SNPs in the IL10 and IL1RL1 genes. Multiple two-gene models were also significant between SNPs in the IL10 and CD86 genes. Additional file 6 shows the two-gene models for atopy in the SLSJ study. Overall, gene-gene interactions were modest for atopy.

To understand the impact of these two-gene models on the risk of asthma, the genotype by genotype odds ratio matrix was calculated and some representative and most significant two-gene models are illustrated in Figure 5. Figure 5a, b and 5c show two-gene models between SNPs in the IL10 and VDR genes. Figure 5a shows that the risk of having asthma is similar for carriers of two rare IL10 alleles irrespective of the VDR genotypes. However, the risk increases with the increasing number of common IL10 alleles for individuals who are homozygous for the common VDR allele. In contrast, the risk tends to decrease with the increasing number of common IL10 alleles for individuals who are homozygous for the rare VDR allele. Figure 5b shows that the risk of asthma increases with the number of rare IL10 alleles, but the effect is greater with an increasing number of common alleles at the VDR locus. Figure 5a and 5b show representative interactions between SNPs located in the promoter region of IL10 and the 3'UTR region of VDR. However, more complex interactions between these two genes were observed between SNPs located in the 3'UTR region of both genes. Figure 5c shows that the effect of the IL10 rare allele goes in the opposite directions depending on whether subjects are homozygous for the common or the rare VDR alleles. Figure 5d shows a representative interaction between promoter polymorphisms in the IL10 gene and nonsynonymous SNPs located in the IL1RL1 gene. In this model, the rare IL10 alleles increase the risk, but the magnitude of the effect is greater with the number of rare alleles at the IL1RL1 gene. Figure 5e shows a representative interaction between promoter polymorphisms in the IL10 and CD86 genes. In this model, the risk of asthma is similar for carriers of two common CD86 alleles irrespective of the IL10 genotypes, however, the risk increases additively with the number of rare alleles in the two genes.

For all the single SNP associations observed in the SLSJ population, the statistical significance was modest and did not survive multiple correction procedures. Accordingly, an effort was made to replicate these findings in three additional studies. A comprehensive set of tagging SNPs in the IL10, CYP24A1, IL1RL1, CD86, and CYP2R1 genes plus the significant (p < 0.05) SNPs in the SLSJ study were genotyped in the CAPPS, SAGE and BHS studies. Table 3 shows SNPs with at least one p value < 0.05 for asthma or atopy in the three studies as well as SNPs with p value < 0.05 in the SLSJ study. Complete results for the replicate samples can be found in Additional file 7. For IL10, none of the SNPs associated with asthma in the SLSJ study were significant in the other studies except the promoter polymorphism rs1800896, with a marginal p value of 0.022 in CAPPS. However, the C allele that was protective in the SLSJ study was the risk allele in the CAPPS study. The direction of the risk allele in the SAGE study was similar to the one observed in the SLSJ, but the effect was not significant. Similar flip-flop phenomena were observed between the SAGE and CAPPS studies for SNPs that were not significant in the SLSJ study. Two SNPs in tight LD, rs3024498 located in the 3'UTR and rs3024492 located in intron 3, showed p values > 0.01 in both replicate samples, but the orientation of the risk allele was reversed. For most IL10 SNPs, it was noticeable that the orientation of the risk alleles is in agreement between the SLSJ and SAGE studies and in the opposite direction for CAPPS. No SNP in the IL10 gene was significantly associated with asthma or atopy in the BHS. Taken together, significant associations in the IL10 gene were observed in three studies. However, different SNPs were associated in different populations and the direction of the risk allele in the CAPPS study was reversed compared with the SLSJ and SAGE studies.

The associations observed for the IL1RL1 and CYP2R1 genes in the SLSJ study were not validated in the replication samples (Table 3). No other SNP in the IL1RL1 gene was associated with asthma or atopy in the replication studies except a non-synonymous coding SNP (rs1041973, Glu78Ala) that was borderline significant for atopy in SAGE. For CYP2R1, two additional SNPs located in the promoter and intron 1 were associated with asthma in the BHS. However, none of these SNPs were replicated in the other studies. For CD86, the promoter SNP (rs2715267) that was associated with atopy in the SLSJ (p = 0.004) could not be assessed in the other studies due to failure of the genotyping assay. However, two other intronic SNPs (rs9831894 and rs2332096, r² = 0.48 in the European HapMap data), not in LD with rs2715267, were significantly associated with asthma in CAPPS. A different SNP (rs9282641) located in non-coding exon 1 was significantly associated with asthma in the BHS. This latter SNP was not in LD with any of the three SNPs associated with asthma in CAPPS and SLSJ. Accordingly, different CD86 polymorphims were associated with asthma in three populations. Finally, significant SNPs were observed in the CYP24A1 gene in all studies. However, there was no consistency with the SLSJ study in terms of phenotype and orientation of the risk alleles.

The two-gene models were also performed in the replication samples for the five genes that were genotyped in CAMP, SAGE and BHS. The significant two-gene models involving IL10 and VDR as well as IL10 and IL1RL1 observed in the SLSJ study is not replicated in the CAPPS, SAGE and BHS. It should be noted that the sample sizes in the SAGE and CAPPS studies limit our ability to replicate the findings. For example, the number of cases contributing to the two-gene model involving IL10-rs1800896 and VDR-rs1544410 (Figure 5a) was approximately halved in SAGE and reduced to a quarter in CAPPS compared with the SLSJ study. However, even with similar power, the significant two-gene models observed in the SLSJ are not replicated in the BHS (see Additional file 8). Taken together, none of the significant two-gene models observed in the SLSJ study were properly replicated in CAPPS, SAGE and BHS.

The significant two-gene models involving IL10 and VDR were also tested in the CAMP study. A different panel of SNPs was genotyped in CAMP and some of the two-gene models are significant (p < 0.05). The most significant two-gene models is observed between VDR-rs7975232 and IL10-rs1800896 (Wald test p = 0.018) or IL10-rs1800872 (Wald-test p = 0.016). Figure 6 shows the comparison of two-gene models observed in the SLSJ and CAMP studies. VDR-rs1544410 is not genotyped in CAMP, but rs7975232 was in moderately high LD with it (D' = 1.0 and r² = 0.59, based on the HapMap CEU genotyping dataset). The gene-gene interaction models were similar between the two studies. The effect of the IL10 variant on asthma seemed dependent on the VDR variant. This can be appreciated in Figure 6 by comparing the solid, dotted and dashed lines between studies. The flip-flop phenomenon previously observed in the VDR gene between the SLSJ and CAMP studies [21, 22] can also be appreciated in Figure 6 (the solid line is at the top in the SLSJ study and at the bottom in the CAMP study).

Single SNP association tests and two-gene models were tested in the combined dataset. Overall, none of the SNPs are significantly associated with asthma or atopy after corrections for multiple testing (see Additional file 9). Similarly, the two-gene models for asthma and atopy were not significant (see Additional files 10 and 11, respectively).