Asthma [MIM 600807] is a chronic respiratory disease that affects over 20 million Americans and 300 million people worldwide [1, 2]. The genetic basis of asthma has been explored for over two decades in candidate gene association, where more than 40 genes have been associated with asthma and replicated in at least one independent population [3, 4]. Recently, genome-wide association (GWA) studies of asthma have found that variants in or near several genes, including ORMDL3 [MIM 610075] [5–12], CHI3L1 [MIM 601525] [13], TLE4 [MIM 605132] [14], PDE4D [MIM 600129] [15], DENND1B [MIM 613292] [16], RAD50-IL13 [MIM 604040, 147683] [17], and the HLA-DR/DQ region on chromosome 6p21.3 [17], contribute to the risk of asthma.

The primary findings of the PDE4D GWA study were obtained in a case-control design consisting of children from an asthma clinical trial and publicly available population controls. Although we used a case-control design, genetic studies of these asthmatic children were originally intended to be part of a family-based design, as DNA was collected from children and their parents. We chose the case-control design in an effort to increase the power to detect genetic associations. Consistent with the approach taken in most GWA studies, we selected the variants with the lowest p-values in our primary population for replication in independent populations. Here, we extensively compare the GWA results of the case-control and family-based designs using probands from the same population, and we attempt to replicate the initial association findings of many SNPs with nominally significant p-values in a two-stage process. Our results help to determine whether the case-control design, the family-based design, or a combination of results from both designs, is more powerful to identify asthma-susceptibility variants.

The CAMP population was designed for, and has often been used in, family-based candidate gene association studies [35–37]. More recently, GWA data have been acquired for a cohort of 403 Caucasian probands and their parents. This cohort has low statistical power to detect associations corresponding to the expected small (i.e. 1.1-1.3) effect sizes that have been observed to underlie many complex traits, including asthma [38]. We sought to increase statistical power by using the CAMP probands as cases in a case-control design, where controls were obtained from Illumina's iControlDB resource. Although we increased statistical power with the case-control design, which is composed of 359 cases and 846 controls, this design also remains underpowered to detect associations of small effect sizes. For example, for a SNP such as rs17572584, which has an MAF near 0.10 and an effect size of 1.6, the power to detect its association to asthma via a TDT in CAMP Trio is 0.37, while the power in CAMP/Illumina is 0.45. For a SNP such as rs4653637, which has an MAF near 0.40 and an effect size of 1.4, the power in CAMP Trio is 0.39, while the power in CAMP/Illumina is 0.47. Thus, despite a significant increase in power, neither design individually has a large enough number of subjects to make it adequate to detect most of the small effect sizes that are expected for asthma. In this work, we attempted to find out whether a combination of the results of the two designs would be helpful to find SNPs that replicated in independent populations, and hence, were likely to be truly associated with asthma.

The top SNPs according to the results for individual designs did not overlap [Figure 2], and only 1/18 of the top SNPs that were genotyped for replication Stage 1 had a p-value < 0.05 [Table 1]. Because of the low p-values in either the CAMP/Illumina or CAMP Trio designs, we were more lenient with the replication threshold for these SNPs to be analyzed for Stage 2, and proceeded to consider the SNPs with p-value < 0.10 in CR. Based on the overall results in Table 1, only the three CAMP/Illumina SNPs on Chromosome 5 (Table 1, CAMP/Illumina Rank 1, 3, 14) show evidence of association to asthma. These three SNPs and the two others near them on Chromosome 5 (Table 1, CAMP/Illumina Rank 1-3, 5, 14) are near and within sequences of the PDE4D gene as we reported previously [15]. In this previous work, we attempted to replicate the top CAMP/Illumina findings in independent populations. We found that in 2/7 independent populations our results replicated with p < 0.05 and that in 5/7 the results did not replicate but had consistent effect directions with CAMP/Illumina, providing overall p-values in the independent populations that supported the association of PDE4D variants with asthma. These previous findings would have been more difficult to identify based on our current replication strategy because the CR population did not convincingly replicate the PDE4D associations. Thus, having the first stage be a filtering stage limits our ability to identify some of the SNPs that may be truly associated with asthma because they may not replicate in an initial independent population but would in other independent populations. This limitation may be overcome in future studies that pool GWA results across multiple populations.

Most of the SNPs that passed replication Stage 1 were distributed closer to the nominal p-value thresholds than to the low p-value extremes of either CAMP/Illumina or CAMP Trio [Figure 2B in green]. The SNPs that passed replication Stage 2 were even closer to the nominal p-value thresholds than those of replication Stage 1 [Figure 2B in red]. Thus, SNPs with consistent results in the two initial study designs were more likely to replicate in independent populations than those with extreme p-values in either individual design. Tables 2, 3 and 4 detail the characteristics of the SNPs that passed replication Stage 2 (i.e. had p-value < 0.05 in CR and either FHS or iCAP, with consistent effect direction across all populations), and hence, are most likely to be truly associated with asthma. Most of these SNPs are in or near genes [Table 2], making them more likely to have biological relevance than SNPs that are often found to be associated with asthma and are in gene deserts. The top SNP (rs17572584) in Table 2 has an overall p-value across the independent populations that is significant after multiple comparisons corrections for the number of SNPs that were genotyped in Stage 1 (p-value = 0.048). If the results for this SNP in the independent populations are combined with those of the primary CAMP studies, then the overall p-value would be 9.7E-07 with CAMP/Illumina and 8.3E-06 with CAMP Trio. The rs17572584 SNP is downstream of the carbohydrate (N-acetylglucosamine-6-O) sulfotransferase 2 gene (CHST2 [MIM 603798]), which encodes a sulfotransferase for 6-sulfated glycan synthesis. Transcription of this gene has been found to be coordinated with that of NF-kappaB and GATA-3, both of which are involved in Th2 cell differentiation [39]. Because Th2 cells are prominent in asthma, especially allergic asthma, genes such as CHST2 that modulate Th2 cell differentiation are likely to play a role in asthma susceptibility. Further, the CHST2 protein product has been shown to be expressed in the human lung [40]. Even though CHST2 is a plausible candidate gene for asthma, rs17572584 is located downstream of this gene, and hence, further study is needed to find out whether the association we measured is stronger in SNPs located nearer or within the gene across multiple populations. However, identification of this association using either CAMP/Illumina or CAMP Trio alone would have been difficult since this SNP ranked 2,539 and 22,513 in these primary studies, respectively. The fact that our strongest associations were located in regions that were not at the lowest p-value extremes, but were at nominally significant levels, reflects the potential of small populations to contain useful genetic associations that can be found by increases in power. As stated previously, this power limitation may be overcome in future studies that pool GWA results across multiple populations.

Currently, among regions reported to be associated with asthma, the one on chromosome 17q21 near ORMDL3 [MIM 610075] [5–12] has been the most consistently replicated. We previously reported [15] that nine CAMP/Illumina SNPs support the original association findings by Moffatt et al. in this region [5], by having p-values < 0.05. Of these nine SNPs, only one (rs9303277) has a CAMP Trio p-value < 0.05. However, this SNP was excluded from our Replication Stage 1 because its CAMP/Illumina p-value was equal to 0.011 (i.e. was greater than our 0.010 cutoff). In a separate study of association in the 17q21 region, rs9303277 was reported to have a p-value = 4.0E-03 in CAMP trios and = 1.5E-05 in CR trios [12]. The lower CAMP p-value reported in [12] was obtained using TaqMan SNP Genotyping Assays (Applied Biosystems) data corresponding to 444 white trios, and not the GWA data corresponding to 403 trios reported in the current study. The difference in CAMP p-values as a result of using a slightly increased number of trios and different genotyping platforms is another reflection of the low power of CAMP Trio, and suggests that additional SNPs with more modest p-values than those considered by our replication strategy are truly associated with asthma. Because the 17q21 region near ORMDL3 has been consistently replicated in so many studies, including CAMP and CR, we evaluated the association for three of the SNPs originally reported by Moffat, et al. in iCAP and FHS. As Table 5 shows, these three SNPs replicated with nominal p-values < 0.05 in these two populations, growing the total number of independent populations replicating this region's results to over 12. Additionally, the replication of the chromosome 17q21 results in FHS and iCAP helps validate the use of these populations, which were not ascertained to have asthma using as strict criteria as CAMP and CR, in asthma association studies.

In previous work that developed a method to identify reproducible associations in small affected trio-based cohorts, two CAMP Trio SNPs were found to reach genome-wide significance according to the developed criteria [41]. These SNPs (rs10863712 and rs1294497) ranked first and second according to an initial power screen and had FBAT p-values that passed the corresponding genome-wide significance thresholds (i.e. p-value < 0.005). The CAMP/Illumina p-values (ranks) for these two SNPs were 0.28 (149891) and 0.049 (27336) for rs10863712 and rs1294497, respectively. The CAMP Trio p-values (ranks) for these two SNPs were 0.0032 (1684) and 0.0048 (2547) for rs10863712 and rs1294497, respectively. Thus, neither of these SNPs met the criteria for replication Stage 1 and neither was tested for replication in our independent cohorts. Future attempts to replicate the top SNPs according to the power-based screening methodology would be helpful to compare its utility with that of the current approach of combining case-control and family-based results.

Our study has limitations similar to those of many other GWA studies. In addition to having low power to detect associations and some missing genotypic data for all of our populations, there were differences in the way our populations were ascertained to have asthma. CAMP and CR are the most similar populations, as they are composed of children with asthma who were carefully ascertained for asthma studies. However, the two populations are ethnically different and have markedly different environmental exposures. The FHS population was not initially gathered for asthma research, but asthma status has been assessed for its participants based on self-reported doctor's diagnosis. The iCAP cohort was selected based on data, including ICD-9 codes for asthma and medication history, extracted from de-identified EMRs. The diversity of our populations increases the certitude that the SNPs that are in Table 2 are general asthma variants, rather than variants that are associated to asthma sub-phenotypes that characterize each of the populations. Any errors in the assignment of cases and controls would decrease our ability to find associations, and thus, does not increase the likelihood that the SNPs in Table 2 are false positive associations.

Our results suggest that SNPs with similar case-control and family-based association results in designs that share probands are those that are most likely to replicate in independent populations, rather than the SNPs with the smallest p-values in either case-control or family-based designs alone. Thus, using probands from family-based studies in case-control designs, and combining results of both approaches, may be a way to augment our ability to find SNPs associated with asthma and other complex diseases.