To our knowledge, this is the largest candidate genes association study that has examined racial differences in childhood asthma. Through this systematic study, we have simultaneously studied both Caucasian and African American asthmatic children and demonstrated that these populations predominantly exhibit different patterns of association between genetic variants and asthma. To accomplish this goal we used well characterized European ancestry and African American children who live in the same geographic region of the greater Cincinnati area. Using both cohorts we have shown that only 1 of 28 genes had associations in both populations, as well as only 2 genes were common across the two races in the recursive partitioning analysis. Indeed, different gene networks were associated with asthma in children with European ancestry versus African Americans suggesting that there may be distinct mechanisms underlying the pathogenesis and expression of asthma in these 2 subgroups. Simultaneous investigation of risk variants across European and African American populations enabled the identification of population specific risk alleles and disease pathways, which may contribute to health disparity. The results from this study may also assist in fine-mapping of genetic associations by exploiting the differences in linkage disequilibrium between populations to narrow the range of marker alleles demarking regions that contain a true biologically relevant variant.
These analyses revealed two major findings. First, we confirmed the importance of IL4 genetic variation in the risk of pediatric asthma, and present evidence of replication among the African-American population. While IL4 has been consistently reported to be associated with asthma in Caucasian, Asian, and Hispanic populations, two of the four SNPs, which reached Bonferroni corrected significance in the Caucasian children (rs2243250 and rs2243274) replicated (p<0.05) in the African American children (). While non-coding IL4 rare variants have been associated with asthma susceptibility in African Americans 
, the association of these two SNPs is novel to this population. This result suggests that some common immunological mechanisms (at these variants) may underlie childhood asthma across different ethnic backgrounds. However, most studied SNPs showed no evidence of replication between Caucasian and African American children. For example, IL4 SNPs, which are highly significant in the Caucasian group such as rs2243282 and rs2243268 didn't reach 5% significance level in African American population. In contrast, SNP rs4448492 in the INSIG2 gene was associated (P
0.002) with asthma in African American population. However, this SNP was not significant even without adjustment (at 5%) in Caucasian population. Several SNPs have shown different allele frequencies between the two races (Figure S1
). This result suggests that these genes do not harbor susceptibility variants common to both races due to a) variation in signatures of natural selection resulting in differences in allele frequencies; b) varying linkage disequilibrium patterns at causal loci across different populations (as shown for IL4 – ); and/or (c) there may be common and distinct pathways that contribute to the development and expression of asthma phenotypes between these two groups. It also remains possible that we do not have sufficient statistical power with the current sample size to detect statistical significance, although this is unlikely for the observed lack of association of INSIG2 in the Caucasian subset. To determine the power of this to detect expected ORs for IL4 coding SNPs in both Caucasian and African American population, we conducted an ad-hoc analysis with the software Quanto 
. With our sample size, we have 96% and 71% power to detect the association of rs2243250 with asthma in Caucasian and African ancestry population, respectively. This ad-hoc power analysis provides sufficient evidence that we have high power in Caucasian and moderate power in African American to detect true effects. The lack of SNP replication in these two populations emphasizes the need to consider ancestry background and detailed examination of population SNPs allele frequency across populations of different and mixed ancestry as well as non-genetic factors.
Secondly, Using RP, we report for the first time an interaction of six genes affecting European ancestry pediatric asthma: rs2243250 (IL4), rs6597 (STUB1) rs11168070 (ADRβ2), rs3024676 (IL4Rα), rs638376 (IL13Rα2) and rs3806446 (CHIA). These SNPs resulted in 62% accuracy of asthmatic and non-allergic classification. Similarly seven SNPs in five genes rs4848492 (INSIG2), rs2243283 (IL4), rs4423003 (CHIA), rs2243283 (IL4), rs12495941 (ADIPOQ), rs2243268 (IL4), and rs2291427 (ALOX5) in African American children had 77% discriminate power between asthmatic and non-allergic individuals. The combination of genotypes in these interactive SNPs can help to pin-point individuals with greater asthma risk (). Importantly, the RP method may elucidate associations, which may be missed using single SNP association. For example, variants in STUB1 and ADRβ2 genes in Caucasian and variants in CHIA and ADIPOQ genes in African American were not associated with asthma in the single SNP analysis; however, in conditional inference framework taking rs2243250 and rs4848492 as a major discriminatory SNPs in Caucasian and African American respectively, variation in these genes is highly associated with asthma (p<0.01). Kabesch et al. 
reported strong gene-gene interactions among genes involving Th3-cell differentiation and signaling pathways. Our study showed that using the RP approach, SNPs that are weakly or not associated in the univariate analysis could discriminate between asthma and non-allergic control individuals in both races. This finding clearly indicates that the effect of one gene may not be disclosed if the effect of another gene is not considered 
, suggesting that the true effect may be driven by gene-gene interaction, rather than by the main effect of each gene by itself.
Further analysis using Ingenuity Pathways Analysis (IPA) revealed that these RP based interactive genes belong to an interconnected and interactive gene network, indicating that they are involved in related biological activities and have functional commonalities (). We also used IPA to characterize the enrichment of specific pathway components into functionally differentiated gene groups 
. The most enriched (p≤3*10−4
) canonical pathway in Caucasian population was IL4 signaling whereas airway inflammation in asthma was the most enriched (p<1.36*10−3
) pathway in African American (data not shown). Differences in the genetic architecture of individuals may have affected determinant pathways in different ways. However, both enriched IPA pathways in both races have essential roles in asthma pathogenesis 
. In network analysis, IL4 was the major hub gene in both Caucasian and African American (). These results were not unexpected given that IL4 is a critical effector in the generation of allergic inflammation and IgE production, and is one of the most relevant genes in regulating the Th2 profile of allergic subjects 
. IL4 is central to B cell heavy class switching from immunoglobulin M (IgM) to IgE, and to the maturation of T helper (Th) cells towards the Th2 phenotype 
. One of the variants (rs2243250), which was most strongly associated with asthma, lies in the IL4 promoter region which has been implicated and replicated in more than 11 studies 
. IL4 rs2243250 is a C-to-T mutation that lies upstream from the open reading frame of the gene. It has previously been shown to increase promoter activity of IL4 transcription and was associated with elevated levels of serum IgE in asthmatic families 
Ingenuity Pathway Analysis (IPA) Interactive network.
In critically evaluating our results, it is important to note that our analyses, and hence interpretations, are subject to several limitations. First, SNP allele frequencies and association were determined by using relatively small sample sizes (see Methods). However, it should be noted that large sample sizes may not help powering genetic studies and improve our understanding of the genetic underpinnings of allergy phenotypes as much as precise phenotyping 
. In the present study we show that use of well-characterized control populations (see Methods) in genetic association studies can overcome relatively small sample sizes to identify risk variants. Further, in order to overcome the relatively small sample size in the AA cohort, we sought to replicate our findings using publically available datasets but found similarly small AA cohorts. Thus, there is a clear need for larger AA cohorts in future studies. Second, to reduce the chance of potential false positive results from multiple testing, we corrected the p-values using Bonferroni adjustment which accounted for the LD among SNPs. As the Bonferroni adjustment is notably conservative, the LD adjustment provides minimization of false positives. We believe that this approach provides a reasonable balance between type I and type II error. Nonetheless, it is likely that we are missing true associations, which may provide insight into racial differences and similarities. Third, the environmental influences between our case and control groups may be different, especially between adults and children. Fourth, our study showed a positive association, but it does not always imply causality. Hence, further studies are needed to confirm the findings and to identify functional variants causally linked to asthma risk. The present study has notable strengths. First, we were able to conduct the analyses separately in each race, and were therefore able to account for the differences in allele frequencies, disease prevalence, and linkage disequilibrium patterns between these subpopulations. Second, our study used a custom designed array that includes more coverage of candidate genes/SNPs of interest and the inclusion of ancestry informative markers (AIMs) to account hidden ethnic variations.
In summary, through our systematic and comprehensive screen of variants in asthmatic children who live in the same geographic region, we have demonstrated the importance of IL4 genetic variation in both Caucasians and African American. Variants found in populations of both African and European ancestry may represent more universally important genes to the disorder 
. The replication of IL4 SNPs in African ancestry can also potentially aid in refining and fine mapping associations due to the unique short range LD in this ethnicity. The use of a population with short LD will result in the greatest localization success rate in distinguishing the causal SNP from its neighbors. Based on the overall lack of SNPs concordance in association between European and African American asthmatic children, we suspect that rare and/or population-specific risk alleles may explain some of the associations in asthma, pointing to genetic heterogeneity in susceptibility alleles. These results also underline the importance of understanding differences in biologic and genetic factors driving asthma in different ancestral populations. Future fine-mapping and deep sequencing studies are needed to determine whether or not other SNPs can be found associated in African Americans as well as to identify both common and/or rare risk-causing alleles in the associated regions.