Trans-ethnic genome-wide association studies: advantages and challenges of mapping in diverse populations

Abstract

Genome-wide association studies (GWASs) are the method most often used by geneticists to interrogate the human genome, and they provide a cost-effective way to identify the genetic variants underpinning complex traits and diseases. Most initial GWASs have focused on genetically homogeneous cohorts from European populations given the limited availability of ethnic minority samples and so as to limit population stratification effects. Transethnic studies have been invaluable in explaining the heritability of common quantitative traits, such as height, and in examining the genetic architecture of complex diseases, such as type 2 diabetes. They provide an opportunity for large-scale signal replication in independent populations and for cross-population meta-analyses to boost statistical power. In addition, transethnic GWASs enable prioritization of candidate genes, fine-mapping of functional variants, and potentially identification of SNPs associated with disease risk in admixed populations, by taking advantage of natural differences in genomic linkage disequilibrium across ethnically diverse populations. Recent efforts to assess the biological function of variants identified by GWAS have highlighted the need for large-scale replication, meta-analyses and fine-mapping across worldwide populations of ethnically diverse genetic ancestries. Here, we review recent advances and new approaches that are important to consider when performing, designing or interpreting transethnic GWASs, and we highlight existing challenges, such as the limited ability to handle heterogeneity in linkage disequilibrium across populations and limitations in dissecting complex architectures, such as those found in recently admixed populations.

Figure 1

Fine-mapping of candidate causal or functional SNPs by transethnic GWAS. The graph shows the results of association testing (in the form of the allele frequencies) for a typical locus in three different populations. In the EUR population, many SNPs in the region are in close LD, leading to a significant signal for a wide set of SNPs. However, LD patterns in the ASN population are different, which enables finer mapping of the causal SNP as being the SNP with the strongest trait association. However, it is rarely obvious in advance which additional populations should be studied, as in some populations (such as AFR in this example) the locus might not be associated with the trait at all, because of epistatic interactions, phenotype heterogeneity, or low minor allele frequency/non-polymorphic markers across the locus. Data shown are based on simulation and do not reflect the result of any published or unpublished studies. Abbreviations: ASN, Asian; AFR, African; EUR, European.

Figure 2

Theoretical basis of admixture GWAS study designs. (a) Populations 1 and 2 are two parental populations in which there has been no gene flow historically. When these populations interbreed the subsequent F1 population includes heterozygotes. Over the course of 5 or 10 generations the chromosome of any given F_n population offspring will include a combination of parental chromosomal `bands'. Some loci are associated with a disease (such as B) and others are not (such as A). (b, c) In a typical GWAS, association testing identifies whether a given allele (such as T at SNP2) is associated with increased risk for having a disease; this is shown as allele frequencies in the table. (c) If the ancestral frequency of T at SNP2 is different in two parental populations (1 and 2) and if it is associated with disease, then the population with higher frequencies of this allele will also have higher risk for disease. One can thus expect to observe higher incidences of disease in individuals carrying the T allele and also higher incidence of disease in individuals from population 1, in which the T allele is more frequent. This is the premise of admixture association studies. By ascertaining local ancestry one can determine if an allele that is much more common in one population may be associated with disease risk. In (b), in a locus with no evidence of association with disease, admixture analysis would find that the minor allele frequencies (and percentages of individuals of either ancestral populations) do not differ between cases and controls. (d) Graph of the allele frequencies along the genome. The relative frequency of the allele from population 1 differs between the cases and the controls only at the locus associated with the disease/phenotype. Thus, in admixed populations, by determining the local ancestry in the cases versus controls, one can determine if there is an association between an allele associated with ancestry and disease liability.