Association Studies for Human Trait Genes

Heritable Traits Many human diseases run in families and are due to a single susceptibility gene. Well-known such mendelian disorders are Huntington disease and cystic fibrosis. These traits tend to be rare with population frequencies much below 1%. On the other hand, common heritable traits such as diabetes and schizophrenia generally do not exhibit mendelian inheritance and are thought to be under the influence of multiple susceptibility genes. Localizing a mendelian disease gene on the human gene map is now relatively straightforward. The challenge is to find genes underlying common so-called complex traits. Here we address statistical issues occurring in case-control gene mapping and microarray expression studies. Genetic mapping is based on large numbers of genetic marker loci that exhibit stable mendelian inheritance and are spread over the whole human genome (chromosomes). For association analyses, the most suitable marker type is the single nucleotide polymorphism (SNP) with two alleles and three genotypes. It is anticipated that in the future, thousands of SNPs will be used for genome-wide case-control association studies. Those SNPs in the vicinity or inside disease-causing genes are expected to show different genotype (and allele) frequencies between case and control individuals. Allele frequencies at SNPs 1 and 2 may be displayed in a 2 × 2 table with rows corresponding to SNP 1 and columns corresponding to SNP 2. For all n SNPs, allele frequencies form a 2 table for cases and an analogous table for control individuals. With n being as large as 1000 or higher, these tables tend to be extremely high-dimensional (“curse of dimensionality”). Because sample sizes tend to be lower than a few hundred at best, these tables are also very sparse. In practice, therefore, researchers analyze one SNP at a time, that is, they evaluate the marginal (main) effect of each SNP, which leads to a large multipletesting problem. In microarray expression experiments, the expression level of several thousand genes is measured on a very small number of observations. For example, a comparison of expression levels between diseased and normal skin in patients with psoriasis may shed light on which genes are involved the etiology of this trait (Oestreicher et al. 2001). Testing one gene after another again leads to a severe multipletesting problem and disregards interactions between genes.