Linkage disequilibrium mapping via cladistic analysis of single-nucleotide polymorphism haplotypes

Abstract

We present a novel approach to disease-gene mapping via cladistic analysis of single-nucleotide polymorphism (SNP) haplotypes obtained from large-scale, population-based association studies, applicable to whole-genome screens, candidate-gene studies, or fine-scale mapping. Clades of haplotypes are tested for association with disease, exploiting the expected similarity of chromosomes with recent shared ancestry in the region flanking the disease gene. The method is developed in a logistic-regression framework and can easily incorporate covariates such as environmental risk factors or additional unlinked loci to allow for population structure. To evaluate the power of this approach to detect disease-marker association, we have developed a simulation algorithm to generate high-density SNP data with short-range linkage disequilibrium based on empirical patterns of haplotype diversity. The results of the simulation study highlight substantial gains in power over single-locus tests for a wide range of disease models, despite overcorrection for multiple testing.

Figure 1

Example of a genealogical tree representing the shared ancestry of chromosomes at the disease gene. Disease chromosomes (D) carrying the same mutation (1 or 2), share more recent common ancestry than normal chromosomes (N) carrying no mutation (0).

Figure 2

Example of a cladogram representing haplotype diversity within a window of SNPs. The cladogram is constructed using hierarchical group average clustering on pairwise haplotype differences, expressed in terms of the proportion of marker mismatches within the window of SNPs.

Figure 3

Generating the SNP haplotype carried by a chromosome carrying allele 1 at the disease SNP, x.

Figure 4

Power to detect disease-marker association in windows of 8 markers overlapping the region flanking a disease gene, under the assumption of a 5% experimentwise significance level, with Bonferroni correction for multiple testing. Power is presented as a function of the disease model, for a disease allele frequency of 0.05. The three analysis methods are the best partition of haplotypes, T[MAX]; the first partition of haplotypes, T[h]; and a single-locus test.

Figure 5

Power of the best partition of haplotypes, T[MAX], to detect disease-marker association in windows of varying size (numbers of markers) overlapping the disease gene, under the assumption of a 5% experimentwise significance level, with Bonferroni correction for multiple testing. Power is presented as a function of the disease model, for a disease allele frequency of 0.05.

Figure 6

Distribution of single-marker association with CF across a 1.8-Mb candidate region flanking the CFTR gene (Kerem et al. 1989). The solid vertical line indicates the true location of ΔF508 at 0.88 Mb, the most common disease mutation identified in the CFTR gene. The dashed line indicates the distribution of association with CF of the best partition of haplotypes, T[MAX], using a sliding window of 6 markers across the candidate region.