Linkage disequilibrium in wild mice

Abstract

Crosses between laboratory strains of mice provide a powerful way of detecting quantitative trait loci for complex traits related to human disease. Hundreds of these loci have been detected, but only a small number of the underlying causative genes have been identified. The main difficulty is the extensive linkage disequilibrium (LD) in intercross progeny and the slow process of fine-scale mapping by traditional methods. Recently, new approaches have been introduced, such as association studies with inbred lines and multigenerational crosses. These approaches are very useful for interval reduction, but generally do not provide single-gene resolution because of strong LD extending over one to several megabases. Here, we investigate the genetic structure of a natural population of mice in Arizona to determine its suitability for fine-scale LD mapping and association studies. There are three main findings: (1) Arizona mice have a high level of genetic variation, which includes a large fraction of the sequence variation present in classical strains of laboratory mice; (2) they show clear evidence of local inbreeding but appear to lack stable population structure across the study area; and (3) LD decays with distance at a rate similar to human populations, which is considerably more rapid than in laboratory populations of mice. Strong associations in Arizona mice are limited primarily to markers less than 100 kb apart, which provides the possibility of fine-scale association mapping at the level of one or a few genes. Although other considerations, such as sample size requirements and marker discovery, are serious issues in the implementation of association studies, the genetic variation and LD results indicate that wild mice could provide a useful tool for identifying genes that cause variation in complex traits.

Figure 1. A Neighbor-Joining Tree of Wild-Caught House Mice Based on Allele-Sharing at 4158 SNP Loci

The individuals include ten M. m. domesticus from Western Europe, seven M. m. musculus from Eastern Europe, nine M. m. castaneus from India, and 94 mice from Tucson, Arizona. The Arizona mice appear closely related to domesticus from Western Europe, as expected from the history of North American colonization and other evidence [1]. The small terminal branch lengths of musculus and castaneus may be due to an ascertainment bias in the SNPs, which were discovered in laboratory strains in which the nuclear genome seems to be predominantly of domesticus origin [4,56].

Figure 2. LD between Pairs of ∼2,900 Autosomal SNPs in Humans and Mice

The r ² variable is the composite (genotypic) measure of LD, which is very similar to the usual gametic measure (see Materials and Methods). The jagged red line is the 95th percentile of the r ² values within a 25 kb sliding window. The horizontal red line is the genome-wide threshold for significance of r ² at α = 0.05 (calculated as the 95th percentile of permuted data). The jagged black line is the mean of the r ² values within a 25 kb sliding window. HS is an outbred laboratory strain.

Figure 3. LD between SNPs Discovered by Resequencing in Selected Regions of the Genome, Measured as the Genotypic r ²

The top panel shows LD in 77 Arizona mice for pairs of SNPs within each of four regions (points) and summary statistics for all four regions combined (lines). The total number is 163 common SNPs (nine Alox15, 53 Apoa2, 70 C3ar1, 31 Nr1h3). The lines connect the midpoint of each distance bin and give either the mean or the 95th percentile of r ² for that bin. The lower panel compares LD in 60 unrelated individuals each of Arizona mice and humans of European ancestry from the CEU sample of the HapMap project. The mouse SNPs are the same as in the top panel, except that the total number is 141 (removing those with MAF < 0.05 in the sample of 60 individuals). The human SNPs are a subset from 10 ENCODE regions, selected to have the same allelic frequency distribution as the mouse SNPs. The human SNP numbers range from 131 to 803 per region, with a total of 3891 over all ten regions. The ranges of r ² statistics (mean or 95th percentile) for the ten ENCODE regions are given by the vertical bars.

Figure 4. Mean LD in 94 Arizona Mice and 90 Asian Humans, Evaluated in a Sliding Window of 0.025 cM

The data are the same as in Figure 2, but physical distance has been converted to genetic distance using genome-wide averages of cM/Mb. The expected value of r ² for a sample size of n diploids is calculated as E (r ²) = (1 / (1 + 4N_ec)) + (1 / n), where N_e is effective population size and c is recombination rate (converted to cM for plotting by the Kosambi mapping function). In this case, n = 92 and N_e = 1,000, 3,000, or 10,000 (from top to bottom).