Haplotype analysis improved evidence for candidate genes for intramuscular fat percentage from a genome wide association study of cattle

Abstract

In genome wide association studies (GWAS), haplotype analyses of SNP data are neglected in favour of single point analysis of associations. In a recent GWAS, we found that none of the known candidate genes for intramuscular fat (IMF) had been identified. In this study, data from the GWAS for these candidate genes were re-analysed as haplotypes. First, we confirmed that the methodology would find evidence for association between haplotypes in candidate genes of the calpain-calpastatin complex and musculus longissimus lumborum peak force (LLPF), because these genes had been confirmed through single point analysis in the GWAS. Then, for intramuscular fat percent (IMF), we found significant partial haplotype substitution effects for the genes ADIPOQ and CXCR4, as well as suggestive associations to the genes CEBPA, FASN, and CAPN1. Haplotypes for these genes explained 80% more of the phenotypic variance compared to the best single SNP. For some genes the analyses suggested that there was more than one causative mutation in some genes, or confirmed that some causative mutations are limited to particular subgroups of a species. Fitting the SNPs and their interactions simultaneously explained a similar amount of the phenotypic variance compared to haplotype analyses. Haplotype analysis is a neglected part of the suite of tools used to analyse GWAS data, would be a useful method to extract more information from these data sets, and may contribute to reducing the missing heritability problem.

Figure 1. Comparison of linkage disequilibrium (LD) measures for the genes in the study.

A. D′ and r ² plotted against distance between SNPs. D′ values are filled black circles, r ² values are open black circles. Least squares fitted regression lines of LD on length of haplotype (D′ solid line, r ² dashed line) are not statistically significant and the slopes are b<−1×10⁻⁵. This is evidence that the length differences between haplotypes are not important in accounting for LD between SNPs in this sample of genes. Values are means of LD estimates for each breed, not calculated from a sample of mixed breed individuals. B. Plot of D′ against r ² for the genes in this study. Most of the comparisons between pairs of SNPs show high D′and low r ² values, a typical result for cattle at this distance between SNPs. High D′values can indicate a reduced number of haplotypes or classes of haplotypes that are missing. r ² values are useful in describing how well the genotypes at one SNP predict the genotypes at the other SNP.

Figure 2. Plot of –logP values for SNPs compared to haplotypes for candidate genes for IMF.

The SNPs are numbered 1, 2, and 3 in order along the chromosome and in the haplotypes, 1 = A and 2 = B alleles at each SNP. Haplotypes were fitted simultaneously. Note that for the gene CXCR4, this gene is the closest gene to the significant SNPs, but these are not located within the gene itself.