Estimating KIR Haplotype Frequencies on a Cohort of 10,000 Individuals: A Comprehensive Study on Population...

The killer cell immunoglobulin-like receptors (KIR) mediate human natural killer (NK) cell cytotoxicity via activating or inhibiting signals. Although informative and functional haplotype patterns have been reported, most genotyping has been performed at resolutions that are structurally ambiguous. In order to leverage structural information given low-resolution genotypes, we performed experiments to quantify the effects of population variations, reference haplotypes, and genotyping resolutions on population-level haplotype frequency estimations as well as predictions of individual haplotypes. We genotyped 10,157 unrelated individuals in 5 populations (518 African American[AFA], 258 Asian or Pacific Islander[API], 8,245 European[EUR], 1,073 Hispanic[HIS], and 63 Native American[NAM]) for KIR gene presence/absence (PA), and additionally half of the AFA samples for KIR gene copy number variation (CNV). A custom EM algorithm was used to estimate haplotype frequencies for each population by interpretation in the context of three sets of reference haplotypes. The algorithm also assigns each individual the haplotype pairs of maximum likelihood. Generally, our haplotype frequency estimates agree with similar previous publications to within <5% difference for all haplotypes. The exception is that estimates for NAM from the U.S. showed higher frequency association of cB02 with tA01 (+14%) instead of tB01 (-8.5%) compared to a previous study of NAM from south of the U.S. The higher-resolution CNV genotyping on the AFA samples allowed unambiguous haplotype-pair assignments for the majority of individuals, resulting in a 22% higher median typing resolution score (TRS), which measures likelihood of self-match in the context of population-specific haplo- and geno-types. The use of TRS to quantify reduced ambiguity with CNV data clearly revealed the few individuals with ambiguous genotypes as outliers. It is observed that typing resolution and reference haplotype set influence haplotype frequency estimates. For example, PA resolution may be used with reference haplotype sets up to the point where certain haplotypes are gene-content subsets of others. At that point, CNV must be used for all genes.