Imputation of KIR Types from SNP Variation Data

Abstract

Large population studies of immune system genes are essential for characterizing their role in diseases, including autoimmune conditions. Of key interest are a group of genes encoding the killer cell immunoglobulin-like receptors (KIRs), which have known and hypothesized roles in autoimmune diseases, resistance to viruses, reproductive conditions, and cancer. These genes are highly polymorphic, which makes typing expensive and time consuming. Consequently, despite their importance, KIRs have been little studied in large cohorts. Statistical imputation methods developed for other complex loci (e.g., human leukocyte antigen [HLA]) on the basis of SNP data provide an inexpensive high-throughput alternative to direct laboratory typing of these loci and have enabled important findings and insights for many diseases. We present KIR∗IMP, a method for imputation of KIR copy number. We show that KIR∗IMP is highly accurate and thus allows the study of KIRs in large cohorts and enables detailed investigation of the role of KIRs in human disease.

Figure 1

The KIR Region (A) Immunochip SNPs and KIR genes. The relative genomic position of Immunochip SNPs and KIR genes on chromosome 19 according to the human reference genome (GRCh37) and the annotation provided with the Illumina Immunochip array. KIR genes are shown as rectangles. Only some KIR genes are present, consistent with the fact that the reference genome is an A haplotype. SNPs are shown as circles. They vary in y coordinates and border color on the basis of whether they are annotated as being within a particular KIR gene (including introns). SNPs that were selected as being most informative are shaded in light green if they had good clustering and magenta if they had poor clustering. SNPs that were monomorphic in the UK reference panel are shaded in orange. (B) Composition of common KIR haplotypes. The common KIR haplotypes are composed of different centromeric (cA01, cB01, and cB02) and telomeric (tA01 and tB01) motifs, each of which differs in the content and arrangement of genes. Framework genes, which are found at the ends and near the middle of the region on nearly all haplotypes, are shaded gray. Some genes (e.g., KIR2DL5) can be located in both motifs. Different centromeric motifs can be paired with different telomeric motifs through the central reciprocal recombination hotspot between KIR3DP1 and KIR2DL4, as indicated by different dashed lines. KIR haplotypes can be classified into two categories: group A haplotypes (shaded gray) and group B haplotypes (unshaded). Group A is composed of only cA01 and tA01 motifs, which have fixed gene-content with one activating gene (KIR2DS4), which on many A haplotypes contains a deletion rendering it non-functional (see Appendix B). Group B comprises at least one motif of type cB01, cB02, or tB01, which has variable gene-content between framework genes and more than one activating KIR gene. Less common KIR haplotypes might differ (with slightly different arrangements and copy number of the KIR genes⁹) from the ones shown here. This panel is adapted from Roberts et al. and relates to haplotypes of European ancestry.

Figure 2

Per-Allele Imputation Accuracy Estimates of the sensitivity and positive predictive value (PPV) for each KIR-type allele (i.e., the different possible values for each of our KIR loci) from the UK cross-validation analysis of KIR^∗IMP. These are plotted against the number of times each allele appeared in the UK reference panel. Each point corresponds to a single allele.

Figure 3

Calibration of Imputation Probabilities A calibration plot of the KIR^∗IMP imputation probabilities. The probabilities used are those associated with the OOB imputations across all KIR loci, imputed on the UK reference panel by KIR^∗IMP trained with the UKsnps set. The imputed KIR types are grouped by their maximum posterior probability (MAP) into ten bins of equal width (on the probability scale) covering the range of the probabilities. For each bin, the observed imputation accuracy and corresponding 95% credible interval (see Material and Methods) are plotted against the mean MAP. Note that the number of values in each bin varies, as reflected by the differing widths of the intervals. Perfect calibration is indicated by the dashed line.

Figure 4

Comparison of Imputation Methods Estimates of the imputation accuracy for the different methods and the associated 95% credible intervals (see Material and Methods). (A) The percentage of correctly imputed haplotypes from the UK cross-validation analysis. (B) The percentage of correctly imputed copy-number types for individuals from the NG validation analysis.

Figure 5

Imputation Accuracy with Different SNP Sets Estimates of the KIR^∗IMP imputation accuracy from the UK cross-validation analysis are compared across different SNP subsets for training: the main set of SNPs used for the cross-validation analyses (UKsnps), the SNPs selected as being highly informative (UKselectedSnps), and the remaining set of SNPs (UKnotSelectedSnps).