Efficient and accurate KIR and HLA genotyping with massively parallel sequencing data

Abstract

Killer cell immunoglobulin like receptor (KIR) genes and human leukocyte antigen (HLA) genes play important roles in innate and adaptive immunity. They are highly polymorphic and cannot be genotyped with standard variant calling pipelines. Compared with HLA genes, many KIR genes are similar to each other in sequences and may be absent in the chromosomes. Therefore, although many tools have been developed to genotype HLA genes using common sequencing data, none of them work for KIR genes. Even specialized KIR genotypers could not resolve all the KIR genes. Here we describe T1K, a novel computational method for the efficient and accurate inference of KIR or HLA alleles from RNA-seq, whole-genome sequencing, or whole-exome sequencing data. T1K jointly considers alleles across all genotyped genes, so it can reliably identify present genes and distinguish homologous genes, including the challenging KIR2DL5A/KIR2DL5B genes. This model also benefits HLA genotyping, where T1K achieves high accuracy in benchmarks. Moreover, T1K can call novel single-nucleotide variants and process single-cell data. Applying T1K to tumor single-cell RNA-seq data, we found that KIR2DL4 expression was enriched in tumor-specific CD8⁺ T cells. T1K may open the opportunity for HLA and KIR genotyping across various sequencing applications.

Figure 1.

Evaluation of T1K. (A) Overview of the T1K workflow. (B) The KIR allele prediction accuracy of T1K and T1K with no multiple-gene mapped reads (mimicking arcasHLA) on simulated reads from KIR mRNAs. (FN) False negative; (FP) false positive. (C) The KIR allele prediction accuracy of T1K on WGS. (HPRC matched) The alleles inferred from the HPRC phased genome that can be found in T1K predictions; (T1K matched) the alleles predicted by T1K that can be found on HPRC genomes. (D) The HLA allele prediction accuracy of T1K, arcasHLA, and OptiType on RNA-seq data validated with 1kPG annotation. (E) The HLA allele prediction accuracy of T1K and OptiType on WES data validated with 1kPG annotation. With T1K_whitelist, T1K only reports alleles present in an allele whitelist coded in OptiType's Python script.

Figure 2.

KIR alleles in CD8⁺ T cells. (A) KIR allele expression fractions in four cells from a colorectal cancer (CRC) patient. (B) The number of expressed KIR alleles in a cell. For example, seven cells expressed two KIR alleles. (C) The number of cells that express the KIR gene, splitting by the cases of single-allele expression or both-allele expression. Only detected KIR genes are displayed. (D) Comparison of the KIR allele fractions between CD39⁻ CD8⁺ T cells and CD39⁺ CD8⁺ T cells. Each line connects a KIR allele in the CD39⁻ cell and the CD39⁺ T cell from the same patient. The P-values are computed with a Wilcoxon signed-rank test and have been adjusted by the Benjamini–Hochberg procedure.