Significant variation between SNP-based HLA imputations in diverse populations: the last mile is the hardest

Abstract

Four SNP-based HLA imputation methods (e-HLA, HIBAG, HLA*IMP:02 and MAGPrediction) were trained using 1000 Genomes SNP and HLA genotypes and assessed for their ability to accurately impute molecular HLA-A, -B, -C, and –DRB1 genotypes in the Human Genome Diversity Project cell panel. Imputation concordance was high (> 89%) across all methods for both HLA-A and HLA-C, but HLA-B and HLA-DRB1 proved generally difficult to impute. Overall, less than 27.8% of subjects were correctly imputed for all HLA loci by any method. Concordance across all loci was not enhanced via the application of confidence thresholds; reliance on confidence scores across methods only led to noticeable improvement (+3.2%) for HLA-DRB1. As the HLA complex is highly relevant to the study of human health and disease, a standardized assessment of SNP-based HLA imputation methods is crucial for advancing genomic research. Considerable room remains for the improvement of HLA-B and especially HLA–DRB1 imputation methods, and no imputation method is as accurate as molecular genotyping. The application of large, ancestrally diverse HLA and SNP reference datasets and multiple imputation methods has the potential to make SNP-based HLA imputation methods a tractable option for determining HLA genotypes.

The PAGE Study: How Genetic Diversity Improves Our Understanding of the Architecture of Complex Traits

“…44 The rs2395029 (G) variant in HCP5, a near perfect tag of HLA-B*57:01 (in Europeans), is used to screen for abacavir hypersensitivity. 45 The tag SNP utility remains relatively high (r ~ 0.92, 46 ) across globally diverse populations in the 1000 Genomes Project. Using PAGE and Global Reference Panel samples, we show that risk allele frequencies for rs2395029 rise above 5% in multiple large South Asian populations, and rise above 1% within some, but not all, admixed populations with Native American ancestry (Figure 4).…”

The complexity and diversity of major histocompatibility complex challenge disease association studies

“…This was evident in a study where HLA‐DRB1 imputation showed very low accuracy rate (<30%) for some alleles such as HLA‐DRB1*01:01, HLA‐DRB1*13:01, and HLA‐DRB1*15:01 belonging to the five most frequent HLA‐DRB1 alleles in the Finnish population when the reference material was from a large heterogeneous European population and not from the Finnish population . Low imputation accuracy for HLA‐DRB1 has also been described by other studies covering heterogeneous study cohorts or diverse ancestry groups . The reasons may be low HLA‐DRB1 tagging of the SNPs used due to the absence of the informative SNPs for HLA‐DRB1 region, inadequate numbers of reference SNPs and its poor representativeness for diverse populations …”

“…HLA‐DRB1*15:01 and HLA‐DQB1*06:02 have also been found to be resistance genes for HIV‐1 infection in Chinese and in European‐descent populations, respectively, corresponding to the protective genes in T1D . HLA‐DRB1*15:01 seems to be protective in Grave's disease and the haplotype DRB1*15:01‐DQA1*01:02‐DQB1*06:02 in European‐descent patients with autoimmune polyglandular syndrome and in Moroccan patients with pemphigus . In the Finnish population, we found the haplotype HLA‐B*07‐DRB1*15:01 to be protective for coronary artery disease .…”

“…44 Low imputation accuracy for HLA-DRB1 has also been described by other studies covering heterogeneous study cohorts or diverse ancestry groups. 45,46 The reasons may be low HLA-DRB1 tagging of the SNPs used due to the absence of the informative SNPs for HLA-DRB1 region, inadequate numbers of reference SNPs and its poor representativeness for diverse populations. 25 Phasing of the HLA haplotypes is performed with computational methods, which rely on probabilistic models and use them to predict the haplotype structure.…”

“…42,43,82,83 HLA-DRB1*15:01 seems to be protective in Grave's disease and the haplotype DRB1*15:01-DQA1*01:02-DQB1*06:02 in European-descent patients with autoimmune polyglandular syndrome and in Moroccan patients with pemphigus. [44][45][46][47] In the Finnish population, we found the haplotype HLA-B*07-DRB1*15:01 to be protective for coronary artery disease 84 . In other studies, however, these same alleles/combinations thereof have been established in neurological disorders such as Alzheimer's disease and multiple sclerosis, Parkinson's disease, narcolepsy, and autism (Table 4).…”