Inaccuracies in epitope repertoire estimations when using multilocus allele-level HLA genotype imputation tools

Limited availability of allele-level HLA genotypes prompts their imputation from allele-group genotypes to estimate epitope mismatches. We evaluated the accuracy of epitope load and repertoire assignment when imputing allele-level HLA genotypes. Analyses were conducted on 175 hematopoietic stem cell (HSC) donors from the Héma-Québec registry (HQR) and 57 HSC donor-recipient pairs from McGill University Health Centre (MUHC), Québec, Canada, genotyped for HLA-A, -B, -C, -DRB1, and -DQB1. Multilocus allele-level imputation was performed using HaploStats. Disagreement in B- and T-cell epitope assignment and epitope mismatches were ascertained for imputed vs measured allele-level HLA genotypes in HSC donors and donor-recipient pairs, respectively. Imputation resulted in no differences in overall eplet mismatches and Predicted Indirectly ReCognizable HLA Epitopes (PIRCHE-II) for HLA-A, -B, and -C in 83.4% and 93.7% of HQR donors and 87.7% and 87.7% of MUHC donors, respectively. HLA-DRB1- and -DQB1-derived eplet mismatches and PIRCHE-II were correctly assigned in 72.0% and 85.1% of HQR donors and 70.2% and 71.9% of MUHC donors, respectively. No discrepancies in eplet load or PIRCHE-II were observed in 96.5% and 86.0% of HSC donor-recipient pairs and in 70.2% and 70.1% of pairs for HLA-A, -B, -C, -DRB1, and -DQB1, respectively. Kappa statistics of 0.9708 and 0.9725, 0.8724 and 0.8177, 0.9827 and 0.9022, 0.5644 and 0.4939, 0.5085 and 0.6361 were estimated when assessing agreement between eplet mismatches and PIRCHE-II of imputed vs measured HLA-A, -B, -C, -DRB1, and -DQB1 types, respectively. To avoid inaccuracies in epitope compatibility estimation, mainly for HLA class II, multilocus allele-level genotype measurement is recommended.