Abstract
The biological processes underlying NK cell alloreactivity in haematopoietic stem cell transplantation (HSCT) remain unclear. Many different models to predict NK alloreactivity through KIR and MHC genotyping exist, raising ambiguities in its utility and application for clinicians. We assessed 27 predictive models, broadly divided into six categories of alloreactivity prediction: ligand-ligand, receptor-ligand, educational, KIR haplotype-based, KIR matching and KIR allelic polymorphism. The models were applied to 78 NGS-typed donor/recipient pairs undergoing allogeneic HSCT in genoidentical (n=43) or haploidentical (n=35) matchings. Correlations between different predictive models differed widely, suggesting that the choice of the model in predicting NK alloreactivity matters. For example, two broadly used models, educational and receptor-ligand, led to opposing predictions especially in the genoidentical cohort. Correlations also depended on the matching fashion, suggesting that this parameter should also be taken into account in the choice of the scoring strategy. The number of centromeric B-motifs was the only model strongly correlated with the incidence of acute graft-versus-host disease in our set of patients in both the genoidentical and the haploidentical cohorts, suggesting that KIR-based alloreactivity, not MHC mismatches, are responsible for it. To our best knowledge, this paper is the first to experimentally compare NK alloreactivity prediction models within a cohort of genoidentical and haploidentical donor-recipient pairs. This study helps to resolve current discrepancies in KIR-based alloreactivity predictions and highlights the need for deeper consideration of the models used in clinical studies as well as in medical practice.