HLA Eplet-based risk stratification for DSA and ABMR after kidney transplantation with validation of immunogenicity and immunoreactivity

Introduction: Antibody-mediated rejection (ABMR) is a major cause of graft dysfunction and graft loss after kidney transplantation. Donor-specific antibodies (DSA) play a key role in ABMR, and donor-recipient HLA mismatch is a major determinant of DSA formation. Although HLA eplet mismatch burden is increasingly used for immunological risk assessment, the clinical relevance of individual mismatched eplets remains incompletely defined. We aimed to develop an eplet-based risk stratification framework and to validate its immunogenicity- and immune reactivity-related basis.
Methods: We conducted a single-center retrospective study in a Chinese population including 268 kidney transplant donor-recipient pairs with complete HLA typing, eplet mismatch data, and clinical follow-up. The primary outcomes were post-transplant DSA and ABMR. HLA eplet mismatches were assessed using HLA Matchmaker Fusion, and eplet co-occurrence patterns were analyzed using Jaccard similarity and hierarchical clustering. Eplet-related features and clinical variables were integrated to build predictive models using Elastic Net regression, logistic regression, and random forest. For mechanistic validation, physicochemical disparities of donor-recipient mismatched amino acids were quantified, and IgG subclasses and Fc glycosylation features were evaluated.
Results: Among the 268 donor-recipient pairs, 82 developed DSA and 45 developed ABMR. Eplet clustering revealed non-random co-occurrence patterns among mismatched eplets. Class II HLA mismatches, particularly HLA-DQ eplet mismatches, showed stronger associations with DSA and ABMR. Elastic Net analysis identified several high-risk eplets, including HLA-DQB1 55PP. The model outperformed conventional HLA molecular mismatch in predicting DSA (AUC 0.82 vs. 0.68, P<0.01) and ABMR (AUC 0.81 vs. 0.65, P<0.01). Mechanistic analyses showed significant differences in physicochemical disparities of mismatched amino acids between the high- and low-risk groups (P<0.05). IgG1 afucosylation (P<0.05) and IgG3 agalactosylation (P<0.01) were also significantly increased in the high-risk group.
Conclusion: The clinical impact of HLA mismatch depends not only on mismatch burden but also on the properties of individual eplets and their differences in immunogenicity and immune reactivity. Integrating eplet clustering, machine learning, and mechanistic validation may provide a practical framework for risk stratification for DSA and ABMR after kidney transplantation.