HLA-DQ molecular mismatch-derived risk score predicts acute rejection and development of de novo donor-specific antibodies in pediatric liver transplantation

Background: Although short-term outcomes after pediatric liver transplantation (PLT) are favorable, long-term graft and patient survival remains limited by alloimmune injury and prolonged immunosuppression exposure. In PLT, the contribution of donor-recipient HLA molecular mismatch (MM) to alloimmune risk remains insufficiently characterized. We prospectively evaluated the impact of HLA MM on biopsy-proven acute rejection (BPAR) and de-novo DSA (dnDSA) development.
Methods: In this single-center prospective longitudinal cohort, PLT recipients transplanted at Hospital de Pediatría JP Garrahan between 2018 and 2023 were followed. High-resolution donor and recipient HLA typing was performed by next-generation sequencing. MM was assessed using eplet mismatch (eMM), PIRCHE-II, and Snow-derived peptide scores. The primary outcomes were BPAR and dnDSA development. Individual mismatched eplets and peptides were encoded as binary variables. To estimate their relative immunogenicity, penalized Cox models were applied and outcome-specific epitope risk scores (ERS) were derived. Multivariate Cox models were used to evaluate the independent contribution of ERS beyond clinical variables.
Results: A total of 114 patients were HLA-typed and prospectively followed. Biopsy results were available in 111 patients and DSA surveillance in 109. BPAR was diagnosed in 74 (69%) patients, including 28 cases detected only in surveillance biopsies. dnDSA developed in 35 (32%) patients, predominantly class II, with DQ dnDSA as the most frequent specificity (n=23). First BPAR occurred significantly earlier than first DSA detection (median 0.97 vs 2.50 years post-transplant, p<0.01). Conventional total molecular mismatch metrics were not associated with BPAR (p>0.05), although HLA-DQ-restricted mismatch showed a trend (p=0.09). In contrast, dnDSA-positive patients had significantly higher total eMM and Snow scores (p<0.05), with stronger associations at the HLA-DQ locus (p<0.01). Ridge-penalized Cox models outperformed lasso and elastic net approaches for feature weighting. DQ eplet-ERS (DQ-eERS) outperformed DQ snow-ERS and conventional cumulative mismatch metrics. In multivariate analysis, tacrolimus variability measured as tortuosity (HR 4.50, 95%CI 3.84-17.79, p<0.001) and DQ-eERS (HR 8.26, 95%CI 2.12-9.56, p<0.001) were independent predictors of BPAR, while DQ-eERS (HR 7.42, 95%CI 3.44-15.99, p<0.001) and prior BPAR (HR 3.03, 95%CI 1.11-8.25, p=0.030) were independent predictors of dnDQ-DSA.
Conclusions: HLA-DQ MM is clinically relevant in PLT, particularly when weighted according to the relative immunogenicity of individual eplets. Epitope-weighted ERS improved alloimmune risk stratification beyond conventional mismatch burden and identified patients at increased risk of BPAR and dnDQ-DSA. These findings support the incorporation of molecular immunological risk assessment into individualized immunosuppression strategies after PLT.