HLA mismatch-derived epitopes predict acute graft-versus-host disease risk after unrelated donor transplantation

Abstract

Abstract Background: Complete and precise HLA matching of unrelated donors reduces the risk of complications following allogeneic hematopoietic cell transplantation (allo-HCT). However, many patients – particularly those of non-European ancestry – lack fully compatible donors. Transplantation from partially matched donors can be life-saving, but the increased risk of graft-versus-host disease (GvHD) limits its broader application. Identifying those immunological features that render certain HLA mismatches more immunogenic remains a critical area of investigation. We hypothesized that quantifying the repertoire of peptides uniquely derived from mismatched HLA alleles, and capable of being indirectly presented on shared or donor-derived HLA molecules could contribute to a more refined and personalized assessment of GvHD risk in the setting of mismatched unrelated (MMUD) allo-HCT. Methods: We conducted a retrospective cohort study of 1,003 patients who received single HLA class I-mismatched (HLA-A, -B and -C) unrelated allo-HCT between 2008-2018, using data from the Center for International Blood and Marrow Transplant Research (CIBMTR). Post-transplant cyclophosphamide (PTCy) transplants were excluded. Starting from the high-resolution HLA typing available for each donor-recipient (D-R) pair under evaluation, our computational pipeline incorporated the following steps: HLA sequence extrapolation, in silico generation of all possible 8-11 amino acid-long peptides (k-mers), and their sequence alignment in order to retain only those specific for the mismatched HLA. The identified k-mers were then subjected to HLA binding prediction, and the resulting HLA mismatch-derived epitope load was calculated as the number of allogenic HLA peptides potentially presented by HLA class I molecules. Findings: The median number of HLA mismatch-derived epitopes was 4 (range: 0-19); as expected, the number of such epitopes was higher in the presence of antigen-level (median: 6, range: 0-12) compared to allele-level mismatches (median: 1, range: 0-19; P <0.0001). To confirm the indirect presentation of allogeneic HLA-derived peptides predicted by our algorithm, we interrogated the HLA Ligand Atlas (Marcu JITC 2021), a publicly available mass spectrometry-based immunopeptidome repository from healthy donor-derived tissues (bone marrow, lymph node, thymus, skin, liver, stomach, small intestine, colon, lung, and brain). HLA ligandome analysis confirmed HLA presentation for 15.5% of predicted HLA mismatch-derived epitopes, a rate higher than that achieved in antigenicity validation of minor histocompatibility antigens (mHAgs) and neoantigens. Grade II-IV acute GvHD was more frequent in patients with an HLA mismatch-derived epitope load above the median (50%, 95% CI: 45-55%) than in those below (39%, 95% CI: 33-45%; p=0.005). To construct a multivariable model, we considered diagnosis, disease status at transplant, patient age, conditioning intensity, graft source, type of GvHD prophylaxis, female-to-male sex mismatch, CMV serostatus, donor age, DPB1 TCE mismatch, and HLA mismatch-derived epitope load. Notably, only HLA mismatch-derived epitope load greater than the median, together with diagnosis and donor age, was associated with an increased risk of developing grades II–IV acute GvHD (hazard ratio = 1.46, 95% confidence interval: 1.15-1.86, P = 0.002). HLA mismatch-derived load above the median was also protective against relapse (P = 0.047), although this did not translate into an overall survival benefit, likely because non-relapse mortality showed a trend toward higher incidence in the high-load group (36%, confidence interval: 30-42%) versus the low-load group (28%, confidence interval: 24-33%; P = 0.051). Patients with an HLA mismatch-derived epitope burden above the median not only relapsed less, but also showed a trend toward delayed relapse, with median not reached versus ~48 months in the low-burden group (log-rank P = 0.09), raising the possibility of a higher incidence of late immune escape events such as HLA loss. Conclusion: HLA mismatch-derived epitope load shows promise as a tool to inform GvHD risk after MMUD allo-HCT and, if validated in independent cohorts, offers an easily implementable approach to support donor selection. Combined with mHAg prediction (Cieri Nat Biotechnol 2024), this framework may help to dissect the interplay and relative contributions of HLA mismatches and mHAgs in shaping alloreactivity, GvHD, and relapse.