Abstract Introduction: HLA allelic diversity allows for the presentation of broad peptide repertoires to alloreactive donor T-cells and can potentially enhance the graft-versus-leukemia effect (GVL) in allogeneic stem cell transplantation (alloSCT). Several studies have reported HLA evolutionary divergence (HED) as a predictor of clinical response to immune checkpoint inhibitors in both solid tumors and alloSCT. Yet, the optimal method for calculating HED and its clinical significance in alloSCT remain undefined. A conventional method for scoring HED is to calculate the Grantham distances of antigen-binding domain amino acid residues. However, the presence of conserved residues among these domains can potentially underestimate clinically meaningful HLA divergence. We hypothesized that an HED calculation method restricted to polymorphic HLA residues might improve HLA divergence estimation, which in turn could enhance the utility of HED scoring for predicting clinical outcomes in alloSCT. Here, we test a new HED scoring approach focused on polymorphic HLA residues and validate its clinical significance in a multi-center cohort of patients with acute myeloid leukemia (AML) who underwent haploidentical alloSCT. Methods: We retrospectively analyzed AML patients who underwent haploidentical alloSCT at the University of Pittsburgh (U. Pitt) and City of Hope (COH) between 2017 and 2023. Conventional total antigen-binding domain (TBD)-based Grantham scores were calculated from antigen-binding residues of HLA class-I/II alleles included with the GranthamDist package https://sourceforge.net/projects/granthamdist/). We also computed Grantham scores from highly polymorphic antigen-binding domain residues (PBD), which were defined as amino acid sequence paralogues with Simpson Diversity indices of =0.2 within a given HLA class (HLA-A, -B, and -C or HLA-DRB1, -DQB1, and- DPB1) Overall survival (OS), relapse rate, and non-relapse mortality (NRM) were estimated by Kaplan Meier curve and compared between low-HED (50 percentile) and high-HED (≥50 percentile) scoring recipients using log-rank analysis. Multivariate analysis was performed with Cox proportional hazard models, including known risk factors predictive of post-transplant relapse: age, conditioning intensity, disease risk index (DRI), and pre-transplant measurable residual diseases. Results: We first compared TBD- and PBD-HED scoring methods by analyzing a U Pitt cohort of 49 patients, in which the median age was 60, with most patients having received reduced intensity conditioning (60.4%). As expected, across all HLA loci, PBD-HED scores were significantly higher than TBD-HED scores. Interestingly, survival of patients with low PBD-HED was significantly inferior to that of patients with high PBD-HED (p=0.02, HR 2.26), whereas TBD-HED did not predict survival (p=0.27). In multivariate analysis, low PBD-HED was an independent risk factor for survival in addition to DRI and conditioning intensity. To validate these findings, we next analyzed a COH cohort (n=145) including younger recipients (median age 56) treated with myeloablative conditioning (67%). Low PBD-HED score was significantly associated with lower survival (P=0.006, HR 2.17) and higher incidence of relapse (P=0.01, HR 2.61), compared to high PBD-HED by univariate analysis. Conclusion: In this study, a new PBD-based HED score significantly predicted the post-transplant survival of AML following haplo-SCT in two independent cohorts. Low HED may foster leukemia escape from GVL in AML in a mechanism similar to what has been described in loss of HLA heterozygosity. Incorporating this new PBD-based HED score into current risk models could improve the prediction of transplant outcomes, thereby identifying patients at high risk for relapse who may benefit from post-transplant prophylactic therapy. Our observations also suggest the importance of HLA evolutionary diversity for optimal GVL and long-term leukemia control.
Cooper et al. (Mon,) studied this question.