Clustering analysis based on HLA-DP polymorphic positions reveals structurally and functionally divergent “core” and “non-core” HLA-DPB1 alleles predictive of clinical outcome. (A) Amino acid variation at 28 polymorphic positions (amino acids 8-215) in HLA-DPB1 coding sequences was used to cluster alleles according to their structural similarity. TCE3 “core” alleles (ie, DPB1*02:01, 04:01, 04:02, 23:01) form a distinct cluster separate from other alleles in this TCE group. For clarity, only the 19 most common alleles in the cohort, with cumulative frequencies of 98.7% and 98.8% in the patients and donors, respectively, are shown. (B) Mean in vitro alloreactive responses (% CD4⁺CD137⁺) from self-TCE3 “core” healthy donors are lowest (18.5% ± 15.2%; n = 47) against permissive TCE3 “core” alloantigens (DPB1*02:01, 04:01, 04:02) and maximal (37.2% ± 12.7%; n = 85) against nonpermissive TCE1 alleles (DPB1*09:01, 10:01, 17:01). Mean alloresponses against representative “non-core” TCE3 alleles (DPB1*01:01, 05:01, 15:01) are intermediate (29.2% ± 16.2%; n = 69) between these 2 extremes. Bars indicate mean with standard deviation. **P < .01; ***P < .001; ****P < .0001. A total of 187/201 cultures were included in Meurer et al¹³ and reanalyzed for this work. (C-D) Forest plots show the HR and 95% CI for (C) aGVHD II-IV and (D) TRM for the DP matching subgroups in the “classic” TCE model and the TCE3 “core” vs “non-core” stratification strategy (model II). HLA-DP nonpermissively mismatched pairs are shown as reference and overall P values of the adjusted models are presented. For statistically significant models (overall P < .01), P values for individual groups are also indicated. (E-F) Cumulative incidence (top panels) and Kaplan-Meier (bottom panels) estimates for (E) TRM and (F) overall survival in the cohort are plotted for the TCE3 “core” and “non-core” permissive subgroups. Results from statistical comparison of the curves with Gray’s test and log-rank are indicated.