Class II HLA supertype diversity predicts poor outcomes with immune checkpoint inhibitors in a pan-cancer cohort

Background: HLA diversity shapes the T-cell repertoire and antigen presentation, with potential effects on immune checkpoint inhibitor (ICI) outcomes. Several diversity metrics have been proposed, including zygosity, HLA evolutionary divergence (HED), and HLA supertype diversity (functional peptide-binding groups). Their relative value in pan-cancer cohorts remains unclear. Methods: We examined the relationship between HLA diversity and clinical outcomes in a pan-tumor cohort of 140 patients treated with ICIs at Johns Hopkins. DNA was extracted from peripheral blood mononuclear cells and genotyped using the Illumina Infinium Global Screening Array v3.0. Class I and Class II HLA alleles were imputed from SNP genotype data using standard reference panels, and from these we derived Class I and Class II zygosity, HLA epitope diversity (Grantham distance matrix-based HED), and Class I and Class II supertype diversity (9 Class I and 7 Class II functional groups from IMGT/HLA). High versus low diversity was defined by cohort median splits. Treatment response was defined as complete or partial response. Multivariate logistic regression adjusted for age, gender, race, BMI, cancer type, MSI/TMB status, line of therapy, therapy class, and all HLA diversity metrics simultaneously. Results: In the fully adjusted model, only Class II supertype diversity remained significant. Patients with at least four distinct supertypes had lower response rates than those with fewer than four (23.2% vs 36.9%, OR 0.199 [0.060-0.608], p=0.006). The effect was unchanged in models without cancer type terms (OR 0.199, p=0.005) and in models conditioning on DR4 and DPmain (OR 0.235, p=0.019). Spline analyses supported the four-supertype cutoff (LR test p=0.274), and no interactions were observed with line of therapy, MSI/TMB, or treatment class. Cox models showed worse OS (HR 1.834 [1.017-3.304], p=0.044) and PFS (HR 1.981 [1.067-3.678], p=0.030), with similar HR in cancer type-stratified and DR4 or DPmain-conditioned analyses. Individual Class II supertypes were consistent with the composite index: DR4 carriers had worse OS (HR 2.062, p=0.007) and PFS (HR 1.872, p=0.025), and DPmain carriers had worse OS (HR 1.524, p=0.043). Alternative constructs including Class I/II zygosity and HED were less informative or nonsignificant. Conclusions: Class II HLA diversity emerged as an unexpected negative predictor of response and survival with ICI therapy. These findings differ from results reported in other ICI-treated cohorts. Although increased HLA diversity can broaden antigen recognition in some settings, prior studies also show that greater Class II diversity can increase thymic deletion of self-reactive T cells. This process reduces the breadth of the peripheral T-cell repertoire and limits tumor neoantigen recognition, providing a biologically plausible explanation for the pattern observed here.