Towards Identifying the Target of Autoimmunity in Aplastic Anemia

Acquired aplastic anemia (AA) is an autoimmune bone marrow aplasia caused by T cell-mediated destruction of hematopoietic stem and progenitor cells. The antigenic target in AA remains unknown. Recently, we and others identified frequent somatic loss of function of several human leukocyte antigen (HLA) class I alleles in hematopoietic cells that survive the AA immune attack, suggesting these alleles ("risk alleles") present autoantigen in the affected patients. We hypothesize that risk alleles share structural features and peptide-binding characteristics that may inform our understanding of the immune mechanism of AA. To identify additional AA HLA class I risk alleles, we have partnered with two multi-institutional consortia, the North American Pediatric Aplastic Anemia Consortium (NAPAAC) and the Center for International Blood and Marrow Transplant Research (CIBMTR), to evaluate 507 AA patients for somatic HLA loss. Using a combination of targeted massively parallel sequencing of HLA class I genes and single nucleotide polymorphism array genotyping, we identified HLA loss in 19% of the NAPAAC cohort (30 of 156 unselected AA patients) and 13% of the CIBMTR cohort (46 of 351 patients selected to maximize discovery of less common risk alleles). Copy number-neutral loss of heterozygosity of chromosome arm 6p (6p CN-LOH) was the most common cause of HLA loss, occurring in 50 patients (10% of total cohort, 12% NAAPAC and 9% CIBMTR). Somatic mutations were present in 45 patients (9% of total cohort, 12% of NAPAAC and 7% CIBMTR). Patients with HLA loss had a median of 1 mutant clone per patient (range 1-7). Of the somatic mutations, 80% were predicted to disrupt expression of the affected allele (through loss of start, nonsense, or frameshift mutations). The other 20% were missense mutations affecting residues in the peptide binding groove (6 mutations) and the α3 domain (9 mutations) of the HLA class I protein. A total of 19 distinct AA HLA risk alleles were identified, of which 13 were newly identified in this study. Several risk alleles (HLA-B*14:02, HLA-B*40:02, HLA-A*02:01) were enriched in AA patients compared to ethnicity matched controls. Among the 507 patients in the study, at least one risk allele was present in 436/493 (88%) of AA patients and in 34/39 (87%) of the 6p CN-LOH events. Mutations clustered in several groups of alleles (supertypes) known to bind overlapping peptide repertoires, with alleles in B27 and B44 supertypes being most commonly affected. There were no mutations in A01, A01A03, A01A24, A24, B58, B62, or in any of the HLA-C alleles analyzed. Using several metrics to estimate the strength of the autoimmune selection on a given allele, including the frequency of somatic loss and similarity to other HLA risk alleles, we developed an AA HLA risk allele pathogenicity index ranging from very high (B*14:02 and B*40:02) and high pathogenicity (A*33:03, B*08:01, B*13:02, B*14:01, B*27:03, B*27:05, B*38:02, B*41:02 and B*49:01) to those unlikely to be pathogenic. Interestingly, even within high risk supertypes such as B44, there was a wide variation in predicted pathogenicity, ranging from high (for B*40:02 and B*41:02) to low pathogenicity (for B*44:02 and B*44:03), suggesting differences in autopeptide binding. Our study provides a comprehensive analysis of AA HLA risk alleles in a large diverse cohort of AA patients. Our results suggest that HLA risk alleles have shared autoantigen binding specificities that define their pathogenicity in AA, which can be used to identify candidate AA autoantigens. HLA risk allele pathogenicity may have future clinical utility as an adjunctive diagnostic test, as well as for prognostic assessment and haploidentical donor selection.