Integrated In Silico Prediction of Minor Histocompatibility Antigen Peptides in HLA-Matched Allogeneic Stem Cell Transplantation for Acute Myeloid Leukemia

Background and Rationale: T-cell responses to minor histocompatibility antigens (mHA) are important drivers of the beneficial graft versus leukemia (GvL) effect and harmful graft versus host disease (GvHD) pathology following HLA-matched allogeneic stem cell transplantation (alloSCT). Despite their importance, Genome Wide Association Studies (GWAS) have failed to elicit a prognostic set of individual mHA associated with the clinically observed GvL and GvH effects of alloSCT [Sato-Otsubo et al. Blood 2015]. This is likely due to a lack of public mHA shared across the global set of donor/recipient pairs (DRPs). Even an optimally frequent single nucleotide polymorphism (SNP) would result in a recipient restricted genetic variant in only 24% of DRPs with a matched unrelated donor (MUD) or 17% of DRPs with a matched related donor (MRD) [Armistead et al. PLoS One 2011]. Moreover, even when DRPs do share a recipient restricted genetic variant, any resulting peptide would have to bind a MHC molecule within the recipient for presentation to T-cells. Methods: To evaluate the role of mHA in alloSCT without requiring public mHA, we developed a bioinformatics pipeline to predict mHA peptides based on SNP differences taking into account peptide/MHC binding estimated by netMHCpan [Nielson et al. Genome Med 2016] and expected GvL vs GvH tissue expression derived from mRNA sequencing data from acute myeloid leukemia (AML) and normal hematopoietic, hepatobiliary, skin, and gastrointestinal tract tissues. In order to understand the distribution of mHA in HLA-matched alloSCT drawn from a diverse pool of DRPs, we applied this analysis to putative DRPs drawn from healthy individuals who underwent whole-exome sequencing as part of the 1000 Genomes Project (n = 1,916) [1000 Genomes Project Consortium et al. Nature 2015]. To evaluate the association of mHA with AML relapse and GvHD incidence following alloSCT, we predicted mHA from SNP data using clinical information from an earlier study [Armistead et al. PLoS One 2011]. Results: To determine a baseline for the number of predicted mHA in an alloSCT, we considered every possible pair of samples in the 1000 Genomes data set as a theoretical alloSCT (n = 3,669,140). We determined each sample's HLA type using PHLAT [Bai et al. BMC Genomics 2014] and then performed mHA predictions for all theoretical transplants with a 10 out of 10 HLA match (n = 10). Within each ethnicity represented in the 1000 Genomes data, the degree of HLA matching was greater than that of all ethnicities pooled, with the lowest HLA diversity in the Finnish, Chinese, and Japanese populations. The number of predicted mHA binding MHC with Kd < 500nM ranged from 6,217 to 13,545 in the 1000 Genomes theoretical HLA-matched DRPs, with mHA contained in genes selectively expressed in AML versus those selectively expressed in GvH target organs numbering 213 to 610 and 367 to 1,135, respectively. HLA-A*02:01 restricted mHA were predicted for 37 actual DRPs in the context of MUD alloSCT and 97 DRPs from MRD alloSCT [Armistead et al. PLoS One 2011]. There were significantly more predicted mHA in MUD transplants (Figure 1). Taking into account both predicted peptide/MHC binding and tissue expression, the number of predicted GvL mHA was significantly associated with remission and the aggregate number of predicted GvH mHA was significantly associated with grade 2-4 GvHD incidence in MUD transplants (Figure 2). Conclusions: Prediction of mHA based on whole-exome sequencing data is feasible and can be used to discover associations of mHA distribution features with clinical outcomes including AML remission and GvHD incidence. Future work in larger datasets will be required to validate these predicted associations and guide development of mHA-directed therapeutics. Disclosures Molldrem: Astellas Pharma: Patents & Royalties.