Common minor histocompatibility antigen discovery based upon patient clinical outcomes and genomic data

Abstract

Background: Minor histocompatibility antigens (mHA) mediate much of the graft vs. leukemia (GvL) effect and graft vs. host disease (GvHD) in patients who undergo allogeneic stem cell transplantation (SCT). Therapeutic decision making and treatments based upon mHAs will require the evaluation of multiple candidate mHAs and the selection of those with the potential to have the greatest impact on clinical outcomes. We hypothesized that common, immunodominant mHAs, which are presented by HLA-A, B, and C molecules, can mediate clinically significant GvL and/or GvHD, and that these mHAs can be identified through association of genomic data with clinical outcomes.

Methodology/principal findings: Because most mHAs result from donor/recipient cSNP disparities, we genotyped 57 myeloid leukemia patients and their donors at 13,917 cSNPs. We correlated the frequency of genetically predicted mHA disparities with clinical evidence of an immune response and then computationally screened all peptides mapping to the highly associated cSNPs for their ability to bind to HLA molecules. As proof-of-concept, we analyzed one predicted antigen, T4A, whose mHA mismatch trended towards improved overall and disease free survival in our cohort. T4A mHA mismatches occurred at the maximum theoretical frequency for any given SCT. T4A-specific CD8+ T lymphocytes (CTLs) were detected in 3 of 4 evaluable post-transplant patients predicted to have a T4A mismatch.

Conclusions/significance: Our method is the first to combine clinical outcomes data with genomics and bioinformatics methods to predict and confirm a mHA. Refinement of this method should enable the discovery of clinically relevant mHAs in the majority of transplant patients and possibly lead to novel immunotherapeutics.

Figure 1. Schematic of mHA selection method.

All 57 patients were classified by GvHD and remission status (1A). The patients in the remission, GvHD+; remission, GvHD-; and relapse, GvHD+ groups were considered as the clinical immune response cohort (cIR+, blue shaded boxes), and the relapse, GvHD- group was considered the clinical non-immune responder group (cIR-, unshaded box). All potential allele combinations for any cSNP in the setting of allogeneic SCT can be represented by a 3×3 table( 1B). Given a major allele ‘A’ and minor allele ‘a’ at any particular cSNP locus, then a donor (D) may be homozygous for the major allele (AA, genotype frequency  =  p²), heterozygous (Aa, genotype frequency  = 2pq), or homozygous for the minor allele (aa, genotype frequency  =  q²). The same assignments can be made for the recipient (R). Assuming potential minor histocompatibility antigens arise from non-synonymous cSNPs, then the genotype frequencies can be multiplied to give the predicted frequencies of a donor/recipient immune reaction against peptide antigens encoded by that locus. For a donor that is homozygous at a given cSNP, 3 potential immune responses can be predicted based on the genotype of the recipient: 1) tolerance, 2) non-tolerance in the GvL/GvHD direction, and 3) non-tolerance in both the GvL/GvHD and rejection directions. Scenarios 2 and 3 are defined as genetically predictive of an immune response (gIR+, red shaded boxes). No gIR is predicted if a donor is heterozygous at a given locus or if the donor and recipient share the same genotype (gIR-, unshaded boxes). The association between gIR+ and cIR+ for each allele in each cSNP was determined using Fisher's exact test (1C).

Figure 2. Peptide binding to HLA-A2.

All synthesized peptides from Table 2- were tested for binding to HLAA2 using a T2 cell binding assay. Unpulsed T2 cells are shown as shaded traces. Test peptide-pulsed T2 cells are shown as traces with dashed lines, and PR1-pulsed T2 cells are shown as solid black lines. The PR1 peptide was used as a positive control because it binds HLA-A2 with high affinity. Five of 17 peptides bound to HLA-A2 with an increase in geometric mean fluorescence of >1.5 fold compared to the non-pulsed control (2A–2E). A representative non-binding peptide is shown (2F).

Figure 3. T4A-specific CD8+ T cell expansion in post-SCT patient samples.

T4A-specific CD8+ T cells are detected in post-transplant samples using T4A loaded, HLA-A2/PE conjugated tetramers. Of four patients with gIR+ to the cSNP at position rs9876490_C with available post-SCT PBMC samples, T4A-tetramer+ CD8+ lymphocytes were identified in 3 patients (3A–3C). The fourth patient with no detectable T4A-specific T lymphocytes was receiving systemic steroids for extensive cGvHD (3D). No T4A/HLA-A2 tetramer+ cells were detected in a control HLA-A2 negative recipient (3E), or in two HLA-A2+ post-transplant T4A gIR- patients (3F, 3G). The mean percentage of T4A-tetramer+ cells in the CD8+ population in samples 3A, 3B, and 3C was 0.073±0.015% compared to 0.021±0.001% for samples 3E, 3F, 3G (student t-test, P = 0.0278).

Figure 4. T4A binding, tissue expression, and antigen frequency.

The iTopia epitope binding assay was used to measure T4A and the alternately encoded T4E peptide (GLYTYWSAGE) binding to HLA-A2 compared to the percentage of maximal binding of the iTopia control peptide (FLPSDFFPSV). T4A binding to HLA-A0201 is 95% of that compared to the positive control; however, T4E binds HLA-A0201 at 24% that of the positive control. According to the manufacturer, peptides binding of >30% that of the control peptide are candidate HLA-A0201 epitopes (4A). The iTopia assay was also used to measure the ED₅₀ and dissociation t_1/2 of T4A to HLA-A2 relative to the iTopia positive control (FLPSDFFPSV) labeled “POS”. T4A binds with a 50% effective dose of ED₅₀ = 1.3 µM and a dissociation half time of t_1/2 = 1.00 hr (4B, 4C). Western blots on whole cell lysates of normal human tissue (colon, heart, liver, skin, testis and PBMC), 4 human AML samples and Jurkat cells (positive control) are shown (4D). Western blotting of Jurkat cells reveals 2 bands of roughly 100 kDa (full length TRIM42) and 50 kDa (unknown protein product). The TRIM42 band is seen at various levels in all AML samples and faintly in PBMC. No TRIM42 protein is detected in the other human tissues. GAPDH expression (40 kDa) was used as a loading control. Subcellular fractionation was performed on AML blasts and isolated healthy donor granulocytes (4E). The fractions are labeled C, cytosolic; N, nuclear; M, membrane; Sk, cytoskeletal. No TRIM42 protein is observed in granulocytes, but the same bands seen in the Jurkat cell positive control are detected in the N and Sk fractions of the AML blasts. The allele frequencies and genotypes for rs9876490, the T4A associated cSNP, in the original ethnic populations used by the International HapMap Project (CEU, HCB, JPT, YRI) are shown (4F). The T4A recipient phenotypes (and frequencies in the CEU population) are AC (0.450) or CC (0.133), and the donor genotype is AA (0.417). The predicted gIR+ in an unrelated transplant scenario can be calculated: P(AA) × P(AC + CC) and applied over a range of allele frequencies yielding the MUD SCT curve (4G). When parental genotypes are considered in the MRD SCT setting (Supplement 1) a similar curve with generally lower gIR+ frequency at each donor allele frequency is generated (4F). In the case of rs9876490 the T4A gIR+ frequency observed in our cohort (•) fits the predicted MRD SCT curve well.

Figure 5. T4A-specific cytokine secretion in post-SCT patient samples.

T2 cells were pulsed with T4A peptide and incubated with a rs9876490 heterozygous control and a T4A gIR+ post-SCT sample. IFN-γ and TNF-α secretion were measured by Luminex analysis and normalized to maximum secretion induced by incubation with OKT3. The T4A gIR+ post-SCT sample had a greater TNF-α (P = 0.0083) secretion and a non-statistically greater IFN-γ secretion (student t-test, P = 0.3296) when incubated with pulsed T2 cells vs. non-pulsed T2 cells (5A). This difference was not observed in the healthy control sample (5A). Two HLA-A0201 expressing EBV-LCL lines were generated and genotyped using Sanger sequencing. One EBV-LCL line was heterozygous for rs9876490 and therefore capable of generating T4A (T4A+), and the other was homozygous for rs9876490_A and therefore incapable of generating T4A (T4A−) (5B). Both EBV-LCL lines produce full length TRIM42 (5C). Two T4A gIR+ patients were identified and post-SCT PBMC samples were split and then incubated with both EBV-LCLs. In both patients an increased T4A/tetramer+, IFN-γ+ population was identified in the sample that was incubated with the T4A+ EBV-LCLs and not the T4A- EBV-LCLs (5D).

Figure 6. T4A gIR+ and clinical outcomes.

A Kaplan-Meier survival analysis for the effect of T4A gIR on OS and DFS was performed on the 57 examined patients. A trend towards improved OS (median OS: NR vs. 26 mos, P = 0.3132) and DFS (median DFS: NR vs. 8 mos, P = 0.1223) was observed in the T4A gIR+ patients compared to T4A gIR- patients (6A, 6B). Also, a trend towards improved relapse rates was observed in the T4A gIR+ patients (11% vs. 46%, P = 0.0911).