Low HLA binding of diabetes-associated CD8+ T-cell epitopes is increased by post translational modifications

To identify diabetes-associated CTL epitopes from the published literature, we employed a bioinformatics tool recently developed in the Sette laboratory [7]. This tool automatically extracts relevant data from the Immune Epitope Database (IEDB; www.​iedb.​org) and generates reference sets of validated epitopes from various disease indications. Here, the tool was applied to epitopes in the IEDB that are derived from a set of human proteins associated with diabetes [8], further modified to take advantage of the IEDB’s search interface that allows for identification of epitope records directly associated with studies related to a specific disease and/or autoimmune context.

Accordingly, using the IEDB disease finder, the query was configured to include epitopes derived from antigens associated with T1D, pre-diabetes, and diabetes mellitus studies (Additional file 1). The query was also structured for epitopes reported to elicit positive T-cell responses in human hosts as determined using multimer/tetramer staining assays or readouts based on intracellular cytokine staining (ICS), enzyme-linked immunospot (ELISPOT), or ⁵¹Cr-release assays. Responses induced following either in vitro or ex vivo stimulation were allowed. Finally, only epitopes between 8 and 11 residues in length were considered, agreeing with the most canonical peptide sizes bound by class I molecules. This generated a set of 138 epitopes (Additional file 2). Notably, 114 of the 138 epitopes (83%) were restricted by HLA-A*02:01 or A2 serological specificities. As A*02 alleles are the most common in almost all major ethnic/geographic populations worldwide [9, 10] and, as a result, have been the most extensively studied, it is unlikely that this bias is related to diabetes incidence.

The 138 peptides were tested for their capacity to bind their respective HLA class I restricting allele(s) in classical competition assays based on the inhibition of the binding of high affinity radiolabeled ligands to purified major histocompatibility (MHC) molecules, as described in the Methods. In instances where the precise restriction was not available, binding to the most common representative subtype from the same allele family was assayed (e.g., A*02:01 for A02; B*07:02 for B7; A*03:01 for A03, etc.). The affinity of each epitope to its reported HLA restricting allele(s) is shown in Table 1. For epitopes reported to be restricted by multiple alleles, each restriction is shown separately. The different alleles assayed, a tally of the number of corresponding peptides tested, and the number that were considered high or intermediate binders, are provided in Table 2.

In total, 56 (48.7%) of the 115 A*02:01-restricted epitopes bound with high affinity (IC50 < 500 nM); another 34 (29.6%) bound with intermediate affinity (IC50 in the 500–5000 nM range). Of the 38 non-A*02 restrictions, only 3 (7.9%) were associated with an affinity of 500 nM or better and another 10 (26.3%) bound with only intermediate affinity. Thus, overall, 103 of 153 (67.3%) of the HLA/epitope combinations were associated with binding at the 5000 nM, or better, level; that is, 59 (38.6%) were associated with high affinity, and another 44 (28.8%) with intermediate affinity.

In terms of thresholds, 75% of the epitopes bound with an IC50 of 8250 nM or better, and 90% of the epitopes bound with an IC50 of 48,000 nM (Fig. 1). These results established a reference threshold for binding affinities of HLA class I-restricted diabetes epitopes, and confirmed that the majority of diabetes epitopes bound with detectable affinity to HLA, consistent with HLA binding being a necessary (albeit not sufficient) requisite for HLA class I-specific immunogenicity.

The binding threshold identified above is notable in that it is somewhat higher (i.e., associated with lower affinity) than the binding threshold (500 nM) previously found to be associated with the vast majority (> 80%) of HLA class I-restricted T-cell epitopes derived from various pathogens [11‐13]. To more directly compare diabetes-associated epitopes to those of pathogenic origin, we next utilized a bioinformatic approach used previously [13] and generated predicted binding affinities for each diabetes-associated epitope to its restricting allele(s) utilizing the SMM algorithm hosted by the IEDB analysis resource. Predictions were also generated for a large panel of over 2200 virus-derived epitopes identified in the IEDB. The cumulative predicted affinity of the two epitope sets were then compared (Fig. 2). As shown, the virus-derived epitopes were predicted to bind at a higher rate than the diabetes-associated epitopes, where a predicted binding threshold of 500 nM captured about 70% of the virus epitopes, but only 60% of the diabetes epitopes. Similarly, 75% of the virus epitopes were expected to bind at 750 nM or better, in comparison to 1260 nM for the diabetes epitopes. The results for the viral epitopes were consistent with previous reports where a binding threshold of 500 nM was associated with approximately 85% of epitopes [11, 12]. These findings indicate that the overall binding affinity is lower for diabetes-associated T-cell epitopes when compared to virus-derived peptides.

These observations were not entirely surprising, as it has been hypothesized that self-epitopes might bind with reduced affinity, and that this is likely characteristic of autoimmunity in general (see, e.g., [2, 14]). However, when we evaluated the predicted binding affinity of 53 non-diabetes autoimmune epitopes retrieved from the IEDB using identical methodology, we found that these epitopes were not only predicted to bind better than diabetes-associated peptides, but even better than the viral epitopes. In fact, a 500 nM threshold identifies 80% of the selected non-diabetes autoimmune-associated epitopes, with 75% of them predicted to bind at the 355 nM level or better (Fig. 2).

The measured and predicted lower affinity of diabetes epitopes could be explained if the epitopes recognized by diabetes-associated T-cells are often PTM products, and the modification is associated with increased HLA binding. To test this hypothesis, which was also previously suggested by McGinty and James [15], we analyzed the binding affinity of diabetes-associated epitopes, specifically from the insulin-B 30-mer, and control non-epitopes, modified by common PTMs, including citrullination, oxidation, deamidation, and chlorination. In each case, as described following, sets of all possible overlapping 9- or 10-mer sequences incorporating the various modifications were synthesized.

In terms of oxidation, Strollo et al. [16] using sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE), three-dimensional fluorescence, and mass spectrometry (MS) demonstrated that histidine (His5), cysteine (Cys7), and L-phenylalanine (Phe24) residues could be oxidized in insulin-B (corresponding to positions 29, 31, and 49, respectively, in the full unspliced insulin sequence). To address the impact of oxidation of these residues on pHLA binding, 66 peptides were synthesized (see Additional file 3), comprising 14 peptides with oxidized Cys7; 14 with oxidized Phe24; and 38 with all residues oxidized (i.e., pan oxidation; see Methods).

Strollo et al. also reported that the tyrosines (Tyr) in positions 16 and 26 in the insulin B-chain (positions 40 and 50 in the unspliced sequence) were chlorinated. Therefore, another 29 peptides were synthesized to evaluate chlorination of both tyrosine residues.

Further, deamination of asparagine (Asn) and glutamine (Gln) residues have been described [4]. Deamination of Asn or Gln leads to and aspartic acid (Asp) or glutamic acid (Glu), respectively, or isoaspartic and isoglutamic acid, respectively. To evaluate the impact of Asn or Gln deamination of insulin B-chain eptiopes, we synthesized another 16 peptides.

In addition to the various published PTMs of the B-chain, we also investigated the effect of citrullination throughout the entire unspliced insulin sequence, which required the synthesis of 70 peptides. Collectively, a total of 276 peptides were synthesized, comprising 181 modified peptides and the corresponding 95 unmodified 9- and 10-mers. All peptides were evaluated for binding to HLA-A*02:01, as described in the Methods. The HLA-A*02:01 binding of each modified peptide is listed, along with its cognate unmodified (i.e., wild type, WT) peptide, in Additional file 3. The location of the various residues subjected to modification in the full-length insulin sequence is shown in Additional file 4.

We next evaluated the impact of these PTMs on the HLA-A*02:01 binding capacity of insulin-derived peptides, including nine that were previously reported as A*02:01 restricted T cell epitopes (see Additional file 3). Of the nine previously reported epitopes, four (44%) were associated with a PTM-dependent improvement of affinity of at least two-fold (average = 3.76 +/− 2.73-fold) compared to the WT peptide (Table 3, top 4 epitopes). These improved peptides include the leader sequence 2–10 9-mer and 2–11 10-mer epitopes; here, modification of the arginine (Arg) in position 6 by citrullination improved binding of the 9-mer and 10-mer 7.83- and 2.76-fold, respectively. Two peptides (one 9- and one 10-mer) from the 33–42 region of the beta chain with chlorination of Tyr40 bound 2- to 2.5-fold better than the corresponding WT sequences. Overall, because in several cases multiple modifications of each epitope were tested, 26.7% (4/15) of the total modifications to known insulin-derived A*02:01-restricted T-cell epitopes resulted in a peptide with increased binding.

In addition to these previously published PTM epitopes, another 13 peptides (Table 3, bottom 13, below line) were found to be associated with improved binding as a result of PTM, compared to the unmodified counterpart. (For the 9-mer B-chain sequence, YLVCGERGF, 2 different modifications led to increased binding). However, only one of the 14 modified peptides — a citrullinated leader sequence peptide —bound with an affinity < 100 nM. By comparison, three of the four modified known HLA-A*02:01 restricted epitopes bound with affinities < 100 nM. The difference in the rate of improvement seen between these two peptide sets is significant (p = 0.0147).

The effects of specific modifications on the binding of A*02 epitopes, non-epitopes, and in total, are summarized in Table 4. With respect to the epitopes, 2/3 (67%) sequences subjected to chlorination, and 2/5 (40%) modified by citrullination, resulted in improved binding. Oxidation did not result in improved binding in any of the seven cases examined. Taken together, however, it should be noted that the number of events probed here are not sufficient to make statements regarding statistical significance of differences between the different modifications. Further, because Asn or Gln residues were not present in the respective epitopes, none of the corresponding peptide sequences were subject to deamidation.

Peptide-MHC affinities were measured using classical competition assays based on the inhibition of binding of a high affinity radiolabeled ligand to purified HLA class I molecules, as previously described [18]. Briefly, 0.1–1 nM of radiolabeled peptide is co-incubated at room temperature with purified MHC in the presence of a cocktail of protease inhibitors. Following a two-day incubation, MHC bound radioactivity is determined by capturing MHC/peptide complexes on W6/32 (anti-HLA class I) mAb coated Lumitrac 600 plates (Greiner Bio-one, Frickenhausen, Germany), and bound cpm measured using the TopCount (Packard Instrument Co., Meriden, CT) microscintillation counter. The concentration of peptide yielding 50% inhibition of binding of the radiolabeled peptide is calculated. Under the conditions utilized, where [label] < [MHC] and IC50 ≥ [MHC], measured IC50 values are reasonable approximations of true K_d. Each competitor peptide is tested at six different concentrations covering a 100,000-fold range, and in three or more independent experiments. As a positive control, the unlabeled version of the radiolabeled probe is also tested in each experiment. Utilizing a previously defined threshold [11, 12], peptides with an affinity of 500 nM or better for their restricting allele were defined as high affinity binders; peptides with affinities in the 500–5000 nM range were defined as intermediate binders.

Binding predictions were performed using the command-line version of the SMM prediction tool available on the Immune Epitope Database website (http://​www.​iedb.​org) [19, 20]. Besides strong performance in predicting A*02:01 binding capacity, SMM also consistently performs as one of the best prediction tools across a wide array of alleles, and also provides predicted IC50 nM values for all of the alleles considered here. In addition to predicted affinity (IC₅₀), the SMM algorithm provides a percentile score expressing the relative capacity of each peptide to bind each specific allele, compared to a universe of potential sequences of the same size.

Peptides were synthesized by A and A (San Diego) as crude material on a 1 mg scale, or purified (> 95%) by reverse phase HPLC. PTM of individual residues, including deamidation, citrullination, chlorination and oxidation of cysteine was performed as part of the standard Fmoc synthesis. Because of the unavailability of Fmoc-Oxo-His, oxidation of histidine was performed post-synthesis. As a result, histidine cannot be individually oxidized, and in all corresponding peptides all other oxidizable residues (e.g., Cys and Met) are similarly oxidized. Throughout, in the corresponding peptide sequences, the respective PTMs are identified as follows. B: Cysteic acid; J: chloryl tyrosine; O: iso-aspartic acid; Z: iso-glutamic acid; U: citrulline. The human insulin sequence utilized was UniProt accession number P01308.