Identification of Novel HLA-A*0201–Restricted Epitopes in Recent-Onset Type 1 Diabetic Subjects and Antibody-Positive Relatives

All subjects consented for study participation according to institutional review board–approved protocols. PBMCs from type 1 diabetic subjects (n = 13, male-to-female ratio 1:1.2, age [means ± SD] 21 ± 7.7 years) were analyzed within 2.5 years of diagnosis. Antibody-positive subjects (n = 6, 1:1, 19.5 ± 8.8 years) were defined as first-degree relatives, except for subject A901, who was a second-degree relative, with at least one of the three diabetes-associated autoantibodies: GAD65 antibody (Ab), insulinoma-associated antigen 2 (IA-2) Ab, or islet cell Ab. Healthy control subjects (n = 8, 1:1.7, 37 ± 13.0 years) had no personal history of type 1 diabetes or other autoimmune disease and had no first- or second-degree relatives with diabetes. In addition, these subjects were negative for GAD65 Ab and IA-2 Ab (Table 1). Peripheral blood was collected by venous puncture into sodium heparin–containing tubes and stored overnight. PBMCs were isolated by Ficoll-hypaque density gradient centrifugation. PBMCs were frozen in a solution of 90% bovine calf serum and 10% DMSO and kept in liquefied nitrogen until use. Patient samples were haplotyped at the HLA-A allele by PCR–sequence-specific primer, using commercially available primers (One Lambda, Canoga Park, CA), by the Puget Sound Blood Center Immunogenetics laboratory or the British Columbia Child and Family Research Institute. All samples were provided for study in a blinded fashion.

Matrix-assisted algorithms from the National Institutes of Health, Center for Information Technology, BioInformatics and Molecular Analysis Section (BIMAS; available at http://bimas.dcrt.nih.gov/molbio/hla_bind/) and SYFPEITHI (available at http://www.syfpeithi.de/) were used to scan the complete amino acid sequences of ppIAPP, IGRP, and GFAP for nonameric peptides predicted to bind to HLA-A2 with a high affinity (18,19). Peptides predicted by each algorithm to bind HLA-A2 with the highest avidity (ranked in the top 2% of putative binders) were then evaluated for the probability of bearing a COOH-terminal proteasomal cleavage site, using the NetChop 2.0 program (available at http://www.cbs.dtu.dk/services/NetChop/) (20). Those peptides with a strongly predicted COOH-terminus and one or fewer internal cleavage sites were synthesized by the Benaroya Research Institute Peptide Synthesis Core Laboratory, using standard fluorenylmethoxycarbonyl chemistry. Three GFAP peptides—GFAP(386-394), GFAP(57-65), and GFAP(147-155)—that were each predicted to contain a COOH-terminal cleavage site, but not bind to HLA-A2 with a strong avidity, by SYFPEITHI or BIMAS algorithms were also synthesized as negative controls. All peptides were tested for purity by electrospray ionization mass spectrometry before use.

We incubated 174 × CEM.T2 cells (1 × 10⁶) in a 100 μmol/l solution of each peptide in 10% pooled human serum for 4 h at 37°C. The cells were then washed with 1% BSA in PBS and stained with 1 μg fluorescein isothiocyanate–conjugated anti–HLA-A2 (clone BB7.2; BD Biosciences Pharmingen, San Diego, CA) for 20 min on ice. After an additional wash, surface levels of HLA-A2 on cells were analyzed using a FACScalibur flow cytometer (BD Biosciences Pharmingen). As negative controls, cells were incubated with either medium alone or with an HLA-A*2401–restricted Epstein-Barr virus Rta peptide (DYCNVLNKEF), which served as the background. High-affinity HLA-A2–restricted influenza MP(58-66) peptide (GILGFVFTL) was used as a positive control to induce maximum HLA-A2 expression levels. Stabilization values induced by the putative epitopes are reported as the percentage of maximum HLA-A2 expression with the background subtracted.

Antigen-reactive T-cells from frozen PBMC samples were quantified using a granzyme B ELISpot assay (BD Biosciences Pharmingen). Frozen PBMC samples were rapidly thawed in a 37°C water bath and pelleted in a solution of 60% FCS and 1 unit/ml DNase (Sigma, St. Louis, MO). Cells were washed twice in 5% FCS in PBS and passed through a cell strainer to remove aggregates (Becton Dickinson Falcon, Franklin Lakes, NJ). ELISpot plates were coated with anti–granzyme B capture antibody (5 μg/ml) overnight at 4°C and then washed once and blocked with 10% pooled human serum in RPMI 1640 for 2 h. Peptides were diluted to a concentration of 5 μg/ml in 10% pooled human serum and added to wells of coated ELISpot plates. Influenza MP(58-66) and human cytomegalovirus pp65 (NLVPMVATV) epitopes were used as positive control peptides (21), and phytohemagglutinin (2.5 μg/ml) was used as an additional positive control treatment. We also included the ppIAPP(5-13) epitope (KLQVFLIVL) as a control because this peptide has been demonstrated to activate CTLs of recent-onset diabetic subjects (14). HLA-A2–restricted PD5 peptide (KLSEGDLLA) derived from pyruvate dehydrogenase was used as a negative control (22). PBMCs were added to wells at 1.0 × 10⁶ per ml, and plates were incubated for 24 h at 37°C in 5.5% CO₂. Plates were washed twice with deionized distilled water and three times with 0.05% Tween 20 in PBS. Wells were coated with biotinylated anti–granzyme B detection antibody (2.0 μg/ml) for 2 h at 25°C. After three washes with 0.05% Tween 20, a 1:100 dilution of streptavidin-conjugated horseradish peroxidase was added to wells for 1 h at 25°C. Wells were then washed three times with 0.05% Tween 20 and then three times with PBS before developing with 3-amino-9-ethylcarbazole substrate (AEC; Vector Laboratories, Burlingame, CA) according to the manufacturer’s protocols. Plates were allowed to dry overnight at room temperature in the dark before quantifying spots per well, using an ImmunoSpot analyzer (Cellular Technology Ltd., Cleveland, OH), and all spot counts per well were also verified manually.

The number of spot-forming cells in wells containing the negative control PD5 epitope was subtracted from the spot-forming cell values of all other wells. Only mean spot-forming cell values with a coefficient of variation <0.4 were reported in order to control for random variation among wells. The cutoff of positivity for each epitope was defined as the mean number of spot-forming cells elicited by incubation of control subject PBMCs with the epitope + 3 SD of the mean. Fisher’s exact test was used to compare the frequencies of epitope responsive samples in each group.

The matrix-assisted algorithms BIMAS and SYFPEITHI were used to compile an initial list of peptides that ranked in the top 2% of the strongest predicted HLA-A2–binding epitopes for GFAP, ppIAPP, and IGRP. Because the COOH-terminus of most class I–restricted epitopes is generated by the proteasome (23), we also analyzed the sequences of each protein using NetChop 2.0, a proteasomal cleavage predictive database that was trained using 1,100 published class I–restricted peptides (20). Those peptides that were predicted to contain a COOH-terminal cleavage site and one or fewer internal cleavage sites were included in the final panel of putative epitopes (Table 2). To test the accuracy of the HLA binding algorithms and ensure that we were not biasing the putative epitope panel based solely on these predictions, we included three peptides in the panel that were not predicted to be strong HLA-A2 binders by either algorithm but were predicted to each contain a COOH-terminal cleavage site: GFAP(386-394), GFAP(57-65), and GFAP(147-155).

Peptides were synthesized, and the ability of each to stabilize HLA-A2 expression on the TAP (transporter associated with antigen processing)-1–and -2–deficient 174 × CEM.T2 cell line was analyzed. These cells express HLA-A2 at appreciable levels only in the presence of exogenously added high-affinity peptide; thus, HLA stabilization is correlated with peptide affinity (24). We normalized the stabilization values in reference to the maximum HLA-A2 expression levels induced by incubation with a high-affinity peptide of influenza, MP(58-66). Figure 1 shows the peptide stabilization values in rank order. Of the nine peptides predicted to bind HLA-A2, seven were able to stabilize HLA-A2 expression at levels ≥50% of that induced by MP(58-66). The three peptides that were predicted to be weak binders by both algorithms—GFAP(386-394), GFAP(57-65), and GFAP(147-155)—stabilized HLA-A2 expression at the lowest levels. IGRP(211-219) was predicted to be a very strong binder by both algorithms, yet it stabilized HLA expression at the lowest level of all predicted binders. Only those peptides that stabilized HLA expression at >50% of the maximal level were used in subsequent experiments.

Each putative epitope was tested for its ability to induce granzyme B secretion by PBMCs drawn from type 1 diabetic patients, antibody-positive relatives, or healthy subjects. Granzyme B is a direct mediator of cytotoxicity, and, unlike γ-interferon, it is more likely to be secreted by CTLs than bystander CD4⁺ T-helper cells. Type 1 diabetic patients had been diagnosed within 2.5 years preceding the blood draw, and antibody-positive subjects were defined as first-degree relatives of type 1 diabetic patients who were positive for the presence of circulating autoantibodies with specificity for GAD65, IA-2, or islet cell Ab. An ELISpot assay was used to quantify granzyme B secretion by PBMCs in response to individual epitopes, and representative results from sample T466 are shown in Fig. 2. The threshold for determining a positive response was defined as 3 SDs above the control population’s mean spot-forming cell value for each epitope. Of the 13 diabetic subjects, 5 recognized at least one novel epitope using this criterion. Three type 1 diabetic subjects (T466, T537, and T413) responded to ppIAPP(9-17), which is an overlapping peptide of the previously described ppIAPP(5-13) epitope (14) (Fig. 3C). However, ppIAPP(5-13) failed to elicit significant numbers of granzyme B–secreting cells from diabetic patients. Other epitopes recognized by multiple diabetic patient samples included GFAP(143-151) and GFAP(214-222). Two type 1 diabetic subjects (T466 and T985) responded to multiple putative epitopes (Table 3). We observed no correlation between the magnitude of response to the positive control influenza and cytomegalovirus epitopes and responses to putative epitopes (data not shown). We also tested ppIAPP(12-20) and IGRP(211-219), peptides that failed to stabilize HLA expression at sufficient levels, for the ability to induce granzyme B secretion by a subset of type 1 diabetic and control subjects. No responses to these peptides were observed by either type 1 diabetic or control subjects (data not shown).

Analyses of PBMCs from antibody-positive relatives demonstrated that four of the six samples responded to ppIAPP(9-17) (Fig. 3C). The frequency of ppIAPP(9-17)-responsive antibody-positive subjects was significantly higher than in the control group (P < 0.05 by Fisher’s exact test). Interestingly, ppIAPP(9-17)-responsive diabetic patients and antibody-positive subjects demonstrated comparable numbers of granzyme B–secreting cells. Other epitopes that were recognized by both subject groups include GFAP(143-151), GFAP(214-222), and IGRP(152-160). Antibody-positive subject A901 responded to multiple epitopes, three of which were not recognized by diabetic patients (Table 3). Moreover, this subject also demonstrated substantially higher numbers of granzyme B–secreting cells to nearly all epitopes and developed diabetes 3 years subsequent to blood collection. There was no correlation observed between antibody status and epitope recognition by either diabetic or antibody-positive subjects. In addition, no relationship between age and epitope responsiveness was observed.

The identification of CTL epitopes in type 1 diabetes is essential for an understanding of disease pathogenesis and could provide additional markers for early detection and therapeutic intervention. The integration of epitope-predictive matrix-assisted algorithms with highly sensitive screening assays has been reported to be an accurate technique for the identification of novel CTL epitopes (25,26). In this study, we used SYFPEITHI and BIMAS algorithms to predict HLA-A2–binding sequences within GFAP, ppIAPP, and IGRP. Strongly predicted binders were analyzed for the presence of COOH-terminal proteasomal cleavage sites, using NetChop 2.0. Because the use of predictive algorithms to screen proteins is not sufficient, by itself, to identify epitopes (27), we tested predicted epitopes for the ability to stabilize HLA-A2 expression on the surface of 174 × CEM.T2 cells. We failed to observe a direct correlation between the predicted binding scores generated by SYFPEITHI or BIMAS algorithms and the magnitude of HLA-A2 stabilization. This is not surprising because predicted HLA binding scores are calculated from matrixes generated with peptide affinity data using cell-free systems rather than stabilization data, which represent a relative measure of peptide affinity. However, stabilization assays are the standard technique for assessing peptide binding strength to a given HLA molecule, and they are rapid and reproducible. Thus, the integration of in silico predictive data with an in vitro assay of peptide binding narrowed the pool of peptides for study and streamlined the protocol for selecting the most likely candidate epitopes.

The association between the stabilization value of a peptide and its ability to induce granzyme B secretion by CTLs was more apparent. Of the seven peptides that stabilized HLA-A2 expression at ≥50% of maximum levels, four were recognized by PBMCs of at least one antibody-positive and type 1 diabetic subject. Interestingly, the peptide that was recognized by the highest number of type 1 diabetic and antibody-positive subjects, ppIAPP(9-17), stabilized HLA-A2 expression at modest levels, indicating a moderate to low affinity for this molecule. Self-peptides with a low affinity for HLA molecules have been reported as immunodominant epitopes in multiple sclerosis (28) and Goodpasture’s disease (29). It has been hypothesized that the fast off-rate of these peptides impedes negative selection, thereby enabling the survival of autoreactive T-cells (30). It is plausible that the lower affinity of ppIAPP(9-17) for HLA-A2 might allow for the escape of antigen-specific CTLs from the thymus, resulting in later activation and expansion in the periphery. We also analyzed a subset of diabetic and control subjects for responsiveness to the overlapping ppIAPP(12-20) epitope and found that it failed to induce granzyme B secretion (data not shown). It is likely that the affinity of this epitope for HLA-A2 is so low as to preclude presentation in the periphery.

Among the type 1 diabetic patients studied, disease duration was a relative predictor of epitope recognition in that four of the five subjects responding to novel epitopes had been diagnosed within ∼1 year preceding blood collection. It is probable that the preferential responsiveness of these patients is the result of antigen presentation in the early stages of disease, before the complete loss of β-cells. Subjects with diabetes of longer duration likely present lower levels of islet antigens, resulting in the contraction of responsive CTLs. This phenomenon was observed in a report of insulin B peptide–specific T-cells that were detectable in patients experiencing recurrent autoimmunity secondary to islet allografts but that were not detectable in long-standing diabetic patients (31). However, in a report by Toma et al. (15), PBMCs from both short- and long-term type 1 diabetic subjects were activated by several proinsulin epitopes. It is plausible that CTLs with specificity for proinsulin peptides represent a memory population, and this accounts for their continued presence in long-term diabetic subjects. The authors used γ-interferon secretion as an indicator of CTL activation and did not report whether these peptides elicit a cytotoxic response to the epitopes. In our study the use of granzyme B ELISpot assays supports the conclusion that CTLs with specificity for multiple islet peptides may play a role in β-cell destruction in individuals with type 1 diabetes, although different subjects show considerable variation in this specificity. Within the samples evaluated in this study, this variation did not appear to correlate with other HLA genotypes present in the individual subjects, but larger studies are needed to fully assess this possibility.

Previously, ppIAPP(5-13) has been shown to elicit γ-interferon secretion by PBMCs of recently diagnosed type 1 diabetic patients but not patients with disease of a longer duration (14). We failed to observe significant numbers of granzyme B–secreting cells arising from recent-onset diabetic patient cultures with this epitope, and we did not assess the ability of novel epitopes to induce cytokine secretion. We are uncertain whether the lack of significant granzyme B response seen in cultures containing ppIAPP(5-13) was attributable to the fact that CTLs specific for this peptide represent a distinct phenotypic subset that differs from ppIAPP(9-17)-responsive cells (central versus effector memory), or whether responsive CTLs from diabetic patients were undetectable because of the high number of spot-forming cells seen in the control subject cultures (Fig. 3C). Further studies of larger populations will need to be performed to assess the differences in these two responsive populations. It is interesting to note that only one of seven subjects responding to ppIAPP(9-17) also recognized ppIAPP(5-13), demonstrating that these epitopes are not cross-reactive.

Two GFAP and one IGRP epitopes were recognized by at least two subjects. This is the first report demonstrating CTL responses in human diabetic or antibody-positive subjects to peptides of these proteins. In a recent report, NOD mice bearing a transgene encoding a chimeric HLA-A2 molecule were shown to generate T-cells with specificity for murine IGRP peptides (32). The predominant CTL response in these animals was directed to IGRP(228-236), and these CTLs were cross-reactive with the homologous human peptide. It is possible that IGRP(228-236) serves as an additional epitope in type 1 diabetes; however, even in NOD mice, a considerable degree of heterogeneity in responses to this epitope was noted (32). We did not identify IGRP(228-236) as a putative epitope because it was predicted to be a low-affinity binder by both predictive algorithms (SYFPEITHI score: 22; BIMAS binding score: 54.6). Our aim in the current study was to rapidly and efficiently identify epitopes by integrating in silico and in vitro techniques. A more exhaustive analysis of peptides derived from ppIAPP, IGRP, and GFAP will likely elucidate additional epitopes that elicit CTL responses in at-risk and diabetic subjects.

Recently, Ouyang et al. (33) used similar techniques to identify epitopes in type 1 diabetes and demonstrated that ppIAPP(9-17), IGRP(152-160), and IGRP(215-223) induced γ-interferon secretion by PBMCs of recent-onset diabetic patients. The observation of CTL responses to identical ppIAPP and IGRP epitopes in separate patient populations, using two different assays of T-cell activation, likely substantiates the hypothesis that these epitopes play an important role in the development of type 1 diabetes. Future studies will be needed to more fully characterize the CTL responses to these epitopes in terms of disease pathogenesis.

Our data demonstrate that ppIAPP(9-17), GFAP(214-222), GFAP(143-151), and, to a lesser degree, IGRP(152-160) represent novel epitopes that induce granzyme B secretion by CTLs of antibody-positive relatives and recent-onset type 1 diabetic subjects. Based on this, we posit that these epitopes play a role in β-cell destruction. The potential use of T-cell responsiveness to these epitopes as an indicator of disease onset or progression will need to be delineated.

This work was supported in part by grants from the National Institutes of Health and the Juvenile Diabetes Research Foundation Center for Translational Research at Benaroya Research Institute and Children’s Hospital and Regional Medical Center.

The authors thank Dr. John Gebe for critical comments, Jason Berger for peptide synthesis, and Susan Masewicz for technical advice.