Identification of a β-Cell-Specific HLA Class I Restricted Epitope in Type 1 Diabetes

To predict nonameric β-cell peptides that would bind to the common MHC allele, HLA-A*0201, we used computer-based programs available at the websites for BioInformatics & Molecular Analysis Section (BIMAS) HLA Peptide Binding Predictions (http://www.bimas.dcrt.nih.gov) and SYFPEITHI, Institute of Cell Biology, University of Tuebingen (http://syfpeithi.bmi-heidelberg.com). All peptides were synthesized by the Nucleic Acid Peptide Synthesis (NAPS) Laboratory (University of British Columbia, Canada), and peptide purity was assessed by analytical high-performance liquid chromatography and mass spectrometry.

T2 cells (5 × 10⁵ cells/well) lacking stable HLA-A*0201 surface expression (unless bound to peptides) were incubated for 16 h at 26°C with synthetic human IAPP precursor protein (preproIAPP) 5-13 (100 μg/ml) or equimolar amounts of a peptide known to bind to HLA-A*0201 (Epstein-Barr virus, EBV BMLF1 lytic cycle antigen, GLCTLVAML) or a control peptide that is known to bind to HLA-B*0801 but not HLA-A*0201 (EBV BZLF1 antigen, RAKFKQLL). The cells were then stained with fluorescein isothiocyanate-labeled anti-HLA-A*0201 (BD Pharmingen) and analyzed on a FACScalibur flow cytometer (BD Biosciences, San Jose, CA) to determine stabilization of HLA expression.

Peripheral blood was obtained from type 1 diabetic HLA-A*0201 patients and nondiabetic HLA-A*0201 healthy control subjects. Study participants were typed at the HLA-A locus by allele-specific DNA amplification (PEL-FREEZ Clinical Systems, Milwaukee, WI). Parents of all participants provided informed written consent, and patients provided written assent. The study protocol was approved by the Clinical Research Ethics Board of the University of British Columbia.

Peripheral blood mononuclear cells (PBMCs) were isolated by Ficoll density centrifugation and screened for peptide recognition using interferon-γ enzyme-linked immunospot (ELISpot) assays (Mabtech, Nacka, Sweden), as previously described (18). In brief, PBMCs (2 × 10⁴ cells/well in duplicate) were incubated overnight with 5 μg/ml preproIAPP 5-13, media (negative control), phytohemagglutinin (PHA, positive control), or nucleocapsid epitope of hepatitis C virus (HCV, negative control, DLMGYIPLV) before transfer to 96-well polyvinylidene fluoride plates (MAIP N 45; Millipore, Bedford, MA). ELISpots were developed (18), scanned at high resolution, and magnified to allow unequivocal identification of spots. The frequency of preproIAPP 5-13 reactive cytotoxic T-cells was calculated by subtracting the number of spots present in wells containing the control HCV epitope from the number of spots in the preproIAPP 5-13 well. Statistical comparison between the three groups was performed using the Mann-Whitney test.

IAPP is a β-cell peptide that is cosecreted with insulin from β-cell secretory granules (19). In human type 1 diabetes, IAPP has attracted little attention as a possible autoantigen because studies have demonstrated no difference in the prevalence of autoantibodies against synthetic IAPP 1-37 in type 1 diabetic patients compared with type 2 diabetic control subjects (20). Interestingly, IAPP has been previously suggested as a candidate autoantigen in the NOD mouse because a CD4+ T-cell clone derived from these animals was shown to recognize a unique autoantigen that mapped to the telomeric region of mouse chromosome 6, where the IAPP gene resides (21).

We therefore revisited the possibility that IAPP, or its precursor molecule, preproIAPP, may be an autoantigen in type 1 diabetes. To this end, the amino acid sequence of human preproIAPP (Fig. 1) was analyzed using the SYFPEITHI and BIMAS algorithms to yield nonameric peptides that might bind to HLA-A*0201 (Table 1). In our experience and those of others (13,14), scores ≥60 (BIMAS) or ≥23 (SYFPEITHI) have the highest potential to bind to the class I heavy chain. To give consideration to the score from each ranking system with similar weighting, we used the product of these two scoring systems and then ranked the peptides according to their weighted score. By this analysis, a peptide within the leader sequence of IAPP (human preproIAPP 5-13: KLQVFLIVL) was found to receive a markedly higher score than any other peptide within preproIAPP (Table 1) and was chosen for further study.

To verify first that preproIAPP 5-13 would bind to HLA-A*0201 molecules, an MHC stabilization assay using T2 cells was performed. PreproIAPP 5-13, but not the negative control peptide, was able to stabilize HLA-A*0201 equally as well as the positive control peptide (Fig. 2), indicating that preproIAPP 5-13 binds to and stabilizes HLA-A*0201 on the cell surface.

After obtaining informed consent, we collected peripheral blood from 45 patients with type 1 diabetes, of whom 18 patients were shown to be HLA-A*0201 positive for at least one allele. Of the 18 HLA-A*0201 patients with type 1 diabetes forming the study population, 9 had recent-onset diabetes, which is defined as <180 days duration (duration of diabetes for each patient: 1, 1, 1, 2, 3, 26, 32, 45, and 120 days). The ages of these recent-onset patients were 4.8, 5.1, 6.9, 7.9, 8.1, 13.2, 13.9, 15.2, and 16.6 years. Nine patients had long-standing diabetes, which is defined as >180 days (duration of diabetes for each patient: 183, 185, 279, 354, 1,233, 1,586, 2,353, 1,650, and 3,273 days). The ages of these long-duration patients were 4.1, 6.5, 11.9, 12.2, 12.6, 13.0, 13.1, 13.4, and 17.3 years. Their insulin requirements ranged from 0.9 to 2.0 units · kg⁻¹ · day⁻¹, and HbA_1c values ranged from 7.2 to 9.1%, suggesting that these long-duration patients had minimal endogenous β-cell function. Nine healthy, nondiabetic, non-age-matched HLA-A*0201 control subjects were also enrolled (ages: 18, 24, 25, 25, 26, 30, 38, 40, and 43 years).

Representative ELISpot assays performed on PBMCs from a patient with type 1 diabetes of recent onset, one of long duration, and one control subject without diabetes are presented in Fig. 3A. The recent-onset type 1 diabetic patient had a marked number of PBMCs responding to preproIAPP 5-13 in contrast to those from the long-standing diabetic patient and the nondiabetic control subject. Neither the diabetic patients nor the control subject had PBMCs that responded to either the negative control peptide (HCV epitope) or medium alone. A summary of all the ELISpot data are presented in Fig. 3B. Of the nine patients with recent-onset type 1 diabetes (<180 days), six had ELISpot responses to preproIAPP 5-13 that were at least 3 SDs above the mean of the nondiabetic control subjects (P = 0.002), although two of these responders were only slightly above this threshold. In contrast, no patients with diabetes for >180 days had a response above this threshold.

In autoimmune diabetes affecting both humans and NOD mice, T-cells and, in particular, CD8+ cytotoxic T-cells are mediators of β-cell destruction (1). Although epitopes recognized by CD4+ T-cells and autoantibodies have been described in type 1 diabetes (22), the nature of the HLA class I epitopes remains largely unknown; one other HLA-A*0201-restricted epitope, derived from glutamic acid decarboxylase, has been previously described (2). In this study, we show that a significant proportion of cytotoxic T-cells from HLA-A*0201 patients with recent-onset type 1 diabetes recognize a peptide derived from the leader sequence of preproIAPP. These cells were not detected in patients who had a longer duration of diabetes, which is consistent with our finding in NOD mice that in the absence of continued antigenic stimulus due to loss of β-cells, the frequency of responding cytotoxic T-cells is low (3).

Like most secretory proteins, the IAPP precursor protein contains a hydrophobic leader sequence to direct its synthesis into the lumen of the endoplasmic reticulum. Once inside the endoplasmic reticulum, leader sequences are trimmed and normally degraded. However, because of their presence in the endoplasmic reticulum where HLA class I molecules are assembled, such sequences may make ideal self-peptides for binding to HLA heavy chains that prefer hydrophobic anchor residues. This mechanism of leader-derived peptide loading has been previously described for peptides binding to HLA-E and QA-1 in humans and mice, respectively (23,24).

In NOD mice, prediction of disease development by quantification of β-cell-specific cytotoxic T-cells (3) and prevention of clinical disease by peptide immunization (1) have been possible because of the identification of immunodominant MHC class I β-cell epitopes. Our identification of an HLA class I β-cell epitope in type 1 diabetes may now enable investigation of these diagnostic and therapeutic approaches in patients. Quantification of autoreactive cytotoxic T-cells may complement current approaches to predicting type 1 diabetes that include detection of both autoantibodies to β-cell proteins (22) and autoreactive CD4+ T-cells (25). In addition, altered or endogenous versions of preproIAPP 5-13 may have therapeutic value, as shown previously for an HLA class II epitope (6). Given the HLA class I heterogeneity of patients with type 1 diabetes, preproIAPP 5-13 is likely to be one of many epitopes recognized by autoreactive cytotoxic T-cells. The development of A*0201 tetramers to recognize preproIAPP 5-13-specific cytotoxic T-cells and the generation of peptide-specific cytotoxic T-cell lines will facilitate future studies aimed at delineating the importance of this epitope in type 1 diabetes pathogenesis. The success of this method suggests that further HLA class I β-cell epitopes may be identified using a similar approach.

This study was funded by a Proof-of-Principle grant (to R.T. and C.B.V.) and a New Investigator Award (to C.B.V.) by the Canadian Institutes of Health Research. C.P. received salary support from the Canadian Pediatric Endocrine Group sponsored by Eli Lilly.

We are grateful to I-Fang Lee for assistance with HLA typing. We are indebted to Dr. Jan Dutz for his thoughtful review of this study and Jacqueline Trudeau for her input into study methodology. We thank Drs. Jean-Pierre Chanoine, Dan Metzger, and Laura Stewart for assistance with patient recruitment.