CD8 T cell autoreactivity to preproinsulin epitopes with very low human leucocyte antigen class I binding affinity

Authors

  • J. R. F. Abreu,

    1. Department of Immunohematology and Blood Transfusion, Leiden University Medical Center, Leiden, the Netherlands
    Search for more papers by this author
  • S. Martina,

    1. Department of Immunohematology and Blood Transfusion, Leiden University Medical Center, Leiden, the Netherlands
    Search for more papers by this author
  • A. A. Verrijn Stuart,

    1. Department of Immunohematology and Blood Transfusion, Leiden University Medical Center, Leiden, the Netherlands
    2. Department of Pediatric Endocrinology and Centre Cellular and Molecular Intervention, UMC Utrecht, Utrecht, the Netherlands
    Search for more papers by this author
  • Y. E. Fillié,

    1. Department of Immunohematology and Blood Transfusion, Leiden University Medical Center, Leiden, the Netherlands
    Search for more papers by this author
  • K. L. M. C. Franken,

    1. Department of Immunohematology and Blood Transfusion, Leiden University Medical Center, Leiden, the Netherlands
    Search for more papers by this author
  • J. W. Drijfhout,

    1. Department of Immunohematology and Blood Transfusion, Leiden University Medical Center, Leiden, the Netherlands
    Search for more papers by this author
  • B. O. Roep

    Corresponding author
    1. Department of Immunohematology and Blood Transfusion, Leiden University Medical Center, Leiden, the Netherlands
    2. JDRF Center for Beta Cell Therapy in Diabetes in Europe, Leiden, the Netherlands
    Search for more papers by this author

B. O. Roep, Leiden University Medical Center, Dept of Immunohematology and Blood Transfusion, E3-Q, PO Box 9600, 2300 RC Leiden, the Netherlands. E-mail: boroep@lumc.nl

Summary

Beta cells presenting islet epitopes are recognized and destroyed by autoreactive CD8 T cells in type 1 diabetes. These islet-specific T cells are believed to react with epitopes binding with high affinity to human leucocyte antigen (HLA) expressed on beta cells. However, this assumption might be flawed in case of islet autoimmunity. We evaluated T cell recognition of the complete array of preproinsulin (PPI) peptides with regard to HLA binding affinity and T cell recognition. In a comprehensive approach, 203 overlapping 9–10mer PPI peptides were tested for HLA-A2 binding and subjected to binding algorithms. Subsequently, a high-throughput assay was employed to detect PPI-specific T cells in patient blood, in which conditional HLA ligands were destabilized by ultraviolet irradiation and HLA molecules refolded with arrays of PPI peptides, followed by quantum-dot labelling and T cell staining. Analysis of patient blood revealed high frequencies of CD8 T cells recognizing very low HLA binding peptides. Of 28 peptides binding to HLA-A2, a majority was predicted not to bind. Unpredicted peptides bound mainly with low affinities. HLA binding affinity and immunogenicity may not correlate in autoimmunity. Algorithms used to predict high-affinity HLA peptide binders discount the majority of low-affinity HLA binding epitopes. Appreciation that peptides binding HLA with very low affinity can act as targets of autoreactive T cells may help to understand loss of tolerance and disease pathogenesis and possibly point to tissue-specific immune intervention targets.

Introduction

Type 1 diabetes (T1D) is a chronic autoimmune disease in which the insulin-producing beta cells are selectively destroyed by autoreactive T cells. It has recently become evident that, in addition to the pathogenic effects mediated by CD4 T cells, cytotoxic CD8 T cells are key players in the beta cell destruction process [1–7]. The epitopes recognized by autoreactive CD8 T cells are considered to be derived primarily from beta cell proteins [8]. With the appreciation of the pivotal role played by CD8 T cell autoreactivity in the pathogenesis of T1D, efforts to search for their beta cell epitopes have increased. Indeed, the discovery of new islet-specific CD8 T cell epitopes could provide new biomarkers to monitor disease progression and responses to therapeutic interventions, and provide novel potential targets for therapy [9]. In attempts to identify novel CD8 T cell epitopes, epitope discovery in T1D has focused strongly on peptides binding with high-affinity to human leucocyte antigen (HLA). As peptide binding motifs for numerous HLA-I molecules have been described, prediction algorithms are used commonly and successfully as a screening tool to predict HLA-I binding of T cell epitopes [10–14]. The use of prediction algorithms allows the preselection of peptides with high binding affinity to HLA-I, thought to be the most important for the induction of a T cell response [15], while low-affinity HLA-binding peptides are normally disregarded. However, such low-affinity HLA binding peptides might prove relevant in the context of autoimmunity. In mice, weak interactions between proinsulin peptides and major histocompatibility complex (MHC)-II have been proposed to lead to defects in negative selection in the thymus and contribute to autoimmunity by allowing the escape of autoreactive T cells from deletion [16,17]. Serendipitously, one study reported an inverse correlation between the binding affinity of beta cell peptides to MHC-I molecules and the corresponding CD8 T cell responses in diabetic patients [18]. Also, in a multi-step discovery approach, we recently uncovered four novel CD8 T cell epitopes derived from preproinsulin (PPI), selected initially for their predicted high HLA binding affinity, but eventually proven to bind with very low to intermediate binding affinities [19]. CD8 T cells recognizing these epitopes could be detected in the peripheral blood of diabetic patients, in spite of their very low HLA binding affinities. These observations underscore the notion that peptides binding with very low affinities to HLA-I may act as important epitopes for autoreactive CD8 T cells.

In this study, we investigated the relation between peptide binding affinity to HLA-A2 and immunogenicity, electing a comprehensive approach in which overlapping peptides spanning the entire PPI molecule were investigated, avoiding selection bias. We focused on PPI as there is increasing evidence that this islet autoantigen acts as an important target of the immune system in T1D. For instance, the level of insulin expression in the thymus is an important risk factor in T1D. Polymorphisms in the variable number of tandem repeat (VNTR) elements in the human insulin promoter, controlling insulin expression in the thymus, are associated with disease susceptibility [20,21]. Accordingly, low insulin expression in the thymus associated with high frequencies of insulin-specific CD4 T cells in the peripheral blood [22]. PPI-specific CD8 T cells have also been found in the peripheral blood of diabetic patients [4,19,23]. Cloned CD8 T cells specific for PPI15–24 were shown to specifically kill human beta cells in vitro[23]. Importantly, this study demonstrated unorthodox processing of PPI peptide epitopes, independent of proteasomal digestion. This notion implies that the epitope repertoire in the context of autoimmune disease is not necessarily restricted by appropriate C-terminal processing by proteasomes.

To address the relation between peptide binding affinity to HLA and immunogenicity, we first tested the HLA-A2 binding capacity of overlapping PPI peptides, spanning the entire molecule. Because HLA-A2, expressed by the majority of T1D patients, is known to accommodate peptides preferably with 9 or 10 amino acid (aa) length, this led to a series of 203 candidate peptides. Strikingly, when testing patient peripheral blood, we observed significantly increased frequencies of CD8 T cells recognizing very low HLA binding PPI peptides, compared to low and intermediate binding peptides. In parallel, we subjected the full PPI aa sequence to commonly used prediction algorithms. While prediction algorithms were capable of predicting binding of peptides for which we confirmed intermediate binding in vitro, this strategy failed to notice peptides binding HLA-A2 with low and very low affinities. Our results indicate that peptides binding with very low affinity to MHC-I can act as epitopes for autoreactive CD8 T cells. Appreciating such peptides in epitope search approaches might disclose islet-specific epitopes with potential as novel diagnostic or therapeutic targets and could be useful in the development of biomarkers of disease progression.

Material and methods

T1D patients

Heparinized blood samples were retrieved from 10 paediatric patients with recent-onset T1D [median age 8 years (range 2–14); median disease duration 0 months (range 0–10); male : female 6:4]. Diagnosis was made according to the International Society of Pediatric and Adolescent Diabetes criteria [24]. All patients were positive for at least one islet autoantibody in their serum. Written informed consent was obtained from all children and/or their parents and the study was approved by all involved local medical ethics review boards. Peripheral blood mononuclear cells (PBMC) were isolated by Ficoll-isopaque density gradient centrifugation, frozen and kept in liquid nitrogen until use. HLA-A2 (HLA-A*0201) typing was confirmed by flow cytometry using fluorescein isothiocyanate (FITC)-conjugated anti-HLA-A2 antibodies (BD Biosciences, Franklin Lakes, NJ, USA).

Peptides

Overlapping 9 and 10aa long PPI peptides and the cytomegalovirus (CMV) peptide were synthesized using solid-phase Fmoc chemistry. All peptides were analysed by reverse-phase high-performance liquid chromatography (RP–HPLC) (purity >85%) and matrix-assisted laser desorption/ionisation-time of flight mass spectrometry (MALDI-TOF) mass spectrometry to confirm the expected mass.

Prediction algorithms

HLA-A2 binding peptides were predicted using the prediction algorithms NetMHC 3.0 and 3.2 ([10], (http://www.cbs.dtu.dk/services/NetMHC) [10], SYFPEITHI (http://www.syfpeithi.de/home.htm) [12,14], BIMAS (http://www-bimas.cit.nih.gov/molbio/hla_bind/) [13] and MOTIFS (available on request) [11].

In-vitro peptide-binding studies

Peptide binding to HLA-A2 was performed as described previously [25]. Briefly, 10-fold dilution series of the peptides were made and incubated with 15 pmol beta 2-microglobulin, 200 fmol recombinant HLA-A2 heavy chain and 100 fmol fluorescein (FL)-labelled indicator peptide. HLA-bound and free FL-labelled peptides were separated by HPLC size-exclusion chromatography with fluorescence detection; 50% inhibitory concentration (IC50) values were calculated using non-linear regression analysis.

Generation of peptide–HLA (pHLA) monomers and quantum-dot (Qdot) labelling

Generation of pHLA monomers was performed as described previously [26]. Briefly, exchange reactions were performed by exposing ultraviolet (UV)-sensitive pHLA monomers [2 µM in phosphate-buffered saline (PBS)] to UV light (366-nm UV lamp; Camag, Berlin, Germany) in the presence or absence (negative control) of exchange peptide (200 µM) for 60 min. Multimeric pHLA complexes were produced by addition of streptavidin-conjugated Qdot-605 and -705 (intermediate, low and very low affinity PPI peptides) or Qdot-585 and -800 (controls) (Invitrogen, Breda, the Netherlands) to achieve a 1 : 20 streptavidin–Qdot : biotinylated–pHLA ratio.

PBMC staining with Qdot-labelled pHLA multimers

PBMC staining was performed as described previously [7]. Briefly, PBMC (2 × 106) were stained with pooled pHLA multimers (0·1 µg of each specific multimer), according to the peptide-binding affinity to HLA-A2 (intermediate, low and very low, see Table 1) in PBS/0·5% bovine serum albumin (BSA) for 15 min at 37°C. Subsequently, allophycocyanin (APC)-labelled anti-CD8, FITC-labelled anti-CD14, -CD20, -CD4, -CD40 and -CD16 (all from BD Biosciences) were added for 30 min at 4°C. After washing, cells were resuspended in PBS/0·5% BSA containing 7-aminoactinomycin D (7-AAD) (eBioscience, San Diego, CA, USA) to exclude dead cells and analysed using a LSRII (BD Biosciences).

Table 1.  In-vitro binding affinity of preproinsulin (PPI) peptides to human leucocyte antigen (HLA)-A2.
Position*SequenceLengthIC50 (nM)HLA-A2 affinityPredicted binder
  • *

    Position in the PPI sequence. N3.0: NetMHC 3.0; N3.2: NetMHC 3.2; S: SYFPEITHI; B: BIMAS; M: MOTIFS. IC50: 50% inhibitory concentration.

34–42HLVEALYLV9133,1IntermediateN3.0,N3.2,S,M
15–24ALWGPDPAAA10490,8IntermediateN3.0,S
2–10ALWMRLLPL9748,8IntermediateN3.0,N3.2,S,B,M
15–23ALWGPDPAA9879,8IntermediateN3.0,N3.2,S,M
6–14RLLPLLALL9884,3IntermediateN3.0,N3.2,S,B,M
2–11ALWMRLLPLL10940,3IntermediateN3.0,N3.2,S,B,M
6–15RLLPLLALLA101296IntermediateN3.0,N3.2
7–15LLPLLALLA91372IntermediateM
13–22LLALWGPDPA101964IntermediateN3.0,N3.2
1–10MALWMRLLPL103216LowN3.0,N3.2,S
17–26WGPDPAAAFV104456Low 
22–30AAAFVNQHL95017Low 
7–16LLPLLALLAL105840LowN3.0,N3.2,S,B,M
101–109SLYQLENYC95860LowB,M
8–16LPLLALLAL911835LowS
35–44LVEALYLVCG1013331Low 
90–99GIVEQCCTSI1029248LowS,M
10–19LLALLALWGP1029623LowN3.2
1–9MALWMRLLP936309Very low 
29–38HLCGSHLVEA1040364Very lowN3.2,S
9–18PLLALLALWG1043458Very low 
4–12WMRLLPLLA953428Very low 
34–43HLVEALYLVC1064329Very low 
35–43LVEALYLVC971249Very low 
4–13WMRLLPLLAL1073485Very lowN3.2,S,M
76–85SLQPLALEGS1087985Very low 
40–49YLVCGERGFF1095026Very low 
96–105CTSICSLYQL1097732Very lowM

Results

Identification of HLA-A2 binding PPI peptides

To investigate the relation between peptide binding affinity to HLA-I and immunogenicity in T1D, we first determined which PPI peptides could actually bind to HLA-A2 molecules. In an unbiased approach, all 203 overlapping 9- and 10-mer PPI peptides were synthesized and tested for binding to purified HLA-A2 molecules. A cell-free peptide–HLA binding assay was used, in which a mixture of unfolded recombinant HLA heavy chain molecules, beta 2-microglobulin and a fluorescent-labelled indicator peptide were allowed to form peptide–HLA complexes [25]. From this assay, the concentration at which the tested peptide reduced by 50% the amount of HLA-bound fluorescent-labelled peptide (IC50 value) was calculated. From 203 PPI peptides, 28 peptides could bind HLA-A2 (Table 1). Binding affinities of PPI peptides ranged from intermediate to very low HLA-A2 binding capacity (Table 1). Interestingly, peptides capable of binding to HLA-A2 were located mainly in the signal peptide of PPI (Fig. 1, upper panels).

Figure 1.

Predicted and in-vitro preproinsulin (PPI) peptide binding to human leucocyte antigen (HLA)-A2 molecules. Overlapping 9- and 10-mer peptides were analysed by prediction algorithms and compared to results of the in-vitro peptide binding assay. (a) All 9- and (b) 10-mer peptides predicted to bind to HLA-A2 by NetMHC 3·0 (black bars), NetMHC 3·2 (diagonally hatched bars), SYFPEITHI (grey bars), BIMAS (white bars) and MOTIFS (horizontal stripes), and the observed in-vitro binding affinity (1/IC50) were plotted against the amino acid sequence of PPI. Boxes on the PPI amino acid sequence indicate the signal peptide, B chain, cleavage site for conversion from proinsulin to insulin, C-peptide, cleavage site and A chain, respectively.

Relation between peptide–HLA binding affinity and T cell recognition

After determining the binding capacity of all 9- and 10-mer PPI peptides to HLA-A2, the relation between peptide-HLA binding affinity and T cell recognition in T1D was investigated. For this, the presence of CD8 T cells recognizing the 28 PPI peptides was examined in PBMC of recent-onset T1D patients. In order to detect PPI-specific CD8 T cells, pHLA complexes were generated for the complete panel of 28 HLA-A2 binding PPI peptides. Conventional techniques to produce HLA-I tetramers are very labour-intensive and time-consuming, as for each peptide a separate in-vitro HLA refolding reaction is required. We elected to apply an UV-mediated ligand exchange technology that we recently developed and validated, where conditional HLA ligands forming stable complexes with HLA molecules disintegrate upon exposure to UV light [27]. These empty HLA molecules were then stabilized by binding of the individual 28 PPI peptide ligands, which rescued HLA molecules from disintegration. Multimeric pHLA complexes were generated by labelling refolded pHLA monomers with Qdot fluorochromes. To investigate the relation between HLA binding affinity and T cell recognition, the 28 pHLA multimers were ranked based on their PPI peptide IC50 values (Table 1) and divided into three groups with intermediate, low and very low HLA-A2 binding peptides (<3·000 nM, 3·000–30·000 nM and >30·000 nM, respectively). Subsequently, we searched for the presence of CD8 T cells recognizing PPI peptides belonging to one of the three groups in PBMC of recent-onset T1D patients. As all pHLA multimers were labelled with the same Qdot combination, any differences in the frequencies of CD8 T cells detected between groups would be attributable to differences in peptide–HLA binding affinity, rather than differences in fluorochrome intensity or sensitivity. For all three groups, CD8 T cells recognizing PPI peptides with intermediate, low and very low HLA-binding affinities were detectable. Strikingly, we observed significantly higher frequencies of CD8 T cells recognizing PPI peptides with very low HLA-A2 binding affinity compared to peptides binding HLA-A2 with low and intermediate affinities (P = 0·0017 and P = 0·0013, respectively; Fig. 2). For reference, the frequencies of autoreactive CD8 T cells in a non-diabetic, HLA-A2 negative blood donor were 0, 0·003 and 0·008% for intermediate, low and very low binding epitopes, respectively. The three patients with the highest frequencies of CD8 T cells recognizing PPI peptides with very low affinity were also the oldest in our paediatric cohort, with a median age of 13 years. Otherwise, no demographic differences between the 10 patients were noted. As negative controls, pHLA multimers that were UV-exchanged in the absence of rescuing PPI peptide and Qdot-labelled in the same manner as PPI pHLA multimers were used, which showed no background staining. Additionally, patient PBMC were stained with CMV-specific pHLA multimers which showed no T cell staining, except for one patient, which may be explained by the young age of the patients in our cohort. The fact that pHLA multimers, loaded with a non-self HLA-A2 peptide, showed no background staining assured the specificity of our reagents.

Figure 2.

High frequency of preproinsulin (PPI)-specific CD8 T cells recognizing peptides with very low human leucocyte antigen (HLA)-A2 binding affinity. PPI peptide HLA (pHLA) multimers were divided into three groups (intermediate, low and very low HLA-A2 binding affinity). (a) Representative dot plots of CD8 T cells recognizing PPI pHLA multimers of intermediate, low and very low affinity (Qdot-605 and -705) and negative control (Qdot-585 and -800). (b) Peripheral blood mononuclear cells (PBMC) of 10 recent-onset type 1 diabetes patients were screened for the presence of CD8 T cells recognizing the PPI peptides or one cytomegalovirus (CMV) epitope. Statistical analysis was performed using the Mann–Whitney U-test.

Accuracy of algorithms in predicting peptide binding to HLA-A2 molecules

To address how accurate algorithms are in predicting binding of peptides that range from very low to high binding affinities to HLA-A2, the complete aa sequence of PPI was applied to commonly used algorithms to predict peptide binding: NetMHC 3·0, NetMHC 3·2, SYFPEITHI, BIMAS and MOTIFS. We focused on the binding of 9- and 10-mer peptides to HLA-A2 molecules. These tools predicted the binding of 10, 12, 14, five and 15 peptides to HLA-A2 when using NetMHC 3·0, NetMHC 3·2, SYFPEITHI, BIMAS or MOTIFS, respectively (Table 2). Next, to determine the accuracy of computer algorithms in predicting peptide binding to HLA-A2, these results were compared to the IC50 results (Fig. 1). Prediction algorithms were mainly capable of predicting binding of peptides with confirmed intermediate binding affinity to HLA-A2. For peptides binding with low to very low affinity to HLA-A2, prediction algorithms were ineffective and inaccurate in predicting peptide binding (Fig. 1 and Table 1). From a total of 28 HLA-binding peptides, prediction algorithms failed to predict more than half of these (Table 3). Unpredicted peptides were mainly of low and very low affinity. Using less stringent binding thresholds for the algorithms did not identify the majority of the HLA-binding PPI peptides, while the number of false positive predicted peptides increased. Thus, these data show that prediction algorithms should not be used to select low-affinity binders, even though they can be used effectively for prediction of high-affinity HLA-A2-binding peptides.

Table 2.  Binding scores of preproinsulin (PPI) peptides to human leucocyte antigen (HLA)-A2, obtained from each prediction algorithm.
Position*SequenceNetMHC 3·0NetMHC 3·2SYFPEITHIBIMASMOTIFS
  • *

    Position in the PPI sequence. Binding thresholds: NetMHC ≤ 500; SYFPEITHI ≥ 20; BIMAS ≥ 60; MOTIFS ≥ 44.

 9-mer     
2–10ALWMRLLPL16132840854
6–14RLLPLLALL19143118258
7–15LLPLLALLA>500>500<20<6052
8–16LPLLALLAL>500>50020<60<44
15–23ALWGPDPAA2034522<6051
34–42HLVEALYLV13827<6063
50–58YTPKTRREA>500>500<20<6044
53–61KTRREAEDL>500>500<20<6048
60–68DLQVGQVEL>500>50025<6054
91–99IVEQCCTSI>500>500<20<6048
101–109SLYQLENYC>500>500<208757
 10-mer     
1–10MALWMRLLPL35710421<60<44
2–11ALWMRLLPLL20682840864
4–13WMRLLPLLAL>50023424<6054
6–15RLLPLLALLA6261<20<60<44
7–16LLPLLALLAL436102288368
10–19LLALLALWGP>500453<20<60<44
13–22LLALWGPDPA273253<20<60<44
15–24ALWGPDPAAA150>50022<60<44
29–38HLCGSHLVEA>50047524<60<44
85–94SLQKRGIVEQ>500>50020<60<44
90–99GIVEQCCTSI>500>50021<6058
96–105CTSICSLYQL>500>500<20<6052
Total predicted binders101214515
Table 3.  Accuracy of algorithms in predicting preproinsulin (PPI) peptide binding to human leucocyte antigen (HLA)-A2.
 PredictedConfirmed bindersFalse positiveUnpredicted
NetMHC 3.01010018
NetMHC 3.21212016
SYFPEITHI1412216
BIMAS55023
MOTIFS1511417

Discussion

Multiple autoantigens have been described for CD4 and CD8 autoreactive T cells in T1D. These antigenic peptides have generally been demonstrated or assumed to bind MHC molecules with high to intermediate affinities. However, observations in recent years that low-affinity MHC-binding peptides can have highly immunogenic potential led us to speculate that low-affinity epitopes may be equally or even more important in the development of autoimmunity. Here, we addressed this hypothesis in an unbiased manner. Overlapping peptides spanning the whole PPI molecule were tested for HLA-A2 binding and recognition by CD8 T cells from the peripheral blood of T1D patients. We report that CD8 autoreactive T cells from T1D patients preferentially recognize peptides binding with very low affinities to HLA-A2. This may reflect a condition in which peptides binding with very low affinity to HLA-A2 are not presented efficiently in the thymus, allowing escape of autoreactive CD8 T cells from negative selection.

During T cell maturation in the thymus, T cells undergo a process of negative selection. Nevertheless, the observation that autoreactive T cells can be detected in peripheral blood of diabetic patients indicates that negative selection is incomplete. High-affinity CD8 T cells have been shown in the case of the islet antigen islet-specific glucose 6 phosphatase (IGRP) to be diabetogenic, while low-affinity IGRP-specific CD8 T cells were shown to be protective [28]. Here, the affinity of the T cell receptor (TCR) for its ligand was explored as a therapeutic opportunity in non-obese diabetic (NOD) mice by specifically deleting the high-affinity, pathogenic CD8 T cells. In human T1D, however, it is undetermined whether high-affinity T cells are the most pathogenic. It has been suggested that T cells escaping thymic deletion interact with self-peptide MHC with affinities below the threshold for deletion. This idea is supported by observations that the affinity of TCR for self-peptide MHC is generally lower than that for foreign peptide MHC [29,30]. In the case of glutamic acid decarboxylase (GAD)-specific CD4 T cells, a diverse range of antigen interactions has been observed, with high-avidity T cells co-existing with low- or moderate-avidity T cells, specific for the same GAD antigen [30–32].

Besides the strength of TCR–MHC interactions, the strength of peptide binding to MHC might influence negative selection in the thymus. While T cell responses in the periphery to non-self peptides appear to favour high HLA-binding affinity peptides to elicit a proper immune response [15], autoreactive epitopes in T1D may actually bind with low affinity to HLA-I. Studies in mice have shown that peptides binding with low affinity to MHC-II can be pathogenic [16,17,33,34]. In humans, low-affinity peptides have been observed to be recognized by autoreactive T cells of T1D and multiple sclerosis patients [18,19,35–37]. In our study, the frequencies of CD8 T cells recognizing self-antigens with a range of HLA-binding affinities was determined. Due to limited amounts of paediatric peripheral blood drawn shortly after diagnosis, PPI peptides were grouped based on their HLA-A2 binding affinity in three peptide pools to test peptide recognition, which precluded determining T cell recognition against each single peptide. Nevertheless, the frequency of CD8 T cells recognizing PPI peptides with very low HLA-A2 binding affinity is significantly higher than T cells recognizing PPI peptides of higher affinities. This implies that such low-affinity peptides, regarded commonly as irrelevant in (auto)immunity, may actually prove relevant with regard to potential immunogenicity and pathogenicity in T1D. Future studies to address the pathogenic potential of each of these peptides in T1D would be of great interest. Baker and colleagues tested overlapping PPI peptides for granzyme B responses [35]. In this case, peptide pools were made with overlapping PPI peptides, regardless of peptide-binding affinity. While combining peptides with high and low binding affinity could lead to competition, they observed none the less that granzyme B responses of T1D patients were directed mainly towards peptide pools containing PPI peptides with low predicted HLA-A2 binding scores. Even though the relation between peptide-binding affinity to HLA and T cell responses was not addressed directly, their conclusion corroborates our current findings and adds potential functionality in terms of cytotoxic potential of autoreactive T cells recognizing peptides with low HLA binding affinity. Perhaps more importantly, Baker and colleagues were able to compare their findings in adult-onset T1D patients with non-diabetic age-matched controls, showing a strong disease association of this phenomenon. Because our study was performed in children (mean age at onset 8 years), we were not allowed to draw blood from properly matched healthy children, precluding defining a relationship with disease.

T cells recognizing self-peptides with high affinity are presumably deleted during T cell education in the thymus [38]. However, the fast off-rates of low-affinity (self-)peptides and the subsequent reduced epitope presentation in the thymus could lead to T cell escape from negative selection and allow survival of autoreactive T cells. When encountering their self-antigen in high quantities in the periphery, such as preproinsulin in the pancreas, these cells could become activated and promote autoimmunity. Our data indicate that peptides binding with very low affinity to HLA-I may allow escape of CD8 T cells from negative deletion, while T cells specific for higher-affinity peptides may be deleted more efficiently and therefore circulate in lower frequencies in peripheral blood.

When testing all overlapping PPI peptides in vitro, we observed that the majority of binding peptides were not predicted to bind HLA-A2 by the algorithms. Notably, the majority of the unpredicted peptides bound HLA-A2 with very low affinity. This may be expected, as prediction algorithms were designed to predict high-affinity HLA-binding peptides. Peptides binding with very low affinity to HLA-A2 were recognized none the less by CD8 T cells from T1D patients in significantly increased frequencies, compared to peptides binding with higher affinities. It is unlikely that currently available algorithms will be capable of predicting binding of very low-affinity peptides. Binding experiments may therefore be required to identify this novel category of potential epitopes in autoimmunity.

Together, our data indicates that peptides binding with very low affinity to HLA molecules allow increased escape of autoreactive CD8 T cells from negative selection. T cells recognizing such epitopes can be found in higher frequencies in the peripheral blood of T1D patients. Our findings open avenues for the discovery of entirely new ranges of epitopes that may hold clues for loss of tolerance and causes of disease and serve as potential targets for tissue-specific immune intervention strategies.

Acknowledgements

We thank Dr Ton N. Schumacher, Division of Immunology, the Netherlands Cancer Institute, Amsterdam, the Netherlands, for providing reagents and reviewing the manuscript, and Mr Robert Cordfunke, Department of Immunohematology and Blood Transfusion, Leiden University Medical Center, Leiden, the Netherlands, for the peptide synthesis. This work was supported by the Juvenile Diabetes Research Foundation, the Dutch Diabetes Research Foundation and ZonMW.

Disclosure

None of the authors has conflicts of interest to declare.

Ancillary