In Vivo Cytotoxicity of Insulin-Specific CD8⁺ T-Cells in HLA-A*0201 Transgenic NOD Mice Free

NOD.β2m^null (7) and NOD.β2m^null.HHD mice (29) have been previously described and are maintained by brother-sister mating at the Albert Einstein College of Medicine. NOD.β2m^null.HHD mice express a monochain chimeric HLA-A*0201 molecule consisting of human β2-microglobulin covalently linked to the α1 and α2 domains of HLA-A*0201, followed by the α3 transmembrane and cytoplasmic portions of H-2D^b. Animal experiments were approved by the institutional animal care and use committee at Albert Einstein College of Medicine.

Peptide libraries containing all of the 8mer, 9mer, 10mer, and 11mer peptides that can be derived from murine preproinsulin 1 and 2 were synthesized by Mimotopes (Raleigh, NC) using their proprietary Truncated PepSet technology. Each mixture in the libraries contained four peptides with a common COOH-terminus, but having a length of 8, 9, 10, or 11 residues. The four peptides in each mixture were present in approximately equimolar amounts. Concentrated peptide stocks (2.75 mmol/l) were prepared in 50% acetonitrile/H₂O, and 40 μmol/l (i.e., ∼10 μmol/l for each peptide in the mixture) working stocks were obtained by serial dilution in PBS (pH 6.5). Individual peptides, having a purity of >90%, were obtained from Mimotopes. Concentrated stocks (10 mmol/l) were prepared in DMSO, and 10 μmol/l working stocks were obtained by dilution in PBS.

Mice were monitored weekly for glucosuria with Diastix reagent strips (Bayer, Elkhart, IN) and were considered diabetic when glucosuria was observed.

Islets were isolated from female NOD.β2m^null.HHD mice by collagenase perfusion of the common bile duct as described (29). Islets were handpicked and cultured for 7 days in 24-well tissue culture plates (∼50 islets/well) in RPMI medium supplemented with 10% FBS (Hyclone, Logan, UT) and 50 units/ml recombinant human IL-2 (PeproTech, Rocky Hill, NJ).

Interferon (IFN)-γ enzyme-linked immunospot (ELISPOT) was performed as previously described (29). Spots were counted using an automated ELISPOT reader system (Autoimmun Diagnostika, Strassberg, Germany). For select experiments, CD8⁺ T-cells were purified by negative selection from cultured islet infiltrates using magnetic separation (Miltenyi Biotec, Bergisch Gladbach, Germany).

Peptide binding to HLA-A*0201 was determined as described (39). Briefly, T2 cells were incubated for 18 h at 26°C. Cells were then incubated with peptide or PBS and human β2-microglobulin (Sigma-Aldrich, St. Louis, MO) for 20 h at 37°C, then incubated at 50°C for 3 min. Cells were stained with anti–HLA-A2 mAb BB7.2 (BD Biosciences Pharmingen, San Jose, CA) and analyzed by flow cytometry.

Splenocytes were prepared from NOD or NOD.β2m^null.HHD mice and divided into three groups. Cells were labeled with 0.5 or 5 μmol/l carboxyfluorescein diacetate succinimidyl ester (CFSE) (Molecular Probes, Carlsbad, CA) or with 10 μmol/l CellTracker Orange (Molecular Probes) for 10 min at 37°C. Labeling was stopped by quenching with complete Dulbecco's modified Eagle's medium on ice for 5 min. After washing, cells were pulsed with 1 μmol/l peptide for 1 h at 37°C. For experiments using NOD mice, control NOD splenocytes (5 μmol/l CFSE) were pulsed with an irrelevant peptide (the HLA-A*0201-binding hepatitis C virus core peptide 132–140; DLMGYIPLV). Target splenocytes labeled with 10 μmol/l CellTracker Orange were pulsed with the I9 variant of Ins1/2 B15–23 (LYLVCGERI), while targets labeled with 0.5 μmol/l CFSE were pulsed with IGRP 206–214 (VYLKTNVFL). For experiments using NOD.β2m^null.HHD mice, control NOD.β2m^null.HHD splenocytes (5 μmol/l CFSE) were pulsed with the irrelevant peptide IGRP 206–214. Target cells labeled with 10 μmol/l CellTracker Orange were pulsed with Ins1 B5–14, while targets labeled with 0.5 μmol/l CFSE were pulsed with Ins1/2 A2–10. After washing with PBS, cells were mixed in equal proportions and 1.5 × 10⁷ total cells in 200 μl injected intravenously into NOD.β2m^null (control) and NOD (8–14 weeks old) or NOD.β2m^null.HHD (15 weeks old) mice. Pancreatic lymph nodes were removed 16 h later, and single-cell suspensions were prepared and analyzed by flow cytometry on a FACSCalibur (BD Biosciences) after addition of 7-AAD (aminoactinomycin D) (BD Biosciences). Data shown are gated on live cells and uncompensated for better identification of the populations of interest. Gating was done for the NOD.β2m^null mouse and applied to all other mice in the experiment. Percent specific lysis was calculated as: {[(control population × A) − target population]/(control population × A)} × 100, where A = target population/control population in the NOD.β2m^null mouse. Islets were prepared from the same mice for propagation of infiltrating T-cells for 7 days, and IFN-γ ELISPOT was performed as described above. For data representation, the number of spots in wells with PBS was subtracted from those with peptide. For select experiments, CD8⁺ T-cell depletion was achieved by intraperitoneal injection of NOD.β2m^null.HHD mice with 0.25 mg of the depleting anti-CD8 mAb YTS169. The in vivo cytotoxicity assay was performed 7 days later, when peripheral blood CD8⁺ T-cells were depleted by 84 ± 4% (n = 5).

The type 1 diabetes–susceptible NOD. β2m^null.HHD stock expresses a monochain chimeric HLA-A*0201 molecule but no murine class I MHC molecules (29). To investigate whether CD8⁺ T-cell responses to insulin-derived peptides develop spontaneously in NOD.β2m^null.HHD mice during the pre-diabetic period, we used IFN-γ ELISPOT analysis and islet-infiltrating T-cells to screen peptide libraries representing murine Ins1 and Ins2. The libraries consisted of mixtures of four peptides each, with all four having a fixed COOH-terminus and ranging in length from 8 to 11 residues. As the libraries were offset by one, they contained every possible 8- to 11-mer peptide that can be derived from Ins1 and Ins2. Four peptide mixtures exerted antigenic activity (designated a–d in Figs. 1 and 2A). We tested T-cells for reactivity to individual components of the mixtures and identified Ins1 L3–11, Ins1 B5–14, and Ins1/2 A2–10 as the minimal epitopes present in mixtures a, b, and c, respectively (Fig. 2B). Reactivity to mixture d (spanning Ins2 L1–11) proved to be extremely rare. Consequently, we were unable to definitively identify the minimal epitope responsible for the activity of this mixture.

NOD.β2m^null.HHD mice express HLA-A*0201 but not the murine class I MHC molecules H-2K^d and H-2D^b encoded by their H2^g7 MHC haplotype (29). Nonetheless, to confirm that recognition of the antigenic insulin peptides was dependent on HLA-A*0201, we performed ELISPOT analysis using either T2 (expressing HLA-A*0201) or RMA-S/K^d cells (expressing H-2K^d, H-2D^b, and H-2K^b) as antigen-presenting cells. These experiments confirmed that peptide recognition was dependent on HLA-A*0201 (Fig. 2C). We also verified that CD8⁺ T-cells are responsible for recognition of the peptides by purifying CD8⁺ T-cells from the islet infiltrates and using both the CD8⁺ and CD8⁻ cell fractions as effectors. As shown in Fig. 2D, peptide-reactive cells were highly enriched in the CD8⁺ T-cell fraction.

Next, we monitored T-cell reactivity to Ins1 L3–11, Ins1 B5–14, and Ins1/2 A2–10 in nondiabetic 12- to 13-week-old NOD.β2m^null.HHD mice (Fig. 3A). These analyses revealed that Ins1/2 A2–10 is the immunodominant HLA-A*0201–restricted insulin epitope among mice in this age-group. We previously reported that HLA-A*0201–restricted islet-infiltrating T-cells from NOD.β2m^null.HHD mice recognize three peptides derived from the β-cell antigen IGRP (29). When we compared T-cell responses to insulin and IGRP in multiple 12- to 13-week-old NOD.β2m^null.HHD mice, four patterns were observed (Fig. 3B). Most mice responded to both insulin and IGRP (e.g., mouse 2), although a minor subset responded to only insulin (e.g., mouse 9) or only IGRP (e.g., mouse 19). A single mouse (mouse 6) responded to neither antigen. These findings support our previous assertion that individuals exhibit distinct patterns of reactivity to islet antigens (40).

Because female NOD.β2m^null.HHD mice can begin to show signs of overt diabetes as early as 12 weeks of age, the process of β-cell destruction can be well advanced by this time. It was therefore of interest to know whether HLA-A*0201–restricted T-cell responses to insulin epitopes were also present in the early pre-diabetic period, which would suggest an important role for these CD8⁺ effectors in the initiation of the pathogenic process. Thus, we examined 5- and 8-week-old female NOD.β2m^null.HHD mice and found that insulin-specific T-cell responses are indeed present in these younger age-groups (Fig. 4). T-cell responses to IGRP peptides in young mice were also noted.

Ins1 B5–14 and Ins1/2 A2–10 differ from their human counterparts by only a single residue (Fig. 5A). For the purpose of evaluating the potential translation of our epitope mapping studies to human type 1 diabetes, it was important to determine whether human (h) Ins B5–14 and A2–10 could also be recognized by T-cells in the context of HLA-A*0201. To investigate Ins2/hIns B5–14 as a potential antigen in type 1 diabetes, we investigated the ability of this peptide to be recognized by islet-infiltrating T-cells isolated from multiple NOD.β2m^null.HHD mice. As shown in Table 1, a subset of the mice that responded to Ins1 B5–14 also showed some, albeit usually less vigorous, reactivity to the Ins2/hIns version. In these cases, reactivity to the Ins2/hIns peptide may represent cross-reaction by Ins1-reactive T-cells or it may represent T-cells specific only for Ins2/hIns present in the islet infiltrates. The reactivity profile of mouse 2 (Table 1), which responded strongly to Ins2/hIns B5–14 and only minimally to its Ins1 equivalent, indicates that Ins2/hIns B5–14 can be independently antigenic in the context of HLA-A*0201. However, in the case of the other mice, it is not possible to discriminate between the two possibilities, which are not mutually exclusive, and it remains possible that in some instances, T-cells can cross-react with both peptides. We next investigated whether T-cells that recognize Ins1/2 A2–10 could cross-react with the human equivalent of this peptide, which differs by a single conservative substitution (Fig. 5A). Four of the five T-cell cultures examined showed appreciable cross-reaction (Table 1), suggesting that, because of the structural similarities of the peptide/MHC complexes, T-cells recognizing Ins1/2 A2–10 can also recognize hIns A2–10. Taken together, these results indicate that hIns B5–14 and A2–10 can be presented by HLA-A*0201 and may be relevant to human type 1 diabetes. Experimental proteasomal cleavage sites within human insulin have been reported at the COOH-termini of both of these peptides (41), lending further support to their potential relevance in humans.

A cell-based MHC stabilization assay was next used to probe the ability of the antigenic peptides identified here to bind to HLA-A*0201. Ins1 L3–11 and Ins1 B5–14 showed readily detectable binding to HLA-A*0201 (Fig. 5), although neither peptide bound as well as the positive control peptide (Flu MP 58–66). Ins2/hIns B5–14 also showed measurable binding. In contrast, Ins2 and hIns L3–11, which differ from Ins1 L3–11 by three or four residues, respectively, did not show detectable binding. Interestingly, Ins1/2 A2–10, the immunodominant insulin epitope, did not bind detectably to HLA-A*0201, nor did its human counterpart, although both peptides are readily recognized by HLA-A*0201–restricted T-cells (Fig. 3 and Table 1). A likely explanation for the inefficient binding of the A2–10 peptides is that they lack preferred residues at both of the HLA-A*0201 anchor positions, i.e., L or M at position 2 and V or L at the COOH-terminus (42). This result is reminiscent of Ins1/2 B15–23, which is targeted by H-2K^d–restricted CD8⁺ T-cells in standard NOD mice, yet binds very poorly to H-2K^d (36,43).

The ability of hIns B5–14 and A2–10 to be recognized by HLA-A*0201–restricted T-cells (Table 1) suggests they may be excellent candidates for applicability to humans, e.g., to form the basis of a peptide-based intervention or for trial monitoring of new-onset patients undergoing intervention therapies. However, we identified these epitopes using T-cells from pre-diabetic NOD.β2m^null.HHD mice. To confirm that they are also targeted after the onset of overt disease, we isolated T-cells from the residual islets of diabetic female NOD.β2m^null.HHD mice that had not received exogenous insulin therapy (Fig. 6). We found that each of the insulin epitopes identified here was recognized by at least one diabetic mouse. This suggests the potential applicability of the human counterparts of these CD8⁺ T-cell epitopes for the development of peptide-based intervention or monitoring strategies in both pre-diabetic and new-onset type 1 diabetic patients.

Our ELISPOT results indicated that T-cells capable of secreting IFN-γ in response to insulin peptides are present in the islets of NOD.β2m^null.HHD mice. The pathogenicity of these T-cells would be further supported if they were shown to be cytotoxic in vivo. To address this, we modified an in vivo assay, previously used to demonstrate cytotoxicity after immunization with a peptide (44,45), to instead examine the ability of spontaneously arising autoreactive T-cells to kill target cells presenting insulin peptides. In our assay, three populations of splenocytes are differentially labeled with fluorescent dyes, allowing their detection by flow cytometry. Two of the populations are pulsed with target peptides and a third with an irrelevant peptide. After mixing, the cells are injected into NOD.β2m^null mice, which lack CD8⁺ T-cells and therefore act as a negative control, and into recipient autoimmune-prone mice. Sixteen hours later, pancreatic lymph nodes, where injected cells would be expected to encounter β-cell autoantigen-specific T-cells, are retrieved and cells are analyzed by flow cytometry. Lysis of a population can be detected as a decrease in the ratio between the target and control populations in the autoimmune-prone mice compared with the ratio in NOD.β2m^null mice. To demonstrate the assay and its applicability, we performed an experiment with well-established targets of CD8⁺ T-cells in NOD mice, i.e., IGRP 206–214 (40) and Ins1/2 B15–23 (36). As seen in Figs. 7A and B, specific lysis of populations pulsed with both target peptides could be detected. These results were further validated by the finding that peptide reactivity of T-cells cultured from the islets of the same mice, measured by IFN-γ ELISPOT (Fig. 7C), correlated with the cytotoxicity observed in the pancreatic lymph nodes. For example, NOD Mouse 4 showed high in vivo cytotoxicity against IGRP 206–214-bearing targets and a strong response to this peptide by islet-infiltrating T-cells. To our knowledge, this is the first time that an in vivo cytotoxicity assay has been used to monitor the activity of spontaneously arising autoreactive T-cells.

Using this system, we asked whether in vivo cytotoxicity could be detected against Ins1 B5–14 and Ins1/2 A2–10 in NOD.β2m^null.HHD mice (Figs. 7D and E). Indeed, we found that T-cells reactive with both epitopes are cytotoxic in vivo. These findings also correlate well with IFN-γ ELISPOT results using T-cells propagated from the islets of the same mice (Fig. 7F). To confirm that CD8⁺ T-cells are required for cytotoxicity, we also performed the assay in five CD8⁺ T-cell–depleted NOD.β2m^null.HHD mice. The average lysis observed in these mice was only 1.5% of that seen in nondepleted animals (data not shown). Thus, Ins1 B5–14 and Ins1/2 A2–10 are CD8⁺ T-cell targets during the HLA-A*0201–restricted pathogenic autoimmune response in vivo.

CD8⁺ T-cells reactive to several islet antigens, including insulin and IGRP, have been detected in the peripheral blood of type 1 diabetic patients directly ex vivo (23–28). The finite, and often small, amount of blood available from subjects requires investigators to limit the number of peptides that can be examined for T-cell reactivity. Thus, several criteria have been developed to select candidate peptides for these investigations. These include predicted or experimental binding to an HLA molecule of interest (23–28,41), predicted or experimental evidence for proteasomal cleavage at the peptide's COOH-terminus (26–28,41), and immunogenicity when administered to an HLA transgenic mouse (23,41). These criteria, often used in combination, have been helpful in allowing the identification of candidate peptides that were subsequently shown to be recognized by peripheral T-cells of type 1 diabetic patients directly ex vivo. Our approach, whereby islet-infiltrating T-cells from diabetes-susceptible HLA transgenic NOD mice are used to identify candidate peptides, has also successfully predicted epitopes relevant to type 1 diabetes in humans (23,29) and presents several advantages. The candidate peptides identified here using spontaneously arising islet infiltrates are naturally processed (albeit by mouse β-cells that may process antigens for presentation differently from human β-cells), since T-cell reactivity to them is detected in the absence of immunization. As they are recognized by T-cells in the islet infiltrate, they are very likely to be disease relevant. Also, the peptides are known to be capable of binding the HLA molecule of interest in an immunologically relevant manner, regardless of the quality of binding revealed by a cell-based or biochemical assay. Importantly, Ouyang et al. (24) recently reported that the experimental binding affinity of β-cell peptides to HLA-A*0201 correlated inversely with the magnitude of the T-cell responses observed to those peptides in type 1 diabetic patients. T-cells reactive to weakly binding self-peptides may escape deletion in the thymus because of poor presentation. The fact that the candidate insulin epitopes reported here were not previously identified as peptides of interest using the aforementioned criteria demonstrates that our strategy is an important and useful complement to those that have been described by others. Furthermore, as shown by this work, our approach has the advantage that it allows us to determine whether the identified peptides are not only presented by HLA-A*0201 and elicit an IFN-γ response, but also whether they are targeted by pathogenic T-cells in vivo.

In NOD mice, Ins2 is expressed in the thymus, while Ins1 is not (46). The finding that Ins2-deficient NOD mice show accelerated disease led to speculation that thymic expression of Ins2 permits negative selection of at least some pathogenic insulin-reactive T-cells (34,47). In the case of CD4⁺ T-cells, evidence in support of this notion has been reported (47). Our studies in the NOD.β2m^null.HHD mice are consistent with the idea that Ins2-reactive CD8⁺ T-cells are also eliminated by central tolerance. For example, two of the epitopes identified here (i.e., Ins1 L3–11 and Ins1 B5–14) are unique to Ins1. Further, in contrast to Ins1 B5–14, reactivity to the Ins2 version of this peptide was rare (Table 1), although the two differ by only a single residue and both show detectable binding to HLA-A*0201 in a cell-based assay (Fig. 5). Collectively, these findings are suggestive of negative selection of CD8⁺ T-cells capable of recognizing Ins2 peptides. Although reactivity to A2–10, an epitope shared between Ins1 and Ins2, was detected in nearly all mice examined (Fig. 3), this peptide does not bind well to HLA-A*0201, at least as measured in our cell-based assay (Fig. 5). Similarly, B15–23 is shared between Ins1 and Ins2, is targeted by islet-infiltrating H-2K^d–restricted CD8⁺ T-cells in standard NOD mice (36), and shows poor class I MHC binding (43). T-cells reactive to A2–10 and B15–23 likely escape central tolerance despite thymic expression of Ins2, as the peptides are unlikely to be well presented in the thymus because of poor MHC binding. In humans, the regulatory region of the insulin gene is polymorphic, with different alleles directing thymic expression to varying degrees (48,49). It is likely that the set of insulin peptides targeted in individual type 1 diabetic patients will also vary depending on the extent of thymic expression conferred by the insulin allele(s) that they carry.

We previously reported that, during the pre-diabetic period, individual NOD mice exhibit their own unique patterns of CD8⁺ T-cell reactivity to different islet antigens, with a given mouse responding to as few as zero or as many as all three of the antigens examined (40). We now document parallel findings for the HLA-A*0201–restricted T-cell response in NOD.β2m^null.HHD mice. Whereas most mice responded to both insulin and IGRP, some responded to only one of the antigens and others to neither (Fig. 3B). Even among mice with insulin-responsive CD8⁺ T-cells, there was variability in the number and identity of the targeted insulin peptides. Diabetic NOD.β2m^null.HHD mice also showed variation in their reactivity to peptides derived from insulin and IGRP (Fig. 6). Interestingly, all diabetic mice did not respond to all of the epitopes, as might have been predicted considering the phenomenon of epitope spreading. Such variation, even among inbred mice, strongly suggests that individual human type 1 diabetic patients will also show unique reactivity patterns to β-cell antigens. These results emphasize the importance of a more complete characterization of the epitopes targeted by CD8⁺ T-cells in patients, since a diagnostic, monitoring, or therapeutic reagent based on a single one is unlikely to offer wide population coverage. Recent work indicates that the peptides targeted during the spontaneous development of the disease may also be important in the context of islet graft rejection (23,24,26,28,50), providing an additional impetus for careful and continued epitope identification and validation.

This work was supported by National Institutes of Health Grants DK64315 (to T.P.D.), DK52956 (to T.P.D.), DK77500 (to T.P.D.), DK69280 (to M.P.M.), DK46266 (to D.V.S.), DK51090 (to D.V.S.), and DK20541 (to Albert Einstein College of Medicine's Diabetes Research and Training Center) and by grants from the Juvenile Diabetes Research Foundation (to T.P.D., M.P.M., and D.V.S.). The flow cytometry facility at Albert Einstein College of Medicine is supported by National Institutes of Health Cancer Center Grant CA13330.