Compensatory Mechanisms Allow Undersized Anchor-Deficient Class I MHC Ligands To Mediate Pathogenic Autoreactive T Cell Responses

Type 1 diabetes (T1D) in humans and the NOD mouse model develops as a result of T cell–mediated autoimmune elimination of pancreatic β cells. Studies in the NOD mouse revealed that both CD4 and CD8 T cells are critical for destructive insulitis leading to T1D development. Evidence specifically implicating CD8 T cells includes a report that NOD mice lacking the β₂-microglobulin (β2m) subunit required for MHC class I expression and CD8 T cell development are T1D resistant (1). NOD mice made CD8 T cell–deficient through genetic ablation of the CD8α gene are similarly protected (2). Furthermore, unlike animal models, human insulitis consists mostly of CD8 T cells (3).

The T cell clones 8.3, G9C8, and AI4 are representative of three β cell–cytotoxic CD8 T cell populations in NOD mice. 8.3-like T cells recognize residues 206–214 of islet-specific glucose-6-phosphatase catalytic subunit-related protein (IGRP)_206–214 in an H-2K^d–restricted manner and are often the most prevalent of the three clonotypes in infiltrated islets (4). NOD mice transgenically expressing the 8.3 TCR develop an accelerated high rate of T1D but that is decreased when the SCID (scid) mutation, which eliminates production of nontransgenic T and B cells, is introduced into the stock, unless the mice are engrafted with polyclonal CD4 T cells from standard NOD donors (5). These results indicate that 8.3-like CD8 T cells require CD4 T cell help to efficiently induce T1D. The G9C8 CD8 T cell clone, which recognizes a peptide derived from the insulin B chain (InsB_15–23) in the context of H-2K^d, is also capable of independently transferring disease, but only after activation in vitro (6). G9C8 TCR-transgenic (Tg) NOD mice develop only a very low rate of T1D (7), indicating that this clonotype on its own is not highly pathogenic. In contrast, AI4 TCR-Tg NOD mice rapidly develop high levels of T1D, even when other lymphocyte populations are eliminated by introduction of the scid mutation or an inactivated Rag1 (Rag1^null) gene (8, 9). Thus, of the diabetogenic CD8 T cells identified to date in NOD mice, the AI4-like population appears to exert the highest level of helper cell–independent pathogenic activity.

The AI4 T cell clone was originally isolated from the islet infiltrate of a 5–6-wk-old prediabetic female NOD mouse (10). The presence of this clonotype in such a young mouse suggests that it may be an important highly pathogenic driver in the very earliest stages of T1D development. We hypothesize that one reason for this may be the antigenic-recognition promiscuity (cross-reactivity) of the AI4 TCR. Indeed, the AI4 TCR was shown to recognize multiple ligands in the context of H-2D^b. These include a peptide derived from the widely expressed protein dystrophia myotonica kinase (DMK)_138–146 (4) and its related mimotopes (Mim and MimA2) (11). AI4 T cells also likely recognize an unknown H-2K^d–restricted epitope, because such effectors only efficiently kill β cells that express both H-2K^d and H-2D^b (11). Furthermore, screening of an H-2K^d–binding positional scanning peptide library revealed the ability of AI4 to recognize peptides presented by this class I molecule (11). Finally, MHC class II variants encoded by the T1D-protective H2^nb1 MHC haplotype can mediate dominant thymic-negative selection of AI4 CD8 T cells (12).

Given the apparent promiscuity of its TCR, the original goal of the current study was to assess, in the presence of CD4 T cell help, the relative hierarchy in NOD mice of AI4-like CD8 T cells in driving T1D development versus those recognizing IGRP or insulin. Our efforts were initiated based on previous studies that explored the hierarchy of T cell responses, which takes place during the phenomenon of epitope spreading that accompanies T1D development (13, 14). In NOD mice, T cell reactivity against proinsulin is required to trigger subsequent IGRP-specific responses (13). Furthermore, Tg overexpression of Ins2 in APCs inhibited T1D development in 8.3 TCR-Tg NOD mice that retained residual non-Tg T cell populations (14). Thus, in the current study we assessed whether Tg overexpression of Ins2 or IGRP in APCs would attenuate AI4 CD8 T cell responses. These analyses revealed that the high diabetogenic activity of AI4 could be attributed to the promiscuity of its TCR extending to recognition, in an H-2D^b–restricted fashion, of an unusually short insulin-derived 7-mer peptide, InsA_14–20, lacking a C-terminal anchor residue normally used for binding to this MHC class I variant. Crystallographic analysis revealed the compensatory mechanisms that enable peptides lacking a C-terminal anchor to nonetheless bind sufficiently well to the MHC to mediate a pathogenic T cell response. Structural studies support the idea that the AI4 TCR primarily focuses its recognition on a short amino acid core of InsA_14–20, shared by its other H-2D^b–binding ligands, thus providing a structural basis for the promiscuity of this highly diabetogenic T cell clone. The poor MHC binding exhibited by InsA_14–20 suggests a mechanism for how cognate self-reactive T cells would escape thymic negative selection and implies that undersized peptides might be particularly important targets of CD8 T cells in autoimmune diseases in general.

NOD.AI4αβ−Tg mice (9) transgenically expressing the TCR from the diabetogenic CD8 T cell clone AI4 (Vα8/Vβ2), similar mice carrying a functionally inactivated Rag1 gene (NOD.Rag1^null.AI4) (8), and NOD.8.3 mice transgenically expressing the diabetogenic TCR 8.3 (5) have been described. NOD/ShiLtDvs mice were maintained by sibling matings. NOD mice expressing mouse proinsulin 2 or IGRP under the control of an MHC class II (I-Eακ) promoter (NOD.PI or NOD.IGRP mice, respectively) were described (13, 15). NOD.INS2-knockout (KO) mice are homozygous for an inactivated proinsulin 2 allele (16). All mice were bred under specific pathogen–free conditions at The Jackson Laboratory or Albert Einstein College of Medicine following protocols approved by their respective Institutional Animal Care and Use Committee.

Bone marrow (BM) from NOD.Rag1^null.AI4, NOD.PI, or NOD.IGRP donors was depleted of mature CD4 and CD8 cells using streptavidin-conjugated magnetic beads (Miltenyi Biotec). NOD female mice were lethally irradiated (1200 R from a [¹³⁷Cs] source) at 5 wk of age and reconstituted i.v. with 5 × 10⁶ single or 1:1 mixtures of BM cells from the indicated donors. Recipient mice were monitored for T1D for up to 30 wk postreconstitution. Chimeras were assessed for proportions of thymic and splenic CD8 T cell populations at 6 wk postreconstitution or upon T1D development.

T1D was assessed by weekly monitoring of glycosuria with Ames Diastix (Bayer, Diagnostics Division), with disease onset defined by two consecutive values ≥ 3.

Peptide libraries containing all of the 8-mer, 9-mer, 10-mer, and 11-mer peptides that can be derived from murine preproinsulin 1 and 2 were synthesized by Mimotopes using their proprietary Truncated PepSet Technology. Each mixture in the libraries contained four peptides with a common C 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 mM) were prepared in 50% acetonitrile/H₂O, and 40 μM (10 μM for each peptide in the mixture) working stocks were prepared by dilution in PBS (pH 6.5).

Individual peptides having a purity ≥ 90% were obtained from Mimotopes. Concentrated stocks (10 mM) were prepared in dimethylformamide, and 10 μM working stocks were obtained by serial dilution in PBS (pH 6.5).

Flow cytometry was performed with FACSCalibur or LSR II instrumentation (BD Biosciences) and FlowJo data analysis software (TreeStar). Fluorochrome-conjugated Abs (BD Biosciences) used specifically recognized Vα8 (B21.14), CD8 (53-6.7), CD4 (GK1.5), IFN-γ (XMG1.2), BP-1 (6C3), LY5.1 (A201.7), CD326 (G8.8), and H-2D^b (KH95).

Ag-presenting dendritic cells (DCs) were purified from collagenase D–digested (Roche Applied Science) spleens or thymi of female NOD, NOD.PI, or NOD.INS2 KO mice using anti-CD11c–conjugated magnetic beads (Miltenyi Biotec) and cultured at various ratios with NOD.Rag1^null.AI4 T cells prelabeled with 2.5 mM CFSE. In certain experiments, porcine insulin (Novo Nordisk) was added. T cell proliferation was assessed after incubation at 37°C for 3 d by flow cytometric detection of CFSE dilution and CD8 staining.

Splenic DCs from NOD mice were cultured overnight (16–20 h) at 37°C with 1 μM porcine insulin, 4 μM peptide mixes from the preproinsulin peptide libraries, or 1 μM individual peptides (except where indicated otherwise) in the presence of LPS. On day 1, 2 × 10⁴ CD8 T cells from spleens of NOD.AI4αβ Tg or NOD.8.3 mice, isolated using MACS MicroBeads (Miltenyi Biotec), were cultured with 2 × 10⁴ peptide-loaded DCs for 72 h, after which BrdU labeling solution (Roche) was added. BrdU incorporation was measured by ELISA after 16–20 h, following the manufacturer’s protocol.

To determine MHC class I restriction, 50 μg/ml H-2K^d (SF1-1.1.1; eBioscience) or H-2D^b (28-14-8; BD Biosciences) blocking Ab was added to the peptide-loaded DCs prior to introduction of the CD8 T cells. T cell proliferation was assessed by BrdU incorporation.

RMA-S cells, cultured overnight at 26°C, were pulsed with 1 μM peptide in DMEM with 10% FBS for 1 h at 26°C, incubated at 37°C for 3 h, washed, stained with anti–H-2D^b mAb, and analyzed by flow cytometry. The fluorescence index was calculated as (mean fluorescence intensity of mAb binding in the presence of peptide)/(mean fluorescence intensity of nonpeptide-pulsed cells).

Escherichia coli strain BL21 (DE3) pLysS cells, expressing the H-2D^b H chain and murine β2m^a as insoluble inclusion bodies, were harvested and suspended in buffer containing 50 mM Tris-HCl (pH 8), 100 mM NaCl, 20% (w/v) sucrose, 1 mM EDTA, and 10 mM DTT. DNase I (10 μg/ml) was added to the suspension, the cells were lysed, and insoluble protein was pelleted by centrifugation. The inclusion bodies were washed three times with buffer containing 10 mM Tris-HCl (pH 8), 100 mM NaCl, 0.5% Triton X-100, 1 mM EDTA, and 10 mM DTT. The detergent was removed by washing the inclusion bodies twice with this buffer but omitting the Triton X-100. Protein purity was confirmed by SDS-PAGE.

The purified, detergent-free inclusion bodies were solubilized in buffer containing 6 M guanidine hydrochloride, 10 mM Na-acetate (pH 4.5), 5 mM EDTA, and 1 mM DTT. The complex was refolded by rapid dilution (17, 18), where solubilized β2m inclusion bodies were injected into the refolding buffer (400 mM arginine hydrochloride, 100 mM Tris-HCl [pH 8], 1 mM EDTA, 5 mM reduced glutathione, and 0.5 mM oxidized glutathione) at a final concentration of 30 μg/ml (2.59 μM), followed sequentially by the pure peptide (50 μg/ml; ∼50 μM) and the H-2D^b H chain inclusion bodies (60 μg/ml; 1.89 μM). The peptide/MHC complex was purified by size-exclusion chromatography using Superdex 200 (GE Healthcare Life Sciences) with a buffer composed of 20 mM HEPES (pH 7), 150 mM NaCl, and 1 mM EDTA.

The refolded and purified MimA2/H-2D^b and YQLENYCGL/H-2D^b complexes (7.5 mg/ml) were subjected to crystallization against ∼600 conditions using the sitting drop vapor diffusion method at room temperature. Good-quality crystals of MimA2/H-2D^b were obtained from the condition 10% (v/v) polyethylene glycol 3000, 0.1 M phosphate/citrate (pH 4.2), and 0.2 M NaCl. MimA2/H-2D^b crystals were flash-frozen in liquid nitrogen with 20% glycerol as the cryoprotectant. Data were collected at the beamline 31-ID (Lilly Research Laboratories Collaborative Access Team) of the Advanced Photon Source (Argonne National Laboratory, Argonne, IL). Diffraction data were processed using the software HKL2000 to 1.98-Å resolution. Molecular replacement was carried out using the Phaser module in the PHENIX suite (19), using SSLENFRAYV/H-2D^b (PDB ID: 1YN6) as the starting model (with the peptide removed). Further refinement rounds using simulated annealing yielded a clear positive, continuous density where the peptide should be bound. The final model has two MimA2/H-2D^b complexes along with two Zn–imidazole–SO₄⁻ clusters at the interfaces of the complexes. The high-quality density helped us to unambiguously build the MimA2 peptide and further refine the structure.

Conjoined, plate-like crystals of YQLENYCGL/H-2D^b were readily obtained in several conditions, and data were collected from one such crystal obtained in the presence of 0.2 M ammonium sulfate, 0.1 M HEPES (pH 7.5), and 25% polyethylene glycol 3350. X-ray diffraction data were collected at the X29A beamline, National Synchrotron Light Source, Brookhaven National Laboratory (Upton, NY), and processed using the iMosflm package. Preliminary scaling trials suggested diffraction anisotropy. The anisotropy resulted in data to ∼2.3 Å in the best orientation and 3.4 Å in the worst. The processed data were subjected to anisotropy analysis using the University of California, Los Angeles anisotropy server (20) to perform appropriate ellipsoidal truncation, anisotropic scaling, and B-factor sharpening to restore the magnitudes of the high-resolution reflections. The resulting isotropic data had an ellipsoidal resolution boundary with limits of 3.4 Å along a*, 2.4 Å along b*, and 2.6 Å along c*. These truncated data were used for further structure solution using molecular replacement and refinement. Two YQLENYCGL/H-2D^b complexes were found in the asymmetric unit, and the YQLENYCGL peptide was unambiguously assigned to the clear difference density that appeared.

The data collection and refinement statistics for both complexes are shown in Table I, and the refined models have been deposited in the Protein Data Bank (http://www.pdb.org) under accession numbers 3WS6 and 3WS3.

To generate peptide-specific CTLs in vitro, splenocytes from NOD.AI4αβ Tg mice were cultured in the presence of mitomycin C–treated NOD splenocytes and 1 μM YQLENYC at a ratio of 1:4 with 12.5 U/ml IL-2. After 4–5 d of culture, CTLs were collected, washed, and used for experiments.

RMA-S cells were loaded with 1 μM test peptides or the negative control H-2D^b–binding peptide TRL9 (TSPRNSTVL) and cultured at 37°C for 1 h. Peptide-pulsed cells were either added directly to 96-well round-bottom plates or added after washing to eliminate free peptide. CTLs were added to the wells at E:T ratios of 5:1 or 2.5:1 in triplicate and cultured for 4 h at 37°C. Lactate dehydrogenase release and specific cytotoxicity were determined according to the manufacturer’s protocol for the LDH Cytotoxicity Assay Kit (Pierce).

Islets were isolated after perfusion of the pancreas with collagenase P (Roche) and cultured for 7 d in RPMI 1640 supplemented with 1 mM sodium pyruvate, 28 μM 2-ME, nonessential amino acids, 10% FBS, and 50 U/ml recombinant human IL-2, as described (21).

IFN-γ ELISPOT was performed as described (21), except that splenic DCs from NOD mice, cultured overnight with test peptides in the presence of LPS, were used as APCs. For most experiments, responses are reported as stimulation index, defined as (number of spots in response to a test peptide)/(number of spots in response to the average of the negative-control peptides) (i.e., H-2D^b binding TRL9 and H-2K^d binding TUM [KYQAVTTTL]). The cutoff for positivity was a stimulation index > 2 and a test peptide spot number > 5 per 1 × 10⁵ T cells (22).

Medullary thymic epithelial cells (mTECs) were isolated from thymi of 5-wk-old female NOD or NOD.PI mice (at least six mice pooled/strain) with collagenase D, DNase I, and Dispase II at 37°C for 30 min, with agitation every 5–10 min. This was followed by the addition of EDTA to 5 mM and incubation for 5 min. Cells were washed and stained for cell surface markers BP-1, LY5.1, and CD326, gated as previously described (23), and sorted using a FACSAria. Briefly, mTECs were defined as LY5.1⁻, CD326⁺, and BP-1⁻. RNA was prepared using an RNeasy Micro Kit, followed by cDNA synthesis using the QuantiTect Reverse Transcription Kit (both from QIAGEN). Primer sequences for real-time quantitative RT-PCR analysis were as follows: Insulin 2, 5′-GCTTCTTCTACACACCCATGT-3′ and 5′-AGCACTGATCTACAATGCCAC-3′; DMK, 5′-CCAACATGTCAGCCGAAGTG-3′ and 5′-TCAGGGGGCGAAGGTGG-3′; and 18S rRNA, 5′-CCGCAGCTAGGAA-3′ and 5′-CGAACCTCCGACT-3′. Samples and primers were combined with Power SYBR Green PCR Master Mix (Applied Biosystems) and acquired on an Applied Biosystems ViiA7 Real Time PCR System. Cycle threshold (CT) values ≥ 35 were defined as zero expression. CT values were normalized to 18S rRNA expression.

Statistical analysis was performed as indicated in the figure legends using GraphPad Prism Version 5.

NOD stocks transgenically expressing either proinsulin 2 (NOD.PI) or IGRP (NOD.IGRP) in class II MHC–expressing APCs are rendered resistant or remain susceptible to T1D development, respectively (13, 15). In NOD.PI mice, T cell tolerance was established to both proinsulin and IGRP, yet in NOD.IGRP mice, T cell tolerance was only established to IGRP and not to proinsulin (13). This suggested that T1D development in NOD mice entails an initial T cell response to proinsulin that must occur to allow for the subsequent activation of pathogenic effectors targeting IGRP. A β cell–specific Ag recognized by the diabetogenic CD8 T cell clonotype AI4, and accounting for its in vivo cytotoxic activity against β cells but not other cell types, has remained elusive. Nonetheless, we were interested in the hierarchy of AI4-like CD8 T cells in naturally driving the course of T1D development relative to those recognizing insulin- or IGRP-derived epitopes. Therefore, we set out to determine how systemic immunological tolerance to proinsulin or IGRP, established by transgenically expressing these proteins in APCs, would affect the ability of AI4 T cells to induce T1D in NOD mice.

Lethally irradiated NOD mice were reconstituted with a 1:1 mix of BM from NOD.Rag1^null.AI4 mice and either NOD.PI (Fig. 1A) mice or NOD.IGRP mice (Fig. 1B) and monitored for T1D development. As controls, additional mice were reconstituted with BM from NOD.Rag1^null.AI4 donors alone or in combination with BM from standard NOD mice. Compared with both groups of controls, T1D development was significantly inhibited in mixed chimeras in which NOD.PI APCs were present (Fig. 1A). Conversely, in mixed BM chimeras in which NOD.IGRP APCs were present, AI4-mediated T1D was not inhibited (Fig. 1B). If interpreted through the prism of the epitope-spreading paradigm, the above results could indicate that T cell responses must be initiated against proinsulin, but not IGRP, to allow for the subsequent efficient activation of the diabetogenic AI4 T cell response. However, we also considered the possibility that AI4’s β cell–specific Ag might be proinsulin itself and that a direct tolerogenic interaction between AI4 and proinsulin-expressing BM-derived APCs was responsible for the disease inhibition observed.

To discriminate between these two possibilities, irradiated NOD recipients were again reconstituted with mixed BM from NOD/NOD.Rag1^null.AI4, NOD.PI/NOD.Rag1^null.AI4, or NOD.IGRP/NOD.Rag1^null.AI4 donors. At 6 wk postreconstitution, both the percentage and the absolute number (Fig. 1C, 1D) of CD4⁺CD8⁺ (double-positive; DP) thymocytes expressing the AI4 TCR (Vα8) were greatly reduced in mixed BM chimeras in which proinsulin-expressing APCs were present. The decreased number of AI4 T cells at the DP stage of thymocyte development suggested that these cells were undergoing negative selection in the presence of proinsulin-expressing APCs. The parallel reduction observed in the frequency and number of splenic AI4 T cells in the presence of proinsulin-expressing APCs is consistent with this notion (Fig. 1E, 1F). These findings revealed proinsulin to be a β cell–specific Ag recognized by AI4.

To more directly test the hypothesis that a proinsulin peptide is recognized by AI4, we compared the ability of T cells from NOD.Rag1^null.AI4 mice to proliferate in vitro when stimulated by thymic DCs from NOD, NOD.PI, or NOD.INS2 KO mice. If AI4 CD8 T cells undergo negative selection through recognition of a proinsulin epitope displayed by tissue-resident APCs, then thymic DCs from NOD.PI mice should be able to induce the highest proliferation of such effectors. Indeed, it was found that both thymic DCs (Fig. 1G) and splenic DC (Fig. 1H) from NOD.PI mice induced robust proliferation of AI4 TCR Tg T cells, in contrast to those from NOD and NOD.INS2 KO mice (Fig. 1G, 1H). These findings suggested the presence of an AI4 epitope derived from proinsulin 2.

To further investigate the existence of an insulin peptide recognized by AI4 T cells, we tested whether thymic DCs isolated from standard NOD mice could process exogenously added porcine insulin to generate an epitope recognized by these diabetogenic effectors. Porcine insulin was used instead of murine insulin 2 because the former is readily available commercially, and there are only four amino acid differences between the two mature insulin proteins. Incubation of NOD thymic DCs with increasing concentrations of insulin led to dose-dependent proliferation of cocultured splenic CD8 T cells from NOD.Rag1^null.AI4 mice (Fig. 2A). This result demonstrates that a peptide capable of stimulating AI4 T cells can be naturally processed from the native, mature insulin protein and presented by DCs. Such responsiveness of CD8 T cells from the Rag1^null-carrying NOD stock indicated that insulin reactivity is truly imparted by the AI4 TCR, rather than another potentially resulting from incomplete allelic exclusion. Similarly, CD8 T cells from standard NOD.AI4αβ Tg mice proliferated when cocultured with NOD splenic DCs in the presence of porcine insulin (Fig. 2B).

To identify the insulin epitope recognized by the AI4 TCR, we did an exhaustive screening of a peptide library spanning the preproinsulin 2 protein. Reactivity to three overlapping peptide mixes (Ins2 101, 102, and 103) was observed (Fig. 2C). The peptides in these mixes are shared between murine preproinsulin 1 and 2 (and with porcine insulin). We then tested for reactivity to the individual peptides in each of the positive mixes and found four to which AI4 T cells responded strongly (Fig. 2D). To ascertain the likely minimal epitope recognized by the AI4 TCR, we aligned the four overlapping Ins2 peptides with the previously described H-2D^b–binding mimotope peptides Mim and MimA2 and the 9-mer peptide FNL9 (DMK_138–146), all recognized by this clonotype (4, 11), and with the peptide-binding motif for H-2D^b, which consists of anchor residues at P5 (N) and P9 (M, I, or L) (24) (Fig. 2E). This alignment revealed a common “ENY” motif present in all of the peptides recognized by AI4. It also suggested that the N terminus of AI4’s Ins2 epitope was Y, and that the N-terminal–extended peptides were trimmed during the assay period. We further found that AI4 has an unprecedented ability among CD8 T cells to recognize the undersized peptides YQLENY (InsA_14–19) and YQLENYC (InsA_14–20), with the 7-mer being the stronger stimulant (Fig. 2F).

Given that the 6-mer and 7-mer peptides were unusually short for binding to H-2D^b, a variant that favors peptides having nine residues (24), we wanted to ensure that TCR specificity was required for recognition and that it occurred in an H-2D^b–restricted manner. This was done using H-2K^d–restricted 8.3 CD8 T cells that recognize IGRP_206–214 (25) or its superagonist mimotope NRP-V7 (26). The InsA_14–20 7-mer could not induce 8.3 T cell proliferation, thus demonstrating the requirement for AI4 TCR specificity in recognizing this epitope (Fig. 3A). A blocking Ab to H-2D^b, but not one to H-2K^d, abolished recognition of InsA_14–20 by AI4 T cells (Fig. 3B), demonstrating that the atypical 7-mer peptide YQLENYC (InsA_14–20) is recognized in the context of H-2D^b.

To assess the potential of the 7-mer peptide to contribute to the pathogenesis of AI4-mediated T1D, we tested whether it could support the in vitro generation of IFN-γ–secreting CTLs from NOD.AI4αβ-Tg mouse splenocytes. We found that culture of YQLENYC-pulsed mitomycin C–treated NOD splenocytes with AI4 splenocytes induced CD8 effectors that released IFN-γ when restimulated with either YQLENYC or MimA2, as monitored by intracellular cytokine staining (Fig. 4A). Note that restimulation with 10 nM MimA2 induced a greater percentage of IFN-γ–secreting cells than did 1–10 μM YQLENYC. This was due, in part, to the fact that MimA2 is an unusually potent superagonist for AI4, showing activity even at subnanomolar concentrations (11). However, we suspected that the comparatively weak response to YQLENYC might also be due to the predicted poor binding of YQLENYC to H-2D^b, because the peptide lacks a P9 anchor residue. To test this, we designed two extended versions of YQLENYC (i.e., YQLENYCAL or YQLENYCGL) because L is one of the preferred P9 anchor residues for H-2D^b, and we wanted the P8 “spacer” residue to have a small or no side chain so it would be unlikely to interfere with T cell recognition. It should be noted that these two extended peptides do not exist naturally. However, we hypothesized that their potential ability to bind H-2D^b with a greater affinity than the InsA_14–20 sequence would allow for an improved ability to detect CD8 T cells recognizing the naturally occurring 7-mer epitope. As predicted, the two extended peptides demonstrated improved binding to H-2D^b compared with InsA_14–20 (Fig. 4B). When the AI4 CTLs generated in vitro with YQLENYC were tested by ELISPOT for their ability to release IFN-γ in response to peptide-pulsed DCs, spots were observed in the case of YQLENYC but increased 4–5-fold when the extended peptides were used (Fig. 4C), supporting their use as high-binding mimotopes of the natural insulin 7-mer. The in vitro–generated AI4 CTLs also exhibited cytotoxic activity against RMA-S target cells pulsed with YQLENYC, the extended peptides, or MimA2 (Fig. 4D, left panel). As expected for a peptide that binds weakly to MHC, if the epitope-pulsed targets were washed to remove free Ag before the CTLs were added, YQLENYC-pulsed targets were no longer recognized (Fig. 4D, right panel). Nevertheless, from these collective data we can conclude that YQLENYC, despite being a 7-mer that binds poorly to H-2D^b, is nonetheless able to induce CTL activity.

In vitro refolding experiments also confirmed the ability of the YQLENYC 7-mer to form a complex with H-2D^b (Fig. 5A), despite the lack of binding observed in the apparently less sensitive cell-based MHC-stabilization assay (Fig. 4B). When bacterially expressed H-2D^b H chain and murine β2m were refolded in the presence of YQLENY or YQLENYC, stable complexes were formed in solution (Fig. 5A). Size-exclusion chromatography showed that the complexes had a similar elution profile as that of a MimA2/H-2D^b complex, although the MimA2 complex was formed at a far greater efficiency (Fig. 5A), as a result of the superior binding of MimA2 to H-2D^b (Fig. 4B). Even after 4 wk of storage at 4°C, the refolded YQLENY and YQLENYC complexes maintained their stability in solution, whereas a similarly refolded metastable complex of the H-2D^b H chain and β2m, without any peptide, rapidly degraded (Fig. 5B). The successful refolding of the 6-mer and 7-mer peptides provides further evidence that these atypical peptides are nonetheless able to form complexes with H-2D^b that are recognizable by the AI4 TCR.

We previously used MimA2/H-2D^b tetramers to enumerate AI4-like T cells in the islet infiltrates of NOD mice (4). This earlier work suggested that AI4-like T cells were often absent from these infiltrates, because a tetramer-positive population was observed in only 25% of mice. However, having identified the unusual YQLENYC 7-mer as a natural β cell peptide mimicked by MimA2, and observing the differences in their sequences and lengths (Fig. 2E), we hypothesized that the use of MimA2 recognition likely underestimated the presence of InsA_14–20-reactive T cells. To test this, we used IFN-γ ELISPOT to screen islet-infiltrating T cells from NOD mice ranging from 11 (Fig. 6A) to 15 wk of age (Fig. 6B) for reactivity to MimA2, YQLENYC, or the extended peptides YQLENYCAL and YQLENYCGL. Multiple peptides were examined, because a single mimotope (or extended peptide) may not be recognized by all T cells specific for the corresponding natural Ag (25). Of all mice examined, 86% (12/14) showed reactivity to at least one of these four peptides. Consistent with our hypothesis, only 25% of the positive mice showed reactivity to MimA2, suggesting that, although it is recognized by the highly promiscuous AI4 TCR, it is not recognized by all T cells specific for YQLENYC. As noted earlier, the extended YQLENYCAL and YQLENYCGL peptides cannot represent naturally occurring insulin-derived antigenic epitopes. Thus, the InsA_14–20 7-mer does not simply represent a weaker CD8 T cell stimulatory variant of a longer naturally occurring insulin-derived antigenic peptide. However, the possibility remained that YQLENYC and its extended versions were recognized by islet-derived CD8 T cells merely because they were acting as mimotopes of the previously identified epitope FNL9 derived from DMK (4). Thus, islet-derived T cells from some of the mice also were tested for reactivity to FNL9 (Fig. 6A). Only two mice responded to FNL9 (Mouse 3 and Mouse 8 from the 11-wk-old group), and most of the mice that responded to YQLENYC or the extended peptides did not respond to FNL9. These results demonstrate that YQLENYC is a true T cell epitope, rather than a mimic of FNL9. For comparison, reactivity to the previously identified H-2K^d–restricted epitopes IGRP_206–214 and InsB_15–23 was monitored in these same mice using the superagonist mimotope peptides NRP-V7 and InsI9, respectively (Fig. 6C, 6D). Of the 14 mice examined, 79% were positive for reactivity to NRP-V7, whereas 14% showed InsI9-reactive cells. These results establish CD8 T cells specific for the atypical H-2D^b–binding peptide YQLENYC as a population present in the pathological insulitic lesions of nearly all NOD mice, regardless of age.

mTECs express otherwise tissue-restricted Ags (27), allowing for negative selection of CD8 T cells specific for these self-proteins either by direct display to T cells by mTECs or through cross-presentation by DCs (28). Purified mTECs from NOD mice express proinsulin mRNA transcripts (Fig. 7). However, InsA_14–20 and the 6-mer InsA_14–19, also recognized by AI4, exhibit poor binding to H-2D^b, likely contributing to the failure of these epitopes to be presented vigorously enough in the thymus of NOD mice to induce the negative selection of AI4-like T cells. However, this deficiency is overridden when thymic APCs transgenically express very high levels of insulin. NOD mTECs did not express mRNA transcripts encoding the previously identified (4) AI4 target Ag DMK (Fig. 7), suggesting that this Ag also cannot contribute to the negative selection of AI4-like T cells. Taken together, these results help to explain our finding of AI4-like T cells in the islet infiltrates of nearly all NOD mice examined.

The finding of MimA2-responsive cells in 21% of all mice (Fig. 6) is consistent with our earlier tetramer study (4) and confirms that T cells sharing the promiscuity of AI4 are present in the islet infiltrates of NOD mice. To obtain a structural explanation for this promiscuity, we determined the crystal structure of H-2D^b/MimA2 at 1.98-Å resolution (PDB ID: 3WS6; Fig. 8A, Table I). Based on the sequence alignment of the natural and mimotope peptides recognized by AI4 (Fig. 2E), we hypothesized that the N at P5 of the peptide would serve as a canonical H-2D^b anchor residue and that the conserved E and Y at P4 and P6, respectively, would be solvent exposed and available for contact with the TCR. Indeed, this is what we observed. Overall, the structural features of MimA2/H-2D^b were consistent with previously reported H-2D^b complexes (29, 30), with N at P5 and L at P9 playing major roles as anchor residues and with P1 (Y), P4 (E), P6 (Y), P7 (L), and P8 (E) being solvent exposed and, thus, potentially contributing to TCR recognition (Fig. 8B, 8C). Of the solvent-exposed residues of MimA2, P4 and P6 exhibited the greatest available surface area (Table II). Taken together with a consideration of the known peptide ligands for AI4 (Fig. 2E), this observation suggests that the AI4 TCR primarily forms important contacts with P4 and P6 of the peptide, while tolerating a variety of residues at the other peptide positions, and it provides a structural explanation for the promiscuity of AI4-like T cells.

Although lacking a P9 anchor residue, the 7-mer peptide YQLENYC was able to serve as a ligand for AI4 T cells (Fig. 2F) and to form a stable complex with H-2D^b in solution (Fig. 5). To identify unique structural features of the interaction between YQLENYC and H-2D^b, we determined the crystal structure of H-2D^b/YQLENYCGL to 2.4-Å resolution (PDB ID: 3WS3; Table I) and compared it with that of H-2D^b/MimA2. The overall structural features of the two complexes were very similar (Fig. 8A, 8B). Structural alignment of the two peptides demonstrated a root-mean-square deviation of 0.207 Å for the nine equivalent Cα atoms, suggesting that the overall conformation of the two peptides was almost the same. The orientations of the side chains of each residue of the two peptides were also comparable (Fig. 8B). In both cases, N at P5 is buried in a polar pocket formed by E9, Q70, Q97, and Y156 of the H chain. The side chain of N at P5 forms hydrogen bonds with Q70 and Q97. L at P9 is deeply buried in a hydrophobic pocket defined by the H chain residues F116, I124, T143, and W147. In addition to these two conserved anchor residues, calculation of available and buried surface areas of each residue of MimA2 and YQLENYCGL revealed that the residue at P3 is also important for MHC binding (Fig. 8C, Table II). The hydrophobic residue at P3 (L in YQLENYCGL and I in MimA2) is deeply buried at the interface, which helps to stabilize the complexes. Importantly, and unique to YQLENYCGL, the Q at P2 is deeply buried in a polar pocket defined by the H chain residues E9, Y22, S24, and Y45 and forms hydrogen bonds with E9 and Y22 (Fig. 8D). In contrast, the nonpolar A at P2 of MimA2 cannot make such contacts and is not involved in anchoring the peptide to the MHC H chain. The presence of the auxiliary anchor at P2 of YQLENY and YQLENYC helps to explain how these peptides, despite lacking a C-terminal anchor residue, can form stable complexes with H-2D^b (Fig. 5) and spontaneously elicit AI4-like T cell responses in nearly all NOD mice (Fig. 6).

Like AI4, other autoreactive T cells were shown to exhibit promiscuous Ag recognition (31–34). It seems likely that a promiscuous TCR capable of recognizing multiple Ags will often be characterized by low-avidity interactions with any single ligand that it can engage, so as to avoid thymic negative selection. This is certainly true for AI4 engagement of the InsA_14–20 epitope, given the low-affinity binding of this peptide to H-2D^b. If expressing a TCR capable of recognizing a diverse array of Ags, but that interacts with each in a low-avidity manner, is a common feature of autoreactive T cells, then an important factor in determining the extent to which such effectors develop and become functionally active may be the relative range of ligands that they can engage when originally undergoing selection versus when encountering their target tissue. A relative paucity of ligands in the thymus may allow autoreactive T cells expressing a promiscuous TCR to escape negative selection. Expression of a larger array of ligands by the target tissue may then allow for pathogenic activation of autoreactive T cells expressing a promiscuous TCR. DMK (4) and insulin are both expressed by β cells, which may be a factor in the high pathogenic effector capacity of AI4 T cells. However, NOD thymic epithelial cells expressed proinsulin, but not DMK, mRNA transcripts, and the lack of the latter Ag at this site could contribute to impaired negative selection of AI4-like CD8 T cells.

Our work provides a structural explanation for the ability of AI4 to recognize multiple peptides in the context of H-2D^b. Of the solvent-exposed peptide residues in MimA2/H-2D^b and YQLENYCGL/H-2D^b, P4 (E) and P6 (Y) exhibit the greatest surface area accessible for TCR contact. These residues are conserved among all of the H-2D^b–binding peptides recognized by the AI4 TCR, further supporting the notion that its promiscuity is explained by a requirement for specific residues at only these two positions of the peptide. This is also consistent with our previous work in which we conducted an alanine-replacement scan of Mim in an attempt to identify the peptide residues most likely involved in recognition by AI4, and we found that alanine substitutions at P4 and P6 were the most deleterious (11). The role of the partially accessible cysteine residue at P7 is less clear. However, because YQLENYC is a stronger agonist than YQLENY (Fig. 2F), it seems likely that the C at P7 contributes to T cell recognition and/or stabilization of the peptide/MHC complex, at least in the context of the natural (i.e., unextended) peptide. In support of this idea, a screen of an H-2D^b–biased positional scanning 9-mer peptide library identified the amino acid residues preferred by AI4 at position 7 as C, I, L, and M, all having a similar (nonpolar) chemical character (35).

The peptide length preference of a given class I MHC molecule is imposed by the relative depth and contour features of its peptide-binding cleft. It was reported that, unlike H-2K^b, H-2D^b cannot bind to a peptide shorter than nine residues because of the presence of a hydrophobic ridge between the α1 and α2 helices that runs nearly perpendicular to the long axis of the peptide-binding cleft (30, 36). This ridge causes a compensatory arch in the backbone of the peptide, requiring at least nine residues for the peptide to fill the cleft. In light of this structural constraint of the peptide-binding groove of H-2D^b, our identification of the 6-mer and 7-mer peptides InsA_14–19 and InsA_14–20 as ligands for AI4-like T cells presents a unique scenario. Although it was shown that the pentapeptide NYPAL (C-terminal part of the Sendai virus nucleoprotein epitope, FAPGNYPAL) can stabilize H-2D^b (37), to our knowledge, this is the first report that a 6-mer or 7-mer peptide can bind sufficiently well to H-2D^b to be recognized by T cells and mediate their proliferation and differentiation into CTLs. Our crystallographic analysis of YQLENYCGL/H-2D^b revealed a heretofore unrecognized interaction between the P2 side chain (Q) of the peptide and residues of the H-2D^b H chain. This interaction likely helps to compensate for the absence of the C-terminal anchor residue (P9).

Although binding of YQLENYC to H-2D^b could not be detected in a standard flow cytometry–based MHC-stabilization assay using RMA-S cells (Fig. 4B), we present several lines of evidence that the peptide can indeed bind sufficiently well to stimulate a T cell response. First, YQLENYC is capable of stimulating the proliferation of AI4 T cells (Figs. 2F, 3A), and this response is abrogated by addition of a blocking Ab to H-2D^b (Fig. 3B). Second, YQLENYC is capable of inducing the differentiation of AI4 T cells into CTLs that can release IFN-γ upon restimulation with YQLENYC, as measured by intracellular cytokine staining (Fig. 4A) and ELISPOT (Fig. 4C). These YQLENYC-generated CTLs also exhibit cytotoxic activity (Fig. 4D). Finally, Fig. 5A and 5B demonstrate that YQLENYC can be refolded with bacterially expressed H-2D^b H chain and β2m. Taken together, these results verify our conclusion that, although the 7-mer peptide binds weakly to H-2D^b, compensatory mechanisms allow it to nonetheless bind sufficiently well to stimulate AI4 T cells.

In addition to InsA_14–19 (YQLENY) and InsA_14–20 (YQLENYC) identified in this study, peptides overlapping with these epitopes were identified as targets of islet-reactive CD8 T cells. These include the H-2D^b–binding InsA_11–19 (CSLYQLENY) and InsA_13–21 (LYQLENYCN), identified using islet infiltrates from NOD.INS2 KO mice (38), as well as the HLA-A2–binding InsA_12–20 (SLYQLENYC), recognized in T1D patients (39). When C57BL/6 mice transgenically express CD80 on their β cells, immunization with a preproinsulin-encoding plasmid induces autoimmune diabetes, and the diabetogenic CD8 T cells recognize InsA_12–21 in an H-2K^b–restricted manner (40). Taken together, these findings suggest that the C-terminal half of the A chain might have particularly good access to class I MHC Ag-processing and -presentation pathways. It might also reflect the need for little or no trimming to achieve the appropriate C-terminal end for these epitopes. The InsA₂₀ cysteine participates in an interchain disulfide bond in the context of the intact protein, which would likely need to be resolved for at least some of the cysteine-containing peptides to be recognized by T cells. We know that this is the case for YQLENYC, because the cysteinylated version of this peptide was not recognized by AI4 (data not shown). Identification of the processing pathways required to produce YQLENY and YQLENYC in the β cells and in cross-presenting APCs will be the objective of future studies.

Although the evidence that InsA_14–20 occurs naturally is indirect, it is nonetheless quite strong. We know from the results shown in Fig. 1 that AI4 T cells recognize a peptide derived from insulin in vivo, because AI4 cells are strongly negatively selected in NOD.PI mice. Our insulin library screen revealed only one area of the insulin molecule that could stimulate AI4, and reactivity was ultimately mapped to YQLENYC. The last residue of YQLENYC is the penultimate residue of the insulin protein, with N being the terminal residue. Thus, a 9-mer version of YQLENYC cannot exist, and the 8-mer YQLENYCN is not recognized by AI4 (Fig. 2D). Finally, nearly all NOD mice recognize YQLENYC or its extended peptide variants (Fig. 6A, 6B), suggesting that YQLENYC is not merely acting as a mimotope of another Ag recognized by AI4. Importantly, T cells from the islet infiltrates of two NOD mice (Mouse 6 in Fig. 6A and Mouse 3 in Fig. 6B) recognized YQLENYC better than MimA2 and the extended versions of InsA_14–20, again supporting the conclusion that the natural Ag being recognized by the T cells is indeed YQLENYC. Although not as definitive as eluting the peptide from purified H-2D^b molecules (which would likely be difficult, because a weakly binding peptide is likely to be lost during the MHC-purification steps), our results, when taken together, strongly support the idea that InsA_14–20 occurs naturally.

In addition to identifying a new prevalent population of islet-reactive CD8 T cells in NOD mice, our work has important implications for autoimmune diseases in general. It provides a structural explanation for the promiscuity of autoreactive T cell populations, such as AI4-like effectors, and a model for how they may escape thymic negative selection and yet have high pathogenic activity in the periphery. Of great importance, it provides proof that unusually short MHC class I–binding peptides can use compensatory mechanisms to allow sufficient MHC binding to support destructive autoreactive T cell responses. Although, to our knowledge, this is the first demonstration of an undersized epitope for a prevalent and pathogenic autoreactive CD8 T cell population, such peptides are likely to be of particular relevance as targets for autoreactive T cells, because their low affinity for MHC would permit cognate T cells to escape negative selection. Our successful use of extended peptides to detect autoreactive T cells specific for such atypical epitopes suggests a strategy for the future discovery of unusually short peptide ligands for CD8 T cells in murine and human autoimmune diseases.

We thank the staff of the X29 beamline at the National Synchrotron Light Source and the 31-ID-D beamline (Lilly Research Laboratories Collaborative Access Team) operated by Eli Lilly and Company for help with data collection.

The authors have no financial conflicts of interest.