Constitutive and Inflammatory Immunopeptidome of Pancreatic β-Cells

NOD mice were purchased and housed in the Bio21 Institute animal facility. Studies were carried out in accordance with accepted standards of humane animal care and were approved by the animal ethics committee, University of Melbourne. Tissues were snap-frozen in liquid nitrogen and stored at −80°C.

The P815 (H-2^d) mastocytoma and EL4 (H-2^b) thymoma were maintained in RPMI 1640 (Invitrogen) medium supplemented with 10% FCS, 2 mmol/L glutamine, 50 IU/mL penicillin, and 50 µg/mL streptomycin. The NIT-1 insulinoma cell line (9) was maintained in low-glucose Dulbecco’s modified Eagle’s medium supplemented as described above. Expanded NIT-1 cells were washed in PBS, harvested by scraping, and snap-frozen. For cytokine treatment, cells were cultured with 100 IU/mL IFN-γ (eBioscience) for 24–72 h. For peptide pulsing, 5 × 10⁷ cells were incubated with 1 μmol/L synthetic peptide in serum-free RPMI 1640 medium for 60 min at 37°C and washed with PBS.

Cells were incubated with fluorescein isothiocyanate–conjugated anti-K^d (SF1.1.10; BD Biosciences) antibody for 30 min on ice. Viable cells (determined by propidium iodide exclusion) were analyzed using a CyAn ADP flow cytometer (Beckman Coulter) in conjunction with FloJo software.

IGRP_206–214 (VY*LKTNVFL; K^d restricted) and Janus kinase (JAK)-1_355–363 (SY*FPEITHI; K^d restricted) absolute quantitation (AQUA) peptides were synthesized to incorporate an isotopically labeled amino acid (*) using standard solid-phase fluorenylmethyloxycarbonyl chemistry and purified to >95% by reversed-phase high-performance liquid chromatography (RP-HPLC). The identity of the peptides was confirmed by mass spectrometry and the concentration of peptides after dissolution in water determined by amino acid analysis as previously described (6).

Cell pellets or tissues were ground in a Retsch Mixer Mill MM 400 under cryogenic conditions; resuspended in 0.5% IGEPAL, 50 mmol/L Tris (pH 8), 150 mmol/L NaCl, and protease inhibitors (Complete Protease Inhibitor Cocktail Tablet; Roche Molecular Biochemicals) at a density of 5 × 10⁷ cells/mL; and incubated for 1 h at 4°C. Lysates were cleared by ultracentrifugation (200,000g) and MHC-peptide complexes immunoaffinity purified using solid-phase bound monoclonal antibodies SF1.1.10 (anti-K^d) and 28.8.6s (anti-D^bK^b) as previously described (10). Bound complexes were eluted with 10% acetic acid. The mixture of peptides and class I heavy-chain and β-2 microglobulin was fractionated on a 4.6-mm internal diameter × 50-mm long reversed-phase C18 HPLC column (Chromolith Speed Rod; Merck) using an ÄKTAmicro HPLC (GE Healthcare) as previously described (6). Fractions were collected, vacuum concentrated, and diluted in 0.1% formic acid.

Concentrated peptide fractions were analyzed by LC-MS/MS using an Eksigent NanoUltra HPLC and an AB SCIEX 5600 TripleTOF mass spectrometer as described in the Supplementary Data online. Data were analyzed with ProteinPilot software (11) and peptide identities determined subject to strict bioinformatic criteria that included the use of a decoy database to calculate the false discovery rate. A false discovery rate cutoff of 10% was applied, and the filtered dataset was further analyzed manually to exclude redundant peptides and known contaminants as described in the Supplementary Data online. Cellular localization of source proteins was analyzed using Software Tool for Researching Annotations of Proteins (12) and pathway analysis performed though the use of Ingenuity Pathway Analysis (IPA) (Ingenuity Systems; www.ingenuity.com).

To determine the validity of the LC-MS/MS dataset generated from the TripleTOF 5600 experiments, 97 peptides from the K^d dataset and 96 peptides from the D^b dataset were selected for MRM detection. For the D^b set, two known subdominant peptides from IGRP were also included (IGRP_225–233, LRLFGIDLL, and IGRP_241–249, KWCANPDWI). MRM transitions for these peptides were designed in silico using Skyline v1.1 (University of Washington, Seattle, WA). Peptide eluates were prepared from 1 × 10⁸ IFN-γ–treated NIT-1 cells in duplicate and analyzed using a Tempo nanoLC-1Dplus in combination with a cHiPLC-nanoflex system (Eksigent) coupled to a 5500 QTRAP mass spectrometer (AB SCIEX) using experimental conditions described in the Supplementary Data online.

Transitions for MRM experiments were designed after inspection of experimental tandem mass spectrometry data. Three transitions were designed per peptide and validated through a series of experiments to investigate specificity, reproducibility, and background (Supplementary Figs. 1–4). Peptide fractions were concentrated and analyzed with a Tempo nanoLC-1Dplus system coupled to a 5500 QTRAP mass spectrometer as described in the Supplementary Data online. The amount of JAK-1_355–363 and IGRP_206–214 peptide present in each HPLC fraction was quantitated by examining the area under each MRM transition peak relative to an internal standard AQUA peptide as previously described (6). The addition of the AQUA peptide occurred immediately after immunoaffinity chromatography, since this is the first practical point for addition. This provides accurate quantitation of the peptide isolated from the immunoaffinity-purified MHC complexes and allows an estimation of the number of complexes per cell (6).

For establishment of a database of peptides presented by β-cells, K^d and D^b molecules were purified from cultured NIT-1 cells grown under basal conditions (3 × 10⁹ cells) or after 48 h IFN-γ treatment (1.5 × 10⁹ cells). For provision of a comparative dataset, peptides were also purified from pooled thymus and spleen from 12-week-old female NOD mice (n = 10). MHC-peptide complexes were isolated from detergent lysates using K^d- or D^b-specific antibodies. Peptides were separated from class I heavy-chain and β-2 microglobulin by RP-HPLC and identified by LC-MS/MS. The complete dataset for IFN-γ–treated NIT-1 cells for both K^d (544 peptides from 484 source proteins) and D^b (339 peptides from 297 source proteins) are shown in Supplementary Tables 1 and 2, respectively, and their presence in untreated NIT-1, spleen, or thymus indicated. The sequences of a further 221 peptides (51 D^b, 170 K^d) identified in untreated NIT-1 cells (not in the thymus, spleen, or cytokine-treated cells by LC-MS/MS) are shown in Supplementary Table 3. These 221 peptides may be present in low abundance in the treated cells; however, manual inspection using extracted ion chromatograms generally did not provide definitive confirmation in the treated data. The predicted peptide-binding affinity was determined using the SYFPEITHI algorithm (13), and scores are recorded for K^d 9-mers and D^b 9-mers and 10-mers. Of those peptides sequenced from the spleen, ∼80% were also detected in the thymus for both K^d and D^b repertoires. The experimental data generated from IFN-γ–treated NIT-1 cells have been deposited with Tranche (https://proteomecommons.org/tranch/) and are Minimum Information About a Proteomics Experiment compliant (14).

No changes were noted in the amino acid preferences within peptides purified from cells (with or without IFN-γ) or tissues. The distribution of predicted peptide-binding scores remained unaltered between the untreated and IFN-γ–treated samples for K^d (Fig. 1) and D^b (not shown). Motif analysis was therefore performed on the entire elution set obtained from NIT-1, NIT-1 plus IFN-γ, spleen, and thymus. A total of 1,146 and 910 peptide sequences for K^d and D^b, respectively, were used to refine the motifs. Peptides bound to K^d were predominantly 9 amino acids in length (80%) with a small number of 10- or 11-mer peptides (Fig. 2A). Consistent with the published motif (15–17), position 2 was dominated by tyrosine and, in a small number of sequences, phenylalanine. The COOH-terminal position was occupied by Ile, Leu, or Val (Fig. 2B). Met and Phe were also tolerated at the COOH terminus; however, this preference was more apparent in peptides of 10 or 11 amino acids. Although the dominant anchor residues (P2 and PΩ) remained consistent, the binding motif for peptides of 9, 10, or 11 amino acids showed differential amino acid preferences in the COOH-terminal half of the peptide (Fig. 3). In particular, a strong preference for Pro at P7 and Gly at P8 in the 11-mer peptides was noted. Pro was also present at a higher frequency (>10%) at P6 in the 11-mer dataset. Ser and Thr remained moderate to strong anchors at all peptide lengths; however, the position of these residues shifted from P7 to P8 in 10-mer peptides and to P9 in the 11-mers.

While the majority of D^b peptides were 9 amino acids in length (46%), more peptides of 10 and 11 amino acids in length (150 and 152 peptides, respectively) were isolated from D^b than K^d (Fig. 2C). The number of peptides longer than 9 amino acids (∼53% of total sequences, 63% of these being 10- or 11-mer peptides) exceeds that previously reported for D^b (18). The dominant anchors of Asn at P5 and Leu at the COOH terminus were present in 9-, 10-, and 11-mer peptide sets (Fig. 2D) as previously described (18). A strong preference for Ala at P2 was seen in the 9-mers, which became dominant in the 10- and 11-mer peptides (Fig. 3). The appearance of Pro at P6 was noted in 10- and 11-mer peptides, and Pro was also a moderate P7 preference in the 10-mer data set. Gly demonstrated a change in frequency, becoming a moderate anchor at P6 in 10-mer peptides and P8 in 11-mer peptides. Of final note, the preference for glutamic acid or aspartic acid shifted from P7 in the short peptide set to P8 in the 10-mers and P9 in the 11-mer dataset.

Peptides were derived largely from ubiquitous proteins, expressed in NIT-1, spleen, and thymus (Supplementary Tables 1 and 2). As expected, the majority of sequences were derived from either nuclear or cytoplasmic proteins (Fig. 4). No change in the localization of source proteins was noted following IFN-γ treatment; however, peptides from IFN-γ–inducible proteins were found in both datasets. Of particular note, peptides from the LMP2 and LMP7 subunits of the immunoproteasome- and the proteasome-associated protein ECM29 were found in IFN-γ–treated cells. This protein binds to the 26S proteasome and has been proposed to couple proteasomes to secretory compartments (19,20). IPA of source proteins within the larger K^d dataset showed a change in the top canonical pathways from caveolar-mediated endocytosis and insulin receptor signaling to ephrin receptor and AMP-activated protein kinase signaling. Phosphatidylinositol 3-kinase/AKT signaling remained a top canonical pathway in both IFNγ–treated and untreated datasets.

Consistent with previous reports, peptides from neuroendocrine or neuronal origin (e.g., α-internexin, neurofilament medium polypeptide, reticulon 4, γ-aminobutyric acid, and glutamate receptors) were identified in both datasets. A number of peptide sequences generated from proteins involved in secretory processes were also identified in untreated and treated NIT-1 cells. Within the IFN-γ–treated dataset, a single peptide derived from the known autoantigen IGRP (D^b-restricted SGVLIIHHL) and a series of overlapping peptides from the potential autoantigen neuropeptide Y (D^b and K^d restricted) were identified (Supplementary Tables 1 and 2). In addition to developing diabetes, NOD mice develop a lupus-like syndrome. Two peptides from proteins associated with Sjögren Syndrome and systemic lupus erythematosus were identified within the D^b elution set, namely Sjögren Syndrome nuclear autoantigen 1 homolog (AALQNYNNEL) and calreticulin (EEESPGQAKDEL). A number of posttranslationally modified peptides were also identified, displaying pyroglutamate formation, deamidation, NH₂-terminal acetylation, and glutathionylation. Such modifications occurred in both untreated and treated datasets.

The robustness of peptide identification was examined using a cohort of peptides identified in the discovery experiments in β-cell lines or in NOD mouse primary tissue. Duplicate samples of NIT-1 cells were solubilized and H-2K^d and D^b peptides isolated. With use of in silico–generated MRMs, 77/97 H-2K^d– and 78/96 H-2D^b–bound peptides were confirmed (representing a minimal 79% validation rate [Supplementary Table 4]). It should be noted that less material was used for the MRM analysis compared with the global LC-MS/MS discovery experiments (15-fold less). The high confirmation rate without the requirement for synthetic peptides provides a high-throughput means of validating initial discovery datasets.

In these global LC-MS/MS experiments, we did not detect any of the immunodominant peptides previously reported in NOD mice. To determine whether this was due to the sensitivity of the approach, we used MRM for the specific detection of the immunodominant K^d-restricted IGRP_206–214 peptide. MRM transitions were established for IGRP_206–214 and the endogenously presented peptide JAK-1_355–363 (Fig. 5A). JAK-1_355–363 (derived from the interferon signaling molecule JAK-1) was selected, as it was readily identified in NIT-1, IFN-γ–treated NIT-1, thymus, and spleen samples and is a high-affinity K^d ligand. For confirmation of the specificity of these peptides for K^d, P815 (K^dD^d) and EL4 (K^bD^b) cells were incubated with a mix of IGRP_206–214 and JAK-1_355–363 AQUA peptides, class I complexes purified, and isolated peptides subjected to MRM analysis. Both IGRP_206–214 and JAK-1_355–363 were detected in the K^d eluate of peptide-pulsed P815 cells, as shown in Fig. 5B. Although P815 cells endogenously present JAK-1_355–363, the use of the AQUA version of the peptide to pulse these cells allows a distinction to be made between endogenous presented peptide and exogenous loading of the mass-shifted AQUA peptide by liquid chromatography–mass spectrometry. Neither peptide was detected in fractions affinity purified from peptide-pulsed EL4 cells (K^b or D^b) or the D^d fractions of P815 cells (Fig. 5C).

We next used this approach to detect peptides presented by NIT-1 cells. NIT-1 cells were treated with IFN-γ and MHC-peptide complexes isolated at 0, 24, 48, and 72 h posttreatment. The increase in K^d expression following cytokine treatment is shown in Fig. 6A. Both JAK-1_355–363 and IGRP_206–214 were detected from NIT-1 cells treated with IFN-γ over 24–72 h (Fig. 6B). A striking difference was observed between the number of MHC/peptide complexes for the two peptides isolated from the same sample. JAK-1_355–363 was quantified at ∼15,000 copies per cell after 72 h of IFN-γ treatment; however, IGRP_206–214 did not exceed >25 copies per cell. In the absence of IFN-γ treatment, JAK-1_355–363 but not IGRP_206–214 could be detected. To determine whether the inability to detect IGRP_206–214 in untreated cells was due to the low level of class I expression, a second larger sample of untreated cells was prepared (3 × 10⁹ cells). JAK-1_355–363 was detected at ∼2,000 copies per cell (Fig. 6C), consistent with the 0-h time point in the IFN-γ dataset. IGRP_206–214 was also detected in this sample at ∼1 copy/cell.

We have established a database of MHC class I peptides endogenously processed and presented by the NIT-1 β-cell line and in tissues of NOD mice. This is the largest dataset of H-2K^d and D^b peptides reported, allowing refinement of the binding motifs for these molecules. Because these peptide sequences are derived from high confidence mass spectrometry identifications of naturally processed and presented peptide ligands, they do not share the inherent bias in amino acid motif definition seen in pool Edman sequencing (21) because of the dominance of highly abundant peptides. In general, the peptide motifs observed in this study were in good agreement with published motifs (15–18) and were further developed to include amino acids present with dominant, strong, and moderate frequencies. Moreover, discrete motifs were also apparent for peptides longer than the canonical 8–9 amino acids. Such long peptides have typically been difficult or impossible to predict because of the lack of a significant number of bona fide longer ligands. In particular, bioinformatics tools have not performed well for D^b in comparison with other alleles (18), attributed to the low number of long natural ligands used to train predictive algorithms. In our study, we identified >400 naturally processed peptides of 10 or 11 amino acids in length, with the selection of such long peptides more frequent in the D^b compared with the K^d repertoire. For both class I molecules, Gly and Pro became more common in peptides of 10 and 11 amino acids, reflecting a requirement for flexibility and kinking or bulging in the peptide backbone of longer peptides (22). For longer peptides bound to H-2D^b, Ala became a dominant motif at P2. Identification of these long, naturally processed peptides and refinement of the current motifs should enhance the success of bioinformatic predictions, providing new opportunities to predict autoantigen-derived epitopes in the NOD mouse.

Treatment of NIT-1 cells with IFN-γ did not significantly change the length of peptides presented by either K^d or D^b or the distribution of predicted binding affinities. This suggests that cytokine treatment does not bias toward high-affinity ligands. Recent studies examining the peptide repertoire of a number of human alleles have documented the presence of NH₂-terminally extended peptides within the endogenous peptide pool (23). Consistent with this observation, we have also identified a small number of nested NH₂-terminally extended peptides. Trimming of the NH₂terminus by the aminopeptidase ERAAP (endoplasmic reticulum aminopeptidase associated with antigen processing) is required for most class I peptides (24). The presence of NH₂-terminally extended peptides is of interest given the reported immunogenicity of such sequences produced in the absence of ERAAP through formation of novel MHC-peptide complexes (25). Although the precise COOH terminus of MHC class I peptides is thought to be generated primarily by the proteasome in the cytoplasm, we also found peptides containing COOH-terminal extensions within the datasets for both K^d and D^b. The identification of these peptides suggests that additional trimming of the COOH terminus may occur after translocation into the endoplasmic reticulum and MHC complex formation.

The majority of peptides sequenced were common to NIT-1, spleen, and thymus. Many of the β-cell–specific peptides were of neuronal origin, consistent with previous data on elution of class II peptides in NIT-1 cells overexpressing the class II transactivator (26). We identified a number of sequences from known autoantigens such as insulin, chromogranin A, and islet amyloid polypeptide. We also identified peptides from other secretory proteins including neuropeptide Y, secretogranin, and proSAAS. These included nested sets of peptides ranging in length from 9 to 23 amino acids, and a number of these were bound to both K^d and D^b (e.g., secretogranin-1, SGKEVKGEEKGENQNSKFEVRLL; secretogranin2, YLNQEQAEQGREHL; neuropeptide Y, RYYSALRHYINLITRQRYG; and chromogranin A, LEGEDDPDRSM). The presence of unusually long peptides bound to class I molecules isolated from β-cells highlights a potential nonconventional antigen-presentation role in NOD mice. Whether these represent cleavage intermediates that are further processed following cytokine treatment and whether peptides derived from these sequences are recognized by autoreactive T cells warrants further investigation given the number of autoantigenic epitopes already identified from secretory granules. Recently, islet amyloid polypeptide was identified as the target of the BDC T-cell clone 5.2.9 (27). The CD4⁺ restricted epitope (KCNTATCATQRLANFLVRSS) was mapped using synthetic peptides spanning the entire molecule. We have identified a peptide that overlaps the COOH terminus of this sequence in untreated NIT-1 cells, bound by K^d (LVRSSNNLGP [Supplementary Table 3]). The propensity of certain immunogenic regions of targeted proteins to yield a number of class I and class II restricted epitopes is well documented. We also sequenced a D^b-restricted peptide derived from IGRP in IFN-γ–treated NIT-1 cells (IGRP_7–15 SGVLIIHHL). This peptide has a moderate binding score of 17 using the SYFPEITHI algorithm. The immunogenicity of this epitope has previously been tested using synthetic peptides (28). IGRP_7–15 failed to elicit a response from islet-infiltrating T cells isolated from NOD mice of varying ages. Although our data demonstrate that IGRP_7–15 is naturally processed and presented by NIT-1 β-cells, it is likely that the levels of presentation are sufficient to induce tolerance.

The ability to detect a tolerogenic peptide from IGRP by LC-MS/MS and not the dominant IGRP_206–214 epitope led us to address the issue of epitope abundance using the sensitive approach of MRM. Using MRM, we were able to detect both the dominant IGRP_206–214 peptide and the subdominant epitope IGRP_225–233 (Supplementary Table 4). Moreover, this targeted method allows quantitation of peptide epitopes through the inclusion of an isotopically labeled standard. We found that IGRP_206–214 was presented at very low levels on NIT-1 cells grown under basal conditions (∼1 copy/cell). Conversely, JAK-1_355–363 was presented on the same cells at ∼2,000 copies/cell. Treatment of cells with IFN-γ increased the level of presentation of both IGRP_206–214 and JAK-1_355–363. Although the total number of IGRP_206–214 peptide MHC complexes remained extremely low (25 copies/cell) compared with JAK-1_355–363 (15,000 copies/cell), the fold increase in presentation of the two epitopes was greater for IGRP_206–214, and did not merely mirror the fold increase in class I expression (fivefold increase in K^d at 72 h). This underscores the complexities of epitope generation and for the first time highlights the disparity between presentation levels of a known autoantigen and a normal endogenous ligand.

We have used a discovery-based approach to characterize the repertoire of peptides presented by pancreatic β-cells in the presence and absence of an inflammatory signal. Global LC-MS/MS analysis has successfully identified peptides from known autoantigens; however, many of these may represent sequences that are tolerogenic and, indeed, a large number of these peptides were present in the thymus. The quantitation of IGRP_206–214 provides the first demonstration of low-level presentation of an immunodominant autoantigenic determinant in absolute terms. Moreover, the disparity in levels of presentation of IGRP_7–15 and the immunodominant IGRP_206–214 highlights differences in epitope liberation even within the same antigen. Such quantitative information can be used to guide discovery-based approaches; finding the threshold at which autoantigenic peptides are likely to be present will allow sequencing of peptides that may normally be masked by more abundant species. We propose that the work flows of LC-MS/MS and MRM analysis can be used in tandem to mine deeper into the immunopeptidome and that comparative analysis of datasets sequenced from cell lines and tissues in the presence and absence of inflammation can be used to guide selection of potential autoantigens. This may be particularly relevant to granule-associated proteins, the presentation of which clearly changed upon cytokine stimulation in this study. One of the most important applications of MRM analysis of MHC-peptide epitopes will lie in the ability to examine hundreds of peptides in a single analysis. This will not only allow changes in the presentation of known autoantigenic peptides to be determined but will also enable changes in the presentation of peptides generated from a single antigen to be assessed. This will have important implications for understanding epitope liberation and the contribution of epitope abundance to immunodominance hierarchies.

The authors acknowledge funding from Diabetes Australia Research Trust and infrastructure funding from the Australian Research Council. A.W.P. is supported by a National Health and Medical Research Council Australia Senior Research Fellowship.

No potential conflicts of interest relevant to this study were reported.

N.L.D. performed the majority of experiments, designed the study, and wrote the manuscript. C.T.T. designed and executed the MRM experiments. D.G.G., N.P.C., and P.T.I. provided technical help and expert guidance in mass spectrometry. A.W.P. designed the study, supervised experimental work, and wrote the manuscript. N.L.D. and A.W.P. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.