CD4⁺ T Cells From Individuals With Type 1 Diabetes Respond to a Novel Class of Deamidated Peptides Formed in Pancreatic Islets Free

Enzymatic deamidation is a posttranslational modification (PTM) mediated by transglutaminase (TGM) enzymes. The result is conversion of a glutamine (Gln) to a glutamic acid (Glu) residue, introducing a negative charge and a mass increase of 0.98 Da. The role of enzymatic deamidation is well established in celiac disease (1), where deamidation enhances the presentation of gliadin peptides by HLA molecules and induces robust T-cell activation (2). Type 1 diabetes is an autoimmune disease that shares susceptible HLA genes with celiac disease and for which a major hallmark is development of islet cell autoantibodies (3). Susceptible HLA class II molecules are thought to promote thymic selection of a potentially autoreactive T-cell repertoire (4). For HLA class I molecules, this hypothesis was borne out by an observed state of benign autoimmunity, i.e., the presence of islet-reactive CD8⁺ T cells at similar circulating frequencies in individuals with type 1 diabetes and HLA-matched control individuals (5,6). Likewise, the type 1 diabetes–associated DR4 haplotype is sufficient to select a CD4⁺ T-cell repertoire that includes high-affinity self-reactive T cells (7). The most common HLA class II haplotype in celiac disease is DR3/DQ2 (>90% of patients with celiac disease harbor at least one copy) (8). The second most predominant haplotype in celiac disease and most highly susceptible haplotype in type 1 diabetes is DR4/DQ8. Both DQ2 and DQ8 anchor negatively charged peptides tightly in their binding groove, with DQ2 preferring negative charges at position 4 (P4) or P6 and DQ8 at P1, P6, and P9 (4,9). HLA-DRB1*04:01, the HLA-DRB1*04 allotype with the highest odds ratio for type 1 diabetes development, accepts negatively charged amino acids at P4 (4). As deamidation induces a negative charge, this modification can increase peptide binding to these susceptible HLA class II haplotypes. A CD4⁺ T-cell line against deamidated insulin C-peptide was isolated from a patient with new-onset type 1 diabetes homozygous for HLA-DQ8, and similar immune responses could be observed in newly diagnosed individuals (10). Of note, a more recent research report has challenged this conclusion, showing no evidence that deamidation increased the immunogenicity of C-peptide (11). T cells specific for a deamidated GAD65 peptide were more frequent in the peripheral blood of individuals with type 1 diabetes compared with healthy control individuals (12), and T-cell receptors that recognized the deamidated peptide were shown to more readily escape thymic selection in HLA transgenic mice (13). The deamidated insulin peptide showed increased binding to HLA-DQ8, while the deamidated GAD65 peptide showed increased binding to HLA-DRB1*04:01 compared with their native forms (10,12).

In addition to enzymatic deamidation, nonenzymatic deamidation occurs, converting asparagine (Asn) to aspartic acid (Asp) (14) or isoAsp and Gln to Glu or isoGlu. Nonenzymatic conversion of Glu or isoGlu occurs through the formation of a six-member-ring glutarimide intermediate, which is less favored than the corresponding succinimide intermediate formed in Asp deamidation, making this reaction much slower (15). We previously showed that nonenzymatic Asn deamidation is significantly increased in cytokine-exposed murine MIN6 β-cells (15). To date, no published study has investigated the role of nonenzymatic Asn deamidation in type 1 diabetes. Given that Asn deamidation is increased in stressed β-cells and results in negative charges that can enhance peptide binding and presentation by type 1 diabetes–susceptible HLA molecules, we hypothesized that this modification may generate neoantigens in type 1 diabetes.

Although deamidation of insulin C-peptide was documented by mass spectrometry (MS) (16), deamidation of GAD65 and other putative PTM autoantigens has never been directly shown in the islets of Langerhans. Here, we exposed human islets of Langerhans (primary islets) to a mixture of inflammatory cytokines or to interferon-α (IFN-α), to mimic inflammation and viral infection, and performed proteomic analysis to identify enzymatically and nonenzymatically deamidated peptides. Tetramer-based assays were then used to show that T cells recognizing deamidated peptides were present in peripheral blood of individuals with type 1 diabetes.

Human islets were obtained from Pisa University with approval by its local ethics committee (#2615). Characteristics of human islet donors are listed in Supplementary Table 1. Islets were cultured in DMEM supplemented with 1% l-glutamine (Gibco), 10% FBS, and 100 units/mL penicillin-streptomycin and exposed to human interleukin 1β (IL-1β) (50 units/mL; R&D Systems), murine tumor necrosis factor-α (TNF-α) (1,000 units/mL; R&D Systems), and human IFN-γ (1,000 units/mL; PeproTech) or human IFN-α (2,000 units/mL; PBL Assay Science) for 24 or 72 h.

Heparinized blood was obtained from control individuals and individuals with type 1 diabetes. All procedures were approved by the ethics committee of UZ Leuven and the institutional review board at the Benaroya Research Institute. Subject attributes are summarized in Supplementary Tables 2 and 3. Peripheral blood mononuclear cells (PBMCs) were isolated using Lymphoprep and frozen in AIM V Medium (Gibco) with 10% DMSO.

RNA was extracted from primary islets using the RNeasy Micro Kit (QIAGEN), and cDNA was made using oligo-d(T) and Superscript II Reverse Transcriptase (Invitrogen). Quantitative RT-PCR was performed on a StepOnePlus RT-PCR System (Applied Biosystems, Santa Clara, CA). The relative fold gene expression was calculated using the delta-delta Ct method.

Human islets were incubated with propidium iodide (Invitrogen) and Hoechst 33342 (Invitrogen) for 15 min at 37°C. At least 10 islets were assessed for apoptosis by two independent researchers on a Nikon Eclipse TI microscope.

Cell Ranger (version 6.1.2) (17) was used for alignment (reference genome hg38) and filtering of raw sequencing reads retrieved from the Human Pancreas Analysis Program (HPAP) (https://hpap.pmacs.upenn.edu) (donor characteristics listed in Supplementary Table 4). Decontamination of ambient mRNA was performed for each sample using SoupX (version 1.6.1) (18), using genes associated with major cell types identified in the initial Seurat (version 4.1.1) (19) clustering. Single-cell gene expression profiles were imported into Seurat for quality control, retaining genes expressed in at least three cells expressing a minimum of 200 genes. Potential doublet cells were identified and removed using the R package scDbIFinder (version 3.16) (20). Additional filtering criteria included nFeature_RNA <9,000, mitochondrial genes <10%, and nCount_RNA <10,000, resulting in high-quality single cells for downstream analysis. Seurat’s SCTransform function was applied to measure differences in sequencing depth per cell and normalize counts by removing variation due to sequencing depth (number of unique molecular identifiers). The top 3,000 variable features were selected for principal component analysis and uniform manifold approximation and projection embedding. To assign cells to known cell types based on marker genes, we used the R package scSorter (version 0.0.2) (21). Subsequently, we extracted pancreatic endocrine cells (α-cells, β-cells, δ-cells, and γ-cells) and pancreatic nonendocrine cells (acinar cells, ductal cells, and endothelial cells) for downstream pseudobulk differential expression analysis. To evaluate the mRNA expression changes of TGM2 within endocrine and nonendocrine cells, we used a pseudobulk analysis strategy as previously described (22). Specifically, we extracted the counts matrix and aggregated the counts for each gene from individual cells to the sample level using the aggregate.Matrix function. This involved combining the counts of all endocrine (13,159 cells for individuals without diabetes and 22,040 cells for individuals with diabetes) or nonendocrine cells (12,417 cells for individuals without diabetes and 18,967 cells for individuals with diabetes) at the sample level. Subsequently, we imported the aggregated counts matrix into DESeq2 (23) for the analysis of differential gene expression across the specified conditions. Statistical significance was assessed by DESeq2, and P values were corrected for multiple testing using the Benjamini-Hochberg method. Finally, the normalized counts matrix by DESeq2 for TGM2 was selected for visualization using the R package ggplot2.

Functional enrichment analysis to identify enriched biological pathways associated with upregulated genes in endocrine cells of individuals with diabetes, focusing on the Reactome database, was conducted using g:Profiler (https://biit.cs.ut.ee/gprofiler/gost) (a list of genes differentially expressed is provided in Supplementary Table 5). The g:SCS algorithm was applied for multiple testing correction of P values obtained from a hypergeometric test–based pathway enrichment analysis. The number of genes contributing to the significance of the gene set enrichment are depicted on the x-axis of Fig. 1K.

Human islets were prepared for liquid chromatography-tandem MS (LC-MS/MS) as previously described (24). Digestion was performed with modified trypsin (trypsin:protein ratio of 1:10 w/w; Pierce) for 90 min to minimize artifactual deamidations (15). Solvent-evaporated peptide mixtures subjected to desalting with C18 ZipTip pipet tips (Millipore) were dissolved in 0.1% formic acid and 5% acetonitrile and run on a Q Exactive Orbitrap Mass Spectrometer (Thermo Fisher Scientific). Peptides were identified by Mascot (Matrix Science) using UniProt (Homo sapiens, 169,779 entries) as a database through Proteome Discoverer 2.2, incorporating Percolator for peptide validation. Oxidation (M), deamidation (N/Q), and citrullination (R), resulting in the same mass increase as deamidation, were included as variable modifications. Carbamidomethylation (C) was included as a fixed modification. Two missed cleavages were allowed, and peptide tolerance was set at 5 parts per million and MS/MS tolerance at 20 millimass units. Spectra of predicted deamidated peptides were manually checked as previously described (15).

Predicted binding of 27 peptides containing deamidated glutamine residues and 30 peptides containing deamidated Asn residues (as detected by LC-MS/MS) to HLA-DRB1*04:01 was performed using a previously described approach (25,26). Supplementary Table 6 shows predicted binding for all deamidated peptides. Peptides with a relative binding affinity score of ≥0.075 were predicted to bind. Selected deamidated peptides were synthesized (Synpeptide), and their binding was measured by incubating increasing concentrations of each peptide with 0.02 μmol/L biotinylated influenza hemagglutinin (HA) 306–318 in wells coated with recombinant DRB1*04:01 protein (purified from insect cell cultures as previously described [27]). Residual biotin-HA was detected after washing with europium-conjugated streptavidin on a Victor Nivo time-resolved fluorometer (PerkinElmer). IC₅₀ values were calculated as the concentration of peptide needed to displace 50% of the biotin-HA peptide. The sequences and IC₅₀ values of peptides with measurable binding are listed in Table 3. For peptides subsequently shown to be immunogenic, the corresponding wild-type peptides were synthesized, and their binding was measured using the same competition assay. Comparisons of IC₅₀ values for deamidated and wild-type peptides are listed in Supplementary Table 7.

Recombinant DRB1*04:01 monomers were purified from insect cell cultures as previously described (28) and loaded with peptide (0.2 mg/mL) in the presence of n-dodecyl-β-maltoside (0.2 mg/mL) and Pefabloc (1 mmol/L; Sigma) for 72 h at 37°C. Peptide-loaded monomers were conjugated into tetramers using streptavidin at a molar ratio of 8:1.

PBMCs (5 × 10⁶) were stimulated with pools of deamidated peptides and a negative control peptide (WGTLTDCVVMXDPQT, with X being a citrulline residue shown to elicit consistent negative T-cell responses) in 48-well plates for 14 days. Starting on day 7, human IL-2 (10 units/mL; Roche) was added every 2 days. Expanded cells were stained with tetramers for 1 h at 37°C and with CD3-BV510, CD4-APC, CD14-PerCPCy5.5, CD19-PerCPCy5.5, and CD25-FITC antibodies for 20 min at 4°C. Cells were acquired on an LSR II Flow Cytometer (BD Biosciences) and analyzed using FlowJo (Treestar Inc.). The gating strategy is shown in Supplementary Fig. 1A. Clones were isolated from cultures by sorting single tetramer-positive CD4⁺ T cells using FACSAria (BD Biosciences) and expanded in the presence of 10⁵ irradiated PBMCs and phyto-HA (2 μg/mL; Remel Inc.) in 96-well plates.

To assess preferential recognition of deamidated peptides, 10⁴ T cells were plated in triplicate in the presence of 10⁵ irradiated HLA-DRB1*04:01⁺ PBMCs as antigen-presenting cells (APCs) and stimulated with 10 μg/mL native or deamidated peptide. After incubation for 72 h at 37°C, cells were pulsed with medium containing ³H-thymidine (1 μCi/well). Incorporation was measured 18 h later with a scintillation counter. Data are represented as stimulation index, calculated by normalizing the ³H-thymidine incorporation of each clone with that of nonstimulated wells.

Cryopreserved PBMCs (20–30 × 10⁶) were thawed and treated with dasatinib (50 nmol/L; Cell Signaling Technology) for 10 min at 37°C and stained with 20 μg/mL phycoerythrin (PE)-labeled, PECF594-labeled, SPRD-labeled, PECy7-labeled, and APC-labeled tetramers for 90 min at room temperature. Cells were then washed and labeled with anti-PE and anti-APC magnetic beads for 20 min at 4°C. A preenriched fraction (2%) of the cells was reserved to determine the total cell number. Remaining cells were enriched using Miltenyi Biotec MS magnetic columns according to the manufacturer’s instructions. Both fractions were stained for 30 min at 4°C with CD4-BUV395, CD14-FITC, CD19-FITC, CD45RA-AF700, CXCR3-BV786, CCR7-BV650, CCR6-BV510, and CCR4-BUV605 antibodies. After washing, cells were labeled with Sytox Green (Thermo Fisher Scientific), run on an LSR II Flow Cytometer, and analyzed using FlowJo. The gating strategy is shown in Supplementary Fig. 1B.

Cell lines from islet-infiltrating T cells were recovered and expanded from Network for Pancreatic Organ Donors With Diabetes (nPOD) donor 6551 (Supplementary Table 8). For generation of lines, 150-μm live slices of pancreas were digested with 1 mg/mL collagenase-P (Sigma-Aldrich) in Hanks’ balanced salt solution with Mg²⁺/Ca²⁺ (Thermo Fisher Scientific) for 10 min in a 37°C water bath with vigorous shaking. Digestion was confirmed by visual inspection. Digested tissue was washed and placed in medium (AIM V supplemented with 2 mmol/L l-glutamine, 5 mmol/L HEPES, 100 units/mL penicillin, 100 μg/mL streptomycin, 0.1 mmol/L of each nonessential amino acid, and 1 mmol/L sodium pyruvate; all from Lonza, Walkersville, MD) with 5% heat-inactivated Human Male AB Serum (Access Biologicals, Vista, CA). Individual islets were handpicked and placed in culture with irradiated allogeneic PBMCs as feeders, a blocking anti-Fas antibody (1 μg/mL; Applied Biosystems); anti-PD1 antibody (1 μg/mL; BD Biosciences), and mifepristone (100 nmol/L; Invitrogen). After 12 days of culture, T-cell lines were collected, further expanded, and phenotyped by flow cytometry for T-cell subsets. Epstein-Barr virus–transformed B-lymphoblastoid cell lines were generated from autologous splenocytes as previously described (29). Responsiveness was assessed as previously described (30). Briefly, irradiated autologous B-lymphoblastoid cell lines were plated (20,000/well) after pulsing with peptides at 50 μg/mL, no peptide, or DMSO volume control and then cultured with T cells (75,000/well) and anti-CD28 (5 μg/mL) for costimulation. After 48 h of coculture, responses were detected by Cytometric Bead Array Human IFN-γ Enhanced Sensitivity Set (BD Biosciences), with a lower limit of 274 fg/mL sensitivity.

Data were analyzed using GraphPad Prism 9 (GraphPad, La Jolla, CA). Statistical tests included Wilcoxon test, paired t test, unpaired t test with Welch correction, and one-way ANOVA. Analysis of quantitative RT-PCR data was performed on delta CT values. All significant differences are reported.

MS proteomics data have been deposited to the ProteomeXchange Consortium via the Proteomics Identifications (PRIDE) (31) partner repository with the data set identifier PXD038758. Other data sets are available from the corresponding author upon reasonable request.

Primary islets from donors without diabetes were exposed to cytokines (IL-1β, TNF-α, and IFN-γ) to mimic inflammation or IFN-α to mimic antiviral response. IFN-α is crucial in innate and adaptive immunity and mediates HLA class I overexpression in human islets in type 1 diabetes (32,33). Inflammatory cytokines induced apoptosis after 72 h (24.90% vs. 5.84% in control islets) (Fig. 1A), and cell death was preceded by endoplasmic reticulum stress, indicated by increased C/EBP homologous protein (CHOP) and activating transcription factor 3 (ATF3) mRNA levels (1.60- and 3.41-fold increase compared with control islets) (Fig. 1B and C). IFN-α induced upregulation of chemokine C-X-C motif chemokine ligand 10 (CXCL10), the antiviral MX dynamin-like GTPase 1 (MX1), and MHC class I mRNA after 24 h (233.85-, 34.18-, and 2.40-fold compared with control islets, respectively) (Fig. 1E–G), as previously described (34). IFN-α did not induce apoptosis (data not shown). Exposure to inflammatory cytokines or IFN-α resulted in significant upregulation of TGM2 mRNA levels, with a 5.17- and 2.15-fold increase compared with control islets, respectively (Fig. 1D and H). Next, TGM2 expression was investigated in RNA sequencing data retrieved from the HPAP database in endocrine and nonendocrine cells from islets obtained from individuals with and without diabetes. While there was no difference in nonendocrine cells, TGM2 expression was significantly increased in islet endocrine cells of individuals with diabetes compared with control individuals (Fig. 1I and J). A functional enrichment analysis of genes enriched in endocrine cells of individuals with diabetes showed the presence of inflammation-induced gene signatures, mostly downstream of IFNs (Fig. 1K), similar to our previous findings (35).

Protein lysates from primary islets exposed to inflammatory cytokines or IFN-α were subjected to LC-MS/MS and searched for Gln and Asn deamidations. Digestion and proteomic analysis of cytokine-exposed islets resulted in the detection of 3,587 ± 823 and 4,245 ± 801 proteins, respectively (identified with high confidence), corresponding to 27,475 ± 5,564 and 30,658 ± 2,930 unique peptides, respectively. Of these unique peptides, 1,510 ± 315 (or 5.52% ± 0.61%) and 1,586 ± 293 (or 5.13% ± 0.55%) were predicted to have one or more deamidated or citrullinated residues. MS/MS spectra were manually inspected to confirm deamidation, as previously described (15). A total of 16 novel enzymatic Gln deamidations were confirmed, with 8 only occurring in the inflammatory cytokine–stressed islets (donors 1–4), 4 only occurring in the IFN-α–stressed islets (donors 5–8), and 4 occurring in both sets of islets (Table 1). Nonenzymatic Asn deamidations were less influenced by the kind of stress; out of 25 confirmed Asn-deamidated peptides, 11 only occurred in inflammatory cytokine–stressed islets, 2 only in IFN-α–stressed islets, and 12 in both sets of islets (Table 2). Of note, nonenzymatic deamidations were restricted to fewer proteins than enzymatic deamidations, but more residues within each protein were deamidated (5 proteins with up to 9 Asn-deamidated residues per protein vs. 14 proteins with a maximum of 2 Gln-deamidated residues per protein).

A previously developed algorithm (25,26) was used to predict binding of deamidated peptide sequences to HLA-DRB1*04:01. Peptides with predicted binding (Supplementary Table 6) were synthesized and their binding affinity measured. Five peptides with enzymatic deamidations and four with nonenzymatic deamidations had detectable binding to DRB1*04:01 (Table 3). Of the five Gln-deamidated peptides, two peptides, including the 78-kDa glucose-regulated protein (GRP78) 445–459 with deamidation at residue 449 (GRP78 445–459 E449) and prolyl-4-hydroxylase β (P4HB) 193–207 with deamidation at residue 197 (P4HB 193–207 E197), bound with high affinity (IC₅₀ <10 μmol/L). One peptide, P4HB 188–202 with deamidation at residue 197 (the same residue as the other P4HB peptide) (P4HB 188–202 E197) had an intermediate binding affinity. Two peptides bound weakly (IC₅₀ >30 μmol/L), including HSP β-1 (HSPB1) 166–180 with deamidation at position 175 (HSPB1 166–180 E175) and 14-3-3 protein ζ/δ (YWHAZ) 138–152 with deamidations at positions 144 and 146 (YWHAZ 138–152 E144E146). Of the four Asn-deamidated peptides, three bound with high affinity: GRP78 547–566 with deamidation at residue 457 (GRP78 447–466 D457), GRP94 114–133 with deamidation at residue 124 (GRP94 114–133 D124), and HSPA9 139–158 with deamidation at residue 149 (HSPA9 139–158 D149). One peptide, GRP94 328–347 with deamidation at residue 337 (GRP94 328–347 D337), had an intermediate binding affinity.

Immunogenicity of the deamidated peptides was assessed by stimulating PBMCs from individuals with type 1 diabetes (summarized in Supplementary Table 2) in vitro with peptides for 14 days, followed by staining with DRB1*04:01 tetramers. The percentage of CD4⁺Tmr⁺ cells was compared with background (a control peptide with measurable DRB1*04:01 binding but no immunogenicity). We observed a significant increase in CD4⁺Tmr⁺ cells compared with background for six deamidated peptides and trends toward a higher percentage for three others (Fig. 2A). Representative staining results for each deamidated peptide are depicted in Fig. 2B. Responses were considered positive when the percentage of CD4⁺Tmr⁺ cells was >0.10% and at least threefold above background tetramer staining (CD4⁻Tmr⁺). Notably, all deamidated peptides with high or intermediate HLA binding elicited positive in vitro T-cell responses except for P4HB 193–207 E197. The two deamidated peptides with weak binding did not generate detectable responses in our assays. Responses against one individual without diabetes are shown in Supplementary Fig. 2.

To confirm that the above-described responses were bona fide, we sorted and expanded Tmr⁺CD4⁺ cells to obtain T-cell clones. Clones specific for P4HB 193–207 E197, GRP78 447–466 D457, GRP94 114–133 D124, and HSPA9 139–158 D149 were successfully isolated. Each clone stained as a single population of T cells with a mean fluorescence intensity above background (Supplementary Fig. 3). Because unmodified versions of several deamidated peptides of interest also had detectable binding to DRB1*04:01, we assessed whether T-cell clones preferentially responded to deamidated peptide by activating each clone in vitro using APCs pulsed with either native or deamidated peptide and assessing their proliferation through thymidine incorporation. In each case, the T-cell clone proliferated more robustly in response to deamidated peptide than the corresponding native peptide (Fig. 3). Specifically, respective proliferation was 23-, 45-, 26-, and 4-fold higher in response to the deamidated vs. native peptide. To clarify whether these observed differences were caused by altered HLA binding, we compared binding of each deamidated and native peptide to DRB1*04:01 (Supplementary Table 7). Deamidation only improved binding for GRP94 328–347, for which the native peptide had no detectable binding. This suggests WDWELMDIK as the minimal binding motif, placing the deamidated residue in P7. For P4HB 193–207, the deamidated residue is predicted to occupy P2, a T-cell receptor contact. For GRP78 547–566, GRP94 114–133, and HSPA9 139–158, deamidation negatively affects binding but is required for recognition. This implies conformational alterations of the bound peptide, leading to altered recognition. In any case, preferred recognition of the deamidated peptides establishes these peptides as bona fide neoepitopes.

We then selected a set of high-interest epitopes based on positivity after in vitro stimulation with peptide and/or the isolation of clones, representing both Gln and Asn-deamidated peptides. Corresponding HLA class II tetramers were used to assess the frequency and phenotype of CD4⁺ T cells specific for these novel deamidated peptides in PBMCs from healthy control individuals and individuals with type 1 diabetes (representative results are shown in Supplementary Fig. 4, positive and negative control individuals in Supplementary Fig. 5, and donor characteristics in Supplementary Table 3). Previous studies set the limit of detection for tetramer staining as 0.5–1 cell per million (36). Detectable frequencies were observed for GRP94 114–133 D124 and HSPA9 139–158 D149, but there was no significant difference between control individuals and individuals with type 1 diabetes (Fig. 4A and B). In contrast, significantly higher frequencies were observed in individuals with type 1 diabetes compared with control individuals for P4HB 193–207 E197, GRP78 447–466 D457, and GRP94 328–347 D337 (Fig. 4A and B). As illustrated by heat map visualization, diverse patterns of T-cell frequency were observed in our cohort. Donors 26 and 27 had high frequencies for GRP94 328–347 D337 but low frequencies for other epitopes. In contrast, donors 35 and 36 had comparatively high frequencies for all deamidated epitopes (Fig. 4C). We observed no significant differences in T-cell phenotype between control individuals and individuals with type 1 diabetes (Supplementary Fig. 6) probably because very few control individuals had adequate frequencies (five or more tetramer-positive cells per million CD4⁺ T cells) to permit phenotypic analysis. Next, to investigate whether the deamidated peptides are present in islets of individuals with diabetes at sufficient concentrations to elicit T-cell responses, cell lines were generated from T cells infiltrating the islets of nPOD donor 6551, and IFN-γ production upon exposure to deamidated peptide was measured. Significant IFN-γ production above background (a control HIV peptide) was measured in one (P4HB 193–207 E197) or two (all other peptides) out of three generated T-cell lines for each deamidated peptide.

We next asked if the frequency of P4HB 193–207 E197–, GRP78 447–466 D457–, and GRP94 328–347 D337–reactive T cells correlates with characteristics such as autoantibody status, age at diagnosis, or disease duration. Individuals positive for insulin autoantibodies (IAAs) at diagnosis (≥0.6% binding) had a trend toward higher frequencies of deamidated peptide–reactive T cells (Fig. 5A). Considering the three deamidated peptides individually, the frequency of T cells responsive against P4HB 193–207 E197 tended to be increased in IAA-positive individuals (Fig. 5B). Because low numbers of individuals were negative for GAD antibody and islet antigen 2 antibodies at diagnosis, we grouped individuals with low (including negative) versus high antibody titers, but no significant differences were observed (Fig. 5A). Zinc transporter 8 antibody levels were not determined and therefore could not be analyzed. A significant negative correlation was observed between Tmr⁺ T-cell frequencies and age at diagnosis (Fig. 5C). A significant positive correlation was present between Tmr⁺ T-cell frequencies and disease duration. No correlation was observed between age and Tmr⁺ T-cell frequencies, suggesting that the correlation between frequencies and disease duration is independent of the generally older age of individuals with longer disease duration (Fig. 5C). Positive individual correlations were also present for P4HB 193–207 E197, GRP78 447–466 D457, and GRP94 328–347 D337, providing extra strength to the observed correlation between disease duration and Tmr⁺ T-cell frequency (Fig. 5D).

Here, we report on the enzymatic formation of novel Gln-deamidated peptides and nonenzymatic formation of Asn-deamidated peptides in stressed primary human islets. Higher frequencies of CD4⁺ T cells reactive against deamidated peptides were present in the peripheral blood of individuals with type 1 diabetes compared with control individuals. In particular, nonenzymatically deamidated targets have not been previously investigated as neoepitopes in this disease. Therefore, our findings implicate a new class of relevant targets of autoimmunity.

Protein deamidation was potentiated in pancreatic islets through treatment with either IFN-α or a blend of inflammatory cytokines (IL-1β, TNF-α, and IFN-γ). Notably, these cytokines also influenced the surface expression of HLA molecules. HLA class I genes harbor interferon consensus sequences in their regulatory region, allowing MHC class I molecules to be induced by both IFN-α and IFN-γ. In contrast, HLA class II genes have a motif present in their regulatory region that does not respond to IFN-α, explaining why IFN-γ is a better inducer of MHC class II proteins (37). Correspondingly, we observed a highly significant induction of the MHC class II DR α-chain in cytokine-stimulated (IL-1β, TNF-α, and IFN-γ) islets (6.63-fold, P < 0.001 [data not shown]) and no significant induction of MHC class II proteins in IFN-α–exposed islets. In parallel, we observed a higher increase in TGM2 expression after cytokine exposure compared with IFN-α alone, as well as more deamidated peptides. These data suggest that inflammatory cytokines promote the formation and presentation of deamidated peptides to T cells. However, a notable limitation of our study is that because data were generated using unfractionated islets, deamidation may occur not only in β-cells but also in other cell types present in the islets. IFN-α is induced early in the immune response, providing a priming mechanism that can orchestrate subsequent pathways in innate and adaptive immunity (32). Since IFN-γ is part of a later stage in type 1 diabetes progression (38), deamidation might contribute to epitope spreading and the acceleration of autoimmune attack following initial immune priming by other classes of antigens.

We previously reported that GRP78 (involved in the unfolded protein response in endoplasmic reticulum stress) is an autoantigen in its citrullinated form in both murine and human diabetes (24,39). Autoantibodies against GRP78 are a diagnostic marker in rheumatoid arthritis, a disease that is also linked to HLA-DRB1*04:01 (40). Furthermore, autoimmunity in rheumatoid arthritis has been observed against citrullinated and carbamylated GRP78 (41,42). Therefore, modified versions of GRP78 are involved in the loss of self-tolerance in multiple disease settings. Notably, GRP78 is translocated to the cell surface upon inflammatory cytokine exposure, followed by secretion in the extracellular space, facilitating presentation to the immune system (43). We previously detected enzymatically deamidated GRP78 in murine islets of Langerhans (44). We now provide evidence that GRP78 is both enzymatically (two Gln residues) and nonenzymatically (nine Asn residues) deamidated in human islets. The frequencies of GRP78 447–466 D457–responsive T cells were increased in individuals with type 1 diabetes compared with healthy control individuals, and there was a positive correlation between Tmr⁺ T-cell frequency and disease duration.

P4HB, a chaperone essential for proper insulin folding, was recently shown to be an antigen in murine models of type 1 diabetes and in humans (45,46). Stressed human islets had amplified levels of carbonylated P4HB, coinciding with altered proinsulin/insulin ratios and decreased glucose-stimulated insulin secretion (GSIS), suggesting a loss of function of P4HB due to carbonylation (45). In our work, the frequencies of deamidated P4HB 193–207 E197–responsive T cells were significantly increased in individuals with type 1 diabetes, and there was a positive correlation between T-cell frequency and disease duration. Deamidation of proteins may alter their structure, stability, or function and contribute to protein degradation. It was recently shown that citrullination, a modification similar to deamidation, of glucokinase altered activity of the enzyme and suppressed GSIS (47). Therefore, in addition to activating autoreactive CD4⁺ T cells, deamidation of P4HB could contribute to altered proinsulin/insulin ratios and decreased GSIS, aggravating the disease.

In vitro responses toward the nonenzymatically deamidated peptides GRP94 114–133 D124 and HSPA9 139–158 D149 were observed, and clones were successfully isolated. However, T cells recognizing these peptides were not present at higher frequencies in individuals with type 1 diabetes than in healthy control individuals, suggesting that these peptides are a component of a benign autoimmune repertoire (as previously noted for islet-reactive CD8⁺ T cells) and not uniquely involved in type 1 diabetes pathogenesis (5,6).

Individuals with type 1 diabetes showed increased CD4⁺ T-cell frequencies directed against GRP94 328–347 D337 and a positive correlation between T-cell frequency and disease duration. GRP94 has not been described as an autoantigen in type 1 diabetes. However, a recent study investigated sequence similarities between GRP94 and known islet antigens, including insulin and islet amyloid polypeptide (48). Interestingly, the region encompassing GRP94 114–133, for which we isolated clones that recognize the deamidated form of the peptide, has high sequence similarity with INS 6–20 (48). This INS signal peptide region is targeted by CD8⁺ T cells (49) and has also been described as a DRB4-restricted CD4⁺ T-cell epitope that is detectable among the circulating T cells of individuals with type 1 diabetes (50). Therefore, GRP94 and other HSP peptides with similarity to islet peptides are of potential immunogenic relevance in type 1 diabetes (48), making GRP94 114–133 D124 an interesting candidate for further study.

Overall, we report on the recognition of five deamidated self-peptides, originating from four different HSPs: GRP78, GRP94, P4HB, and HSPA9. Importantly, several of these specificities were also recognized by islet infiltrating T-cell lines from a pancreatic organ donor. Antibodies against HSPs have been detected in various autoimmune diseases, such as systemic lupus erythematosus, rheumatoid arthritis, and type 1 diabetes (51), suggesting an important shared mechanism in human autoimmune disease. Besides the above-mentioned autoantibodies detected against (modified) GRP78 and P4HB in the context of type 1 diabetes, autoantibodies against HSP65 and HSP90 have also been reported (52,53). Here, we add two more HSPs, GRP94 and HSPA9, and a novel modification to the list, further supporting their importance in autoimmunity and type 1 diabetes.

We observed a negative correlation between T-cell frequencies and age at diagnosis, a positive correlation between frequencies and disease duration, and a trend for lower frequencies in individuals negative for IAA. A similar negative relationship between frequencies and age at diagnosis has been reported for hybrid insulin peptide–reactive cells, as well as lower frequencies in two IAA-negative individuals (54). Although a confirmatory study with more individuals is warranted, the observation of lower frequencies of CD4⁺ T cells reactive against PTM peptides (deamidated peptides and hybrid insulin peptides) in IAA-negative individuals suggests that IAA status at diagnosis might be an indicator of different disease endotypes or pathotypes.

Besides the above-mentioned deamidated peptides, we detected several other deamidated peptides in stressed human islets with LC-MS/MS, including chromogranin-A, neuroendocrine convertase 1, and secretogranin-2. Although none of these were identified as targets of DRB1*04:01-restricted T cells, both HLA-DQ2 and HLA-DQ8 (the HLA-DQ haplotypes most strongly associated with type 1 diabetes) favor presentation of negatively charged peptides (4,9). Consequently, it would be of great interest to investigate the recognition of deamidated targets in the context of HLA-DQ2 and HLA-DQ8.

In summary, we report the detection of novel deamidated peptides in stressed human islets and show that CD4⁺ T cells responsive against those deamidated peptides are present in individuals with type 1 diabetes. These results demonstrate that deamidation of self-proteins occurs within primary islets under conditions of cytokine stress and that deamidated peptides can promote the activation of CD4⁺ T cells. These findings add to the growing list of evidence that PTMs undermine tolerance in type 1 diabetes and may open the road for the development of new diagnostic or therapeutic applications for patients.

Acknowledgments. The authors thank Martine Gilis and Eline Desager (KU Leuven) for expert assistance and the SyBioMa Mass Spectrometry Facility of KU Leuven. The authors also acknowledge Hilde Morobé (UZ Leuven) for patient recruitment and collection of blood samples and thank all the patients and control individuals who donated blood samples. The authors thank the Benaroya Research Institute Flow Cytometry Core Facility for assistance with flow cytometry.

Funding. This work was supported by Innovative Medicines Initiative 2 Joint Undertaking grant 115797 (INNODIA) and 948268 (INNODIA HARVEST). This joint undertaking received support from the European Union’s Horizon 2020 research and innovation program and through the European Federation of Pharmaceutical Industries and Associations (EFPIA), JDRF, and The Leona M. and Harry B. Helmsley Charitable Trust. This research was performed with the support of nPOD (Research Resource Identifier SCR_014641), a collaborative type 1 diabetes research project supported by JDRF grant 5-SRA-2018-557-Q-R and The Leona M. and Harry B. Helmsley Charitable Trust grants 2018PG-T1D053 and G-2108-04793. The work was funded by the KU Leuven grant C16/18/006), Flemish Research Foundation predoctoral fellowship 3-PDF-2018-593-A-N (to A.C.), EFPIA postdoctoral fellowship 1189518N (to M.B.), JDRF grant 2-SRA-2018-551-S-B (to E.A.J.), and National Institute of Diabetes and Digestive and Kidney Diseases grants U01 DK104218 and UC4 DK116284.

The content and views expressed are the responsibility of the authors and do not necessarily reflect the official view of nPOD. Organ Procurement Organizations partnering with nPOD to provide research resources are listed at https://npod.org/for-partners/npod-partners/.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. A.C., P.G., R.D., M.B., F.M.C.S., S.V., and E.A.J. contributed to the design, conduct, analysis, and interpretation of the data. A.M. and S.C.K. conducted the islet T-cell line experiments. X.Y. and D.L.E. conducted the RNA sequencing data analysis. M.S. and P.M. collected human islets. C.M., E.A.J., and L.O. helped design experiments. All authors contributed to the writing of the manuscript and approved the final version. C.M., E.A.J., and L.O. 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.

Prior Presentation. Parts of this study were presented in abstract form at the 19th Immunology of Diabetes Society Congress, Paris, France, 23–27 May 2023. An earlier version of the work appears as a chapter in A.C.’s doctoral dissertation at KU Leuven.