Peptides Derived From Insulin Granule Proteins Are Targeted by CD8⁺ T Cells Across MHC Class I Restrictions in Humans and NOD Mice

Autoimmune CD8⁺ T cells are held as the final effectors of β-cell destruction in type 1 diabetes (T1D). Indirect support for this tenet comes from the observations that the pancreatic immune infiltrates of patients with T1D are dominated by CD8⁺ T cells (1); that at least some of these CD8⁺ T cells recognize islet epitopes (1–3); and that islet-reactive cytotoxic CD8⁺ T-cell clones can lyse β-cells in vitro (4–7). However, clones with similar cytotoxic activity can be derived from healthy donors (2). More generally, recent reports from our laboratory (2,3) and others (8) documented that the frequency of islet-reactive CD8⁺ T cells in peripheral blood is similar between donors with and without T1D, while these T cells are enriched in the pancreas of diseased individuals (2,3). We therefore proposed a novel paradigm of “benign” islet autoimmunity, which is present in all individuals and imprinted in the thymus, due to a marginal impact on clonal deletion of the antigens presented by medullary thymic epithelial cells (mTECs) (2,9,10).

The progression of this benign autoimmune state toward T1D may rely on different, nonexclusive mechanisms. First, regulation may be at play, either T-cell extrinsic (e.g., suppression by regulatory T cells) or intrinsic (e.g., anergy and exhaustion). Indeed, conventional T cells from patients with T1D are more resistant to regulatory T-cell suppression (11,12) and exhaustion (8,13), and assays relying on T-cell activation consistently detected higher frequencies of some islet-reactive fractions in patients with T1D (2,4,14–17). Second, a higher vulnerability of target β-cells may favor T1D progression (18). The lines of evidence for this “β-cell–centric” hypothesis include observations that: 1) some T1D susceptibility gene variants modulate islet inflammation (19–21), which makes β-cells more “visible” to autoimmune CD8⁺ T cells by upregulating HLA class I (HLA-I) expression and peptide–HLA-I (pHLA-I) presentation (3); 2) the benefit of effective immunotherapies is limited in time, which may hint at β-cell–intrinsic components not impacted by treatment (22); 3) contrary to what is found in the blood, islet-reactive CD8⁺ T cells are enriched in the pancreas of those with T1D (2,3), pointing to local factors that promote their disease-specific homing to the target organ; and 4) diseased β-cells display endoplasmic reticulum stress (23) and impaired proinsulin processing (24,25), which may promote their immunogenicity by favoring HLA-I presentation of potentially modified proinsulin peptides.

Overall, a unifying picture is emerging that prompts consideration of T1D as a disease of both the immune system and the β-cell (18,26). The identification of the peptides naturally processed by β-cells that trigger CD8⁺ T-cell recognition through presentation by surface HLA-I molecules may provide novel and relevant information on how β-cells exert an active role in disease pathogenesis. We recently provided a first description of the HLA-I peptidome of human β-cells under basal and inflammatory conditions and detailed novel antigenic peptides recognized by circulating CD8⁺ T cells in the frame of the most prevalent HLA-A*02:01 (HLA-A2) variant (3). As observed for known islet epitopes (2), the frequency of these CD8⁺ T cells was similar in the blood, but higher in the pancreas, of patients with T1D compared with healthy donors (3).

Several other HLA-I–eluted peptides were predicted to bind the other common HLA-A*03:01 variant (HLA-A3 from hereafter) (3), which presents peptides carrying different amino acid (aa) motifs compared with HLA-A2. Whether CD8⁺ T cells recognize these peptides was not assessed. In this study, we looked at this recognition in order to define whether HLA-A2 and HLA-A3, despite their different peptide-binding preferences, present peptides with any shared feature. We report that peptides derived from insulin granule proteins are common for both variants and are also targeted by CD8⁺ T cells that infiltrate the islets of prediabetic NOD mice and transfer disease into NOD/scid recipients.

The first set of candidate HLA-A3–restricted epitopes was previously identified by HLA-I peptidomics (3) (Table 1, top). Potential HLA-A3 binding was assigned based on predicted affinity (NetMHC 4.0) and stability (NetMHCstab 1.0) scores. Candidates derived from mRNA splice variants were identified by a parallel in silico mining of previous RNA-sequencing (RNA-seq) data sets from primary human islets exposed or not to interleukin-1β (IL-1β) and interferon-γ (IFN-γ) and from human mTECs, as described in Gonzalez-Duque et al. (3) and summarized in Supplementary Fig. 1. Relevant mRNA splice variants were first selected based on a median reads per kilobase per million mapped reads (RPKM) >5 in islets, based on the median RPKM of known islet antigens (3,27). Second, mRNA isoforms poorly expressed in mTECs, which might favor T-cell escape from clonal deletion, were selected based on an RPKM <0.1 in mTECs or with a fold decrease >100 versus islets (3). Third, only mRNA isoforms with >10-fold higher expression in islets versus other tissues were retained (3). We then analyzed the predicted aa neosequences encoded by these mRNA variants, taking the longest and/or most prevalent islet mRNA isoform as reference. This neosequence filter yielded 88 out of 166 mRNA variants (53%) and 340 peptide neosequences identified as a potential source of neoepitopes. Thirty-seven peptides predicted to bind HLA-A3 were thus identified. We further restricted this list by focusing on 9–10-aa–long peptides (29 out of 37, 78%) carrying a neosequence stretch ≥3 aa, with 24 out of 29 (83%) candidates finally retained (Table 1, bottom).

Peptides (>90% pure; Synpeptide Co. Ltd.) were tested for binding to HLA-A3 by flow cytometry using biotin-tagged HLA-A3 monomers (immunAware), per the manufacturer’s protocol. Briefly, biotinylated monomers (final concentration 1.2 nmol/L) were folded as described (3) and captured on 6–8-µm streptavidin-coated beads (Spherotech Inc.). Beads were subsequently incubated with an anti–β₂-microglobulin BBM.1 monoclonal antibody (Santa Cruz Biotechnology), followed by an Alexa Fluor (AF) 488–labeled goat polyclonal anti-mouse IgG (RRID:AB_2728715). The HLA-A3–binding peptide Flu NP_265-273 (ILRGSVAHK) and a nonbinding peptide chromogranin-A (CHGA)_382-390 (HPVGEADYF) were included as positive and negative controls, respectively. Following acquisition on a BD LSRFortessa cytometer, results were analyzed by gating on single beads and AF488⁺ events and expressed as the median fluorescence intensity fold increase of the test pHLA complex compared with the negative control. Peptides with a more than fourfold median fluorescence intensity increase were retained as binders.

HLA-A3⁺ patients with new-onset T1D (n = 9; median age 26 years [range 22–40]; median disease duration 1.4 weeks [range 0.3–32]) and age/sex-matched healthy donors (Supplementary Table 1) gave written informed consent under ethics approval DC-2015-2536 Ile-de-France I. HLA-A3 (A*03:01) typing was performed with AmbiSolv primers (Dynal/Thermo Fisher Scientific). Blood was drawn into 9-mL sodium heparin tubes and peripheral blood mononuclear cells (PBMCs) isolated and frozen as described (3,28). Frozen/thawed samples were used throughout the study. HIV serology was assessed with the Alere HIV Combo test (Abbott Laboratories).

HLA-A3 multimers (MMrs) (immunAware) were produced and used as described (3). pHLA-A3 complexes (final concentration 8–27 nmol/L) were conjugated with fluorochrome-labeled streptavidins at a 1:4 ratio. The concentration of each fluorescent MMr was corrected for the variable staining index of each streptavidin, in order to obtain a distinct double-MMr⁺ population for each fluorochrome pair. Compensations were set using fluorescence-minus-one samples (i.e., omitting one streptavidin at a time).

PBMCs were thawed at 37°C and immediately diluted in prewarmed AIM V medium. Following centrifugation and one additional wash, PBMCs were counted and incubated with 50 nmol/L dasatinib for 30 min at 37°C before magnetic depletion of CD8⁻ cells (RRID:AB_2728716). Staining with combinatorial double-coded MMr panels (see results) was performed for 20 min at 20°C in 20 μL PBS-dasatinib for 10⁷ cells followed, without washing, by staining at 4°C for 20 min with monoclonal antibodies CD3-APC-H7 (RRID:AB_1645475), CD8-PE-Cy7 (RRID:AB_396852), CD45RA-FITC (RRID:AB_395879), CCR7-BV421 (RRID:AB_2728119), and Live/Dead Aqua (Thermo Fisher Scientific). After one wash, cells were acquired using an LSRFortessa (screening round) or FACSAria III cytometer (validation round) carrying identical configurations (3) and analyzed with FlowJo v10 and GraphPad Prism 7. HLA-A3–binding candidate epitopes that did not yield any appreciable MMr staining provided negative controls for each panel. The positive control peptides included in each assay were preproinsulin (PPI)_80-88 (LALEGSLQK), a shorter version of the PPI_79-88 peptide (PLALEGSLQK; Immune Epitope Database and Analysis Resource [IEDB] #159292) with superior HLA-A3 binding affinity, HIV nef_84-92 (AVDLSHFLK; IEDB #5295), and Flu NP_265-273 (ILRGSVAHK; IEDB #27283).

Peptides predicted to bind the murine MHC class I (MHC-I) K^d molecule were identified using the SYFPEITHI and NetMHC 4.0 algorithms applied to the entire murine protein sequence (Supplementary Table 3). Islets were isolated from 12–16-week-old female NOD mice by collagenase digestion and cultured for 7 days with recombinant human IL-2 (Proleukin; Novartis), as described (29). Cells exiting the islets were subsequently collected and subjected to recall assays with K^d+ L antigen-presenting cells pulsed with the indicated peptides (10 μmol/L) for 5 h in the presence of brefeldin A, followed by staining for CD45-BV650 (RRID:AB_2738189), TCRβ-PE-Cy7 (RRID:AB_1937310), CD8-AF700 (RRID:AB_396959), CD4-BV711 (RRID:AB_2738389), intracellular IFN-γ (RRID:AB_395376), and Live/Dead-BV510 (Thermo Fisher Scientific). The Tum peptide (KYQAVTTTL) and insulin (Ins)_B15-23 and islet-specific glucose-6-phosphatase catalytic subunit-related protein (Igrp)_206-214 peptides were included as negative and positive controls, respectively.

For transfer experiments, 12-week-old female NOD mice were primed subcutaneously with hepatitis B core antigen (HBcAg)_128-140 helper peptide (TPPAYRPPNAPIL; 1 mmol/L) in complete Freund’s adjuvant containing a pool of five peptides previously identified as targeted by islet-infiltrating CD8⁺ T cells (Ucn3_5-13, Ucn3_32-40, Pcsk2_341-350, Pcsk2_501-510, and Scg5_193-201; 1 mmol/L/each), Ins_B15-23 (1 mmol/L), or PBS. Three intraperitoneal boost immunizations at half dose were performed at 2-week intervals without helper peptide in incomplete Freund’s adjuvant. Splenocytes isolated 2 weeks after the last boost were cultured for 7 days in the presence of the same peptides, followed by intravenous orbital transfer into 9–11-week-old female NOD/scid recipients (4 × 10⁶ cells/mouse). Diabetes development was followed biweekly by monitoring glycosuria, confirmed by glycemia measurement when positive. Mice were sacrificed once diabetic or at the end of the follow-up, and pancreata were fixed in 4% paraformaldehyde, dehydrated for 24 h with ethanol, and embedded in paraffin. Pancreas sections (4 μm) were taken at five different levels at 100-µm intervals, and four sections per level were stained with hematoxylin/eosin. Mononuclear cell infiltration was scored as absent, surrounding islets (peri-insulitis) or infiltrating islets (insulitis; covering >50% of the islet surface area). Sections were further stained with anti-CD8 (RRID:AB_2756376) and anti-insulin (RRID:AB_260137) primary antibodies, followed by AF488-conjugated anti-rabbit (RRID:AB_2576217) and AF594-conjugated anti-mouse (RRID:AB_2338871) secondary antibodies. Images were acquired on a Leica DM4000 B microscope followed by ImageJ software analysis (National Institutes of Health).

All statistical tests were performed using GraphPad Prism 7, as detailed in the legends of each figure. P values <0.05 were considered significant.

The islet RNA-seq data sets used to generate Supplementary Fig. 1 have been deposited in the Gene Expression Omnibus under accession number GSE108413. Other transcriptomics and peptidomics data sets were previously described (3). Raw flow cytometry data are available from the corresponding author upon reasonable request. The study does not involve any noncommercial reagent or tool.

Table 1 (top panel) lists the 19 candidate HLA-A3–restricted epitopes previously eluted from the pHLA-I complexes of the human HLA-A2/A3⁺ ECN90 β-cell line (3). As observed for HLA-A2–restricted epitopes (3), most of these peptides were preferentially eluted from β-cells exposed to IFN-γ, alone or in combination with tumor necrosis factor-α (TNF-α) and IL-1β, compared with the basal condition.

The parallel in silico mining of mRNA splice variants identified in RNA-seq data sets from primary human islets treated or not with cytokines retrieved an additional list of 24 candidates (Table 1, bottom), thus yielding a total of 43 peptides. Most of the source mRNA variants (13 out of 24, 54%) were similarly expressed in untreated and cytokine-treated islets, and only 2 out of 24 (8%) were upregulated in the latter. Notably, the predicted SCG5–009_193-201 peptide was also eluted from IFN-γ–treated ECN90 β-cells.

HLA-A3–binding assays were performed with the same biotinylated HLA-A3 monomeric constructs subsequently used for MMr synthesis, thus directly verifying the efficient folding of each pHLA-A3 complex. HLA-A3 binding was experimentally confirmed for 42 out of 43 candidates (98%; see the last column in Table 1 and Supplementary Fig. 2), which were retained for CD8⁺ T-cell studies.

The combinatorial MMr panel (3) was first set up by staining HLA-A3⁺ PBMCs with the same set of fluorescent streptavidin-labeled MMrs, all loaded with the Flu NP_265-273 epitope, which yielded identical MMr⁺CD8⁺ T-cell frequencies for all of the 15 MMr pairs (Supplementary Fig. 3). The same MMr⁺ population was detected with all pairs, except for the dimmer pairs BV650/BV711, BV650/BV786, and BV711/BV786. These pairs were therefore subsequently used to detect high-frequency CD8⁺ T-cell populations (e.g., those reactive to control viral peptides).

A representative example of the staining obtained with the HLA-A3 MMrs generated with these peptides is shown in Fig. 1. Since the first requirement for a peptide to qualify as a T-cell epitope is the existence of a naive T-cell repertoire capable of recognizing it, we performed a first screening round on seven HLA-A3⁺ healthy donors (Fig. 2 and Supplementary Table 2). Peptides were scored as positive when their cognate CD8⁺ T cells fell into the expected frequency range for naive T-cell precursors (1–50/10⁶ CD8⁺ T cells) and displayed a clustered MMr staining pattern, indicative of specific staining compared with spread staining patterns (Fig. 2A–D). The positive control peptides included in each assay, namely the islet epitope PPI_80-88 (Fig. 2E) and the viral epitope HIV nef_84-92 (Fig. 1), fulfilled both criteria, while the frequency of the positive control recall epitope Flu NP_265-273 (Fig. 2F) was, expectedly, higher in approximately half of the donors. Using the same criteria (along with counting of greater than or equal to five MMr⁺ cells in at least one donor to exclude low counts possibly attributable to assay noise), 19 out of 42 islet peptides were validated (Fig. 2G and H). As expected, only a fraction of MMr⁺CD8⁺ cells recognizing islet peptides displayed an antigen-experienced (CD45RA⁺CCR7⁺) phenotype (Fig. 2I and J) (median 20%, interquartile range 8–41%), while those recognizing the recall Flu peptide were almost exclusively effector/memory (median 98% [interquartile range 88–99%]).

Of the 19 peptides validated in this first screening round, 12 were retained for further comparison between donors with new-onset T1D and healthy donors. This number was chosen to fit the 15-color multiplex capacity of our combinatorial MMr assays (including the PPI, HIV, and Flu control peptides). It included all of the 7 peptides from the HLA-I peptidomics pipeline, namely kinesin family member 1A (KIF1A)_860-868, paraneoplastic antigen Ma2 (PNMA2)_50-58, insulinoma antigen-2 (IA-2, PTPRN)_{576-580/708-711}, reticulon-1 (RTN1)_120-129, secretogranin-3 (SCG3)_166-174, SCG5–009_193-201, and urocortin-3 (UCN3)_46-56; and the 5 out of 12 peptides from the RNA-seq pipeline that displayed the most reproducible frequencies across donors, namely chromosome 15 open reading frame 48 (C15orf48)–003_50-58, catenin δ-1 (CTNND1)-026_852-860, guanine nucleotide-binding protein G(s) subunit α (GNAS)-036_74-83, GNAS-050_159-168, and GNAS-050_477-485.

The results of the second validation round on PBMC samples from donors with T1D and healthy donors are summarized in Fig. 3. Eleven out of the 12 peptides studied were validated as CD8⁺ T-cell epitopes. The exception of PNMA2_50-58 was likely due to a false-positive hit during the screening round, as a third MMr synthesis used to test PBMCs from two healthy donors previously positive confirmed the absence of MMr⁺ events (data not shown). As we previously reported for HLA-A2–restricted islet epitopes (2,3), these circulating CD8⁺ T cells displayed a similar frequency in donors with T1D and healthy donors (Fig. 3A), in the same range previously observed (1–50/10⁶ CD8⁺ T cells). Also in this case, their phenotype was mostly naive irrespective of disease status (Fig. 3B), since antigen-experienced fractions were limited in both donor groups (median 33% and 21% [interquartile range 13–50% and 10–47%] for donors with T1D and healthy donors, respectively). However, large (>75%) fractions of antigen-experienced cells were observed for some peptides in few donors with diabetes and healthy donors (e.g., for SCG5–009_193-201 in patients). Despite negative serological testing, this also applied to HIV-reactive CD8⁺ T cells in two healthy donors (the same as in Fig. 2), likely representing memory-like T cells cross-reactive with unrelated epitopes (30). Overall, antigen-experienced fractions were higher among HIV-reactive CD8⁺ T cells from healthy donors. An opposite, nonsignificant trend was noted for SCG5–009-reactive CD8⁺ T cells, which were more antigen experienced in patients with T1D.

The frequency of islet-reactive CD8⁺ T cells and their antigen-experienced fraction were also similar in donors with T1D and healthy donors when analyzing all islet epitope specificities pooled together (Fig. 3A and B). Since activated islet-reactive CD8⁺ T cells may display different morphological features (31), we also repeated this analysis by applying a more permissive gate that included cells with higher forward and side scatter (Fig. 1), with similar results (Fig. 3C and D).

We further noticed that several proteins previously identified as sources of HLA-A2–restricted epitopes (3) gave positive hits also for the HLA-A3 restriction, namely KIF1A_860-868, secretogranins (SCG3_166-174 and the mRNA splice variant SCG5–009_193-201), UCN3_46-56, GNAS-036_74-83, GNAS-050_159-168, and GNAS-050_477-485. A PTPRN_{576-580/708-711} cis-spliced peptide derived from the fusion of two noncontiguous IA-2 sequences was also identified.

Collectively, these data show that HLA-A3–restricted islet-reactive CD8⁺ T cells circulate at similar frequency irrespective of disease status and share several antigenic targets with their HLA-A2–restricted counterparts.

Besides insulin, other antigens shared by HLA-A3– and HLA-A2–restricted islet-reactive CD8⁺ T cells are secretory granule proteins (i.e., SCG5–009 and UCN3). We previously reported that granule proteins contribute to one-third of the β-cell–specific HLA-I peptidome (3). We therefore asked whether peptides derived from granule proteins other than insulin could also behave as CD8⁺ T-cell epitopes in NOD mice.

To this end, we analyzed the IFN-γ production of islet-infiltrating CD8⁺ T cells from 12–16-week-old prediabetic NOD female mice in response to peptides derived from murine Ucn3, Pcsk2, and Scg5 with a predicted restriction by the MHC-I molecule H-2K^d (Supplementary Table 3). Significant IFN-γ responses were observed for several of these peptides, both in terms of number of responding CD8⁺ T cells and of positive mice (Fig. 4)—Ucn3_5-13 and Ucn3_32-40; Pcsk2_341-350 and Pcsk2_501-510; and Scg5_193-201—and the positive controls Ins_B15-23 and Igrp_206-214. Although lower than for Igrp_206-214, the magnitude of the response was similar to that observed for the immunodominant Ins_B15-23 epitope. No response was observed against peptides derived from the nongranule protein insulin gene enhancer protein-1 (Isl1), except for a low-grade response to Isl1_65-74.

To explore the pathogenic potential of the granule epitopes targeted in islet infiltrates, pooled peptides were used to expand the corresponding CD8⁺ T cells in vivo by biweekly prime-boost immunization of 12-week-old female NOD mice. Splenocytes were then recovered and activated in vitro with the same peptides, followed by intravenous transfer into NOD/scid recipients. While mice receiving control splenocytes from NOD mice remained diabetes free, eight out of nine (89%) mice transferred with Ucn3/Pcsk2/Scg5 peptide-activated splenocytes became diabetic (Fig. 5A) and displayed few residual islets in their pancreata (Fig. 5B); all developed invasive insulitis with CD8⁺ T-cell infiltrates and a weak or near-absent insulin content (Fig. 5B and C). Diabetes development and destructive insulitis were similar in mice transferred with control Ins_B15-23-activated splenocytes (five out of five diabetic mice, 100%). Despite preserved islet counts, invasive CD8⁺ insulitis was also present in the single mouse that did not develop diabetes upon transfer of granule peptide-activated splenocytes (Fig. 5B and C).

Collectively, these results show that some granule proteins are targeted by CD8⁺ T cells across MHC-I restrictions in humans as well as in NOD mice, and that these murine CD8⁺ T cells are diabetogenic upon adoptive transfer.

The key finding of this study is that peptides derived from insulin granule proteins are targeted by CD8⁺ T cells across MHC-I restrictions in humans and NOD mice. The fact that Ucn3-, Pcsk2-, and Scg5-reactive CD8⁺ T cells are part of the autoimmune islet infiltrates of NOD mice suggests a role in disease pathogenesis. This conclusion is further supported by the fact that we analyzed infiltrates at the prediabetic stage of disease, thus making it unlikely that these CD8⁺ T cells represent a simple epiphenomenon of prior islet destruction. More importantly, diabetes transfer into NOD/scid recipients provides direct proof of their pathogenic potential.

The sequences targeted in the NOD mouse were, expectedly, different from the human ones, given the incomplete homology between the mouse and human protein orthologs and the use of a different MHC-I restriction element, namely H-2K^d, in the mouse. Nonetheless, two human HLA-A2 and HLA-A3 and one mouse H-2K^d restrictions carrying different peptide-binding motifs revealed the common feature that the insulin granule is a rich source of epitopes. Besides PPI (2–4,14,32), the granule proteins identified as antigens for CD8⁺ T cells include CHGA, IA-2, islet amyloid polypeptide (IAPP), proconvertase-2 (PCSK2), SCG5, UCN3, and zinc transporter-8 (ZnT8; SLC30A8) for HLA-A2 (2,3,16); IA-2, SCG3, SCG5, and UCN3 for HLA-A3 (this study); and Chga, Pcsk2, Scg5, and Ucn3 for H-2K^d (33 and this study). These granule-derived proteins are dominant yet not exclusive sources of epitopes, the most notable exceptions being 65-kD GAD (GAD65; GAD2) and IGRP/G6PC2.

What may make granule proteins so immunogenic? First, these proteins are among those most abundantly produced by β-cells. This makes them more likely to feed the MHC-I presentation pathway, especially in light of their high turnover and consequent proneness to transcriptional and translational deviations, which are increased under conditions of β-cell stress (18). Indeed, granule peptides account for one-third of the β-cell–specific HLA-I peptidome and are enriched in the peptidome of β-cells exposed to inflammatory cytokines (3). In this respect, CD8⁺ T-cell epitopes derived from the SCG5–009 mRNA splice isoform were identified for both the HLA-A2 and HLA-A3 restriction.

Second, several of these granule proteins, namely CHGA, PCSK2, SCG3, SCG5, and UCN3, share one key feature with the primary islet antigen PPI: all are soluble proteins that are produced as precursors (prepro-proteins), which undergo cleavage of their leader sequence to yield proproteins. This is followed by enzymatic processing, mostly by proconvertases, to give rise to their bioactive products. An impairment of proinsulin processing associated with a reduced expression of proconvertases is described in human islets exposed to inflammatory cytokines (34) and in the islets of both patients with T1D and autoantibody-positive individuals prior to disease onset (24,25). This echoes the increased serum proinsulin/insulin ratio consistently detected in patients with prediabetes (35). Since CHGA, PCSK2, SCG3, SCG5, and UCN3 undergo the same processing pathways of proinsulin, it is possible that they may likewise remain incompletely processed. Whether this incomplete processing represents a defense mechanism of β-cells to reduce their antigenic visibility (25,36) or rather promotes immunogenicity is an open question. Indeed, several of the epitopes identified map to regions that represent byproducts in the processing of these protein precursors. While these byproducts may be immunogenic per se along their physiological processing pathway, their immunogenic potential may be enhanced under conditions of increased availability due to impaired degradation. This altered processing may also divert these proteins toward alternative routes (e.g., crinophagy) (37,38). For some of these proteins, other epitopes map to the leader sequence. Although not detected for the HLA-A3 restriction, such epitopes were common for HLA-A2 (e.g., PPI_2-10, PPI_6-14, PPI_15-24, and UCN3_1-9 [3]) and in this study were also found in the NOD mouse (i.e., Ucn3_5-13). Also in this case, leader sequence peptides may be presented via HLA-I as part of their canonical processing in the endoplasmic reticulum outside of the conventional proteasome route (39), and this pathway may be enhanced or diverted in stressed β-cells.

Third, following granule exocytosis, CHGA, PCSK2, SCG3, SCG5, and UCN3, are released along with (pro)insulin in the bloodstream, in intact or degraded form. They may thus be taken up by dendritic cells outside the pancreas and sensitize lymphoid tissues at distance. This scenario, proposed by Wan et al. (40) for antigenic insulin peptides that are catabolites of proinsulin processing, may also apply to these granule proteins and derived byproducts (18).

Another group of peptides shared by HLA-A2 and HLA-A3 were those derived from GNAS mRNA splice variants. The GNAS gene codes for the ubiquitous protein GαS, which controls the activation of cAMP signals through the receptors of several hormones (41). Loss-of-function mutations are described in several syndromes of hormone resistance, while gain-of-function mutations are found in endocrine and nonendocrine tumors (41). GNAS gives rise to multiple coding and noncoding transcripts (41), which may make it more prone to generate neoepitopes, as suggested for tumor-specific GNAS mutations (42).

This study also reinforces our previous conclusions (2,3) that CD8⁺ T cells reactive to islet epitopes circulate at similar frequencies in those with T1D and healthy individuals, a finding replicated by others (8,43), and that these T cells display a limited antigen-experienced phenotype irrespective of disease status. This phenotype was, however, more largely represented in both donor groups for HLA-A3–restricted CD8⁺ T cells than for their HLA-A2–restricted counterparts previously analyzed (2,3), possibly reflecting cross-reactivity with unrelated epitopes. Indeed, a dominant antigen-experienced fraction was also observed for HIV-reactive CD8⁺ T cells in two HIV-seronegative donors, thus ruling out prior in vivo priming by the nominal epitope. High-throughput technologies to interrogate T-cell cross-reactivity (44,45) will be highly valuable to define TCR promiscuity, possibly pointing to environmental antigen triggers that may favor the development of autoimmunity (46,47). Another major question will be to define the phenotypic and functional differences between the islet-reactive CD8⁺ T cells of donors with diabetes and healthy donors (26).

Antigen-reactive CD8⁺ T cells undergoing clonal expansion can be present in the blood, but may be missed by conventional gating strategies due to their high side scatter profile, which reflects their mitochondrial and chromatin dynamics (31). To rule out this potential pitfall, we included these high-scatter cells in a comparative analysis, but came to similar conclusions. If present in the circulation, these cycling CD8⁺ T cells do not therefore seem to contain significant islet-reactive fractions.

In conclusion, islet-reactive CD8⁺ T cells reveal some common patterns across human and mouse MHC-I restrictions, most notably the key role of the insulin granule in priming diabetogenic CD8⁺ T cells and shaping the autoimmune vulnerability of β-cells. The identification of novel antigens, including neoantigens derived from peptide cis-splicing and mRNA alternative splicing, that are naturally processed and presented by β-cells, targeted by CD8⁺ T cells in the blood and pancreas and pathogenic significantly expands our capacity to track the overall autoimmune “burden” along the natural history of T1D or following immunotherapy (48–50). It invites further studies to define the time of appearance of these T cells along disease progression, their correlation with clinical phenotypes, and the existence of CD4⁺ T cells or autoantibodies directed against the same antigens. More critically, it opens new avenues to elucidate the mechanisms of progression from benign islet autoimmunity to T1D and their associated biomarkers. In this perspective, the relative contribution of T cells and β-cells needs to be considered. Disease pathogenesis should not be viewed as a “T-cell monologue” but rather as a misguided dialogue between the immune system and β-cells (18,26).

Acknowledgments. The authors thank A. Jones and the Diabetology nursing and medical staff of the Cochin (Paris, France) and André Mignot Hospitals (Le Chesnay, France) for patient recruitment; the CyBio and HistIM platforms (Institute Cochin, Paris, France) for assistance with flow cytometry and histology, respectively; G. Sebastiani and F. Dotta (University of Siena, Siena, Italy) for the expert advice on immunofluorescence; A. Carré, J. Perez-Hernandez, and Z. Zhou (Mallone-You Lab, Institut Cochin, Paris, France) for reviewing the manuscript; and all of the patients and healthy volunteers for participating in this study.

Funding. This work was performed within the Département Hospitalo-Universitaire Autoimmune and Hormonal Diseases (AUTHORS) and supported by Legs Borel, a PhD fellowship of the Région Ile-de-France CORDDIM and Aide aux Jeunes Diabétiques (to S.G.-D.), a Master fellowship from the Société Francophone di Diabète (to A.P.), and by grants from JDRF (2-SRA-2016-164-Q-R), the Fondation Francophone pour la Recherche sur le Diabète, the EFSD/JDRF/Lilly European Programme in Type 1 Diabetes Research 2015, the Agence Nationale de la Recherche (ANR-19-CE15-0014-01), the Fondation pour la Recherche Médicale (EQU20193007831), the Association pour la Recherche sur le Diabète (to R.M.), Inserm-Transfert Proof of Concept 2018 (to S.Y.), Welbio/FRFS (Wallonia, Belgium) (WELBIO-CR-2019C-04 to D.L.E.), and Conseil Régional d’Ile-de-France (to J.V.). R.M. and D.L.E. received funding from the H2020 European Research Council Innovative Medicines Initiative 2 Joint Undertaking under grant agreements INNODIA 115797 and INNODIA HARVEST 945268, which also receive support from the European Federation of Pharmaceutical Industries and Associations, JDRF, and the Leona M. and Harry B. Helmsley Charitable Trust.

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

Author Contributions. D.L.E., S.Y., and R.M. were responsible for conceptualization. M.E.A., M.T., G.A., M.L.C., L.A.-H., S.G.-D., Y.V., J.V., S.Y., and R.M. were responsible for methodology. M.E.A., M.T., G.A., A.P., M.L.C., C.M., C.L. L.A.-H., S.G.-D., D.L.E., S.Y., and R.M. were responsible for investigation. S.P., S.B., D.D.-L., E.L., J.-P.B., and G.B. provided resources. M.E.A., M.T., A.P., S.Y., and R.M. were responsible for data curation. M.E.A., S.Y., and R.M. wrote the original draft. G.A., J.V., S.B., D.D.-L., E.L., D.L.E., S.Y., and R.M. were involved in the review and editing of the paper. M.E.A., M.T., A.P., M.L.C., S.Y., and R.M. were responsible for visualization. S.Y. and R.M. provided supervision. D.L.E., S.Y., and R.M. were responsible for funding acquisition. R.M. is the guarantor of this work and, as such, had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented at the annual meetings of the European Association for the Study of Diabetes, Barcelona, Spain, 16–20 September 2019 and the Network for Pancreatic Organ Donors with Diabetes, Hollywood, FL, 19–22 February 2019.