NOD/ShiLtJ mice were from the Jackson Lab (Bar Harbor, ME, USA). 4.1-NOD and 4.1-NOD.Rag2 ^–/– mice have been described (8). All mice were bred and maintained in a specific-pathogen free (SPF) animal facility, and the experimental procedures were approved by the Ethics Committees for Animal Experimentation from the University of Barcelona and University of Calgary.

HEK-293T and JurMA cells were cultured with high-glucose Dulbecco’s modified eagle medium (DMEM) (Sigma) supplemented with 10% fetal bovine serum (FBS) (Sigma), 2 mM L-glutamine (Cultek), 1 mM sodium pyruvate (Sigma), 1% penicillin/streptomycin (Sigma) and 0.05 mg/ml gentamycin sulfate (Sigma). CHO-S cells were grown in PowerCHO-2 media (Lonza) supplemented with 8 mM glutamine and 0.05 mg/ml gentamicin sulfate. E. coli cultures were grown in 100 μg/ml ampicillin (Sigma)–Luria Broth (Conda).

CHO-S cells expressing peptide–I- A β g 7 –Fc-hole/GFP and I- A α d –Fc-knob/CFP were grown in a fed-batch culture (up to 1 L) in a shaking incubator (37°C and 8% CO₂) for 14 days with a temperature shift to 34°C when cell density reached 5 × 10⁶ cells/ml. At day 14 or when cell viability dropped below 60%, cell culture supernatants were harvested and secreted pMHCII heterodimers purified on a protein G affinity column, as previously described (14).

pMHCII monomers (50 μM) were biotinylated overnight with 10 mM biotin-protein ligase (Avidity) in 50 mM bicine buffer (pH 8.3) with 10 mM magnesium acetate, 10 mM ATP, and 5 μM d-biotin. Biotinylation reactions were buffer exchanged with a PD-10 desalting column (GE Healthcare) followed by purification of biotinylated heterodimers with a Monomeric Avidin Kit (Pierce). Biotinylated fractions were dialyzed against 20 mM Tris-HCl buffer pH 8.0 by ultrafiltration using Amicon Ultra-15 (Millipore). Biotinylated heterodimers were tetramerized at room temperature with Streptavidin-Phycoerythrin (PE) (Life Technologies) at a 4:1 molar ratio.

JurMA cells and 4.1-NOD.Rag2 ^–/– splenocytes (after red blood cell lysis) were stained with 400 nM of PE-labeled pMHCII tetramers for 1 h at 37°C in 2% FBS-PBS. Splenocytes were further stained for 20 min at RT with an antibody mix containing Pacific Blue-labeled anti-CD4 (BioLegend), FITC-labeled anti-CD8 (eBioscience), and APC-labeled anti-B220 (BD) (all at 1:100 dilution). The cells were washed twice and analyzed on a Fortessa 5L Flow cytometer.

The HEK-293T-I-A^g7 cell line was generated by sequential transduction of HEK-293T (ATCC, #CRL-11268) cells with lentivirus encoding for I- A α d / C F P , I- A β g 7 / G F P , HLA-DM_α/β/GFP, and human CD74 (invariant chain -Ii-) fused to human CD80 (hCD80) with sequential flow cytometry-based cell-sorting guided by CFP fluorescence (reporting I- A α d expression), GFP (reporting DM_α/β expression), anti-hCD80 (reporting Ii, expression), and anti-I-A^g7 (reporting I- A α d /I- A β g 7 heterodimer expression).

4.1- and BDC2.5-JurMA cells [derived from the TCR_β-null Jurkat cell line J.RT3-T3.5 (ATCC) and carrying a NFAT-driven luciferase reporter] were generated by transducing JurMA cells with retroviruses encoding monocistronic, P2A-tethered TCR_α-TCR_β chains.

Autoantigen expression in artificial APCs was driven by transgenes encoding human invariant chain (Ii_1-80)-β-cell autoantigen-BDC2.5 mimotope (HHPIWARMDA) fusions, or, in a transgene in which insulin (INS) and islet-amyloid polypeptide (IAPP) coding sequences were fused in-frame having their natural cleavage residues mutated to alanine. These transgenes (Ii_1-80-β-cell autoantigen-BDC2.5mi and Ii_1-80-INS-IAPP-BDC2.5mi) were cloned into a eukaryotic expression vector, under the control of a CMV promoter (pCMV).

Plasmids encoding I-A^g7-binding peptides eluted from the CIITA transgenic insulinoma NIT-1 β-cell line (expressing I-A^g7 on the cell surface) (15) or Tandem HIPs (5) were produced by VectorBuilder.

The expression vector encoding for I- A β g 7 (pCMV–GS linker–I- A β g 7 ) was PCR-linearized at the N-terminal end of I- A β g 7 with the primers “Backbone-F” (5’-TCGACAGCGACGTGGGCGAG-3’) and “Backbone-R” (5’-TCCGAGGCTGCTAGCGACCAG-3’). Two additional primers “Insert-F” (5’-GCTAGCAGCCTCGGAGGTNNKCGANNKNNKCTANNKGTANNKNNKGAGGGAGGTGGAGGCGGCTCAGGAGGTGGAAGCGGCGGCTCTGGAGACTCCGAAAGGCATTTCG-3’) and “Insert-Rev” (5’-CCACGTCGCTGTCGAAGCGC-3’) were used to amplify a 250-bp sequence, encoding the amino acid sequence XRXXLXVXXE–GGGGGSGGGSGGS, flanked by a 15-bp long nucleotide sequence that was homologous to the desired integration site in the linearized pCMV-I- A β g 7 product described above. The variable amino acid positions encoded in the epitope-coding nucleotide sequence of this 250-bp fragment used codons with random nucleotides (NNK, where N = any nucleotide, K = T/G), allowing the balanced representation of the 20 possible amino acid residues while minimizing the introduction of stop codons.

PCR-amplified products were digested with DpnI (NEB) to remove potential contaminating plasmid, and DNA fragments were purified by GeneElute Gel extraction kit (Sigma) following the manufacturer’s instructions. The circular plasmid library was generated via a seamless recombination reaction between the linearized pCMV–I- A β g 7 vector and the combinatorial 250-bp sequence, catalyzed by the In-fusion HD reagent (Takara) using 300 ng of vector and a 1:3 vector:insert molar ratio. After pooling three In-fusion reactions, plasmid DNA was precipitated, resuspended in ultrapure water and used to transform electrocompetent DH10B bacteria (ThermoFisher) (exponential protocol; 2 KV, 250 Ω and 25 µF). The plasmid library was then tittered and pools of 500 colony-forming units (CFU)/well grown overnight at 37°C in U-bottomed 96-well plates containing 100 μL of LB media supplemented with 100 μg/ml of ampicillin, in a bacterial shaker at 200 rpm. The bacterial library was preserved as a glycerol stock at −80°C.

Cells were pelleted and lysed with Cell Culture Lysis Reagent (Promega) for 20 min. Next, 30 μl of lysate/well was transferred to white-opaque flat-bottom 96-well plates (Greiner). Luciferase activity in relative light units (RLUs) was determined after injection of 100 μl/well of luciferase assay reagent (Promega), with a delay of 2 s and acquisition for 10 s with a GloMax^® 96 Microplate Luminometer (Promega).

Splenic CD4+ T cells from 4.1-NOD.Rag2^–/– or NOD mice were purified with CD4 (LT34) micro-beads (Miltenyi) and activated for 48 h with Dynabeads™ Mouse T-Activator CD3/CD28 (Gibco) in complete T-cell medium (RPMI-1640) (Hyclone) supplemented with 50 µM 2-mercaptoethanol (Sigma), 10% FBS, 2 mM L-glutamine, 1 mM sodium pyruvate, 1% penicillin/streptomycin, and 30 IU/ml hIL-2 (R&D). After stimulation, dynabeads were removed and expansion was maintained for 6–7 days with 100 IU/ml hIL-2 in complete T-cell medium.

Expanded cells were harvested and cocultured, in a U-bottom 96-well plate, with freshly isolated and red blood cell-lysed 10⁵ NOD splenocytes and 0.0001–10 µg/ml peptide in complete T-cell medium supplemented with 30 IU/ml of hIL-2. Cell culture supernatants were collected after 48 h and IFNγ concentrations were determined by ELISA using ELISA MAX™ Deluxe Set Mouse IFN-γ (BioLegend) according to the manufacturer’s instructions. Synthetic soluble peptides (purity >80%) were obtained from GenScript^®.

We generated an HEK-293T-based artificial APC line (16) expressing transgenic I- A α d , I- A β g 7 , H2-DM_α/β, and invariant chain (CD74 or Ii) to enable the processing and presentation of transiently expressed antigens in the context of I-A^g7 ( Figure 1A ). We also generated a JurMA cell-based responder T-cell line (17) expressing mCD4, and the alpha and beta chains of the 4.1-TCR or the BDC2.5-TCR [as a positive control, recognizing the BDC2.5mi mimotope in the context of I-A^g7 (18)]. This T-cell line carries a NFAT-driven luciferase reporter that is expressed upon TCR ligation ( Figure 1A ).

We first sought to use this approach to ascertain the ability of the 4.1-TCR to recognize previously described T1D-relevant autoantigens (19, 20), including pro-insulin (INS), chromogranin A (ChrA), islet amyloid polypeptide (IAPP), islet-specific glucose-6-phosphatase catalytic subunit-related protein (IGRP), glutamic acid decarboxylase 65 (GAD65), zinc transporter 8 (ZnT8), and islet tyrosine phosphatase 2 (I-A2). This was done by transiently transfecting the artificial APC described above with CMV promoter-driven expression vectors encoding fusions of the first 80 amino acids of murine invariant chain_1-80 (Ii_1-80) to each of the above autoantigens (16) (to promote antigen processing and presentation) as well as a BDC2.5mi mimotope-coding sequence at the C-terminal end as an internal positive control (pCMV-Ii-autoantigen-2.5mi) ( Figure 1A ). Whereas all of the antigen-encoding plasmids described above triggered luciferase activity in BDC2.5-JurMA cells (via recognition of the C-terminal BDC2.5mi epitope encoded in the various plasmids, in the context of I-A^g7), none triggered luciferase activity in 4.1-JurMA cells ( Figure 1B ).

4.1-JurMA cells also failed to respond to artificial APCs transfected with a eukaryotic expression library (100 plasmids/pool) encoding cDNAs cloned from the murine insulinoma beta cell line NIT-1 (10⁵ cDNAs in total, including a plasmid encoding ovalbumin fused to BDC2.5mi as a positive control –pCMV-OVA-BDC2.5mi–) (data not shown). Likewise, 4.1-JurMA cells failed to respond to a single expression vector encoding Ii_1-80 fused to a tandem sequence encoding 21 different peptides eluted from I-A^g7 molecules expressed on the insulinoma cell line NIT-1 ( Supplementary Table S1 ). These peptides were derived from Synaptotagmin 11, Teneurin Transmembrane Protein 1, Neuromodulin, Synapse-associated protein, NCAM, Secretogranin-3, Axonal Transporter of Synaptic Vesicles, Beta-site APP-Cleaving Enzyme, Synaptic Cell Adhesion Molecule, Secretogranin-2 (two peptides), Chromogranin A, NMDA 2A, Gamma-aminobutyric acid receptor-associated protein, Carboxypeptidase H, Lisch 7, Amyloid beta A4 (3 peptides), Solute Carrier Family 12 Member 7, and Reticulon 4 Receptor-Like 1 (15). As expected, the BDC-2.5-JurMA cells responded efficiently in this assay ( Figure 1C ) but not 4.1-JurMA cells.

We next considered the possibility that the 4.1-TCR might recognize a HIP. We therefore generated an expression construct encoding li_1-80 fused to 16 previously published HIP-sequences carrying left and right arms derived from Insulin C and Chromogranin A, Amylin, IAPP, or Secretogranin 2, respectively (5). We also tested two additional expression vectors encoding 63 potential (unpublished) HIPs ( Supplementary Figure S1 ). Whereas the BDC2.5-JurMA cell line responded to all these HIP-encoding constructs and the OVA-BDC2.5mi fusion positive control, 4.1-JurMA cells did not respond to any ( Figure 1D ).

We next sought to develop and test a more systematic, sensitive, higher-throughput approach for T-cell epitope discovery, based on combinatorial epitope display, which could be used in combination with the artificial APC/TCR-transduced JurMA cell system described above.

We cloned combinatorial peptide-coding sequences immediately upstream of a flexible 13-residue linker- and I- A β g 7 −coding sequences into an expression vector. The peptide–linker fusions carrying short arms of homology with the desired plasmid integration site were generated by PCR using degenerate primers and then cloned in-frame (using the In-fusion HD reagent) without intervening nucleotides immediately upstream of I- A β g 7 in the PCR-linearized expression vector with homologous integration sites ( Figure 2A ). We fixed the P1, P4, P6, and P9 I-A^g7-anchor residues (P1=R, P4=V, P6=L, and P9=E) in the peptide-coding sequence, but used degenerate “NNK” codon sequences for potential TCR contact residues (P-1, P2, P3, P5, P7, and P8), where “N” can be any nucleotide and “K” is either a G or a T. By reducing codon degeneracy, we sought to maximize the likelihood that each amino acid will be represented at each open position, while minimizing the chances of introducing stop codons. These plasmids were then transfected into HEK-293T cells constitutively expressing an I- A α d transgene (293.I- A α d ). In this design, the various peptide–linker–I- A β g 7 protein molecules were expected to heterodimerize with the transgenic I- A α d chain to form stable peptide-loaded MHCII complexes on the artificial APC’s surface.

We first tested the sensitivity of this approach by measuring the ability of HEK-293T. I- A α d cells transfected with decreasing ratios of BDC2.5mi-I- A β g 7 : Ins_10-23- I- A β g 7 plasmids (from 1:0 to 1:4,096 and 0:1, respectively) to activate BDC2.5-JurMA cells. As shown in Figure 2B , dilutions lower than 1:1,024 elicited clearly detectable responses.

Having defined the sensitivity threshold of this approach, and to maximize coverage without compromising sensitivity, we screened pools of 500 peptide–linker–I- A β g 7 clones per well distributed in ten 96-well plates, thus representing a total of 5 × 10⁵ peptides ( Figure 2C ). As cutoff luciferase activity values, we used the average of the normalized relative light units for all the screened pools plus three standard deviations. Pools eliciting responses above the cutoff value (each containing 500 clones) were deconvoluted by subcloning and re-screening until single positive clones were identified in each pool. Of the 17/960 positive pools that were identified, we deconvoluted the top 10 ( Figure 2C ). A limitation of these stringent experimental conditions is that it allows for the identification of epitopes with the highest MHC/TCR-binding affinities/avidities, potentially excluding their lower-avidity counterparts. Furthermore, although not investigated herein, it is possible that screening of bacterial pools containing fewer than 500 clones/pool might increase the sensitivity of this approach.

We next compared the agonistic activity of the top 10 peptide–linker–I- A β g 7 plasmids and the corresponding synthetic peptides. As expected, sensitivity was greater with the plasmids than with the peptides, as exogenously added peptides must be able to efficiently displace endogenous, HEK-293T-derived epitopes from surface I-A^g7, to elicit cognate T-cell responses. Nevertheless, these experiments allowed the validation of two homologous peptide sequences (A5D and B2B), differing only at positions P-1 and P3, as being strong 4.1-TCR agonists ( Figure 2D ). Of these, B2B had slightly higher agonistic activity on 4.1-JurMA cells than A5D ( Figure 3A ) and was therefore chosen for further experimentation.

This approach deviates significantly from the strategy used to identify the BDC2.5mi mimotope for the BDC2.5-TCR (18, 21), which involves the screening of a peptide-I-A^g7 library expressed on baculovirus-infected fly cells by several rounds of FACS-based sorting using soluble TCR tetramers. In the approach described herein, we combine the use of an easy to maintain, highly sensitive reporter cell line with an efficient artificial antigen presentation system that employs the highly transfectable HEK-293T cell line. These features make this screening approach more efficient, higher throughput, and less technically demanding, as expression of soluble TCRs is not a trivial undertaking.

We next sought to identify the B2B I-A^g7 binding register(s) that most efficiently trigger(s) 4.1-TCR signaling. To do this, we used plate-bound, knob-into-hole-based B2B-linker- I A α β g 7 monomers secreted from transiently-transduced HEK-293T cells. As described previously, this design results in increased production yields and pMHCII structural stability than those using leucine zipper-heterodimerization domains (14, 22). Although the peptide–linker–I- A β g 7 library that was used to identify the B2B mimotope used four pre-defined anchor residues, it remained possible that the B2B sequence was preferentially displayed to the 4.1-TCR on alternative binding registers, such as those in which B2B uses the negatively charged Glu or Asp at P7 and P8, respectively, to anchor itself on I-A^g7’s pocket 9. This was addressed by producing three different B2B-linker-I-A^g7 pMHCIIs in which the B2B sequence was forced to bind to I-A^g7 on three different registers via the strategic introduction of register-fixing Cysteine residues on the I- A α g 7 chain and the C-terminus of B2B (cys-trap) (14, 22). We then compared the luciferase activity in 4.1-JurMA cells challenged with plate-bound monomers.

As shown in Figure 3B , the construct displaying B2B on register 3 (the Glu at B2B’s P9 docking into I-A^g7’s pocket 9) had superior agonistic activity than those displaying it on registers 1 and 2 (B2B docked into I-A^g7’s P9 pocket via the Glu at P7 and the Asp at P8, respectively). This was subsequently confirmed by staining 4.1-JurMA cells with pMHCII tetramers displaying B2B on register 3 ( Figure 3C ).

To further validate P2, P3, P5, P7, and P8 as 4.1-TCR-contact residues in B2B anchored on I-A^g7 via register 3, we compared the luciferase activity of B2B-like peptides carrying alanine substitutions at these positions. These experiments revealed strong independent contributions of each of these residues to the agonistic activity of B2B ( Figure 3D ). Of note, replacement of the Asp at P8 for a Glu reduced B2B’s agonistic activity, suggesting an important contribution for Asp at P8 on 4.1-TCR agonism.

Database searches using a degenerate amino acid sequence carrying B2B’s 4.1-TCR contact residues and permissive I-A^g7 anchor residues did not yield naturally occurring beta cell-specific autoantigenic targets, suggesting the possibility that the 4.1-TCR might recognize a post-translationally generated and/or modified epitope, such as a HIP. We thus looked for potential HIPs sharing B2B’s 4.1-TCR contact motif XLGXEXE(D/E)X. We focused on HIPs identified in beta cell granules via mass spectrometry (MS) (5, 23, 24), as well as potential new HIPs in which the candidate left and right arms had been previously documented via MS (25). Briefly, we fused these MS-documented left and right peptides in silico and performed a motif search fitting the motif described above, also incorporating in the search other conserved residues for each position. The potential left and right arms included peptides derived from ChrA, Secretogranin-1 (Scg1), Secretogranin-2 (Scg2), Secretogranin-3 (Scg3), Secretogranin-5 (Scg5), IAPP, Ins1, Ins2, Pancreatic Prohormone (Ppy), Peptide YY (Pyy), Urocortin-3 (Ucn3), Neuroendocrine Convertase 1 (Pcsk1), and Neuroendocrine Convertase 2 (Pcsk2). We also considered HIPs fitting the 4.1-TCR-recognition motif upon deamidation of glutamine residues.

Using the above criteria, we identified 13 potential HIP agonists for the 4.1-TCR ( Figure 4A and Supplementary Table S2 ), two of which have been previously identified via MS (HIP 55: QLELGGEVEDPQV and HIP 30: LQTLALEVEDPQV) (25). In all these HIPs, the left arm is donated by the Insulin C-peptide, truncated at seven different carboxyterminal residues (ranging from residues 69 to 85), whereas the right arm corresponds to five different naturally occurring proteolytic products of either ChgA (n = 4) or InsC_57-63 ( Figure 4A and Supplementary Table S2 ). None of the other hybrid peptides that were probed, composed of all random combinations of in silico generated right and left arms from prevalent polypeptides in beta cell granules and crinosomes (including, pro-Ins1, pro-Ins2, ChrgA, IAPP, and Scg1) including variants carrying sequential truncations of one amino acid residue (a total of 7.5 million combinations), conformed to the agonistic motif described above.

Seven of these HIPs (HIPs 15, 18, 30, 30 Q10E, 32, 32 Q9E, and 39) promoted robust IFNγ secretion by anti-CD3/anti-CD28 mAb-expanded 4.1-CD4+ splenocytes ( Figure 4B ), albeit with different functional avidities ( Figure 4C ). Motif-containing HIPs D1, D2, 40, 43, and 55 lacked agonistic activity in these assays, presumably due to disruptive effects of residues at degenerate positions on 4.1-TCR engagement. Interestingly, the deamidated forms of HIPs 30 (HIP 30 Q10E) and 32 (HIP 32 Q9E) had superior agonistic activity than their wild-type counterparts ( Figure 4C ). We note that the deamidated form of the right arm of HIP 32 has been described (25). These two pro-agonistic post-translational modifications resulted in the introduction of highly favorable I-A^g7-anchor residues for I-A^g7’s pocket 9 (E vs. Q).

Experiments using pMHCII tetramers for the seven HIPs indicated that these HIPs can be subclassified into two potential subsets with differences in functional avidity mirrored by differences in physical pMHCII-binding avidity to 4.1-JurMA cells ( Figure 5A ) and CD4+ T cells from 4.1-NOD.Rag2^−/− mice ( Figure 5B ). One of these subsets includes peptides such as HIP 39 and HIP 32 Q9E that exhibit high overall functional and physical avidity. The other subset, including the peptides HIP 15, HIP 18, HIP 30, HIP 30 Q10E, and HIP 32, is recognized with low functional and/or physical avidity. Intrinsic particularities of the different peptides within these two subgroups may account for an imperfect correlation in physical vs. functional avidity. For example, HIP 32 has a Q at P9, which is less favorable than a negatively charged residue at P9 (D in HIP 15). It is possible that, at high concentrations of peptide, the functional assay favors a response against HIP 32 over HIP 15, whereas the presence of D at P9 in HIP15 enhances binding to I-A^g7, thus favoring tetramer staining. Furthermore, although HIP 30 Q10E was more potent than its wild-type counterpart HIP 30 in the IFNγ secretion assay, HIP 30 I-A^g7 tetramers displayed higher physical binding avidity to the 4.1-TCR. This observation suggests that the deamidation Q10E in HIP 30 results in the introduction of a stronger I-A^g7 anchor residue, shifting the register from agonistic to non-agonistic. This change would be expected to have more impact on tetramer staining (requiring a unique register binding) than in the IFNγ secretion assay where register multiplicity occurs at high concentration of soluble peptide.